From fcd21256780fd21c55e72e9338b3992c60db22dc Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Tue, 4 Apr 2023 15:47:53 +0200
Subject: prepare for strategies

---
 include/afl-mutations.h | 992 ++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 992 insertions(+)
 create mode 100644 include/afl-mutations.h

(limited to 'include')

diff --git a/include/afl-mutations.h b/include/afl-mutations.h
new file mode 100644
index 00000000..43b7927d
--- /dev/null
+++ b/include/afl-mutations.h
@@ -0,0 +1,992 @@
+/* Implementation of afl havoc mutation to be used in AFL++ custom mutators and
+   partially in afl-fuzz itself.
+
+   How to use:
+
+   #include "afl-mutations.h"  // needs afl-fuzz.h
+
+   u32 afl_mutate(afl_state_t *afl, u8 *buf, u32 len, u32t steps, bool is_text,
+                  bool is_exploration, u8 *splice_buf, u32 splice_len);
+
+   Returns:
+     u32 - the length of the mutated data return in *buf. 0 = error
+   Parameters:
+     afl_state_t *afl - the *afl state pointer
+     u8 *buf - the input buffer to mutate which will be mutated into.
+               NOTE: must be of MAX_FILE size!
+     u32 len - the length of the input
+     u32 steps - how many mutations to perform on the input
+     bool is_text - is the target expecting text inputs
+     bool is_exploration - mutate for exploration mode (instead of exploitation)
+     splice_buf - a buffer from another corpus item to splice with.
+                  If NULL then no splicing
+     splice_len - the length of the splice buffer. If 0 then no splicing
+*/
+
+#ifndef _ANDROID_ASHMEM_H
+  #define AFL_MUTATIONS_H
+
+  #include <stdbool.h>
+  #include "afl-fuzz.h"
+
+  #define MUT_STRATEGY_ARRAY_SIZE 256
+
+enum {
+
+  /* 00 */ MUT_FLIPBIT,
+  /* 01 */ MUT_INTERESTING8,
+  /* 02 */ MUT_INTERESTING16,
+  /* 03 */ MUT_INTERESTING16BE,
+  /* 04 */ MUT_INTERESTING32,
+  /* 05 */ MUT_INTERESTING32BE,
+  /* 06 */ MUT_ARITH8_,
+  /* 07 */ MUT_ARITH8,
+  /* 08 */ MUT_ARITH16_,
+  /* 09 */ MUT_ARITH16BE_,
+  /* 10 */ MUT_ARITH16,
+  /* 11 */ MUT_ARITH16BE,
+  /* 12 */ MUT_ARITH32_,
+  /* 13 */ MUT_ARITH32BE_,
+  /* 14 */ MUT_ARITH32,
+  /* 15 */ MUT_ARITH32BE,
+  /* 16 */ MUT_RAND8,
+  /* 17 */ MUT_CLONE_OVERWRITE,
+  /* 18 */ MUT_CLONE_INSERT,
+  /* 19 */ MUT_OVERWRITE_COPY,
+  /* 20 */ MUT_OVERWRITE_FIXED,
+  /* 21 */ MUT_BYTEADD,
+  /* 22 */ MUT_BYTESUB,
+  /* 23 */ MUT_FLIP8,
+  /* 24 */ MUT_SWITCH,
+  /* 25 */ MUT_DEL,
+  /* 26 */ MUT_SHUFFLE,
+  /* 27 */ MUT_DELONE,
+  /* 28 */ MUT_INSERTONE,
+  /* 29 */ MUT_ASCIINUM,
+  /* 30 */ MUT_NEG,
+  /* 31 */ MUT_INSERTASCIINUM,
+  /* 32 */ MUT_EXTRA_OVERWRITE,
+  /* 33 */ MUT_EXTRA_INSERT,
+  /* 34 */ MUT_AUTO_EXTRA_OVERWRITE,
+  /* 35 */ MUT_AUTO_EXTRA_INSERT,
+  /* 36 */ MUT_SPLICE_OVERWRITE,
+  /* 37 */ MUT_SPLICE_INSERT,
+
+  MUT_MAX
+
+};
+
+unsigned int mutation_strategy_exploration_text[MUT_STRATEGY_ARRAY_SIZE] = {};
+unsigned int mutation_strategy_exploration_binary[MUT_STRATEGY_ARRAY_SIZE] = {};
+unsigned int mutation_strategy_exploitation_text[MUT_STRATEGY_ARRAY_SIZE] = {};
+unsigned int mutation_strategy_exploitation_binary[MUT_STRATEGY_ARRAY_SIZE] =
+    {};
+
+unsigned int afl_mutate(afl_state_t *, unsigned char *, unsigned int,
+                        unsigned int, bool, bool, unsigned char *,
+                        unsigned int);
+u32          choose_block_len(afl_state_t *, u32);
+
+/* Helper to choose random block len for block operations in fuzz_one().
+   Doesn't return zero, provided that max_len is > 0. */
+
+inline u32 choose_block_len(afl_state_t *afl, u32 limit) {
+
+  u32 min_value, max_value;
+  u32 rlim = MIN(afl->queue_cycle, (u32)3);
+
+  if (unlikely(!afl->run_over10m)) { rlim = 1; }
+
+  switch (rand_below(afl, rlim)) {
+
+    case 0:
+      min_value = 1;
+      max_value = HAVOC_BLK_SMALL;
+      break;
+
+    case 1:
+      min_value = HAVOC_BLK_SMALL;
+      max_value = HAVOC_BLK_MEDIUM;
+      break;
+
+    default:
+
+      if (likely(rand_below(afl, 10))) {
+
+        min_value = HAVOC_BLK_MEDIUM;
+        max_value = HAVOC_BLK_LARGE;
+
+      } else {
+
+        min_value = HAVOC_BLK_LARGE;
+        max_value = HAVOC_BLK_XL;
+
+      }
+
+  }
+
+  if (min_value >= limit) { min_value = 1; }
+
+  return min_value + rand_below(afl, MIN(max_value, limit) - min_value + 1);
+
+}
+
+unsigned int afl_mutate(afl_state_t *afl, unsigned char *buf, unsigned int len,
+                        unsigned int steps, bool is_text, bool is_exploration,
+                        unsigned char *splice_buf, unsigned int splice_len) {
+
+  if (!buf || !len) { return 0; }
+
+  u32                  *mutation_array;
+  static unsigned char *tmp_buf = NULL;
+
+  if (!tmp_buf) {
+
+    if ((tmp_buf = malloc(MAX_FILE)) == NULL) { return 0; }
+
+  }
+
+  if (is_text) {
+
+    if (is_exploration) {
+
+      mutation_array = (unsigned int *)&mutation_strategy_exploration_text;
+
+    } else {
+
+      mutation_array = (unsigned int *)&mutation_strategy_exploitation_text;
+
+    }
+
+  } else {
+
+    if (is_exploration) {
+
+      mutation_array = (unsigned int *)&mutation_strategy_exploration_binary;
+
+    } else {
+
+      mutation_array = (unsigned int *)&mutation_strategy_exploitation_binary;
+
+    }
+
+  }
+
+  for (unsigned int step = 0; step < steps; ++step) {
+
+  retry_havoc_step:
+
+    u32 r = rand_below(afl, MUT_STRATEGY_ARRAY_SIZE), item;
+
+    switch (mutation_array[r]) {
+
+      case MUT_FLIPBIT: {
+
+        /* Flip a single bit somewhere. Spooky! */
+        u8  bit = rand_below(afl, 8);
+        u32 off = rand_below(afl, len);
+        buf[off] ^= 1 << bit;
+
+        break;
+
+      }
+
+      case MUT_INTERESTING8: {
+
+        /* Set byte to interesting value. */
+
+        item = rand_below(afl, sizeof(interesting_8));
+        buf[rand_below(afl, len)] = interesting_8[item];
+        break;
+
+      }
+
+      case MUT_INTERESTING16: {
+
+        /* Set word to interesting value, little endian. */
+
+        if (unlikely(len < 2)) { break; }  // no retry
+
+        item = rand_below(afl, sizeof(interesting_16) >> 1);
+        *(u16 *)(buf + rand_below(afl, len - 1)) = interesting_16[item];
+
+        break;
+
+      }
+
+      case MUT_INTERESTING16BE: {
+
+        /* Set word to interesting value, big endian. */
+
+        if (unlikely(len < 2)) { break; }  // no retry
+
+        item = rand_below(afl, sizeof(interesting_16) >> 1);
+        *(u16 *)(buf + rand_below(afl, len - 1)) = SWAP16(interesting_16[item]);
+
+        break;
+
+      }
+
+      case MUT_INTERESTING32: {
+
+        /* Set dword to interesting value, little endian. */
+
+        if (unlikely(len < 4)) { break; }  // no retry
+
+        item = rand_below(afl, sizeof(interesting_32) >> 2);
+        *(u32 *)(buf + rand_below(afl, len - 3)) = interesting_32[item];
+
+        break;
+
+      }
+
+      case MUT_INTERESTING32BE: {
+
+        /* Set dword to interesting value, big endian. */
+
+        if (unlikely(len < 4)) { break; }  // no retry
+
+        item = rand_below(afl, sizeof(interesting_32) >> 2);
+        *(u32 *)(buf + rand_below(afl, len - 3)) = SWAP32(interesting_32[item]);
+
+        break;
+
+      }
+
+      case MUT_ARITH8_: {
+
+        /* Randomly subtract from byte. */
+
+        item = 1 + rand_below(afl, ARITH_MAX);
+        buf[rand_below(afl, len)] -= item;
+        break;
+
+      }
+
+      case MUT_ARITH8: {
+
+        /* Randomly add to byte. */
+
+        item = 1 + rand_below(afl, ARITH_MAX);
+        buf[rand_below(afl, len)] += item;
+        break;
+
+      }
+
+      case MUT_ARITH16_: {
+
+        /* Randomly subtract from word, little endian. */
+
+        if (unlikely(len < 2)) { break; }  // no retry
+
+        u32 pos = rand_below(afl, len - 1);
+        item = 1 + rand_below(afl, ARITH_MAX);
+        *(u16 *)(buf + pos) -= item;
+
+        break;
+
+      }
+
+      case MUT_ARITH16BE_: {
+
+        /* Randomly subtract from word, big endian. */
+
+        if (unlikely(len < 2)) { break; }  // no retry
+
+        u32 pos = rand_below(afl, len - 1);
+        u16 num = 1 + rand_below(afl, ARITH_MAX);
+        *(u16 *)(buf + pos) = SWAP16(SWAP16(*(u16 *)(buf + pos)) - num);
+
+        break;
+
+      }
+
+      case MUT_ARITH16: {
+
+        /* Randomly add to word, little endian. */
+
+        if (unlikely(len < 2)) { break; }  // no retry
+
+        u32 pos = rand_below(afl, len - 1);
+        item = 1 + rand_below(afl, ARITH_MAX);
+        *(u16 *)(buf + pos) += item;
+
+        break;
+
+      }
+
+      case MUT_ARITH16BE: {
+
+        /* Randomly add to word, big endian. */
+
+        if (unlikely(len < 2)) { break; }  // no retry
+
+        u32 pos = rand_below(afl, len - 1);
+        u16 num = 1 + rand_below(afl, ARITH_MAX);
+        *(u16 *)(buf + pos) = SWAP16(SWAP16(*(u16 *)(buf + pos)) + num);
+
+        break;
+
+      }
+
+      case MUT_ARITH32_: {
+
+        /* Randomly subtract from dword, little endian. */
+
+        if (unlikely(len < 4)) { break; }  // no retry
+
+        u32 pos = rand_below(afl, len - 3);
+        item = 1 + rand_below(afl, ARITH_MAX);
+        *(u32 *)(buf + pos) -= item;
+
+        break;
+
+      }
+
+      case MUT_ARITH32BE_: {
+
+        /* Randomly subtract from dword, big endian. */
+
+        if (unlikely(len < 4)) { break; }  // no retry
+
+        u32 pos = rand_below(afl, len - 3);
+        u32 num = 1 + rand_below(afl, ARITH_MAX);
+        *(u32 *)(buf + pos) = SWAP32(SWAP32(*(u32 *)(buf + pos)) - num);
+
+        break;
+
+      }
+
+      case MUT_ARITH32: {
+
+        /* Randomly add to dword, little endian. */
+
+        if (unlikely(len < 4)) { break; }  // no retry
+
+        u32 pos = rand_below(afl, len - 3);
+        item = 1 + rand_below(afl, ARITH_MAX);
+        *(u32 *)(buf + pos) += item;
+
+        break;
+
+      }
+
+      case MUT_ARITH32BE: {
+
+        /* Randomly add to dword, big endian. */
+
+        if (unlikely(len < 4)) { break; }  // no retry
+
+        u32 pos = rand_below(afl, len - 3);
+        u32 num = 1 + rand_below(afl, ARITH_MAX);
+        *(u32 *)(buf + pos) = SWAP32(SWAP32(*(u32 *)(buf + pos)) + num);
+
+        break;
+
+      }
+
+      case MUT_RAND8: {
+
+        /* Just set a random byte to a random value. Because,
+           why not. We use XOR with 1-255 to eliminate the
+           possibility of a no-op. */
+
+        u32 pos = rand_below(afl, len);
+        item = 1 + rand_below(afl, 255);
+        buf[pos] ^= item;
+        break;
+
+      }
+
+      case MUT_CLONE_OVERWRITE: {
+
+        if (likely(len + HAVOC_BLK_XL < MAX_FILE)) {
+
+          /* Clone bytes. */
+
+          u32 clone_len = choose_block_len(afl, len);
+          u32 clone_from = rand_below(afl, len - clone_len + 1);
+          u32 clone_to = rand_below(afl, len);
+
+          /* Head */
+
+          memcpy(tmp_buf, buf, clone_to);
+
+          /* Inserted part */
+
+          memcpy(tmp_buf + clone_to, buf + clone_from, clone_len);
+
+          /* Tail */
+          memcpy(tmp_buf + clone_to + clone_len, buf + clone_to,
+                 len - clone_to);
+
+          len += clone_len;
+          memcpy(buf, tmp_buf, len);
+
+        } else if (unlikely(len < 8)) {
+
+          break;
+
+        } else {
+
+          goto retry_havoc_step;
+
+        }
+
+        break;
+
+      }
+
+      case MUT_CLONE_INSERT: {
+
+        if (likely(len + HAVOC_BLK_XL < MAX_FILE)) {
+
+          /* Insert a block of constant bytes (25%). */
+
+          u32 clone_len = choose_block_len(afl, HAVOC_BLK_XL);
+          u32 clone_to = rand_below(afl, len);
+          u32 strat = rand_below(afl, 2);
+          u32 clone_from = clone_to ? clone_to - 1 : 0;
+          item = strat ? rand_below(afl, 256) : buf[clone_from];
+
+          /* Head */
+
+          memcpy(tmp_buf, buf, clone_to);
+
+          /* Inserted part */
+
+          memset(tmp_buf + clone_to, item, clone_len);
+
+          /* Tail */
+          memcpy(tmp_buf + clone_to + clone_len, buf + clone_to,
+                 len - clone_to);
+
+          len += clone_len;
+          memcpy(buf, tmp_buf, len);
+
+        } else if (unlikely(len < 8)) {
+
+          break;
+
+        } else {
+
+          goto retry_havoc_step;
+
+        }
+
+        break;
+
+      }
+
+      case MUT_OVERWRITE_COPY: {
+
+        /* Overwrite bytes with a randomly selected chunk bytes. */
+
+        if (unlikely(len < 2)) { break; }  // no retry
+
+        u32 copy_len = choose_block_len(afl, len - 1);
+        u32 copy_from = rand_below(afl, len - copy_len + 1);
+        u32 copy_to = rand_below(afl, len - copy_len + 1);
+
+        if (likely(copy_from != copy_to)) {
+
+          memmove(buf + copy_to, buf + copy_from, copy_len);
+
+        }
+
+        break;
+
+      }
+
+      case MUT_OVERWRITE_FIXED: {
+
+        /* Overwrite bytes with fixed bytes. */
+
+        if (unlikely(len < 2)) { break; }  // no retry
+
+        u32 copy_len = choose_block_len(afl, len - 1);
+        u32 copy_to = rand_below(afl, len - copy_len + 1);
+        u32 strat = rand_below(afl, 2);
+        u32 copy_from = copy_to ? copy_to - 1 : 0;
+        item = strat ? rand_below(afl, 256) : buf[copy_from];
+        memset(buf + copy_to, item, copy_len);
+
+        break;
+
+      }
+
+      case MUT_BYTEADD: {
+
+        /* Increase byte by 1. */
+
+        buf[rand_below(afl, len)]++;
+        break;
+
+      }
+
+      case MUT_BYTESUB: {
+
+        /* Decrease byte by 1. */
+
+        buf[rand_below(afl, len)]--;
+        break;
+
+      }
+
+      case MUT_FLIP8: {
+
+        /* Flip byte. */
+
+        buf[rand_below(afl, len)] ^= 0xff;
+        break;
+
+      }
+
+      case MUT_SWITCH: {
+
+        if (unlikely(len < 4)) { break; }  // no retry
+
+        /* Switch bytes. */
+
+        u32 to_end, switch_to, switch_len, switch_from;
+        switch_from = rand_below(afl, len);
+        do {
+
+          switch_to = rand_below(afl, len);
+
+        } while (unlikely(switch_from == switch_to));
+
+        if (switch_from < switch_to) {
+
+          switch_len = switch_to - switch_from;
+          to_end = len - switch_to;
+
+        } else {
+
+          switch_len = switch_from - switch_to;
+          to_end = len - switch_from;
+
+        }
+
+        switch_len = choose_block_len(afl, MIN(switch_len, to_end));
+
+        /* Backup */
+
+        memcpy(tmp_buf, buf + switch_from, switch_len);
+
+        /* Switch 1 */
+
+        memcpy(buf + switch_from, buf + switch_to, switch_len);
+
+        /* Switch 2 */
+
+        memcpy(buf + switch_to, tmp_buf, switch_len);
+
+        break;
+
+      }
+
+      case MUT_DEL: {
+
+        /* Delete bytes. */
+
+        if (unlikely(len < 2)) { break; }  // no retry
+
+        /* Don't delete too much. */
+
+        u32 del_len = choose_block_len(afl, len - 1);
+        u32 del_from = rand_below(afl, len - del_len + 1);
+        memmove(buf + del_from, buf + del_from + del_len,
+                len - del_from - del_len);
+        len -= del_len;
+
+        break;
+
+      }
+
+      case MUT_SHUFFLE: {
+
+        /* Shuffle bytes. */
+
+        if (unlikely(len < 4)) { break; }  // no retry
+
+        u32 len = choose_block_len(afl, len - 1);
+        u32 off = rand_below(afl, len - len + 1);
+
+        for (u32 i = len - 1; i > 0; i--) {
+
+          u32 j;
+          do {
+
+            j = rand_below(afl, i + 1);
+
+          } while (unlikely(i == j));
+
+          unsigned char temp = buf[off + i];
+          buf[off + i] = buf[off + j];
+          buf[off + j] = temp;
+
+        }
+
+        break;
+
+      }
+
+      case MUT_DELONE: {
+
+        /* Delete bytes. */
+
+        if (unlikely(len < 2)) { break; }  // no retry
+
+        /* Don't delete too much. */
+
+        u32 del_len = 1;
+        u32 del_from = rand_below(afl, len - del_len + 1);
+        memmove(buf + del_from, buf + del_from + del_len,
+                len - del_from - del_len);
+
+        len -= del_len;
+
+        break;
+
+      }
+
+      case MUT_INSERTONE: {
+
+        if (unlikely(len < 2)) { break; }  // no retry
+
+        u32 clone_len = 1;
+        u32 clone_to = rand_below(afl, len);
+        u32 strat = rand_below(afl, 2);
+        u32 clone_from = clone_to ? clone_to - 1 : 0;
+        item = strat ? rand_below(afl, 256) : buf[clone_from];
+
+        /* Head */
+
+        memcpy(tmp_buf, buf, clone_to);
+
+        /* Inserted part */
+
+        memset(tmp_buf + clone_to, item, clone_len);
+
+        /* Tail */
+        memcpy(tmp_buf + clone_to + clone_len, buf + clone_to, len - clone_to);
+
+        len += clone_len;
+        memcpy(buf, tmp_buf, len);
+
+        break;
+
+      }
+
+      case MUT_ASCIINUM: {
+
+        if (unlikely(len < 4)) { break; }  // no retry
+
+        u32 off = rand_below(afl, len), off2 = off, cnt = 0;
+
+        while (off2 + cnt < len && !isdigit(buf[off2 + cnt])) {
+
+          ++cnt;
+
+        }
+
+        // none found, wrap
+        if (off2 + cnt == len) {
+
+          off2 = 0;
+          cnt = 0;
+
+          while (cnt < off && !isdigit(buf[off2 + cnt])) {
+
+            ++cnt;
+
+          }
+
+          if (cnt == off) {
+
+            if (len < 8) {
+
+              break;
+
+            } else {
+
+              goto retry_havoc_step;
+
+            }
+
+          }
+
+        }
+
+        off = off2 + cnt;
+        off2 = off + 1;
+
+        while (off2 < len && isdigit(buf[off2])) {
+
+          ++off2;
+
+        }
+
+        s64 val = buf[off] - '0';
+        for (u32 i = off + 1; i < off2; ++i) {
+
+          val = (val * 10) + buf[i] - '0';
+
+        }
+
+        if (off && buf[off - 1] == '-') { val = -val; }
+
+        u32 strat = rand_below(afl, 8);
+        switch (strat) {
+
+          case 0:
+            val++;
+            break;
+          case 1:
+            val--;
+            break;
+          case 2:
+            val *= 2;
+            break;
+          case 3:
+            val /= 2;
+            break;
+          case 4:
+            if (likely(val && (u64)val < 0x19999999)) {
+
+              val = (u64)rand_next(afl) % (u64)((u64)val * 10);
+
+            } else {
+
+              val = rand_below(afl, 256);
+
+            }
+
+            break;
+          case 5:
+            val += rand_below(afl, 256);
+            break;
+          case 6:
+            val -= rand_below(afl, 256);
+            break;
+          case 7:
+            val = ~(val);
+            break;
+
+        }
+
+        char buf[20];
+        snprintf(buf, sizeof(buf), "%ld", val);
+        u32 old_len = off2 - off;
+        u32 new_len = strlen(buf);
+
+        if (old_len == new_len) {
+
+          memcpy(buf + off, buf, new_len);
+
+        } else {
+
+          /* Head */
+
+          memcpy(tmp_buf, buf, off);
+
+          /* Inserted part */
+
+          memcpy(tmp_buf + off, buf, new_len);
+
+          /* Tail */
+          memcpy(tmp_buf + off + new_len, buf + off2, len - off2);
+
+          len += (new_len - old_len);
+          memcpy(buf, tmp_buf, len);
+
+        }
+
+        // fprintf(stderr, "AFTER : %s\n", buf);
+        break;
+
+      }
+
+      case MUT_NEG: {
+
+        /* Neg byte. */
+
+        item = rand_below(afl, len);
+        buf[item] = ~buf[item];
+
+        break;
+
+      }
+
+      case MUT_INSERTASCIINUM: {
+
+        u32 len = 1 + rand_below(afl, 8);
+        u32 pos = rand_below(afl, len);
+
+        /* Insert ascii number. */
+        if (unlikely(len < pos + len)) {
+
+          if (unlikely(len < 8)) {
+
+            break;
+
+          } else {
+
+            goto retry_havoc_step;
+
+          }
+
+        }
+
+        u64  val = rand_next(afl);
+        char buf[20];
+        snprintf(buf, sizeof(buf), "%llu", val);
+        memcpy(buf + pos, buf, len);
+
+        break;
+
+      }
+
+      case MUT_EXTRA_OVERWRITE: {
+
+        if (unlikely(!afl->extras_cnt)) { goto retry_havoc_step; }
+
+        /* Use the dictionary. */
+
+        u32 use_extra = rand_below(afl, afl->extras_cnt);
+        u32 extra_len = afl->extras[use_extra].len;
+
+        if (unlikely(extra_len > len)) { goto retry_havoc_step; }
+
+        u32 insert_at = rand_below(afl, len - extra_len + 1);
+        memcpy(buf + insert_at, afl->extras[use_extra].data, extra_len);
+
+        break;
+
+      }
+
+      case MUT_EXTRA_INSERT: {
+
+        if (unlikely(!afl->extras_cnt)) { goto retry_havoc_step; }
+
+        u32 use_extra = rand_below(afl, afl->extras_cnt);
+        u32 extra_len = afl->extras[use_extra].len;
+        if (unlikely(len + extra_len >= MAX_FILE)) { goto retry_havoc_step; }
+
+        u8 *ptr = afl->extras[use_extra].data;
+        u32 insert_at = rand_below(afl, len + 1);
+
+        /* Tail */
+        memmove(buf + insert_at + extra_len, buf + insert_at, len - insert_at);
+
+        /* Inserted part */
+        memcpy(buf + insert_at, ptr, extra_len);
+        len += extra_len;
+
+        break;
+
+      }
+
+      case MUT_AUTO_EXTRA_OVERWRITE: {
+
+        if (unlikely(!afl->a_extras_cnt)) { goto retry_havoc_step; }
+
+        /* Use the dictionary. */
+
+        u32 use_extra = rand_below(afl, afl->a_extras_cnt);
+        u32 extra_len = afl->a_extras[use_extra].len;
+
+        if (unlikely(extra_len > len)) { goto retry_havoc_step; }
+
+        u32 insert_at = rand_below(afl, len - extra_len + 1);
+        memcpy(buf + insert_at, afl->a_extras[use_extra].data, extra_len);
+
+        break;
+
+      }
+
+      case MUT_AUTO_EXTRA_INSERT: {
+
+        if (unlikely(!afl->a_extras_cnt)) { goto retry_havoc_step; }
+
+        u32 use_extra = rand_below(afl, afl->a_extras_cnt);
+        u32 extra_len = afl->a_extras[use_extra].len;
+        if (unlikely(len + extra_len >= MAX_FILE)) { goto retry_havoc_step; }
+
+        u8 *ptr = afl->a_extras[use_extra].data;
+        u32 insert_at = rand_below(afl, len + 1);
+
+        /* Tail */
+        memmove(buf + insert_at + extra_len, buf + insert_at, len - insert_at);
+
+        /* Inserted part */
+        memcpy(buf + insert_at, ptr, extra_len);
+        len += extra_len;
+
+        break;
+
+      }
+
+      case MUT_SPLICE_OVERWRITE: {
+
+        if (unlikely(!splice_buf || !splice_len)) { goto retry_havoc_step; }
+
+        /* overwrite mode */
+
+        u32 copy_from, copy_to, copy_len;
+
+        copy_len = choose_block_len(afl, splice_len - 1);
+
+        if (copy_len > len) copy_len = len;
+
+        copy_from = rand_below(afl, splice_len - copy_len + 1);
+        copy_to = rand_below(afl, len - copy_len + 1);
+        memmove(buf + copy_to, splice_buf + copy_from, copy_len);
+
+        break;
+
+      }
+
+      case MUT_SPLICE_INSERT: {
+
+        if (unlikely(!splice_buf || !splice_len)) { goto retry_havoc_step; }
+
+        if (unlikely(len + HAVOC_BLK_XL >= MAX_FILE)) { goto retry_havoc_step; }
+
+        /* insert mode */
+
+        u32 clone_from, clone_to, clone_len;
+
+        clone_len = choose_block_len(afl, splice_len);
+        clone_from = rand_below(afl, splice_len - clone_len + 1);
+        clone_to = rand_below(afl, len + 1);
+
+        /* Head */
+
+        memcpy(tmp_buf, buf, clone_to);
+
+        /* Inserted part */
+
+        memcpy(tmp_buf + clone_to, splice_buf + clone_from, clone_len);
+
+        /* Tail */
+        memcpy(tmp_buf + clone_to + clone_len, buf + clone_to, len - clone_to);
+
+        len += clone_len;
+        memcpy(buf, tmp_buf, len);
+
+        break;
+
+      }
+
+    }
+
+  }
+
+  return len;
+
+}
+
+#endif                                                  /* !AFL_MUTATIONS_H */
+
-- 
cgit 1.4.1


From 32ffa2664cdfa2cc377df12cbf6efdcecbc2e78a Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Tue, 4 Apr 2023 16:23:19 +0200
Subject: max_len support

---
 custom_mutators/aflpp/aflpp.c | 27 +++++++++++++--
 include/afl-mutations.h       | 76 +++++++++++++++++++++++++++----------------
 2 files changed, 72 insertions(+), 31 deletions(-)

(limited to 'include')

diff --git a/custom_mutators/aflpp/aflpp.c b/custom_mutators/aflpp/aflpp.c
index 2b69ad9c..e15d0391 100644
--- a/custom_mutators/aflpp/aflpp.c
+++ b/custom_mutators/aflpp/aflpp.c
@@ -4,6 +4,7 @@ typedef struct my_mutator {
 
   afl_state_t *afl;
   u8          *buf;
+  u32          buf_size;
 
 } my_mutator_t;
 
@@ -19,12 +20,15 @@ my_mutator_t *afl_custom_init(afl_state_t *afl, unsigned int seed) {
 
   }
 
-  data->buf = malloc(MAX_FILE);
-  if (!data->buf) {
+  if ((data->buf = malloc(MAX_FILE)) == NULL) {
 
     perror("afl_custom_init alloc");
     return NULL;
 
+  } else {
+
+    data->buf_size = MAX_FILE;
+
   }
 
   data->afl = afl;
@@ -39,6 +43,23 @@ size_t afl_custom_fuzz(my_mutator_t *data, uint8_t *buf, size_t buf_size,
                        u8 **out_buf, uint8_t *add_buf, size_t add_buf_size,
                        size_t max_size) {
 
+  if (max_size > data->buf_size) {
+
+    u8 *ptr = realloc(data->buf, max_size);
+
+    if (ptr) {
+
+      return 0;
+
+    } else {
+
+      data->buf = ptr;
+      data->buf_size = max_size;
+
+    }
+
+  }
+
   u32 havoc_steps = 1 + rand_below(data->afl, 16);
 
   /* set everything up, costly ... :( */
@@ -46,7 +67,7 @@ size_t afl_custom_fuzz(my_mutator_t *data, uint8_t *buf, size_t buf_size,
 
   /* the mutation */
   u32 out_buf_len = afl_mutate(data->afl, data->buf, buf_size, havoc_steps,
-                               false, true, add_buf, add_buf_size);
+                               false, true, add_buf, add_buf_size, max_size);
 
   /* return size of mutated data */
   *out_buf = data->buf;
diff --git a/include/afl-mutations.h b/include/afl-mutations.h
index 43b7927d..e3f69214 100644
--- a/include/afl-mutations.h
+++ b/include/afl-mutations.h
@@ -6,7 +6,8 @@
    #include "afl-mutations.h"  // needs afl-fuzz.h
 
    u32 afl_mutate(afl_state_t *afl, u8 *buf, u32 len, u32t steps, bool is_text,
-                  bool is_exploration, u8 *splice_buf, u32 splice_len);
+                  bool is_exploration, u8 *splice_buf, u32 splice_len,
+                  u32 max_len);
 
    Returns:
      u32 - the length of the mutated data return in *buf. 0 = error
@@ -21,6 +22,7 @@
      splice_buf - a buffer from another corpus item to splice with.
                   If NULL then no splicing
      splice_len - the length of the splice buffer. If 0 then no splicing
+     u32 max_len - the maximum size the mutated buffer may grow to
 */
 
 #ifndef _ANDROID_ASHMEM_H
@@ -76,16 +78,13 @@ enum {
 
 };
 
-unsigned int mutation_strategy_exploration_text[MUT_STRATEGY_ARRAY_SIZE] = {};
-unsigned int mutation_strategy_exploration_binary[MUT_STRATEGY_ARRAY_SIZE] = {};
-unsigned int mutation_strategy_exploitation_text[MUT_STRATEGY_ARRAY_SIZE] = {};
-unsigned int mutation_strategy_exploitation_binary[MUT_STRATEGY_ARRAY_SIZE] =
-    {};
+u32 mutation_strategy_exploration_text[MUT_STRATEGY_ARRAY_SIZE] = {};
+u32 mutation_strategy_exploration_binary[MUT_STRATEGY_ARRAY_SIZE] = {};
+u32 mutation_strategy_exploitation_text[MUT_STRATEGY_ARRAY_SIZE] = {};
+u32 mutation_strategy_exploitation_binary[MUT_STRATEGY_ARRAY_SIZE] = {};
 
-unsigned int afl_mutate(afl_state_t *, unsigned char *, unsigned int,
-                        unsigned int, bool, bool, unsigned char *,
-                        unsigned int);
-u32          choose_block_len(afl_state_t *, u32);
+u32 afl_mutate(afl_state_t *, u8 *, u32, u32, bool, bool, u8 *, u32, u32);
+u32 choose_block_len(afl_state_t *, u32);
 
 /* Helper to choose random block len for block operations in fuzz_one().
    Doesn't return zero, provided that max_len is > 0. */
@@ -131,18 +130,39 @@ inline u32 choose_block_len(afl_state_t *afl, u32 limit) {
 
 }
 
-unsigned int afl_mutate(afl_state_t *afl, unsigned char *buf, unsigned int len,
-                        unsigned int steps, bool is_text, bool is_exploration,
-                        unsigned char *splice_buf, unsigned int splice_len) {
+inline u32 afl_mutate(afl_state_t *afl, u8 *buf, u32 len, u32 steps,
+                      bool is_text, bool is_exploration, u8 *splice_buf,
+                      u32 splice_len, u32 max_len) {
 
   if (!buf || !len) { return 0; }
 
-  u32                  *mutation_array;
-  static unsigned char *tmp_buf = NULL;
+  u32       *mutation_array;
+  static u8 *tmp_buf = NULL;
+  static u32 tmp_buf_size = 0;
 
-  if (!tmp_buf) {
+  if (max_len > tmp_buf_size) {
 
-    if ((tmp_buf = malloc(MAX_FILE)) == NULL) { return 0; }
+    if (tmp_buf) {
+
+      u8 *ptr = realloc(tmp_buf, max_len);
+
+      if (!ptr) {
+
+        return 0;
+
+      } else {
+
+        tmp_buf = ptr;
+
+      }
+
+    } else {
+
+      if ((tmp_buf = malloc(max_len)) == NULL) { return 0; }
+
+    }
+
+    tmp_buf_size = max_len;
 
   }
 
@@ -150,11 +170,11 @@ unsigned int afl_mutate(afl_state_t *afl, unsigned char *buf, unsigned int len,
 
     if (is_exploration) {
 
-      mutation_array = (unsigned int *)&mutation_strategy_exploration_text;
+      mutation_array = (u32 *)&mutation_strategy_exploration_text;
 
     } else {
 
-      mutation_array = (unsigned int *)&mutation_strategy_exploitation_text;
+      mutation_array = (u32 *)&mutation_strategy_exploitation_text;
 
     }
 
@@ -162,17 +182,17 @@ unsigned int afl_mutate(afl_state_t *afl, unsigned char *buf, unsigned int len,
 
     if (is_exploration) {
 
-      mutation_array = (unsigned int *)&mutation_strategy_exploration_binary;
+      mutation_array = (u32 *)&mutation_strategy_exploration_binary;
 
     } else {
 
-      mutation_array = (unsigned int *)&mutation_strategy_exploitation_binary;
+      mutation_array = (u32 *)&mutation_strategy_exploitation_binary;
 
     }
 
   }
 
-  for (unsigned int step = 0; step < steps; ++step) {
+  for (u32 step = 0; step < steps; ++step) {
 
   retry_havoc_step:
 
@@ -400,7 +420,7 @@ unsigned int afl_mutate(afl_state_t *afl, unsigned char *buf, unsigned int len,
 
       case MUT_CLONE_OVERWRITE: {
 
-        if (likely(len + HAVOC_BLK_XL < MAX_FILE)) {
+        if (likely(len + HAVOC_BLK_XL < max_len)) {
 
           /* Clone bytes. */
 
@@ -439,7 +459,7 @@ unsigned int afl_mutate(afl_state_t *afl, unsigned char *buf, unsigned int len,
 
       case MUT_CLONE_INSERT: {
 
-        if (likely(len + HAVOC_BLK_XL < MAX_FILE)) {
+        if (likely(len + HAVOC_BLK_XL < max_len)) {
 
           /* Insert a block of constant bytes (25%). */
 
@@ -622,7 +642,7 @@ unsigned int afl_mutate(afl_state_t *afl, unsigned char *buf, unsigned int len,
 
           } while (unlikely(i == j));
 
-          unsigned char temp = buf[off + i];
+          u8 temp = buf[off + i];
           buf[off + i] = buf[off + j];
           buf[off + j] = temp;
 
@@ -872,7 +892,7 @@ unsigned int afl_mutate(afl_state_t *afl, unsigned char *buf, unsigned int len,
 
         u32 use_extra = rand_below(afl, afl->extras_cnt);
         u32 extra_len = afl->extras[use_extra].len;
-        if (unlikely(len + extra_len >= MAX_FILE)) { goto retry_havoc_step; }
+        if (unlikely(len + extra_len >= max_len)) { goto retry_havoc_step; }
 
         u8 *ptr = afl->extras[use_extra].data;
         u32 insert_at = rand_below(afl, len + 1);
@@ -912,7 +932,7 @@ unsigned int afl_mutate(afl_state_t *afl, unsigned char *buf, unsigned int len,
 
         u32 use_extra = rand_below(afl, afl->a_extras_cnt);
         u32 extra_len = afl->a_extras[use_extra].len;
-        if (unlikely(len + extra_len >= MAX_FILE)) { goto retry_havoc_step; }
+        if (unlikely(len + extra_len >= max_len)) { goto retry_havoc_step; }
 
         u8 *ptr = afl->a_extras[use_extra].data;
         u32 insert_at = rand_below(afl, len + 1);
@@ -952,7 +972,7 @@ unsigned int afl_mutate(afl_state_t *afl, unsigned char *buf, unsigned int len,
 
         if (unlikely(!splice_buf || !splice_len)) { goto retry_havoc_step; }
 
-        if (unlikely(len + HAVOC_BLK_XL >= MAX_FILE)) { goto retry_havoc_step; }
+        if (unlikely(len + HAVOC_BLK_XL >= max_len)) { goto retry_havoc_step; }
 
         /* insert mode */
 
-- 
cgit 1.4.1


From 2bff92c603463410fa0f97e7c4db7eb14c45e5ed Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Tue, 4 Apr 2023 16:25:05 +0200
Subject: nit

---
 include/afl-mutations.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'include')

diff --git a/include/afl-mutations.h b/include/afl-mutations.h
index e3f69214..707db799 100644
--- a/include/afl-mutations.h
+++ b/include/afl-mutations.h
@@ -14,7 +14,7 @@
    Parameters:
      afl_state_t *afl - the *afl state pointer
      u8 *buf - the input buffer to mutate which will be mutated into.
-               NOTE: must be of MAX_FILE size!
+           NOTE: must be able to contain a size of at least max_len (see below)!
      u32 len - the length of the input
      u32 steps - how many mutations to perform on the input
      bool is_text - is the target expecting text inputs
-- 
cgit 1.4.1


From 3ab18d286142e2e19e37850c051e0b07b9d7b296 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Tue, 4 Apr 2023 19:44:12 +0200
Subject: mode switch

---
 include/afl-fuzz.h      |  69 +++++++------
 include/afl-mutations.h | 259 +++++++++++++++++++++++++++++++++++++++++++++++-
 include/config.h        |   6 ++
 src/afl-fuzz-one.c      |   6 +-
 src/afl-fuzz-state.c    |   1 +
 src/afl-fuzz-stats.c    |   5 +-
 src/afl-fuzz.c          |  51 ++++++++--
 7 files changed, 351 insertions(+), 46 deletions(-)

(limited to 'include')

diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index 58d02af5..6573eabf 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -490,7 +490,9 @@ typedef struct afl_state {
       *orig_cmdline,                    /* Original command line            */
       *infoexec;                       /* Command to execute on a new crash */
 
-  u32 hang_tmout;                       /* Timeout used for hang det (ms)   */
+  u32 hang_tmout,                       /* Timeout used for hang det (ms)   */
+      stats_update_freq,                /* Stats update frequency (execs)   */
+      switch_fuzz_mode;                 /* auto or fixed fuzz mode          */
 
   u8 havoc_stack_pow2,                  /* HAVOC_STACK_POW2                 */
       no_unlink,                        /* do not unlink cur_input          */
@@ -499,40 +501,37 @@ typedef struct afl_state {
       custom_splice_optout,             /* Custom mutator no splice buffer  */
       is_main_node,                     /* if this is the main node         */
       is_secondary_node,                /* if this is a secondary instance  */
-      pizza_is_served;                  /* pizza mode                       */
-
-  u32 stats_update_freq;                /* Stats update frequency (execs)   */
-
-  u8 schedule;                          /* Power schedule (default: EXPLORE)*/
-  u8 havoc_max_mult;
-
-  u8 skip_deterministic,                /* Skip deterministic stages?       */
-      use_splicing,                     /* Recombine input files?           */
-      non_instrumented_mode,            /* Run in non-instrumented mode?    */
-      score_changed,                    /* Scoring for favorites changed?   */
-      resuming_fuzz,                    /* Resuming an older fuzzing job?   */
-      timeout_given,                    /* Specific timeout given?          */
-      not_on_tty,                       /* stdout is not a tty              */
-      term_too_small,                   /* terminal dimensions too small    */
-      no_forkserver,                    /* Disable forkserver?              */
-      crash_mode,                       /* Crash mode! Yeah!                */
-      in_place_resume,                  /* Attempt in-place resume?         */
-      autoresume,                       /* Resume if afl->out_dir exists?   */
-      auto_changed,                     /* Auto-generated tokens changed?   */
-      no_cpu_meter_red,                 /* Feng shui on the status screen   */
-      no_arith,                         /* Skip most arithmetic ops         */
-      shuffle_queue,                    /* Shuffle input queue?             */
-      bitmap_changed,                   /* Time to update bitmap?           */
-      unicorn_mode,                     /* Running in Unicorn mode?         */
-      use_wine,                         /* Use WINE with QEMU mode          */
-      skip_requested,                   /* Skip request, via SIGUSR1        */
-      run_over10m,                      /* Run time over 10 minutes?        */
-      persistent_mode,                  /* Running in persistent mode?      */
-      deferred_mode,                    /* Deferred forkserver mode?        */
-      fixed_seed,                       /* do not reseed                    */
-      fast_cal,                         /* Try to calibrate faster?         */
-      disable_trim,                     /* Never trim in fuzz_one           */
-      shmem_testcase_mode,              /* If sharedmem testcases are used  */
+      pizza_is_served,                  /* pizza mode                       */
+      text_input,                       /* target wants text inputs         */
+      fuzz_mode, /* current mode: coverage/exploration or crash/exploitation */
+      schedule,                         /* Power schedule (default: EXPLORE)*/
+      havoc_max_mult, skip_deterministic, /* Skip deterministic stages?       */
+      use_splicing,                       /* Recombine input files?           */
+      non_instrumented_mode,              /* Run in non-instrumented mode?    */
+      score_changed,                      /* Scoring for favorites changed?   */
+      resuming_fuzz,                      /* Resuming an older fuzzing job?   */
+      timeout_given,                      /* Specific timeout given?          */
+      not_on_tty,                         /* stdout is not a tty              */
+      term_too_small,                     /* terminal dimensions too small    */
+      no_forkserver,                      /* Disable forkserver?              */
+      crash_mode,                         /* Crash mode! Yeah!                */
+      in_place_resume,                    /* Attempt in-place resume?         */
+      autoresume,                         /* Resume if afl->out_dir exists?   */
+      auto_changed,                       /* Auto-generated tokens changed?   */
+      no_cpu_meter_red,                   /* Feng shui on the status screen   */
+      no_arith,                           /* Skip most arithmetic ops         */
+      shuffle_queue,                      /* Shuffle input queue?             */
+      bitmap_changed,                     /* Time to update bitmap?           */
+      unicorn_mode,                       /* Running in Unicorn mode?         */
+      use_wine,                           /* Use WINE with QEMU mode          */
+      skip_requested,                     /* Skip request, via SIGUSR1        */
+      run_over10m,                        /* Run time over 10 minutes?        */
+      persistent_mode,                    /* Running in persistent mode?      */
+      deferred_mode,                      /* Deferred forkserver mode?        */
+      fixed_seed,                         /* do not reseed                    */
+      fast_cal,                           /* Try to calibrate faster?         */
+      disable_trim,                       /* Never trim in fuzz_one           */
+      shmem_testcase_mode,                /* If sharedmem testcases are used  */
       expand_havoc,                /* perform expensive havoc after no find */
       cycle_schedules,                  /* cycle power schedules?           */
       old_seed_selection,               /* use vanilla afl seed selection   */
diff --git a/include/afl-mutations.h b/include/afl-mutations.h
index 707db799..5a1b6356 100644
--- a/include/afl-mutations.h
+++ b/include/afl-mutations.h
@@ -81,7 +81,264 @@ enum {
 u32 mutation_strategy_exploration_text[MUT_STRATEGY_ARRAY_SIZE] = {};
 u32 mutation_strategy_exploration_binary[MUT_STRATEGY_ARRAY_SIZE] = {};
 u32 mutation_strategy_exploitation_text[MUT_STRATEGY_ARRAY_SIZE] = {};
-u32 mutation_strategy_exploitation_binary[MUT_STRATEGY_ARRAY_SIZE] = {};
+u32 mutation_strategy_exploitation_binary[MUT_STRATEGY_ARRAY_SIZE] = {
+
+    MUT_FLIPBIT,
+    MUT_FLIPBIT,
+    MUT_FLIPBIT,
+    MUT_FLIPBIT,
+    MUT_FLIPBIT,
+    MUT_FLIPBIT,
+    MUT_FLIPBIT,
+    MUT_INTERESTING8,
+    MUT_INTERESTING8,
+    MUT_INTERESTING8,
+    MUT_INTERESTING8,
+    MUT_INTERESTING8,
+    MUT_INTERESTING8,
+    MUT_INTERESTING8,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32BE,
+    MUT_INTERESTING32BE,
+    MUT_INTERESTING32BE,
+    MUT_INTERESTING32BE,
+    MUT_INTERESTING32BE,
+    MUT_INTERESTING32BE,
+    MUT_INTERESTING32BE,
+    MUT_INTERESTING32BE,
+    MUT_ARITH8_,
+    MUT_ARITH8_,
+    MUT_ARITH8_,
+    MUT_ARITH8_,
+    MUT_ARITH8_,
+    MUT_ARITH8_,
+    MUT_ARITH8,
+    MUT_ARITH8,
+    MUT_ARITH8,
+    MUT_ARITH8,
+    MUT_ARITH8,
+    MUT_ARITH8,
+    MUT_ARITH8,
+    MUT_ARITH16_,
+    MUT_ARITH16_,
+    MUT_ARITH16_,
+    MUT_ARITH16_,
+    MUT_ARITH16_,
+    MUT_ARITH16_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16,
+    MUT_ARITH16,
+    MUT_ARITH16,
+    MUT_ARITH16,
+    MUT_ARITH16,
+    MUT_ARITH16,
+    MUT_ARITH16,
+    MUT_ARITH16BE,
+    MUT_ARITH16BE,
+    MUT_ARITH16BE,
+    MUT_ARITH16BE,
+    MUT_ARITH16BE,
+    MUT_ARITH16BE,
+    MUT_ARITH16BE,
+    MUT_ARITH32_,
+    MUT_ARITH32_,
+    MUT_ARITH32_,
+    MUT_ARITH32_,
+    MUT_ARITH32_,
+    MUT_ARITH32_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32,
+    MUT_ARITH32,
+    MUT_ARITH32,
+    MUT_ARITH32,
+    MUT_ARITH32,
+    MUT_ARITH32,
+    MUT_ARITH32BE,
+    MUT_ARITH32BE,
+    MUT_ARITH32BE,
+    MUT_ARITH32BE,
+    MUT_ARITH32BE,
+    MUT_ARITH32BE,
+    MUT_ARITH32BE,
+    MUT_RAND8,
+    MUT_RAND8,
+    MUT_RAND8,
+    MUT_RAND8,
+    MUT_RAND8,
+    MUT_RAND8,
+    MUT_RAND8,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_FIXED,
+    MUT_OVERWRITE_FIXED,
+    MUT_OVERWRITE_FIXED,
+    MUT_OVERWRITE_FIXED,
+    MUT_OVERWRITE_FIXED,
+    MUT_OVERWRITE_FIXED,
+    MUT_BYTEADD,
+    MUT_BYTEADD,
+    MUT_BYTEADD,
+    MUT_BYTEADD,
+    MUT_BYTEADD,
+    MUT_BYTEADD,
+    MUT_BYTEADD,
+    MUT_BYTESUB,
+    MUT_BYTESUB,
+    MUT_BYTESUB,
+    MUT_BYTESUB,
+    MUT_BYTESUB,
+    MUT_BYTESUB,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_DEL,
+    MUT_DEL,
+    MUT_DEL,
+    MUT_DEL,
+    MUT_DEL,
+    MUT_DEL,
+    MUT_SHUFFLE,
+    MUT_SHUFFLE,
+    MUT_SHUFFLE,
+    MUT_SHUFFLE,
+    MUT_SHUFFLE,
+    MUT_SHUFFLE,
+    MUT_DELONE,
+    MUT_DELONE,
+    MUT_DELONE,
+    MUT_DELONE,
+    MUT_DELONE,
+    MUT_INSERTONE,
+    MUT_INSERTONE,
+    MUT_INSERTONE,
+    MUT_INSERTONE,
+    MUT_INSERTONE,
+    MUT_INSERTONE,
+    MUT_ASCIINUM,
+    MUT_ASCIINUM,
+    MUT_ASCIINUM,
+    MUT_ASCIINUM,
+    MUT_ASCIINUM,
+    MUT_NEG,
+    MUT_NEG,
+    MUT_NEG,
+    MUT_NEG,
+    MUT_NEG,
+    MUT_NEG,
+    MUT_INSERTASCIINUM,
+    MUT_INSERTASCIINUM,
+    MUT_INSERTASCIINUM,
+    MUT_INSERTASCIINUM,
+    MUT_INSERTASCIINUM,
+    MUT_INSERTASCIINUM,
+    MUT_INSERTASCIINUM,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT};
 
 u32 afl_mutate(afl_state_t *, u8 *, u32, u32, bool, bool, u8 *, u32, u32);
 u32 choose_block_len(afl_state_t *, u32);
diff --git a/include/config.h b/include/config.h
index e46f515a..c1297bdd 100644
--- a/include/config.h
+++ b/include/config.h
@@ -43,6 +43,12 @@
    Default: 8MB (defined in bytes) */
 #define DEFAULT_SHMEM_SIZE (8 * 1024 * 1024)
 
+/* Default time until when no more coverage finds are happening afl-fuzz
+   switches to exploitation mode. It automatically switches back when new
+   coverage is found.
+   Default: 300 (seconds) */
+#define STRATEGY_SWITCH_TIME 300
+
 /* Default file permission umode when creating files (default: 0600) */
 #define DEFAULT_PERMISSION 0600
 
diff --git a/src/afl-fuzz-one.c b/src/afl-fuzz-one.c
index 226fb40e..e6b58713 100644
--- a/src/afl-fuzz-one.c
+++ b/src/afl-fuzz-one.c
@@ -2078,9 +2078,9 @@ havoc_stage:
   u32 *mutation_array;
   u32  stack_max;
 
-  if (afl->queue_cur->is_ascii) {  // is text?
+  if (unlikely(afl->text_input || afl->queue_cur->is_ascii)) {  // is text?
 
-    if (1) {  // is exploration?
+    if (likely(afl->fuzz_mode == 0)) {  // is exploration?
 
       mutation_array = (unsigned int *)&mutation_strategy_exploration_text;
 
@@ -2092,7 +2092,7 @@ havoc_stage:
 
   } else {  // is binary!
 
-    if (1) {  // is exploration?
+    if (likely(afl->fuzz_mode == 0)) {  // is exploration?
 
       mutation_array = (unsigned int *)&mutation_strategy_exploration_binary;
 
diff --git a/src/afl-fuzz-state.c b/src/afl-fuzz-state.c
index f9aa5cfe..907861e9 100644
--- a/src/afl-fuzz-state.c
+++ b/src/afl-fuzz-state.c
@@ -108,6 +108,7 @@ void afl_state_init(afl_state_t *afl, uint32_t map_size) {
   afl->cmplog_lvl = 2;
   afl->min_length = 1;
   afl->max_length = MAX_FILE;
+  afl->switch_fuzz_mode = STRATEGY_SWITCH_TIME;
 #ifndef NO_SPLICING
   afl->use_splicing = 1;
 #endif
diff --git a/src/afl-fuzz-stats.c b/src/afl-fuzz-stats.c
index 25ebe987..de48e10a 100644
--- a/src/afl-fuzz-stats.c
+++ b/src/afl-fuzz-stats.c
@@ -1282,7 +1282,10 @@ void show_stats_normal(afl_state_t *afl) {
   }
 
   /* Last line */
-  SAYF(SET_G1 "\n" bSTG bLB bH30 bH20 bH2 bRB bSTOP cRST RESET_G1);
+
+  SAYF(SET_G1 "\n" bSTG bLB bH cCYA                                bSTOP
+              " strategy:%s %s " bSTG bH20 bH10 bH2 bRB bSTOP cRST RESET_G1,
+       cPIN, afl->fuzz_mode == 0 ? "explore" : "exploit");
 
 #undef IB
 
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index 3380fd90..315107d7 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -128,6 +128,13 @@ static void usage(u8 *argv0, int more_help) {
       "  -o dir        - output directory for fuzzer findings\n\n"
 
       "Execution control settings:\n"
+      "  -P strategy   - set fix mutation strategy: explore (focus on new "
+      "coverage),\n"
+      "                  exploit (focus on triggering crashes). You can also "
+      "set a\n"
+      "                  number of seconds after without any finds it switches "
+      "to\n"
+      "                  exploit mode, and back on new coverage (default: %u)\n"
       "  -p schedule   - power schedules compute a seed's performance score:\n"
       "                  fast(default), explore, exploit, seek, rare, mmopt, "
       "coe, lin\n"
@@ -156,6 +163,7 @@ static void usage(u8 *argv0, int more_help) {
       "\n"
 
       "Mutator settings:\n"
+      "  -a            - target expects ascii text input\n"
       "  -g minlength  - set min length of generated fuzz input (default: 1)\n"
       "  -G maxlength  - set max length of generated fuzz input (default: "
       "%lu)\n"
@@ -212,7 +220,8 @@ static void usage(u8 *argv0, int more_help) {
       "  -e ext        - file extension for the fuzz test input file (if "
       "needed)\n"
       "\n",
-      argv0, EXEC_TIMEOUT, MEM_LIMIT, MAX_FILE, FOREIGN_SYNCS_MAX);
+      argv0, STRATEGY_SWITCH_TIME, EXEC_TIMEOUT, MEM_LIMIT, MAX_FILE,
+      FOREIGN_SYNCS_MAX);
 
   if (more_help > 1) {
 
@@ -553,14 +562,44 @@ int main(int argc, char **argv_orig, char **envp) {
 
   afl->shmem_testcase_mode = 1;  // we always try to perform shmem fuzzing
 
-  while (
-      (opt = getopt(
-           argc, argv,
-           "+Ab:B:c:CdDe:E:hi:I:f:F:g:G:l:L:m:M:nNOo:p:RQs:S:t:T:UV:WXx:YZ")) >
-      0) {
+  // still available: aHjJkKPqruvwz
+  while ((opt = getopt(argc, argv,
+                       "+aAb:B:c:CdDe:E:f:F:g:G:hi:I:l:L:m:M:nNo:Op:P:QRs:S:t:"
+                       "T:UV:WXx:YZ")) > 0) {
 
     switch (opt) {
 
+      case 'a':
+        afl->text_input = 1;
+        break;
+
+      case 'P':
+        if (!stricmp(optarg, "explore") || !stricmp(optarg, "exploration")) {
+
+          afl->fuzz_mode = 0;
+          afl->switch_fuzz_mode = 1;
+
+        } else if (!stricmp(optarg, "exploit") ||
+
+                   !stricmp(optarg, "exploitation")) {
+
+          afl->fuzz_mode = 1;
+          afl->switch_fuzz_mode = 0;
+
+        } else {
+
+          if ((s32)(afl->switch_fuzz_mode = (u32)atoi(optarg)) < 1) {
+
+            FATAL(
+                "Parameter for option -P must be \"explore\", \"exploit\" or a "
+                "number!");
+
+          }
+
+        }
+
+        break;
+
       case 'g':
         afl->min_length = atoi(optarg);
         break;
-- 
cgit 1.4.1


From 41a452d4e8038461f730736439346eb8a7a3968f Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Tue, 4 Apr 2023 21:48:51 +0200
Subject: mutation lists

---
 include/afl-mutations.h | 786 +++++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 783 insertions(+), 3 deletions(-)

(limited to 'include')

diff --git a/include/afl-mutations.h b/include/afl-mutations.h
index 5a1b6356..31d0898a 100644
--- a/include/afl-mutations.h
+++ b/include/afl-mutations.h
@@ -78,9 +78,789 @@ enum {
 
 };
 
-u32 mutation_strategy_exploration_text[MUT_STRATEGY_ARRAY_SIZE] = {};
-u32 mutation_strategy_exploration_binary[MUT_STRATEGY_ARRAY_SIZE] = {};
-u32 mutation_strategy_exploitation_text[MUT_STRATEGY_ARRAY_SIZE] = {};
+u32 mutation_strategy_exploration_text[MUT_STRATEGY_ARRAY_SIZE] = {
+
+    MUT_FLIPBIT,
+    MUT_FLIPBIT,
+    MUT_FLIPBIT,
+    MUT_FLIPBIT,
+    MUT_FLIPBIT,
+    MUT_FLIPBIT,
+    MUT_INTERESTING8,
+    MUT_INTERESTING8,
+    MUT_INTERESTING8,
+    MUT_INTERESTING8,
+    MUT_INTERESTING8,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32BE,
+    MUT_INTERESTING32BE,
+    MUT_INTERESTING32BE,
+    MUT_INTERESTING32BE,
+    MUT_INTERESTING32BE,
+    MUT_ARITH8_,
+    MUT_ARITH8_,
+    MUT_ARITH8_,
+    MUT_ARITH8_,
+    MUT_ARITH8_,
+    MUT_ARITH8_,
+    MUT_ARITH8,
+    MUT_ARITH8,
+    MUT_ARITH8,
+    MUT_ARITH8,
+    MUT_ARITH8,
+    MUT_ARITH8,
+    MUT_ARITH16_,
+    MUT_ARITH16_,
+    MUT_ARITH16_,
+    MUT_ARITH16_,
+    MUT_ARITH16_,
+    MUT_ARITH16_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16,
+    MUT_ARITH16,
+    MUT_ARITH16,
+    MUT_ARITH16,
+    MUT_ARITH16,
+    MUT_ARITH16,
+    MUT_ARITH16BE,
+    MUT_ARITH16BE,
+    MUT_ARITH16BE,
+    MUT_ARITH16BE,
+    MUT_ARITH16BE,
+    MUT_ARITH16BE,
+    MUT_ARITH32_,
+    MUT_ARITH32_,
+    MUT_ARITH32_,
+    MUT_ARITH32_,
+    MUT_ARITH32_,
+    MUT_ARITH32_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32,
+    MUT_ARITH32,
+    MUT_ARITH32,
+    MUT_ARITH32,
+    MUT_ARITH32,
+    MUT_ARITH32,
+    MUT_ARITH32BE,
+    MUT_ARITH32BE,
+    MUT_ARITH32BE,
+    MUT_ARITH32BE,
+    MUT_ARITH32BE,
+    MUT_ARITH32BE,
+    MUT_RAND8,
+    MUT_RAND8,
+    MUT_RAND8,
+    MUT_RAND8,
+    MUT_RAND8,
+    MUT_RAND8,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_FIXED,
+    MUT_OVERWRITE_FIXED,
+    MUT_OVERWRITE_FIXED,
+    MUT_OVERWRITE_FIXED,
+    MUT_OVERWRITE_FIXED,
+    MUT_OVERWRITE_FIXED,
+    MUT_OVERWRITE_FIXED,
+    MUT_OVERWRITE_FIXED,
+    MUT_BYTEADD,
+    MUT_BYTEADD,
+    MUT_BYTEADD,
+    MUT_BYTEADD,
+    MUT_BYTEADD,
+    MUT_BYTEADD,
+    MUT_BYTESUB,
+    MUT_BYTESUB,
+    MUT_BYTESUB,
+    MUT_BYTESUB,
+    MUT_BYTESUB,
+    MUT_BYTESUB,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_DEL,
+    MUT_DEL,
+    MUT_DEL,
+    MUT_DEL,
+    MUT_DEL,
+    MUT_DEL,
+    MUT_DEL,
+    MUT_SHUFFLE,
+    MUT_SHUFFLE,
+    MUT_SHUFFLE,
+    MUT_SHUFFLE,
+    MUT_SHUFFLE,
+    MUT_SHUFFLE,
+    MUT_SHUFFLE,
+    MUT_SHUFFLE,
+    MUT_DELONE,
+    MUT_DELONE,
+    MUT_DELONE,
+    MUT_DELONE,
+    MUT_DELONE,
+    MUT_DELONE,
+    MUT_DELONE,
+    MUT_INSERTONE,
+    MUT_INSERTONE,
+    MUT_INSERTONE,
+    MUT_INSERTONE,
+    MUT_INSERTONE,
+    MUT_INSERTONE,
+    MUT_ASCIINUM,
+    MUT_ASCIINUM,
+    MUT_ASCIINUM,
+    MUT_ASCIINUM,
+    MUT_ASCIINUM,
+    MUT_ASCIINUM,
+    MUT_ASCIINUM,
+    MUT_NEG,
+    MUT_NEG,
+    MUT_NEG,
+    MUT_NEG,
+    MUT_NEG,
+    MUT_NEG,
+    MUT_INSERTASCIINUM,
+    MUT_INSERTASCIINUM,
+    MUT_INSERTASCIINUM,
+    MUT_INSERTASCIINUM,
+    MUT_INSERTASCIINUM,
+    MUT_INSERTASCIINUM,
+    MUT_INSERTASCIINUM,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+
+};
+
+u32 mutation_strategy_exploration_binary[MUT_STRATEGY_ARRAY_SIZE] = {
+
+    MUT_FLIPBIT,
+    MUT_FLIPBIT,
+    MUT_FLIPBIT,
+    MUT_FLIPBIT,
+    MUT_FLIPBIT,
+    MUT_FLIPBIT,
+    MUT_FLIPBIT,
+    MUT_INTERESTING8,
+    MUT_INTERESTING8,
+    MUT_INTERESTING8,
+    MUT_INTERESTING8,
+    MUT_INTERESTING8,
+    MUT_INTERESTING8,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32BE,
+    MUT_INTERESTING32BE,
+    MUT_INTERESTING32BE,
+    MUT_INTERESTING32BE,
+    MUT_INTERESTING32BE,
+    MUT_INTERESTING32BE,
+    MUT_ARITH8_,
+    MUT_ARITH8_,
+    MUT_ARITH8_,
+    MUT_ARITH8_,
+    MUT_ARITH8_,
+    MUT_ARITH8_,
+    MUT_ARITH8_,
+    MUT_ARITH8,
+    MUT_ARITH8,
+    MUT_ARITH8,
+    MUT_ARITH8,
+    MUT_ARITH8,
+    MUT_ARITH8,
+    MUT_ARITH8,
+    MUT_ARITH16_,
+    MUT_ARITH16_,
+    MUT_ARITH16_,
+    MUT_ARITH16_,
+    MUT_ARITH16_,
+    MUT_ARITH16_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16,
+    MUT_ARITH16,
+    MUT_ARITH16,
+    MUT_ARITH16,
+    MUT_ARITH16,
+    MUT_ARITH16,
+    MUT_ARITH16BE,
+    MUT_ARITH16BE,
+    MUT_ARITH16BE,
+    MUT_ARITH16BE,
+    MUT_ARITH16BE,
+    MUT_ARITH16BE,
+    MUT_ARITH32_,
+    MUT_ARITH32_,
+    MUT_ARITH32_,
+    MUT_ARITH32_,
+    MUT_ARITH32_,
+    MUT_ARITH32_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32,
+    MUT_ARITH32,
+    MUT_ARITH32,
+    MUT_ARITH32,
+    MUT_ARITH32,
+    MUT_ARITH32,
+    MUT_ARITH32,
+    MUT_ARITH32BE,
+    MUT_ARITH32BE,
+    MUT_ARITH32BE,
+    MUT_ARITH32BE,
+    MUT_ARITH32BE,
+    MUT_ARITH32BE,
+    MUT_RAND8,
+    MUT_RAND8,
+    MUT_RAND8,
+    MUT_RAND8,
+    MUT_RAND8,
+    MUT_RAND8,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_FIXED,
+    MUT_OVERWRITE_FIXED,
+    MUT_OVERWRITE_FIXED,
+    MUT_OVERWRITE_FIXED,
+    MUT_OVERWRITE_FIXED,
+    MUT_OVERWRITE_FIXED,
+    MUT_OVERWRITE_FIXED,
+    MUT_BYTEADD,
+    MUT_BYTEADD,
+    MUT_BYTEADD,
+    MUT_BYTEADD,
+    MUT_BYTEADD,
+    MUT_BYTEADD,
+    MUT_BYTEADD,
+    MUT_BYTESUB,
+    MUT_BYTESUB,
+    MUT_BYTESUB,
+    MUT_BYTESUB,
+    MUT_BYTESUB,
+    MUT_BYTESUB,
+    MUT_BYTESUB,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_DEL,
+    MUT_DEL,
+    MUT_DEL,
+    MUT_DEL,
+    MUT_DEL,
+    MUT_DEL,
+    MUT_SHUFFLE,
+    MUT_SHUFFLE,
+    MUT_SHUFFLE,
+    MUT_SHUFFLE,
+    MUT_SHUFFLE,
+    MUT_SHUFFLE,
+    MUT_DELONE,
+    MUT_DELONE,
+    MUT_DELONE,
+    MUT_DELONE,
+    MUT_DELONE,
+    MUT_DELONE,
+    MUT_INSERTONE,
+    MUT_INSERTONE,
+    MUT_INSERTONE,
+    MUT_INSERTONE,
+    MUT_INSERTONE,
+    MUT_INSERTONE,
+    MUT_ASCIINUM,
+    MUT_ASCIINUM,
+    MUT_ASCIINUM,
+    MUT_ASCIINUM,
+    MUT_ASCIINUM,
+    MUT_ASCIINUM,
+    MUT_NEG,
+    MUT_NEG,
+    MUT_NEG,
+    MUT_NEG,
+    MUT_NEG,
+    MUT_NEG,
+    MUT_INSERTASCIINUM,
+    MUT_INSERTASCIINUM,
+    MUT_INSERTASCIINUM,
+    MUT_INSERTASCIINUM,
+    MUT_INSERTASCIINUM,
+    MUT_INSERTASCIINUM,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+
+};
+
+u32 mutation_strategy_exploitation_text[MUT_STRATEGY_ARRAY_SIZE] = {
+
+    MUT_FLIPBIT,
+    MUT_FLIPBIT,
+    MUT_FLIPBIT,
+    MUT_FLIPBIT,
+    MUT_FLIPBIT,
+    MUT_FLIPBIT,
+    MUT_FLIPBIT,
+    MUT_INTERESTING8,
+    MUT_INTERESTING8,
+    MUT_INTERESTING8,
+    MUT_INTERESTING8,
+    MUT_INTERESTING8,
+    MUT_INTERESTING8,
+    MUT_INTERESTING8,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING16BE,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32,
+    MUT_INTERESTING32BE,
+    MUT_INTERESTING32BE,
+    MUT_INTERESTING32BE,
+    MUT_INTERESTING32BE,
+    MUT_INTERESTING32BE,
+    MUT_INTERESTING32BE,
+    MUT_INTERESTING32BE,
+    MUT_INTERESTING32BE,
+    MUT_ARITH8_,
+    MUT_ARITH8_,
+    MUT_ARITH8_,
+    MUT_ARITH8_,
+    MUT_ARITH8_,
+    MUT_ARITH8_,
+    MUT_ARITH8,
+    MUT_ARITH8,
+    MUT_ARITH8,
+    MUT_ARITH8,
+    MUT_ARITH8,
+    MUT_ARITH8,
+    MUT_ARITH8,
+    MUT_ARITH16_,
+    MUT_ARITH16_,
+    MUT_ARITH16_,
+    MUT_ARITH16_,
+    MUT_ARITH16_,
+    MUT_ARITH16_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16BE_,
+    MUT_ARITH16,
+    MUT_ARITH16,
+    MUT_ARITH16,
+    MUT_ARITH16,
+    MUT_ARITH16,
+    MUT_ARITH16,
+    MUT_ARITH16,
+    MUT_ARITH16BE,
+    MUT_ARITH16BE,
+    MUT_ARITH16BE,
+    MUT_ARITH16BE,
+    MUT_ARITH16BE,
+    MUT_ARITH16BE,
+    MUT_ARITH16BE,
+    MUT_ARITH32_,
+    MUT_ARITH32_,
+    MUT_ARITH32_,
+    MUT_ARITH32_,
+    MUT_ARITH32_,
+    MUT_ARITH32_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32,
+    MUT_ARITH32,
+    MUT_ARITH32,
+    MUT_ARITH32,
+    MUT_ARITH32,
+    MUT_ARITH32,
+    MUT_ARITH32BE,
+    MUT_ARITH32BE,
+    MUT_ARITH32BE,
+    MUT_ARITH32BE,
+    MUT_ARITH32BE,
+    MUT_ARITH32BE,
+    MUT_ARITH32BE,
+    MUT_RAND8,
+    MUT_RAND8,
+    MUT_RAND8,
+    MUT_RAND8,
+    MUT_RAND8,
+    MUT_RAND8,
+    MUT_RAND8,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_OVERWRITE,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_COPY,
+    MUT_OVERWRITE_FIXED,
+    MUT_OVERWRITE_FIXED,
+    MUT_OVERWRITE_FIXED,
+    MUT_OVERWRITE_FIXED,
+    MUT_OVERWRITE_FIXED,
+    MUT_OVERWRITE_FIXED,
+    MUT_BYTEADD,
+    MUT_BYTEADD,
+    MUT_BYTEADD,
+    MUT_BYTEADD,
+    MUT_BYTEADD,
+    MUT_BYTEADD,
+    MUT_BYTEADD,
+    MUT_BYTESUB,
+    MUT_BYTESUB,
+    MUT_BYTESUB,
+    MUT_BYTESUB,
+    MUT_BYTESUB,
+    MUT_BYTESUB,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_SWITCH,
+    MUT_DEL,
+    MUT_DEL,
+    MUT_DEL,
+    MUT_DEL,
+    MUT_DEL,
+    MUT_DEL,
+    MUT_SHUFFLE,
+    MUT_SHUFFLE,
+    MUT_SHUFFLE,
+    MUT_SHUFFLE,
+    MUT_SHUFFLE,
+    MUT_SHUFFLE,
+    MUT_DELONE,
+    MUT_DELONE,
+    MUT_DELONE,
+    MUT_DELONE,
+    MUT_DELONE,
+    MUT_INSERTONE,
+    MUT_INSERTONE,
+    MUT_INSERTONE,
+    MUT_INSERTONE,
+    MUT_INSERTONE,
+    MUT_INSERTONE,
+    MUT_ASCIINUM,
+    MUT_ASCIINUM,
+    MUT_ASCIINUM,
+    MUT_ASCIINUM,
+    MUT_ASCIINUM,
+    MUT_NEG,
+    MUT_NEG,
+    MUT_NEG,
+    MUT_NEG,
+    MUT_NEG,
+    MUT_NEG,
+    MUT_INSERTASCIINUM,
+    MUT_INSERTASCIINUM,
+    MUT_INSERTASCIINUM,
+    MUT_INSERTASCIINUM,
+    MUT_INSERTASCIINUM,
+    MUT_INSERTASCIINUM,
+    MUT_INSERTASCIINUM,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_OVERWRITE,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_OVERWRITE,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_AUTO_EXTRA_INSERT,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_OVERWRITE,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT,
+
+};
+
 u32 mutation_strategy_exploitation_binary[MUT_STRATEGY_ARRAY_SIZE] = {
 
     MUT_FLIPBIT,
-- 
cgit 1.4.1


From 53b70ef104a334424fd5226c7504130b3bd45625 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Wed, 5 Apr 2023 09:33:09 +0200
Subject: mut changes

---
 include/afl-mutations.h | 83 +++++++++++++++++++++----------------------------
 src/afl-fuzz-one.c      | 17 +---------
 2 files changed, 37 insertions(+), 63 deletions(-)

(limited to 'include')

diff --git a/include/afl-mutations.h b/include/afl-mutations.h
index 31d0898a..9188a37f 100644
--- a/include/afl-mutations.h
+++ b/include/afl-mutations.h
@@ -65,14 +65,13 @@ enum {
   /* 27 */ MUT_DELONE,
   /* 28 */ MUT_INSERTONE,
   /* 29 */ MUT_ASCIINUM,
-  /* 30 */ MUT_NEG,
-  /* 31 */ MUT_INSERTASCIINUM,
-  /* 32 */ MUT_EXTRA_OVERWRITE,
-  /* 33 */ MUT_EXTRA_INSERT,
-  /* 34 */ MUT_AUTO_EXTRA_OVERWRITE,
-  /* 35 */ MUT_AUTO_EXTRA_INSERT,
-  /* 36 */ MUT_SPLICE_OVERWRITE,
-  /* 37 */ MUT_SPLICE_INSERT,
+  /* 30 */ MUT_INSERTASCIINUM,
+  /* 31 */ MUT_EXTRA_OVERWRITE,
+  /* 32 */ MUT_EXTRA_INSERT,
+  /* 33 */ MUT_AUTO_EXTRA_OVERWRITE,
+  /* 34 */ MUT_AUTO_EXTRA_INSERT,
+  /* 35 */ MUT_SPLICE_OVERWRITE,
+  /* 36 */ MUT_SPLICE_INSERT,
 
   MUT_MAX
 
@@ -199,6 +198,7 @@ u32 mutation_strategy_exploration_text[MUT_STRATEGY_ARRAY_SIZE] = {
     MUT_CLONE_INSERT,
     MUT_CLONE_INSERT,
     MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
     MUT_OVERWRITE_COPY,
     MUT_OVERWRITE_COPY,
     MUT_OVERWRITE_COPY,
@@ -233,6 +233,9 @@ u32 mutation_strategy_exploration_text[MUT_STRATEGY_ARRAY_SIZE] = {
     MUT_FLIP8,
     MUT_FLIP8,
     MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
     MUT_SWITCH,
     MUT_SWITCH,
     MUT_SWITCH,
@@ -276,12 +279,8 @@ u32 mutation_strategy_exploration_text[MUT_STRATEGY_ARRAY_SIZE] = {
     MUT_ASCIINUM,
     MUT_ASCIINUM,
     MUT_ASCIINUM,
-    MUT_NEG,
-    MUT_NEG,
-    MUT_NEG,
-    MUT_NEG,
-    MUT_NEG,
-    MUT_NEG,
+    MUT_ASCIINUM,
+    MUT_INSERTASCIINUM,
     MUT_INSERTASCIINUM,
     MUT_INSERTASCIINUM,
     MUT_INSERTASCIINUM,
@@ -335,7 +334,7 @@ u32 mutation_strategy_exploration_text[MUT_STRATEGY_ARRAY_SIZE] = {
     MUT_SPLICE_INSERT,
     MUT_SPLICE_INSERT,
     MUT_SPLICE_INSERT,
-    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT
 
 };
 
@@ -468,6 +467,7 @@ u32 mutation_strategy_exploration_binary[MUT_STRATEGY_ARRAY_SIZE] = {
     MUT_CLONE_INSERT,
     MUT_CLONE_INSERT,
     MUT_CLONE_INSERT,
+    MUT_CLONE_INSERT,
     MUT_OVERWRITE_COPY,
     MUT_OVERWRITE_COPY,
     MUT_OVERWRITE_COPY,
@@ -504,6 +504,10 @@ u32 mutation_strategy_exploration_binary[MUT_STRATEGY_ARRAY_SIZE] = {
     MUT_FLIP8,
     MUT_FLIP8,
     MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
     MUT_SWITCH,
     MUT_SWITCH,
     MUT_SWITCH,
@@ -541,12 +545,6 @@ u32 mutation_strategy_exploration_binary[MUT_STRATEGY_ARRAY_SIZE] = {
     MUT_ASCIINUM,
     MUT_ASCIINUM,
     MUT_ASCIINUM,
-    MUT_NEG,
-    MUT_NEG,
-    MUT_NEG,
-    MUT_NEG,
-    MUT_NEG,
-    MUT_NEG,
     MUT_INSERTASCIINUM,
     MUT_INSERTASCIINUM,
     MUT_INSERTASCIINUM,
@@ -596,7 +594,7 @@ u32 mutation_strategy_exploration_binary[MUT_STRATEGY_ARRAY_SIZE] = {
     MUT_SPLICE_INSERT,
     MUT_SPLICE_INSERT,
     MUT_SPLICE_INSERT,
-    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT
 
 };
 
@@ -766,6 +764,9 @@ u32 mutation_strategy_exploitation_text[MUT_STRATEGY_ARRAY_SIZE] = {
     MUT_FLIP8,
     MUT_FLIP8,
     MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
     MUT_SWITCH,
     MUT_SWITCH,
     MUT_SWITCH,
@@ -785,6 +786,7 @@ u32 mutation_strategy_exploitation_text[MUT_STRATEGY_ARRAY_SIZE] = {
     MUT_SHUFFLE,
     MUT_SHUFFLE,
     MUT_SHUFFLE,
+    MUT_SHUFFLE,
     MUT_DELONE,
     MUT_DELONE,
     MUT_DELONE,
@@ -801,12 +803,8 @@ u32 mutation_strategy_exploitation_text[MUT_STRATEGY_ARRAY_SIZE] = {
     MUT_ASCIINUM,
     MUT_ASCIINUM,
     MUT_ASCIINUM,
-    MUT_NEG,
-    MUT_NEG,
-    MUT_NEG,
-    MUT_NEG,
-    MUT_NEG,
-    MUT_NEG,
+    MUT_ASCIINUM,
+    MUT_INSERTASCIINUM,
     MUT_INSERTASCIINUM,
     MUT_INSERTASCIINUM,
     MUT_INSERTASCIINUM,
@@ -857,7 +855,7 @@ u32 mutation_strategy_exploitation_text[MUT_STRATEGY_ARRAY_SIZE] = {
     MUT_SPLICE_INSERT,
     MUT_SPLICE_INSERT,
     MUT_SPLICE_INSERT,
-    MUT_SPLICE_INSERT,
+    MUT_SPLICE_INSERT
 
 };
 
@@ -959,6 +957,8 @@ u32 mutation_strategy_exploitation_binary[MUT_STRATEGY_ARRAY_SIZE] = {
     MUT_ARITH32BE_,
     MUT_ARITH32BE_,
     MUT_ARITH32BE_,
+    MUT_ARITH32BE_,
+    MUT_ARITH32,
     MUT_ARITH32,
     MUT_ARITH32,
     MUT_ARITH32,
@@ -1027,6 +1027,10 @@ u32 mutation_strategy_exploitation_binary[MUT_STRATEGY_ARRAY_SIZE] = {
     MUT_FLIP8,
     MUT_FLIP8,
     MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
+    MUT_FLIP8,
     MUT_SWITCH,
     MUT_SWITCH,
     MUT_SWITCH,
@@ -1062,12 +1066,6 @@ u32 mutation_strategy_exploitation_binary[MUT_STRATEGY_ARRAY_SIZE] = {
     MUT_ASCIINUM,
     MUT_ASCIINUM,
     MUT_ASCIINUM,
-    MUT_NEG,
-    MUT_NEG,
-    MUT_NEG,
-    MUT_NEG,
-    MUT_NEG,
-    MUT_NEG,
     MUT_INSERTASCIINUM,
     MUT_INSERTASCIINUM,
     MUT_INSERTASCIINUM,
@@ -1118,7 +1116,9 @@ u32 mutation_strategy_exploitation_binary[MUT_STRATEGY_ARRAY_SIZE] = {
     MUT_SPLICE_INSERT,
     MUT_SPLICE_INSERT,
     MUT_SPLICE_INSERT,
-    MUT_SPLICE_INSERT};
+    MUT_SPLICE_INSERT
+
+};
 
 u32 afl_mutate(afl_state_t *, u8 *, u32, u32, bool, bool, u8 *, u32, u32);
 u32 choose_block_len(afl_state_t *, u32);
@@ -1865,17 +1865,6 @@ inline u32 afl_mutate(afl_state_t *afl, u8 *buf, u32 len, u32 steps,
 
       }
 
-      case MUT_NEG: {
-
-        /* Neg byte. */
-
-        item = rand_below(afl, len);
-        buf[item] = ~buf[item];
-
-        break;
-
-      }
-
       case MUT_INSERTASCIINUM: {
 
         u32 len = 1 + rand_below(afl, 8);
diff --git a/src/afl-fuzz-one.c b/src/afl-fuzz-one.c
index e6b58713..bc267b15 100644
--- a/src/afl-fuzz-one.c
+++ b/src/afl-fuzz-one.c
@@ -2659,7 +2659,7 @@ havoc_stage:
 
         case MUT_FLIP8: {
 
-          /* Flip byte. */
+          /* Flip byte with a XOR 0xff. This is the same as NEG. */
 
 #ifdef INTROSPECTION
           snprintf(afl->m_tmp, sizeof(afl->m_tmp), " FLIP8_");
@@ -2987,21 +2987,6 @@ havoc_stage:
 
         }
 
-        case MUT_NEG: {
-
-          /* Neg byte. */
-
-#ifdef INTROSPECTION
-          snprintf(afl->m_tmp, sizeof(afl->m_tmp), " NEG_");
-          strcat(afl->mutation, afl->m_tmp);
-#endif
-          item = rand_below(afl, temp_len);
-
-          out_buf[item] = ~out_buf[item];
-          break;
-
-        }
-
         case MUT_INSERTASCIINUM: {
 
           u32 len = 1 + rand_below(afl, 8);
-- 
cgit 1.4.1


From e313180e4d3f7ba44b773e43af40d4af21088576 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Wed, 5 Apr 2023 10:32:37 +0200
Subject: fix for clang

---
 include/afl-mutations.h | 10 ++++++----
 src/afl-fuzz-one.c      |  9 ++++++---
 2 files changed, 12 insertions(+), 7 deletions(-)

(limited to 'include')

diff --git a/include/afl-mutations.h b/include/afl-mutations.h
index 9188a37f..cc913fb0 100644
--- a/include/afl-mutations.h
+++ b/include/afl-mutations.h
@@ -1231,7 +1231,7 @@ inline u32 afl_mutate(afl_state_t *afl, u8 *buf, u32 len, u32 steps,
 
   for (u32 step = 0; step < steps; ++step) {
 
-  retry_havoc_step:
+  retry_havoc_step : {
 
     u32 r = rand_below(afl, MUT_STRATEGY_ARRAY_SIZE), item;
 
@@ -1667,10 +1667,10 @@ inline u32 afl_mutate(afl_state_t *afl, u8 *buf, u32 len, u32 steps,
 
         if (unlikely(len < 4)) { break; }  // no retry
 
-        u32 len = choose_block_len(afl, len - 1);
-        u32 off = rand_below(afl, len - len + 1);
+        u32 blen = choose_block_len(afl, len - 1);
+        u32 off = rand_below(afl, len - blen + 1);
 
-        for (u32 i = len - 1; i > 0; i--) {
+        for (u32 i = blen - 1; i > 0; i--) {
 
           u32 j;
           do {
@@ -2030,6 +2030,8 @@ inline u32 afl_mutate(afl_state_t *afl, u8 *buf, u32 len, u32 steps,
 
   }
 
+  }
+
   return len;
 
 }
diff --git a/src/afl-fuzz-one.c b/src/afl-fuzz-one.c
index 48aa6eb0..e6ff1d1a 100644
--- a/src/afl-fuzz-one.c
+++ b/src/afl-fuzz-one.c
@@ -2139,8 +2139,8 @@ havoc_stage:
 
         LIST_FOREACH(&afl->custom_mutator_list, struct custom_mutator, {
 
-          if (el->stacked_custom &&
-              rand_below(afl, 100) < el->stacked_custom_prob) {
+          if (unlikely(el->stacked_custom &&
+                       rand_below(afl, 100) < el->stacked_custom_prob)) {
 
             u8    *custom_havoc_buf = NULL;
             size_t new_len = el->afl_custom_havoc_mutation(
@@ -2170,7 +2170,8 @@ havoc_stage:
 
       }
 
-    retry_havoc_step:
+    retry_havoc_step : {
+
       u32 r = rand_below(afl, MUT_STRATEGY_ARRAY_SIZE), item;
 
       switch (mutation_array[r]) {
@@ -3250,6 +3251,8 @@ havoc_stage:
 
     }
 
+    }
+
     if (common_fuzz_stuff(afl, out_buf, temp_len)) { goto abandon_entry; }
 
     /* out_buf might have been mangled a bit, so let's restore it to its
-- 
cgit 1.4.1


From a74561b0e7434282ad214ca634b5c19c2f345e8e Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Wed, 5 Apr 2023 12:12:05 +0200
Subject: implement switch mode

---
 include/afl-fuzz.h    |  6 +++---
 src/afl-fuzz-bitmap.c | 12 ++++++++++++
 src/afl-fuzz-state.c  |  2 +-
 src/afl-fuzz.c        | 30 ++++++++++++++++++++++++++----
 4 files changed, 42 insertions(+), 8 deletions(-)

(limited to 'include')

diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index 6573eabf..23a04f42 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -491,8 +491,7 @@ typedef struct afl_state {
       *infoexec;                       /* Command to execute on a new crash */
 
   u32 hang_tmout,                       /* Timeout used for hang det (ms)   */
-      stats_update_freq,                /* Stats update frequency (execs)   */
-      switch_fuzz_mode;                 /* auto or fixed fuzz mode          */
+      stats_update_freq;                /* Stats update frequency (execs)   */
 
   u8 havoc_stack_pow2,                  /* HAVOC_STACK_POW2                 */
       no_unlink,                        /* do not unlink cur_input          */
@@ -592,7 +591,8 @@ typedef struct afl_state {
       last_hang_time,                   /* Time for most recent hang (ms)   */
       longest_find_time,                /* Longest time taken for a find    */
       exit_on_time,                     /* Delay to exit if no new paths    */
-      sync_time;                        /* Sync time (ms)                   */
+      sync_time,                        /* Sync time (ms)                   */
+      switch_fuzz_mode;                 /* auto or fixed fuzz mode          */
 
   u32 slowest_exec_ms,                  /* Slowest testcase non hang in ms  */
       subseq_tmouts;                    /* Number of timeouts in a row      */
diff --git a/src/afl-fuzz-bitmap.c b/src/afl-fuzz-bitmap.c
index d9c792d1..a937c96d 100644
--- a/src/afl-fuzz-bitmap.c
+++ b/src/afl-fuzz-bitmap.c
@@ -529,6 +529,18 @@ save_if_interesting(afl_state_t *afl, void *mem, u32 len, u8 fault) {
     close(fd);
     add_to_queue(afl, queue_fn, len, 0);
 
+    if (unlikely(afl->fuzz_mode) && likely(afl->switch_fuzz_mode)) {
+
+      if (afl->afl_env.afl_no_ui) {
+
+        ACTF("New coverage found, switching back to exploration mode.");
+
+      }
+
+      afl->fuzz_mode = 0;
+
+    }
+
 #ifdef INTROSPECTION
     if (afl->custom_mutators_count && afl->current_custom_fuzz) {
 
diff --git a/src/afl-fuzz-state.c b/src/afl-fuzz-state.c
index 907861e9..9dc258b1 100644
--- a/src/afl-fuzz-state.c
+++ b/src/afl-fuzz-state.c
@@ -108,7 +108,7 @@ void afl_state_init(afl_state_t *afl, uint32_t map_size) {
   afl->cmplog_lvl = 2;
   afl->min_length = 1;
   afl->max_length = MAX_FILE;
-  afl->switch_fuzz_mode = STRATEGY_SWITCH_TIME;
+  afl->switch_fuzz_mode = STRATEGY_SWITCH_TIME * 1000;
 #ifndef NO_SPLICING
   afl->use_splicing = 1;
 #endif
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index 315107d7..c50b271b 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -577,7 +577,7 @@ int main(int argc, char **argv_orig, char **envp) {
         if (!stricmp(optarg, "explore") || !stricmp(optarg, "exploration")) {
 
           afl->fuzz_mode = 0;
-          afl->switch_fuzz_mode = 1;
+          afl->switch_fuzz_mode = 0;
 
         } else if (!stricmp(optarg, "exploit") ||
 
@@ -588,12 +588,16 @@ int main(int argc, char **argv_orig, char **envp) {
 
         } else {
 
-          if ((s32)(afl->switch_fuzz_mode = (u32)atoi(optarg)) < 1) {
+          if ((afl->switch_fuzz_mode = (u32)atoi(optarg)) > INT_MAX) {
 
             FATAL(
                 "Parameter for option -P must be \"explore\", \"exploit\" or a "
                 "number!");
 
+          } else {
+
+            afl->switch_fuzz_mode *= 1000;
+
           }
 
         }
@@ -2689,13 +2693,31 @@ int main(int argc, char **argv_orig, char **envp) {
 
     } while (skipped_fuzz && afl->queue_cur && !afl->stop_soon);
 
+    u64 cur_time = get_cur_time();
+
+    if (likely(afl->switch_fuzz_mode && afl->fuzz_mode == 0) &&
+        unlikely(cur_time > afl->last_find_time + afl->switch_fuzz_mode)) {
+
+      if (afl->afl_env.afl_no_ui) {
+
+        ACTF(
+            "No new coverage found for %llu seconds, switching to exploitation "
+            "strategy.",
+            afl->switch_fuzz_mode / 1000);
+
+      }
+
+      afl->fuzz_mode = 1;
+
+    }
+
     if (likely(!afl->stop_soon && afl->sync_id)) {
 
       if (likely(afl->skip_deterministic)) {
 
         if (unlikely(afl->is_main_node)) {
 
-          if (unlikely(get_cur_time() >
+          if (unlikely(cur_time >
                        (afl->sync_time >> 1) + afl->last_sync_time)) {
 
             if (!(sync_interval_cnt++ % (SYNC_INTERVAL / 3))) {
@@ -2708,7 +2730,7 @@ int main(int argc, char **argv_orig, char **envp) {
 
         } else {
 
-          if (unlikely(get_cur_time() > afl->sync_time + afl->last_sync_time)) {
+          if (unlikely(cur_time > afl->sync_time + afl->last_sync_time)) {
 
             if (!(sync_interval_cnt++ % SYNC_INTERVAL)) { sync_fuzzers(afl); }
 
-- 
cgit 1.4.1


From 400c5e92cb5ed304a2c14a79597100850cf9f82c Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Fri, 7 Apr 2023 09:41:22 +0200
Subject: renaming

---
 include/afl-mutations.h | 166 ++++++++++++++++++++++++------------------------
 src/afl-fuzz-one.c      |  30 +++++----
 2 files changed, 99 insertions(+), 97 deletions(-)

(limited to 'include')

diff --git a/include/afl-mutations.h b/include/afl-mutations.h
index cc913fb0..8d40855d 100644
--- a/include/afl-mutations.h
+++ b/include/afl-mutations.h
@@ -52,8 +52,8 @@ enum {
   /* 14 */ MUT_ARITH32,
   /* 15 */ MUT_ARITH32BE,
   /* 16 */ MUT_RAND8,
-  /* 17 */ MUT_CLONE_OVERWRITE,
-  /* 18 */ MUT_CLONE_INSERT,
+  /* 17 */ MUT_CLONE_COPY,
+  /* 18 */ MUT_CLONE_FIXED,
   /* 19 */ MUT_OVERWRITE_COPY,
   /* 20 */ MUT_OVERWRITE_FIXED,
   /* 21 */ MUT_BYTEADD,
@@ -176,29 +176,29 @@ u32 mutation_strategy_exploration_text[MUT_STRATEGY_ARRAY_SIZE] = {
     MUT_RAND8,
     MUT_RAND8,
     MUT_RAND8,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
     MUT_OVERWRITE_COPY,
     MUT_OVERWRITE_COPY,
     MUT_OVERWRITE_COPY,
@@ -446,28 +446,28 @@ u32 mutation_strategy_exploration_binary[MUT_STRATEGY_ARRAY_SIZE] = {
     MUT_RAND8,
     MUT_RAND8,
     MUT_RAND8,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
     MUT_OVERWRITE_COPY,
     MUT_OVERWRITE_COPY,
     MUT_OVERWRITE_COPY,
@@ -716,23 +716,23 @@ u32 mutation_strategy_exploitation_text[MUT_STRATEGY_ARRAY_SIZE] = {
     MUT_RAND8,
     MUT_RAND8,
     MUT_RAND8,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
     MUT_OVERWRITE_COPY,
     MUT_OVERWRITE_COPY,
     MUT_OVERWRITE_COPY,
@@ -979,23 +979,23 @@ u32 mutation_strategy_exploitation_binary[MUT_STRATEGY_ARRAY_SIZE] = {
     MUT_RAND8,
     MUT_RAND8,
     MUT_RAND8,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_OVERWRITE,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
-    MUT_CLONE_INSERT,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
+    MUT_CLONE_FIXED,
     MUT_OVERWRITE_COPY,
     MUT_OVERWRITE_COPY,
     MUT_OVERWRITE_COPY,
@@ -1455,7 +1455,7 @@ inline u32 afl_mutate(afl_state_t *afl, u8 *buf, u32 len, u32 steps,
 
       }
 
-      case MUT_CLONE_OVERWRITE: {
+      case MUT_CLONE_COPY: {
 
         if (likely(len + HAVOC_BLK_XL < max_len)) {
 
@@ -1494,7 +1494,7 @@ inline u32 afl_mutate(afl_state_t *afl, u8 *buf, u32 len, u32 steps,
 
       }
 
-      case MUT_CLONE_INSERT: {
+      case MUT_CLONE_FIXED: {
 
         if (likely(len + HAVOC_BLK_XL < max_len)) {
 
diff --git a/src/afl-fuzz-one.c b/src/afl-fuzz-one.c
index e6ff1d1a..f5ddea0e 100644
--- a/src/afl-fuzz-one.c
+++ b/src/afl-fuzz-one.c
@@ -2481,7 +2481,7 @@ havoc_stage:
 
         }
 
-        case MUT_CLONE_OVERWRITE: {
+        case MUT_CLONE_COPY: {
 
           if (likely(temp_len + HAVOC_BLK_XL < MAX_FILE)) {
 
@@ -2493,7 +2493,7 @@ havoc_stage:
 
 #ifdef INTROSPECTION
             snprintf(afl->m_tmp, sizeof(afl->m_tmp), " CLONE-%s_%u_%u_%u",
-                     "overwrite", clone_from, clone_to, clone_len);
+                     "COPY", clone_from, clone_to, clone_len);
             strcat(afl->mutation, afl->m_tmp);
 #endif
             u8 *new_buf =
@@ -2530,7 +2530,7 @@ havoc_stage:
 
         }
 
-        case MUT_CLONE_INSERT: {
+        case MUT_CLONE_FIXED: {
 
           if (likely(temp_len + HAVOC_BLK_XL < MAX_FILE)) {
 
@@ -2544,7 +2544,7 @@ havoc_stage:
 
 #ifdef INTROSPECTION
             snprintf(afl->m_tmp, sizeof(afl->m_tmp), " CLONE-%s_%u_%u_%u",
-                     "insert", strat, clone_to, clone_len);
+                     "FIXED", strat, clone_to, clone_len);
             strcat(afl->mutation, afl->m_tmp);
 #endif
             u8 *new_buf =
@@ -2587,20 +2587,22 @@ havoc_stage:
 
           if (unlikely(temp_len < 2)) { break; }  // no retry
 
-          u32 copy_len = choose_block_len(afl, temp_len - 1);
-          u32 copy_from = rand_below(afl, temp_len - copy_len + 1);
-          u32 copy_to = rand_below(afl, temp_len - copy_len + 1);
+          u32 copy_from, copy_to,
+              copy_len = choose_block_len(afl, temp_len - 1);
+
+          do {
 
-          if (likely(copy_from != copy_to)) {
+            copy_from = rand_below(afl, temp_len - copy_len + 1);
+            copy_to = rand_below(afl, temp_len - copy_len + 1);
+
+          } while (unlikely(copy_from == copy_to));
 
 #ifdef INTROSPECTION
-            snprintf(afl->m_tmp, sizeof(afl->m_tmp), " OVERWRITE-COPY_%u_%u_%u",
-                     copy_from, copy_to, copy_len);
-            strcat(afl->mutation, afl->m_tmp);
+          snprintf(afl->m_tmp, sizeof(afl->m_tmp), " OVERWRITE-COPY_%u_%u_%u",
+                   copy_from, copy_to, copy_len);
+          strcat(afl->mutation, afl->m_tmp);
 #endif
-            memmove(out_buf + copy_to, out_buf + copy_from, copy_len);
-
-          }
+          memmove(out_buf + copy_to, out_buf + copy_from, copy_len);
 
           break;
 
-- 
cgit 1.4.1


From 3e84d6a2ae7df5f6b9073a91ccc6acef50b45aab Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Thu, 27 Apr 2023 11:49:00 +0200
Subject: afl++ -> AFL++

---
 Dockerfile                                 |  4 ++--
 GNUmakefile                                | 12 ++++++------
 GNUmakefile.gcc_plugin                     |  6 +++---
 GNUmakefile.llvm                           |  6 +++---
 afl-cmin                                   |  2 +-
 docs/Changelog.md                          |  2 +-
 docs/INSTALL.md                            |  2 +-
 include/alloc-inl.h                        |  2 +-
 instrumentation/SanitizerCoverageLTO.so.cc | 30 +++++++++++++++---------------
 instrumentation/afl-llvm-common.cc         |  2 +-
 instrumentation/afl-llvm-dict2file.so.cc   |  2 +-
 qemu_mode/build_qemu_support.sh            |  4 ++--
 src/afl-cc.c                               |  2 +-
 src/afl-forkserver.c                       |  6 +++---
 src/afl-fuzz.c                             | 10 +++++-----
 src/afl-ld-lto.c                           |  4 ++--
 test/test-dlopen.c                         |  2 +-
 test/test-gcc-plugin.sh                    |  2 +-
 test/test-performance.sh                   |  4 ++--
 test/test-pre.sh                           |  2 +-
 unicorn_mode/build_unicorn_support.sh      |  2 +-
 21 files changed, 54 insertions(+), 54 deletions(-)

(limited to 'include')

diff --git a/Dockerfile b/Dockerfile
index 4e53de40..1b5ffd28 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -6,7 +6,7 @@
 #
 
 FROM ubuntu:22.04 AS aflplusplus
-LABEL "maintainer"="afl++ team <afl@aflplus.plus>"
+LABEL "maintainer"="AFL++ team <afl@aflplus.plus>"
 LABEL "about"="AFLplusplus container image"
 
 ### Comment out to enable these features
@@ -94,4 +94,4 @@ RUN sed -i.bak 's/^	-/	/g' GNUmakefile && \
 RUN echo "set encoding=utf-8" > /root/.vimrc && \
     echo ". /etc/bash_completion" >> ~/.bashrc && \
     echo 'alias joe="joe --wordwrap --joe_state -nobackup"' >> ~/.bashrc && \
-    echo "export PS1='"'[afl++ \h] \w \$ '"'" >> ~/.bashrc
+    echo "export PS1='"'[AFL++ \h] \w \$ '"'" >> ~/.bashrc
diff --git a/GNUmakefile b/GNUmakefile
index 23cae65d..5900ad61 100644
--- a/GNUmakefile
+++ b/GNUmakefile
@@ -39,7 +39,7 @@ ASAN_OPTIONS=detect_leaks=0
 SYS = $(shell uname -s)
 ARCH = $(shell uname -m)
 
-$(info [*] Compiling afl++ for OS $(SYS) on ARCH $(ARCH))
+$(info [*] Compiling AFL++ for OS $(SYS) on ARCH $(ARCH))
 
 ifdef NO_SPLICING
   override CFLAGS_OPT += -DNO_SPLICING
@@ -359,7 +359,7 @@ performance-test:	source-only
 help:
 	@echo "HELP --- the following make targets exist:"
 	@echo "=========================================="
-	@echo "all: the main afl++ binaries and llvm/gcc instrumentation"
+	@echo "all: the main AFL++ binaries and llvm/gcc instrumentation"
 	@echo "binary-only: everything for binary-only fuzzing: frida_mode, nyx_mode, qemu_mode, frida_mode, unicorn_mode, coresight_mode, libdislocator, libtokencap"
 	@echo "source-only: everything for source code fuzzing: nyx_mode, libdislocator, libtokencap"
 	@echo "distrib: everything (for both binary-only and source code fuzzing)"
@@ -367,7 +367,7 @@ help:
 	@echo "install: installs everything you have compiled with the build option above"
 	@echo "clean: cleans everything compiled (not downloads when on a checkout)"
 	@echo "deepclean: cleans everything including downloads"
-	@echo "uninstall: uninstall afl++ from the system"
+	@echo "uninstall: uninstall AFL++ from the system"
 	@echo "code-format: format the code, do this before you commit and send a PR please!"
 	@echo "tests: this runs the test framework. It is more catered for the developers, but if you run into problems this helps pinpointing the problem"
 	@echo "unit: perform unit tests (based on cmocka and GNU linker)"
@@ -749,7 +749,7 @@ endif
 	@echo
 
 %.8:	%
-	@echo .TH $* 8 $(BUILD_DATE) "afl++" > $@
+	@echo .TH $* 8 $(BUILD_DATE) "AFL++" > $@
 	@echo .SH NAME >> $@
 	@echo .B $* >> $@
 	@echo >> $@
@@ -761,8 +761,8 @@ endif
 	@./$* -hh 2>&1 | tail -n +4 >> $@
 	@echo >> $@
 	@echo .SH AUTHOR >> $@
-	@echo "afl++ was written by Michal \"lcamtuf\" Zalewski and is maintained by Marc \"van Hauser\" Heuse <mh@mh-sec.de>, Heiko \"hexcoder-\" Eissfeldt <heiko.eissfeldt@hexco.de>, Andrea Fioraldi <andreafioraldi@gmail.com> and Dominik Maier <domenukk@gmail.com>" >> $@
-	@echo  The homepage of afl++ is: https://github.com/AFLplusplus/AFLplusplus >> $@
+	@echo "AFL++ was written by Michal \"lcamtuf\" Zalewski and is maintained by Marc \"van Hauser\" Heuse <mh@mh-sec.de>, Dominik Maier <domenukk@gmail.com>, Andrea Fioraldi <andreafioraldi@gmail.com> and Heiko \"hexcoder-\" Eissfeldt <heiko.eissfeldt@hexco.de>" >> $@
+	@echo  The homepage of AFL++ is: https://github.com/AFLplusplus/AFLplusplus >> $@
 	@echo >> $@
 	@echo .SH LICENSE >> $@
 	@echo Apache License Version 2.0, January 2004 >> $@
diff --git a/GNUmakefile.gcc_plugin b/GNUmakefile.gcc_plugin
index 4c4e10c4..41face4c 100644
--- a/GNUmakefile.gcc_plugin
+++ b/GNUmakefile.gcc_plugin
@@ -175,7 +175,7 @@ all_done: test_build
 .NOTPARALLEL: clean
 
 %.8: %
-	@echo .TH $* 8 `date "+%Y-%m-%d"` "afl++" > ./$@
+	@echo .TH $* 8 `date "+%Y-%m-%d"` "AFL++" > ./$@
 	@echo .SH NAME >> ./$@
 	@echo .B $* >> ./$@
 	@echo >> ./$@
@@ -187,8 +187,8 @@ all_done: test_build
 	@./$* -h 2>&1 | tail -n +4 >> ./$@
 	@echo >> ./$@
 	@echo .SH AUTHOR >> ./$@
-	@echo "afl++ was written by Michal \"lcamtuf\" Zalewski and is maintained by Marc \"van Hauser\" Heuse <mh@mh-sec.de>, Heiko \"hexcoder-\" Eissfeldt <heiko.eissfeldt@hexco.de>, Andrea Fioraldi <andreafioraldi@gmail.com> and Dominik Maier <domenukk@gmail.com>" >> ./$@
-	@echo  The homepage of afl++ is: https://github.com/AFLplusplus/AFLplusplus >> ./$@
+	@echo "AFL++ was written by Michal \"lcamtuf\" Zalewski and is maintained by Marc \"van Hauser\" Heuse <mh@mh-sec.de>, Dominik Maier <domenukk@gmail.com>, Andrea Fioraldi <andreafioraldi@gmail.com> and Heiko \"hexcoder-\" Eissfeldt <heiko.eissfeldt@hexco.de>" >> ./$@
+	@echo  The homepage of AFL++ is: https://github.com/AFLplusplus/AFLplusplus >> ./$@
 	@echo >> ./$@
 	@echo .SH LICENSE >> ./$@
 	@echo Apache License Version 2.0, January 2004 >> ./$@
diff --git a/GNUmakefile.llvm b/GNUmakefile.llvm
index a053403b..c1b006ba 100644
--- a/GNUmakefile.llvm
+++ b/GNUmakefile.llvm
@@ -510,7 +510,7 @@ install: all
 	install -m 644 instrumentation/README.*.md $${DESTDIR}$(DOC_PATH)/
 
 %.8: %
-	@echo .TH $* 8 $(BUILD_DATE) "afl++" > ./$@
+	@echo .TH $* 8 $(BUILD_DATE) "AFL++" > ./$@
 	@echo .SH NAME >> ./$@
 	@printf "%s" ".B $* \- " >> ./$@
 	@./$* -h 2>&1 | head -n 1 | sed -e "s/$$(printf '\e')[^m]*m//g" >> ./$@
@@ -524,8 +524,8 @@ install: all
 	@./$* -h 2>&1 | tail -n +4 >> ./$@
 	@echo >> ./$@
 	@echo .SH AUTHOR >> ./$@
-	@echo "afl++ was written by Michal \"lcamtuf\" Zalewski and is maintained by Marc \"van Hauser\" Heuse <mh@mh-sec.de>, Heiko \"hexcoder-\" Eissfeldt <heiko.eissfeldt@hexco.de>, Andrea Fioraldi <andreafioraldi@gmail.com> and Dominik Maier <domenukk@gmail.com>" >> ./$@
-	@echo  The homepage of afl++ is: https://github.com/AFLplusplus/AFLplusplus >> ./$@
+	@echo "AFL++ was written by Michal \"lcamtuf\" Zalewski and is maintained by Marc \"van Hauser\" Heuse <mh@mh-sec.de>, Dominik Maier <domenukk@gmail.com>, Andrea Fioraldi <andreafioraldi@gmail.com> and Heiko \"hexcoder-\" Eissfeldt <heiko.eissfeldt@hexco.de>" >> ./$@
+	@echo  The homepage of AFL++ is: https://github.com/AFLplusplus/AFLplusplus >> ./$@
 	@echo >> ./$@
 	@echo .SH LICENSE >> ./$@
 	@echo Apache License Version 2.0, January 2004 >> ./$@
diff --git a/afl-cmin b/afl-cmin
index 63cfdd7e..ae723c1b 100755
--- a/afl-cmin
+++ b/afl-cmin
@@ -149,7 +149,7 @@ BEGIN {
     redirected = 0
   }
 
-  print "corpus minimization tool for afl++ (awk version)\n"
+  print "corpus minimization tool for AFL++ (awk version)\n"
 
   # defaults
   extra_par = ""
diff --git a/docs/Changelog.md b/docs/Changelog.md
index 20b915fa..cd5ed9fc 100644
--- a/docs/Changelog.md
+++ b/docs/Changelog.md
@@ -229,7 +229,7 @@
     afl-showmap and other tools.
   - afl-cc:
     - detect overflow reads on initial input buffer for asan
-    - new cmplog mode (incompatible with older afl++ versions)
+    - new cmplog mode (incompatible with older AFL++ versions)
     - support llvm IR select instrumentation for default PCGUARD and LTO
     - fix for shared linking on MacOS
     - better selective instrumentation AFL_LLVM_{ALLOW|DENY}LIST
diff --git a/docs/INSTALL.md b/docs/INSTALL.md
index 591b7ded..c54cb9ad 100644
--- a/docs/INSTALL.md
+++ b/docs/INSTALL.md
@@ -51,7 +51,7 @@ make source-only
 
 These build targets exist:
 
-* all: the main afl++ binaries and llvm/gcc instrumentation
+* all: the main AFL++ binaries and llvm/gcc instrumentation
 * binary-only: everything for binary-only fuzzing: frida_mode, nyx_mode,
   qemu_mode, frida_mode, unicorn_mode, coresight_mode, libdislocator,
   libtokencap
diff --git a/include/alloc-inl.h b/include/alloc-inl.h
index ae37028e..bbb42e88 100644
--- a/include/alloc-inl.h
+++ b/include/alloc-inl.h
@@ -42,7 +42,7 @@
 // Be careful! _WANT_ORIGINAL_AFL_ALLOC is not compatible with custom mutators
 
 #ifndef _WANT_ORIGINAL_AFL_ALLOC
-  // afl++ stuff without memory corruption checks - for speed
+  // AFL++ stuff without memory corruption checks - for speed
 
   /* User-facing macro to sprintf() to a dynamically allocated buffer. */
 
diff --git a/instrumentation/SanitizerCoverageLTO.so.cc b/instrumentation/SanitizerCoverageLTO.so.cc
index 42583f9e..6a719737 100644
--- a/instrumentation/SanitizerCoverageLTO.so.cc
+++ b/instrumentation/SanitizerCoverageLTO.so.cc
@@ -1,4 +1,4 @@
-/* SanitizeCoverage.cpp ported to afl++ LTO :-) */
+/* SanitizeCoverage.cpp ported to AFL++ LTO :-) */
 
 #define AFL_LLVM_PASS
 
@@ -234,7 +234,7 @@ class ModuleSanitizerCoverageLTO
 
   SanitizerCoverageOptions Options;
 
-  // afl++ START
+  // AFL++ START
   // const SpecialCaseList *          Allowlist;
   // const SpecialCaseList *          Blocklist;
   uint32_t                         autodictionary = 1;
@@ -260,7 +260,7 @@ class ModuleSanitizerCoverageLTO
   Value                           *MapPtrFixed = NULL;
   std::ofstream                    dFile;
   size_t                           found = 0;
-  // afl++ END
+  // AFL++ END
 
 };
 
@@ -404,7 +404,7 @@ bool ModuleSanitizerCoverageLTO::instrumentModule(
   Int8Ty = IRB.getInt8Ty();
   Int1Ty = IRB.getInt1Ty();
 
-  /* afl++ START */
+  /* AFL++ START */
   char        *ptr;
   LLVMContext &Ctx = M.getContext();
   Ct = &Ctx;
@@ -978,7 +978,7 @@ bool ModuleSanitizerCoverageLTO::instrumentModule(
 
   }
 
-  // afl++ END
+  // AFL++ END
 
   SanCovTracePCIndir =
       M.getOrInsertFunction(SanCovTracePCIndirName, VoidTy, IntptrTy);
@@ -1002,7 +1002,7 @@ bool ModuleSanitizerCoverageLTO::instrumentModule(
   for (auto &F : M)
     instrumentFunction(F, DTCallback, PDTCallback);
 
-  // afl++ START
+  // AFL++ START
   if (dFile.is_open()) dFile.close();
 
   if (!getenv("AFL_LLVM_LTO_SKIPINIT") &&
@@ -1156,7 +1156,7 @@ bool ModuleSanitizerCoverageLTO::instrumentModule(
 
   }
 
-  // afl++ END
+  // AFL++ END
 
   // We don't reference these arrays directly in any of our runtime functions,
   // so we need to prevent them from being dead stripped.
@@ -1213,10 +1213,10 @@ static bool shouldInstrumentBlock(const Function &F, const BasicBlock *BB,
   // (catchswitch blocks).
   if (BB->getFirstInsertionPt() == BB->end()) return false;
 
-  // afl++ START
+  // AFL++ START
   if (!Options.NoPrune && &F.getEntryBlock() == BB && F.size() > 1)
     return false;
-  // afl++ END
+  // AFL++ END
 
   if (Options.NoPrune || &F.getEntryBlock() == BB) return true;
 
@@ -1258,10 +1258,10 @@ void ModuleSanitizerCoverageLTO::instrumentFunction(
   // if (Blocklist && Blocklist->inSection("coverage", "fun", F.getName()))
   // return;
 
-  // afl++ START
+  // AFL++ START
   if (!F.size()) return;
   if (!isInInstrumentList(&F, FMNAME)) return;
-  // afl++ END
+  // AFL++ END
 
   if (Options.CoverageType >= SanitizerCoverageOptions::SCK_Edge)
     SplitAllCriticalEdges(
@@ -1559,7 +1559,7 @@ bool ModuleSanitizerCoverageLTO::InjectCoverage(
 
   for (size_t i = 0, N = AllBlocks.size(); i < N; i++) {
 
-    // afl++ START
+    // AFL++ START
     if (BlockList.size()) {
 
       int skip = 0;
@@ -1581,7 +1581,7 @@ bool ModuleSanitizerCoverageLTO::InjectCoverage(
 
     }
 
-    // afl++ END
+    // AFL++ END
 
     InjectCoverageAtBlock(F, *AllBlocks[i], i, IsLeafFunc);
 
@@ -1647,7 +1647,7 @@ void ModuleSanitizerCoverageLTO::InjectCoverageAtBlock(Function   &F,
 
   if (Options.TracePCGuard) {
 
-    // afl++ START
+    // AFL++ START
     ++afl_global_id;
 
     if (dFile.is_open()) {
@@ -1711,7 +1711,7 @@ void ModuleSanitizerCoverageLTO::InjectCoverageAtBlock(Function   &F,
     // done :)
 
     inst++;
-    // afl++ END
+    // AFL++ END
 
     /*
     XXXXXXXXXXXXXXXXXXX
diff --git a/instrumentation/afl-llvm-common.cc b/instrumentation/afl-llvm-common.cc
index 5d82aa25..7f17b02d 100644
--- a/instrumentation/afl-llvm-common.cc
+++ b/instrumentation/afl-llvm-common.cc
@@ -584,7 +584,7 @@ bool isInInstrumentList(llvm::Function *F, std::string Filename) {
 }
 
 // Calculate the number of average collisions that would occur if all
-// location IDs would be assigned randomly (like normal afl/afl++).
+// location IDs would be assigned randomly (like normal afl/AFL++).
 // This uses the "balls in bins" algorithm.
 unsigned long long int calculateCollisions(uint32_t edges) {
 
diff --git a/instrumentation/afl-llvm-dict2file.so.cc b/instrumentation/afl-llvm-dict2file.so.cc
index 97f1d47f..cf368e35 100644
--- a/instrumentation/afl-llvm-dict2file.so.cc
+++ b/instrumentation/afl-llvm-dict2file.so.cc
@@ -744,7 +744,7 @@ static void registerAFLdict2filePass(const PassManagerBuilder &,
 }
 
 static RegisterPass<AFLdict2filePass> X("afl-dict2file",
-                                        "afl++ dict2file instrumentation pass",
+                                        "AFL++ dict2file instrumentation pass",
                                         false, false);
 
 static RegisterStandardPasses RegisterAFLdict2filePass(
diff --git a/qemu_mode/build_qemu_support.sh b/qemu_mode/build_qemu_support.sh
index a064fe58..f59cba78 100755
--- a/qemu_mode/build_qemu_support.sh
+++ b/qemu_mode/build_qemu_support.sh
@@ -356,7 +356,7 @@ fi
 
 if ! command -v "$CROSS" > /dev/null ; then
   if [ "$CPU_TARGET" = "$(uname -m)" ] ; then
-    echo "[+] Building afl++ qemu support libraries with CC=$CC"
+    echo "[+] Building AFL++ qemu support libraries with CC=$CC"
     echo "[+] Building libcompcov ..."
     make -C libcompcov && echo "[+] libcompcov ready"
     echo "[+] Building unsigaction ..."
@@ -371,7 +371,7 @@ if ! command -v "$CROSS" > /dev/null ; then
     echo "[!] Cross compiler $CROSS could not be found, cannot compile libcompcov libqasan and unsigaction"
   fi
 else
-  echo "[+] Building afl++ qemu support libraries with CC=\"$CROSS $CROSS_FLAGS\""
+  echo "[+] Building AFL++ qemu support libraries with CC=\"$CROSS $CROSS_FLAGS\""
   echo "[+] Building libcompcov ..."
   make -C libcompcov CC="$CROSS $CROSS_FLAGS" && echo "[+] libcompcov ready"
   echo "[+] Building unsigaction ..."
diff --git a/src/afl-cc.c b/src/afl-cc.c
index d1001187..99ce39d4 100644
--- a/src/afl-cc.c
+++ b/src/afl-cc.c
@@ -642,7 +642,7 @@ static void edit_params(u32 argc, char **argv, char **envp) {
     }
 
     //#if LLVM_MAJOR >= 13
-    //    // Use the old pass manager in LLVM 14 which the afl++ passes still
+    //    // Use the old pass manager in LLVM 14 which the AFL++ passes still
     //    use. cc_params[cc_par_cnt++] = "-flegacy-pass-manager";
     //#endif
 
diff --git a/src/afl-forkserver.c b/src/afl-forkserver.c
index aa8c8622..30c8901c 100644
--- a/src/afl-forkserver.c
+++ b/src/afl-forkserver.c
@@ -489,7 +489,7 @@ static void report_error_and_exit(int error) {
       break;
     case FS_ERROR_OLD_CMPLOG:
       FATAL(
-          "the -c cmplog target was instrumented with an too old afl++ "
+          "the -c cmplog target was instrumented with an too old AFL++ "
           "version, you need to recompile it.");
       break;
     case FS_ERROR_OLD_CMPLOG_QEMU:
@@ -987,7 +987,7 @@ void afl_fsrv_start(afl_forkserver_t *fsrv, char **argv,
 
     if ((status & FS_OPT_ENABLED) == FS_OPT_ENABLED) {
 
-      // workaround for recent afl++ versions
+      // workaround for recent AFL++ versions
       if ((status & FS_OPT_OLD_AFLPP_WORKAROUND) == FS_OPT_OLD_AFLPP_WORKAROUND)
         status = (status & 0xf0ffffff);
 
@@ -1059,7 +1059,7 @@ void afl_fsrv_start(afl_forkserver_t *fsrv, char **argv,
 
           FATAL(
               "Target's coverage map size of %u is larger than the one this "
-              "afl++ is set with (%u). Either set AFL_MAP_SIZE=%u and restart "
+              "AFL++ is set with (%u). Either set AFL_MAP_SIZE=%u and restart "
               " afl-fuzz, or change MAP_SIZE_POW2 in config.h and recompile "
               "afl-fuzz",
               tmp_map_size, fsrv->map_size, tmp_map_size);
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index 71d2afd8..646dc50b 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -1280,16 +1280,16 @@ int main(int argc, char **argv_orig, char **envp) {
 
   if (afl->fsrv.mem_limit && afl->shm.cmplog_mode) afl->fsrv.mem_limit += 260;
 
-  OKF("afl++ is maintained by Marc \"van Hauser\" Heuse, Heiko \"hexcoder\" "
-      "Eißfeldt, Andrea Fioraldi and Dominik Maier");
-  OKF("afl++ is open source, get it at "
+  OKF("AFL++ is maintained by Marc \"van Hauser\" Heuse, Dominik Maier, Andrea "
+      "Fioraldi and Heiko \"hexcoder\" Eißfeldt");
+  OKF("AFL++ is open source, get it at "
       "https://github.com/AFLplusplus/AFLplusplus");
-  OKF("NOTE: afl++ >= v3 has changed defaults and behaviours - see README.md");
+  OKF("NOTE: AFL++ >= v3 has changed defaults and behaviours - see README.md");
 
   #ifdef __linux__
   if (afl->fsrv.nyx_mode) {
 
-    OKF("afl++ Nyx mode is enabled (developed and mainted by Sergej Schumilo)");
+    OKF("AFL++ Nyx mode is enabled (developed and mainted by Sergej Schumilo)");
     OKF("Nyx is open source, get it at https://github.com/Nyx-Fuzz");
 
   }
diff --git a/src/afl-ld-lto.c b/src/afl-ld-lto.c
index 5438bd9f..420dd817 100644
--- a/src/afl-ld-lto.c
+++ b/src/afl-ld-lto.c
@@ -2,7 +2,7 @@
   american fuzzy lop++ - wrapper for llvm 11+ lld
   -----------------------------------------------
 
-  Written by Marc Heuse <mh@mh-sec.de> for afl++
+  Written by Marc Heuse <mh@mh-sec.de> for AFL++
 
   Maintained by Marc Heuse <mh@mh-sec.de>,
                 Heiko Eißfeldt <heiko.eissfeldt@hexco.de>
@@ -210,7 +210,7 @@ static void edit_params(int argc, char **argv) {
 
     if (strcmp(argv[i], "--afl") == 0) {
 
-      if (!be_quiet) OKF("afl++ test command line flag detected, exiting.");
+      if (!be_quiet) OKF("AFL++ test command line flag detected, exiting.");
       exit(0);
 
     }
diff --git a/test/test-dlopen.c b/test/test-dlopen.c
index b81bab13..39442f93 100644
--- a/test/test-dlopen.c
+++ b/test/test-dlopen.c
@@ -28,7 +28,7 @@ int main(int argc, char **argv) {
 
   }
 
-  // must use deferred forkserver as otherwise afl++ instrumentation aborts
+  // must use deferred forkserver as otherwise AFL++ instrumentation aborts
   // because all dlopen() of instrumented libs must be before the forkserver
   __AFL_INIT();
 
diff --git a/test/test-gcc-plugin.sh b/test/test-gcc-plugin.sh
index 54e6987f..3690a80a 100755
--- a/test/test-gcc-plugin.sh
+++ b/test/test-gcc-plugin.sh
@@ -23,7 +23,7 @@ test -e ../afl-gcc-fast -a -e ../afl-compiler-rt.o && {
           $ECHO "$GREEN[+] gcc_plugin run reported $TUPLES instrumented locations which is fine"
         } || {
           $ECHO "$RED[!] gcc_plugin instrumentation produces a weird numbers: $TUPLES"
-          $ECHO "$YELLOW[-] this is a known issue in gcc, not afl++. It is not flagged as an error because travis builds would all fail otherwise :-("
+          $ECHO "$YELLOW[-] this is a known issue in gcc, not AFL++. It is not flagged as an error because travis builds would all fail otherwise :-("
           #CODE=1
         }
         test "$TUPLES" -lt 2 && SKIP=1
diff --git a/test/test-performance.sh b/test/test-performance.sh
index d61e2f2a..50957141 100755
--- a/test/test-performance.sh
+++ b/test/test-performance.sh
@@ -7,7 +7,7 @@ FILE=$AFL_PERFORMANCE_FILE
 test -z "$FILE" && FILE=.afl_performance
 
 test -e $FILE || {
-  echo Warning: This script measure the performance of afl++ and saves the result for future comparisons into $FILE
+  echo Warning: This script measure the performance of AFL++ and saves the result for future comparisons into $FILE
   echo Press ENTER to continue or CONTROL-C to abort
   read IN
 }
@@ -74,7 +74,7 @@ afl-system-config > /dev/null 2>&1
 echo Performance settings applied.
 echo
 
-$ECHO "${RESET}${GREY}[*] starting afl++ performance test framework ..."
+$ECHO "${RESET}${GREY}[*] starting AFL++ performance test framework ..."
 
 $ECHO "$BLUE[*] Testing: ${AFL_GCC}"
 GCC=x
diff --git a/test/test-pre.sh b/test/test-pre.sh
index b8b286e5..1ca9dfb5 100755
--- a/test/test-pre.sh
+++ b/test/test-pre.sh
@@ -133,7 +133,7 @@ MEM_LIMIT=none
 
 export PATH="${PATH}:/sbin:/bin:/usr/sbin:/usr/bin:/usr/local/sbin:/usr/local/bin"
 
-$ECHO "${RESET}${GREY}[*] starting afl++ test framework ..."
+$ECHO "${RESET}${GREY}[*] starting AFL++ test framework ..."
 
 test -z "$SYS" && $ECHO "$YELLOW[-] uname -m did not succeed"
 
diff --git a/unicorn_mode/build_unicorn_support.sh b/unicorn_mode/build_unicorn_support.sh
index 53ec2481..d3d16ad5 100755
--- a/unicorn_mode/build_unicorn_support.sh
+++ b/unicorn_mode/build_unicorn_support.sh
@@ -182,7 +182,7 @@ git pull
 sh -c 'git stash && git stash drop' 1>/dev/null 2>/dev/null
 git checkout "$UNICORNAFL_VERSION" || exit 1
 
-echo "[*] making sure afl++ header files match"
+echo "[*] making sure AFL++ header files match"
 cp "../../include/config.h" "./include" || exit 1
 
 echo "[*] Configuring Unicorn build..."
-- 
cgit 1.4.1


From a2daef29f9c323c0a6a7a64013aadb79ffd3e534 Mon Sep 17 00:00:00 2001
From: van Hauser <vh@thc.org>
Date: Thu, 27 Apr 2023 17:57:22 +0200
Subject: slightly different weighting algo (#1719)

* better seed selection

* slightly different weighting calculation

* remove unnecessary memset
---
 include/afl-fuzz.h   |  4 +--
 src/afl-fuzz-queue.c | 92 +++++++++++++++++++++++++++++++++++-----------------
 2 files changed, 65 insertions(+), 31 deletions(-)

(limited to 'include')

diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index 831a0dbc..8fb7ecb1 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -1223,7 +1223,7 @@ double rand_next_percent(afl_state_t *afl);
 
 static inline u32 rand_below(afl_state_t *afl, u32 limit) {
 
-  if (limit <= 1) return 0;
+  if (unlikely(limit <= 1)) return 0;
 
   /* The boundary not being necessarily a power of 2,
      we need to ensure the result uniformity. */
@@ -1256,7 +1256,7 @@ static inline u32 rand_below(afl_state_t *afl, u32 limit) {
    expand havoc mode */
 static inline u32 rand_below_datalen(afl_state_t *afl, u32 limit) {
 
-  if (limit <= 1) return 0;
+  if (unlikely(limit <= 1)) return 0;
 
   switch (rand_below(afl, 3)) {
 
diff --git a/src/afl-fuzz-queue.c b/src/afl-fuzz-queue.c
index 8ad7cd97..b10bf749 100644
--- a/src/afl-fuzz-queue.c
+++ b/src/afl-fuzz-queue.c
@@ -49,11 +49,13 @@ inline u32 select_next_queue_entry(afl_state_t *afl) {
 
   u32    s = rand_below(afl, afl->queued_items);
   double p = rand_next_percent(afl);
+
   /*
   fprintf(stderr, "select: p=%f s=%u ... p < prob[s]=%f ? s=%u : alias[%u]=%u"
   " ==> %u\n", p, s, afl->alias_probability[s], s, s, afl->alias_table[s], p <
   afl->alias_probability[s] ? s : afl->alias_table[s]);
   */
+
   return (p < afl->alias_probability[s] ? s : afl->alias_table[s]);
 
 }
@@ -87,25 +89,28 @@ double compute_weight(afl_state_t *afl, struct queue_entry *q,
 
 void create_alias_table(afl_state_t *afl) {
 
-  u32    n = afl->queued_items, i = 0, a, g;
+  u32    n = afl->queued_items, i = 0, nSmall = 0, nLarge = n - 1;
   double sum = 0;
 
+  double *P = (double *)afl_realloc(AFL_BUF_PARAM(out), n * sizeof(double));
+  u32 *Small = (int *)afl_realloc(AFL_BUF_PARAM(out_scratch), n * sizeof(u32));
+  u32 *Large = (int *)afl_realloc(AFL_BUF_PARAM(in_scratch), n * sizeof(u32));
+
   afl->alias_table =
       (u32 *)afl_realloc((void **)&afl->alias_table, n * sizeof(u32));
   afl->alias_probability = (double *)afl_realloc(
       (void **)&afl->alias_probability, n * sizeof(double));
-  double *P = (double *)afl_realloc(AFL_BUF_PARAM(out), n * sizeof(double));
-  int    *S = (int *)afl_realloc(AFL_BUF_PARAM(out_scratch), n * sizeof(u32));
-  int    *L = (int *)afl_realloc(AFL_BUF_PARAM(in_scratch), n * sizeof(u32));
 
-  if (!P || !S || !L || !afl->alias_table || !afl->alias_probability) {
+  if (!P || !Small || !Large || !afl->alias_table || !afl->alias_probability) {
 
     FATAL("could not acquire memory for alias table");
 
   }
 
-  memset((void *)afl->alias_table, 0, n * sizeof(u32));
   memset((void *)afl->alias_probability, 0, n * sizeof(double));
+  memset((void *)afl->alias_table, 0, n * sizeof(u32));
+  memset((void *)Small, 0, n * sizeof(u32));
+  memset((void *)Large, 0, n * sizeof(u32));
 
   if (likely(afl->schedule < RARE)) {
 
@@ -166,7 +171,15 @@ void create_alias_table(afl_state_t *afl) {
     for (i = 0; i < n; i++) {
 
       // weight is always 0 for disabled entries
-      P[i] = (afl->queue_buf[i]->weight * n) / sum;
+      if (unlikely(afl->queue_buf[i]->disabled)) {
+
+        P[i] = 0;
+
+      } else {
+
+        P[i] = (afl->queue_buf[i]->weight * n) / sum;
+
+      }
 
     }
 
@@ -176,60 +189,81 @@ void create_alias_table(afl_state_t *afl) {
 
       struct queue_entry *q = afl->queue_buf[i];
 
-      if (likely(!q->disabled)) { q->perf_score = calculate_score(afl, q); }
+      if (likely(!q->disabled)) {
+
+        q->perf_score = calculate_score(afl, q);
+        sum += q->perf_score;
 
-      sum += q->perf_score;
+      }
 
     }
 
     for (i = 0; i < n; i++) {
 
       // perf_score is always 0 for disabled entries
-      P[i] = (afl->queue_buf[i]->perf_score * n) / sum;
+      if (unlikely(afl->queue_buf[i]->disabled)) {
+
+        P[i] = 0;
+
+      } else {
+
+        P[i] = (afl->queue_buf[i]->perf_score * n) / sum;
+
+      }
 
     }
 
   }
 
-  int nS = 0, nL = 0, s;
-  for (s = (s32)n - 1; s >= 0; --s) {
+  // Done collecting weightings in P, now create the arrays.
+
+  for (s32 j = (s32)(n - 1); j >= 0; j--) {
 
-    if (P[s] < 1) {
+    if (P[j] < 1) {
 
-      S[nS++] = s;
+      Small[nSmall++] = (u32)j;
 
     } else {
 
-      L[nL++] = s;
+      Large[nLarge--] = (u32)j;
 
     }
 
   }
 
-  while (nS && nL) {
+  while (nSmall && nLarge != n - 1) {
+
+    u32 small = Small[--nSmall];
+    u32 large = Large[++nLarge];
+
+    afl->alias_probability[small] = P[small];
+    afl->alias_table[small] = large;
 
-    a = S[--nS];
-    g = L[--nL];
-    afl->alias_probability[a] = P[a];
-    afl->alias_table[a] = g;
-    P[g] = P[g] + P[a] - 1;
-    if (P[g] < 1) {
+    P[large] = P[large] - (1 - P[small]);
 
-      S[nS++] = g;
+    if (P[large] < 1) {
+
+      Small[nSmall++] = large;
 
     } else {
 
-      L[nL++] = g;
+      Large[nLarge--] = large;
 
     }
 
   }
 
-  while (nL)
-    afl->alias_probability[L[--nL]] = 1;
+  while (nSmall) {
+
+    afl->alias_probability[Small[--nSmall]] = 1;
+
+  }
 
-  while (nS)
-    afl->alias_probability[S[--nS]] = 1;
+  while (nLarge != n - 1) {
+
+    afl->alias_probability[Large[++nLarge]] = 1;
+
+  }
 
   afl->reinit_table = 0;
 
@@ -264,7 +298,7 @@ void create_alias_table(afl_state_t *afl) {
   */
   /*
   fprintf(stderr, "  entry  alias  probability  perf_score   weight
-  filename\n"); for (u32 i = 0; i < n; ++i) fprintf(stderr, "  %5u  %5u  %11u
+  filename\n"); for (i = 0; i < n; ++i) fprintf(stderr, "  %5u  %5u  %11u
   %0.9f  %0.9f  %s\n", i, afl->alias_table[i], afl->alias_probability[i],
   afl->queue_buf[i]->perf_score, afl->queue_buf[i]->weight,
             afl->queue_buf[i]->fname);
-- 
cgit 1.4.1


From 22db79aefafb48fed48199a86a39babdee795870 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Mon, 1 May 2023 15:07:49 +0200
Subject: fix reallocs

---
 include/alloc-inl.h |  7 +++----
 src/afl-fuzz.c      | 39 +++++++++++++++++++++++++++++++++++++++
 2 files changed, 42 insertions(+), 4 deletions(-)

(limited to 'include')

diff --git a/include/alloc-inl.h b/include/alloc-inl.h
index bbb42e88..1e9a192b 100644
--- a/include/alloc-inl.h
+++ b/include/alloc-inl.h
@@ -704,12 +704,11 @@ static inline void *afl_realloc(void **buf, size_t size_needed) {
     *buf = NULL;
     return NULL;
 
-  } else {
-
-    new_buf = newer_buf;
-
   }
 
+  new_buf = newer_buf;
+  memset(((u8 *)new_buf) + current_size, 0, next_size - current_size);
+
   new_buf->complete_size = next_size;
   *buf = (void *)(new_buf->buf);
   return *buf;
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index 646dc50b..c02479cf 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -1979,6 +1979,7 @@ int main(int argc, char **argv_orig, char **envp) {
   if (afl->non_instrumented_mode || afl->fsrv.qemu_mode ||
       afl->fsrv.frida_mode || afl->fsrv.cs_mode || afl->unicorn_mode) {
 
+    u32 old_map_size = map_size;
     map_size = afl->fsrv.real_map_size = afl->fsrv.map_size = MAP_SIZE;
     afl->virgin_bits = ck_realloc(afl->virgin_bits, map_size);
     afl->virgin_tmout = ck_realloc(afl->virgin_tmout, map_size);
@@ -1990,6 +1991,18 @@ int main(int argc, char **argv_orig, char **envp) {
     afl->first_trace = ck_realloc(afl->first_trace, map_size);
     afl->map_tmp_buf = ck_realloc(afl->map_tmp_buf, map_size);
 
+    if (old_map_size < map_size) {
+
+      memset(afl->var_bytes + old_map_size, 0, map_size - old_map_size);
+      memset(afl->top_rated + old_map_size, 0, map_size - old_map_size);
+      memset(afl->clean_trace + old_map_size, 0, map_size - old_map_size);
+      memset(afl->clean_trace_custom + old_map_size, 0,
+             map_size - old_map_size);
+      memset(afl->first_trace + old_map_size, 0, map_size - old_map_size);
+      memset(afl->map_tmp_buf + old_map_size, 0, map_size - old_map_size);
+
+    }
+
   }
 
   afl->argv = use_argv;
@@ -2017,6 +2030,7 @@ int main(int argc, char **argv_orig, char **envp) {
 
       OKF("Re-initializing maps to %u bytes", new_map_size);
 
+      u32 old_map_size = map_size;
       afl->virgin_bits = ck_realloc(afl->virgin_bits, new_map_size);
       afl->virgin_tmout = ck_realloc(afl->virgin_tmout, new_map_size);
       afl->virgin_crash = ck_realloc(afl->virgin_crash, new_map_size);
@@ -2029,6 +2043,18 @@ int main(int argc, char **argv_orig, char **envp) {
       afl->first_trace = ck_realloc(afl->first_trace, new_map_size);
       afl->map_tmp_buf = ck_realloc(afl->map_tmp_buf, new_map_size);
 
+      if (old_map_size < new_map_size) {
+
+        memset(afl->var_bytes + old_map_size, 0, new_map_size - old_map_size);
+        memset(afl->top_rated + old_map_size, 0, new_map_size - old_map_size);
+        memset(afl->clean_trace + old_map_size, 0, new_map_size - old_map_size);
+        memset(afl->clean_trace_custom + old_map_size, 0,
+               new_map_size - old_map_size);
+        memset(afl->first_trace + old_map_size, 0, new_map_size - old_map_size);
+        memset(afl->map_tmp_buf + old_map_size, 0, new_map_size - old_map_size);
+
+      }
+
       afl_fsrv_kill(&afl->fsrv);
       afl_shm_deinit(&afl->shm);
       afl->fsrv.map_size = new_map_size;
@@ -2079,6 +2105,7 @@ int main(int argc, char **argv_orig, char **envp) {
 
       OKF("Re-initializing maps to %u bytes due cmplog", new_map_size);
 
+      u32 old_map_size = map_size;
       afl->virgin_bits = ck_realloc(afl->virgin_bits, new_map_size);
       afl->virgin_tmout = ck_realloc(afl->virgin_tmout, new_map_size);
       afl->virgin_crash = ck_realloc(afl->virgin_crash, new_map_size);
@@ -2091,6 +2118,18 @@ int main(int argc, char **argv_orig, char **envp) {
       afl->first_trace = ck_realloc(afl->first_trace, new_map_size);
       afl->map_tmp_buf = ck_realloc(afl->map_tmp_buf, new_map_size);
 
+      if (old_map_size < new_map_size) {
+
+        memset(afl->var_bytes + old_map_size, 0, new_map_size - old_map_size);
+        memset(afl->top_rated + old_map_size, 0, new_map_size - old_map_size);
+        memset(afl->clean_trace + old_map_size, 0, new_map_size - old_map_size);
+        memset(afl->clean_trace_custom + old_map_size, 0,
+               new_map_size - old_map_size);
+        memset(afl->first_trace + old_map_size, 0, new_map_size - old_map_size);
+        memset(afl->map_tmp_buf + old_map_size, 0, new_map_size - old_map_size);
+
+      }
+
       afl_fsrv_kill(&afl->fsrv);
       afl_fsrv_kill(&afl->cmplog_fsrv);
       afl_shm_deinit(&afl->shm);
-- 
cgit 1.4.1


From 70da0c2e405102dc044cb4bed0f4f1e847c90d0b Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Wed, 10 May 2023 16:09:18 +0200
Subject: better tritondse support

---
 custom_mutators/aflpp_tritondse/aflpp_tritondse.py | 54 ++++++++++---
 docs/custom_mutators.md                            | 28 +++++++
 include/envs.h                                     |  4 +
 src/afl-fuzz.c                                     | 91 ++++++++++++++++------
 4 files changed, 145 insertions(+), 32 deletions(-)

(limited to 'include')

diff --git a/custom_mutators/aflpp_tritondse/aflpp_tritondse.py b/custom_mutators/aflpp_tritondse/aflpp_tritondse.py
index 49f67d75..9584b368 100644
--- a/custom_mutators/aflpp_tritondse/aflpp_tritondse.py
+++ b/custom_mutators/aflpp_tritondse/aflpp_tritondse.py
@@ -7,6 +7,7 @@ from tritondse import Config
 from tritondse import CoverageStrategy
 from tritondse import ProcessState
 from tritondse import Program
+from tritondse import CleLoader
 from tritondse import Seed
 from tritondse import SeedFormat
 from tritondse import SymbolicExecutor
@@ -16,7 +17,7 @@ from tritondse import SymbolicExplorator
 #logging.basicConfig(level=logging.INFO)
 
 is_debug = False
-out_path = "out/tritondse/queue"
+out_path = ""
 input_file = None
 prog = None
 config = None
@@ -29,28 +30,38 @@ def pre_exec_hook(se: SymbolicExecutor, state: ProcessState):
     #logging.info(f"[PRE-EXEC] Processing seed: {se.seed.hash}, \
     #                ({repr(se.seed.content)})")
     global count
-    global hasshes
+    global hashes
+    print('DEBUG - prehook')
     if se.seed.hash not in hashes:
         hashes.add(se.seed.hash)
         filename = out_path + "/id:" + f"{count:06}" + "," + se.seed.hash
         if not os.path.exists(filename):
+            if is_debug:
+                print('Creating queue input ' + filename)
             with open(filename, 'wb') as file:
                 file.write(se.seed.content)
                 count += 1
+    else:
+        print('has hash: ' + se.seed.hash)
     if input_file:
+        if is_debug:
+            print('Writing to ' + input_file + ' the content: ' + str(se.seed.content))
         with open(input_file, 'wb') as file:
             file.write(se.seed.content)
+    else:
+        print('no input!')
 
 
 def init(seed):
     global prog
     global config
     global dse
+    global out_path
     global input_file
     global is_debug
     # Load the program (LIEF-based program loader).
-    prog = Program(os.environ['TRITON_DSE_TARGET'])
-    # Set the configuration.
+    prog = CleLoader(os.environ['AFL_CUSTOM_INFO_PROGRAM'])
+    # Process other configuration environment variables.
     argv = None
     try:
         foo = os.environ['AFL_DEBUG']
@@ -58,15 +69,42 @@ def init(seed):
     except KeyError:
         pass
     try:
-        argv_list = os.environ['TRITON_DSE_TARGET_ARGV']
-        argv = argv_list.split()
+        foo = os.environ['AFL_CUSTOM_INFO_OUT']
+        out_path = foo + '/../tritondse/queue'
     except KeyError:
         pass
     try:
-        foo = os.environ['TRITON_DSE_TARGET_INPUT']
+        foo = os.environ['AFL_CUSTOM_INFO_PROGRAM_INPUT']
         input_file = foo
     except KeyError:
         pass
+    try:
+        argv_list = os.environ['AFL_CUSTOM_INFO_PROGRAM_ARGV']
+        argv_tmp = [ os.environ['AFL_CUSTOM_INFO_PROGRAM'] ]
+        argv_tmp += argv_list.split()
+        argv = []
+        # now check for @@
+        for item in argv_tmp:
+            if "@@" in item:
+                input_file = out_path + '/../.input'
+                argv.append(input_file)
+            else:
+                argv.append(item)
+    except KeyError:
+        pass
+    # Create the output directory
+    os.makedirs(out_path, exist_ok=True)
+    # Debug
+    if is_debug:
+        print('DEBUG target: ' + os.environ['AFL_CUSTOM_INFO_PROGRAM'])
+        if argv:
+            print('DEBUG argv: ')
+            print(argv)
+        if input_file:
+            print('DEBUG input_file: ' + input_file)
+        print('DEBUG out_path: ' + out_path)
+        print('')
+    # Now set up TritonDSE
     config = Config(coverage_strategy = CoverageStrategy.PATH,
                     debug = is_debug,
                     pipe_stdout = is_debug,
@@ -79,8 +117,6 @@ def init(seed):
     dse = SymbolicExplorator(config, prog)
     # Add callbacks.
     dse.callback_manager.register_pre_execution_callback(pre_exec_hook)
-    # Create the output directory
-    os.makedirs(out_path, exist_ok=True)
 
 
 #def fuzz(buf, add_buf, max_size):
diff --git a/docs/custom_mutators.md b/docs/custom_mutators.md
index a1de479e..3f7e9e6e 100644
--- a/docs/custom_mutators.md
+++ b/docs/custom_mutators.md
@@ -304,6 +304,34 @@ Note: for some distributions, you might also need the package `python[3]-apt`.
 In case your setup is different, set the necessary variables like this:
 `PYTHON_INCLUDE=/path/to/python/include LDFLAGS=-L/path/to/python/lib make`.
 
+### Helpers
+
+For C/C++ custom mutators you get a pointer to `afl_state_t *afl` in the
+`afl_custom_init()` which contains all information that you need.
+Note that if you access it, you need to recompile your custom mutator if
+you update AFL++ because the structure might have changed!
+
+For mutators written in Python, Rust, GO, etc. there are a few environment
+variables set to help you to get started:
+
+`AFL_CUSTOM_INFO_PROGRAM` - the program name of the target that is executed.
+If your custom mutator is used with modes like Qemu (`-Q`), this will still
+contain the target program, not afl-qemu-trace.
+
+`AFL_CUSTOM_INFO_PROGRAM_INPUT` - if the `-f` parameter is used with afl-fuzz
+then this value is found in this environment variable.
+
+`AFL_CUSTOM_INFO_PROGRAM_ARGV` - this contains the parameters given to the
+target program and still has the `@@` identifier in there.
+
+Note: If `AFL_CUSTOM_INFO_PROGRAM_INPUT` is empty and `AFL_CUSTOM_INFO_PROGRAM_ARGV`
+is either empty or does not contain `@@` then the target gets the input via
+`stdin`.
+
+`AFL_CUSTOM_INFO_OUT` - This is the output directory for this fuzzer instance,
+so if `afl-fuzz` was called with `-o out -S foobar`, then this will be set to
+`out/foobar`.
+
 ### Custom Mutator Preparation
 
 For C/C++ mutators, the source code must be compiled as a shared object:
diff --git a/include/envs.h b/include/envs.h
index fe5ee0e3..edfd06e4 100644
--- a/include/envs.h
+++ b/include/envs.h
@@ -37,6 +37,10 @@ static char *afl_environment_variables[] = {
     "AFL_CRASH_EXITCODE",
     "AFL_CUSTOM_MUTATOR_LIBRARY",
     "AFL_CUSTOM_MUTATOR_ONLY",
+    "AFL_CUSTOM_INFO_PROGRAM",
+    "AFL_CUSTOM_INFO_PROGRAM_ARGV",
+    "AFL_CUSTOM_INFO_PROGRAM_INPUT",
+    "AFL_CUSTOM_INFO_OUT",
     "AFL_CXX",
     "AFL_CYCLE_SCHEDULES",
     "AFL_DEBUG",
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index f982258f..4339ddd2 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -1530,29 +1530,6 @@ int main(int argc, char **argv_orig, char **envp) {
 
   }
 
-  if (afl->limit_time_sig > 0 && afl->custom_mutators_count) {
-
-    if (afl->custom_only) {
-
-      FATAL("Custom mutators are incompatible with MOpt (-L)");
-
-    }
-
-    u32 custom_fuzz = 0;
-    LIST_FOREACH(&afl->custom_mutator_list, struct custom_mutator, {
-
-      if (el->afl_custom_fuzz) { custom_fuzz = 1; }
-
-    });
-
-    if (custom_fuzz) {
-
-      WARNF("afl_custom_fuzz is incompatible with MOpt (-L)");
-
-    }
-
-  }
-
   if (afl->afl_env.afl_max_det_extras) {
 
     s32 max_det_extras = atoi(afl->afl_env.afl_max_det_extras);
@@ -1827,8 +1804,76 @@ int main(int argc, char **argv_orig, char **envp) {
     printf("DEBUG: rand %06d is %u\n", counter, rand_below(afl, 65536));
   #endif
 
+  if (!getenv("AFL_CUSTOM_INFO_PROGRAM")) {
+
+    setenv("AFL_CUSTOM_INFO_PROGRAM", argv[optind], 1);
+
+  }
+
+  if (!getenv("AFL_CUSTOM_INFO_PROGRAM_INPUT") && afl->fsrv.out_file) {
+
+    setenv("AFL_CUSTOM_INFO_PROGRAM_INPUT", afl->fsrv.out_file, 1);
+
+  }
+
+  {
+
+    u8 envbuf[8096] = "", tmpbuf[8096] = "";
+    for (s32 i = optind + 1; i < argc; ++i) {
+
+      strcpy(tmpbuf, envbuf);
+      if (strchr(argv[i], ' ') && !strchr(argv[i], '"') &&
+          !strchr(argv[i], '\'')) {
+
+        if (!strchr(argv[i], '\'')) {
+
+          snprintf(envbuf, sizeof(tmpbuf), "%s '%s'", tmpbuf, argv[i]);
+
+        } else {
+
+          snprintf(envbuf, sizeof(tmpbuf), "%s \"%s\"", tmpbuf, argv[i]);
+
+        }
+
+      } else {
+
+        snprintf(envbuf, sizeof(tmpbuf), "%s %s", tmpbuf, argv[i]);
+
+      }
+
+    }
+
+    setenv("AFL_CUSTOM_INFO_PROGRAM_ARGV", envbuf + 1, 1);
+
+  }
+
+  setenv("AFL_CUSTOM_INFO_OUT", afl->out_dir, 1);  // same as __AFL_OUT_DIR
+
   setup_custom_mutators(afl);
 
+  if (afl->limit_time_sig > 0 && afl->custom_mutators_count) {
+
+    if (afl->custom_only) {
+
+      FATAL("Custom mutators are incompatible with MOpt (-L)");
+
+    }
+
+    u32 custom_fuzz = 0;
+    LIST_FOREACH(&afl->custom_mutator_list, struct custom_mutator, {
+
+      if (el->afl_custom_fuzz) { custom_fuzz = 1; }
+
+    });
+
+    if (custom_fuzz) {
+
+      WARNF("afl_custom_fuzz is incompatible with MOpt (-L)");
+
+    }
+
+  }
+
   write_setup_file(afl, argc, argv);
 
   setup_cmdline_file(afl, argv + optind);
-- 
cgit 1.4.1


From abd6eace9d767e4db6019e8eb69080d2352015c9 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Thu, 18 May 2023 10:32:15 +0200
Subject: improved symqemu custom mutator

---
 custom_mutators/symqemu/README.md |   2 +-
 custom_mutators/symqemu/symqemu.c | 239 +++++++++++++++-----------------------
 include/afl-fuzz.h                |   1 +
 src/afl-fuzz-one.c                |   1 +
 4 files changed, 98 insertions(+), 145 deletions(-)

(limited to 'include')

diff --git a/custom_mutators/symqemu/README.md b/custom_mutators/symqemu/README.md
index 55ce05c5..b7702c06 100644
--- a/custom_mutators/symqemu/README.md
+++ b/custom_mutators/symqemu/README.md
@@ -8,4 +8,4 @@ on how to build symqemu-x86_x64 and put it in your `PATH`.
 
 just type `make` to build this custom mutator.
 
-```AFL_CUSTOM_MUTATOR_LIBRARY=custom_mutators/symqemu/symqemu-mutator.so AFL_SYNC_TIME=1 AFL_DISABLE_TRIM=1 afl-fuzz ...```
+```AFL_CUSTOM_MUTATOR_LIBRARY=custom_mutators/symqemu/symqemu-mutator.so AFL_DISABLE_TRIM=1 afl-fuzz ...```
diff --git a/custom_mutators/symqemu/symqemu.c b/custom_mutators/symqemu/symqemu.c
index 9030397b..163ae240 100644
--- a/custom_mutators/symqemu/symqemu.c
+++ b/custom_mutators/symqemu/symqemu.c
@@ -13,6 +13,9 @@
 
 afl_state_t *afl_struct;
 static u32   debug = 0;
+static u32   found_items = 0;
+
+#define SYMQEMU_LOCATION "symqemu"
 
 #define DBG(x...) \
   if (debug) { fprintf(stderr, x); }
@@ -22,7 +25,6 @@ typedef struct my_mutator {
   afl_state_t *afl;
   u8          *mutator_buf;
   u8          *out_dir;
-  u8          *queue_dir;
   u8          *target;
   u8          *symqemu;
   u8          *input_file;
@@ -67,8 +69,13 @@ my_mutator_t *afl_custom_init(afl_state_t *afl, unsigned int seed) {
   if (!data->symqemu) FATAL("symqemu binary %s not found", exec_name);
   DBG("Found %s\n", data->symqemu);
 
-  if (getenv("AFL_CUSTOM_MUTATOR_ONLY"))
-    FATAL("the symqemu module cannot be used with AFL_CUSTOM_MUTATOR_ONLY.");
+  if (getenv("AFL_CUSTOM_MUTATOR_ONLY")) {
+
+    WARNF(
+        "the symqemu module is not very effective with "
+        "AFL_CUSTOM_MUTATOR_ONLY.");
+
+  }
 
   if ((data->mutator_buf = malloc(MAX_FILE)) == NULL) {
 
@@ -84,14 +91,11 @@ my_mutator_t *afl_custom_init(afl_state_t *afl, unsigned int seed) {
   u32 len = strlen(path_tmp) + 32;
   u8 *symqemu_path = malloc(len);
   data->out_dir = malloc(len);
-  data->queue_dir = malloc(len);
-  snprintf(symqemu_path, len, "%s/../symqemu", path_tmp);
-  snprintf(data->out_dir, len, "%s/../symqemu/out", path_tmp);
-  snprintf(data->queue_dir, len, "%s/../symqemu/queue", path_tmp);
+  snprintf(symqemu_path, len, "%s/%s", path_tmp, SYMQEMU_LOCATION);
+  snprintf(data->out_dir, len, "%s/out", symqemu_path, path_tmp);
 
-  mkdir(symqemu_path, 0755);
-  mkdir(data->out_dir, 0755);
-  mkdir(data->queue_dir, 0755);
+  (void)mkdir(symqemu_path, 0755);
+  (void)mkdir(data->out_dir, 0755);
 
   setenv("SYMCC_OUTPUT_DIR", data->out_dir, 1);
 
@@ -153,8 +157,8 @@ my_mutator_t *afl_custom_init(afl_state_t *afl, unsigned int seed) {
   data->argv[0] = data->symqemu;
   data->argv[1] = data->target;
 
-  DBG("out_dir=%s, queue_dir=%s, target=%s, input_file=%s, argc=%u\n",
-      data->out_dir, data->queue_dir, data->target,
+  DBG("out_dir=%s, target=%s, input_file=%s, argc=%u\n", data->out_dir,
+      data->target,
       data->input_file ? (char *)data->input_file : (char *)"<stdin>",
       data->argc);
 
@@ -174,29 +178,39 @@ my_mutator_t *afl_custom_init(afl_state_t *afl, unsigned int seed) {
 
   }
 
-  OKF("Custom mutator symqemu loaded - note that the initial startup of "
-      "afl-fuzz will be delayed the more starting seeds are present. This is "
-      "fine, do not worry!");
-
   return data;
 
 }
 
-/* When a new queue entry is added we run this input with the symqemu
-   instrumented binary */
-uint8_t afl_custom_queue_new_entry(my_mutator_t  *data,
-                                   const uint8_t *filename_new_queue,
-                                   const uint8_t *filename_orig_queue) {
+/* No need to receive a splicing item */
+void afl_custom_splice_optout(void *data) {
+
+  (void)(data);
+
+}
+
+u32 afl_custom_fuzz_count(my_mutator_t *data, const u8 *buf, size_t buf_size) {
+
+  if (likely(!afl_struct->queue_cur->favored ||
+             afl_struct->queue_cur->was_fuzzed)) {
+
+    return 0;
+
+  }
 
   int         pipefd[2];
   struct stat st;
-  if (data->afl->afl_env.afl_no_ui)
-    ACTF("Sending to symqemu: %s", filename_new_queue);
-  u8 *fn = alloc_printf("%s", filename_new_queue);
-  if (!(stat(fn, &st) == 0 && S_ISREG(st.st_mode) && st.st_size)) {
 
-    ck_free(fn);
-    PFATAL("Couldn't find enqueued file: %s", fn);
+  if (afl_struct->afl_env.afl_no_ui) {
+
+    ACTF("Sending to symqemu: %s", afl_struct->queue_cur->fname);
+
+  }
+
+  if (!(stat(afl_struct->queue_cur->fname, &st) == 0 && S_ISREG(st.st_mode) &&
+        st.st_size)) {
+
+    PFATAL("Couldn't find enqueued file: %s", afl_struct->queue_cur->fname);
 
   }
 
@@ -204,7 +218,6 @@ uint8_t afl_custom_queue_new_entry(my_mutator_t  *data,
 
     if (pipe(pipefd) == -1) {
 
-      ck_free(fn);
       PFATAL(
           "Couldn't create a pipe for interacting with symqemu child process");
 
@@ -212,19 +225,12 @@ uint8_t afl_custom_queue_new_entry(my_mutator_t  *data,
 
   }
 
-  int fd = open(fn, O_RDONLY);
-  if (fd < 0) return 0;
-  ssize_t r = read(fd, data->mutator_buf, MAX_FILE);
-  DBG("fn=%s, fd=%d, size=%ld\n", fn, fd, r);
-  ck_free(fn);
-  close(fd);
-
   if (data->input_file) {
 
-    fd = open(data->input_file, O_WRONLY | O_CREAT | O_TRUNC, 0644);
-    ssize_t s = write(fd, data->mutator_buf, r);
+    int     fd = open(data->input_file, O_WRONLY | O_CREAT | O_TRUNC, 0644);
+    ssize_t s = write(fd, buf, buf_size);
     close(fd);
-    DBG("wrote %zd/%zd to %s\n", s, r, data->input_file);
+    DBG("wrote %zd/%zd to %s\n", s, buf_size, data->input_file);
 
   }
 
@@ -232,35 +238,20 @@ uint8_t afl_custom_queue_new_entry(my_mutator_t  *data,
 
   if (pid == -1) return 0;
 
-  if (pid) {
+  if (likely(pid)) {
 
     if (!data->input_file || afl_struct->fsrv.use_stdin) {
 
       close(pipefd[0]);
 
-      if (fd >= 0) {
-
-        if (r <= 0) {
-
-          close(pipefd[1]);
-          return 0;
-
-        }
+      if (fcntl(pipefd[1], F_GETPIPE_SZ)) {
 
-        if (r > fcntl(pipefd[1], F_GETPIPE_SZ))
-          fcntl(pipefd[1], F_SETPIPE_SZ, MAX_FILE);
-        ck_write(pipefd[1], data->mutator_buf, r, filename_new_queue);
-
-      } else {
-
-        ck_free(fn);
-        close(pipefd[1]);
-        PFATAL(
-            "Something happened to the enqueued file before sending its "
-            "contents to symqemu binary");
+        fcntl(pipefd[1], F_SETPIPE_SZ, MAX_FILE);
 
       }
 
+      ck_write(pipefd[1], buf, buf_size, data->input_file);
+
       close(pipefd[1]);
 
     }
@@ -268,46 +259,6 @@ uint8_t afl_custom_queue_new_entry(my_mutator_t  *data,
     pid = waitpid(pid, NULL, 0);
     DBG("symqemu finished executing!\n");
 
-    // At this point we need to transfer files to output dir, since their names
-    // collide and symqemu will just overwrite them
-
-    struct dirent **nl;
-    int32_t         items = scandir(data->out_dir, &nl, NULL, NULL);
-    u8             *origin_name = basename(filename_new_queue);
-    u8              source_name[4096], destination_name[4096];
-    int32_t         i;
-
-    if (items > 0) {
-
-      for (i = 0; i < (u32)items; ++i) {
-
-        // symqemu output files start with a digit
-        if (!isdigit(nl[i]->d_name[0])) continue;
-
-        struct stat st;
-        snprintf(source_name, sizeof(source_name), "%s/%s", data->out_dir,
-                 nl[i]->d_name);
-        DBG("file=%s\n", source_name);
-
-        if (stat(source_name, &st) == 0 && S_ISREG(st.st_mode) && st.st_size) {
-
-          snprintf(destination_name, sizeof(destination_name), "%s/id:%06u,%s",
-                   data->queue_dir, data->counter++, nl[i]->d_name);
-          DBG("src=%s dst=%s\n", source_name, destination_name);
-          rename(source_name, destination_name);
-
-        }
-
-        free(nl[i]);
-
-      }
-
-      free(nl);
-
-    }
-
-    DBG("Done!\n");
-
   } else /* (pid == 0) */ {  // child
 
     if (afl_struct->fsrv.use_stdin) {
@@ -338,33 +289,31 @@ uint8_t afl_custom_queue_new_entry(my_mutator_t  *data,
 
   }
 
-  return 0;
-
-}
-
-/*
-uint32_t afl_custom_fuzz_count(my_mutator_t *data, const u8 *buf,
-                               size_t buf_size) {
+  /* back in mother process */
 
-  uint32_t        count = 0, i;
   struct dirent **nl;
-  int32_t         items = scandir(data->out_dir, &nl, NULL, NULL);
+  s32             i, items = scandir(data->out_dir, &nl, NULL, NULL);
+  found_items = 0;
+  char source_name[4096];
 
   if (items > 0) {
 
     for (i = 0; i < (u32)items; ++i) {
 
+      // symqemu output files start with a digit
+      if (!isdigit(nl[i]->d_name[0])) continue;
+
       struct stat st;
-      u8 *        fn = alloc_printf("%s/%s", data->out_dir, nl[i]->d_name);
-      DBG("test=%s\n", fn);
-      if (stat(fn, &st) == 0 && S_ISREG(st.st_mode) && st.st_size) {
+      snprintf(source_name, sizeof(source_name), "%s/%s", data->out_dir,
+               nl[i]->d_name);
+      DBG("file=%s\n", source_name);
+
+      if (stat(source_name, &st) == 0 && S_ISREG(st.st_mode) && st.st_size) {
 
-        DBG("found=%s\n", fn);
-        count++;
+        ++found_items;
 
       }
 
-      ck_free(fn);
       free(nl[i]);
 
     }
@@ -373,65 +322,67 @@ uint32_t afl_custom_fuzz_count(my_mutator_t *data, const u8 *buf,
 
   }
 
-  DBG("dir=%s, count=%u\n", data->out_dir, count);
-  return count;
+  DBG("Done, found %u items!\n", found_items);
 
-}
+  return found_items;
 
-*/
+}
 
-// here we actually just read the files generated from symqemu
-/*
-size_t afl_custom_fuzz(my_mutator_t *data, uint8_t *buf, size_t buf_size,
-                       u8 **out_buf, uint8_t *add_buf, size_t add_buf_size,
+size_t afl_custom_fuzz(my_mutator_t *data, u8 *buf, size_t buf_size,
+                       u8 **out_buf, u8 *add_buf, size_t add_buf_size,
                        size_t max_size) {
 
   struct dirent **nl;
-  int32_t         i, done = 0, items = scandir(data->out_dir, &nl, NULL, NULL);
-  ssize_t         size = 0;
+  s32             done = 0, i, items = scandir(data->out_dir, &nl, NULL, NULL);
+  char            source_name[4096];
 
-  if (items <= 0) return 0;
+  if (items > 0) {
 
-  for (i = 0; i < (u32)items; ++i) {
+    for (i = 0; i < (u32)items; ++i) {
 
-    struct stat st;
-    u8 *        fn = alloc_printf("%s/%s", data->out_dir, nl[i]->d_name);
+      // symqemu output files start with a digit
+      if (!isdigit(nl[i]->d_name[0])) continue;
 
-    if (done == 0) {
+      struct stat st;
+      snprintf(source_name, sizeof(source_name), "%s/%s", data->out_dir,
+               nl[i]->d_name);
+      DBG("file=%s\n", source_name);
 
-      if (stat(fn, &st) == 0 && S_ISREG(st.st_mode) && st.st_size) {
+      if (stat(source_name, &st) == 0 && S_ISREG(st.st_mode) && st.st_size) {
 
-        int fd = open(fn, O_RDONLY);
+        int fd = open(source_name, O_RDONLY);
+        if (fd < 0) { goto got_an_issue; }
 
-        if (fd >= 0) {
+        ssize_t r = read(fd, data->mutator_buf, MAX_FILE);
+        close(fd);
 
-          size = read(fd, data->mutator_buf, max_size);
-          *out_buf = data->mutator_buf;
+        DBG("fn=%s, fd=%d, size=%ld\n", source_name, fd, r);
 
-          close(fd);
-          done = 1;
+        if (r < 1) { goto got_an_issue; }
 
-        }
+        done = 1;
+        --found_items;
+        unlink(source_name);
+
+        *out_buf = data->mutator_buf;
+        return (u32)r;
 
       }
 
-      unlink(fn);
+      free(nl[i]);
 
     }
 
-    ck_free(fn);
-    free(nl[i]);
+    free(nl);
 
   }
 
-  free(nl);
-  DBG("FUZZ size=%lu\n", size);
-  return (uint32_t)size;
+got_an_issue:
+  *out_buf = NULL;
+  return 0;
 
 }
 
-*/
-
 /**
  * Deinitialize everything
  *
diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index 8fb7ecb1..beb2de2a 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -184,6 +184,7 @@ struct queue_entry {
       handicap,                         /* Number of queue cycles behind    */
       depth,                            /* Path depth                       */
       exec_cksum,                       /* Checksum of the execution trace  */
+      custom,                           /* Marker for custom mutators       */
       stats_mutated;                    /* stats: # of mutations performed  */
 
   u8 *trace_mini;                       /* Trace bytes, if kept             */
diff --git a/src/afl-fuzz-one.c b/src/afl-fuzz-one.c
index c6e9a295..5c71fc59 100644
--- a/src/afl-fuzz-one.c
+++ b/src/afl-fuzz-one.c
@@ -1912,6 +1912,7 @@ custom_mutator_stage:
 
   afl->stage_name = "custom mutator";
   afl->stage_short = "custom";
+  afl->stage_cur = 0;
   afl->stage_val_type = STAGE_VAL_NONE;
   bool has_custom_fuzz = false;
   u32  shift = unlikely(afl->custom_only) ? 7 : 8;
-- 
cgit 1.4.1


From 029e039cbcbf9e30f35eb255162679b8d609d25d Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Sun, 21 May 2023 17:49:14 +0200
Subject: code format

---
 .custom-format.py                                 |  2 +-
 frida_mode/src/lib/lib_apple.c                    |  4 +--
 include/config.h                                  |  2 +-
 instrumentation/SanitizerCoverageLTO.so.cc        |  4 +--
 instrumentation/SanitizerCoveragePCGUARD.so.cc    |  4 +--
 instrumentation/afl-compiler-rt.o.c               | 14 +++++-----
 instrumentation/afl-llvm-lto-instrumentlist.so.cc |  2 +-
 instrumentation/cmplog-routines-pass.cc           |  4 +--
 instrumentation/compare-transform-pass.so.cc      |  2 +-
 instrumentation/split-compares-pass.so.cc         |  2 +-
 qemu_mode/libcompcov/libcompcov.so.c              |  6 ++++-
 src/afl-cc.c                                      |  8 +++---
 src/afl-fuzz-redqueen.c                           | 32 +++++++++++------------
 utils/afl_untracer/afl-untracer.c                 |  2 +-
 utils/libtokencap/libtokencap.so.c                |  6 ++++-
 utils/socket_fuzzing/socketfuzz.c                 |  3 ++-
 16 files changed, 53 insertions(+), 44 deletions(-)

(limited to 'include')

diff --git a/.custom-format.py b/.custom-format.py
index 1295ce55..1d5c8839 100755
--- a/.custom-format.py
+++ b/.custom-format.py
@@ -24,7 +24,7 @@ import importlib.metadata
 
 # string_re = re.compile('(\\"(\\\\.|[^"\\\\])*\\")') # TODO: for future use
 
-CURRENT_LLVM = os.getenv('LLVM_VERSION', 14)
+CURRENT_LLVM = os.getenv('LLVM_VERSION', 15)
 CLANG_FORMAT_BIN = os.getenv("CLANG_FORMAT_BIN", "")
 
 
diff --git a/frida_mode/src/lib/lib_apple.c b/frida_mode/src/lib/lib_apple.c
index 634e0e30..d29d0303 100644
--- a/frida_mode/src/lib/lib_apple.c
+++ b/frida_mode/src/lib/lib_apple.c
@@ -17,8 +17,8 @@ static gboolean lib_get_main_module(const GumModuleDetails *details,
 
   GumDarwinModule **ret = (GumDarwinModule **)user_data;
   GumDarwinModule  *module = gum_darwin_module_new_from_memory(
-       details->path, mach_task_self(), details->range->base_address,
-       GUM_DARWIN_MODULE_FLAGS_NONE, NULL);
+      details->path, mach_task_self(), details->range->base_address,
+      GUM_DARWIN_MODULE_FLAGS_NONE, NULL);
 
   FVERBOSE("Found main module: %s", module->name);
 
diff --git a/include/config.h b/include/config.h
index 764c29dc..194786f7 100644
--- a/include/config.h
+++ b/include/config.h
@@ -81,7 +81,7 @@
    will be kept and written to the crash/ directory as RECORD:... files.
    Note that every crash will be written, not only unique ones! */
 
-//#define AFL_PERSISTENT_RECORD
+// #define AFL_PERSISTENT_RECORD
 
 /* console output colors: There are three ways to configure its behavior
  * 1. default: colored outputs fixed on: defined USE_COLOR && defined
diff --git a/instrumentation/SanitizerCoverageLTO.so.cc b/instrumentation/SanitizerCoverageLTO.so.cc
index b3b0d2cd..d7b03634 100644
--- a/instrumentation/SanitizerCoverageLTO.so.cc
+++ b/instrumentation/SanitizerCoverageLTO.so.cc
@@ -1478,8 +1478,8 @@ GlobalVariable *ModuleSanitizerCoverageLTO::CreateFunctionLocalArrayInSection(
 
   ArrayType *ArrayTy = ArrayType::get(Ty, NumElements);
   auto       Array = new GlobalVariable(
-            *CurModule, ArrayTy, false, GlobalVariable::PrivateLinkage,
-            Constant::getNullValue(ArrayTy), "__sancov_gen_");
+      *CurModule, ArrayTy, false, GlobalVariable::PrivateLinkage,
+      Constant::getNullValue(ArrayTy), "__sancov_gen_");
 
 #if LLVM_VERSION_MAJOR >= 13
   if (TargetTriple.supportsCOMDAT() &&
diff --git a/instrumentation/SanitizerCoveragePCGUARD.so.cc b/instrumentation/SanitizerCoveragePCGUARD.so.cc
index 41c38283..8fed2042 100644
--- a/instrumentation/SanitizerCoveragePCGUARD.so.cc
+++ b/instrumentation/SanitizerCoveragePCGUARD.so.cc
@@ -714,8 +714,8 @@ GlobalVariable *ModuleSanitizerCoverageAFL::CreateFunctionLocalArrayInSection(
 
   ArrayType *ArrayTy = ArrayType::get(Ty, NumElements);
   auto       Array = new GlobalVariable(
-            *CurModule, ArrayTy, false, GlobalVariable::PrivateLinkage,
-            Constant::getNullValue(ArrayTy), "__sancov_gen_");
+      *CurModule, ArrayTy, false, GlobalVariable::PrivateLinkage,
+      Constant::getNullValue(ArrayTy), "__sancov_gen_");
 
 #if LLVM_VERSION_MAJOR >= 13
   if (TargetTriple.supportsCOMDAT() &&
diff --git a/instrumentation/afl-compiler-rt.o.c b/instrumentation/afl-compiler-rt.o.c
index 5372fae0..3f8b519b 100644
--- a/instrumentation/afl-compiler-rt.o.c
+++ b/instrumentation/afl-compiler-rt.o.c
@@ -544,12 +544,12 @@ static void __afl_map_shm(void) {
 
     if (__afl_map_size && __afl_map_size > MAP_SIZE) {
 
-             u8 *map_env = (u8 *)getenv("AFL_MAP_SIZE");
-             if (!map_env || atoi((char *)map_env) < MAP_SIZE) {
+      u8 *map_env = (u8 *)getenv("AFL_MAP_SIZE");
+      if (!map_env || atoi((char *)map_env) < MAP_SIZE) {
 
-               fprintf(stderr, "FS_ERROR_MAP_SIZE\n");
-               send_forkserver_error(FS_ERROR_MAP_SIZE);
-               _exit(1);
+        fprintf(stderr, "FS_ERROR_MAP_SIZE\n");
+        send_forkserver_error(FS_ERROR_MAP_SIZE);
+        _exit(1);
 
       }
 
@@ -561,13 +561,13 @@ static void __afl_map_shm(void) {
 
     if (!__afl_area_ptr || __afl_area_ptr == (void *)-1) {
 
-             if (__afl_map_addr)
+      if (__afl_map_addr)
         send_forkserver_error(FS_ERROR_MAP_ADDR);
       else
         send_forkserver_error(FS_ERROR_SHMAT);
 
       perror("shmat for map");
-             _exit(1);
+      _exit(1);
 
     }
 
diff --git a/instrumentation/afl-llvm-lto-instrumentlist.so.cc b/instrumentation/afl-llvm-lto-instrumentlist.so.cc
index db5bd55e..61f97d77 100644
--- a/instrumentation/afl-llvm-lto-instrumentlist.so.cc
+++ b/instrumentation/afl-llvm-lto-instrumentlist.so.cc
@@ -45,7 +45,7 @@
 #include "llvm/IR/Module.h"
 #include "llvm/Pass.h"
 #include "llvm/Support/Debug.h"
-//#include "llvm/Transforms/IPO/PassManagerBuilder.h"
+// #include "llvm/Transforms/IPO/PassManagerBuilder.h"
 #include "llvm/Passes/PassPlugin.h"
 #include "llvm/Passes/PassBuilder.h"
 #include "llvm/IR/PassManager.h"
diff --git a/instrumentation/cmplog-routines-pass.cc b/instrumentation/cmplog-routines-pass.cc
index 39db5aa4..c3fbed8d 100644
--- a/instrumentation/cmplog-routines-pass.cc
+++ b/instrumentation/cmplog-routines-pass.cc
@@ -542,7 +542,7 @@ bool CmpLogRoutines::hookRtns(Module &M) {
     Value               *v1Pcasted = IRB.CreatePointerCast(v1P, i8PtrTy);
     Value               *v2Pcasted = IRB.CreatePointerCast(v2P, i8PtrTy);
     Value               *v3Pbitcast = IRB.CreateBitCast(
-                      v3P, IntegerType::get(C, v3P->getType()->getPrimitiveSizeInBits()));
+        v3P, IntegerType::get(C, v3P->getType()->getPrimitiveSizeInBits()));
     Value *v3Pcasted =
         IRB.CreateIntCast(v3Pbitcast, IntegerType::get(C, 64), false);
     args.push_back(v1Pcasted);
@@ -608,7 +608,7 @@ bool CmpLogRoutines::hookRtns(Module &M) {
     Value               *v1Pcasted = IRB.CreatePointerCast(v1P, i8PtrTy);
     Value               *v2Pcasted = IRB.CreatePointerCast(v2P, i8PtrTy);
     Value               *v3Pbitcast = IRB.CreateBitCast(
-                      v3P, IntegerType::get(C, v3P->getType()->getPrimitiveSizeInBits()));
+        v3P, IntegerType::get(C, v3P->getType()->getPrimitiveSizeInBits()));
     Value *v3Pcasted =
         IRB.CreateIntCast(v3Pbitcast, IntegerType::get(C, 64), false);
     args.push_back(v1Pcasted);
diff --git a/instrumentation/compare-transform-pass.so.cc b/instrumentation/compare-transform-pass.so.cc
index efc99d20..5dd705cf 100644
--- a/instrumentation/compare-transform-pass.so.cc
+++ b/instrumentation/compare-transform-pass.so.cc
@@ -623,7 +623,7 @@ bool CompareTransform::transformCmps(Module &M, const bool processStrcmp,
 
         IRBuilder<> cur_lenchk_IRB(&*(cur_lenchk_bb->getFirstInsertionPt()));
         Value      *icmp = cur_lenchk_IRB.CreateICmpEQ(
-                 sizedValue, ConstantInt::get(sizedValue->getType(), i));
+            sizedValue, ConstantInt::get(sizedValue->getType(), i));
         cur_lenchk_IRB.CreateCondBr(icmp, end_bb, cur_cmp_bb);
         cur_lenchk_bb->getTerminator()->eraseFromParent();
 
diff --git a/instrumentation/split-compares-pass.so.cc b/instrumentation/split-compares-pass.so.cc
index 8a07610c..aec6758e 100644
--- a/instrumentation/split-compares-pass.so.cc
+++ b/instrumentation/split-compares-pass.so.cc
@@ -60,7 +60,7 @@ using namespace llvm;
 
 // uncomment this toggle function verification at each step. horribly slow, but
 // helps to pinpoint a potential problem in the splitting code.
-//#define VERIFY_TOO_MUCH 1
+// #define VERIFY_TOO_MUCH 1
 
 namespace {
 
diff --git a/qemu_mode/libcompcov/libcompcov.so.c b/qemu_mode/libcompcov/libcompcov.so.c
index b6ee0019..b57e9701 100644
--- a/qemu_mode/libcompcov/libcompcov.so.c
+++ b/qemu_mode/libcompcov/libcompcov.so.c
@@ -68,7 +68,11 @@ static int debug_fd = -1;
 
 #define MAX_MAPPINGS 1024
 
-static struct mapping { void *st, *en; } __compcov_ro[MAX_MAPPINGS];
+static struct mapping {
+
+  void *st, *en;
+
+} __compcov_ro[MAX_MAPPINGS];
 
 static u32 __compcov_ro_cnt;
 
diff --git a/src/afl-cc.c b/src/afl-cc.c
index 972ac8cd..e3cc04dd 100644
--- a/src/afl-cc.c
+++ b/src/afl-cc.c
@@ -933,10 +933,10 @@ static void edit_params(u32 argc, char **argv, char **envp) {
 
     }
 
-    //#if LLVM_MAJOR >= 13
-    //    // Use the old pass manager in LLVM 14 which the AFL++ passes still
-    //    use. cc_params[cc_par_cnt++] = "-flegacy-pass-manager";
-    //#endif
+    // #if LLVM_MAJOR >= 13
+    //     // Use the old pass manager in LLVM 14 which the AFL++ passes still
+    //     use. cc_params[cc_par_cnt++] = "-flegacy-pass-manager";
+    // #endif
 
     if (lto_mode && !have_c) {
 
diff --git a/src/afl-fuzz-redqueen.c b/src/afl-fuzz-redqueen.c
index 6e4a655b..d9dc50df 100644
--- a/src/afl-fuzz-redqueen.c
+++ b/src/afl-fuzz-redqueen.c
@@ -28,8 +28,8 @@
 #include "afl-fuzz.h"
 #include "cmplog.h"
 
-//#define _DEBUG
-//#define CMPLOG_INTROSPECTION
+// #define _DEBUG
+// #define CMPLOG_INTROSPECTION
 
 // CMP attribute enum
 enum {
@@ -571,7 +571,7 @@ static u8 its_fuzz(afl_state_t *afl, u8 *buf, u32 len, u8 *status) {
 
 }
 
-//#ifdef CMPLOG_SOLVE_TRANSFORM
+// #ifdef CMPLOG_SOLVE_TRANSFORM
 static int strntoll(const char *str, size_t sz, char **end, int base,
                     long long *out) {
 
@@ -771,7 +771,7 @@ static void to_base64(u8 *src, u8 *dst, u32 dst_len) {
 
 #endif
 
-//#endif
+// #endif
 
 static u8 cmp_extend_encoding(afl_state_t *afl, struct cmp_header *h,
                               u64 pattern, u64 repl, u64 o_pattern,
@@ -803,8 +803,8 @@ static u8 cmp_extend_encoding(afl_state_t *afl, struct cmp_header *h,
   //         o_pattern, pattern, repl, changed_val, idx, taint_len,
   //         hshape, attr);
 
-  //#ifdef CMPLOG_SOLVE_TRANSFORM
-  // reverse atoi()/strnu?toll() is expensive, so we only to it in lvl 3
+  // #ifdef CMPLOG_SOLVE_TRANSFORM
+  //  reverse atoi()/strnu?toll() is expensive, so we only to it in lvl 3
   if (afl->cmplog_enable_transform && (lvl & LVL3)) {
 
     u8                *endptr;
@@ -1120,7 +1120,7 @@ static u8 cmp_extend_encoding(afl_state_t *afl, struct cmp_header *h,
 
   }
 
-  //#endif
+  // #endif
 
   // we only allow this for ascii2integer (above) so leave if this is the case
   if (unlikely(pattern == o_pattern)) { return 0; }
@@ -1275,7 +1275,7 @@ static u8 cmp_extend_encoding(afl_state_t *afl, struct cmp_header *h,
   //       16 = modified float, 32 = modified integer (modified = wont match
   //                                                   in original buffer)
 
-  //#ifdef CMPLOG_SOLVE_ARITHMETIC
+  // #ifdef CMPLOG_SOLVE_ARITHMETIC
   if (!afl->cmplog_enable_arith || lvl < LVL3 || attr == IS_TRANSFORM) {
 
     return 0;
@@ -1440,8 +1440,8 @@ static u8 cmp_extend_encoding(afl_state_t *afl, struct cmp_header *h,
 
   }
 
-  //#endif                                           /*
-  // CMPLOG_SOLVE_ARITHMETIC
+  // #endif                                           /*
+  //  CMPLOG_SOLVE_ARITHMETIC
 
   return 0;
 
@@ -1948,9 +1948,9 @@ static u8 rtn_extend_encoding(afl_state_t *afl, u8 entry,
 #ifndef CMPLOG_COMBINE
   (void)(cbuf);
 #endif
-  //#ifndef CMPLOG_SOLVE_TRANSFORM
-  //  (void)(changed_val);
-  //#endif
+  // #ifndef CMPLOG_SOLVE_TRANSFORM
+  //   (void)(changed_val);
+  // #endif
 
   if (afl->fsrv.total_execs - last_update > screen_update) {
 
@@ -2418,7 +2418,7 @@ static u8 rtn_extend_encoding(afl_state_t *afl, u8 entry,
 
   }
 
-  //#endif
+  // #endif
 
   return 0;
 
@@ -2818,9 +2818,9 @@ u8 input_to_state_stage(afl_state_t *afl, u8 *orig_buf, u8 *buf, u32 len) {
 
     } else if ((lvl & LVL1)
 
-               //#ifdef CMPLOG_SOLVE_TRANSFORM
+               // #ifdef CMPLOG_SOLVE_TRANSFORM
                || ((lvl & LVL3) && afl->cmplog_enable_transform)
-               //#endif
+               // #endif
     ) {
 
       if (unlikely(rtn_fuzz(afl, k, orig_buf, buf, cbuf, len, lvl, taint))) {
diff --git a/utils/afl_untracer/afl-untracer.c b/utils/afl_untracer/afl-untracer.c
index a18e314e..e1038212 100644
--- a/utils/afl_untracer/afl-untracer.c
+++ b/utils/afl_untracer/afl-untracer.c
@@ -288,7 +288,7 @@ library_list_t *find_library(char *name) {
 #pragma GCC optimize("O0")
 void        breakpoint(void) {
 
-         if (debug) fprintf(stderr, "Breakpoint function \"breakpoint\" reached.\n");
+  if (debug) fprintf(stderr, "Breakpoint function \"breakpoint\" reached.\n");
 
 }
 
diff --git a/utils/libtokencap/libtokencap.so.c b/utils/libtokencap/libtokencap.so.c
index 299056ab..b21f3068 100644
--- a/utils/libtokencap/libtokencap.so.c
+++ b/utils/libtokencap/libtokencap.so.c
@@ -81,7 +81,11 @@ void *(*__libc_memmem)(const void *haystack, size_t haystack_len,
 
 #define MAX_MAPPINGS 1024
 
-static struct mapping { void *st, *en; } __tokencap_ro[MAX_MAPPINGS];
+static struct mapping {
+
+  void *st, *en;
+
+} __tokencap_ro[MAX_MAPPINGS];
 
 static u32   __tokencap_ro_cnt;
 static u8    __tokencap_ro_loaded;
diff --git a/utils/socket_fuzzing/socketfuzz.c b/utils/socket_fuzzing/socketfuzz.c
index 3ec8383b..7497519e 100644
--- a/utils/socket_fuzzing/socketfuzz.c
+++ b/utils/socket_fuzzing/socketfuzz.c
@@ -23,7 +23,8 @@
 #include <errno.h>
 #include <stdio.h>
 #include <poll.h>
-//#include "logging.h" // switche from preeny_info() to fprintf(stderr, "Info: "
+// #include "logging.h" // switched from preeny_info() to fprintf(stderr, "Info:
+// "
 
 //
 // originals
-- 
cgit 1.4.1


From 501226c992e5c47672907c5dde7f968f4e8fb001 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Tue, 23 May 2023 14:41:59 +0200
Subject: correct rtn cmplog map size

---
 include/cmplog.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'include')

diff --git a/include/cmplog.h b/include/cmplog.h
index 6e16e6b0..e4821444 100644
--- a/include/cmplog.h
+++ b/include/cmplog.h
@@ -34,7 +34,7 @@
 
 #define CMP_MAP_W 65536
 #define CMP_MAP_H 32
-#define CMP_MAP_RTN_H (CMP_MAP_H / 4)
+#define CMP_MAP_RTN_H (CMP_MAP_H / 2)
 
 #define SHAPE_BYTES(x) (x + 1)
 
-- 
cgit 1.4.1


From 8de7f6131d48e27d53e894b65bd11e0dc3817639 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Tue, 6 Jun 2023 13:12:31 +0200
Subject: add current mutation strategy to include

---
 include/afl-mutations.h | 161 ++++++++++++++++++++++++++++++++++++++++++++++++
 src/afl-fuzz-one.c      |   7 ++-
 2 files changed, 166 insertions(+), 2 deletions(-)

(limited to 'include')

diff --git a/include/afl-mutations.h b/include/afl-mutations.h
index 8d40855d..08037b09 100644
--- a/include/afl-mutations.h
+++ b/include/afl-mutations.h
@@ -77,6 +77,167 @@ enum {
 
 };
 
+  #define MUT_NORMAL_ARRAY_SIZE 77
+u32 normal_splice[MUT_NORMAL_ARRAY_SIZE] = {MUT_FLIPBIT,
+                                            MUT_FLIPBIT,
+                                            MUT_FLIPBIT,
+                                            MUT_FLIPBIT,
+                                            MUT_INTERESTING8,
+                                            MUT_INTERESTING8,
+                                            MUT_INTERESTING8,
+                                            MUT_INTERESTING8,
+                                            MUT_INTERESTING16,
+                                            MUT_INTERESTING16,
+                                            MUT_INTERESTING16BE,
+                                            MUT_INTERESTING16BE,
+                                            MUT_INTERESTING32,
+                                            MUT_INTERESTING32,
+                                            MUT_INTERESTING32BE,
+                                            MUT_INTERESTING32BE,
+                                            MUT_ARITH8_,
+                                            MUT_ARITH8_,
+                                            MUT_ARITH8_,
+                                            MUT_ARITH8_,
+                                            MUT_ARITH8,
+                                            MUT_ARITH8,
+                                            MUT_ARITH8,
+                                            MUT_ARITH8,
+                                            MUT_ARITH16_,
+                                            MUT_ARITH16_,
+                                            MUT_ARITH16BE_,
+                                            MUT_ARITH16BE_,
+                                            MUT_ARITH16,
+                                            MUT_ARITH16,
+                                            MUT_ARITH16BE,
+                                            MUT_ARITH16BE,
+                                            MUT_ARITH32_,
+                                            MUT_ARITH32_,
+                                            MUT_ARITH32BE_,
+                                            MUT_ARITH32BE_,
+                                            MUT_ARITH32,
+                                            MUT_ARITH32,
+                                            MUT_ARITH32BE,
+                                            MUT_ARITH32BE,
+                                            MUT_RAND8,
+                                            MUT_RAND8,
+                                            MUT_RAND8,
+                                            MUT_RAND8,
+                                            MUT_CLONE_COPY,
+                                            MUT_CLONE_COPY,
+                                            MUT_CLONE_COPY,
+                                            MUT_CLONE_FIXED,
+                                            MUT_OVERWRITE_COPY,
+                                            MUT_OVERWRITE_COPY,
+                                            MUT_OVERWRITE_COPY,
+                                            MUT_OVERWRITE_FIXED,
+                                            MUT_BYTEADD,
+                                            MUT_BYTESUB,
+                                            MUT_FLIP8,
+                                            MUT_SWITCH,
+                                            MUT_SWITCH,
+                                            MUT_DEL,
+                                            MUT_DEL,
+                                            MUT_DEL,
+                                            MUT_DEL,
+                                            MUT_DEL,
+                                            MUT_DEL,
+                                            MUT_DEL,
+                                            MUT_DEL,
+                                            MUT_EXTRA_OVERWRITE,
+                                            MUT_EXTRA_OVERWRITE,
+                                            MUT_EXTRA_INSERT,
+                                            MUT_EXTRA_INSERT,
+                                            MUT_AUTO_EXTRA_OVERWRITE,
+                                            MUT_AUTO_EXTRA_OVERWRITE,
+                                            MUT_AUTO_EXTRA_INSERT,
+                                            MUT_AUTO_EXTRA_INSERT,
+                                            MUT_SPLICE_OVERWRITE,
+                                            MUT_SPLICE_OVERWRITE,
+                                            MUT_SPLICE_INSERT,
+                                            MUT_SPLICE_INSERT};
+  #define MUT_SPLICE_ARRAY_SIZE 81
+u32 full_splice_array[MUT_SPLICE_ARRAY_SIZE] = {MUT_FLIPBIT,
+                                                MUT_FLIPBIT,
+                                                MUT_FLIPBIT,
+                                                MUT_FLIPBIT,
+                                                MUT_INTERESTING8,
+                                                MUT_INTERESTING8,
+                                                MUT_INTERESTING8,
+                                                MUT_INTERESTING8,
+                                                MUT_INTERESTING16,
+                                                MUT_INTERESTING16,
+                                                MUT_INTERESTING16BE,
+                                                MUT_INTERESTING16BE,
+                                                MUT_INTERESTING32,
+                                                MUT_INTERESTING32,
+                                                MUT_INTERESTING32BE,
+                                                MUT_INTERESTING32BE,
+                                                MUT_ARITH8_,
+                                                MUT_ARITH8_,
+                                                MUT_ARITH8_,
+                                                MUT_ARITH8_,
+                                                MUT_ARITH8,
+                                                MUT_ARITH8,
+                                                MUT_ARITH8,
+                                                MUT_ARITH8,
+                                                MUT_ARITH16_,
+                                                MUT_ARITH16_,
+                                                MUT_ARITH16BE_,
+                                                MUT_ARITH16BE_,
+                                                MUT_ARITH16,
+                                                MUT_ARITH16,
+                                                MUT_ARITH16BE,
+                                                MUT_ARITH16BE,
+                                                MUT_ARITH32_,
+                                                MUT_ARITH32_,
+                                                MUT_ARITH32BE_,
+                                                MUT_ARITH32BE_,
+                                                MUT_ARITH32,
+                                                MUT_ARITH32,
+                                                MUT_ARITH32BE,
+                                                MUT_ARITH32BE,
+                                                MUT_RAND8,
+                                                MUT_RAND8,
+                                                MUT_RAND8,
+                                                MUT_RAND8,
+                                                MUT_CLONE_COPY,
+                                                MUT_CLONE_COPY,
+                                                MUT_CLONE_COPY,
+                                                MUT_CLONE_FIXED,
+                                                MUT_OVERWRITE_COPY,
+                                                MUT_OVERWRITE_COPY,
+                                                MUT_OVERWRITE_COPY,
+                                                MUT_OVERWRITE_FIXED,
+                                                MUT_BYTEADD,
+                                                MUT_BYTESUB,
+                                                MUT_FLIP8,
+                                                MUT_SWITCH,
+                                                MUT_SWITCH,
+                                                MUT_DEL,
+                                                MUT_DEL,
+                                                MUT_DEL,
+                                                MUT_DEL,
+                                                MUT_DEL,
+                                                MUT_DEL,
+                                                MUT_DEL,
+                                                MUT_DEL,
+                                                MUT_EXTRA_OVERWRITE,
+                                                MUT_EXTRA_OVERWRITE,
+                                                MUT_EXTRA_INSERT,
+                                                MUT_EXTRA_INSERT,
+                                                MUT_AUTO_EXTRA_OVERWRITE,
+                                                MUT_AUTO_EXTRA_OVERWRITE,
+                                                MUT_AUTO_EXTRA_INSERT,
+                                                MUT_AUTO_EXTRA_INSERT,
+                                                MUT_SPLICE_OVERWRITE,
+                                                MUT_SPLICE_OVERWRITE,
+                                                MUT_SPLICE_OVERWRITE,
+                                                MUT_SPLICE_OVERWRITE,
+                                                MUT_SPLICE_INSERT,
+                                                MUT_SPLICE_INSERT,
+                                                MUT_SPLICE_INSERT,
+                                                MUT_SPLICE_INSERT};
+
 u32 mutation_strategy_exploration_text[MUT_STRATEGY_ARRAY_SIZE] = {
 
     MUT_FLIPBIT,
diff --git a/src/afl-fuzz-one.c b/src/afl-fuzz-one.c
index f5ddea0e..312e180d 100644
--- a/src/afl-fuzz-one.c
+++ b/src/afl-fuzz-one.c
@@ -2076,7 +2076,7 @@ havoc_stage:
      where we take the input file and make random stacked tweaks. */
 
   u32 *mutation_array;
-  u32  stack_max, stack_max_pow = afl->havoc_stack_pow2;
+  u32  stack_max;  // stack_max_pow = afl->havoc_stack_pow2;
 
   if (unlikely(afl->text_input || afl->queue_cur->is_ascii)) {  // is text?
 
@@ -2104,6 +2104,7 @@ havoc_stage:
 
   }
 
+  /*
   if (temp_len < 64) {
 
     --stack_max_pow;
@@ -2118,7 +2119,9 @@ havoc_stage:
 
   }
 
-  stack_max = 1 << stack_max_pow;
+  */
+
+  stack_max = 1 << (1 + rand_below(afl, afl->havoc_stack_pow2));
 
   // + (afl->extras_cnt ? 2 : 0) + (afl->a_extras_cnt ? 2 : 0);
 
-- 
cgit 1.4.1


From 9b2c4a2a5a8e54a80bdb82fee39891fbe42544e8 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Tue, 6 Jun 2023 16:54:12 +0200
Subject: nit

---
 include/afl-mutations.h | 155 ++++++++++++++++++++++++------------------------
 1 file changed, 78 insertions(+), 77 deletions(-)

(limited to 'include')

diff --git a/include/afl-mutations.h b/include/afl-mutations.h
index 08037b09..a3c9fd59 100644
--- a/include/afl-mutations.h
+++ b/include/afl-mutations.h
@@ -78,83 +78,84 @@ enum {
 };
 
   #define MUT_NORMAL_ARRAY_SIZE 77
-u32 normal_splice[MUT_NORMAL_ARRAY_SIZE] = {MUT_FLIPBIT,
-                                            MUT_FLIPBIT,
-                                            MUT_FLIPBIT,
-                                            MUT_FLIPBIT,
-                                            MUT_INTERESTING8,
-                                            MUT_INTERESTING8,
-                                            MUT_INTERESTING8,
-                                            MUT_INTERESTING8,
-                                            MUT_INTERESTING16,
-                                            MUT_INTERESTING16,
-                                            MUT_INTERESTING16BE,
-                                            MUT_INTERESTING16BE,
-                                            MUT_INTERESTING32,
-                                            MUT_INTERESTING32,
-                                            MUT_INTERESTING32BE,
-                                            MUT_INTERESTING32BE,
-                                            MUT_ARITH8_,
-                                            MUT_ARITH8_,
-                                            MUT_ARITH8_,
-                                            MUT_ARITH8_,
-                                            MUT_ARITH8,
-                                            MUT_ARITH8,
-                                            MUT_ARITH8,
-                                            MUT_ARITH8,
-                                            MUT_ARITH16_,
-                                            MUT_ARITH16_,
-                                            MUT_ARITH16BE_,
-                                            MUT_ARITH16BE_,
-                                            MUT_ARITH16,
-                                            MUT_ARITH16,
-                                            MUT_ARITH16BE,
-                                            MUT_ARITH16BE,
-                                            MUT_ARITH32_,
-                                            MUT_ARITH32_,
-                                            MUT_ARITH32BE_,
-                                            MUT_ARITH32BE_,
-                                            MUT_ARITH32,
-                                            MUT_ARITH32,
-                                            MUT_ARITH32BE,
-                                            MUT_ARITH32BE,
-                                            MUT_RAND8,
-                                            MUT_RAND8,
-                                            MUT_RAND8,
-                                            MUT_RAND8,
-                                            MUT_CLONE_COPY,
-                                            MUT_CLONE_COPY,
-                                            MUT_CLONE_COPY,
-                                            MUT_CLONE_FIXED,
-                                            MUT_OVERWRITE_COPY,
-                                            MUT_OVERWRITE_COPY,
-                                            MUT_OVERWRITE_COPY,
-                                            MUT_OVERWRITE_FIXED,
-                                            MUT_BYTEADD,
-                                            MUT_BYTESUB,
-                                            MUT_FLIP8,
-                                            MUT_SWITCH,
-                                            MUT_SWITCH,
-                                            MUT_DEL,
-                                            MUT_DEL,
-                                            MUT_DEL,
-                                            MUT_DEL,
-                                            MUT_DEL,
-                                            MUT_DEL,
-                                            MUT_DEL,
-                                            MUT_DEL,
-                                            MUT_EXTRA_OVERWRITE,
-                                            MUT_EXTRA_OVERWRITE,
-                                            MUT_EXTRA_INSERT,
-                                            MUT_EXTRA_INSERT,
-                                            MUT_AUTO_EXTRA_OVERWRITE,
-                                            MUT_AUTO_EXTRA_OVERWRITE,
-                                            MUT_AUTO_EXTRA_INSERT,
-                                            MUT_AUTO_EXTRA_INSERT,
-                                            MUT_SPLICE_OVERWRITE,
-                                            MUT_SPLICE_OVERWRITE,
-                                            MUT_SPLICE_INSERT,
-                                            MUT_SPLICE_INSERT};
+u32 normal_splice_array[MUT_NORMAL_ARRAY_SIZE] = {MUT_FLIPBIT,
+                                                  MUT_FLIPBIT,
+                                                  MUT_FLIPBIT,
+                                                  MUT_FLIPBIT,
+                                                  MUT_INTERESTING8,
+                                                  MUT_INTERESTING8,
+                                                  MUT_INTERESTING8,
+                                                  MUT_INTERESTING8,
+                                                  MUT_INTERESTING16,
+                                                  MUT_INTERESTING16,
+                                                  MUT_INTERESTING16BE,
+                                                  MUT_INTERESTING16BE,
+                                                  MUT_INTERESTING32,
+                                                  MUT_INTERESTING32,
+                                                  MUT_INTERESTING32BE,
+                                                  MUT_INTERESTING32BE,
+                                                  MUT_ARITH8_,
+                                                  MUT_ARITH8_,
+                                                  MUT_ARITH8_,
+                                                  MUT_ARITH8_,
+                                                  MUT_ARITH8,
+                                                  MUT_ARITH8,
+                                                  MUT_ARITH8,
+                                                  MUT_ARITH8,
+                                                  MUT_ARITH16_,
+                                                  MUT_ARITH16_,
+                                                  MUT_ARITH16BE_,
+                                                  MUT_ARITH16BE_,
+                                                  MUT_ARITH16,
+                                                  MUT_ARITH16,
+                                                  MUT_ARITH16BE,
+                                                  MUT_ARITH16BE,
+                                                  MUT_ARITH32_,
+                                                  MUT_ARITH32_,
+                                                  MUT_ARITH32BE_,
+                                                  MUT_ARITH32BE_,
+                                                  MUT_ARITH32,
+                                                  MUT_ARITH32,
+                                                  MUT_ARITH32BE,
+                                                  MUT_ARITH32BE,
+                                                  MUT_RAND8,
+                                                  MUT_RAND8,
+                                                  MUT_RAND8,
+                                                  MUT_RAND8,
+                                                  MUT_CLONE_COPY,
+                                                  MUT_CLONE_COPY,
+                                                  MUT_CLONE_COPY,
+                                                  MUT_CLONE_FIXED,
+                                                  MUT_OVERWRITE_COPY,
+                                                  MUT_OVERWRITE_COPY,
+                                                  MUT_OVERWRITE_COPY,
+                                                  MUT_OVERWRITE_FIXED,
+                                                  MUT_BYTEADD,
+                                                  MUT_BYTESUB,
+                                                  MUT_FLIP8,
+                                                  MUT_SWITCH,
+                                                  MUT_SWITCH,
+                                                  MUT_DEL,
+                                                  MUT_DEL,
+                                                  MUT_DEL,
+                                                  MUT_DEL,
+                                                  MUT_DEL,
+                                                  MUT_DEL,
+                                                  MUT_DEL,
+                                                  MUT_DEL,
+                                                  MUT_EXTRA_OVERWRITE,
+                                                  MUT_EXTRA_OVERWRITE,
+                                                  MUT_EXTRA_INSERT,
+                                                  MUT_EXTRA_INSERT,
+                                                  MUT_AUTO_EXTRA_OVERWRITE,
+                                                  MUT_AUTO_EXTRA_OVERWRITE,
+                                                  MUT_AUTO_EXTRA_INSERT,
+                                                  MUT_AUTO_EXTRA_INSERT,
+                                                  MUT_SPLICE_OVERWRITE,
+                                                  MUT_SPLICE_OVERWRITE,
+                                                  MUT_SPLICE_INSERT,
+                                                  MUT_SPLICE_INSERT};
+
   #define MUT_SPLICE_ARRAY_SIZE 81
 u32 full_splice_array[MUT_SPLICE_ARRAY_SIZE] = {MUT_FLIPBIT,
                                                 MUT_FLIPBIT,
-- 
cgit 1.4.1


From c28779adc543ffd3c68696867eef0f719ecee9d4 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Thu, 8 Jun 2023 12:32:51 +0200
Subject: show fuzzing state

---
 include/afl-fuzz.h   |  1 +
 src/afl-fuzz-one.c   |  5 +++--
 src/afl-fuzz-stats.c | 46 +++++++++++++++++++++++++++++++++++++++++++---
 3 files changed, 47 insertions(+), 5 deletions(-)

(limited to 'include')

diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index e1359dc8..c6c45fbd 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -1202,6 +1202,7 @@ u8     check_if_text_buf(u8 *buf, u32 len);
 #ifndef AFL_SHOWMAP
 void setup_signal_handlers(void);
 #endif
+char *get_fuzzing_state(afl_state_t *afl);
 
 /* CmpLog */
 
diff --git a/src/afl-fuzz-one.c b/src/afl-fuzz-one.c
index 9685885b..af5e57a0 100644
--- a/src/afl-fuzz-one.c
+++ b/src/afl-fuzz-one.c
@@ -402,11 +402,12 @@ u8 fuzz_one_original(afl_state_t *afl) {
   if (unlikely(afl->not_on_tty)) {
 
     ACTF(
-        "Fuzzing test case #%u (%u total, %llu crashes saved, mode=%s, "
+        "Fuzzing test case #%u (%u total, %llu crashes saved, state: %s, "
+        "mode=%s, "
         "perf_score=%0.0f, weight=%0.0f, favorite=%u, was_fuzzed=%u, "
         "exec_us=%llu, hits=%u, map=%u, ascii=%u)...",
         afl->current_entry, afl->queued_items, afl->saved_crashes,
-        afl->fuzz_mode ? "exploit" : "explore",
+        get_fuzzing_state(afl), afl->fuzz_mode ? "exploit" : "explore",
         afl->queue_cur->perf_score, afl->queue_cur->weight,
         afl->queue_cur->favored, afl->queue_cur->was_fuzzed,
         afl->queue_cur->exec_us,
diff --git a/src/afl-fuzz-stats.c b/src/afl-fuzz-stats.c
index 4ffb2536..9a60fd47 100644
--- a/src/afl-fuzz-stats.c
+++ b/src/afl-fuzz-stats.c
@@ -27,6 +27,45 @@
 #include "envs.h"
 #include <limits.h>
 
+static char fuzzing_state[4][12] = {"started :-)", "in progress", "final phase",
+                                    "finished..."};
+
+char *get_fuzzing_state(afl_state_t *afl) {
+
+  u64 cur_ms = get_cur_time();
+  u64 last_find = cur_ms - afl->last_find_time;
+  u64 cur_run_time = cur_ms - afl->start_time;
+  u64 cur_total_run_time = afl->prev_run_time + cur_run_time;
+
+  if (unlikely(cur_run_time < 60 * 3 * 1000 ||
+               cur_total_run_time < 60 * 5 * 1000)) {
+
+    return fuzzing_state[0];
+
+  } else {
+
+    u64 last_find_100 = 100 * last_find;
+    u64 percent_cur = last_find_100 / cur_run_time;
+    u64 percent_total = last_find_100 / cur_total_run_time;
+
+    if (unlikely(percent_cur >= 90 && percent_total >= 90)) {
+
+      return fuzzing_state[3];
+
+    } else if (unlikely(percent_cur >= 75 && percent_total >= 75)) {
+
+      return fuzzing_state[2];
+
+    } else {
+
+      return fuzzing_state[1];
+
+    }
+
+  }
+
+}
+
 /* Write fuzzer setup file */
 
 void write_setup_file(afl_state_t *afl, u32 argc, char **argv) {
@@ -1283,9 +1322,10 @@ void show_stats_normal(afl_state_t *afl) {
 
   /* Last line */
 
-  SAYF(SET_G1 "\n" bSTG bLB bH cCYA                                bSTOP
-              " strategy:%s %s " bSTG bH20 bH10 bH2 bRB bSTOP cRST RESET_G1,
-       cPIN, afl->fuzz_mode == 0 ? "explore" : "exploit");
+  SAYF(SET_G1 "\n" bSTG bLB bH cCYA          bSTOP " strategy:" cPIN
+              " %s " bSTG bH10 cCYA          bSTOP " state:" cPIN
+              " %s " bSTG bH2 bRB bSTOP cRST RESET_G1,
+       afl->fuzz_mode == 0 ? "explore" : "exploit", get_fuzzing_state(afl));
 
 #undef IB
 
-- 
cgit 1.4.1


From bf2727b76366ce4c9cdc723c3f3ccffae3cc3619 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Mon, 12 Jun 2023 08:28:47 +0200
Subject: v4.07c release

---
 README.md         | 4 ++--
 TODO.md           | 2 ++
 docs/Changelog.md | 2 +-
 include/config.h  | 2 +-
 4 files changed, 6 insertions(+), 4 deletions(-)

(limited to 'include')

diff --git a/README.md b/README.md
index 0208a9fe..97fd3997 100644
--- a/README.md
+++ b/README.md
@@ -2,9 +2,9 @@
 
 <img align="right" src="https://raw.githubusercontent.com/AFLplusplus/Website/main/static/aflpp_bg.svg" alt="AFL++ logo" width="250" heigh="250">
 
-Release version: [4.06c](https://github.com/AFLplusplus/AFLplusplus/releases)
+Release version: [4.07c](https://github.com/AFLplusplus/AFLplusplus/releases)
 
-GitHub version: 4.07a
+GitHub version: 4.07c
 
 Repository:
 [https://github.com/AFLplusplus/AFLplusplus](https://github.com/AFLplusplus/AFLplusplus)
diff --git a/TODO.md b/TODO.md
index 2b7e8fcf..26e12cee 100644
--- a/TODO.md
+++ b/TODO.md
@@ -2,6 +2,8 @@
 
 ## Should
 
+ - afl-crash-analysis
+ - show in the UI when fuzzing is "done"
  - test cmplog for less than 16bit
  - support persistent and deferred fork server in afl-showmap?
  - better autodetection of shifting runtime timeout values
diff --git a/docs/Changelog.md b/docs/Changelog.md
index 9ed930b3..c52ddd56 100644
--- a/docs/Changelog.md
+++ b/docs/Changelog.md
@@ -3,7 +3,7 @@
   This is the list of all noteworthy changes made in every public
   release of the tool. See README.md for the general instruction manual.
 
-### Version ++4.07a (dev)
+### Version ++4.07c (release)
   - afl-fuzz:
     - reverse reading the seeds only on restarts (increases performance)
     - new env `AFL_POST_PROCESS_KEEP_ORIGINAL` to keep the orignal
diff --git a/include/config.h b/include/config.h
index 194786f7..53be8549 100644
--- a/include/config.h
+++ b/include/config.h
@@ -26,7 +26,7 @@
 /* Version string: */
 
 // c = release, a = volatile github dev, e = experimental branch
-#define VERSION "++4.07a"
+#define VERSION "++4.07c"
 
 /******************************************************
  *                                                    *
-- 
cgit 1.4.1


From 25eba95bbaf58539c65088fc8bc143ed30ad82b9 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Mon, 12 Jun 2023 08:43:30 +0200
Subject: update new feature config

---
 include/config.h     | 2 +-
 src/afl-fuzz-stats.c | 4 ++--
 2 files changed, 3 insertions(+), 3 deletions(-)

(limited to 'include')

diff --git a/include/config.h b/include/config.h
index b767d526..09d8620d 100644
--- a/include/config.h
+++ b/include/config.h
@@ -47,7 +47,7 @@
    switches to exploitation mode. It automatically switches back when new
    coverage is found.
    Default: 300 (seconds) */
-#define STRATEGY_SWITCH_TIME 300
+#define STRATEGY_SWITCH_TIME 600
 
 /* Default file permission umode when creating files (default: 0600) */
 #define DEFAULT_PERMISSION 0600
diff --git a/src/afl-fuzz-stats.c b/src/afl-fuzz-stats.c
index 9a60fd47..1499a7e4 100644
--- a/src/afl-fuzz-stats.c
+++ b/src/afl-fuzz-stats.c
@@ -48,11 +48,11 @@ char *get_fuzzing_state(afl_state_t *afl) {
     u64 percent_cur = last_find_100 / cur_run_time;
     u64 percent_total = last_find_100 / cur_total_run_time;
 
-    if (unlikely(percent_cur >= 90 && percent_total >= 90)) {
+    if (unlikely(percent_cur >= 80 && percent_total >= 80)) {
 
       return fuzzing_state[3];
 
-    } else if (unlikely(percent_cur >= 75 && percent_total >= 75)) {
+    } else if (unlikely(percent_cur >= 55 && percent_total >= 55)) {
 
       return fuzzing_state[2];
 
-- 
cgit 1.4.1


From 61b6f4ed9e4dce15c39e4350278a95a41ea2522c Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Mon, 12 Jun 2023 09:16:15 +0200
Subject: 4.08a init

---
 README.md         | 2 +-
 docs/Changelog.md | 7 +++++++
 include/config.h  | 2 +-
 3 files changed, 9 insertions(+), 2 deletions(-)

(limited to 'include')

diff --git a/README.md b/README.md
index 97fd3997..05c662c1 100644
--- a/README.md
+++ b/README.md
@@ -4,7 +4,7 @@
 
 Release version: [4.07c](https://github.com/AFLplusplus/AFLplusplus/releases)
 
-GitHub version: 4.07c
+GitHub version: 4.08a
 
 Repository:
 [https://github.com/AFLplusplus/AFLplusplus](https://github.com/AFLplusplus/AFLplusplus)
diff --git a/docs/Changelog.md b/docs/Changelog.md
index c52ddd56..98d59527 100644
--- a/docs/Changelog.md
+++ b/docs/Changelog.md
@@ -3,6 +3,13 @@
   This is the list of all noteworthy changes made in every public
   release of the tool. See README.md for the general instruction manual.
 
+### Version ++4.08a (dev)
+  - new mutation engine: mutations that favor discovery more paths are prefered
+    until no new finds for 10 minutes then switching to mutations that favor
+    triggering crashes. Modes and switch time can be configured wie `-P`.
+  - display the state of the fuzzing run in the UI :-)
+
+
 ### Version ++4.07c (release)
   - afl-fuzz:
     - reverse reading the seeds only on restarts (increases performance)
diff --git a/include/config.h b/include/config.h
index 53be8549..d8153a2c 100644
--- a/include/config.h
+++ b/include/config.h
@@ -26,7 +26,7 @@
 /* Version string: */
 
 // c = release, a = volatile github dev, e = experimental branch
-#define VERSION "++4.07c"
+#define VERSION "++4.08a"
 
 /******************************************************
  *                                                    *
-- 
cgit 1.4.1


From 091d66fa92cd9e4caa5829d579b1b996c49db8c9 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Mon, 12 Jun 2023 13:05:35 +0200
Subject: increase strategy switch

---
 include/config.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'include')

diff --git a/include/config.h b/include/config.h
index 5100d88f..8585041e 100644
--- a/include/config.h
+++ b/include/config.h
@@ -47,7 +47,7 @@
    switches to exploitation mode. It automatically switches back when new
    coverage is found.
    Default: 300 (seconds) */
-#define STRATEGY_SWITCH_TIME 600
+#define STRATEGY_SWITCH_TIME 1000
 
 /* Default file permission umode when creating files (default: 0600) */
 #define DEFAULT_PERMISSION 0600
-- 
cgit 1.4.1


From 90f83c13d08f44fbf50036076a1772909c4d2c86 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Thu, 22 Jun 2023 09:24:00 +0200
Subject: remove dead code, code format

---
 .custom-format.py                              |  2 +-
 docs/Changelog.md                              |  3 ++
 include/alloc-inl.h                            |  8 +++---
 instrumentation/SanitizerCoveragePCGUARD.so.cc | 39 ++------------------------
 qemu_mode/libqasan/dlmalloc.c                  |  2 +-
 src/afl-fuzz-init.c                            |  8 +++---
 src/afl-fuzz.c                                 |  3 +-
 utils/afl_network_proxy/afl-network-server.c   |  2 +-
 8 files changed, 19 insertions(+), 48 deletions(-)

(limited to 'include')

diff --git a/.custom-format.py b/.custom-format.py
index 1d5c8839..3521c05d 100755
--- a/.custom-format.py
+++ b/.custom-format.py
@@ -24,7 +24,7 @@ import importlib.metadata
 
 # string_re = re.compile('(\\"(\\\\.|[^"\\\\])*\\")') # TODO: for future use
 
-CURRENT_LLVM = os.getenv('LLVM_VERSION', 15)
+CURRENT_LLVM = os.getenv('LLVM_VERSION', 16)
 CLANG_FORMAT_BIN = os.getenv("CLANG_FORMAT_BIN", "")
 
 
diff --git a/docs/Changelog.md b/docs/Changelog.md
index 246c3cac..c850c43e 100644
--- a/docs/Changelog.md
+++ b/docs/Changelog.md
@@ -18,6 +18,9 @@
     - fixed a bug inherited from vanilla AFL where a coverage of
       map[123] = 11 would be the same as map[1123] = 1
     - warn on crashing inputs
+  - afl-cc
+    - fixed an off-by-one instrumentation of iselect, hurting coverage a bit.
+      Thanks to @amykweon for spotting and fixing!
 
 
 ### Version ++4.07c (release)
diff --git a/include/alloc-inl.h b/include/alloc-inl.h
index 1e9a192b..cff808b2 100644
--- a/include/alloc-inl.h
+++ b/include/alloc-inl.h
@@ -322,7 +322,7 @@ static inline void DFL_ck_free(void *mem) {
 static inline void *DFL_ck_realloc(void *orig, u32 size) {
 
   void *ret;
-  u32 old_size = 0;
+  u32   old_size = 0;
 
   if (!size) {
 
@@ -392,7 +392,7 @@ static inline void *DFL_ck_realloc(void *orig, u32 size) {
 static inline u8 *DFL_ck_strdup(u8 *str) {
 
   void *ret;
-  u32 size;
+  u32   size;
 
   if (!str) return NULL;
 
@@ -438,14 +438,14 @@ struct TRK_obj {
 
   void *ptr;
   char *file, *func;
-  u32 line;
+  u32   line;
 
 };
 
     #ifdef AFL_MAIN
 
 struct TRK_obj *TRK[ALLOC_BUCKETS];
-u32 TRK_cnt[ALLOC_BUCKETS];
+u32             TRK_cnt[ALLOC_BUCKETS];
 
       #define alloc_report() TRK_report()
 
diff --git a/instrumentation/SanitizerCoveragePCGUARD.so.cc b/instrumentation/SanitizerCoveragePCGUARD.so.cc
index d87af775..57b5d128 100644
--- a/instrumentation/SanitizerCoveragePCGUARD.so.cc
+++ b/instrumentation/SanitizerCoveragePCGUARD.so.cc
@@ -225,49 +225,18 @@ llvmGetPassPluginInfo() {
 
 }
 
-#if LLVM_VERSION_MAJOR == 1
+#if LLVM_VERSION_MAJOR >= 16
 PreservedAnalyses ModuleSanitizerCoverageAFL::run(Module                &M,
                                                   ModuleAnalysisManager &MAM) {
 
-  ModuleSanitizerCoverageAFL ModuleSancov(Options);
-  auto &FAM = MAM.getResult<FunctionAnalysisManagerModuleProxy>(M).getManager();
-  auto  DTCallback = [&FAM](Function &F) -> const DominatorTree  *{
-
-    return &FAM.getResult<DominatorTreeAnalysis>(F);
-
-  };
-
-  auto PDTCallback = [&FAM](Function &F) -> const PostDominatorTree * {
-
-    return &FAM.getResult<PostDominatorTreeAnalysis>(F);
-
-  };
-
-  if (!ModuleSancov.instrumentModule(M, DTCallback, PDTCallback))
-    return PreservedAnalyses::all();
-
-  PreservedAnalyses PA = PreservedAnalyses::none();
-  // GlobalsAA is considered stateless and does not get invalidated unless
-  // explicitly invalidated; PreservedAnalyses::none() is not enough. Sanitizers
-  // make changes that require GlobalsAA to be invalidated.
-  PA.abandon<GlobalsAA>();
-  return PA;
-
-}
-
 #else
-  #if LLVM_VERSION_MAJOR >= 16
-PreservedAnalyses ModuleSanitizerCoverageAFL::run(Module &M,
-                                                  ModuleAnalysisManager &MAM) {
-
-  #else
 PreservedAnalyses ModuleSanitizerCoverageAFL::run(Module                &M,
                                                   ModuleAnalysisManager &MAM) {
 
-  #endif
+#endif
   ModuleSanitizerCoverageAFL ModuleSancov(Options);
   auto &FAM = MAM.getResult<FunctionAnalysisManagerModuleProxy>(M).getManager();
-  auto DTCallback = [&FAM](Function &F) -> const DominatorTree * {
+  auto  DTCallback = [&FAM](Function &F) -> const DominatorTree  *{
 
     return &FAM.getResult<DominatorTreeAnalysis>(F);
 
@@ -285,8 +254,6 @@ PreservedAnalyses ModuleSanitizerCoverageAFL::run(Module                &M,
 
 }
 
-#endif
-
 std::pair<Value *, Value *> ModuleSanitizerCoverageAFL::CreateSecStartEnd(
     Module &M, const char *Section, Type *Ty) {
 
diff --git a/qemu_mode/libqasan/dlmalloc.c b/qemu_mode/libqasan/dlmalloc.c
index 5d0b65ce..b459eb7b 100644
--- a/qemu_mode/libqasan/dlmalloc.c
+++ b/qemu_mode/libqasan/dlmalloc.c
@@ -1762,7 +1762,7 @@ static FORCEINLINE void *win32direct_mmap(size_t size) {
 static FORCEINLINE int win32munmap(void *ptr, size_t size) {
 
   MEMORY_BASIC_INFORMATION minfo;
-  char *cptr = (char *)ptr;
+  char                    *cptr = (char *)ptr;
 
   while (size) {
 
diff --git a/src/afl-fuzz-init.c b/src/afl-fuzz-init.c
index 13802f40..24fd7077 100644
--- a/src/afl-fuzz-init.c
+++ b/src/afl-fuzz-init.c
@@ -1542,8 +1542,8 @@ double get_runnable_processes(void) {
      processes well. */
 
   FILE *f = fopen("/proc/stat", "r");
-  u8 tmp[1024];
-  u32 val = 0;
+  u8    tmp[1024];
+  u32   val = 0;
 
   if (!f) { return 0; }
 
@@ -2226,7 +2226,7 @@ void check_crash_handling(void) {
    *BSD, so we can just let it slide for now. */
 
   s32 fd = open("/proc/sys/kernel/core_pattern", O_RDONLY);
-  u8 fchar;
+  u8  fchar;
 
   if (fd < 0) { return; }
 
@@ -2365,7 +2365,7 @@ void check_cpu_governor(afl_state_t *afl) {
   FATAL("Suboptimal CPU scaling governor");
 
 #elif defined __APPLE__
-  u64 min = 0, max = 0;
+  u64    min = 0, max = 0;
   size_t mlen = sizeof(min);
   if (afl->afl_env.afl_skip_cpufreq) return;
 
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index 8cf786af..79b05da7 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -164,7 +164,8 @@ static void usage(u8 *argv0, int more_help) {
       "\n"
 
       "Mutator settings:\n"
-      "  -a            - target expects ascii text input (prefer text mutators)\n"
+      "  -a            - target expects ascii text input (prefer text "
+      "mutators)\n"
       "  -g minlength  - set min length of generated fuzz input (default: 1)\n"
       "  -G maxlength  - set max length of generated fuzz input (default: "
       "%lu)\n"
diff --git a/utils/afl_network_proxy/afl-network-server.c b/utils/afl_network_proxy/afl-network-server.c
index 7eb3d18e..95b0a551 100644
--- a/utils/afl_network_proxy/afl-network-server.c
+++ b/utils/afl_network_proxy/afl-network-server.c
@@ -173,7 +173,7 @@ static void set_up_environment(afl_forkserver_t *fsrv) {
     }
 
     out_file = alloc_printf("%s/.afl-input-temp-%u", use_dir, getpid());
-	fsrv->out_file = out_file;
+    fsrv->out_file = out_file;
 
   }
 
-- 
cgit 1.4.1


From c2c27349c3d74f79ceb6cd3795862b21d90429ea Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Fri, 23 Jun 2023 17:08:21 +0200
Subject: new mutation weighting

---
 include/afl-mutations.h | 460 ++++++++++++++++++++++++++++++++++++++++++++++++
 src/afl-fuzz-one.c      |  30 +---
 2 files changed, 466 insertions(+), 24 deletions(-)

(limited to 'include')

diff --git a/include/afl-mutations.h b/include/afl-mutations.h
index a3c9fd59..cc4840c8 100644
--- a/include/afl-mutations.h
+++ b/include/afl-mutations.h
@@ -77,6 +77,466 @@ enum {
 
 };
 
+  #define MUT_TXT_ARRAY_SIZE 200
+u32 text_array[MUT_TXT_ARRAY_SIZE] = {MUT_FLIPBIT,
+                                      MUT_FLIPBIT,
+                                      MUT_FLIPBIT,
+                                      MUT_FLIPBIT,
+                                      MUT_FLIPBIT,
+                                      MUT_FLIPBIT,
+                                      MUT_FLIPBIT,
+                                      MUT_FLIPBIT,
+                                      MUT_INTERESTING8,
+                                      MUT_INTERESTING8,
+                                      MUT_INTERESTING8,
+                                      MUT_INTERESTING8,
+                                      MUT_INTERESTING16,
+                                      MUT_INTERESTING16,
+                                      MUT_INTERESTING16BE,
+                                      MUT_INTERESTING16BE,
+                                      MUT_INTERESTING32,
+                                      MUT_INTERESTING32,
+                                      MUT_INTERESTING32BE,
+                                      MUT_INTERESTING32BE,
+                                      MUT_ARITH8_,
+                                      MUT_ARITH8_,
+                                      MUT_ARITH8_,
+                                      MUT_ARITH8_,
+                                      MUT_ARITH8_,
+                                      MUT_ARITH8_,
+                                      MUT_ARITH8,
+                                      MUT_ARITH8,
+                                      MUT_ARITH8,
+                                      MUT_ARITH8,
+                                      MUT_ARITH8,
+                                      MUT_ARITH8,
+                                      MUT_ARITH16_,
+                                      MUT_ARITH16_,
+                                      MUT_ARITH16_,
+                                      MUT_ARITH16_,
+                                      MUT_ARITH16_,
+                                      MUT_ARITH16BE_,
+                                      MUT_ARITH16BE_,
+                                      MUT_ARITH16BE_,
+                                      MUT_ARITH16BE_,
+                                      MUT_ARITH16BE_,
+                                      MUT_ARITH16,
+                                      MUT_ARITH16,
+                                      MUT_ARITH16,
+                                      MUT_ARITH16,
+                                      MUT_ARITH16,
+                                      MUT_ARITH16BE,
+                                      MUT_ARITH16BE,
+                                      MUT_ARITH16BE,
+                                      MUT_ARITH16BE,
+                                      MUT_ARITH16BE,
+                                      MUT_ARITH32_,
+                                      MUT_ARITH32_,
+                                      MUT_ARITH32_,
+                                      MUT_ARITH32_,
+                                      MUT_ARITH32_,
+                                      MUT_ARITH32BE_,
+                                      MUT_ARITH32BE_,
+                                      MUT_ARITH32BE_,
+                                      MUT_ARITH32BE_,
+                                      MUT_ARITH32BE_,
+                                      MUT_ARITH32,
+                                      MUT_ARITH32,
+                                      MUT_ARITH32,
+                                      MUT_ARITH32,
+                                      MUT_ARITH32,
+                                      MUT_ARITH32BE,
+                                      MUT_ARITH32BE,
+                                      MUT_ARITH32BE,
+                                      MUT_ARITH32BE,
+                                      MUT_ARITH32BE,
+                                      MUT_RAND8,
+                                      MUT_RAND8,
+                                      MUT_RAND8,
+                                      MUT_RAND8,
+                                      MUT_RAND8,
+                                      MUT_RAND8,
+                                      MUT_RAND8,
+                                      MUT_RAND8,
+                                      MUT_CLONE_COPY,
+                                      MUT_CLONE_COPY,
+                                      MUT_CLONE_COPY,
+                                      MUT_CLONE_COPY,
+                                      MUT_CLONE_COPY,
+                                      MUT_CLONE_COPY,
+                                      MUT_CLONE_COPY,
+                                      MUT_CLONE_COPY,
+                                      MUT_CLONE_COPY,
+                                      MUT_CLONE_COPY,
+                                      MUT_CLONE_COPY,
+                                      MUT_CLONE_COPY,
+                                      MUT_CLONE_COPY,
+                                      MUT_CLONE_COPY,
+                                      MUT_CLONE_COPY,
+                                      MUT_CLONE_COPY,
+                                      MUT_CLONE_FIXED,
+                                      MUT_CLONE_FIXED,
+                                      MUT_CLONE_FIXED,
+                                      MUT_CLONE_FIXED,
+                                      MUT_CLONE_FIXED,
+                                      MUT_CLONE_FIXED,
+                                      MUT_CLONE_FIXED,
+                                      MUT_CLONE_FIXED,
+                                      MUT_OVERWRITE_COPY,
+                                      MUT_OVERWRITE_COPY,
+                                      MUT_OVERWRITE_COPY,
+                                      MUT_OVERWRITE_COPY,
+                                      MUT_OVERWRITE_COPY,
+                                      MUT_OVERWRITE_COPY,
+                                      MUT_OVERWRITE_COPY,
+                                      MUT_OVERWRITE_COPY,
+                                      MUT_OVERWRITE_COPY,
+                                      MUT_OVERWRITE_COPY,
+                                      MUT_OVERWRITE_FIXED,
+                                      MUT_OVERWRITE_FIXED,
+                                      MUT_OVERWRITE_FIXED,
+                                      MUT_OVERWRITE_FIXED,
+                                      MUT_OVERWRITE_FIXED,
+                                      MUT_BYTEADD,
+                                      MUT_BYTEADD,
+                                      MUT_BYTEADD,
+                                      MUT_BYTEADD,
+                                      MUT_BYTEADD,
+                                      MUT_BYTESUB,
+                                      MUT_BYTESUB,
+                                      MUT_BYTESUB,
+                                      MUT_BYTESUB,
+                                      MUT_BYTESUB,
+                                      MUT_FLIP8,
+                                      MUT_FLIP8,
+                                      MUT_FLIP8,
+                                      MUT_FLIP8,
+                                      MUT_SWITCH,
+                                      MUT_SWITCH,
+                                      MUT_SWITCH,
+                                      MUT_SWITCH,
+                                      MUT_SWITCH,
+                                      MUT_SWITCH,
+                                      MUT_SWITCH,
+                                      MUT_DEL,
+                                      MUT_DEL,
+                                      MUT_DEL,
+                                      MUT_DEL,
+                                      MUT_DEL,
+                                      MUT_DEL,
+                                      MUT_DEL,
+                                      MUT_DEL,
+                                      MUT_DEL,
+                                      MUT_DEL,
+                                      MUT_EXTRA_OVERWRITE,
+                                      MUT_EXTRA_OVERWRITE,
+                                      MUT_EXTRA_OVERWRITE,
+                                      MUT_EXTRA_OVERWRITE,
+                                      MUT_EXTRA_OVERWRITE,
+                                      MUT_EXTRA_OVERWRITE,
+                                      MUT_EXTRA_OVERWRITE,
+                                      MUT_EXTRA_INSERT,
+                                      MUT_EXTRA_INSERT,
+                                      MUT_EXTRA_INSERT,
+                                      MUT_EXTRA_INSERT,
+                                      MUT_EXTRA_INSERT,
+                                      MUT_EXTRA_INSERT,
+                                      MUT_EXTRA_INSERT,
+                                      MUT_EXTRA_INSERT,
+                                      MUT_EXTRA_INSERT,
+                                      MUT_AUTO_EXTRA_OVERWRITE,
+                                      MUT_AUTO_EXTRA_OVERWRITE,
+                                      MUT_AUTO_EXTRA_OVERWRITE,
+                                      MUT_AUTO_EXTRA_OVERWRITE,
+                                      MUT_AUTO_EXTRA_INSERT,
+                                      MUT_AUTO_EXTRA_INSERT,
+                                      MUT_AUTO_EXTRA_INSERT,
+                                      MUT_AUTO_EXTRA_INSERT,
+                                      MUT_AUTO_EXTRA_INSERT,
+                                      MUT_SPLICE_OVERWRITE,
+                                      MUT_SPLICE_OVERWRITE,
+                                      MUT_SPLICE_OVERWRITE,
+                                      MUT_SPLICE_OVERWRITE,
+                                      MUT_SPLICE_OVERWRITE,
+                                      MUT_SPLICE_OVERWRITE,
+                                      MUT_SPLICE_OVERWRITE,
+                                      MUT_SPLICE_OVERWRITE,
+                                      MUT_SPLICE_OVERWRITE,
+                                      MUT_SPLICE_OVERWRITE,
+                                      MUT_SPLICE_OVERWRITE,
+                                      MUT_SPLICE_OVERWRITE,
+                                      MUT_SPLICE_INSERT,
+                                      MUT_SPLICE_INSERT,
+                                      MUT_SPLICE_INSERT,
+                                      MUT_SPLICE_INSERT,
+                                      MUT_SPLICE_INSERT,
+                                      MUT_SPLICE_INSERT,
+                                      MUT_SPLICE_INSERT,
+                                      MUT_SPLICE_INSERT,
+                                      MUT_SPLICE_INSERT,
+                                      MUT_SPLICE_INSERT,
+                                      MUT_SPLICE_INSERT,
+                                      MUT_SPLICE_INSERT,
+                                      MUT_SPLICE_INSERT};
+
+  #define MUT_BIN_ARRAY_SIZE 256
+u32 binary_array[MUT_BIN_ARRAY_SIZE] = {MUT_FLIPBIT,
+                                        MUT_FLIPBIT,
+                                        MUT_FLIPBIT,
+                                        MUT_FLIPBIT,
+                                        MUT_FLIPBIT,
+                                        MUT_FLIPBIT,
+                                        MUT_FLIPBIT,
+                                        MUT_FLIPBIT,
+                                        MUT_FLIPBIT,
+                                        MUT_FLIPBIT,
+                                        MUT_FLIPBIT,
+                                        MUT_INTERESTING8,
+                                        MUT_INTERESTING8,
+                                        MUT_INTERESTING8,
+                                        MUT_INTERESTING8,
+                                        MUT_INTERESTING8,
+                                        MUT_INTERESTING8,
+                                        MUT_INTERESTING8,
+                                        MUT_INTERESTING8,
+                                        MUT_INTERESTING8,
+                                        MUT_INTERESTING16,
+                                        MUT_INTERESTING16,
+                                        MUT_INTERESTING16,
+                                        MUT_INTERESTING16,
+                                        MUT_INTERESTING16,
+                                        MUT_INTERESTING16,
+                                        MUT_INTERESTING16BE,
+                                        MUT_INTERESTING16BE,
+                                        MUT_INTERESTING16BE,
+                                        MUT_INTERESTING16BE,
+                                        MUT_INTERESTING16BE,
+                                        MUT_INTERESTING16BE,
+                                        MUT_INTERESTING32,
+                                        MUT_INTERESTING32,
+                                        MUT_INTERESTING32,
+                                        MUT_INTERESTING32,
+                                        MUT_INTERESTING32,
+                                        MUT_INTERESTING32,
+                                        MUT_INTERESTING32BE,
+                                        MUT_INTERESTING32BE,
+                                        MUT_INTERESTING32BE,
+                                        MUT_INTERESTING32BE,
+                                        MUT_INTERESTING32BE,
+                                        MUT_INTERESTING32BE,
+                                        MUT_ARITH8_,
+                                        MUT_ARITH8_,
+                                        MUT_ARITH8_,
+                                        MUT_ARITH8_,
+                                        MUT_ARITH8_,
+                                        MUT_ARITH8_,
+                                        MUT_ARITH8_,
+                                        MUT_ARITH8_,
+                                        MUT_ARITH8_,
+                                        MUT_ARITH8,
+                                        MUT_ARITH8,
+                                        MUT_ARITH8,
+                                        MUT_ARITH8,
+                                        MUT_ARITH8,
+                                        MUT_ARITH8,
+                                        MUT_ARITH8,
+                                        MUT_ARITH8,
+                                        MUT_ARITH8,
+                                        MUT_ARITH8,
+                                        MUT_ARITH16_,
+                                        MUT_ARITH16_,
+                                        MUT_ARITH16_,
+                                        MUT_ARITH16_,
+                                        MUT_ARITH16_,
+                                        MUT_ARITH16_,
+                                        MUT_ARITH16BE_,
+                                        MUT_ARITH16BE_,
+                                        MUT_ARITH16BE_,
+                                        MUT_ARITH16BE_,
+                                        MUT_ARITH16BE_,
+                                        MUT_ARITH16BE_,
+                                        MUT_ARITH16,
+                                        MUT_ARITH16,
+                                        MUT_ARITH16,
+                                        MUT_ARITH16,
+                                        MUT_ARITH16,
+                                        MUT_ARITH16,
+                                        MUT_ARITH16BE,
+                                        MUT_ARITH16BE,
+                                        MUT_ARITH16BE,
+                                        MUT_ARITH16BE,
+                                        MUT_ARITH16BE,
+                                        MUT_ARITH16BE,
+                                        MUT_ARITH32_,
+                                        MUT_ARITH32_,
+                                        MUT_ARITH32_,
+                                        MUT_ARITH32_,
+                                        MUT_ARITH32_,
+                                        MUT_ARITH32_,
+                                        MUT_ARITH32BE_,
+                                        MUT_ARITH32BE_,
+                                        MUT_ARITH32BE_,
+                                        MUT_ARITH32BE_,
+                                        MUT_ARITH32BE_,
+                                        MUT_ARITH32BE_,
+                                        MUT_ARITH32,
+                                        MUT_ARITH32,
+                                        MUT_ARITH32,
+                                        MUT_ARITH32,
+                                        MUT_ARITH32,
+                                        MUT_ARITH32,
+                                        MUT_ARITH32BE,
+                                        MUT_ARITH32BE,
+                                        MUT_ARITH32BE,
+                                        MUT_ARITH32BE,
+                                        MUT_ARITH32BE,
+                                        MUT_ARITH32BE,
+                                        MUT_RAND8,
+                                        MUT_RAND8,
+                                        MUT_RAND8,
+                                        MUT_RAND8,
+                                        MUT_RAND8,
+                                        MUT_RAND8,
+                                        MUT_RAND8,
+                                        MUT_RAND8,
+                                        MUT_RAND8,
+                                        MUT_CLONE_COPY,
+                                        MUT_CLONE_COPY,
+                                        MUT_CLONE_COPY,
+                                        MUT_CLONE_COPY,
+                                        MUT_CLONE_COPY,
+                                        MUT_CLONE_COPY,
+                                        MUT_CLONE_COPY,
+                                        MUT_CLONE_COPY,
+                                        MUT_CLONE_COPY,
+                                        MUT_CLONE_COPY,
+                                        MUT_CLONE_COPY,
+                                        MUT_CLONE_COPY,
+                                        MUT_CLONE_COPY,
+                                        MUT_CLONE_COPY,
+                                        MUT_CLONE_FIXED,
+                                        MUT_CLONE_FIXED,
+                                        MUT_CLONE_FIXED,
+                                        MUT_CLONE_FIXED,
+                                        MUT_CLONE_FIXED,
+                                        MUT_CLONE_FIXED,
+                                        MUT_CLONE_FIXED,
+                                        MUT_OVERWRITE_COPY,
+                                        MUT_OVERWRITE_COPY,
+                                        MUT_OVERWRITE_COPY,
+                                        MUT_OVERWRITE_COPY,
+                                        MUT_OVERWRITE_COPY,
+                                        MUT_OVERWRITE_COPY,
+                                        MUT_OVERWRITE_COPY,
+                                        MUT_OVERWRITE_COPY,
+                                        MUT_OVERWRITE_COPY,
+                                        MUT_OVERWRITE_COPY,
+                                        MUT_OVERWRITE_FIXED,
+                                        MUT_OVERWRITE_FIXED,
+                                        MUT_OVERWRITE_FIXED,
+                                        MUT_OVERWRITE_FIXED,
+                                        MUT_OVERWRITE_FIXED,
+                                        MUT_BYTEADD,
+                                        MUT_BYTEADD,
+                                        MUT_BYTEADD,
+                                        MUT_BYTEADD,
+                                        MUT_BYTEADD,
+                                        MUT_BYTEADD,
+                                        MUT_BYTESUB,
+                                        MUT_BYTESUB,
+                                        MUT_BYTESUB,
+                                        MUT_BYTESUB,
+                                        MUT_BYTESUB,
+                                        MUT_BYTESUB,
+                                        MUT_FLIP8,
+                                        MUT_FLIP8,
+                                        MUT_FLIP8,
+                                        MUT_FLIP8,
+                                        MUT_SWITCH,
+                                        MUT_SWITCH,
+                                        MUT_SWITCH,
+                                        MUT_SWITCH,
+                                        MUT_SWITCH,
+                                        MUT_SWITCH,
+                                        MUT_DEL,
+                                        MUT_DEL,
+                                        MUT_DEL,
+                                        MUT_DEL,
+                                        MUT_DEL,
+                                        MUT_DEL,
+                                        MUT_DEL,
+                                        MUT_DEL,
+                                        MUT_DEL,
+                                        MUT_EXTRA_OVERWRITE,
+                                        MUT_EXTRA_OVERWRITE,
+                                        MUT_EXTRA_OVERWRITE,
+                                        MUT_EXTRA_OVERWRITE,
+                                        MUT_EXTRA_OVERWRITE,
+                                        MUT_EXTRA_OVERWRITE,
+                                        MUT_EXTRA_OVERWRITE,
+                                        MUT_EXTRA_OVERWRITE,
+                                        MUT_EXTRA_OVERWRITE,
+                                        MUT_EXTRA_OVERWRITE,
+                                        MUT_EXTRA_INSERT,
+                                        MUT_EXTRA_INSERT,
+                                        MUT_EXTRA_INSERT,
+                                        MUT_EXTRA_INSERT,
+                                        MUT_EXTRA_INSERT,
+                                        MUT_EXTRA_INSERT,
+                                        MUT_EXTRA_INSERT,
+                                        MUT_EXTRA_INSERT,
+                                        MUT_EXTRA_INSERT,
+                                        MUT_EXTRA_INSERT,
+                                        MUT_EXTRA_INSERT,
+                                        MUT_EXTRA_INSERT,
+                                        MUT_AUTO_EXTRA_OVERWRITE,
+                                        MUT_AUTO_EXTRA_OVERWRITE,
+                                        MUT_AUTO_EXTRA_OVERWRITE,
+                                        MUT_AUTO_EXTRA_OVERWRITE,
+                                        MUT_AUTO_EXTRA_OVERWRITE,
+                                        MUT_AUTO_EXTRA_OVERWRITE,
+                                        MUT_AUTO_EXTRA_OVERWRITE,
+                                        MUT_AUTO_EXTRA_OVERWRITE,
+                                        MUT_AUTO_EXTRA_OVERWRITE,
+                                        MUT_AUTO_EXTRA_INSERT,
+                                        MUT_AUTO_EXTRA_INSERT,
+                                        MUT_AUTO_EXTRA_INSERT,
+                                        MUT_AUTO_EXTRA_INSERT,
+                                        MUT_AUTO_EXTRA_INSERT,
+                                        MUT_AUTO_EXTRA_INSERT,
+                                        MUT_AUTO_EXTRA_INSERT,
+                                        MUT_AUTO_EXTRA_INSERT,
+                                        MUT_AUTO_EXTRA_INSERT,
+                                        MUT_AUTO_EXTRA_INSERT,
+                                        MUT_AUTO_EXTRA_INSERT,
+                                        MUT_SPLICE_OVERWRITE,
+                                        MUT_SPLICE_OVERWRITE,
+                                        MUT_SPLICE_OVERWRITE,
+                                        MUT_SPLICE_OVERWRITE,
+                                        MUT_SPLICE_OVERWRITE,
+                                        MUT_SPLICE_OVERWRITE,
+                                        MUT_SPLICE_OVERWRITE,
+                                        MUT_SPLICE_OVERWRITE,
+                                        MUT_SPLICE_OVERWRITE,
+                                        MUT_SPLICE_OVERWRITE,
+                                        MUT_SPLICE_OVERWRITE,
+                                        MUT_SPLICE_OVERWRITE,
+                                        MUT_SPLICE_OVERWRITE,
+                                        MUT_SPLICE_INSERT,
+                                        MUT_SPLICE_INSERT,
+                                        MUT_SPLICE_INSERT,
+                                        MUT_SPLICE_INSERT,
+                                        MUT_SPLICE_INSERT,
+                                        MUT_SPLICE_INSERT,
+                                        MUT_SPLICE_INSERT,
+                                        MUT_SPLICE_INSERT,
+                                        MUT_SPLICE_INSERT,
+                                        MUT_SPLICE_INSERT,
+                                        MUT_SPLICE_INSERT,
+                                        MUT_SPLICE_INSERT,
+                                        MUT_SPLICE_INSERT,
+                                        MUT_SPLICE_INSERT};
+
   #define MUT_NORMAL_ARRAY_SIZE 77
 u32 normal_splice_array[MUT_NORMAL_ARRAY_SIZE] = {MUT_FLIPBIT,
                                                   MUT_FLIPBIT,
diff --git a/src/afl-fuzz-one.c b/src/afl-fuzz-one.c
index 32c05182..c6e49653 100644
--- a/src/afl-fuzz-one.c
+++ b/src/afl-fuzz-one.c
@@ -2101,27 +2101,17 @@ havoc_stage:
 
   */
 
-  rand_max = MUT_STRATEGY_ARRAY_SIZE;
-
   if (unlikely(afl->text_input)) {  // is text?
 
     if (likely(afl->fuzz_mode == 0)) {  // is exploration?
 
-      if (unlikely(afl->expand_havoc && afl->ready_for_splicing_count > 1)) {
-
-        mutation_array = full_splice_array;
-        rand_max = MUT_SPLICE_ARRAY_SIZE;
-
-      } else {
-
-        mutation_array = normal_splice_array;
-        rand_max = MUT_NORMAL_ARRAY_SIZE;
-
-      }
+      mutation_array = (unsigned int *)&text_array;
+      rand_max = MUT_TXT_ARRAY_SIZE;
 
     } else {  // is exploitation!
 
       mutation_array = (unsigned int *)&mutation_strategy_exploitation_text;
+      rand_max = MUT_STRATEGY_ARRAY_SIZE;
 
     }
 
@@ -2129,21 +2119,13 @@ havoc_stage:
 
     if (likely(afl->fuzz_mode == 0)) {  // is exploration?
 
-      if (unlikely(afl->expand_havoc && afl->ready_for_splicing_count > 1)) {
-
-        mutation_array = full_splice_array;
-        rand_max = MUT_SPLICE_ARRAY_SIZE;
-
-      } else {
-
-        mutation_array = normal_splice_array;
-        rand_max = MUT_NORMAL_ARRAY_SIZE;
-
-      }
+      mutation_array = (unsigned int *)&binary_array;
+      rand_max = MUT_BIN_ARRAY_SIZE;
 
     } else {  // is exploitation!
 
       mutation_array = (unsigned int *)&mutation_strategy_exploitation_binary;
+      rand_max = MUT_STRATEGY_ARRAY_SIZE;
 
     }
 
-- 
cgit 1.4.1


From 3e1d7941077b1457f702988063d6b9fdd9b80740 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Thu, 29 Jun 2023 16:57:20 +0200
Subject: update mutation strategy

---
 docs/Changelog.md       |  4 +++-
 include/afl-fuzz.h      | 59 +++++++++++++++++++++++++------------------------
 include/afl-mutations.h |  6 ++---
 src/afl-fuzz-one.c      | 56 +++++++++++++++++++++++++++-------------------
 src/afl-fuzz.c          | 26 +++++++++++++++++-----
 5 files changed, 90 insertions(+), 61 deletions(-)

(limited to 'include')

diff --git a/docs/Changelog.md b/docs/Changelog.md
index e6b90d3d..ad58e99e 100644
--- a/docs/Changelog.md
+++ b/docs/Changelog.md
@@ -8,7 +8,8 @@
     - new mutation engine: mutations that favor discovery more paths are
       prefered until no new finds for 10 minutes then switching to mutations
       that favor triggering crashes. Modes and switch time can be configured
-      with `-P`.
+      with `-P`. Also input mode for the target can be defined with `-a` to
+      be `text` or `binary` (defaults to `generic`)
     - new custom mutator that has the new afl++ engine (so it can easily
       incorporated into new custom mutators), and also comes with a standalone
       command line tool! See custom_mutators/aflpp/standalone/
@@ -23,6 +24,7 @@
       Thanks to @amykweon for spotting and fixing!
     - @toka fixed a bug in laf-intel signed integer comparison splitting,
       thanks a lot!!
+    - more LLVM compatability
   - frida_mode:
     - support for long form instrumentation on x86_x64 and arm64
 
diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index c6c45fbd..9da5cc03 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -505,36 +505,37 @@ typedef struct afl_state {
       is_main_node,                     /* if this is the main node         */
       is_secondary_node,                /* if this is a secondary instance  */
       pizza_is_served,                  /* pizza mode                       */
-      text_input,                       /* target wants text inputs         */
-      fuzz_mode, /* current mode: coverage/exploration or crash/exploitation */
+      input_mode,                       /* target wants text inputs         */
+      fuzz_mode,          /* coverage/exploration or crash/exploitation mode */
       schedule,                         /* Power schedule (default: EXPLORE)*/
-      havoc_max_mult, skip_deterministic, /* Skip deterministic stages?       */
-      use_splicing,                       /* Recombine input files?           */
-      non_instrumented_mode,              /* Run in non-instrumented mode?    */
-      score_changed,                      /* Scoring for favorites changed?   */
-      resuming_fuzz,                      /* Resuming an older fuzzing job?   */
-      timeout_given,                      /* Specific timeout given?          */
-      not_on_tty,                         /* stdout is not a tty              */
-      term_too_small,                     /* terminal dimensions too small    */
-      no_forkserver,                      /* Disable forkserver?              */
-      crash_mode,                         /* Crash mode! Yeah!                */
-      in_place_resume,                    /* Attempt in-place resume?         */
-      autoresume,                         /* Resume if afl->out_dir exists?   */
-      auto_changed,                       /* Auto-generated tokens changed?   */
-      no_cpu_meter_red,                   /* Feng shui on the status screen   */
-      no_arith,                           /* Skip most arithmetic ops         */
-      shuffle_queue,                      /* Shuffle input queue?             */
-      bitmap_changed,                     /* Time to update bitmap?           */
-      unicorn_mode,                       /* Running in Unicorn mode?         */
-      use_wine,                           /* Use WINE with QEMU mode          */
-      skip_requested,                     /* Skip request, via SIGUSR1        */
-      run_over10m,                        /* Run time over 10 minutes?        */
-      persistent_mode,                    /* Running in persistent mode?      */
-      deferred_mode,                      /* Deferred forkserver mode?        */
-      fixed_seed,                         /* do not reseed                    */
-      fast_cal,                           /* Try to calibrate faster?         */
-      disable_trim,                       /* Never trim in fuzz_one           */
-      shmem_testcase_mode,                /* If sharedmem testcases are used  */
+      havoc_max_mult,                   /* havoc multiplier                 */
+      skip_deterministic,               /* Skip deterministic stages?       */
+      use_splicing,                     /* Recombine input files?           */
+      non_instrumented_mode,            /* Run in non-instrumented mode?    */
+      score_changed,                    /* Scoring for favorites changed?   */
+      resuming_fuzz,                    /* Resuming an older fuzzing job?   */
+      timeout_given,                    /* Specific timeout given?          */
+      not_on_tty,                       /* stdout is not a tty              */
+      term_too_small,                   /* terminal dimensions too small    */
+      no_forkserver,                    /* Disable forkserver?              */
+      crash_mode,                       /* Crash mode! Yeah!                */
+      in_place_resume,                  /* Attempt in-place resume?         */
+      autoresume,                       /* Resume if afl->out_dir exists?   */
+      auto_changed,                     /* Auto-generated tokens changed?   */
+      no_cpu_meter_red,                 /* Feng shui on the status screen   */
+      no_arith,                         /* Skip most arithmetic ops         */
+      shuffle_queue,                    /* Shuffle input queue?             */
+      bitmap_changed,                   /* Time to update bitmap?           */
+      unicorn_mode,                     /* Running in Unicorn mode?         */
+      use_wine,                         /* Use WINE with QEMU mode          */
+      skip_requested,                   /* Skip request, via SIGUSR1        */
+      run_over10m,                      /* Run time over 10 minutes?        */
+      persistent_mode,                  /* Running in persistent mode?      */
+      deferred_mode,                    /* Deferred forkserver mode?        */
+      fixed_seed,                       /* do not reseed                    */
+      fast_cal,                         /* Try to calibrate faster?         */
+      disable_trim,                     /* Never trim in fuzz_one           */
+      shmem_testcase_mode,              /* If sharedmem testcases are used  */
       expand_havoc,                /* perform expensive havoc after no find */
       cycle_schedules,                  /* cycle power schedules?           */
       old_seed_selection,               /* use vanilla afl seed selection   */
diff --git a/include/afl-mutations.h b/include/afl-mutations.h
index cc4840c8..0a9bbbf4 100644
--- a/include/afl-mutations.h
+++ b/include/afl-mutations.h
@@ -14,14 +14,14 @@
    Parameters:
      afl_state_t *afl - the *afl state pointer
      u8 *buf - the input buffer to mutate which will be mutated into.
-           NOTE: must be able to contain a size of at least max_len (see below)!
+         NOTE: must be able to contain a size of at least max_len!! (see below)
      u32 len - the length of the input
      u32 steps - how many mutations to perform on the input
      bool is_text - is the target expecting text inputs
      bool is_exploration - mutate for exploration mode (instead of exploitation)
      splice_buf - a buffer from another corpus item to splice with.
-                  If NULL then no splicing
-     splice_len - the length of the splice buffer. If 0 then no splicing
+                  If NULL then no splicing is done (obviously).
+     splice_len - the length of the splice buffer. If 0 then no splicing.
      u32 max_len - the maximum size the mutated buffer may grow to
 */
 
diff --git a/src/afl-fuzz-one.c b/src/afl-fuzz-one.c
index c6e49653..0d3c29f2 100644
--- a/src/afl-fuzz-one.c
+++ b/src/afl-fuzz-one.c
@@ -2085,47 +2085,57 @@ havoc_stage:
   u32 *mutation_array;
   u32  stack_max, rand_max;  // stack_max_pow = afl->havoc_stack_pow2;
 
-  /*
+  switch (afl->input_mode) {
 
-  if (unlikely(afl->expand_havoc && afl->ready_for_splicing_count > 1)) {
+    case 1: {  // TEXT
 
-    mutation_array = full_splice_array;
-    rand_max = MUT_SPLICE_ARRAY_SIZE;
+      if (likely(afl->fuzz_mode == 0)) {  // is exploration?
+        mutation_array = (unsigned int *)&binary_array;
+        rand_max = MUT_BIN_ARRAY_SIZE;
 
-  } else {
+      } else {  // exploitation mode
 
-    mutation_array = normal_splice_array;
-    rand_max = MUT_NORMAL_ARRAY_SIZE;
+        mutation_array = (unsigned int *)&mutation_strategy_exploitation_text;
+        rand_max = MUT_STRATEGY_ARRAY_SIZE;
 
-  }
+      }
 
-  */
+      break;
 
-  if (unlikely(afl->text_input)) {  // is text?
+    }
 
-    if (likely(afl->fuzz_mode == 0)) {  // is exploration?
+    case 2: {  // BINARY
 
-      mutation_array = (unsigned int *)&text_array;
-      rand_max = MUT_TXT_ARRAY_SIZE;
+      if (likely(afl->fuzz_mode == 0)) {  // is exploration?
+        mutation_array = (unsigned int *)&mutation_strategy_exploration_binary;
+        rand_max = MUT_STRATEGY_ARRAY_SIZE;
 
-    } else {  // is exploitation!
+      } else {  // exploitation mode
 
-      mutation_array = (unsigned int *)&mutation_strategy_exploitation_text;
-      rand_max = MUT_STRATEGY_ARRAY_SIZE;
+        mutation_array = (unsigned int *)&mutation_strategy_exploitation_binary;
+        rand_max = MUT_STRATEGY_ARRAY_SIZE;
+
+      }
+
+      break;
 
     }
 
-  } else {  // is binary!
+    default: {  // DEFAULT/GENERIC
 
-    if (likely(afl->fuzz_mode == 0)) {  // is exploration?
+      if (likely(afl->fuzz_mode == 0)) {  // is exploration?
+        mutation_array = (unsigned int *)&binary_array;
+        rand_max = MUT_BIN_ARRAY_SIZE;
 
-      mutation_array = (unsigned int *)&binary_array;
-      rand_max = MUT_BIN_ARRAY_SIZE;
+      } else {  // exploitation mode
 
-    } else {  // is exploitation!
+        // this will need to be changed I guess
+        mutation_array = (unsigned int *)&mutation_strategy_exploration_text;
+        rand_max = MUT_STRATEGY_ARRAY_SIZE;
+
+      }
 
-      mutation_array = (unsigned int *)&mutation_strategy_exploitation_binary;
-      rand_max = MUT_STRATEGY_ARRAY_SIZE;
+      break;
 
     }
 
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index 79b05da7..ab7d6534 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -125,7 +125,8 @@ static void usage(u8 *argv0, int more_help) {
 
       "Required parameters:\n"
       "  -i dir        - input directory with test cases (or '-' to resume, "
-      "also see AFL_AUTORESUME)\n"
+      "also see \n"
+      "                  AFL_AUTORESUME)\n"
       "  -o dir        - output directory for fuzzer findings\n\n"
 
       "Execution control settings:\n"
@@ -164,8 +165,8 @@ static void usage(u8 *argv0, int more_help) {
       "\n"
 
       "Mutator settings:\n"
-      "  -a            - target expects ascii text input (prefer text "
-      "mutators)\n"
+      "  -a            - target input format, \"text\" or \"binary\" (default: "
+      "generic)\n"
       "  -g minlength  - set min length of generated fuzz input (default: 1)\n"
       "  -G maxlength  - set max length of generated fuzz input (default: "
       "%lu)\n"
@@ -506,13 +507,28 @@ int main(int argc, char **argv_orig, char **envp) {
 
   // still available: HjJkKqruvwz
   while ((opt = getopt(argc, argv,
-                       "+aAb:B:c:CdDe:E:f:F:g:G:hi:I:l:L:m:M:nNo:Op:P:QRs:S:t:"
+                       "+a:Ab:B:c:CdDe:E:f:F:g:G:hi:I:l:L:m:M:nNo:Op:P:QRs:S:t:"
                        "T:UV:WXx:YZ")) > 0) {
 
     switch (opt) {
 
       case 'a':
-        afl->text_input = 1;
+
+        if (!stricmp(optarg, "text") || !stricmp(optarg, "ascii") ||
+            !stricmp(optarg, "txt") || !stricmp(optarg, "asc")) {
+
+          afl->input_mode = 1;
+
+        } else if (!stricmp(optarg, "bin") || !stricmp(optarg, "binary")) {
+
+          afl->input_mode = 2;
+
+        } else {
+
+          FATAL("-a input mode needs to be \"text\" or \"binary\".");
+
+        }
+
         break;
 
       case 'P':
-- 
cgit 1.4.1


From d5184263350335b24daab635f0bcee455302f990 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Sun, 2 Jul 2023 14:50:18 +0200
Subject: no_ui: display time

---
 include/common.h   |  5 +++++
 src/afl-common.c   | 29 +++++++++++++++++++++++++++++
 src/afl-fuzz-one.c | 10 +++++++---
 3 files changed, 41 insertions(+), 3 deletions(-)

(limited to 'include')

diff --git a/include/common.h b/include/common.h
index 8d85d201..a9739a7d 100644
--- a/include/common.h
+++ b/include/common.h
@@ -115,6 +115,11 @@ u8 *stringify_mem_size(u8 *buf, size_t len, u64 val);
 
 u8 *stringify_time_diff(u8 *buf, size_t len, u64 cur_ms, u64 event_ms);
 
+/* Unsafe describe time delta as simple string.
+   Returns a pointer to buf for convenience. */
+
+u8 *u_simplestring_time_diff(u8 *buf, u64 cur_ms, u64 event_ms);
+
 /* Unsafe Describe integer. The buf sizes are not checked.
    This is unsafe but fast.
    Will return buf for convenience. */
diff --git a/src/afl-common.c b/src/afl-common.c
index 84ddefd8..3e1ec09d 100644
--- a/src/afl-common.c
+++ b/src/afl-common.c
@@ -1298,6 +1298,35 @@ u8 *u_stringify_time_diff(u8 *buf, u64 cur_ms, u64 event_ms) {
 
 }
 
+/* Unsafe describe time delta as simple string.
+   Returns a pointer to buf for convenience. */
+
+u8 *u_simplestring_time_diff(u8 *buf, u64 cur_ms, u64 event_ms) {
+
+  if (!event_ms) {
+
+    sprintf(buf, "00:00:00");
+
+  } else {
+
+    u64 delta;
+    s32 t_d, t_h, t_m, t_s;
+
+    delta = cur_ms - event_ms;
+
+    t_d = delta / 1000 / 60 / 60 / 24;
+    t_h = (delta / 1000 / 60 / 60) % 24;
+    t_m = (delta / 1000 / 60) % 60;
+    t_s = (delta / 1000) % 60;
+
+    sprintf(buf, "%d:%02d:%02d:%02d", t_d, t_h, t_m, t_s);
+
+  }
+
+  return buf;
+
+}
+
 /* Reads the map size from ENV */
 u32 get_map_size(void) {
 
diff --git a/src/afl-fuzz-one.c b/src/afl-fuzz-one.c
index 942381dd..e1ca44ab 100644
--- a/src/afl-fuzz-one.c
+++ b/src/afl-fuzz-one.c
@@ -399,20 +399,24 @@ u8 fuzz_one_original(afl_state_t *afl) {
 
 #endif                                                     /* ^IGNORE_FINDS */
 
-  if (unlikely(afl->not_on_tty)) {
+  if (likely(afl->not_on_tty)) {
 
+    u8 time_tmp[64];
+
+    u_simplestring_time_diff(time_tmp, afl->prev_run_time + get_cur_time(),
+                             afl->start_time);
     ACTF(
         "Fuzzing test case #%u (%u total, %llu crashes saved, state: %s, "
         "mode=%s, "
         "perf_score=%0.0f, weight=%0.0f, favorite=%u, was_fuzzed=%u, "
-        "exec_us=%llu, hits=%u, map=%u, ascii=%u)...",
+        "exec_us=%llu, hits=%u, map=%u, ascii=%u, run_time=%s)...",
         afl->current_entry, afl->queued_items, afl->saved_crashes,
         get_fuzzing_state(afl), afl->fuzz_mode ? "exploit" : "explore",
         afl->queue_cur->perf_score, afl->queue_cur->weight,
         afl->queue_cur->favored, afl->queue_cur->was_fuzzed,
         afl->queue_cur->exec_us,
         likely(afl->n_fuzz) ? afl->n_fuzz[afl->queue_cur->n_fuzz_entry] : 0,
-        afl->queue_cur->bitmap_size, afl->queue_cur->is_ascii);
+        afl->queue_cur->bitmap_size, afl->queue_cur->is_ascii, time_tmp);
     fflush(stdout);
 
   }
-- 
cgit 1.4.1


From 2a34e845072204b29200bf0e480d1d4f2201b332 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Wed, 12 Jul 2023 16:08:22 +0200
Subject: nits

---
 include/afl-fuzz.h       | 2 +-
 include/android-ashmem.h | 4 +++-
 src/afl-ld-lto.c         | 4 +++-
 3 files changed, 7 insertions(+), 3 deletions(-)

(limited to 'include')

diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index 9da5cc03..27668da0 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -31,7 +31,7 @@
 #define MESSAGES_TO_STDOUT
 
 #ifndef _GNU_SOURCE
-  #define _GNU_SOURCE 1
+  #define _GNU_SOURCE
 #endif
 #ifndef _FILE_OFFSET_BITS
   #define _FILE_OFFSET_BITS 64
diff --git a/include/android-ashmem.h b/include/android-ashmem.h
index 1bfd3220..065c213b 100644
--- a/include/android-ashmem.h
+++ b/include/android-ashmem.h
@@ -2,7 +2,9 @@
   #ifndef _ANDROID_ASHMEM_H
     #define _ANDROID_ASHMEM_H
 
-    #define _GNU_SOURCE
+    #ifndef _GNU_SOURCE
+      #define _GNU_SOURCE
+    #endif
     #include <sys/syscall.h>
     #include <unistd.h>
     #include <fcntl.h>
diff --git a/src/afl-ld-lto.c b/src/afl-ld-lto.c
index b306c8d5..b1e6c848 100644
--- a/src/afl-ld-lto.c
+++ b/src/afl-ld-lto.c
@@ -23,7 +23,9 @@
 */
 
 #define AFL_MAIN
-#define _GNU_SOURCE
+#ifndef _GNU_SOURCE
+  #define _GNU_SOURCE
+#endif
 
 #include "config.h"
 #include "types.h"
-- 
cgit 1.4.1


From 127c345161769c513275fed9d64de12536ee979d Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Thu, 13 Jul 2023 14:26:26 +0200
Subject: nts

---
 frida_mode/src/instrument/instrument_arm32.c   | 2 +-
 frida_mode/src/instrument/instrument_arm64.c   | 2 +-
 include/config.h                               | 8 ++++++++
 instrumentation/SanitizerCoveragePCGUARD.so.cc | 1 +
 instrumentation/afl-compiler-rt.o.c            | 6 ------
 5 files changed, 11 insertions(+), 8 deletions(-)

(limited to 'include')

diff --git a/frida_mode/src/instrument/instrument_arm32.c b/frida_mode/src/instrument/instrument_arm32.c
index 2e123247..c1e3f187 100644
--- a/frida_mode/src/instrument/instrument_arm32.c
+++ b/frida_mode/src/instrument/instrument_arm32.c
@@ -15,7 +15,7 @@ static GHashTable *coverage_blocks = NULL;
 
 extern __thread guint64 instrument_previous_pc;
 
-__attribute__((aligned(0x1000))) static guint8 area_ptr_dummy[2UL << 20];
+__attribute__((aligned(0x1000))) static guint8 area_ptr_dummy[MAP_INITIAL_SIZE];
 
   #pragma pack(push, 1)
 typedef struct {
diff --git a/frida_mode/src/instrument/instrument_arm64.c b/frida_mode/src/instrument/instrument_arm64.c
index a8d30dc1..2256f941 100644
--- a/frida_mode/src/instrument/instrument_arm64.c
+++ b/frida_mode/src/instrument/instrument_arm64.c
@@ -22,7 +22,7 @@ gboolean           instrument_cache_enabled = FALSE;
 gsize              instrument_cache_size = 0;
 static GHashTable *coverage_blocks = NULL;
 
-__attribute__((aligned(0x1000))) static guint8 area_ptr_dummy[2UL << 20];
+__attribute__((aligned(0x1000))) static guint8 area_ptr_dummy[MAP_INITIAL_SIZE];
 
   #pragma pack(push, 1)
 typedef struct {
diff --git a/include/config.h b/include/config.h
index 8585041e..16f4b613 100644
--- a/include/config.h
+++ b/include/config.h
@@ -446,7 +446,15 @@
    after changing this - otherwise, SEGVs may ensue. */
 
 #define MAP_SIZE_POW2 16
+
+/* Do not change this unless you really know what you are doing. */
+
 #define MAP_SIZE (1U << MAP_SIZE_POW2)
+#if MAP_SIZE <= 65536
+  #define MAP_INITIAL_SIZE (2 << 20)  // = 2097152
+#else
+  #define MAP_INITIAL_SIZE MAP_SIZE
+#endif
 
 /* Maximum allocator request size (keep well under INT_MAX): */
 
diff --git a/instrumentation/SanitizerCoveragePCGUARD.so.cc b/instrumentation/SanitizerCoveragePCGUARD.so.cc
index 7d614f43..98c5973c 100644
--- a/instrumentation/SanitizerCoveragePCGUARD.so.cc
+++ b/instrumentation/SanitizerCoveragePCGUARD.so.cc
@@ -227,6 +227,7 @@ llvmGetPassPluginInfo() {
 
 PreservedAnalyses ModuleSanitizerCoverageAFL::run(Module                &M,
                                                   ModuleAnalysisManager &MAM) {
+
   ModuleSanitizerCoverageAFL ModuleSancov(Options);
   auto &FAM = MAM.getResult<FunctionAnalysisManagerModuleProxy>(M).getManager();
   auto  DTCallback = [&FAM](Function &F) -> const DominatorTree  *{
diff --git a/instrumentation/afl-compiler-rt.o.c b/instrumentation/afl-compiler-rt.o.c
index 3f8b519b..dd9aae77 100644
--- a/instrumentation/afl-compiler-rt.o.c
+++ b/instrumentation/afl-compiler-rt.o.c
@@ -87,12 +87,6 @@
    is used for instrumentation output before __afl_map_shm() has a chance to
    run. It will end up as .comm, so it shouldn't be too wasteful. */
 
-#if MAP_SIZE <= 65536
-  #define MAP_INITIAL_SIZE 2097152
-#else
-  #define MAP_INITIAL_SIZE MAP_SIZE
-#endif
-
 #if defined(__HAIKU__)
 extern ssize_t _kern_write(int fd, off_t pos, const void *buffer,
                            size_t bufferSize);
-- 
cgit 1.4.1


From 2b8e528a3b5f44df590b8f727983d142857d0433 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Thu, 13 Jul 2023 17:12:55 +0200
Subject: interesting32_float

---
 include/config.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

(limited to 'include')

diff --git a/include/config.h b/include/config.h
index 16f4b613..7c29a674 100644
--- a/include/config.h
+++ b/include/config.h
@@ -360,9 +360,10 @@
       65535,      /* Overflow unsig 16-bit when incremented  */ \
       65536,      /* Overflow unsig 16 bit                   */ \
       100663045,  /* Large positive number (endian-agnostic) */ \
+      2139095040, /* float infinite                          */ \
       2147483647                 /* Overflow signed 32-bit when incremented */
 
-#define INTERESTING_32_LEN 8
+#define INTERESTING_32_LEN 9
 
 /***********************************************************
  *                                                         *
-- 
cgit 1.4.1


From 5f813bbb86e1c9e2480669c44501e9780043728c Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Fri, 21 Jul 2023 18:02:30 +0200
Subject: improve cmplog level 3

---
 docs/Changelog.md       |   1 +
 include/afl-fuzz.h      |   3 +-
 include/config.h        |   8 +--
 src/afl-fuzz-redqueen.c | 171 +++++++++++++++++++++++++-----------------------
 src/afl-fuzz.c          |   7 +-
 5 files changed, 100 insertions(+), 90 deletions(-)

(limited to 'include')

diff --git a/docs/Changelog.md b/docs/Changelog.md
index d61ce8ec..75167172 100644
--- a/docs/Changelog.md
+++ b/docs/Changelog.md
@@ -15,6 +15,7 @@
       command line tool! See custom_mutators/aflpp/standalone/
     - display the state of the fuzzing run in the UI :-)
     - fix timeout setting if '+' is used or a session is restarted
+    - -c X option to enable base64 transformation solving
   - afl-cmin/afl-cmin.bash:
     - fixed a bug inherited from vanilla AFL where a coverage of
       map[123] = 11 would be the same as map[1123] = 1
diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index 27668da0..e114b0fc 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -674,7 +674,8 @@ typedef struct afl_state {
   u32 cmplog_max_filesize;
   u32 cmplog_lvl;
   u32 colorize_success;
-  u8  cmplog_enable_arith, cmplog_enable_transform, cmplog_random_colorization;
+  u8  cmplog_enable_arith, cmplog_enable_transform,
+      cmplog_enable_xtreme_transform, cmplog_random_colorization;
 
   struct afl_pass_stat *pass_stats;
   struct cmp_map       *orig_cmp_map;
diff --git a/include/config.h b/include/config.h
index 7c29a674..df545583 100644
--- a/include/config.h
+++ b/include/config.h
@@ -60,10 +60,6 @@
  *
  */
 
-/* if TRANSFORM is enabled with '-l T', this additionally enables base64
-   encoding/decoding */
-// #define CMPLOG_SOLVE_TRANSFORM_BASE64
-
 /* If a redqueen pass finds more than one solution, try to combine them? */
 #define CMPLOG_COMBINE
 
@@ -71,10 +67,10 @@
 #define CMPLOG_CORPUS_PERCENT 5U
 
 /* Number of potential positions from which we decide if cmplog becomes
-   useless, default 8096 */
+   useless, default 12288 */
 #define CMPLOG_POSITIONS_MAX (12 * 1024)
 
-/* Maximum allowed fails per CMP value. Default: 128 */
+/* Maximum allowed fails per CMP value. Default: 96 */
 #define CMPLOG_FAIL_MAX 96
 
 /* -------------------------------------*/
diff --git a/src/afl-fuzz-redqueen.c b/src/afl-fuzz-redqueen.c
index 73e188e7..5a1f512d 100644
--- a/src/afl-fuzz-redqueen.c
+++ b/src/afl-fuzz-redqueen.c
@@ -571,7 +571,6 @@ static u8 its_fuzz(afl_state_t *afl, u8 *buf, u32 len, u8 *status) {
 
 }
 
-// #ifdef CMPLOG_SOLVE_TRANSFORM
 static int strntoll(const char *str, size_t sz, char **end, int base,
                     long long *out) {
 
@@ -656,7 +655,6 @@ static int is_hex(const char *str) {
 
 }
 
-#ifdef CMPLOG_SOLVE_TRANSFORM_BASE64
 // tests 4 bytes at location
 static int is_base64(const char *str) {
 
@@ -769,10 +767,6 @@ static void to_base64(u8 *src, u8 *dst, u32 dst_len) {
 
 }
 
-#endif
-
-// #endif
-
 static u8 cmp_extend_encoding(afl_state_t *afl, struct cmp_header *h,
                               u64 pattern, u64 repl, u64 o_pattern,
                               u64 changed_val, u8 attr, u32 idx, u32 taint_len,
@@ -797,42 +791,54 @@ static u8 cmp_extend_encoding(afl_state_t *afl, struct cmp_header *h,
 
   }
 
-  // fprintf(stderr,
-  //         "Encode: %llx->%llx into %llx(<-%llx) at idx=%u "
-  //         "taint_len=%u shape=%u attr=%u\n",
-  //         o_pattern, pattern, repl, changed_val, idx, taint_len,
-  //         hshape, attr);
+  /*
+  fprintf(stderr,
+          "Encode: %llx->%llx into %llx(<-%llx) at idx=%u "
+          "taint_len=%u shape=%u attr=%u\n",
+          o_pattern, pattern, repl, changed_val, idx, taint_len,
+          hshape, attr);
+  */
 
-  // #ifdef CMPLOG_SOLVE_TRANSFORM
   //  reverse atoi()/strnu?toll() is expensive, so we only to it in lvl 3
   if (afl->cmplog_enable_transform && (lvl & LVL3)) {
 
     u8                *endptr;
     u8                 use_num = 0, use_unum = 0;
-    unsigned long long unum;
-    long long          num;
+    unsigned long long unum = 0;
+    long long          num = 0;
+
+    // if (afl->queue_cur->is_ascii) {
+
+    // we first check if our input are ascii numbers that are transformed to
+    // an integer and used for comparison:
 
-    if (afl->queue_cur->is_ascii) {
+    endptr = buf_8;
+    if (strntoll(buf_8, len - idx, (char **)&endptr, 0, &num)) {
 
-      endptr = buf_8;
-      if (strntoll(buf_8, len - idx, (char **)&endptr, 0, &num)) {
+      if (!strntoull(buf_8, len - idx, (char **)&endptr, 0, &unum)) {
 
-        if (!strntoull(buf_8, len - idx, (char **)&endptr, 0, &unum))
-          use_unum = 1;
+        use_unum = 1;
 
-      } else
+      }
+
+    } else {
 
-        use_num = 1;
+      use_num = 1;
 
     }
 
+    //}
+
 #ifdef _DEBUG
     if (idx == 0)
-      fprintf(stderr, "ASCII is=%u use_num=%u use_unum=%u idx=%u %llx==%llx\n",
-              afl->queue_cur->is_ascii, use_num, use_unum, idx, num, pattern);
+      fprintf(stderr,
+              "ASCII is=%u use_num=%u>%lld use_unum=%u>%llu idx=%u "
+              "pattern=0x%llx\n",
+              afl->queue_cur->is_ascii, use_num, num, use_unum, unum, idx,
+              pattern);
 #endif
 
-    // num is likely not pattern as atoi("AAA") will be zero...
+    // atoi("AAA") == 0 so !num means we have to investigate
     if (use_num && ((u64)num == pattern || !num)) {
 
       u8     tmp_buf[32];
@@ -961,10 +967,12 @@ static u8 cmp_extend_encoding(afl_state_t *afl, struct cmp_header *h,
       // test for arithmetic, eg. "if ((user_val - 0x1111) == 0x1234) ..."
       s64 diff = pattern - b_val;
       s64 o_diff = o_pattern - o_b_val;
-      /* fprintf(stderr, "DIFF1 idx=%03u shape=%02u %llx-%llx=%lx\n", idx,
-                 hshape, o_pattern, o_b_val, o_diff);
-         fprintf(stderr, "DIFF1 %016llx %llx-%llx=%lx\n", repl, pattern,
-                 b_val, diff); */
+      /*
+             fprintf(stderr, "DIFF1 idx=%03u shape=%02u %llx-%llx=%lx\n", idx,
+                       hshape, o_pattern, o_b_val, o_diff);
+               fprintf(stderr, "DIFF1 %016llx %llx-%llx=%lx\n", repl, pattern,
+                       b_val, diff);
+      */
       if (diff == o_diff && diff) {
 
         // this could be an arithmetic transformation
@@ -1275,7 +1283,6 @@ static u8 cmp_extend_encoding(afl_state_t *afl, struct cmp_header *h,
   //       16 = modified float, 32 = modified integer (modified = wont match
   //                                                   in original buffer)
 
-  // #ifdef CMPLOG_SOLVE_ARITHMETIC
   if (!afl->cmplog_enable_arith || lvl < LVL3 || attr == IS_TRANSFORM) {
 
     return 0;
@@ -2009,8 +2016,12 @@ static u8 rtn_extend_encoding(afl_state_t *afl, u8 entry,
   its_len = MIN(its_len, taint_len);
   u32 saved_its_len = its_len;
 
+  if (its_len <= 1) { return 0; }
+
   if (lvl & LVL3) {
 
+    if (memcmp(changed_val, repl, its_len) != 0) { return 0; }
+
     u32 max_to = MIN(4U, idx);
     if (!(lvl & LVL1) && max_to) { from = 1; }
     to = max_to;
@@ -2089,9 +2100,7 @@ static u8 rtn_extend_encoding(afl_state_t *afl, u8 entry,
   if (afl->cmplog_enable_transform && (lvl & LVL3)) {
 
     u32 toupper = 0, tolower = 0, xor = 0, arith = 0, tohex = 0, fromhex = 0;
-#ifdef CMPLOG_SOLVE_TRANSFORM_BASE64
     u32 tob64 = 0, fromb64 = 0;
-#endif
     u32 from_0 = 0, from_x = 0, from_X = 0, from_slash = 0, from_up = 0;
     u32 to_0 = 0, to_x = 0, to_slash = 0, to_up = 0;
     u8  xor_val[32], arith_val[32], tmp[48];
@@ -2144,7 +2153,8 @@ static u8 rtn_extend_encoding(afl_state_t *afl, u8 entry,
 
       }
 
-      if (i < 16 && is_hex(repl + (i << 1))) {
+      if (afl->cmplog_enable_xtreme_transform && i < 16 &&
+          is_hex(repl + (i << 1))) {
 
         ++tohex;
 
@@ -2163,7 +2173,7 @@ static u8 rtn_extend_encoding(afl_state_t *afl, u8 entry,
 
       }
 
-      if ((i % 2)) {
+      if (afl->cmplog_enable_xtreme_transform && (i % 2)) {
 
         if (len > idx + i + 1 && is_hex(orig_buf + idx + i)) {
 
@@ -2187,20 +2197,21 @@ static u8 rtn_extend_encoding(afl_state_t *afl, u8 entry,
 
       }
 
-#ifdef CMPLOG_SOLVE_TRANSFORM_BASE64
-      if (i % 3 == 2 && i < 24) {
+      if (afl->cmplog_enable_xtreme_transform) {
 
-        if (is_base64(repl + ((i / 3) << 2))) tob64 += 3;
+        if (i % 3 == 2 && i < 24) {
 
-      }
+          if (is_base64(repl + ((i / 3) << 2))) tob64 += 3;
+
+        }
 
-      if (i % 4 == 3 && i < 24) {
+        if (i % 4 == 3 && i < 24) {
 
-        if (is_base64(orig_buf + idx + i - 3)) fromb64 += 4;
+          if (is_base64(orig_buf + idx + i - 3)) fromb64 += 4;
 
-      }
+        }
 
-#endif
+      }
 
       if ((o_pattern[i] ^ orig_buf[idx + i]) == xor_val[i] && xor_val[i]) {
 
@@ -2229,45 +2240,50 @@ static u8 rtn_extend_encoding(afl_state_t *afl, u8 entry,
       }
 
 #ifdef _DEBUG
+      fprintf(stderr, "RTN %s %s %s %s\n", buf, pattern, orig_buf, o_pattern);
       fprintf(stderr,
-              "RTN idx=%u loop=%u xor=%u arith=%u tolower=%u toupper=%u "
+              "RTN idx=%u len=%u loop=%u xor=%u arith=%u tolower=%u toupper=%u "
               "tohex=%u fromhex=%u to_0=%u to_slash=%u to_x=%u "
               "from_0=%u from_slash=%u from_x=%u\n",
-              idx, i, xor, arith, tolower, toupper, tohex, fromhex, to_0,
-              to_slash, to_x, from_0, from_slash, from_x);
-  #ifdef CMPLOG_SOLVE_TRANSFORM_BASE64
-      fprintf(stderr, "RTN idx=%u loop=%u tob64=%u from64=%u\n", tob64,
-              fromb64);
-  #endif
+              idx, its_len, i, xor, arith, tolower, toupper, tohex, fromhex,
+              to_0, to_slash, to_x, from_0, from_slash, from_x);
+      if (afl->cmplog_enable_xtreme_transform) {
+
+        fprintf(stderr, "RTN idx=%u loop=%u tob64=%u from64=%u\n", idx, i,
+                tob64, fromb64);
+
+      }
+
 #endif
 
-#ifdef CMPLOG_SOLVE_TRANSFORM_BASE64
-      // input is base64 and converted to binary? convert repl to base64!
-      if ((i % 4) == 3 && i < 24 && fromb64 > i) {
+      if (afl->cmplog_enable_xtreme_transform) {
 
-        to_base64(repl, tmp, i + 1);
-        memcpy(buf + idx, tmp, i + 1);
-        if (unlikely(its_fuzz(afl, buf, len, status))) { return 1; }
-        // fprintf(stderr, "RTN ATTEMPT fromb64 %u result %u\n", fromb64,
-        // *status);
+        // input is base64 and converted to binary? convert repl to base64!
+        if ((i % 4) == 3 && i < 24 && fromb64 > i) {
 
-      }
+          to_base64(repl, tmp, i + 1);
+          memcpy(buf + idx, tmp, i + 1);
+          if (unlikely(its_fuzz(afl, buf, len, status))) { return 1; }
+          // fprintf(stderr, "RTN ATTEMPT fromb64 %u result %u\n", fromb64,
+          // *status);
+
+        }
 
-      // input is converted to base64? decode repl with base64!
-      if ((i % 3) == 2 && i < 24 && tob64 > i) {
+        // input is converted to base64? decode repl with base64!
+        if ((i % 3) == 2 && i < 24 && tob64 > i) {
 
-        u32 olen = from_base64(repl, tmp, i + 1);
-        memcpy(buf + idx, tmp, olen);
-        if (unlikely(its_fuzz(afl, buf, len, status))) { return 1; }
-        // fprintf(stderr, "RTN ATTEMPT tob64 %u idx=%u result %u\n", tob64,
-        // idx, *status);
+          u32 olen = from_base64(repl, tmp, i + 1);
+          memcpy(buf + idx, tmp, olen);
+          if (unlikely(its_fuzz(afl, buf, len, status))) { return 1; }
+          // fprintf(stderr, "RTN ATTEMPT tob64 %u idx=%u result %u\n", tob64,
+          // idx, *status);
 
-      }
+        }
 
-#endif
+      }
 
       // input is converted to hex? convert repl to binary!
-      if (i < 16 && tohex > i) {
+      if (afl->cmplog_enable_xtreme_transform && i < 16 && tohex > i) {
 
         u32 off;
         if (to_slash + to_x + to_0 == 2) {
@@ -2292,8 +2308,8 @@ static u8 rtn_extend_encoding(afl_state_t *afl, u8 entry,
       }
 
       // input is hex and converted to binary? convert repl to hex!
-      if (i && (i % 2) && i < 16 && fromhex &&
-          fromhex + from_slash + from_x + from_0 > i) {
+      if (afl->cmplog_enable_xtreme_transform && i && (i % 2) && i < 16 &&
+          fromhex && fromhex + from_slash + from_x + from_0 > i) {
 
         u8 off = 0;
         if (from_slash && from_x) {
@@ -2401,11 +2417,9 @@ static u8 rtn_extend_encoding(afl_state_t *afl, u8 entry,
 
       if ((i >= 7 &&
            (i >= xor&&i >= arith &&i >= tolower &&i >= toupper &&i > tohex &&i >
-                (fromhex + from_0 + from_x + from_slash + 1)
-#ifdef CMPLOG_SOLVE_TRANSFORM_BASE64
-            && i > tob64 + 3 && i > fromb64 + 4
-#endif
-            )) ||
+                (fromhex + from_0 + from_x + from_slash + 1) &&
+            (afl->cmplog_enable_xtreme_transform && i > tob64 + 3 &&
+             i > fromb64 + 4))) ||
           repl[i] != changed_val[i] || *status == 1) {
 
         break;
@@ -2418,8 +2432,6 @@ static u8 rtn_extend_encoding(afl_state_t *afl, u8 entry,
 
   }
 
-  // #endif
-
   return 0;
 
 }
@@ -2818,12 +2830,7 @@ u8 input_to_state_stage(afl_state_t *afl, u8 *orig_buf, u8 *buf, u32 len) {
 
       }
 
-    } else if ((lvl & LVL1)
-
-               // #ifdef CMPLOG_SOLVE_TRANSFORM
-               || ((lvl & LVL3) && afl->cmplog_enable_transform)
-               // #endif
-    ) {
+    } else if ((lvl & LVL1) || ((lvl & LVL3) && afl->cmplog_enable_transform)) {
 
       if (unlikely(rtn_fuzz(afl, k, orig_buf, buf, cbuf, len, lvl, taint))) {
 
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index d8a88f00..21a8915c 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -185,7 +185,8 @@ static void usage(u8 *argv0, int more_help) {
       "                  1=small files, 2=larger files (default), 3=all "
       "files,\n"
       "                  A=arithmetic solving, T=transformational solving,\n"
-      "                  R=random colorization bytes.\n\n"
+      "                  X=extreme transform solving, R=random colorization "
+      "bytes.\n\n"
       "Fuzzing behavior settings:\n"
       "  -Z            - sequential queue selection instead of weighted "
       "random\n"
@@ -1120,6 +1121,10 @@ int main(int argc, char **argv_orig, char **envp) {
             case 'T':
               afl->cmplog_enable_transform = 1;
               break;
+            case 'x':
+            case 'X':
+              afl->cmplog_enable_xtreme_transform = 1;
+              break;
             case 'r':
             case 'R':
               afl->cmplog_random_colorization = 1;
-- 
cgit 1.4.1


From 168ade3b70077ec6a24df9fc594e3b8c1db89bd6 Mon Sep 17 00:00:00 2001
From: mark0 <59284400+mark0-cn@users.noreply.github.com>
Date: Mon, 31 Jul 2023 05:51:24 +0800
Subject: Fix format specifiers (#1818)

* Update afl-mutations.h

Fix bug: compilation cannot pass when DEBUG macro is enabled

* Update afl-fuzz-one.c

Fix bug: compilation cannot pass when DEBUG macro is enabled
---
 include/afl-mutations.h | 2 +-
 src/afl-fuzz-one.c      | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

(limited to 'include')

diff --git a/include/afl-mutations.h b/include/afl-mutations.h
index 0a9bbbf4..5dde4473 100644
--- a/include/afl-mutations.h
+++ b/include/afl-mutations.h
@@ -2456,7 +2456,7 @@ inline u32 afl_mutate(afl_state_t *afl, u8 *buf, u32 len, u32 steps,
         }
 
         char buf[20];
-        snprintf(buf, sizeof(buf), "%ld", val);
+        snprintf(buf, sizeof(buf), "%lld", val);
         u32 old_len = off2 - off;
         u32 new_len = strlen(buf);
 
diff --git a/src/afl-fuzz-one.c b/src/afl-fuzz-one.c
index 8ee50bbf..f4ae7bfd 100644
--- a/src/afl-fuzz-one.c
+++ b/src/afl-fuzz-one.c
@@ -2995,7 +2995,7 @@ havoc_stage:
           // fprintf(stderr, "val: %u-%u = %ld\n", off, off2, val);
 
           char buf[20];
-          snprintf(buf, sizeof(buf), "%ld", val);
+          snprintf(buf, sizeof(buf), "%lld", val);
 
           // fprintf(stderr, "BEFORE: %s\n", out_buf);
 
-- 
cgit 1.4.1


From d0782a7f03a23f8323772d189e5b66a4eb086afd Mon Sep 17 00:00:00 2001
From: Dominik Maier <dmnk@google.com>
Date: Sun, 30 Jul 2023 21:59:57 +0000
Subject: Various fixes for warnings, extends #1818

---
 include/afl-mutations.h | 13 +++++++------
 src/afl-fuzz-one.c      |  2 +-
 src/afl-fuzz-redqueen.c |  2 +-
 unicorn_mode/unicornafl |  2 +-
 4 files changed, 10 insertions(+), 9 deletions(-)

(limited to 'include')

diff --git a/include/afl-mutations.h b/include/afl-mutations.h
index 5dde4473..1806790e 100644
--- a/include/afl-mutations.h
+++ b/include/afl-mutations.h
@@ -25,13 +25,14 @@
      u32 max_len - the maximum size the mutated buffer may grow to
 */
 
-#ifndef _ANDROID_ASHMEM_H
-  #define AFL_MUTATIONS_H
+#ifndef AFL_MUTATIONS_H
+#define AFL_MUTATIONS_H
 
-  #include <stdbool.h>
-  #include "afl-fuzz.h"
+#include <stdbool.h>
+#include <inttypes.h>
+#include "afl-fuzz.h"
 
-  #define MUT_STRATEGY_ARRAY_SIZE 256
+#define MUT_STRATEGY_ARRAY_SIZE 256
 
 enum {
 
@@ -2456,7 +2457,7 @@ inline u32 afl_mutate(afl_state_t *afl, u8 *buf, u32 len, u32 steps,
         }
 
         char buf[20];
-        snprintf(buf, sizeof(buf), "%lld", val);
+        snprintf(buf, sizeof(buf), "%" PRId64, val);
         u32 old_len = off2 - off;
         u32 new_len = strlen(buf);
 
diff --git a/src/afl-fuzz-one.c b/src/afl-fuzz-one.c
index f4ae7bfd..2ad4697e 100644
--- a/src/afl-fuzz-one.c
+++ b/src/afl-fuzz-one.c
@@ -2995,7 +2995,7 @@ havoc_stage:
           // fprintf(stderr, "val: %u-%u = %ld\n", off, off2, val);
 
           char buf[20];
-          snprintf(buf, sizeof(buf), "%lld", val);
+          snprintf(buf, sizeof(buf), "%" PRId64, val);
 
           // fprintf(stderr, "BEFORE: %s\n", out_buf);
 
diff --git a/src/afl-fuzz-redqueen.c b/src/afl-fuzz-redqueen.c
index 8a652a9f..54bf4e32 100644
--- a/src/afl-fuzz-redqueen.c
+++ b/src/afl-fuzz-redqueen.c
@@ -2665,7 +2665,7 @@ static u8 rtn_fuzz(afl_state_t *afl, u32 key, u8 *orig_buf, u8 *buf, u8 *cbuf,
 
     }
 
-  rtn_fuzz_next_iter:
+  // rtn_fuzz_next_iter:
     afl->stage_cur++;
 
   }
diff --git a/unicorn_mode/unicornafl b/unicorn_mode/unicornafl
index f2cede37..2df75f3e 160000
--- a/unicorn_mode/unicornafl
+++ b/unicorn_mode/unicornafl
@@ -1 +1 @@
-Subproject commit f2cede37a75bbd4a9b9438f0277727b5d4620572
+Subproject commit 2df75f3e1045367cab95fe3471191b38c1a9f79e
-- 
cgit 1.4.1


From 1429c9724efb62e5ac90ec27d93a64c28632ba5d Mon Sep 17 00:00:00 2001
From: Junwha <qbit@unist.ac.kr>
Date: Wed, 2 Aug 2023 02:59:07 +0900
Subject: Add option for treating crashing input as new crash

Signed-off-by: Junwha Hong <qbit@unist.ac.kr>
---
 include/afl-fuzz.h  |  3 +-
 include/envs.h      |  1 +
 src/afl-fuzz-init.c | 99 +++++++++++++++++++++++++++++++++++++++++++----------
 src/afl-fuzz.c      |  3 +-
 4 files changed, 86 insertions(+), 20 deletions(-)

(limited to 'include')

diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index e114b0fc..7bedc98f 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -539,7 +539,8 @@ typedef struct afl_state {
       expand_havoc,                /* perform expensive havoc after no find */
       cycle_schedules,                  /* cycle power schedules?           */
       old_seed_selection,               /* use vanilla afl seed selection   */
-      reinit_table;                     /* reinit the queue weight table    */
+      reinit_table,                     /* reinit the queue weight table    */
+      crashing_seeds_as_new_crash; /* treat crashing seeds as normal corpus */
 
   u8 *virgin_bits,                      /* Regions yet untouched by fuzzing */
       *virgin_tmout,                    /* Bits we haven't seen in tmouts   */
diff --git a/include/envs.h b/include/envs.h
index edfd06e4..e396acd2 100644
--- a/include/envs.h
+++ b/include/envs.h
@@ -35,6 +35,7 @@ static char *afl_environment_variables[] = {
     "AFL_COMPCOV_BINNAME",
     "AFL_COMPCOV_LEVEL",
     "AFL_CRASH_EXITCODE",
+    "AFL_CRASHING_SEEDS_AS_NEW_CRASH",
     "AFL_CUSTOM_MUTATOR_LIBRARY",
     "AFL_CUSTOM_MUTATOR_ONLY",
     "AFL_CUSTOM_INFO_PROGRAM",
diff --git a/src/afl-fuzz-init.c b/src/afl-fuzz-init.c
index 24fd7077..6b7f3036 100644
--- a/src/afl-fuzz-init.c
+++ b/src/afl-fuzz-init.c
@@ -1056,6 +1056,13 @@ void perform_dry_run(afl_state_t *afl) {
               "skipping",
               fn, (int)(s8)afl->fsrv.crash_exitcode);
 
+        } else if (afl->crashing_seeds_as_new_crash) {
+          
+          WARNF(
+              "Test case '%s' results in a crash,"
+              "as AFL_CRASHING_SEEDS_AS_NEW_CRASH is set, "
+              "saving as a crash", fn);
+
         } else {
 
           WARNF("Test case '%s' results in a crash, skipping", fn);
@@ -1078,38 +1085,94 @@ void perform_dry_run(afl_state_t *afl) {
 
         }
 
-        q->disabled = 1;
-        q->perf_score = 0;
+        /* Crashing corpus will regrad as normal, and categorized as new crash at fuzzing  */
+        if (afl->crashing_seeds_as_new_crash) {
+          
+          ++afl->total_crashes;
 
-        u32 i = 0;
-        while (unlikely(i < afl->queued_items && afl->queue_buf[i] &&
-                        afl->queue_buf[i]->disabled)) {
+          if (likely(!afl->non_instrumented_mode)) {
 
-          ++i;
+            classify_counts(&afl->fsrv);
+          
+            simplify_trace(afl, afl->fsrv.trace_bits);
 
-        }
+            if (!has_new_bits(afl, afl->virgin_crash)) { break; }
+
+          }
+
+          
+          if (unlikely(!afl->saved_crashes) &&
+              (afl->afl_env.afl_no_crash_readme != 1)) {
+
+            write_crash_readme(afl);
+
+          }
+
+          u8  crash_fn[PATH_MAX];
+          u8 *use_name = strstr(q->fname, ",orig:");
+
+          afl->stage_name = "dry_run";
+          afl->stage_short = "dry_run";
+
+    #ifndef SIMPLE_FILES
+
+          snprintf(crash_fn, PATH_MAX, "%s/crashes/id:%06llu,sig:%02u,%s%s", afl->out_dir,
+                  afl->saved_crashes, afl->fsrv.last_kill_signal,
+                  describe_op(afl, 0, NAME_MAX - strlen("id:000000,sig:00,") - strlen(use_name)), use_name);
+
+    #else
+
+          snprintf(crash_fn, PATH_MAX, "%s/crashes/id_%06llu_%02u", afl->out_dir,
+                  afl->saved_crashes, afl->fsrv.last_kill_signal);
+
+    #endif  
 
-        if (i < afl->queued_items && afl->queue_buf[i]) {
+          ++afl->saved_crashes;
 
-          afl->queue = afl->queue_buf[i];
+          fd = open(crash_fn, O_WRONLY | O_CREAT | O_EXCL, DEFAULT_PERMISSION);
+          if (unlikely(fd < 0)) { PFATAL("Unable to create '%s'", crash_fn); }
+          ck_write(fd, use_mem, read_len, crash_fn);
+          close(fd);
+
+          afl->last_crash_time = get_cur_time();
+          afl->last_crash_execs = afl->fsrv.total_execs;
 
         } else {
 
-          afl->queue = afl->queue_buf[0];
+          q->disabled = 1;
+          q->perf_score = 0;
 
-        }
+          u32 i = 0;
+          while (unlikely(i < afl->queued_items && afl->queue_buf[i] &&
+                          afl->queue_buf[i]->disabled)) {
 
-        afl->max_depth = 0;
-        for (i = 0; i < afl->queued_items && likely(afl->queue_buf[i]); i++) {
+            ++i;
 
-          if (!afl->queue_buf[i]->disabled &&
-              afl->queue_buf[i]->depth > afl->max_depth)
-            afl->max_depth = afl->queue_buf[i]->depth;
+          }
 
-        }
+          if (i < afl->queued_items && afl->queue_buf[i]) {
 
-        break;
+            afl->queue = afl->queue_buf[i];
+
+          } else {
 
+            afl->queue = afl->queue_buf[0];
+
+          }
+
+          afl->max_depth = 0;
+          for (i = 0; i < afl->queued_items && likely(afl->queue_buf[i]); i++) {
+
+            if (!afl->queue_buf[i]->disabled &&
+                afl->queue_buf[i]->depth > afl->max_depth)
+              afl->max_depth = afl->queue_buf[i]->depth;
+
+          }
+
+        }
+        
+        break;  
+      
       case FSRV_RUN_ERROR:
 
         FATAL("Unable to execute target application ('%s')", afl->argv[0]);
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index bacbafc4..5cbebb0e 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -1573,7 +1573,8 @@ int main(int argc, char **argv_orig, char **envp) {
   if (get_afl_env("AFL_NO_ARITH")) { afl->no_arith = 1; }
   if (get_afl_env("AFL_SHUFFLE_QUEUE")) { afl->shuffle_queue = 1; }
   if (get_afl_env("AFL_EXPAND_HAVOC_NOW")) { afl->expand_havoc = 1; }
-
+  if (get_afl_env("AFL_CRASHING_SEEDS_AS_NEW_CRASH")) { afl->crashing_seeds_as_new_crash = 1; }
+  
   if (afl->afl_env.afl_autoresume) {
 
     afl->autoresume = 1;
-- 
cgit 1.4.1


From 8f31086a7fa1d7ef9d4dc416f238a10dd140e2d3 Mon Sep 17 00:00:00 2001
From: Sergej Schumilo <sergej@schumilo.de>
Date: Thu, 3 Aug 2023 14:40:10 +0200
Subject: make nyx aux buffer size configurable

---
 include/envs.h       |  1 +
 include/forkserver.h |  1 +
 src/afl-forkserver.c | 10 ++++++++++
 src/afl-fuzz.c       |  3 +++
 4 files changed, 15 insertions(+)

(limited to 'include')

diff --git a/include/envs.h b/include/envs.h
index edfd06e4..0ef331ae 100644
--- a/include/envs.h
+++ b/include/envs.h
@@ -189,6 +189,7 @@ static char *afl_environment_variables[] = {
     "AFL_MAX_DET_EXTRAS",
     "AFL_NO_X86",  // not really an env but we dont want to warn on it
     "AFL_NOOPT",
+    "AFL_NYX_AUX_SIZE",
     "AFL_PASSTHROUGH",
     "AFL_PATH",
     "AFL_PERFORMANCE_FILE",
diff --git a/include/forkserver.h b/include/forkserver.h
index f5069ce2..c93c6f61 100644
--- a/include/forkserver.h
+++ b/include/forkserver.h
@@ -86,6 +86,7 @@ typedef struct {
                                  uint32_t size);
 
   bool (*nyx_remove_work_dir)(const char *workdir);
+  bool (*nyx_config_set_aux_buffer_size)(void *config, uint32_t aux_buffer_size);
 
 } nyx_plugin_handler_t;
 
diff --git a/src/afl-forkserver.c b/src/afl-forkserver.c
index ba7cdd66..957cb2b7 100644
--- a/src/afl-forkserver.c
+++ b/src/afl-forkserver.c
@@ -129,6 +129,9 @@ nyx_plugin_handler_t *afl_load_libnyx_plugin(u8 *libnyx_binary) {
   plugin->nyx_remove_work_dir = dlsym(handle, "nyx_remove_work_dir");
   if (plugin->nyx_remove_work_dir == NULL) { goto fail; }
 
+  plugin->nyx_config_set_aux_buffer_size = dlsym(handle, "nyx_config_set_aux_buffer_size");
+  if (plugin->nyx_config_set_aux_buffer_size == NULL) { goto fail; }
+
   OKF("libnyx plugin is ready!");
   return plugin;
 
@@ -589,6 +592,13 @@ void afl_fsrv_start(afl_forkserver_t *fsrv, char **argv,
 
     }
 
+    if (getenv("AFL_NYX_AUX_SIZE") != NULL) {
+        if(fsrv->nyx_handlers->nyx_config_set_aux_buffer_size(
+            nyx_config, atoi(getenv("AFL_NYX_AUX_SIZE"))) != 1) {  
+            NYX_PRE_FATAL(fsrv, "Invalid AFL_NYX_AUX_SIZE value set (must be a multiple of 4096) ...");
+        }
+    }
+
     if (getenv("NYX_REUSE_SNAPSHOT") != NULL) {
 
       if (access(getenv("NYX_REUSE_SNAPSHOT"), F_OK) == -1) {
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index bacbafc4..9504d908 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -299,6 +299,9 @@ static void usage(u8 *argv0, int more_help) {
       "AFL_NO_SNAPSHOT: do not use the snapshot feature (if the snapshot lkm is loaded)\n"
       "AFL_NO_STARTUP_CALIBRATION: no initial seed calibration, start fuzzing at once\n"
       "AFL_NO_UI: switch status screen off\n"
+      "AFL_NYX_AUX_SIZE: size of the Nyx auxiliary buffer. Must be a multiple of 4096.\n"
+      "                  Increase this value in case the crash reports are truncated.\n"
+      "                  Default value is 4096.\n"
 
       DYN_COLOR
 
-- 
cgit 1.4.1


From 79640acbf1ffff9677ec9094b61ac4a158b1551c Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Fri, 4 Aug 2023 09:25:19 +0200
Subject: nits

---
 include/afl-mutations.h |  8 ++++----
 include/forkserver.h    |  3 ++-
 src/afl-forkserver.c    | 17 ++++++++++++-----
 src/afl-fuzz-redqueen.c |  2 +-
 src/afl-fuzz.c          | 15 +++++++--------
 5 files changed, 26 insertions(+), 19 deletions(-)

(limited to 'include')

diff --git a/include/afl-mutations.h b/include/afl-mutations.h
index 1806790e..98ba6fcf 100644
--- a/include/afl-mutations.h
+++ b/include/afl-mutations.h
@@ -78,7 +78,7 @@ enum {
 
 };
 
-  #define MUT_TXT_ARRAY_SIZE 200
+#define MUT_TXT_ARRAY_SIZE 200
 u32 text_array[MUT_TXT_ARRAY_SIZE] = {MUT_FLIPBIT,
                                       MUT_FLIPBIT,
                                       MUT_FLIPBIT,
@@ -280,7 +280,7 @@ u32 text_array[MUT_TXT_ARRAY_SIZE] = {MUT_FLIPBIT,
                                       MUT_SPLICE_INSERT,
                                       MUT_SPLICE_INSERT};
 
-  #define MUT_BIN_ARRAY_SIZE 256
+#define MUT_BIN_ARRAY_SIZE 256
 u32 binary_array[MUT_BIN_ARRAY_SIZE] = {MUT_FLIPBIT,
                                         MUT_FLIPBIT,
                                         MUT_FLIPBIT,
@@ -538,7 +538,7 @@ u32 binary_array[MUT_BIN_ARRAY_SIZE] = {MUT_FLIPBIT,
                                         MUT_SPLICE_INSERT,
                                         MUT_SPLICE_INSERT};
 
-  #define MUT_NORMAL_ARRAY_SIZE 77
+#define MUT_NORMAL_ARRAY_SIZE 77
 u32 normal_splice_array[MUT_NORMAL_ARRAY_SIZE] = {MUT_FLIPBIT,
                                                   MUT_FLIPBIT,
                                                   MUT_FLIPBIT,
@@ -617,7 +617,7 @@ u32 normal_splice_array[MUT_NORMAL_ARRAY_SIZE] = {MUT_FLIPBIT,
                                                   MUT_SPLICE_INSERT,
                                                   MUT_SPLICE_INSERT};
 
-  #define MUT_SPLICE_ARRAY_SIZE 81
+#define MUT_SPLICE_ARRAY_SIZE 81
 u32 full_splice_array[MUT_SPLICE_ARRAY_SIZE] = {MUT_FLIPBIT,
                                                 MUT_FLIPBIT,
                                                 MUT_FLIPBIT,
diff --git a/include/forkserver.h b/include/forkserver.h
index c93c6f61..1d41d83d 100644
--- a/include/forkserver.h
+++ b/include/forkserver.h
@@ -86,7 +86,8 @@ typedef struct {
                                  uint32_t size);
 
   bool (*nyx_remove_work_dir)(const char *workdir);
-  bool (*nyx_config_set_aux_buffer_size)(void *config, uint32_t aux_buffer_size);
+  bool (*nyx_config_set_aux_buffer_size)(void    *config,
+                                         uint32_t aux_buffer_size);
 
 } nyx_plugin_handler_t;
 
diff --git a/src/afl-forkserver.c b/src/afl-forkserver.c
index 957cb2b7..e90ea460 100644
--- a/src/afl-forkserver.c
+++ b/src/afl-forkserver.c
@@ -129,7 +129,8 @@ nyx_plugin_handler_t *afl_load_libnyx_plugin(u8 *libnyx_binary) {
   plugin->nyx_remove_work_dir = dlsym(handle, "nyx_remove_work_dir");
   if (plugin->nyx_remove_work_dir == NULL) { goto fail; }
 
-  plugin->nyx_config_set_aux_buffer_size = dlsym(handle, "nyx_config_set_aux_buffer_size");
+  plugin->nyx_config_set_aux_buffer_size =
+      dlsym(handle, "nyx_config_set_aux_buffer_size");
   if (plugin->nyx_config_set_aux_buffer_size == NULL) { goto fail; }
 
   OKF("libnyx plugin is ready!");
@@ -593,10 +594,16 @@ void afl_fsrv_start(afl_forkserver_t *fsrv, char **argv,
     }
 
     if (getenv("AFL_NYX_AUX_SIZE") != NULL) {
-        if(fsrv->nyx_handlers->nyx_config_set_aux_buffer_size(
-            nyx_config, atoi(getenv("AFL_NYX_AUX_SIZE"))) != 1) {  
-            NYX_PRE_FATAL(fsrv, "Invalid AFL_NYX_AUX_SIZE value set (must be a multiple of 4096) ...");
-        }
+
+      if (fsrv->nyx_handlers->nyx_config_set_aux_buffer_size(
+              nyx_config, atoi(getenv("AFL_NYX_AUX_SIZE"))) != 1) {
+
+        NYX_PRE_FATAL(fsrv,
+                      "Invalid AFL_NYX_AUX_SIZE value set (must be a multiple "
+                      "of 4096) ...");
+
+      }
+
     }
 
     if (getenv("NYX_REUSE_SNAPSHOT") != NULL) {
diff --git a/src/afl-fuzz-redqueen.c b/src/afl-fuzz-redqueen.c
index 54bf4e32..ca5104c0 100644
--- a/src/afl-fuzz-redqueen.c
+++ b/src/afl-fuzz-redqueen.c
@@ -2665,7 +2665,7 @@ static u8 rtn_fuzz(afl_state_t *afl, u32 key, u8 *orig_buf, u8 *buf, u8 *cbuf,
 
     }
 
-  // rtn_fuzz_next_iter:
+    // rtn_fuzz_next_iter:
     afl->stage_cur++;
 
   }
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index 9504d908..29659013 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -311,8 +311,8 @@ static void usage(u8 *argv0, int more_help) {
 
       PERSISTENT_MSG
 
-      "AFL_POST_PROCESS_KEEP_ORIGINAL: save the file as it was prior post-processing to the queue,\n"
-      "                                but execute the post-processed one\n"
+      "AFL_POST_PROCESS_KEEP_ORIGINAL: save the file as it was prior post-processing to\n"
+      "                                the queue, but execute the post-processed one\n"
       "AFL_PRELOAD: LD_PRELOAD / DYLD_INSERT_LIBRARIES settings for target\n"
       "AFL_TARGET_ENV: pass extra environment variables to target\n"
       "AFL_SHUFFLE_QUEUE: reorder the input queue randomly on startup\n"
@@ -323,18 +323,17 @@ static void usage(u8 *argv0, int more_help) {
       "AFL_STATSD_HOST: change default statsd host (default 127.0.0.1)\n"
       "AFL_STATSD_PORT: change default statsd port (default: 8125)\n"
       "AFL_STATSD_TAGS_FLAVOR: set statsd tags format (default: disable tags)\n"
-      "                        Supported formats are: 'dogstatsd', 'librato',\n"
-      "                        'signalfx' and 'influxdb'\n"
+      "                        suported formats: dogstatsd, librato, signalfx, influxdb\n"
       "AFL_SYNC_TIME: sync time between fuzzing instances (in minutes)\n"
       "AFL_NO_CRASH_README: do not create a README in the crashes directory\n"
       "AFL_TESTCACHE_SIZE: use a cache for testcases, improves performance (in MB)\n"
       "AFL_TMPDIR: directory to use for input file generation (ramdisk recommended)\n"
       "AFL_EARLY_FORKSERVER: force an early forkserver in an afl-clang-fast/\n"
       "                      afl-clang-lto/afl-gcc-fast target\n"
-      "AFL_PERSISTENT: enforce persistent mode (if __AFL_LOOP is in a shared lib\n"
-      "AFL_DEFER_FORKSRV: enforced deferred forkserver (__AFL_INIT is in a .so)\n"
-      "AFL_FUZZER_STATS_UPDATE_INTERVAL: interval to update fuzzer_stats file in seconds, "
-      "(default: 60, minimum: 1)\n"
+      "AFL_PERSISTENT: enforce persistent mode (if __AFL_LOOP is in a shared lib)\n"
+      "AFL_DEFER_FORKSRV: enforced deferred forkserver (__AFL_INIT is in a shared lib)\n"
+      "AFL_FUZZER_STATS_UPDATE_INTERVAL: interval to update fuzzer_stats file in\n"
+      "                                  seconds (default: 60, minimum: 1)\n"
       "\n"
     );
 
-- 
cgit 1.4.1


From fcdfe9e990d84ab477cd3c571cbf540e8bc8e15a Mon Sep 17 00:00:00 2001
From: Junwha <qbit@unist.ac.kr>
Date: Fri, 4 Aug 2023 18:36:58 +0900
Subject: Define AFL_CRASHING_SEEDS_AS_NEW_CRASH as env variable

- and fix typo

Signed-off-by: Junwha <qbit@unist.ac.kr>
---
 include/afl-fuzz.h   |  6 +++---
 src/afl-fuzz-init.c  | 30 ++++++++++++++++--------------
 src/afl-fuzz-state.c |  7 +++++++
 src/afl-fuzz.c       |  1 -
 4 files changed, 26 insertions(+), 18 deletions(-)

(limited to 'include')

diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index 7bedc98f..18352acb 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -1,3 +1,4 @@
+
 /*
    american fuzzy lop++ - fuzzer header
    ------------------------------------
@@ -408,7 +409,7 @@ typedef struct afl_env_vars {
       *afl_max_det_extras, *afl_statsd_host, *afl_statsd_port,
       *afl_crash_exitcode, *afl_statsd_tags_flavor, *afl_testcache_size,
       *afl_testcache_entries, *afl_child_kill_signal, *afl_fsrv_kill_signal,
-      *afl_target_env, *afl_persistent_record, *afl_exit_on_time;
+      *afl_target_env, *afl_persistent_record, *afl_exit_on_time, *afl_crashing_seeds_as_new_crash;
 
   s32 afl_pizza_mode;
 
@@ -539,8 +540,7 @@ typedef struct afl_state {
       expand_havoc,                /* perform expensive havoc after no find */
       cycle_schedules,                  /* cycle power schedules?           */
       old_seed_selection,               /* use vanilla afl seed selection   */
-      reinit_table,                     /* reinit the queue weight table    */
-      crashing_seeds_as_new_crash; /* treat crashing seeds as normal corpus */
+      reinit_table;                     /* reinit the queue weight table    */
 
   u8 *virgin_bits,                      /* Regions yet untouched by fuzzing */
       *virgin_tmout,                    /* Bits we haven't seen in tmouts   */
diff --git a/src/afl-fuzz-init.c b/src/afl-fuzz-init.c
index 6b7f3036..d994d749 100644
--- a/src/afl-fuzz-init.c
+++ b/src/afl-fuzz-init.c
@@ -1056,17 +1056,19 @@ void perform_dry_run(afl_state_t *afl) {
               "skipping",
               fn, (int)(s8)afl->fsrv.crash_exitcode);
 
-        } else if (afl->crashing_seeds_as_new_crash) {
-          
-          WARNF(
-              "Test case '%s' results in a crash,"
-              "as AFL_CRASHING_SEEDS_AS_NEW_CRASH is set, "
-              "saving as a crash", fn);
-
         } else {
+          if (afl->afl_env.afl_crashing_seeds_as_new_crash) {
+          
+            WARNF(
+                "Test case '%s' results in a crash, "
+                "as AFL_CRASHING_SEEDS_AS_NEW_CRASH is set, "
+                "saving as a new crash", fn);
+            
+          } else {
 
-          WARNF("Test case '%s' results in a crash, skipping", fn);
-
+            WARNF("Test case '%s' results in a crash, skipping", fn);
+          
+          }
         }
 
         if (afl->afl_env.afl_exit_on_seed_issues) {
@@ -1085,8 +1087,8 @@ void perform_dry_run(afl_state_t *afl) {
 
         }
 
-        /* Crashing corpus will regrad as normal, and categorized as new crash at fuzzing  */
-        if (afl->crashing_seeds_as_new_crash) {
+        /* Crashing seeds will be regarded as new crashes on startup */
+        if (afl->afl_env.afl_crashing_seeds_as_new_crash) {
           
           ++afl->total_crashes;
 
@@ -1139,9 +1141,6 @@ void perform_dry_run(afl_state_t *afl) {
 
         } else {
 
-          q->disabled = 1;
-          q->perf_score = 0;
-
           u32 i = 0;
           while (unlikely(i < afl->queued_items && afl->queue_buf[i] &&
                           afl->queue_buf[i]->disabled)) {
@@ -1171,6 +1170,9 @@ void perform_dry_run(afl_state_t *afl) {
 
         }
         
+        q->disabled = 1;
+        q->perf_score = 0;
+
         break;  
       
       case FSRV_RUN_ERROR:
diff --git a/src/afl-fuzz-state.c b/src/afl-fuzz-state.c
index 99f69314..5a6b95cf 100644
--- a/src/afl-fuzz-state.c
+++ b/src/afl-fuzz-state.c
@@ -200,6 +200,13 @@ void read_afl_environment(afl_state_t *afl, char **envp) {
             afl->afl_env.afl_exit_on_time =
                 (u8 *)get_afl_env(afl_environment_variables[i]);
 
+          } else if (!strncmp(env, "AFL_CRASHING_SEEDS_AS_NEW_CRASH",
+
+                              afl_environment_variable_len)) {
+
+            afl->afl_env.afl_crashing_seeds_as_new_crash =
+                atoi((u8 *)get_afl_env(afl_environment_variables[i]));
+
           } else if (!strncmp(env, "AFL_NO_AFFINITY",
 
                               afl_environment_variable_len)) {
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index 5cbebb0e..51ca4ee6 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -1573,7 +1573,6 @@ int main(int argc, char **argv_orig, char **envp) {
   if (get_afl_env("AFL_NO_ARITH")) { afl->no_arith = 1; }
   if (get_afl_env("AFL_SHUFFLE_QUEUE")) { afl->shuffle_queue = 1; }
   if (get_afl_env("AFL_EXPAND_HAVOC_NOW")) { afl->expand_havoc = 1; }
-  if (get_afl_env("AFL_CRASHING_SEEDS_AS_NEW_CRASH")) { afl->crashing_seeds_as_new_crash = 1; }
   
   if (afl->afl_env.afl_autoresume) {
 
-- 
cgit 1.4.1


From 08a6fd7c29489f5477f50b94d7a0e425f64fef34 Mon Sep 17 00:00:00 2001
From: Sergej Schumilo <sergej@schumilo.de>
Date: Fri, 4 Aug 2023 12:13:06 +0200
Subject: update the old nyx env var naming scheme (to have a more consistent
 naming overall)

---
 include/envs.h       |  2 ++
 nyx_mode/README.md   |  6 +++---
 src/afl-forkserver.c | 27 ++++++++++++++-------------
 src/afl-fuzz.c       |  3 ++-
 4 files changed, 21 insertions(+), 17 deletions(-)

(limited to 'include')

diff --git a/include/envs.h b/include/envs.h
index 0ef331ae..ff303387 100644
--- a/include/envs.h
+++ b/include/envs.h
@@ -190,6 +190,8 @@ static char *afl_environment_variables[] = {
     "AFL_NO_X86",  // not really an env but we dont want to warn on it
     "AFL_NOOPT",
     "AFL_NYX_AUX_SIZE",
+    "AFL_NYX_DISABLE_SNAPSHOT_MODE",
+    "AFL_NYX_REUSE_SNAPSHOT",
     "AFL_PASSTHROUGH",
     "AFL_PATH",
     "AFL_PERFORMANCE_FILE",
diff --git a/nyx_mode/README.md b/nyx_mode/README.md
index 0565331b..aee9879e 100644
--- a/nyx_mode/README.md
+++ b/nyx_mode/README.md
@@ -150,12 +150,12 @@ afl-cmin -i in_dir -o out_dir -X -- ./PACKAGE-DIRECTORY
 
 On each program startup of one the AFL++ tools in Nyx mode, a Nyx VM is spawned, and a bootstrapping procedure is performed inside the VM to prepare the target environment. As a consequence, due to the bootstrapping procedure, the launch performance is much slower compared to other modes. However, this can be optimized by reusing an existing fuzzing snapshot to avoid the slow re-execution of the bootstrap procedure. 
 
-A fuzzing snapshot is automatically created and stored in the output directory at `out_dir/workdir/snapshot/` by the first parent process of `afl-fuzz` if parallel mode is used. To enable this feature, set the path to an existing snapshot directory in the `NYX_REUSE_SNAPSHOT` environment variable and use the tools as usual:
+A fuzzing snapshot is automatically created and stored in the output directory at `out_dir/workdir/snapshot/` by the first parent process of `afl-fuzz` if parallel mode is used. To enable this feature, set the path to an existing snapshot directory in the `AFL_NYX_REUSE_SNAPSHOT` environment variable and use the tools as usual:
 
 ```shell 
 afl-fuzz -i ./in_dir -o ./out_dir -Y -M 0 ./PACKAGE-DIRECTORY
 
-NYX_REUSE_SNAPSHOT=./out_dir/workdir/snapshot/ afl-analyze -i in_file -X  -- ./PACKAGE-DIRECTORY
+AFL_NYX_REUSE_SNAPSHOT=./out_dir/workdir/snapshot/ afl-analyze -i in_file -X  -- ./PACKAGE-DIRECTORY
 ```
 
 
@@ -311,7 +311,7 @@ command:
 ```
 
 If you want to disable fast snapshots (except for crashes), you can simply set
-the `NYX_DISABLE_SNAPSHOT_MODE` environment variable.
+the `AFL_NYX_DISABLE_SNAPSHOT_MODE` environment variable.
 
 ### Nyx crash reports
 
diff --git a/src/afl-forkserver.c b/src/afl-forkserver.c
index e90ea460..717898d1 100644
--- a/src/afl-forkserver.c
+++ b/src/afl-forkserver.c
@@ -606,23 +606,23 @@ void afl_fsrv_start(afl_forkserver_t *fsrv, char **argv,
 
     }
 
-    if (getenv("NYX_REUSE_SNAPSHOT") != NULL) {
+    if (getenv("AFL_NYX_REUSE_SNAPSHOT") != NULL) {
 
-      if (access(getenv("NYX_REUSE_SNAPSHOT"), F_OK) == -1) {
+      if (access(getenv("AFL_NYX_REUSE_SNAPSHOT"), F_OK) == -1) {
 
-        NYX_PRE_FATAL(fsrv, "NYX_REUSE_SNAPSHOT path does not exist");
+        NYX_PRE_FATAL(fsrv, "AFL_NYX_REUSE_SNAPSHOT path does not exist");
 
       }
 
       /* stupid sanity check to avoid passing an empty or invalid snapshot
        * directory */
       char *snapshot_file_path =
-          alloc_printf("%s/global.state", getenv("NYX_REUSE_SNAPSHOT"));
+          alloc_printf("%s/global.state", getenv("AFL_NYX_REUSE_SNAPSHOT"));
       if (access(snapshot_file_path, R_OK) == -1) {
 
-        NYX_PRE_FATAL(
-            fsrv,
-            "NYX_REUSE_SNAPSHOT path does not contain a valid Nyx snapshot");
+        NYX_PRE_FATAL(fsrv,
+                      "AFL_NYX_REUSE_SNAPSHOT path does not contain a valid "
+                      "Nyx snapshot");
 
       }
 
@@ -634,13 +634,14 @@ void afl_fsrv_start(afl_forkserver_t *fsrv, char **argv,
       char *workdir_snapshot_path =
           alloc_printf("%s/workdir/snapshot", outdir_path_absolute);
       char *reuse_snapshot_path_real =
-          realpath(getenv("NYX_REUSE_SNAPSHOT"), NULL);
+          realpath(getenv("AFL_NYX_REUSE_SNAPSHOT"), NULL);
 
       if (strcmp(workdir_snapshot_path, reuse_snapshot_path_real) == 0) {
 
-        NYX_PRE_FATAL(fsrv,
-                      "NYX_REUSE_SNAPSHOT path is located in current workdir "
-                      "(use another output directory)");
+        NYX_PRE_FATAL(
+            fsrv,
+            "AFL_NYX_REUSE_SNAPSHOT path is located in current workdir "
+            "(use another output directory)");
 
       }
 
@@ -648,7 +649,7 @@ void afl_fsrv_start(afl_forkserver_t *fsrv, char **argv,
       ck_free(workdir_snapshot_path);
 
       fsrv->nyx_handlers->nyx_config_set_reuse_snapshot_path(
-          nyx_config, getenv("NYX_REUSE_SNAPSHOT"));
+          nyx_config, getenv("AFL_NYX_REUSE_SNAPSHOT"));
 
     }
 
@@ -670,7 +671,7 @@ void afl_fsrv_start(afl_forkserver_t *fsrv, char **argv,
         fsrv->nyx_handlers->nyx_get_bitmap_buffer(fsrv->nyx_runner);
 
     fsrv->nyx_handlers->nyx_option_set_reload_mode(
-        fsrv->nyx_runner, getenv("NYX_DISABLE_SNAPSHOT_MODE") == NULL);
+        fsrv->nyx_runner, getenv("AFL_NYX_DISABLE_SNAPSHOT_MODE") == NULL);
     fsrv->nyx_handlers->nyx_option_apply(fsrv->nyx_runner);
 
     fsrv->nyx_handlers->nyx_option_set_timeout(fsrv->nyx_runner, 2, 0);
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index 29659013..3d7adf41 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -302,7 +302,8 @@ static void usage(u8 *argv0, int more_help) {
       "AFL_NYX_AUX_SIZE: size of the Nyx auxiliary buffer. Must be a multiple of 4096.\n"
       "                  Increase this value in case the crash reports are truncated.\n"
       "                  Default value is 4096.\n"
-
+      "AFL_NYX_DISABLE_SNAPSHOT_MODE: disable snapshot mode (must be supported by the agent)\n"
+      "AFL_NYX_REUSE_SNAPSHOT: reuse an existing Nyx root snapshot\n"
       DYN_COLOR
 
       "AFL_PATH: path to AFL support binaries\n"
-- 
cgit 1.4.1


From 247d8539feb0dee3eab80586ee4e32292dc7ca78 Mon Sep 17 00:00:00 2001
From: Jesse Schwartzentruber <truber@mozilla.com>
Date: Fri, 4 Aug 2023 15:29:10 -0400
Subject: Add AFL_NYX_LOG to redirect NYX hprintf messages to a file.

---
 include/envs.h       |  1 +
 include/forkserver.h |  1 +
 src/afl-forkserver.c | 19 +++++++++++++++++++
 src/afl-fuzz.c       |  1 +
 4 files changed, 22 insertions(+)

(limited to 'include')

diff --git a/include/envs.h b/include/envs.h
index ff303387..963e1367 100644
--- a/include/envs.h
+++ b/include/envs.h
@@ -191,6 +191,7 @@ static char *afl_environment_variables[] = {
     "AFL_NOOPT",
     "AFL_NYX_AUX_SIZE",
     "AFL_NYX_DISABLE_SNAPSHOT_MODE",
+    "AFL_NYX_LOG",
     "AFL_NYX_REUSE_SNAPSHOT",
     "AFL_PASSTHROUGH",
     "AFL_PATH",
diff --git a/include/forkserver.h b/include/forkserver.h
index 1d41d83d..5e498c56 100644
--- a/include/forkserver.h
+++ b/include/forkserver.h
@@ -199,6 +199,7 @@ typedef struct afl_forkserver {
   char                 *nyx_aux_string;
   bool                  nyx_use_tmp_workdir;
   char                 *nyx_tmp_workdir_path;
+  s32                   nyx_log_fd;
 #endif
 
 } afl_forkserver_t;
diff --git a/src/afl-forkserver.c b/src/afl-forkserver.c
index 717898d1..9da096f7 100644
--- a/src/afl-forkserver.c
+++ b/src/afl-forkserver.c
@@ -164,6 +164,8 @@ void afl_nyx_runner_kill(afl_forkserver_t *fsrv) {
 
     }
 
+    if (fsrv->nyx_log_fd >= 0) { close(fsrv->nyx_log_fd); }
+
   }
 
 }
@@ -218,6 +220,7 @@ void afl_fsrv_init(afl_forkserver_t *fsrv) {
   fsrv->nyx_bind_cpu_id = 0xFFFFFFFF;
   fsrv->nyx_use_tmp_workdir = false;
   fsrv->nyx_tmp_workdir_path = NULL;
+  fsrv->nyx_log_fd = -1;
 #endif
 
   // this structure needs default so we initialize it if this was not done
@@ -575,6 +578,22 @@ void afl_fsrv_start(afl_forkserver_t *fsrv, char **argv,
     fsrv->nyx_handlers->nyx_config_set_input_buffer_write_protection(nyx_config,
                                                                      true);
 
+    char *nyx_log_path = getenv("AFL_NYX_LOG");
+    if (nyx_log_path) {
+
+      fsrv->nyx_log_fd =
+          open(nyx_log_path, O_CREAT | O_TRUNC | O_WRONLY, DEFAULT_PERMISSION);
+      if (fsrv->nyx_log_fd < 0) {
+
+        NYX_PRE_FATAL(fsrv, "AFL_NYX_LOG path could not be written");
+
+      }
+
+      fsrv->nyx_handlers->nyx_config_set_hprintf_fd(nyx_config,
+                                                    fsrv->nyx_log_fd);
+
+    }
+
     if (fsrv->nyx_standalone) {
 
       fsrv->nyx_handlers->nyx_config_set_process_role(nyx_config, StandAlone);
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index 3d7adf41..e1f93f0d 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -303,6 +303,7 @@ static void usage(u8 *argv0, int more_help) {
       "                  Increase this value in case the crash reports are truncated.\n"
       "                  Default value is 4096.\n"
       "AFL_NYX_DISABLE_SNAPSHOT_MODE: disable snapshot mode (must be supported by the agent)\n"
+      "AFL_NYX_LOG: output NYX hprintf messages to another file\n"
       "AFL_NYX_REUSE_SNAPSHOT: reuse an existing Nyx root snapshot\n"
       DYN_COLOR
 
-- 
cgit 1.4.1


From 55d696fbae435e0e69adf75cb2df1361186fb999 Mon Sep 17 00:00:00 2001
From: marc <marc@marc.srlabs.de>
Date: Wed, 9 Aug 2023 17:14:13 +0200
Subject: code format

---
 include/afl-fuzz.h      |  4 ++--
 src/afl-fuzz-init.c     | 42 ++++++++++++++++++++++++------------------
 src/afl-fuzz-redqueen.c |  2 +-
 src/afl-fuzz.c          |  2 +-
 4 files changed, 28 insertions(+), 22 deletions(-)

(limited to 'include')

diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index 18352acb..ef84a18c 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -402,14 +402,14 @@ typedef struct afl_env_vars {
       afl_exit_on_seed_issues, afl_try_affinity, afl_ignore_problems,
       afl_keep_timeouts, afl_no_crash_readme, afl_ignore_timeouts,
       afl_no_startup_calibration, afl_no_warn_instability,
-      afl_post_process_keep_original;
+      afl_post_process_keep_original, afl_crashing_seeds_as_new_crash;
 
   u8 *afl_tmpdir, *afl_custom_mutator_library, *afl_python_module, *afl_path,
       *afl_hang_tmout, *afl_forksrv_init_tmout, *afl_preload,
       *afl_max_det_extras, *afl_statsd_host, *afl_statsd_port,
       *afl_crash_exitcode, *afl_statsd_tags_flavor, *afl_testcache_size,
       *afl_testcache_entries, *afl_child_kill_signal, *afl_fsrv_kill_signal,
-      *afl_target_env, *afl_persistent_record, *afl_exit_on_time, *afl_crashing_seeds_as_new_crash;
+      *afl_target_env, *afl_persistent_record, *afl_exit_on_time;
 
   s32 afl_pizza_mode;
 
diff --git a/src/afl-fuzz-init.c b/src/afl-fuzz-init.c
index d994d749..5a530821 100644
--- a/src/afl-fuzz-init.c
+++ b/src/afl-fuzz-init.c
@@ -1057,18 +1057,21 @@ void perform_dry_run(afl_state_t *afl) {
               fn, (int)(s8)afl->fsrv.crash_exitcode);
 
         } else {
+
           if (afl->afl_env.afl_crashing_seeds_as_new_crash) {
-          
+
             WARNF(
                 "Test case '%s' results in a crash, "
                 "as AFL_CRASHING_SEEDS_AS_NEW_CRASH is set, "
-                "saving as a new crash", fn);
-            
+                "saving as a new crash",
+                fn);
+
           } else {
 
             WARNF("Test case '%s' results in a crash, skipping", fn);
-          
+
           }
+
         }
 
         if (afl->afl_env.afl_exit_on_seed_issues) {
@@ -1089,20 +1092,19 @@ void perform_dry_run(afl_state_t *afl) {
 
         /* Crashing seeds will be regarded as new crashes on startup */
         if (afl->afl_env.afl_crashing_seeds_as_new_crash) {
-          
+
           ++afl->total_crashes;
 
           if (likely(!afl->non_instrumented_mode)) {
 
             classify_counts(&afl->fsrv);
-          
+
             simplify_trace(afl, afl->fsrv.trace_bits);
 
             if (!has_new_bits(afl, afl->virgin_crash)) { break; }
 
           }
 
-          
           if (unlikely(!afl->saved_crashes) &&
               (afl->afl_env.afl_no_crash_readme != 1)) {
 
@@ -1116,18 +1118,22 @@ void perform_dry_run(afl_state_t *afl) {
           afl->stage_name = "dry_run";
           afl->stage_short = "dry_run";
 
-    #ifndef SIMPLE_FILES
+#ifndef SIMPLE_FILES
 
-          snprintf(crash_fn, PATH_MAX, "%s/crashes/id:%06llu,sig:%02u,%s%s", afl->out_dir,
-                  afl->saved_crashes, afl->fsrv.last_kill_signal,
-                  describe_op(afl, 0, NAME_MAX - strlen("id:000000,sig:00,") - strlen(use_name)), use_name);
+          snprintf(crash_fn, PATH_MAX, "%s/crashes/id:%06llu,sig:%02u,%s%s",
+                   afl->out_dir, afl->saved_crashes, afl->fsrv.last_kill_signal,
+                   describe_op(afl, 0,
+                               NAME_MAX - strlen("id:000000,sig:00,") -
+                                   strlen(use_name)),
+                   use_name);
 
-    #else
+#else
 
-          snprintf(crash_fn, PATH_MAX, "%s/crashes/id_%06llu_%02u", afl->out_dir,
-                  afl->saved_crashes, afl->fsrv.last_kill_signal);
+          snprintf(crash_fn, PATH_MAX, "%s/crashes/id_%06llu_%02u",
+                   afl->out_dir, afl->saved_crashes,
+                   afl->fsrv.last_kill_signal);
 
-    #endif  
+#endif
 
           ++afl->saved_crashes;
 
@@ -1169,12 +1175,12 @@ void perform_dry_run(afl_state_t *afl) {
           }
 
         }
-        
+
         q->disabled = 1;
         q->perf_score = 0;
 
-        break;  
-      
+        break;
+
       case FSRV_RUN_ERROR:
 
         FATAL("Unable to execute target application ('%s')", afl->argv[0]);
diff --git a/src/afl-fuzz-redqueen.c b/src/afl-fuzz-redqueen.c
index 509f66a3..db4991db 100644
--- a/src/afl-fuzz-redqueen.c
+++ b/src/afl-fuzz-redqueen.c
@@ -2525,7 +2525,7 @@ static u8 rtn_fuzz(afl_state_t *afl, u32 key, u8 *orig_buf, u8 *buf, u8 *cbuf,
     */
 
 #ifdef _DEBUG
-    u32 j;
+    u32                j;
     struct cmp_header *hh = &afl->orig_cmp_map->headers[key];
     fprintf(stderr, "RTN N hits=%u id=%u shape=%u attr=%u v0=", h->hits, h->id,
             hshape, h->attribute);
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index 733c7429..cdb3f996 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -1592,7 +1592,7 @@ int main(int argc, char **argv_orig, char **envp) {
   if (get_afl_env("AFL_NO_ARITH")) { afl->no_arith = 1; }
   if (get_afl_env("AFL_SHUFFLE_QUEUE")) { afl->shuffle_queue = 1; }
   if (get_afl_env("AFL_EXPAND_HAVOC_NOW")) { afl->expand_havoc = 1; }
-  
+
   if (afl->afl_env.afl_autoresume) {
 
     afl->autoresume = 1;
-- 
cgit 1.4.1


From 3721c65a0b7fdf2b24713f8009030c6c241e200b Mon Sep 17 00:00:00 2001
From: marc <marc@marc.srlabs.de>
Date: Thu, 10 Aug 2023 10:41:55 +0200
Subject: v4.08c release

---
 README.md         | 10 +++++-----
 docs/Changelog.md |  2 +-
 include/config.h  |  6 +++---
 3 files changed, 9 insertions(+), 9 deletions(-)

(limited to 'include')

diff --git a/README.md b/README.md
index b73cf2c1..951efe59 100644
--- a/README.md
+++ b/README.md
@@ -2,9 +2,9 @@
 
 <img align="right" src="https://raw.githubusercontent.com/AFLplusplus/Website/main/static/aflpp_bg.svg" alt="AFL++ logo" width="250" heigh="250">
 
-Release version: [4.07c](https://github.com/AFLplusplus/AFLplusplus/releases)
+Release version: [4.08c](https://github.com/AFLplusplus/AFLplusplus/releases)
 
-GitHub version: 4.08a
+GitHub version: 4.08c
 
 Repository:
 [https://github.com/AFLplusplus/AFLplusplus](https://github.com/AFLplusplus/AFLplusplus)
@@ -12,13 +12,13 @@ Repository:
 AFL++ is maintained by:
 
 * Marc "van Hauser" Heuse <mh@mh-sec.de>
-* Andrea Fioraldi <andreafioraldi@gmail.com>
 * Dominik Maier <mail@dmnk.co>
-* Heiko "hexcoder-" Eißfeldt <heiko.eissfeldt@hexco.de>
+* Andrea Fioraldi <andreafioraldi@gmail.com>
+* Heiko "hexcoder-" Eissfeldt <heiko.eissfeldt@hexco.de>
 * frida_mode is maintained by @Worksbutnottested
 * Documentation: Jana Aydinbas <jana.aydinbas@gmail.com>
 
-Originally developed by Michał "lcamtuf" Zalewski.
+Originally developed by Michal "lcamtuf" Zalewski.
 
 AFL++ is a superior fork to Google's AFL - more speed, more and better
 mutations, more and better instrumentation, custom module support, etc.
diff --git a/docs/Changelog.md b/docs/Changelog.md
index 76f98547..2c747e42 100644
--- a/docs/Changelog.md
+++ b/docs/Changelog.md
@@ -3,7 +3,7 @@
   This is the list of all noteworthy changes made in every public
   release of the tool. See README.md for the general instruction manual.
 
-### Version ++4.08a (dev)
+### Version ++4.08c (release)
   - afl-fuzz:
     - new mutation engine: mutations that favor discovery more paths are
       prefered until no new finds for 10 minutes then switching to mutations
diff --git a/include/config.h b/include/config.h
index df545583..5a81c4e2 100644
--- a/include/config.h
+++ b/include/config.h
@@ -5,9 +5,9 @@
    Originally written by Michal Zalewski
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                     Heiko Eißfeldt <heiko.eissfeldt@hexco.de>,
-                     Andrea Fioraldi <andreafioraldi@gmail.com>,
                      Dominik Maier <mail@dmnk.co>
+                     Andrea Fioraldi <andreafioraldi@gmail.com>,
+                     Heiko Eissfeldt <heiko.eissfeldt@hexco.de>,
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
    Copyright 2019-2023 AFLplusplus Project. All rights reserved.
@@ -26,7 +26,7 @@
 /* Version string: */
 
 // c = release, a = volatile github dev, e = experimental branch
-#define VERSION "++4.08a"
+#define VERSION "++4.08c"
 
 /******************************************************
  *                                                    *
-- 
cgit 1.4.1


From 9607d1db06ebfc2fe1ba565a0ef0123ab3f3e76c Mon Sep 17 00:00:00 2001
From: marc <marc@marc.srlabs.de>
Date: Thu, 10 Aug 2023 10:56:20 +0200
Subject: v4.09a init

---
 README.md         | 2 +-
 docs/Changelog.md | 4 ++++
 include/config.h  | 2 +-
 3 files changed, 6 insertions(+), 2 deletions(-)

(limited to 'include')

diff --git a/README.md b/README.md
index 951efe59..322ebcf2 100644
--- a/README.md
+++ b/README.md
@@ -4,7 +4,7 @@
 
 Release version: [4.08c](https://github.com/AFLplusplus/AFLplusplus/releases)
 
-GitHub version: 4.08c
+GitHub version: 4.09a
 
 Repository:
 [https://github.com/AFLplusplus/AFLplusplus](https://github.com/AFLplusplus/AFLplusplus)
diff --git a/docs/Changelog.md b/docs/Changelog.md
index 2c747e42..94b4c502 100644
--- a/docs/Changelog.md
+++ b/docs/Changelog.md
@@ -3,6 +3,10 @@
   This is the list of all noteworthy changes made in every public
   release of the tool. See README.md for the general instruction manual.
 
+### Version ++4.09a (dev)
+  - something cool :-)
+
+
 ### Version ++4.08c (release)
   - afl-fuzz:
     - new mutation engine: mutations that favor discovery more paths are
diff --git a/include/config.h b/include/config.h
index 5a81c4e2..6a75737f 100644
--- a/include/config.h
+++ b/include/config.h
@@ -26,7 +26,7 @@
 /* Version string: */
 
 // c = release, a = volatile github dev, e = experimental branch
-#define VERSION "++4.08c"
+#define VERSION "++4.09a"
 
 /******************************************************
  *                                                    *
-- 
cgit 1.4.1


From 8823f22a9c87123c1bfcc5bff10044de4c7a4a1f Mon Sep 17 00:00:00 2001
From: marc <marc@marc.srlabs.de>
Date: Fri, 11 Aug 2023 11:22:18 +0200
Subject: add AFL_FINAL_SYNC

---
 docs/Changelog.md     |  7 +++----
 docs/env_variables.md | 13 +++++++++----
 include/afl-fuzz.h    |  3 ++-
 include/envs.h        |  1 +
 src/afl-fuzz-state.c  |  7 +++++++
 src/afl-fuzz.c        |  9 +++++++++
 6 files changed, 31 insertions(+), 9 deletions(-)

(limited to 'include')

diff --git a/docs/Changelog.md b/docs/Changelog.md
index 94b4c502..8f2b2545 100644
--- a/docs/Changelog.md
+++ b/docs/Changelog.md
@@ -4,7 +4,9 @@
   release of the tool. See README.md for the general instruction manual.
 
 ### Version ++4.09a (dev)
-  - something cool :-)
+  - afl-fuzz:
+    - added `AFL_FINAL_SYNC` which forces a final fuzzer sync (also for `-F`)
+      before terminating.
 
 
 ### Version ++4.08c (release)
@@ -22,7 +24,6 @@
     - -l X option to enable base64 transformation solving
     - allow to disable CMPLOG with '-c -' (e.g. afl.rs enforces '-c 0' on
       every instance which is counterproductive).
-
   - afl-cmin/afl-cmin.bash:
     - fixed a bug inherited from vanilla AFL where a coverage of
       map[123] = 11 would be the same as map[1123] = 1
@@ -40,7 +41,6 @@
   - qemu_mode:
     - added qemu_mode/utils/qemu_get_symbol_addr.sh
 
-
 ### Version ++4.07c (release)
   - afl-fuzz:
     - reverse reading the seeds only on restarts (increases performance)
@@ -69,7 +69,6 @@
     - TritonDSE in custom_mutators/aflpp_tritondse
     - SymQEMU in custom_mutators/symqemu
 
-
 ### Version ++4.06c (release)
   - afl-fuzz:
     - ensure temporary file descriptor is closed when not used
diff --git a/docs/env_variables.md b/docs/env_variables.md
index affc9e3c..2ce274d3 100644
--- a/docs/env_variables.md
+++ b/docs/env_variables.md
@@ -412,10 +412,15 @@ checks or alter some of the more exotic semantics of the tool:
     set `AFL_IGNORE_PROBLEMS`. If you additionally want to also ignore coverage
     from late loaded libraries, you can set `AFL_IGNORE_PROBLEMS_COVERAGE`.
 
-  - When running in the `-M` or `-S` mode, setting `AFL_IMPORT_FIRST` causes the
-    fuzzer to import test cases from other instances before doing anything else.
-    This makes the "own finds" counter in the UI more accurate. Beyond counter
-    aesthetics, not much else should change.
+  - When running with multiple afl-fuzz or with `-F`,  setting `AFL_IMPORT_FIRST`
+    causes the fuzzer to import test cases from other instances before doing
+    anything else. This makes the "own finds" counter in the UI more accurate.
+
+  - When running with multiple afl-fuzz or with `-F`,  setting `AFL_FINAL_SYNC`
+    will cause the fuzzer to perform a final import of test cases when
+    terminating. This is beneficial for `-M` main fuzzers to ensure it has all
+    unique test cases and hence you only need to `afl-cmin` this single
+    queue.
 
   - Setting `AFL_INPUT_LEN_MIN` and `AFL_INPUT_LEN_MAX` are an alternative to
     the afl-fuzz -g/-G command line option to control the minimum/maximum
diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index ef84a18c..1f89bbd8 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -402,7 +402,8 @@ typedef struct afl_env_vars {
       afl_exit_on_seed_issues, afl_try_affinity, afl_ignore_problems,
       afl_keep_timeouts, afl_no_crash_readme, afl_ignore_timeouts,
       afl_no_startup_calibration, afl_no_warn_instability,
-      afl_post_process_keep_original, afl_crashing_seeds_as_new_crash;
+      afl_post_process_keep_original, afl_crashing_seeds_as_new_crash,
+      afl_final_sync;
 
   u8 *afl_tmpdir, *afl_custom_mutator_library, *afl_python_module, *afl_path,
       *afl_hang_tmout, *afl_forksrv_init_tmout, *afl_preload,
diff --git a/include/envs.h b/include/envs.h
index 0007d5a8..3f5a9e1c 100644
--- a/include/envs.h
+++ b/include/envs.h
@@ -59,6 +59,7 @@ static char *afl_environment_variables[] = {
     "AFL_EXIT_ON_TIME",
     "AFL_EXIT_ON_SEED_ISSUES",
     "AFL_FAST_CAL",
+    "AFL_FINAL_SYNC",
     "AFL_FORCE_UI",
     "AFL_FRIDA_DEBUG_MAPS",
     "AFL_FRIDA_DRIVER_NO_HOOK",
diff --git a/src/afl-fuzz-state.c b/src/afl-fuzz-state.c
index 5a6b95cf..97e00415 100644
--- a/src/afl-fuzz-state.c
+++ b/src/afl-fuzz-state.c
@@ -269,6 +269,13 @@ void read_afl_environment(afl_state_t *afl, char **envp) {
             afl->afl_env.afl_import_first =
                 get_afl_env(afl_environment_variables[i]) ? 1 : 0;
 
+          } else if (!strncmp(env, "AFL_FINAL_SYNC",
+
+                              afl_environment_variable_len)) {
+
+            afl->afl_env.afl_final_sync =
+                get_afl_env(afl_environment_variables[i]) ? 1 : 0;
+
           } else if (!strncmp(env, "AFL_CUSTOM_MUTATOR_ONLY",
 
                               afl_environment_variable_len)) {
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index cdb3f996..c2ec4a1d 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -2899,6 +2899,15 @@ stop_fuzzing:
        time_spent_working / afl->fsrv.total_execs);
   #endif
 
+  if (afl->afl_env.afl_final_sync) {
+
+    SAYF(cYEL "[!] " cRST "\nPerforming final sync, this make take some time ...\n");
+    sync_fuzzers(afl);
+    write_bitmap(afl);
+    SAYF(cYEL "[!] " cRST "Done!\n\n");
+
+  }
+
   if (afl->is_main_node) {
 
     u8 path[PATH_MAX];
-- 
cgit 1.4.1


From 4d8d8633ff39cda2f1d48b66c45e5ae6cd2af477 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Sun, 13 Aug 2023 11:44:37 +0200
Subject: update faq

---
 docs/FAQ.md        | 40 ++++++++++++++++++++++++++++++++++++++--
 include/afl-fuzz.h |  9 ++++-----
 2 files changed, 42 insertions(+), 7 deletions(-)

(limited to 'include')

diff --git a/docs/FAQ.md b/docs/FAQ.md
index 9275eb94..242a379b 100644
--- a/docs/FAQ.md
+++ b/docs/FAQ.md
@@ -29,8 +29,8 @@ If you find an interesting or important question missing, submit it via
   which then implemented their own research and features, making it now by far
   the most flexible and feature rich guided fuzzer available as open source. And
   in independent fuzzing benchmarks it is one of the best fuzzers available,
-  e.g., [Fuzzbench
-  Report](https://www.fuzzbench.com/reports/2020-08-03/index.html).
+  e.g.,
+  [Fuzzbench Report](https://www.fuzzbench.com/reports/2020-08-03/index.html).
 </p></details>
 
 <details>
@@ -103,6 +103,42 @@ If you find an interesting or important question missing, submit it via
   to itself, this too would be an edge.
 </p></details>
 
+<details>
+  <summary id="should-you-ever-stop-afl-fuzz-minimize-the-corpus-and-restart">Should you ever stop afl-fuzz, minimize the corpus and restart?</summary><p>
+
+  To stop afl-fuzz, minimize it's corpus and restart you would usually do:
+
+  ```
+  Control-C  # to terminate afl-fuzz
+  $ afl-cmin -T nproc -i out/default/queue -o minimized_queue -- ./target
+  $ AFL_FAST_CAL=1 AFL_CMPLOG_ONLY_NEW=1 afl-fuzz -i minimized_queue -o out2 [other options] -- ./target
+  ```
+
+  If this improves fuzzing or not is debated and no consensus has been reached
+  or in-depth analysis been performed.
+
+  On the pro side:
+    * The queue/corpus is reduced (up to 20%) by removing intermediate paths
+      that are maybe not needed anymore.
+
+  On the con side:
+    * Fuzzing time is lost for the time the fuzzing is stopped, minimized and
+      restarted.
+
+  The the big question:
+    * Does a minimized queue/corpus improve finding new coverage or does it
+      hinder it?
+
+  The AFL++ team's own limited analysis seem to to show that keeping
+  intermediate paths help to find more coverage, at least for afl-fuzz.
+
+  For honggfuzz in comparison it is a good idea to restart it from time to
+  time if you have other fuzzers (e.g: AFL++) running in parallel to sync
+  the finds of other fuzzers to honggfuzz as it has no syncing feature like
+  AFL++ or libfuzzer.
+
+</p></details>
+
 ## Targets
 
 <details>
diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index 1f89bbd8..3dfd2b2c 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -158,6 +158,7 @@ struct queue_entry {
 
   u8 colorized,                         /* Do not run redqueen stage again  */
       cal_failed;                       /* Calibration failed?              */
+
   bool trim_done,                       /* Trimmed?                         */
       was_fuzzed,                       /* historical, but needed for MOpt  */
       passed_det,                       /* Deterministic stages passed?     */
@@ -169,17 +170,15 @@ struct queue_entry {
       disabled;                         /* Is disabled from fuzz selection  */
 
   u32 bitmap_size,                      /* Number of bits set in bitmap     */
-      fuzz_level,                       /* Number of fuzzing iterations     */
-      n_fuzz_entry                      /* offset in n_fuzz                 */
 #ifdef INTROSPECTION
-      ,
       stats_selected,                   /* stats: how often selected        */
       stats_skipped,                    /* stats: how often skipped         */
       stats_finds,                      /* stats: # of saved finds          */
       stats_crashes,                    /* stats: # of saved crashes        */
-      stats_tmouts                      /* stats: # of saved timeouts       */
+      stats_tmouts,                      /* stats: # of saved timeouts       */
 #endif
-      ;
+      fuzz_level,                       /* Number of fuzzing iterations     */
+      n_fuzz_entry;                      /* offset in n_fuzz                 */
 
   u64 exec_us,                          /* Execution time (us)              */
       handicap,                         /* Number of queue cycles behind    */
-- 
cgit 1.4.1


From 549e5dd9269238ac43ff482d439f7f671946185c Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Wed, 23 Aug 2023 18:02:33 +0200
Subject: AFL_IGNORE_SEED_PROBLEMS

---
 docs/Changelog.md     |  2 ++
 docs/env_variables.md |  3 +++
 include/afl-fuzz.h    |  7 +++----
 include/envs.h        |  1 +
 src/afl-fuzz-init.c   | 53 +++++++++++++++++++++++++++++++++++++++------------
 src/afl-fuzz-state.c  |  7 +++++++
 src/afl-fuzz.c        |  2 ++
 7 files changed, 59 insertions(+), 16 deletions(-)

(limited to 'include')

diff --git a/docs/Changelog.md b/docs/Changelog.md
index 961b2940..87c01f21 100644
--- a/docs/Changelog.md
+++ b/docs/Changelog.md
@@ -7,6 +7,8 @@
   - afl-fuzz:
     - added `AFL_FINAL_SYNC` which forces a final fuzzer sync (also for `-F`)
       before terminating.
+    - added AFL_IGNORE_SEED_PROBLEMS to skip over seeds that time out instead
+      of exiting with an error message
   - afl-whatsup:
     - detect instanced that are starting up and show them as such as not dead
     - now also shows coverage reached
diff --git a/docs/env_variables.md b/docs/env_variables.md
index 2ce274d3..3bb4e844 100644
--- a/docs/env_variables.md
+++ b/docs/env_variables.md
@@ -327,6 +327,9 @@ checks or alter some of the more exotic semantics of the tool:
     (`-i in`). This is an important feature to set when resuming a fuzzing
     session.
 
+  - `AFL_IGNORE_SEED_PROBLEMS` will skip over crashes and timeouts in the seeds
+    instead of exiting.
+
   - Setting `AFL_CRASH_EXITCODE` sets the exit code AFL++ treats as crash. For
     example, if `AFL_CRASH_EXITCODE='-1'` is set, each input resulting in a `-1`
     return code (i.e. `exit(-1)` got called), will be treated as if a crash had
diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index 3dfd2b2c..d02e852e 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -1,4 +1,3 @@
-
 /*
    american fuzzy lop++ - fuzzer header
    ------------------------------------
@@ -175,10 +174,10 @@ struct queue_entry {
       stats_skipped,                    /* stats: how often skipped         */
       stats_finds,                      /* stats: # of saved finds          */
       stats_crashes,                    /* stats: # of saved crashes        */
-      stats_tmouts,                      /* stats: # of saved timeouts       */
+      stats_tmouts,                     /* stats: # of saved timeouts       */
 #endif
       fuzz_level,                       /* Number of fuzzing iterations     */
-      n_fuzz_entry;                      /* offset in n_fuzz                 */
+      n_fuzz_entry;                     /* offset in n_fuzz                 */
 
   u64 exec_us,                          /* Execution time (us)              */
       handicap,                         /* Number of queue cycles behind    */
@@ -402,7 +401,7 @@ typedef struct afl_env_vars {
       afl_keep_timeouts, afl_no_crash_readme, afl_ignore_timeouts,
       afl_no_startup_calibration, afl_no_warn_instability,
       afl_post_process_keep_original, afl_crashing_seeds_as_new_crash,
-      afl_final_sync;
+      afl_final_sync, afl_ignore_seed_problems;
 
   u8 *afl_tmpdir, *afl_custom_mutator_library, *afl_python_module, *afl_path,
       *afl_hang_tmout, *afl_forksrv_init_tmout, *afl_preload,
diff --git a/include/envs.h b/include/envs.h
index 3f5a9e1c..4259d6dd 100644
--- a/include/envs.h
+++ b/include/envs.h
@@ -113,6 +113,7 @@ static char *afl_environment_variables[] = {
     "AFL_I_DONT_CARE_ABOUT_MISSING_CRASHES",
     "AFL_IGNORE_PROBLEMS",
     "AFL_IGNORE_PROBLEMS_COVERAGE",
+    "AFL_IGNORE_SEED_PROBLEMS",
     "AFL_IGNORE_TIMEOUTS",
     "AFL_IGNORE_UNKNOWN_ENVS",
     "AFL_IMPORT_FIRST",
diff --git a/src/afl-fuzz-init.c b/src/afl-fuzz-init.c
index 4c09fab7..9fc0cc57 100644
--- a/src/afl-fuzz-init.c
+++ b/src/afl-fuzz-init.c
@@ -951,19 +951,47 @@ void perform_dry_run(afl_state_t *afl) {
 
         } else {
 
-          SAYF("\n" cLRD "[-] " cRST
-               "The program took more than %u ms to process one of the initial "
-               "test cases.\n"
-               "    This is bad news; raising the limit with the -t option is "
-               "possible, but\n"
-               "    will probably make the fuzzing process extremely slow.\n\n"
+          static int say_once = 0;
+
+          if (!say_once) {
+
+            SAYF(
+                "\n" cLRD "[-] " cRST
+                "The program took more than %u ms to process one of the "
+                "initial "
+                "test cases.\n"
+                "    This is bad news; raising the limit with the -t option is "
+                "possible, but\n"
+                "    will probably make the fuzzing process extremely slow.\n\n"
+
+                "    If this test case is just a fluke, the other option is to "
+                "just avoid it\n"
+                "    altogether, and find one that is less of a CPU hog.\n",
+                afl->fsrv.exec_tmout);
+
+            if (!afl->afl_env.afl_ignore_seed_problems) {
+
+              FATAL("Test case '%s' results in a timeout", fn);
+
+            }
+
+            say_once = 1;
+
+          }
+
+          if (!q->was_fuzzed) {
 
-               "    If this test case is just a fluke, the other option is to "
-               "just avoid it\n"
-               "    altogether, and find one that is less of a CPU hog.\n",
-               afl->fsrv.exec_tmout);
+            q->was_fuzzed = 1;
+            --afl->pending_not_fuzzed;
+            --afl->active_items;
 
-          FATAL("Test case '%s' results in a timeout", fn);
+          }
+
+          q->disabled = 1;
+          q->perf_score = 0;
+
+          WARNF("Test case '%s' results in a timeout, skipping", fn);
+          break;
 
         }
 
@@ -2270,7 +2298,8 @@ void check_crash_handling(void) {
      reporting the awful way. */
 
   #if !TARGET_OS_IPHONE
-  if (system("launchctl list 2>/dev/null | grep -q '\\.ReportCrash\\>'")) return;
+  if (system("launchctl list 2>/dev/null | grep -q '\\.ReportCrash\\>'"))
+    return;
 
   SAYF(
       "\n" cLRD "[-] " cRST
diff --git a/src/afl-fuzz-state.c b/src/afl-fuzz-state.c
index 97e00415..db82536d 100644
--- a/src/afl-fuzz-state.c
+++ b/src/afl-fuzz-state.c
@@ -316,6 +316,13 @@ void read_afl_environment(afl_state_t *afl, char **envp) {
             afl->afl_env.afl_ignore_problems =
                 get_afl_env(afl_environment_variables[i]) ? 1 : 0;
 
+          } else if (!strncmp(env, "AFL_IGNORE_SEED_PROBLEMS",
+
+                              afl_environment_variable_len)) {
+
+            afl->afl_env.afl_ignore_seed_problems =
+                get_afl_env(afl_environment_variables[i]) ? 1 : 0;
+
           } else if (!strncmp(env, "AFL_IGNORE_TIMEOUTS",
 
                               afl_environment_variable_len)) {
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index 43834172..08960ac6 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -275,6 +275,8 @@ static void usage(u8 *argv0, int more_help) {
       "AFL_IGNORE_PROBLEMS: do not abort fuzzing if an incorrect setup is detected\n"
       "AFL_IGNORE_PROBLEMS_COVERAGE: if set in addition to AFL_IGNORE_PROBLEMS - also\n"
       "                              ignore those libs for coverage\n"
+      "AFL_IGNORE_SEED_PROBLEMS: skip over crashes and timeouts in the seeds instead of\n"
+      "                          exiting\n"
       "AFL_IGNORE_TIMEOUTS: do not process or save any timeouts\n"
       "AFL_IGNORE_UNKNOWN_ENVS: don't warn on unknown env vars\n"
       "AFL_IMPORT_FIRST: sync and import test cases from other fuzzer instances first\n"
-- 
cgit 1.4.1


From a809c3c50ce32fde390769b607b020dd68730474 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Sun, 3 Sep 2023 11:22:54 +0200
Subject: less impact

---
 frida_mode/src/instrument/instrument_arm64.c | 14 +++++---------
 include/afl-fuzz.h                           |  1 +
 src/afl-fuzz-one.c                           | 10 ++++++++--
 src/afl-fuzz-queue.c                         | 13 ++++++++++++-
 src/afl-fuzz.c                               | 28 +++++++++++++++++++---------
 5 files changed, 45 insertions(+), 21 deletions(-)

(limited to 'include')

diff --git a/frida_mode/src/instrument/instrument_arm64.c b/frida_mode/src/instrument/instrument_arm64.c
index a0c66697..1147275f 100644
--- a/frida_mode/src/instrument/instrument_arm64.c
+++ b/frida_mode/src/instrument/instrument_arm64.c
@@ -402,17 +402,13 @@ bool instrument_write_inline(GumArm64Writer *cw, GumAddress code_addr,
 
   }
 
-  /* 
-   * The mov instruction supports up to a 16-bit offset. If our offset is out of 
-   * range, then it can end up clobbering the op-code portion of the instruction 
-   * rather than just the operands. So return false and fall back to the 
+  /*
+   * The mov instruction supports up to a 16-bit offset. If our offset is out of
+   * range, then it can end up clobbering the op-code portion of the instruction
+   * rather than just the operands. So return false and fall back to the
    * alternative instrumentation.
    */
-  if (area_offset > UINT16_MAX) {
-
-    return false;
-    
-  }
+  if (area_offset > UINT16_MAX) { return false; }
 
   code.code.mov_x0_curr_loc |= area_offset << 5;
 
diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index d02e852e..217a720a 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -610,6 +610,7 @@ typedef struct afl_state {
 
   u32 stage_cur, stage_max;             /* Stage progression                */
   s32 splicing_with;                    /* Splicing with which test case?   */
+  s64 smallest_favored;                 /* smallest queue id favored        */
 
   u32 main_node_id, main_node_max;      /*   Main instance job splitting    */
 
diff --git a/src/afl-fuzz-one.c b/src/afl-fuzz-one.c
index 2ad4697e..ae39abe8 100644
--- a/src/afl-fuzz-one.c
+++ b/src/afl-fuzz-one.c
@@ -3442,7 +3442,12 @@ abandon_entry:
     --afl->pending_not_fuzzed;
     afl->queue_cur->was_fuzzed = 1;
     afl->reinit_table = 1;
-    if (afl->queue_cur->favored) { --afl->pending_favored; }
+    if (afl->queue_cur->favored) {
+
+      --afl->pending_favored;
+      afl->smallest_favored = -1;
+
+    }
 
   }
 
@@ -5905,7 +5910,8 @@ pacemaker_fuzzing:
 
             --afl->pending_not_fuzzed;
             afl->queue_cur->was_fuzzed = 1;
-            if (afl->queue_cur->favored) { --afl->pending_favored; }
+            if (afl->queue_cur->favored) { --afl->pending_favored;
+        afl->smallest_favored = -1; }
 
           }
 
diff --git a/src/afl-fuzz-queue.c b/src/afl-fuzz-queue.c
index 14ba1ace..5f915c9a 100644
--- a/src/afl-fuzz-queue.c
+++ b/src/afl-fuzz-queue.c
@@ -826,6 +826,8 @@ void cull_queue(afl_state_t *afl) {
   /* Let's see if anything in the bitmap isn't captured in temp_v.
      If yes, and if it has a afl->top_rated[] contender, let's use it. */
 
+  afl->smallest_favored = -1;
+
   for (i = 0; i < afl->fsrv.map_size; ++i) {
 
     if (afl->top_rated[i] && (temp_v[i >> 3] & (1 << (i & 7)))) {
@@ -849,7 +851,16 @@ void cull_queue(afl_state_t *afl) {
         afl->top_rated[i]->favored = 1;
         ++afl->queued_favored;
 
-        if (!afl->top_rated[i]->was_fuzzed) { ++afl->pending_favored; }
+        if (!afl->top_rated[i]->was_fuzzed) {
+
+          ++afl->pending_favored;
+          if (unlikely(afl->smallest_favored > (s64)afl->top_rated[i]->id)) {
+
+            afl->smallest_favored = (s64)afl->top_rated[i]->id;
+
+          }
+
+        }
 
       }
 
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index c8cc7da6..d34b52db 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -2707,20 +2707,30 @@ int main(int argc, char **argv_orig, char **envp) {
 
       if (likely(!afl->old_seed_selection)) {
 
-        if (likely(afl->pending_favored)) {
+        if (likely(afl->pending_favored && afl->smallest_favored >= 0)) {
 
-          for (u32 iter = 0; iter < afl->queued_items; ++iter) {
+          afl->current_entry = afl->smallest_favored;
 
-            if (unlikely(afl->queue_buf[iter]->favored &&
-                !afl->queue_buf[iter]->was_fuzzed)) {
+          /*
 
-              afl->current_entry = iter;
-              afl->queue_cur = afl->queue_buf[afl->current_entry];
-              break;
+                    } else {
 
-            }
+                      for (s32 iter = afl->queued_items - 1; iter >= 0; --iter)
+             {
 
-          }
+                        if (unlikely(afl->queue_buf[iter]->favored &&
+                                     !afl->queue_buf[iter]->was_fuzzed)) {
+
+                          afl->current_entry = iter;
+                          break;
+
+                        }
+
+                      }
+
+          */
+
+          afl->queue_cur = afl->queue_buf[afl->current_entry];
 
         } else {
 
-- 
cgit 1.4.1


From 0b6e74eeb099ac045932e5d3603af899268b48d0 Mon Sep 17 00:00:00 2001
From: Thomas Rooijakkers <thomas.rooijakkers@tno.nl>
Date: Wed, 13 Sep 2023 15:49:04 +0200
Subject: Add support for UTF-8 line rendering

---
 GNUmakefile     |  5 ++++
 docs/INSTALL.md |  1 +
 include/debug.h | 81 ++++++++++++++++++++++++++++++++++++---------------------
 3 files changed, 57 insertions(+), 30 deletions(-)

(limited to 'include')

diff --git a/GNUmakefile b/GNUmakefile
index 88816e85..4a234c51 100644
--- a/GNUmakefile
+++ b/GNUmakefile
@@ -45,6 +45,10 @@ ifdef NO_SPLICING
   override CFLAGS_OPT += -DNO_SPLICING
 endif
 
+ifdef UTF
+  override CFLAGS_OPT += -DFANCY_BOXES_UTF
+endif
+
 ifdef ASAN_BUILD
   $(info Compiling ASAN version of binaries)
   override CFLAGS += $(ASAN_CFLAGS)
@@ -391,6 +395,7 @@ help:
 	@echo INTROSPECTION - compile afl-fuzz with mutation introspection
 	@echo NO_PYTHON - disable python support
 	@echo NO_SPLICING - disables splicing mutation in afl-fuzz, not recommended for normal fuzzing
+	@echo UTF - use UTF-8 for line rendering in status screen
 	@echo NO_NYX - disable building nyx mode dependencies
 	@echo "NO_CORESIGHT - disable building coresight (arm64 only)"
 	@echo NO_UNICORN_ARM64 - disable building unicorn on arm64
diff --git a/docs/INSTALL.md b/docs/INSTALL.md
index 9005a7eb..7c04d7d8 100644
--- a/docs/INSTALL.md
+++ b/docs/INSTALL.md
@@ -87,6 +87,7 @@ These build options exist:
 * INTROSPECTION - compile afl-fuzz with mutation introspection
 * NO_PYTHON - disable python support
 * NO_SPLICING - disables splicing mutation in afl-fuzz, not recommended for normal fuzzing
+* UTF - use UTF-8 for line rendering in status screen
 * NO_NYX - disable building nyx mode dependencies
 * NO_CORESIGHT - disable building coresight (arm64 only)
 * NO_UNICORN_ARM64 - disable building unicorn on arm64
diff --git a/include/debug.h b/include/debug.h
index cd621a72..a9179329 100644
--- a/include/debug.h
+++ b/include/debug.h
@@ -116,42 +116,63 @@
  * Box drawing sequences *
  *************************/
 
-#ifdef FANCY_BOXES
-
-  #define SET_G1 "\x1b)0"                      /* Set G1 for box drawing    */
-  #define RESET_G1 "\x1b)B"                    /* Reset G1 to ASCII         */
-  #define bSTART "\x0e"                        /* Enter G1 drawing mode     */
-  #define bSTOP "\x0f"                         /* Leave G1 drawing mode     */
-  #define bH "q"                               /* Horizontal line           */
-  #define bV "x"                               /* Vertical line             */
-  #define bLT "l"                              /* Left top corner           */
-  #define bRT "k"                              /* Right top corner          */
-  #define bLB "m"                              /* Left bottom corner        */
-  #define bRB "j"                              /* Right bottom corner       */
-  #define bX "n"                               /* Cross                     */
-  #define bVR "t"                              /* Vertical, branch right    */
-  #define bVL "u"                              /* Vertical, branch left     */
-  #define bHT "v"                              /* Horizontal, branch top    */
-  #define bHB "w"                              /* Horizontal, branch bottom */
-
-#else
+#ifdef FANCY_BOXES_UTF
 
   #define SET_G1 ""
   #define RESET_G1 ""
   #define bSTART ""
   #define bSTOP ""
-  #define bH "-"
-  #define bV "|"
-  #define bLT "+"
-  #define bRT "+"
-  #define bLB "+"
-  #define bRB "+"
-  #define bX "+"
-  #define bVR "+"
-  #define bVL "+"
-  #define bHT "+"
-  #define bHB "+"
+  #define bH "\u2500"                          /* Horizontal line           */
+  #define bV "\u2502"                          /* Vertical line             */
+  #define bLT "\u250c"                         /* Left top corner           */
+  #define bRT "\u2510"                         /* Right top corner          */
+  #define bLB "\u2514"                         /* Left bottom corner        */
+  #define bRB "\u2518"                         /* Right bottom corner       */
+  #define bX "\u253c"                          /* Cross                     */
+  #define bVR "\u251c"                         /* Vertical, branch right    */
+  #define bVL "\u2524"                         /* Vertical, branch left     */
+  #define bHT "\u2534"                         /* Horizontal, branch top    */
+  #define bHB "\u252c"                         /* Horizontal, branch bottom */
+
+#else
 
+  #ifdef FANCY_BOXES
+
+    #define SET_G1 "\x1b)0"                    /* Set G1 for box drawing    */
+    #define RESET_G1 "\x1b)B"                  /* Reset G1 to ASCII         */
+    #define bSTART "\x0e"                      /* Enter G1 drawing mode     */
+    #define bSTOP "\x0f"                       /* Leave G1 drawing mode     */
+    #define bH "q"                             /* Horizontal line           */
+    #define bV "x"                             /* Vertical line             */
+    #define bLT "l"                            /* Left top corner           */
+    #define bRT "k"                            /* Right top corner          */
+    #define bLB "m"                            /* Left bottom corner        */
+    #define bRB "j"                            /* Right bottom corner       */
+    #define bX "n"                             /* Cross                     */
+    #define bVR "t"                            /* Vertical, branch right    */
+    #define bVL "u"                            /* Vertical, branch left     */
+    #define bHT "v"                            /* Horizontal, branch top    */
+    #define bHB "w"                            /* Horizontal, branch bottom */
+
+  #else
+
+    #define SET_G1 ""
+    #define RESET_G1 ""
+    #define bSTART ""
+    #define bSTOP ""
+    #define bH "-"
+    #define bV "|"
+    #define bLT "+"
+    #define bRT "+"
+    #define bLB "+"
+    #define bRB "+"
+    #define bX "+"
+    #define bVR "+"
+    #define bVL "+"
+    #define bHT "+"
+    #define bHB "+"
+
+  #endif
 #endif                                                      /* ^FANCY_BOXES */
 
 /***********************
-- 
cgit 1.4.1


From 54f01481571ba3a7c05a5e37b9f5021c1304834e Mon Sep 17 00:00:00 2001
From: Thomas Rooijakkers <thomas.rooijakkers@tno.nl>
Date: Tue, 19 Sep 2023 13:31:29 +0200
Subject: UTF-8 line rendering for status screen as default

---
 GNUmakefile     |  6 +++---
 docs/INSTALL.md |  2 +-
 include/debug.h | 64 ++++++++++++++++++++++++++++-----------------------------
 3 files changed, 36 insertions(+), 36 deletions(-)

(limited to 'include')

diff --git a/GNUmakefile b/GNUmakefile
index 4a234c51..fadf20bd 100644
--- a/GNUmakefile
+++ b/GNUmakefile
@@ -45,8 +45,8 @@ ifdef NO_SPLICING
   override CFLAGS_OPT += -DNO_SPLICING
 endif
 
-ifdef UTF
-  override CFLAGS_OPT += -DFANCY_BOXES_UTF
+ifdef NO_UTF
+  override CFLAGS_OPT += -DFANCY_BOXES_NO_UTF
 endif
 
 ifdef ASAN_BUILD
@@ -395,7 +395,7 @@ help:
 	@echo INTROSPECTION - compile afl-fuzz with mutation introspection
 	@echo NO_PYTHON - disable python support
 	@echo NO_SPLICING - disables splicing mutation in afl-fuzz, not recommended for normal fuzzing
-	@echo UTF - use UTF-8 for line rendering in status screen
+	@echo NO_UTF - do not use UTF-8 for line rendering in status screen (fallback to G1 box drawing, of vanilla AFL)
 	@echo NO_NYX - disable building nyx mode dependencies
 	@echo "NO_CORESIGHT - disable building coresight (arm64 only)"
 	@echo NO_UNICORN_ARM64 - disable building unicorn on arm64
diff --git a/docs/INSTALL.md b/docs/INSTALL.md
index 7c04d7d8..41f512ed 100644
--- a/docs/INSTALL.md
+++ b/docs/INSTALL.md
@@ -87,7 +87,7 @@ These build options exist:
 * INTROSPECTION - compile afl-fuzz with mutation introspection
 * NO_PYTHON - disable python support
 * NO_SPLICING - disables splicing mutation in afl-fuzz, not recommended for normal fuzzing
-* UTF - use UTF-8 for line rendering in status screen
+* NO_UTF - do not use UTF-8 for line rendering in status screen (fallback to G1 box drawing, of vanilla AFL)
 * NO_NYX - disable building nyx mode dependencies
 * NO_CORESIGHT - disable building coresight (arm64 only)
 * NO_UNICORN_ARM64 - disable building unicorn on arm64
diff --git a/include/debug.h b/include/debug.h
index a9179329..234d8fc4 100644
--- a/include/debug.h
+++ b/include/debug.h
@@ -116,43 +116,43 @@
  * Box drawing sequences *
  *************************/
 
-#ifdef FANCY_BOXES_UTF
-
-  #define SET_G1 ""
-  #define RESET_G1 ""
-  #define bSTART ""
-  #define bSTOP ""
-  #define bH "\u2500"                          /* Horizontal line           */
-  #define bV "\u2502"                          /* Vertical line             */
-  #define bLT "\u250c"                         /* Left top corner           */
-  #define bRT "\u2510"                         /* Right top corner          */
-  #define bLB "\u2514"                         /* Left bottom corner        */
-  #define bRB "\u2518"                         /* Right bottom corner       */
-  #define bX "\u253c"                          /* Cross                     */
-  #define bVR "\u251c"                         /* Vertical, branch right    */
-  #define bVL "\u2524"                         /* Vertical, branch left     */
-  #define bHT "\u2534"                         /* Horizontal, branch top    */
-  #define bHB "\u252c"                         /* Horizontal, branch bottom */
+#ifdef FANCY_BOXES_NO_UTF
+
+  #define SET_G1 "\x1b)0"                      /* Set G1 for box drawing    */
+  #define RESET_G1 "\x1b)B"                    /* Reset G1 to ASCII         */
+  #define bSTART "\x0e"                        /* Enter G1 drawing mode     */
+  #define bSTOP "\x0f"                         /* Leave G1 drawing mode     */
+  #define bH "q"                               /* Horizontal line           */
+  #define bV "x"                               /* Vertical line             */
+  #define bLT "l"                              /* Left top corner           */
+  #define bRT "k"                              /* Right top corner          */
+  #define bLB "m"                              /* Left bottom corner        */
+  #define bRB "j"                              /* Right bottom corner       */
+  #define bX "n"                               /* Cross                     */
+  #define bVR "t"                              /* Vertical, branch right    */
+  #define bVL "u"                              /* Vertical, branch left     */
+  #define bHT "v"                              /* Horizontal, branch top    */
+  #define bHB "w"                              /* Horizontal, branch bottom */
 
 #else
 
   #ifdef FANCY_BOXES
 
-    #define SET_G1 "\x1b)0"                    /* Set G1 for box drawing    */
-    #define RESET_G1 "\x1b)B"                  /* Reset G1 to ASCII         */
-    #define bSTART "\x0e"                      /* Enter G1 drawing mode     */
-    #define bSTOP "\x0f"                       /* Leave G1 drawing mode     */
-    #define bH "q"                             /* Horizontal line           */
-    #define bV "x"                             /* Vertical line             */
-    #define bLT "l"                            /* Left top corner           */
-    #define bRT "k"                            /* Right top corner          */
-    #define bLB "m"                            /* Left bottom corner        */
-    #define bRB "j"                            /* Right bottom corner       */
-    #define bX "n"                             /* Cross                     */
-    #define bVR "t"                            /* Vertical, branch right    */
-    #define bVL "u"                            /* Vertical, branch left     */
-    #define bHT "v"                            /* Horizontal, branch top    */
-    #define bHB "w"                            /* Horizontal, branch bottom */
+    #define SET_G1 ""
+    #define RESET_G1 ""
+    #define bSTART ""
+    #define bSTOP ""
+    #define bH "\u2500"                        /* Horizontal line           */
+    #define bV "\u2502"                        /* Vertical line             */
+    #define bLT "\u250c"                       /* Left top corner           */
+    #define bRT "\u2510"                       /* Right top corner          */
+    #define bLB "\u2514"                       /* Left bottom corner        */
+    #define bRB "\u2518"                       /* Right bottom corner       */
+    #define bX "\u253c"                        /* Cross                     */
+    #define bVR "\u251c"                       /* Vertical, branch right    */
+    #define bVL "\u2524"                       /* Vertical, branch left     */
+    #define bHT "\u2534"                       /* Horizontal, branch top    */
+    #define bHB "\u252c"                       /* Horizontal, branch bottom */
 
   #else
 
-- 
cgit 1.4.1


From a6b7da880852ab33e54c4c0eb55570eb5ee4aede Mon Sep 17 00:00:00 2001
From: Thomas Rooijakkers <thomas.rooijakkers@tno.nl>
Date: Wed, 20 Sep 2023 09:41:49 +0200
Subject: Remove Android specifics

---
 include/config.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

(limited to 'include')

diff --git a/include/config.h b/include/config.h
index 6a75737f..988e536e 100644
--- a/include/config.h
+++ b/include/config.h
@@ -120,9 +120,9 @@
 
 // #define _WANT_ORIGINAL_AFL_ALLOC
 
-/* Comment out to disable fancy ANSI boxes and use poor man's 7-bit UI: */
+/* Comment out to disable fancy boxes and use poor man's 7-bit UI: */
 
-#ifndef ANDROID_DISABLE_FANCY  // Fancy boxes are ugly from adb
+#ifndef DISABLE_FANCY
   #define FANCY_BOXES
 #endif
 
-- 
cgit 1.4.1


From 48bff70cdd7fb7aa8333533e01a372205c670a4f Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Fri, 6 Oct 2023 18:53:44 +0200
Subject: add AFL_NO_CFG_FUZZING to env list

---
 TODO.md        | 1 +
 include/envs.h | 1 +
 2 files changed, 2 insertions(+)

(limited to 'include')

diff --git a/TODO.md b/TODO.md
index ac24fe07..12da6026 100644
--- a/TODO.md
+++ b/TODO.md
@@ -2,6 +2,7 @@
 
 ## Should
 
+ - cmplog rtn sanity check on fixed length
  - afl-showmap -f support
  - afl-fuzz multicore wrapper script
  - add value_profile but only enable after 15 minutes without finds
diff --git a/include/envs.h b/include/envs.h
index 4259d6dd..734b1707 100644
--- a/include/envs.h
+++ b/include/envs.h
@@ -179,6 +179,7 @@ static char *afl_environment_variables[] = {
     "AFL_NO_COLOUR",
 #endif
     "AFL_NO_CPU_RED",
+    "AFL_NO_CFG_FUZZING",  // afl.rs rust crate option
     "AFL_NO_CRASH_README",
     "AFL_NO_FORKSRV",
     "AFL_NO_UI",
-- 
cgit 1.4.1


From bfb841d01383a4801a28b007c5f7039f2f28bef9 Mon Sep 17 00:00:00 2001
From: "Christian Holler (:decoder)" <choller@mozilla.com>
Date: Fri, 20 Oct 2023 00:07:35 +0200
Subject: Use proper AFL_NYX_AUX_SIZE for nyx_aux_string

---
 include/forkserver.h  |  1 +
 src/afl-forkserver.c  | 12 +++++++++---
 src/afl-fuzz-bitmap.c |  2 +-
 3 files changed, 11 insertions(+), 4 deletions(-)

(limited to 'include')

diff --git a/include/forkserver.h b/include/forkserver.h
index 5e498c56..f6230fe8 100644
--- a/include/forkserver.h
+++ b/include/forkserver.h
@@ -197,6 +197,7 @@ typedef struct afl_forkserver {
   u32                   nyx_id;          /* nyx runner id (0 -> master)      */
   u32                   nyx_bind_cpu_id; /* nyx runner cpu id                */
   char                 *nyx_aux_string;
+  u32                   nyx_aux_string_len;
   bool                  nyx_use_tmp_workdir;
   char                 *nyx_tmp_workdir_path;
   s32                   nyx_log_fd;
diff --git a/src/afl-forkserver.c b/src/afl-forkserver.c
index 07f5a1a9..9b710733 100644
--- a/src/afl-forkserver.c
+++ b/src/afl-forkserver.c
@@ -615,8 +615,10 @@ void afl_fsrv_start(afl_forkserver_t *fsrv, char **argv,
 
     if (getenv("AFL_NYX_AUX_SIZE") != NULL) {
 
+      fsrv->nyx_aux_string_len = atoi(getenv("AFL_NYX_AUX_SIZE"));
+
       if (fsrv->nyx_handlers->nyx_config_set_aux_buffer_size(
-              nyx_config, atoi(getenv("AFL_NYX_AUX_SIZE"))) != 1) {
+              nyx_config, fsrv->nyx_aux_string_len) != 1) {
 
         NYX_PRE_FATAL(fsrv,
                       "Invalid AFL_NYX_AUX_SIZE value set (must be a multiple "
@@ -624,6 +626,10 @@ void afl_fsrv_start(afl_forkserver_t *fsrv, char **argv,
 
       }
 
+    } else {
+
+      fsrv->nyx_aux_string_len = 0x1000;
+
     }
 
     if (getenv("AFL_NYX_REUSE_SNAPSHOT") != NULL) {
@@ -697,8 +703,8 @@ void afl_fsrv_start(afl_forkserver_t *fsrv, char **argv,
     fsrv->nyx_handlers->nyx_option_set_timeout(fsrv->nyx_runner, 2, 0);
     fsrv->nyx_handlers->nyx_option_apply(fsrv->nyx_runner);
 
-    fsrv->nyx_aux_string = malloc(0x1000);
-    memset(fsrv->nyx_aux_string, 0, 0x1000);
+    fsrv->nyx_aux_string = malloc(fsrv->nyx_aux_string_len);
+    memset(fsrv->nyx_aux_string, 0, fsrv->nyx_aux_string_len);
 
     /* dry run */
     fsrv->nyx_handlers->nyx_set_afl_input(fsrv->nyx_runner, "INIT", 4);
diff --git a/src/afl-fuzz-bitmap.c b/src/afl-fuzz-bitmap.c
index 0429db34..d76158ce 100644
--- a/src/afl-fuzz-bitmap.c
+++ b/src/afl-fuzz-bitmap.c
@@ -866,7 +866,7 @@ save_if_interesting(afl_state_t *afl, void *mem, u32 len, u8 fault) {
     if (unlikely(fd < 0)) { PFATAL("Unable to create '%s'", fn_log); }
 
     u32 nyx_aux_string_len = afl->fsrv.nyx_handlers->nyx_get_aux_string(
-        afl->fsrv.nyx_runner, afl->fsrv.nyx_aux_string, 0x1000);
+        afl->fsrv.nyx_runner, afl->fsrv.nyx_aux_string, afl->fsrv.nyx_aux_string_len);
 
     ck_write(fd, afl->fsrv.nyx_aux_string, nyx_aux_string_len, fn_log);
     close(fd);
-- 
cgit 1.4.1


From 4cdf7a1e3e351f10537683d49a08181b6c1576cc Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Mon, 23 Oct 2023 18:03:59 +0200
Subject: add scale encode mode for cmplog

---
 include/afl-fuzz.h      |   2 +-
 src/afl-fuzz-redqueen.c | 263 +++++++++++++++++++++++++++++++++++++++++++-----
 src/afl-fuzz.c          |   4 +
 3 files changed, 244 insertions(+), 25 deletions(-)

(limited to 'include')

diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index 217a720a..8112d430 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -675,7 +675,7 @@ typedef struct afl_state {
   u32 cmplog_max_filesize;
   u32 cmplog_lvl;
   u32 colorize_success;
-  u8  cmplog_enable_arith, cmplog_enable_transform,
+  u8  cmplog_enable_arith, cmplog_enable_transform, cmplog_enable_scale,
       cmplog_enable_xtreme_transform, cmplog_random_colorization;
 
   struct afl_pass_stat *pass_stats;
diff --git a/src/afl-fuzz-redqueen.c b/src/afl-fuzz-redqueen.c
index db4991db..b6c54df2 100644
--- a/src/afl-fuzz-redqueen.c
+++ b/src/afl-fuzz-redqueen.c
@@ -40,7 +40,7 @@ enum {
   IS_FP = 8,       // is a floating point, not an integer
   /* --- below are internal settings, not from target cmplog */
   IS_FP_MOD = 16,    // arithemtic changed floating point
-  IS_INT_MOD = 32,   // arithmetic changed interger
+  IS_INT_MOD = 32,   // arithmetic changed integer
   IS_TRANSFORM = 64  // transformed integer
 
 };
@@ -775,6 +775,13 @@ static u32 to_base64(u8 *src, u8 *dst, u32 dst_len) {
 
 }
 
+#ifdef WORD_SIZE_64
+static u8 cmp_extend_encodingN(afl_state_t *afl, struct cmp_header *h,
+                               u128 pattern, u128 repl, u128 o_pattern,
+                               u128 changed_val, u8 attr, u32 idx,
+                               u32 taint_len, u8 *orig_buf, u8 *buf, u8 *cbuf,
+                               u32 len, u8 do_reverse, u8 lvl, u8 *status);
+#endif
 static u8 cmp_extend_encoding(afl_state_t *afl, struct cmp_header *h,
                               u64 pattern, u64 repl, u64 o_pattern,
                               u64 changed_val, u8 attr, u32 idx, u32 taint_len,
@@ -807,6 +814,29 @@ static u8 cmp_extend_encoding(afl_state_t *afl, struct cmp_header *h,
           hshape, attr);
   */
 
+  u8 bytes;
+
+  switch (hshape) {
+
+    case 0:
+    case 1:
+      bytes = 1;
+      break;
+    case 2:
+      bytes = 2;
+      break;
+    case 3:
+    case 4:
+      bytes = 4;
+      break;
+    default:
+      bytes = 8;
+
+  }
+
+  // necessary for preventing heap access overflow
+  bytes = MIN(bytes, len - idx);
+
   //  reverse atoi()/strnu?toll() is expensive, so we only to it in lvl 3
   if (afl->cmplog_enable_transform && (lvl & LVL3)) {
 
@@ -895,29 +925,6 @@ static u8 cmp_extend_encoding(afl_state_t *afl, struct cmp_header *h,
     if (pattern != o_pattern && repl == changed_val && attr <= IS_EQUAL) {
 
       u64 b_val, o_b_val, mask;
-      u8  bytes;
-
-      switch (hshape) {
-
-        case 0:
-        case 1:
-          bytes = 1;
-          break;
-        case 2:
-          bytes = 2;
-          break;
-        case 3:
-        case 4:
-          bytes = 4;
-          break;
-        default:
-          bytes = 8;
-
-      }
-
-      // necessary for preventing heap access overflow
-      bytes = MIN(bytes, len - idx);
-
       switch (bytes) {
 
         case 0:                        // cannot happen
@@ -1285,6 +1292,125 @@ static u8 cmp_extend_encoding(afl_state_t *afl, struct cmp_header *h,
 
   }
 
+  // If 'S' is set for cmplog mode then we try a scale encoding of the value.
+  // Currently we can only handle bytes up to 1 << 55
+
+  if (attr < IS_FP && attr < 32 && (afl->cmplog_enable_scale || lvl >= LVL3)) {
+
+    u8  do_call = 1;
+    u64 new_val = repl << 2;
+    u32 saved_hshape = hshape;
+
+    if (changed_val <= 255) {
+
+      // nothing
+
+    } else if (new_val <= 65535) {
+
+      new_val += 1;  // two byte mode
+      hshape = 2;
+
+    } else if (new_val <= 4294967295) {
+
+      new_val += 2;  // four byte mode
+      hshape = 4;
+
+    } else {
+
+#ifndef WORD_SIZE_64
+      if (repl <= 0x00ffffffffffffff {
+
+        new_val = repl << 8;
+        u8  scale_len = 0;
+        u64 tmp_val = repl;
+        while (tmp_val) {
+
+          tmp_val >>= 8;
+          ++scale_len;
+
+        }  // scale_len will be >= 4;
+
+        if (scale_len >= 4) {
+
+          scale_len -= 4;
+
+        } else {
+
+          scale_len = 0;
+
+        };
+
+        new_val += (scale_len << 2) + 3;
+        hshape = 8;
+
+      } else {
+
+        do_call = 0;
+
+      }
+
+#else
+      {
+
+        u128 new_val = ((u128)repl) << 8;
+        u8   scale_len = 0;
+        u128 tmp_val = (u128)repl;
+
+        while (tmp_val) {
+
+          tmp_val >>= 8;
+          ++scale_len;
+
+        }  // scale_len will be >= 4;
+
+        if (scale_len >= 4) {
+
+          scale_len -= 4;
+
+        } else {
+
+          scale_len = 0;
+
+        };
+
+        new_val += (scale_len << 2) + 3;
+        hshape = scale_len + 5;
+
+        if (unlikely(cmp_extend_encodingN(afl, h, (u128)pattern, new_val,
+                                          (u128)o_pattern, (u128)changed_val,
+                                          32, idx, taint_len, orig_buf, buf,
+                                          cbuf, len, 1, lvl, status))) {
+
+          hshape = saved_hshape;
+          return 1;
+
+        }
+
+        do_call = 0;
+
+      }
+
+#endif
+
+    }
+
+    if (do_call) {
+
+      if (unlikely(cmp_extend_encoding(
+              afl, h, pattern, new_val, o_pattern, changed_val, 32, idx,
+              taint_len, orig_buf, buf, cbuf, len, 1, lvl, status))) {
+
+        hshape = saved_hshape;
+        return 1;
+
+      }
+
+    }
+
+    hshape = saved_hshape;
+
+  }
+
   // here we add and subract 1 from the value, but only if it is not an
   // == or != comparison
   // Bits: 1 = Equal, 2 = Greater, 4 = Lesser, 8 = Float
@@ -1551,6 +1677,95 @@ static u8 cmp_extend_encodingN(afl_state_t *afl, struct cmp_header *h,
 
     }
 
+    if (attr < IS_FP && attr < 32 &&
+        (afl->cmplog_enable_scale || lvl >= LVL3)) {
+
+      u128 new_val = repl << 2;
+      u128 max_scale = (u128)1 << 120;
+      u32  saved_hshape = hshape;
+
+      if (new_val <= 255) {
+
+        hshape = 1;
+        if (unlikely(cmp_extend_encoding(afl, h, (u64)pattern, new_val,
+                                         (u64)o_pattern, (u64)changed_val, 32,
+                                         idx, taint_len, orig_buf, buf, cbuf,
+                                         len, 1, lvl, status))) {
+
+          hshape = saved_hshape;
+          return 1;
+
+        }
+
+      } else if (new_val <= 65535) {
+
+        new_val += 1;  // two byte mode
+        hshape = 2;
+        if (unlikely(cmp_extend_encoding(afl, h, (u64)pattern, new_val,
+                                         (u64)o_pattern, (u64)changed_val, 32,
+                                         idx, taint_len, orig_buf, buf, cbuf,
+                                         len, 1, lvl, status))) {
+
+          hshape = saved_hshape;
+          return 1;
+
+        }
+
+      } else if (new_val <= 4294967295) {
+
+        new_val += 2;  // four byte mode
+        hshape = 4;
+        if (unlikely(cmp_extend_encoding(afl, h, (u64)pattern, new_val,
+                                         (u64)o_pattern, (u64)changed_val, 32,
+                                         idx, taint_len, orig_buf, buf, cbuf,
+                                         len, 1, lvl, status))) {
+
+          hshape = saved_hshape;
+          return 1;
+
+        }
+
+      } else if (repl < max_scale) {
+
+        u128 new_val = (u128)repl << 8;
+        u8   scale_len = 0;
+        u128 tmp_val = (u128)repl;
+        while (tmp_val) {
+
+          tmp_val >>= 8;
+          ++scale_len;
+
+        }  // scale_len will be >= 4;
+
+        if (scale_len >= 4) {
+
+          scale_len -= 4;
+
+        } else {
+
+          scale_len = 0;
+
+        };
+
+        new_val += (scale_len << 2) + 3;
+        hshape = scale_len + 5;
+
+        if (unlikely(cmp_extend_encodingN(afl, h, (u128)pattern, new_val,
+                                          (u128)o_pattern, (u128)changed_val,
+                                          32, idx, taint_len, orig_buf, buf,
+                                          cbuf, len, 1, lvl, status))) {
+
+          hshape = saved_hshape;
+          return 1;
+
+        }
+
+      }
+
+      hshape = saved_hshape;
+
+    }
+
   }
 
   return 0;
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index 2538f4a4..9fdd2193 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -1152,6 +1152,10 @@ int main(int argc, char **argv_orig, char **envp) {
             case 'A':
               afl->cmplog_enable_arith = 1;
               break;
+            case 's':
+            case 'S':
+              afl->cmplog_enable_scale = 1;
+              break;
             case 't':
             case 'T':
               afl->cmplog_enable_transform = 1;
-- 
cgit 1.4.1


From cf458a7d25dc3448b94ffe08d3d89531fc8d4818 Mon Sep 17 00:00:00 2001
From: Jesse Schwartzentruber <truber@mozilla.com>
Date: Thu, 19 Oct 2023 17:14:31 -0400
Subject: Add an env to afl-clang-fast to disable setting rpath if LLVM path
 isn't recognized

---
 docs/env_variables.md |  6 ++++++
 include/envs.h        |  1 +
 src/afl-cc.c          | 27 ++++++++++++++++-----------
 3 files changed, 23 insertions(+), 11 deletions(-)

(limited to 'include')

diff --git a/docs/env_variables.md b/docs/env_variables.md
index 3bb4e844..a7636511 100644
--- a/docs/env_variables.md
+++ b/docs/env_variables.md
@@ -135,6 +135,12 @@ subset of the settings discussed in section 1, with the exception of:
   - `TMPDIR` and `AFL_KEEP_ASSEMBLY`, since no temporary assembly files are
     created.
 
+  - LLVM modes compiling C++ will normally set rpath in the binary if LLVM is
+    not in a usual location (/usr or /lib). Setting `AFL_LLVM_NO_RPATH=1`
+    disables this behaviour in case it isn't desired. For example, the compiling
+    toolchain might be in a custom location, but the target machine has LLVM
+    runtime libs in the search path.
+
 Then there are a few specific features that are only available in
 instrumentation mode:
 
diff --git a/include/envs.h b/include/envs.h
index 734b1707..93e49e34 100644
--- a/include/envs.h
+++ b/include/envs.h
@@ -162,6 +162,7 @@ static char *afl_environment_variables[] = {
     "AFL_LLVM_MAP_DYNAMIC",
     "AFL_LLVM_NGRAM_SIZE",
     "AFL_NGRAM_SIZE",
+    "AFL_LLVM_NO_RPATH",
     "AFL_LLVM_NOT_ZERO",
     "AFL_LLVM_INSTRUMENT_FILE",
     "AFL_LLVM_THREADSAFE_INST",
diff --git a/src/afl-cc.c b/src/afl-cc.c
index 037a5c30..5f8f278f 100644
--- a/src/afl-cc.c
+++ b/src/afl-cc.c
@@ -1144,19 +1144,22 @@ static void edit_params(u32 argc, char **argv, char **envp) {
 
   if (!have_pic) { cc_params[cc_par_cnt++] = "-fPIC"; }
 
-  // in case LLVM is installed not via a package manager or "make install"
-  // e.g. compiled download or compiled from github then its ./lib directory
-  // might not be in the search path. Add it if so.
-  u8 *libdir = strdup(LLVM_LIBDIR);
-  if (plusplus_mode && strlen(libdir) && strncmp(libdir, "/usr", 4) &&
-      strncmp(libdir, "/lib", 4)) {
+  if (!getenv("AFL_LLVM_NO_RPATH")) {
+    // in case LLVM is installed not via a package manager or "make install"
+    // e.g. compiled download or compiled from github then its ./lib directory
+    // might not be in the search path. Add it if so.
+    u8 *libdir = strdup(LLVM_LIBDIR);
+    if (plusplus_mode && strlen(libdir) && strncmp(libdir, "/usr", 4) &&
+        strncmp(libdir, "/lib", 4)) {
 
-    cc_params[cc_par_cnt++] = "-Wl,-rpath";
-    cc_params[cc_par_cnt++] = libdir;
+      cc_params[cc_par_cnt++] = "-Wl,-rpath";
+      cc_params[cc_par_cnt++] = libdir;
 
-  } else {
+    } else {
 
-    free(libdir);
+      free(libdir);
+
+    }
 
   }
 
@@ -2289,7 +2292,9 @@ int main(int argc, char **argv, char **envp) {
             "  AFL_LLVM_CTX: use full context sensitive coverage (for "
             "CLASSIC)\n"
             "  AFL_LLVM_NGRAM_SIZE: use ngram prev_loc count coverage (for "
-            "CLASSIC)\n");
+            "CLASSIC)\n"
+            "  AFL_LLVM_NO_RPATH: disable rpath setting for custom LLVM "
+            "locations\n");
 
 #ifdef AFL_CLANG_FLTO
       if (have_lto)
-- 
cgit 1.4.1


From f3d2127fd815bed2ec9dfab981123898d11cea65 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Mon, 6 Nov 2023 10:13:59 +0100
Subject: clang-format 16->17

---
 .custom-format.py                           |  2 +-
 frida_mode/src/main.c                       |  8 +++---
 include/afl-mutations.h                     |  2 +-
 include/xxhash.h                            | 30 +++++++++++------------
 instrumentation/cmplog-instructions-pass.cc |  2 +-
 instrumentation/cmplog-routines-pass.cc     |  2 +-
 instrumentation/cmplog-switches-pass.cc     |  2 +-
 instrumentation/split-switches-pass.so.cc   |  2 +-
 qemu_mode/libqasan/dlmalloc.c               | 38 ++++++++++++++---------------
 qemu_mode/libqasan/malloc.c                 |  4 +--
 src/afl-fuzz-one.c                          | 30 +++++++++++------------
 src/afl-fuzz-redqueen.c                     |  2 +-
 utils/libtokencap/libtokencap.so.c          |  6 ++---
 13 files changed, 65 insertions(+), 65 deletions(-)

(limited to 'include')

diff --git a/.custom-format.py b/.custom-format.py
index 3521c05d..c8075ace 100755
--- a/.custom-format.py
+++ b/.custom-format.py
@@ -24,7 +24,7 @@ import importlib.metadata
 
 # string_re = re.compile('(\\"(\\\\.|[^"\\\\])*\\")') # TODO: for future use
 
-CURRENT_LLVM = os.getenv('LLVM_VERSION', 16)
+CURRENT_LLVM = os.getenv('LLVM_VERSION', 17)
 CLANG_FORMAT_BIN = os.getenv("CLANG_FORMAT_BIN", "")
 
 
diff --git a/frida_mode/src/main.c b/frida_mode/src/main.c
index f11c4b25..bd7b1351 100644
--- a/frida_mode/src/main.c
+++ b/frida_mode/src/main.c
@@ -49,10 +49,10 @@ extern void __libc_init(void *raw_args, void (*onexit)(void) __unused,
                         int (*slingshot)(int, char **, char **),
                         structors_array_t const *const structors);
 #else
-extern int  __libc_start_main(int (*main)(int, char **, char **), int argc,
-                              char **ubp_av, void (*init)(void),
-                              void (*fini)(void), void (*rtld_fini)(void),
-                              void(*stack_end));
+extern int __libc_start_main(int (*main)(int, char **, char **), int argc,
+                             char **ubp_av, void (*init)(void),
+                             void (*fini)(void), void (*rtld_fini)(void),
+                             void(*stack_end));
 #endif
 
 typedef int (*main_fn_t)(int argc, char **argv, char **envp);
diff --git a/include/afl-mutations.h b/include/afl-mutations.h
index 98ba6fcf..d709b90d 100644
--- a/include/afl-mutations.h
+++ b/include/afl-mutations.h
@@ -1854,7 +1854,7 @@ inline u32 afl_mutate(afl_state_t *afl, u8 *buf, u32 len, u32 steps,
 
   for (u32 step = 0; step < steps; ++step) {
 
-  retry_havoc_step : {
+  retry_havoc_step: {
 
     u32 r = rand_below(afl, MUT_STRATEGY_ARRAY_SIZE), item;
 
diff --git a/include/xxhash.h b/include/xxhash.h
index 7bc0a14e..a8bd6f27 100644
--- a/include/xxhash.h
+++ b/include/xxhash.h
@@ -365,7 +365,7 @@ typedef uint32_t XXH32_hash_t;
       (defined(__cplusplus) || \
        (defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */))
     #include <stdint.h>
-typedef uint32_t      XXH32_hash_t;
+typedef uint32_t XXH32_hash_t;
 
   #else
     #include <limits.h>
@@ -1082,7 +1082,7 @@ struct XXH64_state_s {
       #include <stdalign.h>
       #define XXH_ALIGN(n) alignas(n)
     #elif defined(__cplusplus) && (__cplusplus >= 201103L)      /* >= C++11 */
-      /* In C++ alignas() is a keyword */
+    /* In C++ alignas() is a keyword */
       #define XXH_ALIGN(n) alignas(n)
     #elif defined(__GNUC__)
       #define XXH_ALIGN(n) __attribute__((aligned(n)))
@@ -3031,8 +3031,8 @@ XXH64_hashFromCanonical(const XXH64_canonical_t *src) {
           __STDC_VERSION__ >= 199901L                             /* >= C99 */
         #define XXH_RESTRICT restrict
       #else
-        /* Note: it might be useful to define __restrict or __restrict__ for
-         * some C++ compilers */
+      /* Note: it might be useful to define __restrict or __restrict__ for
+       * some C++ compilers */
         #define XXH_RESTRICT                                     /* disable */
       #endif
 
@@ -3492,8 +3492,8 @@ XXH_FORCE_INLINE xxh_u64x2 XXH_vec_loadu(const void *ptr) {
           #define XXH_vec_mulo vec_mulo
           #define XXH_vec_mule vec_mule
         #elif defined(__clang__) && XXH_HAS_BUILTIN(__builtin_altivec_vmuleuw)
-          /* Clang has a better way to control this, we can just use the builtin
-           * which doesn't swap. */
+        /* Clang has a better way to control this, we can just use the builtin
+         * which doesn't swap. */
           #define XXH_vec_mulo __builtin_altivec_vmulouw
           #define XXH_vec_mule __builtin_altivec_vmuleuw
         #else
@@ -3604,15 +3604,15 @@ XXH_FORCE_INLINE xxh_u64 XXH_mult32to64(xxh_u64 x, xxh_u64 y) {
         #include <intrin.h>
         #define XXH_mult32to64(x, y) __emulu((unsigned)(x), (unsigned)(y))
       #else
-        /*
-         * Downcast + upcast is usually better than masking on older compilers
-         * like GCC 4.2 (especially 32-bit ones), all without affecting newer
-         * compilers.
-         *
-         * The other method, (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF), will AND both
-         * operands and perform a full 64x64 multiply -- entirely redundant on
-         * 32-bit.
-         */
+      /*
+       * Downcast + upcast is usually better than masking on older compilers
+       * like GCC 4.2 (especially 32-bit ones), all without affecting newer
+       * compilers.
+       *
+       * The other method, (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF), will AND both
+       * operands and perform a full 64x64 multiply -- entirely redundant on
+       * 32-bit.
+       */
         #define XXH_mult32to64(x, y) \
           ((xxh_u64)(xxh_u32)(x) * (xxh_u64)(xxh_u32)(y))
       #endif
diff --git a/instrumentation/cmplog-instructions-pass.cc b/instrumentation/cmplog-instructions-pass.cc
index bca1f927..9cd1dc59 100644
--- a/instrumentation/cmplog-instructions-pass.cc
+++ b/instrumentation/cmplog-instructions-pass.cc
@@ -90,7 +90,7 @@ class CmpLogInstructions : public ModulePass {
 #if LLVM_MAJOR >= 11                                /* use new pass manager */
   PreservedAnalyses run(Module &M, ModuleAnalysisManager &MAM);
 #else
-  bool      runOnModule(Module &M) override;
+  bool runOnModule(Module &M) override;
 
   #if LLVM_VERSION_MAJOR >= 4
   StringRef getPassName() const override {
diff --git a/instrumentation/cmplog-routines-pass.cc b/instrumentation/cmplog-routines-pass.cc
index c3fbed8d..54e9ddf3 100644
--- a/instrumentation/cmplog-routines-pass.cc
+++ b/instrumentation/cmplog-routines-pass.cc
@@ -85,7 +85,7 @@ class CmpLogRoutines : public ModulePass {
 #if LLVM_VERSION_MAJOR >= 11                        /* use new pass manager */
   PreservedAnalyses run(Module &M, ModuleAnalysisManager &MAM);
 #else
-  bool      runOnModule(Module &M) override;
+  bool runOnModule(Module &M) override;
 
   #if LLVM_VERSION_MAJOR >= 4
   StringRef getPassName() const override {
diff --git a/instrumentation/cmplog-switches-pass.cc b/instrumentation/cmplog-switches-pass.cc
index 38de669d..01da6da7 100644
--- a/instrumentation/cmplog-switches-pass.cc
+++ b/instrumentation/cmplog-switches-pass.cc
@@ -85,7 +85,7 @@ class CmplogSwitches : public ModulePass {
 #if LLVM_VERSION_MAJOR >= 11                        /* use new pass manager */
   PreservedAnalyses run(Module &M, ModuleAnalysisManager &MAM);
 #else
-  bool        runOnModule(Module &M) override;
+  bool runOnModule(Module &M) override;
 
   #if LLVM_VERSION_MAJOR < 4
   const char *getPassName() const override {
diff --git a/instrumentation/split-switches-pass.so.cc b/instrumentation/split-switches-pass.so.cc
index dcd89652..e3dfea0d 100644
--- a/instrumentation/split-switches-pass.so.cc
+++ b/instrumentation/split-switches-pass.so.cc
@@ -84,7 +84,7 @@ class SplitSwitchesTransform : public ModulePass {
 #if LLVM_VERSION_MAJOR >= 11                        /* use new pass manager */
   PreservedAnalyses run(Module &M, ModuleAnalysisManager &MAM);
 #else
-  bool      runOnModule(Module &M) override;
+  bool runOnModule(Module &M) override;
 
   #if LLVM_VERSION_MAJOR >= 4
   StringRef getPassName() const override {
diff --git a/qemu_mode/libqasan/dlmalloc.c b/qemu_mode/libqasan/dlmalloc.c
index b459eb7b..1919ae26 100644
--- a/qemu_mode/libqasan/dlmalloc.c
+++ b/qemu_mode/libqasan/dlmalloc.c
@@ -771,8 +771,8 @@ MAX_RELEASE_CHECK_RATE   default: 4095 unless not HAVE_MMAP
       #include "/usr/include/malloc.h"
     #else                                      /* HAVE_USR_INCLUDE_MALLOC_H */
       #ifndef STRUCT_MALLINFO_DECLARED
-        /* HP-UX (and others?) redefines mallinfo unless _STRUCT_MALLINFO is
-         * defined */
+      /* HP-UX (and others?) redefines mallinfo unless _STRUCT_MALLINFO is
+       * defined */
         #define _STRUCT_MALLINFO
         #define STRUCT_MALLINFO_DECLARED 1
 struct mallinfo {
@@ -1660,10 +1660,10 @@ extern size_t getpagesize();
   #define is_aligned(A) (((size_t)((A)) & (CHUNK_ALIGN_MASK)) == 0)
 
   /* the number of bytes to offset an address to align it */
-  #define align_offset(A)                                         \
-    ((((size_t)(A)&CHUNK_ALIGN_MASK) == 0)                        \
-         ? 0                                                      \
-         : ((MALLOC_ALIGNMENT - ((size_t)(A)&CHUNK_ALIGN_MASK)) & \
+  #define align_offset(A)                                           \
+    ((((size_t)(A) & CHUNK_ALIGN_MASK) == 0)                        \
+         ? 0                                                        \
+         : ((MALLOC_ALIGNMENT - ((size_t)(A) & CHUNK_ALIGN_MASK)) & \
             CHUNK_ALIGN_MASK))
 
   /* -------------------------- MMAP preliminaries ------------------------- */
@@ -1715,10 +1715,10 @@ static FORCEINLINE int unixmunmap(void *ptr, size_t size) {
         #define MUNMAP_DEFAULT(a, s) unixmunmap((a), (s))
 
       #else                                                /* MAP_ANONYMOUS */
-        /*
-           Nearly all versions of mmap support MAP_ANONYMOUS, so the following
-           is unlikely to be needed, but is supplied just in case.
-        */
+      /*
+         Nearly all versions of mmap support MAP_ANONYMOUS, so the following
+         is unlikely to be needed, but is supplied just in case.
+      */
         #define MMAP_FLAGS (MAP_PRIVATE)
 static int dev_zero_fd = -1;       /* Cached file descriptor for /dev/zero. */
         #define MMAP_DEFAULT(s)                                        \
@@ -1965,7 +1965,7 @@ static FORCEINLINE void x86_clear_lock(int *sl) {
 
       #endif                                     /* ... gcc spins locks ... */
 
-      /* How to yield for a spin lock */
+    /* How to yield for a spin lock */
       #define SPINS_PER_YIELD 63
       #if defined(_MSC_VER)
         #define SLEEP_EX_DURATION 50               /* delay for yield/sleep */
@@ -2008,11 +2008,11 @@ static MLOCK_T malloc_global_mutex = 0;
           #define CURRENT_THREAD GetCurrentThreadId()
           #define EQ_OWNER(X, Y) ((X) == (Y))
         #else
-          /*
-            Note: the following assume that pthread_t is a type that can be
-            initialized to (casted) zero. If this is not the case, you will need
-            to somehow redefine these or not use spin locks.
-          */
+        /*
+          Note: the following assume that pthread_t is a type that can be
+          initialized to (casted) zero. If this is not the case, you will need
+          to somehow redefine these or not use spin locks.
+        */
           #define THREAD_ID_T pthread_t
           #define CURRENT_THREAD pthread_self()
           #define EQ_OWNER(X, Y) pthread_equal(X, Y)
@@ -2169,7 +2169,7 @@ static int pthread_init_lock(MLOCK_T *lk) {
 
     #endif                                            /* ... lock types ... */
 
-    /* Common code for all lock types */
+  /* Common code for all lock types */
     #define USE_LOCK_BIT (2U)
 
     #ifndef ACQUIRE_MALLOC_GLOBAL_LOCK
@@ -3077,7 +3077,7 @@ static size_t traverse_and_check(mstate m);
   /* The size of the smallest chunk held in bin with index i */
   #define minsize_for_tree_index(i)                 \
     ((SIZE_T_ONE << (((i) >> 1) + TREEBIN_SHIFT)) | \
-     (((size_t)((i)&SIZE_T_ONE)) << (((i) >> 1) + TREEBIN_SHIFT - 1)))
+     (((size_t)((i) & SIZE_T_ONE)) << (((i) >> 1) + TREEBIN_SHIFT - 1)))
 
   /* ------------------------ Operations on bin maps ----------------------- */
 
@@ -3245,7 +3245,7 @@ static size_t traverse_and_check(mstate m);
 
   #else                                                          /* FOOTERS */
 
-    /* Set foot of inuse chunk to be xor of mstate and seed */
+  /* Set foot of inuse chunk to be xor of mstate and seed */
     #define mark_inuse_foot(M, p, s)                 \
       (((mchunkptr)((char *)(p) + (s)))->prev_foot = \
            ((size_t)(M) ^ mparams.magic))
diff --git a/qemu_mode/libqasan/malloc.c b/qemu_mode/libqasan/malloc.c
index d2db3856..4448f480 100644
--- a/qemu_mode/libqasan/malloc.c
+++ b/qemu_mode/libqasan/malloc.c
@@ -80,8 +80,8 @@ static unsigned char __tmp_alloc_zone[TMP_ZONE_SIZE];
 #else
 
 // From dlmalloc.c
-void                     *dlmalloc(size_t);
-void                      dlfree(void *);
+void *dlmalloc(size_t);
+void  dlfree(void *);
   #define backend_malloc dlmalloc
   #define backend_free dlfree
 
diff --git a/src/afl-fuzz-one.c b/src/afl-fuzz-one.c
index 2003be1f..b2306996 100644
--- a/src/afl-fuzz-one.c
+++ b/src/afl-fuzz-one.c
@@ -577,13 +577,13 @@ u8 fuzz_one_original(afl_state_t *afl) {
    * SIMPLE BITFLIP (+dictionary construction) *
    *********************************************/
 
-#define FLIP_BIT(_ar, _b)                   \
-  do {                                      \
-                                            \
-    u8 *_arf = (u8 *)(_ar);                 \
-    u32 _bf = (_b);                         \
-    _arf[(_bf) >> 3] ^= (128 >> ((_bf)&7)); \
-                                            \
+#define FLIP_BIT(_ar, _b)                     \
+  do {                                        \
+                                              \
+    u8 *_arf = (u8 *)(_ar);                   \
+    u32 _bf = (_b);                           \
+    _arf[(_bf) >> 3] ^= (128 >> ((_bf) & 7)); \
+                                              \
   } while (0)
 
   /* Single walking bit. */
@@ -2216,7 +2216,7 @@ havoc_stage:
 
       }
 
-    retry_havoc_step : {
+    retry_havoc_step: {
 
       u32 r = rand_below(afl, rand_max), item;
 
@@ -3703,13 +3703,13 @@ static u8 mopt_common_fuzzing(afl_state_t *afl, MOpt_globals_t MOpt_globals) {
    * SIMPLE BITFLIP (+dictionary construction) *
    *********************************************/
 
-#define FLIP_BIT(_ar, _b)                   \
-  do {                                      \
-                                            \
-    u8 *_arf = (u8 *)(_ar);                 \
-    u32 _bf = (_b);                         \
-    _arf[(_bf) >> 3] ^= (128 >> ((_bf)&7)); \
-                                            \
+#define FLIP_BIT(_ar, _b)                     \
+  do {                                        \
+                                              \
+    u8 *_arf = (u8 *)(_ar);                   \
+    u32 _bf = (_b);                           \
+    _arf[(_bf) >> 3] ^= (128 >> ((_bf) & 7)); \
+                                              \
   } while (0)
 
   /* Single walking bit. */
diff --git a/src/afl-fuzz-redqueen.c b/src/afl-fuzz-redqueen.c
index 43b5c8bd..86e7f1cf 100644
--- a/src/afl-fuzz-redqueen.c
+++ b/src/afl-fuzz-redqueen.c
@@ -1828,7 +1828,7 @@ static void try_to_add_to_dictN(afl_state_t *afl, u128 v, u8 size) {
   for (k = 0; k < size; ++k) {
 
   #else
-  u32    off = 16 - size;
+  u32 off = 16 - size;
   for (k = 16 - size; k < 16; ++k) {
 
   #endif
diff --git a/utils/libtokencap/libtokencap.so.c b/utils/libtokencap/libtokencap.so.c
index b21f3068..f4024799 100644
--- a/utils/libtokencap/libtokencap.so.c
+++ b/utils/libtokencap/libtokencap.so.c
@@ -55,7 +55,7 @@
 #elif defined __HAIKU__
   #include <kernel/image.h>
 #elif defined __sun
-  /* For map addresses the old struct is enough */
+/* For map addresses the old struct is enough */
   #include <sys/procfs.h>
   #include <limits.h>
 #endif
@@ -168,7 +168,7 @@ static void __tokencap_load_mappings(void) {
 #elif defined __FreeBSD__ || defined __OpenBSD__ || defined __NetBSD__
 
   #if defined   __FreeBSD__
-  int    mib[] = {CTL_KERN, KERN_PROC, KERN_PROC_VMMAP, __tokencap_pid};
+  int mib[] = {CTL_KERN, KERN_PROC, KERN_PROC_VMMAP, __tokencap_pid};
   #elif defined __OpenBSD__
   int mib[] = {CTL_KERN, KERN_PROC_VMMAP, __tokencap_pid};
   #elif defined __NetBSD__
@@ -209,7 +209,7 @@ static void __tokencap_load_mappings(void) {
   #if defined __FreeBSD__ || defined __NetBSD__
 
     #if defined   __FreeBSD__
-    size_t                size = region->kve_structsize;
+    size_t size = region->kve_structsize;
 
     if (size == 0) break;
     #elif defined __NetBSD__
-- 
cgit 1.4.1


From 770e868d04c0f52a1c57e5471e459dd24a002748 Mon Sep 17 00:00:00 2001
From: yangzao <yangzaocn@outlook.com>
Date: Fri, 24 Nov 2023 11:06:06 -0700
Subject: add custom_post_run.c

---
 custom_mutators/examples/custom_post_run.c | 53 ++++++++++++++++++++++++++++++
 include/afl-fuzz.h                         | 12 +++++++
 src/afl-fuzz-mutators.c                    | 12 +++++++
 src/afl-fuzz-run.c                         |  2 ++
 4 files changed, 79 insertions(+)
 create mode 100644 custom_mutators/examples/custom_post_run.c

(limited to 'include')

diff --git a/custom_mutators/examples/custom_post_run.c b/custom_mutators/examples/custom_post_run.c
new file mode 100644
index 00000000..073aac96
--- /dev/null
+++ b/custom_mutators/examples/custom_post_run.c
@@ -0,0 +1,53 @@
+//
+// This is an example on how to use afl_custom_post_run
+// It executes custom code each time after AFL++ executes the target
+//
+// cc -O3 -fPIC -shared -g -o custom_send.so -I../../include custom_send.c /////////////////////to_be_edited
+// cd ../..
+// afl-cc -o test-instr test-instr.c
+// AFL_CUSTOM_MUTATOR_LIBRARY=custom_mutators/examples/custom_send.so \
+//   afl-fuzz -i in -o out -- ./test-instr -f /tmp/foo
+//
+
+
+#include "afl-fuzz.h"
+
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdio.h>
+
+typedef struct my_mutator {
+
+  afl_state_t *afl;
+
+} my_mutator_t;
+
+my_mutator_t *afl_custom_init(afl_state_t *afl, unsigned int seed) {
+
+  my_mutator_t *data = calloc(1, sizeof(my_mutator_t));
+  if (!data) {
+
+    perror("afl_custom_init alloc");
+    return NULL;
+
+  }
+
+  data->afl = afl;
+
+  return data;
+
+}
+
+void afl_custom_post_run(my_mutator_t *data, uint8_t *buf, size_t buf_size) {
+
+  printf("hello from afl_custom_post_run\n");
+  return;
+}
+
+
+void afl_custom_deinit(my_mutator_t *data) {
+
+  free(data);
+
+}
\ No newline at end of file
diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index 8112d430..7e91dc03 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -1020,6 +1020,18 @@ struct custom_mutator {
    */
   void (*afl_custom_fuzz_send)(void *data, const u8 *buf, size_t buf_size);
 
+  /**
+   * This method can be used if you want to run some code or scripts each time
+   * AFL++ executes the target with afl-fuzz.
+   *
+   * (Optional)
+   *
+   * @param data pointer returned in afl_custom_init by this custom mutator
+   * @param buf Buffer containing the test case
+   * @param buf_size Size of the test case
+   */
+  void (*afl_custom_post_run)(void *data, const u8 *buf, size_t buf_size);
+
   /**
    * Allow for additional analysis (e.g. calling a different tool that does a
    * different kind of coverage and saves this for the custom mutator).
diff --git a/src/afl-fuzz-mutators.c b/src/afl-fuzz-mutators.c
index 64dbe7c6..17fb9368 100644
--- a/src/afl-fuzz-mutators.c
+++ b/src/afl-fuzz-mutators.c
@@ -397,6 +397,18 @@ struct custom_mutator *load_custom_mutator(afl_state_t *afl, const char *fn) {
 
   }
 
+  /* "afl_custom_post_run", optional */
+  mutator->afl_custom_post_run = dlsym(dh, "afl_custom_post_run");
+  if (!mutator->afl_custom_post_run) {
+
+    ACTF("optional symbol 'afl_custom_post_run' not found.");
+
+  } else {
+
+    OKF("Found 'afl_custom_post_run'.");
+
+  }
+
   /* "afl_custom_queue_new_entry", optional */
   mutator->afl_custom_queue_new_entry = dlsym(dh, "afl_custom_queue_new_entry");
   if (!mutator->afl_custom_queue_new_entry) {
diff --git a/src/afl-fuzz-run.c b/src/afl-fuzz-run.c
index ac4fb4a9..29cc5352 100644
--- a/src/afl-fuzz-run.c
+++ b/src/afl-fuzz-run.c
@@ -60,6 +60,8 @@ fuzz_run_target(afl_state_t *afl, afl_forkserver_t *fsrv, u32 timeout) {
 
   fsrv_run_result_t res = afl_fsrv_run_target(fsrv, timeout, &afl->stop_soon);
 
+  
+
 #ifdef PROFILING
   clock_gettime(CLOCK_REALTIME, &spec);
   time_spent_start = (spec.tv_sec * 1000000000) + spec.tv_nsec;
-- 
cgit 1.4.1


From 8af74bcaeebbe2407006333024d8803baacdb4e2 Mon Sep 17 00:00:00 2001
From: yangzao <yangzaocn@outlook.com>
Date: Fri, 24 Nov 2023 22:47:50 -0700
Subject: update afl-fuzz-run

---
 custom_mutators/examples/custom_post_run.c |  6 +++---
 include/afl-fuzz.h                         |  4 +---
 src/afl-fuzz-run.c                         | 19 ++++++++++++++++++-
 3 files changed, 22 insertions(+), 7 deletions(-)

(limited to 'include')

diff --git a/custom_mutators/examples/custom_post_run.c b/custom_mutators/examples/custom_post_run.c
index 073aac96..828216ea 100644
--- a/custom_mutators/examples/custom_post_run.c
+++ b/custom_mutators/examples/custom_post_run.c
@@ -2,10 +2,10 @@
 // This is an example on how to use afl_custom_post_run
 // It executes custom code each time after AFL++ executes the target
 //
-// cc -O3 -fPIC -shared -g -o custom_send.so -I../../include custom_send.c /////////////////////to_be_edited
+// cc -O3 -fPIC -shared -g -o custom_post_run.so -I../../include custom_post_run.c
 // cd ../..
 // afl-cc -o test-instr test-instr.c
-// AFL_CUSTOM_MUTATOR_LIBRARY=custom_mutators/examples/custom_send.so \
+// AFL_CUSTOM_MUTATOR_LIBRARY=custom_mutators/examples/custom_post_run.so \
 //   afl-fuzz -i in -o out -- ./test-instr -f /tmp/foo
 //
 
@@ -39,7 +39,7 @@ my_mutator_t *afl_custom_init(afl_state_t *afl, unsigned int seed) {
 
 }
 
-void afl_custom_post_run(my_mutator_t *data, uint8_t *buf, size_t buf_size) {
+void afl_custom_post_run(my_mutator_t *data) {
 
   printf("hello from afl_custom_post_run\n");
   return;
diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index 7e91dc03..94f48009 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -1027,10 +1027,8 @@ struct custom_mutator {
    * (Optional)
    *
    * @param data pointer returned in afl_custom_init by this custom mutator
-   * @param buf Buffer containing the test case
-   * @param buf_size Size of the test case
    */
-  void (*afl_custom_post_run)(void *data, const u8 *buf, size_t buf_size);
+  void (*afl_custom_post_run)(void *data);
 
   /**
    * Allow for additional analysis (e.g. calling a different tool that does a
diff --git a/src/afl-fuzz-run.c b/src/afl-fuzz-run.c
index 29cc5352..ac346b86 100644
--- a/src/afl-fuzz-run.c
+++ b/src/afl-fuzz-run.c
@@ -60,7 +60,7 @@ fuzz_run_target(afl_state_t *afl, afl_forkserver_t *fsrv, u32 timeout) {
 
   fsrv_run_result_t res = afl_fsrv_run_target(fsrv, timeout, &afl->stop_soon);
 
-  
+  post_run(afl);
 
 #ifdef PROFILING
   clock_gettime(CLOCK_REALTIME, &spec);
@@ -1113,3 +1113,20 @@ common_fuzz_stuff(afl_state_t *afl, u8 *out_buf, u32 len) {
 
 }
 
+/* Run some code each time scripts each time AFL++ executes the target
+   with afl-fuzz. */
+
+void post_run(afl_state_t *afl) {
+  if (unlikely(afl->custom_mutators_count)) {
+
+      LIST_FOREACH(&afl->custom_mutator_list, struct custom_mutator, {
+
+        if (el->afl_custom_post_run) {
+
+          el->afl_custom_post_run(el->data);
+
+        }
+
+      });
+    }
+}
\ No newline at end of file
-- 
cgit 1.4.1


From faedb3fb29186c29a4f0cf28daa5d07350ed8094 Mon Sep 17 00:00:00 2001
From: yangzao <yangzaocn@outlook.com>
Date: Sat, 25 Nov 2023 21:18:32 -0700
Subject: update python module

---
 custom_mutators/examples/example.py |  5 +++++
 include/afl-fuzz.h                  |  2 ++
 src/afl-fuzz-python.c               | 32 ++++++++++++++++++++++++++++++++
 3 files changed, 39 insertions(+)

(limited to 'include')

diff --git a/custom_mutators/examples/example.py b/custom_mutators/examples/example.py
index 3a6d22e4..830f302f 100644
--- a/custom_mutators/examples/example.py
+++ b/custom_mutators/examples/example.py
@@ -133,6 +133,11 @@ def fuzz(buf, add_buf, max_size):
 #     @return: The buffer containing the test case after
 #     '''
 #     return buf
+# def post_run():
+#     '''
+#     Called after each time the execution of the target program by AFL++
+#     '''
+#     pass
 #
 # def havoc_mutation(buf, max_size):
 #     '''
diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index 94f48009..f1813df6 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -345,6 +345,7 @@ enum {
   /* 13 */ PY_FUNC_DESCRIBE,
   /* 14 */ PY_FUNC_FUZZ_SEND,
   /* 15 */ PY_FUNC_SPLICE_OPTOUT,
+  /* 16 */ PY_FUNC_POST_RUN,
   PY_FUNC_COUNT
 
 };
@@ -1085,6 +1086,7 @@ void                   finalize_py_module(void *);
 
 u32         fuzz_count_py(void *, const u8 *, size_t);
 void        fuzz_send_py(void *, const u8 *, size_t);
+void        post_run_py(void *);
 size_t      post_process_py(void *, u8 *, size_t, u8 **);
 s32         init_trim_py(void *, u8 *, size_t);
 s32         post_trim_py(void *, u8);
diff --git a/src/afl-fuzz-python.c b/src/afl-fuzz-python.c
index 7dad0770..1b287405 100644
--- a/src/afl-fuzz-python.c
+++ b/src/afl-fuzz-python.c
@@ -249,6 +249,8 @@ static py_mutator_t *init_py_module(afl_state_t *afl, u8 *module_name) {
         PyObject_GetAttrString(py_module, "queue_get");
     py_functions[PY_FUNC_FUZZ_SEND] =
         PyObject_GetAttrString(py_module, "fuzz_send");
+    py_functions[PY_FUNC_POST_RUN] =
+        PyObject_GetAttrString(py_module, "post_run");
     py_functions[PY_FUNC_SPLICE_OPTOUT] =
         PyObject_GetAttrString(py_module, "splice_optout");
     if (py_functions[PY_FUNC_SPLICE_OPTOUT]) { afl->custom_splice_optout = 1; }
@@ -468,6 +470,12 @@ struct custom_mutator *load_custom_mutator_py(afl_state_t *afl,
 
   }
 
+  if (py_functions[PY_FUNC_POST_RUN]) {
+
+    mutator->afl_custom_post_run = post_run_py;
+
+  }
+
   if (py_functions[PY_FUNC_SPLICE_OPTOUT]) {
 
     mutator->afl_custom_splice_optout = splice_optout_py;
@@ -925,6 +933,30 @@ void fuzz_send_py(void *py_mutator, const u8 *buf, size_t buf_size) {
 
 }
 
+void post_run_py(void *py_mutator) {
+
+  PyObject *py_args, *py_value;
+
+  py_args = PyTuple_New(0);
+  py_value = PyObject_CallObject(
+      ((py_mutator_t *)py_mutator)
+          ->py_functions[PY_FUNC_POST_RUN],
+      py_args);
+  Py_DECREF(py_args);
+
+  if (py_value != NULL) {
+
+    Py_DECREF(py_value);
+
+  } else {
+
+    PyErr_Print();
+    FATAL("Call failed");
+
+  }
+
+}
+
 u8 queue_new_entry_py(void *py_mutator, const u8 *filename_new_queue,
                       const u8 *filename_orig_queue) {
 
-- 
cgit 1.4.1


From f290bdd83ba1b540396db2215d133e5a40570419 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Thu, 14 Dec 2023 16:00:57 +0100
Subject: add AFL_FUZZER_LOOPCOUNT

---
 include/envs.h | 1 +
 1 file changed, 1 insertion(+)

(limited to 'include')

diff --git a/include/envs.h b/include/envs.h
index 93e49e34..560092d9 100644
--- a/include/envs.h
+++ b/include/envs.h
@@ -172,6 +172,7 @@ static char *afl_environment_variables[] = {
     "AFL_LLVM_LTO_DONTWRITEID",
     "AFL_LLVM_LTO_SKIPINIT"
     "AFL_LLVM_LTO_STARTID",
+    "AFL_FUZZER_LOOPCOUNT",
     "AFL_NO_ARITH",
     "AFL_NO_AUTODICT",
     "AFL_NO_BUILTIN",
-- 
cgit 1.4.1


From 37505928bcec63a08fe50cdebdbf7b9b28b952d0 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Fri, 15 Dec 2023 09:23:30 +0100
Subject: fix 2 mutation bugs

---
 docs/Changelog.md       |  3 +++
 include/afl-mutations.h | 16 ++++++++--------
 2 files changed, 11 insertions(+), 8 deletions(-)

(limited to 'include')

diff --git a/docs/Changelog.md b/docs/Changelog.md
index 7faa0ab3..0d75782d 100644
--- a/docs/Changelog.md
+++ b/docs/Changelog.md
@@ -5,6 +5,7 @@
 
 ### Version ++4.09a (dev)
   - afl-fuzz:
+    - fixed the new mutation implementation for two bugs
     - added `AFL_FINAL_SYNC` which forces a final fuzzer sync (also for `-F`)
       before terminating.
     - added AFL_IGNORE_SEED_PROBLEMS to skip over seeds that time out instead
@@ -23,6 +24,8 @@
     - option -n will not use color in the output
   - instrumentation:
     - fix for a few string compare transform functions for LAF
+    - we are instrumenting __cxx internal functions again. this might break
+      a few targets, please report if so.
   - frida_mode:
     - fixes support for large map offsets
   - support for AFL_FUZZER_LOOPCOUNT for afl.rs and LLVMFuzzerTestOneInput
diff --git a/include/afl-mutations.h b/include/afl-mutations.h
index d709b90d..6338c93c 100644
--- a/include/afl-mutations.h
+++ b/include/afl-mutations.h
@@ -2456,14 +2456,14 @@ inline u32 afl_mutate(afl_state_t *afl, u8 *buf, u32 len, u32 steps,
 
         }
 
-        char buf[20];
-        snprintf(buf, sizeof(buf), "%" PRId64, val);
+        char numbuf[32];
+        snprintf(numbuf, sizeof(buf), "%" PRId64, val);
         u32 old_len = off2 - off;
-        u32 new_len = strlen(buf);
+        u32 new_len = strlen(numbuf);
 
         if (old_len == new_len) {
 
-          memcpy(buf + off, buf, new_len);
+          memcpy(buf + off, numbuf, new_len);
 
         } else {
 
@@ -2473,7 +2473,7 @@ inline u32 afl_mutate(afl_state_t *afl, u8 *buf, u32 len, u32 steps,
 
           /* Inserted part */
 
-          memcpy(tmp_buf + off, buf, new_len);
+          memcpy(tmp_buf + off, numbuf, new_len);
 
           /* Tail */
           memcpy(tmp_buf + off + new_len, buf + off2, len - off2);
@@ -2509,9 +2509,9 @@ inline u32 afl_mutate(afl_state_t *afl, u8 *buf, u32 len, u32 steps,
         }
 
         u64  val = rand_next(afl);
-        char buf[20];
-        snprintf(buf, sizeof(buf), "%llu", val);
-        memcpy(buf + pos, buf, len);
+        char numbuf[32];
+        snprintf(numbuf, sizeof(numbuf), "%llu", val);
+        memcpy(buf + pos, numbuf, len);
 
         break;
 
-- 
cgit 1.4.1


From 8a7705aedbb759dd8ff331d47a99cc6bbc17902b Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Fri, 15 Dec 2023 09:28:39 +0100
Subject: v4.09c release

---
 README.md         | 4 ++--
 docs/Changelog.md | 2 +-
 include/config.h  | 2 +-
 3 files changed, 4 insertions(+), 4 deletions(-)

(limited to 'include')

diff --git a/README.md b/README.md
index 322ebcf2..a09147c5 100644
--- a/README.md
+++ b/README.md
@@ -2,9 +2,9 @@
 
 <img align="right" src="https://raw.githubusercontent.com/AFLplusplus/Website/main/static/aflpp_bg.svg" alt="AFL++ logo" width="250" heigh="250">
 
-Release version: [4.08c](https://github.com/AFLplusplus/AFLplusplus/releases)
+Release version: [4.09c](https://github.com/AFLplusplus/AFLplusplus/releases)
 
-GitHub version: 4.09a
+GitHub version: 4.09c
 
 Repository:
 [https://github.com/AFLplusplus/AFLplusplus](https://github.com/AFLplusplus/AFLplusplus)
diff --git a/docs/Changelog.md b/docs/Changelog.md
index 0d75782d..2dfcb482 100644
--- a/docs/Changelog.md
+++ b/docs/Changelog.md
@@ -3,7 +3,7 @@
   This is the list of all noteworthy changes made in every public
   release of the tool. See README.md for the general instruction manual.
 
-### Version ++4.09a (dev)
+### Version ++4.09c (release)
   - afl-fuzz:
     - fixed the new mutation implementation for two bugs
     - added `AFL_FINAL_SYNC` which forces a final fuzzer sync (also for `-F`)
diff --git a/include/config.h b/include/config.h
index 988e536e..b346d7b4 100644
--- a/include/config.h
+++ b/include/config.h
@@ -26,7 +26,7 @@
 /* Version string: */
 
 // c = release, a = volatile github dev, e = experimental branch
-#define VERSION "++4.09a"
+#define VERSION "++4.09c"
 
 /******************************************************
  *                                                    *
-- 
cgit 1.4.1


From ca0c9f6d1797bac121996c3b2ac50423f6e67b8f Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Fri, 15 Dec 2023 09:44:02 +0100
Subject: v4.10a init

---
 README.md         | 2 +-
 docs/Changelog.md | 5 ++++-
 include/config.h  | 2 +-
 3 files changed, 6 insertions(+), 3 deletions(-)

(limited to 'include')

diff --git a/README.md b/README.md
index a09147c5..fd48cb14 100644
--- a/README.md
+++ b/README.md
@@ -4,7 +4,7 @@
 
 Release version: [4.09c](https://github.com/AFLplusplus/AFLplusplus/releases)
 
-GitHub version: 4.09c
+GitHub version: 4.10a
 
 Repository:
 [https://github.com/AFLplusplus/AFLplusplus](https://github.com/AFLplusplus/AFLplusplus)
diff --git a/docs/Changelog.md b/docs/Changelog.md
index 2dfcb482..2ac87f47 100644
--- a/docs/Changelog.md
+++ b/docs/Changelog.md
@@ -3,6 +3,10 @@
   This is the list of all noteworthy changes made in every public
   release of the tool. See README.md for the general instruction manual.
 
+### Version ++4.10a (dev)
+  - ...
+
+
 ### Version ++4.09c (release)
   - afl-fuzz:
     - fixed the new mutation implementation for two bugs
@@ -34,7 +38,6 @@
   - added benchmark/benchmark.py if you want to see how good your fuzzing
     speed is in comparison to other setups.
 
-
 ### Version ++4.08c (release)
   - afl-fuzz:
     - new mutation engine: mutations that favor discovery more paths are
diff --git a/include/config.h b/include/config.h
index b346d7b4..63340650 100644
--- a/include/config.h
+++ b/include/config.h
@@ -26,7 +26,7 @@
 /* Version string: */
 
 // c = release, a = volatile github dev, e = experimental branch
-#define VERSION "++4.09c"
+#define VERSION "++4.10a"
 
 /******************************************************
  *                                                    *
-- 
cgit 1.4.1


From 7fabe5052bd41deec72fad43acd5219b5f506ac0 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Tue, 19 Dec 2023 09:26:11 +0100
Subject: fix MUT_INSERTASCIINUM

---
 include/afl-mutations.h | 20 +++++++++++++++++---
 1 file changed, 17 insertions(+), 3 deletions(-)

(limited to 'include')

diff --git a/include/afl-mutations.h b/include/afl-mutations.h
index 6338c93c..24c6b8ff 100644
--- a/include/afl-mutations.h
+++ b/include/afl-mutations.h
@@ -2490,12 +2490,13 @@ inline u32 afl_mutate(afl_state_t *afl, u8 *buf, u32 len, u32 steps,
 
       case MUT_INSERTASCIINUM: {
 
-        u32 len = 1 + rand_below(afl, 8);
+        u32 ins_len = 1 + rand_below(afl, 8);
         u32 pos = rand_below(afl, len);
 
         /* Insert ascii number. */
-        if (unlikely(len < pos + len)) {
+        if (unlikely(len < pos + ins_len)) {
 
+          // no retry if we have a small input
           if (unlikely(len < 8)) {
 
             break;
@@ -2511,7 +2512,20 @@ inline u32 afl_mutate(afl_state_t *afl, u8 *buf, u32 len, u32 steps,
         u64  val = rand_next(afl);
         char numbuf[32];
         snprintf(numbuf, sizeof(numbuf), "%llu", val);
-        memcpy(buf + pos, numbuf, len);
+        size_t val_len = strlen(numbuf), off;
+
+        if (ins_len > val_len) {
+
+          ins_len = val_len;
+          off = 0;
+
+        } else {
+
+          off = val_len - ins_len;
+
+        }
+
+        memcpy(buf + pos, numbuf + off, ins_len);
 
         break;
 
-- 
cgit 1.4.1


From f822cdeb747fb7aad8be7a9d9472331e36f3dd83 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Tue, 19 Dec 2023 09:29:12 +0100
Subject: fix MUT_STRATEGY_ARRAY_SIZE

---
 include/afl-mutations.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'include')

diff --git a/include/afl-mutations.h b/include/afl-mutations.h
index 24c6b8ff..dcc62d0b 100644
--- a/include/afl-mutations.h
+++ b/include/afl-mutations.h
@@ -32,7 +32,7 @@
 #include <inttypes.h>
 #include "afl-fuzz.h"
 
-#define MUT_STRATEGY_ARRAY_SIZE 256
+#define MUT_STRATEGY_ARRAY_SIZE 255
 
 enum {
 
-- 
cgit 1.4.1


From 806a76afaeb1e1c99847df95af4181b3d1b48a91 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Tue, 19 Dec 2023 11:15:33 +0100
Subject: fix bad fix for MUT_STRATEGY_ARRAY_SIZE

---
 docs/Changelog.md       | 7 +++++--
 include/afl-mutations.h | 3 ++-
 2 files changed, 7 insertions(+), 3 deletions(-)

(limited to 'include')

diff --git a/docs/Changelog.md b/docs/Changelog.md
index 150ce6c7..133e460b 100644
--- a/docs/Changelog.md
+++ b/docs/Changelog.md
@@ -4,8 +4,11 @@
   release of the tool. See README.md for the general instruction manual.
 
 ### Version ++4.10a (dev)
-  - default power schedule is now EXPLORE, due a fix in fast schedules
-    explore is slightly better now.
+  - afl-fuzz:
+    - default power schedule is now EXPLORE, due a fix in fast schedules
+      explore is slightly better now.
+    - fixed minor issues in the mutation engine, thanks to @futhewo for
+      reporting!
 
 
 ### Version ++4.09c (release)
diff --git a/include/afl-mutations.h b/include/afl-mutations.h
index dcc62d0b..75e66484 100644
--- a/include/afl-mutations.h
+++ b/include/afl-mutations.h
@@ -32,7 +32,7 @@
 #include <inttypes.h>
 #include "afl-fuzz.h"
 
-#define MUT_STRATEGY_ARRAY_SIZE 255
+#define MUT_STRATEGY_ARRAY_SIZE 256
 
 enum {
 
@@ -1082,6 +1082,7 @@ u32 mutation_strategy_exploration_binary[MUT_STRATEGY_ARRAY_SIZE] = {
     MUT_CLONE_COPY,
     MUT_CLONE_COPY,
     MUT_CLONE_COPY,
+    MUT_CLONE_COPY,
     MUT_CLONE_FIXED,
     MUT_CLONE_FIXED,
     MUT_CLONE_FIXED,
-- 
cgit 1.4.1


From 1fc1b32db261b27cf14f0d1d7f77a06854b7376c Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Wed, 27 Dec 2023 13:53:11 +0100
Subject: initial simple injection detection support

---
 GNUmakefile                         |  5 ++--
 GNUmakefile.llvm                    |  5 +++-
 include/envs.h                      |  4 ++++
 injections.dic                      |  7 ++++++
 instrumentation/afl-compiler-rt.o.c | 48 +++++++++++++++++++++++++++++++++++++
 src/afl-cc.c                        | 20 ++++++++++++++++
 6 files changed, 86 insertions(+), 3 deletions(-)
 create mode 100644 injections.dic

(limited to 'include')

diff --git a/GNUmakefile b/GNUmakefile
index 364cdde1..b67f9c15 100644
--- a/GNUmakefile
+++ b/GNUmakefile
@@ -782,7 +782,7 @@ install: all $(MANPAGES)
 	@rm -f $${DESTDIR}$(BIN_PATH)/afl-plot.sh
 	@rm -f $${DESTDIR}$(BIN_PATH)/afl-as
 	@rm -f $${DESTDIR}$(HELPER_PATH)/afl-llvm-rt.o $${DESTDIR}$(HELPER_PATH)/afl-llvm-rt-32.o $${DESTDIR}$(HELPER_PATH)/afl-llvm-rt-64.o $${DESTDIR}$(HELPER_PATH)/afl-gcc-rt.o
-	@for i in afl-llvm-dict2file.so afl-llvm-lto-instrumentlist.so afl-llvm-pass.so cmplog-instructions-pass.so cmplog-routines-pass.so cmplog-switches-pass.so compare-transform-pass.so libcompcov.so libdislocator.so libnyx.so libqasan.so libtokencap.so SanitizerCoverageLTO.so SanitizerCoveragePCGUARD.so split-compares-pass.so split-switches-pass.so; do echo rm -fv $${DESTDIR}$(HELPER_PATH)/$${i}; done
+	@for i in afl-llvm-dict2file.so afl-llvm-lto-instrumentlist.so afl-llvm-pass.so cmplog-instructions-pass.so cmplog-routines-pass.so cmplog-switches-pass.so compare-transform-pass.so libcompcov.so libdislocator.so libnyx.so libqasan.so libtokencap.so SanitizerCoverageLTO.so SanitizerCoveragePCGUARD.so split-compares-pass.so split-switches-pass.so injection-pass.so; do echo rm -fv $${DESTDIR}$(HELPER_PATH)/$${i}; done
 	install -m 755 $(PROGS) $(SH_PROGS) $${DESTDIR}$(BIN_PATH)
 	@if [ -f afl-qemu-trace ]; then install -m 755 afl-qemu-trace $${DESTDIR}$(BIN_PATH); fi
 	@if [ -f utils/plot_ui/afl-plot-ui ]; then install -m 755 utils/plot_ui/afl-plot-ui $${DESTDIR}$(BIN_PATH); fi
@@ -813,11 +813,12 @@ endif
 	install -m 644 docs/*.md $${DESTDIR}$(DOC_PATH)
 	cp -r testcases/ $${DESTDIR}$(MISC_PATH)
 	cp -r dictionaries/ $${DESTDIR}$(MISC_PATH)
+	cp injections.dic $${DESTDIR}$(MISC_PATH)
 
 .PHONY: uninstall
 uninstall:
 	-cd $${DESTDIR}$(BIN_PATH) && rm -f $(PROGS) $(SH_PROGS) afl-cs-proxy afl-qemu-trace afl-plot-ui afl-fuzz-document afl-network-client afl-network-server afl-g* afl-plot.sh afl-as afl-ld-lto afl-c* afl-lto*
-	-cd $${DESTDIR}$(HELPER_PATH) && rm -f afl-g*.*o afl-llvm-*.*o afl-compiler-*.*o libdislocator.so libtokencap.so libcompcov.so libqasan.so afl-frida-trace.so libnyx.so socketfuzz*.so argvfuzz*.so libAFLDriver.a libAFLQemuDriver.a as afl-as SanitizerCoverage*.so compare-transform-pass.so cmplog-*-pass.so split-*-pass.so dynamic_list.txt
+	-cd $${DESTDIR}$(HELPER_PATH) && rm -f afl-g*.*o afl-llvm-*.*o afl-compiler-*.*o libdislocator.so libtokencap.so libcompcov.so libqasan.so afl-frida-trace.so libnyx.so socketfuzz*.so argvfuzz*.so libAFLDriver.a libAFLQemuDriver.a as afl-as SanitizerCoverage*.so compare-transform-pass.so cmplog-*-pass.so split-*-pass.so dynamic_list.txt injections.dic
 	-rm -rf $${DESTDIR}$(MISC_PATH)/testcases $${DESTDIR}$(MISC_PATH)/dictionaries
 	-sh -c "ls docs/*.md | sed 's|^docs/|$${DESTDIR}$(DOC_PATH)/|' | xargs rm -f"
 	-cd $${DESTDIR}$(MAN_PATH) && rm -f $(MANPAGES)
diff --git a/GNUmakefile.llvm b/GNUmakefile.llvm
index 0845ae3a..c704d772 100644
--- a/GNUmakefile.llvm
+++ b/GNUmakefile.llvm
@@ -341,7 +341,7 @@ ifeq "$(TEST_MMAP)" "1"
 endif
 
 PROGS_ALWAYS = ./afl-cc ./afl-compiler-rt.o ./afl-compiler-rt-32.o ./afl-compiler-rt-64.o 
-PROGS        = $(PROGS_ALWAYS) ./afl-llvm-pass.so ./SanitizerCoveragePCGUARD.so ./split-compares-pass.so ./split-switches-pass.so ./cmplog-routines-pass.so ./cmplog-instructions-pass.so ./cmplog-switches-pass.so ./afl-llvm-dict2file.so ./compare-transform-pass.so ./afl-ld-lto ./afl-llvm-lto-instrumentlist.so ./SanitizerCoverageLTO.so
+PROGS        = $(PROGS_ALWAYS) ./afl-llvm-pass.so ./SanitizerCoveragePCGUARD.so ./split-compares-pass.so ./split-switches-pass.so ./cmplog-routines-pass.so ./cmplog-instructions-pass.so ./cmplog-switches-pass.so ./afl-llvm-dict2file.so ./compare-transform-pass.so ./afl-ld-lto ./afl-llvm-lto-instrumentlist.so ./SanitizerCoverageLTO.so ./injection-pass.so
 
 # If prerequisites are not given, warn, do not build anything, and exit with code 0
 ifeq "$(LLVMVER)" ""
@@ -469,6 +469,9 @@ endif
 afl-llvm-dict2file.so:	instrumentation/afl-llvm-dict2file.so.cc instrumentation/afl-llvm-common.o | test_deps
 	$(CXX) $(CLANG_CPPFL) -shared $< -o $@ $(CLANG_LFL) instrumentation/afl-llvm-common.o
 
+./injection-pass.so:	instrumentation/injection-pass.cc instrumentation/afl-llvm-common.o | test_deps
+	$(CXX) $(CLANG_CPPFL) -shared $< -o $@ $(CLANG_LFL) instrumentation/afl-llvm-common.o
+
 .PHONY: document
 document:
 	$(CLANG_BIN) -D_AFL_DOCUMENT_MUTATIONS $(CFLAGS_SAFE) $(CPPFLAGS) $(CLANG_CFL) -O3 -Wno-unused-result -fPIC -c instrumentation/afl-compiler-rt.o.c -o ./afl-compiler-rt.o
diff --git a/include/envs.h b/include/envs.h
index 560092d9..75b2e13d 100644
--- a/include/envs.h
+++ b/include/envs.h
@@ -151,6 +151,10 @@ static char *afl_environment_variables[] = {
     "AFL_LLVM_LTO_AUTODICTIONARY",
     "AFL_LLVM_AUTODICTIONARY",
     "AFL_LLVM_SKIPSINGLEBLOCK",
+    "AFL_LLVM_INJECTIONS_ALL",
+    "AFL_LLVM_INJECTIONS_SQL",
+    "AFL_LLVM_INJECTIONS_LDAP",
+    "AFL_LLVM_INJECTIONS_XSS",
     "AFL_LLVM_INSTRIM_SKIPSINGLEBLOCK",
     "AFL_LLVM_LAF_SPLIT_COMPARES",
     "AFL_LLVM_LAF_SPLIT_COMPARES_BITW",
diff --git a/injections.dic b/injections.dic
new file mode 100644
index 00000000..4063cd17
--- /dev/null
+++ b/injections.dic
@@ -0,0 +1,7 @@
+"1'\" OR \"1\"=\"1"
+"1\"' OR '1'='1"
+"'\"><FUZZ"
+"*)(FUZZ=*))(|"
+"\";FUZZ;\""
+"';FUZZ;'"
+"$(FUZZ)"
diff --git a/instrumentation/afl-compiler-rt.o.c b/instrumentation/afl-compiler-rt.o.c
index def59b6b..50bafb9e 100644
--- a/instrumentation/afl-compiler-rt.o.c
+++ b/instrumentation/afl-compiler-rt.o.c
@@ -92,6 +92,8 @@ extern ssize_t _kern_write(int fd, off_t pos, const void *buffer,
                            size_t bufferSize);
 #endif  // HAIKU
 
+char *strcasestr(const char *haystack, const char *needle);
+
 static u8  __afl_area_initial[MAP_INITIAL_SIZE];
 static u8 *__afl_area_ptr_dummy = __afl_area_initial;
 static u8 *__afl_area_ptr_backup = __afl_area_initial;
@@ -2670,5 +2672,51 @@ void __afl_set_persistent_mode(u8 mode) {
 
 }
 
+void __afl_injection_sql(u8 *buf) {
+
+  if (likely(buf)) {
+
+    if (unlikely(strcasestr((char *)buf, "1'\" OR \"1\"=\"1") ||
+                 strcasestr((char *)buf, "1\"' OR '1'='1"))) {
+
+      fprintf(stderr, "ALERT: Detected SQL injection in query: %s\n", buf);
+      abort();
+
+    }
+
+  }
+
+}
+
+void __afl_injection_ldap(u8 *buf) {
+
+  if (likely(buf)) {
+
+    if (unlikely(strcasestr((char *)buf, "*)(FUZZ=*))(|"))) {
+
+      fprintf(stderr, "ALERT: Detected LDAP injection in query: %s\n", buf);
+      abort();
+
+    }
+
+  }
+
+}
+
+void __afl_injection_xss(u8 *buf) {
+
+  if (likely(buf)) {
+
+    if (unlikely(strcasestr((char *)buf, "\";FUZZ;\""))) {
+
+      fprintf(stderr, "ALERT: Detected XSS injection in content: %s\n", buf);
+      abort();
+
+    }
+
+  }
+
+}
+
 #undef write_error
 
diff --git a/src/afl-cc.c b/src/afl-cc.c
index 22cce2cd..a46facc7 100644
--- a/src/afl-cc.c
+++ b/src/afl-cc.c
@@ -1139,6 +1139,26 @@ static void edit_params(u32 argc, char **argv, char **envp) {
 
     }
 
+    if (getenv("AFL_LLVM_INJECTIONS_ALL") ||
+        getenv("AFL_LLVM_INJECTIONS_SQL") ||
+        getenv("AFL_LLVM_INJECTIONS_LDAP") ||
+        getenv("AFL_LLVM_INJECTIONS_XSS")) {
+
+#if LLVM_MAJOR >= 11
+  #if LLVM_MAJOR < 16
+      cc_params[cc_par_cnt++] = "-fexperimental-new-pass-manager";
+  #endif
+      cc_params[cc_par_cnt++] =
+          alloc_printf("-fpass-plugin=%s/injection-pass.so", obj_path);
+#else
+      cc_params[cc_par_cnt++] = "-Xclang";
+      cc_params[cc_par_cnt++] = "-load";
+      cc_params[cc_par_cnt++] = "-Xclang";
+      cc_params[cc_par_cnt++] = alloc_printf("%s/injection-pass.so", obj_path);
+#endif
+
+    }
+
     // cc_params[cc_par_cnt++] = "-Qunused-arguments";
 
   }
-- 
cgit 1.4.1


From 1eb54c4c3eb4ab4bc12f7f1f80f5ece15b238ef0 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Sat, 30 Dec 2023 10:49:00 +0100
Subject: finish injection implementation

---
 include/envs.h                       |  1 +
 instrumentation/README.injections.md | 48 ++++++++++++++++++++++++++++++++++++
 instrumentation/afl-compiler-rt.o.c  |  9 ++++---
 src/afl-fuzz.c                       | 28 +++++++++++++++++++++
 4 files changed, 82 insertions(+), 4 deletions(-)
 create mode 100644 instrumentation/README.injections.md

(limited to 'include')

diff --git a/include/envs.h b/include/envs.h
index 75b2e13d..aa5c658e 100644
--- a/include/envs.h
+++ b/include/envs.h
@@ -151,6 +151,7 @@ static char *afl_environment_variables[] = {
     "AFL_LLVM_LTO_AUTODICTIONARY",
     "AFL_LLVM_AUTODICTIONARY",
     "AFL_LLVM_SKIPSINGLEBLOCK",
+    // Marker: ADD_TO_INJECTIONS
     "AFL_LLVM_INJECTIONS_ALL",
     "AFL_LLVM_INJECTIONS_SQL",
     "AFL_LLVM_INJECTIONS_LDAP",
diff --git a/instrumentation/README.injections.md b/instrumentation/README.injections.md
new file mode 100644
index 00000000..16cc3713
--- /dev/null
+++ b/instrumentation/README.injections.md
@@ -0,0 +1,48 @@
+# Injection fuzzing
+
+Coverage guided fuzzing so far is only able to detect crashes, so usually
+memory corruption issues, or - if implemented by hand in the harness -
+invariants.
+
+This is a proof-of-concept implementation to additionally hunt for injection
+vulnerabilities.
+It works by instrumenting calls to specific functions and parsing the
+query parameter for a specific unescaped dictionary string, and if detected,
+crashes the target.
+
+This has a very low false positive rate.
+But obviously this can only find injection vulnerailities that are suspectible
+to this specific (but most common) issue. Hence in a rare kind of injection
+vulnerability this won't find the bug - and be a false negative.
+But this can be tweaked by the user - see the HOW TO MODIFY section below.
+
+## How to use
+
+Set one or more of the following environment variables for **compiling**
+the target and - *this is important* - when **fuzzing** the target:
+
+ - `AFL_LLVM_INJECTIONS_SQL`
+ - `AFL_LLVM_INJECTIONS_LDAP`
+ - `AFL_LLVM_INJECTIONS_XSS`
+
+Alternatively you can set `AFL_LLVM_INJECTIONS_ALL` to enable all.
+
+## How to modify
+
+If you want to add more fuctions to check for e.g. SQL injections:
+Add these to `instrumentation/injection-pass.cc` and recompile.
+
+If you want to test for more injection inputs:
+Add the dictionary tokens to `src/afl-fuzz.c` and the check for them to
+`instrumentation/afl-compiler-rt.o.c`.
+
+If you want to add new injection targets:
+You will have to edit all three files.
+
+Just search for:
+```
+// Marker: ADD_TO_INJECTIONS
+```
+in the files to see where this needs to be added.
+
+**NOTE:** pull requests to improve this feature are highly welcome :-)
diff --git a/instrumentation/afl-compiler-rt.o.c b/instrumentation/afl-compiler-rt.o.c
index 50bafb9e..39a762b6 100644
--- a/instrumentation/afl-compiler-rt.o.c
+++ b/instrumentation/afl-compiler-rt.o.c
@@ -2672,12 +2672,13 @@ void __afl_set_persistent_mode(u8 mode) {
 
 }
 
+// Marker: ADD_TO_INJECTIONS
+
 void __afl_injection_sql(u8 *buf) {
 
   if (likely(buf)) {
 
-    if (unlikely(strcasestr((char *)buf, "1'\" OR \"1\"=\"1") ||
-                 strcasestr((char *)buf, "1\"' OR '1'='1"))) {
+    if (unlikely(strstr((char *)buf, "'\"\"'"))) {
 
       fprintf(stderr, "ALERT: Detected SQL injection in query: %s\n", buf);
       abort();
@@ -2692,7 +2693,7 @@ void __afl_injection_ldap(u8 *buf) {
 
   if (likely(buf)) {
 
-    if (unlikely(strcasestr((char *)buf, "*)(FUZZ=*))(|"))) {
+    if (unlikely(strstr((char *)buf, "*)(1=*))(|"))) {
 
       fprintf(stderr, "ALERT: Detected LDAP injection in query: %s\n", buf);
       abort();
@@ -2707,7 +2708,7 @@ void __afl_injection_xss(u8 *buf) {
 
   if (likely(buf)) {
 
-    if (unlikely(strcasestr((char *)buf, "\";FUZZ;\""))) {
+    if (unlikely(strstr((char *)buf, "1\"><\""))) {
 
       fprintf(stderr, "ALERT: Detected XSS injection in content: %s\n", buf);
       abort();
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index dd990e71..17949fd7 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -1749,6 +1749,34 @@ int main(int argc, char **argv_orig, char **envp) {
 
   }
 
+  // Marker: ADD_TO_INJECTIONS
+  if (getenv("AFL_LLVM_INJECTIONS_ALL") || getenv("AFL_LLVM_INJECTIONS_SQL") ||
+      getenv("AFL_LLVM_INJECTIONS_LDAP") || getenv("AFL_LLVM_INJECTIONS_XSS")) {
+
+    OKF("Adding injection tokens to dictionary.");
+    if (getenv("AFL_LLVM_INJECTIONS_ALL") ||
+        getenv("AFL_LLVM_INJECTIONS_SQL")) {
+
+      add_extra(afl, "'\"\"'", 4);
+
+    }
+
+    if (getenv("AFL_LLVM_INJECTIONS_ALL") ||
+        getenv("AFL_LLVM_INJECTIONS_LDAP")) {
+
+      add_extra(afl, "*)(1=*))(|", 10);
+
+    }
+
+    if (getenv("AFL_LLVM_INJECTIONS_ALL") ||
+        getenv("AFL_LLVM_INJECTIONS_XSS")) {
+
+      add_extra(afl, "1\"><\"", 5);
+
+    }
+
+  }
+
   OKF("Generating fuzz data with a length of min=%u max=%u", afl->min_length,
       afl->max_length);
   u32 min_alloc = MAX(64U, afl->min_length);
-- 
cgit 1.4.1


From ee7d69b8175f31f2efb2b3bf15f2bbb29f61aa14 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Thu, 4 Jan 2024 15:44:28 +0100
Subject: changelog

---
 docs/Changelog.md |   6 +-
 include/envs.h    | 319 ++++++++++++++----------------------------------------
 2 files changed, 88 insertions(+), 237 deletions(-)

(limited to 'include')

diff --git a/docs/Changelog.md b/docs/Changelog.md
index adc81d64..69a369e3 100644
--- a/docs/Changelog.md
+++ b/docs/Changelog.md
@@ -9,15 +9,17 @@
       explore is slightly better now.
     - fixed minor issues in the mutation engine, thanks to @futhewo for
       reporting!
+  - afl-cc:
+    - large rewrite by @SonicStark which fixes a few corner cases, thanks!
   - instrumentation:
     - LLVM 18 support, thanks to @devnexen!
     - Injection (SQL, LDAP, XSS) feature now available, see
       `instrumentation/README.injections.md` how to activate/use/expand.
     - compcov/LAF-intel:
       - floating point splitting bug fix by @hexcoder
-      - due a bug in LLVM 17 integer splitting is disabled!
+      - due a bug in LLVM 17 integer splitting is disabled there!
       - when splitting floats was selected, integers were always split as well,
-        fixed to require AFL_LLVM_LAF_SPLIT_COMPARES as it should
+        fixed to require AFL_LLVM_LAF_SPLIT_COMPARES or _ALL as it should
 
 
 ### Version ++4.09c (release)
diff --git a/include/envs.h b/include/envs.h
index aa5c658e..0f645d23 100644
--- a/include/envs.h
+++ b/include/envs.h
@@ -16,255 +16,104 @@ static char *afl_environment_deprecated[] = {
 
 static char *afl_environment_variables[] = {
 
-    "AFL_ALIGNED_ALLOC",
-    "AFL_ALLOW_TMP",
-    "AFL_ANALYZE_HEX",
-    "AFL_AS",
-    "AFL_AUTORESUME",
-    "AFL_AS_FORCE_INSTRUMENT",
-    "AFL_BENCH_JUST_ONE",
-    "AFL_BENCH_UNTIL_CRASH",
-    "AFL_CAL_FAST",
-    "AFL_CC",
-    "AFL_CC_COMPILER",
-    "AFL_CMIN_ALLOW_ANY",
-    "AFL_CMIN_CRASHES_ONLY",
-    "AFL_CMPLOG_ONLY_NEW",
-    "AFL_CODE_END",
-    "AFL_CODE_START",
-    "AFL_COMPCOV_BINNAME",
-    "AFL_COMPCOV_LEVEL",
-    "AFL_CRASH_EXITCODE",
-    "AFL_CRASHING_SEEDS_AS_NEW_CRASH",
-    "AFL_CUSTOM_MUTATOR_LIBRARY",
-    "AFL_CUSTOM_MUTATOR_ONLY",
-    "AFL_CUSTOM_INFO_PROGRAM",
-    "AFL_CUSTOM_INFO_PROGRAM_ARGV",
-    "AFL_CUSTOM_INFO_PROGRAM_INPUT",
-    "AFL_CUSTOM_INFO_OUT",
-    "AFL_CXX",
-    "AFL_CYCLE_SCHEDULES",
-    "AFL_DEBUG",
-    "AFL_DEBUG_CHILD",
-    "AFL_DEBUG_GDB",
-    "AFL_DEBUG_UNICORN",
-    "AFL_DISABLE_TRIM",
-    "AFL_DISABLE_LLVM_INSTRUMENTATION",
-    "AFL_DONT_OPTIMIZE",
-    "AFL_DRIVER_STDERR_DUPLICATE_FILENAME",
-    "AFL_DUMB_FORKSRV",
-    "AFL_EARLY_FORKSERVER",
-    "AFL_ENTRYPOINT",
-    "AFL_EXIT_WHEN_DONE",
-    "AFL_EXIT_ON_TIME",
-    "AFL_EXIT_ON_SEED_ISSUES",
-    "AFL_FAST_CAL",
-    "AFL_FINAL_SYNC",
-    "AFL_FORCE_UI",
-    "AFL_FRIDA_DEBUG_MAPS",
-    "AFL_FRIDA_DRIVER_NO_HOOK",
-    "AFL_FRIDA_EXCLUDE_RANGES",
-    "AFL_FRIDA_INST_CACHE_SIZE",
-    "AFL_FRIDA_INST_COVERAGE_ABSOLUTE",
-    "AFL_FRIDA_INST_COVERAGE_FILE",
-    "AFL_FRIDA_INST_DEBUG_FILE",
-    "AFL_FRIDA_INST_INSN",
-    "AFL_FRIDA_INST_JIT",
-    "AFL_FRIDA_INST_NO_CACHE",
-    "AFL_FRIDA_INST_NO_DYNAMIC_LOAD",
-    "AFL_FRIDA_INST_NO_OPTIMIZE",
-    "AFL_FRIDA_INST_NO_PREFETCH",
-    "AFL_FRIDA_INST_NO_PREFETCH_BACKPATCH",
+    "AFL_ALIGNED_ALLOC", "AFL_ALLOW_TMP", "AFL_ANALYZE_HEX", "AFL_AS",
+    "AFL_AUTORESUME", "AFL_AS_FORCE_INSTRUMENT", "AFL_BENCH_JUST_ONE",
+    "AFL_BENCH_UNTIL_CRASH", "AFL_CAL_FAST", "AFL_CC", "AFL_CC_COMPILER",
+    "AFL_CMIN_ALLOW_ANY", "AFL_CMIN_CRASHES_ONLY", "AFL_CMPLOG_ONLY_NEW",
+    "AFL_CODE_END", "AFL_CODE_START", "AFL_COMPCOV_BINNAME",
+    "AFL_COMPCOV_LEVEL", "AFL_CRASH_EXITCODE",
+    "AFL_CRASHING_SEEDS_AS_NEW_CRASH", "AFL_CUSTOM_MUTATOR_LIBRARY",
+    "AFL_CUSTOM_MUTATOR_ONLY", "AFL_CUSTOM_INFO_PROGRAM",
+    "AFL_CUSTOM_INFO_PROGRAM_ARGV", "AFL_CUSTOM_INFO_PROGRAM_INPUT",
+    "AFL_CUSTOM_INFO_OUT", "AFL_CXX", "AFL_CYCLE_SCHEDULES", "AFL_DEBUG",
+    "AFL_DEBUG_CHILD", "AFL_DEBUG_GDB", "AFL_DEBUG_UNICORN", "AFL_DISABLE_TRIM",
+    "AFL_DISABLE_LLVM_INSTRUMENTATION", "AFL_DONT_OPTIMIZE",
+    "AFL_DRIVER_STDERR_DUPLICATE_FILENAME", "AFL_DUMB_FORKSRV",
+    "AFL_EARLY_FORKSERVER", "AFL_ENTRYPOINT", "AFL_EXIT_WHEN_DONE",
+    "AFL_EXIT_ON_TIME", "AFL_EXIT_ON_SEED_ISSUES", "AFL_FAST_CAL",
+    "AFL_FINAL_SYNC", "AFL_FORCE_UI", "AFL_FRIDA_DEBUG_MAPS",
+    "AFL_FRIDA_DRIVER_NO_HOOK", "AFL_FRIDA_EXCLUDE_RANGES",
+    "AFL_FRIDA_INST_CACHE_SIZE", "AFL_FRIDA_INST_COVERAGE_ABSOLUTE",
+    "AFL_FRIDA_INST_COVERAGE_FILE", "AFL_FRIDA_INST_DEBUG_FILE",
+    "AFL_FRIDA_INST_INSN", "AFL_FRIDA_INST_JIT", "AFL_FRIDA_INST_NO_CACHE",
+    "AFL_FRIDA_INST_NO_DYNAMIC_LOAD", "AFL_FRIDA_INST_NO_OPTIMIZE",
+    "AFL_FRIDA_INST_NO_PREFETCH", "AFL_FRIDA_INST_NO_PREFETCH_BACKPATCH",
     "AFL_FRIDA_INST_NO_SUPPRESS"
     "AFL_FRIDA_INST_RANGES",
-    "AFL_FRIDA_INST_REGS_FILE",
-    "AFL_FRIDA_INST_SEED",
-    "AFL_FRIDA_INST_TRACE",
-    "AFL_FRIDA_INST_TRACE_UNIQUE",
-    "AFL_FRIDA_INST_UNSTABLE_COVERAGE_FILE",
-    "AFL_FRIDA_JS_SCRIPT",
-    "AFL_FRIDA_OUTPUT_STDOUT",
-    "AFL_FRIDA_OUTPUT_STDERR",
-    "AFL_FRIDA_PERSISTENT_ADDR",
-    "AFL_FRIDA_PERSISTENT_CNT",
-    "AFL_FRIDA_PERSISTENT_DEBUG",
-    "AFL_FRIDA_PERSISTENT_HOOK",
-    "AFL_FRIDA_PERSISTENT_RET",
-    "AFL_FRIDA_STALKER_ADJACENT_BLOCKS",
-    "AFL_FRIDA_STALKER_IC_ENTRIES",
-    "AFL_FRIDA_STALKER_NO_BACKPATCH",
-    "AFL_FRIDA_STATS_FILE",
-    "AFL_FRIDA_STATS_INTERVAL",
-    "AFL_FRIDA_TRACEABLE",
+    "AFL_FRIDA_INST_REGS_FILE", "AFL_FRIDA_INST_SEED", "AFL_FRIDA_INST_TRACE",
+    "AFL_FRIDA_INST_TRACE_UNIQUE", "AFL_FRIDA_INST_UNSTABLE_COVERAGE_FILE",
+    "AFL_FRIDA_JS_SCRIPT", "AFL_FRIDA_OUTPUT_STDOUT", "AFL_FRIDA_OUTPUT_STDERR",
+    "AFL_FRIDA_PERSISTENT_ADDR", "AFL_FRIDA_PERSISTENT_CNT",
+    "AFL_FRIDA_PERSISTENT_DEBUG", "AFL_FRIDA_PERSISTENT_HOOK",
+    "AFL_FRIDA_PERSISTENT_RET", "AFL_FRIDA_STALKER_ADJACENT_BLOCKS",
+    "AFL_FRIDA_STALKER_IC_ENTRIES", "AFL_FRIDA_STALKER_NO_BACKPATCH",
+    "AFL_FRIDA_STATS_FILE", "AFL_FRIDA_STATS_INTERVAL", "AFL_FRIDA_TRACEABLE",
     "AFL_FRIDA_VERBOSE",
     "AFL_FUZZER_ARGS",  // oss-fuzz
-    "AFL_FUZZER_STATS_UPDATE_INTERVAL",
-    "AFL_GDB",
-    "AFL_GCC_ALLOWLIST",
-    "AFL_GCC_DENYLIST",
-    "AFL_GCC_BLOCKLIST",
-    "AFL_GCC_INSTRUMENT_FILE",
-    "AFL_GCC_OUT_OF_LINE",
-    "AFL_GCC_SKIP_NEVERZERO",
-    "AFL_GCJ",
-    "AFL_HANG_TMOUT",
-    "AFL_FORKSRV_INIT_TMOUT",
-    "AFL_HARDEN",
-    "AFL_I_DONT_CARE_ABOUT_MISSING_CRASHES",
-    "AFL_IGNORE_PROBLEMS",
-    "AFL_IGNORE_PROBLEMS_COVERAGE",
-    "AFL_IGNORE_SEED_PROBLEMS",
-    "AFL_IGNORE_TIMEOUTS",
-    "AFL_IGNORE_UNKNOWN_ENVS",
-    "AFL_IMPORT_FIRST",
-    "AFL_INPUT_LEN_MIN",
-    "AFL_INPUT_LEN_MAX",
-    "AFL_INST_LIBS",
-    "AFL_INST_RATIO",
-    "AFL_KEEP_TIMEOUTS",
-    "AFL_KILL_SIGNAL",
-    "AFL_FORK_SERVER_KILL_SIGNAL",
-    "AFL_KEEP_TRACES",
-    "AFL_KEEP_ASSEMBLY",
-    "AFL_LD_HARD_FAIL",
-    "AFL_LD_LIMIT_MB",
-    "AFL_LD_NO_CALLOC_OVER",
-    "AFL_LD_PASSTHROUGH",
-    "AFL_REAL_LD",
-    "AFL_LD_PRELOAD",
-    "AFL_LD_VERBOSE",
-    "AFL_LLVM_ALLOWLIST",
-    "AFL_LLVM_DENYLIST",
-    "AFL_LLVM_BLOCKLIST",
-    "AFL_CMPLOG",
-    "AFL_LLVM_CMPLOG",
-    "AFL_GCC_CMPLOG",
-    "AFL_LLVM_INSTRIM",
-    "AFL_LLVM_CALLER",
-    "AFL_LLVM_CTX",
-    "AFL_LLVM_CTX_K",
-    "AFL_LLVM_DICT2FILE",
-    "AFL_LLVM_DICT2FILE_NO_MAIN",
-    "AFL_LLVM_DOCUMENT_IDS",
-    "AFL_LLVM_INSTRIM_LOOPHEAD",
-    "AFL_LLVM_INSTRUMENT",
-    "AFL_LLVM_LTO_AUTODICTIONARY",
-    "AFL_LLVM_AUTODICTIONARY",
+    "AFL_FUZZER_STATS_UPDATE_INTERVAL", "AFL_GDB", "AFL_GCC_ALLOWLIST",
+    "AFL_GCC_DENYLIST", "AFL_GCC_BLOCKLIST", "AFL_GCC_INSTRUMENT_FILE",
+    "AFL_GCC_OUT_OF_LINE", "AFL_GCC_SKIP_NEVERZERO", "AFL_GCJ",
+    "AFL_HANG_TMOUT", "AFL_FORKSRV_INIT_TMOUT", "AFL_HARDEN",
+    "AFL_I_DONT_CARE_ABOUT_MISSING_CRASHES", "AFL_IGNORE_PROBLEMS",
+    "AFL_IGNORE_PROBLEMS_COVERAGE", "AFL_IGNORE_SEED_PROBLEMS",
+    "AFL_IGNORE_TIMEOUTS", "AFL_IGNORE_UNKNOWN_ENVS", "AFL_IMPORT_FIRST",
+    "AFL_INPUT_LEN_MIN", "AFL_INPUT_LEN_MAX", "AFL_INST_LIBS", "AFL_INST_RATIO",
+    "AFL_KEEP_TIMEOUTS", "AFL_KILL_SIGNAL", "AFL_FORK_SERVER_KILL_SIGNAL",
+    "AFL_KEEP_TRACES", "AFL_KEEP_ASSEMBLY", "AFL_LD_HARD_FAIL",
+    "AFL_LD_LIMIT_MB", "AFL_LD_NO_CALLOC_OVER", "AFL_LD_PASSTHROUGH",
+    "AFL_REAL_LD", "AFL_LD_PRELOAD", "AFL_LD_VERBOSE", "AFL_LLVM_ALLOWLIST",
+    "AFL_LLVM_DENYLIST", "AFL_LLVM_BLOCKLIST", "AFL_CMPLOG", "AFL_LLVM_CMPLOG",
+    "AFL_GCC_CMPLOG", "AFL_LLVM_INSTRIM", "AFL_LLVM_CALLER", "AFL_LLVM_CTX",
+    "AFL_LLVM_CTX_K", "AFL_LLVM_DICT2FILE", "AFL_LLVM_DICT2FILE_NO_MAIN",
+    "AFL_LLVM_DOCUMENT_IDS", "AFL_LLVM_INSTRIM_LOOPHEAD", "AFL_LLVM_INSTRUMENT",
+    "AFL_LLVM_LTO_AUTODICTIONARY", "AFL_LLVM_AUTODICTIONARY",
     "AFL_LLVM_SKIPSINGLEBLOCK",
     // Marker: ADD_TO_INJECTIONS
-    "AFL_LLVM_INJECTIONS_ALL",
-    "AFL_LLVM_INJECTIONS_SQL",
-    "AFL_LLVM_INJECTIONS_LDAP",
-    "AFL_LLVM_INJECTIONS_XSS",
-    "AFL_LLVM_INSTRIM_SKIPSINGLEBLOCK",
-    "AFL_LLVM_LAF_SPLIT_COMPARES",
-    "AFL_LLVM_LAF_SPLIT_COMPARES_BITW",
-    "AFL_LLVM_LAF_SPLIT_FLOATS",
-    "AFL_LLVM_LAF_SPLIT_SWITCHES",
-    "AFL_LLVM_LAF_ALL",
-    "AFL_LLVM_LAF_TRANSFORM_COMPARES",
-    "AFL_LLVM_MAP_ADDR",
-    "AFL_LLVM_MAP_DYNAMIC",
-    "AFL_LLVM_NGRAM_SIZE",
-    "AFL_NGRAM_SIZE",
-    "AFL_LLVM_NO_RPATH",
-    "AFL_LLVM_NOT_ZERO",
-    "AFL_LLVM_INSTRUMENT_FILE",
-    "AFL_LLVM_THREADSAFE_INST",
-    "AFL_LLVM_SKIP_NEVERZERO",
-    "AFL_NO_AFFINITY",
-    "AFL_TRY_AFFINITY",
-    "AFL_LLVM_LTO_DONTWRITEID",
+    "AFL_LLVM_INJECTIONS_ALL", "AFL_LLVM_INJECTIONS_SQL",
+    "AFL_LLVM_INJECTIONS_LDAP", "AFL_LLVM_INJECTIONS_XSS",
+    "AFL_LLVM_INSTRIM_SKIPSINGLEBLOCK", "AFL_LLVM_LAF_SPLIT_COMPARES",
+    "AFL_LLVM_LAF_SPLIT_COMPARES_BITW", "AFL_LLVM_LAF_SPLIT_FLOATS",
+    "AFL_LLVM_LAF_SPLIT_SWITCHES", "AFL_LLVM_LAF_ALL",
+    "AFL_LLVM_LAF_TRANSFORM_COMPARES", "AFL_LLVM_MAP_ADDR",
+    "AFL_LLVM_MAP_DYNAMIC", "AFL_LLVM_NGRAM_SIZE", "AFL_NGRAM_SIZE",
+    "AFL_LLVM_NO_RPATH", "AFL_LLVM_NOT_ZERO", "AFL_LLVM_INSTRUMENT_FILE",
+    "AFL_LLVM_THREADSAFE_INST", "AFL_LLVM_SKIP_NEVERZERO", "AFL_NO_AFFINITY",
+    "AFL_TRY_AFFINITY", "AFL_LLVM_LTO_DONTWRITEID",
     "AFL_LLVM_LTO_SKIPINIT"
     "AFL_LLVM_LTO_STARTID",
-    "AFL_FUZZER_LOOPCOUNT",
-    "AFL_NO_ARITH",
-    "AFL_NO_AUTODICT",
-    "AFL_NO_BUILTIN",
+    "AFL_FUZZER_LOOPCOUNT", "AFL_NO_ARITH", "AFL_NO_AUTODICT", "AFL_NO_BUILTIN",
 #if defined USE_COLOR && !defined ALWAYS_COLORED
-    "AFL_NO_COLOR",
-    "AFL_NO_COLOUR",
+    "AFL_NO_COLOR", "AFL_NO_COLOUR",
 #endif
     "AFL_NO_CPU_RED",
     "AFL_NO_CFG_FUZZING",  // afl.rs rust crate option
-    "AFL_NO_CRASH_README",
-    "AFL_NO_FORKSRV",
-    "AFL_NO_UI",
-    "AFL_NO_PYTHON",
-    "AFL_NO_STARTUP_CALIBRATION",
-    "AFL_NO_WARN_INSTABILITY",
-    "AFL_UNTRACER_FILE",
-    "AFL_LLVM_USE_TRACE_PC",
-    "AFL_MAP_SIZE",
-    "AFL_MAPSIZE",
+    "AFL_NO_CRASH_README", "AFL_NO_FORKSRV", "AFL_NO_UI", "AFL_NO_PYTHON",
+    "AFL_NO_STARTUP_CALIBRATION", "AFL_NO_WARN_INSTABILITY",
+    "AFL_UNTRACER_FILE", "AFL_LLVM_USE_TRACE_PC", "AFL_MAP_SIZE", "AFL_MAPSIZE",
     "AFL_MAX_DET_EXTRAS",
     "AFL_NO_X86",  // not really an env but we dont want to warn on it
-    "AFL_NOOPT",
-    "AFL_NYX_AUX_SIZE",
-    "AFL_NYX_DISABLE_SNAPSHOT_MODE",
-    "AFL_NYX_LOG",
-    "AFL_NYX_REUSE_SNAPSHOT",
-    "AFL_PASSTHROUGH",
-    "AFL_PATH",
-    "AFL_PERFORMANCE_FILE",
-    "AFL_PERSISTENT_RECORD",
-    "AFL_POST_PROCESS_KEEP_ORIGINAL",
-    "AFL_PRELOAD",
-    "AFL_TARGET_ENV",
-    "AFL_PYTHON_MODULE",
-    "AFL_QEMU_CUSTOM_BIN",
-    "AFL_QEMU_COMPCOV",
-    "AFL_QEMU_COMPCOV_DEBUG",
-    "AFL_QEMU_DEBUG_MAPS",
-    "AFL_QEMU_DISABLE_CACHE",
-    "AFL_QEMU_DRIVER_NO_HOOK",
-    "AFL_QEMU_FORCE_DFL",
-    "AFL_QEMU_PERSISTENT_ADDR",
-    "AFL_QEMU_PERSISTENT_CNT",
-    "AFL_QEMU_PERSISTENT_GPR",
-    "AFL_QEMU_PERSISTENT_HOOK",
-    "AFL_QEMU_PERSISTENT_MEM",
-    "AFL_QEMU_PERSISTENT_RET",
-    "AFL_QEMU_PERSISTENT_RETADDR_OFFSET",
-    "AFL_QEMU_PERSISTENT_EXITS",
-    "AFL_QEMU_INST_RANGES",
-    "AFL_QEMU_EXCLUDE_RANGES",
-    "AFL_QEMU_SNAPSHOT",
-    "AFL_QEMU_TRACK_UNSTABLE",
-    "AFL_QUIET",
-    "AFL_RANDOM_ALLOC_CANARY",
-    "AFL_REAL_PATH",
-    "AFL_SHUFFLE_QUEUE",
-    "AFL_SKIP_BIN_CHECK",
-    "AFL_SKIP_CPUFREQ",
-    "AFL_SKIP_CRASHES",
-    "AFL_SKIP_OSSFUZZ",
-    "AFL_STATSD",
-    "AFL_STATSD_HOST",
-    "AFL_STATSD_PORT",
-    "AFL_STATSD_TAGS_FLAVOR",
-    "AFL_SYNC_TIME",
-    "AFL_TESTCACHE_SIZE",
-    "AFL_TESTCACHE_ENTRIES",
-    "AFL_TMIN_EXACT",
-    "AFL_TMPDIR",
-    "AFL_TOKEN_FILE",
-    "AFL_TRACE_PC",
-    "AFL_USE_ASAN",
-    "AFL_USE_MSAN",
-    "AFL_USE_TRACE_PC",
-    "AFL_USE_UBSAN",
-    "AFL_USE_TSAN",
-    "AFL_USE_CFISAN",
-    "AFL_USE_LSAN",
-    "AFL_WINE_PATH",
-    "AFL_NO_SNAPSHOT",
-    "AFL_EXPAND_HAVOC_NOW",
-    "AFL_USE_FASAN",
-    "AFL_USE_QASAN",
-    "AFL_PRINT_FILENAMES",
-    "AFL_PIZZA_MODE",
-    NULL
+    "AFL_NOOPT", "AFL_NYX_AUX_SIZE", "AFL_NYX_DISABLE_SNAPSHOT_MODE",
+    "AFL_NYX_LOG", "AFL_NYX_REUSE_SNAPSHOT", "AFL_PASSTHROUGH", "AFL_PATH",
+    "AFL_PERFORMANCE_FILE", "AFL_PERSISTENT_RECORD",
+    "AFL_POST_PROCESS_KEEP_ORIGINAL", "AFL_PRELOAD", "AFL_TARGET_ENV",
+    "AFL_PYTHON_MODULE", "AFL_QEMU_CUSTOM_BIN", "AFL_QEMU_COMPCOV",
+    "AFL_QEMU_COMPCOV_DEBUG", "AFL_QEMU_DEBUG_MAPS", "AFL_QEMU_DISABLE_CACHE",
+    "AFL_QEMU_DRIVER_NO_HOOK", "AFL_QEMU_FORCE_DFL", "AFL_QEMU_PERSISTENT_ADDR",
+    "AFL_QEMU_PERSISTENT_CNT", "AFL_QEMU_PERSISTENT_GPR",
+    "AFL_QEMU_PERSISTENT_HOOK", "AFL_QEMU_PERSISTENT_MEM",
+    "AFL_QEMU_PERSISTENT_RET", "AFL_QEMU_PERSISTENT_RETADDR_OFFSET",
+    "AFL_QEMU_PERSISTENT_EXITS", "AFL_QEMU_INST_RANGES",
+    "AFL_QEMU_EXCLUDE_RANGES", "AFL_QEMU_SNAPSHOT", "AFL_QEMU_TRACK_UNSTABLE",
+    "AFL_QUIET", "AFL_RANDOM_ALLOC_CANARY", "AFL_REAL_PATH",
+    "AFL_SHUFFLE_QUEUE", "AFL_SKIP_BIN_CHECK", "AFL_SKIP_CPUFREQ",
+    "AFL_SKIP_CRASHES", "AFL_SKIP_OSSFUZZ", "AFL_STATSD", "AFL_STATSD_HOST",
+    "AFL_STATSD_PORT", "AFL_STATSD_TAGS_FLAVOR", "AFL_SYNC_TIME",
+    "AFL_TESTCACHE_SIZE", "AFL_TESTCACHE_ENTRIES", "AFL_TMIN_EXACT",
+    "AFL_TMPDIR", "AFL_TOKEN_FILE", "AFL_TRACE_PC", "AFL_USE_ASAN",
+    "AFL_USE_MSAN", "AFL_USE_TRACE_PC", "AFL_USE_UBSAN", "AFL_USE_TSAN",
+    "AFL_USE_CFISAN", "AFL_USE_LSAN", "AFL_WINE_PATH", "AFL_NO_SNAPSHOT",
+    "AFL_EXPAND_HAVOC_NOW", "AFL_USE_FASAN", "AFL_USE_QASAN",
+    "AFL_PRINT_FILENAMES", "AFL_PIZZA_MODE", NULL
 
 };
 
-- 
cgit 1.4.1


From 8fedf4998449d5b6b909a1118fc2e152e4d2e6e7 Mon Sep 17 00:00:00 2001
From: Davide Quarta <quarta@qti.qualcomm.com>
Date: Tue, 23 Jan 2024 19:36:49 +0100
Subject: replay mode support

---
 .gitignore                                  |   1 +
 include/afl-fuzz.h                          |   4 +
 include/config.h                            |   5 +-
 include/persistent_replay.h                 | 149 ++++++++++++++++++++++++++++
 instrumentation/afl-compiler-rt.o.c         |  36 +++++++
 src/afl-forkserver.c                        |  79 +++++++++------
 src/afl-fuzz-init.c                         |   6 ++
 src/afl-fuzz.c                              |   2 +-
 utils/persistent_mode/Makefile              |   3 +-
 utils/persistent_mode/persistent_demo_new.c |  15 +--
 10 files changed, 257 insertions(+), 43 deletions(-)
 create mode 100644 include/persistent_replay.h

(limited to 'include')

diff --git a/.gitignore b/.gitignore
index f76a86fc..891ced9f 100644
--- a/.gitignore
+++ b/.gitignore
@@ -103,6 +103,7 @@ utils/optimin/build
 utils/optimin/optimin
 utils/persistent_mode/persistent_demo
 utils/persistent_mode/persistent_demo_new
+utils/persistent_mode/persistent_demo_new_compat
 utils/persistent_mode/test-instr
 utils/plot_ui/afl-plot-ui
 vuln_prog
diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index f1813df6..864bc6b6 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -124,6 +124,10 @@
   #define CASE_PREFIX "id_"
 #endif                                                    /* ^!SIMPLE_FILES */
 
+#ifdef AFL_PERSISTENT_RECORD
+  #define RECORD_PREFIX "RECORD:" 
+#endif
+
 #define STAGE_BUF_SIZE (64)  /* usable size for stage name buf in afl_state */
 
 // Little helper to access the ptr to afl->##name_buf - for use in afl_realloc.
diff --git a/include/config.h b/include/config.h
index 63340650..1649f110 100644
--- a/include/config.h
+++ b/include/config.h
@@ -83,7 +83,10 @@
    will be kept and written to the crash/ directory as RECORD:... files.
    Note that every crash will be written, not only unique ones! */
 
-// #define AFL_PERSISTENT_RECORD
+//  #define AFL_PERSISTENT_RECORD
+
+/* Builds compiler-rt with support to replay persistent records */
+//  #define AFL_PERSISTENT_REPLAY
 
 /* console output colors: There are three ways to configure its behavior
  * 1. default: colored outputs fixed on: defined USE_COLOR && defined
diff --git a/include/persistent_replay.h b/include/persistent_replay.h
new file mode 100644
index 00000000..b1a55e9f
--- /dev/null
+++ b/include/persistent_replay.h
@@ -0,0 +1,149 @@
+#ifndef _HAVE_PERSISTENT_REPLAY_H
+#define _HAVE_PERSISTENT_REPLAY_H
+
+#include <dirent.h>
+#include <string.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <malloc.h>
+#include <unistd.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+
+static unsigned short int is_replay_record;
+static unsigned int replay_record;
+static unsigned int replay_record_cnt;
+static char replay_record_path[PATH_MAX];
+static char **record_arg;
+static char *replay_record_dir;
+static struct dirent **record_list;
+
+static int select_files(const struct dirent *dirbuf) {
+
+  char fn[4096];
+
+  if (dirbuf->d_name[0] == '.'){
+    return 0;
+  } else {
+    snprintf(fn, sizeof(fn), "RECORD:%06u", replay_record);
+    return !!strstr(dirbuf->d_name, fn);
+  }
+}
+  
+static int compare_files(const struct dirent **da, const struct dirent **db) {
+  
+  unsigned int c1=0, c2=0;
+
+  sscanf((*da)->d_name, "RECORD:%*u,cnt:%06u", &c1);
+  sscanf((*db)->d_name, "RECORD:%*u,cnt:%06u", &c2);
+
+  return c1-c2;
+}
+
+__attribute__((destructor)) static void __afl_record_replay_destroy(void){
+  for (int i=0; i < replay_record_cnt; i++) {
+    free(record_list[i]);
+  }
+  free(record_list);
+}
+
+__attribute__((constructor)) static void __afl_record_replay_init(int argc, char **argv) {
+  
+  char **argp;
+
+  /* caveat: if harness uses @@ and we don't pass it, it will regardless loop the number of iterations defined for AFL_LOOP (on the same file)*/
+  if(!(is_replay_record = !!getenv("AFL_PERSISTENT_REPLAY"))){
+    // printf("[warning] AFL_PERSISTENT_REPLAY not set.\n");
+    return;
+  }
+
+  replay_record = atoi(getenv("AFL_PERSISTENT_REPLAY"));
+  replay_record_dir = getenv("AFL_PERSISTENT_DIR");
+  replay_record_cnt = scandir(replay_record_dir ? replay_record_dir : "./", &record_list, select_files, compare_files);  
+
+  if (!replay_record_cnt){
+    printf("[error] Can't find the requested record!\n");
+    is_replay_record = 0;
+  }
+
+  argp = argv;
+  while (*argp){
+    if (!strcmp(*argp, "@@")){
+      record_arg = argp;
+      *record_arg = replay_record_path;
+      break;
+    }
+    ++argp;
+  }
+
+}
+
+/* only used if explictly included for compatibility
+   compiling without afl-cc */
+
+#ifdef AFL_COMPAT
+
+#ifndef PATH_MAX
+  #define PATH_MAX 4096
+#endif
+
+#define FUZZ_BUF_SIZE 1024000
+
+// extern ssize_t read(int fildes, void *buf, size_t nbyte);
+
+//extern int __afl_persistent_loop(unsigned int max_cnt);
+//extern unsigned char fuzz_buf[];
+
+#ifndef __AFL_HAVE_MANUAL_CONTROL
+  #define __AFL_HAVE_MANUAL_CONTROL
+#endif
+
+#define __AFL_FUZZ_TESTCASE_LEN (read(0, fuzz_buf, FUZZ_BUF_SIZE))
+#define __AFL_FUZZ_TESTCASE_BUF fuzz_buf
+#define __AFL_FUZZ_INIT() void sync(void); 
+#define __AFL_INIT() sync() 
+#define __AFL_LOOP(x) __afl_persistent_loop(x)
+
+unsigned char fuzz_buf[FUZZ_BUF_SIZE];
+
+int __afl_persistent_loop(unsigned int max_cnt) {
+
+  static unsigned int cycle_cnt = 1;
+  static unsigned short int inited = 0;
+  char tcase[PATH_MAX];
+
+  if( is_replay_record ){
+
+    if (!inited){
+      cycle_cnt = replay_record_cnt;
+      inited = 1;
+    }
+
+    snprintf(tcase, PATH_MAX, "%s/%s",
+                  replay_record_dir ? replay_record_dir : "./",
+                  record_list[replay_record_cnt-cycle_cnt]->d_name);
+
+
+    if (record_arg) {
+      *record_arg = tcase;
+    } else {
+      int fd = open(tcase, O_RDONLY);
+      dup2(fd, 0);
+      close(fd);
+    }
+
+  } else {
+
+    if (!inited){
+      cycle_cnt = max_cnt;
+      inited = 1;
+    }
+
+  }
+
+  return cycle_cnt--;
+}
+
+#endif  // AFL_COMPAT
+
+#endif  // _HAVE_PERSISTENT_REPLAY_H
\ No newline at end of file
diff --git a/instrumentation/afl-compiler-rt.o.c b/instrumentation/afl-compiler-rt.o.c
index 39a762b6..0fa22aee 100644
--- a/instrumentation/afl-compiler-rt.o.c
+++ b/instrumentation/afl-compiler-rt.o.c
@@ -83,6 +83,10 @@
 #include <sys/mman.h>
 #include <fcntl.h>
 
+#ifdef AFL_PERSISTENT_REPLAY
+#include "persistent_replay.h"
+#endif
+
 /* Globals needed by the injected instrumentation. The __afl_area_initial region
    is used for instrumentation output before __afl_map_shm() has a chance to
    run. It will end up as .comm, so it shouldn't be too wasteful. */
@@ -1338,6 +1342,38 @@ int __afl_persistent_loop(unsigned int max_cnt) {
   static u8  first_pass = 1;
   static u32 cycle_cnt;
 
+#ifdef AFL_PERSISTENT_REPLAY
+
+#ifndef PATH_MAX
+  #define PATH_MAX 4096
+#endif
+
+  static u8  inited = 0;
+  char tcase[PATH_MAX];
+
+  if( unlikely(is_replay_record) ){
+
+      if (!inited){
+        cycle_cnt = replay_record_cnt;
+        inited = 1;
+      }
+
+      snprintf(tcase, PATH_MAX, "%s/%s",
+                  replay_record_dir ? replay_record_dir : "./",
+                  record_list[replay_record_cnt-cycle_cnt]->d_name);
+
+      if (record_arg) {
+        *record_arg = tcase;
+      } else {
+        int fd = open(tcase, O_RDONLY);
+        dup2(fd, 0);
+        close(fd);
+      }
+    return cycle_cnt--;
+  } else
+
+#endif  
+
   if (first_pass) {
 
     /* Make sure that every iteration of __AFL_LOOP() starts with a clean slate.
diff --git a/src/afl-forkserver.c b/src/afl-forkserver.c
index 3f9bfa72..f8dd783f 100644
--- a/src/afl-forkserver.c
+++ b/src/afl-forkserver.c
@@ -1591,6 +1591,11 @@ afl_fsrv_run_target(afl_forkserver_t *fsrv, u32 timeout,
   u32 exec_ms;
   u32 write_value = fsrv->last_run_timed_out;
 
+#ifdef AFL_PERSISTENT_RECORD
+  fsrv_run_result_t retval = FSRV_RUN_OK;
+  char *persistent_out_fmt;
+#endif
+
 #ifdef __linux__
   if (fsrv->nyx_mode) {
 
@@ -1684,7 +1689,7 @@ afl_fsrv_run_target(afl_forkserver_t *fsrv, u32 timeout,
 
   }
 
-#ifdef AFL_PERSISTENT_RECORD
+#ifdef AFL_eERSISTENT_RECORD
   // end of persistent loop?
   if (unlikely(fsrv->persistent_record &&
                fsrv->persistent_record_pid != fsrv->child_pid)) {
@@ -1790,8 +1795,14 @@ afl_fsrv_run_target(afl_forkserver_t *fsrv, u32 timeout,
   if (unlikely(fsrv->last_run_timed_out)) {
 
     fsrv->last_kill_signal = fsrv->child_kill_signal;
-    return FSRV_RUN_TMOUT;
 
+#ifndef AFL_PERSISTENT_RECORD
+    return FSRV_RUN_TMOUT;
+#else
+    retval = FSRV_RUN_TMOUT;
+    persistent_out_fmt = "%s/hangs/RECORD:%06u,cnt:%06u";
+    goto store_persistent_record;
+#endif
   }
 
   /* Did we crash?
@@ -1811,48 +1822,58 @@ afl_fsrv_run_target(afl_forkserver_t *fsrv, u32 timeout,
           (fsrv->uses_crash_exitcode &&
            WEXITSTATUS(fsrv->child_status) == fsrv->crash_exitcode))) {
 
-#ifdef AFL_PERSISTENT_RECORD
-    if (unlikely(fsrv->persistent_record)) {
+    /* For a proper crash, set last_kill_signal to WTERMSIG, else set it to 0 */
+    fsrv->last_kill_signal =
+        WIFSIGNALED(fsrv->child_status) ? WTERMSIG(fsrv->child_status) : 0;
 
-      char fn[PATH_MAX];
-      u32  i, writecnt = 0;
-      for (i = 0; i < fsrv->persistent_record; ++i) {
+#ifndef AFL_PERSISTENT_RECORD
+    return FSRV_RUN_CRASH;
+#else
+    retval = FSRV_RUN_CRASH;
+    persistent_out_fmt = "%s/crashes/RECORD:%06u,cnt:%06u";
+    goto store_persistent_record;
+#endif
 
-        u32 entry = (i + fsrv->persistent_record_idx) % fsrv->persistent_record;
-        u8 *data = fsrv->persistent_record_data[entry];
-        u32 len = fsrv->persistent_record_len[entry];
-        if (likely(len && data)) {
+  }
 
-          snprintf(fn, sizeof(fn), "%s/RECORD:%06u,cnt:%06u",
-                   fsrv->persistent_record_dir, fsrv->persistent_record_cnt,
-                   writecnt++);
-          int fd = open(fn, O_CREAT | O_TRUNC | O_WRONLY, 0644);
-          if (fd >= 0) {
+  /* success :) */
+  return FSRV_RUN_OK;
+
+#ifdef AFL_PERSISTENT_RECORD
+store_persistent_record:
+  if (unlikely(retval == FSRV_RUN_CRASH || retval == FSRV_RUN_TMOUT)  &&
+      unlikely(fsrv->persistent_record)) {
 
-            ck_write(fd, data, len, fn);
-            close(fd);
+    char fn[PATH_MAX];
+    u32  i, writecnt = 0;
+    for (i = 0; i < fsrv->persistent_record; ++i) {
 
-          }
+      u32 entry = (i + fsrv->persistent_record_idx) % fsrv->persistent_record;
+      u8 *data = fsrv->persistent_record_data[entry];
+      u32 len = fsrv->persistent_record_len[entry];
+      if (likely(len && data)) {
+
+        snprintf(fn, sizeof(fn), persistent_out_fmt,
+                 fsrv->persistent_record_dir, fsrv->persistent_record_cnt,
+                 writecnt++);
+        int fd = open(fn, O_CREAT | O_TRUNC | O_WRONLY, 0644);
+        if (fd >= 0) {
+
+          ck_write(fd, data, len, fn);
+          close(fd);
 
         }
 
       }
 
-      ++fsrv->persistent_record_cnt;
-
     }
 
-#endif
-
-    /* For a proper crash, set last_kill_signal to WTERMSIG, else set it to 0 */
-    fsrv->last_kill_signal =
-        WIFSIGNALED(fsrv->child_status) ? WTERMSIG(fsrv->child_status) : 0;
-    return FSRV_RUN_CRASH;
+    ++fsrv->persistent_record_cnt;
 
   }
 
-  /* success :) */
-  return FSRV_RUN_OK;
+  return retval;
+#endif
 
 }
 
diff --git a/src/afl-fuzz-init.c b/src/afl-fuzz-init.c
index 35932913..5b7dc4c1 100644
--- a/src/afl-fuzz-init.c
+++ b/src/afl-fuzz-init.c
@@ -1915,6 +1915,9 @@ static void handle_existing_out_dir(afl_state_t *afl) {
 
   }
 
+#ifdef AFL_PERSISTENT_RECORD
+  delete_files(fn, RECORD_PREFIX);
+#endif
   if (delete_files(fn, CASE_PREFIX)) { goto dir_cleanup_failed; }
   ck_free(fn);
 
@@ -1947,6 +1950,9 @@ static void handle_existing_out_dir(afl_state_t *afl) {
 
   }
 
+#ifdef AFL_PERSISTENT_RECORD
+  delete_files(fn, RECORD_PREFIX);
+#endif
   if (delete_files(fn, CASE_PREFIX)) { goto dir_cleanup_failed; }
   ck_free(fn);
 
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index 17949fd7..40c30472 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -2163,7 +2163,7 @@ int main(int argc, char **argv_orig, char **envp) {
 
     }
 
-    afl->fsrv.persistent_record_dir = alloc_printf("%s/crashes", afl->out_dir);
+    afl->fsrv.persistent_record_dir = alloc_printf("%s", afl->out_dir);
 
   }
 
diff --git a/utils/persistent_mode/Makefile b/utils/persistent_mode/Makefile
index e348c46c..64de82a7 100644
--- a/utils/persistent_mode/Makefile
+++ b/utils/persistent_mode/Makefile
@@ -1,10 +1,11 @@
 all:
 	../../afl-clang-fast -o persistent_demo persistent_demo.c
 	../../afl-clang-fast -o persistent_demo_new persistent_demo_new.c
+	gcc -g -I ../../include -o persistent_demo_new_compat persistent_demo_new.c
 	AFL_DONT_OPTIMIZE=1 ../../afl-clang-fast -o test-instr test-instr.c
 
 document:
 	AFL_DONT_OPTIMIZE=1 ../../afl-clang-fast -D_AFL_DOCUMENT_MUTATIONS -o test-instr test-instr.c
 
 clean:
-	rm -f persistent_demo persistent_demo_new test-instr
+	rm -f persistent_demo persistent_demo_new persistent_demo_new_compat test-instr
diff --git a/utils/persistent_mode/persistent_demo_new.c b/utils/persistent_mode/persistent_demo_new.c
index 285f50aa..40ada9e1 100644
--- a/utils/persistent_mode/persistent_demo_new.c
+++ b/utils/persistent_mode/persistent_demo_new.c
@@ -31,17 +31,8 @@
 
 /* this lets the source compile without afl-clang-fast/lto */
 #ifndef __AFL_FUZZ_TESTCASE_LEN
-
-ssize_t       fuzz_len;
-unsigned char fuzz_buf[1024000];
-
-  #define __AFL_FUZZ_TESTCASE_LEN fuzz_len
-  #define __AFL_FUZZ_TESTCASE_BUF fuzz_buf
-  #define __AFL_FUZZ_INIT() void sync(void);
-  #define __AFL_LOOP(x) \
-    ((fuzz_len = read(0, fuzz_buf, sizeof(fuzz_buf))) > 0 ? 1 : 0)
-  #define __AFL_INIT() sync()
-
+#define AFL_COMPAT
+#include "persistent_replay.h"
 #endif
 
 __AFL_FUZZ_INIT();
@@ -95,6 +86,8 @@ int main(int argc, char **argv) {
               if (buf[5] == '!') {
 
                 printf("six\n");
+                char *nullo = NULL+1;
+                *nullo = 'p';
                 abort();
 
               }
-- 
cgit 1.4.1


From 06f0982f0f4506e18872efb86b97993f2518988c Mon Sep 17 00:00:00 2001
From: Han Zheng <35988108+kdsjZh@users.noreply.github.com>
Date: Thu, 1 Feb 2024 15:13:21 +0100
Subject: Enhancement on Deterministic stage (#1972)

* fuzzer: init commit based on aflpp 60dc37a8cf09f8e9048e4b6a2204d6c90b27655a

* fuzzers: adding the skip variables and initialize

* log: profile the det/havoc finding

* log: add profile log output

* fuzzers: sperate log/skipdet module

* fuzzers: add quick eff_map calc

* fuzzers: add skip_eff_map in fuzz_one

* fuzzers: mark whole input space in eff_map

* fuzzers: add undet bit threshold to skip some seeds

* fuzzers: fix one byte overflow

* fuzzers: fix overflow

* fix code format

* add havoc only again

* code format

* remove log to INTROSPECTION, rename skipdet module

* rename skipdet module

* remove log to stats

* clean redundant code

* code format

* remove redundant code format check

* remove redundant doc

* remove redundant objects

* clean files

* change -d to default skipdet

* disable deterministic when using CUSTOM_MUTATOR

* revert fix
---
 include/afl-fuzz.h     |  58 +++++++
 include/config.h       |  12 ++
 include/forkserver.h   |   3 +-
 src/afl-fuzz-init.c    |  15 ++
 src/afl-fuzz-one.c     | 166 ++++++++++++++------
 src/afl-fuzz-queue.c   |  11 ++
 src/afl-fuzz-skipdet.c | 403 +++++++++++++++++++++++++++++++++++++++++++++++++
 src/afl-fuzz-state.c   |  10 +-
 src/afl-fuzz-stats.c   |  38 +++++
 src/afl-fuzz.c         |  13 +-
 10 files changed, 680 insertions(+), 49 deletions(-)
 create mode 100644 src/afl-fuzz-skipdet.c

(limited to 'include')

diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index f1813df6..c2b09b2e 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -149,6 +149,48 @@ struct tainted {
 
 };
 
+struct inf_profile {
+
+  u32 inf_skipped_bytes;               /* Inference Stage Profiling         */
+  u64 inf_execs_cost, inf_time_cost;
+
+};
+
+/* ToDo: add cmplog profile as well */
+struct havoc_profile {
+
+  u32 queued_det_stage,                 /* Det/Havoc Stage Profiling        */
+      queued_havoc_stage, total_queued_det, edge_det_stage, edge_havoc_stage,
+      total_det_edge;
+
+  u64 det_stage_time, havoc_stage_time, total_det_time;
+
+};
+
+struct skipdet_entry {
+
+  u8  continue_inf, done_eff;
+  u32 undet_bits, quick_eff_bytes;
+
+  u8 *skip_eff_map,                     /* we'v finish the eff_map          */
+      *done_inf_map;                    /* some bytes are not done yet      */
+
+};
+
+struct skipdet_global {
+
+  u8 use_skip_havoc;
+
+  u32 undet_bits_threshold;
+
+  u64 last_cov_undet;
+
+  u8 *virgin_det_bits;                  /* global fuzzed bits               */
+
+  struct inf_profile *inf_prof;
+
+};
+
 struct queue_entry {
 
   u8 *fname;                            /* File name for the test case      */
@@ -203,6 +245,8 @@ struct queue_entry {
 
   struct queue_entry *mother;           /* queue entry this based on        */
 
+  struct skipdet_entry *skipdet_e;
+
 };
 
 struct extra_data {
@@ -247,6 +291,8 @@ enum {
   /* 19 */ STAGE_CUSTOM_MUTATOR,
   /* 20 */ STAGE_COLORIZATION,
   /* 21 */ STAGE_ITS,
+  /* 22 */ STAGE_INF,
+  /* 23 */ STAGE_QUICK,
 
   STAGE_NUM_MAX
 
@@ -782,6 +828,11 @@ typedef struct afl_state {
    * is too large) */
   struct queue_entry **q_testcase_cache;
 
+  /* Global Profile Data for deterministic/havoc-splice stage */
+  struct havoc_profile *havoc_prof;
+
+  struct skipdet_global *skipdet_g;
+
 #ifdef INTROSPECTION
   char  mutation[8072];
   char  m_tmp[4096];
@@ -1232,6 +1283,13 @@ AFL_RAND_RETURN rand_next(afl_state_t *afl);
 /* probability between 0.0 and 1.0 */
 double rand_next_percent(afl_state_t *afl);
 
+/* SkipDet Functions */
+
+u8 skip_deterministic_stage(afl_state_t *, u8 *, u8 *, u32, u64);
+u8 is_det_timeout(u64, u8);
+
+void plot_profile_data(afl_state_t *, struct queue_entry *);
+
 /**** Inline routines ****/
 
 /* Generate a random number (from 0 to limit - 1). This may
diff --git a/include/config.h b/include/config.h
index 63340650..7ad73c2f 100644
--- a/include/config.h
+++ b/include/config.h
@@ -52,6 +52,18 @@
 /* Default file permission umode when creating files (default: 0600) */
 #define DEFAULT_PERMISSION 0600
 
+/* SkipDet's global configuration */
+
+#define MINIMAL_BLOCK_SIZE 64
+#define SMALL_DET_TIME (60 * 1000 * 1000U)
+#define MAXIMUM_INF_EXECS (16 * 1024U)
+#define MAXIMUM_QUICK_EFF_EXECS (64 * 1024U)
+#define THRESHOLD_DEC_TIME (20 * 60 * 1000U)
+
+/* Set the Prob of selecting eff_bytes 3 times more than original,
+   Now disabled */
+#define EFF_HAVOC_RATE 3
+
 /* CMPLOG/REDQUEEN TUNING
  *
  * Here you can modify tuning and solving options for CMPLOG.
diff --git a/include/forkserver.h b/include/forkserver.h
index f6230fe8..f1d3b5b1 100644
--- a/include/forkserver.h
+++ b/include/forkserver.h
@@ -126,7 +126,8 @@ typedef struct afl_forkserver {
   u8 *out_file,                         /* File to fuzz, if any             */
       *target_path;                     /* Path of the target               */
 
-  FILE *plot_file;                      /* Gnuplot output file              */
+  FILE *plot_file,                      /* Gnuplot output file              */
+      *det_plot_file;
 
   /* Note: last_run_timed_out is u32 to send it to the child as 4 byte array */
   u32 last_run_timed_out;               /* Traced process timed out?        */
diff --git a/src/afl-fuzz-init.c b/src/afl-fuzz-init.c
index 8ab44a3b..057d8cf5 100644
--- a/src/afl-fuzz-init.c
+++ b/src/afl-fuzz-init.c
@@ -2236,6 +2236,21 @@ void setup_dirs_fds(afl_state_t *afl) {
 
   fflush(afl->fsrv.plot_file);
 
+#ifdef INTROSPECTION
+
+  tmp = alloc_printf("%s/plot_det_data", afl->out_dir);
+
+  int fd = open(tmp, O_WRONLY | O_CREAT, DEFAULT_PERMISSION);
+  if (fd < 0) { PFATAL("Unable to create '%s'", tmp); }
+  ck_free(tmp);
+
+  afl->fsrv.det_plot_file = fdopen(fd, "w");
+  if (!afl->fsrv.det_plot_file) { PFATAL("fdopen() failed"); }
+
+  if (afl->in_place_resume) { fseek(afl->fsrv.det_plot_file, 0, SEEK_END); }
+
+#endif
+
   /* ignore errors */
 
 }
diff --git a/src/afl-fuzz-one.c b/src/afl-fuzz-one.c
index 01e34b69..4a7d3fad 100644
--- a/src/afl-fuzz-one.c
+++ b/src/afl-fuzz-one.c
@@ -545,12 +545,37 @@ u8 fuzz_one_original(afl_state_t *afl) {
 
   }
 
+  u64 before_det_time = get_cur_time();
+#ifdef INTROSPECTION
+
+  u64 before_havoc_time;
+  u32 before_det_findings = afl->queued_items,
+      before_det_edges = count_non_255_bytes(afl, afl->virgin_bits),
+      before_havoc_findings, before_havoc_edges;
+  u8 is_logged = 0;
+
+#endif
+  if (!afl->skip_deterministic) {
+
+    if (!skip_deterministic_stage(afl, in_buf, out_buf, len, before_det_time)) {
+
+      goto abandon_entry;
+
+    }
+
+  }
+
+  u8 *skip_eff_map = afl->queue_cur->skipdet_e->skip_eff_map;
+
   /* Skip right away if -d is given, if it has not been chosen sufficiently
      often to warrant the expensive deterministic stage (fuzz_level), or
      if it has gone through deterministic testing in earlier, resumed runs
      (passed_det). */
+  /* if skipdet decide to skip the seed or no interesting bytes found,
+     we skip the whole deterministic stage as well */
 
   if (likely(afl->skip_deterministic) || likely(afl->queue_cur->passed_det) ||
+      likely(!afl->queue_cur->skipdet_e->quick_eff_bytes) ||
       likely(perf_score <
              (afl->queue_cur->depth * 30 <= afl->havoc_max_mult * 100
                   ? afl->queue_cur->depth * 30
@@ -609,6 +634,10 @@ u8 fuzz_one_original(afl_state_t *afl) {
 
     afl->stage_cur_byte = afl->stage_cur >> 3;
 
+    if (!skip_eff_map[afl->stage_cur_byte]) continue;
+
+    if (is_det_timeout(before_det_time, 0)) { goto custom_mutator_stage; }
+
     FLIP_BIT(out_buf, afl->stage_cur);
 
 #ifdef INTROSPECTION
@@ -725,6 +754,10 @@ u8 fuzz_one_original(afl_state_t *afl) {
 
     afl->stage_cur_byte = afl->stage_cur >> 3;
 
+    if (!skip_eff_map[afl->stage_cur_byte]) continue;
+
+    if (is_det_timeout(before_det_time, 0)) { goto custom_mutator_stage; }
+
     FLIP_BIT(out_buf, afl->stage_cur);
     FLIP_BIT(out_buf, afl->stage_cur + 1);
 
@@ -760,6 +793,10 @@ u8 fuzz_one_original(afl_state_t *afl) {
 
     afl->stage_cur_byte = afl->stage_cur >> 3;
 
+    if (!skip_eff_map[afl->stage_cur_byte]) continue;
+
+    if (is_det_timeout(before_det_time, 0)) { goto custom_mutator_stage; }
+
     FLIP_BIT(out_buf, afl->stage_cur);
     FLIP_BIT(out_buf, afl->stage_cur + 1);
     FLIP_BIT(out_buf, afl->stage_cur + 2);
@@ -828,6 +865,10 @@ u8 fuzz_one_original(afl_state_t *afl) {
 
     afl->stage_cur_byte = afl->stage_cur;
 
+    if (!skip_eff_map[afl->stage_cur_byte]) continue;
+
+    if (is_det_timeout(before_det_time, 0)) { goto custom_mutator_stage; }
+
     out_buf[afl->stage_cur] ^= 0xFF;
 
 #ifdef INTROSPECTION
@@ -837,37 +878,6 @@ u8 fuzz_one_original(afl_state_t *afl) {
 
     if (common_fuzz_stuff(afl, out_buf, len)) { goto abandon_entry; }
 
-    /* We also use this stage to pull off a simple trick: we identify
-       bytes that seem to have no effect on the current execution path
-       even when fully flipped - and we skip them during more expensive
-       deterministic stages, such as arithmetics or known ints. */
-
-    if (!eff_map[EFF_APOS(afl->stage_cur)]) {
-
-      u64 cksum;
-
-      /* If in non-instrumented mode or if the file is very short, just flag
-         everything without wasting time on checksums. */
-
-      if (!afl->non_instrumented_mode && len >= EFF_MIN_LEN) {
-
-        cksum = hash64(afl->fsrv.trace_bits, afl->fsrv.map_size, HASH_CONST);
-
-      } else {
-
-        cksum = ~prev_cksum;
-
-      }
-
-      if (cksum != prev_cksum) {
-
-        eff_map[EFF_APOS(afl->stage_cur)] = 1;
-        ++eff_cnt;
-
-      }
-
-    }
-
     out_buf[afl->stage_cur] ^= 0xFF;
 
   }
@@ -876,18 +886,8 @@ u8 fuzz_one_original(afl_state_t *afl) {
      whole thing as worth fuzzing, since we wouldn't be saving much time
      anyway. */
 
-  if (eff_cnt != (u32)EFF_ALEN(len) &&
-      eff_cnt * 100 / EFF_ALEN(len) > EFF_MAX_PERC) {
-
-    memset(eff_map, 1, EFF_ALEN(len));
-
-    afl->blocks_eff_select += EFF_ALEN(len);
-
-  } else {
-
-    afl->blocks_eff_select += eff_cnt;
-
-  }
+  memset(eff_map, 1, EFF_ALEN(len));
+  afl->blocks_eff_select += EFF_ALEN(len);
 
   afl->blocks_eff_total += EFF_ALEN(len);
 
@@ -921,6 +921,10 @@ u8 fuzz_one_original(afl_state_t *afl) {
 
     }
 
+    if (!skip_eff_map[i]) continue;
+
+    if (is_det_timeout(before_det_time, 0)) { goto custom_mutator_stage; }
+
     afl->stage_cur_byte = i;
 
     *(u16 *)(out_buf + i) ^= 0xFFFF;
@@ -967,6 +971,10 @@ u8 fuzz_one_original(afl_state_t *afl) {
 
     }
 
+    if (!skip_eff_map[i]) continue;
+
+    if (is_det_timeout(before_det_time, 0)) { goto custom_mutator_stage; }
+
     afl->stage_cur_byte = i;
 
     *(u32 *)(out_buf + i) ^= 0xFFFFFFFF;
@@ -1023,6 +1031,10 @@ skip_bitflip:
 
     }
 
+    if (!skip_eff_map[i]) continue;
+
+    if (is_det_timeout(before_det_time, 0)) { goto custom_mutator_stage; }
+
     afl->stage_cur_byte = i;
 
     for (j = 1; j <= ARITH_MAX; ++j) {
@@ -1110,6 +1122,10 @@ skip_bitflip:
 
     }
 
+    if (!skip_eff_map[i]) continue;
+
+    if (is_det_timeout(before_det_time, 0)) { goto custom_mutator_stage; }
+
     afl->stage_cur_byte = i;
 
     for (j = 1; j <= ARITH_MAX; ++j) {
@@ -1244,6 +1260,10 @@ skip_bitflip:
 
     }
 
+    if (!skip_eff_map[i]) continue;
+
+    if (is_det_timeout(before_det_time, 0)) { goto custom_mutator_stage; }
+
     afl->stage_cur_byte = i;
 
     for (j = 1; j <= ARITH_MAX; ++j) {
@@ -1381,6 +1401,10 @@ skip_arith:
 
     }
 
+    if (!skip_eff_map[i]) continue;
+
+    if (is_det_timeout(before_det_time, 0)) { goto custom_mutator_stage; }
+
     afl->stage_cur_byte = i;
 
     for (j = 0; j < (u32)sizeof(interesting_8); ++j) {
@@ -1444,6 +1468,10 @@ skip_arith:
 
     }
 
+    if (!skip_eff_map[i]) continue;
+
+    if (is_det_timeout(before_det_time, 0)) { goto custom_mutator_stage; }
+
     afl->stage_cur_byte = i;
 
     for (j = 0; j < sizeof(interesting_16) / 2; ++j) {
@@ -1536,6 +1564,10 @@ skip_arith:
 
     }
 
+    if (!skip_eff_map[i]) continue;
+
+    if (is_det_timeout(before_det_time, 0)) { goto custom_mutator_stage; }
+
     afl->stage_cur_byte = i;
 
     for (j = 0; j < sizeof(interesting_32) / 4; ++j) {
@@ -1626,6 +1658,10 @@ skip_interest:
 
     u32 last_len = 0;
 
+    if (!skip_eff_map[i]) continue;
+
+    if (is_det_timeout(before_det_time, 0)) { goto custom_mutator_stage; }
+
     afl->stage_cur_byte = i;
 
     /* Extras are sorted by size, from smallest to largest. This means
@@ -1693,6 +1729,10 @@ skip_interest:
 
   for (i = 0; i <= (u32)len; ++i) {
 
+    if (!skip_eff_map[i % len]) continue;
+
+    if (is_det_timeout(before_det_time, 0)) { goto custom_mutator_stage; }
+
     afl->stage_cur_byte = i;
 
     for (j = 0; j < afl->extras_cnt; ++j) {
@@ -1755,6 +1795,10 @@ skip_user_extras:
 
     u32 last_len = 0;
 
+    if (!skip_eff_map[i]) continue;
+
+    if (is_det_timeout(before_det_time, 0)) { goto custom_mutator_stage; }
+
     afl->stage_cur_byte = i;
 
     u32 min_extra_len = MIN(afl->a_extras_cnt, (u32)USE_AUTO_EXTRAS);
@@ -1813,6 +1857,10 @@ skip_user_extras:
 
   for (i = 0; i <= (u32)len; ++i) {
 
+    if (!skip_eff_map[i % len]) continue;
+
+    if (is_det_timeout(before_det_time, 0)) { goto custom_mutator_stage; }
+
     afl->stage_cur_byte = i;
 
     for (j = 0; j < afl->a_extras_cnt; ++j) {
@@ -2020,6 +2068,19 @@ custom_mutator_stage:
 
 havoc_stage:
 
+#ifdef INTROSPECTION
+
+  if (!is_logged) {
+
+    is_logged = 1;
+    before_havoc_findings = afl->queued_items;
+    before_havoc_edges = count_non_255_bytes(afl, afl->virgin_bits);
+    before_havoc_time = get_cur_time();
+
+  }
+
+#endif
+
   if (unlikely(afl->custom_only)) {
 
     /* Force UI update */
@@ -3430,6 +3491,25 @@ retry_splicing:
 
   ret_val = 0;
 
+#ifdef INTROSPECTION
+
+  afl->havoc_prof->queued_det_stage =
+      before_havoc_findings - before_det_findings;
+  afl->havoc_prof->queued_havoc_stage =
+      afl->queued_items - before_havoc_findings;
+  afl->havoc_prof->total_queued_det += afl->havoc_prof->queued_det_stage;
+  afl->havoc_prof->edge_det_stage = before_havoc_edges - before_det_edges;
+  afl->havoc_prof->edge_havoc_stage =
+      count_non_255_bytes(afl, afl->virgin_bits) - before_havoc_edges;
+  afl->havoc_prof->total_det_edge += afl->havoc_prof->edge_det_stage;
+  afl->havoc_prof->det_stage_time = before_havoc_time - before_det_time;
+  afl->havoc_prof->havoc_stage_time = get_cur_time() - before_havoc_time;
+  afl->havoc_prof->total_det_time += afl->havoc_prof->det_stage_time;
+
+  plot_profile_data(afl, afl->queue_cur);
+
+#endif
+
 /* we are through with this queue entry - for this iteration */
 abandon_entry:
 
diff --git a/src/afl-fuzz-queue.c b/src/afl-fuzz-queue.c
index 4b9627f7..67931bba 100644
--- a/src/afl-fuzz-queue.c
+++ b/src/afl-fuzz-queue.c
@@ -664,6 +664,8 @@ void add_to_queue(afl_state_t *afl, u8 *fname, u32 len, u8 passed_det) {
 
   }
 
+  q->skipdet_e = (struct skipdet_entry *)ck_alloc(sizeof(struct skipdet_entry));
+
 }
 
 /* Destroy the entire queue. */
@@ -679,6 +681,15 @@ void destroy_queue(afl_state_t *afl) {
     q = afl->queue_buf[i];
     ck_free(q->fname);
     ck_free(q->trace_mini);
+    if (q->skipdet_e) {
+
+      if (q->skipdet_e->done_inf_map) ck_free(q->skipdet_e->done_inf_map);
+      if (q->skipdet_e->skip_eff_map) ck_free(q->skipdet_e->skip_eff_map);
+
+      ck_free(q->skipdet_e);
+
+    }
+
     ck_free(q);
 
   }
diff --git a/src/afl-fuzz-skipdet.c b/src/afl-fuzz-skipdet.c
new file mode 100644
index 00000000..e52d59a3
--- /dev/null
+++ b/src/afl-fuzz-skipdet.c
@@ -0,0 +1,403 @@
+
+
+#include "afl-fuzz.h"
+
+void flip_range(u8 *input, u32 pos, u32 size) {
+
+  for (u32 i = 0; i < size; i++)
+    input[pos + i] ^= 0xFF;
+
+  return;
+
+}
+
+#define MAX_EFF_TIMEOUT (10 * 60 * 1000)
+#define MAX_DET_TIMEOUT (15 * 60 * 1000)
+u8 is_det_timeout(u64 cur_ms, u8 is_flip) {
+
+  if (is_flip) {
+
+    if (unlikely(get_cur_time() - cur_ms > MAX_EFF_TIMEOUT)) return 1;
+
+  } else {
+
+    if (unlikely(get_cur_time() - cur_ms > MAX_DET_TIMEOUT)) return 1;
+
+  }
+
+  return 0;
+
+}
+
+/* decide if the seed should be deterministically fuzzed */
+
+u8 should_det_fuzz(afl_state_t *afl, struct queue_entry *q) {
+
+  if (!afl->skipdet_g->virgin_det_bits) {
+
+    afl->skipdet_g->virgin_det_bits =
+        (u8 *)ck_alloc(sizeof(u8) * afl->fsrv.map_size);
+
+  }
+
+  if (!q->favored || q->passed_det) return 0;
+  if (!q->trace_mini) return 0;
+
+  if (!afl->skipdet_g->last_cov_undet)
+    afl->skipdet_g->last_cov_undet = get_cur_time();
+
+  if (get_cur_time() - afl->skipdet_g->last_cov_undet >= THRESHOLD_DEC_TIME) {
+
+    if (afl->skipdet_g->undet_bits_threshold >= 2) {
+
+      afl->skipdet_g->undet_bits_threshold *= 0.75;
+      afl->skipdet_g->last_cov_undet = get_cur_time();
+
+    }
+
+  }
+
+  u32 new_det_bits = 0;
+
+  for (u32 i = 0; i < afl->fsrv.map_size; i++) {
+
+    if (unlikely(q->trace_mini[i >> 3] & (1 << (i & 7)))) {
+
+      if (!afl->skipdet_g->virgin_det_bits[i]) { new_det_bits++; }
+
+    }
+
+  }
+
+  if (!afl->skipdet_g->undet_bits_threshold)
+    afl->skipdet_g->undet_bits_threshold = new_det_bits * 0.05;
+
+  if (new_det_bits >= afl->skipdet_g->undet_bits_threshold) {
+
+    afl->skipdet_g->last_cov_undet = get_cur_time();
+    q->skipdet_e->undet_bits = new_det_bits;
+
+    for (u32 i = 0; i < afl->fsrv.map_size; i++) {
+
+      if (unlikely(q->trace_mini[i >> 3] & (1 << (i & 7)))) {
+
+        if (!afl->skipdet_g->virgin_det_bits[i])
+          afl->skipdet_g->virgin_det_bits[i] = 1;
+
+      }
+
+    }
+
+    return 1;
+
+  }
+
+  return 0;
+
+}
+
+/*
+  consists of two stages that
+  return 0 if exec failed.
+*/
+
+u8 skip_deterministic_stage(afl_state_t *afl, u8 *orig_buf, u8 *out_buf,
+                            u32 len, u64 before_det_time) {
+
+  u64 orig_hit_cnt, new_hit_cnt;
+
+  if (afl->queue_cur->skipdet_e->done_eff) return 1;
+
+  if (!should_det_fuzz(afl, afl->queue_cur)) return 1;
+
+  /* Add check to make sure that for seeds without too much undet bits,
+     we ignore them */
+
+  /******************
+   * SKIP INFERENCE *
+   ******************/
+
+  afl->stage_short = "inf";
+  afl->stage_name = "inference";
+  afl->stage_cur = 0;
+  orig_hit_cnt = afl->queued_items + afl->saved_crashes;
+
+  u8 *inf_eff_map = (u8 *)ck_alloc(sizeof(u8) * len);
+  memset(inf_eff_map, 1, sizeof(u8) * len);
+
+  if (common_fuzz_stuff(afl, orig_buf, len)) { return 0; }
+
+  u64 prev_cksum = hash64(afl->fsrv.trace_bits, afl->fsrv.map_size, HASH_CONST);
+  u64 _prev_cksum = prev_cksum;
+
+  if (MINIMAL_BLOCK_SIZE * 8 < len) {
+
+    // u64 size_skiped = 0, quick_skip_exec = total_execs, quick_skip_time =
+    // get_cur_time();
+    u64 pre_inf_exec = afl->fsrv.total_execs, pre_inf_time = get_cur_time();
+
+    /* if determine stage time / input size is too small, just go ahead */
+
+    u32 pos = 0, cur_block_size = MINIMAL_BLOCK_SIZE, max_block_size = len / 8;
+
+    while (pos < len - 1) {
+
+      cur_block_size = MINIMAL_BLOCK_SIZE;
+
+      while (cur_block_size < max_block_size) {
+
+        u32 flip_block_size =
+            (cur_block_size + pos < len) ? cur_block_size : len - 1 - pos;
+
+        afl->stage_cur += 1;
+
+        flip_range(out_buf, pos, flip_block_size);
+
+        if (common_fuzz_stuff(afl, out_buf, len)) return 0;
+
+        flip_range(out_buf, pos, flip_block_size);
+
+        u64 cksum =
+            hash64(afl->fsrv.trace_bits, afl->fsrv.map_size, HASH_CONST);
+
+        // printf("Now trying range %d with %d, %s.\n", pos, cur_block_size,
+        //     (cksum == prev_cksum) ? (u8*)"Yes" : (u8*) "Not");
+
+        /* continue until we fail or exceed length */
+        if (cksum == _prev_cksum) {
+
+          cur_block_size *= 2;
+
+          if (cur_block_size >= len - 1 - pos) break;
+
+        } else {
+
+          break;
+
+        }
+
+      }
+
+      if (cur_block_size == MINIMAL_BLOCK_SIZE) {
+
+        /* we failed early on*/
+
+        pos += cur_block_size;
+
+      } else {
+
+        u32 cur_skip_len = (cur_block_size / 2 + pos < len)
+                               ? (cur_block_size / 2)
+                               : (len - pos - 1);
+
+        memset(inf_eff_map + pos, 0, cur_skip_len);
+
+        afl->skipdet_g->inf_prof->inf_skipped_bytes += cur_skip_len;
+
+        pos += cur_skip_len;
+
+      }
+
+    }
+
+    afl->skipdet_g->inf_prof->inf_execs_cost +=
+        (afl->fsrv.total_execs - pre_inf_exec);
+    afl->skipdet_g->inf_prof->inf_time_cost += (get_cur_time() - pre_inf_time);
+    // PFATAL("Done, now have %d bytes skipped, with exec %lld, time %lld.\n",
+    // afl->inf_skipped_bytes, afl->inf_execs_cost, afl->inf_time_cost);
+
+  } else
+
+    memset(inf_eff_map, 1, len);
+
+  new_hit_cnt = afl->queued_items + afl->saved_crashes;
+
+  afl->stage_finds[STAGE_INF] += new_hit_cnt - orig_hit_cnt;
+  afl->stage_cycles[STAGE_INF] += afl->stage_cur;
+
+  /****************************
+   * Quick Skip Effective Map *
+   ****************************/
+
+  /* Quick Effective Map Calculation */
+
+  afl->stage_short = "quick";
+  afl->stage_name = "quick eff";
+  afl->stage_cur = 0;
+  afl->stage_max = 32 * 1024;
+
+  orig_hit_cnt = afl->queued_items + afl->saved_crashes;
+
+  u32 before_skip_inf = afl->queued_items;
+
+  /* clean all the eff bytes, since previous eff bytes are already fuzzed */
+  u8 *skip_eff_map = afl->queue_cur->skipdet_e->skip_eff_map,
+     *done_inf_map = afl->queue_cur->skipdet_e->done_inf_map;
+
+  if (!skip_eff_map) {
+
+    skip_eff_map = (u8 *)ck_alloc(sizeof(u8) * len);
+    afl->queue_cur->skipdet_e->skip_eff_map = skip_eff_map;
+
+  } else {
+
+    memset(skip_eff_map, 0, sizeof(u8) * len);
+
+  }
+
+  /* restore the starting point */
+  if (!done_inf_map) {
+
+    done_inf_map = (u8 *)ck_alloc(sizeof(u8) * len);
+    afl->queue_cur->skipdet_e->done_inf_map = done_inf_map;
+
+  } else {
+
+    for (afl->stage_cur = 0; afl->stage_cur < len; afl->stage_cur++) {
+
+      if (done_inf_map[afl->stage_cur] == 0) break;
+
+    }
+
+  }
+
+  /* depending on the seed's performance, we could search eff bytes
+     for multiple rounds */
+
+  u8 eff_round_continue = 1, eff_round_done = 0, done_eff = 0, repeat_eff = 0,
+     fuzz_nearby = 0, *non_eff_bytes = 0;
+
+  u64 before_eff_execs = afl->fsrv.total_execs;
+
+  if (getenv("REPEAT_EFF")) repeat_eff = 1;
+  if (getenv("FUZZ_NEARBY")) fuzz_nearby = 1;
+
+  if (fuzz_nearby) {
+
+    non_eff_bytes = (u8 *)ck_alloc(sizeof(u8) * len);
+
+    // clean exec cksum
+    if (common_fuzz_stuff(afl, out_buf, len)) { return 0; }
+    prev_cksum = hash64(afl->fsrv.trace_bits, afl->fsrv.map_size, HASH_CONST);
+
+  }
+
+  do {
+
+    eff_round_continue = 0;
+    afl->stage_max = 32 * 1024;
+
+    for (; afl->stage_cur < afl->stage_max && afl->stage_cur < len;
+         ++afl->stage_cur) {
+
+      afl->stage_cur_byte = afl->stage_cur;
+
+      if (!inf_eff_map[afl->stage_cur_byte] ||
+          skip_eff_map[afl->stage_cur_byte])
+        continue;
+
+      if (is_det_timeout(before_det_time, 1)) { goto cleanup_skipdet; }
+
+      u8 orig = out_buf[afl->stage_cur_byte], replace = rand_below(afl, 256);
+
+      while (replace == orig) {
+
+        replace = rand_below(afl, 256);
+
+      }
+
+      out_buf[afl->stage_cur_byte] = replace;
+
+      before_skip_inf = afl->queued_items;
+
+      if (common_fuzz_stuff(afl, out_buf, len)) { return 0; }
+
+      out_buf[afl->stage_cur_byte] = orig;
+
+      if (fuzz_nearby) {
+
+        if (prev_cksum ==
+            hash64(afl->fsrv.trace_bits, afl->fsrv.map_size, HASH_CONST)) {
+
+          non_eff_bytes[afl->stage_cur_byte] = 1;
+
+        }
+
+      }
+
+      if (afl->queued_items != before_skip_inf) {
+
+        skip_eff_map[afl->stage_cur_byte] = 1;
+        afl->queue_cur->skipdet_e->quick_eff_bytes += 1;
+
+        if (afl->stage_max < MAXIMUM_QUICK_EFF_EXECS) { afl->stage_max *= 2; }
+
+        if (afl->stage_max == MAXIMUM_QUICK_EFF_EXECS && repeat_eff)
+          eff_round_continue = 1;
+
+      }
+
+      done_inf_map[afl->stage_cur_byte] = 1;
+
+    }
+
+    afl->stage_cur = 0;
+    done_eff = 1;
+
+    if (++eff_round_done >= 8) break;
+
+  } while (eff_round_continue);
+
+  new_hit_cnt = afl->queued_items + afl->saved_crashes;
+
+  afl->stage_finds[STAGE_QUICK] += new_hit_cnt - orig_hit_cnt;
+  afl->stage_cycles[STAGE_QUICK] += (afl->fsrv.total_execs - before_eff_execs);
+
+cleanup_skipdet:
+
+  if (fuzz_nearby) {
+
+    u8 *nearby_bytes = (u8 *)ck_alloc(sizeof(u8) * len);
+
+    u32 i = 3;
+    while (i < len) {
+
+      // assume DWORD size, from i - 3 -> i + 3
+      if (skip_eff_map[i]) {
+
+        u32 fill_length = (i + 3 < len) ? 7 : len - i + 2;
+        memset(nearby_bytes + i - 3, 1, fill_length);
+        i += 3;
+
+      } else
+
+        i += 1;
+
+    }
+
+    for (i = 0; i < len; i++) {
+
+      if (nearby_bytes[i] && !non_eff_bytes[i]) skip_eff_map[i] = 1;
+
+    }
+
+    ck_free(nearby_bytes);
+    ck_free(non_eff_bytes);
+
+  }
+
+  if (done_eff) {
+
+    afl->queue_cur->skipdet_e->continue_inf = 0;
+    afl->queue_cur->skipdet_e->done_eff = 1;
+
+  } else {
+
+    afl->queue_cur->skipdet_e->continue_inf = 1;
+
+  }
+
+  return 1;
+
+}
+
diff --git a/src/afl-fuzz-state.c b/src/afl-fuzz-state.c
index 7d6fdfb9..6cf580ce 100644
--- a/src/afl-fuzz-state.c
+++ b/src/afl-fuzz-state.c
@@ -102,7 +102,7 @@ void afl_state_init(afl_state_t *afl, uint32_t map_size) {
   afl->stats_update_freq = 1;
   afl->stats_file_update_freq_msecs = STATS_UPDATE_SEC * 1000;
   afl->stats_avg_exec = 0;
-  afl->skip_deterministic = 1;
+  afl->skip_deterministic = 0;
   afl->sync_time = SYNC_TIME;
   afl->cmplog_lvl = 2;
   afl->min_length = 1;
@@ -140,6 +140,14 @@ void afl_state_init(afl_state_t *afl, uint32_t map_size) {
   afl->fsrv.child_pid = -1;
   afl->fsrv.out_dir_fd = -1;
 
+  /* Init SkipDet */
+  afl->skipdet_g =
+      (struct skipdet_global *)ck_alloc(sizeof(struct skipdet_global));
+  afl->skipdet_g->inf_prof =
+      (struct inf_profile *)ck_alloc(sizeof(struct inf_profile));
+  afl->havoc_prof =
+      (struct havoc_profile *)ck_alloc(sizeof(struct havoc_profile));
+
   init_mopt_globals(afl);
 
   list_append(&afl_states, afl);
diff --git a/src/afl-fuzz-stats.c b/src/afl-fuzz-stats.c
index deb28b7a..4b83ad29 100644
--- a/src/afl-fuzz-stats.c
+++ b/src/afl-fuzz-stats.c
@@ -502,6 +502,44 @@ void maybe_update_plot_file(afl_state_t *afl, u32 t_bytes, double bitmap_cvg,
 
 }
 
+/* Log deterministic stage efficiency */
+
+void plot_profile_data(afl_state_t *afl, struct queue_entry *q) {
+
+  u64 current_ms = get_cur_time() - afl->start_time;
+
+  u32    current_edges = count_non_255_bytes(afl, afl->virgin_bits);
+  double det_finding_rate = (double)afl->havoc_prof->total_det_edge * 100.0 /
+                            (double)current_edges,
+         det_time_rate = (double)afl->havoc_prof->total_det_time * 100.0 /
+                         (double)current_ms;
+
+  u32 ndet_bits = 0;
+  for (u32 i = 0; i < afl->fsrv.map_size; i++) {
+
+    if (afl->skipdet_g->virgin_det_bits[i]) ndet_bits += 1;
+
+  }
+
+  double det_fuzzed_rate = (double)ndet_bits * 100.0 / (double)current_edges;
+
+  fprintf(afl->fsrv.det_plot_file,
+          "[%02lld:%02lld:%02lld] fuzz %d (%d), find %d/%d among %d(%02.2f) "
+          "and spend %lld/%lld(%02.2f), cover %02.2f yet, %d/%d undet bits, "
+          "continue %d.\n",
+          current_ms / 1000 / 3600, (current_ms / 1000 / 60) % 60,
+          (current_ms / 1000) % 60, afl->current_entry, q->fuzz_level,
+          afl->havoc_prof->edge_det_stage, afl->havoc_prof->edge_havoc_stage,
+          current_edges, det_finding_rate,
+          afl->havoc_prof->det_stage_time / 1000,
+          afl->havoc_prof->havoc_stage_time / 1000, det_time_rate,
+          det_fuzzed_rate, q->skipdet_e->undet_bits,
+          afl->skipdet_g->undet_bits_threshold, q->skipdet_e->continue_inf);
+
+  fflush(afl->fsrv.det_plot_file);
+
+}
+
 /* Check terminal dimensions after resize. */
 
 static void check_term_size(afl_state_t *afl) {
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index 8cf6c735..7db1aeb3 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -955,14 +955,14 @@ int main(int argc, char **argv_orig, char **envp) {
 
       break;
 
-      case 'D':                                    /* enforce deterministic */
+      case 'D':                                         /* no deterministic */
 
-        afl->skip_deterministic = 0;
+        afl->skip_deterministic = 1;
         break;
 
-      case 'd':                                       /* skip deterministic */
+      case 'd':                                    /* partial deterministic */
 
-        afl->skip_deterministic = 1;
+        afl->skip_deterministic = 0;
         break;
 
       case 'B':                                              /* load bitmap */
@@ -3031,6 +3031,11 @@ stop_fuzzing:
   if (frida_afl_preload) { ck_free(frida_afl_preload); }
 
   fclose(afl->fsrv.plot_file);
+
+  #ifdef INTROSPECTION
+  fclose(afl->fsrv.det_plot_file);
+  #endif
+
   destroy_queue(afl);
   destroy_extras(afl);
   destroy_custom_mutators(afl);
-- 
cgit 1.4.1


From ed1a6f8a570c6fcabee962f402d8d58f6cea77b7 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Sat, 3 Feb 2024 11:01:31 +0100
Subject: 2024 v4.10c release

---
 GNUmakefile.gcc_plugin                            | 2 +-
 README.md                                         | 4 ++--
 afl-cmin.bash                                     | 2 +-
 afl-whatsup                                       | 2 +-
 docs/Changelog.md                                 | 3 +--
 frida_mode/Scripting.md                           | 2 +-
 frida_mode/test/cmplog/cmplog.c                   | 2 +-
 frida_mode/test/deferred/testinstr.c              | 2 +-
 frida_mode/test/dynamic/testinstr.c               | 2 +-
 frida_mode/test/entry_point/testinstr.c           | 2 +-
 frida_mode/test/exe/testinstr.c                   | 2 +-
 frida_mode/test/js/test.c                         | 2 +-
 frida_mode/test/js/test2.c                        | 2 +-
 frida_mode/test/output/testinstr.c                | 2 +-
 frida_mode/test/perf/perf.c                       | 2 +-
 frida_mode/test/persistent_ret/testinstr.c        | 2 +-
 frida_mode/test/testinstr/testinstr.c             | 2 +-
 frida_mode/test/unstable/unstable.c               | 2 +-
 frida_mode/util/frida_get_symbol_addr.sh          | 2 +-
 include/afl-as.h                                  | 2 +-
 include/afl-fuzz.h                                | 2 +-
 include/afl-prealloc.h                            | 2 +-
 include/alloc-inl.h                               | 2 +-
 include/cmplog.h                                  | 2 +-
 include/common.h                                  | 2 +-
 include/config.h                                  | 4 ++--
 include/debug.h                                   | 2 +-
 include/forkserver.h                              | 2 +-
 include/hash.h                                    | 2 +-
 include/list.h                                    | 2 +-
 include/sharedmem.h                               | 2 +-
 include/snapshot-inl.h                            | 2 +-
 include/types.h                                   | 2 +-
 include/xxhash.h                                  | 2 +-
 instrumentation/afl-compiler-rt.o.c               | 2 +-
 instrumentation/afl-gcc-cmplog-pass.so.cc         | 2 +-
 instrumentation/afl-gcc-cmptrs-pass.so.cc         | 2 +-
 instrumentation/afl-gcc-common.h                  | 2 +-
 instrumentation/afl-gcc-pass.so.cc                | 2 +-
 instrumentation/afl-llvm-dict2file.so.cc          | 2 +-
 instrumentation/afl-llvm-lto-instrumentlist.so.cc | 2 +-
 instrumentation/afl-llvm-pass.so.cc               | 2 +-
 instrumentation/cmplog-instructions-pass.cc       | 2 +-
 instrumentation/cmplog-routines-pass.cc           | 2 +-
 instrumentation/cmplog-switches-pass.cc           | 2 +-
 instrumentation/injection-pass.cc                 | 2 +-
 qemu_mode/build_qemu_support.sh                   | 2 +-
 qemu_mode/fastexit/Makefile                       | 2 +-
 qemu_mode/libcompcov/Makefile                     | 2 +-
 qemu_mode/libcompcov/compcovtest.cc               | 2 +-
 qemu_mode/libcompcov/libcompcov.so.c              | 2 +-
 qemu_mode/libqasan/Makefile                       | 2 +-
 qemu_mode/libqasan/hooks.c                        | 2 +-
 qemu_mode/libqasan/libqasan.c                     | 2 +-
 qemu_mode/libqasan/libqasan.h                     | 2 +-
 qemu_mode/libqasan/malloc.c                       | 2 +-
 qemu_mode/libqasan/patch.c                        | 2 +-
 qemu_mode/libqasan/string.c                       | 2 +-
 qemu_mode/libqasan/uninstrument.c                 | 2 +-
 qemu_mode/unsigaction/Makefile                    | 2 +-
 qemu_mode/util/qemu_get_symbol_addr.sh            | 2 +-
 src/afl-analyze.c                                 | 2 +-
 src/afl-as.c                                      | 2 +-
 src/afl-cc.c                                      | 2 +-
 src/afl-common.c                                  | 2 +-
 src/afl-forkserver.c                              | 2 +-
 src/afl-fuzz-bitmap.c                             | 2 +-
 src/afl-fuzz-cmplog.c                             | 2 +-
 src/afl-fuzz-extras.c                             | 2 +-
 src/afl-fuzz-init.c                               | 2 +-
 src/afl-fuzz-mutators.c                           | 2 +-
 src/afl-fuzz-one.c                                | 2 +-
 src/afl-fuzz-python.c                             | 2 +-
 src/afl-fuzz-queue.c                              | 2 +-
 src/afl-fuzz-redqueen.c                           | 2 +-
 src/afl-fuzz-run.c                                | 2 +-
 src/afl-fuzz-state.c                              | 2 +-
 src/afl-fuzz-stats.c                              | 2 +-
 src/afl-fuzz.c                                    | 2 +-
 src/afl-gotcpu.c                                  | 2 +-
 src/afl-ld-lto.c                                  | 2 +-
 src/afl-sharedmem.c                               | 2 +-
 src/afl-showmap.c                                 | 2 +-
 src/afl-tmin.c                                    | 2 +-
 test-instr.c                                      | 2 +-
 unicorn_mode/build_unicorn_support.sh             | 2 +-
 utils/afl_network_proxy/afl-network-client.c      | 2 +-
 utils/afl_network_proxy/afl-network-server.c      | 2 +-
 utils/afl_proxy/afl-proxy.c                       | 2 +-
 utils/afl_untracer/afl-untracer.c                 | 2 +-
 utils/afl_untracer/libtestinstr.c                 | 2 +-
 utils/argv_fuzzing/Makefile                       | 2 +-
 utils/argv_fuzzing/argvfuzz.c                     | 2 +-
 utils/distributed_fuzzing/sync_script.sh          | 2 +-
 utils/libdislocator/libdislocator.so.c            | 2 +-
 utils/libtokencap/libtokencap.so.c                | 2 +-
 utils/persistent_mode/test-instr.c                | 2 +-
 97 files changed, 99 insertions(+), 100 deletions(-)

(limited to 'include')

diff --git a/GNUmakefile.gcc_plugin b/GNUmakefile.gcc_plugin
index 16c98399..8f06792d 100644
--- a/GNUmakefile.gcc_plugin
+++ b/GNUmakefile.gcc_plugin
@@ -11,7 +11,7 @@
 # from Laszlo Szekeres.
 #
 # Copyright 2015 Google Inc. All rights reserved.
-# Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+# Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/README.md b/README.md
index fd48cb14..f713e971 100644
--- a/README.md
+++ b/README.md
@@ -2,9 +2,9 @@
 
 <img align="right" src="https://raw.githubusercontent.com/AFLplusplus/Website/main/static/aflpp_bg.svg" alt="AFL++ logo" width="250" heigh="250">
 
-Release version: [4.09c](https://github.com/AFLplusplus/AFLplusplus/releases)
+Release version: [4.10c](https://github.com/AFLplusplus/AFLplusplus/releases)
 
-GitHub version: 4.10a
+GitHub version: 4.10c
 
 Repository:
 [https://github.com/AFLplusplus/AFLplusplus](https://github.com/AFLplusplus/AFLplusplus)
diff --git a/afl-cmin.bash b/afl-cmin.bash
index fda48fb4..6c271220 100755
--- a/afl-cmin.bash
+++ b/afl-cmin.bash
@@ -7,7 +7,7 @@
 #
 # Copyright 2014, 2015 Google Inc. All rights reserved.
 #
-# Copyright 2019-2023 AFLplusplus
+# Copyright 2019-2024 AFLplusplus
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/afl-whatsup b/afl-whatsup
index 5b7cbcd6..aa081e41 100755
--- a/afl-whatsup
+++ b/afl-whatsup
@@ -6,7 +6,7 @@
 # Originally written by Michal Zalewski
 #
 # Copyright 2015 Google Inc. All rights reserved.
-# Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+# Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/docs/Changelog.md b/docs/Changelog.md
index 29081549..48003f4b 100644
--- a/docs/Changelog.md
+++ b/docs/Changelog.md
@@ -3,7 +3,7 @@
   This is the list of all noteworthy changes made in every public
   release of the tool. See README.md for the general instruction manual.
 
-### Version ++4.10a (dev)
+### Version ++4.10c (release)
   - afl-fuzz:
     - default power schedule is now EXPLORE, due a fix in fast schedules
       explore is slightly better now.
@@ -34,7 +34,6 @@
   - updated the custom grammar mutator
   - document afl-cmin does not work on macOS (but afl-cmin.bash does)
 
-
 ### Version ++4.09c (release)
   - afl-fuzz:
     - fixed the new mutation implementation for two bugs
diff --git a/frida_mode/Scripting.md b/frida_mode/Scripting.md
index dfd09e7b..653687f0 100644
--- a/frida_mode/Scripting.md
+++ b/frida_mode/Scripting.md
@@ -390,7 +390,7 @@ Consider the [following](test/js/test2.c) test code...
    --------------------------------------------------------
    Originally written by Michal Zalewski
    Copyright 2014 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
    You may obtain a copy of the License at:
diff --git a/frida_mode/test/cmplog/cmplog.c b/frida_mode/test/cmplog/cmplog.c
index 2565b35c..d397f36e 100644
--- a/frida_mode/test/cmplog/cmplog.c
+++ b/frida_mode/test/cmplog/cmplog.c
@@ -2,7 +2,7 @@
 //
 // Author: Mateusz Jurczyk (mjurczyk@google.com)
 //
-// Copyright 2019-2023 Google LLC
+// Copyright 2019-2024 Google LLC
 //
 // Licensed under the Apache License, Version 2.0 (the "License");
 // you may not use this file except in compliance with the License.
diff --git a/frida_mode/test/deferred/testinstr.c b/frida_mode/test/deferred/testinstr.c
index 0ab44582..4e5124ed 100644
--- a/frida_mode/test/deferred/testinstr.c
+++ b/frida_mode/test/deferred/testinstr.c
@@ -3,7 +3,7 @@
    --------------------------------------------------------
    Originally written by Michal Zalewski
    Copyright 2014 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
    You may obtain a copy of the License at:
diff --git a/frida_mode/test/dynamic/testinstr.c b/frida_mode/test/dynamic/testinstr.c
index 8b285f6d..0abc61fd 100644
--- a/frida_mode/test/dynamic/testinstr.c
+++ b/frida_mode/test/dynamic/testinstr.c
@@ -3,7 +3,7 @@
    --------------------------------------------------------
    Originally written by Michal Zalewski
    Copyright 2014 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
    You may obtain a copy of the License at:
diff --git a/frida_mode/test/entry_point/testinstr.c b/frida_mode/test/entry_point/testinstr.c
index 24d9a615..75e71bda 100644
--- a/frida_mode/test/entry_point/testinstr.c
+++ b/frida_mode/test/entry_point/testinstr.c
@@ -3,7 +3,7 @@
    --------------------------------------------------------
    Originally written by Michal Zalewski
    Copyright 2014 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
    You may obtain a copy of the License at:
diff --git a/frida_mode/test/exe/testinstr.c b/frida_mode/test/exe/testinstr.c
index d965502e..7b603659 100644
--- a/frida_mode/test/exe/testinstr.c
+++ b/frida_mode/test/exe/testinstr.c
@@ -3,7 +3,7 @@
    --------------------------------------------------------
    Originally written by Michal Zalewski
    Copyright 2014 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
    You may obtain a copy of the License at:
diff --git a/frida_mode/test/js/test.c b/frida_mode/test/js/test.c
index 87c9cdf6..9799bf3b 100644
--- a/frida_mode/test/js/test.c
+++ b/frida_mode/test/js/test.c
@@ -3,7 +3,7 @@
    --------------------------------------------------------
    Originally written by Michal Zalewski
    Copyright 2014 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
    You may obtain a copy of the License at:
diff --git a/frida_mode/test/js/test2.c b/frida_mode/test/js/test2.c
index 6b680a24..60b30eb5 100644
--- a/frida_mode/test/js/test2.c
+++ b/frida_mode/test/js/test2.c
@@ -3,7 +3,7 @@
    --------------------------------------------------------
    Originally written by Michal Zalewski
    Copyright 2014 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
    You may obtain a copy of the License at:
diff --git a/frida_mode/test/output/testinstr.c b/frida_mode/test/output/testinstr.c
index d965502e..7b603659 100644
--- a/frida_mode/test/output/testinstr.c
+++ b/frida_mode/test/output/testinstr.c
@@ -3,7 +3,7 @@
    --------------------------------------------------------
    Originally written by Michal Zalewski
    Copyright 2014 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
    You may obtain a copy of the License at:
diff --git a/frida_mode/test/perf/perf.c b/frida_mode/test/perf/perf.c
index d9626974..55efba26 100644
--- a/frida_mode/test/perf/perf.c
+++ b/frida_mode/test/perf/perf.c
@@ -3,7 +3,7 @@
    --------------------------------------------------------
    Originally written by Michal Zalewski
    Copyright 2014 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
    You may obtain a copy of the License at:
diff --git a/frida_mode/test/persistent_ret/testinstr.c b/frida_mode/test/persistent_ret/testinstr.c
index 12365ceb..85aa2b80 100644
--- a/frida_mode/test/persistent_ret/testinstr.c
+++ b/frida_mode/test/persistent_ret/testinstr.c
@@ -3,7 +3,7 @@
    --------------------------------------------------------
    Originally written by Michal Zalewski
    Copyright 2014 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
    You may obtain a copy of the License at:
diff --git a/frida_mode/test/testinstr/testinstr.c b/frida_mode/test/testinstr/testinstr.c
index d965502e..7b603659 100644
--- a/frida_mode/test/testinstr/testinstr.c
+++ b/frida_mode/test/testinstr/testinstr.c
@@ -3,7 +3,7 @@
    --------------------------------------------------------
    Originally written by Michal Zalewski
    Copyright 2014 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
    You may obtain a copy of the License at:
diff --git a/frida_mode/test/unstable/unstable.c b/frida_mode/test/unstable/unstable.c
index a87b6c74..16978e7e 100644
--- a/frida_mode/test/unstable/unstable.c
+++ b/frida_mode/test/unstable/unstable.c
@@ -3,7 +3,7 @@
    --------------------------------------------------------
    Originally written by Michal Zalewski
    Copyright 2014 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
    You may obtain a copy of the License at:
diff --git a/frida_mode/util/frida_get_symbol_addr.sh b/frida_mode/util/frida_get_symbol_addr.sh
index 2e682255..53d5b802 100755
--- a/frida_mode/util/frida_get_symbol_addr.sh
+++ b/frida_mode/util/frida_get_symbol_addr.sh
@@ -1,5 +1,5 @@
 #!/bin/bash
-# Copyright 2023 AFLplusplus
+# Copyright 2024 AFLplusplus
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/include/afl-as.h b/include/afl-as.h
index 486314e2..612f34f4 100644
--- a/include/afl-as.h
+++ b/include/afl-as.h
@@ -10,7 +10,7 @@
                      Dominik Maier <mail@dmnk.co>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index c2b09b2e..c24f39e2 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -10,7 +10,7 @@
                      Dominik Maier <mail@dmnk.co>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/include/afl-prealloc.h b/include/afl-prealloc.h
index d19a7b52..3c621d79 100644
--- a/include/afl-prealloc.h
+++ b/include/afl-prealloc.h
@@ -10,7 +10,7 @@
                      Dominik Maier <mail@dmnk.co>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/include/alloc-inl.h b/include/alloc-inl.h
index cff808b2..0aa417be 100644
--- a/include/alloc-inl.h
+++ b/include/alloc-inl.h
@@ -10,7 +10,7 @@
                      Dominik Maier <mail@dmnk.co>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/include/cmplog.h b/include/cmplog.h
index e4821444..6bfc146b 100644
--- a/include/cmplog.h
+++ b/include/cmplog.h
@@ -12,7 +12,7 @@
                      Dominik Maier <mail@dmnk.co>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/include/common.h b/include/common.h
index a9739a7d..0df07dee 100644
--- a/include/common.h
+++ b/include/common.h
@@ -10,7 +10,7 @@
                      Dominik Maier <mail@dmnk.co>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/include/config.h b/include/config.h
index 7ad73c2f..9349828f 100644
--- a/include/config.h
+++ b/include/config.h
@@ -10,7 +10,7 @@
                      Heiko Eissfeldt <heiko.eissfeldt@hexco.de>,
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
@@ -26,7 +26,7 @@
 /* Version string: */
 
 // c = release, a = volatile github dev, e = experimental branch
-#define VERSION "++4.10a"
+#define VERSION "++4.10c"
 
 /******************************************************
  *                                                    *
diff --git a/include/debug.h b/include/debug.h
index 234d8fc4..4b812f8e 100644
--- a/include/debug.h
+++ b/include/debug.h
@@ -10,7 +10,7 @@
                      Dominik Maier <mail@dmnk.co>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/include/forkserver.h b/include/forkserver.h
index f1d3b5b1..be7f9e8d 100644
--- a/include/forkserver.h
+++ b/include/forkserver.h
@@ -12,7 +12,7 @@
                      Dominik Maier <mail@dmnk.co>>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/include/hash.h b/include/hash.h
index 0243c5b7..5d56a108 100644
--- a/include/hash.h
+++ b/include/hash.h
@@ -15,7 +15,7 @@
    Other code written by Michal Zalewski
 
    Copyright 2016 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/include/list.h b/include/list.h
index 283bf035..441eccd3 100644
--- a/include/list.h
+++ b/include/list.h
@@ -10,7 +10,7 @@
                      Dominik Maier <mail@dmnk.co>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/include/sharedmem.h b/include/sharedmem.h
index d32bd845..4484066e 100644
--- a/include/sharedmem.h
+++ b/include/sharedmem.h
@@ -12,7 +12,7 @@
                      Dominik Maier <mail@dmnk.co>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/include/snapshot-inl.h b/include/snapshot-inl.h
index 3864e473..b2c81402 100644
--- a/include/snapshot-inl.h
+++ b/include/snapshot-inl.h
@@ -12,7 +12,7 @@
                      Dominik Maier <mail@dmnk.co>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/include/types.h b/include/types.h
index d6476d82..22332135 100644
--- a/include/types.h
+++ b/include/types.h
@@ -10,7 +10,7 @@
                      Dominik Maier <mail@dmnk.co>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/include/xxhash.h b/include/xxhash.h
index a8bd6f27..9a880470 100644
--- a/include/xxhash.h
+++ b/include/xxhash.h
@@ -1,7 +1,7 @@
 /*
  * xxHash - Extremely Fast Hash algorithm
  * Header File
- * Copyright (C) 2012-2023 Yann Collet
+ * Copyright (C) 2012-2024 Yann Collet
  *
  * BSD 2-Clause License (https://www.opensource.org/licenses/bsd-license.php)
  *
diff --git a/instrumentation/afl-compiler-rt.o.c b/instrumentation/afl-compiler-rt.o.c
index 8e55d6a0..caa3c3a8 100644
--- a/instrumentation/afl-compiler-rt.o.c
+++ b/instrumentation/afl-compiler-rt.o.c
@@ -3,7 +3,7 @@
    ------------------------------------------------
 
    Copyright 2015, 2016 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/instrumentation/afl-gcc-cmplog-pass.so.cc b/instrumentation/afl-gcc-cmplog-pass.so.cc
index b4e6fda9..774dd5fd 100644
--- a/instrumentation/afl-gcc-cmplog-pass.so.cc
+++ b/instrumentation/afl-gcc-cmplog-pass.so.cc
@@ -3,7 +3,7 @@
    Copyright 2014-2019 Free Software Foundation, Inc
    Copyright 2015, 2016 Google Inc. All rights reserved.
    Copyright 2019-2020 AFLplusplus Project. All rights reserved.
-   Copyright 2019-2023 AdaCore
+   Copyright 2019-2024 AdaCore
 
    Written by Alexandre Oliva <oliva@adacore.com>, based on the AFL++
    LLVM CmpLog pass by Andrea Fioraldi <andreafioraldi@gmail.com>, and
diff --git a/instrumentation/afl-gcc-cmptrs-pass.so.cc b/instrumentation/afl-gcc-cmptrs-pass.so.cc
index c56263dd..929a9d7a 100644
--- a/instrumentation/afl-gcc-cmptrs-pass.so.cc
+++ b/instrumentation/afl-gcc-cmptrs-pass.so.cc
@@ -3,7 +3,7 @@
    Copyright 2014-2019 Free Software Foundation, Inc
    Copyright 2015, 2016 Google Inc. All rights reserved.
    Copyright 2019-2020 AFLplusplus Project. All rights reserved.
-   Copyright 2019-2023 AdaCore
+   Copyright 2019-2024 AdaCore
 
    Written by Alexandre Oliva <oliva@adacore.com>, based on the AFL++
    LLVM CmpLog Routines pass by Andrea Fioraldi
diff --git a/instrumentation/afl-gcc-common.h b/instrumentation/afl-gcc-common.h
index 1d5eb466..80ded57d 100644
--- a/instrumentation/afl-gcc-common.h
+++ b/instrumentation/afl-gcc-common.h
@@ -2,7 +2,7 @@
 
    Copyright 2014-2019 Free Software Foundation, Inc
    Copyright 2015, 2016 Google Inc. All rights reserved.
-   Copyright 2019-2023 AdaCore
+   Copyright 2019-2024 AdaCore
 
    Written by Alexandre Oliva <oliva@adacore.com>, based on the AFL++
    GCC plugin.
diff --git a/instrumentation/afl-gcc-pass.so.cc b/instrumentation/afl-gcc-pass.so.cc
index 4d7fd0ef..41b1e5af 100644
--- a/instrumentation/afl-gcc-pass.so.cc
+++ b/instrumentation/afl-gcc-pass.so.cc
@@ -2,7 +2,7 @@
 
    Copyright 2014-2019 Free Software Foundation, Inc
    Copyright 2015, 2016 Google Inc. All rights reserved.
-   Copyright 2019-2023 AdaCore
+   Copyright 2019-2024 AdaCore
 
    Written by Alexandre Oliva <oliva@adacore.com>, based on the AFL
    LLVM pass by Laszlo Szekeres <lszekeres@google.com> and Michal
diff --git a/instrumentation/afl-llvm-dict2file.so.cc b/instrumentation/afl-llvm-dict2file.so.cc
index c60f3e06..ac497b5b 100644
--- a/instrumentation/afl-llvm-dict2file.so.cc
+++ b/instrumentation/afl-llvm-dict2file.so.cc
@@ -4,7 +4,7 @@
 
    Written by Marc Heuse <mh@mh-sec.de>
 
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/instrumentation/afl-llvm-lto-instrumentlist.so.cc b/instrumentation/afl-llvm-lto-instrumentlist.so.cc
index 61f97d77..e0899cd3 100644
--- a/instrumentation/afl-llvm-lto-instrumentlist.so.cc
+++ b/instrumentation/afl-llvm-lto-instrumentlist.so.cc
@@ -9,7 +9,7 @@
    from afl-as.c are Michal's fault.
 
    Copyright 2015, 2016 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/instrumentation/afl-llvm-pass.so.cc b/instrumentation/afl-llvm-pass.so.cc
index 052488a9..62f5023d 100644
--- a/instrumentation/afl-llvm-pass.so.cc
+++ b/instrumentation/afl-llvm-pass.so.cc
@@ -12,7 +12,7 @@
    NGRAM previous location coverage comes from Adrian Herrera.
 
    Copyright 2015, 2016 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/instrumentation/cmplog-instructions-pass.cc b/instrumentation/cmplog-instructions-pass.cc
index 8be8c294..dc60221e 100644
--- a/instrumentation/cmplog-instructions-pass.cc
+++ b/instrumentation/cmplog-instructions-pass.cc
@@ -5,7 +5,7 @@
    Written by Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2015, 2016 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/instrumentation/cmplog-routines-pass.cc b/instrumentation/cmplog-routines-pass.cc
index b27e06e0..78317d5d 100644
--- a/instrumentation/cmplog-routines-pass.cc
+++ b/instrumentation/cmplog-routines-pass.cc
@@ -5,7 +5,7 @@
    Written by Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2015, 2016 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/instrumentation/cmplog-switches-pass.cc b/instrumentation/cmplog-switches-pass.cc
index 01da6da7..3e05c13d 100644
--- a/instrumentation/cmplog-switches-pass.cc
+++ b/instrumentation/cmplog-switches-pass.cc
@@ -5,7 +5,7 @@
    Written by Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2015, 2016 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/instrumentation/injection-pass.cc b/instrumentation/injection-pass.cc
index 971b103b..2280208b 100644
--- a/instrumentation/injection-pass.cc
+++ b/instrumentation/injection-pass.cc
@@ -5,7 +5,7 @@
    Written by Marc Heuse <mh@mh-sec.de>
 
    Copyright 2015, 2016 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/qemu_mode/build_qemu_support.sh b/qemu_mode/build_qemu_support.sh
index 3f8a88f2..45019cc8 100755
--- a/qemu_mode/build_qemu_support.sh
+++ b/qemu_mode/build_qemu_support.sh
@@ -13,7 +13,7 @@
 # counters by Andrea Fioraldi <andreafioraldi@gmail.com>
 #
 # Copyright 2015, 2016, 2017 Google Inc. All rights reserved.
-# Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+# Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/qemu_mode/fastexit/Makefile b/qemu_mode/fastexit/Makefile
index c7b79277..be80207d 100644
--- a/qemu_mode/fastexit/Makefile
+++ b/qemu_mode/fastexit/Makefile
@@ -4,7 +4,7 @@
 #
 # Written by Andrea Fioraldi <andreafioraldi@gmail.com>
 #
-# Copyright 2019-2023 Andrea Fioraldi. All rights reserved.
+# Copyright 2019-2024 Andrea Fioraldi. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/qemu_mode/libcompcov/Makefile b/qemu_mode/libcompcov/Makefile
index 7260df87..4761ac02 100644
--- a/qemu_mode/libcompcov/Makefile
+++ b/qemu_mode/libcompcov/Makefile
@@ -4,7 +4,7 @@
 #
 # Written by Andrea Fioraldi <andreafioraldi@gmail.com>
 #
-# Copyright 2019-2023 Andrea Fioraldi. All rights reserved.
+# Copyright 2019-2024 Andrea Fioraldi. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/qemu_mode/libcompcov/compcovtest.cc b/qemu_mode/libcompcov/compcovtest.cc
index 23215013..11797091 100644
--- a/qemu_mode/libcompcov/compcovtest.cc
+++ b/qemu_mode/libcompcov/compcovtest.cc
@@ -2,7 +2,7 @@
 //
 // Author: Mateusz Jurczyk (mjurczyk@google.com)
 //
-// Copyright 2019-2023 Google LLC
+// Copyright 2019-2024 Google LLC
 //
 // Licensed under the Apache License, Version 2.0 (the "License");
 // you may not use this file except in compliance with the License.
diff --git a/qemu_mode/libcompcov/libcompcov.so.c b/qemu_mode/libcompcov/libcompcov.so.c
index b57e9701..36f7b2e2 100644
--- a/qemu_mode/libcompcov/libcompcov.so.c
+++ b/qemu_mode/libcompcov/libcompcov.so.c
@@ -5,7 +5,7 @@
 
    Written and maintained by Andrea Fioraldi <andreafioraldi@gmail.com>
 
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/qemu_mode/libqasan/Makefile b/qemu_mode/libqasan/Makefile
index 61782894..7366d6f6 100644
--- a/qemu_mode/libqasan/Makefile
+++ b/qemu_mode/libqasan/Makefile
@@ -4,7 +4,7 @@
 #
 # Written by Andrea Fioraldi <andreafioraldi@gmail.com>
 #
-# Copyright 2019-2023 Andrea Fioraldi. All rights reserved.
+# Copyright 2019-2024 Andrea Fioraldi. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/qemu_mode/libqasan/hooks.c b/qemu_mode/libqasan/hooks.c
index a9fd0ce9..cf1b0820 100644
--- a/qemu_mode/libqasan/hooks.c
+++ b/qemu_mode/libqasan/hooks.c
@@ -1,5 +1,5 @@
 /*******************************************************************************
-Copyright (c) 2019-2023, Andrea Fioraldi
+Copyright (c) 2019-2024, Andrea Fioraldi
 
 
 Redistribution and use in source and binary forms, with or without
diff --git a/qemu_mode/libqasan/libqasan.c b/qemu_mode/libqasan/libqasan.c
index 12be7778..45f47d5a 100644
--- a/qemu_mode/libqasan/libqasan.c
+++ b/qemu_mode/libqasan/libqasan.c
@@ -1,5 +1,5 @@
 /*******************************************************************************
-Copyright (c) 2019-2023, Andrea Fioraldi
+Copyright (c) 2019-2024, Andrea Fioraldi
 
 
 Redistribution and use in source and binary forms, with or without
diff --git a/qemu_mode/libqasan/libqasan.h b/qemu_mode/libqasan/libqasan.h
index a430c868..f0844e23 100644
--- a/qemu_mode/libqasan/libqasan.h
+++ b/qemu_mode/libqasan/libqasan.h
@@ -1,5 +1,5 @@
 /*******************************************************************************
-Copyright (c) 2019-2023, Andrea Fioraldi
+Copyright (c) 2019-2024, Andrea Fioraldi
 
 
 Redistribution and use in source and binary forms, with or without
diff --git a/qemu_mode/libqasan/malloc.c b/qemu_mode/libqasan/malloc.c
index 4448f480..ae470b56 100644
--- a/qemu_mode/libqasan/malloc.c
+++ b/qemu_mode/libqasan/malloc.c
@@ -1,5 +1,5 @@
 /*******************************************************************************
-Copyright (c) 2019-2023, Andrea Fioraldi
+Copyright (c) 2019-2024, Andrea Fioraldi
 
 
 Redistribution and use in source and binary forms, with or without
diff --git a/qemu_mode/libqasan/patch.c b/qemu_mode/libqasan/patch.c
index 38e0903b..4ce8c3d8 100644
--- a/qemu_mode/libqasan/patch.c
+++ b/qemu_mode/libqasan/patch.c
@@ -1,5 +1,5 @@
 /*******************************************************************************
-Copyright (c) 2019-2023, Andrea Fioraldi
+Copyright (c) 2019-2024, Andrea Fioraldi
 
 
 Redistribution and use in source and binary forms, with or without
diff --git a/qemu_mode/libqasan/string.c b/qemu_mode/libqasan/string.c
index e17cff4b..cd14d57b 100644
--- a/qemu_mode/libqasan/string.c
+++ b/qemu_mode/libqasan/string.c
@@ -1,5 +1,5 @@
 /*******************************************************************************
-Copyright (c) 2019-2023, Andrea Fioraldi
+Copyright (c) 2019-2024, Andrea Fioraldi
 
 
 Redistribution and use in source and binary forms, with or without
diff --git a/qemu_mode/libqasan/uninstrument.c b/qemu_mode/libqasan/uninstrument.c
index e37a9b46..996f2a74 100644
--- a/qemu_mode/libqasan/uninstrument.c
+++ b/qemu_mode/libqasan/uninstrument.c
@@ -7,7 +7,7 @@ for some strange reason.
 */
 
 /*******************************************************************************
-Copyright (c) 2019-2023, Andrea Fioraldi
+Copyright (c) 2019-2024, Andrea Fioraldi
 
 
 Redistribution and use in source and binary forms, with or without
diff --git a/qemu_mode/unsigaction/Makefile b/qemu_mode/unsigaction/Makefile
index c1a7397f..d5e807d8 100644
--- a/qemu_mode/unsigaction/Makefile
+++ b/qemu_mode/unsigaction/Makefile
@@ -4,7 +4,7 @@
 #
 # Written by Andrea Fioraldi <andreafioraldi@gmail.com>
 #
-# Copyright 2019-2023 Andrea Fioraldi. All rights reserved.
+# Copyright 2019-2024 Andrea Fioraldi. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/qemu_mode/util/qemu_get_symbol_addr.sh b/qemu_mode/util/qemu_get_symbol_addr.sh
index e0a7ae80..5e00f1b2 100755
--- a/qemu_mode/util/qemu_get_symbol_addr.sh
+++ b/qemu_mode/util/qemu_get_symbol_addr.sh
@@ -1,5 +1,5 @@
 #!/bin/bash
-# Copyright 2023 AFLplusplus
+# Copyright 2024 AFLplusplus
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/src/afl-analyze.c b/src/afl-analyze.c
index 5b122741..95f32fee 100644
--- a/src/afl-analyze.c
+++ b/src/afl-analyze.c
@@ -9,7 +9,7 @@
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/src/afl-as.c b/src/afl-as.c
index 772e31b3..09ba75bf 100644
--- a/src/afl-as.c
+++ b/src/afl-as.c
@@ -9,7 +9,7 @@
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/src/afl-cc.c b/src/afl-cc.c
index 98310545..e9564277 100644
--- a/src/afl-cc.c
+++ b/src/afl-cc.c
@@ -5,7 +5,7 @@
    Written by Michal Zalewski, Laszlo Szekeres and Marc Heuse
 
    Copyright 2015, 2016 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/src/afl-common.c b/src/afl-common.c
index ba498b3b..87003b03 100644
--- a/src/afl-common.c
+++ b/src/afl-common.c
@@ -9,7 +9,7 @@
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/src/afl-forkserver.c b/src/afl-forkserver.c
index ded0c21d..0a77d61c 100644
--- a/src/afl-forkserver.c
+++ b/src/afl-forkserver.c
@@ -13,7 +13,7 @@
 
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/src/afl-fuzz-bitmap.c b/src/afl-fuzz-bitmap.c
index 5f67347c..d056ac9f 100644
--- a/src/afl-fuzz-bitmap.c
+++ b/src/afl-fuzz-bitmap.c
@@ -9,7 +9,7 @@
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/src/afl-fuzz-cmplog.c b/src/afl-fuzz-cmplog.c
index 3e6432ca..21f34e12 100644
--- a/src/afl-fuzz-cmplog.c
+++ b/src/afl-fuzz-cmplog.c
@@ -11,7 +11,7 @@
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/src/afl-fuzz-extras.c b/src/afl-fuzz-extras.c
index 905431d1..3b1d13f1 100644
--- a/src/afl-fuzz-extras.c
+++ b/src/afl-fuzz-extras.c
@@ -9,7 +9,7 @@
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/src/afl-fuzz-init.c b/src/afl-fuzz-init.c
index 057d8cf5..76291cc4 100644
--- a/src/afl-fuzz-init.c
+++ b/src/afl-fuzz-init.c
@@ -9,7 +9,7 @@
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/src/afl-fuzz-mutators.c b/src/afl-fuzz-mutators.c
index 17fb9368..ae4d6668 100644
--- a/src/afl-fuzz-mutators.c
+++ b/src/afl-fuzz-mutators.c
@@ -10,7 +10,7 @@
                         Dominik Maier <mail@dmnk.co>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/src/afl-fuzz-one.c b/src/afl-fuzz-one.c
index c163a420..d9c074ec 100644
--- a/src/afl-fuzz-one.c
+++ b/src/afl-fuzz-one.c
@@ -9,7 +9,7 @@
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/src/afl-fuzz-python.c b/src/afl-fuzz-python.c
index 4c7da774..16a398fd 100644
--- a/src/afl-fuzz-python.c
+++ b/src/afl-fuzz-python.c
@@ -9,7 +9,7 @@
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/src/afl-fuzz-queue.c b/src/afl-fuzz-queue.c
index 67931bba..1ea50418 100644
--- a/src/afl-fuzz-queue.c
+++ b/src/afl-fuzz-queue.c
@@ -9,7 +9,7 @@
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
    You may obtain a copy of the License at:
diff --git a/src/afl-fuzz-redqueen.c b/src/afl-fuzz-redqueen.c
index 9e9b3822..eead7a8b 100644
--- a/src/afl-fuzz-redqueen.c
+++ b/src/afl-fuzz-redqueen.c
@@ -11,7 +11,7 @@
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/src/afl-fuzz-run.c b/src/afl-fuzz-run.c
index 1ee8ebe7..d764952c 100644
--- a/src/afl-fuzz-run.c
+++ b/src/afl-fuzz-run.c
@@ -10,7 +10,7 @@
                         Dominik Maier <mail@dmnk.co>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/src/afl-fuzz-state.c b/src/afl-fuzz-state.c
index b647ac84..4467cae8 100644
--- a/src/afl-fuzz-state.c
+++ b/src/afl-fuzz-state.c
@@ -9,7 +9,7 @@
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/src/afl-fuzz-stats.c b/src/afl-fuzz-stats.c
index 4b83ad29..76577081 100644
--- a/src/afl-fuzz-stats.c
+++ b/src/afl-fuzz-stats.c
@@ -9,7 +9,7 @@
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index 69064d51..12d67fe7 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -9,7 +9,7 @@
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/src/afl-gotcpu.c b/src/afl-gotcpu.c
index 4f851099..7aee2985 100644
--- a/src/afl-gotcpu.c
+++ b/src/afl-gotcpu.c
@@ -9,7 +9,7 @@
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/src/afl-ld-lto.c b/src/afl-ld-lto.c
index 7ce5de41..513c1ae9 100644
--- a/src/afl-ld-lto.c
+++ b/src/afl-ld-lto.c
@@ -9,7 +9,7 @@
                 Andrea Fioraldi <andreafioraldi@gmail.com>
                 Dominik Maier <domenukk@gmail.com>
 
-  Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+  Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
   Licensed under the Apache License, Version 2.0 (the "License");
   you may not use this file except in compliance with the License.
diff --git a/src/afl-sharedmem.c b/src/afl-sharedmem.c
index a2c81586..daea8f46 100644
--- a/src/afl-sharedmem.c
+++ b/src/afl-sharedmem.c
@@ -11,7 +11,7 @@
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/src/afl-showmap.c b/src/afl-showmap.c
index 7a639cf6..20ba5a5e 100644
--- a/src/afl-showmap.c
+++ b/src/afl-showmap.c
@@ -12,7 +12,7 @@
                         Dominik Maier <mail@dmnk.co>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/src/afl-tmin.c b/src/afl-tmin.c
index e7442d1d..4e5dab41 100644
--- a/src/afl-tmin.c
+++ b/src/afl-tmin.c
@@ -12,7 +12,7 @@
                         Dominik Maier <mail@dmnk.co>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/test-instr.c b/test-instr.c
index eda5189c..28552893 100644
--- a/test-instr.c
+++ b/test-instr.c
@@ -3,7 +3,7 @@
    --------------------------------------------------------
    Originally written by Michal Zalewski
    Copyright 2014 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
    You may obtain a copy of the License at:
diff --git a/unicorn_mode/build_unicorn_support.sh b/unicorn_mode/build_unicorn_support.sh
index d3d16ad5..baca2171 100755
--- a/unicorn_mode/build_unicorn_support.sh
+++ b/unicorn_mode/build_unicorn_support.sh
@@ -14,7 +14,7 @@
 #                                <andreafioraldi@gmail.com>
 #
 # Copyright 2017 Battelle Memorial Institute. All rights reserved.
-# Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+# Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/utils/afl_network_proxy/afl-network-client.c b/utils/afl_network_proxy/afl-network-client.c
index 0416f0f9..1f04dd87 100644
--- a/utils/afl_network_proxy/afl-network-client.c
+++ b/utils/afl_network_proxy/afl-network-client.c
@@ -4,7 +4,7 @@
 
    Written by Marc Heuse <mh@mh-sec.de>
 
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/utils/afl_network_proxy/afl-network-server.c b/utils/afl_network_proxy/afl-network-server.c
index 95b0a551..c4a700f4 100644
--- a/utils/afl_network_proxy/afl-network-server.c
+++ b/utils/afl_network_proxy/afl-network-server.c
@@ -12,7 +12,7 @@
                         Dominik Maier <mail@dmnk.co>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/utils/afl_proxy/afl-proxy.c b/utils/afl_proxy/afl-proxy.c
index 531a97a2..6cf47636 100644
--- a/utils/afl_proxy/afl-proxy.c
+++ b/utils/afl_proxy/afl-proxy.c
@@ -4,7 +4,7 @@
 
    Written by Marc Heuse <mh@mh-sec.de>
 
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/utils/afl_untracer/afl-untracer.c b/utils/afl_untracer/afl-untracer.c
index 0e3f8a45..e6a74518 100644
--- a/utils/afl_untracer/afl-untracer.c
+++ b/utils/afl_untracer/afl-untracer.c
@@ -4,7 +4,7 @@
 
    Written by Marc Heuse <mh@mh-sec.de>
 
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/utils/afl_untracer/libtestinstr.c b/utils/afl_untracer/libtestinstr.c
index b7afc325..0a98778a 100644
--- a/utils/afl_untracer/libtestinstr.c
+++ b/utils/afl_untracer/libtestinstr.c
@@ -3,7 +3,7 @@
    --------------------------------------------------------
    Originally written by Michal Zalewski
    Copyright 2014 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
    You may obtain a copy of the License at:
diff --git a/utils/argv_fuzzing/Makefile b/utils/argv_fuzzing/Makefile
index 6786467a..ba977e5f 100644
--- a/utils/argv_fuzzing/Makefile
+++ b/utils/argv_fuzzing/Makefile
@@ -2,7 +2,7 @@
 # american fuzzy lop++ - argvfuzz
 # --------------------------------
 #
-# Copyright 2019-2023 Kjell Braden <afflux@pentabarf.de>
+# Copyright 2019-2024 Kjell Braden <afflux@pentabarf.de>
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/utils/argv_fuzzing/argvfuzz.c b/utils/argv_fuzzing/argvfuzz.c
index 41eead0c..47383138 100644
--- a/utils/argv_fuzzing/argvfuzz.c
+++ b/utils/argv_fuzzing/argvfuzz.c
@@ -2,7 +2,7 @@
    american fuzzy lop++ - LD_PRELOAD for fuzzing argv in binaries
    ------------------------------------------------------------
 
-   Copyright 2019-2023 Kjell Braden <afflux@pentabarf.de>
+   Copyright 2019-2024 Kjell Braden <afflux@pentabarf.de>
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/utils/distributed_fuzzing/sync_script.sh b/utils/distributed_fuzzing/sync_script.sh
index b22816f1..861b65c8 100755
--- a/utils/distributed_fuzzing/sync_script.sh
+++ b/utils/distributed_fuzzing/sync_script.sh
@@ -6,7 +6,7 @@
 # Originally written by Michal Zalewski
 #
 # Copyright 2014 Google Inc. All rights reserved.
-# Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+# Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/utils/libdislocator/libdislocator.so.c b/utils/libdislocator/libdislocator.so.c
index 1cd7abc6..b80be1a1 100644
--- a/utils/libdislocator/libdislocator.so.c
+++ b/utils/libdislocator/libdislocator.so.c
@@ -6,7 +6,7 @@
    Originally written by Michal Zalewski
 
    Copyright 2016 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/utils/libtokencap/libtokencap.so.c b/utils/libtokencap/libtokencap.so.c
index f4024799..cc499150 100644
--- a/utils/libtokencap/libtokencap.so.c
+++ b/utils/libtokencap/libtokencap.so.c
@@ -6,7 +6,7 @@
    Originally written by Michal Zalewski
 
    Copyright 2016 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
diff --git a/utils/persistent_mode/test-instr.c b/utils/persistent_mode/test-instr.c
index 4ead6577..72e26e93 100644
--- a/utils/persistent_mode/test-instr.c
+++ b/utils/persistent_mode/test-instr.c
@@ -3,7 +3,7 @@
    --------------------------------------------------------
    Originally written by Michal Zalewski
    Copyright 2014 Google Inc. All rights reserved.
-   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
    You may obtain a copy of the License at:
-- 
cgit 1.4.1


From 9fab7e892d4e2ba09305aac40392a4df598464c9 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Sat, 3 Feb 2024 16:48:38 +0100
Subject: new forkserver - server part

---
 include/types.h      |   7 +
 src/afl-forkserver.c | 368 +++++++++++++++++++++++++++++++++++++--------------
 2 files changed, 274 insertions(+), 101 deletions(-)

(limited to 'include')

diff --git a/include/types.h b/include/types.h
index 22332135..d0a2d124 100644
--- a/include/types.h
+++ b/include/types.h
@@ -49,6 +49,13 @@ typedef uint128_t         u128;
 #define FS_ERROR_OLD_CMPLOG 32
 #define FS_ERROR_OLD_CMPLOG_QEMU 64
 
+/* New Forkserver */
+#define FS_NEW_VERSION_MIN 1
+#define FS_NEW_VERSION_MAX 1
+#define FS_NEW_OPT_MAPSIZE     0x00000001  // parameter: 32 bit value
+#define FS_NEW_OPT_SHDMEM_FUZZ 0x00000002  // paramter: none
+#define FS_NEW_OPT_AUTODICT    0x00000800  // autodictionary data
+
 /* Reporting options */
 #define FS_OPT_ENABLED 0x80000001
 #define FS_OPT_MAPSIZE 0x40000000
diff --git a/src/afl-forkserver.c b/src/afl-forkserver.c
index 0a77d61c..1f796e53 100644
--- a/src/afl-forkserver.c
+++ b/src/afl-forkserver.c
@@ -389,7 +389,7 @@ static void afl_fauxsrv_execv(afl_forkserver_t *fsrv, char **argv) {
   while (1) {
 
     uint32_t was_killed;
-    int      status;
+    u32      status;
 
     /* Wait for parent by reading from the pipe. Exit if read fails. */
 
@@ -524,7 +524,7 @@ void afl_fsrv_start(afl_forkserver_t *fsrv, char **argv,
                     volatile u8 *stop_soon_p, u8 debug_child_output) {
 
   int   st_pipe[2], ctl_pipe[2];
-  s32   status;
+  u32   status;
   s32   rlen;
   char *ignore_autodict = getenv("AFL_NO_AUTODICT");
 
@@ -1017,69 +1017,95 @@ void afl_fsrv_start(afl_forkserver_t *fsrv, char **argv,
 
   if (rlen == 4) {
 
+    /*
+     *  The new fork server model works like this:
+     *    Client: sends "AFLx" in little endian, with x being the forkserver
+     *            protocol version.
+     *    Server: replies with XOR of the message or exits with an error if it
+     *            is not a supported version.
+     *    Client: sends 32 bit of options and then sends all parameters of
+     *            the options, one after another, increasing by option number.
+     *            Ends with "AFLx".
+     *  After the initial protocol version confirmation the server does not
+     *  send any data anymore - except a future option requires this.
+     */
+
     if (status >= 0x41464c00 && status <= 0x41464cff) {
 
-      FATAL(
-          "Target uses the new forkserver model, you need to switch to a newer "
-          "afl-fuzz too!");
+      u32 version = status - 0x41464c00;
 
-    }
+      if (!version) {
 
-    if (!be_quiet) { OKF("All right - fork server is up."); }
+        FATAL(
+            "Fork server version is not assigned, this should not happen. "
+            "Recompile target.");
 
-    if (getenv("AFL_DEBUG")) {
+      } else if (version < FS_NEW_VERSION_MIN || version > FS_NEW_VERSION_MAX) {
 
-      ACTF("Extended forkserver functions received (%08x).", status);
+        FATAL(
+            "Fork server version is not not supported.  Recompile the target.");
 
-    }
+      }
 
-    if ((status & FS_OPT_ERROR) == FS_OPT_ERROR)
-      report_error_and_exit(FS_OPT_GET_ERROR(status));
+      status ^= 0xffffffff;
+      if (write(fsrv->fsrv_ctl_fd, &status, 4) != 4) {
 
-    if ((status & FS_OPT_ENABLED) == FS_OPT_ENABLED) {
+        FATAL("Writing to forkserver failed.");
 
-      // workaround for recent AFL++ versions
-      if ((status & FS_OPT_OLD_AFLPP_WORKAROUND) == FS_OPT_OLD_AFLPP_WORKAROUND)
-        status = (status & 0xf0ffffff);
+      }
 
-      if ((status & FS_OPT_NEWCMPLOG) == 0 && fsrv->cmplog_binary) {
+      if (!be_quiet) {
 
-        if (fsrv->qemu_mode || fsrv->frida_mode) {
+        OKF("All right - new fork server model v%u is up.", version);
 
-          report_error_and_exit(FS_ERROR_OLD_CMPLOG_QEMU);
+      }
 
-        } else {
+      rlen = read(fsrv->fsrv_st_fd, &status, 4);
 
-          report_error_and_exit(FS_ERROR_OLD_CMPLOG);
+      if (getenv("AFL_DEBUG")) {
 
-        }
+        ACTF("Forkserver options received: (%08x)", status);
 
       }
 
-      if ((status & FS_OPT_SNAPSHOT) == FS_OPT_SNAPSHOT) {
+      if ((status & FS_NEW_OPT_MAPSIZE)) {
 
-        fsrv->snapshot = 1;
-        if (!be_quiet) { ACTF("Using SNAPSHOT feature."); }
+        u32 tmp_map_size;
+        rlen = read(fsrv->fsrv_st_fd, &tmp_map_size, 4);
 
-      }
+        if (!fsrv->map_size) { fsrv->map_size = MAP_SIZE; }
 
-      if ((status & FS_OPT_SHDMEM_FUZZ) == FS_OPT_SHDMEM_FUZZ) {
+        fsrv->real_map_size = tmp_map_size;
 
-        if (fsrv->support_shmem_fuzz) {
+        if (tmp_map_size % 64) {
 
-          fsrv->use_shmem_fuzz = 1;
-          if (!be_quiet) { ACTF("Using SHARED MEMORY FUZZING feature."); }
+          tmp_map_size = (((tmp_map_size + 63) >> 6) << 6);
+
+        }
 
-          if ((status & FS_OPT_AUTODICT) == 0 || ignore_autodict) {
+        if (!be_quiet) { ACTF("Target map size: %u", fsrv->real_map_size); }
+        if (tmp_map_size > fsrv->map_size) {
 
-            u32 send_status = (FS_OPT_ENABLED | FS_OPT_SHDMEM_FUZZ);
-            if (write(fsrv->fsrv_ctl_fd, &send_status, 4) != 4) {
+          FATAL(
+              "Target's coverage map size of %u is larger than the one this "
+              "AFL++ is set with (%u). Either set AFL_MAP_SIZE=%u and "
+              "restart "
+              " afl-fuzz, or change MAP_SIZE_POW2 in config.h and recompile "
+              "afl-fuzz",
+              tmp_map_size, fsrv->map_size, tmp_map_size);
 
-              FATAL("Writing to forkserver failed.");
+        }
 
-            }
+        fsrv->map_size = tmp_map_size;
 
-          }
+      }
+
+      if ((status & FS_NEW_OPT_SHDMEM_FUZZ)) {
+
+        if (fsrv->support_shmem_fuzz) {
+
+          fsrv->use_shmem_fuzz = 1;
+          if (!be_quiet) { ACTF("Using SHARED MEMORY FUZZING feature."); }
 
         } else {
 
@@ -1091,134 +1117,274 @@ void afl_fsrv_start(afl_forkserver_t *fsrv, char **argv,
 
       }
 
-      if ((status & FS_OPT_MAPSIZE) == FS_OPT_MAPSIZE) {
+      if ((status & FS_NEW_OPT_AUTODICT)) {
 
-        u32 tmp_map_size = FS_OPT_GET_MAPSIZE(status);
+        u32 dict_size;
+        if (read(fsrv->fsrv_st_fd, &dict_size, 4) != 4) {
 
-        if (!fsrv->map_size) { fsrv->map_size = MAP_SIZE; }
+          FATAL("Reading from forkserver failed.");
 
-        fsrv->real_map_size = tmp_map_size;
+        }
 
-        if (tmp_map_size % 64) {
+        if (dict_size < 2 || dict_size > 0xffffff) {
 
-          tmp_map_size = (((tmp_map_size + 63) >> 6) << 6);
+          FATAL("Dictionary has an illegal size: %d", dict_size);
 
         }
 
-        if (!be_quiet) { ACTF("Target map size: %u", fsrv->real_map_size); }
-        if (tmp_map_size > fsrv->map_size) {
+        u32 offset = 0, count = 0;
+        u8 *dict = ck_alloc(dict_size);
+        if (dict == NULL) {
 
-          FATAL(
-              "Target's coverage map size of %u is larger than the one this "
-              "AFL++ is set with (%u). Either set AFL_MAP_SIZE=%u and restart "
-              " afl-fuzz, or change MAP_SIZE_POW2 in config.h and recompile "
-              "afl-fuzz",
-              tmp_map_size, fsrv->map_size, tmp_map_size);
+          FATAL("Could not allocate %u bytes of autodictionary memory",
+                dict_size);
 
         }
 
-        fsrv->map_size = tmp_map_size;
+        while (dict_size != 0) {
+
+          rlen = read(fsrv->fsrv_st_fd, dict + offset, dict_size);
+          if (rlen > 0) {
+
+            dict_size -= rlen;
+            offset += rlen;
+
+          } else {
+
+            FATAL(
+                "Reading autodictionary fail at position %u with %u bytes "
+                "left.",
+                offset, dict_size);
+
+          }
+
+        }
+
+        offset = 0;
+        while (offset < dict_size && (u8)dict[offset] + offset < dict_size) {
+
+          fsrv->add_extra_func(fsrv->afl_ptr, dict + offset + 1,
+                               (u8)dict[offset]);
+          offset += (1 + dict[offset]);
+          count++;
+
+        }
+
+        if (!be_quiet) { ACTF("Loaded %u autodictionary entries", count); }
+        ck_free(dict);
 
       }
 
-      if ((status & FS_OPT_AUTODICT) == FS_OPT_AUTODICT) {
+      u32 status2;
+      rlen = read(fsrv->fsrv_st_fd, &status2, 4);
 
-        if (!ignore_autodict) {
+      if (status2 != status) { FATAL("Error in forkserver communication"); }
 
-          if (fsrv->add_extra_func == NULL || fsrv->afl_ptr == NULL) {
+    } else {
 
-            // this is not afl-fuzz - or it is cmplog - we deny and return
-            if (fsrv->use_shmem_fuzz) {
+      WARNF(
+          "Old fork server model is used by the target, this still works "
+          "though.");
 
-              status = (FS_OPT_ENABLED | FS_OPT_SHDMEM_FUZZ);
+      if (!be_quiet) { OKF("All right - old fork server is up."); }
 
-            } else {
+      if (getenv("AFL_DEBUG")) {
 
-              status = (FS_OPT_ENABLED);
+        ACTF("Extended forkserver functions received (%08x).", status);
 
-            }
+      }
 
-            if (write(fsrv->fsrv_ctl_fd, &status, 4) != 4) {
+      if ((status & FS_OPT_ERROR) == FS_OPT_ERROR)
+        report_error_and_exit(FS_OPT_GET_ERROR(status));
 
-              FATAL("Writing to forkserver failed.");
+      if ((status & FS_OPT_ENABLED) == FS_OPT_ENABLED) {
 
-            }
+        // workaround for recent AFL++ versions
+        if ((status & FS_OPT_OLD_AFLPP_WORKAROUND) ==
+            FS_OPT_OLD_AFLPP_WORKAROUND)
+          status = (status & 0xf0ffffff);
+
+        if ((status & FS_OPT_NEWCMPLOG) == 0 && fsrv->cmplog_binary) {
+
+          if (fsrv->qemu_mode || fsrv->frida_mode) {
+
+            report_error_and_exit(FS_ERROR_OLD_CMPLOG_QEMU);
+
+          } else {
 
-            return;
+            report_error_and_exit(FS_ERROR_OLD_CMPLOG);
 
           }
 
-          if (!be_quiet) { ACTF("Using AUTODICT feature."); }
+        }
+
+        if ((status & FS_OPT_SNAPSHOT) == FS_OPT_SNAPSHOT) {
+
+          fsrv->snapshot = 1;
+          if (!be_quiet) { ACTF("Using SNAPSHOT feature."); }
+
+        }
+
+        if ((status & FS_OPT_SHDMEM_FUZZ) == FS_OPT_SHDMEM_FUZZ) {
+
+          if (fsrv->support_shmem_fuzz) {
+
+            fsrv->use_shmem_fuzz = 1;
+            if (!be_quiet) { ACTF("Using SHARED MEMORY FUZZING feature."); }
+
+            if ((status & FS_OPT_AUTODICT) == 0 || ignore_autodict) {
 
-          if (fsrv->use_shmem_fuzz) {
+              u32 send_status = (FS_OPT_ENABLED | FS_OPT_SHDMEM_FUZZ);
+              if (write(fsrv->fsrv_ctl_fd, &send_status, 4) != 4) {
 
-            status = (FS_OPT_ENABLED | FS_OPT_AUTODICT | FS_OPT_SHDMEM_FUZZ);
+                FATAL("Writing to forkserver failed.");
+
+              }
+
+            }
 
           } else {
 
-            status = (FS_OPT_ENABLED | FS_OPT_AUTODICT);
+            FATAL(
+                "Target requested sharedmem fuzzing, but we failed to enable "
+                "it.");
 
           }
 
-          if (write(fsrv->fsrv_ctl_fd, &status, 4) != 4) {
+        }
 
-            FATAL("Writing to forkserver failed.");
+        if ((status & FS_OPT_MAPSIZE) == FS_OPT_MAPSIZE) {
 
-          }
+          u32 tmp_map_size = FS_OPT_GET_MAPSIZE(status);
 
-          if (read(fsrv->fsrv_st_fd, &status, 4) != 4) {
+          if (!fsrv->map_size) { fsrv->map_size = MAP_SIZE; }
 
-            FATAL("Reading from forkserver failed.");
+          fsrv->real_map_size = tmp_map_size;
+
+          if (tmp_map_size % 64) {
+
+            tmp_map_size = (((tmp_map_size + 63) >> 6) << 6);
 
           }
 
-          if (status < 2 || (u32)status > 0xffffff) {
+          if (!be_quiet) { ACTF("Target map size: %u", fsrv->real_map_size); }
+          if (tmp_map_size > fsrv->map_size) {
 
-            FATAL("Dictionary has an illegal size: %d", status);
+            FATAL(
+                "Target's coverage map size of %u is larger than the one this "
+                "AFL++ is set with (%u). Either set AFL_MAP_SIZE=%u and "
+                "restart "
+                " afl-fuzz, or change MAP_SIZE_POW2 in config.h and recompile "
+                "afl-fuzz",
+                tmp_map_size, fsrv->map_size, tmp_map_size);
 
           }
 
-          u32 offset = 0, count = 0;
-          u32 len = status;
-          u8 *dict = ck_alloc(len);
-          if (dict == NULL) {
+          fsrv->map_size = tmp_map_size;
+
+        }
 
-            FATAL("Could not allocate %u bytes of autodictionary memory", len);
+        if ((status & FS_OPT_AUTODICT) == FS_OPT_AUTODICT) {
 
-          }
+          if (!ignore_autodict) {
 
-          while (len != 0) {
+            if (fsrv->add_extra_func == NULL || fsrv->afl_ptr == NULL) {
 
-            rlen = read(fsrv->fsrv_st_fd, dict + offset, len);
-            if (rlen > 0) {
+              // this is not afl-fuzz - or it is cmplog - we deny and return
+              if (fsrv->use_shmem_fuzz) {
 
-              len -= rlen;
-              offset += rlen;
+                status = (FS_OPT_ENABLED | FS_OPT_SHDMEM_FUZZ);
+
+              } else {
+
+                status = (FS_OPT_ENABLED);
+
+              }
+
+              if (write(fsrv->fsrv_ctl_fd, &status, 4) != 4) {
+
+                FATAL("Writing to forkserver failed.");
+
+              }
+
+              return;
+
+            }
+
+            if (!be_quiet) { ACTF("Using AUTODICT feature."); }
+
+            if (fsrv->use_shmem_fuzz) {
+
+              status = (FS_OPT_ENABLED | FS_OPT_AUTODICT | FS_OPT_SHDMEM_FUZZ);
 
             } else {
 
-              FATAL(
-                  "Reading autodictionary fail at position %u with %u bytes "
-                  "left.",
-                  offset, len);
+              status = (FS_OPT_ENABLED | FS_OPT_AUTODICT);
 
             }
 
-          }
+            if (write(fsrv->fsrv_ctl_fd, &status, 4) != 4) {
 
-          offset = 0;
-          while (offset < (u32)status &&
-                 (u8)dict[offset] + offset < (u32)status) {
+              FATAL("Writing to forkserver failed.");
 
-            fsrv->add_extra_func(fsrv->afl_ptr, dict + offset + 1,
-                                 (u8)dict[offset]);
-            offset += (1 + dict[offset]);
-            count++;
+            }
 
-          }
+            if (read(fsrv->fsrv_st_fd, &status, 4) != 4) {
 
-          if (!be_quiet) { ACTF("Loaded %u autodictionary entries", count); }
-          ck_free(dict);
+              FATAL("Reading from forkserver failed.");
+
+            }
+
+            if (status < 2 || (u32)status > 0xffffff) {
+
+              FATAL("Dictionary has an illegal size: %d", status);
+
+            }
+
+            u32 offset = 0, count = 0;
+            u32 len = status;
+            u8 *dict = ck_alloc(len);
+            if (dict == NULL) {
+
+              FATAL("Could not allocate %u bytes of autodictionary memory",
+                    len);
+
+            }
+
+            while (len != 0) {
+
+              rlen = read(fsrv->fsrv_st_fd, dict + offset, len);
+              if (rlen > 0) {
+
+                len -= rlen;
+                offset += rlen;
+
+              } else {
+
+                FATAL(
+                    "Reading autodictionary fail at position %u with %u bytes "
+                    "left.",
+                    offset, len);
+
+              }
+
+            }
+
+            offset = 0;
+            while (offset < (u32)status &&
+                   (u8)dict[offset] + offset < (u32)status) {
+
+              fsrv->add_extra_func(fsrv->afl_ptr, dict + offset + 1,
+                                   (u8)dict[offset]);
+              offset += (1 + dict[offset]);
+              count++;
+
+            }
+
+            if (!be_quiet) { ACTF("Loaded %u autodictionary entries", count); }
+            ck_free(dict);
+
+          }
 
         }
 
-- 
cgit 1.4.1


From 27338fcef121c7700a1e2e99cb31cb7106159293 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Sat, 3 Feb 2024 18:27:01 +0100
Subject: new forkserver - client side

---
 include/types.h                     |   7 +-
 instrumentation/afl-compiler-rt.o.c | 414 ++++++------------------------------
 src/afl-forkserver.c                |  15 +-
 3 files changed, 80 insertions(+), 356 deletions(-)

(limited to 'include')

diff --git a/include/types.h b/include/types.h
index d0a2d124..18c5df91 100644
--- a/include/types.h
+++ b/include/types.h
@@ -52,9 +52,10 @@ typedef uint128_t         u128;
 /* New Forkserver */
 #define FS_NEW_VERSION_MIN 1
 #define FS_NEW_VERSION_MAX 1
-#define FS_NEW_OPT_MAPSIZE     0x00000001  // parameter: 32 bit value
-#define FS_NEW_OPT_SHDMEM_FUZZ 0x00000002  // paramter: none
-#define FS_NEW_OPT_AUTODICT    0x00000800  // autodictionary data
+#define FS_NEW_ERROR 0xeffe0000
+#define FS_NEW_OPT_MAPSIZE 0x00000001      // parameter: 32 bit value
+#define FS_NEW_OPT_SHDMEM_FUZZ 0x00000002  // parameter: none
+#define FS_NEW_OPT_AUTODICT 0x00000800     // autodictionary data
 
 /* Reporting options */
 #define FS_OPT_ENABLED 0x80000001
diff --git a/instrumentation/afl-compiler-rt.o.c b/instrumentation/afl-compiler-rt.o.c
index caa3c3a8..c342334c 100644
--- a/instrumentation/afl-compiler-rt.o.c
+++ b/instrumentation/afl-compiler-rt.o.c
@@ -264,7 +264,7 @@ static void send_forkserver_error(int error) {
 
   u32 status;
   if (!error || error > 0xffff) return;
-  status = (FS_OPT_ERROR | FS_OPT_SET_ERROR(error));
+  status = (FS_NEW_ERROR | error);
   if (write(FORKSRV_FD + 1, (char *)&status, 4) != 4) { return; }
 
 }
@@ -367,32 +367,13 @@ static void __afl_map_shm(void) {
       if ((ptr = getenv("AFL_MAP_SIZE")) != NULL) { val = atoi(ptr); }
       if (val < __afl_final_loc) {
 
-        if (__afl_final_loc > FS_OPT_MAX_MAPSIZE) {
+        if (__afl_final_loc > MAP_INITIAL_SIZE && !getenv("AFL_QUIET")) {
 
-          if (!getenv("AFL_QUIET"))
-            fprintf(stderr,
-                    "Error: AFL++ tools *require* to set AFL_MAP_SIZE to %u "
-                    "to be able to run this instrumented program!\n",
-                    __afl_final_loc);
-
-          if (id_str) {
-
-            send_forkserver_error(FS_ERROR_MAP_SIZE);
-            exit(-1);
-
-          }
-
-        } else {
-
-          if (__afl_final_loc > MAP_INITIAL_SIZE && !getenv("AFL_QUIET")) {
-
-            fprintf(stderr,
-                    "Warning: AFL++ tools might need to set AFL_MAP_SIZE to %u "
-                    "to be able to run this instrumented program if this "
-                    "crashes!\n",
-                    __afl_final_loc);
-
-          }
+          fprintf(stderr,
+                  "Warning: AFL++ tools might need to set AFL_MAP_SIZE to %u "
+                  "to be able to run this instrumented program if this "
+                  "crashes!\n",
+                  __afl_final_loc);
 
         }
 
@@ -400,15 +381,6 @@ static void __afl_map_shm(void) {
 
     }
 
-  } else {
-
-    if (getenv("AFL_DUMP_MAP_SIZE")) {
-
-      printf("%u\n", MAP_SIZE);
-      exit(-1);
-
-    }
-
   }
 
   if (__afl_sharedmem_fuzzing && (!id_str || !getenv(SHM_FUZZ_ENV_VAR) ||
@@ -474,14 +446,13 @@ static void __afl_map_shm(void) {
 
   if (__afl_debug) {
 
-    fprintf(
-        stderr,
-        "DEBUG: (1) id_str %s, __afl_area_ptr %p, __afl_area_initial %p, "
-        "__afl_area_ptr_dummy %p, __afl_map_addr 0x%llx, MAP_SIZE %u, "
-        "__afl_final_loc %u, __afl_map_size %u, max_size_forkserver %u/0x%x\n",
-        id_str == NULL ? "<null>" : id_str, __afl_area_ptr, __afl_area_initial,
-        __afl_area_ptr_dummy, __afl_map_addr, MAP_SIZE, __afl_final_loc,
-        __afl_map_size, FS_OPT_MAX_MAPSIZE, FS_OPT_MAX_MAPSIZE);
+    fprintf(stderr,
+            "DEBUG: (1) id_str %s, __afl_area_ptr %p, __afl_area_initial %p, "
+            "__afl_area_ptr_dummy %p, __afl_map_addr 0x%llx, MAP_SIZE %u, "
+            "__afl_final_loc %u, __afl_map_size %u\n",
+            id_str == NULL ? "<null>" : id_str, __afl_area_ptr,
+            __afl_area_initial, __afl_area_ptr_dummy, __afl_map_addr, MAP_SIZE,
+            __afl_final_loc, __afl_map_size);
 
   }
 
@@ -639,12 +610,10 @@ static void __afl_map_shm(void) {
     fprintf(stderr,
             "DEBUG: (2) id_str %s, __afl_area_ptr %p, __afl_area_initial %p, "
             "__afl_area_ptr_dummy %p, __afl_map_addr 0x%llx, MAP_SIZE "
-            "%u, __afl_final_loc %u, __afl_map_size %u, "
-            "max_size_forkserver %u/0x%x\n",
+            "%u, __afl_final_loc %u, __afl_map_size %u",
             id_str == NULL ? "<null>" : id_str, __afl_area_ptr,
             __afl_area_initial, __afl_area_ptr_dummy, __afl_map_addr, MAP_SIZE,
-            __afl_final_loc, __afl_map_size, FS_OPT_MAX_MAPSIZE,
-            FS_OPT_MAX_MAPSIZE);
+            __afl_final_loc, __afl_map_size);
 
   }
 
@@ -855,242 +824,6 @@ void write_error_with_location(char *text, char *filename, int linenumber) {
 
 }
 
-#ifdef __linux__
-static void __afl_start_snapshots(void) {
-
-  static u8 tmp[4] = {0, 0, 0, 0};
-  u32       status = 0;
-  u32       already_read_first = 0;
-  u32       was_killed;
-
-  u8 child_stopped = 0;
-
-  void (*old_sigchld_handler)(int) = signal(SIGCHLD, SIG_DFL);
-
-  /* Phone home and tell the parent that we're OK. If parent isn't there,
-     assume we're not running in forkserver mode and just execute program. */
-
-  status |= (FS_OPT_ENABLED | FS_OPT_SNAPSHOT | FS_OPT_NEWCMPLOG);
-  if (__afl_sharedmem_fuzzing) { status |= FS_OPT_SHDMEM_FUZZ; }
-  if (__afl_map_size <= FS_OPT_MAX_MAPSIZE)
-    status |= (FS_OPT_SET_MAPSIZE(__afl_map_size) | FS_OPT_MAPSIZE);
-  if (__afl_dictionary_len && __afl_dictionary) { status |= FS_OPT_AUTODICT; }
-  memcpy(tmp, &status, 4);
-
-  if (write(FORKSRV_FD + 1, tmp, 4) != 4) { return; }
-
-  if (__afl_sharedmem_fuzzing || (__afl_dictionary_len && __afl_dictionary)) {
-
-    if (read(FORKSRV_FD, &was_killed, 4) != 4) {
-
-      write_error("read to afl-fuzz");
-      _exit(1);
-
-    }
-
-    if (__afl_debug) {
-
-      fprintf(stderr, "DEBUG: target forkserver recv: %08x\n", was_killed);
-
-    }
-
-    if ((was_killed & (FS_OPT_ENABLED | FS_OPT_SHDMEM_FUZZ)) ==
-        (FS_OPT_ENABLED | FS_OPT_SHDMEM_FUZZ)) {
-
-      __afl_map_shm_fuzz();
-
-    }
-
-    if ((was_killed & (FS_OPT_ENABLED | FS_OPT_AUTODICT)) ==
-            (FS_OPT_ENABLED | FS_OPT_AUTODICT) &&
-        __afl_dictionary_len && __afl_dictionary) {
-
-      // great lets pass the dictionary through the forkserver FD
-      u32 len = __afl_dictionary_len, offset = 0;
-      s32 ret;
-
-      if (write(FORKSRV_FD + 1, &len, 4) != 4) {
-
-        write(2, "Error: could not send dictionary len\n",
-              strlen("Error: could not send dictionary len\n"));
-        _exit(1);
-
-      }
-
-      while (len != 0) {
-
-        ret = write(FORKSRV_FD + 1, __afl_dictionary + offset, len);
-
-        if (ret < 1) {
-
-          write(2, "Error: could not send dictionary\n",
-                strlen("Error: could not send dictionary\n"));
-          _exit(1);
-
-        }
-
-        len -= ret;
-        offset += ret;
-
-      }
-
-    } else {
-
-      // uh this forkserver does not understand extended option passing
-      // or does not want the dictionary
-      if (!__afl_fuzz_ptr) already_read_first = 1;
-
-    }
-
-  }
-
-  while (1) {
-
-    int status;
-
-    if (already_read_first) {
-
-      already_read_first = 0;
-
-    } else {
-
-      /* Wait for parent by reading from the pipe. Abort if read fails. */
-      if (read(FORKSRV_FD, &was_killed, 4) != 4) {
-
-        write_error("reading from afl-fuzz");
-        _exit(1);
-
-      }
-
-    }
-
-  #ifdef _AFL_DOCUMENT_MUTATIONS
-    if (__afl_fuzz_ptr) {
-
-      static uint32_t counter = 0;
-      char            fn[32];
-      sprintf(fn, "%09u:forkserver", counter);
-      s32 fd_doc = open(fn, O_WRONLY | O_CREAT | O_TRUNC, DEFAULT_PERMISSION);
-      if (fd_doc >= 0) {
-
-        if (write(fd_doc, __afl_fuzz_ptr, *__afl_fuzz_len) != *__afl_fuzz_len) {
-
-          fprintf(stderr, "write of mutation file failed: %s\n", fn);
-          unlink(fn);
-
-        }
-
-        close(fd_doc);
-
-      }
-
-      counter++;
-
-    }
-
-  #endif
-
-    /* If we stopped the child in persistent mode, but there was a race
-       condition and afl-fuzz already issued SIGKILL, write off the old
-       process. */
-
-    if (child_stopped && was_killed) {
-
-      child_stopped = 0;
-      if (waitpid(child_pid, &status, 0) < 0) {
-
-        write_error("child_stopped && was_killed");
-        _exit(1);  // TODO why exit?
-
-      }
-
-    }
-
-    if (!child_stopped) {
-
-      /* Once woken up, create a clone of our process. */
-
-      child_pid = fork();
-      if (child_pid < 0) {
-
-        write_error("fork");
-        _exit(1);
-
-      }
-
-      /* In child process: close fds, resume execution. */
-
-      if (!child_pid) {
-
-        //(void)nice(-20);  // does not seem to improve
-
-        signal(SIGCHLD, old_sigchld_handler);
-        signal(SIGTERM, old_sigterm_handler);
-
-        close(FORKSRV_FD);
-        close(FORKSRV_FD + 1);
-
-        if (!afl_snapshot_take(AFL_SNAPSHOT_MMAP | AFL_SNAPSHOT_FDS |
-                               AFL_SNAPSHOT_REGS | AFL_SNAPSHOT_EXIT)) {
-
-          raise(SIGSTOP);
-
-        }
-
-        __afl_area_ptr[0] = 1;
-        memset(__afl_prev_loc, 0, NGRAM_SIZE_MAX * sizeof(PREV_LOC_T));
-
-        return;
-
-      }
-
-    } else {
-
-      /* Special handling for persistent mode: if the child is alive but
-         currently stopped, simply restart it with SIGCONT. */
-
-      kill(child_pid, SIGCONT);
-      child_stopped = 0;
-
-    }
-
-    /* In parent process: write PID to pipe, then wait for child. */
-
-    if (write(FORKSRV_FD + 1, &child_pid, 4) != 4) {
-
-      write_error("write to afl-fuzz");
-      _exit(1);
-
-    }
-
-    if (waitpid(child_pid, &status, WUNTRACED) < 0) {
-
-      write_error("waitpid");
-      _exit(1);
-
-    }
-
-    /* In persistent mode, the child stops itself with SIGSTOP to indicate
-       a successful run. In this case, we want to wake it up without forking
-       again. */
-
-    if (WIFSTOPPED(status)) child_stopped = 1;
-
-    /* Relay wait status to pipe, then loop back. */
-
-    if (write(FORKSRV_FD + 1, &status, 4) != 4) {
-
-      write_error("writing to afl-fuzz");
-      _exit(1);
-
-    }
-
-  }
-
-}
-
-#endif
-
 /* Fork server logic. */
 
 static void __afl_start_forkserver(void) {
@@ -1103,113 +836,92 @@ static void __afl_start_forkserver(void) {
   old_sigterm_handler = orig_action.sa_handler;
   signal(SIGTERM, at_exit);
 
-#ifdef __linux__
-  if (/*!is_persistent &&*/ !__afl_cmp_map && !getenv("AFL_NO_SNAPSHOT") &&
-      afl_snapshot_init() >= 0) {
-
-    __afl_start_snapshots();
-    return;
-
-  }
-
-#endif
-
-  u8  tmp[4] = {0, 0, 0, 0};
-  u32 status_for_fsrv = 0;
   u32 already_read_first = 0;
   u32 was_killed;
+  u32 version = 0x41464c00 + FS_NEW_VERSION_MAX;
+  u32 tmp = version ^ 0xffffffff, status2, status = version;
+  u8 *msg = (u8 *)&status;
+  u8 *reply = (u8 *)&status2;
 
   u8 child_stopped = 0;
 
   void (*old_sigchld_handler)(int) = signal(SIGCHLD, SIG_DFL);
 
-  if (__afl_map_size <= FS_OPT_MAX_MAPSIZE) {
-
-    status_for_fsrv |= (FS_OPT_SET_MAPSIZE(__afl_map_size) | FS_OPT_MAPSIZE);
+  /* Phone home and tell the parent that we're OK. If parent isn't there,
+     assume we're not running in forkserver mode and just execute program. */
 
-  }
+  // return because possible non-forkserver usage
+  if (write(FORKSRV_FD + 1, msg, 4) != 4) { return; }
 
-  if (__afl_dictionary_len && __afl_dictionary) {
+  if (read(FORKSRV_FD, reply, 4) != 4) { _exit(1); }
+  if (tmp != status2) {
 
-    status_for_fsrv |= FS_OPT_AUTODICT;
+    write_error("wrong forkserver message from AFL++ tool");
+    _exit(1);
 
   }
 
-  if (__afl_sharedmem_fuzzing) { status_for_fsrv |= FS_OPT_SHDMEM_FUZZ; }
-  if (status_for_fsrv) {
+  // send the set/requested options to forkserver
+  status = FS_NEW_OPT_MAPSIZE;  // we always send the map size
+  if (__afl_sharedmem_fuzzing) { status |= FS_NEW_OPT_SHDMEM_FUZZ; }
+  if (__afl_dictionary_len && __afl_dictionary) {
 
-    status_for_fsrv |= (FS_OPT_ENABLED | FS_OPT_NEWCMPLOG);
+    status |= FS_NEW_OPT_AUTODICT;
 
   }
 
-  memcpy(tmp, &status_for_fsrv, 4);
+  if (write(FORKSRV_FD + 1, msg, 4) != 4) { _exit(1); }
 
-  /* Phone home and tell the parent that we're OK. If parent isn't there,
-     assume we're not running in forkserver mode and just execute program. */
-
-  if (write(FORKSRV_FD + 1, tmp, 4) != 4) { return; }
-
-  __afl_connected = 1;
+  // Now send the parameters for the set options, increasing by option number
 
-  if (__afl_sharedmem_fuzzing || (__afl_dictionary_len && __afl_dictionary)) {
+  // FS_NEW_OPT_MAPSIZE - we always send the map size
+  status = __afl_map_size;
+  if (write(FORKSRV_FD + 1, msg, 4) != 4) { _exit(1); }
 
-    if (read(FORKSRV_FD, &was_killed, 4) != 4) _exit(1);
+  // FS_NEW_OPT_SHDMEM_FUZZ - no data
 
-    if (__afl_debug) {
-
-      fprintf(stderr, "DEBUG: target forkserver recv: %08x\n", was_killed);
+  // FS_NEW_OPT_AUTODICT - send autodictionary
+  if (__afl_dictionary_len && __afl_dictionary) {
 
-    }
+    // pass the dictionary through the forkserver FD
+    u32 len = __afl_dictionary_len, offset = 0;
 
-    if ((was_killed & (FS_OPT_ENABLED | FS_OPT_SHDMEM_FUZZ)) ==
-        (FS_OPT_ENABLED | FS_OPT_SHDMEM_FUZZ)) {
+    if (write(FORKSRV_FD + 1, &len, 4) != 4) {
 
-      __afl_map_shm_fuzz();
+      write(2, "Error: could not send dictionary len\n",
+            strlen("Error: could not send dictionary len\n"));
+      _exit(1);
 
     }
 
-    if ((was_killed & (FS_OPT_ENABLED | FS_OPT_AUTODICT)) ==
-            (FS_OPT_ENABLED | FS_OPT_AUTODICT) &&
-        __afl_dictionary_len && __afl_dictionary) {
+    while (len != 0) {
 
-      // great lets pass the dictionary through the forkserver FD
-      u32 len = __afl_dictionary_len, offset = 0;
+      s32 ret;
+      ret = write(FORKSRV_FD + 1, __afl_dictionary + offset, len);
 
-      if (write(FORKSRV_FD + 1, &len, 4) != 4) {
+      if (ret < 1) {
 
-        write(2, "Error: could not send dictionary len\n",
-              strlen("Error: could not send dictionary len\n"));
+        write_error("could not send dictionary");
         _exit(1);
 
       }
 
-      while (len != 0) {
-
-        s32 ret;
-        ret = write(FORKSRV_FD + 1, __afl_dictionary + offset, len);
-
-        if (ret < 1) {
-
-          write(2, "Error: could not send dictionary\n",
-                strlen("Error: could not send dictionary\n"));
-          _exit(1);
-
-        }
+      len -= ret;
+      offset += ret;
 
-        len -= ret;
-        offset += ret;
+    }
 
-      }
+  }
 
-    } else {
+  // send welcome message as final message
+  status = version;
+  if (write(FORKSRV_FD + 1, msg, 4) != 4) { _exit(1); }
 
-      // uh this forkserver does not understand extended option passing
-      // or does not want the dictionary
-      if (!__afl_fuzz_ptr) already_read_first = 1;
+  // END forkserver handshake
 
-    }
+  __afl_connected = 1;
 
-  }
+  if (__afl_sharedmem_fuzzing) { __afl_map_shm_fuzz(); }
 
   while (1) {
 
@@ -1225,7 +937,7 @@ static void __afl_start_forkserver(void) {
 
       if (read(FORKSRV_FD, &was_killed, 4) != 4) {
 
-        // write_error("read from afl-fuzz");
+        write_error("read from AFL++ tool");
         _exit(1);
 
       }
diff --git a/src/afl-forkserver.c b/src/afl-forkserver.c
index 1f796e53..a3a869d7 100644
--- a/src/afl-forkserver.c
+++ b/src/afl-forkserver.c
@@ -1030,6 +1030,12 @@ void afl_fsrv_start(afl_forkserver_t *fsrv, char **argv,
      *  send any data anymore - except a future option requires this.
      */
 
+    if ((status & FS_NEW_ERROR) == FS_NEW_ERROR) {
+
+      report_error_and_exit(status & 0x0000ffff);
+
+    }
+
     if (status >= 0x41464c00 && status <= 0x41464cff) {
 
       u32 version = status - 0x41464c00;
@@ -1047,6 +1053,7 @@ void afl_fsrv_start(afl_forkserver_t *fsrv, char **argv,
 
       }
 
+      u32 keep = status;
       status ^= 0xffffffff;
       if (write(fsrv->fsrv_ctl_fd, &status, 4) != 4) {
 
@@ -1064,7 +1071,7 @@ void afl_fsrv_start(afl_forkserver_t *fsrv, char **argv,
 
       if (getenv("AFL_DEBUG")) {
 
-        ACTF("Forkserver options received: (%08x)", status);
+        ACTF("Forkserver options received: (0x%08x)", status);
 
       }
 
@@ -1178,7 +1185,11 @@ void afl_fsrv_start(afl_forkserver_t *fsrv, char **argv,
       u32 status2;
       rlen = read(fsrv->fsrv_st_fd, &status2, 4);
 
-      if (status2 != status) { FATAL("Error in forkserver communication"); }
+      if (status2 != keep) {
+
+        FATAL("Error in forkserver communication (%08x=>%08x)", keep, status2);
+
+      }
 
     } else {
 
-- 
cgit 1.4.1


From c77709cdd9b50832ed537dfd65d30bc7ffa79e7b Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Sun, 4 Feb 2024 16:03:12 +0100
Subject: add U256/32byte support

---
 include/cmplog.h                    | 19 ++++-----
 instrumentation/afl-compiler-rt.o.c | 78 +++++++++++++++++++++++++++----------
 src/afl-forkserver.c                | 10 +++++
 src/afl-fuzz-redqueen.c             | 10 ++---
 src/afl-fuzz.c                      |  6 ++-
 test/test-llvm.sh                   |  2 +-
 6 files changed, 87 insertions(+), 38 deletions(-)

(limited to 'include')

diff --git a/include/cmplog.h b/include/cmplog.h
index 6bfc146b..91c2a665 100644
--- a/include/cmplog.h
+++ b/include/cmplog.h
@@ -43,13 +43,11 @@
 
 struct cmp_header {
 
-  unsigned hits : 24;
-  unsigned id : 24;
-  unsigned shape : 5;
-  unsigned type : 2;
-  unsigned attribute : 4;
-  unsigned overflow : 1;
-  unsigned reserved : 4;
+  unsigned hits : 6;       // up to 63 entries, we have CMP_MAP_H = 32
+  unsigned shape : 6;      // 63 bytes, we support 32 max
+  unsigned type : 2;       // 4, we use 3: none, rtn, cmp
+  unsigned attribute : 4;  // 16 for arithmetic comparison types
+  unsigned reserved : 6;
 
 } __attribute__((packed));
 
@@ -59,14 +57,17 @@ struct cmp_operands {
   u64 v1;
   u64 v0_128;
   u64 v1_128;
+  u64 unused;
+  u8  unused1;
+  u8  unused2;
 
 } __attribute__((packed));
 
 struct cmpfn_operands {
 
-  u8 v0[31];
+  u8 v0[32];
   u8 v0_len;
-  u8 v1[31];
+  u8 v1[32];
   u8 v1_len;
 
 } __attribute__((packed));
diff --git a/instrumentation/afl-compiler-rt.o.c b/instrumentation/afl-compiler-rt.o.c
index c342334c..a154bcf7 100644
--- a/instrumentation/afl-compiler-rt.o.c
+++ b/instrumentation/afl-compiler-rt.o.c
@@ -186,6 +186,8 @@ __thread u32        __afl_prev_ctx;
 struct cmp_map *__afl_cmp_map;
 struct cmp_map *__afl_cmp_map_backup;
 
+static u8 __afl_cmplog_max_len = 16;
+
 /* Child pid? */
 
 static s32 child_pid;
@@ -730,6 +732,12 @@ static void __afl_map_shm(void) {
 
 #endif  // __AFL_CODE_COVERAGE
 
+  if (!__afl_cmp_map && getenv("AFL_CMPLOG_DEBUG")) {
+
+    __afl_cmp_map_backup = __afl_cmp_map = malloc(sizeof(struct cmp_map));
+
+  }
+
 }
 
 /* unmap SHM. */
@@ -1893,7 +1901,8 @@ void __cmplog_ins_hook1(uint8_t arg1, uint8_t arg2, uint8_t attr) {
 
 void __cmplog_ins_hook2(uint16_t arg1, uint16_t arg2, uint8_t attr) {
 
-  if (unlikely(!__afl_cmp_map || arg1 == arg2)) return;
+  if (likely(!__afl_cmp_map)) return;
+  if (unlikely(arg1 == arg2)) return;
 
   uintptr_t k = (uintptr_t)__builtin_return_address(0);
   k = (uintptr_t)(default_hash((u8 *)&k, sizeof(uintptr_t)) & (CMP_MAP_W - 1));
@@ -1931,7 +1940,8 @@ void __cmplog_ins_hook4(uint32_t arg1, uint32_t arg2, uint8_t attr) {
 
   // fprintf(stderr, "hook4 arg0=%x arg1=%x attr=%u\n", arg1, arg2, attr);
 
-  if (unlikely(!__afl_cmp_map || arg1 == arg2)) return;
+  if (likely(!__afl_cmp_map)) return;
+  if (unlikely(arg1 == arg2)) return;
 
   uintptr_t k = (uintptr_t)__builtin_return_address(0);
   k = (uintptr_t)(default_hash((u8 *)&k, sizeof(uintptr_t)) & (CMP_MAP_W - 1));
@@ -1969,7 +1979,8 @@ void __cmplog_ins_hook8(uint64_t arg1, uint64_t arg2, uint8_t attr) {
 
   // fprintf(stderr, "hook8 arg0=%lx arg1=%lx attr=%u\n", arg1, arg2, attr);
 
-  if (unlikely(!__afl_cmp_map || arg1 == arg2)) return;
+  if (likely(!__afl_cmp_map)) return;
+  if (unlikely(arg1 == arg2)) return;
 
   uintptr_t k = (uintptr_t)__builtin_return_address(0);
   k = (uintptr_t)(default_hash((u8 *)&k, sizeof(uintptr_t)) & (CMP_MAP_W - 1));
@@ -2012,7 +2023,8 @@ void __cmplog_ins_hookN(uint128_t arg1, uint128_t arg2, uint8_t attr,
   // (u64)(arg1 >> 64), (u64)arg1, (u64)(arg2 >> 64), (u64)arg2, size + 1,
   // attr);
 
-  if (unlikely(!__afl_cmp_map || arg1 == arg2)) return;
+  if (likely(!__afl_cmp_map)) return;
+  if (unlikely(arg1 == arg2 || size > __afl_cmplog_max_len)) return;
 
   uintptr_t k = (uintptr_t)__builtin_return_address(0);
   k = (uintptr_t)(default_hash((u8 *)&k, sizeof(uintptr_t)) & (CMP_MAP_W - 1));
@@ -2056,6 +2068,7 @@ void __cmplog_ins_hookN(uint128_t arg1, uint128_t arg2, uint8_t attr,
 void __cmplog_ins_hook16(uint128_t arg1, uint128_t arg2, uint8_t attr) {
 
   if (likely(!__afl_cmp_map)) return;
+  if (16 > __afl_cmplog_max_len) return;
 
   uintptr_t k = (uintptr_t)__builtin_return_address(0);
   k = (uintptr_t)(default_hash((u8 *)&k, sizeof(uintptr_t)) & (CMP_MAP_W - 1));
@@ -2249,13 +2262,25 @@ void __cmplog_rtn_hook_strn(u8 *ptr1, u8 *ptr2, u64 len) {
 
   // fprintf(stderr, "RTN1 %p %p %u\n", ptr1, ptr2, len);
   if (likely(!__afl_cmp_map)) return;
-  if (unlikely(!len)) return;
-  int len0 = MIN(len, 31);
+  if (unlikely(!len || len > __afl_cmplog_max_len)) return;
+
+  int len0 = MIN(len, 32);
+
   int len1 = strnlen(ptr1, len0);
-  if (len1 < 31) len1 = area_is_valid(ptr1, len1 + 1);
+  if (len1 <= 32) len1 = area_is_valid(ptr1, len1 + 1);
+  if (len1 > __afl_cmplog_max_len) len1 = 0;
+
   int len2 = strnlen(ptr2, len0);
-  if (len2 < 31) len2 = area_is_valid(ptr2, len2 + 1);
-  int l = MAX(len1, len2);
+  if (len2 <= 32) len2 = area_is_valid(ptr2, len2 + 1);
+  if (len2 > __afl_cmplog_max_len) len2 = 0;
+
+  int l;
+  if (!len1)
+    l = len2;
+  else if (!len2)
+    l = len1;
+  else
+    l = MAX(len1, len2);
   if (l < 2) return;
 
   uintptr_t k = (uintptr_t)__builtin_return_address(0);
@@ -2299,10 +2324,18 @@ void __cmplog_rtn_hook_str(u8 *ptr1, u8 *ptr2) {
   // fprintf(stderr, "RTN1 %p %p\n", ptr1, ptr2);
   if (likely(!__afl_cmp_map)) return;
   if (unlikely(!ptr1 || !ptr2)) return;
-  int len1 = strnlen(ptr1, 30) + 1;
-  int len2 = strnlen(ptr2, 30) + 1;
-  int l = MAX(len1, len2);
-  if (l < 3) return;
+  int len1 = strnlen(ptr1, 31) + 1;
+  int len2 = strnlen(ptr2, 31) + 1;
+  if (len1 > __afl_cmplog_max_len) len1 = 0;
+  if (len2 > __afl_cmplog_max_len) len2 = 0;
+  int l;
+  if (!len1)
+    l = len2;
+  else if (!len2)
+    l = len1;
+  else
+    l = MAX(len1, len2);
+  if (l < 2) return;
 
   uintptr_t k = (uintptr_t)__builtin_return_address(0);
   k = (uintptr_t)(default_hash((u8 *)&k, sizeof(uintptr_t)) & (CMP_MAP_W - 1));
@@ -2344,7 +2377,7 @@ void __cmplog_rtn_hook(u8 *ptr1, u8 *ptr2) {
 
   /*
     u32 i;
-    if (area_is_valid(ptr1, 31) <= 0 || area_is_valid(ptr2, 31) <= 0) return;
+    if (area_is_valid(ptr1, 32) <= 0 || area_is_valid(ptr2, 32) <= 0) return;
     fprintf(stderr, "rtn arg0=");
     for (i = 0; i < 32; i++)
       fprintf(stderr, "%02x", ptr1[i]);
@@ -2357,10 +2390,10 @@ void __cmplog_rtn_hook(u8 *ptr1, u8 *ptr2) {
   // fprintf(stderr, "RTN1 %p %p\n", ptr1, ptr2);
   if (likely(!__afl_cmp_map)) return;
   int l1, l2;
-  if ((l1 = area_is_valid(ptr1, 31)) <= 0 ||
-      (l2 = area_is_valid(ptr2, 31)) <= 0)
+  if ((l1 = area_is_valid(ptr1, 32)) <= 0 ||
+      (l2 = area_is_valid(ptr2, 32)) <= 0)
     return;
-  int len = MIN(31, MIN(l1, l2));
+  int len = MIN(__afl_cmplog_max_len, MIN(l1, l2));
 
   // fprintf(stderr, "RTN2 %u\n", len);
   uintptr_t k = (uintptr_t)__builtin_return_address(0);
@@ -2409,7 +2442,7 @@ void __cmplog_rtn_hook_n(u8 *ptr1, u8 *ptr2, u64 len) {
 #if 0
   /*
     u32 i;
-    if (area_is_valid(ptr1, 31) <= 0 || area_is_valid(ptr2, 31) <= 0) return;
+    if (area_is_valid(ptr1, 32) <= 0 || area_is_valid(ptr2, 32) <= 0) return;
     fprintf(stderr, "rtn_n len=%u arg0=", len);
     for (i = 0; i < len; i++)
       fprintf(stderr, "%02x", ptr1[i]);
@@ -2421,12 +2454,15 @@ void __cmplog_rtn_hook_n(u8 *ptr1, u8 *ptr2, u64 len) {
 
   // fprintf(stderr, "RTN1 %p %p %u\n", ptr1, ptr2, len);
   if (likely(!__afl_cmp_map)) return;
-  if (unlikely(!len)) return;
-  int l = MIN(31, len);
+  if (!len) return;
+  int l = MIN(32, len), l1, l2;
 
-  if ((l = area_is_valid(ptr1, l)) <= 0 || (l = area_is_valid(ptr2, l)) <= 0)
+  if ((l1 = area_is_valid(ptr1, l)) <= 0 || (l2 = area_is_valid(ptr2, l)) <= 0)
     return;
 
+  len = MIN(l1, l2);
+  if (len > __afl_cmplog_max_len) return;
+
   // fprintf(stderr, "RTN2 %u\n", l);
   uintptr_t k = (uintptr_t)__builtin_return_address(0);
   k = (uintptr_t)(default_hash((u8 *)&k, sizeof(uintptr_t)) & (CMP_MAP_W - 1));
diff --git a/src/afl-forkserver.c b/src/afl-forkserver.c
index a3a869d7..c5184639 100644
--- a/src/afl-forkserver.c
+++ b/src/afl-forkserver.c
@@ -1105,6 +1105,10 @@ void afl_fsrv_start(afl_forkserver_t *fsrv, char **argv,
 
         fsrv->map_size = tmp_map_size;
 
+      } else {
+
+        fsrv->real_map_size = fsrv->map_size = MAP_SIZE;
+
       }
 
       if ((status & FS_NEW_OPT_SHDMEM_FUZZ)) {
@@ -1208,6 +1212,12 @@ void afl_fsrv_start(afl_forkserver_t *fsrv, char **argv,
       if ((status & FS_OPT_ERROR) == FS_OPT_ERROR)
         report_error_and_exit(FS_OPT_GET_ERROR(status));
 
+      if (fsrv->cmplog_binary) {
+
+        FATAL("Target was recompiled with outdated CMPLOG, recompile it!\n");
+
+      }
+
       if ((status & FS_OPT_ENABLED) == FS_OPT_ENABLED) {
 
         // workaround for recent AFL++ versions
diff --git a/src/afl-fuzz-redqueen.c b/src/afl-fuzz-redqueen.c
index eead7a8b..eb96de68 100644
--- a/src/afl-fuzz-redqueen.c
+++ b/src/afl-fuzz-redqueen.c
@@ -11,7 +11,7 @@
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
-   Copyright 2019-2024 AFLplusplus Project. All rights reserved.
+   Copyright 2019-2023 AFLplusplus Project. All rights reserved.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
@@ -2219,15 +2219,15 @@ static u8 rtn_extend_encoding(afl_state_t *afl, u8 entry,
 
   }
 
-  if (l0 == 0 || l1 == 0 || ol0 == 0 || ol1 == 0 || l0 > 31 || l1 > 31 ||
-      ol0 > 31 || ol1 > 31) {
+  if (l0 == 0 || l1 == 0 || ol0 == 0 || ol1 == 0 || l0 > 32 || l1 > 32 ||
+      ol0 > 32 || ol1 > 32) {
 
     l0 = ol0 = hshape;
 
   }
 
   u8  lmax = MAX(l0, ol0);
-  u8  save[40];
+  u8  save[80];
   u32 saved_idx = idx, pre, from = 0, to = 0, i, j;
   u32 its_len = MIN(MIN(lmax, hshape), len - idx);
   its_len = MIN(its_len, taint_len);
@@ -2330,7 +2330,7 @@ static u8 rtn_extend_encoding(afl_state_t *afl, u8 entry,
     u32 tob64 = 0, fromb64 = 0;
     u32 from_0 = 0, from_x = 0, from_X = 0, from_slash = 0, from_up = 0;
     u32 to_0 = 0, to_x = 0, to_slash = 0, to_up = 0;
-    u8  xor_val[32], arith_val[32], tmp[48];
+    u8  xor_val[64], arith_val[64], tmp[64];
 
     idx = saved_idx;
     its_len = saved_its_len;
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index b556b4b6..34268113 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -956,9 +956,11 @@ int main(int argc, char **argv_orig, char **envp) {
       break;
 
       case 'd':
-      case 'D':                                    /* old deterministic */
+      case 'D':                                        /* old deterministic */
 
-        WARNF("Parameters -d and -D are deprecated, a new enhanced deterministic fuzzing is active by default, to disable it use -z");
+        WARNF(
+            "Parameters -d and -D are deprecated, a new enhanced deterministic "
+            "fuzzing is active by default, to disable it use -z");
         break;
 
       case 'z':                                         /* no deterministic */
diff --git a/test/test-llvm.sh b/test/test-llvm.sh
index 53bbd7b4..aef7a5e2 100755
--- a/test/test-llvm.sh
+++ b/test/test-llvm.sh
@@ -62,7 +62,7 @@ test -e ../afl-clang-fast -a -e ../split-switches-pass.so && {
       $ECHO "$RED[!] llvm_mode threadsafe instrumentation failed"
       CODE=1
     }
-    rm -f test-instr.ts.0 test-instr.ts.1
+    rm -f test-instr.ts.0 test-instr.ts.1 test-instr.ts
   } || {
     $ECHO "$RED[!] llvm_mode (threadsafe) failed"
     CODE=1
-- 
cgit 1.4.1


From 34a3060b0fee900331decce0bae55d4caefafc01 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Sun, 4 Feb 2024 16:08:57 +0100
Subject: config __afl_cmplog_max_len

---
 include/envs.h                      | 2 +-
 instrumentation/afl-compiler-rt.o.c | 9 ++++++++-
 2 files changed, 9 insertions(+), 2 deletions(-)

(limited to 'include')

diff --git a/include/envs.h b/include/envs.h
index 0f645d23..8f342553 100644
--- a/include/envs.h
+++ b/include/envs.h
@@ -21,7 +21,7 @@ static char *afl_environment_variables[] = {
     "AFL_BENCH_UNTIL_CRASH", "AFL_CAL_FAST", "AFL_CC", "AFL_CC_COMPILER",
     "AFL_CMIN_ALLOW_ANY", "AFL_CMIN_CRASHES_ONLY", "AFL_CMPLOG_ONLY_NEW",
     "AFL_CODE_END", "AFL_CODE_START", "AFL_COMPCOV_BINNAME",
-    "AFL_COMPCOV_LEVEL", "AFL_CRASH_EXITCODE",
+    "AFL_CMPLOG_MAX_LEN", "AFL_COMPCOV_LEVEL", "AFL_CRASH_EXITCODE",
     "AFL_CRASHING_SEEDS_AS_NEW_CRASH", "AFL_CUSTOM_MUTATOR_LIBRARY",
     "AFL_CUSTOM_MUTATOR_ONLY", "AFL_CUSTOM_INFO_PROGRAM",
     "AFL_CUSTOM_INFO_PROGRAM_ARGV", "AFL_CUSTOM_INFO_PROGRAM_INPUT",
diff --git a/instrumentation/afl-compiler-rt.o.c b/instrumentation/afl-compiler-rt.o.c
index a154bcf7..c4177b08 100644
--- a/instrumentation/afl-compiler-rt.o.c
+++ b/instrumentation/afl-compiler-rt.o.c
@@ -186,7 +186,7 @@ __thread u32        __afl_prev_ctx;
 struct cmp_map *__afl_cmp_map;
 struct cmp_map *__afl_cmp_map_backup;
 
-static u8 __afl_cmplog_max_len = 16;
+static u8 __afl_cmplog_max_len = 32;  // 16-32
 
 /* Child pid? */
 
@@ -738,6 +738,13 @@ static void __afl_map_shm(void) {
 
   }
 
+  if (getenv("AFL_CMPLOG_MAX_LEN")) {
+
+    int tmp = atoi(getenv("AFL_CMPLOG_MAX_LEN"));
+    if (tmp >= 16 && tmp <= 32) { __afl_cmplog_max_len = tmp; }
+
+  }
+
 }
 
 /* unmap SHM. */
-- 
cgit 1.4.1


From 47e7d243f7522b21beeb7bb9e91f3f312a9659f0 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Sun, 4 Feb 2024 16:18:21 +0100
Subject: increase version

---
 README.md        | 2 +-
 include/config.h | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

(limited to 'include')

diff --git a/README.md b/README.md
index f713e971..f15089c2 100644
--- a/README.md
+++ b/README.md
@@ -4,7 +4,7 @@
 
 Release version: [4.10c](https://github.com/AFLplusplus/AFLplusplus/releases)
 
-GitHub version: 4.10c
+GitHub version: 4.20a
 
 Repository:
 [https://github.com/AFLplusplus/AFLplusplus](https://github.com/AFLplusplus/AFLplusplus)
diff --git a/include/config.h b/include/config.h
index 9349828f..f3392431 100644
--- a/include/config.h
+++ b/include/config.h
@@ -26,7 +26,7 @@
 /* Version string: */
 
 // c = release, a = volatile github dev, e = experimental branch
-#define VERSION "++4.10c"
+#define VERSION "++4.20a"
 
 /******************************************************
  *                                                    *
-- 
cgit 1.4.1


From 40df85d1e6fb80e9d641064e645a48b623aee681 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Mon, 5 Feb 2024 15:05:46 +0100
Subject: adjust cmplog header

---
 .github/workflows/ci.yml |   1 +
 GNUmakefile              |   4 +-
 include/cmplog.h         |  10 ++--
 src/afl-fuzz-redqueen.c  |  43 ++++++++++++--
 src/hashmap.c            | 149 +++++++++++++++++++++++++++++++++++++++++++++++
 5 files changed, 194 insertions(+), 13 deletions(-)
 create mode 100644 src/hashmap.c

(limited to 'include')

diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index ed382fbb..dd0d13e9 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -5,6 +5,7 @@ on:
     branches:
       - stable
       - dev
+      - 420
   pull_request:
     branches:
       - dev # No need for stable-pull-request, as that equals dev-push
diff --git a/GNUmakefile b/GNUmakefile
index 283c57c2..d3cf2674 100644
--- a/GNUmakefile
+++ b/GNUmakefile
@@ -464,8 +464,8 @@ src/afl-forkserver.o : $(COMM_HDR) src/afl-forkserver.c include/forkserver.h
 src/afl-sharedmem.o : $(COMM_HDR) src/afl-sharedmem.c include/sharedmem.h
 	$(CC) $(CFLAGS) $(CFLAGS_FLTO) -c src/afl-sharedmem.c -o src/afl-sharedmem.o
 
-afl-fuzz: $(COMM_HDR) include/afl-fuzz.h $(AFL_FUZZ_FILES) src/afl-common.o src/afl-sharedmem.o src/afl-forkserver.o src/afl-performance.o | test_x86
-	$(CC) $(CFLAGS) $(COMPILE_STATIC) $(CFLAGS_FLTO) $(AFL_FUZZ_FILES) src/afl-common.o src/afl-sharedmem.o src/afl-forkserver.o src/afl-performance.o -o $@ $(PYFLAGS) $(LDFLAGS) -lm
+afl-fuzz: $(COMM_HDR) include/afl-fuzz.h $(AFL_FUZZ_FILES) src/afl-common.o src/afl-sharedmem.o src/afl-forkserver.o src/afl-performance.o src/hashmap.c | test_x86
+	$(CC) $(CFLAGS) $(COMPILE_STATIC) $(CFLAGS_FLTO) -Wno-shift-count-overflow $(AFL_FUZZ_FILES) src/afl-common.o src/afl-sharedmem.o src/afl-forkserver.o src/afl-performance.o src/hashmap.c -o $@ $(PYFLAGS) $(LDFLAGS) -lm
 
 afl-showmap: src/afl-showmap.c src/afl-common.o src/afl-sharedmem.o src/afl-forkserver.o src/afl-performance.o $(COMM_HDR) | test_x86
 	$(CC) $(CFLAGS) $(COMPILE_STATIC) $(CFLAGS_FLTO) src/$@.c src/afl-fuzz-mutators.c src/afl-fuzz-python.c src/afl-common.o src/afl-sharedmem.o src/afl-forkserver.o src/afl-performance.o -o $@ $(PYFLAGS) $(LDFLAGS)
diff --git a/include/cmplog.h b/include/cmplog.h
index 91c2a665..589570fe 100644
--- a/include/cmplog.h
+++ b/include/cmplog.h
@@ -38,16 +38,16 @@
 
 #define SHAPE_BYTES(x) (x + 1)
 
-#define CMP_TYPE_INS 1
-#define CMP_TYPE_RTN 2
+#define CMP_TYPE_INS 0
+#define CMP_TYPE_RTN 1
 
 struct cmp_header {
 
   unsigned hits : 6;       // up to 63 entries, we have CMP_MAP_H = 32
-  unsigned shape : 6;      // 63 bytes, we support 32 max
-  unsigned type : 2;       // 4, we use 3: none, rtn, cmp
+  unsigned shape : 5;      // 31+1 bytes
+  unsigned type : 1;       // 4, we use 3: none, rtn, cmp
   unsigned attribute : 4;  // 16 for arithmetic comparison types
-  unsigned reserved : 6;
+  //unsigned reserved : 6;
 
 } __attribute__((packed));
 
diff --git a/src/afl-fuzz-redqueen.c b/src/afl-fuzz-redqueen.c
index eb96de68..bc83c9ed 100644
--- a/src/afl-fuzz-redqueen.c
+++ b/src/afl-fuzz-redqueen.c
@@ -87,6 +87,11 @@ static u32 hshape;
 static u64 screen_update;
 static u64 last_update;
 
+// hashmap functions
+void hashmap_reset();
+bool hashmap_search_and_add(uint8_t type, uint64_t key);
+bool hashmap_search_and_add_ptr(uint8_t type, u8 *key);
+
 static struct range *add_range(struct range *ranges, u32 start, u32 end) {
 
   struct range *r = ck_alloc_nozero(sizeof(struct range));
@@ -795,7 +800,7 @@ static u8 cmp_extend_encoding(afl_state_t *afl, struct cmp_header *h,
   u64 *o_buf_64 = (u64 *)&orig_buf[idx];
   u32 *o_buf_32 = (u32 *)&orig_buf[idx];
   u16 *o_buf_16 = (u16 *)&orig_buf[idx];
-  u8  *o_buf_8 = &orig_buf[idx];
+  // u8  *o_buf_8 = &orig_buf[idx];
 
   u32 its_len = MIN(len - idx, taint_len);
 
@@ -836,6 +841,7 @@ static u8 cmp_extend_encoding(afl_state_t *afl, struct cmp_header *h,
 
   // necessary for preventing heap access overflow
   bytes = MIN(bytes, len - idx);
+  if (unlikely(bytes <= 1)) { return 0; }
 
   //  reverse atoi()/strnu?toll() is expensive, so we only to it in lvl 3
   if (afl->cmplog_enable_transform && (lvl & LVL3)) {
@@ -1266,6 +1272,7 @@ static u8 cmp_extend_encoding(afl_state_t *afl, struct cmp_header *h,
 
     }
 
+    /*
     if (*status != 1) {  // u8
 
       // if (its_len >= 1)
@@ -1290,6 +1297,8 @@ static u8 cmp_extend_encoding(afl_state_t *afl, struct cmp_header *h,
 
     }
 
+    */
+
   }
 
   // If 'S' is set for cmplog mode then we try a scale encoding of the value.
@@ -1881,6 +1890,8 @@ static u8 cmp_fuzz(afl_state_t *afl, u32 key, u8 *orig_buf, u8 *buf, u8 *cbuf,
 
   hshape = SHAPE_BYTES(h->shape);
 
+  if (hshape < 2) { return 0; }
+
   if (h->hits > CMP_MAP_H) {
 
     loggeds = CMP_MAP_H;
@@ -1906,8 +1917,6 @@ static u8 cmp_fuzz(afl_state_t *afl, u32 key, u8 *orig_buf, u8 *buf, u8 *cbuf,
 
 #endif
 
-  if (hshape < 2) { return 0; }
-
   for (i = 0; i < loggeds; ++i) {
 
     struct cmp_operands *o = &afl->shm.cmp_map->log[key][i];
@@ -1945,6 +1954,16 @@ static u8 cmp_fuzz(afl_state_t *afl, u32 key, u8 *orig_buf, u8 *buf, u8 *cbuf,
 
     }
 
+    // TODO: add attribute? not sure
+    if (hshape <= 8 && !hashmap_search_and_add(hshape - 1, o->v0) &&
+        !hashmap_search_and_add(hshape - 1, orig_o->v0) &&
+        !hashmap_search_and_add(hshape - 1, o->v1) &&
+        !hashmap_search_and_add(hshape - 1, orig_o->v1)) {
+
+      continue;
+
+    }
+
 #ifdef _DEBUG
     fprintf(stderr, "Handling: %llx->%llx vs %llx->%llx attr=%u shape=%u\n",
             orig_o->v0, o->v0, orig_o->v1, o->v1, h->attribute, hshape);
@@ -2615,12 +2634,13 @@ static u8 rtn_extend_encoding(afl_state_t *afl, u8 entry,
 
           }
 
-          memcpy(buf + idx, tmp, hlen + 1 + off);
+          u32 tmp_l = hlen + 1 + off;
+          memcpy(buf + idx, tmp, tmp_l);
           if (unlikely(its_fuzz(afl, buf, len, status))) { return 1; }
-          tmp[hlen + 1 + off] = 0;
+          tmp[tmp_l] = 0;
           // fprintf(stderr, "RTN ATTEMPT idx=%u len=%u fromhex %u %s %s result
           // %u\n", idx, len, fromhex, tmp, repl, *status);
-          memcpy(buf + idx, save, hlen + 1 + off);
+          memcpy(buf + idx, save, tmp_l);
 
         }
 
@@ -2755,6 +2775,15 @@ static u8 rtn_fuzz(afl_state_t *afl, u32 key, u8 *orig_buf, u8 *buf, u8 *cbuf,
     fprintf(stderr, "\n");
 #endif
 
+    if (hshape <= 8 && !hashmap_search_and_add_ptr(hshape - 1, o->v0) &&
+        !hashmap_search_and_add_ptr(hshape - 1, orig_o->v0) &&
+        !hashmap_search_and_add_ptr(hshape - 1, o->v1) &&
+        !hashmap_search_and_add_ptr(hshape - 1, orig_o->v1)) {
+
+      continue;
+
+    }
+
     t = taint;
     while (t->next) {
 
@@ -3021,6 +3050,8 @@ u8 input_to_state_stage(afl_state_t *afl, u8 *orig_buf, u8 *buf, u32 len) {
 
   // Start insertion loop
 
+  hashmap_reset();
+
   u64 orig_hit_cnt, new_hit_cnt;
   u64 orig_execs = afl->fsrv.total_execs;
   orig_hit_cnt = afl->queued_items + afl->saved_crashes;
diff --git a/src/hashmap.c b/src/hashmap.c
new file mode 100644
index 00000000..a0a9283c
--- /dev/null
+++ b/src/hashmap.c
@@ -0,0 +1,149 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdbool.h>
+#include "types.h"
+#define TABLE_SIZE 10007  // Use a prime number for better distribution
+
+typedef struct HashNode {
+
+  uint64_t         key;
+  struct HashNode *next;
+
+} HashNode;
+
+typedef struct HashMap {
+
+  HashNode **table;
+
+} HashMap;
+
+static HashMap *_hashmap;
+
+void hashmap_reset() {
+
+  if (unlikely(!_hashmap)) {
+
+    _hashmap = (HashMap *)malloc(sizeof(HashMap));
+    _hashmap->table = (HashNode **)malloc(sizeof(HashNode *) * TABLE_SIZE);
+    memset((char *)_hashmap->table, 0, sizeof(HashNode *) * TABLE_SIZE);
+
+  } else {
+
+    for (int i = 0; i < TABLE_SIZE; i++) {
+
+      HashNode *node = _hashmap->table[i];
+      while (node) {
+
+        HashNode *temp = node;
+        node = node->next;
+        free(temp);
+
+      }
+
+    }
+
+    memset((char *)_hashmap->table, 0, sizeof(HashNode *) * TABLE_SIZE);
+
+  }
+
+}
+
+static inline unsigned int hash(uint64_t key) {
+
+  return key % TABLE_SIZE;
+
+}
+
+// type must be below 8
+bool hashmap_search_and_add(uint8_t type, uint64_t key) {
+
+  if (unlikely(type >= 8)) return false;
+  uint64_t     val = (key & 0xf8ffffffffffffff) + (type << 56);
+  unsigned int index = hash(val);
+  HashNode    *node = _hashmap->table[index];
+  while (node) {
+
+    if (node->key == val) return true;
+    node = node->next;
+
+  }
+
+  // not found so add it
+  node = (HashNode *)malloc(sizeof(HashNode));
+  node->key = val;
+  node->next = _hashmap->table[index];
+  _hashmap->table[index] = node;
+
+  return false;
+
+}
+
+// type must be below 8
+bool hashmap_search_and_add_ptr(uint8_t type, u8 *key) {
+
+  if (unlikely(type >= 8)) return false;
+  uint64_t key_t = 0;
+  memcpy(((char *)key_t) + (7 - type), key, type + 1);
+  return hashmap_search_and_add(type, key_t);
+
+}
+
+/* below is not used */
+
+void hashmap_insert(uint64_t key) {
+
+  unsigned int index = hash(key);
+  HashNode    *node = (HashNode *)malloc(sizeof(HashNode));
+  node->key = key;
+  node->next = _hashmap->table[index];
+  _hashmap->table[index] = node;
+
+}
+
+bool hashmap_search(uint64_t key) {
+
+  unsigned int index = hash(key);
+  HashNode    *node = _hashmap->table[index];
+  while (node) {
+
+    if (node->key == key) return true;
+    node = node->next;
+
+  }
+
+  return false;
+
+}
+
+void delete(uint64_t key) {
+
+  unsigned int index = hash(key);
+  HashNode    *prev = NULL, *node = _hashmap->table[index];
+  while (node) {
+
+    if (node->key == key) {
+
+      if (prev)
+        prev->next = node->next;
+      else
+        _hashmap->table[index] = node->next;
+      free(node);
+      return;
+
+    }
+
+    prev = node;
+    node = node->next;
+
+  }
+
+}
+
+void freeHashMap(HashMap *map) {
+
+  free(_hashmap->table);
+  free(map);
+
+}
+
-- 
cgit 1.4.1


From 023fc19ce04bffcbd623e27a1f2d1810c3ec0c3c Mon Sep 17 00:00:00 2001
From: Davide Quarta <quarta@qti.qualcomm.com>
Date: Mon, 5 Feb 2024 18:26:46 +0100
Subject: better replay mode error handling, added replay mode documentation,
 code formatting

---
 include/afl-fuzz.h                          |   2 +-
 include/config.h                            |  10 +-
 include/persistent_replay.h                 | 152 +++++++++++++++++++---------
 instrumentation/README.persistent_mode.md   |  30 +++++-
 instrumentation/afl-compiler-rt.o.c         |  58 ++++++-----
 src/afl-forkserver.c                        |   5 +-
 utils/persistent_mode/persistent_demo_new.c |   6 +-
 7 files changed, 182 insertions(+), 81 deletions(-)

(limited to 'include')

diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index 864bc6b6..f95dcc20 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -125,7 +125,7 @@
 #endif                                                    /* ^!SIMPLE_FILES */
 
 #ifdef AFL_PERSISTENT_RECORD
-  #define RECORD_PREFIX "RECORD:" 
+  #define RECORD_PREFIX "RECORD:"
 #endif
 
 #define STAGE_BUF_SIZE (64)  /* usable size for stage name buf in afl_state */
diff --git a/include/config.h b/include/config.h
index 1649f110..d44cda9c 100644
--- a/include/config.h
+++ b/include/config.h
@@ -83,10 +83,14 @@
    will be kept and written to the crash/ directory as RECORD:... files.
    Note that every crash will be written, not only unique ones! */
 
-//  #define AFL_PERSISTENT_RECORD
+#define AFL_PERSISTENT_RECORD
 
-/* Builds compiler-rt with support to replay persistent records */
-//  #define AFL_PERSISTENT_REPLAY
+/* Adds support in compiler-rt to replay persistent records */
+#define AFL_PERSISTENT_REPLAY
+
+/* Adds support in compiler-rt to replay persistent records in @@-style
+ * harnesses */
+//  #define AFL_PERSISTENT_REPLAY_ARGPARSE
 
 /* console output colors: There are three ways to configure its behavior
  * 1. default: colored outputs fixed on: defined USE_COLOR && defined
diff --git a/include/persistent_replay.h b/include/persistent_replay.h
index b1a55e9f..58b22fb4 100644
--- a/include/persistent_replay.h
+++ b/include/persistent_replay.h
@@ -11,71 +11,116 @@
 #include <fcntl.h>
 
 static unsigned short int is_replay_record;
-static unsigned int replay_record;
-static unsigned int replay_record_cnt;
-static char replay_record_path[PATH_MAX];
-static char **record_arg;
-static char *replay_record_dir;
-static struct dirent **record_list;
+static unsigned int       replay_record;
+static unsigned int       replay_record_cnt;
+static char               replay_record_path[PATH_MAX];
+static char              *replay_record_dir;
+static struct dirent    **record_list;
+
+#ifdef AFL_PERSISTENT_REPLAY_ARGPARSE
+static char **record_arg = NULL;
+#endif  // AFL_PERSISTENT_REPLAY_ARGPARSE
 
 static int select_files(const struct dirent *dirbuf) {
 
   char fn[4096];
 
-  if (dirbuf->d_name[0] == '.'){
+  if (dirbuf->d_name[0] == '.') {
+
     return 0;
+
   } else {
+
     snprintf(fn, sizeof(fn), "RECORD:%06u", replay_record);
     return !!strstr(dirbuf->d_name, fn);
+
   }
+
 }
-  
+
 static int compare_files(const struct dirent **da, const struct dirent **db) {
-  
-  unsigned int c1=0, c2=0;
+
+  unsigned int c1 = 0, c2 = 0;
 
   sscanf((*da)->d_name, "RECORD:%*u,cnt:%06u", &c1);
   sscanf((*db)->d_name, "RECORD:%*u,cnt:%06u", &c2);
 
-  return c1-c2;
+  return c1 - c2;
+
 }
 
-__attribute__((destructor)) static void __afl_record_replay_destroy(void){
-  for (int i=0; i < replay_record_cnt; i++) {
+__attribute__((destructor)) static void __afl_record_replay_destroy(void) {
+
+  for (int i = 0; i < replay_record_cnt; i++) {
+
     free(record_list[i]);
+
   }
+
   free(record_list);
+
 }
 
-__attribute__((constructor)) static void __afl_record_replay_init(int argc, char **argv) {
-  
+__attribute__((constructor)) static void __afl_record_replay_init(
+#ifdef AFL_PERSISTENT_REPLAY_ARGPARSE
+    int argc, char **argv
+#endif  // AFL_PERSISTENT_REPLAY_ARGPARSE
+) {
+
+#ifdef AFL_PERSISTENT_REPLAY_ARGPARSE
   char **argp;
+#endif  // AFL_PERSISTENT_REPLAY_ARGPARSE
+
+  struct stat sb;
+
+  /* caveat: if harness uses @@ and we don't pass it, it will regardless loop
+   * the number of iterations defined for AFL_LOOP (on the same file)*/
+  if (!(is_replay_record = !!getenv("AFL_PERSISTENT_REPLAY"))) {
 
-  /* caveat: if harness uses @@ and we don't pass it, it will regardless loop the number of iterations defined for AFL_LOOP (on the same file)*/
-  if(!(is_replay_record = !!getenv("AFL_PERSISTENT_REPLAY"))){
     // printf("[warning] AFL_PERSISTENT_REPLAY not set.\n");
     return;
+
   }
 
   replay_record = atoi(getenv("AFL_PERSISTENT_REPLAY"));
   replay_record_dir = getenv("AFL_PERSISTENT_DIR");
-  replay_record_cnt = scandir(replay_record_dir ? replay_record_dir : "./", &record_list, select_files, compare_files);  
 
-  if (!replay_record_cnt){
-    printf("[error] Can't find the requested record!\n");
+  if (!(stat(replay_record_dir, &sb) == 0 && S_ISDIR(sb.st_mode))) {
+
+    fprintf(stderr, "[error] Can't find the requested record directory!\n");
     is_replay_record = 0;
+    return;
+
   }
 
+  replay_record_cnt = scandir(replay_record_dir ? replay_record_dir : "./",
+                              &record_list, select_files, compare_files);
+
+  if (!replay_record_cnt) {
+
+    fprintf(stderr, "[error] Can't find the requested record!\n");
+    is_replay_record = 0;
+
+  }
+
+#ifdef AFL_PERSISTENT_REPLAY_ARGPARSE
   argp = argv;
-  while (*argp){
-    if (!strcmp(*argp, "@@")){
+  while (*argp) {
+
+    if (!strcmp(*argp, "@@")) {
+
       record_arg = argp;
       *record_arg = replay_record_path;
       break;
+
     }
+
     ++argp;
+
   }
 
+#endif  // AFL_PERSISTENT_REPLAY_ARGPARSE
+
 }
 
 /* only used if explictly included for compatibility
@@ -83,67 +128,80 @@ __attribute__((constructor)) static void __afl_record_replay_init(int argc, char
 
 #ifdef AFL_COMPAT
 
-#ifndef PATH_MAX
-  #define PATH_MAX 4096
-#endif
+  #ifndef PATH_MAX
+    #define PATH_MAX 4096
+  #endif
 
-#define FUZZ_BUF_SIZE 1024000
+  #define FUZZ_BUF_SIZE 1024000
 
-// extern ssize_t read(int fildes, void *buf, size_t nbyte);
+  // extern ssize_t read(int fildes, void *buf, size_t nbyte);
 
-//extern int __afl_persistent_loop(unsigned int max_cnt);
-//extern unsigned char fuzz_buf[];
+  // extern int __afl_persistent_loop(unsigned int max_cnt);
+  // extern unsigned char fuzz_buf[];
 
-#ifndef __AFL_HAVE_MANUAL_CONTROL
-  #define __AFL_HAVE_MANUAL_CONTROL
-#endif
+  #ifndef __AFL_HAVE_MANUAL_CONTROL
+    #define __AFL_HAVE_MANUAL_CONTROL
+  #endif
 
-#define __AFL_FUZZ_TESTCASE_LEN (read(0, fuzz_buf, FUZZ_BUF_SIZE))
-#define __AFL_FUZZ_TESTCASE_BUF fuzz_buf
-#define __AFL_FUZZ_INIT() void sync(void); 
-#define __AFL_INIT() sync() 
-#define __AFL_LOOP(x) __afl_persistent_loop(x)
+  #define __AFL_FUZZ_TESTCASE_LEN (read(0, fuzz_buf, FUZZ_BUF_SIZE))
+  #define __AFL_FUZZ_TESTCASE_BUF fuzz_buf
+  #define __AFL_FUZZ_INIT() void sync(void);
+  #define __AFL_INIT() sync()
+  #define __AFL_LOOP(x) __afl_persistent_loop(x)
 
 unsigned char fuzz_buf[FUZZ_BUF_SIZE];
 
 int __afl_persistent_loop(unsigned int max_cnt) {
 
-  static unsigned int cycle_cnt = 1;
+  static unsigned int       cycle_cnt = 1;
   static unsigned short int inited = 0;
-  char tcase[PATH_MAX];
+  char                      tcase[PATH_MAX];
+
+  if (is_replay_record) {
 
-  if( is_replay_record ){
+    if (!inited) {
 
-    if (!inited){
       cycle_cnt = replay_record_cnt;
       inited = 1;
+
     }
 
     snprintf(tcase, PATH_MAX, "%s/%s",
-                  replay_record_dir ? replay_record_dir : "./",
-                  record_list[replay_record_cnt-cycle_cnt]->d_name);
-
+             replay_record_dir ? replay_record_dir : "./",
+             record_list[replay_record_cnt - cycle_cnt]->d_name);
 
+  #ifdef AFL_PERSISTENT_REPLAY_ARGPARSE
     if (record_arg) {
+
       *record_arg = tcase;
-    } else {
+
+    } else
+
+  #endif  // AFL_PERSISTENT_REPLAY_ARGPARSE
+    {
+
       int fd = open(tcase, O_RDONLY);
       dup2(fd, 0);
       close(fd);
+
     }
 
   } else {
 
-    if (!inited){
+    if (!inited) {
+
       cycle_cnt = max_cnt;
       inited = 1;
+
     }
 
   }
 
   return cycle_cnt--;
+
 }
 
 #endif  // AFL_COMPAT
 
-#endif  // _HAVE_PERSISTENT_REPLAY_H
\ No newline at end of file
+#endif  // _HAVE_PERSISTENT_REPLAY_H
+
diff --git a/instrumentation/README.persistent_mode.md b/instrumentation/README.persistent_mode.md
index 14e59f4a..b5d982b0 100644
--- a/instrumentation/README.persistent_mode.md
+++ b/instrumentation/README.persistent_mode.md
@@ -195,4 +195,32 @@ Then as first line after the `__AFL_LOOP` while loop:
   int len = __AFL_FUZZ_TESTCASE_LEN;
 ```
 
-And that is all!
\ No newline at end of file
+And that is all!
+
+## 6) Persistent record, and replay
+
+If your software under test requires keeping a state between persistent loop iterations (i.e., a stateful network stack), you can use the `AFL_PERSISTENT_RECORD` variable as described in the [environment variables documentation](../docs/env_variables.md).
+
+To easily replay a crashing, or hanging record, you can use the persistent replay functionality by compiling AFL++ after uncommenting the `AFL_PERSISTENT_REPLAY` define  in [config.h](../include/config.h).
+
+You can then run the test binary specifying the record number via the AFL_PERSISTENT_REPLAY environment variable (i.e., `RECORD:XXXXX`` -> `AFL_PERSISTENT_REPLAY=XXXXX`).
+The directory where the record files live can be specified via the `AFL_PERSISTENT_DIR` environment varilable, otherwise by default it will be considered the current directory (`./`).
+
+If your harness reads the input files from arguments using the special `@@` argument you will need to define `AFL_PERSISTENT_ARGPARSE` in  `config.h`, or before including the `persistent_replay.h` header file as show before.
+In order to offer transparent support to harnesses using the `@@` command line argument, arguments are parsed by the `__afl_record_replay_init` init function. Since not all systems support passing arguments to initializers, this functionality is disabled by default, it's recommendable to use the `__AFL_FUZZ_TESTCASE_BUF/__AFL_FUZZ_TESTCASE_LEN` shared memory mechanism instead.
+
+### 7) Drop in replay functionality
+
+To use the replay functionality without having to use `afl-cc` you can just define `AFL_COMPAT` and include the [include/persistent_replay.h](../include/persistent_replay.h) self contained header file that provides a drop-in replacement for the persistent loop mechanism.
+
+```c
+#ifndef __AFL_FUZZ_TESTCASE_LEN
+  #define AFL_COMPAT
+  // #define AFL_PERSISTENT_REPLAY_ARGPARSE
+  #include "persistent_replay.h"
+#endif
+
+__AFL_FUZZ_INIT();
+```
+
+A simple example is provided in [persistent_demo_new.c](../utils/persistent_mode/persistent_demo_new.c).
\ No newline at end of file
diff --git a/instrumentation/afl-compiler-rt.o.c b/instrumentation/afl-compiler-rt.o.c
index 0fa22aee..037caaf0 100644
--- a/instrumentation/afl-compiler-rt.o.c
+++ b/instrumentation/afl-compiler-rt.o.c
@@ -84,7 +84,7 @@
 #include <fcntl.h>
 
 #ifdef AFL_PERSISTENT_REPLAY
-#include "persistent_replay.h"
+  #include "persistent_replay.h"
 #endif
 
 /* Globals needed by the injected instrumentation. The __afl_area_initial region
@@ -1344,37 +1344,49 @@ int __afl_persistent_loop(unsigned int max_cnt) {
 
 #ifdef AFL_PERSISTENT_REPLAY
 
-#ifndef PATH_MAX
-  #define PATH_MAX 4096
-#endif
+  #ifndef PATH_MAX
+    #define PATH_MAX 4096
+  #endif
 
-  static u8  inited = 0;
-  char tcase[PATH_MAX];
+  static u8 inited = 0;
+  char      tcase[PATH_MAX];
 
-  if( unlikely(is_replay_record) ){
+  if (unlikely(is_replay_record)) {
 
-      if (!inited){
-        cycle_cnt = replay_record_cnt;
-        inited = 1;
-      }
+    if (!inited) {
 
-      snprintf(tcase, PATH_MAX, "%s/%s",
-                  replay_record_dir ? replay_record_dir : "./",
-                  record_list[replay_record_cnt-cycle_cnt]->d_name);
+      cycle_cnt = replay_record_cnt;
+      inited = 1;
+
+    }
+
+    snprintf(tcase, PATH_MAX, "%s/%s",
+             replay_record_dir ? replay_record_dir : "./",
+             record_list[replay_record_cnt - cycle_cnt]->d_name);
+
+  #ifdef AFL_PERSISTENT_REPLAY_ARGPARSE
+    if (record_arg) {
+
+      *record_arg = tcase;
+
+    } else
+
+  #endif  // AFL_PERSISTENT_REPLAY_ARGPARSE
+    {
+
+      int fd = open(tcase, O_RDONLY);
+      dup2(fd, 0);
+      close(fd);
+
+    }
 
-      if (record_arg) {
-        *record_arg = tcase;
-      } else {
-        int fd = open(tcase, O_RDONLY);
-        dup2(fd, 0);
-        close(fd);
-      }
     return cycle_cnt--;
+
   } else
 
-#endif  
+#endif
 
-  if (first_pass) {
+      if (first_pass) {
 
     /* Make sure that every iteration of __AFL_LOOP() starts with a clean slate.
        On subsequent calls, the parent will take care of that, but on the first
diff --git a/src/afl-forkserver.c b/src/afl-forkserver.c
index f8dd783f..36e46444 100644
--- a/src/afl-forkserver.c
+++ b/src/afl-forkserver.c
@@ -1593,7 +1593,7 @@ afl_fsrv_run_target(afl_forkserver_t *fsrv, u32 timeout,
 
 #ifdef AFL_PERSISTENT_RECORD
   fsrv_run_result_t retval = FSRV_RUN_OK;
-  char *persistent_out_fmt;
+  char             *persistent_out_fmt;
 #endif
 
 #ifdef __linux__
@@ -1803,6 +1803,7 @@ afl_fsrv_run_target(afl_forkserver_t *fsrv, u32 timeout,
     persistent_out_fmt = "%s/hangs/RECORD:%06u,cnt:%06u";
     goto store_persistent_record;
 #endif
+
   }
 
   /* Did we crash?
@@ -1841,7 +1842,7 @@ afl_fsrv_run_target(afl_forkserver_t *fsrv, u32 timeout,
 
 #ifdef AFL_PERSISTENT_RECORD
 store_persistent_record:
-  if (unlikely(retval == FSRV_RUN_CRASH || retval == FSRV_RUN_TMOUT)  &&
+  if (unlikely(retval == FSRV_RUN_CRASH || retval == FSRV_RUN_TMOUT) &&
       unlikely(fsrv->persistent_record)) {
 
     char fn[PATH_MAX];
diff --git a/utils/persistent_mode/persistent_demo_new.c b/utils/persistent_mode/persistent_demo_new.c
index 40ada9e1..3d9d90a6 100644
--- a/utils/persistent_mode/persistent_demo_new.c
+++ b/utils/persistent_mode/persistent_demo_new.c
@@ -31,8 +31,8 @@
 
 /* this lets the source compile without afl-clang-fast/lto */
 #ifndef __AFL_FUZZ_TESTCASE_LEN
-#define AFL_COMPAT
-#include "persistent_replay.h"
+  #define AFL_COMPAT
+  #include "persistent_replay.h"
 #endif
 
 __AFL_FUZZ_INIT();
@@ -86,8 +86,6 @@ int main(int argc, char **argv) {
               if (buf[5] == '!') {
 
                 printf("six\n");
-                char *nullo = NULL+1;
-                *nullo = 'p';
                 abort();
 
               }
-- 
cgit 1.4.1


From e405e721fad46e594b633147a6940cfdd602e4c2 Mon Sep 17 00:00:00 2001
From: Davide Quarta <quarta@qti.qualcomm.com>
Date: Tue, 6 Feb 2024 18:19:52 +0100
Subject: reuse first_pass aux var in persistent loop for record replay mode,
 keep area ptr and loc logic intact in record replay mode, move replay record
 example to own dir in utils, update docs, move record compat layer to
 separate header file

---
 include/afl-persistent-replay.h              | 131 +++++++++++++++++
 include/afl-record-compat.h                  |  67 +++++++++
 include/config.h                             |   6 +-
 include/persistent_replay.h                  | 207 ---------------------------
 instrumentation/README.persistent_mode.md    |  20 +--
 instrumentation/afl-compiler-rt.o.c          |  87 +++++------
 utils/persistent_mode/Makefile               |   1 -
 utils/replay_record/Makefile                 |   8 ++
 utils/replay_record/persistent_demo_replay.c | 148 +++++++++++++++++++
 9 files changed, 413 insertions(+), 262 deletions(-)
 create mode 100644 include/afl-persistent-replay.h
 create mode 100644 include/afl-record-compat.h
 delete mode 100644 include/persistent_replay.h
 create mode 100644 utils/replay_record/Makefile
 create mode 100644 utils/replay_record/persistent_demo_replay.c

(limited to 'include')

diff --git a/include/afl-persistent-replay.h b/include/afl-persistent-replay.h
new file mode 100644
index 00000000..9e60ff9c
--- /dev/null
+++ b/include/afl-persistent-replay.h
@@ -0,0 +1,131 @@
+#ifndef _HAVE_PERSISTENT_REPLAY_H
+#define _HAVE_PERSISTENT_REPLAY_H
+
+#include <dirent.h>
+#include <string.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <malloc.h>
+#include <unistd.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+
+#ifndef PATH_MAX
+  #define PATH_MAX 4096
+#endif
+
+static unsigned short int is_replay_record;
+static unsigned int       replay_record;
+static unsigned int       replay_record_cnt;
+static char               replay_record_path[PATH_MAX];
+static char              *replay_record_dir;
+static struct dirent    **record_list;
+
+#ifdef AFL_PERSISTENT_REPLAY_ARGPARSE
+static char **record_arg = NULL;
+#endif  // AFL_PERSISTENT_REPLAY_ARGPARSE
+
+static int select_files(const struct dirent *dirbuf) {
+
+  char fn[PATH_MAX];
+
+  if (dirbuf->d_name[0] == '.') {
+
+    return 0;
+
+  } else {
+
+    snprintf(fn, sizeof(fn), "RECORD:%06u", replay_record);
+    return !!strstr(dirbuf->d_name, fn);
+
+  }
+
+}
+
+static int compare_files(const struct dirent **da, const struct dirent **db) {
+
+  unsigned int c1 = 0, c2 = 0;
+
+  sscanf((*da)->d_name, "RECORD:%*u,cnt:%06u", &c1);
+  sscanf((*db)->d_name, "RECORD:%*u,cnt:%06u", &c2);
+
+  return c1 - c2;
+
+}
+
+__attribute__((destructor)) static void __afl_record_replay_destroy(void) {
+
+  for (int i = 0; i < replay_record_cnt; i++) {
+
+    free(record_list[i]);
+
+  }
+
+  free(record_list);
+
+}
+
+__attribute__((constructor)) static void __afl_record_replay_init(
+#ifdef AFL_PERSISTENT_REPLAY_ARGPARSE
+    int argc, char **argv
+#endif  // AFL_PERSISTENT_REPLAY_ARGPARSE
+) {
+
+#ifdef AFL_PERSISTENT_REPLAY_ARGPARSE
+  char **argp;
+#endif  // AFL_PERSISTENT_REPLAY_ARGPARSE
+
+  struct stat sb;
+
+  /* caveat: if harness uses @@ and we don't pass it, it will regardless loop
+   * the number of iterations defined for AFL_LOOP (on the same file)*/
+  if (!(is_replay_record = !!getenv("AFL_PERSISTENT_REPLAY"))) {
+
+    // printf("[warning] AFL_PERSISTENT_REPLAY not set.\n");
+    return;
+
+  }
+
+  replay_record = atoi(getenv("AFL_PERSISTENT_REPLAY"));
+  replay_record_dir = getenv("AFL_PERSISTENT_DIR");
+
+  if (!(stat(replay_record_dir, &sb) == 0 && S_ISDIR(sb.st_mode))) {
+
+    fprintf(stderr, "[error] Can't find the requested record directory!\n");
+    is_replay_record = 0;
+    return;
+
+  }
+
+  replay_record_cnt = scandir(replay_record_dir ? replay_record_dir : "./",
+                              &record_list, select_files, compare_files);
+
+  if (!replay_record_cnt) {
+
+    fprintf(stderr, "[error] Can't find the requested record!\n");
+    is_replay_record = 0;
+
+  }
+
+#ifdef AFL_PERSISTENT_REPLAY_ARGPARSE
+  argp = argv;
+  while (*argp) {
+
+    if (!strcmp(*argp, "@@")) {
+
+      record_arg = argp;
+      *record_arg = replay_record_path;
+      break;
+
+    }
+
+    ++argp;
+
+  }
+
+#endif  // AFL_PERSISTENT_REPLAY_ARGPARSE
+
+}
+
+#endif  // _HAVE_PERSISTENT_REPLAY_H
+
diff --git a/include/afl-record-compat.h b/include/afl-record-compat.h
new file mode 100644
index 00000000..2c79595d
--- /dev/null
+++ b/include/afl-record-compat.h
@@ -0,0 +1,67 @@
+#ifndef _HAVE_AFL_COMPAT_H
+#define _HAVE_AFL_COMPAT_H
+
+#include <afl-persistent-replay.h>
+
+#define FUZZ_BUF_SIZE 1024000
+
+// extern ssize_t read(int fildes, void *buf, size_t nbyte);
+
+// extern int __afl_persistent_loop(unsigned int max_cnt);
+// extern unsigned char fuzz_buf[];
+
+#ifndef __AFL_HAVE_MANUAL_CONTROL
+  #define __AFL_HAVE_MANUAL_CONTROL
+#endif
+
+#define __AFL_FUZZ_TESTCASE_LEN (read(0, fuzz_buf, FUZZ_BUF_SIZE))
+#define __AFL_FUZZ_TESTCASE_BUF fuzz_buf
+#define __AFL_FUZZ_INIT() void sync(void);
+#define __AFL_INIT() sync()
+#define __AFL_LOOP(x) __afl_persistent_loop(x)
+
+unsigned char fuzz_buf[FUZZ_BUF_SIZE];
+
+int __afl_persistent_loop(unsigned int max_cnt) {
+
+  static unsigned int       cycle_cnt = 1;
+  static unsigned short int inited = 0;
+  char                      tcase[PATH_MAX];
+
+  if (is_replay_record) {
+
+    if (!inited) {
+
+      cycle_cnt = replay_record_cnt;
+      inited = 1;
+
+    }
+
+    snprintf(tcase, PATH_MAX, "%s/%s",
+             replay_record_dir ? replay_record_dir : "./",
+             record_list[replay_record_cnt - cycle_cnt]->d_name);
+
+#ifdef AFL_PERSISTENT_REPLAY_ARGPARSE
+    if (record_arg) {
+
+      *record_arg = tcase;
+
+    } else
+
+#endif  // AFL_PERSISTENT_REPLAY_ARGPARSE
+    {
+
+      int fd = open(tcase, O_RDONLY);
+      dup2(fd, 0);
+      close(fd);
+
+    }
+
+  }
+
+  return --cycle_cnt;
+
+}
+
+#endif  // _HAVE_AFL_COMPAT_H
+
diff --git a/include/config.h b/include/config.h
index d44cda9c..a5b6eba1 100644
--- a/include/config.h
+++ b/include/config.h
@@ -83,13 +83,11 @@
    will be kept and written to the crash/ directory as RECORD:... files.
    Note that every crash will be written, not only unique ones! */
 
-#define AFL_PERSISTENT_RECORD
-
-/* Adds support in compiler-rt to replay persistent records */
-#define AFL_PERSISTENT_REPLAY
+// #define AFL_PERSISTENT_RECORD
 
 /* Adds support in compiler-rt to replay persistent records in @@-style
  * harnesses */
+
 //  #define AFL_PERSISTENT_REPLAY_ARGPARSE
 
 /* console output colors: There are three ways to configure its behavior
diff --git a/include/persistent_replay.h b/include/persistent_replay.h
deleted file mode 100644
index 58b22fb4..00000000
--- a/include/persistent_replay.h
+++ /dev/null
@@ -1,207 +0,0 @@
-#ifndef _HAVE_PERSISTENT_REPLAY_H
-#define _HAVE_PERSISTENT_REPLAY_H
-
-#include <dirent.h>
-#include <string.h>
-#include <stdio.h>
-#include <stdlib.h>
-#include <malloc.h>
-#include <unistd.h>
-#include <sys/stat.h>
-#include <fcntl.h>
-
-static unsigned short int is_replay_record;
-static unsigned int       replay_record;
-static unsigned int       replay_record_cnt;
-static char               replay_record_path[PATH_MAX];
-static char              *replay_record_dir;
-static struct dirent    **record_list;
-
-#ifdef AFL_PERSISTENT_REPLAY_ARGPARSE
-static char **record_arg = NULL;
-#endif  // AFL_PERSISTENT_REPLAY_ARGPARSE
-
-static int select_files(const struct dirent *dirbuf) {
-
-  char fn[4096];
-
-  if (dirbuf->d_name[0] == '.') {
-
-    return 0;
-
-  } else {
-
-    snprintf(fn, sizeof(fn), "RECORD:%06u", replay_record);
-    return !!strstr(dirbuf->d_name, fn);
-
-  }
-
-}
-
-static int compare_files(const struct dirent **da, const struct dirent **db) {
-
-  unsigned int c1 = 0, c2 = 0;
-
-  sscanf((*da)->d_name, "RECORD:%*u,cnt:%06u", &c1);
-  sscanf((*db)->d_name, "RECORD:%*u,cnt:%06u", &c2);
-
-  return c1 - c2;
-
-}
-
-__attribute__((destructor)) static void __afl_record_replay_destroy(void) {
-
-  for (int i = 0; i < replay_record_cnt; i++) {
-
-    free(record_list[i]);
-
-  }
-
-  free(record_list);
-
-}
-
-__attribute__((constructor)) static void __afl_record_replay_init(
-#ifdef AFL_PERSISTENT_REPLAY_ARGPARSE
-    int argc, char **argv
-#endif  // AFL_PERSISTENT_REPLAY_ARGPARSE
-) {
-
-#ifdef AFL_PERSISTENT_REPLAY_ARGPARSE
-  char **argp;
-#endif  // AFL_PERSISTENT_REPLAY_ARGPARSE
-
-  struct stat sb;
-
-  /* caveat: if harness uses @@ and we don't pass it, it will regardless loop
-   * the number of iterations defined for AFL_LOOP (on the same file)*/
-  if (!(is_replay_record = !!getenv("AFL_PERSISTENT_REPLAY"))) {
-
-    // printf("[warning] AFL_PERSISTENT_REPLAY not set.\n");
-    return;
-
-  }
-
-  replay_record = atoi(getenv("AFL_PERSISTENT_REPLAY"));
-  replay_record_dir = getenv("AFL_PERSISTENT_DIR");
-
-  if (!(stat(replay_record_dir, &sb) == 0 && S_ISDIR(sb.st_mode))) {
-
-    fprintf(stderr, "[error] Can't find the requested record directory!\n");
-    is_replay_record = 0;
-    return;
-
-  }
-
-  replay_record_cnt = scandir(replay_record_dir ? replay_record_dir : "./",
-                              &record_list, select_files, compare_files);
-
-  if (!replay_record_cnt) {
-
-    fprintf(stderr, "[error] Can't find the requested record!\n");
-    is_replay_record = 0;
-
-  }
-
-#ifdef AFL_PERSISTENT_REPLAY_ARGPARSE
-  argp = argv;
-  while (*argp) {
-
-    if (!strcmp(*argp, "@@")) {
-
-      record_arg = argp;
-      *record_arg = replay_record_path;
-      break;
-
-    }
-
-    ++argp;
-
-  }
-
-#endif  // AFL_PERSISTENT_REPLAY_ARGPARSE
-
-}
-
-/* only used if explictly included for compatibility
-   compiling without afl-cc */
-
-#ifdef AFL_COMPAT
-
-  #ifndef PATH_MAX
-    #define PATH_MAX 4096
-  #endif
-
-  #define FUZZ_BUF_SIZE 1024000
-
-  // extern ssize_t read(int fildes, void *buf, size_t nbyte);
-
-  // extern int __afl_persistent_loop(unsigned int max_cnt);
-  // extern unsigned char fuzz_buf[];
-
-  #ifndef __AFL_HAVE_MANUAL_CONTROL
-    #define __AFL_HAVE_MANUAL_CONTROL
-  #endif
-
-  #define __AFL_FUZZ_TESTCASE_LEN (read(0, fuzz_buf, FUZZ_BUF_SIZE))
-  #define __AFL_FUZZ_TESTCASE_BUF fuzz_buf
-  #define __AFL_FUZZ_INIT() void sync(void);
-  #define __AFL_INIT() sync()
-  #define __AFL_LOOP(x) __afl_persistent_loop(x)
-
-unsigned char fuzz_buf[FUZZ_BUF_SIZE];
-
-int __afl_persistent_loop(unsigned int max_cnt) {
-
-  static unsigned int       cycle_cnt = 1;
-  static unsigned short int inited = 0;
-  char                      tcase[PATH_MAX];
-
-  if (is_replay_record) {
-
-    if (!inited) {
-
-      cycle_cnt = replay_record_cnt;
-      inited = 1;
-
-    }
-
-    snprintf(tcase, PATH_MAX, "%s/%s",
-             replay_record_dir ? replay_record_dir : "./",
-             record_list[replay_record_cnt - cycle_cnt]->d_name);
-
-  #ifdef AFL_PERSISTENT_REPLAY_ARGPARSE
-    if (record_arg) {
-
-      *record_arg = tcase;
-
-    } else
-
-  #endif  // AFL_PERSISTENT_REPLAY_ARGPARSE
-    {
-
-      int fd = open(tcase, O_RDONLY);
-      dup2(fd, 0);
-      close(fd);
-
-    }
-
-  } else {
-
-    if (!inited) {
-
-      cycle_cnt = max_cnt;
-      inited = 1;
-
-    }
-
-  }
-
-  return cycle_cnt--;
-
-}
-
-#endif  // AFL_COMPAT
-
-#endif  // _HAVE_PERSISTENT_REPLAY_H
-
diff --git a/instrumentation/README.persistent_mode.md b/instrumentation/README.persistent_mode.md
index b5d982b0..8e4f6ae4 100644
--- a/instrumentation/README.persistent_mode.md
+++ b/instrumentation/README.persistent_mode.md
@@ -201,26 +201,28 @@ And that is all!
 
 If your software under test requires keeping a state between persistent loop iterations (i.e., a stateful network stack), you can use the `AFL_PERSISTENT_RECORD` variable as described in the [environment variables documentation](../docs/env_variables.md).
 
-To easily replay a crashing, or hanging record, you can use the persistent replay functionality by compiling AFL++ after uncommenting the `AFL_PERSISTENT_REPLAY` define  in [config.h](../include/config.h).
-
-You can then run the test binary specifying the record number via the AFL_PERSISTENT_REPLAY environment variable (i.e., `RECORD:XXXXX`` -> `AFL_PERSISTENT_REPLAY=XXXXX`).
+When `AFL_PERSISTENT_RECORD` is enabled, replay functionality is also included in the compiler-rt library. To replay a specific record, assign the record number to the AFL_PERSISTENT_REPLAY environment variable (i.e., `RECORD:XXXXX`` -> `AFL_PERSISTENT_REPLAY=XXXXX`), and run the test binary as you would normally do.
 The directory where the record files live can be specified via the `AFL_PERSISTENT_DIR` environment varilable, otherwise by default it will be considered the current directory (`./`).
 
-If your harness reads the input files from arguments using the special `@@` argument you will need to define `AFL_PERSISTENT_ARGPARSE` in  `config.h`, or before including the `persistent_replay.h` header file as show before.
+If your harness reads the input files from arguments using the special `@@` argument you will need to include support by enabling `AFL_PERSISTENT_ARGPARSE` in  `config.h`.
+
 In order to offer transparent support to harnesses using the `@@` command line argument, arguments are parsed by the `__afl_record_replay_init` init function. Since not all systems support passing arguments to initializers, this functionality is disabled by default, it's recommendable to use the `__AFL_FUZZ_TESTCASE_BUF/__AFL_FUZZ_TESTCASE_LEN` shared memory mechanism instead.
 
-### 7) Drop in replay functionality
+## 7) Drop-in persistent loop replay replacement
 
-To use the replay functionality without having to use `afl-cc` you can just define `AFL_COMPAT` and include the [include/persistent_replay.h](../include/persistent_replay.h) self contained header file that provides a drop-in replacement for the persistent loop mechanism.
+To use the replay functionality without having to use `afl-cc`, include the [include/record_compat.h](../include/afl-record_compat.h) header file. Together with the [include/afl-persistent-replay.h](../include/afl-persistent-replay.h) header included in it, `afl-record-compat.h` provides a drop-in replacement for the persistent loop mechanism.
 
 ```c
 #ifndef __AFL_FUZZ_TESTCASE_LEN
-  #define AFL_COMPAT
   // #define AFL_PERSISTENT_REPLAY_ARGPARSE
-  #include "persistent_replay.h"
+  #include "afl-record-compat.h"
 #endif
 
 __AFL_FUZZ_INIT();
 ```
 
-A simple example is provided in [persistent_demo_new.c](../utils/persistent_mode/persistent_demo_new.c).
\ No newline at end of file
+A simple example is provided in [persistent_demo_replay.c](../utils/replay_record/persistent_demo_replay.c).
+
+Be aware that the [afl-record-compat.h](../include/afl-record-compat.h) header should only be included in a single compilation unit, or you will end up with clobbered functions and variables.
+
+If you need a cleaner solution, you'll have to move the functions and variables defined in [include/record_compat.h](../include/afl-record-compat.h) and [include/afl-persistent-replay.h](../include/afl-persistent-replay.h) in a C file, and add the relevant declarations to a header file. After including the new header file, the compilation unit resulting from compiling the C file can then be linked with your program.
\ No newline at end of file
diff --git a/instrumentation/afl-compiler-rt.o.c b/instrumentation/afl-compiler-rt.o.c
index 037caaf0..4c5d4e79 100644
--- a/instrumentation/afl-compiler-rt.o.c
+++ b/instrumentation/afl-compiler-rt.o.c
@@ -83,8 +83,8 @@
 #include <sys/mman.h>
 #include <fcntl.h>
 
-#ifdef AFL_PERSISTENT_REPLAY
-  #include "persistent_replay.h"
+#ifdef AFL_PERSISTENT_RECORD
+  #include "afl-persistent-replay.h"
 #endif
 
 /* Globals needed by the injected instrumentation. The __afl_area_initial region
@@ -1342,68 +1342,73 @@ int __afl_persistent_loop(unsigned int max_cnt) {
   static u8  first_pass = 1;
   static u32 cycle_cnt;
 
-#ifdef AFL_PERSISTENT_REPLAY
+#ifdef AFL_PERSISTENT_RECORD
+  char tcase[PATH_MAX];
+#endif
 
-  #ifndef PATH_MAX
-    #define PATH_MAX 4096
-  #endif
+  if (first_pass) {
 
-  static u8 inited = 0;
-  char      tcase[PATH_MAX];
+    /* Make sure that every iteration of __AFL_LOOP() starts with a clean slate.
+       On subsequent calls, the parent will take care of that, but on the first
+       iteration, it's our job to erase any trace of whatever happened
+       before the loop. */
 
-  if (unlikely(is_replay_record)) {
+    memset(__afl_area_ptr, 0, __afl_map_size);
+    __afl_area_ptr[0] = 1;
+    memset(__afl_prev_loc, 0, NGRAM_SIZE_MAX * sizeof(PREV_LOC_T));
 
-    if (!inited) {
+    first_pass = 0;
+    __afl_selective_coverage_temp = 1;
+
+#ifdef AFL_PERSISTENT_RECORD
+    if (unlikely(is_replay_record)) {
 
       cycle_cnt = replay_record_cnt;
-      inited = 1;
+      goto persistent_record;
 
-    }
+    } else
 
-    snprintf(tcase, PATH_MAX, "%s/%s",
-             replay_record_dir ? replay_record_dir : "./",
-             record_list[replay_record_cnt - cycle_cnt]->d_name);
+#endif
+    {
 
-  #ifdef AFL_PERSISTENT_REPLAY_ARGPARSE
-    if (record_arg) {
+      cycle_cnt = max_cnt;
 
-      *record_arg = tcase;
+    }
 
-    } else
+    return 1;
 
-  #endif  // AFL_PERSISTENT_REPLAY_ARGPARSE
-    {
+  } else if (--cycle_cnt) {
 
-      int fd = open(tcase, O_RDONLY);
-      dup2(fd, 0);
-      close(fd);
+#ifdef AFL_PERSISTENT_RECORD
+    if (unlikely(is_replay_record)) {
 
-    }
+    persistent_record:
 
-    return cycle_cnt--;
+      snprintf(tcase, PATH_MAX, "%s/%s",
+               replay_record_dir ? replay_record_dir : "./",
+               record_list[replay_record_cnt - cycle_cnt]->d_name);
 
-  } else
+  #ifdef AFL_PERSISTENT_REPLAY_ARGPARSE
+      if (unlikely(record_arg)) {
 
-#endif
+        *record_arg = tcase;
 
-      if (first_pass) {
+      } else
 
-    /* Make sure that every iteration of __AFL_LOOP() starts with a clean slate.
-       On subsequent calls, the parent will take care of that, but on the first
-       iteration, it's our job to erase any trace of whatever happened
-       before the loop. */
+  #endif  // AFL_PERSISTENT_REPLAY_ARGPARSE
+      {
 
-    memset(__afl_area_ptr, 0, __afl_map_size);
-    __afl_area_ptr[0] = 1;
-    memset(__afl_prev_loc, 0, NGRAM_SIZE_MAX * sizeof(PREV_LOC_T));
+        int fd = open(tcase, O_RDONLY);
+        dup2(fd, 0);
+        close(fd);
 
-    cycle_cnt = max_cnt;
-    first_pass = 0;
-    __afl_selective_coverage_temp = 1;
+      }
 
-    return 1;
+      return 1;
 
-  } else if (--cycle_cnt) {
+    }
+
+#endif
 
     raise(SIGSTOP);
 
diff --git a/utils/persistent_mode/Makefile b/utils/persistent_mode/Makefile
index 64de82a7..498aa3f8 100644
--- a/utils/persistent_mode/Makefile
+++ b/utils/persistent_mode/Makefile
@@ -1,7 +1,6 @@
 all:
 	../../afl-clang-fast -o persistent_demo persistent_demo.c
 	../../afl-clang-fast -o persistent_demo_new persistent_demo_new.c
-	gcc -g -I ../../include -o persistent_demo_new_compat persistent_demo_new.c
 	AFL_DONT_OPTIMIZE=1 ../../afl-clang-fast -o test-instr test-instr.c
 
 document:
diff --git a/utils/replay_record/Makefile b/utils/replay_record/Makefile
new file mode 100644
index 00000000..0d1cba92
--- /dev/null
+++ b/utils/replay_record/Makefile
@@ -0,0 +1,8 @@
+all:
+	test `grep '//[\s\t ]*#define[\s\t ]*AFL_PERSISTENT_RECORD' ../../include/config.h | wc -l` -eq 0 || (echo "AFL_PERSISTENT_RECORD must be enabled in config.h"; exit 1)
+	../../afl-clang-fast -o persistent_demo_replay persistent_demo_replay.c
+	${CC} -I ../../include -o persistent_demo_replay_compat persistent_demo_replay.c
+	${CC} -g -I ../../include -DAFL_PERSISTENT_REPLAY_ARGPARSE -o persistent_demo_replay_argparse persistent_demo_replay.c
+
+clean:
+	rm -f persistent_demo_replay persistent_demo_replay_argparse persistent_demo_replay_compat
diff --git a/utils/replay_record/persistent_demo_replay.c b/utils/replay_record/persistent_demo_replay.c
new file mode 100644
index 00000000..6f6648f1
--- /dev/null
+++ b/utils/replay_record/persistent_demo_replay.c
@@ -0,0 +1,148 @@
+/*
+   american fuzzy lop++ - persistent mode example
+   --------------------------------------------
+
+   Originally written by Michal Zalewski
+
+   Copyright 2015 Google Inc. All rights reserved.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at:
+
+     http://www.apache.org/licenses/LICENSE-2.0
+
+   This file demonstrates the high-performance "persistent mode" that may be
+   suitable for fuzzing certain fast and well-behaved libraries, provided that
+   they are stateless or that their internal state can be easily reset
+   across runs.
+
+   To make this work, the library and this shim need to be compiled in LLVM
+   mode using afl-clang-fast (other compiler wrappers will *not* work).
+
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <signal.h>
+#include <string.h>
+#include <limits.h>
+
+#ifdef AFL_PERSISTENT_REPLAY_ARGPARSE
+  #include <sys/stat.h>
+  #include <fcntl.h>
+#endif
+
+/* this lets the source compile without afl-clang-fast/lto */
+#ifndef __AFL_FUZZ_TESTCASE_LEN
+  #include "afl-record-compat.h"
+#endif
+
+__AFL_FUZZ_INIT();
+
+/* Main entry point. */
+
+/* To ensure checks are not optimized out it is recommended to disable
+   code optimization for the fuzzer harness main() */
+#pragma clang optimize off
+#pragma GCC            optimize("O0")
+
+int main(int argc, char **argv) {
+
+  ssize_t        len;                        /* how much input did we read? */
+  unsigned char *buf;                        /* test case buffer pointer    */
+
+#ifdef AFL_PERSISTENT_REPLAY_ARGPARSE
+  int fd;
+
+  if (argc < 2) { printf("Need an input file!"); }
+#endif
+
+  /* The number passed to __AFL_LOOP() controls the maximum number of
+     iterations before the loop exits and the program is allowed to
+     terminate normally. This limits the impact of accidental memory leaks
+     and similar hiccups. */
+
+  __AFL_INIT();
+
+#ifdef AFL_PERSISTENT_REPLAY_ARGPARSE
+  buf = malloc(1000);
+#else
+  buf = __AFL_FUZZ_TESTCASE_BUF;  // this must be assigned before __AFL_LOOP!
+#endif
+
+  while (__AFL_LOOP(UINT_MAX)) {  // increase if you have good stability
+
+#ifdef AFL_PERSISTENT_REPLAY_ARGPARSE
+    fd = open(argv[1], O_RDONLY);
+    len = read(fd, buf, 1000);
+    close(fd);
+#else
+    len = __AFL_FUZZ_TESTCASE_LEN;  // do not use the macro directly in a call!
+#endif
+
+    // fprintf(stderr, "input: %zd \"%s\"\n", len, buf);
+
+    /* do we have enough data? */
+    if (len < 8) continue;
+
+    if (strcmp((char *)buf, "thisisateststring") == 0) printf("teststring\n");
+
+    if (buf[0] == 'f') {
+
+      printf("one\n");
+      if (buf[1] == 'o') {
+
+        printf("two\n");
+        if (buf[2] == 'o') {
+
+          printf("three\n");
+          if (buf[3] == '!') {
+
+            printf("four\n");
+            if (buf[4] == '!') {
+
+              printf("five\n");
+              if (buf[5] == '!') {
+
+                printf("six\n");
+                abort();
+
+              } else {
+
+                if (buf[5] == 'O') {
+
+                  // hang
+                  while (1) {
+
+                    continue;
+
+                  };
+
+                }
+
+              }
+
+            }
+
+          }
+
+        }
+
+      }
+
+    }
+
+    /*** END PLACEHOLDER CODE ***/
+
+  }
+
+  /* Once the loop is exited, terminate normally - AFL will restart the process
+     when this happens, with a clean slate when it comes to allocated memory,
+     leftover file descriptors, etc. */
+
+  return 0;
+
+}
+
-- 
cgit 1.4.1


From 038fef962c3d85fe7e37fcd8717270654f927881 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Thu, 8 Feb 2024 12:46:00 +0100
Subject: performance

---
 .gitignore                |     1 +
 GNUmakefile               |    59 +-
 docs/INSTALL.md           |     5 +-
 include/config.h          |     2 +-
 include/t1ha.h            |   719 +++
 include/t1ha0_ia32aes_b.h |   167 +
 include/t1ha_bits.h       |  1254 +++++
 include/t1ha_selfcheck.h  |    76 +
 include/xxhash.h          | 11013 ++++++++++++++++++++++++--------------------
 src/afl-fuzz.c            |     4 +
 src/afl-performance.c     |     9 +
 utils/bench/Makefile      |     8 +
 utils/bench/README.md     |     2 +
 utils/bench/hash.c        |    42 +
 14 files changed, 8215 insertions(+), 5146 deletions(-)
 create mode 100644 include/t1ha.h
 create mode 100644 include/t1ha0_ia32aes_b.h
 create mode 100644 include/t1ha_bits.h
 create mode 100644 include/t1ha_selfcheck.h
 create mode 100644 utils/bench/Makefile
 create mode 100644 utils/bench/README.md
 create mode 100644 utils/bench/hash.c

(limited to 'include')

diff --git a/.gitignore b/.gitignore
index 67feb240..8e191e29 100644
--- a/.gitignore
+++ b/.gitignore
@@ -99,6 +99,7 @@ unicorn_mode/samples/*/\.test-*
 utils/afl_network_proxy/afl-network-client
 utils/afl_network_proxy/afl-network-server
 utils/afl_proxy/afl-proxy
+utils/bench/hash
 utils/optimin/build
 utils/optimin/optimin
 utils/persistent_mode/persistent_demo
diff --git a/GNUmakefile b/GNUmakefile
index 283c57c2..64dfc37f 100644
--- a/GNUmakefile
+++ b/GNUmakefile
@@ -84,21 +84,27 @@ else
 endif
 endif
 
-#ifeq "$(shell echo 'int main() {return 0; }' | $(CC) -fno-move-loop-invariants -fdisable-tree-cunrolli -x c - -o .test 2>/dev/null && echo 1 || echo 0 ; rm -f .test )" "1"
-#	SPECIAL_PERFORMANCE += -fno-move-loop-invariants -fdisable-tree-cunrolli
-#endif
-
-#ifeq "$(shell echo 'int main() {return 0; }' | $(CC) $(CFLAGS) -Werror -x c - -march=native -o .test 2>/dev/null && echo 1 || echo 0 ; rm -f .test )" "1"
-#  ifndef SOURCE_DATE_EPOCH
-#    HAVE_MARCHNATIVE = 1
-#    CFLAGS_OPT += -march=native
-#  endif
-#endif
+ifdef PERFORMANCE
+  SPECIAL_PERFORMANCE := -D_AFL_SPECIAL_PERFORMANCE
+  ifeq "$(SYS)" "Linux"
+    ifeq "$(shell grep avx2 /proc/cpuinfo)" ""
+    else
+  	SPECIAL_PERFORMANCE += -mavx2 -D_HAVE_AVX2
+    endif
+  endif
+  ifeq "$(shell echo 'int main() {return 0; }' | $(CC) $(CFLAGS) -Werror -x c - -march=native -o .test 2>/dev/null && echo 1 || echo 0 ; rm -f .test )" "1"
+	HAVE_MARCHNATIVE = 1
+	SPECIAL_PERFORMANCE += -march=native
+  endif
+  $(info SPECIAL_PERFORMANCE=$(SPECIAL_PERFORMANCE))
+else
+  SPECIAL_PERFORMANCE :=
+endif
 
 ifneq "$(SYS)" "Darwin"
-  #ifeq "$(HAVE_MARCHNATIVE)" "1"
-  #  SPECIAL_PERFORMANCE += -march=native
-  #endif
+ #ifeq "$(HAVE_MARCHNATIVE)" "1"
+ #  SPECIAL_PERFORMANCE += -march=native
+ #endif
  #ifndef DEBUG
  #  CFLAGS_OPT += -D_FORTIFY_SOURCE=1
  #endif
@@ -389,6 +395,7 @@ help:
 	@echo
 	@echo Known build environment options:
 	@echo "=========================================="
+	@echo "PERFORMANCE - compile with performance options that make the binary not transferable to other systems. Recommended!"
 	@echo STATIC - compile AFL++ static
 	@echo "CODE_COVERAGE - compile the target for code coverage (see docs/instrumentation/README.llvm.md)"
 	@echo ASAN_BUILD - compiles AFL++ with memory sanitizer for debug purposes
@@ -453,31 +460,31 @@ afl-as: src/afl-as.c include/afl-as.h $(COMM_HDR) | test_x86
 	@ln -sf afl-as as
 
 src/afl-performance.o : $(COMM_HDR) src/afl-performance.c include/hash.h
-	$(CC) $(CFLAGS) $(CFLAGS_OPT) -Iinclude -c src/afl-performance.c -o src/afl-performance.o
+	$(CC) $(CFLAGS) $(CFLAGS_OPT) $(SPECIAL_PERFORMANCE) -Iinclude -c src/afl-performance.c -o src/afl-performance.o
 
 src/afl-common.o : $(COMM_HDR) src/afl-common.c include/common.h
-	$(CC) $(CFLAGS) $(CFLAGS_FLTO) -c src/afl-common.c -o src/afl-common.o
+	$(CC) $(CFLAGS) $(CFLAGS_FLTO) $(SPECIAL_PERFORMANCE) -c src/afl-common.c -o src/afl-common.o
 
 src/afl-forkserver.o : $(COMM_HDR) src/afl-forkserver.c include/forkserver.h
-	$(CC) $(CFLAGS) $(CFLAGS_FLTO) -c src/afl-forkserver.c -o src/afl-forkserver.o
+	$(CC) $(CFLAGS) $(CFLAGS_FLTO) $(SPECIAL_PERFORMANCE) -c src/afl-forkserver.c -o src/afl-forkserver.o
 
 src/afl-sharedmem.o : $(COMM_HDR) src/afl-sharedmem.c include/sharedmem.h
-	$(CC) $(CFLAGS) $(CFLAGS_FLTO) -c src/afl-sharedmem.c -o src/afl-sharedmem.o
+	$(CC) $(CFLAGS) $(CFLAGS_FLTO) $(SPECIAL_PERFORMANCE) -c src/afl-sharedmem.c -o src/afl-sharedmem.o
 
 afl-fuzz: $(COMM_HDR) include/afl-fuzz.h $(AFL_FUZZ_FILES) src/afl-common.o src/afl-sharedmem.o src/afl-forkserver.o src/afl-performance.o | test_x86
-	$(CC) $(CFLAGS) $(COMPILE_STATIC) $(CFLAGS_FLTO) $(AFL_FUZZ_FILES) src/afl-common.o src/afl-sharedmem.o src/afl-forkserver.o src/afl-performance.o -o $@ $(PYFLAGS) $(LDFLAGS) -lm
+	$(CC) $(CFLAGS) $(COMPILE_STATIC) $(CFLAGS_FLTO) $(SPECIAL_PERFORMANCE) $(AFL_FUZZ_FILES) src/afl-common.o src/afl-sharedmem.o src/afl-forkserver.o src/afl-performance.o -o $@ $(PYFLAGS) $(LDFLAGS) -lm
 
 afl-showmap: src/afl-showmap.c src/afl-common.o src/afl-sharedmem.o src/afl-forkserver.o src/afl-performance.o $(COMM_HDR) | test_x86
-	$(CC) $(CFLAGS) $(COMPILE_STATIC) $(CFLAGS_FLTO) src/$@.c src/afl-fuzz-mutators.c src/afl-fuzz-python.c src/afl-common.o src/afl-sharedmem.o src/afl-forkserver.o src/afl-performance.o -o $@ $(PYFLAGS) $(LDFLAGS)
+	$(CC) $(CFLAGS) $(COMPILE_STATIC) $(CFLAGS_FLTO) $(SPECIAL_PERFORMANCE) src/$@.c src/afl-fuzz-mutators.c src/afl-fuzz-python.c src/afl-common.o src/afl-sharedmem.o src/afl-forkserver.o src/afl-performance.o -o $@ $(PYFLAGS) $(LDFLAGS)
 
 afl-tmin: src/afl-tmin.c src/afl-common.o src/afl-sharedmem.o src/afl-forkserver.o src/afl-performance.o $(COMM_HDR) | test_x86
-	$(CC) $(CFLAGS) $(COMPILE_STATIC) $(CFLAGS_FLTO) src/$@.c src/afl-common.o src/afl-sharedmem.o src/afl-forkserver.o src/afl-performance.o -o $@ $(LDFLAGS)
+	$(CC) $(CFLAGS) $(COMPILE_STATIC) $(CFLAGS_FLTO) $(SPECIAL_PERFORMANCE) src/$@.c src/afl-common.o src/afl-sharedmem.o src/afl-forkserver.o src/afl-performance.o -o $@ $(LDFLAGS)
 
 afl-analyze: src/afl-analyze.c src/afl-common.o src/afl-sharedmem.o src/afl-performance.o src/afl-forkserver.o $(COMM_HDR) | test_x86
-	$(CC) $(CFLAGS) $(COMPILE_STATIC) $(CFLAGS_FLTO) src/$@.c src/afl-common.o src/afl-sharedmem.o src/afl-performance.o src/afl-forkserver.o -o $@ $(LDFLAGS)
+	$(CC) $(CFLAGS) $(COMPILE_STATIC) $(CFLAGS_FLTO) $(SPECIAL_PERFORMANCE) src/$@.c src/afl-common.o src/afl-sharedmem.o src/afl-performance.o src/afl-forkserver.o -o $@ $(LDFLAGS)
 
 afl-gotcpu: src/afl-gotcpu.c src/afl-common.o $(COMM_HDR) | test_x86
-	$(CC) $(CFLAGS) $(COMPILE_STATIC) $(CFLAGS_FLTO) src/$@.c src/afl-common.o -o $@ $(LDFLAGS)
+	$(CC) $(CFLAGS) $(COMPILE_STATIC) $(CFLAGS_FLTO) $(SPECIAL_PERFORMANCE) src/$@.c src/afl-common.o -o $@ $(LDFLAGS)
 
 .PHONY: document
 document:	afl-fuzz-document
@@ -494,17 +501,17 @@ unit_maybe_alloc: test/unittests/unit_maybe_alloc.o
 	./test/unittests/unit_maybe_alloc
 
 test/unittests/unit_hash.o : $(COMM_HDR) include/alloc-inl.h test/unittests/unit_hash.c $(AFL_FUZZ_FILES) src/afl-performance.o
-	@$(CC) $(CFLAGS) $(ASAN_CFLAGS) -c test/unittests/unit_hash.c -o test/unittests/unit_hash.o
+	@$(CC) $(CFLAGS) $(ASAN_CFLAGS) $(SPECIAL_PERFORMANCE) -c test/unittests/unit_hash.c -o test/unittests/unit_hash.o
 
 unit_hash: test/unittests/unit_hash.o src/afl-performance.o
-	@$(CC) $(CFLAGS) -Wl,--wrap=exit -Wl,--wrap=printf $^ -o test/unittests/unit_hash $(LDFLAGS) $(ASAN_LDFLAGS) -lcmocka
+	@$(CC) $(CFLAGS) $(SPECIAL_PERFORMANCE) -Wl,--wrap=exit -Wl,--wrap=printf $^ -o test/unittests/unit_hash $(LDFLAGS) $(ASAN_LDFLAGS) -lcmocka
 	./test/unittests/unit_hash
 
 test/unittests/unit_rand.o : $(COMM_HDR) include/alloc-inl.h test/unittests/unit_rand.c $(AFL_FUZZ_FILES) src/afl-performance.o
-	@$(CC) $(CFLAGS) $(ASAN_CFLAGS) -c test/unittests/unit_rand.c -o test/unittests/unit_rand.o
+	@$(CC) $(CFLAGS) $(ASAN_CFLAGS) $(SPECIAL_PERFORMANCE) -c test/unittests/unit_rand.c -o test/unittests/unit_rand.o
 
 unit_rand: test/unittests/unit_rand.o src/afl-common.o src/afl-performance.o
-	@$(CC) $(CFLAGS) $(ASAN_CFLAGS) -Wl,--wrap=exit -Wl,--wrap=printf $^ -o test/unittests/unit_rand  $(LDFLAGS) $(ASAN_LDFLAGS) -lcmocka
+	@$(CC) $(CFLAGS) $(ASAN_CFLAGS) $(SPECIAL_PERFORMANCE) -Wl,--wrap=exit -Wl,--wrap=printf $^ -o test/unittests/unit_rand  $(LDFLAGS) $(ASAN_LDFLAGS) -lcmocka
 	./test/unittests/unit_rand
 
 test/unittests/unit_list.o : $(COMM_HDR) include/list.h test/unittests/unit_list.c $(AFL_FUZZ_FILES)
diff --git a/docs/INSTALL.md b/docs/INSTALL.md
index 84bbe3ea..9f53afed 100644
--- a/docs/INSTALL.md
+++ b/docs/INSTALL.md
@@ -69,14 +69,15 @@ These build targets exist:
 
 [Unless you are on Mac OS X](https://developer.apple.com/library/archive/qa/qa1118/_index.html),
 you can also build statically linked versions of the AFL++ binaries by passing
-the `STATIC=1` argument to make:
+the `PERFORMANCE=1` argument to make:
 
 ```shell
-make STATIC=1
+make PERFORMANCE=1
 ```
 
 These build options exist:
 
+* PERFORMANCE - compile with performance options that make the binary not transferable to other systems. Recommended!
 * STATIC - compile AFL++ static
 * CODE_COVERAGE - compile the target for code coverage (see docs/instrumentation/README.llvm.md)
 * ASAN_BUILD - compiles AFL++ with memory sanitizer for debug purposes
diff --git a/include/config.h b/include/config.h
index 70ce2ae3..31d66b14 100644
--- a/include/config.h
+++ b/include/config.h
@@ -26,7 +26,7 @@
 /* Version string: */
 
 // c = release, a = volatile github dev, e = experimental branch
-#define VERSION "++4.10c"
+#define VERSION "++4.20a"
 
 /******************************************************
  *                                                    *
diff --git a/include/t1ha.h b/include/t1ha.h
new file mode 100644
index 00000000..498f0dd6
--- /dev/null
+++ b/include/t1ha.h
@@ -0,0 +1,719 @@
+/*
+ *  Copyright (c) 2016-2020 Positive Technologies, https://www.ptsecurity.com,
+ *  Fast Positive Hash.
+ *
+ *  Portions Copyright (c) 2010-2020 Leonid Yuriev <leo@yuriev.ru>,
+ *  The 1Hippeus project (t1h).
+ *
+ *  This software is provided 'as-is', without any express or implied
+ *  warranty. In no event will the authors be held liable for any damages
+ *  arising from the use of this software.
+ *
+ *  Permission is granted to anyone to use this software for any purpose,
+ *  including commercial applications, and to alter it and redistribute it
+ *  freely, subject to the following restrictions:
+ *
+ *  1. The origin of this software must not be misrepresented; you must not
+ *     claim that you wrote the original software. If you use this software
+ *     in a product, an acknowledgement in the product documentation would be
+ *     appreciated but is not required.
+ *  2. Altered source versions must be plainly marked as such, and must not be
+ *     misrepresented as being the original software.
+ *  3. This notice may not be removed or altered from any source distribution.
+ */
+
+/*
+ * t1ha = { Fast Positive Hash, aka "Позитивный Хэш" }
+ * by [Positive Technologies](https://www.ptsecurity.ru)
+ *
+ * Briefly, it is a 64-bit Hash Function:
+ *  1. Created for 64-bit little-endian platforms, in predominantly for x86_64,
+ *     but portable and without penalties it can run on any 64-bit CPU.
+ *  2. In most cases up to 15% faster than City64, xxHash, mum-hash, metro-hash
+ *     and all others portable hash-functions (which do not use specific
+ *     hardware tricks).
+ *  3. Not suitable for cryptography.
+ *
+ * The Future will (be) Positive. Всё будет хорошо.
+ *
+ * ACKNOWLEDGEMENT:
+ * The t1ha was originally developed by Leonid Yuriev (Леонид Юрьев)
+ * for The 1Hippeus project - zerocopy messaging in the spirit of Sparta!
+ */
+
+#pragma once
+
+/*****************************************************************************
+ *
+ * PLEASE PAY ATTENTION TO THE FOLLOWING NOTES
+ * about macros definitions which controls t1ha behaviour and/or performance.
+ *
+ *
+ * 1) T1HA_SYS_UNALIGNED_ACCESS = Defines the system/platform/CPU/architecture
+ *                                abilities for unaligned data access.
+ *
+ *    By default, when the T1HA_SYS_UNALIGNED_ACCESS not defined,
+ *    it will defined on the basis hardcoded knowledge about of capabilities
+ *    of most common CPU architectures. But you could override this
+ *    default behavior when build t1ha library itself:
+ *
+ *      // To disable unaligned access at all.
+ *      #define T1HA_SYS_UNALIGNED_ACCESS 0
+ *
+ *      // To enable unaligned access, but indicate that it significantly slow.
+ *      #define T1HA_SYS_UNALIGNED_ACCESS 1
+ *
+ *      // To enable unaligned access, and indicate that it effecient.
+ *      #define T1HA_SYS_UNALIGNED_ACCESS 2
+ *
+ *
+ * 2) T1HA_USE_FAST_ONESHOT_READ = Controls the data reads at the end of buffer.
+ *
+ *    When defined to non-zero, t1ha will use 'one shot' method for reading
+ *    up to 8 bytes at the end of data. In this case just the one 64-bit read
+ *    will be performed even when the available less than 8 bytes.
+ *
+ *    This is little bit faster that switching by length of data tail.
+ *    Unfortunately this will triggering a false-positive alarms from Valgrind,
+ *    AddressSanitizer and other similar tool.
+ *
+ *    By default, t1ha defines it to 1, but you could override this
+ *    default behavior when build t1ha library itself:
+ *
+ *      // For little bit faster and small code.
+ *      #define T1HA_USE_FAST_ONESHOT_READ 1
+ *
+ *      // For calmness if doubt.
+ *      #define T1HA_USE_FAST_ONESHOT_READ 0
+ *
+ *
+ * 3) T1HA0_RUNTIME_SELECT = Controls choice fastest function in runtime.
+ *
+ *    t1ha library offers the t1ha0() function as the fastest for current CPU.
+ *    But actual CPU's features/capabilities and may be significantly different,
+ *    especially on x86 platform. Therefore, internally, t1ha0() may require
+ *    dynamic dispatching for choice best implementation.
+ *
+ *    By default, t1ha enables such runtime choice and (may be) corresponding
+ *    indirect calls if it reasonable, but you could override this default
+ *    behavior when build t1ha library itself:
+ *
+ *      // To enable runtime choice of fastest implementation.
+ *      #define T1HA0_RUNTIME_SELECT 1
+ *
+ *      // To disable runtime choice of fastest implementation.
+ *      #define T1HA0_RUNTIME_SELECT 0
+ *
+ *    When T1HA0_RUNTIME_SELECT is nonzero the t1ha0_resolve() function could
+ *    be used to get actual t1ha0() implementation address at runtime. This is
+ *    useful for two cases:
+ *      - calling by local pointer-to-function usually is little
+ *        bit faster (less overhead) than via a PLT thru the DSO boundary.
+ *      - GNU Indirect functions (see below) don't supported by environment
+ *        and calling by t1ha0_funcptr is not available and/or expensive.
+ *
+ * 4) T1HA_USE_INDIRECT_FUNCTIONS = Controls usage of GNU Indirect functions.
+ *
+ *    In continue of T1HA0_RUNTIME_SELECT the T1HA_USE_INDIRECT_FUNCTIONS
+ *    controls usage of ELF indirect functions feature. In general, when
+ *    available, this reduces overhead of indirect function's calls though
+ *    a DSO-bundary (https://sourceware.org/glibc/wiki/GNU_IFUNC).
+ *
+ *    By default, t1ha engage GNU Indirect functions when it available
+ *    and useful, but you could override this default behavior when build
+ *    t1ha library itself:
+ *
+ *      // To enable use of GNU ELF Indirect functions.
+ *      #define T1HA_USE_INDIRECT_FUNCTIONS 1
+ *
+ *      // To disable use of GNU ELF Indirect functions. This may be useful
+ *      // if the actual toolchain or the system's loader don't support ones.
+ *      #define T1HA_USE_INDIRECT_FUNCTIONS 0
+ *
+ * 5) T1HA0_AESNI_AVAILABLE = Controls AES-NI detection and dispatching on x86.
+ *
+ *    In continue of T1HA0_RUNTIME_SELECT the T1HA0_AESNI_AVAILABLE controls
+ *    detection and usage of AES-NI CPU's feature. On the other hand, this
+ *    requires compiling parts of t1ha library with certain properly options,
+ *    and could be difficult or inconvenient in some cases.
+ *
+ *    By default, t1ha engade AES-NI for t1ha0() on the x86 platform, but
+ *    you could override this default behavior when build t1ha library itself:
+ *
+ *      // To disable detection and usage of AES-NI instructions for t1ha0().
+ *      // This may be useful when you unable to build t1ha library properly
+ *      // or known that AES-NI will be unavailable at the deploy.
+ *      #define T1HA0_AESNI_AVAILABLE 0
+ *
+ *      // To force detection and usage of AES-NI instructions for t1ha0(),
+ *      // but I don't known reasons to anybody would need this.
+ *      #define T1HA0_AESNI_AVAILABLE 1
+ *
+ * 6) T1HA0_DISABLED, T1HA1_DISABLED, T1HA2_DISABLED = Controls availability of
+ *    t1ha functions.
+ *
+ *    In some cases could be useful to import/use only few of t1ha functions
+ *    or just the one. So, this definitions allows disable corresponding parts
+ *    of t1ha library.
+ *
+ *      // To disable t1ha0(), t1ha0_32le(), t1ha0_32be() and all AES-NI.
+ *      #define T1HA0_DISABLED
+ *
+ *      // To disable t1ha1_le() and t1ha1_be().
+ *      #define T1HA1_DISABLED
+ *
+ *      // To disable t1ha2_atonce(), t1ha2_atonce128() and so on.
+ *      #define T1HA2_DISABLED
+ *
+ *****************************************************************************/
+
+#define T1HA_VERSION_MAJOR 2
+#define T1HA_VERSION_MINOR 1
+#define T1HA_VERSION_RELEASE 1
+
+#ifndef __has_attribute
+#define __has_attribute(x) (0)
+#endif
+
+#ifndef __has_include
+#define __has_include(x) (0)
+#endif
+
+#ifndef __GNUC_PREREQ
+#if defined(__GNUC__) && defined(__GNUC_MINOR__)
+#define __GNUC_PREREQ(maj, min)                                                \
+  ((__GNUC__ << 16) + __GNUC_MINOR__ >= ((maj) << 16) + (min))
+#else
+#define __GNUC_PREREQ(maj, min) 0
+#endif
+#endif /* __GNUC_PREREQ */
+
+#ifndef __CLANG_PREREQ
+#ifdef __clang__
+#define __CLANG_PREREQ(maj, min)                                               \
+  ((__clang_major__ << 16) + __clang_minor__ >= ((maj) << 16) + (min))
+#else
+#define __CLANG_PREREQ(maj, min) (0)
+#endif
+#endif /* __CLANG_PREREQ */
+
+#ifndef __LCC_PREREQ
+#ifdef __LCC__
+#define __LCC_PREREQ(maj, min)                                                 \
+  ((__LCC__ << 16) + __LCC_MINOR__ >= ((maj) << 16) + (min))
+#else
+#define __LCC_PREREQ(maj, min) (0)
+#endif
+#endif /* __LCC_PREREQ */
+
+/*****************************************************************************/
+
+#ifdef _MSC_VER
+/* Avoid '16' bytes padding added after data member 't1ha_context::total'
+ * and other warnings from std-headers if warning-level > 3. */
+#pragma warning(push, 3)
+#endif
+
+#if defined(__cplusplus) && __cplusplus >= 201103L
+#include <climits>
+#include <cstddef>
+#include <cstdint>
+#else
+#include <limits.h>
+#include <stddef.h>
+#include <stdint.h>
+#endif
+
+/*****************************************************************************/
+
+#if defined(i386) || defined(__386) || defined(__i386) || defined(__i386__) || \
+    defined(i486) || defined(__i486) || defined(__i486__) ||                   \
+    defined(i586) | defined(__i586) || defined(__i586__) || defined(i686) ||   \
+    defined(__i686) || defined(__i686__) || defined(_M_IX86) ||                \
+    defined(_X86_) || defined(__THW_INTEL__) || defined(__I86__) ||            \
+    defined(__INTEL__) || defined(__x86_64) || defined(__x86_64__) ||          \
+    defined(__amd64__) || defined(__amd64) || defined(_M_X64) ||               \
+    defined(_M_AMD64) || defined(__IA32__) || defined(__INTEL__)
+#ifndef __ia32__
+/* LY: define neutral __ia32__ for x86 and x86-64 archs */
+#define __ia32__ 1
+#endif /* __ia32__ */
+#if !defined(__amd64__) && (defined(__x86_64) || defined(__x86_64__) ||        \
+                            defined(__amd64) || defined(_M_X64))
+/* LY: define trusty __amd64__ for all AMD64/x86-64 arch */
+#define __amd64__ 1
+#endif /* __amd64__ */
+#endif /* all x86 */
+
+#if !defined(__BYTE_ORDER__) || !defined(__ORDER_LITTLE_ENDIAN__) ||           \
+    !defined(__ORDER_BIG_ENDIAN__)
+
+/* *INDENT-OFF* */
+/* clang-format off */
+
+#if defined(__GLIBC__) || defined(__GNU_LIBRARY__) || defined(__ANDROID__) ||  \
+    defined(HAVE_ENDIAN_H) || __has_include(<endian.h>)
+#include <endian.h>
+#elif defined(__APPLE__) || defined(__MACH__) || defined(__OpenBSD__) ||       \
+    defined(HAVE_MACHINE_ENDIAN_H) || __has_include(<machine/endian.h>)
+#include <machine/endian.h>
+#elif defined(HAVE_SYS_ISA_DEFS_H) || __has_include(<sys/isa_defs.h>)
+#include <sys/isa_defs.h>
+#elif (defined(HAVE_SYS_TYPES_H) && defined(HAVE_SYS_ENDIAN_H)) ||             \
+    (__has_include(<sys/types.h>) && __has_include(<sys/endian.h>))
+#include <sys/endian.h>
+#include <sys/types.h>
+#elif defined(__bsdi__) || defined(__DragonFly__) || defined(__FreeBSD__) ||   \
+    defined(__NETBSD__) || defined(__NetBSD__) ||                              \
+    defined(HAVE_SYS_PARAM_H) || __has_include(<sys/param.h>)
+#include <sys/param.h>
+#endif /* OS */
+
+/* *INDENT-ON* */
+/* clang-format on */
+
+#if defined(__BYTE_ORDER) && defined(__LITTLE_ENDIAN) && defined(__BIG_ENDIAN)
+#define __ORDER_LITTLE_ENDIAN__ __LITTLE_ENDIAN
+#define __ORDER_BIG_ENDIAN__ __BIG_ENDIAN
+#define __BYTE_ORDER__ __BYTE_ORDER
+#elif defined(_BYTE_ORDER) && defined(_LITTLE_ENDIAN) && defined(_BIG_ENDIAN)
+#define __ORDER_LITTLE_ENDIAN__ _LITTLE_ENDIAN
+#define __ORDER_BIG_ENDIAN__ _BIG_ENDIAN
+#define __BYTE_ORDER__ _BYTE_ORDER
+#else
+#define __ORDER_LITTLE_ENDIAN__ 1234
+#define __ORDER_BIG_ENDIAN__ 4321
+
+#if defined(__LITTLE_ENDIAN__) ||                                              \
+    (defined(_LITTLE_ENDIAN) && !defined(_BIG_ENDIAN)) ||                      \
+    defined(__ARMEL__) || defined(__THUMBEL__) || defined(__AARCH64EL__) ||    \
+    defined(__MIPSEL__) || defined(_MIPSEL) || defined(__MIPSEL) ||            \
+    defined(_M_ARM) || defined(_M_ARM64) || defined(__e2k__) ||                \
+    defined(__elbrus_4c__) || defined(__elbrus_8c__) || defined(__bfin__) ||   \
+    defined(__BFIN__) || defined(__ia64__) || defined(_IA64) ||                \
+    defined(__IA64__) || defined(__ia64) || defined(_M_IA64) ||                \
+    defined(__itanium__) || defined(__ia32__) || defined(__CYGWIN__) ||        \
+    defined(_WIN64) || defined(_WIN32) || defined(__TOS_WIN__) ||              \
+    defined(__WINDOWS__)
+#define __BYTE_ORDER__ __ORDER_LITTLE_ENDIAN__
+
+#elif defined(__BIG_ENDIAN__) ||                                               \
+    (defined(_BIG_ENDIAN) && !defined(_LITTLE_ENDIAN)) ||                      \
+    defined(__ARMEB__) || defined(__THUMBEB__) || defined(__AARCH64EB__) ||    \
+    defined(__MIPSEB__) || defined(_MIPSEB) || defined(__MIPSEB) ||            \
+    defined(__m68k__) || defined(M68000) || defined(__hppa__) ||               \
+    defined(__hppa) || defined(__HPPA__) || defined(__sparc__) ||              \
+    defined(__sparc) || defined(__370__) || defined(__THW_370__) ||            \
+    defined(__s390__) || defined(__s390x__) || defined(__SYSC_ZARCH__)
+#define __BYTE_ORDER__ __ORDER_BIG_ENDIAN__
+
+#else
+#error __BYTE_ORDER__ should be defined.
+#endif /* Arch */
+
+#endif
+#endif /* __BYTE_ORDER__ || __ORDER_LITTLE_ENDIAN__ || __ORDER_BIG_ENDIAN__ */
+
+/*****************************************************************************/
+
+#ifndef __dll_export
+#if defined(_WIN32) || defined(_WIN64) || defined(__CYGWIN__)
+#if defined(__GNUC__) || __has_attribute(dllexport)
+#define __dll_export __attribute__((dllexport))
+#else
+#define __dll_export __declspec(dllexport)
+#endif
+#elif defined(__GNUC__) || __has_attribute(__visibility__)
+#define __dll_export __attribute__((__visibility__("default")))
+#else
+#define __dll_export
+#endif
+#endif /* __dll_export */
+
+#ifndef __dll_import
+#if defined(_WIN32) || defined(_WIN64) || defined(__CYGWIN__)
+#if defined(__GNUC__) || __has_attribute(dllimport)
+#define __dll_import __attribute__((dllimport))
+#else
+#define __dll_import __declspec(dllimport)
+#endif
+#elif defined(__GNUC__) || __has_attribute(__visibility__)
+#define __dll_import __attribute__((__visibility__("default")))
+#else
+#define __dll_import
+#endif
+#endif /* __dll_import */
+
+#ifndef __force_inline
+#ifdef _MSC_VER
+#define __force_inline __forceinline
+#elif __GNUC_PREREQ(3, 2) || __has_attribute(__always_inline__)
+#define __force_inline __inline __attribute__((__always_inline__))
+#else
+#define __force_inline __inline
+#endif
+#endif /* __force_inline */
+
+#ifndef T1HA_API
+#if defined(t1ha_EXPORTS)
+#define T1HA_API __dll_export
+#elif defined(t1ha_IMPORTS)
+#define T1HA_API __dll_import
+#else
+#define T1HA_API
+#endif
+#endif /* T1HA_API */
+
+#if defined(_MSC_VER) && defined(__ia32__)
+#define T1HA_ALIGN_PREFIX __declspec(align(32)) /* required only for SIMD */
+#else
+#define T1HA_ALIGN_PREFIX
+#endif /* _MSC_VER */
+
+#if defined(__GNUC__) && defined(__ia32__)
+#define T1HA_ALIGN_SUFFIX                                                      \
+  __attribute__((__aligned__(32))) /* required only for SIMD */
+#else
+#define T1HA_ALIGN_SUFFIX
+#endif /* GCC x86 */
+
+#ifndef T1HA_USE_INDIRECT_FUNCTIONS
+/* GNU ELF indirect functions usage control. For more info please see
+ * https://en.wikipedia.org/wiki/Executable_and_Linkable_Format
+ * and https://sourceware.org/glibc/wiki/GNU_IFUNC */
+#if defined(__ELF__) && defined(__amd64__) &&                                  \
+    (__has_attribute(__ifunc__) ||                                             \
+     (!defined(__clang__) && defined(__GNUC__) && __GNUC__ >= 4 &&             \
+      !defined(__SANITIZE_ADDRESS__) && !defined(__SSP_ALL__)))
+/* Enable gnu_indirect_function by default if :
+ *  - ELF AND x86_64
+ *  - attribute(__ifunc__) is available OR
+ *    GCC >= 4 WITHOUT -fsanitize=address NOR -fstack-protector-all */
+#define T1HA_USE_INDIRECT_FUNCTIONS 1
+#else
+#define T1HA_USE_INDIRECT_FUNCTIONS 0
+#endif
+#endif /* T1HA_USE_INDIRECT_FUNCTIONS */
+
+#if __GNUC_PREREQ(4, 0)
+#pragma GCC visibility push(hidden)
+#endif /* __GNUC_PREREQ(4,0) */
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef union T1HA_ALIGN_PREFIX t1ha_state256 {
+  uint8_t bytes[32];
+  uint32_t u32[8];
+  uint64_t u64[4];
+  struct {
+    uint64_t a, b, c, d;
+  } n;
+} t1ha_state256_t T1HA_ALIGN_SUFFIX;
+
+typedef struct t1ha_context {
+  t1ha_state256_t state;
+  t1ha_state256_t buffer;
+  size_t partial;
+  uint64_t total;
+} t1ha_context_t;
+
+#ifdef _MSC_VER
+#pragma warning(pop)
+#endif
+
+/******************************************************************************
+ *
+ * Self-testing API.
+ *
+ * Unfortunately, some compilers (exactly only Microsoft Visual C/C++) has
+ * a bugs which leads t1ha-functions to produce wrong results. This API allows
+ * check the correctness of the actual code in runtime.
+ *
+ * All check-functions returns 0 on success, or -1 in case the corresponding
+ * hash-function failed verification. PLEASE, always perform such checking at
+ * initialization of your code, if you using MSVC or other troubleful compilers.
+ */
+
+T1HA_API int t1ha_selfcheck__all_enabled(void);
+
+#ifndef T1HA2_DISABLED
+T1HA_API int t1ha_selfcheck__t1ha2_atonce(void);
+T1HA_API int t1ha_selfcheck__t1ha2_atonce128(void);
+T1HA_API int t1ha_selfcheck__t1ha2_stream(void);
+T1HA_API int t1ha_selfcheck__t1ha2(void);
+#endif /* T1HA2_DISABLED */
+
+#ifndef T1HA1_DISABLED
+T1HA_API int t1ha_selfcheck__t1ha1_le(void);
+T1HA_API int t1ha_selfcheck__t1ha1_be(void);
+T1HA_API int t1ha_selfcheck__t1ha1(void);
+#endif /* T1HA1_DISABLED */
+
+#ifndef T1HA0_DISABLED
+T1HA_API int t1ha_selfcheck__t1ha0_32le(void);
+T1HA_API int t1ha_selfcheck__t1ha0_32be(void);
+T1HA_API int t1ha_selfcheck__t1ha0(void);
+
+/* Define T1HA0_AESNI_AVAILABLE to 0 for disable AES-NI support. */
+#ifndef T1HA0_AESNI_AVAILABLE
+#if defined(__e2k__) ||                                                        \
+    (defined(__ia32__) && (!defined(_M_IX86) || _MSC_VER > 1800))
+#define T1HA0_AESNI_AVAILABLE 1
+#else
+#define T1HA0_AESNI_AVAILABLE 0
+#endif
+#endif /* ifndef T1HA0_AESNI_AVAILABLE */
+
+#if T1HA0_AESNI_AVAILABLE
+T1HA_API int t1ha_selfcheck__t1ha0_ia32aes_noavx(void);
+T1HA_API int t1ha_selfcheck__t1ha0_ia32aes_avx(void);
+#ifndef __e2k__
+T1HA_API int t1ha_selfcheck__t1ha0_ia32aes_avx2(void);
+#endif
+#endif /* if T1HA0_AESNI_AVAILABLE */
+#endif /* T1HA0_DISABLED */
+
+/******************************************************************************
+ *
+ *  t1ha2 = 64 and 128-bit, SLIGHTLY MORE ATTENTION FOR QUALITY AND STRENGTH.
+ *
+ *    - The recommended version of "Fast Positive Hash" with good quality
+ *      for checksum, hash tables and fingerprinting.
+ *    - Portable and extremely efficiency on modern 64-bit CPUs.
+ *      Designed for 64-bit little-endian platforms,
+ *      in other cases will runs slowly.
+ *    - Great quality of hashing and still faster than other non-t1ha hashes.
+ *      Provides streaming mode and 128-bit result.
+ *
+ * Note: Due performance reason 64- and 128-bit results are completely
+ *       different each other, i.e. 64-bit result is NOT any part of 128-bit.
+ */
+#ifndef T1HA2_DISABLED
+
+/* The at-once variant with 64-bit result */
+T1HA_API uint64_t t1ha2_atonce(const void *data, size_t length, uint64_t seed);
+
+/* The at-once variant with 128-bit result.
+ * Argument `extra_result` is NOT optional and MUST be valid.
+ * The high 64-bit part of 128-bit hash will be always unconditionally
+ * stored to the address given by `extra_result` argument. */
+T1HA_API uint64_t t1ha2_atonce128(uint64_t *__restrict extra_result,
+                                  const void *__restrict data, size_t length,
+                                  uint64_t seed);
+
+/* The init/update/final trinity for streaming.
+ * Return 64 or 128-bit result depentently from `extra_result` argument. */
+T1HA_API void t1ha2_init(t1ha_context_t *ctx, uint64_t seed_x, uint64_t seed_y);
+T1HA_API void t1ha2_update(t1ha_context_t *__restrict ctx,
+                           const void *__restrict data, size_t length);
+
+/* Argument `extra_result` is optional and MAY be NULL.
+ *  - If `extra_result` is NOT NULL then the 128-bit hash will be calculated,
+ *    and high 64-bit part of it will be stored to the address given
+ *    by `extra_result` argument.
+ *  - Otherwise the 64-bit hash will be calculated
+ *    and returned from function directly.
+ *
+ * Note: Due performance reason 64- and 128-bit results are completely
+ *       different each other, i.e. 64-bit result is NOT any part of 128-bit. */
+T1HA_API uint64_t t1ha2_final(t1ha_context_t *__restrict ctx,
+                              uint64_t *__restrict extra_result /* optional */);
+
+#endif /* T1HA2_DISABLED */
+
+/******************************************************************************
+ *
+ *  t1ha1 = 64-bit, BASELINE FAST PORTABLE HASH:
+ *
+ *    - Runs faster on 64-bit platforms in other cases may runs slowly.
+ *    - Portable and stable, returns same 64-bit result
+ *      on all architectures and CPUs.
+ *    - Unfortunately it fails the "strict avalanche criteria",
+ *      see test results at https://github.com/demerphq/smhasher.
+ *
+ *      This flaw is insignificant for the t1ha1() purposes and imperceptible
+ *      from a practical point of view.
+ *      However, nowadays this issue has resolved in the next t1ha2(),
+ *      that was initially planned to providing a bit more quality.
+ */
+#ifndef T1HA1_DISABLED
+
+/* The little-endian variant. */
+T1HA_API uint64_t t1ha1_le(const void *data, size_t length, uint64_t seed);
+
+/* The big-endian variant. */
+T1HA_API uint64_t t1ha1_be(const void *data, size_t length, uint64_t seed);
+
+#endif /* T1HA1_DISABLED */
+
+/******************************************************************************
+ *
+ *  t1ha0 = 64-bit, JUST ONLY FASTER:
+ *
+ *    - Provides fast-as-possible hashing for current CPU, including
+ *      32-bit systems and engaging the available hardware acceleration.
+ *    - It is a facade that selects most quick-and-dirty hash
+ *      for the current processor. For instance, on IA32 (x86) actual function
+ *      will be selected in runtime, depending on current CPU capabilities
+ *
+ * BE CAREFUL!!!  THIS IS MEANS:
+ *
+ *   1. The quality of hash is a subject for tradeoffs with performance.
+ *      So, the quality and strength of t1ha0() may be lower than t1ha1(),
+ *      especially on 32-bit targets, but then much faster.
+ *      However, guaranteed that it passes all SMHasher tests.
+ *
+ *   2. No warranty that the hash result will be same for particular
+ *      key on another machine or another version of libt1ha.
+ *
+ *      Briefly, such hash-results and their derivatives, should be
+ *      used only in runtime, but should not be persist or transferred
+ *      over a network.
+ *
+ *
+ *  When T1HA0_RUNTIME_SELECT is nonzero the t1ha0_resolve() function could
+ *  be used to get actual t1ha0() implementation address at runtime. This is
+ *  useful for two cases:
+ *    - calling by local pointer-to-function usually is little
+ *      bit faster (less overhead) than via a PLT thru the DSO boundary.
+ *    - GNU Indirect functions (see below) don't supported by environment
+ *      and calling by t1ha0_funcptr is not available and/or expensive.
+ */
+
+#ifndef T1HA0_DISABLED
+
+/* The little-endian variant for 32-bit CPU. */
+uint64_t t1ha0_32le(const void *data, size_t length, uint64_t seed);
+/* The big-endian variant for 32-bit CPU. */
+uint64_t t1ha0_32be(const void *data, size_t length, uint64_t seed);
+
+/* Define T1HA0_AESNI_AVAILABLE to 0 for disable AES-NI support. */
+#ifndef T1HA0_AESNI_AVAILABLE
+#if defined(__e2k__) ||                                                        \
+    (defined(__ia32__) && (!defined(_M_IX86) || _MSC_VER > 1800))
+#define T1HA0_AESNI_AVAILABLE 1
+#else
+#define T1HA0_AESNI_AVAILABLE 0
+#endif
+#endif /* T1HA0_AESNI_AVAILABLE */
+
+/* Define T1HA0_RUNTIME_SELECT to 0 for disable dispatching t1ha0 at runtime. */
+#ifndef T1HA0_RUNTIME_SELECT
+#if T1HA0_AESNI_AVAILABLE && !defined(__e2k__)
+#define T1HA0_RUNTIME_SELECT 1
+#else
+#define T1HA0_RUNTIME_SELECT 0
+#endif
+#endif /* T1HA0_RUNTIME_SELECT */
+
+#if !T1HA0_RUNTIME_SELECT && !defined(T1HA0_USE_DEFINE)
+#if defined(__LCC__)
+#define T1HA0_USE_DEFINE 1
+#else
+#define T1HA0_USE_DEFINE 0
+#endif
+#endif /* T1HA0_USE_DEFINE */
+
+#if T1HA0_AESNI_AVAILABLE
+uint64_t t1ha0_ia32aes_noavx(const void *data, size_t length, uint64_t seed);
+uint64_t t1ha0_ia32aes_avx(const void *data, size_t length, uint64_t seed);
+#ifndef __e2k__
+uint64_t t1ha0_ia32aes_avx2(const void *data, size_t length, uint64_t seed);
+#endif
+#endif /* T1HA0_AESNI_AVAILABLE */
+
+#if T1HA0_RUNTIME_SELECT
+typedef uint64_t (*t1ha0_function_t)(const void *, size_t, uint64_t);
+T1HA_API t1ha0_function_t t1ha0_resolve(void);
+#if T1HA_USE_INDIRECT_FUNCTIONS
+T1HA_API uint64_t t1ha0(const void *data, size_t length, uint64_t seed);
+#else
+/* Otherwise function pointer will be used.
+ * Unfortunately this may cause some overhead calling. */
+T1HA_API extern uint64_t (*t1ha0_funcptr)(const void *data, size_t length,
+                                          uint64_t seed);
+static __force_inline uint64_t t1ha0(const void *data, size_t length,
+                                     uint64_t seed) {
+  return t1ha0_funcptr(data, length, seed);
+}
+#endif /* T1HA_USE_INDIRECT_FUNCTIONS */
+
+#elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+
+#if T1HA0_USE_DEFINE
+
+#if (UINTPTR_MAX > 0xffffFFFFul || ULONG_MAX > 0xffffFFFFul) &&                \
+    (!defined(T1HA1_DISABLED) || !defined(T1HA2_DISABLED))
+#if defined(T1HA1_DISABLED)
+#define t1ha0 t1ha2_atonce
+#else
+#define t1ha0 t1ha1_be
+#endif /* T1HA1_DISABLED */
+#else  /* 32/64 */
+#define t1ha0 t1ha0_32be
+#endif /* 32/64 */
+
+#else /* T1HA0_USE_DEFINE */
+
+static __force_inline uint64_t t1ha0(const void *data, size_t length,
+                                     uint64_t seed) {
+#if (UINTPTR_MAX > 0xffffFFFFul || ULONG_MAX > 0xffffFFFFul) &&                \
+    (!defined(T1HA1_DISABLED) || !defined(T1HA2_DISABLED))
+#if defined(T1HA1_DISABLED)
+  return t1ha2_atonce(data, length, seed);
+#else
+  return t1ha1_be(data, length, seed);
+#endif /* T1HA1_DISABLED */
+#else  /* 32/64 */
+  return t1ha0_32be(data, length, seed);
+#endif /* 32/64 */
+}
+
+#endif /* !T1HA0_USE_DEFINE */
+
+#else /* !T1HA0_RUNTIME_SELECT && __BYTE_ORDER__ != __ORDER_BIG_ENDIAN__ */
+
+#if T1HA0_USE_DEFINE
+
+#if (UINTPTR_MAX > 0xffffFFFFul || ULONG_MAX > 0xffffFFFFul) &&                \
+    (!defined(T1HA1_DISABLED) || !defined(T1HA2_DISABLED))
+#if defined(T1HA1_DISABLED)
+#define t1ha0 t1ha2_atonce
+#else
+#define t1ha0 t1ha1_le
+#endif /* T1HA1_DISABLED */
+#else  /* 32/64 */
+#define t1ha0 t1ha0_32le
+#endif /* 32/64 */
+
+#else
+
+static __force_inline uint64_t t1ha0(const void *data, size_t length,
+                                     uint64_t seed) {
+#if (UINTPTR_MAX > 0xffffFFFFul || ULONG_MAX > 0xffffFFFFul) &&                \
+    (!defined(T1HA1_DISABLED) || !defined(T1HA2_DISABLED))
+#if defined(T1HA1_DISABLED)
+  return t1ha2_atonce(data, length, seed);
+#else
+  return t1ha1_le(data, length, seed);
+#endif /* T1HA1_DISABLED */
+#else  /* 32/64 */
+  return t1ha0_32le(data, length, seed);
+#endif /* 32/64 */
+}
+
+#endif /* !T1HA0_USE_DEFINE */
+
+#endif /* !T1HA0_RUNTIME_SELECT */
+
+#endif /* T1HA0_DISABLED */
+
+#ifdef __cplusplus
+}
+#endif
+
+#if __GNUC_PREREQ(4, 0)
+#pragma GCC visibility pop
+#endif /* __GNUC_PREREQ(4,0) */
diff --git a/include/t1ha0_ia32aes_b.h b/include/t1ha0_ia32aes_b.h
new file mode 100644
index 00000000..e8e52638
--- /dev/null
+++ b/include/t1ha0_ia32aes_b.h
@@ -0,0 +1,167 @@
+/*
+ *  Copyright (c) 2016-2020 Positive Technologies, https://www.ptsecurity.com,
+ *  Fast Positive Hash.
+ *
+ *  Portions Copyright (c) 2010-2020 Leonid Yuriev <leo@yuriev.ru>,
+ *  The 1Hippeus project (t1h).
+ *
+ *  This software is provided 'as-is', without any express or implied
+ *  warranty. In no event will the authors be held liable for any damages
+ *  arising from the use of this software.
+ *
+ *  Permission is granted to anyone to use this software for any purpose,
+ *  including commercial applications, and to alter it and redistribute it
+ *  freely, subject to the following restrictions:
+ *
+ *  1. The origin of this software must not be misrepresented; you must not
+ *     claim that you wrote the original software. If you use this software
+ *     in a product, an acknowledgement in the product documentation would be
+ *     appreciated but is not required.
+ *  2. Altered source versions must be plainly marked as such, and must not be
+ *     misrepresented as being the original software.
+ *  3. This notice may not be removed or altered from any source distribution.
+ */
+
+/*
+ * t1ha = { Fast Positive Hash, aka "Позитивный Хэш" }
+ * by [Positive Technologies](https://www.ptsecurity.ru)
+ *
+ * Briefly, it is a 64-bit Hash Function:
+ *  1. Created for 64-bit little-endian platforms, in predominantly for x86_64,
+ *     but portable and without penalties it can run on any 64-bit CPU.
+ *  2. In most cases up to 15% faster than City64, xxHash, mum-hash, metro-hash
+ *     and all others portable hash-functions (which do not use specific
+ *     hardware tricks).
+ *  3. Not suitable for cryptography.
+ *
+ * The Future will (be) Positive. Всё будет хорошо.
+ *
+ * ACKNOWLEDGEMENT:
+ * The t1ha was originally developed by Leonid Yuriev (Леонид Юрьев)
+ * for The 1Hippeus project - zerocopy messaging in the spirit of Sparta!
+ */
+
+#include "t1ha_bits.h"
+#include "t1ha_selfcheck.h"
+
+#if T1HA0_AESNI_AVAILABLE
+
+uint64_t T1HA_IA32AES_NAME(const void *data, uint32_t len) {
+  uint64_t a = 0;
+  uint64_t b = len;
+
+  if (likely(len > 32)) {
+    __m128i x = _mm_set_epi64x(a, b);
+    __m128i y = _mm_aesenc_si128(x, _mm_set_epi64x(prime_0, prime_1));
+
+    const __m128i *v = (const __m128i *)data;
+    const __m128i *const detent =
+        (const __m128i *)((const uint8_t *)data + (len & ~15ul));
+    data = detent;
+
+    if (len & 16) {
+      x = _mm_add_epi64(x, _mm_loadu_si128(v++));
+      y = _mm_aesenc_si128(x, y);
+    }
+    len &= 15;
+
+    if (v + 7 < detent) {
+      __m128i salt = y;
+      do {
+        __m128i t = _mm_aesenc_si128(_mm_loadu_si128(v++), salt);
+        t = _mm_aesdec_si128(t, _mm_loadu_si128(v++));
+        t = _mm_aesdec_si128(t, _mm_loadu_si128(v++));
+        t = _mm_aesdec_si128(t, _mm_loadu_si128(v++));
+
+        t = _mm_aesdec_si128(t, _mm_loadu_si128(v++));
+        t = _mm_aesdec_si128(t, _mm_loadu_si128(v++));
+        t = _mm_aesdec_si128(t, _mm_loadu_si128(v++));
+        t = _mm_aesdec_si128(t, _mm_loadu_si128(v++));
+
+        salt = _mm_add_epi64(salt, _mm_set_epi64x(prime_5, prime_6));
+        t = _mm_aesenc_si128(x, t);
+        x = _mm_add_epi64(y, x);
+        y = t;
+      } while (v + 7 < detent);
+    }
+
+    while (v < detent) {
+      __m128i v0y = _mm_add_epi64(y, _mm_loadu_si128(v++));
+      __m128i v1x = _mm_sub_epi64(x, _mm_loadu_si128(v++));
+      x = _mm_aesdec_si128(x, v0y);
+      y = _mm_aesdec_si128(y, v1x);
+    }
+
+    x = _mm_add_epi64(_mm_aesdec_si128(x, _mm_aesenc_si128(y, x)), y);
+#if defined(__x86_64__) || defined(_M_X64)
+#if defined(__SSE4_1__) || defined(__AVX__)
+    a = _mm_extract_epi64(x, 0);
+    b = _mm_extract_epi64(x, 1);
+#else
+    a = _mm_cvtsi128_si64(x);
+    b = _mm_cvtsi128_si64(_mm_unpackhi_epi64(x, x));
+#endif
+#else
+#if defined(__SSE4_1__) || defined(__AVX__)
+    a = (uint32_t)_mm_extract_epi32(x, 0) | (uint64_t)_mm_extract_epi32(x, 1)
+                                                << 32;
+    b = (uint32_t)_mm_extract_epi32(x, 2) | (uint64_t)_mm_extract_epi32(x, 3)
+                                                << 32;
+#else
+    a = (uint32_t)_mm_cvtsi128_si32(x);
+    a |= (uint64_t)_mm_cvtsi128_si32(_mm_shuffle_epi32(x, 1)) << 32;
+    x = _mm_unpackhi_epi64(x, x);
+    b = (uint32_t)_mm_cvtsi128_si32(x);
+    b |= (uint64_t)_mm_cvtsi128_si32(_mm_shuffle_epi32(x, 1)) << 32;
+#endif
+#endif
+#ifdef __AVX__
+    _mm256_zeroupper();
+#elif !(defined(_X86_64_) || defined(__x86_64__) || defined(_M_X64) ||         \
+        defined(__e2k__))
+    _mm_empty();
+#endif
+  }
+
+  const uint64_t *v = (const uint64_t *)data;
+  switch (len) {
+  default:
+    mixup64(&a, &b, fetch64_le_unaligned(v++), prime_4);
+  /* fall through */
+  case 24:
+  case 23:
+  case 22:
+  case 21:
+  case 20:
+  case 19:
+  case 18:
+  case 17:
+    mixup64(&b, &a, fetch64_le_unaligned(v++), prime_3);
+  /* fall through */
+  case 16:
+  case 15:
+  case 14:
+  case 13:
+  case 12:
+  case 11:
+  case 10:
+  case 9:
+    mixup64(&a, &b, fetch64_le_unaligned(v++), prime_2);
+  /* fall through */
+  case 8:
+  case 7:
+  case 6:
+  case 5:
+  case 4:
+  case 3:
+  case 2:
+  case 1:
+    mixup64(&b, &a, tail64_le_unaligned(v, len), prime_1);
+  /* fall through */
+  case 0:
+    return final64(a, b);
+  }
+}
+
+#endif /* T1HA0_AESNI_AVAILABLE */
+#undef T1HA_IA32AES_NAME
diff --git a/include/t1ha_bits.h b/include/t1ha_bits.h
new file mode 100644
index 00000000..539369aa
--- /dev/null
+++ b/include/t1ha_bits.h
@@ -0,0 +1,1254 @@
+/*
+ *  Copyright (c) 2016-2020 Positive Technologies, https://www.ptsecurity.com,
+ *  Fast Positive Hash.
+ *
+ *  Portions Copyright (c) 2010-2020 Leonid Yuriev <leo@yuriev.ru>,
+ *  The 1Hippeus project (t1h).
+ *
+ *  This software is provided 'as-is', without any express or implied
+ *  warranty. In no event will the authors be held liable for any damages
+ *  arising from the use of this software.
+ *
+ *  Permission is granted to anyone to use this software for any purpose,
+ *  including commercial applications, and to alter it and redistribute it
+ *  freely, subject to the following restrictions:
+ *
+ *  1. The origin of this software must not be misrepresented; you must not
+ *     claim that you wrote the original software. If you use this software
+ *     in a product, an acknowledgement in the product documentation would be
+ *     appreciated but is not required.
+ *  2. Altered source versions must be plainly marked as such, and must not be
+ *     misrepresented as being the original software.
+ *  3. This notice may not be removed or altered from any source distribution.
+ */
+
+/*
+ * t1ha = { Fast Positive Hash, aka "Позитивный Хэш" }
+ * by [Positive Technologies](https://www.ptsecurity.ru)
+ *
+ * Briefly, it is a 64-bit Hash Function:
+ *  1. Created for 64-bit little-endian platforms, in predominantly for x86_64,
+ *     but portable and without penalties it can run on any 64-bit CPU.
+ *  2. In most cases up to 15% faster than City64, xxHash, mum-hash, metro-hash
+ *     and all others portable hash-functions (which do not use specific
+ *     hardware tricks).
+ *  3. Not suitable for cryptography.
+ *
+ * The Future will (be) Positive. Всё будет хорошо.
+ *
+ * ACKNOWLEDGEMENT:
+ * The t1ha was originally developed by Leonid Yuriev (Леонид Юрьев)
+ * for The 1Hippeus project - zerocopy messaging in the spirit of Sparta!
+ */
+
+#pragma once
+
+#if defined(_MSC_VER)
+#pragma warning(disable : 4201) /* nameless struct/union */
+#if _MSC_VER > 1800
+#pragma warning(disable : 4464) /* relative include path contains '..' */
+#endif                          /* 1800 */
+#endif                          /* MSVC */
+#include "t1ha.h"
+
+#ifndef T1HA_USE_FAST_ONESHOT_READ
+/* Define it to 1 for little bit faster code.
+ * Unfortunately this may triggering a false-positive alarms from Valgrind,
+ * AddressSanitizer and other similar tool.
+ * So, define it to 0 for calmness if doubt. */
+#define T1HA_USE_FAST_ONESHOT_READ 1
+#endif /* T1HA_USE_FAST_ONESHOT_READ */
+
+/*****************************************************************************/
+
+#include <assert.h>  /* for assert() */
+#include <stdbool.h> /* for bool */
+#include <string.h>  /* for memcpy() */
+
+#if __BYTE_ORDER__ != __ORDER_LITTLE_ENDIAN__ &&                               \
+    __BYTE_ORDER__ != __ORDER_BIG_ENDIAN__
+#error Unsupported byte order.
+#endif
+
+#define T1HA_UNALIGNED_ACCESS__UNABLE 0
+#define T1HA_UNALIGNED_ACCESS__SLOW 1
+#define T1HA_UNALIGNED_ACCESS__EFFICIENT 2
+
+#ifndef T1HA_SYS_UNALIGNED_ACCESS
+#if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)
+#define T1HA_SYS_UNALIGNED_ACCESS T1HA_UNALIGNED_ACCESS__EFFICIENT
+#elif defined(__ia32__)
+#define T1HA_SYS_UNALIGNED_ACCESS T1HA_UNALIGNED_ACCESS__EFFICIENT
+#elif defined(__e2k__)
+#define T1HA_SYS_UNALIGNED_ACCESS T1HA_UNALIGNED_ACCESS__SLOW
+#elif defined(__ARM_FEATURE_UNALIGNED)
+#define T1HA_SYS_UNALIGNED_ACCESS T1HA_UNALIGNED_ACCESS__EFFICIENT
+#else
+#define T1HA_SYS_UNALIGNED_ACCESS T1HA_UNALIGNED_ACCESS__UNABLE
+#endif
+#endif /* T1HA_SYS_UNALIGNED_ACCESS */
+
+#define ALIGNMENT_16 2
+#define ALIGNMENT_32 4
+#if UINTPTR_MAX > 0xffffFFFFul || ULONG_MAX > 0xffffFFFFul
+#define ALIGNMENT_64 8
+#else
+#define ALIGNMENT_64 4
+#endif
+
+#ifndef PAGESIZE
+#define PAGESIZE 4096
+#endif /* PAGESIZE */
+
+/***************************************************************************/
+
+#ifndef __has_builtin
+#define __has_builtin(x) (0)
+#endif
+
+#ifndef __has_warning
+#define __has_warning(x) (0)
+#endif
+
+#ifndef __has_feature
+#define __has_feature(x) (0)
+#endif
+
+#ifndef __has_extension
+#define __has_extension(x) (0)
+#endif
+
+#if __has_feature(address_sanitizer)
+#define __SANITIZE_ADDRESS__ 1
+#endif
+
+#ifndef __optimize
+#if defined(__clang__) && !__has_attribute(__optimize__)
+#define __optimize(ops)
+#elif defined(__GNUC__) || __has_attribute(__optimize__)
+#define __optimize(ops) __attribute__((__optimize__(ops)))
+#else
+#define __optimize(ops)
+#endif
+#endif /* __optimize */
+
+#ifndef __cold
+#if defined(__OPTIMIZE__)
+#if defined(__e2k__)
+#define __cold __optimize(1) __attribute__((__cold__))
+#elif defined(__clang__) && !__has_attribute(__cold__) &&                      \
+    __has_attribute(__section__)
+/* just put infrequently used functions in separate section */
+#define __cold __attribute__((__section__("text.unlikely"))) __optimize("Os")
+#elif defined(__GNUC__) || __has_attribute(__cold__)
+#define __cold __attribute__((__cold__)) __optimize("Os")
+#else
+#define __cold __optimize("Os")
+#endif
+#else
+#define __cold
+#endif
+#endif /* __cold */
+
+#if __GNUC_PREREQ(4, 4) || defined(__clang__)
+
+#if defined(__ia32__) || defined(__e2k__)
+#include <x86intrin.h>
+#endif
+
+#if defined(__ia32__) && !defined(__cpuid_count)
+#include <cpuid.h>
+#endif
+
+#if defined(__e2k__)
+#include <e2kbuiltin.h>
+#endif
+
+#ifndef likely
+#define likely(cond) __builtin_expect(!!(cond), 1)
+#endif
+
+#ifndef unlikely
+#define unlikely(cond) __builtin_expect(!!(cond), 0)
+#endif
+
+#if __GNUC_PREREQ(4, 5) || __has_builtin(__builtin_unreachable)
+#define unreachable() __builtin_unreachable()
+#endif
+
+#define bswap64(v) __builtin_bswap64(v)
+#define bswap32(v) __builtin_bswap32(v)
+#if __GNUC_PREREQ(4, 8) || __has_builtin(__builtin_bswap16)
+#define bswap16(v) __builtin_bswap16(v)
+#endif
+
+#if !defined(__maybe_unused) &&                                                \
+    (__GNUC_PREREQ(4, 3) || __has_attribute(__unused__))
+#define __maybe_unused __attribute__((__unused__))
+#endif
+
+#if !defined(__always_inline) &&                                               \
+    (__GNUC_PREREQ(3, 2) || __has_attribute(__always_inline__))
+#define __always_inline __inline __attribute__((__always_inline__))
+#endif
+
+#if defined(__e2k__)
+
+#if __iset__ >= 3
+#define mul_64x64_high(a, b) __builtin_e2k_umulhd(a, b)
+#endif /* __iset__ >= 3 */
+
+#if __iset__ >= 5
+static __maybe_unused __always_inline unsigned
+e2k_add64carry_first(uint64_t base, uint64_t addend, uint64_t *sum) {
+  *sum = base + addend;
+  return (unsigned)__builtin_e2k_addcd_c(base, addend, 0);
+}
+#define add64carry_first(base, addend, sum)                                    \
+  e2k_add64carry_first(base, addend, sum)
+
+static __maybe_unused __always_inline unsigned
+e2k_add64carry_next(unsigned carry, uint64_t base, uint64_t addend,
+                    uint64_t *sum) {
+  *sum = __builtin_e2k_addcd(base, addend, carry);
+  return (unsigned)__builtin_e2k_addcd_c(base, addend, carry);
+}
+#define add64carry_next(carry, base, addend, sum)                              \
+  e2k_add64carry_next(carry, base, addend, sum)
+
+static __maybe_unused __always_inline void e2k_add64carry_last(unsigned carry,
+                                                               uint64_t base,
+                                                               uint64_t addend,
+                                                               uint64_t *sum) {
+  *sum = __builtin_e2k_addcd(base, addend, carry);
+}
+#define add64carry_last(carry, base, addend, sum)                              \
+  e2k_add64carry_last(carry, base, addend, sum)
+#endif /* __iset__ >= 5 */
+
+#define fetch64_be_aligned(ptr) ((uint64_t)__builtin_e2k_ld_64s_be(ptr))
+#define fetch32_be_aligned(ptr) ((uint32_t)__builtin_e2k_ld_32u_be(ptr))
+
+#endif /* __e2k__ Elbrus */
+
+#elif defined(_MSC_VER)
+
+#if _MSC_FULL_VER < 190024234 && defined(_M_IX86)
+#pragma message(                                                               \
+    "For AES-NI at least \"Microsoft C/C++ Compiler\" version 19.00.24234 (Visual Studio 2015 Update 3) is required.")
+#endif
+#if _MSC_FULL_VER < 191526730
+#pragma message(                                                               \
+    "It is recommended to use \"Microsoft C/C++ Compiler\" version 19.15.26730 (Visual Studio 2017 15.8) or newer.")
+#endif
+#if _MSC_FULL_VER < 180040629
+#error At least "Microsoft C/C++ Compiler" version 18.00.40629 (Visual Studio 2013 Update 5) is required.
+#endif
+
+#pragma warning(push, 1)
+
+#include <intrin.h>
+#include <stdlib.h>
+#define likely(cond) (cond)
+#define unlikely(cond) (cond)
+#define unreachable() __assume(0)
+#define bswap64(v) _byteswap_uint64(v)
+#define bswap32(v) _byteswap_ulong(v)
+#define bswap16(v) _byteswap_ushort(v)
+#define rot64(v, s) _rotr64(v, s)
+#define rot32(v, s) _rotr(v, s)
+#define __always_inline __forceinline
+
+#if defined(_M_X64) || defined(_M_IA64)
+#pragma intrinsic(_umul128)
+#define mul_64x64_128(a, b, ph) _umul128(a, b, ph)
+#pragma intrinsic(_addcarry_u64)
+#define add64carry_first(base, addend, sum) _addcarry_u64(0, base, addend, sum)
+#define add64carry_next(carry, base, addend, sum)                              \
+  _addcarry_u64(carry, base, addend, sum)
+#define add64carry_last(carry, base, addend, sum)                              \
+  (void)_addcarry_u64(carry, base, addend, sum)
+#endif
+
+#if defined(_M_ARM64) || defined(_M_X64) || defined(_M_IA64)
+#pragma intrinsic(__umulh)
+#define mul_64x64_high(a, b) __umulh(a, b)
+#endif
+
+#if defined(_M_IX86)
+#pragma intrinsic(__emulu)
+#define mul_32x32_64(a, b) __emulu(a, b)
+
+#if _MSC_VER >= 1915 /* LY: workaround for SSA-optimizer bug */
+#pragma intrinsic(_addcarry_u32)
+#define add32carry_first(base, addend, sum) _addcarry_u32(0, base, addend, sum)
+#define add32carry_next(carry, base, addend, sum)                              \
+  _addcarry_u32(carry, base, addend, sum)
+#define add32carry_last(carry, base, addend, sum)                              \
+  (void)_addcarry_u32(carry, base, addend, sum)
+
+static __forceinline char
+msvc32_add64carry_first(uint64_t base, uint64_t addend, uint64_t *sum) {
+  uint32_t *const sum32 = (uint32_t *)sum;
+  const uint32_t base_32l = (uint32_t)base;
+  const uint32_t base_32h = (uint32_t)(base >> 32);
+  const uint32_t addend_32l = (uint32_t)addend;
+  const uint32_t addend_32h = (uint32_t)(addend >> 32);
+  return add32carry_next(add32carry_first(base_32l, addend_32l, sum32),
+                         base_32h, addend_32h, sum32 + 1);
+}
+#define add64carry_first(base, addend, sum)                                    \
+  msvc32_add64carry_first(base, addend, sum)
+
+static __forceinline char msvc32_add64carry_next(char carry, uint64_t base,
+                                                 uint64_t addend,
+                                                 uint64_t *sum) {
+  uint32_t *const sum32 = (uint32_t *)sum;
+  const uint32_t base_32l = (uint32_t)base;
+  const uint32_t base_32h = (uint32_t)(base >> 32);
+  const uint32_t addend_32l = (uint32_t)addend;
+  const uint32_t addend_32h = (uint32_t)(addend >> 32);
+  return add32carry_next(add32carry_next(carry, base_32l, addend_32l, sum32),
+                         base_32h, addend_32h, sum32 + 1);
+}
+#define add64carry_next(carry, base, addend, sum)                              \
+  msvc32_add64carry_next(carry, base, addend, sum)
+
+static __forceinline void msvc32_add64carry_last(char carry, uint64_t base,
+                                                 uint64_t addend,
+                                                 uint64_t *sum) {
+  uint32_t *const sum32 = (uint32_t *)sum;
+  const uint32_t base_32l = (uint32_t)base;
+  const uint32_t base_32h = (uint32_t)(base >> 32);
+  const uint32_t addend_32l = (uint32_t)addend;
+  const uint32_t addend_32h = (uint32_t)(addend >> 32);
+  add32carry_last(add32carry_next(carry, base_32l, addend_32l, sum32), base_32h,
+                  addend_32h, sum32 + 1);
+}
+#define add64carry_last(carry, base, addend, sum)                              \
+  msvc32_add64carry_last(carry, base, addend, sum)
+#endif /* _MSC_FULL_VER >= 190024231 */
+
+#elif defined(_M_ARM)
+#define mul_32x32_64(a, b) _arm_umull(a, b)
+#endif
+
+#pragma warning(pop)
+#pragma warning(disable : 4514) /* 'xyz': unreferenced inline function         \
+                                   has been removed */
+#pragma warning(disable : 4710) /* 'xyz': function not inlined */
+#pragma warning(disable : 4711) /* function 'xyz' selected for                 \
+                                   automatic inline expansion */
+#pragma warning(disable : 4127) /* conditional expression is constant */
+#pragma warning(disable : 4702) /* unreachable code */
+#endif                          /* Compiler */
+
+#ifndef likely
+#define likely(cond) (cond)
+#endif
+#ifndef unlikely
+#define unlikely(cond) (cond)
+#endif
+#ifndef __maybe_unused
+#define __maybe_unused
+#endif
+#ifndef __always_inline
+#define __always_inline __inline
+#endif
+#ifndef unreachable
+#define unreachable()                                                          \
+  do {                                                                         \
+  } while (1)
+#endif
+
+#ifndef bswap64
+#if defined(bswap_64)
+#define bswap64 bswap_64
+#elif defined(__bswap_64)
+#define bswap64 __bswap_64
+#else
+static __always_inline uint64_t bswap64(uint64_t v) {
+  return v << 56 | v >> 56 | ((v << 40) & UINT64_C(0x00ff000000000000)) |
+         ((v << 24) & UINT64_C(0x0000ff0000000000)) |
+         ((v << 8) & UINT64_C(0x000000ff00000000)) |
+         ((v >> 8) & UINT64_C(0x00000000ff000000)) |
+         ((v >> 24) & UINT64_C(0x0000000000ff0000)) |
+         ((v >> 40) & UINT64_C(0x000000000000ff00));
+}
+#endif
+#endif /* bswap64 */
+
+#ifndef bswap32
+#if defined(bswap_32)
+#define bswap32 bswap_32
+#elif defined(__bswap_32)
+#define bswap32 __bswap_32
+#else
+static __always_inline uint32_t bswap32(uint32_t v) {
+  return v << 24 | v >> 24 | ((v << 8) & UINT32_C(0x00ff0000)) |
+         ((v >> 8) & UINT32_C(0x0000ff00));
+}
+#endif
+#endif /* bswap32 */
+
+#ifndef bswap16
+#if defined(bswap_16)
+#define bswap16 bswap_16
+#elif defined(__bswap_16)
+#define bswap16 __bswap_16
+#else
+static __always_inline uint16_t bswap16(uint16_t v) { return v << 8 | v >> 8; }
+#endif
+#endif /* bswap16 */
+
+#if defined(__ia32__) ||                                                       \
+    T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__EFFICIENT
+/* The __builtin_assume_aligned() leads gcc/clang to load values into the
+ * registers, even when it is possible to directly use an operand from memory.
+ * This can lead to a shortage of registers and a significant slowdown.
+ * Therefore avoid unnecessary use of  __builtin_assume_aligned() for x86. */
+#define read_unaligned(ptr, bits) (*(const uint##bits##_t *__restrict)(ptr))
+#define read_aligned(ptr, bits) (*(const uint##bits##_t *__restrict)(ptr))
+#endif /* __ia32__ */
+
+#ifndef read_unaligned
+#if defined(__GNUC__) || __has_attribute(__packed__)
+typedef struct {
+  uint8_t unaligned_8;
+  uint16_t unaligned_16;
+  uint32_t unaligned_32;
+  uint64_t unaligned_64;
+} __attribute__((__packed__)) t1ha_unaligned_proxy;
+#define read_unaligned(ptr, bits)                                              \
+  (((const t1ha_unaligned_proxy *)((const uint8_t *)(ptr)-offsetof(            \
+        t1ha_unaligned_proxy, unaligned_##bits)))                              \
+       ->unaligned_##bits)
+#elif defined(_MSC_VER)
+#pragma warning(                                                               \
+    disable : 4235) /* nonstandard extension used: '__unaligned'               \
+                     * keyword not supported on this architecture */
+#define read_unaligned(ptr, bits) (*(const __unaligned uint##bits##_t *)(ptr))
+#else
+#pragma pack(push, 1)
+typedef struct {
+  uint8_t unaligned_8;
+  uint16_t unaligned_16;
+  uint32_t unaligned_32;
+  uint64_t unaligned_64;
+} t1ha_unaligned_proxy;
+#pragma pack(pop)
+#define read_unaligned(ptr, bits)                                              \
+  (((const t1ha_unaligned_proxy *)((const uint8_t *)(ptr)-offsetof(            \
+        t1ha_unaligned_proxy, unaligned_##bits)))                              \
+       ->unaligned_##bits)
+#endif
+#endif /* read_unaligned */
+
+#ifndef read_aligned
+#if __GNUC_PREREQ(4, 8) || __has_builtin(__builtin_assume_aligned)
+#define read_aligned(ptr, bits)                                                \
+  (*(const uint##bits##_t *)__builtin_assume_aligned(ptr, ALIGNMENT_##bits))
+#elif (__GNUC_PREREQ(3, 3) || __has_attribute(__aligned__)) &&                 \
+    !defined(__clang__)
+#define read_aligned(ptr, bits)                                                \
+  (*(const uint##bits##_t                                                      \
+     __attribute__((__aligned__(ALIGNMENT_##bits))) *)(ptr))
+#elif __has_attribute(__assume_aligned__)
+
+static __always_inline const
+    uint16_t *__attribute__((__assume_aligned__(ALIGNMENT_16)))
+    cast_aligned_16(const void *ptr) {
+  return (const uint16_t *)ptr;
+}
+static __always_inline const
+    uint32_t *__attribute__((__assume_aligned__(ALIGNMENT_32)))
+    cast_aligned_32(const void *ptr) {
+  return (const uint32_t *)ptr;
+}
+static __always_inline const
+    uint64_t *__attribute__((__assume_aligned__(ALIGNMENT_64)))
+    cast_aligned_64(const void *ptr) {
+  return (const uint64_t *)ptr;
+}
+
+#define read_aligned(ptr, bits) (*cast_aligned_##bits(ptr))
+
+#elif defined(_MSC_VER)
+#define read_aligned(ptr, bits)                                                \
+  (*(const __declspec(align(ALIGNMENT_##bits)) uint##bits##_t *)(ptr))
+#else
+#define read_aligned(ptr, bits) (*(const uint##bits##_t *)(ptr))
+#endif
+#endif /* read_aligned */
+
+#ifndef prefetch
+#if (__GNUC_PREREQ(4, 0) || __has_builtin(__builtin_prefetch)) &&              \
+    !defined(__ia32__)
+#define prefetch(ptr) __builtin_prefetch(ptr)
+#elif defined(_M_ARM64) || defined(_M_ARM)
+#define prefetch(ptr) __prefetch(ptr)
+#else
+#define prefetch(ptr)                                                          \
+  do {                                                                         \
+    (void)(ptr);                                                               \
+  } while (0)
+#endif
+#endif /* prefetch */
+
+#if __has_warning("-Wconstant-logical-operand")
+#if defined(__clang__)
+#pragma clang diagnostic ignored "-Wconstant-logical-operand"
+#elif defined(__GNUC__)
+#pragma GCC diagnostic ignored "-Wconstant-logical-operand"
+#else
+#pragma warning disable "constant-logical-operand"
+#endif
+#endif /* -Wconstant-logical-operand */
+
+#if __has_warning("-Wtautological-pointer-compare")
+#if defined(__clang__)
+#pragma clang diagnostic ignored "-Wtautological-pointer-compare"
+#elif defined(__GNUC__)
+#pragma GCC diagnostic ignored "-Wtautological-pointer-compare"
+#else
+#pragma warning disable "tautological-pointer-compare"
+#endif
+#endif /* -Wtautological-pointer-compare */
+
+/***************************************************************************/
+
+#if __GNUC_PREREQ(4, 0)
+#pragma GCC visibility push(hidden)
+#endif /* __GNUC_PREREQ(4,0) */
+
+/*---------------------------------------------------------- Little Endian */
+
+#ifndef fetch16_le_aligned
+static __maybe_unused __always_inline uint16_t
+fetch16_le_aligned(const void *v) {
+  assert(((uintptr_t)v) % ALIGNMENT_16 == 0);
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+  return read_aligned(v, 16);
+#else
+  return bswap16(read_aligned(v, 16));
+#endif
+}
+#endif /* fetch16_le_aligned */
+
+#ifndef fetch16_le_unaligned
+static __maybe_unused __always_inline uint16_t
+fetch16_le_unaligned(const void *v) {
+#if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__UNABLE
+  const uint8_t *p = (const uint8_t *)v;
+  return p[0] | (uint16_t)p[1] << 8;
+#elif __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+  return read_unaligned(v, 16);
+#else
+  return bswap16(read_unaligned(v, 16));
+#endif
+}
+#endif /* fetch16_le_unaligned */
+
+#ifndef fetch32_le_aligned
+static __maybe_unused __always_inline uint32_t
+fetch32_le_aligned(const void *v) {
+  assert(((uintptr_t)v) % ALIGNMENT_32 == 0);
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+  return read_aligned(v, 32);
+#else
+  return bswap32(read_aligned(v, 32));
+#endif
+}
+#endif /* fetch32_le_aligned */
+
+#ifndef fetch32_le_unaligned
+static __maybe_unused __always_inline uint32_t
+fetch32_le_unaligned(const void *v) {
+#if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__UNABLE
+  return fetch16_le_unaligned(v) |
+         (uint32_t)fetch16_le_unaligned((const uint8_t *)v + 2) << 16;
+#elif __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+  return read_unaligned(v, 32);
+#else
+  return bswap32(read_unaligned(v, 32));
+#endif
+}
+#endif /* fetch32_le_unaligned */
+
+#ifndef fetch64_le_aligned
+static __maybe_unused __always_inline uint64_t
+fetch64_le_aligned(const void *v) {
+  assert(((uintptr_t)v) % ALIGNMENT_64 == 0);
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+  return read_aligned(v, 64);
+#else
+  return bswap64(read_aligned(v, 64));
+#endif
+}
+#endif /* fetch64_le_aligned */
+
+#ifndef fetch64_le_unaligned
+static __maybe_unused __always_inline uint64_t
+fetch64_le_unaligned(const void *v) {
+#if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__UNABLE
+  return fetch32_le_unaligned(v) |
+         (uint64_t)fetch32_le_unaligned((const uint8_t *)v + 4) << 32;
+#elif __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+  return read_unaligned(v, 64);
+#else
+  return bswap64(read_unaligned(v, 64));
+#endif
+}
+#endif /* fetch64_le_unaligned */
+
+static __maybe_unused __always_inline uint64_t tail64_le_aligned(const void *v,
+                                                                 size_t tail) {
+  const uint8_t *const p = (const uint8_t *)v;
+#if T1HA_USE_FAST_ONESHOT_READ && !defined(__SANITIZE_ADDRESS__)
+  /* We can perform a 'oneshot' read, which is little bit faster. */
+  const unsigned shift = ((8 - tail) & 7) << 3;
+  return fetch64_le_aligned(p) & ((~UINT64_C(0)) >> shift);
+#else
+  uint64_t r = 0;
+  switch (tail & 7) {
+  default:
+    unreachable();
+/* fall through */
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+  /* For most CPUs this code is better when not needed byte reordering. */
+  case 0:
+    return fetch64_le_aligned(p);
+  case 7:
+    r = (uint64_t)p[6] << 8;
+  /* fall through */
+  case 6:
+    r += p[5];
+    r <<= 8;
+  /* fall through */
+  case 5:
+    r += p[4];
+    r <<= 32;
+  /* fall through */
+  case 4:
+    return r + fetch32_le_aligned(p);
+  case 3:
+    r = (uint64_t)p[2] << 16;
+  /* fall through */
+  case 2:
+    return r + fetch16_le_aligned(p);
+  case 1:
+    return p[0];
+#else
+  case 0:
+    r = p[7] << 8;
+  /* fall through */
+  case 7:
+    r += p[6];
+    r <<= 8;
+  /* fall through */
+  case 6:
+    r += p[5];
+    r <<= 8;
+  /* fall through */
+  case 5:
+    r += p[4];
+    r <<= 8;
+  /* fall through */
+  case 4:
+    r += p[3];
+    r <<= 8;
+  /* fall through */
+  case 3:
+    r += p[2];
+    r <<= 8;
+  /* fall through */
+  case 2:
+    r += p[1];
+    r <<= 8;
+  /* fall through */
+  case 1:
+    return r + p[0];
+#endif
+  }
+#endif /* T1HA_USE_FAST_ONESHOT_READ */
+}
+
+#if T1HA_USE_FAST_ONESHOT_READ &&                                              \
+    T1HA_SYS_UNALIGNED_ACCESS != T1HA_UNALIGNED_ACCESS__UNABLE &&              \
+    defined(PAGESIZE) && PAGESIZE > 42 && !defined(__SANITIZE_ADDRESS__)
+#define can_read_underside(ptr, size)                                          \
+  (((PAGESIZE - (size)) & (uintptr_t)(ptr)) != 0)
+#endif /* T1HA_USE_FAST_ONESHOT_READ */
+
+static __maybe_unused __always_inline uint64_t
+tail64_le_unaligned(const void *v, size_t tail) {
+  const uint8_t *p = (const uint8_t *)v;
+#if defined(can_read_underside) &&                                             \
+    (UINTPTR_MAX > 0xffffFFFFul || ULONG_MAX > 0xffffFFFFul)
+  /* On some systems (e.g. x86_64) we can perform a 'oneshot' read, which
+   * is little bit faster. Thanks Marcin Żukowski <marcin.zukowski@gmail.com>
+   * for the reminder. */
+  const unsigned offset = (8 - tail) & 7;
+  const unsigned shift = offset << 3;
+  if (likely(can_read_underside(p, 8))) {
+    p -= offset;
+    return fetch64_le_unaligned(p) >> shift;
+  }
+  return fetch64_le_unaligned(p) & ((~UINT64_C(0)) >> shift);
+#else
+  uint64_t r = 0;
+  switch (tail & 7) {
+  default:
+    unreachable();
+/* fall through */
+#if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__EFFICIENT &&           \
+    __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+  /* For most CPUs this code is better when not needed
+   * copying for alignment or byte reordering. */
+  case 0:
+    return fetch64_le_unaligned(p);
+  case 7:
+    r = (uint64_t)p[6] << 8;
+  /* fall through */
+  case 6:
+    r += p[5];
+    r <<= 8;
+  /* fall through */
+  case 5:
+    r += p[4];
+    r <<= 32;
+  /* fall through */
+  case 4:
+    return r + fetch32_le_unaligned(p);
+  case 3:
+    r = (uint64_t)p[2] << 16;
+  /* fall through */
+  case 2:
+    return r + fetch16_le_unaligned(p);
+  case 1:
+    return p[0];
+#else
+  /* For most CPUs this code is better than a
+   * copying for alignment and/or byte reordering. */
+  case 0:
+    r = p[7] << 8;
+  /* fall through */
+  case 7:
+    r += p[6];
+    r <<= 8;
+  /* fall through */
+  case 6:
+    r += p[5];
+    r <<= 8;
+  /* fall through */
+  case 5:
+    r += p[4];
+    r <<= 8;
+  /* fall through */
+  case 4:
+    r += p[3];
+    r <<= 8;
+  /* fall through */
+  case 3:
+    r += p[2];
+    r <<= 8;
+  /* fall through */
+  case 2:
+    r += p[1];
+    r <<= 8;
+  /* fall through */
+  case 1:
+    return r + p[0];
+#endif
+  }
+#endif /* can_read_underside */
+}
+
+/*------------------------------------------------------------- Big Endian */
+
+#ifndef fetch16_be_aligned
+static __maybe_unused __always_inline uint16_t
+fetch16_be_aligned(const void *v) {
+  assert(((uintptr_t)v) % ALIGNMENT_16 == 0);
+#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+  return read_aligned(v, 16);
+#else
+  return bswap16(read_aligned(v, 16));
+#endif
+}
+#endif /* fetch16_be_aligned */
+
+#ifndef fetch16_be_unaligned
+static __maybe_unused __always_inline uint16_t
+fetch16_be_unaligned(const void *v) {
+#if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__UNABLE
+  const uint8_t *p = (const uint8_t *)v;
+  return (uint16_t)p[0] << 8 | p[1];
+#elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+  return read_unaligned(v, 16);
+#else
+  return bswap16(read_unaligned(v, 16));
+#endif
+}
+#endif /* fetch16_be_unaligned */
+
+#ifndef fetch32_be_aligned
+static __maybe_unused __always_inline uint32_t
+fetch32_be_aligned(const void *v) {
+  assert(((uintptr_t)v) % ALIGNMENT_32 == 0);
+#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+  return read_aligned(v, 32);
+#else
+  return bswap32(read_aligned(v, 32));
+#endif
+}
+#endif /* fetch32_be_aligned */
+
+#ifndef fetch32_be_unaligned
+static __maybe_unused __always_inline uint32_t
+fetch32_be_unaligned(const void *v) {
+#if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__UNABLE
+  return (uint32_t)fetch16_be_unaligned(v) << 16 |
+         fetch16_be_unaligned((const uint8_t *)v + 2);
+#elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+  return read_unaligned(v, 32);
+#else
+  return bswap32(read_unaligned(v, 32));
+#endif
+}
+#endif /* fetch32_be_unaligned */
+
+#ifndef fetch64_be_aligned
+static __maybe_unused __always_inline uint64_t
+fetch64_be_aligned(const void *v) {
+  assert(((uintptr_t)v) % ALIGNMENT_64 == 0);
+#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+  return read_aligned(v, 64);
+#else
+  return bswap64(read_aligned(v, 64));
+#endif
+}
+#endif /* fetch64_be_aligned */
+
+#ifndef fetch64_be_unaligned
+static __maybe_unused __always_inline uint64_t
+fetch64_be_unaligned(const void *v) {
+#if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__UNABLE
+  return (uint64_t)fetch32_be_unaligned(v) << 32 |
+         fetch32_be_unaligned((const uint8_t *)v + 4);
+#elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+  return read_unaligned(v, 64);
+#else
+  return bswap64(read_unaligned(v, 64));
+#endif
+}
+#endif /* fetch64_be_unaligned */
+
+static __maybe_unused __always_inline uint64_t tail64_be_aligned(const void *v,
+                                                                 size_t tail) {
+  const uint8_t *const p = (const uint8_t *)v;
+#if T1HA_USE_FAST_ONESHOT_READ && !defined(__SANITIZE_ADDRESS__)
+  /* We can perform a 'oneshot' read, which is little bit faster. */
+  const unsigned shift = ((8 - tail) & 7) << 3;
+  return fetch64_be_aligned(p) >> shift;
+#else
+  switch (tail & 7) {
+  default:
+    unreachable();
+/* fall through */
+#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+  /* For most CPUs this code is better when not byte reordering. */
+  case 1:
+    return p[0];
+  case 2:
+    return fetch16_be_aligned(p);
+  case 3:
+    return (uint32_t)fetch16_be_aligned(p) << 8 | p[2];
+  case 4:
+    return fetch32_be_aligned(p);
+  case 5:
+    return (uint64_t)fetch32_be_aligned(p) << 8 | p[4];
+  case 6:
+    return (uint64_t)fetch32_be_aligned(p) << 16 | fetch16_be_aligned(p + 4);
+  case 7:
+    return (uint64_t)fetch32_be_aligned(p) << 24 |
+           (uint32_t)fetch16_be_aligned(p + 4) << 8 | p[6];
+  case 0:
+    return fetch64_be_aligned(p);
+#else
+  case 1:
+    return p[0];
+  case 2:
+    return p[1] | (uint32_t)p[0] << 8;
+  case 3:
+    return p[2] | (uint32_t)p[1] << 8 | (uint32_t)p[0] << 16;
+  case 4:
+    return p[3] | (uint32_t)p[2] << 8 | (uint32_t)p[1] << 16 |
+           (uint32_t)p[0] << 24;
+  case 5:
+    return p[4] | (uint32_t)p[3] << 8 | (uint32_t)p[2] << 16 |
+           (uint32_t)p[1] << 24 | (uint64_t)p[0] << 32;
+  case 6:
+    return p[5] | (uint32_t)p[4] << 8 | (uint32_t)p[3] << 16 |
+           (uint32_t)p[2] << 24 | (uint64_t)p[1] << 32 | (uint64_t)p[0] << 40;
+  case 7:
+    return p[6] | (uint32_t)p[5] << 8 | (uint32_t)p[4] << 16 |
+           (uint32_t)p[3] << 24 | (uint64_t)p[2] << 32 | (uint64_t)p[1] << 40 |
+           (uint64_t)p[0] << 48;
+  case 0:
+    return p[7] | (uint32_t)p[6] << 8 | (uint32_t)p[5] << 16 |
+           (uint32_t)p[4] << 24 | (uint64_t)p[3] << 32 | (uint64_t)p[2] << 40 |
+           (uint64_t)p[1] << 48 | (uint64_t)p[0] << 56;
+#endif
+  }
+#endif /* T1HA_USE_FAST_ONESHOT_READ */
+}
+
+static __maybe_unused __always_inline uint64_t
+tail64_be_unaligned(const void *v, size_t tail) {
+  const uint8_t *p = (const uint8_t *)v;
+#if defined(can_read_underside) &&                                             \
+    (UINTPTR_MAX > 0xffffFFFFul || ULONG_MAX > 0xffffFFFFul)
+  /* On some systems (e.g. x86_64) we can perform a 'oneshot' read, which
+   * is little bit faster. Thanks Marcin Żukowski <marcin.zukowski@gmail.com>
+   * for the reminder. */
+  const unsigned offset = (8 - tail) & 7;
+  const unsigned shift = offset << 3;
+  if (likely(can_read_underside(p, 8))) {
+    p -= offset;
+    return fetch64_be_unaligned(p) & ((~UINT64_C(0)) >> shift);
+  }
+  return fetch64_be_unaligned(p) >> shift;
+#else
+  switch (tail & 7) {
+  default:
+    unreachable();
+/* fall through */
+#if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__EFFICIENT &&           \
+    __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+  /* For most CPUs this code is better when not needed
+   * copying for alignment or byte reordering. */
+  case 1:
+    return p[0];
+  case 2:
+    return fetch16_be_unaligned(p);
+  case 3:
+    return (uint32_t)fetch16_be_unaligned(p) << 8 | p[2];
+  case 4:
+    return fetch32_be(p);
+  case 5:
+    return (uint64_t)fetch32_be_unaligned(p) << 8 | p[4];
+  case 6:
+    return (uint64_t)fetch32_be_unaligned(p) << 16 |
+           fetch16_be_unaligned(p + 4);
+  case 7:
+    return (uint64_t)fetch32_be_unaligned(p) << 24 |
+           (uint32_t)fetch16_be_unaligned(p + 4) << 8 | p[6];
+  case 0:
+    return fetch64_be_unaligned(p);
+#else
+  /* For most CPUs this code is better than a
+   * copying for alignment and/or byte reordering. */
+  case 1:
+    return p[0];
+  case 2:
+    return p[1] | (uint32_t)p[0] << 8;
+  case 3:
+    return p[2] | (uint32_t)p[1] << 8 | (uint32_t)p[0] << 16;
+  case 4:
+    return p[3] | (uint32_t)p[2] << 8 | (uint32_t)p[1] << 16 |
+           (uint32_t)p[0] << 24;
+  case 5:
+    return p[4] | (uint32_t)p[3] << 8 | (uint32_t)p[2] << 16 |
+           (uint32_t)p[1] << 24 | (uint64_t)p[0] << 32;
+  case 6:
+    return p[5] | (uint32_t)p[4] << 8 | (uint32_t)p[3] << 16 |
+           (uint32_t)p[2] << 24 | (uint64_t)p[1] << 32 | (uint64_t)p[0] << 40;
+  case 7:
+    return p[6] | (uint32_t)p[5] << 8 | (uint32_t)p[4] << 16 |
+           (uint32_t)p[3] << 24 | (uint64_t)p[2] << 32 | (uint64_t)p[1] << 40 |
+           (uint64_t)p[0] << 48;
+  case 0:
+    return p[7] | (uint32_t)p[6] << 8 | (uint32_t)p[5] << 16 |
+           (uint32_t)p[4] << 24 | (uint64_t)p[3] << 32 | (uint64_t)p[2] << 40 |
+           (uint64_t)p[1] << 48 | (uint64_t)p[0] << 56;
+#endif
+  }
+#endif /* can_read_underside */
+}
+
+/***************************************************************************/
+
+#ifndef rot64
+static __maybe_unused __always_inline uint64_t rot64(uint64_t v, unsigned s) {
+  return (v >> s) | (v << (64 - s));
+}
+#endif /* rot64 */
+
+#ifndef mul_32x32_64
+static __maybe_unused __always_inline uint64_t mul_32x32_64(uint32_t a,
+                                                            uint32_t b) {
+  return a * (uint64_t)b;
+}
+#endif /* mul_32x32_64 */
+
+#ifndef add64carry_first
+static __maybe_unused __always_inline unsigned
+add64carry_first(uint64_t base, uint64_t addend, uint64_t *sum) {
+#if __has_builtin(__builtin_addcll)
+  unsigned long long carryout;
+  *sum = __builtin_addcll(base, addend, 0, &carryout);
+  return (unsigned)carryout;
+#else
+  *sum = base + addend;
+  return *sum < addend;
+#endif /* __has_builtin(__builtin_addcll) */
+}
+#endif /* add64carry_fist */
+
+#ifndef add64carry_next
+static __maybe_unused __always_inline unsigned
+add64carry_next(unsigned carry, uint64_t base, uint64_t addend, uint64_t *sum) {
+#if __has_builtin(__builtin_addcll)
+  unsigned long long carryout;
+  *sum = __builtin_addcll(base, addend, carry, &carryout);
+  return (unsigned)carryout;
+#else
+  *sum = base + addend + carry;
+  return *sum < addend || (carry && *sum == addend);
+#endif /* __has_builtin(__builtin_addcll) */
+}
+#endif /* add64carry_next */
+
+#ifndef add64carry_last
+static __maybe_unused __always_inline void
+add64carry_last(unsigned carry, uint64_t base, uint64_t addend, uint64_t *sum) {
+#if __has_builtin(__builtin_addcll)
+  unsigned long long carryout;
+  *sum = __builtin_addcll(base, addend, carry, &carryout);
+  (void)carryout;
+#else
+  *sum = base + addend + carry;
+#endif /* __has_builtin(__builtin_addcll) */
+}
+#endif /* add64carry_last */
+
+#ifndef mul_64x64_128
+static __maybe_unused __always_inline uint64_t mul_64x64_128(uint64_t a,
+                                                             uint64_t b,
+                                                             uint64_t *h) {
+#if (defined(__SIZEOF_INT128__) ||                                             \
+     (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)) &&            \
+    (!defined(__LCC__) || __LCC__ != 124)
+  __uint128_t r = (__uint128_t)a * (__uint128_t)b;
+  /* modern GCC could nicely optimize this */
+  *h = (uint64_t)(r >> 64);
+  return (uint64_t)r;
+#elif defined(mul_64x64_high)
+  *h = mul_64x64_high(a, b);
+  return a * b;
+#else
+  /* performs 64x64 to 128 bit multiplication */
+  const uint64_t ll = mul_32x32_64((uint32_t)a, (uint32_t)b);
+  const uint64_t lh = mul_32x32_64(a >> 32, (uint32_t)b);
+  const uint64_t hl = mul_32x32_64((uint32_t)a, b >> 32);
+  const uint64_t hh = mul_32x32_64(a >> 32, b >> 32);
+
+  /* Few simplification are possible here for 32-bit architectures,
+   * but thus we would lost compatibility with the original 64-bit
+   * version.  Think is very bad idea, because then 32-bit t1ha will
+   * still (relatively) very slowly and well yet not compatible. */
+  uint64_t l;
+  add64carry_last(add64carry_first(ll, lh << 32, &l), hh, lh >> 32, h);
+  add64carry_last(add64carry_first(l, hl << 32, &l), *h, hl >> 32, h);
+  return l;
+#endif
+}
+#endif /* mul_64x64_128() */
+
+#ifndef mul_64x64_high
+static __maybe_unused __always_inline uint64_t mul_64x64_high(uint64_t a,
+                                                              uint64_t b) {
+  uint64_t h;
+  mul_64x64_128(a, b, &h);
+  return h;
+}
+#endif /* mul_64x64_high */
+
+/***************************************************************************/
+
+/* 'magic' primes */
+static const uint64_t prime_0 = UINT64_C(0xEC99BF0D8372CAAB);
+static const uint64_t prime_1 = UINT64_C(0x82434FE90EDCEF39);
+static const uint64_t prime_2 = UINT64_C(0xD4F06DB99D67BE4B);
+static const uint64_t prime_3 = UINT64_C(0xBD9CACC22C6E9571);
+static const uint64_t prime_4 = UINT64_C(0x9C06FAF4D023E3AB);
+static const uint64_t prime_5 = UINT64_C(0xC060724A8424F345);
+static const uint64_t prime_6 = UINT64_C(0xCB5AF53AE3AAAC31);
+
+/* xor high and low parts of full 128-bit product */
+static __maybe_unused __always_inline uint64_t mux64(uint64_t v,
+                                                     uint64_t prime) {
+  uint64_t l, h;
+  l = mul_64x64_128(v, prime, &h);
+  return l ^ h;
+}
+
+static __maybe_unused __always_inline uint64_t final64(uint64_t a, uint64_t b) {
+  uint64_t x = (a + rot64(b, 41)) * prime_0;
+  uint64_t y = (rot64(a, 23) + b) * prime_6;
+  return mux64(x ^ y, prime_5);
+}
+
+static __maybe_unused __always_inline void mixup64(uint64_t *__restrict a,
+                                                   uint64_t *__restrict b,
+                                                   uint64_t v, uint64_t prime) {
+  uint64_t h;
+  *a ^= mul_64x64_128(*b + v, prime, &h);
+  *b += h;
+}
+
+/***************************************************************************/
+
+typedef union t1ha_uint128 {
+#if defined(__SIZEOF_INT128__) ||                                              \
+    (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
+  __uint128_t v;
+#endif
+  struct {
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+    uint64_t l, h;
+#else
+    uint64_t h, l;
+#endif
+  };
+} t1ha_uint128_t;
+
+static __maybe_unused __always_inline t1ha_uint128_t
+not128(const t1ha_uint128_t v) {
+  t1ha_uint128_t r;
+#if defined(__SIZEOF_INT128__) ||                                              \
+    (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
+  r.v = ~v.v;
+#else
+  r.l = ~v.l;
+  r.h = ~v.h;
+#endif
+  return r;
+}
+
+static __maybe_unused __always_inline t1ha_uint128_t
+left128(const t1ha_uint128_t v, unsigned s) {
+  t1ha_uint128_t r;
+  assert(s < 128);
+#if defined(__SIZEOF_INT128__) ||                                              \
+    (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
+  r.v = v.v << s;
+#else
+  r.l = (s < 64) ? v.l << s : 0;
+  r.h = (s < 64) ? (v.h << s) | (s ? v.l >> (64 - s) : 0) : v.l << (s - 64);
+#endif
+  return r;
+}
+
+static __maybe_unused __always_inline t1ha_uint128_t
+right128(const t1ha_uint128_t v, unsigned s) {
+  t1ha_uint128_t r;
+  assert(s < 128);
+#if defined(__SIZEOF_INT128__) ||                                              \
+    (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
+  r.v = v.v >> s;
+#else
+  r.l = (s < 64) ? (s ? v.h << (64 - s) : 0) | (v.l >> s) : v.h >> (s - 64);
+  r.h = (s < 64) ? v.h >> s : 0;
+#endif
+  return r;
+}
+
+static __maybe_unused __always_inline t1ha_uint128_t or128(t1ha_uint128_t x,
+                                                           t1ha_uint128_t y) {
+  t1ha_uint128_t r;
+#if defined(__SIZEOF_INT128__) ||                                              \
+    (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
+  r.v = x.v | y.v;
+#else
+  r.l = x.l | y.l;
+  r.h = x.h | y.h;
+#endif
+  return r;
+}
+
+static __maybe_unused __always_inline t1ha_uint128_t xor128(t1ha_uint128_t x,
+                                                            t1ha_uint128_t y) {
+  t1ha_uint128_t r;
+#if defined(__SIZEOF_INT128__) ||                                              \
+    (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
+  r.v = x.v ^ y.v;
+#else
+  r.l = x.l ^ y.l;
+  r.h = x.h ^ y.h;
+#endif
+  return r;
+}
+
+static __maybe_unused __always_inline t1ha_uint128_t rot128(t1ha_uint128_t v,
+                                                            unsigned s) {
+  s &= 127;
+#if defined(__SIZEOF_INT128__) ||                                              \
+    (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
+  v.v = (v.v << (128 - s)) | (v.v >> s);
+  return v;
+#else
+  return s ? or128(left128(v, 128 - s), right128(v, s)) : v;
+#endif
+}
+
+static __maybe_unused __always_inline t1ha_uint128_t add128(t1ha_uint128_t x,
+                                                            t1ha_uint128_t y) {
+  t1ha_uint128_t r;
+#if defined(__SIZEOF_INT128__) ||                                              \
+    (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
+  r.v = x.v + y.v;
+#else
+  add64carry_last(add64carry_first(x.l, y.l, &r.l), x.h, y.h, &r.h);
+#endif
+  return r;
+}
+
+static __maybe_unused __always_inline t1ha_uint128_t mul128(t1ha_uint128_t x,
+                                                            t1ha_uint128_t y) {
+  t1ha_uint128_t r;
+#if defined(__SIZEOF_INT128__) ||                                              \
+    (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
+  r.v = x.v * y.v;
+#else
+  r.l = mul_64x64_128(x.l, y.l, &r.h);
+  r.h += x.l * y.h + y.l * x.h;
+#endif
+  return r;
+}
+
+/***************************************************************************/
+
+#if T1HA0_AESNI_AVAILABLE && defined(__ia32__)
+uint64_t t1ha_ia32cpu_features(void);
+
+static __maybe_unused __always_inline bool
+t1ha_ia32_AESNI_avail(uint64_t ia32cpu_features) {
+  /* check for AES-NI */
+  return (ia32cpu_features & UINT32_C(0x02000000)) != 0;
+}
+
+static __maybe_unused __always_inline bool
+t1ha_ia32_AVX_avail(uint64_t ia32cpu_features) {
+  /* check for any AVX */
+  return (ia32cpu_features & UINT32_C(0x1A000000)) == UINT32_C(0x1A000000);
+}
+
+static __maybe_unused __always_inline bool
+t1ha_ia32_AVX2_avail(uint64_t ia32cpu_features) {
+  /* check for 'Advanced Vector Extensions 2' */
+  return ((ia32cpu_features >> 32) & 32) != 0;
+}
+
+#endif /* T1HA0_AESNI_AVAILABLE && __ia32__ */
diff --git a/include/t1ha_selfcheck.h b/include/t1ha_selfcheck.h
new file mode 100644
index 00000000..ff7c589c
--- /dev/null
+++ b/include/t1ha_selfcheck.h
@@ -0,0 +1,76 @@
+/*
+ *  Copyright (c) 2016-2020 Positive Technologies, https://www.ptsecurity.com,
+ *  Fast Positive Hash.
+ *
+ *  Portions Copyright (c) 2010-2020 Leonid Yuriev <leo@yuriev.ru>,
+ *  The 1Hippeus project (t1h).
+ *
+ *  This software is provided 'as-is', without any express or implied
+ *  warranty. In no event will the authors be held liable for any damages
+ *  arising from the use of this software.
+ *
+ *  Permission is granted to anyone to use this software for any purpose,
+ *  including commercial applications, and to alter it and redistribute it
+ *  freely, subject to the following restrictions:
+ *
+ *  1. The origin of this software must not be misrepresented; you must not
+ *     claim that you wrote the original software. If you use this software
+ *     in a product, an acknowledgement in the product documentation would be
+ *     appreciated but is not required.
+ *  2. Altered source versions must be plainly marked as such, and must not be
+ *     misrepresented as being the original software.
+ *  3. This notice may not be removed or altered from any source distribution.
+ */
+
+/*
+ * t1ha = { Fast Positive Hash, aka "Позитивный Хэш" }
+ * by [Positive Technologies](https://www.ptsecurity.ru)
+ *
+ * Briefly, it is a 64-bit Hash Function:
+ *  1. Created for 64-bit little-endian platforms, in predominantly for x86_64,
+ *     but portable and without penalties it can run on any 64-bit CPU.
+ *  2. In most cases up to 15% faster than City64, xxHash, mum-hash, metro-hash
+ *     and all others portable hash-functions (which do not use specific
+ *     hardware tricks).
+ *  3. Not suitable for cryptography.
+ *
+ * The Future will (be) Positive. Всё будет хорошо.
+ *
+ * ACKNOWLEDGEMENT:
+ * The t1ha was originally developed by Leonid Yuriev (Леонид Юрьев)
+ * for The 1Hippeus project - zerocopy messaging in the spirit of Sparta!
+ */
+
+#pragma once
+#if defined(_MSC_VER) && _MSC_VER > 1800
+#pragma warning(disable : 4464) /* relative include path contains '..' */
+#endif                          /* MSVC */
+#include "t1ha.h"
+
+/***************************************************************************/
+/* Self-checking */
+
+extern const uint8_t t1ha_test_pattern[64];
+int t1ha_selfcheck(uint64_t (*hash)(const void *, size_t, uint64_t),
+                   const uint64_t *reference_values);
+
+#ifndef T1HA2_DISABLED
+extern const uint64_t t1ha_refval_2atonce[81];
+extern const uint64_t t1ha_refval_2atonce128[81];
+extern const uint64_t t1ha_refval_2stream[81];
+extern const uint64_t t1ha_refval_2stream128[81];
+#endif /* T1HA2_DISABLED */
+
+#ifndef T1HA1_DISABLED
+extern const uint64_t t1ha_refval_64le[81];
+extern const uint64_t t1ha_refval_64be[81];
+#endif /* T1HA1_DISABLED */
+
+#ifndef T1HA0_DISABLED
+extern const uint64_t t1ha_refval_32le[81];
+extern const uint64_t t1ha_refval_32be[81];
+#if T1HA0_AESNI_AVAILABLE
+extern const uint64_t t1ha_refval_ia32aes_a[81];
+extern const uint64_t t1ha_refval_ia32aes_b[81];
+#endif /* T1HA0_AESNI_AVAILABLE */
+#endif /* T1HA0_DISABLED */
diff --git a/include/xxhash.h b/include/xxhash.h
index 9a880470..d11f0f63 100644
--- a/include/xxhash.h
+++ b/include/xxhash.h
@@ -1,7 +1,7 @@
 /*
  * xxHash - Extremely Fast Hash algorithm
  * Header File
- * Copyright (C) 2012-2024 Yann Collet
+ * Copyright (C) 2012-2023 Yann Collet
  *
  * BSD 2-Clause License (https://www.opensource.org/licenses/bsd-license.php)
  *
@@ -32,328 +32,553 @@
  *   - xxHash homepage: https://www.xxhash.com
  *   - xxHash source repository: https://github.com/Cyan4973/xxHash
  */
+
 /*!
  * @mainpage xxHash
  *
+ * xxHash is an extremely fast non-cryptographic hash algorithm, working at RAM speed
+ * limits.
+ *
+ * It is proposed in four flavors, in three families:
+ * 1. @ref XXH32_family
+ *   - Classic 32-bit hash function. Simple, compact, and runs on almost all
+ *     32-bit and 64-bit systems.
+ * 2. @ref XXH64_family
+ *   - Classic 64-bit adaptation of XXH32. Just as simple, and runs well on most
+ *     64-bit systems (but _not_ 32-bit systems).
+ * 3. @ref XXH3_family
+ *   - Modern 64-bit and 128-bit hash function family which features improved
+ *     strength and performance across the board, especially on smaller data.
+ *     It benefits greatly from SIMD and 64-bit without requiring it.
+ *
+ * Benchmarks
+ * ---
+ * The reference system uses an Intel i7-9700K CPU, and runs Ubuntu x64 20.04.
+ * The open source benchmark program is compiled with clang v10.0 using -O3 flag.
+ *
+ * | Hash Name            | ISA ext | Width | Large Data Speed | Small Data Velocity |
+ * | -------------------- | ------- | ----: | ---------------: | ------------------: |
+ * | XXH3_64bits()        | @b AVX2 |    64 |        59.4 GB/s |               133.1 |
+ * | MeowHash             | AES-NI  |   128 |        58.2 GB/s |                52.5 |
+ * | XXH3_128bits()       | @b AVX2 |   128 |        57.9 GB/s |               118.1 |
+ * | CLHash               | PCLMUL  |    64 |        37.1 GB/s |                58.1 |
+ * | XXH3_64bits()        | @b SSE2 |    64 |        31.5 GB/s |               133.1 |
+ * | XXH3_128bits()       | @b SSE2 |   128 |        29.6 GB/s |               118.1 |
+ * | RAM sequential read  |         |   N/A |        28.0 GB/s |                 N/A |
+ * | ahash                | AES-NI  |    64 |        22.5 GB/s |               107.2 |
+ * | City64               |         |    64 |        22.0 GB/s |                76.6 |
+ * | T1ha2                |         |    64 |        22.0 GB/s |                99.0 |
+ * | City128              |         |   128 |        21.7 GB/s |                57.7 |
+ * | FarmHash             | AES-NI  |    64 |        21.3 GB/s |                71.9 |
+ * | XXH64()              |         |    64 |        19.4 GB/s |                71.0 |
+ * | SpookyHash           |         |    64 |        19.3 GB/s |                53.2 |
+ * | Mum                  |         |    64 |        18.0 GB/s |                67.0 |
+ * | CRC32C               | SSE4.2  |    32 |        13.0 GB/s |                57.9 |
+ * | XXH32()              |         |    32 |         9.7 GB/s |                71.9 |
+ * | City32               |         |    32 |         9.1 GB/s |                66.0 |
+ * | Blake3*              | @b AVX2 |   256 |         4.4 GB/s |                 8.1 |
+ * | Murmur3              |         |    32 |         3.9 GB/s |                56.1 |
+ * | SipHash*             |         |    64 |         3.0 GB/s |                43.2 |
+ * | Blake3*              | @b SSE2 |   256 |         2.4 GB/s |                 8.1 |
+ * | HighwayHash          |         |    64 |         1.4 GB/s |                 6.0 |
+ * | FNV64                |         |    64 |         1.2 GB/s |                62.7 |
+ * | Blake2*              |         |   256 |         1.1 GB/s |                 5.1 |
+ * | SHA1*                |         |   160 |         0.8 GB/s |                 5.6 |
+ * | MD5*                 |         |   128 |         0.6 GB/s |                 7.8 |
+ * @note
+ *   - Hashes which require a specific ISA extension are noted. SSE2 is also noted,
+ *     even though it is mandatory on x64.
+ *   - Hashes with an asterisk are cryptographic. Note that MD5 is non-cryptographic
+ *     by modern standards.
+ *   - Small data velocity is a rough average of algorithm's efficiency for small
+ *     data. For more accurate information, see the wiki.
+ *   - More benchmarks and strength tests are found on the wiki:
+ *         https://github.com/Cyan4973/xxHash/wiki
+ *
+ * Usage
+ * ------
+ * All xxHash variants use a similar API. Changing the algorithm is a trivial
+ * substitution.
+ *
+ * @pre
+ *    For functions which take an input and length parameter, the following
+ *    requirements are assumed:
+ *    - The range from [`input`, `input + length`) is valid, readable memory.
+ *      - The only exception is if the `length` is `0`, `input` may be `NULL`.
+ *    - For C++, the objects must have the *TriviallyCopyable* property, as the
+ *      functions access bytes directly as if it was an array of `unsigned char`.
+ *
+ * @anchor single_shot_example
+ * **Single Shot**
+ *
+ * These functions are stateless functions which hash a contiguous block of memory,
+ * immediately returning the result. They are the easiest and usually the fastest
+ * option.
+ *
+ * XXH32(), XXH64(), XXH3_64bits(), XXH3_128bits()
+ *
+ * @code{.c}
+ *   #include <string.h>
+ *   #include "xxhash.h"
+ *
+ *   // Example for a function which hashes a null terminated string with XXH32().
+ *   XXH32_hash_t hash_string(const char* string, XXH32_hash_t seed)
+ *   {
+ *       // NULL pointers are only valid if the length is zero
+ *       size_t length = (string == NULL) ? 0 : strlen(string);
+ *       return XXH32(string, length, seed);
+ *   }
+ * @endcode
+ *
+ *
+ * @anchor streaming_example
+ * **Streaming**
+ *
+ * These groups of functions allow incremental hashing of unknown size, even
+ * more than what would fit in a size_t.
+ *
+ * XXH32_reset(), XXH64_reset(), XXH3_64bits_reset(), XXH3_128bits_reset()
+ *
+ * @code{.c}
+ *   #include <stdio.h>
+ *   #include <assert.h>
+ *   #include "xxhash.h"
+ *   // Example for a function which hashes a FILE incrementally with XXH3_64bits().
+ *   XXH64_hash_t hashFile(FILE* f)
+ *   {
+ *       // Allocate a state struct. Do not just use malloc() or new.
+ *       XXH3_state_t* state = XXH3_createState();
+ *       assert(state != NULL && "Out of memory!");
+ *       // Reset the state to start a new hashing session.
+ *       XXH3_64bits_reset(state);
+ *       char buffer[4096];
+ *       size_t count;
+ *       // Read the file in chunks
+ *       while ((count = fread(buffer, 1, sizeof(buffer), f)) != 0) {
+ *           // Run update() as many times as necessary to process the data
+ *           XXH3_64bits_update(state, buffer, count);
+ *       }
+ *       // Retrieve the finalized hash. This will not change the state.
+ *       XXH64_hash_t result = XXH3_64bits_digest(state);
+ *       // Free the state. Do not use free().
+ *       XXH3_freeState(state);
+ *       return result;
+ *   }
+ * @endcode
+ *
+ * Streaming functions generate the xxHash value from an incremental input.
+ * This method is slower than single-call functions, due to state management.
+ * For small inputs, prefer `XXH32()` and `XXH64()`, which are better optimized.
+ *
+ * An XXH state must first be allocated using `XXH*_createState()`.
+ *
+ * Start a new hash by initializing the state with a seed using `XXH*_reset()`.
+ *
+ * Then, feed the hash state by calling `XXH*_update()` as many times as necessary.
+ *
+ * The function returns an error code, with 0 meaning OK, and any other value
+ * meaning there is an error.
+ *
+ * Finally, a hash value can be produced anytime, by using `XXH*_digest()`.
+ * This function returns the nn-bits hash as an int or long long.
+ *
+ * It's still possible to continue inserting input into the hash state after a
+ * digest, and generate new hash values later on by invoking `XXH*_digest()`.
+ *
+ * When done, release the state using `XXH*_freeState()`.
+ *
+ *
+ * @anchor canonical_representation_example
+ * **Canonical Representation**
+ *
+ * The default return values from XXH functions are unsigned 32, 64 and 128 bit
+ * integers.
+ * This the simplest and fastest format for further post-processing.
+ *
+ * However, this leaves open the question of what is the order on the byte level,
+ * since little and big endian conventions will store the same number differently.
+ *
+ * The canonical representation settles this issue by mandating big-endian
+ * convention, the same convention as human-readable numbers (large digits first).
+ *
+ * When writing hash values to storage, sending them over a network, or printing
+ * them, it's highly recommended to use the canonical representation to ensure
+ * portability across a wider range of systems, present and future.
+ *
+ * The following functions allow transformation of hash values to and from
+ * canonical format.
+ *
+ * XXH32_canonicalFromHash(), XXH32_hashFromCanonical(),
+ * XXH64_canonicalFromHash(), XXH64_hashFromCanonical(),
+ * XXH128_canonicalFromHash(), XXH128_hashFromCanonical(),
+ *
+ * @code{.c}
+ *   #include <stdio.h>
+ *   #include "xxhash.h"
+ *
+ *   // Example for a function which prints XXH32_hash_t in human readable format
+ *   void printXxh32(XXH32_hash_t hash)
+ *   {
+ *       XXH32_canonical_t cano;
+ *       XXH32_canonicalFromHash(&cano, hash);
+ *       size_t i;
+ *       for(i = 0; i < sizeof(cano.digest); ++i) {
+ *           printf("%02x", cano.digest[i]);
+ *       }
+ *       printf("\n");
+ *   }
+ *
+ *   // Example for a function which converts XXH32_canonical_t to XXH32_hash_t
+ *   XXH32_hash_t convertCanonicalToXxh32(XXH32_canonical_t cano)
+ *   {
+ *       XXH32_hash_t hash = XXH32_hashFromCanonical(&cano);
+ *       return hash;
+ *   }
+ * @endcode
+ *
+ *
  * @file xxhash.h
  * xxHash prototypes and implementation
  */
-/* TODO: update */
-/* Notice extracted from xxHash homepage:
-
-xxHash is an extremely fast hash algorithm, running at RAM speed limits.
-It also successfully passes all tests from the SMHasher suite.
-
-Comparison (single thread, Windows Seven 32 bits, using SMHasher on a Core 2 Duo
-@3GHz)
-
-Name            Speed       Q.Score   Author
-xxHash          5.4 GB/s     10
-CrapWow         3.2 GB/s      2       Andrew
-MurmurHash 3a   2.7 GB/s     10       Austin Appleby
-SpookyHash      2.0 GB/s     10       Bob Jenkins
-SBox            1.4 GB/s      9       Bret Mulvey
-Lookup3         1.2 GB/s      9       Bob Jenkins
-SuperFastHash   1.2 GB/s      1       Paul Hsieh
-CityHash64      1.05 GB/s    10       Pike & Alakuijala
-FNV             0.55 GB/s     5       Fowler, Noll, Vo
-CRC32           0.43 GB/s     9
-MD5-32          0.33 GB/s    10       Ronald L. Rivest
-SHA1-32         0.28 GB/s    10
-
-Q.Score is a measure of quality of the hash function.
-It depends on successfully passing SMHasher test set.
-10 is a perfect score.
-
-Note: SMHasher's CRC32 implementation is not the fastest one.
-Other speed-oriented implementations can be faster,
-especially in combination with PCLMUL instruction:
-https://fastcompression.blogspot.com/2019/03/presenting-xxh3.html?showComment=1552696407071#c3490092340461170735
-
-A 64-bit version, named XXH64, is available since r35.
-It offers much better speed, but for 64-bit applications only.
-Name     Speed on 64 bits    Speed on 32 bits
-XXH64       13.8 GB/s            1.9 GB/s
-XXH32        6.8 GB/s            6.0 GB/s
-*/
-
-#if defined(__cplusplus)
-extern "C" {
 
+#if defined (__cplusplus)
+extern "C" {
 #endif
 
 /* ****************************
  *  INLINE mode
  ******************************/
 /*!
- * XXH_INLINE_ALL (and XXH_PRIVATE_API)
+ * @defgroup public Public API
+ * Contains details on the public xxHash functions.
+ * @{
+ */
+#ifdef XXH_DOXYGEN
+/*!
+ * @brief Gives access to internal state declaration, required for static allocation.
+ *
+ * Incompatible with dynamic linking, due to risks of ABI changes.
+ *
+ * Usage:
+ * @code{.c}
+ *     #define XXH_STATIC_LINKING_ONLY
+ *     #include "xxhash.h"
+ * @endcode
+ */
+#  define XXH_STATIC_LINKING_ONLY
+/* Do not undef XXH_STATIC_LINKING_ONLY for Doxygen */
+
+/*!
+ * @brief Gives access to internal definitions.
+ *
+ * Usage:
+ * @code{.c}
+ *     #define XXH_STATIC_LINKING_ONLY
+ *     #define XXH_IMPLEMENTATION
+ *     #include "xxhash.h"
+ * @endcode
+ */
+#  define XXH_IMPLEMENTATION
+/* Do not undef XXH_IMPLEMENTATION for Doxygen */
+
+/*!
+ * @brief Exposes the implementation and marks all functions as `inline`.
+ *
  * Use these build macros to inline xxhash into the target unit.
  * Inlining improves performance on small inputs, especially when the length is
  * expressed as a compile-time constant:
  *
- *      https://fastcompression.blogspot.com/2018/03/xxhash-for-small-keys-impressive-power.html
+ *  https://fastcompression.blogspot.com/2018/03/xxhash-for-small-keys-impressive-power.html
  *
  * It also keeps xxHash symbols private to the unit, so they are not exported.
  *
  * Usage:
+ * @code{.c}
  *     #define XXH_INLINE_ALL
  *     #include "xxhash.h"
- *
+ * @endcode
  * Do not compile and link xxhash.o as a separate object, as it is not useful.
  */
-#if (defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API)) && \
-    !defined(XXH_INLINE_ALL_31684351384)
-/* this section should be traversed only once */
-  #define XXH_INLINE_ALL_31684351384
-/* give access to the advanced API, required to compile implementations */
-  #undef XXH_STATIC_LINKING_ONLY                       /* avoid macro redef */
-  #define XXH_STATIC_LINKING_ONLY
-/* make all functions private */
-  #undef XXH_PUBLIC_API
-  #if defined(__GNUC__)
-    #define XXH_PUBLIC_API static __inline __attribute__((unused))
-  #elif defined(__cplusplus) || \
-      (defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
-    #define XXH_PUBLIC_API static inline
-  #elif defined(_MSC_VER)
-    #define XXH_PUBLIC_API static __inline
-  #else
-  /* note: this version may generate warnings for unused static functions */
-    #define XXH_PUBLIC_API static
-  #endif
+#  define XXH_INLINE_ALL
+#  undef XXH_INLINE_ALL
+/*!
+ * @brief Exposes the implementation without marking functions as inline.
+ */
+#  define XXH_PRIVATE_API
+#  undef XXH_PRIVATE_API
+/*!
+ * @brief Emulate a namespace by transparently prefixing all symbols.
+ *
+ * If you want to include _and expose_ xxHash functions from within your own
+ * library, but also want to avoid symbol collisions with other libraries which
+ * may also include xxHash, you can use @ref XXH_NAMESPACE to automatically prefix
+ * any public symbol from xxhash library with the value of @ref XXH_NAMESPACE
+ * (therefore, avoid empty or numeric values).
+ *
+ * Note that no change is required within the calling program as long as it
+ * includes `xxhash.h`: Regular symbol names will be automatically translated
+ * by this header.
+ */
+#  define XXH_NAMESPACE /* YOUR NAME HERE */
+#  undef XXH_NAMESPACE
+#endif
 
-/*
- * This part deals with the special case where a unit wants to inline xxHash,
- * but "xxhash.h" has previously been included without XXH_INLINE_ALL,
- * such as part of some previously included *.h header file.
- * Without further action, the new include would just be ignored,
- * and functions would effectively _not_ be inlined (silent failure).
- * The following macros solve this situation by prefixing all inlined names,
- * avoiding naming collision with previous inclusions.
- */
-/* Before that, we unconditionally #undef all symbols,
- * in case they were already defined with XXH_NAMESPACE.
- * They will then be redefined for XXH_INLINE_ALL
- */
-  #undef XXH_versionNumber
-/* XXH32 */
-  #undef XXH32
-  #undef XXH32_createState
-  #undef XXH32_freeState
-  #undef XXH32_reset
-  #undef XXH32_update
-  #undef XXH32_digest
-  #undef XXH32_copyState
-  #undef XXH32_canonicalFromHash
-  #undef XXH32_hashFromCanonical
-/* XXH64 */
-  #undef XXH64
-  #undef XXH64_createState
-  #undef XXH64_freeState
-  #undef XXH64_reset
-  #undef XXH64_update
-  #undef XXH64_digest
-  #undef XXH64_copyState
-  #undef XXH64_canonicalFromHash
-  #undef XXH64_hashFromCanonical
-/* XXH3_64bits */
-  #undef XXH3_64bits
-  #undef XXH3_64bits_withSecret
-  #undef XXH3_64bits_withSeed
-  #undef XXH3_createState
-  #undef XXH3_freeState
-  #undef XXH3_copyState
-  #undef XXH3_64bits_reset
-  #undef XXH3_64bits_reset_withSeed
-  #undef XXH3_64bits_reset_withSecret
-  #undef XXH3_64bits_update
-  #undef XXH3_64bits_digest
-  #undef XXH3_generateSecret
-/* XXH3_128bits */
-  #undef XXH128
-  #undef XXH3_128bits
-  #undef XXH3_128bits_withSeed
-  #undef XXH3_128bits_withSecret
-  #undef XXH3_128bits_reset
-  #undef XXH3_128bits_reset_withSeed
-  #undef XXH3_128bits_reset_withSecret
-  #undef XXH3_128bits_update
-  #undef XXH3_128bits_digest
-  #undef XXH128_isEqual
-  #undef XXH128_cmp
-  #undef XXH128_canonicalFromHash
-  #undef XXH128_hashFromCanonical
-/* Finally, free the namespace itself */
-  #undef XXH_NAMESPACE
-
-/* employ the namespace for XXH_INLINE_ALL */
-  #define XXH_NAMESPACE XXH_INLINE_
-/*
- * Some identifiers (enums, type names) are not symbols,
- * but they must nonetheless be renamed to avoid redeclaration.
- * Alternative solution: do not redeclare them.
- * However, this requires some #ifdefs, and has a more dispersed impact.
- * Meanwhile, renaming can be achieved in a single place.
- */
-  #define XXH_IPREF(Id) XXH_NAMESPACE##Id
-  #define XXH_OK XXH_IPREF(XXH_OK)
-  #define XXH_ERROR XXH_IPREF(XXH_ERROR)
-  #define XXH_errorcode XXH_IPREF(XXH_errorcode)
-  #define XXH32_canonical_t XXH_IPREF(XXH32_canonical_t)
-  #define XXH64_canonical_t XXH_IPREF(XXH64_canonical_t)
-  #define XXH128_canonical_t XXH_IPREF(XXH128_canonical_t)
-  #define XXH32_state_s XXH_IPREF(XXH32_state_s)
-  #define XXH32_state_t XXH_IPREF(XXH32_state_t)
-  #define XXH64_state_s XXH_IPREF(XXH64_state_s)
-  #define XXH64_state_t XXH_IPREF(XXH64_state_t)
-  #define XXH3_state_s XXH_IPREF(XXH3_state_s)
-  #define XXH3_state_t XXH_IPREF(XXH3_state_t)
-  #define XXH128_hash_t XXH_IPREF(XXH128_hash_t)
-/* Ensure the header is parsed again, even if it was previously included */
-  #undef XXHASH_H_5627135585666179
-  #undef XXHASH_H_STATIC_13879238742
-#endif                                 /* XXH_INLINE_ALL || XXH_PRIVATE_API */
+#if (defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API)) \
+    && !defined(XXH_INLINE_ALL_31684351384)
+   /* this section should be traversed only once */
+#  define XXH_INLINE_ALL_31684351384
+   /* give access to the advanced API, required to compile implementations */
+#  undef XXH_STATIC_LINKING_ONLY   /* avoid macro redef */
+#  define XXH_STATIC_LINKING_ONLY
+   /* make all functions private */
+#  undef XXH_PUBLIC_API
+#  if defined(__GNUC__)
+#    define XXH_PUBLIC_API static __inline __attribute__((unused))
+#  elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#    define XXH_PUBLIC_API static inline
+#  elif defined(_MSC_VER)
+#    define XXH_PUBLIC_API static __inline
+#  else
+     /* note: this version may generate warnings for unused static functions */
+#    define XXH_PUBLIC_API static
+#  endif
+
+   /*
+    * This part deals with the special case where a unit wants to inline xxHash,
+    * but "xxhash.h" has previously been included without XXH_INLINE_ALL,
+    * such as part of some previously included *.h header file.
+    * Without further action, the new include would just be ignored,
+    * and functions would effectively _not_ be inlined (silent failure).
+    * The following macros solve this situation by prefixing all inlined names,
+    * avoiding naming collision with previous inclusions.
+    */
+   /* Before that, we unconditionally #undef all symbols,
+    * in case they were already defined with XXH_NAMESPACE.
+    * They will then be redefined for XXH_INLINE_ALL
+    */
+#  undef XXH_versionNumber
+    /* XXH32 */
+#  undef XXH32
+#  undef XXH32_createState
+#  undef XXH32_freeState
+#  undef XXH32_reset
+#  undef XXH32_update
+#  undef XXH32_digest
+#  undef XXH32_copyState
+#  undef XXH32_canonicalFromHash
+#  undef XXH32_hashFromCanonical
+    /* XXH64 */
+#  undef XXH64
+#  undef XXH64_createState
+#  undef XXH64_freeState
+#  undef XXH64_reset
+#  undef XXH64_update
+#  undef XXH64_digest
+#  undef XXH64_copyState
+#  undef XXH64_canonicalFromHash
+#  undef XXH64_hashFromCanonical
+    /* XXH3_64bits */
+#  undef XXH3_64bits
+#  undef XXH3_64bits_withSecret
+#  undef XXH3_64bits_withSeed
+#  undef XXH3_64bits_withSecretandSeed
+#  undef XXH3_createState
+#  undef XXH3_freeState
+#  undef XXH3_copyState
+#  undef XXH3_64bits_reset
+#  undef XXH3_64bits_reset_withSeed
+#  undef XXH3_64bits_reset_withSecret
+#  undef XXH3_64bits_update
+#  undef XXH3_64bits_digest
+#  undef XXH3_generateSecret
+    /* XXH3_128bits */
+#  undef XXH128
+#  undef XXH3_128bits
+#  undef XXH3_128bits_withSeed
+#  undef XXH3_128bits_withSecret
+#  undef XXH3_128bits_reset
+#  undef XXH3_128bits_reset_withSeed
+#  undef XXH3_128bits_reset_withSecret
+#  undef XXH3_128bits_reset_withSecretandSeed
+#  undef XXH3_128bits_update
+#  undef XXH3_128bits_digest
+#  undef XXH128_isEqual
+#  undef XXH128_cmp
+#  undef XXH128_canonicalFromHash
+#  undef XXH128_hashFromCanonical
+    /* Finally, free the namespace itself */
+#  undef XXH_NAMESPACE
+
+    /* employ the namespace for XXH_INLINE_ALL */
+#  define XXH_NAMESPACE XXH_INLINE_
+   /*
+    * Some identifiers (enums, type names) are not symbols,
+    * but they must nonetheless be renamed to avoid redeclaration.
+    * Alternative solution: do not redeclare them.
+    * However, this requires some #ifdefs, and has a more dispersed impact.
+    * Meanwhile, renaming can be achieved in a single place.
+    */
+#  define XXH_IPREF(Id)   XXH_NAMESPACE ## Id
+#  define XXH_OK XXH_IPREF(XXH_OK)
+#  define XXH_ERROR XXH_IPREF(XXH_ERROR)
+#  define XXH_errorcode XXH_IPREF(XXH_errorcode)
+#  define XXH32_canonical_t  XXH_IPREF(XXH32_canonical_t)
+#  define XXH64_canonical_t  XXH_IPREF(XXH64_canonical_t)
+#  define XXH128_canonical_t XXH_IPREF(XXH128_canonical_t)
+#  define XXH32_state_s XXH_IPREF(XXH32_state_s)
+#  define XXH32_state_t XXH_IPREF(XXH32_state_t)
+#  define XXH64_state_s XXH_IPREF(XXH64_state_s)
+#  define XXH64_state_t XXH_IPREF(XXH64_state_t)
+#  define XXH3_state_s  XXH_IPREF(XXH3_state_s)
+#  define XXH3_state_t  XXH_IPREF(XXH3_state_t)
+#  define XXH128_hash_t XXH_IPREF(XXH128_hash_t)
+   /* Ensure the header is parsed again, even if it was previously included */
+#  undef XXHASH_H_5627135585666179
+#  undef XXHASH_H_STATIC_13879238742
+#endif /* XXH_INLINE_ALL || XXH_PRIVATE_API */
 
 /* ****************************************************************
  *  Stable API
  *****************************************************************/
 #ifndef XXHASH_H_5627135585666179
-  #define XXHASH_H_5627135585666179 1
-
-  /*!
-   * @defgroup public Public API
-   * Contains details on the public xxHash functions.
-   * @{
-
-   */
-  /* specific declaration modes for Windows */
-  #if !defined(XXH_INLINE_ALL) && !defined(XXH_PRIVATE_API)
-    #if defined(WIN32) && defined(_MSC_VER) && \
-        (defined(XXH_IMPORT) || defined(XXH_EXPORT))
-      #ifdef XXH_EXPORT
-        #define XXH_PUBLIC_API __declspec(dllexport)
-      #elif XXH_IMPORT
-        #define XXH_PUBLIC_API __declspec(dllimport)
-      #endif
-    #else
-      #define XXH_PUBLIC_API                                  /* do nothing */
-    #endif
-  #endif
-
-  #ifdef XXH_DOXYGEN
-    /*!
-     * @brief Emulate a namespace by transparently prefixing all symbols.
-     *
-     * If you want to include _and expose_ xxHash functions from within your own
-     * library, but also want to avoid symbol collisions with other libraries
-     * which may also include xxHash, you can use XXH_NAMESPACE to automatically
-     * prefix any public symbol from xxhash library with the value of
-     * XXH_NAMESPACE (therefore, avoid empty or numeric values).
-     *
-     * Note that no change is required within the calling program as long as it
-     * includes `xxhash.h`: Regular symbol names will be automatically
-     * translated by this header.
-     */
-    #define XXH_NAMESPACE                                 /* YOUR NAME HERE */
-    #undef XXH_NAMESPACE
-  #endif
-
-  #ifdef XXH_NAMESPACE
-    #define XXH_CAT(A, B) A##B
-    #define XXH_NAME2(A, B) XXH_CAT(A, B)
-    #define XXH_versionNumber XXH_NAME2(XXH_NAMESPACE, XXH_versionNumber)
-    /* XXH32 */
-    #define XXH32 XXH_NAME2(XXH_NAMESPACE, XXH32)
-    #define XXH32_createState XXH_NAME2(XXH_NAMESPACE, XXH32_createState)
-    #define XXH32_freeState XXH_NAME2(XXH_NAMESPACE, XXH32_freeState)
-    #define XXH32_reset XXH_NAME2(XXH_NAMESPACE, XXH32_reset)
-    #define XXH32_update XXH_NAME2(XXH_NAMESPACE, XXH32_update)
-    #define XXH32_digest XXH_NAME2(XXH_NAMESPACE, XXH32_digest)
-    #define XXH32_copyState XXH_NAME2(XXH_NAMESPACE, XXH32_copyState)
-    #define XXH32_canonicalFromHash \
-      XXH_NAME2(XXH_NAMESPACE, XXH32_canonicalFromHash)
-    #define XXH32_hashFromCanonical \
-      XXH_NAME2(XXH_NAMESPACE, XXH32_hashFromCanonical)
-    /* XXH64 */
-    #define XXH64 XXH_NAME2(XXH_NAMESPACE, XXH64)
-    #define XXH64_createState XXH_NAME2(XXH_NAMESPACE, XXH64_createState)
-    #define XXH64_freeState XXH_NAME2(XXH_NAMESPACE, XXH64_freeState)
-    #define XXH64_reset XXH_NAME2(XXH_NAMESPACE, XXH64_reset)
-    #define XXH64_update XXH_NAME2(XXH_NAMESPACE, XXH64_update)
-    #define XXH64_digest XXH_NAME2(XXH_NAMESPACE, XXH64_digest)
-    #define XXH64_copyState XXH_NAME2(XXH_NAMESPACE, XXH64_copyState)
-    #define XXH64_canonicalFromHash \
-      XXH_NAME2(XXH_NAMESPACE, XXH64_canonicalFromHash)
-    #define XXH64_hashFromCanonical \
-      XXH_NAME2(XXH_NAMESPACE, XXH64_hashFromCanonical)
-    /* XXH3_64bits */
-    #define XXH3_64bits XXH_NAME2(XXH_NAMESPACE, XXH3_64bits)
-    #define XXH3_64bits_withSecret \
-      XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSecret)
-    #define XXH3_64bits_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSeed)
-    #define XXH3_createState XXH_NAME2(XXH_NAMESPACE, XXH3_createState)
-    #define XXH3_freeState XXH_NAME2(XXH_NAMESPACE, XXH3_freeState)
-    #define XXH3_copyState XXH_NAME2(XXH_NAMESPACE, XXH3_copyState)
-    #define XXH3_64bits_reset XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset)
-    #define XXH3_64bits_reset_withSeed \
-      XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSeed)
-    #define XXH3_64bits_reset_withSecret \
-      XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSecret)
-    #define XXH3_64bits_update XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_update)
-    #define XXH3_64bits_digest XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_digest)
-    #define XXH3_generateSecret XXH_NAME2(XXH_NAMESPACE, XXH3_generateSecret)
-    /* XXH3_128bits */
-    #define XXH128 XXH_NAME2(XXH_NAMESPACE, XXH128)
-    #define XXH3_128bits XXH_NAME2(XXH_NAMESPACE, XXH3_128bits)
-    #define XXH3_128bits_withSeed \
-      XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSeed)
-    #define XXH3_128bits_withSecret \
-      XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSecret)
-    #define XXH3_128bits_reset XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset)
-    #define XXH3_128bits_reset_withSeed \
-      XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSeed)
-    #define XXH3_128bits_reset_withSecret \
-      XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSecret)
-    #define XXH3_128bits_update XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_update)
-    #define XXH3_128bits_digest XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_digest)
-    #define XXH128_isEqual XXH_NAME2(XXH_NAMESPACE, XXH128_isEqual)
-    #define XXH128_cmp XXH_NAME2(XXH_NAMESPACE, XXH128_cmp)
-    #define XXH128_canonicalFromHash \
-      XXH_NAME2(XXH_NAMESPACE, XXH128_canonicalFromHash)
-    #define XXH128_hashFromCanonical \
-      XXH_NAME2(XXH_NAMESPACE, XXH128_hashFromCanonical)
-  #endif
-
-  /* *************************************
-   *  Version
-   ***************************************/
-  #define XXH_VERSION_MAJOR 0
-  #define XXH_VERSION_MINOR 8
-  #define XXH_VERSION_RELEASE 1
-  #define XXH_VERSION_NUMBER                                   \
-    (XXH_VERSION_MAJOR * 100 * 100 + XXH_VERSION_MINOR * 100 + \
-     XXH_VERSION_RELEASE)
+#define XXHASH_H_5627135585666179 1
+
+/*! @brief Marks a global symbol. */
+#if !defined(XXH_INLINE_ALL) && !defined(XXH_PRIVATE_API)
+#  if defined(WIN32) && defined(_MSC_VER) && (defined(XXH_IMPORT) || defined(XXH_EXPORT))
+#    ifdef XXH_EXPORT
+#      define XXH_PUBLIC_API __declspec(dllexport)
+#    elif XXH_IMPORT
+#      define XXH_PUBLIC_API __declspec(dllimport)
+#    endif
+#  else
+#    define XXH_PUBLIC_API   /* do nothing */
+#  endif
+#endif
+
+#ifdef XXH_NAMESPACE
+#  define XXH_CAT(A,B) A##B
+#  define XXH_NAME2(A,B) XXH_CAT(A,B)
+#  define XXH_versionNumber XXH_NAME2(XXH_NAMESPACE, XXH_versionNumber)
+/* XXH32 */
+#  define XXH32 XXH_NAME2(XXH_NAMESPACE, XXH32)
+#  define XXH32_createState XXH_NAME2(XXH_NAMESPACE, XXH32_createState)
+#  define XXH32_freeState XXH_NAME2(XXH_NAMESPACE, XXH32_freeState)
+#  define XXH32_reset XXH_NAME2(XXH_NAMESPACE, XXH32_reset)
+#  define XXH32_update XXH_NAME2(XXH_NAMESPACE, XXH32_update)
+#  define XXH32_digest XXH_NAME2(XXH_NAMESPACE, XXH32_digest)
+#  define XXH32_copyState XXH_NAME2(XXH_NAMESPACE, XXH32_copyState)
+#  define XXH32_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH32_canonicalFromHash)
+#  define XXH32_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH32_hashFromCanonical)
+/* XXH64 */
+#  define XXH64 XXH_NAME2(XXH_NAMESPACE, XXH64)
+#  define XXH64_createState XXH_NAME2(XXH_NAMESPACE, XXH64_createState)
+#  define XXH64_freeState XXH_NAME2(XXH_NAMESPACE, XXH64_freeState)
+#  define XXH64_reset XXH_NAME2(XXH_NAMESPACE, XXH64_reset)
+#  define XXH64_update XXH_NAME2(XXH_NAMESPACE, XXH64_update)
+#  define XXH64_digest XXH_NAME2(XXH_NAMESPACE, XXH64_digest)
+#  define XXH64_copyState XXH_NAME2(XXH_NAMESPACE, XXH64_copyState)
+#  define XXH64_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH64_canonicalFromHash)
+#  define XXH64_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH64_hashFromCanonical)
+/* XXH3_64bits */
+#  define XXH3_64bits XXH_NAME2(XXH_NAMESPACE, XXH3_64bits)
+#  define XXH3_64bits_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSecret)
+#  define XXH3_64bits_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSeed)
+#  define XXH3_64bits_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSecretandSeed)
+#  define XXH3_createState XXH_NAME2(XXH_NAMESPACE, XXH3_createState)
+#  define XXH3_freeState XXH_NAME2(XXH_NAMESPACE, XXH3_freeState)
+#  define XXH3_copyState XXH_NAME2(XXH_NAMESPACE, XXH3_copyState)
+#  define XXH3_64bits_reset XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset)
+#  define XXH3_64bits_reset_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSeed)
+#  define XXH3_64bits_reset_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSecret)
+#  define XXH3_64bits_reset_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSecretandSeed)
+#  define XXH3_64bits_update XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_update)
+#  define XXH3_64bits_digest XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_digest)
+#  define XXH3_generateSecret XXH_NAME2(XXH_NAMESPACE, XXH3_generateSecret)
+#  define XXH3_generateSecret_fromSeed XXH_NAME2(XXH_NAMESPACE, XXH3_generateSecret_fromSeed)
+/* XXH3_128bits */
+#  define XXH128 XXH_NAME2(XXH_NAMESPACE, XXH128)
+#  define XXH3_128bits XXH_NAME2(XXH_NAMESPACE, XXH3_128bits)
+#  define XXH3_128bits_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSeed)
+#  define XXH3_128bits_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSecret)
+#  define XXH3_128bits_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSecretandSeed)
+#  define XXH3_128bits_reset XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset)
+#  define XXH3_128bits_reset_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSeed)
+#  define XXH3_128bits_reset_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSecret)
+#  define XXH3_128bits_reset_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSecretandSeed)
+#  define XXH3_128bits_update XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_update)
+#  define XXH3_128bits_digest XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_digest)
+#  define XXH128_isEqual XXH_NAME2(XXH_NAMESPACE, XXH128_isEqual)
+#  define XXH128_cmp     XXH_NAME2(XXH_NAMESPACE, XXH128_cmp)
+#  define XXH128_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH128_canonicalFromHash)
+#  define XXH128_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH128_hashFromCanonical)
+#endif
+
+
+/* *************************************
+*  Compiler specifics
+***************************************/
+
+/* specific declaration modes for Windows */
+#if !defined(XXH_INLINE_ALL) && !defined(XXH_PRIVATE_API)
+#  if defined(WIN32) && defined(_MSC_VER) && (defined(XXH_IMPORT) || defined(XXH_EXPORT))
+#    ifdef XXH_EXPORT
+#      define XXH_PUBLIC_API __declspec(dllexport)
+#    elif XXH_IMPORT
+#      define XXH_PUBLIC_API __declspec(dllimport)
+#    endif
+#  else
+#    define XXH_PUBLIC_API   /* do nothing */
+#  endif
+#endif
+
+#if defined (__GNUC__)
+# define XXH_CONSTF  __attribute__((const))
+# define XXH_PUREF   __attribute__((pure))
+# define XXH_MALLOCF __attribute__((malloc))
+#else
+# define XXH_CONSTF  /* disable */
+# define XXH_PUREF
+# define XXH_MALLOCF
+#endif
+
+/* *************************************
+*  Version
+***************************************/
+#define XXH_VERSION_MAJOR    0
+#define XXH_VERSION_MINOR    8
+#define XXH_VERSION_RELEASE  2
+/*! @brief Version number, encoded as two digits each */
+#define XXH_VERSION_NUMBER  (XXH_VERSION_MAJOR *100*100 + XXH_VERSION_MINOR *100 + XXH_VERSION_RELEASE)
 
 /*!
  * @brief Obtains the xxHash version.
  *
- * This is only useful when xxHash is compiled as a shared library, as it is
- * independent of the version defined in the header.
+ * This is mostly useful when xxHash is compiled as a shared library,
+ * since the returned value comes from the library, as opposed to header file.
  *
- * @return `XXH_VERSION_NUMBER` as of when the libray was compiled.
+ * @return @ref XXH_VERSION_NUMBER of the invoked library.
+ */
+XXH_PUBLIC_API XXH_CONSTF unsigned XXH_versionNumber (void);
+
+
+/* ****************************
+*  Common basic types
+******************************/
+#include <stddef.h>   /* size_t */
+/*!
+ * @brief Exit code for the streaming API.
  */
-XXH_PUBLIC_API unsigned XXH_versionNumber(void);
+typedef enum {
+    XXH_OK = 0, /*!< OK */
+    XXH_ERROR   /*!< Error */
+} XXH_errorcode;
 
-  /* ****************************
-   *  Definitions
-   ******************************/
-  #include <stddef.h>                                             /* size_t */
-typedef enum { XXH_OK = 0, XXH_ERROR } XXH_errorcode;
 
-  /*-**********************************************************************
-   *  32-bit hash
-   ************************************************************************/
-  #if defined(XXH_DOXYGEN)                 /* Don't show <stdint.h> include */
+/*-**********************************************************************
+*  32-bit hash
+************************************************************************/
+#if defined(XXH_DOXYGEN) /* Don't show <stdint.h> include */
 /*!
  * @brief An unsigned 32-bit integer.
  *
@@ -361,51 +586,44 @@ typedef enum { XXH_OK = 0, XXH_ERROR } XXH_errorcode;
  */
 typedef uint32_t XXH32_hash_t;
 
-  #elif !defined(__VMS) &&     \
-      (defined(__cplusplus) || \
-       (defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */))
-    #include <stdint.h>
-typedef uint32_t XXH32_hash_t;
-
-  #else
-    #include <limits.h>
-    #if UINT_MAX == 0xFFFFFFFFUL
-typedef unsigned int XXH32_hash_t;
-    #else
-      #if ULONG_MAX == 0xFFFFFFFFUL
-typedef unsigned long XXH32_hash_t;
-      #else
-        #error "unsupported platform: need a 32-bit type"
-      #endif
-    #endif
-  #endif
+#elif !defined (__VMS) \
+  && (defined (__cplusplus) \
+  || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+#   include <stdint.h>
+    typedef uint32_t XXH32_hash_t;
+
+#else
+#   include <limits.h>
+#   if UINT_MAX == 0xFFFFFFFFUL
+      typedef unsigned int XXH32_hash_t;
+#   elif ULONG_MAX == 0xFFFFFFFFUL
+      typedef unsigned long XXH32_hash_t;
+#   else
+#     error "unsupported platform: need a 32-bit type"
+#   endif
+#endif
 
 /*!
  * @}
  *
- * @defgroup xxh32_family XXH32 family
+ * @defgroup XXH32_family XXH32 family
  * @ingroup public
  * Contains functions used in the classic 32-bit xxHash algorithm.
  *
  * @note
- *   XXH32 is considered rather weak by today's standards.
- *   The @ref xxh3_family provides competitive speed for both 32-bit and 64-bit
- *   systems, and offers true 64/128 bit hash results. It provides a superior
- *   level of dispersion, and greatly reduces the risks of collisions.
+ *   XXH32 is useful for older platforms, with no or poor 64-bit performance.
+ *   Note that the @ref XXH3_family provides competitive speed for both 32-bit
+ *   and 64-bit systems, and offers true 64/128 bit hash results.
  *
- * @see @ref xxh64_family, @ref xxh3_family : Other xxHash families
- * @see @ref xxh32_impl for implementation details
+ * @see @ref XXH64_family, @ref XXH3_family : Other xxHash families
+ * @see @ref XXH32_impl for implementation details
  * @{
-
  */
 
 /*!
  * @brief Calculates the 32-bit hash of @p input using xxHash32.
  *
- * Speed on Core 2 Duo @ 3 GHz (single thread, SMHasher benchmark): 5.4 GB/s
- *
- * @param input The block of data to be hashed, at least @p length bytes in
- * size.
+ * @param input The block of data to be hashed, at least @p length bytes in size.
  * @param length The length of @p input, in bytes.
  * @param seed The 32-bit seed to alter the hash's output predictably.
  *
@@ -414,94 +632,46 @@ typedef unsigned long XXH32_hash_t;
  *   readable, contiguous memory. However, if @p length is `0`, @p input may be
  *   `NULL`. In C++, this also must be *TriviallyCopyable*.
  *
- * @return The calculated 32-bit hash value.
- *
- * @see
- *    XXH64(), XXH3_64bits_withSeed(), XXH3_128bits_withSeed(), XXH128():
- *    Direct equivalents for the other variants of xxHash.
- * @see
- *    XXH32_createState(), XXH32_update(), XXH32_digest(): Streaming version.
- */
-XXH_PUBLIC_API XXH32_hash_t XXH32(const void *input, size_t length,
-                                  XXH32_hash_t seed);
-
-/*!
- * Streaming functions generate the xxHash value from an incremental input.
- * This method is slower than single-call functions, due to state management.
- * For small inputs, prefer `XXH32()` and `XXH64()`, which are better optimized.
- *
- * An XXH state must first be allocated using `XXH*_createState()`.
- *
- * Start a new hash by initializing the state with a seed using `XXH*_reset()`.
- *
- * Then, feed the hash state by calling `XXH*_update()` as many times as
- * necessary.
- *
- * The function returns an error code, with 0 meaning OK, and any other value
- * meaning there is an error.
- *
- * Finally, a hash value can be produced anytime, by using `XXH*_digest()`.
- * This function returns the nn-bits hash as an int or long long.
- *
- * It's still possible to continue inserting input into the hash state after a
- * digest, and generate new hash values later on by invoking `XXH*_digest()`.
- *
- * When done, release the state using `XXH*_freeState()`.
- *
- * Example code for incrementally hashing a file:
- * @code{.c}
- *    #include <stdio.h>
- *    #include <xxhash.h>
- *    #define BUFFER_SIZE 256
- *
- *    // Note: XXH64 and XXH3 use the same interface.
- *    XXH32_hash_t
- *    hashFile(FILE* stream)
- *    {
-
- *        XXH32_state_t* state;
- *        unsigned char buf[BUFFER_SIZE];
- *        size_t amt;
- *        XXH32_hash_t hash;
+ * @return The calculated 32-bit xxHash32 value.
  *
- *        state = XXH32_createState();       // Create a state
- *        assert(state != NULL);             // Error check here
- *        XXH32_reset(state, 0xbaad5eed);    // Reset state with our seed
- *        while ((amt = fread(buf, 1, sizeof(buf), stream)) != 0) {
-
- *            XXH32_update(state, buf, amt); // Hash the file in chunks
- *        }
- *        hash = XXH32_digest(state);        // Finalize the hash
- *        XXH32_freeState(state);            // Clean up
- *        return hash;
- *    }
- * @endcode
+ * @see @ref single_shot_example "Single Shot Example" for an example.
  */
+XXH_PUBLIC_API XXH_PUREF XXH32_hash_t XXH32 (const void* input, size_t length, XXH32_hash_t seed);
 
+#ifndef XXH_NO_STREAM
 /*!
  * @typedef struct XXH32_state_s XXH32_state_t
  * @brief The opaque state struct for the XXH32 streaming API.
  *
  * @see XXH32_state_s for details.
+ * @see @ref streaming_example "Streaming Example"
  */
 typedef struct XXH32_state_s XXH32_state_t;
 
 /*!
  * @brief Allocates an @ref XXH32_state_t.
  *
- * Must be freed with XXH32_freeState().
- * @return An allocated XXH32_state_t on success, `NULL` on failure.
+ * @return An allocated pointer of @ref XXH32_state_t on success.
+ * @return `NULL` on failure.
+ *
+ * @note Must be freed with XXH32_freeState().
+ *
+ * @see @ref streaming_example "Streaming Example"
  */
-XXH_PUBLIC_API XXH32_state_t *XXH32_createState(void);
+XXH_PUBLIC_API XXH_MALLOCF XXH32_state_t* XXH32_createState(void);
 /*!
  * @brief Frees an @ref XXH32_state_t.
  *
- * Must be allocated with XXH32_createState().
- * @param statePtr A pointer to an @ref XXH32_state_t allocated with @ref
- * XXH32_createState().
- * @return XXH_OK.
+ * @param statePtr A pointer to an @ref XXH32_state_t allocated with @ref XXH32_createState().
+ *
+ * @return @ref XXH_OK.
+ *
+ * @note @p statePtr must be allocated with XXH32_createState().
+ *
+ * @see @ref streaming_example "Streaming Example"
+ *
  */
-XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t *statePtr);
+XXH_PUBLIC_API XXH_errorcode  XXH32_freeState(XXH32_state_t* statePtr);
 /*!
  * @brief Copies one @ref XXH32_state_t to another.
  *
@@ -510,33 +680,31 @@ XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t *statePtr);
  * @pre
  *   @p dst_state and @p src_state must not be `NULL` and must not overlap.
  */
-XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t       *dst_state,
-                                    const XXH32_state_t *src_state);
+XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dst_state, const XXH32_state_t* src_state);
 
 /*!
  * @brief Resets an @ref XXH32_state_t to begin a new hash.
  *
- * This function resets and seeds a state. Call it before @ref XXH32_update().
- *
  * @param statePtr The state struct to reset.
  * @param seed The 32-bit seed to alter the hash result predictably.
  *
  * @pre
  *   @p statePtr must not be `NULL`.
  *
- * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
+ * @return @ref XXH_OK on success.
+ * @return @ref XXH_ERROR on failure.
+ *
+ * @note This function resets and seeds a state. Call it before @ref XXH32_update().
+ *
+ * @see @ref streaming_example "Streaming Example"
  */
-XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t *statePtr,
-                                         XXH32_hash_t   seed);
+XXH_PUBLIC_API XXH_errorcode XXH32_reset  (XXH32_state_t* statePtr, XXH32_hash_t seed);
 
 /*!
  * @brief Consumes a block of @p input to an @ref XXH32_state_t.
  *
- * Call this to incrementally consume blocks of data.
- *
  * @param statePtr The state struct to update.
- * @param input The block of data to be hashed, at least @p length bytes in
- * size.
+ * @param input The block of data to be hashed, at least @p length bytes in size.
  * @param length The length of @p input, in bytes.
  *
  * @pre
@@ -546,70 +714,55 @@ XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t *statePtr,
  *   readable, contiguous memory. However, if @p length is `0`, @p input may be
  *   `NULL`. In C++, this also must be *TriviallyCopyable*.
  *
- * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
+ * @return @ref XXH_OK on success.
+ * @return @ref XXH_ERROR on failure.
+ *
+ * @note Call this to incrementally consume blocks of data.
+ *
+ * @see @ref streaming_example "Streaming Example"
  */
-XXH_PUBLIC_API XXH_errorcode XXH32_update(XXH32_state_t *statePtr,
-                                          const void *input, size_t length);
+XXH_PUBLIC_API XXH_errorcode XXH32_update (XXH32_state_t* statePtr, const void* input, size_t length);
 
 /*!
  * @brief Returns the calculated hash value from an @ref XXH32_state_t.
  *
- * @note
- *   Calling XXH32_digest() will not affect @p statePtr, so you can update,
- *   digest, and update again.
- *
  * @param statePtr The state struct to calculate the hash from.
  *
  * @pre
  *  @p statePtr must not be `NULL`.
  *
- * @return The calculated xxHash32 value from that state.
- */
-XXH_PUBLIC_API XXH32_hash_t XXH32_digest(const XXH32_state_t *statePtr);
-
-/*******   Canonical representation   *******/
-
-/*
- * The default return values from XXH functions are unsigned 32 and 64 bit
- * integers.
- * This the simplest and fastest format for further post-processing.
- *
- * However, this leaves open the question of what is the order on the byte
- * level, since little and big endian conventions will store the same number
- * differently.
- *
- * The canonical representation settles this issue by mandating big-endian
- * convention, the same convention as human-readable numbers (large digits
- * first).
+ * @return The calculated 32-bit xxHash32 value from that state.
  *
- * When writing hash values to storage, sending them over a network, or printing
- * them, it's highly recommended to use the canonical representation to ensure
- * portability across a wider range of systems, present and future.
+ * @note
+ *   Calling XXH32_digest() will not affect @p statePtr, so you can update,
+ *   digest, and update again.
  *
- * The following functions allow transformation of hash values to and from
- * canonical format.
+ * @see @ref streaming_example "Streaming Example"
  */
+XXH_PUBLIC_API XXH_PUREF XXH32_hash_t XXH32_digest (const XXH32_state_t* statePtr);
+#endif /* !XXH_NO_STREAM */
+
+/*******   Canonical representation   *******/
 
 /*!
  * @brief Canonical (big endian) representation of @ref XXH32_hash_t.
  */
 typedef struct {
-
-  unsigned char digest[4];                      /*!< Hash bytes, big endian */
-
+    unsigned char digest[4]; /*!< Hash bytes, big endian */
 } XXH32_canonical_t;
 
 /*!
  * @brief Converts an @ref XXH32_hash_t to a big endian @ref XXH32_canonical_t.
  *
- * @param dst The @ref XXH32_canonical_t pointer to be stored to.
+ * @param dst  The @ref XXH32_canonical_t pointer to be stored to.
  * @param hash The @ref XXH32_hash_t to be converted.
  *
  * @pre
  *   @p dst must not be `NULL`.
+ *
+ * @see @ref canonical_representation_example "Canonical Representation Example"
  */
-XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t *dst,
-                                            XXH32_hash_t       hash);
+XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash);
 
 /*!
  * @brief Converts an @ref XXH32_canonical_t to a native @ref XXH32_hash_t.
@@ -620,103 +773,127 @@ XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t *dst,
  *   @p src must not be `NULL`.
  *
  * @return The converted hash.
+ *
+ * @see @ref canonical_representation_example "Canonical Representation Example"
+ */
+XXH_PUBLIC_API XXH_PUREF XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src);
+
+
+/*! @cond Doxygen ignores this part */
+#ifdef __has_attribute
+# define XXH_HAS_ATTRIBUTE(x) __has_attribute(x)
+#else
+# define XXH_HAS_ATTRIBUTE(x) 0
+#endif
+/*! @endcond */
+
+/*! @cond Doxygen ignores this part */
+/*
+ * C23 __STDC_VERSION__ number hasn't been specified yet. For now
+ * leave as `201711L` (C17 + 1).
+ * TODO: Update to correct value when its been specified.
+ */
+#define XXH_C23_VN 201711L
+/*! @endcond */
+
+/*! @cond Doxygen ignores this part */
+/* C-language Attributes are added in C23. */
+#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= XXH_C23_VN) && defined(__has_c_attribute)
+# define XXH_HAS_C_ATTRIBUTE(x) __has_c_attribute(x)
+#else
+# define XXH_HAS_C_ATTRIBUTE(x) 0
+#endif
+/*! @endcond */
+
+/*! @cond Doxygen ignores this part */
+#if defined(__cplusplus) && defined(__has_cpp_attribute)
+# define XXH_HAS_CPP_ATTRIBUTE(x) __has_cpp_attribute(x)
+#else
+# define XXH_HAS_CPP_ATTRIBUTE(x) 0
+#endif
+/*! @endcond */
+
+/*! @cond Doxygen ignores this part */
+/*
+ * Define XXH_FALLTHROUGH macro for annotating switch case with the 'fallthrough' attribute
+ * introduced in CPP17 and C23.
+ * CPP17 : https://en.cppreference.com/w/cpp/language/attributes/fallthrough
+ * C23   : https://en.cppreference.com/w/c/language/attributes/fallthrough
+ */
+#if XXH_HAS_C_ATTRIBUTE(fallthrough) || XXH_HAS_CPP_ATTRIBUTE(fallthrough)
+# define XXH_FALLTHROUGH [[fallthrough]]
+#elif XXH_HAS_ATTRIBUTE(__fallthrough__)
+# define XXH_FALLTHROUGH __attribute__ ((__fallthrough__))
+#else
+# define XXH_FALLTHROUGH /* fallthrough */
+#endif
+/*! @endcond */
+
+/*! @cond Doxygen ignores this part */
+/*
+ * Define XXH_NOESCAPE for annotated pointers in public API.
+ * https://clang.llvm.org/docs/AttributeReference.html#noescape
+ * As of writing this, only supported by clang.
  */
-XXH_PUBLIC_API XXH32_hash_t
-XXH32_hashFromCanonical(const XXH32_canonical_t *src);
-
-  #ifdef __has_attribute
-    #define XXH_HAS_ATTRIBUTE(x) __has_attribute(x)
-  #else
-    #define XXH_HAS_ATTRIBUTE(x) 0
-  #endif
-
-  /* C-language Attributes are added in C23. */
-  #if defined(__STDC_VERSION__) && (__STDC_VERSION__ > 201710L) && \
-      defined(__has_c_attribute)
-    #define XXH_HAS_C_ATTRIBUTE(x) __has_c_attribute(x)
-  #else
-    #define XXH_HAS_C_ATTRIBUTE(x) 0
-  #endif
-
-  #if defined(__cplusplus) && defined(__has_cpp_attribute)
-    #define XXH_HAS_CPP_ATTRIBUTE(x) __has_cpp_attribute(x)
-  #else
-    #define XXH_HAS_CPP_ATTRIBUTE(x) 0
-  #endif
-
-  /*
-  Define XXH_FALLTHROUGH macro for annotating switch case with the 'fallthrough'
-  attribute introduced in CPP17 and C23. CPP17 :
-  https://en.cppreference.com/w/cpp/language/attributes/fallthrough C23   :
-  https://en.cppreference.com/w/c/language/attributes/fallthrough
-  */
-  #if XXH_HAS_C_ATTRIBUTE(x)
-    #define XXH_FALLTHROUGH [[fallthrough]]
-  #elif XXH_HAS_CPP_ATTRIBUTE(x)
-    #define XXH_FALLTHROUGH [[fallthrough]]
-  #elif XXH_HAS_ATTRIBUTE(__fallthrough__)
-    #define XXH_FALLTHROUGH __attribute__((fallthrough))
-  #else
-    #define XXH_FALLTHROUGH
-  #endif
+#if XXH_HAS_ATTRIBUTE(noescape)
+# define XXH_NOESCAPE __attribute__((noescape))
+#else
+# define XXH_NOESCAPE
+#endif
+/*! @endcond */
+
 
 /*!
  * @}
  * @ingroup public
  * @{
-
  */
 
-  #ifndef XXH_NO_LONG_LONG
-    /*-**********************************************************************
-     *  64-bit hash
-     ************************************************************************/
-    #if defined(XXH_DOXYGEN)                    /* don't include <stdint.h> */
+#ifndef XXH_NO_LONG_LONG
+/*-**********************************************************************
+*  64-bit hash
+************************************************************************/
+#if defined(XXH_DOXYGEN) /* don't include <stdint.h> */
 /*!
  * @brief An unsigned 64-bit integer.
  *
  * Not necessarily defined to `uint64_t` but functionally equivalent.
  */
 typedef uint64_t XXH64_hash_t;
-    #elif !defined(__VMS) &&                                   \
-        (defined(__cplusplus) || (defined(__STDC_VERSION__) && \
-                                  (__STDC_VERSION__ >= 199901L) /* C99 */))
-      #include <stdint.h>
-typedef uint64_t XXH64_hash_t;
-    #else
-      #include <limits.h>
-      #if defined(__LP64__) && ULONG_MAX == 0xFFFFFFFFFFFFFFFFULL
-/* LP64 ABI says uint64_t is unsigned long */
-typedef unsigned long XXH64_hash_t;
-      #else
-/* the following type must have a width of 64-bit */
-typedef unsigned long long XXH64_hash_t;
-      #endif
-    #endif
+#elif !defined (__VMS) \
+  && (defined (__cplusplus) \
+  || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+#  include <stdint.h>
+   typedef uint64_t XXH64_hash_t;
+#else
+#  include <limits.h>
+#  if defined(__LP64__) && ULONG_MAX == 0xFFFFFFFFFFFFFFFFULL
+     /* LP64 ABI says uint64_t is unsigned long */
+     typedef unsigned long XXH64_hash_t;
+#  else
+     /* the following type must have a width of 64-bit */
+     typedef unsigned long long XXH64_hash_t;
+#  endif
+#endif
 
 /*!
  * @}
  *
- * @defgroup xxh64_family XXH64 family
+ * @defgroup XXH64_family XXH64 family
  * @ingroup public
  * @{
-
  * Contains functions used in the classic 64-bit xxHash algorithm.
  *
  * @note
  *   XXH3 provides competitive speed for both 32-bit and 64-bit systems,
- *   and offers true 64/128 bit hash results. It provides a superior level of
- *   dispersion, and greatly reduces the risks of collisions.
+ *   and offers true 64/128 bit hash results.
+ *   It provides better speed for systems with vector processing capabilities.
  */
 
 /*!
  * @brief Calculates the 64-bit hash of @p input using xxHash64.
  *
- * This function usually runs faster on 64-bit systems, but slower on 32-bit
- * systems (see benchmark).
- *
- * @param input The block of data to be hashed, at least @p length bytes in
- * size.
+ * @param input The block of data to be hashed, at least @p length bytes in size.
  * @param length The length of @p input, in bytes.
  * @param seed The 64-bit seed to alter the hash's output predictably.
  *
@@ -725,62 +902,166 @@ typedef unsigned long long XXH64_hash_t;
  *   readable, contiguous memory. However, if @p length is `0`, @p input may be
  *   `NULL`. In C++, this also must be *TriviallyCopyable*.
  *
- * @return The calculated 64-bit hash.
+ * @return The calculated 64-bit xxHash64 value.
  *
- * @see
- *    XXH32(), XXH3_64bits_withSeed(), XXH3_128bits_withSeed(), XXH128():
- *    Direct equivalents for the other variants of xxHash.
- * @see
- *    XXH64_createState(), XXH64_update(), XXH64_digest(): Streaming version.
+ * @see @ref single_shot_example "Single Shot Example" for an example.
  */
-XXH_PUBLIC_API XXH64_hash_t XXH64(const void *input, size_t length,
-                                  XXH64_hash_t seed);
+XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH64(XXH_NOESCAPE const void* input, size_t length, XXH64_hash_t seed);
 
 /*******   Streaming   *******/
+#ifndef XXH_NO_STREAM
 /*!
  * @brief The opaque state struct for the XXH64 streaming API.
  *
  * @see XXH64_state_s for details.
+ * @see @ref streaming_example "Streaming Example"
  */
-typedef struct XXH64_state_s  XXH64_state_t;             /* incomplete type */
-XXH_PUBLIC_API XXH64_state_t *XXH64_createState(void);
-XXH_PUBLIC_API XXH_errorcode  XXH64_freeState(XXH64_state_t *statePtr);
-XXH_PUBLIC_API void           XXH64_copyState(XXH64_state_t       *dst_state,
-                                              const XXH64_state_t *src_state);
-
-XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH64_state_t *statePtr,
-                                         XXH64_hash_t   seed);
-XXH_PUBLIC_API XXH_errorcode XXH64_update(XXH64_state_t *statePtr,
-                                          const void *input, size_t length);
-XXH_PUBLIC_API XXH64_hash_t  XXH64_digest(const XXH64_state_t *statePtr);
-
-/*******   Canonical representation   *******/
-typedef struct {
-
-  unsigned char digest[sizeof(XXH64_hash_t)];
-
-} XXH64_canonical_t;
-
-XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t *dst,
-                                            XXH64_hash_t       hash);
-XXH_PUBLIC_API XXH64_hash_t
-XXH64_hashFromCanonical(const XXH64_canonical_t *src);
+typedef struct XXH64_state_s XXH64_state_t;   /* incomplete type */
 
 /*!
- * @}
- * ************************************************************************
- * @defgroup xxh3_family XXH3 family
- * @ingroup public
- * @{
-
+ * @brief Allocates an @ref XXH64_state_t.
  *
- * XXH3 is a more recent hash algorithm featuring:
- *  - Improved speed for both small and large inputs
- *  - True 64-bit and 128-bit outputs
- *  - SIMD acceleration
- *  - Improved 32-bit viability
+ * @return An allocated pointer of @ref XXH64_state_t on success.
+ * @return `NULL` on failure.
  *
- * Speed analysis methodology is explained here:
+ * @note Must be freed with XXH64_freeState().
+ *
+ * @see @ref streaming_example "Streaming Example"
+ */
+XXH_PUBLIC_API XXH_MALLOCF XXH64_state_t* XXH64_createState(void);
+
+/*!
+ * @brief Frees an @ref XXH64_state_t.
+ *
+ * @param statePtr A pointer to an @ref XXH64_state_t allocated with @ref XXH64_createState().
+ *
+ * @return @ref XXH_OK.
+ *
+ * @note @p statePtr must be allocated with XXH64_createState().
+ *
+ * @see @ref streaming_example "Streaming Example"
+ */
+XXH_PUBLIC_API XXH_errorcode  XXH64_freeState(XXH64_state_t* statePtr);
+
+/*!
+ * @brief Copies one @ref XXH64_state_t to another.
+ *
+ * @param dst_state The state to copy to.
+ * @param src_state The state to copy from.
+ * @pre
+ *   @p dst_state and @p src_state must not be `NULL` and must not overlap.
+ */
+XXH_PUBLIC_API void XXH64_copyState(XXH_NOESCAPE XXH64_state_t* dst_state, const XXH64_state_t* src_state);
+
+/*!
+ * @brief Resets an @ref XXH64_state_t to begin a new hash.
+ *
+ * @param statePtr The state struct to reset.
+ * @param seed The 64-bit seed to alter the hash result predictably.
+ *
+ * @pre
+ *   @p statePtr must not be `NULL`.
+ *
+ * @return @ref XXH_OK on success.
+ * @return @ref XXH_ERROR on failure.
+ *
+ * @note This function resets and seeds a state. Call it before @ref XXH64_update().
+ *
+ * @see @ref streaming_example "Streaming Example"
+ */
+XXH_PUBLIC_API XXH_errorcode XXH64_reset  (XXH_NOESCAPE XXH64_state_t* statePtr, XXH64_hash_t seed);
+
+/*!
+ * @brief Consumes a block of @p input to an @ref XXH64_state_t.
+ *
+ * @param statePtr The state struct to update.
+ * @param input The block of data to be hashed, at least @p length bytes in size.
+ * @param length The length of @p input, in bytes.
+ *
+ * @pre
+ *   @p statePtr must not be `NULL`.
+ * @pre
+ *   The memory between @p input and @p input + @p length must be valid,
+ *   readable, contiguous memory. However, if @p length is `0`, @p input may be
+ *   `NULL`. In C++, this also must be *TriviallyCopyable*.
+ *
+ * @return @ref XXH_OK on success.
+ * @return @ref XXH_ERROR on failure.
+ *
+ * @note Call this to incrementally consume blocks of data.
+ *
+ * @see @ref streaming_example "Streaming Example"
+ */
+XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH_NOESCAPE XXH64_state_t* statePtr, XXH_NOESCAPE const void* input, size_t length);
+
+/*!
+ * @brief Returns the calculated hash value from an @ref XXH64_state_t.
+ *
+ * @param statePtr The state struct to calculate the hash from.
+ *
+ * @pre
+ *  @p statePtr must not be `NULL`.
+ *
+ * @return The calculated 64-bit xxHash64 value from that state.
+ *
+ * @note
+ *   Calling XXH64_digest() will not affect @p statePtr, so you can update,
+ *   digest, and update again.
+ *
+ * @see @ref streaming_example "Streaming Example"
+ */
+XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH64_digest (XXH_NOESCAPE const XXH64_state_t* statePtr);
+#endif /* !XXH_NO_STREAM */
+/*******   Canonical representation   *******/
+
+/*!
+ * @brief Canonical (big endian) representation of @ref XXH64_hash_t.
+ */
+typedef struct { unsigned char digest[sizeof(XXH64_hash_t)]; } XXH64_canonical_t;
+
+/*!
+ * @brief Converts an @ref XXH64_hash_t to a big endian @ref XXH64_canonical_t.
+ *
+ * @param dst The @ref XXH64_canonical_t pointer to be stored to.
+ * @param hash The @ref XXH64_hash_t to be converted.
+ *
+ * @pre
+ *   @p dst must not be `NULL`.
+ *
+ * @see @ref canonical_representation_example "Canonical Representation Example"
+ */
+XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH_NOESCAPE XXH64_canonical_t* dst, XXH64_hash_t hash);
+
+/*!
+ * @brief Converts an @ref XXH64_canonical_t to a native @ref XXH64_hash_t.
+ *
+ * @param src The @ref XXH64_canonical_t to convert.
+ *
+ * @pre
+ *   @p src must not be `NULL`.
+ *
+ * @return The converted hash.
+ *
+ * @see @ref canonical_representation_example "Canonical Representation Example"
+ */
+XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH64_hashFromCanonical(XXH_NOESCAPE const XXH64_canonical_t* src);
+
+#ifndef XXH_NO_XXH3
+
+/*!
+ * @}
+ * ************************************************************************
+ * @defgroup XXH3_family XXH3 family
+ * @ingroup public
+ * @{
+ *
+ * XXH3 is a more recent hash algorithm featuring:
+ *  - Improved speed for both small and large inputs
+ *  - True 64-bit and 128-bit outputs
+ *  - SIMD acceleration
+ *  - Improved 32-bit viability
+ *
+ * Speed analysis methodology is explained here:
  *
  *    https://fastcompression.blogspot.com/2019/03/presenting-xxh3.html
  *
@@ -790,16 +1071,26 @@ XXH64_hashFromCanonical(const XXH64_canonical_t *src);
  *
  * XXH3's speed benefits greatly from SIMD and 64-bit arithmetic,
  * but does not require it.
- * Any 32-bit and 64-bit targets that can run XXH32 smoothly
- * can run XXH3 at competitive speeds, even without vector support.
- * Further details are explained in the implementation.
- *
- * Optimized implementations are provided for AVX512, AVX2, SSE2, NEON, POWER8,
- * ZVector and scalar targets. This can be controlled via the XXH_VECTOR macro.
+ * Most 32-bit and 64-bit targets that can run XXH32 smoothly can run XXH3
+ * at competitive speeds, even without vector support. Further details are
+ * explained in the implementation.
+ *
+ * XXH3 has a fast scalar implementation, but it also includes accelerated SIMD
+ * implementations for many common platforms:
+ *   - AVX512
+ *   - AVX2
+ *   - SSE2
+ *   - ARM NEON
+ *   - WebAssembly SIMD128
+ *   - POWER8 VSX
+ *   - s390x ZVector
+ * This can be controlled via the @ref XXH_VECTOR macro, but it automatically
+ * selects the best version according to predefined macros. For the x86 family, an
+ * automatic runtime dispatcher is included separately in @ref xxh_x86dispatch.c.
  *
  * XXH3 implementation is portable:
  * it has a generic C90 formulation that can be compiled on any platform,
- * all implementations generage exactly the same hash value on all platforms.
+ * all implementations generate exactly the same hash value on all platforms.
  * Starting from v0.8.0, it's also labelled "stable", meaning that
  * any future version will also generate the same hash value.
  *
@@ -811,53 +1102,106 @@ XXH64_hashFromCanonical(const XXH64_canonical_t *src);
  *
  * The API supports one-shot hashing, streaming mode, and custom secrets.
  */
-
 /*-**********************************************************************
- *  XXH3 64-bit variant
- ************************************************************************/
+*  XXH3 64-bit variant
+************************************************************************/
 
-/* XXH3_64bits():
- * default 64-bit variant, using default secret and default seed of 0.
- * It's the fastest variant. */
-XXH_PUBLIC_API XXH64_hash_t XXH3_64bits(const void *data, size_t len);
+/*!
+ * @brief Calculates 64-bit unseeded variant of XXH3 hash of @p input.
+ *
+ * @param input  The block of data to be hashed, at least @p length bytes in size.
+ * @param length The length of @p input, in bytes.
+ *
+ * @pre
+ *   The memory between @p input and @p input + @p length must be valid,
+ *   readable, contiguous memory. However, if @p length is `0`, @p input may be
+ *   `NULL`. In C++, this also must be *TriviallyCopyable*.
+ *
+ * @return The calculated 64-bit XXH3 hash value.
+ *
+ * @note
+ *   This is equivalent to @ref XXH3_64bits_withSeed() with a seed of `0`, however
+ *   it may have slightly better performance due to constant propagation of the
+ *   defaults.
+ *
+ * @see
+ *    XXH3_64bits_withSeed(), XXH3_64bits_withSecret(): other seeding variants
+ * @see @ref single_shot_example "Single Shot Example" for an example.
+ */
+XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits(XXH_NOESCAPE const void* input, size_t length);
 
-/*
- * XXH3_64bits_withSeed():
- * This variant generates a custom secret on the fly
- * based on default secret altered using the `seed` value.
+/*!
+ * @brief Calculates 64-bit seeded variant of XXH3 hash of @p input.
+ *
+ * @param input  The block of data to be hashed, at least @p length bytes in size.
+ * @param length The length of @p input, in bytes.
+ * @param seed   The 64-bit seed to alter the hash result predictably.
+ *
+ * @pre
+ *   The memory between @p input and @p input + @p length must be valid,
+ *   readable, contiguous memory. However, if @p length is `0`, @p input may be
+ *   `NULL`. In C++, this also must be *TriviallyCopyable*.
+ *
+ * @return The calculated 64-bit XXH3 hash value.
+ *
+ * @note
+ *    seed == 0 produces the same results as @ref XXH3_64bits().
+ *
+ * This variant generates a custom secret on the fly based on default secret
+ * altered using the @p seed value.
+ *
  * While this operation is decently fast, note that it's not completely free.
- * Note: seed==0 produces the same results as XXH3_64bits().
+ *
+ * @see @ref single_shot_example "Single Shot Example" for an example.
  */
-XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_withSeed(const void *data, size_t len,
-                                                 XXH64_hash_t seed);
+XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits_withSeed(XXH_NOESCAPE const void* input, size_t length, XXH64_hash_t seed);
 
-    /*!
-     * The bare minimum size for a custom secret.
-     *
-     * @see
-     *  XXH3_64bits_withSecret(), XXH3_64bits_reset_withSecret(),
-     *  XXH3_128bits_withSecret(), XXH3_128bits_reset_withSecret().
-     */
-    #define XXH3_SECRET_SIZE_MIN 136
+/*!
+ * The bare minimum size for a custom secret.
+ *
+ * @see
+ *  XXH3_64bits_withSecret(), XXH3_64bits_reset_withSecret(),
+ *  XXH3_128bits_withSecret(), XXH3_128bits_reset_withSecret().
+ */
+#define XXH3_SECRET_SIZE_MIN 136
+
+/*!
+ * @brief Calculates 64-bit variant of XXH3 with a custom "secret".
+ *
+ * @param data       The block of data to be hashed, at least @p len bytes in size.
+ * @param len        The length of @p data, in bytes.
+ * @param secret     The secret data.
+ * @param secretSize The length of @p secret, in bytes.
+ *
+ * @return The calculated 64-bit XXH3 hash value.
+ *
+ * @pre
+ *   The memory between @p data and @p data + @p len must be valid,
+ *   readable, contiguous memory. However, if @p length is `0`, @p data may be
+ *   `NULL`. In C++, this also must be *TriviallyCopyable*.
+ *
+ * It's possible to provide any blob of bytes as a "secret" to generate the hash.
+ * This makes it more difficult for an external actor to prepare an intentional collision.
+ * The main condition is that @p secretSize *must* be large enough (>= @ref XXH3_SECRET_SIZE_MIN).
+ * However, the quality of the secret impacts the dispersion of the hash algorithm.
+ * Therefore, the secret _must_ look like a bunch of random bytes.
+ * Avoid "trivial" or structured data such as repeated sequences or a text document.
+ * Whenever in doubt about the "randomness" of the blob of bytes,
+ * consider employing @ref XXH3_generateSecret() instead (see below).
+ * It will generate a proper high entropy secret derived from the blob of bytes.
+ * Another advantage of using XXH3_generateSecret() is that
+ * it guarantees that all bits within the initial blob of bytes
+ * will impact every bit of the output.
+ * This is not necessarily the case when using the blob of bytes directly
+ * because, when hashing _small_ inputs, only a portion of the secret is employed.
+ *
+ * @see @ref single_shot_example "Single Shot Example" for an example.
+ */
+XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits_withSecret(XXH_NOESCAPE const void* data, size_t len, XXH_NOESCAPE const void* secret, size_t secretSize);
 
-/*
- * XXH3_64bits_withSecret():
- * It's possible to provide any blob of bytes as a "secret" to generate the
- * hash. This makes it more difficult for an external actor to prepare an
- * intentional collision. The main condition is that secretSize *must* be large
- * enough (>= XXH3_SECRET_SIZE_MIN). However, the quality of produced hash
- * values depends on secret's entropy. Technically, the secret must look like a
- * bunch of random bytes. Avoid "trivial" or structured data such as repeated
- * sequences or a text document. Whenever unsure about the "randomness" of the
- * blob of bytes, consider relabelling it as a "custom seed" instead, and employ
- * "XXH3_generateSecret()" (see below) to generate a high entropy secret derived
- * from the custom seed.
- */
-XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_withSecret(const void *data, size_t len,
-                                                   const void *secret,
-                                                   size_t      secretSize);
 
 /*******   Streaming   *******/
+#ifndef XXH_NO_STREAM
 /*
  * Streaming requires state maintenance.
  * This operation costs memory and CPU.
@@ -866,52 +1210,143 @@ XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_withSecret(const void *data, size_t len,
  */
 
 /*!
- * @brief The state struct for the XXH3 streaming API.
+ * @brief The opaque state struct for the XXH3 streaming API.
  *
  * @see XXH3_state_s for details.
+ * @see @ref streaming_example "Streaming Example"
  */
-typedef struct XXH3_state_s  XXH3_state_t;
-XXH_PUBLIC_API XXH3_state_t *XXH3_createState(void);
-XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t *statePtr);
-XXH_PUBLIC_API void          XXH3_copyState(XXH3_state_t       *dst_state,
-                                            const XXH3_state_t *src_state);
+typedef struct XXH3_state_s XXH3_state_t;
+XXH_PUBLIC_API XXH_MALLOCF XXH3_state_t* XXH3_createState(void);
+XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t* statePtr);
 
-/*
- * XXH3_64bits_reset():
- * Initialize with default parameters.
- * digest will be equivalent to `XXH3_64bits()`.
+/*!
+ * @brief Copies one @ref XXH3_state_t to another.
+ *
+ * @param dst_state The state to copy to.
+ * @param src_state The state to copy from.
+ * @pre
+ *   @p dst_state and @p src_state must not be `NULL` and must not overlap.
  */
-XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset(XXH3_state_t *statePtr);
-/*
- * XXH3_64bits_reset_withSeed():
- * Generate a custom secret from `seed`, and store it into `statePtr`.
- * digest will be equivalent to `XXH3_64bits_withSeed()`.
+XXH_PUBLIC_API void XXH3_copyState(XXH_NOESCAPE XXH3_state_t* dst_state, XXH_NOESCAPE const XXH3_state_t* src_state);
+
+/*!
+ * @brief Resets an @ref XXH3_state_t to begin a new hash.
+ *
+ * @param statePtr The state struct to reset.
+ *
+ * @pre
+ *   @p statePtr must not be `NULL`.
+ *
+ * @return @ref XXH_OK on success.
+ * @return @ref XXH_ERROR on failure.
+ *
+ * @note
+ *   - This function resets `statePtr` and generate a secret with default parameters.
+ *   - Call this function before @ref XXH3_64bits_update().
+ *   - Digest will be equivalent to `XXH3_64bits()`.
+ *
+ * @see @ref streaming_example "Streaming Example"
+ *
  */
-XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSeed(XXH3_state_t *statePtr,
-                                                        XXH64_hash_t  seed);
-/*
- * XXH3_64bits_reset_withSecret():
- * `secret` is referenced, it _must outlive_ the hash streaming session.
- * Similar to one-shot API, `secretSize` must be >= `XXH3_SECRET_SIZE_MIN`,
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset(XXH_NOESCAPE XXH3_state_t* statePtr);
+
+/*!
+ * @brief Resets an @ref XXH3_state_t with 64-bit seed to begin a new hash.
+ *
+ * @param statePtr The state struct to reset.
+ * @param seed     The 64-bit seed to alter the hash result predictably.
+ *
+ * @pre
+ *   @p statePtr must not be `NULL`.
+ *
+ * @return @ref XXH_OK on success.
+ * @return @ref XXH_ERROR on failure.
+ *
+ * @note
+ *   - This function resets `statePtr` and generate a secret from `seed`.
+ *   - Call this function before @ref XXH3_64bits_update().
+ *   - Digest will be equivalent to `XXH3_64bits_withSeed()`.
+ *
+ * @see @ref streaming_example "Streaming Example"
+ *
+ */
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH64_hash_t seed);
+
+/*!
+ * @brief Resets an @ref XXH3_state_t with secret data to begin a new hash.
+ *
+ * @param statePtr The state struct to reset.
+ * @param secret     The secret data.
+ * @param secretSize The length of @p secret, in bytes.
+ *
+ * @pre
+ *   @p statePtr must not be `NULL`.
+ *
+ * @return @ref XXH_OK on success.
+ * @return @ref XXH_ERROR on failure.
+ *
+ * @note
+ *   `secret` is referenced, it _must outlive_ the hash streaming session.
+ *
+ * Similar to one-shot API, `secretSize` must be >= @ref XXH3_SECRET_SIZE_MIN,
  * and the quality of produced hash values depends on secret's entropy
  * (secret's content should look like a bunch of random bytes).
  * When in doubt about the randomness of a candidate `secret`,
  * consider employing `XXH3_generateSecret()` instead (see below).
+ *
+ * @see @ref streaming_example "Streaming Example"
+ */
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSecret(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize);
+
+/*!
+ * @brief Consumes a block of @p input to an @ref XXH3_state_t.
+ *
+ * @param statePtr The state struct to update.
+ * @param input The block of data to be hashed, at least @p length bytes in size.
+ * @param length The length of @p input, in bytes.
+ *
+ * @pre
+ *   @p statePtr must not be `NULL`.
+ * @pre
+ *   The memory between @p input and @p input + @p length must be valid,
+ *   readable, contiguous memory. However, if @p length is `0`, @p input may be
+ *   `NULL`. In C++, this also must be *TriviallyCopyable*.
+ *
+ * @return @ref XXH_OK on success.
+ * @return @ref XXH_ERROR on failure.
+ *
+ * @note Call this to incrementally consume blocks of data.
+ *
+ * @see @ref streaming_example "Streaming Example"
  */
-XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSecret(
-    XXH3_state_t *statePtr, const void *secret, size_t secretSize);
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_update (XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* input, size_t length);
 
-XXH_PUBLIC_API XXH_errorcode XXH3_64bits_update(XXH3_state_t *statePtr,
-                                                const void   *input,
-                                                size_t        length);
-XXH_PUBLIC_API XXH64_hash_t  XXH3_64bits_digest(const XXH3_state_t *statePtr);
+/*!
+ * @brief Returns the calculated XXH3 64-bit hash value from an @ref XXH3_state_t.
+ *
+ * @param statePtr The state struct to calculate the hash from.
+ *
+ * @pre
+ *  @p statePtr must not be `NULL`.
+ *
+ * @return The calculated XXH3 64-bit hash value from that state.
+ *
+ * @note
+ *   Calling XXH3_64bits_digest() will not affect @p statePtr, so you can update,
+ *   digest, and update again.
+ *
+ * @see @ref streaming_example "Streaming Example"
+ */
+XXH_PUBLIC_API XXH_PUREF XXH64_hash_t  XXH3_64bits_digest (XXH_NOESCAPE const XXH3_state_t* statePtr);
+#endif /* !XXH_NO_STREAM */
 
 /* note : canonical representation of XXH3 is the same as XXH64
  * since they both produce XXH64_hash_t values */
 
+
 /*-**********************************************************************
- *  XXH3 128-bit variant
- ************************************************************************/
+*  XXH3 128-bit variant
+************************************************************************/
 
 /*!
  * @brief The return value from 128-bit hashes.
@@ -920,21 +1355,80 @@ XXH_PUBLIC_API XXH64_hash_t  XXH3_64bits_digest(const XXH3_state_t *statePtr);
  * endianness.
  */
 typedef struct {
-
-  XXH64_hash_t low64;                     /*!< `value & 0xFFFFFFFFFFFFFFFF` */
-  XXH64_hash_t high64;                                   /*!< `value >> 64` */
-
+    XXH64_hash_t low64;   /*!< `value & 0xFFFFFFFFFFFFFFFF` */
+    XXH64_hash_t high64;  /*!< `value >> 64` */
 } XXH128_hash_t;
 
-XXH_PUBLIC_API XXH128_hash_t XXH3_128bits(const void *data, size_t len);
-XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_withSeed(const void *data, size_t len,
-                                                   XXH64_hash_t seed);
-XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_withSecret(const void *data,
-                                                     size_t      len,
-                                                     const void *secret,
-                                                     size_t      secretSize);
+/*!
+ * @brief Calculates 128-bit unseeded variant of XXH3 of @p data.
+ *
+ * @param data The block of data to be hashed, at least @p length bytes in size.
+ * @param len  The length of @p data, in bytes.
+ *
+ * @return The calculated 128-bit variant of XXH3 value.
+ *
+ * The 128-bit variant of XXH3 has more strength, but it has a bit of overhead
+ * for shorter inputs.
+ *
+ * This is equivalent to @ref XXH3_128bits_withSeed() with a seed of `0`, however
+ * it may have slightly better performance due to constant propagation of the
+ * defaults.
+ *
+ * @see XXH3_128bits_withSeed(), XXH3_128bits_withSecret(): other seeding variants
+ * @see @ref single_shot_example "Single Shot Example" for an example.
+ */
+XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits(XXH_NOESCAPE const void* data, size_t len);
+/*! @brief Calculates 128-bit seeded variant of XXH3 hash of @p data.
+ *
+ * @param data The block of data to be hashed, at least @p length bytes in size.
+ * @param len  The length of @p data, in bytes.
+ * @param seed The 64-bit seed to alter the hash result predictably.
+ *
+ * @return The calculated 128-bit variant of XXH3 value.
+ *
+ * @note
+ *    seed == 0 produces the same results as @ref XXH3_64bits().
+ *
+ * This variant generates a custom secret on the fly based on default secret
+ * altered using the @p seed value.
+ *
+ * While this operation is decently fast, note that it's not completely free.
+ *
+ * @see XXH3_128bits(), XXH3_128bits_withSecret(): other seeding variants
+ * @see @ref single_shot_example "Single Shot Example" for an example.
+ */
+XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits_withSeed(XXH_NOESCAPE const void* data, size_t len, XXH64_hash_t seed);
+/*!
+ * @brief Calculates 128-bit variant of XXH3 with a custom "secret".
+ *
+ * @param data       The block of data to be hashed, at least @p len bytes in size.
+ * @param len        The length of @p data, in bytes.
+ * @param secret     The secret data.
+ * @param secretSize The length of @p secret, in bytes.
+ *
+ * @return The calculated 128-bit variant of XXH3 value.
+ *
+ * It's possible to provide any blob of bytes as a "secret" to generate the hash.
+ * This makes it more difficult for an external actor to prepare an intentional collision.
+ * The main condition is that @p secretSize *must* be large enough (>= @ref XXH3_SECRET_SIZE_MIN).
+ * However, the quality of the secret impacts the dispersion of the hash algorithm.
+ * Therefore, the secret _must_ look like a bunch of random bytes.
+ * Avoid "trivial" or structured data such as repeated sequences or a text document.
+ * Whenever in doubt about the "randomness" of the blob of bytes,
+ * consider employing @ref XXH3_generateSecret() instead (see below).
+ * It will generate a proper high entropy secret derived from the blob of bytes.
+ * Another advantage of using XXH3_generateSecret() is that
+ * it guarantees that all bits within the initial blob of bytes
+ * will impact every bit of the output.
+ * This is not necessarily the case when using the blob of bytes directly
+ * because, when hashing _small_ inputs, only a portion of the secret is employed.
+ *
+ * @see @ref single_shot_example "Single Shot Example" for an example.
+ */
+XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits_withSecret(XXH_NOESCAPE const void* data, size_t len, XXH_NOESCAPE const void* secret, size_t secretSize);
 
 /*******   Streaming   *******/
+#ifndef XXH_NO_STREAM
 /*
  * Streaming requires state maintenance.
  * This operation costs memory and CPU.
@@ -944,73 +1438,193 @@ XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_withSecret(const void *data,
  * XXH3_128bits uses the same XXH3_state_t as XXH3_64bits().
  * Use already declared XXH3_createState() and XXH3_freeState().
  *
- * All reset and streaming functions have same meaning as their 64-bit
- * counterpart.
+ * All reset and streaming functions have same meaning as their 64-bit counterpart.
  */
 
-XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset(XXH3_state_t *statePtr);
-XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSeed(XXH3_state_t *statePtr,
-                                                         XXH64_hash_t  seed);
-XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSecret(
-    XXH3_state_t *statePtr, const void *secret, size_t secretSize);
-
-XXH_PUBLIC_API XXH_errorcode XXH3_128bits_update(XXH3_state_t *statePtr,
-                                                 const void   *input,
-                                                 size_t        length);
-XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_digest(const XXH3_state_t *statePtr);
-
-/* Following helper functions make it possible to compare XXH128_hast_t values.
- * Since XXH128_hash_t is a structure, this capability is not offered by the
- * language.
- * Note: For better performance, these functions can be inlined using
- * XXH_INLINE_ALL */
-
 /*!
- * XXH128_isEqual():
- * Return: 1 if `h1` and `h2` are equal, 0 if they are not.
+ * @brief Resets an @ref XXH3_state_t to begin a new hash.
+ *
+ * @param statePtr The state struct to reset.
+ *
+ * @pre
+ *   @p statePtr must not be `NULL`.
+ *
+ * @return @ref XXH_OK on success.
+ * @return @ref XXH_ERROR on failure.
+ *
+ * @note
+ *   - This function resets `statePtr` and generate a secret with default parameters.
+ *   - Call it before @ref XXH3_128bits_update().
+ *   - Digest will be equivalent to `XXH3_128bits()`.
+ *
+ * @see @ref streaming_example "Streaming Example"
  */
-XXH_PUBLIC_API int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2);
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset(XXH_NOESCAPE XXH3_state_t* statePtr);
 
 /*!
- * XXH128_cmp():
+ * @brief Resets an @ref XXH3_state_t with 64-bit seed to begin a new hash.
  *
- * This comparator is compatible with stdlib's `qsort()`/`bsearch()`.
+ * @param statePtr The state struct to reset.
+ * @param seed     The 64-bit seed to alter the hash result predictably.
  *
- * return: >0 if *h128_1  > *h128_2
- *         =0 if *h128_1 == *h128_2
- *         <0 if *h128_1  < *h128_2
- */
-XXH_PUBLIC_API int XXH128_cmp(const void *h128_1, const void *h128_2);
-
+ * @pre
+ *   @p statePtr must not be `NULL`.
+ *
+ * @return @ref XXH_OK on success.
+ * @return @ref XXH_ERROR on failure.
+ *
+ * @note
+ *   - This function resets `statePtr` and generate a secret from `seed`.
+ *   - Call it before @ref XXH3_128bits_update().
+ *   - Digest will be equivalent to `XXH3_128bits_withSeed()`.
+ *
+ * @see @ref streaming_example "Streaming Example"
+ */
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH64_hash_t seed);
+/*!
+ * @brief Resets an @ref XXH3_state_t with secret data to begin a new hash.
+ *
+ * @param statePtr   The state struct to reset.
+ * @param secret     The secret data.
+ * @param secretSize The length of @p secret, in bytes.
+ *
+ * @pre
+ *   @p statePtr must not be `NULL`.
+ *
+ * @return @ref XXH_OK on success.
+ * @return @ref XXH_ERROR on failure.
+ *
+ * `secret` is referenced, it _must outlive_ the hash streaming session.
+ * Similar to one-shot API, `secretSize` must be >= @ref XXH3_SECRET_SIZE_MIN,
+ * and the quality of produced hash values depends on secret's entropy
+ * (secret's content should look like a bunch of random bytes).
+ * When in doubt about the randomness of a candidate `secret`,
+ * consider employing `XXH3_generateSecret()` instead (see below).
+ *
+ * @see @ref streaming_example "Streaming Example"
+ */
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSecret(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize);
+
+/*!
+ * @brief Consumes a block of @p input to an @ref XXH3_state_t.
+ *
+ * Call this to incrementally consume blocks of data.
+ *
+ * @param statePtr The state struct to update.
+ * @param input The block of data to be hashed, at least @p length bytes in size.
+ * @param length The length of @p input, in bytes.
+ *
+ * @pre
+ *   @p statePtr must not be `NULL`.
+ *
+ * @return @ref XXH_OK on success.
+ * @return @ref XXH_ERROR on failure.
+ *
+ * @note
+ *   The memory between @p input and @p input + @p length must be valid,
+ *   readable, contiguous memory. However, if @p length is `0`, @p input may be
+ *   `NULL`. In C++, this also must be *TriviallyCopyable*.
+ *
+ */
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_update (XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* input, size_t length);
+
+/*!
+ * @brief Returns the calculated XXH3 128-bit hash value from an @ref XXH3_state_t.
+ *
+ * @param statePtr The state struct to calculate the hash from.
+ *
+ * @pre
+ *  @p statePtr must not be `NULL`.
+ *
+ * @return The calculated XXH3 128-bit hash value from that state.
+ *
+ * @note
+ *   Calling XXH3_128bits_digest() will not affect @p statePtr, so you can update,
+ *   digest, and update again.
+ *
+ */
+XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits_digest (XXH_NOESCAPE const XXH3_state_t* statePtr);
+#endif /* !XXH_NO_STREAM */
+
+/* Following helper functions make it possible to compare XXH128_hast_t values.
+ * Since XXH128_hash_t is a structure, this capability is not offered by the language.
+ * Note: For better performance, these functions can be inlined using XXH_INLINE_ALL */
+
+/*!
+ * @brief Check equality of two XXH128_hash_t values
+ *
+ * @param h1 The 128-bit hash value.
+ * @param h2 Another 128-bit hash value.
+ *
+ * @return `1` if `h1` and `h2` are equal.
+ * @return `0` if they are not.
+ */
+XXH_PUBLIC_API XXH_PUREF int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2);
+
+/*!
+ * @brief Compares two @ref XXH128_hash_t
+ *
+ * This comparator is compatible with stdlib's `qsort()`/`bsearch()`.
+ *
+ * @param h128_1 Left-hand side value
+ * @param h128_2 Right-hand side value
+ *
+ * @return >0 if @p h128_1  > @p h128_2
+ * @return =0 if @p h128_1 == @p h128_2
+ * @return <0 if @p h128_1  < @p h128_2
+ */
+XXH_PUBLIC_API XXH_PUREF int XXH128_cmp(XXH_NOESCAPE const void* h128_1, XXH_NOESCAPE const void* h128_2);
+
+
 /*******   Canonical representation   *******/
-typedef struct {
+typedef struct { unsigned char digest[sizeof(XXH128_hash_t)]; } XXH128_canonical_t;
 
-  unsigned char digest[sizeof(XXH128_hash_t)];
 
-} XXH128_canonical_t;
+/*!
+ * @brief Converts an @ref XXH128_hash_t to a big endian @ref XXH128_canonical_t.
+ *
+ * @param dst  The @ref XXH128_canonical_t pointer to be stored to.
+ * @param hash The @ref XXH128_hash_t to be converted.
+ *
+ * @pre
+ *   @p dst must not be `NULL`.
+ * @see @ref canonical_representation_example "Canonical Representation Example"
+ */
+XXH_PUBLIC_API void XXH128_canonicalFromHash(XXH_NOESCAPE XXH128_canonical_t* dst, XXH128_hash_t hash);
 
-XXH_PUBLIC_API void XXH128_canonicalFromHash(XXH128_canonical_t *dst,
-                                             XXH128_hash_t       hash);
-XXH_PUBLIC_API XXH128_hash_t
-XXH128_hashFromCanonical(const XXH128_canonical_t *src);
+/*!
+ * @brief Converts an @ref XXH128_canonical_t to a native @ref XXH128_hash_t.
+ *
+ * @param src The @ref XXH128_canonical_t to convert.
+ *
+ * @pre
+ *   @p src must not be `NULL`.
+ *
+ * @return The converted hash.
+ * @see @ref canonical_representation_example "Canonical Representation Example"
+ */
+XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH128_hashFromCanonical(XXH_NOESCAPE const XXH128_canonical_t* src);
 
-  #endif                                                /* XXH_NO_LONG_LONG */
+
+#endif  /* !XXH_NO_XXH3 */
+#endif  /* XXH_NO_LONG_LONG */
 
 /*!
  * @}
  */
-#endif                                         /* XXHASH_H_5627135585666179 */
+#endif /* XXHASH_H_5627135585666179 */
+
+
 
 #if defined(XXH_STATIC_LINKING_ONLY) && !defined(XXHASH_H_STATIC_13879238742)
-  #define XXHASH_H_STATIC_13879238742
+#define XXHASH_H_STATIC_13879238742
 /* ****************************************************************************
  * This section contains declarations which are not guaranteed to remain stable.
  * They may change in future versions, becoming incompatible with a different
  * version of the library.
  * These declarations should only be used with static linking.
  * Never use them in association with dynamic linking!
- *****************************************************************************
- */
+ ***************************************************************************** */
 
 /*
  * These definitions are only present to allow static allocation
@@ -1031,23 +1645,16 @@ XXH128_hashFromCanonical(const XXH128_canonical_t *src);
  * @see XXH64_state_s, XXH3_state_s
  */
 struct XXH32_state_s {
+   XXH32_hash_t total_len_32; /*!< Total length hashed, modulo 2^32 */
+   XXH32_hash_t large_len;    /*!< Whether the hash is >= 16 (handles @ref total_len_32 overflow) */
+   XXH32_hash_t v[4];         /*!< Accumulator lanes */
+   XXH32_hash_t mem32[4];     /*!< Internal buffer for partial reads. Treated as unsigned char[16]. */
+   XXH32_hash_t memsize;      /*!< Amount of data in @ref mem32 */
+   XXH32_hash_t reserved;     /*!< Reserved field. Do not read nor write to it. */
+};   /* typedef'd to XXH32_state_t */
 
-  XXH32_hash_t total_len_32;          /*!< Total length hashed, modulo 2^32 */
-  XXH32_hash_t large_len;    /*!< Whether the hash is >= 16 (handles @ref
-                                total_len_32 overflow) */
-  XXH32_hash_t v1;                              /*!< First accumulator lane */
-  XXH32_hash_t v2;                             /*!< Second accumulator lane */
-  XXH32_hash_t v3;                              /*!< Third accumulator lane */
-  XXH32_hash_t v4;                             /*!< Fourth accumulator lane */
-  XXH32_hash_t mem32[4];     /*!< Internal buffer for partial reads. Treated as
-                                unsigned char[16]. */
-  XXH32_hash_t memsize;                   /*!< Amount of data in @ref mem32 */
-  XXH32_hash_t reserved; /*!< Reserved field. Do not read or write to it, it may
-                            be removed. */
 
-};                                            /* typedef'd to XXH32_state_t */
-
-  #ifndef XXH_NO_LONG_LONG       /* defined when there is no 64-bit support */
+#ifndef XXH_NO_LONG_LONG  /* defined when there is no 64-bit support */
 
 /*!
  * @internal
@@ -1062,64 +1669,57 @@ struct XXH32_state_s {
  * @see XXH32_state_s, XXH3_state_s
  */
 struct XXH64_state_s {
+   XXH64_hash_t total_len;    /*!< Total length hashed. This is always 64-bit. */
+   XXH64_hash_t v[4];         /*!< Accumulator lanes */
+   XXH64_hash_t mem64[4];     /*!< Internal buffer for partial reads. Treated as unsigned char[32]. */
+   XXH32_hash_t memsize;      /*!< Amount of data in @ref mem64 */
+   XXH32_hash_t reserved32;   /*!< Reserved field, needed for padding anyways*/
+   XXH64_hash_t reserved64;   /*!< Reserved field. Do not read or write to it. */
+};   /* typedef'd to XXH64_state_t */
+
+#ifndef XXH_NO_XXH3
+
+#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L) /* >= C11 */
+#  include <stdalign.h>
+#  define XXH_ALIGN(n)      alignas(n)
+#elif defined(__cplusplus) && (__cplusplus >= 201103L) /* >= C++11 */
+/* In C++ alignas() is a keyword */
+#  define XXH_ALIGN(n)      alignas(n)
+#elif defined(__GNUC__)
+#  define XXH_ALIGN(n)      __attribute__ ((aligned(n)))
+#elif defined(_MSC_VER)
+#  define XXH_ALIGN(n)      __declspec(align(n))
+#else
+#  define XXH_ALIGN(n)   /* disabled */
+#endif
 
-  XXH64_hash_t total_len;  /*!< Total length hashed. This is always 64-bit. */
-  XXH64_hash_t v1;                              /*!< First accumulator lane */
-  XXH64_hash_t v2;                             /*!< Second accumulator lane */
-  XXH64_hash_t v3;                              /*!< Third accumulator lane */
-  XXH64_hash_t v4;                             /*!< Fourth accumulator lane */
-  XXH64_hash_t mem64[4];   /*!< Internal buffer for partial reads. Treated as
-                              unsigned char[32]. */
-  XXH32_hash_t memsize;                   /*!< Amount of data in @ref mem64 */
-  XXH32_hash_t reserved32;   /*!< Reserved field, needed for padding anyways*/
-  XXH64_hash_t reserved64; /*!< Reserved field. Do not read or write to it, it
-                              may be removed. */
-
-};                                            /* typedef'd to XXH64_state_t */
-
-    #if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L) /* >= C11 \
-                                                                    */
-      #include <stdalign.h>
-      #define XXH_ALIGN(n) alignas(n)
-    #elif defined(__cplusplus) && (__cplusplus >= 201103L)      /* >= C++11 */
-    /* In C++ alignas() is a keyword */
-      #define XXH_ALIGN(n) alignas(n)
-    #elif defined(__GNUC__)
-      #define XXH_ALIGN(n) __attribute__((aligned(n)))
-    #elif defined(_MSC_VER)
-      #define XXH_ALIGN(n) __declspec(align(n))
-    #else
-      #define XXH_ALIGN(n)                                      /* disabled */
-    #endif
-
-    /* Old GCC versions only accept the attribute after the type in structures.
-     */
-    #if !(defined(__STDC_VERSION__) &&                                        \
-          (__STDC_VERSION__ >= 201112L))                       /* C11+ */     \
-        && !(defined(__cplusplus) && (__cplusplus >= 201103L)) /* >= C++11 */ \
-        && defined(__GNUC__)
-      #define XXH_ALIGN_MEMBER(align, type) type XXH_ALIGN(align)
-    #else
-      #define XXH_ALIGN_MEMBER(align, type) XXH_ALIGN(align) type
-    #endif
-
-    /*!
-     * @brief The size of the internal XXH3 buffer.
-     *
-     * This is the optimal update size for incremental hashing.
-     *
-     * @see XXH3_64b_update(), XXH3_128b_update().
-     */
-    #define XXH3_INTERNALBUFFER_SIZE 256
+/* Old GCC versions only accept the attribute after the type in structures. */
+#if !(defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L))   /* C11+ */ \
+    && ! (defined(__cplusplus) && (__cplusplus >= 201103L)) /* >= C++11 */ \
+    && defined(__GNUC__)
+#   define XXH_ALIGN_MEMBER(align, type) type XXH_ALIGN(align)
+#else
+#   define XXH_ALIGN_MEMBER(align, type) XXH_ALIGN(align) type
+#endif
 
-    /*!
-     * @brief Default size of the secret buffer (and @ref XXH3_kSecret).
-     *
-     * This is the size used in @ref XXH3_kSecret and the seeded functions.
-     *
-     * Not to be confused with @ref XXH3_SECRET_SIZE_MIN.
-     */
-    #define XXH3_SECRET_DEFAULT_SIZE 192
+/*!
+ * @brief The size of the internal XXH3 buffer.
+ *
+ * This is the optimal update size for incremental hashing.
+ *
+ * @see XXH3_64b_update(), XXH3_128b_update().
+ */
+#define XXH3_INTERNALBUFFER_SIZE 256
+
+/*!
+ * @internal
+ * @brief Default size of the secret buffer (and @ref XXH3_kSecret).
+ *
+ * This is the size used in @ref XXH3_kSecret and the seeded functions.
+ *
+ * Not to be confused with @ref XXH3_SECRET_SIZE_MIN.
+ */
+#define XXH3_SECRET_DEFAULT_SIZE 192
 
 /*!
  * @internal
@@ -1144,111 +1744,284 @@ struct XXH64_state_s {
  * @see XXH32_state_s, XXH64_state_s
  */
 struct XXH3_state_s {
+   XXH_ALIGN_MEMBER(64, XXH64_hash_t acc[8]);
+       /*!< The 8 accumulators. See @ref XXH32_state_s::v and @ref XXH64_state_s::v */
+   XXH_ALIGN_MEMBER(64, unsigned char customSecret[XXH3_SECRET_DEFAULT_SIZE]);
+       /*!< Used to store a custom secret generated from a seed. */
+   XXH_ALIGN_MEMBER(64, unsigned char buffer[XXH3_INTERNALBUFFER_SIZE]);
+       /*!< The internal buffer. @see XXH32_state_s::mem32 */
+   XXH32_hash_t bufferedSize;
+       /*!< The amount of memory in @ref buffer, @see XXH32_state_s::memsize */
+   XXH32_hash_t useSeed;
+       /*!< Reserved field. Needed for padding on 64-bit. */
+   size_t nbStripesSoFar;
+       /*!< Number or stripes processed. */
+   XXH64_hash_t totalLen;
+       /*!< Total length hashed. 64-bit even on 32-bit targets. */
+   size_t nbStripesPerBlock;
+       /*!< Number of stripes per block. */
+   size_t secretLimit;
+       /*!< Size of @ref customSecret or @ref extSecret */
+   XXH64_hash_t seed;
+       /*!< Seed for _withSeed variants. Must be zero otherwise, @see XXH3_INITSTATE() */
+   XXH64_hash_t reserved64;
+       /*!< Reserved field. */
+   const unsigned char* extSecret;
+       /*!< Reference to an external secret for the _withSecret variants, NULL
+        *   for other variants. */
+   /* note: there may be some padding at the end due to alignment on 64 bytes */
+}; /* typedef'd to XXH3_state_t */
+
+#undef XXH_ALIGN_MEMBER
 
-  XXH_ALIGN_MEMBER(64, XXH64_hash_t acc[8]);
-  /*!< The 8 accumulators. Similar to `vN` in @ref XXH32_state_s::v1 and @ref
-   * XXH64_state_s */
-  XXH_ALIGN_MEMBER(64, unsigned char customSecret[XXH3_SECRET_DEFAULT_SIZE]);
-  /*!< Used to store a custom secret generated from a seed. */
-  XXH_ALIGN_MEMBER(64, unsigned char buffer[XXH3_INTERNALBUFFER_SIZE]);
-  /*!< The internal buffer. @see XXH32_state_s::mem32 */
-  XXH32_hash_t bufferedSize;
-  /*!< The amount of memory in @ref buffer, @see XXH32_state_s::memsize */
-  XXH32_hash_t reserved32;
-  /*!< Reserved field. Needed for padding on 64-bit. */
-  size_t nbStripesSoFar;
-  /*!< Number or stripes processed. */
-  XXH64_hash_t totalLen;
-  /*!< Total length hashed. 64-bit even on 32-bit targets. */
-  size_t nbStripesPerBlock;
-  /*!< Number of stripes per block. */
-  size_t secretLimit;
-  /*!< Size of @ref customSecret or @ref extSecret */
-  XXH64_hash_t seed;
-  /*!< Seed for _withSeed variants. Must be zero otherwise, @see
-   * XXH3_INITSTATE() */
-  XXH64_hash_t reserved64;
-  /*!< Reserved field. */
-  const unsigned char *extSecret;
-  /*!< Reference to an external secret for the _withSecret variants, NULL
-   *   for other variants. */
-  /* note: there may be some padding at the end due to alignment on 64 bytes */
-
-};                                             /* typedef'd to XXH3_state_t */
-
-    #undef XXH_ALIGN_MEMBER
-
-    /*!
-     * @brief Initializes a stack-allocated `XXH3_state_s`.
-     *
-     * When the @ref XXH3_state_t structure is merely emplaced on stack,
-     * it should be initialized with XXH3_INITSTATE() or a memset()
-     * in case its first reset uses XXH3_NNbits_reset_withSeed().
-     * This init can be omitted if the first reset uses default or _withSecret
-     * mode. This operation isn't necessary when the state is created with
-     * XXH3_createState(). Note that this doesn't prepare the state for a
-     * streaming operation, it's still necessary to use XXH3_NNbits_reset*()
-     * afterwards.
-     */
-    #define XXH3_INITSTATE(XXH3_state_ptr) \
-      { (XXH3_state_ptr)->seed = 0; }
+/*!
+ * @brief Initializes a stack-allocated `XXH3_state_s`.
+ *
+ * When the @ref XXH3_state_t structure is merely emplaced on stack,
+ * it should be initialized with XXH3_INITSTATE() or a memset()
+ * in case its first reset uses XXH3_NNbits_reset_withSeed().
+ * This init can be omitted if the first reset uses default or _withSecret mode.
+ * This operation isn't necessary when the state is created with XXH3_createState().
+ * Note that this doesn't prepare the state for a streaming operation,
+ * it's still necessary to use XXH3_NNbits_reset*() afterwards.
+ */
+#define XXH3_INITSTATE(XXH3_state_ptr)                       \
+    do {                                                     \
+        XXH3_state_t* tmp_xxh3_state_ptr = (XXH3_state_ptr); \
+        tmp_xxh3_state_ptr->seed = 0;                        \
+        tmp_xxh3_state_ptr->extSecret = NULL;                \
+    } while(0)
 
-/* ===   Experimental API   === */
-/* Symbols defined below must be considered tied to a specific library version.
+
+/*!
+ * @brief Calculates the 128-bit hash of @p data using XXH3.
+ *
+ * @param data The block of data to be hashed, at least @p len bytes in size.
+ * @param len  The length of @p data, in bytes.
+ * @param seed The 64-bit seed to alter the hash's output predictably.
+ *
+ * @pre
+ *   The memory between @p data and @p data + @p len must be valid,
+ *   readable, contiguous memory. However, if @p len is `0`, @p data may be
+ *   `NULL`. In C++, this also must be *TriviallyCopyable*.
+ *
+ * @return The calculated 128-bit XXH3 value.
+ *
+ * @see @ref single_shot_example "Single Shot Example" for an example.
  */
+XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH128(XXH_NOESCAPE const void* data, size_t len, XXH64_hash_t seed);
 
-/*
- * XXH3_generateSecret():
+
+/* ===   Experimental API   === */
+/* Symbols defined below must be considered tied to a specific library version. */
+
+/*!
+ * @brief Derive a high-entropy secret from any user-defined content, named customSeed.
+ *
+ * @param secretBuffer    A writable buffer for derived high-entropy secret data.
+ * @param secretSize      Size of secretBuffer, in bytes.  Must be >= XXH3_SECRET_DEFAULT_SIZE.
+ * @param customSeed      A user-defined content.
+ * @param customSeedSize  Size of customSeed, in bytes.
  *
- * Derive a high-entropy secret from any user-defined content, named customSeed.
- * The generated secret can be used in combination with `*_withSecret()`
- * functions. The `_withSecret()` variants are useful to provide a higher level
- * of protection than 64-bit seed, as it becomes much more difficult for an
- * external actor to guess how to impact the calculation logic.
+ * @return @ref XXH_OK on success.
+ * @return @ref XXH_ERROR on failure.
+ *
+ * The generated secret can be used in combination with `*_withSecret()` functions.
+ * The `_withSecret()` variants are useful to provide a higher level of protection
+ * than 64-bit seed, as it becomes much more difficult for an external actor to
+ * guess how to impact the calculation logic.
  *
  * The function accepts as input a custom seed of any length and any content,
- * and derives from it a high-entropy secret of length XXH3_SECRET_DEFAULT_SIZE
- * into an already allocated buffer secretBuffer.
- * The generated secret is _always_ XXH_SECRET_DEFAULT_SIZE bytes long.
+ * and derives from it a high-entropy secret of length @p secretSize into an
+ * already allocated buffer @p secretBuffer.
  *
  * The generated secret can then be used with any `*_withSecret()` variant.
- * Functions `XXH3_128bits_withSecret()`, `XXH3_64bits_withSecret()`,
- * `XXH3_128bits_reset_withSecret()` and `XXH3_64bits_reset_withSecret()`
+ * The functions @ref XXH3_128bits_withSecret(), @ref XXH3_64bits_withSecret(),
+ * @ref XXH3_128bits_reset_withSecret() and @ref XXH3_64bits_reset_withSecret()
  * are part of this list. They all accept a `secret` parameter
- * which must be very long for implementation reasons (>= XXH3_SECRET_SIZE_MIN)
+ * which must be large enough for implementation reasons (>= @ref XXH3_SECRET_SIZE_MIN)
  * _and_ feature very high entropy (consist of random-looking bytes).
- * These conditions can be a high bar to meet, so
- * this function can be used to generate a secret of proper quality.
+ * These conditions can be a high bar to meet, so @ref XXH3_generateSecret() can
+ * be employed to ensure proper quality.
+ *
+ * @p customSeed can be anything. It can have any size, even small ones,
+ * and its content can be anything, even "poor entropy" sources such as a bunch
+ * of zeroes. The resulting `secret` will nonetheless provide all required qualities.
  *
- * customSeed can be anything. It can have any size, even small ones,
- * and its content can be anything, even stupidly "low entropy" source such as a
- * bunch of zeroes. The resulting `secret` will nonetheless provide all expected
- * qualities.
+ * @pre
+ *   - @p secretSize must be >= @ref XXH3_SECRET_SIZE_MIN
+ *   - When @p customSeedSize > 0, supplying NULL as customSeed is undefined behavior.
  *
- * Supplying NULL as the customSeed copies the default secret into
- * `secretBuffer`. When customSeedSize > 0, supplying NULL as customSeed is
- * undefined behavior.
+ * Example code:
+ * @code{.c}
+ *    #include <stdio.h>
+ *    #include <stdlib.h>
+ *    #include <string.h>
+ *    #define XXH_STATIC_LINKING_ONLY // expose unstable API
+ *    #include "xxhash.h"
+ *    // Hashes argv[2] using the entropy from argv[1].
+ *    int main(int argc, char* argv[])
+ *    {
+ *        char secret[XXH3_SECRET_SIZE_MIN];
+ *        if (argv != 3) { return 1; }
+ *        XXH3_generateSecret(secret, sizeof(secret), argv[1], strlen(argv[1]));
+ *        XXH64_hash_t h = XXH3_64bits_withSecret(
+ *             argv[2], strlen(argv[2]),
+ *             secret, sizeof(secret)
+ *        );
+ *        printf("%016llx\n", (unsigned long long) h);
+ *    }
+ * @endcode
  */
-XXH_PUBLIC_API void XXH3_generateSecret(void       *secretBuffer,
-                                        const void *customSeed,
-                                        size_t      customSeedSize);
+XXH_PUBLIC_API XXH_errorcode XXH3_generateSecret(XXH_NOESCAPE void* secretBuffer, size_t secretSize, XXH_NOESCAPE const void* customSeed, size_t customSeedSize);
 
-/* simple short-cut to pre-selected XXH3_128bits variant */
-XXH_PUBLIC_API XXH128_hash_t XXH128(const void *data, size_t len,
-                                    XXH64_hash_t seed);
+/*!
+ * @brief Generate the same secret as the _withSeed() variants.
+ *
+ * @param secretBuffer A writable buffer of @ref XXH3_SECRET_SIZE_MIN bytes
+ * @param seed         The 64-bit seed to alter the hash result predictably.
+ *
+ * The generated secret can be used in combination with
+ *`*_withSecret()` and `_withSecretandSeed()` variants.
+ *
+ * Example C++ `std::string` hash class:
+ * @code{.cpp}
+ *    #include <string>
+ *    #define XXH_STATIC_LINKING_ONLY // expose unstable API
+ *    #include "xxhash.h"
+ *    // Slow, seeds each time
+ *    class HashSlow {
+ *        XXH64_hash_t seed;
+ *    public:
+ *        HashSlow(XXH64_hash_t s) : seed{s} {}
+ *        size_t operator()(const std::string& x) const {
+ *            return size_t{XXH3_64bits_withSeed(x.c_str(), x.length(), seed)};
+ *        }
+ *    };
+ *    // Fast, caches the seeded secret for future uses.
+ *    class HashFast {
+ *        unsigned char secret[XXH3_SECRET_SIZE_MIN];
+ *    public:
+ *        HashFast(XXH64_hash_t s) {
+ *            XXH3_generateSecret_fromSeed(secret, seed);
+ *        }
+ *        size_t operator()(const std::string& x) const {
+ *            return size_t{
+ *                XXH3_64bits_withSecret(x.c_str(), x.length(), secret, sizeof(secret))
+ *            };
+ *        }
+ *    };
+ * @endcode
+ */
+XXH_PUBLIC_API void XXH3_generateSecret_fromSeed(XXH_NOESCAPE void* secretBuffer, XXH64_hash_t seed);
+
+/*!
+ * @brief Calculates 64/128-bit seeded variant of XXH3 hash of @p data.
+ *
+ * @param data       The block of data to be hashed, at least @p len bytes in size.
+ * @param len        The length of @p data, in bytes.
+ * @param secret     The secret data.
+ * @param secretSize The length of @p secret, in bytes.
+ * @param seed       The 64-bit seed to alter the hash result predictably.
+ *
+ * These variants generate hash values using either
+ * @p seed for "short" keys (< @ref XXH3_MIDSIZE_MAX = 240 bytes)
+ * or @p secret for "large" keys (>= @ref XXH3_MIDSIZE_MAX).
+ *
+ * This generally benefits speed, compared to `_withSeed()` or `_withSecret()`.
+ * `_withSeed()` has to generate the secret on the fly for "large" keys.
+ * It's fast, but can be perceptible for "not so large" keys (< 1 KB).
+ * `_withSecret()` has to generate the masks on the fly for "small" keys,
+ * which requires more instructions than _withSeed() variants.
+ * Therefore, _withSecretandSeed variant combines the best of both worlds.
+ *
+ * When @p secret has been generated by XXH3_generateSecret_fromSeed(),
+ * this variant produces *exactly* the same results as `_withSeed()` variant,
+ * hence offering only a pure speed benefit on "large" input,
+ * by skipping the need to regenerate the secret for every large input.
+ *
+ * Another usage scenario is to hash the secret to a 64-bit hash value,
+ * for example with XXH3_64bits(), which then becomes the seed,
+ * and then employ both the seed and the secret in _withSecretandSeed().
+ * On top of speed, an added benefit is that each bit in the secret
+ * has a 50% chance to swap each bit in the output, via its impact to the seed.
+ *
+ * This is not guaranteed when using the secret directly in "small data" scenarios,
+ * because only portions of the secret are employed for small data.
+ */
+XXH_PUBLIC_API XXH_PUREF XXH64_hash_t
+XXH3_64bits_withSecretandSeed(XXH_NOESCAPE const void* data, size_t len,
+                              XXH_NOESCAPE const void* secret, size_t secretSize,
+                              XXH64_hash_t seed);
+/*!
+ * @brief Calculates 128-bit seeded variant of XXH3 hash of @p data.
+ *
+ * @param input      The block of data to be hashed, at least @p len bytes in size.
+ * @param length     The length of @p data, in bytes.
+ * @param secret     The secret data.
+ * @param secretSize The length of @p secret, in bytes.
+ * @param seed64     The 64-bit seed to alter the hash result predictably.
+ *
+ * @return @ref XXH_OK on success.
+ * @return @ref XXH_ERROR on failure.
+ *
+ * @see XXH3_64bits_withSecretandSeed()
+ */
+XXH_PUBLIC_API XXH_PUREF XXH128_hash_t
+XXH3_128bits_withSecretandSeed(XXH_NOESCAPE const void* input, size_t length,
+                               XXH_NOESCAPE const void* secret, size_t secretSize,
+                               XXH64_hash_t seed64);
+#ifndef XXH_NO_STREAM
+/*!
+ * @brief Resets an @ref XXH3_state_t with secret data to begin a new hash.
+ *
+ * @param statePtr   A pointer to an @ref XXH3_state_t allocated with @ref XXH3_createState().
+ * @param secret     The secret data.
+ * @param secretSize The length of @p secret, in bytes.
+ * @param seed64     The 64-bit seed to alter the hash result predictably.
+ *
+ * @return @ref XXH_OK on success.
+ * @return @ref XXH_ERROR on failure.
+ *
+ * @see XXH3_64bits_withSecretandSeed()
+ */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr,
+                                    XXH_NOESCAPE const void* secret, size_t secretSize,
+                                    XXH64_hash_t seed64);
+/*!
+ * @brief Resets an @ref XXH3_state_t with secret data to begin a new hash.
+ *
+ * @param statePtr   A pointer to an @ref XXH3_state_t allocated with @ref XXH3_createState().
+ * @param secret     The secret data.
+ * @param secretSize The length of @p secret, in bytes.
+ * @param seed64     The 64-bit seed to alter the hash result predictably.
+ *
+ * @return @ref XXH_OK on success.
+ * @return @ref XXH_ERROR on failure.
+ *
+ * @see XXH3_64bits_withSecretandSeed()
+ */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr,
+                                     XXH_NOESCAPE const void* secret, size_t secretSize,
+                                     XXH64_hash_t seed64);
+#endif /* !XXH_NO_STREAM */
+
+#endif  /* !XXH_NO_XXH3 */
+#endif  /* XXH_NO_LONG_LONG */
+#if defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API)
+#  define XXH_IMPLEMENTATION
+#endif
 
-  #endif                                                /* XXH_NO_LONG_LONG */
-  #if defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API)
-    #define XXH_IMPLEMENTATION
-  #endif
+#endif  /* defined(XXH_STATIC_LINKING_ONLY) && !defined(XXHASH_H_STATIC_13879238742) */
 
-#endif /* defined(XXH_STATIC_LINKING_ONLY) && \
-          !defined(XXHASH_H_STATIC_13879238742) */
 
 /* ======================================================================== */
 /* ======================================================================== */
 /* ======================================================================== */
 
+
 /*-**********************************************************************
  * xxHash implementation
  *-**********************************************************************
@@ -1271,424 +2044,477 @@ XXH_PUBLIC_API XXH128_hash_t XXH128(const void *data, size_t len,
  * which can then be linked into the final binary.
  ************************************************************************/
 
-#if (defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API) || \
-     defined(XXH_IMPLEMENTATION)) &&                        \
-    !defined(XXH_IMPLEM_13a8737387)
-  #define XXH_IMPLEM_13a8737387
-
-  /* *************************************
-   *  Tuning parameters
-   ***************************************/
-
-  /*!
-   * @defgroup tuning Tuning parameters
-   * @{
-
-   *
-   * Various macros to control xxHash's behavior.
-   */
-  #ifdef XXH_DOXYGEN
-    /*!
-     * @brief Define this to disable 64-bit code.
-     *
-     * Useful if only using the @ref xxh32_family and you have a strict C90
-     * compiler.
-     */
-    #define XXH_NO_LONG_LONG
-    #undef XXH_NO_LONG_LONG                               /* don't actually */
-    /*!
-     * @brief Controls how unaligned memory is accessed.
-     *
-     * By default, access to unaligned memory is controlled by `memcpy()`, which
-     * is safe and portable.
-     *
-     * Unfortunately, on some target/compiler combinations, the generated
-     * assembly is sub-optimal.
-     *
-     * The below switch allow selection of a different access method
-     * in the search for improved performance.
-     *
-     * @par Possible options:
-     *
-     *  - `XXH_FORCE_MEMORY_ACCESS=0` (default): `memcpy`
-     *   @par
-     *     Use `memcpy()`. Safe and portable. Note that most modern compilers
-     * will eliminate the function call and treat it as an unaligned access.
-     *
-     *  - `XXH_FORCE_MEMORY_ACCESS=1`: `__attribute__((packed))`
-     *   @par
-     *     Depends on compiler extensions and is therefore not portable.
-     *     This method is safe _if_ your compiler supports it,
-     *     and *generally* as fast or faster than `memcpy`.
-     *
-     *  - `XXH_FORCE_MEMORY_ACCESS=2`: Direct cast
-     *  @par
-     *     Casts directly and dereferences. This method doesn't depend on the
-     *     compiler, but it violates the C standard as it directly dereferences
-     * an unaligned pointer. It can generate buggy code on targets which do not
-     *     support unaligned memory accesses, but in some circumstances, it's
-     * the only known way to get the most performance.
-     *
-     *  - `XXH_FORCE_MEMORY_ACCESS=3`: Byteshift
-     *  @par
-     *     Also portable. This can generate the best code on old compilers which
-     * don't inline small `memcpy()` calls, and it might also be faster on
-     * big-endian systems which lack a native byteswap instruction. However,
-     * some compilers will emit literal byteshifts even if the target supports
-     * unaligned access.
-     *  .
-     *
-     * @warning
-     *   Methods 1 and 2 rely on implementation-defined behavior. Use these with
-     *   care, as what works on one compiler/platform/optimization level may
-     * cause another to read garbage data or even crash.
-     *
-     * See https://stackoverflow.com/a/32095106/646947 for details.
-     *
-     * Prefer these methods in priority order (0 > 3 > 1 > 2)
-     */
-    #define XXH_FORCE_MEMORY_ACCESS 0
-    /*!
-     * @def XXH_ACCEPT_NULL_INPUT_POINTER
-     * @brief Whether to add explicit `NULL` checks.
-     *
-     * If the input pointer is `NULL` and the length is non-zero, xxHash's
-     * default behavior is to dereference it, triggering a segfault.
-     *
-     * When this macro is enabled, xxHash actively checks the input for a null
-     * pointer. If it is, the result for null input pointers is the same as a
-     * zero-length input.
-     */
-    #define XXH_ACCEPT_NULL_INPUT_POINTER 0
-    /*!
-     * @def XXH_FORCE_ALIGN_CHECK
-     * @brief If defined to non-zero, adds a special path for aligned inputs
-     * (XXH32() and XXH64() only).
-     *
-     * This is an important performance trick for architectures without decent
-     * unaligned memory access performance.
-     *
-     * It checks for input alignment, and when conditions are met, uses a "fast
-     * path" employing direct 32-bit/64-bit reads, resulting in _dramatically
-     * faster_ read speed.
-     *
-     * The check costs one initial branch per hash, which is generally
-     * negligible, but not zero.
-     *
-     * Moreover, it's not useful to generate an additional code path if memory
-     * access uses the same instruction for both aligned and unaligned
-     * addresses (e.g. x86 and aarch64).
-     *
-     * In these cases, the alignment check can be removed by setting this macro
-     * to 0. Then the code will always use unaligned memory access. Align check
-     * is automatically disabled on x86, x64 & arm64, which are platforms known
-     * to offer good unaligned memory accesses performance.
-     *
-     * This option does not affect XXH3 (only XXH32 and XXH64).
-     */
-    #define XXH_FORCE_ALIGN_CHECK 0
-
-    /*!
-     * @def XXH_NO_INLINE_HINTS
-     * @brief When non-zero, sets all functions to `static`.
-     *
-     * By default, xxHash tries to force the compiler to inline almost all
-     * internal functions.
-     *
-     * This can usually improve performance due to reduced jumping and improved
-     * constant folding, but significantly increases the size of the binary
-     * which might not be favorable.
-     *
-     * Additionally, sometimes the forced inlining can be detrimental to
-     * performance, depending on the architecture.
-     *
-     * XXH_NO_INLINE_HINTS marks all internal functions as static, giving the
-     * compiler full control on whether to inline or not.
-     *
-     * When not optimizing (-O0), optimizing for size (-Os, -Oz), or using
-     * -fno-inline with GCC or Clang, this will automatically be defined.
-     */
-    #define XXH_NO_INLINE_HINTS 0
-
-    /*!
-     * @def XXH_REROLL
-     * @brief Whether to reroll `XXH32_finalize`.
-     *
-     * For performance, `XXH32_finalize` uses an unrolled loop
-     * in the form of a switch statement.
-     *
-     * This is not always desirable, as it generates larger code,
-     * and depending on the architecture, may even be slower
-     *
-     * This is automatically defined with `-Os`/`-Oz` on GCC and Clang.
-     */
-    #define XXH_REROLL 0
+#if ( defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API) \
+   || defined(XXH_IMPLEMENTATION) ) && !defined(XXH_IMPLEM_13a8737387)
+#  define XXH_IMPLEM_13a8737387
 
-    /*!
-     * @internal
-     * @brief Redefines old internal names.
-     *
-     * For compatibility with code that uses xxHash's internals before the names
-     * were changed to improve namespacing. There is no other reason to use
-     * this.
-     */
-    #define XXH_OLD_NAMES
-    #undef XXH_OLD_NAMES                 /* don't actually use, it is ugly. */
-  #endif                                                     /* XXH_DOXYGEN */
-                         /*!
-                          * @}
-                          */
-
-  #ifndef XXH_FORCE_MEMORY_ACCESS /* can be defined externally, on command \
-                                     line for example */
-  /* prefer __packed__ structures (method 1) for gcc on armv7+ and mips */
-    #if !defined(__clang__) &&                                          \
-        ((defined(__INTEL_COMPILER) && !defined(_WIN32)) ||             \
-         (defined(__GNUC__) &&                                          \
-          ((defined(__ARM_ARCH) && __ARM_ARCH >= 7) ||                  \
-           (defined(__mips__) && (__mips <= 5 || __mips_isa_rev < 6) && \
-            (!defined(__mips16) || defined(__mips_mips16e2))))))
-      #define XXH_FORCE_MEMORY_ACCESS 1
-    #endif
-  #endif
-
-  #ifndef XXH_ACCEPT_NULL_INPUT_POINTER        /* can be defined externally */
-    #define XXH_ACCEPT_NULL_INPUT_POINTER 0
-  #endif
-
-  #ifndef XXH_FORCE_ALIGN_CHECK                /* can be defined externally */
-    #if defined(__i386) || defined(__x86_64__) || defined(__aarch64__) || \
-        defined(_M_IX86) || defined(_M_X64) || defined(_M_ARM64)  /* visual */
-      #define XXH_FORCE_ALIGN_CHECK 0
-    #else
-      #define XXH_FORCE_ALIGN_CHECK 1
-    #endif
-  #endif
-
-  #ifndef XXH_NO_INLINE_HINTS
-    #if defined(__OPTIMIZE_SIZE__) /* -Os, -Oz */ \
-        || defined(__NO_INLINE__)                       /* -O0, -fno-inline */
-      #define XXH_NO_INLINE_HINTS 1
-    #else
-      #define XXH_NO_INLINE_HINTS 0
-    #endif
-  #endif
-
-  #ifndef XXH_REROLL
-    #if defined(__OPTIMIZE_SIZE__) /* -Os, -Oz */ || \
-        (defined(__GNUC__) && !defined(__clang__))
-    /* The if/then loop is preferable to switch/case on gcc (on x64) */
-      #define XXH_REROLL 1
-    #else
-      #define XXH_REROLL 0
-    #endif
-  #endif
-
-  /*!
-   * @defgroup impl Implementation
-   * @{
-
-   */
-
-  /* *************************************
-   *  Includes & Memory related functions
-   ***************************************/
-  /*
-   * Modify the local functions below should you wish to use
-   * different memory routines for malloc() and free()
-   */
-  #include <stdlib.h>
+/* *************************************
+*  Tuning parameters
+***************************************/
 
 /*!
- * @internal
- * @brief Modify this function to use a different routine than malloc().
+ * @defgroup tuning Tuning parameters
+ * @{
+ *
+ * Various macros to control xxHash's behavior.
  */
-static void *XXH_malloc(size_t s) {
-
-  return malloc(s);
-
-}
-
+#ifdef XXH_DOXYGEN
 /*!
- * @internal
- * @brief Modify this function to use a different routine than free().
+ * @brief Define this to disable 64-bit code.
+ *
+ * Useful if only using the @ref XXH32_family and you have a strict C90 compiler.
  */
-static void XXH_free(void *p) {
-
-  free(p);
-
-}
-
-  #include <string.h>
-
+#  define XXH_NO_LONG_LONG
+#  undef XXH_NO_LONG_LONG /* don't actually */
 /*!
- * @internal
- * @brief Modify this function to use a different routine than memcpy().
+ * @brief Controls how unaligned memory is accessed.
+ *
+ * By default, access to unaligned memory is controlled by `memcpy()`, which is
+ * safe and portable.
+ *
+ * Unfortunately, on some target/compiler combinations, the generated assembly
+ * is sub-optimal.
+ *
+ * The below switch allow selection of a different access method
+ * in the search for improved performance.
+ *
+ * @par Possible options:
+ *
+ *  - `XXH_FORCE_MEMORY_ACCESS=0` (default): `memcpy`
+ *   @par
+ *     Use `memcpy()`. Safe and portable. Note that most modern compilers will
+ *     eliminate the function call and treat it as an unaligned access.
+ *
+ *  - `XXH_FORCE_MEMORY_ACCESS=1`: `__attribute__((aligned(1)))`
+ *   @par
+ *     Depends on compiler extensions and is therefore not portable.
+ *     This method is safe _if_ your compiler supports it,
+ *     and *generally* as fast or faster than `memcpy`.
+ *
+ *  - `XXH_FORCE_MEMORY_ACCESS=2`: Direct cast
+ *  @par
+ *     Casts directly and dereferences. This method doesn't depend on the
+ *     compiler, but it violates the C standard as it directly dereferences an
+ *     unaligned pointer. It can generate buggy code on targets which do not
+ *     support unaligned memory accesses, but in some circumstances, it's the
+ *     only known way to get the most performance.
+ *
+ *  - `XXH_FORCE_MEMORY_ACCESS=3`: Byteshift
+ *  @par
+ *     Also portable. This can generate the best code on old compilers which don't
+ *     inline small `memcpy()` calls, and it might also be faster on big-endian
+ *     systems which lack a native byteswap instruction. However, some compilers
+ *     will emit literal byteshifts even if the target supports unaligned access.
+ *
+ *
+ * @warning
+ *   Methods 1 and 2 rely on implementation-defined behavior. Use these with
+ *   care, as what works on one compiler/platform/optimization level may cause
+ *   another to read garbage data or even crash.
+ *
+ * See https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html for details.
+ *
+ * Prefer these methods in priority order (0 > 3 > 1 > 2)
  */
-static void *XXH_memcpy(void *dest, const void *src, size_t size) {
-
-  return memcpy(dest, src, size);
-
-}
-
-  #include <limits.h>                                         /* ULLONG_MAX */
-
-  /* *************************************
-   *  Compiler Specific Options
-   ***************************************/
-  #ifdef _MSC_VER                              /* Visual Studio warning fix */
-    #pragma warning(disable : 4127) /* disable: C4127: conditional expression \
-                                       is constant */
-  #endif
-
-  #if XXH_NO_INLINE_HINTS                         /* disable inlining hints */
-    #if defined(__GNUC__)
-      #define XXH_FORCE_INLINE static __attribute__((unused))
-    #else
-      #define XXH_FORCE_INLINE static
-    #endif
-    #define XXH_NO_INLINE static
-  /* enable inlining hints */
-  #elif defined(_MSC_VER)                                  /* Visual Studio */
-    #define XXH_FORCE_INLINE static __forceinline
-    #define XXH_NO_INLINE static __declspec(noinline)
-  #elif defined(__GNUC__)
-    #define XXH_FORCE_INLINE \
-      static __inline__ __attribute__((always_inline, unused))
-    #define XXH_NO_INLINE static __attribute__((noinline))
-  #elif defined(__cplusplus) || \
-      (defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L))   /* C99 */
-    #define XXH_FORCE_INLINE static inline
-    #define XXH_NO_INLINE static
-  #else
-    #define XXH_FORCE_INLINE static
-    #define XXH_NO_INLINE static
-  #endif
-
-  /* *************************************
-   *  Debug
-   ***************************************/
-  /*!
-   * @ingroup tuning
-   * @def XXH_DEBUGLEVEL
-   * @brief Sets the debugging level.
-   *
-   * XXH_DEBUGLEVEL is expected to be defined externally, typically via the
-   * compiler's command line options. The value must be a number.
-   */
-  #ifndef XXH_DEBUGLEVEL
-    #ifdef DEBUGLEVEL                                   /* backwards compat */
-      #define XXH_DEBUGLEVEL DEBUGLEVEL
-    #else
-      #define XXH_DEBUGLEVEL 0
-    #endif
-  #endif
-
-  #if (XXH_DEBUGLEVEL >= 1)
-    #include <assert.h>          /* note: can still be disabled with NDEBUG */
-    #define XXH_ASSERT(c) assert(c)
-  #else
-    #define XXH_ASSERT(c) ((void)0)
-  #endif
-
-  /* note: use after variable declarations */
-  #ifndef XXH_STATIC_ASSERT
-    #if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L)   /* C11 */
-      #include <assert.h>
-      #define XXH_STATIC_ASSERT_WITH_MESSAGE(c, m) \
-        do {                                       \
-                                                   \
-          static_assert((c), m);                   \
-                                                   \
-        } while (0)
-
-    #elif defined(__cplusplus) && (__cplusplus >= 201103L)         /* C++11 */
-      #define XXH_STATIC_ASSERT_WITH_MESSAGE(c, m) \
-        do {                                       \
-                                                   \
-          static_assert((c), m);                   \
-                                                   \
-        } while (0)
-
-    #else
-      #define XXH_STATIC_ASSERT_WITH_MESSAGE(c, m) \
-        do {                                       \
-                                                   \
-          struct xxh_sa {                          \
-                                                   \
-            char x[(c) ? 1 : -1];                  \
-                                                   \
-          };                                       \
-                                                   \
-        } while (0)
-
-    #endif
-    #define XXH_STATIC_ASSERT(c) XXH_STATIC_ASSERT_WITH_MESSAGE((c), #c)
-  #endif
-
-  /*!
-   * @internal
-   * @def XXH_COMPILER_GUARD(var)
-   * @brief Used to prevent unwanted optimizations for @p var.
-   *
-   * It uses an empty GCC inline assembly statement with a register constraint
-   * which forces @p var into a general purpose register (eg eax, ebx, ecx
-   * on x86) and marks it as modified.
-   *
-   * This is used in a few places to avoid unwanted autovectorization (e.g.
-   * XXH32_round()). All vectorization we want is explicit via intrinsics,
-   * and _usually_ isn't wanted elsewhere.
-   *
-   * We also use it to prevent unwanted constant folding for AArch64 in
-   * XXH3_initCustomSecret_scalar().
-   */
-  #ifdef __GNUC__
-    #define XXH_COMPILER_GUARD(var) __asm__ __volatile__("" : "+r"(var))
-  #else
-    #define XXH_COMPILER_GUARD(var) ((void)0)
-  #endif
-
-  /* *************************************
-   *  Basic Types
-   ***************************************/
-  #if !defined(__VMS) &&       \
-      (defined(__cplusplus) || \
-       (defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */))
-    #include <stdint.h>
-typedef uint8_t xxh_u8;
-  #else
-typedef unsigned char xxh_u8;
-  #endif
-typedef XXH32_hash_t xxh_u32;
+#  define XXH_FORCE_MEMORY_ACCESS 0
 
-  #ifdef XXH_OLD_NAMES
-    #define BYTE xxh_u8
-    #define U8 xxh_u8
-    #define U32 xxh_u32
-  #endif
-
-/* ***   Memory access   *** */
+/*!
+ * @def XXH_SIZE_OPT
+ * @brief Controls how much xxHash optimizes for size.
+ *
+ * xxHash, when compiled, tends to result in a rather large binary size. This
+ * is mostly due to heavy usage to forced inlining and constant folding of the
+ * @ref XXH3_family to increase performance.
+ *
+ * However, some developers prefer size over speed. This option can
+ * significantly reduce the size of the generated code. When using the `-Os`
+ * or `-Oz` options on GCC or Clang, this is defined to 1 by default,
+ * otherwise it is defined to 0.
+ *
+ * Most of these size optimizations can be controlled manually.
+ *
+ * This is a number from 0-2.
+ *  - `XXH_SIZE_OPT` == 0: Default. xxHash makes no size optimizations. Speed
+ *    comes first.
+ *  - `XXH_SIZE_OPT` == 1: Default for `-Os` and `-Oz`. xxHash is more
+ *    conservative and disables hacks that increase code size. It implies the
+ *    options @ref XXH_NO_INLINE_HINTS == 1, @ref XXH_FORCE_ALIGN_CHECK == 0,
+ *    and @ref XXH3_NEON_LANES == 8 if they are not already defined.
+ *  - `XXH_SIZE_OPT` == 2: xxHash tries to make itself as small as possible.
+ *    Performance may cry. For example, the single shot functions just use the
+ *    streaming API.
+ */
+#  define XXH_SIZE_OPT 0
 
 /*!
- * @internal
- * @fn xxh_u32 XXH_read32(const void* ptr)
- * @brief Reads an unaligned 32-bit integer from @p ptr in native endianness.
+ * @def XXH_FORCE_ALIGN_CHECK
+ * @brief If defined to non-zero, adds a special path for aligned inputs (XXH32()
+ * and XXH64() only).
  *
- * Affected by @ref XXH_FORCE_MEMORY_ACCESS.
+ * This is an important performance trick for architectures without decent
+ * unaligned memory access performance.
  *
- * @param ptr The pointer to read from.
- * @return The 32-bit native endian integer from the bytes at @p ptr.
+ * It checks for input alignment, and when conditions are met, uses a "fast
+ * path" employing direct 32-bit/64-bit reads, resulting in _dramatically
+ * faster_ read speed.
+ *
+ * The check costs one initial branch per hash, which is generally negligible,
+ * but not zero.
+ *
+ * Moreover, it's not useful to generate an additional code path if memory
+ * access uses the same instruction for both aligned and unaligned
+ * addresses (e.g. x86 and aarch64).
+ *
+ * In these cases, the alignment check can be removed by setting this macro to 0.
+ * Then the code will always use unaligned memory access.
+ * Align check is automatically disabled on x86, x64, ARM64, and some ARM chips
+ * which are platforms known to offer good unaligned memory accesses performance.
+ *
+ * It is also disabled by default when @ref XXH_SIZE_OPT >= 1.
+ *
+ * This option does not affect XXH3 (only XXH32 and XXH64).
  */
+#  define XXH_FORCE_ALIGN_CHECK 0
 
 /*!
- * @internal
- * @fn xxh_u32 XXH_readLE32(const void* ptr)
- * @brief Reads an unaligned 32-bit little endian integer from @p ptr.
+ * @def XXH_NO_INLINE_HINTS
+ * @brief When non-zero, sets all functions to `static`.
  *
- * Affected by @ref XXH_FORCE_MEMORY_ACCESS.
+ * By default, xxHash tries to force the compiler to inline almost all internal
+ * functions.
  *
- * @param ptr The pointer to read from.
- * @return The 32-bit little endian integer from the bytes at @p ptr.
+ * This can usually improve performance due to reduced jumping and improved
+ * constant folding, but significantly increases the size of the binary which
+ * might not be favorable.
+ *
+ * Additionally, sometimes the forced inlining can be detrimental to performance,
+ * depending on the architecture.
+ *
+ * XXH_NO_INLINE_HINTS marks all internal functions as static, giving the
+ * compiler full control on whether to inline or not.
+ *
+ * When not optimizing (-O0), using `-fno-inline` with GCC or Clang, or if
+ * @ref XXH_SIZE_OPT >= 1, this will automatically be defined.
+ */
+#  define XXH_NO_INLINE_HINTS 0
+
+/*!
+ * @def XXH3_INLINE_SECRET
+ * @brief Determines whether to inline the XXH3 withSecret code.
+ *
+ * When the secret size is known, the compiler can improve the performance
+ * of XXH3_64bits_withSecret() and XXH3_128bits_withSecret().
+ *
+ * However, if the secret size is not known, it doesn't have any benefit. This
+ * happens when xxHash is compiled into a global symbol. Therefore, if
+ * @ref XXH_INLINE_ALL is *not* defined, this will be defined to 0.
+ *
+ * Additionally, this defaults to 0 on GCC 12+, which has an issue with function pointers
+ * that are *sometimes* force inline on -Og, and it is impossible to automatically
+ * detect this optimization level.
+ */
+#  define XXH3_INLINE_SECRET 0
+
+/*!
+ * @def XXH32_ENDJMP
+ * @brief Whether to use a jump for `XXH32_finalize`.
+ *
+ * For performance, `XXH32_finalize` uses multiple branches in the finalizer.
+ * This is generally preferable for performance,
+ * but depending on exact architecture, a jmp may be preferable.
+ *
+ * This setting is only possibly making a difference for very small inputs.
+ */
+#  define XXH32_ENDJMP 0
+
+/*!
+ * @internal
+ * @brief Redefines old internal names.
+ *
+ * For compatibility with code that uses xxHash's internals before the names
+ * were changed to improve namespacing. There is no other reason to use this.
+ */
+#  define XXH_OLD_NAMES
+#  undef XXH_OLD_NAMES /* don't actually use, it is ugly. */
+
+/*!
+ * @def XXH_NO_STREAM
+ * @brief Disables the streaming API.
+ *
+ * When xxHash is not inlined and the streaming functions are not used, disabling
+ * the streaming functions can improve code size significantly, especially with
+ * the @ref XXH3_family which tends to make constant folded copies of itself.
+ */
+#  define XXH_NO_STREAM
+#  undef XXH_NO_STREAM /* don't actually */
+#endif /* XXH_DOXYGEN */
+/*!
+ * @}
+ */
+
+#ifndef XXH_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
+   /* prefer __packed__ structures (method 1) for GCC
+    * < ARMv7 with unaligned access (e.g. Raspbian armhf) still uses byte shifting, so we use memcpy
+    * which for some reason does unaligned loads. */
+#  if defined(__GNUC__) && !(defined(__ARM_ARCH) && __ARM_ARCH < 7 && defined(__ARM_FEATURE_UNALIGNED))
+#    define XXH_FORCE_MEMORY_ACCESS 1
+#  endif
+#endif
+
+#ifndef XXH_SIZE_OPT
+   /* default to 1 for -Os or -Oz */
+#  if (defined(__GNUC__) || defined(__clang__)) && defined(__OPTIMIZE_SIZE__)
+#    define XXH_SIZE_OPT 1
+#  else
+#    define XXH_SIZE_OPT 0
+#  endif
+#endif
+
+#ifndef XXH_FORCE_ALIGN_CHECK  /* can be defined externally */
+   /* don't check on sizeopt, x86, aarch64, or arm when unaligned access is available */
+#  if XXH_SIZE_OPT >= 1 || \
+      defined(__i386)  || defined(__x86_64__) || defined(__aarch64__) || defined(__ARM_FEATURE_UNALIGNED) \
+   || defined(_M_IX86) || defined(_M_X64)     || defined(_M_ARM64)    || defined(_M_ARM) /* visual */
+#    define XXH_FORCE_ALIGN_CHECK 0
+#  else
+#    define XXH_FORCE_ALIGN_CHECK 1
+#  endif
+#endif
+
+#ifndef XXH_NO_INLINE_HINTS
+#  if XXH_SIZE_OPT >= 1 || defined(__NO_INLINE__)  /* -O0, -fno-inline */
+#    define XXH_NO_INLINE_HINTS 1
+#  else
+#    define XXH_NO_INLINE_HINTS 0
+#  endif
+#endif
+
+#ifndef XXH3_INLINE_SECRET
+#  if (defined(__GNUC__) && !defined(__clang__) && __GNUC__ >= 12) \
+     || !defined(XXH_INLINE_ALL)
+#    define XXH3_INLINE_SECRET 0
+#  else
+#    define XXH3_INLINE_SECRET 1
+#  endif
+#endif
+
+#ifndef XXH32_ENDJMP
+/* generally preferable for performance */
+#  define XXH32_ENDJMP 0
+#endif
+
+/*!
+ * @defgroup impl Implementation
+ * @{
+ */
+
+
+/* *************************************
+*  Includes & Memory related functions
+***************************************/
+#if defined(XXH_NO_STREAM)
+/* nothing */
+#elif defined(XXH_NO_STDLIB)
+
+/* When requesting to disable any mention of stdlib,
+ * the library loses the ability to invoked malloc / free.
+ * In practice, it means that functions like `XXH*_createState()`
+ * will always fail, and return NULL.
+ * This flag is useful in situations where
+ * xxhash.h is integrated into some kernel, embedded or limited environment
+ * without access to dynamic allocation.
+ */
+
+static XXH_CONSTF void* XXH_malloc(size_t s) { (void)s; return NULL; }
+static void XXH_free(void* p) { (void)p; }
+
+#else
+
+/*
+ * Modify the local functions below should you wish to use
+ * different memory routines for malloc() and free()
+ */
+#include <stdlib.h>
+
+/*!
+ * @internal
+ * @brief Modify this function to use a different routine than malloc().
+ */
+static XXH_MALLOCF void* XXH_malloc(size_t s) { return malloc(s); }
+
+/*!
+ * @internal
+ * @brief Modify this function to use a different routine than free().
+ */
+static void XXH_free(void* p) { free(p); }
+
+#endif  /* XXH_NO_STDLIB */
+
+#include <string.h>
+
+/*!
+ * @internal
+ * @brief Modify this function to use a different routine than memcpy().
+ */
+static void* XXH_memcpy(void* dest, const void* src, size_t size)
+{
+    return memcpy(dest,src,size);
+}
+
+#include <limits.h>   /* ULLONG_MAX */
+
+
+/* *************************************
+*  Compiler Specific Options
+***************************************/
+#ifdef _MSC_VER /* Visual Studio warning fix */
+#  pragma warning(disable : 4127) /* disable: C4127: conditional expression is constant */
+#endif
+
+#if XXH_NO_INLINE_HINTS  /* disable inlining hints */
+#  if defined(__GNUC__) || defined(__clang__)
+#    define XXH_FORCE_INLINE static __attribute__((unused))
+#  else
+#    define XXH_FORCE_INLINE static
+#  endif
+#  define XXH_NO_INLINE static
+/* enable inlining hints */
+#elif defined(__GNUC__) || defined(__clang__)
+#  define XXH_FORCE_INLINE static __inline__ __attribute__((always_inline, unused))
+#  define XXH_NO_INLINE static __attribute__((noinline))
+#elif defined(_MSC_VER)  /* Visual Studio */
+#  define XXH_FORCE_INLINE static __forceinline
+#  define XXH_NO_INLINE static __declspec(noinline)
+#elif defined (__cplusplus) \
+  || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L))   /* C99 */
+#  define XXH_FORCE_INLINE static inline
+#  define XXH_NO_INLINE static
+#else
+#  define XXH_FORCE_INLINE static
+#  define XXH_NO_INLINE static
+#endif
+
+#if XXH3_INLINE_SECRET
+#  define XXH3_WITH_SECRET_INLINE XXH_FORCE_INLINE
+#else
+#  define XXH3_WITH_SECRET_INLINE XXH_NO_INLINE
+#endif
+
+
+/* *************************************
+*  Debug
+***************************************/
+/*!
+ * @ingroup tuning
+ * @def XXH_DEBUGLEVEL
+ * @brief Sets the debugging level.
+ *
+ * XXH_DEBUGLEVEL is expected to be defined externally, typically via the
+ * compiler's command line options. The value must be a number.
+ */
+#ifndef XXH_DEBUGLEVEL
+#  ifdef DEBUGLEVEL /* backwards compat */
+#    define XXH_DEBUGLEVEL DEBUGLEVEL
+#  else
+#    define XXH_DEBUGLEVEL 0
+#  endif
+#endif
+
+#if (XXH_DEBUGLEVEL>=1)
+#  include <assert.h>   /* note: can still be disabled with NDEBUG */
+#  define XXH_ASSERT(c)   assert(c)
+#else
+#  if defined(__INTEL_COMPILER)
+#    define XXH_ASSERT(c)   XXH_ASSUME((unsigned char) (c))
+#  else
+#    define XXH_ASSERT(c)   XXH_ASSUME(c)
+#  endif
+#endif
+
+/* note: use after variable declarations */
+#ifndef XXH_STATIC_ASSERT
+#  if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L)    /* C11 */
+#    define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { _Static_assert((c),m); } while(0)
+#  elif defined(__cplusplus) && (__cplusplus >= 201103L)            /* C++11 */
+#    define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { static_assert((c),m); } while(0)
+#  else
+#    define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { struct xxh_sa { char x[(c) ? 1 : -1]; }; } while(0)
+#  endif
+#  define XXH_STATIC_ASSERT(c) XXH_STATIC_ASSERT_WITH_MESSAGE((c),#c)
+#endif
+
+/*!
+ * @internal
+ * @def XXH_COMPILER_GUARD(var)
+ * @brief Used to prevent unwanted optimizations for @p var.
+ *
+ * It uses an empty GCC inline assembly statement with a register constraint
+ * which forces @p var into a general purpose register (eg eax, ebx, ecx
+ * on x86) and marks it as modified.
+ *
+ * This is used in a few places to avoid unwanted autovectorization (e.g.
+ * XXH32_round()). All vectorization we want is explicit via intrinsics,
+ * and _usually_ isn't wanted elsewhere.
+ *
+ * We also use it to prevent unwanted constant folding for AArch64 in
+ * XXH3_initCustomSecret_scalar().
+ */
+#if defined(__GNUC__) || defined(__clang__)
+#  define XXH_COMPILER_GUARD(var) __asm__("" : "+r" (var))
+#else
+#  define XXH_COMPILER_GUARD(var) ((void)0)
+#endif
+
+/* Specifically for NEON vectors which use the "w" constraint, on
+ * Clang. */
+#if defined(__clang__) && defined(__ARM_ARCH) && !defined(__wasm__)
+#  define XXH_COMPILER_GUARD_CLANG_NEON(var) __asm__("" : "+w" (var))
+#else
+#  define XXH_COMPILER_GUARD_CLANG_NEON(var) ((void)0)
+#endif
+
+/* *************************************
+*  Basic Types
+***************************************/
+#if !defined (__VMS) \
+ && (defined (__cplusplus) \
+ || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+# include <stdint.h>
+  typedef uint8_t xxh_u8;
+#else
+  typedef unsigned char xxh_u8;
+#endif
+typedef XXH32_hash_t xxh_u32;
+
+#ifdef XXH_OLD_NAMES
+#  warning "XXH_OLD_NAMES is planned to be removed starting v0.9. If the program depends on it, consider moving away from it by employing newer type names directly"
+#  define BYTE xxh_u8
+#  define U8   xxh_u8
+#  define U32  xxh_u32
+#endif
+
+/* ***   Memory access   *** */
+
+/*!
+ * @internal
+ * @fn xxh_u32 XXH_read32(const void* ptr)
+ * @brief Reads an unaligned 32-bit integer from @p ptr in native endianness.
+ *
+ * Affected by @ref XXH_FORCE_MEMORY_ACCESS.
+ *
+ * @param ptr The pointer to read from.
+ * @return The 32-bit native endian integer from the bytes at @p ptr.
+ */
+
+/*!
+ * @internal
+ * @fn xxh_u32 XXH_readLE32(const void* ptr)
+ * @brief Reads an unaligned 32-bit little endian integer from @p ptr.
+ *
+ * Affected by @ref XXH_FORCE_MEMORY_ACCESS.
+ *
+ * @param ptr The pointer to read from.
+ * @return The 32-bit little endian integer from the bytes at @p ptr.
  */
 
 /*!
@@ -1719,288 +2545,304 @@ typedef XXH32_hash_t xxh_u32;
  * @return The 32-bit little endian integer from the bytes at @p ptr.
  */
 
-  #if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS == 3))
-  /*
-   * Manual byteshift. Best for old compilers which don't inline memcpy.
-   * We actually directly use XXH_readLE32 and XXH_readBE32.
-   */
-  #elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS == 2))
+#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
+/*
+ * Manual byteshift. Best for old compilers which don't inline memcpy.
+ * We actually directly use XXH_readLE32 and XXH_readBE32.
+ */
+#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))
 
 /*
  * Force direct memory access. Only works on CPU which support unaligned memory
  * access in hardware.
  */
-static xxh_u32 XXH_read32(const void *memPtr) {
+static xxh_u32 XXH_read32(const void* memPtr) { return *(const xxh_u32*) memPtr; }
 
-  return *(const xxh_u32 *)memPtr;
+#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))
 
+/*
+ * __attribute__((aligned(1))) is supported by gcc and clang. Originally the
+ * documentation claimed that it only increased the alignment, but actually it
+ * can decrease it on gcc, clang, and icc:
+ * https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69502,
+ * https://gcc.godbolt.org/z/xYez1j67Y.
+ */
+#ifdef XXH_OLD_NAMES
+typedef union { xxh_u32 u32; } __attribute__((packed)) unalign;
+#endif
+static xxh_u32 XXH_read32(const void* ptr)
+{
+    typedef __attribute__((aligned(1))) xxh_u32 xxh_unalign32;
+    return *((const xxh_unalign32*)ptr);
 }
 
-  #elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS == 1))
+#else
 
-    /*
-     * __pack instructions are safer but compiler specific, hence potentially
-     * problematic for some compilers.
-     *
-     * Currently only defined for GCC and ICC.
-     */
-    #ifdef XXH_OLD_NAMES
-typedef union {
+/*
+ * Portable and safe solution. Generally efficient.
+ * see: https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html
+ */
+static xxh_u32 XXH_read32(const void* memPtr)
+{
+    xxh_u32 val;
+    XXH_memcpy(&val, memPtr, sizeof(val));
+    return val;
+}
 
-  xxh_u32 u32;
+#endif   /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
 
-} __attribute__((packed)) unalign;
 
-    #endif
-static xxh_u32 XXH_read32(const void *ptr) {
+/* ***   Endianness   *** */
 
-  typedef union {
+/*!
+ * @ingroup tuning
+ * @def XXH_CPU_LITTLE_ENDIAN
+ * @brief Whether the target is little endian.
+ *
+ * Defined to 1 if the target is little endian, or 0 if it is big endian.
+ * It can be defined externally, for example on the compiler command line.
+ *
+ * If it is not defined,
+ * a runtime check (which is usually constant folded) is used instead.
+ *
+ * @note
+ *   This is not necessarily defined to an integer constant.
+ *
+ * @see XXH_isLittleEndian() for the runtime check.
+ */
+#ifndef XXH_CPU_LITTLE_ENDIAN
+/*
+ * Try to detect endianness automatically, to avoid the nonstandard behavior
+ * in `XXH_isLittleEndian()`
+ */
+#  if defined(_WIN32) /* Windows is always little endian */ \
+     || defined(__LITTLE_ENDIAN__) \
+     || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
+#    define XXH_CPU_LITTLE_ENDIAN 1
+#  elif defined(__BIG_ENDIAN__) \
+     || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
+#    define XXH_CPU_LITTLE_ENDIAN 0
+#  else
+/*!
+ * @internal
+ * @brief Runtime check for @ref XXH_CPU_LITTLE_ENDIAN.
+ *
+ * Most compilers will constant fold this.
+ */
+static int XXH_isLittleEndian(void)
+{
+    /*
+     * Portable and well-defined behavior.
+     * Don't use static: it is detrimental to performance.
+     */
+    const union { xxh_u32 u; xxh_u8 c[4]; } one = { 1 };
+    return one.c[0];
+}
+#   define XXH_CPU_LITTLE_ENDIAN   XXH_isLittleEndian()
+#  endif
+#endif
 
-    xxh_u32 u32;
 
-  } __attribute__((packed)) xxh_unalign;
 
-  return ((const xxh_unalign *)ptr)->u32;
 
-}
+/* ****************************************
+*  Compiler-specific Functions and Macros
+******************************************/
+#define XXH_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
+
+#ifdef __has_builtin
+#  define XXH_HAS_BUILTIN(x) __has_builtin(x)
+#else
+#  define XXH_HAS_BUILTIN(x) 0
+#endif
+
 
-  #else
 
 /*
- * Portable and safe solution. Generally efficient.
- * see: https://stackoverflow.com/a/32095106/646947
+ * C23 and future versions have standard "unreachable()".
+ * Once it has been implemented reliably we can add it as an
+ * additional case:
+ *
+ * ```
+ * #if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= XXH_C23_VN)
+ * #  include <stddef.h>
+ * #  ifdef unreachable
+ * #    define XXH_UNREACHABLE() unreachable()
+ * #  endif
+ * #endif
+ * ```
+ *
+ * Note C++23 also has std::unreachable() which can be detected
+ * as follows:
+ * ```
+ * #if defined(__cpp_lib_unreachable) && (__cpp_lib_unreachable >= 202202L)
+ * #  include <utility>
+ * #  define XXH_UNREACHABLE() std::unreachable()
+ * #endif
+ * ```
+ * NB: `__cpp_lib_unreachable` is defined in the `<version>` header.
+ * We don't use that as including `<utility>` in `extern "C"` blocks
+ * doesn't work on GCC12
  */
-static xxh_u32 XXH_read32(const void *memPtr) {
 
-  xxh_u32 val;
-  memcpy(&val, memPtr, sizeof(val));
-  return val;
+#if XXH_HAS_BUILTIN(__builtin_unreachable)
+#  define XXH_UNREACHABLE() __builtin_unreachable()
 
-}
+#elif defined(_MSC_VER)
+#  define XXH_UNREACHABLE() __assume(0)
 
-  #endif                                  /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
+#else
+#  define XXH_UNREACHABLE()
+#endif
 
-  /* ***   Endianness   *** */
+#if XXH_HAS_BUILTIN(__builtin_assume)
+#  define XXH_ASSUME(c) __builtin_assume(c)
+#else
+#  define XXH_ASSUME(c) if (!(c)) { XXH_UNREACHABLE(); }
+#endif
 
-  /*!
-   * @ingroup tuning
-   * @def XXH_CPU_LITTLE_ENDIAN
-   * @brief Whether the target is little endian.
-   *
-   * Defined to 1 if the target is little endian, or 0 if it is big endian.
-   * It can be defined externally, for example on the compiler command line.
-   *
-   * If it is not defined,
-   * a runtime check (which is usually constant folded) is used instead.
-   *
-   * @note
-   *   This is not necessarily defined to an integer constant.
-   *
-   * @see XXH_isLittleEndian() for the runtime check.
-   */
-  #ifndef XXH_CPU_LITTLE_ENDIAN
-    /*
-     * Try to detect endianness automatically, to avoid the nonstandard behavior
-     * in `XXH_isLittleEndian()`
-     */
-    #if defined(_WIN32) /* Windows is always little endian */ \
-        || defined(__LITTLE_ENDIAN__) ||                      \
-        (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
-      #define XXH_CPU_LITTLE_ENDIAN 1
-    #elif defined(__BIG_ENDIAN__) || \
-        (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
-      #define XXH_CPU_LITTLE_ENDIAN 0
-    #else
 /*!
  * @internal
- * @brief Runtime check for @ref XXH_CPU_LITTLE_ENDIAN.
+ * @def XXH_rotl32(x,r)
+ * @brief 32-bit rotate left.
  *
- * Most compilers will constant fold this.
+ * @param x The 32-bit integer to be rotated.
+ * @param r The number of bits to rotate.
+ * @pre
+ *   @p r > 0 && @p r < 32
+ * @note
+ *   @p x and @p r may be evaluated multiple times.
+ * @return The rotated result.
+ */
+#if !defined(NO_CLANG_BUILTIN) && XXH_HAS_BUILTIN(__builtin_rotateleft32) \
+                               && XXH_HAS_BUILTIN(__builtin_rotateleft64)
+#  define XXH_rotl32 __builtin_rotateleft32
+#  define XXH_rotl64 __builtin_rotateleft64
+/* Note: although _rotl exists for minGW (GCC under windows), performance seems poor */
+#elif defined(_MSC_VER)
+#  define XXH_rotl32(x,r) _rotl(x,r)
+#  define XXH_rotl64(x,r) _rotl64(x,r)
+#else
+#  define XXH_rotl32(x,r) (((x) << (r)) | ((x) >> (32 - (r))))
+#  define XXH_rotl64(x,r) (((x) << (r)) | ((x) >> (64 - (r))))
+#endif
+
+/*!
+ * @internal
+ * @fn xxh_u32 XXH_swap32(xxh_u32 x)
+ * @brief A 32-bit byteswap.
+ *
+ * @param x The 32-bit integer to byteswap.
+ * @return @p x, byteswapped.
  */
-static int XXH_isLittleEndian(void) {
-
-  /*
-   * Portable and well-defined behavior.
-   * Don't use static: it is detrimental to performance.
-   */
-  const union {
-
-    xxh_u32 u;
-    xxh_u8  c[4];
-
-  } one = {1};
-
-  return one.c[0];
-
-}
-
-      #define XXH_CPU_LITTLE_ENDIAN XXH_isLittleEndian()
-    #endif
-  #endif
-
-  /* ****************************************
-   *  Compiler-specific Functions and Macros
-   ******************************************/
-  #define XXH_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
-
-  #ifdef __has_builtin
-    #define XXH_HAS_BUILTIN(x) __has_builtin(x)
-  #else
-    #define XXH_HAS_BUILTIN(x) 0
-  #endif
-
-  /*!
-   * @internal
-   * @def XXH_rotl32(x,r)
-   * @brief 32-bit rotate left.
-   *
-   * @param x The 32-bit integer to be rotated.
-   * @param r The number of bits to rotate.
-   * @pre
-   *   @p r > 0 && @p r < 32
-   * @note
-   *   @p x and @p r may be evaluated multiple times.
-   * @return The rotated result.
-   */
-  #if !defined(NO_CLANG_BUILTIN) && XXH_HAS_BUILTIN(__builtin_rotateleft32) && \
-      XXH_HAS_BUILTIN(__builtin_rotateleft64)
-    #define XXH_rotl32 __builtin_rotateleft32
-    #define XXH_rotl64 __builtin_rotateleft64
-  /* Note: although _rotl exists for minGW (GCC under windows), performance
-   * seems poor */
-  #elif defined(_MSC_VER)
-    #define XXH_rotl32(x, r) _rotl(x, r)
-    #define XXH_rotl64(x, r) _rotl64(x, r)
-  #else
-    #define XXH_rotl32(x, r) (((x) << (r)) | ((x) >> (32 - (r))))
-    #define XXH_rotl64(x, r) (((x) << (r)) | ((x) >> (64 - (r))))
-  #endif
-
-  /*!
-   * @internal
-   * @fn xxh_u32 XXH_swap32(xxh_u32 x)
-   * @brief A 32-bit byteswap.
-   *
-   * @param x The 32-bit integer to byteswap.
-   * @return @p x, byteswapped.
-   */
-  #if defined(_MSC_VER)                                    /* Visual Studio */
-    #define XXH_swap32 _byteswap_ulong
-  #elif XXH_GCC_VERSION >= 403
-    #define XXH_swap32 __builtin_bswap32
-  #else
-static xxh_u32 XXH_swap32(xxh_u32 x) {
-
-  return ((x << 24) & 0xff000000) | ((x << 8) & 0x00ff0000) |
-         ((x >> 8) & 0x0000ff00) | ((x >> 24) & 0x000000ff);
-
-}
-
-  #endif
+#if defined(_MSC_VER)     /* Visual Studio */
+#  define XXH_swap32 _byteswap_ulong
+#elif XXH_GCC_VERSION >= 403
+#  define XXH_swap32 __builtin_bswap32
+#else
+static xxh_u32 XXH_swap32 (xxh_u32 x)
+{
+    return  ((x << 24) & 0xff000000 ) |
+            ((x <<  8) & 0x00ff0000 ) |
+            ((x >>  8) & 0x0000ff00 ) |
+            ((x >> 24) & 0x000000ff );
+}
+#endif
+
 
 /* ***************************
- *  Memory reads
- *****************************/
+*  Memory reads
+*****************************/
 
 /*!
  * @internal
  * @brief Enum to indicate whether a pointer is aligned.
  */
 typedef enum {
-
-  XXH_aligned,                                                 /*!< Aligned */
-  XXH_unaligned                                     /*!< Possibly unaligned */
-
+    XXH_aligned,  /*!< Aligned */
+    XXH_unaligned /*!< Possibly unaligned */
 } XXH_alignment;
 
-  /*
-   * XXH_FORCE_MEMORY_ACCESS==3 is an endian-independent byteshift load.
-   *
-   * This is ideal for older compilers which don't inline memcpy.
-   */
-  #if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS == 3))
-
-XXH_FORCE_INLINE xxh_u32 XXH_readLE32(const void *memPtr) {
-
-  const xxh_u8 *bytePtr = (const xxh_u8 *)memPtr;
-  return bytePtr[0] | ((xxh_u32)bytePtr[1] << 8) | ((xxh_u32)bytePtr[2] << 16) |
-         ((xxh_u32)bytePtr[3] << 24);
+/*
+ * XXH_FORCE_MEMORY_ACCESS==3 is an endian-independent byteshift load.
+ *
+ * This is ideal for older compilers which don't inline memcpy.
+ */
+#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
 
+XXH_FORCE_INLINE xxh_u32 XXH_readLE32(const void* memPtr)
+{
+    const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
+    return bytePtr[0]
+         | ((xxh_u32)bytePtr[1] << 8)
+         | ((xxh_u32)bytePtr[2] << 16)
+         | ((xxh_u32)bytePtr[3] << 24);
 }
 
-XXH_FORCE_INLINE xxh_u32 XXH_readBE32(const void *memPtr) {
-
-  const xxh_u8 *bytePtr = (const xxh_u8 *)memPtr;
-  return bytePtr[3] | ((xxh_u32)bytePtr[2] << 8) | ((xxh_u32)bytePtr[1] << 16) |
-         ((xxh_u32)bytePtr[0] << 24);
-
+XXH_FORCE_INLINE xxh_u32 XXH_readBE32(const void* memPtr)
+{
+    const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
+    return bytePtr[3]
+         | ((xxh_u32)bytePtr[2] << 8)
+         | ((xxh_u32)bytePtr[1] << 16)
+         | ((xxh_u32)bytePtr[0] << 24);
 }
 
-  #else
-XXH_FORCE_INLINE xxh_u32 XXH_readLE32(const void *ptr) {
-
-  return XXH_CPU_LITTLE_ENDIAN ? XXH_read32(ptr) : XXH_swap32(XXH_read32(ptr));
-
+#else
+XXH_FORCE_INLINE xxh_u32 XXH_readLE32(const void* ptr)
+{
+    return XXH_CPU_LITTLE_ENDIAN ? XXH_read32(ptr) : XXH_swap32(XXH_read32(ptr));
 }
 
-static xxh_u32 XXH_readBE32(const void *ptr) {
-
-  return XXH_CPU_LITTLE_ENDIAN ? XXH_swap32(XXH_read32(ptr)) : XXH_read32(ptr);
-
+static xxh_u32 XXH_readBE32(const void* ptr)
+{
+    return XXH_CPU_LITTLE_ENDIAN ? XXH_swap32(XXH_read32(ptr)) : XXH_read32(ptr);
 }
+#endif
 
-  #endif
-
-XXH_FORCE_INLINE xxh_u32 XXH_readLE32_align(const void   *ptr,
-                                            XXH_alignment align) {
-
-  if (align == XXH_unaligned) {
-
-    return XXH_readLE32(ptr);
-
-  } else {
-
-    return XXH_CPU_LITTLE_ENDIAN ? *(const xxh_u32 *)ptr
-                                 : XXH_swap32(*(const xxh_u32 *)ptr);
-
-  }
-
+XXH_FORCE_INLINE xxh_u32
+XXH_readLE32_align(const void* ptr, XXH_alignment align)
+{
+    if (align==XXH_unaligned) {
+        return XXH_readLE32(ptr);
+    } else {
+        return XXH_CPU_LITTLE_ENDIAN ? *(const xxh_u32*)ptr : XXH_swap32(*(const xxh_u32*)ptr);
+    }
 }
 
+
 /* *************************************
- *  Misc
- ***************************************/
+*  Misc
+***************************************/
 /*! @ingroup public */
-XXH_PUBLIC_API unsigned XXH_versionNumber(void) {
+XXH_PUBLIC_API unsigned XXH_versionNumber (void) { return XXH_VERSION_NUMBER; }
 
-  return XXH_VERSION_NUMBER;
-
-}
 
 /* *******************************************************************
- *  32-bit hash functions
- *********************************************************************/
+*  32-bit hash functions
+*********************************************************************/
 /*!
  * @}
- * @defgroup xxh32_impl XXH32 implementation
+ * @defgroup XXH32_impl XXH32 implementation
  * @ingroup impl
+ *
+ * Details on the XXH32 implementation.
  * @{
-
  */
-/* #define instead of static const, to be used as initializers */
-  #define XXH_PRIME32_1 0x9E3779B1U /*!< 0b10011110001101110111100110110001 */
-  #define XXH_PRIME32_2 0x85EBCA77U /*!< 0b10000101111010111100101001110111 */
-  #define XXH_PRIME32_3 0xC2B2AE3DU /*!< 0b11000010101100101010111000111101 */
-  #define XXH_PRIME32_4 0x27D4EB2FU /*!< 0b00100111110101001110101100101111 */
-  #define XXH_PRIME32_5 0x165667B1U /*!< 0b00010110010101100110011110110001 */
-
-  #ifdef XXH_OLD_NAMES
-    #define PRIME32_1 XXH_PRIME32_1
-    #define PRIME32_2 XXH_PRIME32_2
-    #define PRIME32_3 XXH_PRIME32_3
-    #define PRIME32_4 XXH_PRIME32_4
-    #define PRIME32_5 XXH_PRIME32_5
-  #endif
+ /* #define instead of static const, to be used as initializers */
+#define XXH_PRIME32_1  0x9E3779B1U  /*!< 0b10011110001101110111100110110001 */
+#define XXH_PRIME32_2  0x85EBCA77U  /*!< 0b10000101111010111100101001110111 */
+#define XXH_PRIME32_3  0xC2B2AE3DU  /*!< 0b11000010101100101010111000111101 */
+#define XXH_PRIME32_4  0x27D4EB2FU  /*!< 0b00100111110101001110101100101111 */
+#define XXH_PRIME32_5  0x165667B1U  /*!< 0b00010110010101100110011110110001 */
+
+#ifdef XXH_OLD_NAMES
+#  define PRIME32_1 XXH_PRIME32_1
+#  define PRIME32_2 XXH_PRIME32_2
+#  define PRIME32_3 XXH_PRIME32_3
+#  define PRIME32_4 XXH_PRIME32_4
+#  define PRIME32_5 XXH_PRIME32_5
+#endif
 
 /*!
  * @internal
@@ -2013,50 +2855,51 @@ XXH_PUBLIC_API unsigned XXH_versionNumber(void) {
  * @param input The stripe of input to mix.
  * @return The mixed accumulator lane.
  */
-static xxh_u32 XXH32_round(xxh_u32 acc, xxh_u32 input) {
-
-  acc += input * XXH_PRIME32_2;
-  acc = XXH_rotl32(acc, 13);
-  acc *= XXH_PRIME32_1;
-  #if (defined(__SSE4_1__) || defined(__aarch64__)) && \
-      !defined(XXH_ENABLE_AUTOVECTORIZE)
-  /*
-   * UGLY HACK:
-   * A compiler fence is the only thing that prevents GCC and Clang from
-   * autovectorizing the XXH32 loop (pragmas and attributes don't work for some
-   * reason) without globally disabling SSE4.1.
-   *
-   * The reason we want to avoid vectorization is because despite working on
-   * 4 integers at a time, there are multiple factors slowing XXH32 down on
-   * SSE4:
-   * - There's a ridiculous amount of lag from pmulld (10 cycles of latency on
-   *   newer chips!) making it slightly slower to multiply four integers at
-   *   once compared to four integers independently. Even when pmulld was
-   *   fastest, Sandy/Ivy Bridge, it is still not worth it to go into SSE
-   *   just to multiply unless doing a long operation.
-   *
-   * - Four instructions are required to rotate,
-   *      movqda tmp,  v // not required with VEX encoding
-   *      pslld  tmp, 13 // tmp <<= 13
-   *      psrld  v,   19 // x >>= 19
-   *      por    v,  tmp // x |= tmp
-   *   compared to one for scalar:
-   *      roll   v, 13    // reliably fast across the board
-   *      shldl  v, v, 13 // Sandy Bridge and later prefer this for some reason
-   *
-   * - Instruction level parallelism is actually more beneficial here because
-   *   the SIMD actually serializes this operation: While v1 is rotating, v2
-   *   can load data, while v3 can multiply. SSE forces them to operate
-   *   together.
-   *
-   * This is also enabled on AArch64, as Clang autovectorizes it incorrectly
-   * and it is pointless writing a NEON implementation that is basically the
-   * same speed as scalar for XXH32.
-   */
-  XXH_COMPILER_GUARD(acc);
-  #endif
-  return acc;
-
+static xxh_u32 XXH32_round(xxh_u32 acc, xxh_u32 input)
+{
+    acc += input * XXH_PRIME32_2;
+    acc  = XXH_rotl32(acc, 13);
+    acc *= XXH_PRIME32_1;
+#if (defined(__SSE4_1__) || defined(__aarch64__) || defined(__wasm_simd128__)) && !defined(XXH_ENABLE_AUTOVECTORIZE)
+    /*
+     * UGLY HACK:
+     * A compiler fence is the only thing that prevents GCC and Clang from
+     * autovectorizing the XXH32 loop (pragmas and attributes don't work for some
+     * reason) without globally disabling SSE4.1.
+     *
+     * The reason we want to avoid vectorization is because despite working on
+     * 4 integers at a time, there are multiple factors slowing XXH32 down on
+     * SSE4:
+     * - There's a ridiculous amount of lag from pmulld (10 cycles of latency on
+     *   newer chips!) making it slightly slower to multiply four integers at
+     *   once compared to four integers independently. Even when pmulld was
+     *   fastest, Sandy/Ivy Bridge, it is still not worth it to go into SSE
+     *   just to multiply unless doing a long operation.
+     *
+     * - Four instructions are required to rotate,
+     *      movqda tmp,  v // not required with VEX encoding
+     *      pslld  tmp, 13 // tmp <<= 13
+     *      psrld  v,   19 // x >>= 19
+     *      por    v,  tmp // x |= tmp
+     *   compared to one for scalar:
+     *      roll   v, 13    // reliably fast across the board
+     *      shldl  v, v, 13 // Sandy Bridge and later prefer this for some reason
+     *
+     * - Instruction level parallelism is actually more beneficial here because
+     *   the SIMD actually serializes this operation: While v1 is rotating, v2
+     *   can load data, while v3 can multiply. SSE forces them to operate
+     *   together.
+     *
+     * This is also enabled on AArch64, as Clang is *very aggressive* in vectorizing
+     * the loop. NEON is only faster on the A53, and with the newer cores, it is less
+     * than half the speed.
+     *
+     * Additionally, this is used on WASM SIMD128 because it JITs to the same
+     * SIMD instructions and has the same issue.
+     */
+    XXH_COMPILER_GUARD(acc);
+#endif
+    return acc;
 }
 
 /*!
@@ -2066,38 +2909,20 @@ static xxh_u32 XXH32_round(xxh_u32 acc, xxh_u32 input) {
  * The final mix ensures that all input bits have a chance to impact any bit in
  * the output digest, resulting in an unbiased distribution.
  *
- * @param h32 The hash to avalanche.
+ * @param hash The hash to avalanche.
  * @return The avalanched hash.
  */
-static xxh_u32 XXH32_avalanche(xxh_u32 h32) {
-
-  h32 ^= h32 >> 15;
-  h32 *= XXH_PRIME32_2;
-  h32 ^= h32 >> 13;
-  h32 *= XXH_PRIME32_3;
-  h32 ^= h32 >> 16;
-  return (h32);
-
+static xxh_u32 XXH32_avalanche(xxh_u32 hash)
+{
+    hash ^= hash >> 15;
+    hash *= XXH_PRIME32_2;
+    hash ^= hash >> 13;
+    hash *= XXH_PRIME32_3;
+    hash ^= hash >> 16;
+    return hash;
 }
 
-  #define XXH_get32bits(p) XXH_readLE32_align(p, align)
-
-  #define XXH_PROCESS1                           \
-    do {                                         \
-                                                 \
-      h32 += (*ptr++) * XXH_PRIME32_5;           \
-      h32 = XXH_rotl32(h32, 11) * XXH_PRIME32_1; \
-                                                 \
-    } while (0)
-
-  #define XXH_PROCESS4                           \
-    do {                                         \
-                                                 \
-      h32 += XXH_get32bits(ptr) * XXH_PRIME32_3; \
-      ptr += 4;                                  \
-      h32 = XXH_rotl32(h32, 17) * XXH_PRIME32_4; \
-                                                 \
-    } while (0)
+#define XXH_get32bits(p) XXH_readLE32_align(p, align)
 
 /*!
  * @internal
@@ -2107,1515 +2932,1366 @@ static xxh_u32 XXH32_avalanche(xxh_u32 h32) {
  * This final stage will digest them to ensure that all input bytes are present
  * in the final mix.
  *
- * @param h32 The hash to finalize.
+ * @param hash The hash to finalize.
  * @param ptr The pointer to the remaining input.
  * @param len The remaining length, modulo 16.
  * @param align Whether @p ptr is aligned.
  * @return The finalized hash.
+ * @see XXH64_finalize().
  */
-static xxh_u32 XXH32_finalize(xxh_u32 h32, const xxh_u8 *ptr, size_t len,
-                              XXH_alignment align) {
-
-  /* Compact rerolled version */
-  if (XXH_REROLL) {
-
-    len &= 15;
-    while (len >= 4) {
-
-      XXH_PROCESS4;
-      len -= 4;
-
-    }
-
-    while (len > 0) {
-
-      XXH_PROCESS1;
-      --len;
-
-    }
-
-    return XXH32_avalanche(h32);
-
-  } else {
-
-    switch (len & 15) /* or switch(bEnd - p) */ {
-
-      case 12:
-        XXH_PROCESS4;
-        XXH_FALLTHROUGH;
-      case 8:
-        XXH_PROCESS4;
-        XXH_FALLTHROUGH;
-      case 4:
-        XXH_PROCESS4;
-        return XXH32_avalanche(h32);
-
-      case 13:
-        XXH_PROCESS4;
-        XXH_FALLTHROUGH;
-      case 9:
-        XXH_PROCESS4;
-        XXH_FALLTHROUGH;
-      case 5:
-        XXH_PROCESS4;
-        XXH_PROCESS1;
-        return XXH32_avalanche(h32);
-
-      case 14:
-        XXH_PROCESS4;
-        XXH_FALLTHROUGH;
-      case 10:
-        XXH_PROCESS4;
-        XXH_FALLTHROUGH;
-      case 6:
-        XXH_PROCESS4;
-        XXH_PROCESS1;
-        XXH_PROCESS1;
-        return XXH32_avalanche(h32);
-
-      case 15:
-        XXH_PROCESS4;
-        XXH_FALLTHROUGH;
-      case 11:
-        XXH_PROCESS4;
-        XXH_FALLTHROUGH;
-      case 7:
-        XXH_PROCESS4;
-        XXH_FALLTHROUGH;
-      case 3:
-        XXH_PROCESS1;
-        XXH_FALLTHROUGH;
-      case 2:
-        XXH_PROCESS1;
-        XXH_FALLTHROUGH;
-      case 1:
-        XXH_PROCESS1;
-        XXH_FALLTHROUGH;
-      case 0:
-        return XXH32_avalanche(h32);
-
+static XXH_PUREF xxh_u32
+XXH32_finalize(xxh_u32 hash, const xxh_u8* ptr, size_t len, XXH_alignment align)
+{
+#define XXH_PROCESS1 do {                             \
+    hash += (*ptr++) * XXH_PRIME32_5;                 \
+    hash = XXH_rotl32(hash, 11) * XXH_PRIME32_1;      \
+} while (0)
+
+#define XXH_PROCESS4 do {                             \
+    hash += XXH_get32bits(ptr) * XXH_PRIME32_3;       \
+    ptr += 4;                                         \
+    hash  = XXH_rotl32(hash, 17) * XXH_PRIME32_4;     \
+} while (0)
+
+    if (ptr==NULL) XXH_ASSERT(len == 0);
+
+    /* Compact rerolled version; generally faster */
+    if (!XXH32_ENDJMP) {
+        len &= 15;
+        while (len >= 4) {
+            XXH_PROCESS4;
+            len -= 4;
+        }
+        while (len > 0) {
+            XXH_PROCESS1;
+            --len;
+        }
+        return XXH32_avalanche(hash);
+    } else {
+         switch(len&15) /* or switch(bEnd - p) */ {
+           case 12:      XXH_PROCESS4;
+                         XXH_FALLTHROUGH;  /* fallthrough */
+           case 8:       XXH_PROCESS4;
+                         XXH_FALLTHROUGH;  /* fallthrough */
+           case 4:       XXH_PROCESS4;
+                         return XXH32_avalanche(hash);
+
+           case 13:      XXH_PROCESS4;
+                         XXH_FALLTHROUGH;  /* fallthrough */
+           case 9:       XXH_PROCESS4;
+                         XXH_FALLTHROUGH;  /* fallthrough */
+           case 5:       XXH_PROCESS4;
+                         XXH_PROCESS1;
+                         return XXH32_avalanche(hash);
+
+           case 14:      XXH_PROCESS4;
+                         XXH_FALLTHROUGH;  /* fallthrough */
+           case 10:      XXH_PROCESS4;
+                         XXH_FALLTHROUGH;  /* fallthrough */
+           case 6:       XXH_PROCESS4;
+                         XXH_PROCESS1;
+                         XXH_PROCESS1;
+                         return XXH32_avalanche(hash);
+
+           case 15:      XXH_PROCESS4;
+                         XXH_FALLTHROUGH;  /* fallthrough */
+           case 11:      XXH_PROCESS4;
+                         XXH_FALLTHROUGH;  /* fallthrough */
+           case 7:       XXH_PROCESS4;
+                         XXH_FALLTHROUGH;  /* fallthrough */
+           case 3:       XXH_PROCESS1;
+                         XXH_FALLTHROUGH;  /* fallthrough */
+           case 2:       XXH_PROCESS1;
+                         XXH_FALLTHROUGH;  /* fallthrough */
+           case 1:       XXH_PROCESS1;
+                         XXH_FALLTHROUGH;  /* fallthrough */
+           case 0:       return XXH32_avalanche(hash);
+        }
+        XXH_ASSERT(0);
+        return hash;   /* reaching this point is deemed impossible */
     }
-
-    XXH_ASSERT(0);
-    return h32;                 /* reaching this point is deemed impossible */
-
-  }
-
 }
 
-  #ifdef XXH_OLD_NAMES
-    #define PROCESS1 XXH_PROCESS1
-    #define PROCESS4 XXH_PROCESS4
-  #else
-    #undef XXH_PROCESS1
-    #undef XXH_PROCESS4
-  #endif
+#ifdef XXH_OLD_NAMES
+#  define PROCESS1 XXH_PROCESS1
+#  define PROCESS4 XXH_PROCESS4
+#else
+#  undef XXH_PROCESS1
+#  undef XXH_PROCESS4
+#endif
 
 /*!
  * @internal
  * @brief The implementation for @ref XXH32().
  *
- * @param input, len, seed Directly passed from @ref XXH32().
+ * @param input , len , seed Directly passed from @ref XXH32().
  * @param align Whether @p input is aligned.
  * @return The calculated hash.
  */
-XXH_FORCE_INLINE xxh_u32 XXH32_endian_align(const xxh_u8 *input, size_t len,
-                                            xxh_u32 seed, XXH_alignment align) {
+XXH_FORCE_INLINE XXH_PUREF xxh_u32
+XXH32_endian_align(const xxh_u8* input, size_t len, xxh_u32 seed, XXH_alignment align)
+{
+    xxh_u32 h32;
+
+    if (input==NULL) XXH_ASSERT(len == 0);
+
+    if (len>=16) {
+        const xxh_u8* const bEnd = input + len;
+        const xxh_u8* const limit = bEnd - 15;
+        xxh_u32 v1 = seed + XXH_PRIME32_1 + XXH_PRIME32_2;
+        xxh_u32 v2 = seed + XXH_PRIME32_2;
+        xxh_u32 v3 = seed + 0;
+        xxh_u32 v4 = seed - XXH_PRIME32_1;
+
+        do {
+            v1 = XXH32_round(v1, XXH_get32bits(input)); input += 4;
+            v2 = XXH32_round(v2, XXH_get32bits(input)); input += 4;
+            v3 = XXH32_round(v3, XXH_get32bits(input)); input += 4;
+            v4 = XXH32_round(v4, XXH_get32bits(input)); input += 4;
+        } while (input < limit);
+
+        h32 = XXH_rotl32(v1, 1)  + XXH_rotl32(v2, 7)
+            + XXH_rotl32(v3, 12) + XXH_rotl32(v4, 18);
+    } else {
+        h32  = seed + XXH_PRIME32_5;
+    }
 
-  const xxh_u8 *bEnd = input ? input + len : NULL;
-  xxh_u32       h32;
+    h32 += (xxh_u32)len;
 
-  #if defined(XXH_ACCEPT_NULL_INPUT_POINTER) && \
-      (XXH_ACCEPT_NULL_INPUT_POINTER >= 1)
-  if (input == NULL) {
+    return XXH32_finalize(h32, input, len&15, align);
+}
 
-    len = 0;
-    bEnd = input = (const xxh_u8 *)(size_t)16;
+/*! @ingroup XXH32_family */
+XXH_PUBLIC_API XXH32_hash_t XXH32 (const void* input, size_t len, XXH32_hash_t seed)
+{
+#if !defined(XXH_NO_STREAM) && XXH_SIZE_OPT >= 2
+    /* Simple version, good for code maintenance, but unfortunately slow for small inputs */
+    XXH32_state_t state;
+    XXH32_reset(&state, seed);
+    XXH32_update(&state, (const xxh_u8*)input, len);
+    return XXH32_digest(&state);
+#else
+    if (XXH_FORCE_ALIGN_CHECK) {
+        if ((((size_t)input) & 3) == 0) {   /* Input is 4-bytes aligned, leverage the speed benefit */
+            return XXH32_endian_align((const xxh_u8*)input, len, seed, XXH_aligned);
+    }   }
 
-  }
+    return XXH32_endian_align((const xxh_u8*)input, len, seed, XXH_unaligned);
+#endif
+}
 
-  #endif
 
-  if (len >= 16) {
-
-    const xxh_u8 *const limit = bEnd - 15;
-    xxh_u32             v1 = seed + XXH_PRIME32_1 + XXH_PRIME32_2;
-    xxh_u32             v2 = seed + XXH_PRIME32_2;
-    xxh_u32             v3 = seed + 0;
-    xxh_u32             v4 = seed - XXH_PRIME32_1;
-
-    do {
-
-      v1 = XXH32_round(v1, XXH_get32bits(input));
-      input += 4;
-      v2 = XXH32_round(v2, XXH_get32bits(input));
-      input += 4;
-      v3 = XXH32_round(v3, XXH_get32bits(input));
-      input += 4;
-      v4 = XXH32_round(v4, XXH_get32bits(input));
-      input += 4;
-
-    } while (input < limit);
-
-    h32 = XXH_rotl32(v1, 1) + XXH_rotl32(v2, 7) + XXH_rotl32(v3, 12) +
-          XXH_rotl32(v4, 18);
-
-  } else {
-
-    h32 = seed + XXH_PRIME32_5;
-
-  }
-
-  h32 += (xxh_u32)len;
-
-  return XXH32_finalize(h32, input, len & 15, align);
-
-}
-
-/*! @ingroup xxh32_family */
-XXH_PUBLIC_API XXH32_hash_t XXH32(const void *input, size_t len,
-                                  XXH32_hash_t seed) {
-
-  #if 0
-    /* Simple version, good for code maintenance, but unfortunately slow for small inputs */
-    XXH32_state_t state;
-    XXH32_reset(&state, seed);
-    XXH32_update(&state, (const xxh_u8*)input, len);
-    return XXH32_digest(&state);
-  #else
-  if (XXH_FORCE_ALIGN_CHECK) {
-
-    if ((((size_t)input) & 3) ==
-        0) {        /* Input is 4-bytes aligned, leverage the speed benefit */
-      return XXH32_endian_align((const xxh_u8 *)input, len, seed, XXH_aligned);
-
-    }
-
-  }
-
-  return XXH32_endian_align((const xxh_u8 *)input, len, seed, XXH_unaligned);
-  #endif
-
-}
 
 /*******   Hash streaming   *******/
-/*!
- * @ingroup xxh32_family
- */
-XXH_PUBLIC_API XXH32_state_t *XXH32_createState(void) {
-
-  return (XXH32_state_t *)XXH_malloc(sizeof(XXH32_state_t));
-
-}
-
-/*! @ingroup xxh32_family */
-XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t *statePtr) {
-
-  XXH_free(statePtr);
-  return XXH_OK;
-
-}
-
-/*! @ingroup xxh32_family */
-XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t       *dstState,
-                                    const XXH32_state_t *srcState) {
-
-  memcpy(dstState, srcState, sizeof(*dstState));
-
+#ifndef XXH_NO_STREAM
+/*! @ingroup XXH32_family */
+XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void)
+{
+    return (XXH32_state_t*)XXH_malloc(sizeof(XXH32_state_t));
+}
+/*! @ingroup XXH32_family */
+XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr)
+{
+    XXH_free(statePtr);
+    return XXH_OK;
 }
 
-/*! @ingroup xxh32_family */
-XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t *statePtr,
-                                         XXH32_hash_t   seed) {
-
-  XXH32_state_t state; /* using a local state to memcpy() in order to avoid
-                          strict-aliasing warnings */
-  memset(&state, 0, sizeof(state));
-  state.v1 = seed + XXH_PRIME32_1 + XXH_PRIME32_2;
-  state.v2 = seed + XXH_PRIME32_2;
-  state.v3 = seed + 0;
-  state.v4 = seed - XXH_PRIME32_1;
-  /* do not write into reserved, planned to be removed in a future version */
-  memcpy(statePtr, &state, sizeof(state) - sizeof(state.reserved));
-  return XXH_OK;
-
+/*! @ingroup XXH32_family */
+XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dstState, const XXH32_state_t* srcState)
+{
+    XXH_memcpy(dstState, srcState, sizeof(*dstState));
 }
 
-/*! @ingroup xxh32_family */
-XXH_PUBLIC_API XXH_errorcode XXH32_update(XXH32_state_t *state,
-                                          const void *input, size_t len) {
-
-  if (input == NULL)
-  #if defined(XXH_ACCEPT_NULL_INPUT_POINTER) && \
-      (XXH_ACCEPT_NULL_INPUT_POINTER >= 1)
+/*! @ingroup XXH32_family */
+XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t* statePtr, XXH32_hash_t seed)
+{
+    XXH_ASSERT(statePtr != NULL);
+    memset(statePtr, 0, sizeof(*statePtr));
+    statePtr->v[0] = seed + XXH_PRIME32_1 + XXH_PRIME32_2;
+    statePtr->v[1] = seed + XXH_PRIME32_2;
+    statePtr->v[2] = seed + 0;
+    statePtr->v[3] = seed - XXH_PRIME32_1;
     return XXH_OK;
-  #else
-    return XXH_ERROR;
-  #endif
-
-  {
-
-    const xxh_u8       *p = (const xxh_u8 *)input;
-    const xxh_u8 *const bEnd = p + len;
-
-    state->total_len_32 += (XXH32_hash_t)len;
-    state->large_len |=
-        (XXH32_hash_t)((len >= 16) | (state->total_len_32 >= 16));
-
-    if (state->memsize + len < 16) {                  /* fill in tmp buffer */
-      XXH_memcpy((xxh_u8 *)(state->mem32) + state->memsize, input, len);
-      state->memsize += (XXH32_hash_t)len;
-      return XXH_OK;
-
-    }
-
-    if (state->memsize) {            /* some data left from previous update */
-      XXH_memcpy((xxh_u8 *)(state->mem32) + state->memsize, input,
-                 16 - state->memsize);
-      {
-
-        const xxh_u32 *p32 = state->mem32;
-        state->v1 = XXH32_round(state->v1, XXH_readLE32(p32));
-        p32++;
-        state->v2 = XXH32_round(state->v2, XXH_readLE32(p32));
-        p32++;
-        state->v3 = XXH32_round(state->v3, XXH_readLE32(p32));
-        p32++;
-        state->v4 = XXH32_round(state->v4, XXH_readLE32(p32));
-
-      }
+}
 
-      p += 16 - state->memsize;
-      state->memsize = 0;
 
+/*! @ingroup XXH32_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH32_update(XXH32_state_t* state, const void* input, size_t len)
+{
+    if (input==NULL) {
+        XXH_ASSERT(len == 0);
+        return XXH_OK;
     }
 
-    if (p <= bEnd - 16) {
+    {   const xxh_u8* p = (const xxh_u8*)input;
+        const xxh_u8* const bEnd = p + len;
 
-      const xxh_u8 *const limit = bEnd - 16;
-      xxh_u32             v1 = state->v1;
-      xxh_u32             v2 = state->v2;
-      xxh_u32             v3 = state->v3;
-      xxh_u32             v4 = state->v4;
+        state->total_len_32 += (XXH32_hash_t)len;
+        state->large_len |= (XXH32_hash_t)((len>=16) | (state->total_len_32>=16));
 
-      do {
-
-        v1 = XXH32_round(v1, XXH_readLE32(p));
-        p += 4;
-        v2 = XXH32_round(v2, XXH_readLE32(p));
-        p += 4;
-        v3 = XXH32_round(v3, XXH_readLE32(p));
-        p += 4;
-        v4 = XXH32_round(v4, XXH_readLE32(p));
-        p += 4;
-
-      } while (p <= limit);
+        if (state->memsize + len < 16)  {   /* fill in tmp buffer */
+            XXH_memcpy((xxh_u8*)(state->mem32) + state->memsize, input, len);
+            state->memsize += (XXH32_hash_t)len;
+            return XXH_OK;
+        }
 
-      state->v1 = v1;
-      state->v2 = v2;
-      state->v3 = v3;
-      state->v4 = v4;
+        if (state->memsize) {   /* some data left from previous update */
+            XXH_memcpy((xxh_u8*)(state->mem32) + state->memsize, input, 16-state->memsize);
+            {   const xxh_u32* p32 = state->mem32;
+                state->v[0] = XXH32_round(state->v[0], XXH_readLE32(p32)); p32++;
+                state->v[1] = XXH32_round(state->v[1], XXH_readLE32(p32)); p32++;
+                state->v[2] = XXH32_round(state->v[2], XXH_readLE32(p32)); p32++;
+                state->v[3] = XXH32_round(state->v[3], XXH_readLE32(p32));
+            }
+            p += 16-state->memsize;
+            state->memsize = 0;
+        }
 
-    }
+        if (p <= bEnd-16) {
+            const xxh_u8* const limit = bEnd - 16;
 
-    if (p < bEnd) {
+            do {
+                state->v[0] = XXH32_round(state->v[0], XXH_readLE32(p)); p+=4;
+                state->v[1] = XXH32_round(state->v[1], XXH_readLE32(p)); p+=4;
+                state->v[2] = XXH32_round(state->v[2], XXH_readLE32(p)); p+=4;
+                state->v[3] = XXH32_round(state->v[3], XXH_readLE32(p)); p+=4;
+            } while (p<=limit);
 
-      XXH_memcpy(state->mem32, p, (size_t)(bEnd - p));
-      state->memsize = (unsigned)(bEnd - p);
+        }
 
+        if (p < bEnd) {
+            XXH_memcpy(state->mem32, p, (size_t)(bEnd-p));
+            state->memsize = (unsigned)(bEnd-p);
+        }
     }
 
-  }
-
-  return XXH_OK;
-
+    return XXH_OK;
 }
 
-/*! @ingroup xxh32_family */
-XXH_PUBLIC_API XXH32_hash_t XXH32_digest(const XXH32_state_t *state) {
-
-  xxh_u32 h32;
-
-  if (state->large_len) {
 
-    h32 = XXH_rotl32(state->v1, 1) + XXH_rotl32(state->v2, 7) +
-          XXH_rotl32(state->v3, 12) + XXH_rotl32(state->v4, 18);
+/*! @ingroup XXH32_family */
+XXH_PUBLIC_API XXH32_hash_t XXH32_digest(const XXH32_state_t* state)
+{
+    xxh_u32 h32;
 
-  } else {
-
-    h32 = state->v3 /* == seed */ + XXH_PRIME32_5;
-
-  }
-
-  h32 += state->total_len_32;
+    if (state->large_len) {
+        h32 = XXH_rotl32(state->v[0], 1)
+            + XXH_rotl32(state->v[1], 7)
+            + XXH_rotl32(state->v[2], 12)
+            + XXH_rotl32(state->v[3], 18);
+    } else {
+        h32 = state->v[2] /* == seed */ + XXH_PRIME32_5;
+    }
 
-  return XXH32_finalize(h32, (const xxh_u8 *)state->mem32, state->memsize,
-                        XXH_aligned);
+    h32 += state->total_len_32;
 
+    return XXH32_finalize(h32, (const xxh_u8*)state->mem32, state->memsize, XXH_aligned);
 }
+#endif /* !XXH_NO_STREAM */
 
 /*******   Canonical representation   *******/
 
-/*!
- * @ingroup xxh32_family
- * The default return values from XXH functions are unsigned 32 and 64 bit
- * integers.
- *
- * The canonical representation uses big endian convention, the same convention
- * as human-readable numbers (large digits first).
- *
- * This way, hash values can be written into a file or buffer, remaining
- * comparable across different systems.
- *
- * The following functions allow transformation of hash values to and from their
- * canonical format.
- */
-XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t *dst,
-                                            XXH32_hash_t       hash) {
-
-  XXH_STATIC_ASSERT(sizeof(XXH32_canonical_t) == sizeof(XXH32_hash_t));
-  if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap32(hash);
-  memcpy(dst, &hash, sizeof(*dst));
-
+/*! @ingroup XXH32_family */
+XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash)
+{
+    XXH_STATIC_ASSERT(sizeof(XXH32_canonical_t) == sizeof(XXH32_hash_t));
+    if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap32(hash);
+    XXH_memcpy(dst, &hash, sizeof(*dst));
 }
-
-/*! @ingroup xxh32_family */
-XXH_PUBLIC_API XXH32_hash_t
-XXH32_hashFromCanonical(const XXH32_canonical_t *src) {
-
-  return XXH_readBE32(src);
-
+/*! @ingroup XXH32_family */
+XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src)
+{
+    return XXH_readBE32(src);
 }
 
-  #ifndef XXH_NO_LONG_LONG
+
+#ifndef XXH_NO_LONG_LONG
 
 /* *******************************************************************
- *  64-bit hash functions
- *********************************************************************/
+*  64-bit hash functions
+*********************************************************************/
 /*!
  * @}
  * @ingroup impl
  * @{
-
  */
 /*******   Memory access   *******/
 
 typedef XXH64_hash_t xxh_u64;
 
-    #ifdef XXH_OLD_NAMES
-      #define U64 xxh_u64
-    #endif
-
-    #if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS == 3))
-    /*
-     * Manual byteshift. Best for old compilers which don't inline memcpy.
-     * We actually directly use XXH_readLE64 and XXH_readBE64.
-     */
-    #elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS == 2))
-
-/* Force direct memory access. Only works on CPU which support unaligned memory
- * access in hardware */
-static xxh_u64 XXH_read64(const void *memPtr) {
+#ifdef XXH_OLD_NAMES
+#  define U64 xxh_u64
+#endif
 
-  return *(const xxh_u64 *)memPtr;
+#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
+/*
+ * Manual byteshift. Best for old compilers which don't inline memcpy.
+ * We actually directly use XXH_readLE64 and XXH_readBE64.
+ */
+#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))
 
+/* Force direct memory access. Only works on CPU which support unaligned memory access in hardware */
+static xxh_u64 XXH_read64(const void* memPtr)
+{
+    return *(const xxh_u64*) memPtr;
 }
 
-    #elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS == 1))
-
-      /*
-       * __pack instructions are safer, but compiler specific, hence potentially
-       * problematic for some compilers.
-       *
-       * Currently only defined for GCC and ICC.
-       */
-      #ifdef XXH_OLD_NAMES
-typedef union {
-
-  xxh_u32 u32;
-  xxh_u64 u64;
-
-} __attribute__((packed)) unalign64;
-
-      #endif
-static xxh_u64 XXH_read64(const void *ptr) {
-
-  typedef union {
-
-    xxh_u32 u32;
-    xxh_u64 u64;
-
-  } __attribute__((packed)) xxh_unalign64;
-
-  return ((const xxh_unalign64 *)ptr)->u64;
+#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))
 
+/*
+ * __attribute__((aligned(1))) is supported by gcc and clang. Originally the
+ * documentation claimed that it only increased the alignment, but actually it
+ * can decrease it on gcc, clang, and icc:
+ * https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69502,
+ * https://gcc.godbolt.org/z/xYez1j67Y.
+ */
+#ifdef XXH_OLD_NAMES
+typedef union { xxh_u32 u32; xxh_u64 u64; } __attribute__((packed)) unalign64;
+#endif
+static xxh_u64 XXH_read64(const void* ptr)
+{
+    typedef __attribute__((aligned(1))) xxh_u64 xxh_unalign64;
+    return *((const xxh_unalign64*)ptr);
 }
 
-    #else
+#else
 
 /*
  * Portable and safe solution. Generally efficient.
- * see: https://stackoverflow.com/a/32095106/646947
+ * see: https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html
  */
-static xxh_u64 XXH_read64(const void *memPtr) {
-
-  xxh_u64 val;
-  memcpy(&val, memPtr, sizeof(val));
-  return val;
-
-}
-
-    #endif                                /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
-
-    #if defined(_MSC_VER)                                  /* Visual Studio */
-      #define XXH_swap64 _byteswap_uint64
-    #elif XXH_GCC_VERSION >= 403
-      #define XXH_swap64 __builtin_bswap64
-    #else
-static xxh_u64 XXH_swap64(xxh_u64 x) {
-
-  return ((x << 56) & 0xff00000000000000ULL) |
-         ((x << 40) & 0x00ff000000000000ULL) |
-         ((x << 24) & 0x0000ff0000000000ULL) |
-         ((x << 8) & 0x000000ff00000000ULL) |
-         ((x >> 8) & 0x00000000ff000000ULL) |
-         ((x >> 24) & 0x0000000000ff0000ULL) |
-         ((x >> 40) & 0x000000000000ff00ULL) |
-         ((x >> 56) & 0x00000000000000ffULL);
-
+static xxh_u64 XXH_read64(const void* memPtr)
+{
+    xxh_u64 val;
+    XXH_memcpy(&val, memPtr, sizeof(val));
+    return val;
+}
+
+#endif   /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
+
+#if defined(_MSC_VER)     /* Visual Studio */
+#  define XXH_swap64 _byteswap_uint64
+#elif XXH_GCC_VERSION >= 403
+#  define XXH_swap64 __builtin_bswap64
+#else
+static xxh_u64 XXH_swap64(xxh_u64 x)
+{
+    return  ((x << 56) & 0xff00000000000000ULL) |
+            ((x << 40) & 0x00ff000000000000ULL) |
+            ((x << 24) & 0x0000ff0000000000ULL) |
+            ((x << 8)  & 0x000000ff00000000ULL) |
+            ((x >> 8)  & 0x00000000ff000000ULL) |
+            ((x >> 24) & 0x0000000000ff0000ULL) |
+            ((x >> 40) & 0x000000000000ff00ULL) |
+            ((x >> 56) & 0x00000000000000ffULL);
 }
+#endif
 
-    #endif
-
-    /* XXH_FORCE_MEMORY_ACCESS==3 is an endian-independent byteshift load. */
-    #if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS == 3))
-
-XXH_FORCE_INLINE xxh_u64 XXH_readLE64(const void *memPtr) {
 
-  const xxh_u8 *bytePtr = (const xxh_u8 *)memPtr;
-  return bytePtr[0] | ((xxh_u64)bytePtr[1] << 8) | ((xxh_u64)bytePtr[2] << 16) |
-         ((xxh_u64)bytePtr[3] << 24) | ((xxh_u64)bytePtr[4] << 32) |
-         ((xxh_u64)bytePtr[5] << 40) | ((xxh_u64)bytePtr[6] << 48) |
-         ((xxh_u64)bytePtr[7] << 56);
+/* XXH_FORCE_MEMORY_ACCESS==3 is an endian-independent byteshift load. */
+#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
 
+XXH_FORCE_INLINE xxh_u64 XXH_readLE64(const void* memPtr)
+{
+    const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
+    return bytePtr[0]
+         | ((xxh_u64)bytePtr[1] << 8)
+         | ((xxh_u64)bytePtr[2] << 16)
+         | ((xxh_u64)bytePtr[3] << 24)
+         | ((xxh_u64)bytePtr[4] << 32)
+         | ((xxh_u64)bytePtr[5] << 40)
+         | ((xxh_u64)bytePtr[6] << 48)
+         | ((xxh_u64)bytePtr[7] << 56);
 }
 
-XXH_FORCE_INLINE xxh_u64 XXH_readBE64(const void *memPtr) {
-
-  const xxh_u8 *bytePtr = (const xxh_u8 *)memPtr;
-  return bytePtr[7] | ((xxh_u64)bytePtr[6] << 8) | ((xxh_u64)bytePtr[5] << 16) |
-         ((xxh_u64)bytePtr[4] << 24) | ((xxh_u64)bytePtr[3] << 32) |
-         ((xxh_u64)bytePtr[2] << 40) | ((xxh_u64)bytePtr[1] << 48) |
-         ((xxh_u64)bytePtr[0] << 56);
-
+XXH_FORCE_INLINE xxh_u64 XXH_readBE64(const void* memPtr)
+{
+    const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
+    return bytePtr[7]
+         | ((xxh_u64)bytePtr[6] << 8)
+         | ((xxh_u64)bytePtr[5] << 16)
+         | ((xxh_u64)bytePtr[4] << 24)
+         | ((xxh_u64)bytePtr[3] << 32)
+         | ((xxh_u64)bytePtr[2] << 40)
+         | ((xxh_u64)bytePtr[1] << 48)
+         | ((xxh_u64)bytePtr[0] << 56);
 }
 
-    #else
-XXH_FORCE_INLINE xxh_u64 XXH_readLE64(const void *ptr) {
-
-  return XXH_CPU_LITTLE_ENDIAN ? XXH_read64(ptr) : XXH_swap64(XXH_read64(ptr));
-
+#else
+XXH_FORCE_INLINE xxh_u64 XXH_readLE64(const void* ptr)
+{
+    return XXH_CPU_LITTLE_ENDIAN ? XXH_read64(ptr) : XXH_swap64(XXH_read64(ptr));
 }
 
-static xxh_u64 XXH_readBE64(const void *ptr) {
-
-  return XXH_CPU_LITTLE_ENDIAN ? XXH_swap64(XXH_read64(ptr)) : XXH_read64(ptr);
-
+static xxh_u64 XXH_readBE64(const void* ptr)
+{
+    return XXH_CPU_LITTLE_ENDIAN ? XXH_swap64(XXH_read64(ptr)) : XXH_read64(ptr);
 }
+#endif
 
-    #endif
-
-XXH_FORCE_INLINE xxh_u64 XXH_readLE64_align(const void   *ptr,
-                                            XXH_alignment align) {
-
-  if (align == XXH_unaligned)
-    return XXH_readLE64(ptr);
-  else
-    return XXH_CPU_LITTLE_ENDIAN ? *(const xxh_u64 *)ptr
-                                 : XXH_swap64(*(const xxh_u64 *)ptr);
-
+XXH_FORCE_INLINE xxh_u64
+XXH_readLE64_align(const void* ptr, XXH_alignment align)
+{
+    if (align==XXH_unaligned)
+        return XXH_readLE64(ptr);
+    else
+        return XXH_CPU_LITTLE_ENDIAN ? *(const xxh_u64*)ptr : XXH_swap64(*(const xxh_u64*)ptr);
 }
 
-    /*******   xxh64   *******/
-    /*!
-     * @}
-     * @defgroup xxh64_impl XXH64 implementation
-     * @ingroup impl
-     * @{
-
-     */
-    /* #define rather that static const, to be used as initializers */
-    #define XXH_PRIME64_1                                                                         \
-      0x9E3779B185EBCA87ULL /*!<                                                                  \
-                               0b1001111000110111011110011011000110000101111010111100101010000111 \
-                             */
-    #define XXH_PRIME64_2                                                                         \
-      0xC2B2AE3D27D4EB4FULL /*!<                                                                  \
-                               0b1100001010110010101011100011110100100111110101001110101101001111 \
-                             */
-    #define XXH_PRIME64_3                                                                         \
-      0x165667B19E3779F9ULL /*!<                                                                  \
-                               0b0001011001010110011001111011000110011110001101110111100111111001 \
-                             */
-    #define XXH_PRIME64_4                                                                         \
-      0x85EBCA77C2B2AE63ULL /*!<                                                                  \
-                               0b1000010111101011110010100111011111000010101100101010111001100011 \
-                             */
-    #define XXH_PRIME64_5                                                                         \
-      0x27D4EB2F165667C5ULL /*!<                                                                  \
-                               0b0010011111010100111010110010111100010110010101100110011111000101 \
-                             */
-
-    #ifdef XXH_OLD_NAMES
-      #define PRIME64_1 XXH_PRIME64_1
-      #define PRIME64_2 XXH_PRIME64_2
-      #define PRIME64_3 XXH_PRIME64_3
-      #define PRIME64_4 XXH_PRIME64_4
-      #define PRIME64_5 XXH_PRIME64_5
-    #endif
 
-static xxh_u64 XXH64_round(xxh_u64 acc, xxh_u64 input) {
-
-  acc += input * XXH_PRIME64_2;
-  acc = XXH_rotl64(acc, 31);
-  acc *= XXH_PRIME64_1;
-  return acc;
+/*******   xxh64   *******/
+/*!
+ * @}
+ * @defgroup XXH64_impl XXH64 implementation
+ * @ingroup impl
+ *
+ * Details on the XXH64 implementation.
+ * @{
+ */
+/* #define rather that static const, to be used as initializers */
+#define XXH_PRIME64_1  0x9E3779B185EBCA87ULL  /*!< 0b1001111000110111011110011011000110000101111010111100101010000111 */
+#define XXH_PRIME64_2  0xC2B2AE3D27D4EB4FULL  /*!< 0b1100001010110010101011100011110100100111110101001110101101001111 */
+#define XXH_PRIME64_3  0x165667B19E3779F9ULL  /*!< 0b0001011001010110011001111011000110011110001101110111100111111001 */
+#define XXH_PRIME64_4  0x85EBCA77C2B2AE63ULL  /*!< 0b1000010111101011110010100111011111000010101100101010111001100011 */
+#define XXH_PRIME64_5  0x27D4EB2F165667C5ULL  /*!< 0b0010011111010100111010110010111100010110010101100110011111000101 */
+
+#ifdef XXH_OLD_NAMES
+#  define PRIME64_1 XXH_PRIME64_1
+#  define PRIME64_2 XXH_PRIME64_2
+#  define PRIME64_3 XXH_PRIME64_3
+#  define PRIME64_4 XXH_PRIME64_4
+#  define PRIME64_5 XXH_PRIME64_5
+#endif
 
+/*! @copydoc XXH32_round */
+static xxh_u64 XXH64_round(xxh_u64 acc, xxh_u64 input)
+{
+    acc += input * XXH_PRIME64_2;
+    acc  = XXH_rotl64(acc, 31);
+    acc *= XXH_PRIME64_1;
+    return acc;
 }
 
-static xxh_u64 XXH64_mergeRound(xxh_u64 acc, xxh_u64 val) {
-
-  val = XXH64_round(0, val);
-  acc ^= val;
-  acc = acc * XXH_PRIME64_1 + XXH_PRIME64_4;
-  return acc;
-
+static xxh_u64 XXH64_mergeRound(xxh_u64 acc, xxh_u64 val)
+{
+    val  = XXH64_round(0, val);
+    acc ^= val;
+    acc  = acc * XXH_PRIME64_1 + XXH_PRIME64_4;
+    return acc;
 }
 
-static xxh_u64 XXH64_avalanche(xxh_u64 h64) {
-
-  h64 ^= h64 >> 33;
-  h64 *= XXH_PRIME64_2;
-  h64 ^= h64 >> 29;
-  h64 *= XXH_PRIME64_3;
-  h64 ^= h64 >> 32;
-  return h64;
-
+/*! @copydoc XXH32_avalanche */
+static xxh_u64 XXH64_avalanche(xxh_u64 hash)
+{
+    hash ^= hash >> 33;
+    hash *= XXH_PRIME64_2;
+    hash ^= hash >> 29;
+    hash *= XXH_PRIME64_3;
+    hash ^= hash >> 32;
+    return hash;
 }
 
-    #define XXH_get64bits(p) XXH_readLE64_align(p, align)
-
-static xxh_u64 XXH64_finalize(xxh_u64 h64, const xxh_u8 *ptr, size_t len,
-                              XXH_alignment align) {
-
-  len &= 31;
-  while (len >= 8) {
-
-    xxh_u64 const k1 = XXH64_round(0, XXH_get64bits(ptr));
-    ptr += 8;
-    h64 ^= k1;
-    h64 = XXH_rotl64(h64, 27) * XXH_PRIME64_1 + XXH_PRIME64_4;
-    len -= 8;
-
-  }
-
-  if (len >= 4) {
-
-    h64 ^= (xxh_u64)(XXH_get32bits(ptr)) * XXH_PRIME64_1;
-    ptr += 4;
-    h64 = XXH_rotl64(h64, 23) * XXH_PRIME64_2 + XXH_PRIME64_3;
-    len -= 4;
-
-  }
-
-  while (len > 0) {
 
-    h64 ^= (*ptr++) * XXH_PRIME64_5;
-    h64 = XXH_rotl64(h64, 11) * XXH_PRIME64_1;
-    --len;
-
-  }
-
-  return XXH64_avalanche(h64);
+#define XXH_get64bits(p) XXH_readLE64_align(p, align)
 
+/*!
+ * @internal
+ * @brief Processes the last 0-31 bytes of @p ptr.
+ *
+ * There may be up to 31 bytes remaining to consume from the input.
+ * This final stage will digest them to ensure that all input bytes are present
+ * in the final mix.
+ *
+ * @param hash The hash to finalize.
+ * @param ptr The pointer to the remaining input.
+ * @param len The remaining length, modulo 32.
+ * @param align Whether @p ptr is aligned.
+ * @return The finalized hash
+ * @see XXH32_finalize().
+ */
+static XXH_PUREF xxh_u64
+XXH64_finalize(xxh_u64 hash, const xxh_u8* ptr, size_t len, XXH_alignment align)
+{
+    if (ptr==NULL) XXH_ASSERT(len == 0);
+    len &= 31;
+    while (len >= 8) {
+        xxh_u64 const k1 = XXH64_round(0, XXH_get64bits(ptr));
+        ptr += 8;
+        hash ^= k1;
+        hash  = XXH_rotl64(hash,27) * XXH_PRIME64_1 + XXH_PRIME64_4;
+        len -= 8;
+    }
+    if (len >= 4) {
+        hash ^= (xxh_u64)(XXH_get32bits(ptr)) * XXH_PRIME64_1;
+        ptr += 4;
+        hash = XXH_rotl64(hash, 23) * XXH_PRIME64_2 + XXH_PRIME64_3;
+        len -= 4;
+    }
+    while (len > 0) {
+        hash ^= (*ptr++) * XXH_PRIME64_5;
+        hash = XXH_rotl64(hash, 11) * XXH_PRIME64_1;
+        --len;
+    }
+    return  XXH64_avalanche(hash);
 }
 
-    #ifdef XXH_OLD_NAMES
-      #define PROCESS1_64 XXH_PROCESS1_64
-      #define PROCESS4_64 XXH_PROCESS4_64
-      #define PROCESS8_64 XXH_PROCESS8_64
-    #else
-      #undef XXH_PROCESS1_64
-      #undef XXH_PROCESS4_64
-      #undef XXH_PROCESS8_64
-    #endif
-
-XXH_FORCE_INLINE xxh_u64 XXH64_endian_align(const xxh_u8 *input, size_t len,
-                                            xxh_u64 seed, XXH_alignment align) {
-
-  const xxh_u8 *bEnd = input ? input + len : NULL;
-  xxh_u64       h64;
-
-    #if defined(XXH_ACCEPT_NULL_INPUT_POINTER) && \
-        (XXH_ACCEPT_NULL_INPUT_POINTER >= 1)
-  if (input == NULL) {
-
-    len = 0;
-    bEnd = input = (const xxh_u8 *)(size_t)32;
-
-  }
-
-    #endif
-
-  if (len >= 32) {
-
-    const xxh_u8 *const limit = bEnd - 32;
-    xxh_u64             v1 = seed + XXH_PRIME64_1 + XXH_PRIME64_2;
-    xxh_u64             v2 = seed + XXH_PRIME64_2;
-    xxh_u64             v3 = seed + 0;
-    xxh_u64             v4 = seed - XXH_PRIME64_1;
-
-    do {
-
-      v1 = XXH64_round(v1, XXH_get64bits(input));
-      input += 8;
-      v2 = XXH64_round(v2, XXH_get64bits(input));
-      input += 8;
-      v3 = XXH64_round(v3, XXH_get64bits(input));
-      input += 8;
-      v4 = XXH64_round(v4, XXH_get64bits(input));
-      input += 8;
-
-    } while (input <= limit);
-
-    h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) +
-          XXH_rotl64(v4, 18);
-    h64 = XXH64_mergeRound(h64, v1);
-    h64 = XXH64_mergeRound(h64, v2);
-    h64 = XXH64_mergeRound(h64, v3);
-    h64 = XXH64_mergeRound(h64, v4);
-
-  } else {
-
-    h64 = seed + XXH_PRIME64_5;
+#ifdef XXH_OLD_NAMES
+#  define PROCESS1_64 XXH_PROCESS1_64
+#  define PROCESS4_64 XXH_PROCESS4_64
+#  define PROCESS8_64 XXH_PROCESS8_64
+#else
+#  undef XXH_PROCESS1_64
+#  undef XXH_PROCESS4_64
+#  undef XXH_PROCESS8_64
+#endif
 
-  }
+/*!
+ * @internal
+ * @brief The implementation for @ref XXH64().
+ *
+ * @param input , len , seed Directly passed from @ref XXH64().
+ * @param align Whether @p input is aligned.
+ * @return The calculated hash.
+ */
+XXH_FORCE_INLINE XXH_PUREF xxh_u64
+XXH64_endian_align(const xxh_u8* input, size_t len, xxh_u64 seed, XXH_alignment align)
+{
+    xxh_u64 h64;
+    if (input==NULL) XXH_ASSERT(len == 0);
+
+    if (len>=32) {
+        const xxh_u8* const bEnd = input + len;
+        const xxh_u8* const limit = bEnd - 31;
+        xxh_u64 v1 = seed + XXH_PRIME64_1 + XXH_PRIME64_2;
+        xxh_u64 v2 = seed + XXH_PRIME64_2;
+        xxh_u64 v3 = seed + 0;
+        xxh_u64 v4 = seed - XXH_PRIME64_1;
+
+        do {
+            v1 = XXH64_round(v1, XXH_get64bits(input)); input+=8;
+            v2 = XXH64_round(v2, XXH_get64bits(input)); input+=8;
+            v3 = XXH64_round(v3, XXH_get64bits(input)); input+=8;
+            v4 = XXH64_round(v4, XXH_get64bits(input)); input+=8;
+        } while (input<limit);
+
+        h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) + XXH_rotl64(v4, 18);
+        h64 = XXH64_mergeRound(h64, v1);
+        h64 = XXH64_mergeRound(h64, v2);
+        h64 = XXH64_mergeRound(h64, v3);
+        h64 = XXH64_mergeRound(h64, v4);
 
-  h64 += (xxh_u64)len;
+    } else {
+        h64  = seed + XXH_PRIME64_5;
+    }
 
-  return XXH64_finalize(h64, input, len, align);
+    h64 += (xxh_u64) len;
 
+    return XXH64_finalize(h64, input, len, align);
 }
 
-/*! @ingroup xxh64_family */
-XXH_PUBLIC_API XXH64_hash_t XXH64(const void *input, size_t len,
-                                  XXH64_hash_t seed) {
 
-    #if 0
+/*! @ingroup XXH64_family */
+XXH_PUBLIC_API XXH64_hash_t XXH64 (XXH_NOESCAPE const void* input, size_t len, XXH64_hash_t seed)
+{
+#if !defined(XXH_NO_STREAM) && XXH_SIZE_OPT >= 2
     /* Simple version, good for code maintenance, but unfortunately slow for small inputs */
     XXH64_state_t state;
     XXH64_reset(&state, seed);
     XXH64_update(&state, (const xxh_u8*)input, len);
     return XXH64_digest(&state);
-    #else
-  if (XXH_FORCE_ALIGN_CHECK) {
-
-    if ((((size_t)input) & 7) ==
-        0) {        /* Input is aligned, let's leverage the speed advantage */
-      return XXH64_endian_align((const xxh_u8 *)input, len, seed, XXH_aligned);
-
-    }
-
-  }
+#else
+    if (XXH_FORCE_ALIGN_CHECK) {
+        if ((((size_t)input) & 7)==0) {  /* Input is aligned, let's leverage the speed advantage */
+            return XXH64_endian_align((const xxh_u8*)input, len, seed, XXH_aligned);
+    }   }
 
-  return XXH64_endian_align((const xxh_u8 *)input, len, seed, XXH_unaligned);
-
-    #endif
+    return XXH64_endian_align((const xxh_u8*)input, len, seed, XXH_unaligned);
 
+#endif
 }
 
 /*******   Hash Streaming   *******/
-
-/*! @ingroup xxh64_family*/
-XXH_PUBLIC_API XXH64_state_t *XXH64_createState(void) {
-
-  return (XXH64_state_t *)XXH_malloc(sizeof(XXH64_state_t));
-
-}
-
-/*! @ingroup xxh64_family */
-XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t *statePtr) {
-
-  XXH_free(statePtr);
-  return XXH_OK;
-
-}
-
-/*! @ingroup xxh64_family */
-XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t       *dstState,
-                                    const XXH64_state_t *srcState) {
-
-  memcpy(dstState, srcState, sizeof(*dstState));
-
+#ifndef XXH_NO_STREAM
+/*! @ingroup XXH64_family*/
+XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void)
+{
+    return (XXH64_state_t*)XXH_malloc(sizeof(XXH64_state_t));
+}
+/*! @ingroup XXH64_family */
+XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr)
+{
+    XXH_free(statePtr);
+    return XXH_OK;
 }
 
-/*! @ingroup xxh64_family */
-XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH64_state_t *statePtr,
-                                         XXH64_hash_t   seed) {
-
-  XXH64_state_t state; /* use a local state to memcpy() in order to avoid
-                          strict-aliasing warnings */
-  memset(&state, 0, sizeof(state));
-  state.v1 = seed + XXH_PRIME64_1 + XXH_PRIME64_2;
-  state.v2 = seed + XXH_PRIME64_2;
-  state.v3 = seed + 0;
-  state.v4 = seed - XXH_PRIME64_1;
-  /* do not write into reserved64, might be removed in a future version */
-  memcpy(statePtr, &state, sizeof(state) - sizeof(state.reserved64));
-  return XXH_OK;
-
+/*! @ingroup XXH64_family */
+XXH_PUBLIC_API void XXH64_copyState(XXH_NOESCAPE XXH64_state_t* dstState, const XXH64_state_t* srcState)
+{
+    XXH_memcpy(dstState, srcState, sizeof(*dstState));
 }
 
-/*! @ingroup xxh64_family */
-XXH_PUBLIC_API XXH_errorcode XXH64_update(XXH64_state_t *state,
-                                          const void *input, size_t len) {
-
-  if (input == NULL)
-    #if defined(XXH_ACCEPT_NULL_INPUT_POINTER) && \
-        (XXH_ACCEPT_NULL_INPUT_POINTER >= 1)
+/*! @ingroup XXH64_family */
+XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH_NOESCAPE XXH64_state_t* statePtr, XXH64_hash_t seed)
+{
+    XXH_ASSERT(statePtr != NULL);
+    memset(statePtr, 0, sizeof(*statePtr));
+    statePtr->v[0] = seed + XXH_PRIME64_1 + XXH_PRIME64_2;
+    statePtr->v[1] = seed + XXH_PRIME64_2;
+    statePtr->v[2] = seed + 0;
+    statePtr->v[3] = seed - XXH_PRIME64_1;
     return XXH_OK;
-    #else
-    return XXH_ERROR;
-    #endif
-
-  {
-
-    const xxh_u8       *p = (const xxh_u8 *)input;
-    const xxh_u8 *const bEnd = p + len;
-
-    state->total_len += len;
-
-    if (state->memsize + len < 32) {                  /* fill in tmp buffer */
-      XXH_memcpy(((xxh_u8 *)state->mem64) + state->memsize, input, len);
-      state->memsize += (xxh_u32)len;
-      return XXH_OK;
-
-    }
-
-    if (state->memsize) {                             /* tmp buffer is full */
-      XXH_memcpy(((xxh_u8 *)state->mem64) + state->memsize, input,
-                 32 - state->memsize);
-      state->v1 = XXH64_round(state->v1, XXH_readLE64(state->mem64 + 0));
-      state->v2 = XXH64_round(state->v2, XXH_readLE64(state->mem64 + 1));
-      state->v3 = XXH64_round(state->v3, XXH_readLE64(state->mem64 + 2));
-      state->v4 = XXH64_round(state->v4, XXH_readLE64(state->mem64 + 3));
-      p += 32 - state->memsize;
-      state->memsize = 0;
+}
 
+/*! @ingroup XXH64_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH64_update (XXH_NOESCAPE XXH64_state_t* state, XXH_NOESCAPE const void* input, size_t len)
+{
+    if (input==NULL) {
+        XXH_ASSERT(len == 0);
+        return XXH_OK;
     }
 
-    if (p + 32 <= bEnd) {
-
-      const xxh_u8 *const limit = bEnd - 32;
-      xxh_u64             v1 = state->v1;
-      xxh_u64             v2 = state->v2;
-      xxh_u64             v3 = state->v3;
-      xxh_u64             v4 = state->v4;
+    {   const xxh_u8* p = (const xxh_u8*)input;
+        const xxh_u8* const bEnd = p + len;
 
-      do {
+        state->total_len += len;
 
-        v1 = XXH64_round(v1, XXH_readLE64(p));
-        p += 8;
-        v2 = XXH64_round(v2, XXH_readLE64(p));
-        p += 8;
-        v3 = XXH64_round(v3, XXH_readLE64(p));
-        p += 8;
-        v4 = XXH64_round(v4, XXH_readLE64(p));
-        p += 8;
-
-      } while (p <= limit);
+        if (state->memsize + len < 32) {  /* fill in tmp buffer */
+            XXH_memcpy(((xxh_u8*)state->mem64) + state->memsize, input, len);
+            state->memsize += (xxh_u32)len;
+            return XXH_OK;
+        }
 
-      state->v1 = v1;
-      state->v2 = v2;
-      state->v3 = v3;
-      state->v4 = v4;
+        if (state->memsize) {   /* tmp buffer is full */
+            XXH_memcpy(((xxh_u8*)state->mem64) + state->memsize, input, 32-state->memsize);
+            state->v[0] = XXH64_round(state->v[0], XXH_readLE64(state->mem64+0));
+            state->v[1] = XXH64_round(state->v[1], XXH_readLE64(state->mem64+1));
+            state->v[2] = XXH64_round(state->v[2], XXH_readLE64(state->mem64+2));
+            state->v[3] = XXH64_round(state->v[3], XXH_readLE64(state->mem64+3));
+            p += 32 - state->memsize;
+            state->memsize = 0;
+        }
 
-    }
+        if (p+32 <= bEnd) {
+            const xxh_u8* const limit = bEnd - 32;
 
-    if (p < bEnd) {
+            do {
+                state->v[0] = XXH64_round(state->v[0], XXH_readLE64(p)); p+=8;
+                state->v[1] = XXH64_round(state->v[1], XXH_readLE64(p)); p+=8;
+                state->v[2] = XXH64_round(state->v[2], XXH_readLE64(p)); p+=8;
+                state->v[3] = XXH64_round(state->v[3], XXH_readLE64(p)); p+=8;
+            } while (p<=limit);
 
-      XXH_memcpy(state->mem64, p, (size_t)(bEnd - p));
-      state->memsize = (unsigned)(bEnd - p);
+        }
 
+        if (p < bEnd) {
+            XXH_memcpy(state->mem64, p, (size_t)(bEnd-p));
+            state->memsize = (unsigned)(bEnd-p);
+        }
     }
 
-  }
-
-  return XXH_OK;
-
+    return XXH_OK;
 }
 
-/*! @ingroup xxh64_family */
-XXH_PUBLIC_API XXH64_hash_t XXH64_digest(const XXH64_state_t *state) {
-
-  xxh_u64 h64;
-
-  if (state->total_len >= 32) {
-
-    xxh_u64 const v1 = state->v1;
-    xxh_u64 const v2 = state->v2;
-    xxh_u64 const v3 = state->v3;
-    xxh_u64 const v4 = state->v4;
 
-    h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) +
-          XXH_rotl64(v4, 18);
-    h64 = XXH64_mergeRound(h64, v1);
-    h64 = XXH64_mergeRound(h64, v2);
-    h64 = XXH64_mergeRound(h64, v3);
-    h64 = XXH64_mergeRound(h64, v4);
+/*! @ingroup XXH64_family */
+XXH_PUBLIC_API XXH64_hash_t XXH64_digest(XXH_NOESCAPE const XXH64_state_t* state)
+{
+    xxh_u64 h64;
 
-  } else {
-
-    h64 = state->v3 /*seed*/ + XXH_PRIME64_5;
-
-  }
-
-  h64 += (xxh_u64)state->total_len;
+    if (state->total_len >= 32) {
+        h64 = XXH_rotl64(state->v[0], 1) + XXH_rotl64(state->v[1], 7) + XXH_rotl64(state->v[2], 12) + XXH_rotl64(state->v[3], 18);
+        h64 = XXH64_mergeRound(h64, state->v[0]);
+        h64 = XXH64_mergeRound(h64, state->v[1]);
+        h64 = XXH64_mergeRound(h64, state->v[2]);
+        h64 = XXH64_mergeRound(h64, state->v[3]);
+    } else {
+        h64  = state->v[2] /*seed*/ + XXH_PRIME64_5;
+    }
 
-  return XXH64_finalize(h64, (const xxh_u8 *)state->mem64,
-                        (size_t)state->total_len, XXH_aligned);
+    h64 += (xxh_u64) state->total_len;
 
+    return XXH64_finalize(h64, (const xxh_u8*)state->mem64, (size_t)state->total_len, XXH_aligned);
 }
+#endif /* !XXH_NO_STREAM */
 
 /******* Canonical representation   *******/
 
-/*! @ingroup xxh64_family */
-XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t *dst,
-                                            XXH64_hash_t       hash) {
-
-  XXH_STATIC_ASSERT(sizeof(XXH64_canonical_t) == sizeof(XXH64_hash_t));
-  if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap64(hash);
-  memcpy(dst, &hash, sizeof(*dst));
-
+/*! @ingroup XXH64_family */
+XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH_NOESCAPE XXH64_canonical_t* dst, XXH64_hash_t hash)
+{
+    XXH_STATIC_ASSERT(sizeof(XXH64_canonical_t) == sizeof(XXH64_hash_t));
+    if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap64(hash);
+    XXH_memcpy(dst, &hash, sizeof(*dst));
 }
 
-/*! @ingroup xxh64_family */
-XXH_PUBLIC_API XXH64_hash_t
-XXH64_hashFromCanonical(const XXH64_canonical_t *src) {
-
-  return XXH_readBE64(src);
-
+/*! @ingroup XXH64_family */
+XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(XXH_NOESCAPE const XXH64_canonical_t* src)
+{
+    return XXH_readBE64(src);
 }
 
-    #ifndef XXH_NO_XXH3
-
-    /* *********************************************************************
-     *  XXH3
-     *  New generation hash designed for speed on small keys and vectorization
-     ************************************************************************ */
-    /*!
-     * @}
-     * @defgroup xxh3_impl XXH3 implementation
-     * @ingroup impl
-     * @{
-
-     */
+#ifndef XXH_NO_XXH3
 
-    /* ===   Compiler specifics   === */
-
-      #if ((defined(sun) || defined(__sun)) &&                                \
-           __cplusplus) /* Solaris includes __STDC_VERSION__ with C++. Tested \
-                           with GCC 5.5 */
-        #define XXH_RESTRICT                                     /* disable */
-      #elif defined(__STDC_VERSION__) && \
-          __STDC_VERSION__ >= 199901L                             /* >= C99 */
-        #define XXH_RESTRICT restrict
-      #else
-      /* Note: it might be useful to define __restrict or __restrict__ for
-       * some C++ compilers */
-        #define XXH_RESTRICT                                     /* disable */
-      #endif
-
-      #if (defined(__GNUC__) && (__GNUC__ >= 3)) ||                   \
-          (defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 800)) || \
-          defined(__clang__)
-        #define XXH_likely(x) __builtin_expect(x, 1)
-        #define XXH_unlikely(x) __builtin_expect(x, 0)
-      #else
-        #define XXH_likely(x) (x)
-        #define XXH_unlikely(x) (x)
-      #endif
-
-      #if defined(__GNUC__)
-        #if defined(__AVX2__)
-          #include <immintrin.h>
-        #elif defined(__SSE2__)
-          #include <emmintrin.h>
-        #elif defined(__ARM_NEON__) || defined(__ARM_NEON)
-          #define inline __inline__               /* circumvent a clang bug */
-          #include <arm_neon.h>
-          #undef inline
-        #endif
-      #elif defined(_MSC_VER)
-        #include <intrin.h>
-      #endif
-
-      /*
-       * One goal of XXH3 is to make it fast on both 32-bit and 64-bit, while
-       * remaining a true 64-bit/128-bit hash function.
-       *
-       * This is done by prioritizing a subset of 64-bit operations that can be
-       * emulated without too many steps on the average 32-bit machine.
-       *
-       * For example, these two lines seem similar, and run equally fast on
-       * 64-bit:
-       *
-       *   xxh_u64 x;
-       *   x ^= (x >> 47); // good
-       *   x ^= (x >> 13); // bad
-       *
-       * However, to a 32-bit machine, there is a major difference.
-       *
-       * x ^= (x >> 47) looks like this:
-       *
-       *   x.lo ^= (x.hi >> (47 - 32));
-       *
-       * while x ^= (x >> 13) looks like this:
-       *
-       *   // note: funnel shifts are not usually cheap.
-       *   x.lo ^= (x.lo >> 13) | (x.hi << (32 - 13));
-       *   x.hi ^= (x.hi >> 13);
-       *
-       * The first one is significantly faster than the second, simply because
-       * the shift is larger than 32. This means:
-       *  - All the bits we need are in the upper 32 bits, so we can ignore the
-       * lower 32 bits in the shift.
-       *  - The shift result will always fit in the lower 32 bits, and
-       * therefore, we can ignore the upper 32 bits in the xor.
-       *
-       * Thanks to this optimization, XXH3 only requires these features to be
-       * efficient:
-       *
-       *  - Usable unaligned access
-       *  - A 32-bit or 64-bit ALU
-       *      - If 32-bit, a decent ADC instruction
-       *  - A 32 or 64-bit multiply with a 64-bit result
-       *  - For the 128-bit variant, a decent byteswap helps short inputs.
-       *
-       * The first two are already required by XXH32, and almost all 32-bit and
-       * 64-bit platforms which can run XXH32 can run XXH3 efficiently.
-       *
-       * Thumb-1, the classic 16-bit only subset of ARM's instruction set, is
-       * one notable exception.
-       *
-       * First of all, Thumb-1 lacks support for the UMULL instruction which
-       * performs the important long multiply. This means numerous __aeabi_lmul
-       * calls.
-       *
-       * Second of all, the 8 functional registers are just not enough.
-       * Setup for __aeabi_lmul, byteshift loads, pointers, and all arithmetic
-       * need Lo registers, and this shuffling results in thousands more MOVs
-       * than A32.
-       *
-       * A32 and T32 don't have this limitation. They can access all 14
-       * registers, do a 32->64 multiply with UMULL, and the flexible operand
-       * allowing free shifts is helpful, too.
-       *
-       * Therefore, we do a quick sanity check.
-       *
-       * If compiling Thumb-1 for a target which supports ARM instructions, we
-       * will emit a warning, as it is not a "sane" platform to compile for.
-       *
-       * Usually, if this happens, it is because of an accident and you probably
-       * need to specify -march, as you likely meant to compile for a newer
-       * architecture.
-       *
-       * Credit: large sections of the vectorial and asm source code paths
-       *         have been contributed by @easyaspi314
-       */
-      #if defined(__thumb__) && !defined(__thumb2__) && \
-          defined(__ARM_ARCH_ISA_ARM)
-        #warning "XXH3 is highly inefficient without ARM or Thumb-2."
-      #endif
-
-    /* ==========================================
-     * Vectorization detection
-     * ========================================== */
-
-      #ifdef XXH_DOXYGEN
-        /*!
-         * @ingroup tuning
-         * @brief Overrides the vectorization implementation chosen for XXH3.
-         *
-         * Can be defined to 0 to disable SIMD or any of the values mentioned in
-         * @ref XXH_VECTOR_TYPE.
-         *
-         * If this is not defined, it uses predefined macros to determine the
-         * best implementation.
-         */
-        #define XXH_VECTOR XXH_SCALAR
+/* *********************************************************************
+*  XXH3
+*  New generation hash designed for speed on small keys and vectorization
+************************************************************************ */
 /*!
- * @ingroup tuning
- * @brief Possible values for @ref XXH_VECTOR.
- *
- * Note that these are actually implemented as macros.
- *
- * If this is not defined, it is detected automatically.
- * @ref XXH_X86DISPATCH overrides this.
+ * @}
+ * @defgroup XXH3_impl XXH3 implementation
+ * @ingroup impl
+ * @{
  */
-enum XXH_VECTOR_TYPE /* fake enum */ {
 
-  XXH_SCALAR = 0,                              /*!< Portable scalar version */
-  XXH_SSE2 = 1,   /*!<
-                   * SSE2 for Pentium 4, Opteron, all x86_64.
-                   *
-                   * @note SSE2 is also guaranteed on Windows 10, macOS, and
-                   * Android x86.
-                   */
-  XXH_AVX2 = 2,                         /*!< AVX2 for Haswell and Bulldozer */
-  XXH_AVX512 = 3,                       /*!< AVX512 for Skylake and Icelake */
-  XXH_NEON = 4,                  /*!< NEON for most ARMv7-A and all AArch64 */
-  XXH_VSX = 5,                 /*!< VSX and ZVector for POWER8/z13 (64-bit) */
+/* ===   Compiler specifics   === */
 
-};
+#if ((defined(sun) || defined(__sun)) && __cplusplus) /* Solaris includes __STDC_VERSION__ with C++. Tested with GCC 5.5 */
+#  define XXH_RESTRICT   /* disable */
+#elif defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* >= C99 */
+#  define XXH_RESTRICT   restrict
+#elif (defined (__GNUC__) && ((__GNUC__ > 3) || (__GNUC__ == 3 && __GNUC_MINOR__ >= 1))) \
+   || (defined (__clang__)) \
+   || (defined (_MSC_VER) && (_MSC_VER >= 1400)) \
+   || (defined (__INTEL_COMPILER) && (__INTEL_COMPILER >= 1300))
+/*
+ * There are a LOT more compilers that recognize __restrict but this
+ * covers the major ones.
+ */
+#  define XXH_RESTRICT   __restrict
+#else
+#  define XXH_RESTRICT   /* disable */
+#endif
 
-        /*!
-         * @ingroup tuning
-         * @brief Selects the minimum alignment for XXH3's accumulators.
-         *
-         * When using SIMD, this should match the alignment reqired for said
-         * vector type, so, for example, 32 for AVX2.
-         *
-         * Default: Auto detected.
-         */
-        #define XXH_ACC_ALIGN 8
-      #endif
-
-      /* Actual definition */
-      #ifndef XXH_DOXYGEN
-        #define XXH_SCALAR 0
-        #define XXH_SSE2 1
-        #define XXH_AVX2 2
-        #define XXH_AVX512 3
-        #define XXH_NEON 4
-        #define XXH_VSX 5
-      #endif
-
-      #ifndef XXH_VECTOR                  /* can be defined on command line */
-        #if defined(__AVX512F__)
-          #define XXH_VECTOR XXH_AVX512
-        #elif defined(__AVX2__)
-          #define XXH_VECTOR XXH_AVX2
-        #elif defined(__SSE2__) || defined(_M_AMD64) || defined(_M_X64) || \
-            (defined(_M_IX86_FP) && (_M_IX86_FP == 2))
-          #define XXH_VECTOR XXH_SSE2
-        #elif defined(__GNUC__) /* msvc support maybe later */               \
-            && (defined(__ARM_NEON__) || defined(__ARM_NEON)) &&             \
-            (defined(                                                        \
-                 __LITTLE_ENDIAN__) /* We only support little endian NEON */ \
-             || (defined(__BYTE_ORDER__) &&                                  \
-                 __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__))
-          #define XXH_VECTOR XXH_NEON
-        #elif (defined(__PPC64__) && defined(__POWER8_VECTOR__)) || \
-            (defined(__s390x__) && defined(__VEC__)) &&             \
-                defined(__GNUC__)                           /* TODO: IBM XL */
-          #define XXH_VECTOR XXH_VSX
-        #else
-          #define XXH_VECTOR XXH_SCALAR
-        #endif
-      #endif
-
-      /*
-       * Controls the alignment of the accumulator,
-       * for compatibility with aligned vector loads, which are usually faster.
-       */
-      #ifndef XXH_ACC_ALIGN
-        #if defined(XXH_X86DISPATCH)
-          #define XXH_ACC_ALIGN 64         /* for compatibility with avx512 */
-        #elif XXH_VECTOR == XXH_SCALAR                            /* scalar */
-          #define XXH_ACC_ALIGN 8
-        #elif XXH_VECTOR == XXH_SSE2                                /* sse2 */
-          #define XXH_ACC_ALIGN 16
-        #elif XXH_VECTOR == XXH_AVX2                                /* avx2 */
-          #define XXH_ACC_ALIGN 32
-        #elif XXH_VECTOR == XXH_NEON                                /* neon */
-          #define XXH_ACC_ALIGN 16
-        #elif XXH_VECTOR == XXH_VSX                                  /* vsx */
-          #define XXH_ACC_ALIGN 16
-        #elif XXH_VECTOR == XXH_AVX512                            /* avx512 */
-          #define XXH_ACC_ALIGN 64
-        #endif
-      #endif
-
-      #if defined(XXH_X86DISPATCH) || XXH_VECTOR == XXH_SSE2 || \
-          XXH_VECTOR == XXH_AVX2 || XXH_VECTOR == XXH_AVX512
-        #define XXH_SEC_ALIGN XXH_ACC_ALIGN
-      #else
-        #define XXH_SEC_ALIGN 8
-      #endif
-
-      /*
-       * UGLY HACK:
-       * GCC usually generates the best code with -O3 for xxHash.
-       *
-       * However, when targeting AVX2, it is overzealous in its unrolling
-       * resulting in code roughly 3/4 the speed of Clang.
-       *
-       * There are other issues, such as GCC splitting _mm256_loadu_si256 into
-       * _mm_loadu_si128 + _mm256_inserti128_si256. This is an optimization
-       * which only applies to Sandy and Ivy Bridge... which don't even support
-       * AVX2.
-       *
-       * That is why when compiling the AVX2 version, it is recommended to use
-       * either -O2 -mavx2 -march=haswell or -O2 -mavx2
-       * -mno-avx256-split-unaligned-load for decent performance, or to use
-       * Clang instead.
-       *
-       * Fortunately, we can control the first one with a pragma that forces GCC
-       * into -O2, but the other one we can't control without "failed to inline
-       * always inline function due to target mismatch" warnings.
-       */
-      #if XXH_VECTOR == XXH_AVX2                      /* AVX2 */           \
-          && defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
-          && defined(__OPTIMIZE__) &&                                      \
-          !defined(__OPTIMIZE_SIZE__)                /* respect -O0 and -Os */
-        #pragma GCC push_options
-        #pragma GCC optimize("-O2")
-      #endif
-
-      #if XXH_VECTOR == XXH_NEON
-        /*
-         * NEON's setup for vmlal_u32 is a little more complicated than it is on
-         * SSE2, AVX2, and VSX.
-         *
-         * While PMULUDQ and VMULEUW both perform a mask, VMLAL.U32 performs an
-         * upcast.
-         *
-         * To do the same operation, the 128-bit 'Q' register needs to be split
-         * into two 64-bit 'D' registers, performing this operation::
-         *
-         *   [                a                 |                 b ] |
-         * '---------. .--------'                | |                         x |
-         *            |              .---------' '--------.                |
-         *   [ a & 0xFFFFFFFF | b & 0xFFFFFFFF ],[    a >> 32     |     b >> 32
-         * ]
-         *
-         * Due to significant changes in aarch64, the fastest method for aarch64
-         * is completely different than the fastest method for ARMv7-A.
-         *
-         * ARMv7-A treats D registers as unions overlaying Q registers, so
-         * modifying D11 will modify the high half of Q5. This is similar to how
-         * modifying AH will only affect bits 8-15 of AX on x86.
-         *
-         * VZIP takes two registers, and puts even lanes in one register and odd
-         * lanes in the other.
-         *
-         * On ARMv7-A, this strangely modifies both parameters in place instead
-         * of taking the usual 3-operand form.
-         *
-         * Therefore, if we want to do this, we can simply use a D-form VZIP.32
-         * on the lower and upper halves of the Q register to end up with the
-         * high and low halves where we want - all in one instruction.
-         *
-         *   vzip.32   d10, d11       @ d10 = { d10[0], d11[0] }; d11 = {
+#if (defined(__GNUC__) && (__GNUC__ >= 3))  \
+  || (defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 800)) \
+  || defined(__clang__)
+#    define XXH_likely(x) __builtin_expect(x, 1)
+#    define XXH_unlikely(x) __builtin_expect(x, 0)
+#else
+#    define XXH_likely(x) (x)
+#    define XXH_unlikely(x) (x)
+#endif
 
-         * d10[1], d11[1] }
-         *
-         * Unfortunately we need inline assembly for this: Instructions
-         * modifying two registers at once is not possible in GCC or Clang's IR,
-         * and they have to create a copy.
-         *
-         * aarch64 requires a different approach.
-         *
-         * In order to make it easier to write a decent compiler for aarch64,
-         * many quirks were removed, such as conditional execution.
-         *
-         * NEON was also affected by this.
-         *
-         * aarch64 cannot access the high bits of a Q-form register, and writes
-         * to a D-form register zero the high bits, similar to how writes to
-         * W-form scalar registers (or DWORD registers on x86_64) work.
-         *
-         * The formerly free vget_high intrinsics now require a vext (with a few
-         * exceptions)
-         *
-         * Additionally, VZIP was replaced by ZIP1 and ZIP2, which are the
-         * equivalent of PUNPCKL* and PUNPCKH* in SSE, respectively, in order to
-         * only modify one operand.
-         *
-         * The equivalent of the VZIP.32 on the lower and upper halves would be
-         * this mess:
-         *
-         *   ext     v2.4s, v0.4s, v0.4s, #2 // v2 = { v0[2], v0[3], v0[0],
-         * v0[1] } zip1    v1.2s, v0.2s, v2.2s     // v1 = { v0[0], v2[0] } zip2
-         * v0.2s, v0.2s, v1.2s     // v0 = { v0[1], v2[1] }
-         *
-         * Instead, we use a literal downcast, vmovn_u64 (XTN), and vshrn_n_u64
-         * (SHRN):
-         *
-         *   shrn    v1.2s, v0.2d, #32  // v1 = (uint32x2_t)(v0 >> 32);
-         *   xtn     v0.2s, v0.2d       // v0 = (uint32x2_t)(v0 & 0xFFFFFFFF);
-         *
-         * This is available on ARMv7-A, but is less efficient than a single
-         * VZIP.32.
-         */
+#ifndef XXH_HAS_INCLUDE
+#  ifdef __has_include
+/*
+ * Not defined as XXH_HAS_INCLUDE(x) (function-like) because
+ * this causes segfaults in Apple Clang 4.2 (on Mac OS X 10.7 Lion)
+ */
+#    define XXH_HAS_INCLUDE __has_include
+#  else
+#    define XXH_HAS_INCLUDE(x) 0
+#  endif
+#endif
 
-        /*!
-         * Function-like macro:
-         * void XXH_SPLIT_IN_PLACE(uint64x2_t &in, uint32x2_t &outLo, uint32x2_t
-         * &outHi)
-         * {
+#if defined(__GNUC__) || defined(__clang__)
+#  if defined(__ARM_FEATURE_SVE)
+#    include <arm_sve.h>
+#  endif
+#  if defined(__ARM_NEON__) || defined(__ARM_NEON) \
+   || (defined(_M_ARM) && _M_ARM >= 7) \
+   || defined(_M_ARM64) || defined(_M_ARM64EC) \
+   || (defined(__wasm_simd128__) && XXH_HAS_INCLUDE(<arm_neon.h>)) /* WASM SIMD128 via SIMDe */
+#    define inline __inline__  /* circumvent a clang bug */
+#    include <arm_neon.h>
+#    undef inline
+#  elif defined(__AVX2__)
+#    include <immintrin.h>
+#  elif defined(__SSE2__)
+#    include <emmintrin.h>
+#  endif
+#endif
 
-         *     outLo = (uint32x2_t)(in & 0xFFFFFFFF);
-         *     outHi = (uint32x2_t)(in >> 32);
-         *     in = UNDEFINED;
-         * }
-         */
-        #if !defined(XXH_NO_VZIP_HACK) /* define to disable */ \
-            && defined(__GNUC__) && !defined(__aarch64__) &&   \
-            !defined(__arm64__)
-          #define XXH_SPLIT_IN_PLACE(in, outLo, outHi)                                                   \
-            do {                                                                                         \
-                                                                                                         \
-              /* Undocumented GCC/Clang operand modifier: %e0 = lower D half,                            \
-               * %f0 = upper D half */                                                                   \
-              /* https://github.com/gcc-mirror/gcc/blob/38cf91e5/gcc/config/arm/arm.c#L22486             \
-               */                                                                                        \
-              /* https://github.com/llvm-mirror/llvm/blob/2c4ca683/lib/Target/ARM/ARMAsmPrinter.cpp#L399 \
-               */                                                                                        \
-              __asm__("vzip.32  %e0, %f0" : "+w"(in));                                                   \
-              (outLo) = vget_low_u32(vreinterpretq_u32_u64(in));                                         \
-              (outHi) = vget_high_u32(vreinterpretq_u32_u64(in));                                        \
-                                                                                                         \
-            } while (0)
-
-        #else
-          #define XXH_SPLIT_IN_PLACE(in, outLo, outHi) \
-            do {                                       \
-                                                       \
-              (outLo) = vmovn_u64(in);                 \
-              (outHi) = vshrn_n_u64((in), 32);         \
-                                                       \
-            } while (0)
-
-        #endif
-      #endif                                      /* XXH_VECTOR == XXH_NEON */
-
-      /*
-       * VSX and Z Vector helpers.
-       *
-       * This is very messy, and any pull requests to clean this up are welcome.
-       *
-       * There are a lot of problems with supporting VSX and s390x, due to
-       * inconsistent intrinsics, spotty coverage, and multiple endiannesses.
-       */
-      #if XXH_VECTOR == XXH_VSX
-        #if defined(__s390x__)
-          #include <s390intrin.h>
-        #else
-          /* gcc's altivec.h can have the unwanted consequence to
-           * unconditionally #define bool, vector, and pixel keywords, with bad
-           * consequences for programs already using these keywords for other
-           * purposes. The paragraph defining these macros is skipped when
-           * __APPLE_ALTIVEC__ is defined.
-           * __APPLE_ALTIVEC__ is _generally_ defined automatically by the
-           * compiler, but it seems that, in some cases, it isn't. Force the
-           * build macro to be defined, so that keywords are not altered.
-           */
-          #if defined(__GNUC__) && !defined(__APPLE_ALTIVEC__)
-            #define __APPLE_ALTIVEC__
-          #endif
-          #include <altivec.h>
-        #endif
+#if defined(_MSC_VER)
+#  include <intrin.h>
+#endif
 
-typedef __vector unsigned long long xxh_u64x2;
-typedef __vector unsigned char      xxh_u8x16;
-typedef __vector unsigned           xxh_u32x4;
-
-        #ifndef XXH_VSX_BE
-          #if defined(__BIG_ENDIAN__) ||  \
-              (defined(__BYTE_ORDER__) && \
-               __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
-            #define XXH_VSX_BE 1
-          #elif defined(__VEC_ELEMENT_REG_ORDER__) && \
-              __VEC_ELEMENT_REG_ORDER__ == __ORDER_BIG_ENDIAN__
-            #warning \
-                "-maltivec=be is not recommended. Please use native endianness."
-            #define XXH_VSX_BE 1
-          #else
-            #define XXH_VSX_BE 0
-          #endif
-        #endif                                      /* !defined(XXH_VSX_BE) */
-
-        #if XXH_VSX_BE
-          #if defined(__POWER9_VECTOR__) || \
-              (defined(__clang__) && defined(__s390x__))
-            #define XXH_vec_revb vec_revb
-          #else
+/*
+ * One goal of XXH3 is to make it fast on both 32-bit and 64-bit, while
+ * remaining a true 64-bit/128-bit hash function.
+ *
+ * This is done by prioritizing a subset of 64-bit operations that can be
+ * emulated without too many steps on the average 32-bit machine.
+ *
+ * For example, these two lines seem similar, and run equally fast on 64-bit:
+ *
+ *   xxh_u64 x;
+ *   x ^= (x >> 47); // good
+ *   x ^= (x >> 13); // bad
+ *
+ * However, to a 32-bit machine, there is a major difference.
+ *
+ * x ^= (x >> 47) looks like this:
+ *
+ *   x.lo ^= (x.hi >> (47 - 32));
+ *
+ * while x ^= (x >> 13) looks like this:
+ *
+ *   // note: funnel shifts are not usually cheap.
+ *   x.lo ^= (x.lo >> 13) | (x.hi << (32 - 13));
+ *   x.hi ^= (x.hi >> 13);
+ *
+ * The first one is significantly faster than the second, simply because the
+ * shift is larger than 32. This means:
+ *  - All the bits we need are in the upper 32 bits, so we can ignore the lower
+ *    32 bits in the shift.
+ *  - The shift result will always fit in the lower 32 bits, and therefore,
+ *    we can ignore the upper 32 bits in the xor.
+ *
+ * Thanks to this optimization, XXH3 only requires these features to be efficient:
+ *
+ *  - Usable unaligned access
+ *  - A 32-bit or 64-bit ALU
+ *      - If 32-bit, a decent ADC instruction
+ *  - A 32 or 64-bit multiply with a 64-bit result
+ *  - For the 128-bit variant, a decent byteswap helps short inputs.
+ *
+ * The first two are already required by XXH32, and almost all 32-bit and 64-bit
+ * platforms which can run XXH32 can run XXH3 efficiently.
+ *
+ * Thumb-1, the classic 16-bit only subset of ARM's instruction set, is one
+ * notable exception.
+ *
+ * First of all, Thumb-1 lacks support for the UMULL instruction which
+ * performs the important long multiply. This means numerous __aeabi_lmul
+ * calls.
+ *
+ * Second of all, the 8 functional registers are just not enough.
+ * Setup for __aeabi_lmul, byteshift loads, pointers, and all arithmetic need
+ * Lo registers, and this shuffling results in thousands more MOVs than A32.
+ *
+ * A32 and T32 don't have this limitation. They can access all 14 registers,
+ * do a 32->64 multiply with UMULL, and the flexible operand allowing free
+ * shifts is helpful, too.
+ *
+ * Therefore, we do a quick sanity check.
+ *
+ * If compiling Thumb-1 for a target which supports ARM instructions, we will
+ * emit a warning, as it is not a "sane" platform to compile for.
+ *
+ * Usually, if this happens, it is because of an accident and you probably need
+ * to specify -march, as you likely meant to compile for a newer architecture.
+ *
+ * Credit: large sections of the vectorial and asm source code paths
+ *         have been contributed by @easyaspi314
+ */
+#if defined(__thumb__) && !defined(__thumb2__) && defined(__ARM_ARCH_ISA_ARM)
+#   warning "XXH3 is highly inefficient without ARM or Thumb-2."
+#endif
+
+/* ==========================================
+ * Vectorization detection
+ * ========================================== */
+
+#ifdef XXH_DOXYGEN
 /*!
- * A polyfill for POWER9's vec_revb().
+ * @ingroup tuning
+ * @brief Overrides the vectorization implementation chosen for XXH3.
+ *
+ * Can be defined to 0 to disable SIMD or any of the values mentioned in
+ * @ref XXH_VECTOR_TYPE.
+ *
+ * If this is not defined, it uses predefined macros to determine the best
+ * implementation.
+ */
+#  define XXH_VECTOR XXH_SCALAR
+/*!
+ * @ingroup tuning
+ * @brief Possible values for @ref XXH_VECTOR.
+ *
+ * Note that these are actually implemented as macros.
+ *
+ * If this is not defined, it is detected automatically.
+ * internal macro XXH_X86DISPATCH overrides this.
+ */
+enum XXH_VECTOR_TYPE /* fake enum */ {
+    XXH_SCALAR = 0,  /*!< Portable scalar version */
+    XXH_SSE2   = 1,  /*!<
+                      * SSE2 for Pentium 4, Opteron, all x86_64.
+                      *
+                      * @note SSE2 is also guaranteed on Windows 10, macOS, and
+                      * Android x86.
+                      */
+    XXH_AVX2   = 2,  /*!< AVX2 for Haswell and Bulldozer */
+    XXH_AVX512 = 3,  /*!< AVX512 for Skylake and Icelake */
+    XXH_NEON   = 4,  /*!<
+                       * NEON for most ARMv7-A, all AArch64, and WASM SIMD128
+                       * via the SIMDeverywhere polyfill provided with the
+                       * Emscripten SDK.
+                       */
+    XXH_VSX    = 5,  /*!< VSX and ZVector for POWER8/z13 (64-bit) */
+    XXH_SVE    = 6,  /*!< SVE for some ARMv8-A and ARMv9-A */
+};
+/*!
+ * @ingroup tuning
+ * @brief Selects the minimum alignment for XXH3's accumulators.
+ *
+ * When using SIMD, this should match the alignment required for said vector
+ * type, so, for example, 32 for AVX2.
+ *
+ * Default: Auto detected.
  */
-XXH_FORCE_INLINE xxh_u64x2 XXH_vec_revb(xxh_u64x2 val) {
+#  define XXH_ACC_ALIGN 8
+#endif
 
-  xxh_u8x16 const vByteSwap = {0x07, 0x06, 0x05, 0x04, 0x03, 0x02, 0x01, 0x00,
-                               0x0F, 0x0E, 0x0D, 0x0C, 0x0B, 0x0A, 0x09, 0x08};
-  return vec_perm(val, val, vByteSwap);
+/* Actual definition */
+#ifndef XXH_DOXYGEN
+#  define XXH_SCALAR 0
+#  define XXH_SSE2   1
+#  define XXH_AVX2   2
+#  define XXH_AVX512 3
+#  define XXH_NEON   4
+#  define XXH_VSX    5
+#  define XXH_SVE    6
+#endif
 
-}
+#ifndef XXH_VECTOR    /* can be defined on command line */
+#  if defined(__ARM_FEATURE_SVE)
+#    define XXH_VECTOR XXH_SVE
+#  elif ( \
+        defined(__ARM_NEON__) || defined(__ARM_NEON) /* gcc */ \
+     || defined(_M_ARM) || defined(_M_ARM64) || defined(_M_ARM64EC) /* msvc */ \
+     || (defined(__wasm_simd128__) && XXH_HAS_INCLUDE(<arm_neon.h>)) /* wasm simd128 via SIMDe */ \
+   ) && ( \
+        defined(_WIN32) || defined(__LITTLE_ENDIAN__) /* little endian only */ \
+    || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
+   )
+#    define XXH_VECTOR XXH_NEON
+#  elif defined(__AVX512F__)
+#    define XXH_VECTOR XXH_AVX512
+#  elif defined(__AVX2__)
+#    define XXH_VECTOR XXH_AVX2
+#  elif defined(__SSE2__) || defined(_M_AMD64) || defined(_M_X64) || (defined(_M_IX86_FP) && (_M_IX86_FP == 2))
+#    define XXH_VECTOR XXH_SSE2
+#  elif (defined(__PPC64__) && defined(__POWER8_VECTOR__)) \
+     || (defined(__s390x__) && defined(__VEC__)) \
+     && defined(__GNUC__) /* TODO: IBM XL */
+#    define XXH_VECTOR XXH_VSX
+#  else
+#    define XXH_VECTOR XXH_SCALAR
+#  endif
+#endif
 
-          #endif
-        #endif                                                /* XXH_VSX_BE */
+/* __ARM_FEATURE_SVE is only supported by GCC & Clang. */
+#if (XXH_VECTOR == XXH_SVE) && !defined(__ARM_FEATURE_SVE)
+#  ifdef _MSC_VER
+#    pragma warning(once : 4606)
+#  else
+#    warning "__ARM_FEATURE_SVE isn't supported. Use SCALAR instead."
+#  endif
+#  undef XXH_VECTOR
+#  define XXH_VECTOR XXH_SCALAR
+#endif
 
-/*!
- * Performs an unaligned vector load and byte swaps it on big endian.
+/*
+ * Controls the alignment of the accumulator,
+ * for compatibility with aligned vector loads, which are usually faster.
  */
-XXH_FORCE_INLINE xxh_u64x2 XXH_vec_loadu(const void *ptr) {
+#ifndef XXH_ACC_ALIGN
+#  if defined(XXH_X86DISPATCH)
+#     define XXH_ACC_ALIGN 64  /* for compatibility with avx512 */
+#  elif XXH_VECTOR == XXH_SCALAR  /* scalar */
+#     define XXH_ACC_ALIGN 8
+#  elif XXH_VECTOR == XXH_SSE2  /* sse2 */
+#     define XXH_ACC_ALIGN 16
+#  elif XXH_VECTOR == XXH_AVX2  /* avx2 */
+#     define XXH_ACC_ALIGN 32
+#  elif XXH_VECTOR == XXH_NEON  /* neon */
+#     define XXH_ACC_ALIGN 16
+#  elif XXH_VECTOR == XXH_VSX   /* vsx */
+#     define XXH_ACC_ALIGN 16
+#  elif XXH_VECTOR == XXH_AVX512  /* avx512 */
+#     define XXH_ACC_ALIGN 64
+#  elif XXH_VECTOR == XXH_SVE   /* sve */
+#     define XXH_ACC_ALIGN 64
+#  endif
+#endif
 
-  xxh_u64x2 ret;
-  memcpy(&ret, ptr, sizeof(xxh_u64x2));
-        #if XXH_VSX_BE
-  ret = XXH_vec_revb(ret);
-        #endif
-  return ret;
+#if defined(XXH_X86DISPATCH) || XXH_VECTOR == XXH_SSE2 \
+    || XXH_VECTOR == XXH_AVX2 || XXH_VECTOR == XXH_AVX512
+#  define XXH_SEC_ALIGN XXH_ACC_ALIGN
+#elif XXH_VECTOR == XXH_SVE
+#  define XXH_SEC_ALIGN XXH_ACC_ALIGN
+#else
+#  define XXH_SEC_ALIGN 8
+#endif
 
-}
+#if defined(__GNUC__) || defined(__clang__)
+#  define XXH_ALIASING __attribute__((may_alias))
+#else
+#  define XXH_ALIASING /* nothing */
+#endif
 
-        /*
-         * vec_mulo and vec_mule are very problematic intrinsics on PowerPC
-         *
-         * These intrinsics weren't added until GCC 8, despite existing for a
-         * while, and they are endian dependent. Also, their meaning swap
-         * depending on version.
-         * */
-        #if defined(__s390x__)
-        /* s390x is always big endian, no issue on this platform */
-          #define XXH_vec_mulo vec_mulo
-          #define XXH_vec_mule vec_mule
-        #elif defined(__clang__) && XXH_HAS_BUILTIN(__builtin_altivec_vmuleuw)
-        /* Clang has a better way to control this, we can just use the builtin
-         * which doesn't swap. */
-          #define XXH_vec_mulo __builtin_altivec_vmulouw
-          #define XXH_vec_mule __builtin_altivec_vmuleuw
-        #else
-/* gcc needs inline assembly */
-/* Adapted from
- * https://github.com/google/highwayhash/blob/master/highwayhash/hh_vsx.h. */
-XXH_FORCE_INLINE xxh_u64x2 XXH_vec_mulo(xxh_u32x4 a, xxh_u32x4 b) {
+/*
+ * UGLY HACK:
+ * GCC usually generates the best code with -O3 for xxHash.
+ *
+ * However, when targeting AVX2, it is overzealous in its unrolling resulting
+ * in code roughly 3/4 the speed of Clang.
+ *
+ * There are other issues, such as GCC splitting _mm256_loadu_si256 into
+ * _mm_loadu_si128 + _mm256_inserti128_si256. This is an optimization which
+ * only applies to Sandy and Ivy Bridge... which don't even support AVX2.
+ *
+ * That is why when compiling the AVX2 version, it is recommended to use either
+ *   -O2 -mavx2 -march=haswell
+ * or
+ *   -O2 -mavx2 -mno-avx256-split-unaligned-load
+ * for decent performance, or to use Clang instead.
+ *
+ * Fortunately, we can control the first one with a pragma that forces GCC into
+ * -O2, but the other one we can't control without "failed to inline always
+ * inline function due to target mismatch" warnings.
+ */
+#if XXH_VECTOR == XXH_AVX2 /* AVX2 */ \
+  && defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
+  && defined(__OPTIMIZE__) && XXH_SIZE_OPT <= 0 /* respect -O0 and -Os */
+#  pragma GCC push_options
+#  pragma GCC optimize("-O2")
+#endif
+
+#if XXH_VECTOR == XXH_NEON
+
+/*
+ * UGLY HACK: While AArch64 GCC on Linux does not seem to care, on macOS, GCC -O3
+ * optimizes out the entire hashLong loop because of the aliasing violation.
+ *
+ * However, GCC is also inefficient at load-store optimization with vld1q/vst1q,
+ * so the only option is to mark it as aliasing.
+ */
+typedef uint64x2_t xxh_aliasing_uint64x2_t XXH_ALIASING;
 
-  xxh_u64x2 result;
-  __asm__("vmulouw %0, %1, %2" : "=v"(result) : "v"(a), "v"(b));
-  return result;
+/*!
+ * @internal
+ * @brief `vld1q_u64` but faster and alignment-safe.
+ *
+ * On AArch64, unaligned access is always safe, but on ARMv7-a, it is only
+ * *conditionally* safe (`vld1` has an alignment bit like `movdq[ua]` in x86).
+ *
+ * GCC for AArch64 sees `vld1q_u8` as an intrinsic instead of a load, so it
+ * prohibits load-store optimizations. Therefore, a direct dereference is used.
+ *
+ * Otherwise, `vld1q_u8` is used with `vreinterpretq_u8_u64` to do a safe
+ * unaligned load.
+ */
+#if defined(__aarch64__) && defined(__GNUC__) && !defined(__clang__)
+XXH_FORCE_INLINE uint64x2_t XXH_vld1q_u64(void const* ptr) /* silence -Wcast-align */
+{
+    return *(xxh_aliasing_uint64x2_t const *)ptr;
+}
+#else
+XXH_FORCE_INLINE uint64x2_t XXH_vld1q_u64(void const* ptr)
+{
+    return vreinterpretq_u64_u8(vld1q_u8((uint8_t const*)ptr));
+}
+#endif
 
+/*!
+ * @internal
+ * @brief `vmlal_u32` on low and high halves of a vector.
+ *
+ * This is a workaround for AArch64 GCC < 11 which implemented arm_neon.h with
+ * inline assembly and were therefore incapable of merging the `vget_{low, high}_u32`
+ * with `vmlal_u32`.
+ */
+#if defined(__aarch64__) && defined(__GNUC__) && !defined(__clang__) && __GNUC__ < 11
+XXH_FORCE_INLINE uint64x2_t
+XXH_vmlal_low_u32(uint64x2_t acc, uint32x4_t lhs, uint32x4_t rhs)
+{
+    /* Inline assembly is the only way */
+    __asm__("umlal   %0.2d, %1.2s, %2.2s" : "+w" (acc) : "w" (lhs), "w" (rhs));
+    return acc;
+}
+XXH_FORCE_INLINE uint64x2_t
+XXH_vmlal_high_u32(uint64x2_t acc, uint32x4_t lhs, uint32x4_t rhs)
+{
+    /* This intrinsic works as expected */
+    return vmlal_high_u32(acc, lhs, rhs);
+}
+#else
+/* Portable intrinsic versions */
+XXH_FORCE_INLINE uint64x2_t
+XXH_vmlal_low_u32(uint64x2_t acc, uint32x4_t lhs, uint32x4_t rhs)
+{
+    return vmlal_u32(acc, vget_low_u32(lhs), vget_low_u32(rhs));
+}
+/*! @copydoc XXH_vmlal_low_u32
+ * Assume the compiler converts this to vmlal_high_u32 on aarch64 */
+XXH_FORCE_INLINE uint64x2_t
+XXH_vmlal_high_u32(uint64x2_t acc, uint32x4_t lhs, uint32x4_t rhs)
+{
+    return vmlal_u32(acc, vget_high_u32(lhs), vget_high_u32(rhs));
 }
+#endif
+
+/*!
+ * @ingroup tuning
+ * @brief Controls the NEON to scalar ratio for XXH3
+ *
+ * This can be set to 2, 4, 6, or 8.
+ *
+ * ARM Cortex CPUs are _very_ sensitive to how their pipelines are used.
+ *
+ * For example, the Cortex-A73 can dispatch 3 micro-ops per cycle, but only 2 of those
+ * can be NEON. If you are only using NEON instructions, you are only using 2/3 of the CPU
+ * bandwidth.
+ *
+ * This is even more noticeable on the more advanced cores like the Cortex-A76 which
+ * can dispatch 8 micro-ops per cycle, but still only 2 NEON micro-ops at once.
+ *
+ * Therefore, to make the most out of the pipeline, it is beneficial to run 6 NEON lanes
+ * and 2 scalar lanes, which is chosen by default.
+ *
+ * This does not apply to Apple processors or 32-bit processors, which run better with
+ * full NEON. These will default to 8. Additionally, size-optimized builds run 8 lanes.
+ *
+ * This change benefits CPUs with large micro-op buffers without negatively affecting
+ * most other CPUs:
+ *
+ *  | Chipset               | Dispatch type       | NEON only | 6:2 hybrid | Diff. |
+ *  |:----------------------|:--------------------|----------:|-----------:|------:|
+ *  | Snapdragon 730 (A76)  | 2 NEON/8 micro-ops  |  8.8 GB/s |  10.1 GB/s |  ~16% |
+ *  | Snapdragon 835 (A73)  | 2 NEON/3 micro-ops  |  5.1 GB/s |   5.3 GB/s |   ~5% |
+ *  | Marvell PXA1928 (A53) | In-order dual-issue |  1.9 GB/s |   1.9 GB/s |    0% |
+ *  | Apple M1              | 4 NEON/8 micro-ops  | 37.3 GB/s |  36.1 GB/s |  ~-3% |
+ *
+ * It also seems to fix some bad codegen on GCC, making it almost as fast as clang.
+ *
+ * When using WASM SIMD128, if this is 2 or 6, SIMDe will scalarize 2 of the lanes meaning
+ * it effectively becomes worse 4.
+ *
+ * @see XXH3_accumulate_512_neon()
+ */
+# ifndef XXH3_NEON_LANES
+#  if (defined(__aarch64__) || defined(__arm64__) || defined(_M_ARM64) || defined(_M_ARM64EC)) \
+   && !defined(__APPLE__) && XXH_SIZE_OPT <= 0
+#   define XXH3_NEON_LANES 6
+#  else
+#   define XXH3_NEON_LANES XXH_ACC_NB
+#  endif
+# endif
+#endif  /* XXH_VECTOR == XXH_NEON */
+
+/*
+ * VSX and Z Vector helpers.
+ *
+ * This is very messy, and any pull requests to clean this up are welcome.
+ *
+ * There are a lot of problems with supporting VSX and s390x, due to
+ * inconsistent intrinsics, spotty coverage, and multiple endiannesses.
+ */
+#if XXH_VECTOR == XXH_VSX
+/* Annoyingly, these headers _may_ define three macros: `bool`, `vector`,
+ * and `pixel`. This is a problem for obvious reasons.
+ *
+ * These keywords are unnecessary; the spec literally says they are
+ * equivalent to `__bool`, `__vector`, and `__pixel` and may be undef'd
+ * after including the header.
+ *
+ * We use pragma push_macro/pop_macro to keep the namespace clean. */
+#  pragma push_macro("bool")
+#  pragma push_macro("vector")
+#  pragma push_macro("pixel")
+/* silence potential macro redefined warnings */
+#  undef bool
+#  undef vector
+#  undef pixel
+
+#  if defined(__s390x__)
+#    include <s390intrin.h>
+#  else
+#    include <altivec.h>
+#  endif
+
+/* Restore the original macro values, if applicable. */
+#  pragma pop_macro("pixel")
+#  pragma pop_macro("vector")
+#  pragma pop_macro("bool")
 
-XXH_FORCE_INLINE xxh_u64x2 XXH_vec_mule(xxh_u32x4 a, xxh_u32x4 b) {
+typedef __vector unsigned long long xxh_u64x2;
+typedef __vector unsigned char xxh_u8x16;
+typedef __vector unsigned xxh_u32x4;
 
-  xxh_u64x2 result;
-  __asm__("vmuleuw %0, %1, %2" : "=v"(result) : "v"(a), "v"(b));
-  return result;
+/*
+ * UGLY HACK: Similar to aarch64 macOS GCC, s390x GCC has the same aliasing issue.
+ */
+typedef xxh_u64x2 xxh_aliasing_u64x2 XXH_ALIASING;
+
+# ifndef XXH_VSX_BE
+#  if defined(__BIG_ENDIAN__) \
+  || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
+#    define XXH_VSX_BE 1
+#  elif defined(__VEC_ELEMENT_REG_ORDER__) && __VEC_ELEMENT_REG_ORDER__ == __ORDER_BIG_ENDIAN__
+#    warning "-maltivec=be is not recommended. Please use native endianness."
+#    define XXH_VSX_BE 1
+#  else
+#    define XXH_VSX_BE 0
+#  endif
+# endif /* !defined(XXH_VSX_BE) */
+
+# if XXH_VSX_BE
+#  if defined(__POWER9_VECTOR__) || (defined(__clang__) && defined(__s390x__))
+#    define XXH_vec_revb vec_revb
+#  else
+/*!
+ * A polyfill for POWER9's vec_revb().
+ */
+XXH_FORCE_INLINE xxh_u64x2 XXH_vec_revb(xxh_u64x2 val)
+{
+    xxh_u8x16 const vByteSwap = { 0x07, 0x06, 0x05, 0x04, 0x03, 0x02, 0x01, 0x00,
+                                  0x0F, 0x0E, 0x0D, 0x0C, 0x0B, 0x0A, 0x09, 0x08 };
+    return vec_perm(val, val, vByteSwap);
+}
+#  endif
+# endif /* XXH_VSX_BE */
 
+/*!
+ * Performs an unaligned vector load and byte swaps it on big endian.
+ */
+XXH_FORCE_INLINE xxh_u64x2 XXH_vec_loadu(const void *ptr)
+{
+    xxh_u64x2 ret;
+    XXH_memcpy(&ret, ptr, sizeof(xxh_u64x2));
+# if XXH_VSX_BE
+    ret = XXH_vec_revb(ret);
+# endif
+    return ret;
 }
 
-        #endif                                /* XXH_vec_mulo, XXH_vec_mule */
-      #endif                                       /* XXH_VECTOR == XXH_VSX */
+/*
+ * vec_mulo and vec_mule are very problematic intrinsics on PowerPC
+ *
+ * These intrinsics weren't added until GCC 8, despite existing for a while,
+ * and they are endian dependent. Also, their meaning swap depending on version.
+ * */
+# if defined(__s390x__)
+ /* s390x is always big endian, no issue on this platform */
+#  define XXH_vec_mulo vec_mulo
+#  define XXH_vec_mule vec_mule
+# elif defined(__clang__) && XXH_HAS_BUILTIN(__builtin_altivec_vmuleuw) && !defined(__ibmxl__)
+/* Clang has a better way to control this, we can just use the builtin which doesn't swap. */
+ /* The IBM XL Compiler (which defined __clang__) only implements the vec_* operations */
+#  define XXH_vec_mulo __builtin_altivec_vmulouw
+#  define XXH_vec_mule __builtin_altivec_vmuleuw
+# else
+/* gcc needs inline assembly */
+/* Adapted from https://github.com/google/highwayhash/blob/master/highwayhash/hh_vsx.h. */
+XXH_FORCE_INLINE xxh_u64x2 XXH_vec_mulo(xxh_u32x4 a, xxh_u32x4 b)
+{
+    xxh_u64x2 result;
+    __asm__("vmulouw %0, %1, %2" : "=v" (result) : "v" (a), "v" (b));
+    return result;
+}
+XXH_FORCE_INLINE xxh_u64x2 XXH_vec_mule(xxh_u32x4 a, xxh_u32x4 b)
+{
+    xxh_u64x2 result;
+    __asm__("vmuleuw %0, %1, %2" : "=v" (result) : "v" (a), "v" (b));
+    return result;
+}
+# endif /* XXH_vec_mulo, XXH_vec_mule */
+#endif /* XXH_VECTOR == XXH_VSX */
+
+#if XXH_VECTOR == XXH_SVE
+#define ACCRND(acc, offset) \
+do { \
+    svuint64_t input_vec = svld1_u64(mask, xinput + offset);         \
+    svuint64_t secret_vec = svld1_u64(mask, xsecret + offset);       \
+    svuint64_t mixed = sveor_u64_x(mask, secret_vec, input_vec);     \
+    svuint64_t swapped = svtbl_u64(input_vec, kSwap);                \
+    svuint64_t mixed_lo = svextw_u64_x(mask, mixed);                 \
+    svuint64_t mixed_hi = svlsr_n_u64_x(mask, mixed, 32);            \
+    svuint64_t mul = svmad_u64_x(mask, mixed_lo, mixed_hi, swapped); \
+    acc = svadd_u64_x(mask, acc, mul);                               \
+} while (0)
+#endif /* XXH_VECTOR == XXH_SVE */
+
+/* prefetch
+ * can be disabled, by declaring XXH_NO_PREFETCH build macro */
+#if defined(XXH_NO_PREFETCH)
+#  define XXH_PREFETCH(ptr)  (void)(ptr)  /* disabled */
+#else
+#  if XXH_SIZE_OPT >= 1
+#    define XXH_PREFETCH(ptr) (void)(ptr)
+#  elif defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86))  /* _mm_prefetch() not defined outside of x86/x64 */
+#    include <mmintrin.h>   /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
+#    define XXH_PREFETCH(ptr)  _mm_prefetch((const char*)(ptr), _MM_HINT_T0)
+#  elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) )
+#    define XXH_PREFETCH(ptr)  __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */)
+#  else
+#    define XXH_PREFETCH(ptr) (void)(ptr)  /* disabled */
+#  endif
+#endif  /* XXH_NO_PREFETCH */
 
-      /* prefetch
-       * can be disabled, by declaring XXH_NO_PREFETCH build macro */
-      #if defined(XXH_NO_PREFETCH)
-        #define XXH_PREFETCH(ptr) (void)(ptr)                   /* disabled */
-      #else
-        #if defined(_MSC_VER) && \
-            (defined(_M_X64) ||  \
-             defined(            \
-                 _M_IX86)) /* _mm_prefetch() not defined outside of x86/x64 */
-          #include <mmintrin.h> /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
-          #define XXH_PREFETCH(ptr) \
-            _mm_prefetch((const char *)(ptr), _MM_HINT_T0)
-        #elif defined(__GNUC__) && \
-            ((__GNUC__ >= 4) || ((__GNUC__ == 3) && (__GNUC_MINOR__ >= 1)))
-          #define XXH_PREFETCH(ptr) \
-            __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */)
-        #else
-          #define XXH_PREFETCH(ptr) (void)(ptr)                 /* disabled */
-        #endif
-      #endif                                             /* XXH_NO_PREFETCH */
 
-    /* ==========================================
-     * XXH3 default settings
-     * ========================================== */
+/* ==========================================
+ * XXH3 default settings
+ * ========================================== */
 
-      #define XXH_SECRET_DEFAULT_SIZE 192   /* minimum XXH3_SECRET_SIZE_MIN */
+#define XXH_SECRET_DEFAULT_SIZE 192   /* minimum XXH3_SECRET_SIZE_MIN */
 
-      #if (XXH_SECRET_DEFAULT_SIZE < XXH3_SECRET_SIZE_MIN)
-        #error "default keyset is not large enough"
-      #endif
+#if (XXH_SECRET_DEFAULT_SIZE < XXH3_SECRET_SIZE_MIN)
+#  error "default keyset is not large enough"
+#endif
 
 /*! Pseudorandom secret taken directly from FARSH. */
-XXH_ALIGN(64)
-static const xxh_u8 XXH3_kSecret[XXH_SECRET_DEFAULT_SIZE] = {
-
-    0xb8, 0xfe, 0x6c, 0x39, 0x23, 0xa4, 0x4b, 0xbe, 0x7c, 0x01, 0x81, 0x2c,
-    0xf7, 0x21, 0xad, 0x1c, 0xde, 0xd4, 0x6d, 0xe9, 0x83, 0x90, 0x97, 0xdb,
-    0x72, 0x40, 0xa4, 0xa4, 0xb7, 0xb3, 0x67, 0x1f, 0xcb, 0x79, 0xe6, 0x4e,
-    0xcc, 0xc0, 0xe5, 0x78, 0x82, 0x5a, 0xd0, 0x7d, 0xcc, 0xff, 0x72, 0x21,
-    0xb8, 0x08, 0x46, 0x74, 0xf7, 0x43, 0x24, 0x8e, 0xe0, 0x35, 0x90, 0xe6,
-    0x81, 0x3a, 0x26, 0x4c, 0x3c, 0x28, 0x52, 0xbb, 0x91, 0xc3, 0x00, 0xcb,
-    0x88, 0xd0, 0x65, 0x8b, 0x1b, 0x53, 0x2e, 0xa3, 0x71, 0x64, 0x48, 0x97,
-    0xa2, 0x0d, 0xf9, 0x4e, 0x38, 0x19, 0xef, 0x46, 0xa9, 0xde, 0xac, 0xd8,
-    0xa8, 0xfa, 0x76, 0x3f, 0xe3, 0x9c, 0x34, 0x3f, 0xf9, 0xdc, 0xbb, 0xc7,
-    0xc7, 0x0b, 0x4f, 0x1d, 0x8a, 0x51, 0xe0, 0x4b, 0xcd, 0xb4, 0x59, 0x31,
-    0xc8, 0x9f, 0x7e, 0xc9, 0xd9, 0x78, 0x73, 0x64, 0xea, 0xc5, 0xac, 0x83,
-    0x34, 0xd3, 0xeb, 0xc3, 0xc5, 0x81, 0xa0, 0xff, 0xfa, 0x13, 0x63, 0xeb,
-    0x17, 0x0d, 0xdd, 0x51, 0xb7, 0xf0, 0xda, 0x49, 0xd3, 0x16, 0x55, 0x26,
-    0x29, 0xd4, 0x68, 0x9e, 0x2b, 0x16, 0xbe, 0x58, 0x7d, 0x47, 0xa1, 0xfc,
-    0x8f, 0xf8, 0xb8, 0xd1, 0x7a, 0xd0, 0x31, 0xce, 0x45, 0xcb, 0x3a, 0x8f,
-    0x95, 0x16, 0x04, 0x28, 0xaf, 0xd7, 0xfb, 0xca, 0xbb, 0x4b, 0x40, 0x7e,
-
+XXH_ALIGN(64) static const xxh_u8 XXH3_kSecret[XXH_SECRET_DEFAULT_SIZE] = {
+    0xb8, 0xfe, 0x6c, 0x39, 0x23, 0xa4, 0x4b, 0xbe, 0x7c, 0x01, 0x81, 0x2c, 0xf7, 0x21, 0xad, 0x1c,
+    0xde, 0xd4, 0x6d, 0xe9, 0x83, 0x90, 0x97, 0xdb, 0x72, 0x40, 0xa4, 0xa4, 0xb7, 0xb3, 0x67, 0x1f,
+    0xcb, 0x79, 0xe6, 0x4e, 0xcc, 0xc0, 0xe5, 0x78, 0x82, 0x5a, 0xd0, 0x7d, 0xcc, 0xff, 0x72, 0x21,
+    0xb8, 0x08, 0x46, 0x74, 0xf7, 0x43, 0x24, 0x8e, 0xe0, 0x35, 0x90, 0xe6, 0x81, 0x3a, 0x26, 0x4c,
+    0x3c, 0x28, 0x52, 0xbb, 0x91, 0xc3, 0x00, 0xcb, 0x88, 0xd0, 0x65, 0x8b, 0x1b, 0x53, 0x2e, 0xa3,
+    0x71, 0x64, 0x48, 0x97, 0xa2, 0x0d, 0xf9, 0x4e, 0x38, 0x19, 0xef, 0x46, 0xa9, 0xde, 0xac, 0xd8,
+    0xa8, 0xfa, 0x76, 0x3f, 0xe3, 0x9c, 0x34, 0x3f, 0xf9, 0xdc, 0xbb, 0xc7, 0xc7, 0x0b, 0x4f, 0x1d,
+    0x8a, 0x51, 0xe0, 0x4b, 0xcd, 0xb4, 0x59, 0x31, 0xc8, 0x9f, 0x7e, 0xc9, 0xd9, 0x78, 0x73, 0x64,
+    0xea, 0xc5, 0xac, 0x83, 0x34, 0xd3, 0xeb, 0xc3, 0xc5, 0x81, 0xa0, 0xff, 0xfa, 0x13, 0x63, 0xeb,
+    0x17, 0x0d, 0xdd, 0x51, 0xb7, 0xf0, 0xda, 0x49, 0xd3, 0x16, 0x55, 0x26, 0x29, 0xd4, 0x68, 0x9e,
+    0x2b, 0x16, 0xbe, 0x58, 0x7d, 0x47, 0xa1, 0xfc, 0x8f, 0xf8, 0xb8, 0xd1, 0x7a, 0xd0, 0x31, 0xce,
+    0x45, 0xcb, 0x3a, 0x8f, 0x95, 0x16, 0x04, 0x28, 0xaf, 0xd7, 0xfb, 0xca, 0xbb, 0x4b, 0x40, 0x7e,
 };
 
-      #ifdef XXH_OLD_NAMES
-        #define kSecret XXH3_kSecret
-      #endif
+static const xxh_u64 PRIME_MX1 = 0x165667919E3779F9ULL;  /*!< 0b0001011001010110011001111001000110011110001101110111100111111001 */
+static const xxh_u64 PRIME_MX2 = 0x9FB21C651E98DF25ULL;  /*!< 0b1001111110110010000111000110010100011110100110001101111100100101 */
+
+#ifdef XXH_OLD_NAMES
+#  define kSecret XXH3_kSecret
+#endif
 
-      #ifdef XXH_DOXYGEN
+#ifdef XXH_DOXYGEN
 /*!
  * @brief Calculates a 32-bit to 64-bit long multiply.
  *
  * Implemented as a macro.
  *
- * Wraps `__emulu` on MSVC x86 because it tends to call `__allmul` when it
- * doesn't need to (but it shouldn't need to anyways, it is about 7 instructions
- * to do a 64x64 multiply...). Since we know that this will _always_ emit
- * `MULL`, we use that instead of the normal method.
+ * Wraps `__emulu` on MSVC x86 because it tends to call `__allmul` when it doesn't
+ * need to (but it shouldn't need to anyways, it is about 7 instructions to do
+ * a 64x64 multiply...). Since we know that this will _always_ emit `MULL`, we
+ * use that instead of the normal method.
  *
- * If you are compiling for platforms like Thumb-1 and don't have a better
- * option, you may also want to write your own long multiply routine here.
+ * If you are compiling for platforms like Thumb-1 and don't have a better option,
+ * you may also want to write your own long multiply routine here.
  *
  * @param x, y Numbers to be multiplied
  * @return 64-bit product of the low 32 bits of @p x and @p y.
  */
-XXH_FORCE_INLINE xxh_u64 XXH_mult32to64(xxh_u64 x, xxh_u64 y) {
-
-  return (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF);
-
-}
-
-      #elif defined(_MSC_VER) && defined(_M_IX86)
-        #include <intrin.h>
-        #define XXH_mult32to64(x, y) __emulu((unsigned)(x), (unsigned)(y))
-      #else
-      /*
-       * Downcast + upcast is usually better than masking on older compilers
-       * like GCC 4.2 (especially 32-bit ones), all without affecting newer
-       * compilers.
-       *
-       * The other method, (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF), will AND both
-       * operands and perform a full 64x64 multiply -- entirely redundant on
-       * 32-bit.
-       */
-        #define XXH_mult32to64(x, y) \
-          ((xxh_u64)(xxh_u32)(x) * (xxh_u64)(xxh_u32)(y))
-      #endif
+XXH_FORCE_INLINE xxh_u64
+XXH_mult32to64(xxh_u64 x, xxh_u64 y)
+{
+   return (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF);
+}
+#elif defined(_MSC_VER) && defined(_M_IX86)
+#    define XXH_mult32to64(x, y) __emulu((unsigned)(x), (unsigned)(y))
+#else
+/*
+ * Downcast + upcast is usually better than masking on older compilers like
+ * GCC 4.2 (especially 32-bit ones), all without affecting newer compilers.
+ *
+ * The other method, (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF), will AND both operands
+ * and perform a full 64x64 multiply -- entirely redundant on 32-bit.
+ */
+#    define XXH_mult32to64(x, y) ((xxh_u64)(xxh_u32)(x) * (xxh_u64)(xxh_u32)(y))
+#endif
 
 /*!
  * @brief Calculates a 64->128-bit long multiply.
@@ -3623,157 +4299,167 @@ XXH_FORCE_INLINE xxh_u64 XXH_mult32to64(xxh_u64 x, xxh_u64 y) {
  * Uses `__uint128_t` and `_umul128` if available, otherwise uses a scalar
  * version.
  *
- * @param lhs, rhs The 64-bit integers to be multiplied
+ * @param lhs , rhs The 64-bit integers to be multiplied
  * @return The 128-bit result represented in an @ref XXH128_hash_t.
  */
-static XXH128_hash_t XXH_mult64to128(xxh_u64 lhs, xxh_u64 rhs) {
-
-      /*
-       * GCC/Clang __uint128_t method.
-       *
-       * On most 64-bit targets, GCC and Clang define a __uint128_t type.
-       * This is usually the best way as it usually uses a native long 64-bit
-       * multiply, such as MULQ on x86_64 or MUL + UMULH on aarch64.
-       *
-       * Usually.
-       *
-       * Despite being a 32-bit platform, Clang (and emscripten) define this
-       * type despite not having the arithmetic for it. This results in a laggy
-       * compiler builtin call which calculates a full 128-bit multiply.
-       * In that case it is best to use the portable one.
-       * https://github.com/Cyan4973/xxHash/issues/211#issuecomment-515575677
-       */
-      #if defined(__GNUC__) && !defined(__wasm__) && \
-              defined(__SIZEOF_INT128__) ||          \
-          (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
-
-  __uint128_t const product = (__uint128_t)lhs * (__uint128_t)rhs;
-  XXH128_hash_t     r128;
-  r128.low64 = (xxh_u64)(product);
-  r128.high64 = (xxh_u64)(product >> 64);
-  return r128;
+static XXH128_hash_t
+XXH_mult64to128(xxh_u64 lhs, xxh_u64 rhs)
+{
+    /*
+     * GCC/Clang __uint128_t method.
+     *
+     * On most 64-bit targets, GCC and Clang define a __uint128_t type.
+     * This is usually the best way as it usually uses a native long 64-bit
+     * multiply, such as MULQ on x86_64 or MUL + UMULH on aarch64.
+     *
+     * Usually.
+     *
+     * Despite being a 32-bit platform, Clang (and emscripten) define this type
+     * despite not having the arithmetic for it. This results in a laggy
+     * compiler builtin call which calculates a full 128-bit multiply.
+     * In that case it is best to use the portable one.
+     * https://github.com/Cyan4973/xxHash/issues/211#issuecomment-515575677
+     */
+#if (defined(__GNUC__) || defined(__clang__)) && !defined(__wasm__) \
+    && defined(__SIZEOF_INT128__) \
+    || (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
 
-        /*
-         * MSVC for x64's _umul128 method.
-         *
-         * xxh_u64 _umul128(xxh_u64 Multiplier, xxh_u64 Multiplicand, xxh_u64
-         * *HighProduct);
-         *
-         * This compiles to single operand MUL on x64.
-         */
-      #elif defined(_M_X64) || defined(_M_IA64)
-
-        #ifndef _MSC_VER
-          #pragma intrinsic(_umul128)
-        #endif
-  xxh_u64       product_high;
-  xxh_u64 const product_low = _umul128(lhs, rhs, &product_high);
-  XXH128_hash_t r128;
-  r128.low64 = product_low;
-  r128.high64 = product_high;
-  return r128;
-
-      #else
-  /*
-   * Portable scalar method. Optimized for 32-bit and 64-bit ALUs.
-   *
-   * This is a fast and simple grade school multiply, which is shown below
-   * with base 10 arithmetic instead of base 0x100000000.
-   *
-   *           9 3 // D2 lhs = 93
-   *         x 7 5 // D2 rhs = 75
-   *     ----------
-   *           1 5 // D2 lo_lo = (93 % 10) * (75 % 10) = 15
-   *         4 5 | // D2 hi_lo = (93 / 10) * (75 % 10) = 45
-   *         2 1 | // D2 lo_hi = (93 % 10) * (75 / 10) = 21
-   *     + 6 3 | | // D2 hi_hi = (93 / 10) * (75 / 10) = 63
-   *     ---------
-   *         2 7 | // D2 cross = (15 / 10) + (45 % 10) + 21 = 27
-   *     + 6 7 | | // D2 upper = (27 / 10) + (45 / 10) + 63 = 67
-   *     ---------
-   *       6 9 7 5 // D4 res = (27 * 10) + (15 % 10) + (67 * 100) = 6975
-   *
-   * The reasons for adding the products like this are:
-   *  1. It avoids manual carry tracking. Just like how
-   *     (9 * 9) + 9 + 9 = 99, the same applies with this for UINT64_MAX.
-   *     This avoids a lot of complexity.
-   *
-   *  2. It hints for, and on Clang, compiles to, the powerful UMAAL
-   *     instruction available in ARM's Digital Signal Processing extension
-   *     in 32-bit ARMv6 and later, which is shown below:
-   *
-   *         void UMAAL(xxh_u32 *RdLo, xxh_u32 *RdHi, xxh_u32 Rn, xxh_u32 Rm)
-   *         {
-
-   *             xxh_u64 product = (xxh_u64)*RdLo * (xxh_u64)*RdHi + Rn + Rm;
-   *             *RdLo = (xxh_u32)(product & 0xFFFFFFFF);
-   *             *RdHi = (xxh_u32)(product >> 32);
-   *         }
-   *
-   *     This instruction was designed for efficient long multiplication, and
-   *     allows this to be calculated in only 4 instructions at speeds
-   *     comparable to some 64-bit ALUs.
-   *
-   *  3. It isn't terrible on other platforms. Usually this will be a couple
-   *     of 32-bit ADD/ADCs.
-   */
-
-  /* First calculate all of the cross products. */
-  xxh_u64 const lo_lo = XXH_mult32to64(lhs & 0xFFFFFFFF, rhs & 0xFFFFFFFF);
-  xxh_u64 const hi_lo = XXH_mult32to64(lhs >> 32, rhs & 0xFFFFFFFF);
-  xxh_u64 const lo_hi = XXH_mult32to64(lhs & 0xFFFFFFFF, rhs >> 32);
-  xxh_u64 const hi_hi = XXH_mult32to64(lhs >> 32, rhs >> 32);
-
-  /* Now add the products together. These will never overflow. */
-  xxh_u64 const cross = (lo_lo >> 32) + (hi_lo & 0xFFFFFFFF) + lo_hi;
-  xxh_u64 const upper = (hi_lo >> 32) + (cross >> 32) + hi_hi;
-  xxh_u64 const lower = (cross << 32) | (lo_lo & 0xFFFFFFFF);
-
-  XXH128_hash_t r128;
-  r128.low64 = lower;
-  r128.high64 = upper;
-  return r128;
-      #endif
+    __uint128_t const product = (__uint128_t)lhs * (__uint128_t)rhs;
+    XXH128_hash_t r128;
+    r128.low64  = (xxh_u64)(product);
+    r128.high64 = (xxh_u64)(product >> 64);
+    return r128;
+
+    /*
+     * MSVC for x64's _umul128 method.
+     *
+     * xxh_u64 _umul128(xxh_u64 Multiplier, xxh_u64 Multiplicand, xxh_u64 *HighProduct);
+     *
+     * This compiles to single operand MUL on x64.
+     */
+#elif (defined(_M_X64) || defined(_M_IA64)) && !defined(_M_ARM64EC)
+
+#ifndef _MSC_VER
+#   pragma intrinsic(_umul128)
+#endif
+    xxh_u64 product_high;
+    xxh_u64 const product_low = _umul128(lhs, rhs, &product_high);
+    XXH128_hash_t r128;
+    r128.low64  = product_low;
+    r128.high64 = product_high;
+    return r128;
+
+    /*
+     * MSVC for ARM64's __umulh method.
+     *
+     * This compiles to the same MUL + UMULH as GCC/Clang's __uint128_t method.
+     */
+#elif defined(_M_ARM64) || defined(_M_ARM64EC)
+
+#ifndef _MSC_VER
+#   pragma intrinsic(__umulh)
+#endif
+    XXH128_hash_t r128;
+    r128.low64  = lhs * rhs;
+    r128.high64 = __umulh(lhs, rhs);
+    return r128;
+
+#else
+    /*
+     * Portable scalar method. Optimized for 32-bit and 64-bit ALUs.
+     *
+     * This is a fast and simple grade school multiply, which is shown below
+     * with base 10 arithmetic instead of base 0x100000000.
+     *
+     *           9 3 // D2 lhs = 93
+     *         x 7 5 // D2 rhs = 75
+     *     ----------
+     *           1 5 // D2 lo_lo = (93 % 10) * (75 % 10) = 15
+     *         4 5 | // D2 hi_lo = (93 / 10) * (75 % 10) = 45
+     *         2 1 | // D2 lo_hi = (93 % 10) * (75 / 10) = 21
+     *     + 6 3 | | // D2 hi_hi = (93 / 10) * (75 / 10) = 63
+     *     ---------
+     *         2 7 | // D2 cross = (15 / 10) + (45 % 10) + 21 = 27
+     *     + 6 7 | | // D2 upper = (27 / 10) + (45 / 10) + 63 = 67
+     *     ---------
+     *       6 9 7 5 // D4 res = (27 * 10) + (15 % 10) + (67 * 100) = 6975
+     *
+     * The reasons for adding the products like this are:
+     *  1. It avoids manual carry tracking. Just like how
+     *     (9 * 9) + 9 + 9 = 99, the same applies with this for UINT64_MAX.
+     *     This avoids a lot of complexity.
+     *
+     *  2. It hints for, and on Clang, compiles to, the powerful UMAAL
+     *     instruction available in ARM's Digital Signal Processing extension
+     *     in 32-bit ARMv6 and later, which is shown below:
+     *
+     *         void UMAAL(xxh_u32 *RdLo, xxh_u32 *RdHi, xxh_u32 Rn, xxh_u32 Rm)
+     *         {
+     *             xxh_u64 product = (xxh_u64)*RdLo * (xxh_u64)*RdHi + Rn + Rm;
+     *             *RdLo = (xxh_u32)(product & 0xFFFFFFFF);
+     *             *RdHi = (xxh_u32)(product >> 32);
+     *         }
+     *
+     *     This instruction was designed for efficient long multiplication, and
+     *     allows this to be calculated in only 4 instructions at speeds
+     *     comparable to some 64-bit ALUs.
+     *
+     *  3. It isn't terrible on other platforms. Usually this will be a couple
+     *     of 32-bit ADD/ADCs.
+     */
 
+    /* First calculate all of the cross products. */
+    xxh_u64 const lo_lo = XXH_mult32to64(lhs & 0xFFFFFFFF, rhs & 0xFFFFFFFF);
+    xxh_u64 const hi_lo = XXH_mult32to64(lhs >> 32,        rhs & 0xFFFFFFFF);
+    xxh_u64 const lo_hi = XXH_mult32to64(lhs & 0xFFFFFFFF, rhs >> 32);
+    xxh_u64 const hi_hi = XXH_mult32to64(lhs >> 32,        rhs >> 32);
+
+    /* Now add the products together. These will never overflow. */
+    xxh_u64 const cross = (lo_lo >> 32) + (hi_lo & 0xFFFFFFFF) + lo_hi;
+    xxh_u64 const upper = (hi_lo >> 32) + (cross >> 32)        + hi_hi;
+    xxh_u64 const lower = (cross << 32) | (lo_lo & 0xFFFFFFFF);
+
+    XXH128_hash_t r128;
+    r128.low64  = lower;
+    r128.high64 = upper;
+    return r128;
+#endif
 }
 
 /*!
  * @brief Calculates a 64-bit to 128-bit multiply, then XOR folds it.
  *
  * The reason for the separate function is to prevent passing too many structs
- * around by value. This will hopefully inline the multiply, but we don't force
- * it.
+ * around by value. This will hopefully inline the multiply, but we don't force it.
  *
- * @param lhs, rhs The 64-bit integers to multiply
+ * @param lhs , rhs The 64-bit integers to multiply
  * @return The low 64 bits of the product XOR'd by the high 64 bits.
  * @see XXH_mult64to128()
  */
-static xxh_u64 XXH3_mul128_fold64(xxh_u64 lhs, xxh_u64 rhs) {
-
-  XXH128_hash_t product = XXH_mult64to128(lhs, rhs);
-  return product.low64 ^ product.high64;
-
+static xxh_u64
+XXH3_mul128_fold64(xxh_u64 lhs, xxh_u64 rhs)
+{
+    XXH128_hash_t product = XXH_mult64to128(lhs, rhs);
+    return product.low64 ^ product.high64;
 }
 
 /*! Seems to produce slightly better code on GCC for some reason. */
-XXH_FORCE_INLINE xxh_u64 XXH_xorshift64(xxh_u64 v64, int shift) {
-
-  XXH_ASSERT(0 <= shift && shift < 64);
-  return v64 ^ (v64 >> shift);
-
+XXH_FORCE_INLINE XXH_CONSTF xxh_u64 XXH_xorshift64(xxh_u64 v64, int shift)
+{
+    XXH_ASSERT(0 <= shift && shift < 64);
+    return v64 ^ (v64 >> shift);
 }
 
 /*
  * This is a fast avalanche stage,
  * suitable when input bits are already partially mixed
  */
-static XXH64_hash_t XXH3_avalanche(xxh_u64 h64) {
-
-  h64 = XXH_xorshift64(h64, 37);
-  h64 *= 0x165667919E3779F9ULL;
-  h64 = XXH_xorshift64(h64, 32);
-  return h64;
-
+static XXH64_hash_t XXH3_avalanche(xxh_u64 h64)
+{
+    h64 = XXH_xorshift64(h64, 37);
+    h64 *= PRIME_MX1;
+    h64 = XXH_xorshift64(h64, 32);
+    return h64;
 }
 
 /*
@@ -3781,17 +4467,17 @@ static XXH64_hash_t XXH3_avalanche(xxh_u64 h64) {
  * inspired by Pelle Evensen's rrmxmx
  * preferable when input has not been previously mixed
  */
-static XXH64_hash_t XXH3_rrmxmx(xxh_u64 h64, xxh_u64 len) {
-
-  /* this mix is inspired by Pelle Evensen's rrmxmx */
-  h64 ^= XXH_rotl64(h64, 49) ^ XXH_rotl64(h64, 24);
-  h64 *= 0x9FB21C651E98DF25ULL;
-  h64 ^= (h64 >> 35) + len;
-  h64 *= 0x9FB21C651E98DF25ULL;
-  return XXH_xorshift64(h64, 28);
-
+static XXH64_hash_t XXH3_rrmxmx(xxh_u64 h64, xxh_u64 len)
+{
+    /* this mix is inspired by Pelle Evensen's rrmxmx */
+    h64 ^= XXH_rotl64(h64, 49) ^ XXH_rotl64(h64, 24);
+    h64 *= PRIME_MX2;
+    h64 ^= (h64 >> 35) + len ;
+    h64 *= PRIME_MX2;
+    return XXH_xorshift64(h64, 28);
 }
 
+
 /* ==========================================
  * Short keys
  * ==========================================
@@ -3800,8 +4486,7 @@ static XXH64_hash_t XXH3_rrmxmx(xxh_u64 h64, xxh_u64 len) {
  * favored lengths that were a multiple of 4 or 8.
  *
  * Instead of iterating over individual inputs, we use a set of single shot
- * functions which piece together a range of lengths and operate in constant
- * time.
+ * functions which piece together a range of lengths and operate in constant time.
  *
  * Additionally, the number of multiplies has been significantly reduced. This
  * reduces latency, especially when emulating 64-bit multiplies on 32-bit.
@@ -3826,98 +4511,70 @@ static XXH64_hash_t XXH3_rrmxmx(xxh_u64 h64, xxh_u64 len) {
  *
  * This adds an extra layer of strength for custom secrets.
  */
-XXH_FORCE_INLINE XXH64_hash_t XXH3_len_1to3_64b(const xxh_u8 *input, size_t len,
-                                                const xxh_u8 *secret,
-                                                XXH64_hash_t  seed) {
-
-  XXH_ASSERT(input != NULL);
-  XXH_ASSERT(1 <= len && len <= 3);
-  XXH_ASSERT(secret != NULL);
-  /*
-   * len = 1: combined = { input[0], 0x01, input[0], input[0] }
-   * len = 2: combined = { input[1], 0x02, input[0], input[1] }
-   * len = 3: combined = { input[2], 0x03, input[0], input[1] }
-   */
-  {
-
-    xxh_u8 const  c1 = input[0];
-    xxh_u8 const  c2 = input[len >> 1];
-    xxh_u8 const  c3 = input[len - 1];
-    xxh_u32 const combined = ((xxh_u32)c1 << 16) | ((xxh_u32)c2 << 24) |
-                             ((xxh_u32)c3 << 0) | ((xxh_u32)len << 8);
-    xxh_u64 const bitflip =
-        (XXH_readLE32(secret) ^ XXH_readLE32(secret + 4)) + seed;
-    xxh_u64 const keyed = (xxh_u64)combined ^ bitflip;
-    return XXH64_avalanche(keyed);
-
-  }
-
+XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t
+XXH3_len_1to3_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(input != NULL);
+    XXH_ASSERT(1 <= len && len <= 3);
+    XXH_ASSERT(secret != NULL);
+    /*
+     * len = 1: combined = { input[0], 0x01, input[0], input[0] }
+     * len = 2: combined = { input[1], 0x02, input[0], input[1] }
+     * len = 3: combined = { input[2], 0x03, input[0], input[1] }
+     */
+    {   xxh_u8  const c1 = input[0];
+        xxh_u8  const c2 = input[len >> 1];
+        xxh_u8  const c3 = input[len - 1];
+        xxh_u32 const combined = ((xxh_u32)c1 << 16) | ((xxh_u32)c2  << 24)
+                               | ((xxh_u32)c3 <<  0) | ((xxh_u32)len << 8);
+        xxh_u64 const bitflip = (XXH_readLE32(secret) ^ XXH_readLE32(secret+4)) + seed;
+        xxh_u64 const keyed = (xxh_u64)combined ^ bitflip;
+        return XXH64_avalanche(keyed);
+    }
 }
 
-XXH_FORCE_INLINE XXH64_hash_t XXH3_len_4to8_64b(const xxh_u8 *input, size_t len,
-                                                const xxh_u8 *secret,
-                                                XXH64_hash_t  seed) {
-
-  XXH_ASSERT(input != NULL);
-  XXH_ASSERT(secret != NULL);
-  XXH_ASSERT(4 <= len && len <= 8);
-  seed ^= (xxh_u64)XXH_swap32((xxh_u32)seed) << 32;
-  {
-
-    xxh_u32 const input1 = XXH_readLE32(input);
-    xxh_u32 const input2 = XXH_readLE32(input + len - 4);
-    xxh_u64 const bitflip =
-        (XXH_readLE64(secret + 8) ^ XXH_readLE64(secret + 16)) - seed;
-    xxh_u64 const input64 = input2 + (((xxh_u64)input1) << 32);
-    xxh_u64 const keyed = input64 ^ bitflip;
-    return XXH3_rrmxmx(keyed, len);
-
-  }
-
+XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t
+XXH3_len_4to8_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(input != NULL);
+    XXH_ASSERT(secret != NULL);
+    XXH_ASSERT(4 <= len && len <= 8);
+    seed ^= (xxh_u64)XXH_swap32((xxh_u32)seed) << 32;
+    {   xxh_u32 const input1 = XXH_readLE32(input);
+        xxh_u32 const input2 = XXH_readLE32(input + len - 4);
+        xxh_u64 const bitflip = (XXH_readLE64(secret+8) ^ XXH_readLE64(secret+16)) - seed;
+        xxh_u64 const input64 = input2 + (((xxh_u64)input1) << 32);
+        xxh_u64 const keyed = input64 ^ bitflip;
+        return XXH3_rrmxmx(keyed, len);
+    }
 }
 
-XXH_FORCE_INLINE XXH64_hash_t XXH3_len_9to16_64b(const xxh_u8 *input,
-                                                 size_t        len,
-                                                 const xxh_u8 *secret,
-                                                 XXH64_hash_t  seed) {
-
-  XXH_ASSERT(input != NULL);
-  XXH_ASSERT(secret != NULL);
-  XXH_ASSERT(9 <= len && len <= 16);
-  {
-
-    xxh_u64 const bitflip1 =
-        (XXH_readLE64(secret + 24) ^ XXH_readLE64(secret + 32)) + seed;
-    xxh_u64 const bitflip2 =
-        (XXH_readLE64(secret + 40) ^ XXH_readLE64(secret + 48)) - seed;
-    xxh_u64 const input_lo = XXH_readLE64(input) ^ bitflip1;
-    xxh_u64 const input_hi = XXH_readLE64(input + len - 8) ^ bitflip2;
-    xxh_u64 const acc = len + XXH_swap64(input_lo) + input_hi +
-                        XXH3_mul128_fold64(input_lo, input_hi);
-    return XXH3_avalanche(acc);
-
-  }
-
+XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t
+XXH3_len_9to16_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(input != NULL);
+    XXH_ASSERT(secret != NULL);
+    XXH_ASSERT(9 <= len && len <= 16);
+    {   xxh_u64 const bitflip1 = (XXH_readLE64(secret+24) ^ XXH_readLE64(secret+32)) + seed;
+        xxh_u64 const bitflip2 = (XXH_readLE64(secret+40) ^ XXH_readLE64(secret+48)) - seed;
+        xxh_u64 const input_lo = XXH_readLE64(input)           ^ bitflip1;
+        xxh_u64 const input_hi = XXH_readLE64(input + len - 8) ^ bitflip2;
+        xxh_u64 const acc = len
+                          + XXH_swap64(input_lo) + input_hi
+                          + XXH3_mul128_fold64(input_lo, input_hi);
+        return XXH3_avalanche(acc);
+    }
 }
 
-XXH_FORCE_INLINE XXH64_hash_t XXH3_len_0to16_64b(const xxh_u8 *input,
-                                                 size_t        len,
-                                                 const xxh_u8 *secret,
-                                                 XXH64_hash_t  seed) {
-
-  XXH_ASSERT(len <= 16);
-  {
-
-    if (XXH_likely(len > 8))
-      return XXH3_len_9to16_64b(input, len, secret, seed);
-    if (XXH_likely(len >= 4))
-      return XXH3_len_4to8_64b(input, len, secret, seed);
-    if (len) return XXH3_len_1to3_64b(input, len, secret, seed);
-    return XXH64_avalanche(
-        seed ^ (XXH_readLE64(secret + 56) ^ XXH_readLE64(secret + 64)));
-
-  }
-
+XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t
+XXH3_len_0to16_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(len <= 16);
+    {   if (XXH_likely(len >  8)) return XXH3_len_9to16_64b(input, len, secret, seed);
+        if (XXH_likely(len >= 4)) return XXH3_len_4to8_64b(input, len, secret, seed);
+        if (len) return XXH3_len_1to3_64b(input, len, secret, seed);
+        return XXH64_avalanche(seed ^ (XXH_readLE64(secret+56) ^ XXH_readLE64(secret+64)));
+    }
 }
 
 /*
@@ -3946,113 +4603,106 @@ XXH_FORCE_INLINE XXH64_hash_t XXH3_len_0to16_64b(const xxh_u8 *input,
  * by this, although it is always a good idea to use a proper seed if you care
  * about strength.
  */
-XXH_FORCE_INLINE xxh_u64 XXH3_mix16B(const xxh_u8 *XXH_RESTRICT input,
-                                     const xxh_u8 *XXH_RESTRICT secret,
-                                     xxh_u64                    seed64) {
-
-      #if defined(__GNUC__) && !defined(__clang__)  /* GCC, not Clang */      \
-          && defined(__i386__) && defined(__SSE2__) /* x86 + SSE2 */          \
-          && !defined(XXH_ENABLE_AUTOVECTORIZE)     /* Define to disable like \
-                                                       XXH32 hack */
-  /*
-   * UGLY HACK:
-   * GCC for x86 tends to autovectorize the 128-bit multiply, resulting in
-   * slower code.
-   *
-   * By forcing seed64 into a register, we disrupt the cost model and
-   * cause it to scalarize. See `XXH32_round()`
-   *
-   * FIXME: Clang's output is still _much_ faster -- On an AMD Ryzen 3600,
-   * XXH3_64bits @ len=240 runs at 4.6 GB/s with Clang 9, but 3.3 GB/s on
-   * GCC 9.2, despite both emitting scalar code.
-   *
-   * GCC generates much better scalar code than Clang for the rest of XXH3,
-   * which is why finding a more optimal codepath is an interest.
-   */
-  XXH_COMPILER_GUARD(seed64);
-      #endif
-  {
-
-    xxh_u64 const input_lo = XXH_readLE64(input);
-    xxh_u64 const input_hi = XXH_readLE64(input + 8);
-    return XXH3_mul128_fold64(input_lo ^ (XXH_readLE64(secret) + seed64),
-                              input_hi ^ (XXH_readLE64(secret + 8) - seed64));
-
-  }
-
+XXH_FORCE_INLINE xxh_u64 XXH3_mix16B(const xxh_u8* XXH_RESTRICT input,
+                                     const xxh_u8* XXH_RESTRICT secret, xxh_u64 seed64)
+{
+#if defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
+  && defined(__i386__) && defined(__SSE2__)  /* x86 + SSE2 */ \
+  && !defined(XXH_ENABLE_AUTOVECTORIZE)      /* Define to disable like XXH32 hack */
+    /*
+     * UGLY HACK:
+     * GCC for x86 tends to autovectorize the 128-bit multiply, resulting in
+     * slower code.
+     *
+     * By forcing seed64 into a register, we disrupt the cost model and
+     * cause it to scalarize. See `XXH32_round()`
+     *
+     * FIXME: Clang's output is still _much_ faster -- On an AMD Ryzen 3600,
+     * XXH3_64bits @ len=240 runs at 4.6 GB/s with Clang 9, but 3.3 GB/s on
+     * GCC 9.2, despite both emitting scalar code.
+     *
+     * GCC generates much better scalar code than Clang for the rest of XXH3,
+     * which is why finding a more optimal codepath is an interest.
+     */
+    XXH_COMPILER_GUARD(seed64);
+#endif
+    {   xxh_u64 const input_lo = XXH_readLE64(input);
+        xxh_u64 const input_hi = XXH_readLE64(input+8);
+        return XXH3_mul128_fold64(
+            input_lo ^ (XXH_readLE64(secret)   + seed64),
+            input_hi ^ (XXH_readLE64(secret+8) - seed64)
+        );
+    }
 }
 
 /* For mid range keys, XXH3 uses a Mum-hash variant. */
-XXH_FORCE_INLINE XXH64_hash_t XXH3_len_17to128_64b(
-    const xxh_u8 *XXH_RESTRICT input, size_t len,
-    const xxh_u8 *XXH_RESTRICT secret, size_t secretSize, XXH64_hash_t seed) {
-
-  XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
-  (void)secretSize;
-  XXH_ASSERT(16 < len && len <= 128);
-
-  {
-
-    xxh_u64 acc = len * XXH_PRIME64_1;
-    if (len > 32) {
-
-      if (len > 64) {
-
-        if (len > 96) {
-
-          acc += XXH3_mix16B(input + 48, secret + 96, seed);
-          acc += XXH3_mix16B(input + len - 64, secret + 112, seed);
-
+XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t
+XXH3_len_17to128_64b(const xxh_u8* XXH_RESTRICT input, size_t len,
+                     const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
+                     XXH64_hash_t seed)
+{
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
+    XXH_ASSERT(16 < len && len <= 128);
+
+    {   xxh_u64 acc = len * XXH_PRIME64_1;
+#if XXH_SIZE_OPT >= 1
+        /* Smaller and cleaner, but slightly slower. */
+        unsigned int i = (unsigned int)(len - 1) / 32;
+        do {
+            acc += XXH3_mix16B(input+16 * i, secret+32*i, seed);
+            acc += XXH3_mix16B(input+len-16*(i+1), secret+32*i+16, seed);
+        } while (i-- != 0);
+#else
+        if (len > 32) {
+            if (len > 64) {
+                if (len > 96) {
+                    acc += XXH3_mix16B(input+48, secret+96, seed);
+                    acc += XXH3_mix16B(input+len-64, secret+112, seed);
+                }
+                acc += XXH3_mix16B(input+32, secret+64, seed);
+                acc += XXH3_mix16B(input+len-48, secret+80, seed);
+            }
+            acc += XXH3_mix16B(input+16, secret+32, seed);
+            acc += XXH3_mix16B(input+len-32, secret+48, seed);
         }
-
-        acc += XXH3_mix16B(input + 32, secret + 64, seed);
-        acc += XXH3_mix16B(input + len - 48, secret + 80, seed);
-
-      }
-
-      acc += XXH3_mix16B(input + 16, secret + 32, seed);
-      acc += XXH3_mix16B(input + len - 32, secret + 48, seed);
-
+        acc += XXH3_mix16B(input+0, secret+0, seed);
+        acc += XXH3_mix16B(input+len-16, secret+16, seed);
+#endif
+        return XXH3_avalanche(acc);
     }
-
-    acc += XXH3_mix16B(input + 0, secret + 0, seed);
-    acc += XXH3_mix16B(input + len - 16, secret + 16, seed);
-
-    return XXH3_avalanche(acc);
-
-  }
-
 }
 
-      #define XXH3_MIDSIZE_MAX 240
-
-XXH_NO_INLINE XXH64_hash_t XXH3_len_129to240_64b(
-    const xxh_u8 *XXH_RESTRICT input, size_t len,
-    const xxh_u8 *XXH_RESTRICT secret, size_t secretSize, XXH64_hash_t seed) {
-
-  XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
-  (void)secretSize;
-  XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX);
-
-      #define XXH3_MIDSIZE_STARTOFFSET 3
-      #define XXH3_MIDSIZE_LASTOFFSET 17
-
-  {
-
-    xxh_u64   acc = len * XXH_PRIME64_1;
-    int const nbRounds = (int)len / 16;
-    int       i;
-    for (i = 0; i < 8; i++) {
-
-      acc += XXH3_mix16B(input + (16 * i), secret + (16 * i), seed);
-
-    }
-
-    acc = XXH3_avalanche(acc);
-    XXH_ASSERT(nbRounds >= 8);
-      #if defined(__clang__)                                /* Clang */ \
-          && (defined(__ARM_NEON) || defined(__ARM_NEON__)) /* NEON */  \
-          && !defined(XXH_ENABLE_AUTOVECTORIZE)        /* Define to disable */
+/*!
+ * @brief Maximum size of "short" key in bytes.
+ */
+#define XXH3_MIDSIZE_MAX 240
+
+XXH_NO_INLINE XXH_PUREF XXH64_hash_t
+XXH3_len_129to240_64b(const xxh_u8* XXH_RESTRICT input, size_t len,
+                      const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
+                      XXH64_hash_t seed)
+{
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
+    XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX);
+
+    #define XXH3_MIDSIZE_STARTOFFSET 3
+    #define XXH3_MIDSIZE_LASTOFFSET  17
+
+    {   xxh_u64 acc = len * XXH_PRIME64_1;
+        xxh_u64 acc_end;
+        unsigned int const nbRounds = (unsigned int)len / 16;
+        unsigned int i;
+        XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX);
+        for (i=0; i<8; i++) {
+            acc += XXH3_mix16B(input+(16*i), secret+(16*i), seed);
+        }
+        /* last bytes */
+        acc_end = XXH3_mix16B(input + len - 16, secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET, seed);
+        XXH_ASSERT(nbRounds >= 8);
+        acc = XXH3_avalanche(acc);
+#if defined(__clang__)                                /* Clang */ \
+    && (defined(__ARM_NEON) || defined(__ARM_NEON__)) /* NEON */ \
+    && !defined(XXH_ENABLE_AUTOVECTORIZE)             /* Define to disable */
         /*
          * UGLY HACK:
          * Clang for ARMv7-A tries to vectorize this loop, similar to GCC x86.
@@ -4065,2205 +4715,2334 @@ XXH_NO_INLINE XXH64_hash_t XXH3_len_129to240_64b(
          * converts them to the nonexistent "vmulq_u64" intrinsic, which is then
          * scalarized into an ugly mess of VMOV.32 instructions.
          *
-         * This mess is difficult to avoid without turning autovectorization
-         * off completely, but they are usually relatively minor and/or not
-         * worth it to fix.
-         *
-         * This loop is the easiest to fix, as unlike XXH32, this pragma
-         * _actually works_ because it is a loop vectorization instead of an
-         * SLP vectorization.
-         */
-        #pragma clang loop vectorize(disable)
-      #endif
-    for (i = 8; i < nbRounds; i++) {
-
-      acc +=
-          XXH3_mix16B(input + (16 * i),
-                      secret + (16 * (i - 8)) + XXH3_MIDSIZE_STARTOFFSET, seed);
-
-    }
-
-    /* last bytes */
-    acc += XXH3_mix16B(input + len - 16,
-                       secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET,
-                       seed);
-    return XXH3_avalanche(acc);
-
-  }
-
-}
-
-    /* =======     Long Keys     ======= */
-
-      #define XXH_STRIPE_LEN 64
-      #define XXH_SECRET_CONSUME_RATE \
-        8               /* nb of secret bytes consumed at each accumulation */
-      #define XXH_ACC_NB (XXH_STRIPE_LEN / sizeof(xxh_u64))
-
-      #ifdef XXH_OLD_NAMES
-        #define STRIPE_LEN XXH_STRIPE_LEN
-        #define ACC_NB XXH_ACC_NB
-      #endif
-
-XXH_FORCE_INLINE void XXH_writeLE64(void *dst, xxh_u64 v64) {
-
-  if (!XXH_CPU_LITTLE_ENDIAN) v64 = XXH_swap64(v64);
-  memcpy(dst, &v64, sizeof(v64));
-
-}
-
-      /* Several intrinsic functions below are supposed to accept __int64 as
-       * argument, as documented in
-       * https://software.intel.com/sites/landingpage/IntrinsicsGuide/ .
-       * However, several environments do not define __int64 type,
-       * requiring a workaround.
-       */
-      #if !defined(__VMS) &&                                     \
-          (defined(__cplusplus) || (defined(__STDC_VERSION__) && \
-                                    (__STDC_VERSION__ >= 199901L) /* C99 */))
-typedef int64_t xxh_i64;
-      #else
-/* the following type must have a width of 64-bit */
-typedef long long xxh_i64;
-      #endif
-
-    /*
-     * XXH3_accumulate_512 is the tightest loop for long inputs, and it is the
-     * most optimized.
-     *
-     * It is a hardened version of UMAC, based off of FARSH's implementation.
-     *
-     * This was chosen because it adapts quite well to 32-bit, 64-bit, and SIMD
-     * implementations, and it is ridiculously fast.
-     *
-     * We harden it by mixing the original input to the accumulators as well as
-     * the product.
-     *
-     * This means that in the (relatively likely) case of a multiply by zero,
-     * the original input is preserved.
-     *
-     * On 128-bit inputs, we swap 64-bit pairs when we add the input to improve
-     * cross-pollination, as otherwise the upper and lower halves would be
-     * essentially independent.
-     *
-     * This doesn't matter on 64-bit hashes since they all get merged together
-     * in the end, so we skip the extra step.
-     *
-     * Both XXH3_64bits and XXH3_128bits use this subroutine.
-     */
-
-      #if (XXH_VECTOR == XXH_AVX512) || \
-          (defined(XXH_DISPATCH_AVX512) && XXH_DISPATCH_AVX512 != 0)
-
-        #ifndef XXH_TARGET_AVX512
-          #define XXH_TARGET_AVX512             /* disable attribute target */
-        #endif
-
-XXH_FORCE_INLINE XXH_TARGET_AVX512 void XXH3_accumulate_512_avx512(
-    void *XXH_RESTRICT acc, const void *XXH_RESTRICT input,
-    const void *XXH_RESTRICT secret) {
-
-  __m512i *const xacc = (__m512i *)acc;
-  XXH_ASSERT((((size_t)acc) & 63) == 0);
-  XXH_STATIC_ASSERT(XXH_STRIPE_LEN == sizeof(__m512i));
-
-  {
-
-    /* data_vec    = input[0]; */
-    __m512i const data_vec = _mm512_loadu_si512(input);
-    /* key_vec     = secret[0]; */
-    __m512i const key_vec = _mm512_loadu_si512(secret);
-    /* data_key    = data_vec ^ key_vec; */
-    __m512i const data_key = _mm512_xor_si512(data_vec, key_vec);
-    /* data_key_lo = data_key >> 32; */
-    __m512i const data_key_lo =
-        _mm512_shuffle_epi32(data_key, (_MM_PERM_ENUM)_MM_SHUFFLE(0, 3, 0, 1));
-    /* product     = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
-    __m512i const product = _mm512_mul_epu32(data_key, data_key_lo);
-    /* xacc[0] += swap(data_vec); */
-    __m512i const data_swap =
-        _mm512_shuffle_epi32(data_vec, (_MM_PERM_ENUM)_MM_SHUFFLE(1, 0, 3, 2));
-    __m512i const sum = _mm512_add_epi64(*xacc, data_swap);
-    /* xacc[0] += product; */
-    *xacc = _mm512_add_epi64(product, sum);
-
-  }
-
-}
-
-/*
- * XXH3_scrambleAcc: Scrambles the accumulators to improve mixing.
- *
- * Multiplication isn't perfect, as explained by Google in HighwayHash:
- *
- *  // Multiplication mixes/scrambles bytes 0-7 of the 64-bit result to
- *  // varying degrees. In descending order of goodness, bytes
- *  // 3 4 2 5 1 6 0 7 have quality 228 224 164 160 100 96 36 32.
- *  // As expected, the upper and lower bytes are much worse.
- *
- * Source:
- * https://github.com/google/highwayhash/blob/0aaf66b/highwayhash/hh_avx2.h#L291
- *
- * Since our algorithm uses a pseudorandom secret to add some variance into the
- * mix, we don't need to (or want to) mix as often or as much as HighwayHash
- * does.
- *
- * This isn't as tight as XXH3_accumulate, but still written in SIMD to avoid
- * extraction.
- *
- * Both XXH3_64bits and XXH3_128bits use this subroutine.
- */
-
-XXH_FORCE_INLINE XXH_TARGET_AVX512 void XXH3_scrambleAcc_avx512(
-    void *XXH_RESTRICT acc, const void *XXH_RESTRICT secret) {
-
-  XXH_ASSERT((((size_t)acc) & 63) == 0);
-  XXH_STATIC_ASSERT(XXH_STRIPE_LEN == sizeof(__m512i));
-  {
-
-    __m512i *const xacc = (__m512i *)acc;
-    const __m512i  prime32 = _mm512_set1_epi32((int)XXH_PRIME32_1);
-
-    /* xacc[0] ^= (xacc[0] >> 47) */
-    __m512i const acc_vec = *xacc;
-    __m512i const shifted = _mm512_srli_epi64(acc_vec, 47);
-    __m512i const data_vec = _mm512_xor_si512(acc_vec, shifted);
-    /* xacc[0] ^= secret; */
-    __m512i const key_vec = _mm512_loadu_si512(secret);
-    __m512i const data_key = _mm512_xor_si512(data_vec, key_vec);
-
-    /* xacc[0] *= XXH_PRIME32_1; */
-    __m512i const data_key_hi =
-        _mm512_shuffle_epi32(data_key, (_MM_PERM_ENUM)_MM_SHUFFLE(0, 3, 0, 1));
-    __m512i const prod_lo = _mm512_mul_epu32(data_key, prime32);
-    __m512i const prod_hi = _mm512_mul_epu32(data_key_hi, prime32);
-    *xacc = _mm512_add_epi64(prod_lo, _mm512_slli_epi64(prod_hi, 32));
-
-  }
-
-}
-
-XXH_FORCE_INLINE XXH_TARGET_AVX512 void XXH3_initCustomSecret_avx512(
-    void *XXH_RESTRICT customSecret, xxh_u64 seed64) {
-
-  XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 63) == 0);
-  XXH_STATIC_ASSERT(XXH_SEC_ALIGN == 64);
-  XXH_ASSERT(((size_t)customSecret & 63) == 0);
-  (void)(&XXH_writeLE64);
-  {
-
-    int const     nbRounds = XXH_SECRET_DEFAULT_SIZE / sizeof(__m512i);
-    __m512i const seed = _mm512_mask_set1_epi64(
-        _mm512_set1_epi64((xxh_i64)seed64), 0xAA, (xxh_i64)(0U - seed64));
-
-    const __m512i *const src = (const __m512i *)((const void *)XXH3_kSecret);
-    __m512i *const       dest = (__m512i *)customSecret;
-    int                  i;
-    XXH_ASSERT(((size_t)src & 63) == 0);               /* control alignment */
-    XXH_ASSERT(((size_t)dest & 63) == 0);
-    for (i = 0; i < nbRounds; ++i) {
-
-      /* GCC has a bug, _mm512_stream_load_si512 accepts 'void*', not 'void
-       * const*', this will warn "discards 'const' qualifier". */
-      union {
-
-        const __m512i *cp;
-        void          *p;
-
-      } remote_const_void;
-
-      remote_const_void.cp = src + i;
-      dest[i] =
-          _mm512_add_epi64(_mm512_stream_load_si512(remote_const_void.p), seed);
-
-    }
-
-  }
-
-}
-
-      #endif
-
-      #if (XXH_VECTOR == XXH_AVX2) || \
-          (defined(XXH_DISPATCH_AVX2) && XXH_DISPATCH_AVX2 != 0)
-
-        #ifndef XXH_TARGET_AVX2
-          #define XXH_TARGET_AVX2               /* disable attribute target */
-        #endif
-
-XXH_FORCE_INLINE XXH_TARGET_AVX2 void XXH3_accumulate_512_avx2(
-    void *XXH_RESTRICT acc, const void *XXH_RESTRICT input,
-    const void *XXH_RESTRICT secret) {
-
-  XXH_ASSERT((((size_t)acc) & 31) == 0);
-  {
-
-    __m256i *const xacc = (__m256i *)acc;
-    /* Unaligned. This is mainly for pointer arithmetic, and because
-     * _mm256_loadu_si256 requires  a const __m256i * pointer for some reason.
-     */
-    const __m256i *const xinput = (const __m256i *)input;
-    /* Unaligned. This is mainly for pointer arithmetic, and because
-     * _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */
-    const __m256i *const xsecret = (const __m256i *)secret;
-
-    size_t i;
-    for (i = 0; i < XXH_STRIPE_LEN / sizeof(__m256i); i++) {
-
-      /* data_vec    = xinput[i]; */
-      __m256i const data_vec = _mm256_loadu_si256(xinput + i);
-      /* key_vec     = xsecret[i]; */
-      __m256i const key_vec = _mm256_loadu_si256(xsecret + i);
-      /* data_key    = data_vec ^ key_vec; */
-      __m256i const data_key = _mm256_xor_si256(data_vec, key_vec);
-      /* data_key_lo = data_key >> 32; */
-      __m256i const data_key_lo =
-          _mm256_shuffle_epi32(data_key, _MM_SHUFFLE(0, 3, 0, 1));
-      /* product     = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
-      __m256i const product = _mm256_mul_epu32(data_key, data_key_lo);
-      /* xacc[i] += swap(data_vec); */
-      __m256i const data_swap =
-          _mm256_shuffle_epi32(data_vec, _MM_SHUFFLE(1, 0, 3, 2));
-      __m256i const sum = _mm256_add_epi64(xacc[i], data_swap);
-      /* xacc[i] += product; */
-      xacc[i] = _mm256_add_epi64(product, sum);
-
-    }
-
-  }
-
-}
-
-XXH_FORCE_INLINE XXH_TARGET_AVX2 void XXH3_scrambleAcc_avx2(
-    void *XXH_RESTRICT acc, const void *XXH_RESTRICT secret) {
-
-  XXH_ASSERT((((size_t)acc) & 31) == 0);
-  {
-
-    __m256i *const xacc = (__m256i *)acc;
-    /* Unaligned. This is mainly for pointer arithmetic, and because
-     * _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */
-    const __m256i *const xsecret = (const __m256i *)secret;
-    const __m256i        prime32 = _mm256_set1_epi32((int)XXH_PRIME32_1);
-
-    size_t i;
-    for (i = 0; i < XXH_STRIPE_LEN / sizeof(__m256i); i++) {
-
-      /* xacc[i] ^= (xacc[i] >> 47) */
-      __m256i const acc_vec = xacc[i];
-      __m256i const shifted = _mm256_srli_epi64(acc_vec, 47);
-      __m256i const data_vec = _mm256_xor_si256(acc_vec, shifted);
-      /* xacc[i] ^= xsecret; */
-      __m256i const key_vec = _mm256_loadu_si256(xsecret + i);
-      __m256i const data_key = _mm256_xor_si256(data_vec, key_vec);
-
-      /* xacc[i] *= XXH_PRIME32_1; */
-      __m256i const data_key_hi =
-          _mm256_shuffle_epi32(data_key, _MM_SHUFFLE(0, 3, 0, 1));
-      __m256i const prod_lo = _mm256_mul_epu32(data_key, prime32);
-      __m256i const prod_hi = _mm256_mul_epu32(data_key_hi, prime32);
-      xacc[i] = _mm256_add_epi64(prod_lo, _mm256_slli_epi64(prod_hi, 32));
-
-    }
-
-  }
-
-}
-
-XXH_FORCE_INLINE XXH_TARGET_AVX2 void XXH3_initCustomSecret_avx2(
-    void *XXH_RESTRICT customSecret, xxh_u64 seed64) {
-
-  XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 31) == 0);
-  XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE / sizeof(__m256i)) == 6);
-  XXH_STATIC_ASSERT(XXH_SEC_ALIGN <= 64);
-  (void)(&XXH_writeLE64);
-  XXH_PREFETCH(customSecret);
-  {
-
-    __m256i const seed =
-        _mm256_set_epi64x((xxh_i64)(0U - seed64), (xxh_i64)seed64,
-                          (xxh_i64)(0U - seed64), (xxh_i64)seed64);
-
-    const __m256i *const src = (const __m256i *)((const void *)XXH3_kSecret);
-    __m256i             *dest = (__m256i *)customSecret;
-
-        #if defined(__GNUC__) || defined(__clang__)
-    /*
-     * On GCC & Clang, marking 'dest' as modified will cause the compiler:
-     *   - do not extract the secret from sse registers in the internal loop
-     *   - use less common registers, and avoid pushing these reg into stack
-     */
-    XXH_COMPILER_GUARD(dest);
-        #endif
-    XXH_ASSERT(((size_t)src & 31) == 0);               /* control alignment */
-    XXH_ASSERT(((size_t)dest & 31) == 0);
-
-    /* GCC -O2 need unroll loop manually */
-    dest[0] = _mm256_add_epi64(_mm256_stream_load_si256(src + 0), seed);
-    dest[1] = _mm256_add_epi64(_mm256_stream_load_si256(src + 1), seed);
-    dest[2] = _mm256_add_epi64(_mm256_stream_load_si256(src + 2), seed);
-    dest[3] = _mm256_add_epi64(_mm256_stream_load_si256(src + 3), seed);
-    dest[4] = _mm256_add_epi64(_mm256_stream_load_si256(src + 4), seed);
-    dest[5] = _mm256_add_epi64(_mm256_stream_load_si256(src + 5), seed);
-
-  }
-
-}
-
-      #endif
-
-      /* x86dispatch always generates SSE2 */
-      #if (XXH_VECTOR == XXH_SSE2) || defined(XXH_X86DISPATCH)
-
-        #ifndef XXH_TARGET_SSE2
-          #define XXH_TARGET_SSE2               /* disable attribute target */
-        #endif
-
-XXH_FORCE_INLINE XXH_TARGET_SSE2 void XXH3_accumulate_512_sse2(
-    void *XXH_RESTRICT acc, const void *XXH_RESTRICT input,
-    const void *XXH_RESTRICT secret) {
-
-  /* SSE2 is just a half-scale version of the AVX2 version. */
-  XXH_ASSERT((((size_t)acc) & 15) == 0);
-  {
-
-    __m128i *const xacc = (__m128i *)acc;
-    /* Unaligned. This is mainly for pointer arithmetic, and because
-     * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
-    const __m128i *const xinput = (const __m128i *)input;
-    /* Unaligned. This is mainly for pointer arithmetic, and because
-     * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
-    const __m128i *const xsecret = (const __m128i *)secret;
-
-    size_t i;
-    for (i = 0; i < XXH_STRIPE_LEN / sizeof(__m128i); i++) {
-
-      /* data_vec    = xinput[i]; */
-      __m128i const data_vec = _mm_loadu_si128(xinput + i);
-      /* key_vec     = xsecret[i]; */
-      __m128i const key_vec = _mm_loadu_si128(xsecret + i);
-      /* data_key    = data_vec ^ key_vec; */
-      __m128i const data_key = _mm_xor_si128(data_vec, key_vec);
-      /* data_key_lo = data_key >> 32; */
-      __m128i const data_key_lo =
-          _mm_shuffle_epi32(data_key, _MM_SHUFFLE(0, 3, 0, 1));
-      /* product     = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
-      __m128i const product = _mm_mul_epu32(data_key, data_key_lo);
-      /* xacc[i] += swap(data_vec); */
-      __m128i const data_swap =
-          _mm_shuffle_epi32(data_vec, _MM_SHUFFLE(1, 0, 3, 2));
-      __m128i const sum = _mm_add_epi64(xacc[i], data_swap);
-      /* xacc[i] += product; */
-      xacc[i] = _mm_add_epi64(product, sum);
-
-    }
-
-  }
-
-}
-
-XXH_FORCE_INLINE XXH_TARGET_SSE2 void XXH3_scrambleAcc_sse2(
-    void *XXH_RESTRICT acc, const void *XXH_RESTRICT secret) {
-
-  XXH_ASSERT((((size_t)acc) & 15) == 0);
-  {
-
-    __m128i *const xacc = (__m128i *)acc;
-    /* Unaligned. This is mainly for pointer arithmetic, and because
-     * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
-    const __m128i *const xsecret = (const __m128i *)secret;
-    const __m128i        prime32 = _mm_set1_epi32((int)XXH_PRIME32_1);
-
-    size_t i;
-    for (i = 0; i < XXH_STRIPE_LEN / sizeof(__m128i); i++) {
-
-      /* xacc[i] ^= (xacc[i] >> 47) */
-      __m128i const acc_vec = xacc[i];
-      __m128i const shifted = _mm_srli_epi64(acc_vec, 47);
-      __m128i const data_vec = _mm_xor_si128(acc_vec, shifted);
-      /* xacc[i] ^= xsecret[i]; */
-      __m128i const key_vec = _mm_loadu_si128(xsecret + i);
-      __m128i const data_key = _mm_xor_si128(data_vec, key_vec);
-
-      /* xacc[i] *= XXH_PRIME32_1; */
-      __m128i const data_key_hi =
-          _mm_shuffle_epi32(data_key, _MM_SHUFFLE(0, 3, 0, 1));
-      __m128i const prod_lo = _mm_mul_epu32(data_key, prime32);
-      __m128i const prod_hi = _mm_mul_epu32(data_key_hi, prime32);
-      xacc[i] = _mm_add_epi64(prod_lo, _mm_slli_epi64(prod_hi, 32));
-
-    }
-
-  }
-
-}
-
-XXH_FORCE_INLINE XXH_TARGET_SSE2 void XXH3_initCustomSecret_sse2(
-    void *XXH_RESTRICT customSecret, xxh_u64 seed64) {
-
-  XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 15) == 0);
-  (void)(&XXH_writeLE64);
-  {
-
-    int const nbRounds = XXH_SECRET_DEFAULT_SIZE / sizeof(__m128i);
-
-        #if defined(_MSC_VER) && defined(_M_IX86) && _MSC_VER < 1900
-    /* MSVC 32bit mode does not support _mm_set_epi64x before 2015 */
-    XXH_ALIGN(16)
-    const xxh_i64 seed64x2[2] = {(xxh_i64)seed64, (xxh_i64)(0U - seed64)};
-    __m128i const seed = _mm_load_si128((__m128i const *)seed64x2);
-        #else
-    __m128i const seed =
-        _mm_set_epi64x((xxh_i64)(0U - seed64), (xxh_i64)seed64);
-        #endif
-    int i;
-
-    const void *const src16 = XXH3_kSecret;
-    __m128i          *dst16 = (__m128i *)customSecret;
-        #if defined(__GNUC__) || defined(__clang__)
-    /*
-     * On GCC & Clang, marking 'dest' as modified will cause the compiler:
-     *   - do not extract the secret from sse registers in the internal loop
-     *   - use less common registers, and avoid pushing these reg into stack
-     */
-    XXH_COMPILER_GUARD(dst16);
-        #endif
-    XXH_ASSERT(((size_t)src16 & 15) == 0);             /* control alignment */
-    XXH_ASSERT(((size_t)dst16 & 15) == 0);
-
-    for (i = 0; i < nbRounds; ++i) {
-
-      dst16[i] =
-          _mm_add_epi64(_mm_load_si128((const __m128i *)src16 + i), seed);
-
-    }
-
-  }
-
-}
-
-      #endif
-
-      #if (XXH_VECTOR == XXH_NEON)
-
-XXH_FORCE_INLINE void XXH3_accumulate_512_neon(
-    void *XXH_RESTRICT acc, const void *XXH_RESTRICT input,
-    const void *XXH_RESTRICT secret) {
-
-  XXH_ASSERT((((size_t)acc) & 15) == 0);
-  {
-
-    uint64x2_t *const xacc = (uint64x2_t *)acc;
-    /* We don't use a uint32x4_t pointer because it causes bus errors on ARMv7.
-     */
-    uint8_t const *const xinput = (const uint8_t *)input;
-    uint8_t const *const xsecret = (const uint8_t *)secret;
-
-    size_t i;
-    for (i = 0; i < XXH_STRIPE_LEN / sizeof(uint64x2_t); i++) {
-
-      /* data_vec = xinput[i]; */
-      uint8x16_t data_vec = vld1q_u8(xinput + (i * 16));
-      /* key_vec  = xsecret[i];  */
-      uint8x16_t key_vec = vld1q_u8(xsecret + (i * 16));
-      uint64x2_t data_key;
-      uint32x2_t data_key_lo, data_key_hi;
-      /* xacc[i] += swap(data_vec); */
-      uint64x2_t const data64 = vreinterpretq_u64_u8(data_vec);
-      uint64x2_t const swapped = vextq_u64(data64, data64, 1);
-      xacc[i] = vaddq_u64(xacc[i], swapped);
-      /* data_key = data_vec ^ key_vec; */
-      data_key = vreinterpretq_u64_u8(veorq_u8(data_vec, key_vec));
-      /* data_key_lo = (uint32x2_t) (data_key & 0xFFFFFFFF);
-       * data_key_hi = (uint32x2_t) (data_key >> 32);
-       * data_key = UNDEFINED; */
-      XXH_SPLIT_IN_PLACE(data_key, data_key_lo, data_key_hi);
-      /* xacc[i] += (uint64x2_t) data_key_lo * (uint64x2_t) data_key_hi; */
-      xacc[i] = vmlal_u32(xacc[i], data_key_lo, data_key_hi);
-
-    }
-
-  }
-
-}
-
-XXH_FORCE_INLINE void XXH3_scrambleAcc_neon(void *XXH_RESTRICT       acc,
-                                            const void *XXH_RESTRICT secret) {
-
-  XXH_ASSERT((((size_t)acc) & 15) == 0);
-
-  {
-
-    uint64x2_t    *xacc = (uint64x2_t *)acc;
-    uint8_t const *xsecret = (uint8_t const *)secret;
-    uint32x2_t     prime = vdup_n_u32(XXH_PRIME32_1);
-
-    size_t i;
-    for (i = 0; i < XXH_STRIPE_LEN / sizeof(uint64x2_t); i++) {
-
-      /* xacc[i] ^= (xacc[i] >> 47); */
-      uint64x2_t acc_vec = xacc[i];
-      uint64x2_t shifted = vshrq_n_u64(acc_vec, 47);
-      uint64x2_t data_vec = veorq_u64(acc_vec, shifted);
-
-      /* xacc[i] ^= xsecret[i]; */
-      uint8x16_t key_vec = vld1q_u8(xsecret + (i * 16));
-      uint64x2_t data_key = veorq_u64(data_vec, vreinterpretq_u64_u8(key_vec));
-
-      /* xacc[i] *= XXH_PRIME32_1 */
-      uint32x2_t data_key_lo, data_key_hi;
-      /* data_key_lo = (uint32x2_t) (xacc[i] & 0xFFFFFFFF);
-       * data_key_hi = (uint32x2_t) (xacc[i] >> 32);
-       * xacc[i] = UNDEFINED; */
-      XXH_SPLIT_IN_PLACE(data_key, data_key_lo, data_key_hi);
-      { /*
-         * prod_hi = (data_key >> 32) * XXH_PRIME32_1;
-         *
-         * Avoid vmul_u32 + vshll_n_u32 since Clang 6 and 7 will
-         * incorrectly "optimize" this:
-         *   tmp     = vmul_u32(vmovn_u64(a), vmovn_u64(b));
-         *   shifted = vshll_n_u32(tmp, 32);
-         * to this:
-         *   tmp     = "vmulq_u64"(a, b); // no such thing!
-         *   shifted = vshlq_n_u64(tmp, 32);
-         *
-         * However, unlike SSE, Clang lacks a 64-bit multiply routine
-         * for NEON, and it scalarizes two 64-bit multiplies instead.
-         *
-         * vmull_u32 has the same timing as vmul_u32, and it avoids
-         * this bug completely.
-         * See https://bugs.llvm.org/show_bug.cgi?id=39967
-         */
-        uint64x2_t prod_hi = vmull_u32(data_key_hi, prime);
-        /* xacc[i] = prod_hi << 32; */
-        xacc[i] = vshlq_n_u64(prod_hi, 32);
-        /* xacc[i] += (prod_hi & 0xFFFFFFFF) * XXH_PRIME32_1; */
-        xacc[i] = vmlal_u32(xacc[i], data_key_lo, prime);
-
-      }
-
-    }
-
-  }
-
-}
-
-      #endif
-
-      #if (XXH_VECTOR == XXH_VSX)
-
-XXH_FORCE_INLINE void XXH3_accumulate_512_vsx(void *XXH_RESTRICT       acc,
-                                              const void *XXH_RESTRICT input,
-                                              const void *XXH_RESTRICT secret) {
-
-  xxh_u64x2 *const       xacc = (xxh_u64x2 *)acc;       /* presumed aligned */
-  xxh_u64x2 const *const xinput =
-      (xxh_u64x2 const *)input;                 /* no alignment restriction */
-  xxh_u64x2 const *const xsecret =
-      (xxh_u64x2 const *)secret;                /* no alignment restriction */
-  xxh_u64x2 const v32 = {32, 32};
-  size_t          i;
-  for (i = 0; i < XXH_STRIPE_LEN / sizeof(xxh_u64x2); i++) {
-
-    /* data_vec = xinput[i]; */
-    xxh_u64x2 const data_vec = XXH_vec_loadu(xinput + i);
-    /* key_vec = xsecret[i]; */
-    xxh_u64x2 const key_vec = XXH_vec_loadu(xsecret + i);
-    xxh_u64x2 const data_key = data_vec ^ key_vec;
-    /* shuffled = (data_key << 32) | (data_key >> 32); */
-    xxh_u32x4 const shuffled = (xxh_u32x4)vec_rl(data_key, v32);
-    /* product = ((xxh_u64x2)data_key & 0xFFFFFFFF) * ((xxh_u64x2)shuffled &
-     * 0xFFFFFFFF); */
-    xxh_u64x2 const product = XXH_vec_mulo((xxh_u32x4)data_key, shuffled);
-    xacc[i] += product;
-
-          /* swap high and low halves */
-        #ifdef __s390x__
-    xacc[i] += vec_permi(data_vec, data_vec, 2);
-        #else
-    xacc[i] += vec_xxpermdi(data_vec, data_vec, 2);
-        #endif
-
-  }
-
-}
-
-XXH_FORCE_INLINE void XXH3_scrambleAcc_vsx(void *XXH_RESTRICT       acc,
-                                           const void *XXH_RESTRICT secret) {
-
-  XXH_ASSERT((((size_t)acc) & 15) == 0);
-
-  {
-
-    xxh_u64x2 *const       xacc = (xxh_u64x2 *)acc;
-    const xxh_u64x2 *const xsecret = (const xxh_u64x2 *)secret;
-    /* constants */
-    xxh_u64x2 const v32 = {32, 32};
-    xxh_u64x2 const v47 = {47, 47};
-    xxh_u32x4 const prime = {XXH_PRIME32_1, XXH_PRIME32_1, XXH_PRIME32_1,
-                             XXH_PRIME32_1};
-    size_t          i;
-    for (i = 0; i < XXH_STRIPE_LEN / sizeof(xxh_u64x2); i++) {
-
-      /* xacc[i] ^= (xacc[i] >> 47); */
-      xxh_u64x2 const acc_vec = xacc[i];
-      xxh_u64x2 const data_vec = acc_vec ^ (acc_vec >> v47);
-
-      /* xacc[i] ^= xsecret[i]; */
-      xxh_u64x2 const key_vec = XXH_vec_loadu(xsecret + i);
-      xxh_u64x2 const data_key = data_vec ^ key_vec;
-
-      /* xacc[i] *= XXH_PRIME32_1 */
-      /* prod_lo = ((xxh_u64x2)data_key & 0xFFFFFFFF) * ((xxh_u64x2)prime &
-       * 0xFFFFFFFF);  */
-      xxh_u64x2 const prod_even = XXH_vec_mule((xxh_u32x4)data_key, prime);
-      /* prod_hi = ((xxh_u64x2)data_key >> 32) * ((xxh_u64x2)prime >> 32);  */
-      xxh_u64x2 const prod_odd = XXH_vec_mulo((xxh_u32x4)data_key, prime);
-      xacc[i] = prod_odd + (prod_even << v32);
-
-    }
-
-  }
-
-}
-
-      #endif
-
-/* scalar variants - universal */
-
-XXH_FORCE_INLINE void XXH3_accumulate_512_scalar(
-    void *XXH_RESTRICT acc, const void *XXH_RESTRICT input,
-    const void *XXH_RESTRICT secret) {
-
-  xxh_u64 *const      xacc = (xxh_u64 *)acc;            /* presumed aligned */
-  const xxh_u8 *const xinput =
-      (const xxh_u8 *)input;                    /* no alignment restriction */
-  const xxh_u8 *const xsecret =
-      (const xxh_u8 *)secret;                   /* no alignment restriction */
-  size_t i;
-  XXH_ASSERT(((size_t)acc & (XXH_ACC_ALIGN - 1)) == 0);
-  for (i = 0; i < XXH_ACC_NB; i++) {
-
-    xxh_u64 const data_val = XXH_readLE64(xinput + 8 * i);
-    xxh_u64 const data_key = data_val ^ XXH_readLE64(xsecret + i * 8);
-    xacc[i ^ 1] += data_val;                         /* swap adjacent lanes */
-    xacc[i] += XXH_mult32to64(data_key & 0xFFFFFFFF, data_key >> 32);
-
-  }
-
-}
-
-XXH_FORCE_INLINE void XXH3_scrambleAcc_scalar(void *XXH_RESTRICT       acc,
-                                              const void *XXH_RESTRICT secret) {
-
-  xxh_u64 *const      xacc = (xxh_u64 *)acc;            /* presumed aligned */
-  const xxh_u8 *const xsecret =
-      (const xxh_u8 *)secret;                   /* no alignment restriction */
-  size_t i;
-  XXH_ASSERT((((size_t)acc) & (XXH_ACC_ALIGN - 1)) == 0);
-  for (i = 0; i < XXH_ACC_NB; i++) {
-
-    xxh_u64 const key64 = XXH_readLE64(xsecret + 8 * i);
-    xxh_u64       acc64 = xacc[i];
-    acc64 = XXH_xorshift64(acc64, 47);
-    acc64 ^= key64;
-    acc64 *= XXH_PRIME32_1;
-    xacc[i] = acc64;
-
-  }
-
-}
-
-XXH_FORCE_INLINE void XXH3_initCustomSecret_scalar(
-    void *XXH_RESTRICT customSecret, xxh_u64 seed64) {
-
-  /*
-   * We need a separate pointer for the hack below,
-   * which requires a non-const pointer.
-   * Any decent compiler will optimize this out otherwise.
-   */
-  const xxh_u8 *kSecretPtr = XXH3_kSecret;
-  XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 15) == 0);
-
-      #if defined(__clang__) && defined(__aarch64__)
-  /*
-   * UGLY HACK:
-   * Clang generates a bunch of MOV/MOVK pairs for aarch64, and they are
-   * placed sequentially, in order, at the top of the unrolled loop.
-   *
-   * While MOVK is great for generating constants (2 cycles for a 64-bit
-   * constant compared to 4 cycles for LDR), long MOVK chains stall the
-   * integer pipelines:
-   *   I   L   S
-   * MOVK
-   * MOVK
-   * MOVK
-   * MOVK
-   * ADD
-   * SUB      STR
-   *          STR
-   * By forcing loads from memory (as the asm line causes Clang to assume
-   * that XXH3_kSecretPtr has been changed), the pipelines are used more
-   * efficiently:
-   *   I   L   S
-   *      LDR
-   *  ADD LDR
-   *  SUB     STR
-   *          STR
-   * XXH3_64bits_withSeed, len == 256, Snapdragon 835
-   *   without hack: 2654.4 MB/s
-   *   with hack:    3202.9 MB/s
-   */
-  XXH_COMPILER_GUARD(kSecretPtr);
-      #endif
-  /*
-   * Note: in debug mode, this overrides the asm optimization
-   * and Clang will emit MOVK chains again.
-   */
-  XXH_ASSERT(kSecretPtr == XXH3_kSecret);
-
-  {
-
-    int const nbRounds = XXH_SECRET_DEFAULT_SIZE / 16;
-    int       i;
-    for (i = 0; i < nbRounds; i++) {
-
-      /*
-       * The asm hack causes Clang to assume that kSecretPtr aliases with
-       * customSecret, and on aarch64, this prevented LDP from merging two
-       * loads together for free. Putting the loads together before the stores
-       * properly generates LDP.
-       */
-      xxh_u64 lo = XXH_readLE64(kSecretPtr + 16 * i) + seed64;
-      xxh_u64 hi = XXH_readLE64(kSecretPtr + 16 * i + 8) - seed64;
-      XXH_writeLE64((xxh_u8 *)customSecret + 16 * i, lo);
-      XXH_writeLE64((xxh_u8 *)customSecret + 16 * i + 8, hi);
-
-    }
-
-  }
-
-}
-
-typedef void (*XXH3_f_accumulate_512)(void *XXH_RESTRICT, const void *,
-                                      const void *);
-typedef void (*XXH3_f_scrambleAcc)(void *XXH_RESTRICT, const void *);
-typedef void (*XXH3_f_initCustomSecret)(void *XXH_RESTRICT, xxh_u64);
-
-      #if (XXH_VECTOR == XXH_AVX512)
-
-        #define XXH3_accumulate_512 XXH3_accumulate_512_avx512
-        #define XXH3_scrambleAcc XXH3_scrambleAcc_avx512
-        #define XXH3_initCustomSecret XXH3_initCustomSecret_avx512
-
-      #elif (XXH_VECTOR == XXH_AVX2)
-
-        #define XXH3_accumulate_512 XXH3_accumulate_512_avx2
-        #define XXH3_scrambleAcc XXH3_scrambleAcc_avx2
-        #define XXH3_initCustomSecret XXH3_initCustomSecret_avx2
-
-      #elif (XXH_VECTOR == XXH_SSE2)
-
-        #define XXH3_accumulate_512 XXH3_accumulate_512_sse2
-        #define XXH3_scrambleAcc XXH3_scrambleAcc_sse2
-        #define XXH3_initCustomSecret XXH3_initCustomSecret_sse2
-
-      #elif (XXH_VECTOR == XXH_NEON)
-
-        #define XXH3_accumulate_512 XXH3_accumulate_512_neon
-        #define XXH3_scrambleAcc XXH3_scrambleAcc_neon
-        #define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
-
-      #elif (XXH_VECTOR == XXH_VSX)
-
-        #define XXH3_accumulate_512 XXH3_accumulate_512_vsx
-        #define XXH3_scrambleAcc XXH3_scrambleAcc_vsx
-        #define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
-
-      #else                                                       /* scalar */
-
-        #define XXH3_accumulate_512 XXH3_accumulate_512_scalar
-        #define XXH3_scrambleAcc XXH3_scrambleAcc_scalar
-        #define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
-
-      #endif
-
-      #ifndef XXH_PREFETCH_DIST
-        #ifdef __clang__
-          #define XXH_PREFETCH_DIST 320
-        #else
-          #if (XXH_VECTOR == XXH_AVX512)
-            #define XXH_PREFETCH_DIST 512
-          #else
-            #define XXH_PREFETCH_DIST 384
-          #endif
-        #endif                                                 /* __clang__ */
-      #endif                                           /* XXH_PREFETCH_DIST */
-
-/*
- * XXH3_accumulate()
- * Loops over XXH3_accumulate_512().
- * Assumption: nbStripes will not overflow the secret size
- */
-XXH_FORCE_INLINE void XXH3_accumulate(xxh_u64 *XXH_RESTRICT      acc,
-                                      const xxh_u8 *XXH_RESTRICT input,
-                                      const xxh_u8 *XXH_RESTRICT secret,
-                                      size_t                     nbStripes,
-                                      XXH3_f_accumulate_512      f_acc512) {
-
-  size_t n;
-  for (n = 0; n < nbStripes; n++) {
-
-    const xxh_u8 *const in = input + n * XXH_STRIPE_LEN;
-    XXH_PREFETCH(in + XXH_PREFETCH_DIST);
-    f_acc512(acc, in, secret + n * XXH_SECRET_CONSUME_RATE);
-
-  }
-
-}
-
-XXH_FORCE_INLINE void XXH3_hashLong_internal_loop(
-    xxh_u64 *XXH_RESTRICT acc, const xxh_u8 *XXH_RESTRICT input, size_t len,
-    const xxh_u8 *XXH_RESTRICT secret, size_t secretSize,
-    XXH3_f_accumulate_512 f_acc512, XXH3_f_scrambleAcc f_scramble) {
-
-  size_t const nbStripesPerBlock =
-      (secretSize - XXH_STRIPE_LEN) / XXH_SECRET_CONSUME_RATE;
-  size_t const block_len = XXH_STRIPE_LEN * nbStripesPerBlock;
-  size_t const nb_blocks = (len - 1) / block_len;
-
-  size_t n;
-
-  XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
-
-  for (n = 0; n < nb_blocks; n++) {
-
-    XXH3_accumulate(acc, input + n * block_len, secret, nbStripesPerBlock,
-                    f_acc512);
-    f_scramble(acc, secret + secretSize - XXH_STRIPE_LEN);
-
-  }
-
-  /* last partial block */
-  XXH_ASSERT(len > XXH_STRIPE_LEN);
-  {
-
-    size_t const nbStripes =
-        ((len - 1) - (block_len * nb_blocks)) / XXH_STRIPE_LEN;
-    XXH_ASSERT(nbStripes <= (secretSize / XXH_SECRET_CONSUME_RATE));
-    XXH3_accumulate(acc, input + nb_blocks * block_len, secret, nbStripes,
-                    f_acc512);
-
-    /* last stripe */
-    {
-
-      const xxh_u8 *const p = input + len - XXH_STRIPE_LEN;
-      #define XXH_SECRET_LASTACC_START                                       \
-        7 /* not aligned on 8, last secret is different from acc & scrambler \
-           */
-      f_acc512(acc, p,
-               secret + secretSize - XXH_STRIPE_LEN - XXH_SECRET_LASTACC_START);
-
-    }
-
-  }
-
-}
-
-XXH_FORCE_INLINE xxh_u64 XXH3_mix2Accs(const xxh_u64 *XXH_RESTRICT acc,
-                                       const xxh_u8 *XXH_RESTRICT  secret) {
-
-  return XXH3_mul128_fold64(acc[0] ^ XXH_readLE64(secret),
-                            acc[1] ^ XXH_readLE64(secret + 8));
-
-}
-
-static XXH64_hash_t XXH3_mergeAccs(const xxh_u64 *XXH_RESTRICT acc,
-                                   const xxh_u8 *XXH_RESTRICT  secret,
-                                   xxh_u64                     start) {
-
-  xxh_u64 result64 = start;
-  size_t  i = 0;
-
-  for (i = 0; i < 4; i++) {
-
-    result64 += XXH3_mix2Accs(acc + 2 * i, secret + 16 * i);
-      #if defined(__clang__)                                /* Clang */ \
-          && (defined(__arm__) || defined(__thumb__))       /* ARMv7 */ \
-          && (defined(__ARM_NEON) || defined(__ARM_NEON__)) /* NEON */  \
-          && !defined(XXH_ENABLE_AUTOVECTORIZE)        /* Define to disable */
-    /*
-     * UGLY HACK:
-     * Prevent autovectorization on Clang ARMv7-a. Exact same problem as
-     * the one in XXH3_len_129to240_64b. Speeds up shorter keys > 240b.
-     * XXH3_64bits, len == 256, Snapdragon 835:
-     *   without hack: 2063.7 MB/s
-     *   with hack:    2560.7 MB/s
-     */
-    XXH_COMPILER_GUARD(result64);
-      #endif
-
-  }
-
-  return XXH3_avalanche(result64);
-
-}
-
-      #define XXH3_INIT_ACC                                              \
-        {                                                                \
-                                                                         \
-          XXH_PRIME32_3, XXH_PRIME64_1, XXH_PRIME64_2, XXH_PRIME64_3,    \
-              XXH_PRIME64_4, XXH_PRIME32_2, XXH_PRIME64_5, XXH_PRIME32_1 \
-                                                                         \
-        }
-
-XXH_FORCE_INLINE XXH64_hash_t XXH3_hashLong_64b_internal(
-    const void *XXH_RESTRICT input, size_t len, const void *XXH_RESTRICT secret,
-    size_t secretSize, XXH3_f_accumulate_512 f_acc512,
-    XXH3_f_scrambleAcc f_scramble) {
-
-  XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[XXH_ACC_NB] = XXH3_INIT_ACC;
-
-  XXH3_hashLong_internal_loop(acc, (const xxh_u8 *)input, len,
-                              (const xxh_u8 *)secret, secretSize, f_acc512,
-                              f_scramble);
-
-  /* converge into final hash */
-  XXH_STATIC_ASSERT(sizeof(acc) == 64);
-      /* do not align on 8, so that the secret is different from the accumulator
-       */
-      #define XXH_SECRET_MERGEACCS_START 11
-  XXH_ASSERT(secretSize >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
-  return XXH3_mergeAccs(acc,
-                        (const xxh_u8 *)secret + XXH_SECRET_MERGEACCS_START,
-                        (xxh_u64)len * XXH_PRIME64_1);
-
-}
-
-/*
- * It's important for performance that XXH3_hashLong is not inlined.
- */
-XXH_NO_INLINE XXH64_hash_t XXH3_hashLong_64b_withSecret(
-    const void *XXH_RESTRICT input, size_t len, XXH64_hash_t seed64,
-    const xxh_u8 *XXH_RESTRICT secret, size_t secretLen) {
-
-  (void)seed64;
-  return XXH3_hashLong_64b_internal(input, len, secret, secretLen,
-                                    XXH3_accumulate_512, XXH3_scrambleAcc);
-
-}
-
-/*
- * It's important for performance that XXH3_hashLong is not inlined.
- * Since the function is not inlined, the compiler may not be able to understand
- * that, in some scenarios, its `secret` argument is actually a compile time
- * constant. This variant enforces that the compiler can detect that, and uses
- * this opportunity to streamline the generated code for better performance.
- */
-XXH_NO_INLINE XXH64_hash_t XXH3_hashLong_64b_default(
-    const void *XXH_RESTRICT input, size_t len, XXH64_hash_t seed64,
-    const xxh_u8 *XXH_RESTRICT secret, size_t secretLen) {
-
-  (void)seed64;
-  (void)secret;
-  (void)secretLen;
-  return XXH3_hashLong_64b_internal(input, len, XXH3_kSecret,
-                                    sizeof(XXH3_kSecret), XXH3_accumulate_512,
-                                    XXH3_scrambleAcc);
-
-}
-
-/*
- * XXH3_hashLong_64b_withSeed():
- * Generate a custom key based on alteration of default XXH3_kSecret with the
- * seed, and then use this key for long mode hashing.
- *
- * This operation is decently fast but nonetheless costs a little bit of time.
- * Try to avoid it whenever possible (typically when seed==0).
- *
- * It's important for performance that XXH3_hashLong is not inlined. Not sure
- * why (uop cache maybe?), but the difference is large and easily measurable.
- */
-XXH_FORCE_INLINE XXH64_hash_t XXH3_hashLong_64b_withSeed_internal(
-    const void *input, size_t len, XXH64_hash_t seed,
-    XXH3_f_accumulate_512 f_acc512, XXH3_f_scrambleAcc f_scramble,
-    XXH3_f_initCustomSecret f_initSec) {
-
-  if (seed == 0)
-    return XXH3_hashLong_64b_internal(
-        input, len, XXH3_kSecret, sizeof(XXH3_kSecret), f_acc512, f_scramble);
-  {
-
-    XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
-    f_initSec(secret, seed);
-    return XXH3_hashLong_64b_internal(input, len, secret, sizeof(secret),
-                                      f_acc512, f_scramble);
-
-  }
-
-}
-
-/*
- * It's important for performance that XXH3_hashLong is not inlined.
- */
-XXH_NO_INLINE XXH64_hash_t XXH3_hashLong_64b_withSeed(const void   *input,
-                                                      size_t        len,
-                                                      XXH64_hash_t  seed,
-                                                      const xxh_u8 *secret,
-                                                      size_t        secretLen) {
-
-  (void)secret;
-  (void)secretLen;
-  return XXH3_hashLong_64b_withSeed_internal(
-      input, len, seed, XXH3_accumulate_512, XXH3_scrambleAcc,
-      XXH3_initCustomSecret);
-
-}
-
-typedef XXH64_hash_t (*XXH3_hashLong64_f)(const void *XXH_RESTRICT, size_t,
-                                          XXH64_hash_t,
-                                          const xxh_u8 *XXH_RESTRICT, size_t);
-
-XXH_FORCE_INLINE XXH64_hash_t
-XXH3_64bits_internal(const void *XXH_RESTRICT input, size_t len,
-                     XXH64_hash_t seed64, const void *XXH_RESTRICT secret,
-                     size_t secretLen, XXH3_hashLong64_f f_hashLong) {
-
-  XXH_ASSERT(secretLen >= XXH3_SECRET_SIZE_MIN);
-  /*
-   * If an action is to be taken if `secretLen` condition is not respected,
-   * it should be done here.
-   * For now, it's a contract pre-condition.
-   * Adding a check and a branch here would cost performance at every hash.
-   * Also, note that function signature doesn't offer room to return an error.
-   */
-  if (len <= 16)
-    return XXH3_len_0to16_64b((const xxh_u8 *)input, len,
-                              (const xxh_u8 *)secret, seed64);
-  if (len <= 128)
-    return XXH3_len_17to128_64b((const xxh_u8 *)input, len,
-                                (const xxh_u8 *)secret, secretLen, seed64);
-  if (len <= XXH3_MIDSIZE_MAX)
-    return XXH3_len_129to240_64b((const xxh_u8 *)input, len,
-                                 (const xxh_u8 *)secret, secretLen, seed64);
-  return f_hashLong(input, len, seed64, (const xxh_u8 *)secret, secretLen);
-
-}
-
-/* ===   Public entry point   === */
-
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API XXH64_hash_t XXH3_64bits(const void *input, size_t len) {
-
-  return XXH3_64bits_internal(input, len, 0, XXH3_kSecret, sizeof(XXH3_kSecret),
-                              XXH3_hashLong_64b_default);
-
+         * This mess is difficult to avoid without turning autovectorization
+         * off completely, but they are usually relatively minor and/or not
+         * worth it to fix.
+         *
+         * This loop is the easiest to fix, as unlike XXH32, this pragma
+         * _actually works_ because it is a loop vectorization instead of an
+         * SLP vectorization.
+         */
+        #pragma clang loop vectorize(disable)
+#endif
+        for (i=8 ; i < nbRounds; i++) {
+            /*
+             * Prevents clang for unrolling the acc loop and interleaving with this one.
+             */
+            XXH_COMPILER_GUARD(acc);
+            acc_end += XXH3_mix16B(input+(16*i), secret+(16*(i-8)) + XXH3_MIDSIZE_STARTOFFSET, seed);
+        }
+        return XXH3_avalanche(acc + acc_end);
+    }
 }
 
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_withSecret(const void *input,
-                                                   size_t      len,
-                                                   const void *secret,
-                                                   size_t      secretSize) {
 
-  return XXH3_64bits_internal(input, len, 0, secret, secretSize,
-                              XXH3_hashLong_64b_withSecret);
+/* =======     Long Keys     ======= */
 
-}
+#define XXH_STRIPE_LEN 64
+#define XXH_SECRET_CONSUME_RATE 8   /* nb of secret bytes consumed at each accumulation */
+#define XXH_ACC_NB (XXH_STRIPE_LEN / sizeof(xxh_u64))
 
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_withSeed(const void *input, size_t len,
-                                                 XXH64_hash_t seed) {
+#ifdef XXH_OLD_NAMES
+#  define STRIPE_LEN XXH_STRIPE_LEN
+#  define ACC_NB XXH_ACC_NB
+#endif
 
-  return XXH3_64bits_internal(input, len, seed, XXH3_kSecret,
-                              sizeof(XXH3_kSecret), XXH3_hashLong_64b_withSeed);
+#ifndef XXH_PREFETCH_DIST
+#  ifdef __clang__
+#    define XXH_PREFETCH_DIST 320
+#  else
+#    if (XXH_VECTOR == XXH_AVX512)
+#      define XXH_PREFETCH_DIST 512
+#    else
+#      define XXH_PREFETCH_DIST 384
+#    endif
+#  endif  /* __clang__ */
+#endif  /* XXH_PREFETCH_DIST */
 
-}
+/*
+ * These macros are to generate an XXH3_accumulate() function.
+ * The two arguments select the name suffix and target attribute.
+ *
+ * The name of this symbol is XXH3_accumulate_<name>() and it calls
+ * XXH3_accumulate_512_<name>().
+ *
+ * It may be useful to hand implement this function if the compiler fails to
+ * optimize the inline function.
+ */
+#define XXH3_ACCUMULATE_TEMPLATE(name)                      \
+void                                                        \
+XXH3_accumulate_##name(xxh_u64* XXH_RESTRICT acc,           \
+                       const xxh_u8* XXH_RESTRICT input,    \
+                       const xxh_u8* XXH_RESTRICT secret,   \
+                       size_t nbStripes)                    \
+{                                                           \
+    size_t n;                                               \
+    for (n = 0; n < nbStripes; n++ ) {                      \
+        const xxh_u8* const in = input + n*XXH_STRIPE_LEN;  \
+        XXH_PREFETCH(in + XXH_PREFETCH_DIST);               \
+        XXH3_accumulate_512_##name(                         \
+                 acc,                                       \
+                 in,                                        \
+                 secret + n*XXH_SECRET_CONSUME_RATE);       \
+    }                                                       \
+}
+
+
+XXH_FORCE_INLINE void XXH_writeLE64(void* dst, xxh_u64 v64)
+{
+    if (!XXH_CPU_LITTLE_ENDIAN) v64 = XXH_swap64(v64);
+    XXH_memcpy(dst, &v64, sizeof(v64));
+}
+
+/* Several intrinsic functions below are supposed to accept __int64 as argument,
+ * as documented in https://software.intel.com/sites/landingpage/IntrinsicsGuide/ .
+ * However, several environments do not define __int64 type,
+ * requiring a workaround.
+ */
+#if !defined (__VMS) \
+  && (defined (__cplusplus) \
+  || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+    typedef int64_t xxh_i64;
+#else
+    /* the following type must have a width of 64-bit */
+    typedef long long xxh_i64;
+#endif
 
-/* ===   XXH3 streaming   === */
 
 /*
- * Malloc's a pointer that is always aligned to align.
+ * XXH3_accumulate_512 is the tightest loop for long inputs, and it is the most optimized.
  *
- * This must be freed with `XXH_alignedFree()`.
+ * It is a hardened version of UMAC, based off of FARSH's implementation.
  *
- * malloc typically guarantees 16 byte alignment on 64-bit systems and 8 byte
- * alignment on 32-bit. This isn't enough for the 32 byte aligned loads in AVX2
- * or on 32-bit, the 16 byte aligned loads in SSE2 and NEON.
+ * This was chosen because it adapts quite well to 32-bit, 64-bit, and SIMD
+ * implementations, and it is ridiculously fast.
  *
- * This underalignment previously caused a rather obvious crash which went
- * completely unnoticed due to XXH3_createState() not actually being tested.
- * Credit to RedSpah for noticing this bug.
+ * We harden it by mixing the original input to the accumulators as well as the product.
  *
- * The alignment is done manually: Functions like posix_memalign or _mm_malloc
- * are avoided: To maintain portability, we would have to write a fallback
- * like this anyways, and besides, testing for the existence of library
- * functions without relying on external build tools is impossible.
+ * This means that in the (relatively likely) case of a multiply by zero, the
+ * original input is preserved.
  *
- * The method is simple: Overallocate, manually align, and store the offset
- * to the original behind the returned pointer.
+ * On 128-bit inputs, we swap 64-bit pairs when we add the input to improve
+ * cross-pollination, as otherwise the upper and lower halves would be
+ * essentially independent.
  *
- * Align must be a power of 2 and 8 <= align <= 128.
+ * This doesn't matter on 64-bit hashes since they all get merged together in
+ * the end, so we skip the extra step.
+ *
+ * Both XXH3_64bits and XXH3_128bits use this subroutine.
  */
-static void *XXH_alignedMalloc(size_t s, size_t align) {
-
-  XXH_ASSERT(align <= 128 && align >= 8);                    /* range check */
-  XXH_ASSERT((align & (align - 1)) == 0);                     /* power of 2 */
-  XXH_ASSERT(s != 0 && s < (s + align));                  /* empty/overflow */
-  {  /* Overallocate to make room for manual realignment and an offset byte */
-    xxh_u8 *base = (xxh_u8 *)XXH_malloc(s + align);
-    if (base != NULL) {
 
-      /*
-       * Get the offset needed to align this pointer.
-       *
-       * Even if the returned pointer is aligned, there will always be
-       * at least one byte to store the offset to the original pointer.
-       */
-      size_t offset = align - ((size_t)base & (align - 1)); /* base % align */
-      /* Add the offset for the now-aligned pointer */
-      xxh_u8 *ptr = base + offset;
+#if (XXH_VECTOR == XXH_AVX512) \
+     || (defined(XXH_DISPATCH_AVX512) && XXH_DISPATCH_AVX512 != 0)
 
-      XXH_ASSERT((size_t)ptr % align == 0);
+#ifndef XXH_TARGET_AVX512
+# define XXH_TARGET_AVX512  /* disable attribute target */
+#endif
 
-      /* Store the offset immediately before the returned pointer. */
-      ptr[-1] = (xxh_u8)offset;
-      return ptr;
+XXH_FORCE_INLINE XXH_TARGET_AVX512 void
+XXH3_accumulate_512_avx512(void* XXH_RESTRICT acc,
+                     const void* XXH_RESTRICT input,
+                     const void* XXH_RESTRICT secret)
+{
+    __m512i* const xacc = (__m512i *) acc;
+    XXH_ASSERT((((size_t)acc) & 63) == 0);
+    XXH_STATIC_ASSERT(XXH_STRIPE_LEN == sizeof(__m512i));
 
+    {
+        /* data_vec    = input[0]; */
+        __m512i const data_vec    = _mm512_loadu_si512   (input);
+        /* key_vec     = secret[0]; */
+        __m512i const key_vec     = _mm512_loadu_si512   (secret);
+        /* data_key    = data_vec ^ key_vec; */
+        __m512i const data_key    = _mm512_xor_si512     (data_vec, key_vec);
+        /* data_key_lo = data_key >> 32; */
+        __m512i const data_key_lo = _mm512_srli_epi64 (data_key, 32);
+        /* product     = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
+        __m512i const product     = _mm512_mul_epu32     (data_key, data_key_lo);
+        /* xacc[0] += swap(data_vec); */
+        __m512i const data_swap = _mm512_shuffle_epi32(data_vec, (_MM_PERM_ENUM)_MM_SHUFFLE(1, 0, 3, 2));
+        __m512i const sum       = _mm512_add_epi64(*xacc, data_swap);
+        /* xacc[0] += product; */
+        *xacc = _mm512_add_epi64(product, sum);
     }
-
-    return NULL;
-
-  }
-
 }
+XXH_FORCE_INLINE XXH_TARGET_AVX512 XXH3_ACCUMULATE_TEMPLATE(avx512)
 
 /*
- * Frees an aligned pointer allocated by XXH_alignedMalloc(). Don't pass
- * normal malloc'd pointers, XXH_alignedMalloc has a specific data layout.
+ * XXH3_scrambleAcc: Scrambles the accumulators to improve mixing.
+ *
+ * Multiplication isn't perfect, as explained by Google in HighwayHash:
+ *
+ *  // Multiplication mixes/scrambles bytes 0-7 of the 64-bit result to
+ *  // varying degrees. In descending order of goodness, bytes
+ *  // 3 4 2 5 1 6 0 7 have quality 228 224 164 160 100 96 36 32.
+ *  // As expected, the upper and lower bytes are much worse.
+ *
+ * Source: https://github.com/google/highwayhash/blob/0aaf66b/highwayhash/hh_avx2.h#L291
+ *
+ * Since our algorithm uses a pseudorandom secret to add some variance into the
+ * mix, we don't need to (or want to) mix as often or as much as HighwayHash does.
+ *
+ * This isn't as tight as XXH3_accumulate, but still written in SIMD to avoid
+ * extraction.
+ *
+ * Both XXH3_64bits and XXH3_128bits use this subroutine.
  */
-static void XXH_alignedFree(void *p) {
-
-  if (p != NULL) {
-
-    xxh_u8 *ptr = (xxh_u8 *)p;
-    /* Get the offset byte we added in XXH_malloc. */
-    xxh_u8 offset = ptr[-1];
-    /* Free the original malloc'd pointer */
-    xxh_u8 *base = ptr - offset;
-    XXH_free(base);
-
-  }
 
+XXH_FORCE_INLINE XXH_TARGET_AVX512 void
+XXH3_scrambleAcc_avx512(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 63) == 0);
+    XXH_STATIC_ASSERT(XXH_STRIPE_LEN == sizeof(__m512i));
+    {   __m512i* const xacc = (__m512i*) acc;
+        const __m512i prime32 = _mm512_set1_epi32((int)XXH_PRIME32_1);
+
+        /* xacc[0] ^= (xacc[0] >> 47) */
+        __m512i const acc_vec     = *xacc;
+        __m512i const shifted     = _mm512_srli_epi64    (acc_vec, 47);
+        /* xacc[0] ^= secret; */
+        __m512i const key_vec     = _mm512_loadu_si512   (secret);
+        __m512i const data_key    = _mm512_ternarylogic_epi32(key_vec, acc_vec, shifted, 0x96 /* key_vec ^ acc_vec ^ shifted */);
+
+        /* xacc[0] *= XXH_PRIME32_1; */
+        __m512i const data_key_hi = _mm512_srli_epi64 (data_key, 32);
+        __m512i const prod_lo     = _mm512_mul_epu32     (data_key, prime32);
+        __m512i const prod_hi     = _mm512_mul_epu32     (data_key_hi, prime32);
+        *xacc = _mm512_add_epi64(prod_lo, _mm512_slli_epi64(prod_hi, 32));
+    }
 }
 
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API XXH3_state_t *XXH3_createState(void) {
-
-  XXH3_state_t *const state =
-      (XXH3_state_t *)XXH_alignedMalloc(sizeof(XXH3_state_t), 64);
-  if (state == NULL) return NULL;
-  XXH3_INITSTATE(state);
-  return state;
+XXH_FORCE_INLINE XXH_TARGET_AVX512 void
+XXH3_initCustomSecret_avx512(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
+{
+    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 63) == 0);
+    XXH_STATIC_ASSERT(XXH_SEC_ALIGN == 64);
+    XXH_ASSERT(((size_t)customSecret & 63) == 0);
+    (void)(&XXH_writeLE64);
+    {   int const nbRounds = XXH_SECRET_DEFAULT_SIZE / sizeof(__m512i);
+        __m512i const seed_pos = _mm512_set1_epi64((xxh_i64)seed64);
+        __m512i const seed     = _mm512_mask_sub_epi64(seed_pos, 0xAA, _mm512_set1_epi8(0), seed_pos);
 
+        const __m512i* const src  = (const __m512i*) ((const void*) XXH3_kSecret);
+              __m512i* const dest = (      __m512i*) customSecret;
+        int i;
+        XXH_ASSERT(((size_t)src & 63) == 0); /* control alignment */
+        XXH_ASSERT(((size_t)dest & 63) == 0);
+        for (i=0; i < nbRounds; ++i) {
+            dest[i] = _mm512_add_epi64(_mm512_load_si512(src + i), seed);
+    }   }
 }
 
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t *statePtr) {
-
-  XXH_alignedFree(statePtr);
-  return XXH_OK;
-
-}
+#endif
 
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API void XXH3_copyState(XXH3_state_t       *dst_state,
-                                   const XXH3_state_t *src_state) {
+#if (XXH_VECTOR == XXH_AVX2) \
+    || (defined(XXH_DISPATCH_AVX2) && XXH_DISPATCH_AVX2 != 0)
 
-  memcpy(dst_state, src_state, sizeof(*dst_state));
+#ifndef XXH_TARGET_AVX2
+# define XXH_TARGET_AVX2  /* disable attribute target */
+#endif
 
+XXH_FORCE_INLINE XXH_TARGET_AVX2 void
+XXH3_accumulate_512_avx2( void* XXH_RESTRICT acc,
+                    const void* XXH_RESTRICT input,
+                    const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 31) == 0);
+    {   __m256i* const xacc    =       (__m256i *) acc;
+        /* Unaligned. This is mainly for pointer arithmetic, and because
+         * _mm256_loadu_si256 requires  a const __m256i * pointer for some reason. */
+        const         __m256i* const xinput  = (const __m256i *) input;
+        /* Unaligned. This is mainly for pointer arithmetic, and because
+         * _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */
+        const         __m256i* const xsecret = (const __m256i *) secret;
+
+        size_t i;
+        for (i=0; i < XXH_STRIPE_LEN/sizeof(__m256i); i++) {
+            /* data_vec    = xinput[i]; */
+            __m256i const data_vec    = _mm256_loadu_si256    (xinput+i);
+            /* key_vec     = xsecret[i]; */
+            __m256i const key_vec     = _mm256_loadu_si256   (xsecret+i);
+            /* data_key    = data_vec ^ key_vec; */
+            __m256i const data_key    = _mm256_xor_si256     (data_vec, key_vec);
+            /* data_key_lo = data_key >> 32; */
+            __m256i const data_key_lo = _mm256_srli_epi64 (data_key, 32);
+            /* product     = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
+            __m256i const product     = _mm256_mul_epu32     (data_key, data_key_lo);
+            /* xacc[i] += swap(data_vec); */
+            __m256i const data_swap = _mm256_shuffle_epi32(data_vec, _MM_SHUFFLE(1, 0, 3, 2));
+            __m256i const sum       = _mm256_add_epi64(xacc[i], data_swap);
+            /* xacc[i] += product; */
+            xacc[i] = _mm256_add_epi64(product, sum);
+    }   }
+}
+XXH_FORCE_INLINE XXH_TARGET_AVX2 XXH3_ACCUMULATE_TEMPLATE(avx2)
+
+XXH_FORCE_INLINE XXH_TARGET_AVX2 void
+XXH3_scrambleAcc_avx2(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 31) == 0);
+    {   __m256i* const xacc = (__m256i*) acc;
+        /* Unaligned. This is mainly for pointer arithmetic, and because
+         * _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */
+        const         __m256i* const xsecret = (const __m256i *) secret;
+        const __m256i prime32 = _mm256_set1_epi32((int)XXH_PRIME32_1);
+
+        size_t i;
+        for (i=0; i < XXH_STRIPE_LEN/sizeof(__m256i); i++) {
+            /* xacc[i] ^= (xacc[i] >> 47) */
+            __m256i const acc_vec     = xacc[i];
+            __m256i const shifted     = _mm256_srli_epi64    (acc_vec, 47);
+            __m256i const data_vec    = _mm256_xor_si256     (acc_vec, shifted);
+            /* xacc[i] ^= xsecret; */
+            __m256i const key_vec     = _mm256_loadu_si256   (xsecret+i);
+            __m256i const data_key    = _mm256_xor_si256     (data_vec, key_vec);
+
+            /* xacc[i] *= XXH_PRIME32_1; */
+            __m256i const data_key_hi = _mm256_srli_epi64 (data_key, 32);
+            __m256i const prod_lo     = _mm256_mul_epu32     (data_key, prime32);
+            __m256i const prod_hi     = _mm256_mul_epu32     (data_key_hi, prime32);
+            xacc[i] = _mm256_add_epi64(prod_lo, _mm256_slli_epi64(prod_hi, 32));
+        }
+    }
 }
 
-static void XXH3_reset_internal(XXH3_state_t *statePtr, XXH64_hash_t seed,
-                                const void *secret, size_t secretSize) {
+XXH_FORCE_INLINE XXH_TARGET_AVX2 void XXH3_initCustomSecret_avx2(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
+{
+    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 31) == 0);
+    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE / sizeof(__m256i)) == 6);
+    XXH_STATIC_ASSERT(XXH_SEC_ALIGN <= 64);
+    (void)(&XXH_writeLE64);
+    XXH_PREFETCH(customSecret);
+    {   __m256i const seed = _mm256_set_epi64x((xxh_i64)(0U - seed64), (xxh_i64)seed64, (xxh_i64)(0U - seed64), (xxh_i64)seed64);
 
-  size_t const initStart = offsetof(XXH3_state_t, bufferedSize);
-  size_t const initLength =
-      offsetof(XXH3_state_t, nbStripesPerBlock) - initStart;
-  XXH_ASSERT(offsetof(XXH3_state_t, nbStripesPerBlock) > initStart);
-  XXH_ASSERT(statePtr != NULL);
-  /* set members from bufferedSize to nbStripesPerBlock (excluded) to 0 */
-  memset((char *)statePtr + initStart, 0, initLength);
-  statePtr->acc[0] = XXH_PRIME32_3;
-  statePtr->acc[1] = XXH_PRIME64_1;
-  statePtr->acc[2] = XXH_PRIME64_2;
-  statePtr->acc[3] = XXH_PRIME64_3;
-  statePtr->acc[4] = XXH_PRIME64_4;
-  statePtr->acc[5] = XXH_PRIME32_2;
-  statePtr->acc[6] = XXH_PRIME64_5;
-  statePtr->acc[7] = XXH_PRIME32_1;
-  statePtr->seed = seed;
-  statePtr->extSecret = (const unsigned char *)secret;
-  XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
-  statePtr->secretLimit = secretSize - XXH_STRIPE_LEN;
-  statePtr->nbStripesPerBlock = statePtr->secretLimit / XXH_SECRET_CONSUME_RATE;
+        const __m256i* const src  = (const __m256i*) ((const void*) XXH3_kSecret);
+              __m256i*       dest = (      __m256i*) customSecret;
 
+#       if defined(__GNUC__) || defined(__clang__)
+        /*
+         * On GCC & Clang, marking 'dest' as modified will cause the compiler:
+         *   - do not extract the secret from sse registers in the internal loop
+         *   - use less common registers, and avoid pushing these reg into stack
+         */
+        XXH_COMPILER_GUARD(dest);
+#       endif
+        XXH_ASSERT(((size_t)src & 31) == 0); /* control alignment */
+        XXH_ASSERT(((size_t)dest & 31) == 0);
+
+        /* GCC -O2 need unroll loop manually */
+        dest[0] = _mm256_add_epi64(_mm256_load_si256(src+0), seed);
+        dest[1] = _mm256_add_epi64(_mm256_load_si256(src+1), seed);
+        dest[2] = _mm256_add_epi64(_mm256_load_si256(src+2), seed);
+        dest[3] = _mm256_add_epi64(_mm256_load_si256(src+3), seed);
+        dest[4] = _mm256_add_epi64(_mm256_load_si256(src+4), seed);
+        dest[5] = _mm256_add_epi64(_mm256_load_si256(src+5), seed);
+    }
 }
 
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset(XXH3_state_t *statePtr) {
-
-  if (statePtr == NULL) return XXH_ERROR;
-  XXH3_reset_internal(statePtr, 0, XXH3_kSecret, XXH_SECRET_DEFAULT_SIZE);
-  return XXH_OK;
-
-}
+#endif
 
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSecret(
-    XXH3_state_t *statePtr, const void *secret, size_t secretSize) {
+/* x86dispatch always generates SSE2 */
+#if (XXH_VECTOR == XXH_SSE2) || defined(XXH_X86DISPATCH)
 
-  if (statePtr == NULL) return XXH_ERROR;
-  XXH3_reset_internal(statePtr, 0, secret, secretSize);
-  if (secret == NULL) return XXH_ERROR;
-  if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
-  return XXH_OK;
+#ifndef XXH_TARGET_SSE2
+# define XXH_TARGET_SSE2  /* disable attribute target */
+#endif
 
+XXH_FORCE_INLINE XXH_TARGET_SSE2 void
+XXH3_accumulate_512_sse2( void* XXH_RESTRICT acc,
+                    const void* XXH_RESTRICT input,
+                    const void* XXH_RESTRICT secret)
+{
+    /* SSE2 is just a half-scale version of the AVX2 version. */
+    XXH_ASSERT((((size_t)acc) & 15) == 0);
+    {   __m128i* const xacc    =       (__m128i *) acc;
+        /* Unaligned. This is mainly for pointer arithmetic, and because
+         * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
+        const         __m128i* const xinput  = (const __m128i *) input;
+        /* Unaligned. This is mainly for pointer arithmetic, and because
+         * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
+        const         __m128i* const xsecret = (const __m128i *) secret;
+
+        size_t i;
+        for (i=0; i < XXH_STRIPE_LEN/sizeof(__m128i); i++) {
+            /* data_vec    = xinput[i]; */
+            __m128i const data_vec    = _mm_loadu_si128   (xinput+i);
+            /* key_vec     = xsecret[i]; */
+            __m128i const key_vec     = _mm_loadu_si128   (xsecret+i);
+            /* data_key    = data_vec ^ key_vec; */
+            __m128i const data_key    = _mm_xor_si128     (data_vec, key_vec);
+            /* data_key_lo = data_key >> 32; */
+            __m128i const data_key_lo = _mm_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1));
+            /* product     = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
+            __m128i const product     = _mm_mul_epu32     (data_key, data_key_lo);
+            /* xacc[i] += swap(data_vec); */
+            __m128i const data_swap = _mm_shuffle_epi32(data_vec, _MM_SHUFFLE(1,0,3,2));
+            __m128i const sum       = _mm_add_epi64(xacc[i], data_swap);
+            /* xacc[i] += product; */
+            xacc[i] = _mm_add_epi64(product, sum);
+    }   }
+}
+XXH_FORCE_INLINE XXH_TARGET_SSE2 XXH3_ACCUMULATE_TEMPLATE(sse2)
+
+XXH_FORCE_INLINE XXH_TARGET_SSE2 void
+XXH3_scrambleAcc_sse2(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 15) == 0);
+    {   __m128i* const xacc = (__m128i*) acc;
+        /* Unaligned. This is mainly for pointer arithmetic, and because
+         * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
+        const         __m128i* const xsecret = (const __m128i *) secret;
+        const __m128i prime32 = _mm_set1_epi32((int)XXH_PRIME32_1);
+
+        size_t i;
+        for (i=0; i < XXH_STRIPE_LEN/sizeof(__m128i); i++) {
+            /* xacc[i] ^= (xacc[i] >> 47) */
+            __m128i const acc_vec     = xacc[i];
+            __m128i const shifted     = _mm_srli_epi64    (acc_vec, 47);
+            __m128i const data_vec    = _mm_xor_si128     (acc_vec, shifted);
+            /* xacc[i] ^= xsecret[i]; */
+            __m128i const key_vec     = _mm_loadu_si128   (xsecret+i);
+            __m128i const data_key    = _mm_xor_si128     (data_vec, key_vec);
+
+            /* xacc[i] *= XXH_PRIME32_1; */
+            __m128i const data_key_hi = _mm_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1));
+            __m128i const prod_lo     = _mm_mul_epu32     (data_key, prime32);
+            __m128i const prod_hi     = _mm_mul_epu32     (data_key_hi, prime32);
+            xacc[i] = _mm_add_epi64(prod_lo, _mm_slli_epi64(prod_hi, 32));
+        }
+    }
 }
 
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSeed(XXH3_state_t *statePtr,
-                                                        XXH64_hash_t  seed) {
-
-  if (statePtr == NULL) return XXH_ERROR;
-  if (seed == 0) return XXH3_64bits_reset(statePtr);
-  if (seed != statePtr->seed)
-    XXH3_initCustomSecret(statePtr->customSecret, seed);
-  XXH3_reset_internal(statePtr, seed, NULL, XXH_SECRET_DEFAULT_SIZE);
-  return XXH_OK;
+XXH_FORCE_INLINE XXH_TARGET_SSE2 void XXH3_initCustomSecret_sse2(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
+{
+    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 15) == 0);
+    (void)(&XXH_writeLE64);
+    {   int const nbRounds = XXH_SECRET_DEFAULT_SIZE / sizeof(__m128i);
+
+#       if defined(_MSC_VER) && defined(_M_IX86) && _MSC_VER < 1900
+        /* MSVC 32bit mode does not support _mm_set_epi64x before 2015 */
+        XXH_ALIGN(16) const xxh_i64 seed64x2[2] = { (xxh_i64)seed64, (xxh_i64)(0U - seed64) };
+        __m128i const seed = _mm_load_si128((__m128i const*)seed64x2);
+#       else
+        __m128i const seed = _mm_set_epi64x((xxh_i64)(0U - seed64), (xxh_i64)seed64);
+#       endif
+        int i;
+
+        const void* const src16 = XXH3_kSecret;
+        __m128i* dst16 = (__m128i*) customSecret;
+#       if defined(__GNUC__) || defined(__clang__)
+        /*
+         * On GCC & Clang, marking 'dest' as modified will cause the compiler:
+         *   - do not extract the secret from sse registers in the internal loop
+         *   - use less common registers, and avoid pushing these reg into stack
+         */
+        XXH_COMPILER_GUARD(dst16);
+#       endif
+        XXH_ASSERT(((size_t)src16 & 15) == 0); /* control alignment */
+        XXH_ASSERT(((size_t)dst16 & 15) == 0);
 
+        for (i=0; i < nbRounds; ++i) {
+            dst16[i] = _mm_add_epi64(_mm_load_si128((const __m128i *)src16+i), seed);
+    }   }
 }
 
-/* Note : when XXH3_consumeStripes() is invoked,
- * there must be a guarantee that at least one more byte must be consumed from
- * input
- * so that the function can blindly consume all stripes using the "normal"
- * secret segment */
-XXH_FORCE_INLINE void XXH3_consumeStripes(
-    xxh_u64 *XXH_RESTRICT acc, size_t *XXH_RESTRICT nbStripesSoFarPtr,
-    size_t nbStripesPerBlock, const xxh_u8 *XXH_RESTRICT input,
-    size_t nbStripes, const xxh_u8 *XXH_RESTRICT secret, size_t secretLimit,
-    XXH3_f_accumulate_512 f_acc512, XXH3_f_scrambleAcc f_scramble) {
-
-  XXH_ASSERT(nbStripes <=
-             nbStripesPerBlock); /* can handle max 1 scramble per invocation */
-  XXH_ASSERT(*nbStripesSoFarPtr < nbStripesPerBlock);
-  if (nbStripesPerBlock - *nbStripesSoFarPtr <= nbStripes) {
-
-    /* need a scrambling operation */
-    size_t const nbStripesToEndofBlock = nbStripesPerBlock - *nbStripesSoFarPtr;
-    size_t const nbStripesAfterBlock = nbStripes - nbStripesToEndofBlock;
-    XXH3_accumulate(acc, input,
-                    secret + nbStripesSoFarPtr[0] * XXH_SECRET_CONSUME_RATE,
-                    nbStripesToEndofBlock, f_acc512);
-    f_scramble(acc, secret + secretLimit);
-    XXH3_accumulate(acc, input + nbStripesToEndofBlock * XXH_STRIPE_LEN, secret,
-                    nbStripesAfterBlock, f_acc512);
-    *nbStripesSoFarPtr = nbStripesAfterBlock;
-
-  } else {
+#endif
 
-    XXH3_accumulate(acc, input,
-                    secret + nbStripesSoFarPtr[0] * XXH_SECRET_CONSUME_RATE,
-                    nbStripes, f_acc512);
-    *nbStripesSoFarPtr += nbStripes;
+#if (XXH_VECTOR == XXH_NEON)
 
-  }
+/* forward declarations for the scalar routines */
+XXH_FORCE_INLINE void
+XXH3_scalarRound(void* XXH_RESTRICT acc, void const* XXH_RESTRICT input,
+                 void const* XXH_RESTRICT secret, size_t lane);
 
-}
+XXH_FORCE_INLINE void
+XXH3_scalarScrambleRound(void* XXH_RESTRICT acc,
+                         void const* XXH_RESTRICT secret, size_t lane);
 
-/*
- * Both XXH3_64bits_update and XXH3_128bits_update use this routine.
+/*!
+ * @internal
+ * @brief The bulk processing loop for NEON and WASM SIMD128.
+ *
+ * The NEON code path is actually partially scalar when running on AArch64. This
+ * is to optimize the pipelining and can have up to 15% speedup depending on the
+ * CPU, and it also mitigates some GCC codegen issues.
+ *
+ * @see XXH3_NEON_LANES for configuring this and details about this optimization.
+ *
+ * NEON's 32-bit to 64-bit long multiply takes a half vector of 32-bit
+ * integers instead of the other platforms which mask full 64-bit vectors,
+ * so the setup is more complicated than just shifting right.
+ *
+ * Additionally, there is an optimization for 4 lanes at once noted below.
+ *
+ * Since, as stated, the most optimal amount of lanes for Cortexes is 6,
+ * there needs to be *three* versions of the accumulate operation used
+ * for the remaining 2 lanes.
+ *
+ * WASM's SIMD128 uses SIMDe's arm_neon.h polyfill because the intrinsics overlap
+ * nearly perfectly.
  */
-XXH_FORCE_INLINE XXH_errorcode XXH3_update(XXH3_state_t *state,
-                                           const xxh_u8 *input, size_t len,
-                                           XXH3_f_accumulate_512 f_acc512,
-                                           XXH3_f_scrambleAcc    f_scramble) {
-
-  if (input == NULL)
-      #if defined(XXH_ACCEPT_NULL_INPUT_POINTER) && \
-          (XXH_ACCEPT_NULL_INPUT_POINTER >= 1)
-    return XXH_OK;
-      #else
-    return XXH_ERROR;
-      #endif
-
-  {
-
-    const xxh_u8 *const        bEnd = input + len;
-    const unsigned char *const secret =
-        (state->extSecret == NULL) ? state->customSecret : state->extSecret;
-
-    state->totalLen += len;
-    XXH_ASSERT(state->bufferedSize <= XXH3_INTERNALBUFFER_SIZE);
-
-    if (state->bufferedSize + len <=
-        XXH3_INTERNALBUFFER_SIZE) {                   /* fill in tmp buffer */
-      XXH_memcpy(state->buffer + state->bufferedSize, input, len);
-      state->bufferedSize += (XXH32_hash_t)len;
-      return XXH_OK;
 
+XXH_FORCE_INLINE void
+XXH3_accumulate_512_neon( void* XXH_RESTRICT acc,
+                    const void* XXH_RESTRICT input,
+                    const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 15) == 0);
+    XXH_STATIC_ASSERT(XXH3_NEON_LANES > 0 && XXH3_NEON_LANES <= XXH_ACC_NB && XXH3_NEON_LANES % 2 == 0);
+    {   /* GCC for darwin arm64 does not like aliasing here */
+        xxh_aliasing_uint64x2_t* const xacc = (xxh_aliasing_uint64x2_t*) acc;
+        /* We don't use a uint32x4_t pointer because it causes bus errors on ARMv7. */
+        uint8_t const* xinput = (const uint8_t *) input;
+        uint8_t const* xsecret  = (const uint8_t *) secret;
+
+        size_t i;
+#ifdef __wasm_simd128__
+        /*
+         * On WASM SIMD128, Clang emits direct address loads when XXH3_kSecret
+         * is constant propagated, which results in it converting it to this
+         * inside the loop:
+         *
+         *    a = v128.load(XXH3_kSecret +  0 + $secret_offset, offset = 0)
+         *    b = v128.load(XXH3_kSecret + 16 + $secret_offset, offset = 0)
+         *    ...
+         *
+         * This requires a full 32-bit address immediate (and therefore a 6 byte
+         * instruction) as well as an add for each offset.
+         *
+         * Putting an asm guard prevents it from folding (at the cost of losing
+         * the alignment hint), and uses the free offset in `v128.load` instead
+         * of adding secret_offset each time which overall reduces code size by
+         * about a kilobyte and improves performance.
+         */
+        XXH_COMPILER_GUARD(xsecret);
+#endif
+        /* Scalar lanes use the normal scalarRound routine */
+        for (i = XXH3_NEON_LANES; i < XXH_ACC_NB; i++) {
+            XXH3_scalarRound(acc, input, secret, i);
+        }
+        i = 0;
+        /* 4 NEON lanes at a time. */
+        for (; i+1 < XXH3_NEON_LANES / 2; i+=2) {
+            /* data_vec = xinput[i]; */
+            uint64x2_t data_vec_1 = XXH_vld1q_u64(xinput  + (i * 16));
+            uint64x2_t data_vec_2 = XXH_vld1q_u64(xinput  + ((i+1) * 16));
+            /* key_vec  = xsecret[i];  */
+            uint64x2_t key_vec_1  = XXH_vld1q_u64(xsecret + (i * 16));
+            uint64x2_t key_vec_2  = XXH_vld1q_u64(xsecret + ((i+1) * 16));
+            /* data_swap = swap(data_vec) */
+            uint64x2_t data_swap_1 = vextq_u64(data_vec_1, data_vec_1, 1);
+            uint64x2_t data_swap_2 = vextq_u64(data_vec_2, data_vec_2, 1);
+            /* data_key = data_vec ^ key_vec; */
+            uint64x2_t data_key_1 = veorq_u64(data_vec_1, key_vec_1);
+            uint64x2_t data_key_2 = veorq_u64(data_vec_2, key_vec_2);
+
+            /*
+             * If we reinterpret the 64x2 vectors as 32x4 vectors, we can use a
+             * de-interleave operation for 4 lanes in 1 step with `vuzpq_u32` to
+             * get one vector with the low 32 bits of each lane, and one vector
+             * with the high 32 bits of each lane.
+             *
+             * The intrinsic returns a double vector because the original ARMv7-a
+             * instruction modified both arguments in place. AArch64 and SIMD128 emit
+             * two instructions from this intrinsic.
+             *
+             *  [ dk11L | dk11H | dk12L | dk12H ] -> [ dk11L | dk12L | dk21L | dk22L ]
+             *  [ dk21L | dk21H | dk22L | dk22H ] -> [ dk11H | dk12H | dk21H | dk22H ]
+             */
+            uint32x4x2_t unzipped = vuzpq_u32(
+                vreinterpretq_u32_u64(data_key_1),
+                vreinterpretq_u32_u64(data_key_2)
+            );
+            /* data_key_lo = data_key & 0xFFFFFFFF */
+            uint32x4_t data_key_lo = unzipped.val[0];
+            /* data_key_hi = data_key >> 32 */
+            uint32x4_t data_key_hi = unzipped.val[1];
+            /*
+             * Then, we can split the vectors horizontally and multiply which, as for most
+             * widening intrinsics, have a variant that works on both high half vectors
+             * for free on AArch64. A similar instruction is available on SIMD128.
+             *
+             * sum = data_swap + (u64x2) data_key_lo * (u64x2) data_key_hi
+             */
+            uint64x2_t sum_1 = XXH_vmlal_low_u32(data_swap_1, data_key_lo, data_key_hi);
+            uint64x2_t sum_2 = XXH_vmlal_high_u32(data_swap_2, data_key_lo, data_key_hi);
+            /*
+             * Clang reorders
+             *    a += b * c;     // umlal   swap.2d, dkl.2s, dkh.2s
+             *    c += a;         // add     acc.2d, acc.2d, swap.2d
+             * to
+             *    c += a;         // add     acc.2d, acc.2d, swap.2d
+             *    c += b * c;     // umlal   acc.2d, dkl.2s, dkh.2s
+             *
+             * While it would make sense in theory since the addition is faster,
+             * for reasons likely related to umlal being limited to certain NEON
+             * pipelines, this is worse. A compiler guard fixes this.
+             */
+            XXH_COMPILER_GUARD_CLANG_NEON(sum_1);
+            XXH_COMPILER_GUARD_CLANG_NEON(sum_2);
+            /* xacc[i] = acc_vec + sum; */
+            xacc[i]   = vaddq_u64(xacc[i], sum_1);
+            xacc[i+1] = vaddq_u64(xacc[i+1], sum_2);
+        }
+        /* Operate on the remaining NEON lanes 2 at a time. */
+        for (; i < XXH3_NEON_LANES / 2; i++) {
+            /* data_vec = xinput[i]; */
+            uint64x2_t data_vec = XXH_vld1q_u64(xinput  + (i * 16));
+            /* key_vec  = xsecret[i];  */
+            uint64x2_t key_vec  = XXH_vld1q_u64(xsecret + (i * 16));
+            /* acc_vec_2 = swap(data_vec) */
+            uint64x2_t data_swap = vextq_u64(data_vec, data_vec, 1);
+            /* data_key = data_vec ^ key_vec; */
+            uint64x2_t data_key = veorq_u64(data_vec, key_vec);
+            /* For two lanes, just use VMOVN and VSHRN. */
+            /* data_key_lo = data_key & 0xFFFFFFFF; */
+            uint32x2_t data_key_lo = vmovn_u64(data_key);
+            /* data_key_hi = data_key >> 32; */
+            uint32x2_t data_key_hi = vshrn_n_u64(data_key, 32);
+            /* sum = data_swap + (u64x2) data_key_lo * (u64x2) data_key_hi; */
+            uint64x2_t sum = vmlal_u32(data_swap, data_key_lo, data_key_hi);
+            /* Same Clang workaround as before */
+            XXH_COMPILER_GUARD_CLANG_NEON(sum);
+            /* xacc[i] = acc_vec + sum; */
+            xacc[i] = vaddq_u64 (xacc[i], sum);
+        }
     }
+}
+XXH_FORCE_INLINE XXH3_ACCUMULATE_TEMPLATE(neon)
 
-        /* total input is now > XXH3_INTERNALBUFFER_SIZE */
-
-      #define XXH3_INTERNALBUFFER_STRIPES \
-        (XXH3_INTERNALBUFFER_SIZE / XXH_STRIPE_LEN)
-    XXH_STATIC_ASSERT(XXH3_INTERNALBUFFER_SIZE % XXH_STRIPE_LEN ==
-                      0);                                 /* clean multiple */
+XXH_FORCE_INLINE void
+XXH3_scrambleAcc_neon(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 15) == 0);
 
-    /*
-     * Internal buffer is partially filled (always, except at beginning)
-     * Complete it, then consume it.
-     */
-    if (state->bufferedSize) {
+    {   xxh_aliasing_uint64x2_t* xacc       = (xxh_aliasing_uint64x2_t*) acc;
+        uint8_t const* xsecret = (uint8_t const*) secret;
 
-      size_t const loadSize = XXH3_INTERNALBUFFER_SIZE - state->bufferedSize;
-      XXH_memcpy(state->buffer + state->bufferedSize, input, loadSize);
-      input += loadSize;
-      XXH3_consumeStripes(state->acc, &state->nbStripesSoFar,
-                          state->nbStripesPerBlock, state->buffer,
-                          XXH3_INTERNALBUFFER_STRIPES, secret,
-                          state->secretLimit, f_acc512, f_scramble);
-      state->bufferedSize = 0;
+        size_t i;
+        /* WASM uses operator overloads and doesn't need these. */
+#ifndef __wasm_simd128__
+        /* { prime32_1, prime32_1 } */
+        uint32x2_t const kPrimeLo = vdup_n_u32(XXH_PRIME32_1);
+        /* { 0, prime32_1, 0, prime32_1 } */
+        uint32x4_t const kPrimeHi = vreinterpretq_u32_u64(vdupq_n_u64((xxh_u64)XXH_PRIME32_1 << 32));
+#endif
 
+        /* AArch64 uses both scalar and neon at the same time */
+        for (i = XXH3_NEON_LANES; i < XXH_ACC_NB; i++) {
+            XXH3_scalarScrambleRound(acc, secret, i);
+        }
+        for (i=0; i < XXH3_NEON_LANES / 2; i++) {
+            /* xacc[i] ^= (xacc[i] >> 47); */
+            uint64x2_t acc_vec  = xacc[i];
+            uint64x2_t shifted  = vshrq_n_u64(acc_vec, 47);
+            uint64x2_t data_vec = veorq_u64(acc_vec, shifted);
+
+            /* xacc[i] ^= xsecret[i]; */
+            uint64x2_t key_vec  = XXH_vld1q_u64(xsecret + (i * 16));
+            uint64x2_t data_key = veorq_u64(data_vec, key_vec);
+            /* xacc[i] *= XXH_PRIME32_1 */
+#ifdef __wasm_simd128__
+            /* SIMD128 has multiply by u64x2, use it instead of expanding and scalarizing */
+            xacc[i] = data_key * XXH_PRIME32_1;
+#else
+            /*
+             * Expanded version with portable NEON intrinsics
+             *
+             *    lo(x) * lo(y) + (hi(x) * lo(y) << 32)
+             *
+             * prod_hi = hi(data_key) * lo(prime) << 32
+             *
+             * Since we only need 32 bits of this multiply a trick can be used, reinterpreting the vector
+             * as a uint32x4_t and multiplying by { 0, prime, 0, prime } to cancel out the unwanted bits
+             * and avoid the shift.
+             */
+            uint32x4_t prod_hi = vmulq_u32 (vreinterpretq_u32_u64(data_key), kPrimeHi);
+            /* Extract low bits for vmlal_u32  */
+            uint32x2_t data_key_lo = vmovn_u64(data_key);
+            /* xacc[i] = prod_hi + lo(data_key) * XXH_PRIME32_1; */
+            xacc[i] = vmlal_u32(vreinterpretq_u64_u32(prod_hi), data_key_lo, kPrimeLo);
+#endif
+        }
     }
+}
+#endif
 
-    XXH_ASSERT(input < bEnd);
-
-    /* Consume input by a multiple of internal buffer size */
-    if (bEnd - input > XXH3_INTERNALBUFFER_SIZE) {
-
-      const xxh_u8 *const limit = bEnd - XXH3_INTERNALBUFFER_SIZE;
-      do {
-
-        XXH3_consumeStripes(state->acc, &state->nbStripesSoFar,
-                            state->nbStripesPerBlock, input,
-                            XXH3_INTERNALBUFFER_STRIPES, secret,
-                            state->secretLimit, f_acc512, f_scramble);
-        input += XXH3_INTERNALBUFFER_SIZE;
-
-      } while (input < limit);
-
-      /* for last partial stripe */
-      memcpy(state->buffer + sizeof(state->buffer) - XXH_STRIPE_LEN,
-             input - XXH_STRIPE_LEN, XXH_STRIPE_LEN);
-
+#if (XXH_VECTOR == XXH_VSX)
+
+XXH_FORCE_INLINE void
+XXH3_accumulate_512_vsx(  void* XXH_RESTRICT acc,
+                    const void* XXH_RESTRICT input,
+                    const void* XXH_RESTRICT secret)
+{
+    /* presumed aligned */
+    xxh_aliasing_u64x2* const xacc = (xxh_aliasing_u64x2*) acc;
+    xxh_u8 const* const xinput   = (xxh_u8 const*) input;   /* no alignment restriction */
+    xxh_u8 const* const xsecret  = (xxh_u8 const*) secret;    /* no alignment restriction */
+    xxh_u64x2 const v32 = { 32, 32 };
+    size_t i;
+    for (i = 0; i < XXH_STRIPE_LEN / sizeof(xxh_u64x2); i++) {
+        /* data_vec = xinput[i]; */
+        xxh_u64x2 const data_vec = XXH_vec_loadu(xinput + 16*i);
+        /* key_vec = xsecret[i]; */
+        xxh_u64x2 const key_vec  = XXH_vec_loadu(xsecret + 16*i);
+        xxh_u64x2 const data_key = data_vec ^ key_vec;
+        /* shuffled = (data_key << 32) | (data_key >> 32); */
+        xxh_u32x4 const shuffled = (xxh_u32x4)vec_rl(data_key, v32);
+        /* product = ((xxh_u64x2)data_key & 0xFFFFFFFF) * ((xxh_u64x2)shuffled & 0xFFFFFFFF); */
+        xxh_u64x2 const product  = XXH_vec_mulo((xxh_u32x4)data_key, shuffled);
+        /* acc_vec = xacc[i]; */
+        xxh_u64x2 acc_vec        = xacc[i];
+        acc_vec += product;
+
+        /* swap high and low halves */
+#ifdef __s390x__
+        acc_vec += vec_permi(data_vec, data_vec, 2);
+#else
+        acc_vec += vec_xxpermdi(data_vec, data_vec, 2);
+#endif
+        xacc[i] = acc_vec;
     }
-
-    XXH_ASSERT(input < bEnd);
-
-    /* Some remaining input (always) : buffer it */
-    XXH_memcpy(state->buffer, input, (size_t)(bEnd - input));
-    state->bufferedSize = (XXH32_hash_t)(bEnd - input);
-
-  }
-
-  return XXH_OK;
-
 }
-
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API XXH_errorcode XXH3_64bits_update(XXH3_state_t *state,
-                                                const void *input, size_t len) {
-
-  return XXH3_update(state, (const xxh_u8 *)input, len, XXH3_accumulate_512,
-                     XXH3_scrambleAcc);
-
+XXH_FORCE_INLINE XXH3_ACCUMULATE_TEMPLATE(vsx)
+
+XXH_FORCE_INLINE void
+XXH3_scrambleAcc_vsx(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 15) == 0);
+
+    {   xxh_aliasing_u64x2* const xacc = (xxh_aliasing_u64x2*) acc;
+        const xxh_u8* const xsecret = (const xxh_u8*) secret;
+        /* constants */
+        xxh_u64x2 const v32  = { 32, 32 };
+        xxh_u64x2 const v47 = { 47, 47 };
+        xxh_u32x4 const prime = { XXH_PRIME32_1, XXH_PRIME32_1, XXH_PRIME32_1, XXH_PRIME32_1 };
+        size_t i;
+        for (i = 0; i < XXH_STRIPE_LEN / sizeof(xxh_u64x2); i++) {
+            /* xacc[i] ^= (xacc[i] >> 47); */
+            xxh_u64x2 const acc_vec  = xacc[i];
+            xxh_u64x2 const data_vec = acc_vec ^ (acc_vec >> v47);
+
+            /* xacc[i] ^= xsecret[i]; */
+            xxh_u64x2 const key_vec  = XXH_vec_loadu(xsecret + 16*i);
+            xxh_u64x2 const data_key = data_vec ^ key_vec;
+
+            /* xacc[i] *= XXH_PRIME32_1 */
+            /* prod_lo = ((xxh_u64x2)data_key & 0xFFFFFFFF) * ((xxh_u64x2)prime & 0xFFFFFFFF);  */
+            xxh_u64x2 const prod_even  = XXH_vec_mule((xxh_u32x4)data_key, prime);
+            /* prod_hi = ((xxh_u64x2)data_key >> 32) * ((xxh_u64x2)prime >> 32);  */
+            xxh_u64x2 const prod_odd  = XXH_vec_mulo((xxh_u32x4)data_key, prime);
+            xacc[i] = prod_odd + (prod_even << v32);
+    }   }
 }
 
-XXH_FORCE_INLINE void XXH3_digest_long(XXH64_hash_t        *acc,
-                                       const XXH3_state_t  *state,
-                                       const unsigned char *secret) {
-
-  /*
-   * Digest on a local copy. This way, the state remains unaltered, and it can
-   * continue ingesting more input afterwards.
-   */
-  memcpy(acc, state->acc, sizeof(state->acc));
-  if (state->bufferedSize >= XXH_STRIPE_LEN) {
-
-    size_t const nbStripes = (state->bufferedSize - 1) / XXH_STRIPE_LEN;
-    size_t       nbStripesSoFar = state->nbStripesSoFar;
-    XXH3_consumeStripes(acc, &nbStripesSoFar, state->nbStripesPerBlock,
-                        state->buffer, nbStripes, secret, state->secretLimit,
-                        XXH3_accumulate_512, XXH3_scrambleAcc);
-    /* last stripe */
-    XXH3_accumulate_512(acc,
-                        state->buffer + state->bufferedSize - XXH_STRIPE_LEN,
-                        secret + state->secretLimit - XXH_SECRET_LASTACC_START);
-
-  } else {                                 /* bufferedSize < XXH_STRIPE_LEN */
-
-    xxh_u8       lastStripe[XXH_STRIPE_LEN];
-    size_t const catchupSize = XXH_STRIPE_LEN - state->bufferedSize;
-    XXH_ASSERT(state->bufferedSize >
-               0);                   /* there is always some input buffered */
-    memcpy(lastStripe, state->buffer + sizeof(state->buffer) - catchupSize,
-           catchupSize);
-    memcpy(lastStripe + catchupSize, state->buffer, state->bufferedSize);
-    XXH3_accumulate_512(acc, lastStripe,
-                        secret + state->secretLimit - XXH_SECRET_LASTACC_START);
-
-  }
+#endif
 
+#if (XXH_VECTOR == XXH_SVE)
+
+XXH_FORCE_INLINE void
+XXH3_accumulate_512_sve( void* XXH_RESTRICT acc,
+                   const void* XXH_RESTRICT input,
+                   const void* XXH_RESTRICT secret)
+{
+    uint64_t *xacc = (uint64_t *)acc;
+    const uint64_t *xinput = (const uint64_t *)(const void *)input;
+    const uint64_t *xsecret = (const uint64_t *)(const void *)secret;
+    svuint64_t kSwap = sveor_n_u64_z(svptrue_b64(), svindex_u64(0, 1), 1);
+    uint64_t element_count = svcntd();
+    if (element_count >= 8) {
+        svbool_t mask = svptrue_pat_b64(SV_VL8);
+        svuint64_t vacc = svld1_u64(mask, xacc);
+        ACCRND(vacc, 0);
+        svst1_u64(mask, xacc, vacc);
+    } else if (element_count == 2) {   /* sve128 */
+        svbool_t mask = svptrue_pat_b64(SV_VL2);
+        svuint64_t acc0 = svld1_u64(mask, xacc + 0);
+        svuint64_t acc1 = svld1_u64(mask, xacc + 2);
+        svuint64_t acc2 = svld1_u64(mask, xacc + 4);
+        svuint64_t acc3 = svld1_u64(mask, xacc + 6);
+        ACCRND(acc0, 0);
+        ACCRND(acc1, 2);
+        ACCRND(acc2, 4);
+        ACCRND(acc3, 6);
+        svst1_u64(mask, xacc + 0, acc0);
+        svst1_u64(mask, xacc + 2, acc1);
+        svst1_u64(mask, xacc + 4, acc2);
+        svst1_u64(mask, xacc + 6, acc3);
+    } else {
+        svbool_t mask = svptrue_pat_b64(SV_VL4);
+        svuint64_t acc0 = svld1_u64(mask, xacc + 0);
+        svuint64_t acc1 = svld1_u64(mask, xacc + 4);
+        ACCRND(acc0, 0);
+        ACCRND(acc1, 4);
+        svst1_u64(mask, xacc + 0, acc0);
+        svst1_u64(mask, xacc + 4, acc1);
+    }
 }
 
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_digest(const XXH3_state_t *state) {
-
-  const unsigned char *const secret =
-      (state->extSecret == NULL) ? state->customSecret : state->extSecret;
-  if (state->totalLen > XXH3_MIDSIZE_MAX) {
-
-    XXH_ALIGN(XXH_ACC_ALIGN) XXH64_hash_t acc[XXH_ACC_NB];
-    XXH3_digest_long(acc, state, secret);
-    return XXH3_mergeAccs(acc, secret + XXH_SECRET_MERGEACCS_START,
-                          (xxh_u64)state->totalLen * XXH_PRIME64_1);
-
-  }
-
-  /* totalLen <= XXH3_MIDSIZE_MAX: digesting a short input */
-  if (state->seed)
-    return XXH3_64bits_withSeed(state->buffer, (size_t)state->totalLen,
-                                state->seed);
-  return XXH3_64bits_withSecret(state->buffer, (size_t)(state->totalLen),
-                                secret, state->secretLimit + XXH_STRIPE_LEN);
-
+XXH_FORCE_INLINE void
+XXH3_accumulate_sve(xxh_u64* XXH_RESTRICT acc,
+               const xxh_u8* XXH_RESTRICT input,
+               const xxh_u8* XXH_RESTRICT secret,
+               size_t nbStripes)
+{
+    if (nbStripes != 0) {
+        uint64_t *xacc = (uint64_t *)acc;
+        const uint64_t *xinput = (const uint64_t *)(const void *)input;
+        const uint64_t *xsecret = (const uint64_t *)(const void *)secret;
+        svuint64_t kSwap = sveor_n_u64_z(svptrue_b64(), svindex_u64(0, 1), 1);
+        uint64_t element_count = svcntd();
+        if (element_count >= 8) {
+            svbool_t mask = svptrue_pat_b64(SV_VL8);
+            svuint64_t vacc = svld1_u64(mask, xacc + 0);
+            do {
+                /* svprfd(svbool_t, void *, enum svfprop); */
+                svprfd(mask, xinput + 128, SV_PLDL1STRM);
+                ACCRND(vacc, 0);
+                xinput += 8;
+                xsecret += 1;
+                nbStripes--;
+           } while (nbStripes != 0);
+
+           svst1_u64(mask, xacc + 0, vacc);
+        } else if (element_count == 2) { /* sve128 */
+            svbool_t mask = svptrue_pat_b64(SV_VL2);
+            svuint64_t acc0 = svld1_u64(mask, xacc + 0);
+            svuint64_t acc1 = svld1_u64(mask, xacc + 2);
+            svuint64_t acc2 = svld1_u64(mask, xacc + 4);
+            svuint64_t acc3 = svld1_u64(mask, xacc + 6);
+            do {
+                svprfd(mask, xinput + 128, SV_PLDL1STRM);
+                ACCRND(acc0, 0);
+                ACCRND(acc1, 2);
+                ACCRND(acc2, 4);
+                ACCRND(acc3, 6);
+                xinput += 8;
+                xsecret += 1;
+                nbStripes--;
+           } while (nbStripes != 0);
+
+           svst1_u64(mask, xacc + 0, acc0);
+           svst1_u64(mask, xacc + 2, acc1);
+           svst1_u64(mask, xacc + 4, acc2);
+           svst1_u64(mask, xacc + 6, acc3);
+        } else {
+            svbool_t mask = svptrue_pat_b64(SV_VL4);
+            svuint64_t acc0 = svld1_u64(mask, xacc + 0);
+            svuint64_t acc1 = svld1_u64(mask, xacc + 4);
+            do {
+                svprfd(mask, xinput + 128, SV_PLDL1STRM);
+                ACCRND(acc0, 0);
+                ACCRND(acc1, 4);
+                xinput += 8;
+                xsecret += 1;
+                nbStripes--;
+           } while (nbStripes != 0);
+
+           svst1_u64(mask, xacc + 0, acc0);
+           svst1_u64(mask, xacc + 4, acc1);
+       }
+    }
 }
 
-      #define XXH_MIN(x, y) (((x) > (y)) ? (y) : (x))
-
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API void XXH3_generateSecret(void       *secretBuffer,
-                                        const void *customSeed,
-                                        size_t      customSeedSize) {
-
-  XXH_ASSERT(secretBuffer != NULL);
-  if (customSeedSize == 0) {
-
-    memcpy(secretBuffer, XXH3_kSecret, XXH_SECRET_DEFAULT_SIZE);
-    return;
-
-  }
-
-  XXH_ASSERT(customSeed != NULL);
+#endif
 
-  {
+/* scalar variants - universal */
 
-    size_t const       segmentSize = sizeof(XXH128_hash_t);
-    size_t const       nbSegments = XXH_SECRET_DEFAULT_SIZE / segmentSize;
-    XXH128_canonical_t scrambler;
-    XXH64_hash_t       seeds[12];
-    size_t             segnb;
-    XXH_ASSERT(nbSegments == 12);
-    XXH_ASSERT(segmentSize * nbSegments ==
-               XXH_SECRET_DEFAULT_SIZE);                  /* exact multiple */
-    XXH128_canonicalFromHash(&scrambler, XXH128(customSeed, customSeedSize, 0));
+#if defined(__aarch64__) && (defined(__GNUC__) || defined(__clang__))
+/*
+ * In XXH3_scalarRound(), GCC and Clang have a similar codegen issue, where they
+ * emit an excess mask and a full 64-bit multiply-add (MADD X-form).
+ *
+ * While this might not seem like much, as AArch64 is a 64-bit architecture, only
+ * big Cortex designs have a full 64-bit multiplier.
+ *
+ * On the little cores, the smaller 32-bit multiplier is used, and full 64-bit
+ * multiplies expand to 2-3 multiplies in microcode. This has a major penalty
+ * of up to 4 latency cycles and 2 stall cycles in the multiply pipeline.
+ *
+ * Thankfully, AArch64 still provides the 32-bit long multiply-add (UMADDL) which does
+ * not have this penalty and does the mask automatically.
+ */
+XXH_FORCE_INLINE xxh_u64
+XXH_mult32to64_add64(xxh_u64 lhs, xxh_u64 rhs, xxh_u64 acc)
+{
+    xxh_u64 ret;
+    /* note: %x = 64-bit register, %w = 32-bit register */
+    __asm__("umaddl %x0, %w1, %w2, %x3" : "=r" (ret) : "r" (lhs), "r" (rhs), "r" (acc));
+    return ret;
+}
+#else
+XXH_FORCE_INLINE xxh_u64
+XXH_mult32to64_add64(xxh_u64 lhs, xxh_u64 rhs, xxh_u64 acc)
+{
+    return XXH_mult32to64((xxh_u32)lhs, (xxh_u32)rhs) + acc;
+}
+#endif
 
-    /*
-     * Copy customSeed to seeds[], truncating or repeating as necessary.
-     */
+/*!
+ * @internal
+ * @brief Scalar round for @ref XXH3_accumulate_512_scalar().
+ *
+ * This is extracted to its own function because the NEON path uses a combination
+ * of NEON and scalar.
+ */
+XXH_FORCE_INLINE void
+XXH3_scalarRound(void* XXH_RESTRICT acc,
+                 void const* XXH_RESTRICT input,
+                 void const* XXH_RESTRICT secret,
+                 size_t lane)
+{
+    xxh_u64* xacc = (xxh_u64*) acc;
+    xxh_u8 const* xinput  = (xxh_u8 const*) input;
+    xxh_u8 const* xsecret = (xxh_u8 const*) secret;
+    XXH_ASSERT(lane < XXH_ACC_NB);
+    XXH_ASSERT(((size_t)acc & (XXH_ACC_ALIGN-1)) == 0);
     {
-
-      size_t toFill = XXH_MIN(customSeedSize, sizeof(seeds));
-      size_t filled = toFill;
-      memcpy(seeds, customSeed, toFill);
-      while (filled < sizeof(seeds)) {
-
-        toFill = XXH_MIN(filled, sizeof(seeds) - filled);
-        memcpy((char *)seeds + filled, seeds, toFill);
-        filled += toFill;
-
-      }
-
+        xxh_u64 const data_val = XXH_readLE64(xinput + lane * 8);
+        xxh_u64 const data_key = data_val ^ XXH_readLE64(xsecret + lane * 8);
+        xacc[lane ^ 1] += data_val; /* swap adjacent lanes */
+        xacc[lane] = XXH_mult32to64_add64(data_key /* & 0xFFFFFFFF */, data_key >> 32, xacc[lane]);
     }
+}
 
-    /* generate secret */
-    memcpy(secretBuffer, &scrambler, sizeof(scrambler));
-    for (segnb = 1; segnb < nbSegments; segnb++) {
-
-      size_t const       segmentStart = segnb * segmentSize;
-      XXH128_canonical_t segment;
-      XXH128_canonicalFromHash(&segment,
-                               XXH128(&scrambler, sizeof(scrambler),
-                                      XXH_readLE64(seeds + segnb) + segnb));
-      memcpy((char *)secretBuffer + segmentStart, &segment, sizeof(segment));
-
+/*!
+ * @internal
+ * @brief Processes a 64 byte block of data using the scalar path.
+ */
+XXH_FORCE_INLINE void
+XXH3_accumulate_512_scalar(void* XXH_RESTRICT acc,
+                     const void* XXH_RESTRICT input,
+                     const void* XXH_RESTRICT secret)
+{
+    size_t i;
+    /* ARM GCC refuses to unroll this loop, resulting in a 24% slowdown on ARMv6. */
+#if defined(__GNUC__) && !defined(__clang__) \
+  && (defined(__arm__) || defined(__thumb2__)) \
+  && defined(__ARM_FEATURE_UNALIGNED) /* no unaligned access just wastes bytes */ \
+  && XXH_SIZE_OPT <= 0
+#  pragma GCC unroll 8
+#endif
+    for (i=0; i < XXH_ACC_NB; i++) {
+        XXH3_scalarRound(acc, input, secret, i);
     }
-
-  }
-
 }
+XXH_FORCE_INLINE XXH3_ACCUMULATE_TEMPLATE(scalar)
 
-/* ==========================================
- * XXH3 128 bits (a.k.a XXH128)
- * ==========================================
- * XXH3's 128-bit variant has better mixing and strength than the 64-bit
- * variant, even without counting the significantly larger output size.
- *
- * For example, extra steps are taken to avoid the seed-dependent collisions
- * in 17-240 byte inputs (See XXH3_mix16B and XXH128_mix32B).
- *
- * This strength naturally comes at the cost of some speed, especially on short
- * lengths. Note that longer hashes are about as fast as the 64-bit version
- * due to it using only a slight modification of the 64-bit loop.
+/*!
+ * @internal
+ * @brief Scalar scramble step for @ref XXH3_scrambleAcc_scalar().
  *
- * XXH128 is also more oriented towards 64-bit machines. It is still extremely
- * fast for a _128-bit_ hash on 32-bit (it usually clears XXH64).
+ * This is extracted to its own function because the NEON path uses a combination
+ * of NEON and scalar.
  */
-
-XXH_FORCE_INLINE XXH128_hash_t XXH3_len_1to3_128b(const xxh_u8 *input,
-                                                  size_t        len,
-                                                  const xxh_u8 *secret,
-                                                  XXH64_hash_t  seed) {
-
-  /* A doubled version of 1to3_64b with different constants. */
-  XXH_ASSERT(input != NULL);
-  XXH_ASSERT(1 <= len && len <= 3);
-  XXH_ASSERT(secret != NULL);
-  /*
-   * len = 1: combinedl = { input[0], 0x01, input[0], input[0] }
-   * len = 2: combinedl = { input[1], 0x02, input[0], input[1] }
-   * len = 3: combinedl = { input[2], 0x03, input[0], input[1] }
-   */
-  {
-
-    xxh_u8 const  c1 = input[0];
-    xxh_u8 const  c2 = input[len >> 1];
-    xxh_u8 const  c3 = input[len - 1];
-    xxh_u32 const combinedl = ((xxh_u32)c1 << 16) | ((xxh_u32)c2 << 24) |
-                              ((xxh_u32)c3 << 0) | ((xxh_u32)len << 8);
-    xxh_u32 const combinedh = XXH_rotl32(XXH_swap32(combinedl), 13);
-    xxh_u64 const bitflipl =
-        (XXH_readLE32(secret) ^ XXH_readLE32(secret + 4)) + seed;
-    xxh_u64 const bitfliph =
-        (XXH_readLE32(secret + 8) ^ XXH_readLE32(secret + 12)) - seed;
-    xxh_u64 const keyed_lo = (xxh_u64)combinedl ^ bitflipl;
-    xxh_u64 const keyed_hi = (xxh_u64)combinedh ^ bitfliph;
-    XXH128_hash_t h128;
-    h128.low64 = XXH64_avalanche(keyed_lo);
-    h128.high64 = XXH64_avalanche(keyed_hi);
-    return h128;
-
-  }
-
-}
-
-XXH_FORCE_INLINE XXH128_hash_t XXH3_len_4to8_128b(const xxh_u8 *input,
-                                                  size_t        len,
-                                                  const xxh_u8 *secret,
-                                                  XXH64_hash_t  seed) {
-
-  XXH_ASSERT(input != NULL);
-  XXH_ASSERT(secret != NULL);
-  XXH_ASSERT(4 <= len && len <= 8);
-  seed ^= (xxh_u64)XXH_swap32((xxh_u32)seed) << 32;
-  {
-
-    xxh_u32 const input_lo = XXH_readLE32(input);
-    xxh_u32 const input_hi = XXH_readLE32(input + len - 4);
-    xxh_u64 const input_64 = input_lo + ((xxh_u64)input_hi << 32);
-    xxh_u64 const bitflip =
-        (XXH_readLE64(secret + 16) ^ XXH_readLE64(secret + 24)) + seed;
-    xxh_u64 const keyed = input_64 ^ bitflip;
-
-    /* Shift len to the left to ensure it is even, this avoids even multiplies.
-     */
-    XXH128_hash_t m128 = XXH_mult64to128(keyed, XXH_PRIME64_1 + (len << 2));
-
-    m128.high64 += (m128.low64 << 1);
-    m128.low64 ^= (m128.high64 >> 3);
-
-    m128.low64 = XXH_xorshift64(m128.low64, 35);
-    m128.low64 *= 0x9FB21C651E98DF25ULL;
-    m128.low64 = XXH_xorshift64(m128.low64, 28);
-    m128.high64 = XXH3_avalanche(m128.high64);
-    return m128;
-
-  }
-
+XXH_FORCE_INLINE void
+XXH3_scalarScrambleRound(void* XXH_RESTRICT acc,
+                         void const* XXH_RESTRICT secret,
+                         size_t lane)
+{
+    xxh_u64* const xacc = (xxh_u64*) acc;   /* presumed aligned */
+    const xxh_u8* const xsecret = (const xxh_u8*) secret;   /* no alignment restriction */
+    XXH_ASSERT((((size_t)acc) & (XXH_ACC_ALIGN-1)) == 0);
+    XXH_ASSERT(lane < XXH_ACC_NB);
+    {
+        xxh_u64 const key64 = XXH_readLE64(xsecret + lane * 8);
+        xxh_u64 acc64 = xacc[lane];
+        acc64 = XXH_xorshift64(acc64, 47);
+        acc64 ^= key64;
+        acc64 *= XXH_PRIME32_1;
+        xacc[lane] = acc64;
+    }
 }
 
-XXH_FORCE_INLINE XXH128_hash_t XXH3_len_9to16_128b(const xxh_u8 *input,
-                                                   size_t        len,
-                                                   const xxh_u8 *secret,
-                                                   XXH64_hash_t  seed) {
-
-  XXH_ASSERT(input != NULL);
-  XXH_ASSERT(secret != NULL);
-  XXH_ASSERT(9 <= len && len <= 16);
-  {
+/*!
+ * @internal
+ * @brief Scrambles the accumulators after a large chunk has been read
+ */
+XXH_FORCE_INLINE void
+XXH3_scrambleAcc_scalar(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+    size_t i;
+    for (i=0; i < XXH_ACC_NB; i++) {
+        XXH3_scalarScrambleRound(acc, secret, i);
+    }
+}
 
-    xxh_u64 const bitflipl =
-        (XXH_readLE64(secret + 32) ^ XXH_readLE64(secret + 40)) - seed;
-    xxh_u64 const bitfliph =
-        (XXH_readLE64(secret + 48) ^ XXH_readLE64(secret + 56)) + seed;
-    xxh_u64 const input_lo = XXH_readLE64(input);
-    xxh_u64       input_hi = XXH_readLE64(input + len - 8);
-    XXH128_hash_t m128 =
-        XXH_mult64to128(input_lo ^ input_hi ^ bitflipl, XXH_PRIME64_1);
+XXH_FORCE_INLINE void
+XXH3_initCustomSecret_scalar(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
+{
     /*
-     * Put len in the middle of m128 to ensure that the length gets mixed to
-     * both the low and high bits in the 128x64 multiply below.
+     * We need a separate pointer for the hack below,
+     * which requires a non-const pointer.
+     * Any decent compiler will optimize this out otherwise.
      */
-    m128.low64 += (xxh_u64)(len - 1) << 54;
-    input_hi ^= bitfliph;
+    const xxh_u8* kSecretPtr = XXH3_kSecret;
+    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 15) == 0);
+
+#if defined(__GNUC__) && defined(__aarch64__)
     /*
-     * Add the high 32 bits of input_hi to the high 32 bits of m128, then
-     * add the long product of the low 32 bits of input_hi and XXH_PRIME32_2 to
-     * the high 64 bits of m128.
+     * UGLY HACK:
+     * GCC and Clang generate a bunch of MOV/MOVK pairs for aarch64, and they are
+     * placed sequentially, in order, at the top of the unrolled loop.
      *
-     * The best approach to this operation is different on 32-bit and 64-bit.
+     * While MOVK is great for generating constants (2 cycles for a 64-bit
+     * constant compared to 4 cycles for LDR), it fights for bandwidth with
+     * the arithmetic instructions.
+     *
+     *   I   L   S
+     * MOVK
+     * MOVK
+     * MOVK
+     * MOVK
+     * ADD
+     * SUB      STR
+     *          STR
+     * By forcing loads from memory (as the asm line causes the compiler to assume
+     * that XXH3_kSecretPtr has been changed), the pipelines are used more
+     * efficiently:
+     *   I   L   S
+     *      LDR
+     *  ADD LDR
+     *  SUB     STR
+     *          STR
+     *
+     * See XXH3_NEON_LANES for details on the pipsline.
+     *
+     * XXH3_64bits_withSeed, len == 256, Snapdragon 835
+     *   without hack: 2654.4 MB/s
+     *   with hack:    3202.9 MB/s
      */
-    if (sizeof(void *) < sizeof(xxh_u64)) {                       /* 32-bit */
-      /*
-       * 32-bit optimized version, which is more readable.
-       *
-       * On 32-bit, it removes an ADC and delays a dependency between the two
-       * halves of m128.high64, but it generates an extra mask on 64-bit.
-       */
-      m128.high64 += (input_hi & 0xFFFFFFFF00000000ULL) +
-                     XXH_mult32to64((xxh_u32)input_hi, XXH_PRIME32_2);
-
-    } else {
-
-      /*
-       * 64-bit optimized (albeit more confusing) version.
-       *
-       * Uses some properties of addition and multiplication to remove the mask:
-       *
-       * Let:
-       *    a = input_hi.lo = (input_hi & 0x00000000FFFFFFFF)
-       *    b = input_hi.hi = (input_hi & 0xFFFFFFFF00000000)
-       *    c = XXH_PRIME32_2
-       *
-       *    a + (b * c)
-       * Inverse Property: x + y - x == y
-       *    a + (b * (1 + c - 1))
-       * Distributive Property: x * (y + z) == (x * y) + (x * z)
-       *    a + (b * 1) + (b * (c - 1))
-       * Identity Property: x * 1 == x
-       *    a + b + (b * (c - 1))
-       *
-       * Substitute a, b, and c:
-       *    input_hi.hi + input_hi.lo + ((xxh_u64)input_hi.lo * (XXH_PRIME32_2 -
-       * 1))
-       *
-       * Since input_hi.hi + input_hi.lo == input_hi, we get this:
-       *    input_hi + ((xxh_u64)input_hi.lo * (XXH_PRIME32_2 - 1))
-       */
-      m128.high64 +=
-          input_hi + XXH_mult32to64((xxh_u32)input_hi, XXH_PRIME32_2 - 1);
-
-    }
-
-    /* m128 ^= XXH_swap64(m128 >> 64); */
-    m128.low64 ^= XXH_swap64(m128.high64);
-
-    {                      /* 128x64 multiply: h128 = m128 * XXH_PRIME64_2; */
-      XXH128_hash_t h128 = XXH_mult64to128(m128.low64, XXH_PRIME64_2);
-      h128.high64 += m128.high64 * XXH_PRIME64_2;
-
-      h128.low64 = XXH3_avalanche(h128.low64);
-      h128.high64 = XXH3_avalanche(h128.high64);
-      return h128;
-
-    }
-
-  }
-
+    XXH_COMPILER_GUARD(kSecretPtr);
+#endif
+    {   int const nbRounds = XXH_SECRET_DEFAULT_SIZE / 16;
+        int i;
+        for (i=0; i < nbRounds; i++) {
+            /*
+             * The asm hack causes the compiler to assume that kSecretPtr aliases with
+             * customSecret, and on aarch64, this prevented LDP from merging two
+             * loads together for free. Putting the loads together before the stores
+             * properly generates LDP.
+             */
+            xxh_u64 lo = XXH_readLE64(kSecretPtr + 16*i)     + seed64;
+            xxh_u64 hi = XXH_readLE64(kSecretPtr + 16*i + 8) - seed64;
+            XXH_writeLE64((xxh_u8*)customSecret + 16*i,     lo);
+            XXH_writeLE64((xxh_u8*)customSecret + 16*i + 8, hi);
+    }   }
 }
 
-/*
- * Assumption: `secret` size is >= XXH3_SECRET_SIZE_MIN
- */
-XXH_FORCE_INLINE XXH128_hash_t XXH3_len_0to16_128b(const xxh_u8 *input,
-                                                   size_t        len,
-                                                   const xxh_u8 *secret,
-                                                   XXH64_hash_t  seed) {
-
-  XXH_ASSERT(len <= 16);
-  {
 
-    if (len > 8) return XXH3_len_9to16_128b(input, len, secret, seed);
-    if (len >= 4) return XXH3_len_4to8_128b(input, len, secret, seed);
-    if (len) return XXH3_len_1to3_128b(input, len, secret, seed);
-    {
-
-      XXH128_hash_t h128;
-      xxh_u64 const bitflipl =
-          XXH_readLE64(secret + 64) ^ XXH_readLE64(secret + 72);
-      xxh_u64 const bitfliph =
-          XXH_readLE64(secret + 80) ^ XXH_readLE64(secret + 88);
-      h128.low64 = XXH64_avalanche(seed ^ bitflipl);
-      h128.high64 = XXH64_avalanche(seed ^ bitfliph);
-      return h128;
-
-    }
+typedef void (*XXH3_f_accumulate)(xxh_u64* XXH_RESTRICT, const xxh_u8* XXH_RESTRICT, const xxh_u8* XXH_RESTRICT, size_t);
+typedef void (*XXH3_f_scrambleAcc)(void* XXH_RESTRICT, const void*);
+typedef void (*XXH3_f_initCustomSecret)(void* XXH_RESTRICT, xxh_u64);
 
-  }
 
-}
+#if (XXH_VECTOR == XXH_AVX512)
 
-/*
- * A bit slower than XXH3_mix16B, but handles multiply by zero better.
- */
-XXH_FORCE_INLINE XXH128_hash_t XXH128_mix32B(XXH128_hash_t acc,
-                                             const xxh_u8 *input_1,
-                                             const xxh_u8 *input_2,
-                                             const xxh_u8 *secret,
-                                             XXH64_hash_t  seed) {
+#define XXH3_accumulate_512 XXH3_accumulate_512_avx512
+#define XXH3_accumulate     XXH3_accumulate_avx512
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_avx512
+#define XXH3_initCustomSecret XXH3_initCustomSecret_avx512
 
-  acc.low64 += XXH3_mix16B(input_1, secret + 0, seed);
-  acc.low64 ^= XXH_readLE64(input_2) + XXH_readLE64(input_2 + 8);
-  acc.high64 += XXH3_mix16B(input_2, secret + 16, seed);
-  acc.high64 ^= XXH_readLE64(input_1) + XXH_readLE64(input_1 + 8);
-  return acc;
+#elif (XXH_VECTOR == XXH_AVX2)
 
-}
+#define XXH3_accumulate_512 XXH3_accumulate_512_avx2
+#define XXH3_accumulate     XXH3_accumulate_avx2
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_avx2
+#define XXH3_initCustomSecret XXH3_initCustomSecret_avx2
 
-XXH_FORCE_INLINE XXH128_hash_t XXH3_len_17to128_128b(
-    const xxh_u8 *XXH_RESTRICT input, size_t len,
-    const xxh_u8 *XXH_RESTRICT secret, size_t secretSize, XXH64_hash_t seed) {
+#elif (XXH_VECTOR == XXH_SSE2)
 
-  XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
-  (void)secretSize;
-  XXH_ASSERT(16 < len && len <= 128);
+#define XXH3_accumulate_512 XXH3_accumulate_512_sse2
+#define XXH3_accumulate     XXH3_accumulate_sse2
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_sse2
+#define XXH3_initCustomSecret XXH3_initCustomSecret_sse2
 
-  {
+#elif (XXH_VECTOR == XXH_NEON)
 
-    XXH128_hash_t acc;
-    acc.low64 = len * XXH_PRIME64_1;
-    acc.high64 = 0;
-    if (len > 32) {
+#define XXH3_accumulate_512 XXH3_accumulate_512_neon
+#define XXH3_accumulate     XXH3_accumulate_neon
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_neon
+#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
 
-      if (len > 64) {
+#elif (XXH_VECTOR == XXH_VSX)
 
-        if (len > 96) {
+#define XXH3_accumulate_512 XXH3_accumulate_512_vsx
+#define XXH3_accumulate     XXH3_accumulate_vsx
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_vsx
+#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
 
-          acc = XXH128_mix32B(acc, input + 48, input + len - 64, secret + 96,
-                              seed);
+#elif (XXH_VECTOR == XXH_SVE)
+#define XXH3_accumulate_512 XXH3_accumulate_512_sve
+#define XXH3_accumulate     XXH3_accumulate_sve
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_scalar
+#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
 
-        }
+#else /* scalar */
 
-        acc =
-            XXH128_mix32B(acc, input + 32, input + len - 48, secret + 64, seed);
+#define XXH3_accumulate_512 XXH3_accumulate_512_scalar
+#define XXH3_accumulate     XXH3_accumulate_scalar
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_scalar
+#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
 
-      }
+#endif
 
-      acc = XXH128_mix32B(acc, input + 16, input + len - 32, secret + 32, seed);
+#if XXH_SIZE_OPT >= 1 /* don't do SIMD for initialization */
+#  undef XXH3_initCustomSecret
+#  define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
+#endif
 
-    }
+XXH_FORCE_INLINE void
+XXH3_hashLong_internal_loop(xxh_u64* XXH_RESTRICT acc,
+                      const xxh_u8* XXH_RESTRICT input, size_t len,
+                      const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
+                            XXH3_f_accumulate f_acc,
+                            XXH3_f_scrambleAcc f_scramble)
+{
+    size_t const nbStripesPerBlock = (secretSize - XXH_STRIPE_LEN) / XXH_SECRET_CONSUME_RATE;
+    size_t const block_len = XXH_STRIPE_LEN * nbStripesPerBlock;
+    size_t const nb_blocks = (len - 1) / block_len;
 
-    acc = XXH128_mix32B(acc, input, input + len - 16, secret, seed);
-    {
+    size_t n;
 
-      XXH128_hash_t h128;
-      h128.low64 = acc.low64 + acc.high64;
-      h128.high64 = (acc.low64 * XXH_PRIME64_1) + (acc.high64 * XXH_PRIME64_4) +
-                    ((len - seed) * XXH_PRIME64_2);
-      h128.low64 = XXH3_avalanche(h128.low64);
-      h128.high64 = (XXH64_hash_t)0 - XXH3_avalanche(h128.high64);
-      return h128;
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
 
+    for (n = 0; n < nb_blocks; n++) {
+        f_acc(acc, input + n*block_len, secret, nbStripesPerBlock);
+        f_scramble(acc, secret + secretSize - XXH_STRIPE_LEN);
     }
 
-  }
+    /* last partial block */
+    XXH_ASSERT(len > XXH_STRIPE_LEN);
+    {   size_t const nbStripes = ((len - 1) - (block_len * nb_blocks)) / XXH_STRIPE_LEN;
+        XXH_ASSERT(nbStripes <= (secretSize / XXH_SECRET_CONSUME_RATE));
+        f_acc(acc, input + nb_blocks*block_len, secret, nbStripes);
 
+        /* last stripe */
+        {   const xxh_u8* const p = input + len - XXH_STRIPE_LEN;
+#define XXH_SECRET_LASTACC_START 7  /* not aligned on 8, last secret is different from acc & scrambler */
+            XXH3_accumulate_512(acc, p, secret + secretSize - XXH_STRIPE_LEN - XXH_SECRET_LASTACC_START);
+    }   }
 }
 
-XXH_NO_INLINE XXH128_hash_t XXH3_len_129to240_128b(
-    const xxh_u8 *XXH_RESTRICT input, size_t len,
-    const xxh_u8 *XXH_RESTRICT secret, size_t secretSize, XXH64_hash_t seed) {
-
-  XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
-  (void)secretSize;
-  XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX);
+XXH_FORCE_INLINE xxh_u64
+XXH3_mix2Accs(const xxh_u64* XXH_RESTRICT acc, const xxh_u8* XXH_RESTRICT secret)
+{
+    return XXH3_mul128_fold64(
+               acc[0] ^ XXH_readLE64(secret),
+               acc[1] ^ XXH_readLE64(secret+8) );
+}
 
-  {
+static XXH64_hash_t
+XXH3_mergeAccs(const xxh_u64* XXH_RESTRICT acc, const xxh_u8* XXH_RESTRICT secret, xxh_u64 start)
+{
+    xxh_u64 result64 = start;
+    size_t i = 0;
 
-    XXH128_hash_t acc;
-    int const     nbRounds = (int)len / 32;
-    int           i;
-    acc.low64 = len * XXH_PRIME64_1;
-    acc.high64 = 0;
     for (i = 0; i < 4; i++) {
-
-      acc = XXH128_mix32B(acc, input + (32 * i), input + (32 * i) + 16,
-                          secret + (32 * i), seed);
-
+        result64 += XXH3_mix2Accs(acc+2*i, secret + 16*i);
+#if defined(__clang__)                                /* Clang */ \
+    && (defined(__arm__) || defined(__thumb__))       /* ARMv7 */ \
+    && (defined(__ARM_NEON) || defined(__ARM_NEON__)) /* NEON */  \
+    && !defined(XXH_ENABLE_AUTOVECTORIZE)             /* Define to disable */
+        /*
+         * UGLY HACK:
+         * Prevent autovectorization on Clang ARMv7-a. Exact same problem as
+         * the one in XXH3_len_129to240_64b. Speeds up shorter keys > 240b.
+         * XXH3_64bits, len == 256, Snapdragon 835:
+         *   without hack: 2063.7 MB/s
+         *   with hack:    2560.7 MB/s
+         */
+        XXH_COMPILER_GUARD(result64);
+#endif
     }
 
-    acc.low64 = XXH3_avalanche(acc.low64);
-    acc.high64 = XXH3_avalanche(acc.high64);
-    XXH_ASSERT(nbRounds >= 4);
-    for (i = 4; i < nbRounds; i++) {
-
-      acc = XXH128_mix32B(acc, input + (32 * i), input + (32 * i) + 16,
-                          secret + XXH3_MIDSIZE_STARTOFFSET + (32 * (i - 4)),
-                          seed);
+    return XXH3_avalanche(result64);
+}
 
-    }
+#define XXH3_INIT_ACC { XXH_PRIME32_3, XXH_PRIME64_1, XXH_PRIME64_2, XXH_PRIME64_3, \
+                        XXH_PRIME64_4, XXH_PRIME32_2, XXH_PRIME64_5, XXH_PRIME32_1 }
 
-    /* last bytes */
-    acc = XXH128_mix32B(
-        acc, input + len - 16, input + len - 32,
-        secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET - 16,
-        0ULL - seed);
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_hashLong_64b_internal(const void* XXH_RESTRICT input, size_t len,
+                           const void* XXH_RESTRICT secret, size_t secretSize,
+                           XXH3_f_accumulate f_acc,
+                           XXH3_f_scrambleAcc f_scramble)
+{
+    XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[XXH_ACC_NB] = XXH3_INIT_ACC;
 
-    {
+    XXH3_hashLong_internal_loop(acc, (const xxh_u8*)input, len, (const xxh_u8*)secret, secretSize, f_acc, f_scramble);
 
-      XXH128_hash_t h128;
-      h128.low64 = acc.low64 + acc.high64;
-      h128.high64 = (acc.low64 * XXH_PRIME64_1) + (acc.high64 * XXH_PRIME64_4) +
-                    ((len - seed) * XXH_PRIME64_2);
-      h128.low64 = XXH3_avalanche(h128.low64);
-      h128.high64 = (XXH64_hash_t)0 - XXH3_avalanche(h128.high64);
-      return h128;
+    /* converge into final hash */
+    XXH_STATIC_ASSERT(sizeof(acc) == 64);
+    /* do not align on 8, so that the secret is different from the accumulator */
+#define XXH_SECRET_MERGEACCS_START 11
+    XXH_ASSERT(secretSize >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
+    return XXH3_mergeAccs(acc, (const xxh_u8*)secret + XXH_SECRET_MERGEACCS_START, (xxh_u64)len * XXH_PRIME64_1);
+}
 
-    }
+/*
+ * It's important for performance to transmit secret's size (when it's static)
+ * so that the compiler can properly optimize the vectorized loop.
+ * This makes a big performance difference for "medium" keys (<1 KB) when using AVX instruction set.
+ * When the secret size is unknown, or on GCC 12 where the mix of NO_INLINE and FORCE_INLINE
+ * breaks -Og, this is XXH_NO_INLINE.
+ */
+XXH3_WITH_SECRET_INLINE XXH64_hash_t
+XXH3_hashLong_64b_withSecret(const void* XXH_RESTRICT input, size_t len,
+                             XXH64_hash_t seed64, const xxh_u8* XXH_RESTRICT secret, size_t secretLen)
+{
+    (void)seed64;
+    return XXH3_hashLong_64b_internal(input, len, secret, secretLen, XXH3_accumulate, XXH3_scrambleAcc);
+}
 
-  }
+/*
+ * It's preferable for performance that XXH3_hashLong is not inlined,
+ * as it results in a smaller function for small data, easier to the instruction cache.
+ * Note that inside this no_inline function, we do inline the internal loop,
+ * and provide a statically defined secret size to allow optimization of vector loop.
+ */
+XXH_NO_INLINE XXH_PUREF XXH64_hash_t
+XXH3_hashLong_64b_default(const void* XXH_RESTRICT input, size_t len,
+                          XXH64_hash_t seed64, const xxh_u8* XXH_RESTRICT secret, size_t secretLen)
+{
+    (void)seed64; (void)secret; (void)secretLen;
+    return XXH3_hashLong_64b_internal(input, len, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_accumulate, XXH3_scrambleAcc);
+}
 
+/*
+ * XXH3_hashLong_64b_withSeed():
+ * Generate a custom key based on alteration of default XXH3_kSecret with the seed,
+ * and then use this key for long mode hashing.
+ *
+ * This operation is decently fast but nonetheless costs a little bit of time.
+ * Try to avoid it whenever possible (typically when seed==0).
+ *
+ * It's important for performance that XXH3_hashLong is not inlined. Not sure
+ * why (uop cache maybe?), but the difference is large and easily measurable.
+ */
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_hashLong_64b_withSeed_internal(const void* input, size_t len,
+                                    XXH64_hash_t seed,
+                                    XXH3_f_accumulate f_acc,
+                                    XXH3_f_scrambleAcc f_scramble,
+                                    XXH3_f_initCustomSecret f_initSec)
+{
+#if XXH_SIZE_OPT <= 0
+    if (seed == 0)
+        return XXH3_hashLong_64b_internal(input, len,
+                                          XXH3_kSecret, sizeof(XXH3_kSecret),
+                                          f_acc, f_scramble);
+#endif
+    {   XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
+        f_initSec(secret, seed);
+        return XXH3_hashLong_64b_internal(input, len, secret, sizeof(secret),
+                                          f_acc, f_scramble);
+    }
 }
 
-XXH_FORCE_INLINE XXH128_hash_t XXH3_hashLong_128b_internal(
-    const void *XXH_RESTRICT input, size_t len,
-    const xxh_u8 *XXH_RESTRICT secret, size_t secretSize,
-    XXH3_f_accumulate_512 f_acc512, XXH3_f_scrambleAcc f_scramble) {
+/*
+ * It's important for performance that XXH3_hashLong is not inlined.
+ */
+XXH_NO_INLINE XXH64_hash_t
+XXH3_hashLong_64b_withSeed(const void* XXH_RESTRICT input, size_t len,
+                           XXH64_hash_t seed, const xxh_u8* XXH_RESTRICT secret, size_t secretLen)
+{
+    (void)secret; (void)secretLen;
+    return XXH3_hashLong_64b_withSeed_internal(input, len, seed,
+                XXH3_accumulate, XXH3_scrambleAcc, XXH3_initCustomSecret);
+}
 
-  XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[XXH_ACC_NB] = XXH3_INIT_ACC;
 
-  XXH3_hashLong_internal_loop(acc, (const xxh_u8 *)input, len, secret,
-                              secretSize, f_acc512, f_scramble);
+typedef XXH64_hash_t (*XXH3_hashLong64_f)(const void* XXH_RESTRICT, size_t,
+                                          XXH64_hash_t, const xxh_u8* XXH_RESTRICT, size_t);
 
-  /* converge into final hash */
-  XXH_STATIC_ASSERT(sizeof(acc) == 64);
-  XXH_ASSERT(secretSize >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
-  {
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_64bits_internal(const void* XXH_RESTRICT input, size_t len,
+                     XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen,
+                     XXH3_hashLong64_f f_hashLong)
+{
+    XXH_ASSERT(secretLen >= XXH3_SECRET_SIZE_MIN);
+    /*
+     * If an action is to be taken if `secretLen` condition is not respected,
+     * it should be done here.
+     * For now, it's a contract pre-condition.
+     * Adding a check and a branch here would cost performance at every hash.
+     * Also, note that function signature doesn't offer room to return an error.
+     */
+    if (len <= 16)
+        return XXH3_len_0to16_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, seed64);
+    if (len <= 128)
+        return XXH3_len_17to128_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
+    if (len <= XXH3_MIDSIZE_MAX)
+        return XXH3_len_129to240_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
+    return f_hashLong(input, len, seed64, (const xxh_u8*)secret, secretLen);
+}
 
-    XXH128_hash_t h128;
-    h128.low64 = XXH3_mergeAccs(acc, secret + XXH_SECRET_MERGEACCS_START,
-                                (xxh_u64)len * XXH_PRIME64_1);
-    h128.high64 = XXH3_mergeAccs(
-        acc, secret + secretSize - sizeof(acc) - XXH_SECRET_MERGEACCS_START,
-        ~((xxh_u64)len * XXH_PRIME64_2));
-    return h128;
 
-  }
+/* ===   Public entry point   === */
 
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH64_hash_t XXH3_64bits(XXH_NOESCAPE const void* input, size_t length)
+{
+    return XXH3_64bits_internal(input, length, 0, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_default);
 }
 
-/*
- * It's important for performance that XXH3_hashLong is not inlined.
- */
-XXH_NO_INLINE XXH128_hash_t XXH3_hashLong_128b_default(
-    const void *XXH_RESTRICT input, size_t len, XXH64_hash_t seed64,
-    const void *XXH_RESTRICT secret, size_t secretLen) {
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH64_hash_t
+XXH3_64bits_withSecret(XXH_NOESCAPE const void* input, size_t length, XXH_NOESCAPE const void* secret, size_t secretSize)
+{
+    return XXH3_64bits_internal(input, length, 0, secret, secretSize, XXH3_hashLong_64b_withSecret);
+}
 
-  (void)seed64;
-  (void)secret;
-  (void)secretLen;
-  return XXH3_hashLong_128b_internal(input, len, XXH3_kSecret,
-                                     sizeof(XXH3_kSecret), XXH3_accumulate_512,
-                                     XXH3_scrambleAcc);
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH64_hash_t
+XXH3_64bits_withSeed(XXH_NOESCAPE const void* input, size_t length, XXH64_hash_t seed)
+{
+    return XXH3_64bits_internal(input, length, seed, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_withSeed);
+}
 
+XXH_PUBLIC_API XXH64_hash_t
+XXH3_64bits_withSecretandSeed(XXH_NOESCAPE const void* input, size_t length, XXH_NOESCAPE const void* secret, size_t secretSize, XXH64_hash_t seed)
+{
+    if (length <= XXH3_MIDSIZE_MAX)
+        return XXH3_64bits_internal(input, length, seed, XXH3_kSecret, sizeof(XXH3_kSecret), NULL);
+    return XXH3_hashLong_64b_withSecret(input, length, seed, (const xxh_u8*)secret, secretSize);
 }
 
+
+/* ===   XXH3 streaming   === */
+#ifndef XXH_NO_STREAM
 /*
- * It's important for performance that XXH3_hashLong is not inlined.
+ * Malloc's a pointer that is always aligned to align.
+ *
+ * This must be freed with `XXH_alignedFree()`.
+ *
+ * malloc typically guarantees 16 byte alignment on 64-bit systems and 8 byte
+ * alignment on 32-bit. This isn't enough for the 32 byte aligned loads in AVX2
+ * or on 32-bit, the 16 byte aligned loads in SSE2 and NEON.
+ *
+ * This underalignment previously caused a rather obvious crash which went
+ * completely unnoticed due to XXH3_createState() not actually being tested.
+ * Credit to RedSpah for noticing this bug.
+ *
+ * The alignment is done manually: Functions like posix_memalign or _mm_malloc
+ * are avoided: To maintain portability, we would have to write a fallback
+ * like this anyways, and besides, testing for the existence of library
+ * functions without relying on external build tools is impossible.
+ *
+ * The method is simple: Overallocate, manually align, and store the offset
+ * to the original behind the returned pointer.
+ *
+ * Align must be a power of 2 and 8 <= align <= 128.
  */
-XXH_NO_INLINE XXH128_hash_t XXH3_hashLong_128b_withSecret(
-    const void *XXH_RESTRICT input, size_t len, XXH64_hash_t seed64,
-    const void *XXH_RESTRICT secret, size_t secretLen) {
-
-  (void)seed64;
-  return XXH3_hashLong_128b_internal(input, len, (const xxh_u8 *)secret,
-                                     secretLen, XXH3_accumulate_512,
-                                     XXH3_scrambleAcc);
+static XXH_MALLOCF void* XXH_alignedMalloc(size_t s, size_t align)
+{
+    XXH_ASSERT(align <= 128 && align >= 8); /* range check */
+    XXH_ASSERT((align & (align-1)) == 0);   /* power of 2 */
+    XXH_ASSERT(s != 0 && s < (s + align));  /* empty/overflow */
+    {   /* Overallocate to make room for manual realignment and an offset byte */
+        xxh_u8* base = (xxh_u8*)XXH_malloc(s + align);
+        if (base != NULL) {
+            /*
+             * Get the offset needed to align this pointer.
+             *
+             * Even if the returned pointer is aligned, there will always be
+             * at least one byte to store the offset to the original pointer.
+             */
+            size_t offset = align - ((size_t)base & (align - 1)); /* base % align */
+            /* Add the offset for the now-aligned pointer */
+            xxh_u8* ptr = base + offset;
+
+            XXH_ASSERT((size_t)ptr % align == 0);
+
+            /* Store the offset immediately before the returned pointer. */
+            ptr[-1] = (xxh_u8)offset;
+            return ptr;
+        }
+        return NULL;
+    }
+}
+/*
+ * Frees an aligned pointer allocated by XXH_alignedMalloc(). Don't pass
+ * normal malloc'd pointers, XXH_alignedMalloc has a specific data layout.
+ */
+static void XXH_alignedFree(void* p)
+{
+    if (p != NULL) {
+        xxh_u8* ptr = (xxh_u8*)p;
+        /* Get the offset byte we added in XXH_malloc. */
+        xxh_u8 offset = ptr[-1];
+        /* Free the original malloc'd pointer */
+        xxh_u8* base = ptr - offset;
+        XXH_free(base);
+    }
+}
+/*! @ingroup XXH3_family */
+/*!
+ * @brief Allocate an @ref XXH3_state_t.
+ *
+ * @return An allocated pointer of @ref XXH3_state_t on success.
+ * @return `NULL` on failure.
+ *
+ * @note Must be freed with XXH3_freeState().
+ *
+ * @see @ref streaming_example "Streaming Example"
+ */
+XXH_PUBLIC_API XXH3_state_t* XXH3_createState(void)
+{
+    XXH3_state_t* const state = (XXH3_state_t*)XXH_alignedMalloc(sizeof(XXH3_state_t), 64);
+    if (state==NULL) return NULL;
+    XXH3_INITSTATE(state);
+    return state;
+}
 
+/*! @ingroup XXH3_family */
+/*!
+ * @brief Frees an @ref XXH3_state_t.
+ *
+ * @param statePtr A pointer to an @ref XXH3_state_t allocated with @ref XXH3_createState().
+ *
+ * @return @ref XXH_OK.
+ *
+ * @note Must be allocated with XXH3_createState().
+ *
+ * @see @ref streaming_example "Streaming Example"
+ */
+XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t* statePtr)
+{
+    XXH_alignedFree(statePtr);
+    return XXH_OK;
 }
 
-XXH_FORCE_INLINE XXH128_hash_t XXH3_hashLong_128b_withSeed_internal(
-    const void *XXH_RESTRICT input, size_t len, XXH64_hash_t seed64,
-    XXH3_f_accumulate_512 f_acc512, XXH3_f_scrambleAcc f_scramble,
-    XXH3_f_initCustomSecret f_initSec) {
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API void
+XXH3_copyState(XXH_NOESCAPE XXH3_state_t* dst_state, XXH_NOESCAPE const XXH3_state_t* src_state)
+{
+    XXH_memcpy(dst_state, src_state, sizeof(*dst_state));
+}
+
+static void
+XXH3_reset_internal(XXH3_state_t* statePtr,
+                    XXH64_hash_t seed,
+                    const void* secret, size_t secretSize)
+{
+    size_t const initStart = offsetof(XXH3_state_t, bufferedSize);
+    size_t const initLength = offsetof(XXH3_state_t, nbStripesPerBlock) - initStart;
+    XXH_ASSERT(offsetof(XXH3_state_t, nbStripesPerBlock) > initStart);
+    XXH_ASSERT(statePtr != NULL);
+    /* set members from bufferedSize to nbStripesPerBlock (excluded) to 0 */
+    memset((char*)statePtr + initStart, 0, initLength);
+    statePtr->acc[0] = XXH_PRIME32_3;
+    statePtr->acc[1] = XXH_PRIME64_1;
+    statePtr->acc[2] = XXH_PRIME64_2;
+    statePtr->acc[3] = XXH_PRIME64_3;
+    statePtr->acc[4] = XXH_PRIME64_4;
+    statePtr->acc[5] = XXH_PRIME32_2;
+    statePtr->acc[6] = XXH_PRIME64_5;
+    statePtr->acc[7] = XXH_PRIME32_1;
+    statePtr->seed = seed;
+    statePtr->useSeed = (seed != 0);
+    statePtr->extSecret = (const unsigned char*)secret;
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
+    statePtr->secretLimit = secretSize - XXH_STRIPE_LEN;
+    statePtr->nbStripesPerBlock = statePtr->secretLimit / XXH_SECRET_CONSUME_RATE;
+}
+
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_reset(XXH_NOESCAPE XXH3_state_t* statePtr)
+{
+    if (statePtr == NULL) return XXH_ERROR;
+    XXH3_reset_internal(statePtr, 0, XXH3_kSecret, XXH_SECRET_DEFAULT_SIZE);
+    return XXH_OK;
+}
 
-  if (seed64 == 0)
-    return XXH3_hashLong_128b_internal(
-        input, len, XXH3_kSecret, sizeof(XXH3_kSecret), f_acc512, f_scramble);
-  {
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_reset_withSecret(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize)
+{
+    if (statePtr == NULL) return XXH_ERROR;
+    XXH3_reset_internal(statePtr, 0, secret, secretSize);
+    if (secret == NULL) return XXH_ERROR;
+    if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
+    return XXH_OK;
+}
 
-    XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
-    f_initSec(secret, seed64);
-    return XXH3_hashLong_128b_internal(input, len, (const xxh_u8 *)secret,
-                                       sizeof(secret), f_acc512, f_scramble);
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_reset_withSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH64_hash_t seed)
+{
+    if (statePtr == NULL) return XXH_ERROR;
+    if (seed==0) return XXH3_64bits_reset(statePtr);
+    if ((seed != statePtr->seed) || (statePtr->extSecret != NULL))
+        XXH3_initCustomSecret(statePtr->customSecret, seed);
+    XXH3_reset_internal(statePtr, seed, NULL, XXH_SECRET_DEFAULT_SIZE);
+    return XXH_OK;
+}
 
-  }
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize, XXH64_hash_t seed64)
+{
+    if (statePtr == NULL) return XXH_ERROR;
+    if (secret == NULL) return XXH_ERROR;
+    if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
+    XXH3_reset_internal(statePtr, seed64, secret, secretSize);
+    statePtr->useSeed = 1; /* always, even if seed64==0 */
+    return XXH_OK;
+}
 
+/*!
+ * @internal
+ * @brief Processes a large input for XXH3_update() and XXH3_digest_long().
+ *
+ * Unlike XXH3_hashLong_internal_loop(), this can process data that overlaps a block.
+ *
+ * @param acc                Pointer to the 8 accumulator lanes
+ * @param nbStripesSoFarPtr  In/out pointer to the number of leftover stripes in the block*
+ * @param nbStripesPerBlock  Number of stripes in a block
+ * @param input              Input pointer
+ * @param nbStripes          Number of stripes to process
+ * @param secret             Secret pointer
+ * @param secretLimit        Offset of the last block in @p secret
+ * @param f_acc              Pointer to an XXH3_accumulate implementation
+ * @param f_scramble         Pointer to an XXH3_scrambleAcc implementation
+ * @return                   Pointer past the end of @p input after processing
+ */
+XXH_FORCE_INLINE const xxh_u8 *
+XXH3_consumeStripes(xxh_u64* XXH_RESTRICT acc,
+                    size_t* XXH_RESTRICT nbStripesSoFarPtr, size_t nbStripesPerBlock,
+                    const xxh_u8* XXH_RESTRICT input, size_t nbStripes,
+                    const xxh_u8* XXH_RESTRICT secret, size_t secretLimit,
+                    XXH3_f_accumulate f_acc,
+                    XXH3_f_scrambleAcc f_scramble)
+{
+    const xxh_u8* initialSecret = secret + *nbStripesSoFarPtr * XXH_SECRET_CONSUME_RATE;
+    /* Process full blocks */
+    if (nbStripes >= (nbStripesPerBlock - *nbStripesSoFarPtr)) {
+        /* Process the initial partial block... */
+        size_t nbStripesThisIter = nbStripesPerBlock - *nbStripesSoFarPtr;
+
+        do {
+            /* Accumulate and scramble */
+            f_acc(acc, input, initialSecret, nbStripesThisIter);
+            f_scramble(acc, secret + secretLimit);
+            input += nbStripesThisIter * XXH_STRIPE_LEN;
+            nbStripes -= nbStripesThisIter;
+            /* Then continue the loop with the full block size */
+            nbStripesThisIter = nbStripesPerBlock;
+            initialSecret = secret;
+        } while (nbStripes >= nbStripesPerBlock);
+        *nbStripesSoFarPtr = 0;
+    }
+    /* Process a partial block */
+    if (nbStripes > 0) {
+        f_acc(acc, input, initialSecret, nbStripes);
+        input += nbStripes * XXH_STRIPE_LEN;
+        *nbStripesSoFarPtr += nbStripes;
+    }
+    /* Return end pointer */
+    return input;
 }
 
+#ifndef XXH3_STREAM_USE_STACK
+# if XXH_SIZE_OPT <= 0 && !defined(__clang__) /* clang doesn't need additional stack space */
+#   define XXH3_STREAM_USE_STACK 1
+# endif
+#endif
 /*
- * It's important for performance that XXH3_hashLong is not inlined.
+ * Both XXH3_64bits_update and XXH3_128bits_update use this routine.
  */
-XXH_NO_INLINE XXH128_hash_t
-XXH3_hashLong_128b_withSeed(const void *input, size_t len, XXH64_hash_t seed64,
-                            const void *XXH_RESTRICT secret, size_t secretLen) {
+XXH_FORCE_INLINE XXH_errorcode
+XXH3_update(XXH3_state_t* XXH_RESTRICT const state,
+            const xxh_u8* XXH_RESTRICT input, size_t len,
+            XXH3_f_accumulate f_acc,
+            XXH3_f_scrambleAcc f_scramble)
+{
+    if (input==NULL) {
+        XXH_ASSERT(len == 0);
+        return XXH_OK;
+    }
+
+    XXH_ASSERT(state != NULL);
+    {   const xxh_u8* const bEnd = input + len;
+        const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret;
+#if defined(XXH3_STREAM_USE_STACK) && XXH3_STREAM_USE_STACK >= 1
+        /* For some reason, gcc and MSVC seem to suffer greatly
+         * when operating accumulators directly into state.
+         * Operating into stack space seems to enable proper optimization.
+         * clang, on the other hand, doesn't seem to need this trick */
+        XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[8];
+        XXH_memcpy(acc, state->acc, sizeof(acc));
+#else
+        xxh_u64* XXH_RESTRICT const acc = state->acc;
+#endif
+        state->totalLen += len;
+        XXH_ASSERT(state->bufferedSize <= XXH3_INTERNALBUFFER_SIZE);
+
+        /* small input : just fill in tmp buffer */
+        if (len <= XXH3_INTERNALBUFFER_SIZE - state->bufferedSize) {
+            XXH_memcpy(state->buffer + state->bufferedSize, input, len);
+            state->bufferedSize += (XXH32_hash_t)len;
+            return XXH_OK;
+        }
 
-  (void)secret;
-  (void)secretLen;
-  return XXH3_hashLong_128b_withSeed_internal(
-      input, len, seed64, XXH3_accumulate_512, XXH3_scrambleAcc,
-      XXH3_initCustomSecret);
+        /* total input is now > XXH3_INTERNALBUFFER_SIZE */
+        #define XXH3_INTERNALBUFFER_STRIPES (XXH3_INTERNALBUFFER_SIZE / XXH_STRIPE_LEN)
+        XXH_STATIC_ASSERT(XXH3_INTERNALBUFFER_SIZE % XXH_STRIPE_LEN == 0);   /* clean multiple */
 
-}
+        /*
+         * Internal buffer is partially filled (always, except at beginning)
+         * Complete it, then consume it.
+         */
+        if (state->bufferedSize) {
+            size_t const loadSize = XXH3_INTERNALBUFFER_SIZE - state->bufferedSize;
+            XXH_memcpy(state->buffer + state->bufferedSize, input, loadSize);
+            input += loadSize;
+            XXH3_consumeStripes(acc,
+                               &state->nbStripesSoFar, state->nbStripesPerBlock,
+                                state->buffer, XXH3_INTERNALBUFFER_STRIPES,
+                                secret, state->secretLimit,
+                                f_acc, f_scramble);
+            state->bufferedSize = 0;
+        }
+        XXH_ASSERT(input < bEnd);
+        if (bEnd - input > XXH3_INTERNALBUFFER_SIZE) {
+            size_t nbStripes = (size_t)(bEnd - 1 - input) / XXH_STRIPE_LEN;
+            input = XXH3_consumeStripes(acc,
+                                       &state->nbStripesSoFar, state->nbStripesPerBlock,
+                                       input, nbStripes,
+                                       secret, state->secretLimit,
+                                       f_acc, f_scramble);
+            XXH_memcpy(state->buffer + sizeof(state->buffer) - XXH_STRIPE_LEN, input - XXH_STRIPE_LEN, XXH_STRIPE_LEN);
 
-typedef XXH128_hash_t (*XXH3_hashLong128_f)(const void *XXH_RESTRICT, size_t,
-                                            XXH64_hash_t,
-                                            const void *XXH_RESTRICT, size_t);
+        }
+        /* Some remaining input (always) : buffer it */
+        XXH_ASSERT(input < bEnd);
+        XXH_ASSERT(bEnd - input <= XXH3_INTERNALBUFFER_SIZE);
+        XXH_ASSERT(state->bufferedSize == 0);
+        XXH_memcpy(state->buffer, input, (size_t)(bEnd-input));
+        state->bufferedSize = (XXH32_hash_t)(bEnd-input);
+#if defined(XXH3_STREAM_USE_STACK) && XXH3_STREAM_USE_STACK >= 1
+        /* save stack accumulators into state */
+        XXH_memcpy(state->acc, acc, sizeof(acc));
+#endif
+    }
 
-XXH_FORCE_INLINE XXH128_hash_t
-XXH3_128bits_internal(const void *input, size_t len, XXH64_hash_t seed64,
-                      const void *XXH_RESTRICT secret, size_t secretLen,
-                      XXH3_hashLong128_f f_hl128) {
-
-  XXH_ASSERT(secretLen >= XXH3_SECRET_SIZE_MIN);
-  /*
-   * If an action is to be taken if `secret` conditions are not respected,
-   * it should be done here.
-   * For now, it's a contract pre-condition.
-   * Adding a check and a branch here would cost performance at every hash.
-   */
-  if (len <= 16)
-    return XXH3_len_0to16_128b((const xxh_u8 *)input, len,
-                               (const xxh_u8 *)secret, seed64);
-  if (len <= 128)
-    return XXH3_len_17to128_128b((const xxh_u8 *)input, len,
-                                 (const xxh_u8 *)secret, secretLen, seed64);
-  if (len <= XXH3_MIDSIZE_MAX)
-    return XXH3_len_129to240_128b((const xxh_u8 *)input, len,
-                                  (const xxh_u8 *)secret, secretLen, seed64);
-  return f_hl128(input, len, seed64, secret, secretLen);
+    return XXH_OK;
+}
 
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_update(XXH_NOESCAPE XXH3_state_t* state, XXH_NOESCAPE const void* input, size_t len)
+{
+    return XXH3_update(state, (const xxh_u8*)input, len,
+                       XXH3_accumulate, XXH3_scrambleAcc);
 }
 
-/* ===   Public XXH128 API   === */
 
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API XXH128_hash_t XXH3_128bits(const void *input, size_t len) {
+XXH_FORCE_INLINE void
+XXH3_digest_long (XXH64_hash_t* acc,
+                  const XXH3_state_t* state,
+                  const unsigned char* secret)
+{
+    xxh_u8 lastStripe[XXH_STRIPE_LEN];
+    const xxh_u8* lastStripePtr;
 
-  return XXH3_128bits_internal(input, len, 0, XXH3_kSecret,
-                               sizeof(XXH3_kSecret),
-                               XXH3_hashLong_128b_default);
+    /*
+     * Digest on a local copy. This way, the state remains unaltered, and it can
+     * continue ingesting more input afterwards.
+     */
+    XXH_memcpy(acc, state->acc, sizeof(state->acc));
+    if (state->bufferedSize >= XXH_STRIPE_LEN) {
+        /* Consume remaining stripes then point to remaining data in buffer */
+        size_t const nbStripes = (state->bufferedSize - 1) / XXH_STRIPE_LEN;
+        size_t nbStripesSoFar = state->nbStripesSoFar;
+        XXH3_consumeStripes(acc,
+                           &nbStripesSoFar, state->nbStripesPerBlock,
+                            state->buffer, nbStripes,
+                            secret, state->secretLimit,
+                            XXH3_accumulate, XXH3_scrambleAcc);
+        lastStripePtr = state->buffer + state->bufferedSize - XXH_STRIPE_LEN;
+    } else {  /* bufferedSize < XXH_STRIPE_LEN */
+        /* Copy to temp buffer */
+        size_t const catchupSize = XXH_STRIPE_LEN - state->bufferedSize;
+        XXH_ASSERT(state->bufferedSize > 0);  /* there is always some input buffered */
+        XXH_memcpy(lastStripe, state->buffer + sizeof(state->buffer) - catchupSize, catchupSize);
+        XXH_memcpy(lastStripe + catchupSize, state->buffer, state->bufferedSize);
+        lastStripePtr = lastStripe;
+    }
+    /* Last stripe */
+    XXH3_accumulate_512(acc,
+                        lastStripePtr,
+                        secret + state->secretLimit - XXH_SECRET_LASTACC_START);
+}
 
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_digest (XXH_NOESCAPE const XXH3_state_t* state)
+{
+    const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret;
+    if (state->totalLen > XXH3_MIDSIZE_MAX) {
+        XXH_ALIGN(XXH_ACC_ALIGN) XXH64_hash_t acc[XXH_ACC_NB];
+        XXH3_digest_long(acc, state, secret);
+        return XXH3_mergeAccs(acc,
+                              secret + XXH_SECRET_MERGEACCS_START,
+                              (xxh_u64)state->totalLen * XXH_PRIME64_1);
+    }
+    /* totalLen <= XXH3_MIDSIZE_MAX: digesting a short input */
+    if (state->useSeed)
+        return XXH3_64bits_withSeed(state->buffer, (size_t)state->totalLen, state->seed);
+    return XXH3_64bits_withSecret(state->buffer, (size_t)(state->totalLen),
+                                  secret, state->secretLimit + XXH_STRIPE_LEN);
 }
+#endif /* !XXH_NO_STREAM */
 
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_withSecret(const void *input,
-                                                     size_t      len,
-                                                     const void *secret,
-                                                     size_t      secretSize) {
 
-  return XXH3_128bits_internal(input, len, 0, (const xxh_u8 *)secret,
-                               secretSize, XXH3_hashLong_128b_withSecret);
+/* ==========================================
+ * XXH3 128 bits (a.k.a XXH128)
+ * ==========================================
+ * XXH3's 128-bit variant has better mixing and strength than the 64-bit variant,
+ * even without counting the significantly larger output size.
+ *
+ * For example, extra steps are taken to avoid the seed-dependent collisions
+ * in 17-240 byte inputs (See XXH3_mix16B and XXH128_mix32B).
+ *
+ * This strength naturally comes at the cost of some speed, especially on short
+ * lengths. Note that longer hashes are about as fast as the 64-bit version
+ * due to it using only a slight modification of the 64-bit loop.
+ *
+ * XXH128 is also more oriented towards 64-bit machines. It is still extremely
+ * fast for a _128-bit_ hash on 32-bit (it usually clears XXH64).
+ */
 
+XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t
+XXH3_len_1to3_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    /* A doubled version of 1to3_64b with different constants. */
+    XXH_ASSERT(input != NULL);
+    XXH_ASSERT(1 <= len && len <= 3);
+    XXH_ASSERT(secret != NULL);
+    /*
+     * len = 1: combinedl = { input[0], 0x01, input[0], input[0] }
+     * len = 2: combinedl = { input[1], 0x02, input[0], input[1] }
+     * len = 3: combinedl = { input[2], 0x03, input[0], input[1] }
+     */
+    {   xxh_u8 const c1 = input[0];
+        xxh_u8 const c2 = input[len >> 1];
+        xxh_u8 const c3 = input[len - 1];
+        xxh_u32 const combinedl = ((xxh_u32)c1 <<16) | ((xxh_u32)c2 << 24)
+                                | ((xxh_u32)c3 << 0) | ((xxh_u32)len << 8);
+        xxh_u32 const combinedh = XXH_rotl32(XXH_swap32(combinedl), 13);
+        xxh_u64 const bitflipl = (XXH_readLE32(secret) ^ XXH_readLE32(secret+4)) + seed;
+        xxh_u64 const bitfliph = (XXH_readLE32(secret+8) ^ XXH_readLE32(secret+12)) - seed;
+        xxh_u64 const keyed_lo = (xxh_u64)combinedl ^ bitflipl;
+        xxh_u64 const keyed_hi = (xxh_u64)combinedh ^ bitfliph;
+        XXH128_hash_t h128;
+        h128.low64  = XXH64_avalanche(keyed_lo);
+        h128.high64 = XXH64_avalanche(keyed_hi);
+        return h128;
+    }
 }
 
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_withSeed(const void  *input,
-                                                   size_t       len,
-                                                   XXH64_hash_t seed) {
-
-  return XXH3_128bits_internal(input, len, seed, XXH3_kSecret,
-                               sizeof(XXH3_kSecret),
-                               XXH3_hashLong_128b_withSeed);
-
+XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t
+XXH3_len_4to8_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(input != NULL);
+    XXH_ASSERT(secret != NULL);
+    XXH_ASSERT(4 <= len && len <= 8);
+    seed ^= (xxh_u64)XXH_swap32((xxh_u32)seed) << 32;
+    {   xxh_u32 const input_lo = XXH_readLE32(input);
+        xxh_u32 const input_hi = XXH_readLE32(input + len - 4);
+        xxh_u64 const input_64 = input_lo + ((xxh_u64)input_hi << 32);
+        xxh_u64 const bitflip = (XXH_readLE64(secret+16) ^ XXH_readLE64(secret+24)) + seed;
+        xxh_u64 const keyed = input_64 ^ bitflip;
+
+        /* Shift len to the left to ensure it is even, this avoids even multiplies. */
+        XXH128_hash_t m128 = XXH_mult64to128(keyed, XXH_PRIME64_1 + (len << 2));
+
+        m128.high64 += (m128.low64 << 1);
+        m128.low64  ^= (m128.high64 >> 3);
+
+        m128.low64   = XXH_xorshift64(m128.low64, 35);
+        m128.low64  *= PRIME_MX2;
+        m128.low64   = XXH_xorshift64(m128.low64, 28);
+        m128.high64  = XXH3_avalanche(m128.high64);
+        return m128;
+    }
 }
 
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API XXH128_hash_t XXH128(const void *input, size_t len,
-                                    XXH64_hash_t seed) {
+XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t
+XXH3_len_9to16_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(input != NULL);
+    XXH_ASSERT(secret != NULL);
+    XXH_ASSERT(9 <= len && len <= 16);
+    {   xxh_u64 const bitflipl = (XXH_readLE64(secret+32) ^ XXH_readLE64(secret+40)) - seed;
+        xxh_u64 const bitfliph = (XXH_readLE64(secret+48) ^ XXH_readLE64(secret+56)) + seed;
+        xxh_u64 const input_lo = XXH_readLE64(input);
+        xxh_u64       input_hi = XXH_readLE64(input + len - 8);
+        XXH128_hash_t m128 = XXH_mult64to128(input_lo ^ input_hi ^ bitflipl, XXH_PRIME64_1);
+        /*
+         * Put len in the middle of m128 to ensure that the length gets mixed to
+         * both the low and high bits in the 128x64 multiply below.
+         */
+        m128.low64 += (xxh_u64)(len - 1) << 54;
+        input_hi   ^= bitfliph;
+        /*
+         * Add the high 32 bits of input_hi to the high 32 bits of m128, then
+         * add the long product of the low 32 bits of input_hi and XXH_PRIME32_2 to
+         * the high 64 bits of m128.
+         *
+         * The best approach to this operation is different on 32-bit and 64-bit.
+         */
+        if (sizeof(void *) < sizeof(xxh_u64)) { /* 32-bit */
+            /*
+             * 32-bit optimized version, which is more readable.
+             *
+             * On 32-bit, it removes an ADC and delays a dependency between the two
+             * halves of m128.high64, but it generates an extra mask on 64-bit.
+             */
+            m128.high64 += (input_hi & 0xFFFFFFFF00000000ULL) + XXH_mult32to64((xxh_u32)input_hi, XXH_PRIME32_2);
+        } else {
+            /*
+             * 64-bit optimized (albeit more confusing) version.
+             *
+             * Uses some properties of addition and multiplication to remove the mask:
+             *
+             * Let:
+             *    a = input_hi.lo = (input_hi & 0x00000000FFFFFFFF)
+             *    b = input_hi.hi = (input_hi & 0xFFFFFFFF00000000)
+             *    c = XXH_PRIME32_2
+             *
+             *    a + (b * c)
+             * Inverse Property: x + y - x == y
+             *    a + (b * (1 + c - 1))
+             * Distributive Property: x * (y + z) == (x * y) + (x * z)
+             *    a + (b * 1) + (b * (c - 1))
+             * Identity Property: x * 1 == x
+             *    a + b + (b * (c - 1))
+             *
+             * Substitute a, b, and c:
+             *    input_hi.hi + input_hi.lo + ((xxh_u64)input_hi.lo * (XXH_PRIME32_2 - 1))
+             *
+             * Since input_hi.hi + input_hi.lo == input_hi, we get this:
+             *    input_hi + ((xxh_u64)input_hi.lo * (XXH_PRIME32_2 - 1))
+             */
+            m128.high64 += input_hi + XXH_mult32to64((xxh_u32)input_hi, XXH_PRIME32_2 - 1);
+        }
+        /* m128 ^= XXH_swap64(m128 >> 64); */
+        m128.low64  ^= XXH_swap64(m128.high64);
 
-  return XXH3_128bits_withSeed(input, len, seed);
+        {   /* 128x64 multiply: h128 = m128 * XXH_PRIME64_2; */
+            XXH128_hash_t h128 = XXH_mult64to128(m128.low64, XXH_PRIME64_2);
+            h128.high64 += m128.high64 * XXH_PRIME64_2;
 
+            h128.low64   = XXH3_avalanche(h128.low64);
+            h128.high64  = XXH3_avalanche(h128.high64);
+            return h128;
+    }   }
 }
 
-/* ===   XXH3 128-bit streaming   === */
+/*
+ * Assumption: `secret` size is >= XXH3_SECRET_SIZE_MIN
+ */
+XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t
+XXH3_len_0to16_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(len <= 16);
+    {   if (len > 8) return XXH3_len_9to16_128b(input, len, secret, seed);
+        if (len >= 4) return XXH3_len_4to8_128b(input, len, secret, seed);
+        if (len) return XXH3_len_1to3_128b(input, len, secret, seed);
+        {   XXH128_hash_t h128;
+            xxh_u64 const bitflipl = XXH_readLE64(secret+64) ^ XXH_readLE64(secret+72);
+            xxh_u64 const bitfliph = XXH_readLE64(secret+80) ^ XXH_readLE64(secret+88);
+            h128.low64 = XXH64_avalanche(seed ^ bitflipl);
+            h128.high64 = XXH64_avalanche( seed ^ bitfliph);
+            return h128;
+    }   }
+}
 
 /*
- * All the functions are actually the same as for 64-bit streaming variant.
- * The only difference is the finalization routine.
+ * A bit slower than XXH3_mix16B, but handles multiply by zero better.
  */
+XXH_FORCE_INLINE XXH128_hash_t
+XXH128_mix32B(XXH128_hash_t acc, const xxh_u8* input_1, const xxh_u8* input_2,
+              const xxh_u8* secret, XXH64_hash_t seed)
+{
+    acc.low64  += XXH3_mix16B (input_1, secret+0, seed);
+    acc.low64  ^= XXH_readLE64(input_2) + XXH_readLE64(input_2 + 8);
+    acc.high64 += XXH3_mix16B (input_2, secret+16, seed);
+    acc.high64 ^= XXH_readLE64(input_1) + XXH_readLE64(input_1 + 8);
+    return acc;
+}
+
+
+XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t
+XXH3_len_17to128_128b(const xxh_u8* XXH_RESTRICT input, size_t len,
+                      const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
+                      XXH64_hash_t seed)
+{
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
+    XXH_ASSERT(16 < len && len <= 128);
+
+    {   XXH128_hash_t acc;
+        acc.low64 = len * XXH_PRIME64_1;
+        acc.high64 = 0;
+
+#if XXH_SIZE_OPT >= 1
+        {
+            /* Smaller, but slightly slower. */
+            unsigned int i = (unsigned int)(len - 1) / 32;
+            do {
+                acc = XXH128_mix32B(acc, input+16*i, input+len-16*(i+1), secret+32*i, seed);
+            } while (i-- != 0);
+        }
+#else
+        if (len > 32) {
+            if (len > 64) {
+                if (len > 96) {
+                    acc = XXH128_mix32B(acc, input+48, input+len-64, secret+96, seed);
+                }
+                acc = XXH128_mix32B(acc, input+32, input+len-48, secret+64, seed);
+            }
+            acc = XXH128_mix32B(acc, input+16, input+len-32, secret+32, seed);
+        }
+        acc = XXH128_mix32B(acc, input, input+len-16, secret, seed);
+#endif
+        {   XXH128_hash_t h128;
+            h128.low64  = acc.low64 + acc.high64;
+            h128.high64 = (acc.low64    * XXH_PRIME64_1)
+                        + (acc.high64   * XXH_PRIME64_4)
+                        + ((len - seed) * XXH_PRIME64_2);
+            h128.low64  = XXH3_avalanche(h128.low64);
+            h128.high64 = (XXH64_hash_t)0 - XXH3_avalanche(h128.high64);
+            return h128;
+        }
+    }
+}
 
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset(XXH3_state_t *statePtr) {
+XXH_NO_INLINE XXH_PUREF XXH128_hash_t
+XXH3_len_129to240_128b(const xxh_u8* XXH_RESTRICT input, size_t len,
+                       const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
+                       XXH64_hash_t seed)
+{
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
+    XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX);
 
-  if (statePtr == NULL) return XXH_ERROR;
-  XXH3_reset_internal(statePtr, 0, XXH3_kSecret, XXH_SECRET_DEFAULT_SIZE);
-  return XXH_OK;
+    {   XXH128_hash_t acc;
+        unsigned i;
+        acc.low64 = len * XXH_PRIME64_1;
+        acc.high64 = 0;
+        /*
+         *  We set as `i` as offset + 32. We do this so that unchanged
+         * `len` can be used as upper bound. This reaches a sweet spot
+         * where both x86 and aarch64 get simple agen and good codegen
+         * for the loop.
+         */
+        for (i = 32; i < 160; i += 32) {
+            acc = XXH128_mix32B(acc,
+                                input  + i - 32,
+                                input  + i - 16,
+                                secret + i - 32,
+                                seed);
+        }
+        acc.low64 = XXH3_avalanche(acc.low64);
+        acc.high64 = XXH3_avalanche(acc.high64);
+        /*
+         * NB: `i <= len` will duplicate the last 32-bytes if
+         * len % 32 was zero. This is an unfortunate necessity to keep
+         * the hash result stable.
+         */
+        for (i=160; i <= len; i += 32) {
+            acc = XXH128_mix32B(acc,
+                                input + i - 32,
+                                input + i - 16,
+                                secret + XXH3_MIDSIZE_STARTOFFSET + i - 160,
+                                seed);
+        }
+        /* last bytes */
+        acc = XXH128_mix32B(acc,
+                            input + len - 16,
+                            input + len - 32,
+                            secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET - 16,
+                            (XXH64_hash_t)0 - seed);
+
+        {   XXH128_hash_t h128;
+            h128.low64  = acc.low64 + acc.high64;
+            h128.high64 = (acc.low64    * XXH_PRIME64_1)
+                        + (acc.high64   * XXH_PRIME64_4)
+                        + ((len - seed) * XXH_PRIME64_2);
+            h128.low64  = XXH3_avalanche(h128.low64);
+            h128.high64 = (XXH64_hash_t)0 - XXH3_avalanche(h128.high64);
+            return h128;
+        }
+    }
+}
 
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_hashLong_128b_internal(const void* XXH_RESTRICT input, size_t len,
+                            const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
+                            XXH3_f_accumulate f_acc,
+                            XXH3_f_scrambleAcc f_scramble)
+{
+    XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[XXH_ACC_NB] = XXH3_INIT_ACC;
+
+    XXH3_hashLong_internal_loop(acc, (const xxh_u8*)input, len, secret, secretSize, f_acc, f_scramble);
+
+    /* converge into final hash */
+    XXH_STATIC_ASSERT(sizeof(acc) == 64);
+    XXH_ASSERT(secretSize >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
+    {   XXH128_hash_t h128;
+        h128.low64  = XXH3_mergeAccs(acc,
+                                     secret + XXH_SECRET_MERGEACCS_START,
+                                     (xxh_u64)len * XXH_PRIME64_1);
+        h128.high64 = XXH3_mergeAccs(acc,
+                                     secret + secretSize
+                                            - sizeof(acc) - XXH_SECRET_MERGEACCS_START,
+                                     ~((xxh_u64)len * XXH_PRIME64_2));
+        return h128;
+    }
 }
 
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSecret(
-    XXH3_state_t *statePtr, const void *secret, size_t secretSize) {
+/*
+ * It's important for performance that XXH3_hashLong() is not inlined.
+ */
+XXH_NO_INLINE XXH_PUREF XXH128_hash_t
+XXH3_hashLong_128b_default(const void* XXH_RESTRICT input, size_t len,
+                           XXH64_hash_t seed64,
+                           const void* XXH_RESTRICT secret, size_t secretLen)
+{
+    (void)seed64; (void)secret; (void)secretLen;
+    return XXH3_hashLong_128b_internal(input, len, XXH3_kSecret, sizeof(XXH3_kSecret),
+                                       XXH3_accumulate, XXH3_scrambleAcc);
+}
 
-  if (statePtr == NULL) return XXH_ERROR;
-  XXH3_reset_internal(statePtr, 0, secret, secretSize);
-  if (secret == NULL) return XXH_ERROR;
-  if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
-  return XXH_OK;
+/*
+ * It's important for performance to pass @p secretLen (when it's static)
+ * to the compiler, so that it can properly optimize the vectorized loop.
+ *
+ * When the secret size is unknown, or on GCC 12 where the mix of NO_INLINE and FORCE_INLINE
+ * breaks -Og, this is XXH_NO_INLINE.
+ */
+XXH3_WITH_SECRET_INLINE XXH128_hash_t
+XXH3_hashLong_128b_withSecret(const void* XXH_RESTRICT input, size_t len,
+                              XXH64_hash_t seed64,
+                              const void* XXH_RESTRICT secret, size_t secretLen)
+{
+    (void)seed64;
+    return XXH3_hashLong_128b_internal(input, len, (const xxh_u8*)secret, secretLen,
+                                       XXH3_accumulate, XXH3_scrambleAcc);
+}
 
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_hashLong_128b_withSeed_internal(const void* XXH_RESTRICT input, size_t len,
+                                XXH64_hash_t seed64,
+                                XXH3_f_accumulate f_acc,
+                                XXH3_f_scrambleAcc f_scramble,
+                                XXH3_f_initCustomSecret f_initSec)
+{
+    if (seed64 == 0)
+        return XXH3_hashLong_128b_internal(input, len,
+                                           XXH3_kSecret, sizeof(XXH3_kSecret),
+                                           f_acc, f_scramble);
+    {   XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
+        f_initSec(secret, seed64);
+        return XXH3_hashLong_128b_internal(input, len, (const xxh_u8*)secret, sizeof(secret),
+                                           f_acc, f_scramble);
+    }
 }
 
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSeed(XXH3_state_t *statePtr,
-                                                         XXH64_hash_t  seed) {
+/*
+ * It's important for performance that XXH3_hashLong is not inlined.
+ */
+XXH_NO_INLINE XXH128_hash_t
+XXH3_hashLong_128b_withSeed(const void* input, size_t len,
+                            XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen)
+{
+    (void)secret; (void)secretLen;
+    return XXH3_hashLong_128b_withSeed_internal(input, len, seed64,
+                XXH3_accumulate, XXH3_scrambleAcc, XXH3_initCustomSecret);
+}
 
-  if (statePtr == NULL) return XXH_ERROR;
-  if (seed == 0) return XXH3_128bits_reset(statePtr);
-  if (seed != statePtr->seed)
-    XXH3_initCustomSecret(statePtr->customSecret, seed);
-  XXH3_reset_internal(statePtr, seed, NULL, XXH_SECRET_DEFAULT_SIZE);
-  return XXH_OK;
+typedef XXH128_hash_t (*XXH3_hashLong128_f)(const void* XXH_RESTRICT, size_t,
+                                            XXH64_hash_t, const void* XXH_RESTRICT, size_t);
 
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_128bits_internal(const void* input, size_t len,
+                      XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen,
+                      XXH3_hashLong128_f f_hl128)
+{
+    XXH_ASSERT(secretLen >= XXH3_SECRET_SIZE_MIN);
+    /*
+     * If an action is to be taken if `secret` conditions are not respected,
+     * it should be done here.
+     * For now, it's a contract pre-condition.
+     * Adding a check and a branch here would cost performance at every hash.
+     */
+    if (len <= 16)
+        return XXH3_len_0to16_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, seed64);
+    if (len <= 128)
+        return XXH3_len_17to128_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
+    if (len <= XXH3_MIDSIZE_MAX)
+        return XXH3_len_129to240_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
+    return f_hl128(input, len, seed64, secret, secretLen);
 }
 
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API XXH_errorcode XXH3_128bits_update(XXH3_state_t *state,
-                                                 const void   *input,
-                                                 size_t        len) {
 
-  return XXH3_update(state, (const xxh_u8 *)input, len, XXH3_accumulate_512,
-                     XXH3_scrambleAcc);
+/* ===   Public XXH128 API   === */
 
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits(XXH_NOESCAPE const void* input, size_t len)
+{
+    return XXH3_128bits_internal(input, len, 0,
+                                 XXH3_kSecret, sizeof(XXH3_kSecret),
+                                 XXH3_hashLong_128b_default);
 }
 
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_digest(const XXH3_state_t *state) {
-
-  const unsigned char *const secret =
-      (state->extSecret == NULL) ? state->customSecret : state->extSecret;
-  if (state->totalLen > XXH3_MIDSIZE_MAX) {
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH128_hash_t
+XXH3_128bits_withSecret(XXH_NOESCAPE const void* input, size_t len, XXH_NOESCAPE const void* secret, size_t secretSize)
+{
+    return XXH3_128bits_internal(input, len, 0,
+                                 (const xxh_u8*)secret, secretSize,
+                                 XXH3_hashLong_128b_withSecret);
+}
 
-    XXH_ALIGN(XXH_ACC_ALIGN) XXH64_hash_t acc[XXH_ACC_NB];
-    XXH3_digest_long(acc, state, secret);
-    XXH_ASSERT(state->secretLimit + XXH_STRIPE_LEN >=
-               sizeof(acc) + XXH_SECRET_MERGEACCS_START);
-    {
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH128_hash_t
+XXH3_128bits_withSeed(XXH_NOESCAPE const void* input, size_t len, XXH64_hash_t seed)
+{
+    return XXH3_128bits_internal(input, len, seed,
+                                 XXH3_kSecret, sizeof(XXH3_kSecret),
+                                 XXH3_hashLong_128b_withSeed);
+}
 
-      XXH128_hash_t h128;
-      h128.low64 = XXH3_mergeAccs(acc, secret + XXH_SECRET_MERGEACCS_START,
-                                  (xxh_u64)state->totalLen * XXH_PRIME64_1);
-      h128.high64 =
-          XXH3_mergeAccs(acc,
-                         secret + state->secretLimit + XXH_STRIPE_LEN -
-                             sizeof(acc) - XXH_SECRET_MERGEACCS_START,
-                         ~((xxh_u64)state->totalLen * XXH_PRIME64_2));
-      return h128;
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH128_hash_t
+XXH3_128bits_withSecretandSeed(XXH_NOESCAPE const void* input, size_t len, XXH_NOESCAPE const void* secret, size_t secretSize, XXH64_hash_t seed)
+{
+    if (len <= XXH3_MIDSIZE_MAX)
+        return XXH3_128bits_internal(input, len, seed, XXH3_kSecret, sizeof(XXH3_kSecret), NULL);
+    return XXH3_hashLong_128b_withSecret(input, len, seed, secret, secretSize);
+}
 
-    }
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH128_hash_t
+XXH128(XXH_NOESCAPE const void* input, size_t len, XXH64_hash_t seed)
+{
+    return XXH3_128bits_withSeed(input, len, seed);
+}
 
-  }
 
-  /* len <= XXH3_MIDSIZE_MAX : short code */
-  if (state->seed)
-    return XXH3_128bits_withSeed(state->buffer, (size_t)state->totalLen,
-                                 state->seed);
-  return XXH3_128bits_withSecret(state->buffer, (size_t)(state->totalLen),
-                                 secret, state->secretLimit + XXH_STRIPE_LEN);
+/* ===   XXH3 128-bit streaming   === */
+#ifndef XXH_NO_STREAM
+/*
+ * All initialization and update functions are identical to 64-bit streaming variant.
+ * The only difference is the finalization routine.
+ */
 
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_reset(XXH_NOESCAPE XXH3_state_t* statePtr)
+{
+    return XXH3_64bits_reset(statePtr);
+}
+
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_reset_withSecret(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize)
+{
+    return XXH3_64bits_reset_withSecret(statePtr, secret, secretSize);
+}
+
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_reset_withSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH64_hash_t seed)
+{
+    return XXH3_64bits_reset_withSeed(statePtr, seed);
+}
+
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize, XXH64_hash_t seed)
+{
+    return XXH3_64bits_reset_withSecretandSeed(statePtr, secret, secretSize, seed);
+}
+
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_update(XXH_NOESCAPE XXH3_state_t* state, XXH_NOESCAPE const void* input, size_t len)
+{
+    return XXH3_64bits_update(state, input, len);
+}
+
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_digest (XXH_NOESCAPE const XXH3_state_t* state)
+{
+    const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret;
+    if (state->totalLen > XXH3_MIDSIZE_MAX) {
+        XXH_ALIGN(XXH_ACC_ALIGN) XXH64_hash_t acc[XXH_ACC_NB];
+        XXH3_digest_long(acc, state, secret);
+        XXH_ASSERT(state->secretLimit + XXH_STRIPE_LEN >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
+        {   XXH128_hash_t h128;
+            h128.low64  = XXH3_mergeAccs(acc,
+                                         secret + XXH_SECRET_MERGEACCS_START,
+                                         (xxh_u64)state->totalLen * XXH_PRIME64_1);
+            h128.high64 = XXH3_mergeAccs(acc,
+                                         secret + state->secretLimit + XXH_STRIPE_LEN
+                                                - sizeof(acc) - XXH_SECRET_MERGEACCS_START,
+                                         ~((xxh_u64)state->totalLen * XXH_PRIME64_2));
+            return h128;
+        }
+    }
+    /* len <= XXH3_MIDSIZE_MAX : short code */
+    if (state->seed)
+        return XXH3_128bits_withSeed(state->buffer, (size_t)state->totalLen, state->seed);
+    return XXH3_128bits_withSecret(state->buffer, (size_t)(state->totalLen),
+                                   secret, state->secretLimit + XXH_STRIPE_LEN);
 }
+#endif /* !XXH_NO_STREAM */
+/* 128-bit utility functions */
 
-    /* 128-bit utility functions */
-
-      #include <string.h>                                 /* memcmp, memcpy */
+#include <string.h>   /* memcmp, memcpy */
 
 /* return : 1 is equal, 0 if different */
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2) {
-
-  /* note : XXH128_hash_t is compact, it has no padding byte */
-  return !(memcmp(&h1, &h2, sizeof(h1)));
-
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2)
+{
+    /* note : XXH128_hash_t is compact, it has no padding byte */
+    return !(memcmp(&h1, &h2, sizeof(h1)));
 }
 
 /* This prototype is compatible with stdlib's qsort().
- * return : >0 if *h128_1  > *h128_2
- *          <0 if *h128_1  < *h128_2
- *          =0 if *h128_1 == *h128_2  */
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API int XXH128_cmp(const void *h128_1, const void *h128_2) {
+ * @return : >0 if *h128_1  > *h128_2
+ *           <0 if *h128_1  < *h128_2
+ *           =0 if *h128_1 == *h128_2  */
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API int XXH128_cmp(XXH_NOESCAPE const void* h128_1, XXH_NOESCAPE const void* h128_2)
+{
+    XXH128_hash_t const h1 = *(const XXH128_hash_t*)h128_1;
+    XXH128_hash_t const h2 = *(const XXH128_hash_t*)h128_2;
+    int const hcmp = (h1.high64 > h2.high64) - (h2.high64 > h1.high64);
+    /* note : bets that, in most cases, hash values are different */
+    if (hcmp) return hcmp;
+    return (h1.low64 > h2.low64) - (h2.low64 > h1.low64);
+}
 
-  XXH128_hash_t const h1 = *(const XXH128_hash_t *)h128_1;
-  XXH128_hash_t const h2 = *(const XXH128_hash_t *)h128_2;
-  int const           hcmp = (h1.high64 > h2.high64) - (h2.high64 > h1.high64);
-  /* note : bets that, in most cases, hash values are different */
-  if (hcmp) return hcmp;
-  return (h1.low64 > h2.low64) - (h2.low64 > h1.low64);
 
+/*======   Canonical representation   ======*/
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API void
+XXH128_canonicalFromHash(XXH_NOESCAPE XXH128_canonical_t* dst, XXH128_hash_t hash)
+{
+    XXH_STATIC_ASSERT(sizeof(XXH128_canonical_t) == sizeof(XXH128_hash_t));
+    if (XXH_CPU_LITTLE_ENDIAN) {
+        hash.high64 = XXH_swap64(hash.high64);
+        hash.low64  = XXH_swap64(hash.low64);
+    }
+    XXH_memcpy(dst, &hash.high64, sizeof(hash.high64));
+    XXH_memcpy((char*)dst + sizeof(hash.high64), &hash.low64, sizeof(hash.low64));
 }
 
-/*======   Canonical representation   ======*/
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API void XXH128_canonicalFromHash(XXH128_canonical_t *dst,
-                                             XXH128_hash_t       hash) {
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH128_hash_t
+XXH128_hashFromCanonical(XXH_NOESCAPE const XXH128_canonical_t* src)
+{
+    XXH128_hash_t h;
+    h.high64 = XXH_readBE64(src);
+    h.low64  = XXH_readBE64(src->digest + 8);
+    return h;
+}
 
-  XXH_STATIC_ASSERT(sizeof(XXH128_canonical_t) == sizeof(XXH128_hash_t));
-  if (XXH_CPU_LITTLE_ENDIAN) {
 
-    hash.high64 = XXH_swap64(hash.high64);
-    hash.low64 = XXH_swap64(hash.low64);
 
-  }
+/* ==========================================
+ * Secret generators
+ * ==========================================
+ */
+#define XXH_MIN(x, y) (((x) > (y)) ? (y) : (x))
+
+XXH_FORCE_INLINE void XXH3_combine16(void* dst, XXH128_hash_t h128)
+{
+    XXH_writeLE64( dst, XXH_readLE64(dst) ^ h128.low64 );
+    XXH_writeLE64( (char*)dst+8, XXH_readLE64((char*)dst+8) ^ h128.high64 );
+}
+
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_generateSecret(XXH_NOESCAPE void* secretBuffer, size_t secretSize, XXH_NOESCAPE const void* customSeed, size_t customSeedSize)
+{
+#if (XXH_DEBUGLEVEL >= 1)
+    XXH_ASSERT(secretBuffer != NULL);
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
+#else
+    /* production mode, assert() are disabled */
+    if (secretBuffer == NULL) return XXH_ERROR;
+    if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
+#endif
 
-  memcpy(dst, &hash.high64, sizeof(hash.high64));
-  memcpy((char *)dst + sizeof(hash.high64), &hash.low64, sizeof(hash.low64));
+    if (customSeedSize == 0) {
+        customSeed = XXH3_kSecret;
+        customSeedSize = XXH_SECRET_DEFAULT_SIZE;
+    }
+#if (XXH_DEBUGLEVEL >= 1)
+    XXH_ASSERT(customSeed != NULL);
+#else
+    if (customSeed == NULL) return XXH_ERROR;
+#endif
 
+    /* Fill secretBuffer with a copy of customSeed - repeat as needed */
+    {   size_t pos = 0;
+        while (pos < secretSize) {
+            size_t const toCopy = XXH_MIN((secretSize - pos), customSeedSize);
+            memcpy((char*)secretBuffer + pos, customSeed, toCopy);
+            pos += toCopy;
+    }   }
+
+    {   size_t const nbSeg16 = secretSize / 16;
+        size_t n;
+        XXH128_canonical_t scrambler;
+        XXH128_canonicalFromHash(&scrambler, XXH128(customSeed, customSeedSize, 0));
+        for (n=0; n<nbSeg16; n++) {
+            XXH128_hash_t const h128 = XXH128(&scrambler, sizeof(scrambler), n);
+            XXH3_combine16((char*)secretBuffer + n*16, h128);
+        }
+        /* last segment */
+        XXH3_combine16((char*)secretBuffer + secretSize - 16, XXH128_hashFromCanonical(&scrambler));
+    }
+    return XXH_OK;
 }
 
-/*! @ingroup xxh3_family */
-XXH_PUBLIC_API XXH128_hash_t
-XXH128_hashFromCanonical(const XXH128_canonical_t *src) {
+/*! @ingroup XXH3_family */
+XXH_PUBLIC_API void
+XXH3_generateSecret_fromSeed(XXH_NOESCAPE void* secretBuffer, XXH64_hash_t seed)
+{
+    XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
+    XXH3_initCustomSecret(secret, seed);
+    XXH_ASSERT(secretBuffer != NULL);
+    memcpy(secretBuffer, secret, XXH_SECRET_DEFAULT_SIZE);
+}
 
-  XXH128_hash_t h;
-  h.high64 = XXH_readBE64(src);
-  h.low64 = XXH_readBE64(src->digest + 8);
-  return h;
 
-}
 
-      /* Pop our optimization override from above */
-      #if XXH_VECTOR == XXH_AVX2                      /* AVX2 */           \
-          && defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
-          && defined(__OPTIMIZE__) &&                                      \
-          !defined(__OPTIMIZE_SIZE__)                /* respect -O0 and -Os */
-        #pragma GCC pop_options
-      #endif
+/* Pop our optimization override from above */
+#if XXH_VECTOR == XXH_AVX2 /* AVX2 */ \
+  && defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
+  && defined(__OPTIMIZE__) && XXH_SIZE_OPT <= 0 /* respect -O0 and -Os */
+#  pragma GCC pop_options
+#endif
 
-    #endif                                              /* XXH_NO_LONG_LONG */
+#endif  /* XXH_NO_LONG_LONG */
 
-  #endif                                                     /* XXH_NO_XXH3 */
+#endif  /* XXH_NO_XXH3 */
 
 /*!
  * @}
  */
-#endif                                                /* XXH_IMPLEMENTATION */
-
-#if defined(__cplusplus)
+#endif  /* XXH_IMPLEMENTATION */
 
-}
 
+#if defined (__cplusplus)
+} /* extern "C" */
 #endif
-
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index ea8f1423..48e32996 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -401,6 +401,10 @@ static void usage(u8 *argv0, int more_help) {
   SAYF("Compiled with _AFL_DOCUMENT_MUTATIONS.\n");
 #endif
 
+#ifdef _AFL_SPECIAL_PERFORMANCE
+  SAYF("Compiled with special performance options for this specific system, it might not work on other platforms!\n");
+#endif
+
   SAYF("For additional help please consult %s/README.md :)\n\n", doc_path);
 
   exit(1);
diff --git a/src/afl-performance.c b/src/afl-performance.c
index 07c1b527..22cf4dec 100644
--- a/src/afl-performance.c
+++ b/src/afl-performance.c
@@ -2,9 +2,18 @@
 #include "afl-fuzz.h"
 #include "types.h"
 
+#ifdef _HAVE_AVX2
+#define T1HA0_AESNI_AVAILABLE 1
+#define T1HA_USE_FAST_ONESHOT_READ 1
+#define T1HA_USE_INDIRECT_FUNCTIONS 1
+#define T1HA_IA32AES_NAME XXH3_64bits
+#include "t1ha0_ia32aes_b.h"
+#else
 #define XXH_INLINE_ALL
 #include "xxhash.h"
 #undef XXH_INLINE_ALL
+#endif
+
 
 void rand_set_seed(afl_state_t *afl, s64 init_seed) {
 
diff --git a/utils/bench/Makefile b/utils/bench/Makefile
new file mode 100644
index 00000000..e7d2f3a1
--- /dev/null
+++ b/utils/bench/Makefile
@@ -0,0 +1,8 @@
+all:	hash
+
+hash:	hash.c
+	gcc -O3 -mavx2 -march=native -I../../include -o hash hash.c
+
+clean:
+	rm -f hash
+
diff --git a/utils/bench/README.md b/utils/bench/README.md
new file mode 100644
index 00000000..772c117b
--- /dev/null
+++ b/utils/bench/README.md
@@ -0,0 +1,2 @@
+# Internal AFL++ benchmarking
+
diff --git a/utils/bench/hash.c b/utils/bench/hash.c
new file mode 100644
index 00000000..013a5321
--- /dev/null
+++ b/utils/bench/hash.c
@@ -0,0 +1,42 @@
+#include <stdio.h>
+#include <stdint.h>
+#include <time.h>
+
+#define T1HA0_AESNI_AVAILABLE 1
+#define T1HA_USE_FAST_ONESHOT_READ 1
+#define T1HA_USE_INDIRECT_FUNCTIONS 1
+#define T1HA_IA32AES_NAME t1ha0_ia32aes
+#include "t1ha0_ia32aes_b.h"
+
+#define XXH_INLINE_ALL
+#include "xxhash.h"
+#undef XXH_INLINE_ALL
+
+int main() {
+  char *data = malloc(4097);
+  struct timespec start, end;
+  long long duration;
+  int i;
+  uint64_t res;
+
+  clock_gettime(CLOCK_MONOTONIC, &start);
+  for (i = 0; i < 100000000; ++i) {
+     res = XXH3_64bits(data, 4097);
+     memcpy(data + 16, (char*)&res, 8);
+  }
+  clock_gettime(CLOCK_MONOTONIC, &end);
+  duration = (end.tv_sec - start.tv_sec) * 1000000000LL + (end.tv_nsec - start.tv_nsec);
+  printf("xxh3 duration:          %lld ns\n", duration);
+
+  memset(data, 0, 4097);
+  clock_gettime(CLOCK_MONOTONIC, &start);
+  for (i = 0; i < 100000000; ++i) {
+     res = t1ha0_ia32aes(data, 4097);
+     memcpy(data + 16, (char*)&res, 8);
+  }
+  clock_gettime(CLOCK_MONOTONIC, &end);
+  duration = (end.tv_sec - start.tv_sec) * 1000000000LL + (end.tv_nsec - start.tv_nsec);
+  printf("t1ha0_ia32aes duration: %lld ns\n", duration);
+
+  return 0;
+}
-- 
cgit 1.4.1


From 369fce9c85bf3b850a7109e4604fee71f694d2cb Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Thu, 8 Feb 2024 15:13:46 +0100
Subject: code format

---
 TODO.md                   |     4 +-
 docs/Changelog.md         |    18 +
 include/cmplog.h          |     7 +-
 include/envs.h            |     2 +
 include/t1ha.h            |   527 +--
 include/t1ha0_ia32aes_b.h |   116 +-
 include/t1ha_bits.h       |  1466 ++++---
 include/t1ha_selfcheck.h  |    15 +-
 include/xxhash.h          | 10283 ++++++++++++++++++++++++--------------------
 src/afl-fuzz-redqueen.c   |     2 +
 src/afl-fuzz.c            |     4 +-
 src/afl-performance.c     |    17 +-
 utils/bench/hash.c        |    31 +-
 13 files changed, 6853 insertions(+), 5639 deletions(-)

(limited to 'include')

diff --git a/TODO.md b/TODO.md
index f2e3963f..d47372b8 100644
--- a/TODO.md
+++ b/TODO.md
@@ -2,17 +2,15 @@
 
 ## Must
 
- - UI revamp
  - hardened_usercopy=0 page_alloc.shuffle=0
  - add value_profile but only enable after 15 minutes without finds
- - cmplog max len, cmplog max items envs?
+ - cmplog max items env?
  - adapt MOpt to new mutation engine
    - Update afl->pending_not_fuzzed for MOpt
  - cmplog rtn sanity check on fixed length? currently we ignore the length
  - afl-showmap -f support
  - afl-fuzz multicore wrapper script
  - when trimming then perform crash detection
- - problem: either -L0 and/or -p mmopt results in zero new coverage
 
 
 ## Should
diff --git a/docs/Changelog.md b/docs/Changelog.md
index e5169daf..3415150a 100644
--- a/docs/Changelog.md
+++ b/docs/Changelog.md
@@ -4,13 +4,31 @@
   release of the tool. See README.md for the general instruction manual.
 
 ### Version ++4.20a (dev)
+  ! A new forkserver communication model is now introduced. afl-fuzz is
+    backward compatible to old compiled targets if they are not built
+    for CMPLOG/Redqueen, but new compiled targets will not work with
+    old afl-fuzz versions!
+  ! Recompiled all targets that are instrumented for CMPLOG/Redqueen!
+  - AFL++ now supports up to 4 billion coverage edges, up from 6 million.
+  - New compile option: `make PERFORMANCE=1` - this will enable special
+    CPU dependent optimizations that make everything more performant - but
+    the binaries will likely won't work on different platforms. Also
+    enables a faster hasher if the CPU requirements are met.
+  - The persistent record feature (see config.h) was expanded to also
+    support replay, thanks to @quarta-qti !
   - afl-fuzz:
     - the new deterministic fuzzing feature is now activated by default,
       deactivate with -z. Parameters -d and -D are ignored.
+    - small improvements to CMPLOG/redqueen
+    - workround for a bug with MOpt -L when used with -M - in the future
+      we will either remove or rewrite MOpt.
   - afl-cc:
     - added collision free caller instrumentation to LTO mode. activate with
       `AFL_LLVM_LTO_CALLER=1`. You can set a max depth to go through single
       block functions with `AFL_LLVM_LTO_CALLER_DEPTH` (default 0)
+  - Minor edits to afl-persistent-config
+  - Prevent temporary files being left behind on aborted afl-whatsup
+  - More CPU benchmarks added to benchmark/
 
 
 ### Version ++4.10c (release)
diff --git a/include/cmplog.h b/include/cmplog.h
index 589570fe..a6162b59 100644
--- a/include/cmplog.h
+++ b/include/cmplog.h
@@ -41,13 +41,12 @@
 #define CMP_TYPE_INS 0
 #define CMP_TYPE_RTN 1
 
-struct cmp_header {
+struct cmp_header {  // 16 bit = 2 bytes
 
   unsigned hits : 6;       // up to 63 entries, we have CMP_MAP_H = 32
-  unsigned shape : 5;      // 31+1 bytes
-  unsigned type : 1;       // 4, we use 3: none, rtn, cmp
+  unsigned shape : 5;      // 31+1 bytes max
+  unsigned type : 1;       // 2: cmp, rtn
   unsigned attribute : 4;  // 16 for arithmetic comparison types
-  //unsigned reserved : 6;
 
 } __attribute__((packed));
 
diff --git a/include/envs.h b/include/envs.h
index 8f342553..d32e2f92 100644
--- a/include/envs.h
+++ b/include/envs.h
@@ -64,6 +64,8 @@ static char *afl_environment_variables[] = {
     "AFL_REAL_LD", "AFL_LD_PRELOAD", "AFL_LD_VERBOSE", "AFL_LLVM_ALLOWLIST",
     "AFL_LLVM_DENYLIST", "AFL_LLVM_BLOCKLIST", "AFL_CMPLOG", "AFL_LLVM_CMPLOG",
     "AFL_GCC_CMPLOG", "AFL_LLVM_INSTRIM", "AFL_LLVM_CALLER", "AFL_LLVM_CTX",
+    "AFL_LLVM_LTO_CALLER", "AFL_LLVM_LTO_CTX", "AFL_LLVM_LTO_CALLER_DEPTH",
+    "AFL_LLVM_LTO_CTX_DEPTH", "AFL_LLVM_CALLER_DEPTH", "AFL_LLVM_CTX_DEPTH",
     "AFL_LLVM_CTX_K", "AFL_LLVM_DICT2FILE", "AFL_LLVM_DICT2FILE_NO_MAIN",
     "AFL_LLVM_DOCUMENT_IDS", "AFL_LLVM_INSTRIM_LOOPHEAD", "AFL_LLVM_INSTRUMENT",
     "AFL_LLVM_LTO_AUTODICTIONARY", "AFL_LLVM_AUTODICTIONARY",
diff --git a/include/t1ha.h b/include/t1ha.h
index 498f0dd6..1af29395 100644
--- a/include/t1ha.h
+++ b/include/t1ha.h
@@ -172,56 +172,56 @@
 #define T1HA_VERSION_RELEASE 1
 
 #ifndef __has_attribute
-#define __has_attribute(x) (0)
+  #define __has_attribute(x) (0)
 #endif
 
 #ifndef __has_include
-#define __has_include(x) (0)
+  #define __has_include(x) (0)
 #endif
 
 #ifndef __GNUC_PREREQ
-#if defined(__GNUC__) && defined(__GNUC_MINOR__)
-#define __GNUC_PREREQ(maj, min)                                                \
-  ((__GNUC__ << 16) + __GNUC_MINOR__ >= ((maj) << 16) + (min))
-#else
-#define __GNUC_PREREQ(maj, min) 0
-#endif
-#endif /* __GNUC_PREREQ */
+  #if defined(__GNUC__) && defined(__GNUC_MINOR__)
+    #define __GNUC_PREREQ(maj, min) \
+      ((__GNUC__ << 16) + __GNUC_MINOR__ >= ((maj) << 16) + (min))
+  #else
+    #define __GNUC_PREREQ(maj, min) 0
+  #endif
+#endif                                                     /* __GNUC_PREREQ */
 
 #ifndef __CLANG_PREREQ
-#ifdef __clang__
-#define __CLANG_PREREQ(maj, min)                                               \
-  ((__clang_major__ << 16) + __clang_minor__ >= ((maj) << 16) + (min))
-#else
-#define __CLANG_PREREQ(maj, min) (0)
-#endif
-#endif /* __CLANG_PREREQ */
+  #ifdef __clang__
+    #define __CLANG_PREREQ(maj, min) \
+      ((__clang_major__ << 16) + __clang_minor__ >= ((maj) << 16) + (min))
+  #else
+    #define __CLANG_PREREQ(maj, min) (0)
+  #endif
+#endif                                                    /* __CLANG_PREREQ */
 
 #ifndef __LCC_PREREQ
-#ifdef __LCC__
-#define __LCC_PREREQ(maj, min)                                                 \
-  ((__LCC__ << 16) + __LCC_MINOR__ >= ((maj) << 16) + (min))
-#else
-#define __LCC_PREREQ(maj, min) (0)
-#endif
-#endif /* __LCC_PREREQ */
+  #ifdef __LCC__
+    #define __LCC_PREREQ(maj, min) \
+      ((__LCC__ << 16) + __LCC_MINOR__ >= ((maj) << 16) + (min))
+  #else
+    #define __LCC_PREREQ(maj, min) (0)
+  #endif
+#endif                                                      /* __LCC_PREREQ */
 
 /*****************************************************************************/
 
 #ifdef _MSC_VER
-/* Avoid '16' bytes padding added after data member 't1ha_context::total'
- * and other warnings from std-headers if warning-level > 3. */
-#pragma warning(push, 3)
+  /* Avoid '16' bytes padding added after data member 't1ha_context::total'
+   * and other warnings from std-headers if warning-level > 3. */
+  #pragma warning(push, 3)
 #endif
 
 #if defined(__cplusplus) && __cplusplus >= 201103L
-#include <climits>
-#include <cstddef>
-#include <cstdint>
+  #include <climits>
+  #include <cstddef>
+  #include <cstdint>
 #else
-#include <limits.h>
-#include <stddef.h>
-#include <stdint.h>
+  #include <limits.h>
+  #include <stddef.h>
+  #include <stdint.h>
 #endif
 
 /*****************************************************************************/
@@ -234,18 +234,18 @@
     defined(__INTEL__) || defined(__x86_64) || defined(__x86_64__) ||          \
     defined(__amd64__) || defined(__amd64) || defined(_M_X64) ||               \
     defined(_M_AMD64) || defined(__IA32__) || defined(__INTEL__)
-#ifndef __ia32__
-/* LY: define neutral __ia32__ for x86 and x86-64 archs */
-#define __ia32__ 1
-#endif /* __ia32__ */
-#if !defined(__amd64__) && (defined(__x86_64) || defined(__x86_64__) ||        \
-                            defined(__amd64) || defined(_M_X64))
-/* LY: define trusty __amd64__ for all AMD64/x86-64 arch */
-#define __amd64__ 1
-#endif /* __amd64__ */
-#endif /* all x86 */
-
-#if !defined(__BYTE_ORDER__) || !defined(__ORDER_LITTLE_ENDIAN__) ||           \
+  #ifndef __ia32__
+    /* LY: define neutral __ia32__ for x86 and x86-64 archs */
+    #define __ia32__ 1
+  #endif                                                        /* __ia32__ */
+  #if !defined(__amd64__) && (defined(__x86_64) || defined(__x86_64__) || \
+                              defined(__amd64) || defined(_M_X64))
+    /* LY: define trusty __amd64__ for all AMD64/x86-64 arch */
+    #define __amd64__ 1
+  #endif                                                       /* __amd64__ */
+#endif                                                           /* all x86 */
+
+#if !defined(__BYTE_ORDER__) || !defined(__ORDER_LITTLE_ENDIAN__) || \
     !defined(__ORDER_BIG_ENDIAN__)
 
 /* *INDENT-OFF* */
@@ -267,160 +267,168 @@
     defined(__NETBSD__) || defined(__NetBSD__) ||                              \
     defined(HAVE_SYS_PARAM_H) || __has_include(<sys/param.h>)
 #include <sys/param.h>
-#endif /* OS */
+#endif                                                                /* OS */
 
 /* *INDENT-ON* */
 /* clang-format on */
 
-#if defined(__BYTE_ORDER) && defined(__LITTLE_ENDIAN) && defined(__BIG_ENDIAN)
-#define __ORDER_LITTLE_ENDIAN__ __LITTLE_ENDIAN
-#define __ORDER_BIG_ENDIAN__ __BIG_ENDIAN
-#define __BYTE_ORDER__ __BYTE_ORDER
-#elif defined(_BYTE_ORDER) && defined(_LITTLE_ENDIAN) && defined(_BIG_ENDIAN)
-#define __ORDER_LITTLE_ENDIAN__ _LITTLE_ENDIAN
-#define __ORDER_BIG_ENDIAN__ _BIG_ENDIAN
-#define __BYTE_ORDER__ _BYTE_ORDER
-#else
-#define __ORDER_LITTLE_ENDIAN__ 1234
-#define __ORDER_BIG_ENDIAN__ 4321
-
-#if defined(__LITTLE_ENDIAN__) ||                                              \
-    (defined(_LITTLE_ENDIAN) && !defined(_BIG_ENDIAN)) ||                      \
-    defined(__ARMEL__) || defined(__THUMBEL__) || defined(__AARCH64EL__) ||    \
-    defined(__MIPSEL__) || defined(_MIPSEL) || defined(__MIPSEL) ||            \
-    defined(_M_ARM) || defined(_M_ARM64) || defined(__e2k__) ||                \
-    defined(__elbrus_4c__) || defined(__elbrus_8c__) || defined(__bfin__) ||   \
-    defined(__BFIN__) || defined(__ia64__) || defined(_IA64) ||                \
-    defined(__IA64__) || defined(__ia64) || defined(_M_IA64) ||                \
-    defined(__itanium__) || defined(__ia32__) || defined(__CYGWIN__) ||        \
-    defined(_WIN64) || defined(_WIN32) || defined(__TOS_WIN__) ||              \
-    defined(__WINDOWS__)
-#define __BYTE_ORDER__ __ORDER_LITTLE_ENDIAN__
-
-#elif defined(__BIG_ENDIAN__) ||                                               \
-    (defined(_BIG_ENDIAN) && !defined(_LITTLE_ENDIAN)) ||                      \
-    defined(__ARMEB__) || defined(__THUMBEB__) || defined(__AARCH64EB__) ||    \
-    defined(__MIPSEB__) || defined(_MIPSEB) || defined(__MIPSEB) ||            \
-    defined(__m68k__) || defined(M68000) || defined(__hppa__) ||               \
-    defined(__hppa) || defined(__HPPA__) || defined(__sparc__) ||              \
-    defined(__sparc) || defined(__370__) || defined(__THW_370__) ||            \
-    defined(__s390__) || defined(__s390x__) || defined(__SYSC_ZARCH__)
-#define __BYTE_ORDER__ __ORDER_BIG_ENDIAN__
-
-#else
-#error __BYTE_ORDER__ should be defined.
-#endif /* Arch */
-
-#endif
+  #if defined(__BYTE_ORDER) && defined(__LITTLE_ENDIAN) && defined(__BIG_ENDIAN)
+    #define __ORDER_LITTLE_ENDIAN__ __LITTLE_ENDIAN
+    #define __ORDER_BIG_ENDIAN__ __BIG_ENDIAN
+    #define __BYTE_ORDER__ __BYTE_ORDER
+  #elif defined(_BYTE_ORDER) && defined(_LITTLE_ENDIAN) && defined(_BIG_ENDIAN)
+    #define __ORDER_LITTLE_ENDIAN__ _LITTLE_ENDIAN
+    #define __ORDER_BIG_ENDIAN__ _BIG_ENDIAN
+    #define __BYTE_ORDER__ _BYTE_ORDER
+  #else
+    #define __ORDER_LITTLE_ENDIAN__ 1234
+    #define __ORDER_BIG_ENDIAN__ 4321
+
+    #if defined(__LITTLE_ENDIAN__) ||                                        \
+        (defined(_LITTLE_ENDIAN) && !defined(_BIG_ENDIAN)) ||                \
+        defined(__ARMEL__) || defined(__THUMBEL__) ||                        \
+        defined(__AARCH64EL__) || defined(__MIPSEL__) || defined(_MIPSEL) || \
+        defined(__MIPSEL) || defined(_M_ARM) || defined(_M_ARM64) ||         \
+        defined(__e2k__) || defined(__elbrus_4c__) ||                        \
+        defined(__elbrus_8c__) || defined(__bfin__) || defined(__BFIN__) ||  \
+        defined(__ia64__) || defined(_IA64) || defined(__IA64__) ||          \
+        defined(__ia64) || defined(_M_IA64) || defined(__itanium__) ||       \
+        defined(__ia32__) || defined(__CYGWIN__) || defined(_WIN64) ||       \
+        defined(_WIN32) || defined(__TOS_WIN__) || defined(__WINDOWS__)
+      #define __BYTE_ORDER__ __ORDER_LITTLE_ENDIAN__
+
+    #elif defined(__BIG_ENDIAN__) ||                                         \
+        (defined(_BIG_ENDIAN) && !defined(_LITTLE_ENDIAN)) ||                \
+        defined(__ARMEB__) || defined(__THUMBEB__) ||                        \
+        defined(__AARCH64EB__) || defined(__MIPSEB__) || defined(_MIPSEB) || \
+        defined(__MIPSEB) || defined(__m68k__) || defined(M68000) ||         \
+        defined(__hppa__) || defined(__hppa) || defined(__HPPA__) ||         \
+        defined(__sparc__) || defined(__sparc) || defined(__370__) ||        \
+        defined(__THW_370__) || defined(__s390__) || defined(__s390x__) ||   \
+        defined(__SYSC_ZARCH__)
+      #define __BYTE_ORDER__ __ORDER_BIG_ENDIAN__
+
+    #else
+      #error __BYTE_ORDER__ should be defined.
+    #endif                                                          /* Arch */
+
+  #endif
 #endif /* __BYTE_ORDER__ || __ORDER_LITTLE_ENDIAN__ || __ORDER_BIG_ENDIAN__ */
 
 /*****************************************************************************/
 
 #ifndef __dll_export
-#if defined(_WIN32) || defined(_WIN64) || defined(__CYGWIN__)
-#if defined(__GNUC__) || __has_attribute(dllexport)
-#define __dll_export __attribute__((dllexport))
-#else
-#define __dll_export __declspec(dllexport)
-#endif
-#elif defined(__GNUC__) || __has_attribute(__visibility__)
-#define __dll_export __attribute__((__visibility__("default")))
-#else
-#define __dll_export
-#endif
-#endif /* __dll_export */
+  #if defined(_WIN32) || defined(_WIN64) || defined(__CYGWIN__)
+    #if defined(__GNUC__) || __has_attribute(dllexport)
+      #define __dll_export __attribute__((dllexport))
+    #else
+      #define __dll_export __declspec(dllexport)
+    #endif
+  #elif defined(__GNUC__) || __has_attribute(__visibility__)
+    #define __dll_export __attribute__((__visibility__("default")))
+  #else
+    #define __dll_export
+  #endif
+#endif                                                      /* __dll_export */
 
 #ifndef __dll_import
-#if defined(_WIN32) || defined(_WIN64) || defined(__CYGWIN__)
-#if defined(__GNUC__) || __has_attribute(dllimport)
-#define __dll_import __attribute__((dllimport))
-#else
-#define __dll_import __declspec(dllimport)
-#endif
-#elif defined(__GNUC__) || __has_attribute(__visibility__)
-#define __dll_import __attribute__((__visibility__("default")))
-#else
-#define __dll_import
-#endif
-#endif /* __dll_import */
+  #if defined(_WIN32) || defined(_WIN64) || defined(__CYGWIN__)
+    #if defined(__GNUC__) || __has_attribute(dllimport)
+      #define __dll_import __attribute__((dllimport))
+    #else
+      #define __dll_import __declspec(dllimport)
+    #endif
+  #elif defined(__GNUC__) || __has_attribute(__visibility__)
+    #define __dll_import __attribute__((__visibility__("default")))
+  #else
+    #define __dll_import
+  #endif
+#endif                                                      /* __dll_import */
 
 #ifndef __force_inline
-#ifdef _MSC_VER
-#define __force_inline __forceinline
-#elif __GNUC_PREREQ(3, 2) || __has_attribute(__always_inline__)
-#define __force_inline __inline __attribute__((__always_inline__))
-#else
-#define __force_inline __inline
-#endif
-#endif /* __force_inline */
+  #ifdef _MSC_VER
+    #define __force_inline __forceinline
+  #elif __GNUC_PREREQ(3, 2) || __has_attribute(__always_inline__)
+    #define __force_inline __inline __attribute__((__always_inline__))
+  #else
+    #define __force_inline __inline
+  #endif
+#endif                                                    /* __force_inline */
 
 #ifndef T1HA_API
-#if defined(t1ha_EXPORTS)
-#define T1HA_API __dll_export
-#elif defined(t1ha_IMPORTS)
-#define T1HA_API __dll_import
-#else
-#define T1HA_API
-#endif
-#endif /* T1HA_API */
+  #if defined(t1ha_EXPORTS)
+    #define T1HA_API __dll_export
+  #elif defined(t1ha_IMPORTS)
+    #define T1HA_API __dll_import
+  #else
+    #define T1HA_API
+  #endif
+#endif                                                          /* T1HA_API */
 
 #if defined(_MSC_VER) && defined(__ia32__)
-#define T1HA_ALIGN_PREFIX __declspec(align(32)) /* required only for SIMD */
+  #define T1HA_ALIGN_PREFIX __declspec(align(32)) /* required only for SIMD */
 #else
-#define T1HA_ALIGN_PREFIX
-#endif /* _MSC_VER */
+  #define T1HA_ALIGN_PREFIX
+#endif                                                          /* _MSC_VER */
 
 #if defined(__GNUC__) && defined(__ia32__)
-#define T1HA_ALIGN_SUFFIX                                                      \
-  __attribute__((__aligned__(32))) /* required only for SIMD */
+  #define T1HA_ALIGN_SUFFIX \
+    __attribute__((__aligned__(32)))              /* required only for SIMD */
 #else
-#define T1HA_ALIGN_SUFFIX
-#endif /* GCC x86 */
+  #define T1HA_ALIGN_SUFFIX
+#endif                                                           /* GCC x86 */
 
 #ifndef T1HA_USE_INDIRECT_FUNCTIONS
-/* GNU ELF indirect functions usage control. For more info please see
- * https://en.wikipedia.org/wiki/Executable_and_Linkable_Format
- * and https://sourceware.org/glibc/wiki/GNU_IFUNC */
-#if defined(__ELF__) && defined(__amd64__) &&                                  \
-    (__has_attribute(__ifunc__) ||                                             \
-     (!defined(__clang__) && defined(__GNUC__) && __GNUC__ >= 4 &&             \
-      !defined(__SANITIZE_ADDRESS__) && !defined(__SSP_ALL__)))
-/* Enable gnu_indirect_function by default if :
- *  - ELF AND x86_64
- *  - attribute(__ifunc__) is available OR
- *    GCC >= 4 WITHOUT -fsanitize=address NOR -fstack-protector-all */
-#define T1HA_USE_INDIRECT_FUNCTIONS 1
-#else
-#define T1HA_USE_INDIRECT_FUNCTIONS 0
-#endif
-#endif /* T1HA_USE_INDIRECT_FUNCTIONS */
+  /* GNU ELF indirect functions usage control. For more info please see
+   * https://en.wikipedia.org/wiki/Executable_and_Linkable_Format
+   * and https://sourceware.org/glibc/wiki/GNU_IFUNC */
+  #if defined(__ELF__) && defined(__amd64__) &&                      \
+      (__has_attribute(__ifunc__) ||                                 \
+       (!defined(__clang__) && defined(__GNUC__) && __GNUC__ >= 4 && \
+        !defined(__SANITIZE_ADDRESS__) && !defined(__SSP_ALL__)))
+    /* Enable gnu_indirect_function by default if :
+     *  - ELF AND x86_64
+     *  - attribute(__ifunc__) is available OR
+     *    GCC >= 4 WITHOUT -fsanitize=address NOR -fstack-protector-all */
+    #define T1HA_USE_INDIRECT_FUNCTIONS 1
+  #else
+    #define T1HA_USE_INDIRECT_FUNCTIONS 0
+  #endif
+#endif                                       /* T1HA_USE_INDIRECT_FUNCTIONS */
 
 #if __GNUC_PREREQ(4, 0)
-#pragma GCC visibility push(hidden)
-#endif /* __GNUC_PREREQ(4,0) */
+  #pragma GCC visibility push(hidden)
+#endif                                                /* __GNUC_PREREQ(4,0) */
 
 #ifdef __cplusplus
 extern "C" {
+
 #endif
 
 typedef union T1HA_ALIGN_PREFIX t1ha_state256 {
-  uint8_t bytes[32];
+
+  uint8_t  bytes[32];
   uint32_t u32[8];
   uint64_t u64[4];
   struct {
+
     uint64_t a, b, c, d;
+
   } n;
+
 } t1ha_state256_t T1HA_ALIGN_SUFFIX;
 
 typedef struct t1ha_context {
+
   t1ha_state256_t state;
   t1ha_state256_t buffer;
-  size_t partial;
-  uint64_t total;
+  size_t          partial;
+  uint64_t        total;
+
 } t1ha_context_t;
 
 #ifdef _MSC_VER
-#pragma warning(pop)
+  #pragma warning(pop)
 #endif
 
 /******************************************************************************
@@ -443,37 +451,37 @@ T1HA_API int t1ha_selfcheck__t1ha2_atonce(void);
 T1HA_API int t1ha_selfcheck__t1ha2_atonce128(void);
 T1HA_API int t1ha_selfcheck__t1ha2_stream(void);
 T1HA_API int t1ha_selfcheck__t1ha2(void);
-#endif /* T1HA2_DISABLED */
+#endif                                                    /* T1HA2_DISABLED */
 
 #ifndef T1HA1_DISABLED
 T1HA_API int t1ha_selfcheck__t1ha1_le(void);
 T1HA_API int t1ha_selfcheck__t1ha1_be(void);
 T1HA_API int t1ha_selfcheck__t1ha1(void);
-#endif /* T1HA1_DISABLED */
+#endif                                                    /* T1HA1_DISABLED */
 
 #ifndef T1HA0_DISABLED
 T1HA_API int t1ha_selfcheck__t1ha0_32le(void);
 T1HA_API int t1ha_selfcheck__t1ha0_32be(void);
 T1HA_API int t1ha_selfcheck__t1ha0(void);
 
-/* Define T1HA0_AESNI_AVAILABLE to 0 for disable AES-NI support. */
-#ifndef T1HA0_AESNI_AVAILABLE
-#if defined(__e2k__) ||                                                        \
-    (defined(__ia32__) && (!defined(_M_IX86) || _MSC_VER > 1800))
-#define T1HA0_AESNI_AVAILABLE 1
-#else
-#define T1HA0_AESNI_AVAILABLE 0
-#endif
-#endif /* ifndef T1HA0_AESNI_AVAILABLE */
-
-#if T1HA0_AESNI_AVAILABLE
+  /* Define T1HA0_AESNI_AVAILABLE to 0 for disable AES-NI support. */
+  #ifndef T1HA0_AESNI_AVAILABLE
+    #if defined(__e2k__) || \
+        (defined(__ia32__) && (!defined(_M_IX86) || _MSC_VER > 1800))
+      #define T1HA0_AESNI_AVAILABLE 1
+    #else
+      #define T1HA0_AESNI_AVAILABLE 0
+    #endif
+  #endif                                    /* ifndef T1HA0_AESNI_AVAILABLE */
+
+  #if T1HA0_AESNI_AVAILABLE
 T1HA_API int t1ha_selfcheck__t1ha0_ia32aes_noavx(void);
 T1HA_API int t1ha_selfcheck__t1ha0_ia32aes_avx(void);
-#ifndef __e2k__
+    #ifndef __e2k__
 T1HA_API int t1ha_selfcheck__t1ha0_ia32aes_avx2(void);
-#endif
-#endif /* if T1HA0_AESNI_AVAILABLE */
-#endif /* T1HA0_DISABLED */
+    #endif
+  #endif                                        /* if T1HA0_AESNI_AVAILABLE */
+#endif                                                    /* T1HA0_DISABLED */
 
 /******************************************************************************
  *
@@ -521,7 +529,7 @@ T1HA_API void t1ha2_update(t1ha_context_t *__restrict ctx,
 T1HA_API uint64_t t1ha2_final(t1ha_context_t *__restrict ctx,
                               uint64_t *__restrict extra_result /* optional */);
 
-#endif /* T1HA2_DISABLED */
+#endif                                                    /* T1HA2_DISABLED */
 
 /******************************************************************************
  *
@@ -546,7 +554,7 @@ T1HA_API uint64_t t1ha1_le(const void *data, size_t length, uint64_t seed);
 /* The big-endian variant. */
 T1HA_API uint64_t t1ha1_be(const void *data, size_t length, uint64_t seed);
 
-#endif /* T1HA1_DISABLED */
+#endif                                                    /* T1HA1_DISABLED */
 
 /******************************************************************************
  *
@@ -589,131 +597,142 @@ uint64_t t1ha0_32le(const void *data, size_t length, uint64_t seed);
 /* The big-endian variant for 32-bit CPU. */
 uint64_t t1ha0_32be(const void *data, size_t length, uint64_t seed);
 
-/* Define T1HA0_AESNI_AVAILABLE to 0 for disable AES-NI support. */
-#ifndef T1HA0_AESNI_AVAILABLE
-#if defined(__e2k__) ||                                                        \
-    (defined(__ia32__) && (!defined(_M_IX86) || _MSC_VER > 1800))
-#define T1HA0_AESNI_AVAILABLE 1
-#else
-#define T1HA0_AESNI_AVAILABLE 0
-#endif
-#endif /* T1HA0_AESNI_AVAILABLE */
-
-/* Define T1HA0_RUNTIME_SELECT to 0 for disable dispatching t1ha0 at runtime. */
-#ifndef T1HA0_RUNTIME_SELECT
-#if T1HA0_AESNI_AVAILABLE && !defined(__e2k__)
-#define T1HA0_RUNTIME_SELECT 1
-#else
-#define T1HA0_RUNTIME_SELECT 0
-#endif
-#endif /* T1HA0_RUNTIME_SELECT */
-
-#if !T1HA0_RUNTIME_SELECT && !defined(T1HA0_USE_DEFINE)
-#if defined(__LCC__)
-#define T1HA0_USE_DEFINE 1
-#else
-#define T1HA0_USE_DEFINE 0
-#endif
-#endif /* T1HA0_USE_DEFINE */
-
-#if T1HA0_AESNI_AVAILABLE
+  /* Define T1HA0_AESNI_AVAILABLE to 0 for disable AES-NI support. */
+  #ifndef T1HA0_AESNI_AVAILABLE
+    #if defined(__e2k__) || \
+        (defined(__ia32__) && (!defined(_M_IX86) || _MSC_VER > 1800))
+      #define T1HA0_AESNI_AVAILABLE 1
+    #else
+      #define T1HA0_AESNI_AVAILABLE 0
+    #endif
+  #endif                                           /* T1HA0_AESNI_AVAILABLE */
+
+  /* Define T1HA0_RUNTIME_SELECT to 0 for disable dispatching t1ha0 at runtime.
+   */
+  #ifndef T1HA0_RUNTIME_SELECT
+    #if T1HA0_AESNI_AVAILABLE && !defined(__e2k__)
+      #define T1HA0_RUNTIME_SELECT 1
+    #else
+      #define T1HA0_RUNTIME_SELECT 0
+    #endif
+  #endif                                            /* T1HA0_RUNTIME_SELECT */
+
+  #if !T1HA0_RUNTIME_SELECT && !defined(T1HA0_USE_DEFINE)
+    #if defined(__LCC__)
+      #define T1HA0_USE_DEFINE 1
+    #else
+      #define T1HA0_USE_DEFINE 0
+    #endif
+  #endif                                                /* T1HA0_USE_DEFINE */
+
+  #if T1HA0_AESNI_AVAILABLE
 uint64_t t1ha0_ia32aes_noavx(const void *data, size_t length, uint64_t seed);
 uint64_t t1ha0_ia32aes_avx(const void *data, size_t length, uint64_t seed);
-#ifndef __e2k__
+    #ifndef __e2k__
 uint64_t t1ha0_ia32aes_avx2(const void *data, size_t length, uint64_t seed);
-#endif
-#endif /* T1HA0_AESNI_AVAILABLE */
+    #endif
+  #endif                                           /* T1HA0_AESNI_AVAILABLE */
 
-#if T1HA0_RUNTIME_SELECT
+  #if T1HA0_RUNTIME_SELECT
 typedef uint64_t (*t1ha0_function_t)(const void *, size_t, uint64_t);
 T1HA_API t1ha0_function_t t1ha0_resolve(void);
-#if T1HA_USE_INDIRECT_FUNCTIONS
+    #if T1HA_USE_INDIRECT_FUNCTIONS
 T1HA_API uint64_t t1ha0(const void *data, size_t length, uint64_t seed);
-#else
+    #else
 /* Otherwise function pointer will be used.
  * Unfortunately this may cause some overhead calling. */
 T1HA_API extern uint64_t (*t1ha0_funcptr)(const void *data, size_t length,
                                           uint64_t seed);
 static __force_inline uint64_t t1ha0(const void *data, size_t length,
                                      uint64_t seed) {
+
   return t1ha0_funcptr(data, length, seed);
+
 }
-#endif /* T1HA_USE_INDIRECT_FUNCTIONS */
 
-#elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+    #endif                                   /* T1HA_USE_INDIRECT_FUNCTIONS */
 
-#if T1HA0_USE_DEFINE
+  #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
 
-#if (UINTPTR_MAX > 0xffffFFFFul || ULONG_MAX > 0xffffFFFFul) &&                \
-    (!defined(T1HA1_DISABLED) || !defined(T1HA2_DISABLED))
-#if defined(T1HA1_DISABLED)
-#define t1ha0 t1ha2_atonce
-#else
-#define t1ha0 t1ha1_be
-#endif /* T1HA1_DISABLED */
-#else  /* 32/64 */
-#define t1ha0 t1ha0_32be
-#endif /* 32/64 */
+    #if T1HA0_USE_DEFINE
+
+      #if (UINTPTR_MAX > 0xffffFFFFul || ULONG_MAX > 0xffffFFFFul) && \
+          (!defined(T1HA1_DISABLED) || !defined(T1HA2_DISABLED))
+        #if defined(T1HA1_DISABLED)
+          #define t1ha0 t1ha2_atonce
+        #else
+          #define t1ha0 t1ha1_be
+        #endif                                            /* T1HA1_DISABLED */
+      #else                                                        /* 32/64 */
+        #define t1ha0 t1ha0_32be
+      #endif                                                       /* 32/64 */
 
-#else /* T1HA0_USE_DEFINE */
+    #else                                               /* T1HA0_USE_DEFINE */
 
 static __force_inline uint64_t t1ha0(const void *data, size_t length,
                                      uint64_t seed) {
-#if (UINTPTR_MAX > 0xffffFFFFul || ULONG_MAX > 0xffffFFFFul) &&                \
-    (!defined(T1HA1_DISABLED) || !defined(T1HA2_DISABLED))
-#if defined(T1HA1_DISABLED)
+
+      #if (UINTPTR_MAX > 0xffffFFFFul || ULONG_MAX > 0xffffFFFFul) && \
+          (!defined(T1HA1_DISABLED) || !defined(T1HA2_DISABLED))
+        #if defined(T1HA1_DISABLED)
   return t1ha2_atonce(data, length, seed);
-#else
+        #else
   return t1ha1_be(data, length, seed);
-#endif /* T1HA1_DISABLED */
-#else  /* 32/64 */
+        #endif                                            /* T1HA1_DISABLED */
+      #else                                                        /* 32/64 */
   return t1ha0_32be(data, length, seed);
-#endif /* 32/64 */
+      #endif                                                       /* 32/64 */
+
 }
 
-#endif /* !T1HA0_USE_DEFINE */
+    #endif                                             /* !T1HA0_USE_DEFINE */
 
-#else /* !T1HA0_RUNTIME_SELECT && __BYTE_ORDER__ != __ORDER_BIG_ENDIAN__ */
+  #else  /* !T1HA0_RUNTIME_SELECT && __BYTE_ORDER__ != __ORDER_BIG_ENDIAN__ */
 
-#if T1HA0_USE_DEFINE
+    #if T1HA0_USE_DEFINE
 
-#if (UINTPTR_MAX > 0xffffFFFFul || ULONG_MAX > 0xffffFFFFul) &&                \
-    (!defined(T1HA1_DISABLED) || !defined(T1HA2_DISABLED))
-#if defined(T1HA1_DISABLED)
-#define t1ha0 t1ha2_atonce
-#else
-#define t1ha0 t1ha1_le
-#endif /* T1HA1_DISABLED */
-#else  /* 32/64 */
-#define t1ha0 t1ha0_32le
-#endif /* 32/64 */
+      #if (UINTPTR_MAX > 0xffffFFFFul || ULONG_MAX > 0xffffFFFFul) && \
+          (!defined(T1HA1_DISABLED) || !defined(T1HA2_DISABLED))
+        #if defined(T1HA1_DISABLED)
+          #define t1ha0 t1ha2_atonce
+        #else
+          #define t1ha0 t1ha1_le
+        #endif                                            /* T1HA1_DISABLED */
+      #else                                                        /* 32/64 */
+        #define t1ha0 t1ha0_32le
+      #endif                                                       /* 32/64 */
 
-#else
+    #else
 
 static __force_inline uint64_t t1ha0(const void *data, size_t length,
                                      uint64_t seed) {
-#if (UINTPTR_MAX > 0xffffFFFFul || ULONG_MAX > 0xffffFFFFul) &&                \
-    (!defined(T1HA1_DISABLED) || !defined(T1HA2_DISABLED))
-#if defined(T1HA1_DISABLED)
+
+      #if (UINTPTR_MAX > 0xffffFFFFul || ULONG_MAX > 0xffffFFFFul) && \
+          (!defined(T1HA1_DISABLED) || !defined(T1HA2_DISABLED))
+        #if defined(T1HA1_DISABLED)
   return t1ha2_atonce(data, length, seed);
-#else
+        #else
   return t1ha1_le(data, length, seed);
-#endif /* T1HA1_DISABLED */
-#else  /* 32/64 */
+        #endif                                            /* T1HA1_DISABLED */
+      #else                                                        /* 32/64 */
   return t1ha0_32le(data, length, seed);
-#endif /* 32/64 */
+      #endif                                                       /* 32/64 */
+
 }
 
-#endif /* !T1HA0_USE_DEFINE */
+    #endif                                             /* !T1HA0_USE_DEFINE */
 
-#endif /* !T1HA0_RUNTIME_SELECT */
+  #endif                                           /* !T1HA0_RUNTIME_SELECT */
 
-#endif /* T1HA0_DISABLED */
+#endif                                                    /* T1HA0_DISABLED */
 
 #ifdef __cplusplus
+
 }
+
 #endif
 
 #if __GNUC_PREREQ(4, 0)
-#pragma GCC visibility pop
-#endif /* __GNUC_PREREQ(4,0) */
+  #pragma GCC visibility pop
+#endif                                                /* __GNUC_PREREQ(4,0) */
+
diff --git a/include/t1ha0_ia32aes_b.h b/include/t1ha0_ia32aes_b.h
index e8e52638..93b16771 100644
--- a/include/t1ha0_ia32aes_b.h
+++ b/include/t1ha0_ia32aes_b.h
@@ -47,27 +47,34 @@
 #if T1HA0_AESNI_AVAILABLE
 
 uint64_t T1HA_IA32AES_NAME(const void *data, uint32_t len) {
+
   uint64_t a = 0;
   uint64_t b = len;
 
   if (likely(len > 32)) {
+
     __m128i x = _mm_set_epi64x(a, b);
     __m128i y = _mm_aesenc_si128(x, _mm_set_epi64x(prime_0, prime_1));
 
-    const __m128i *v = (const __m128i *)data;
+    const __m128i       *v = (const __m128i *)data;
     const __m128i *const detent =
         (const __m128i *)((const uint8_t *)data + (len & ~15ul));
     data = detent;
 
     if (len & 16) {
+
       x = _mm_add_epi64(x, _mm_loadu_si128(v++));
       y = _mm_aesenc_si128(x, y);
+
     }
+
     len &= 15;
 
     if (v + 7 < detent) {
+
       __m128i salt = y;
       do {
+
         __m128i t = _mm_aesenc_si128(_mm_loadu_si128(v++), salt);
         t = _mm_aesdec_si128(t, _mm_loadu_si128(v++));
         t = _mm_aesdec_si128(t, _mm_loadu_si128(v++));
@@ -82,86 +89,95 @@ uint64_t T1HA_IA32AES_NAME(const void *data, uint32_t len) {
         t = _mm_aesenc_si128(x, t);
         x = _mm_add_epi64(y, x);
         y = t;
+
       } while (v + 7 < detent);
+
     }
 
     while (v < detent) {
+
       __m128i v0y = _mm_add_epi64(y, _mm_loadu_si128(v++));
       __m128i v1x = _mm_sub_epi64(x, _mm_loadu_si128(v++));
       x = _mm_aesdec_si128(x, v0y);
       y = _mm_aesdec_si128(y, v1x);
+
     }
 
     x = _mm_add_epi64(_mm_aesdec_si128(x, _mm_aesenc_si128(y, x)), y);
-#if defined(__x86_64__) || defined(_M_X64)
-#if defined(__SSE4_1__) || defined(__AVX__)
+  #if defined(__x86_64__) || defined(_M_X64)
+    #if defined(__SSE4_1__) || defined(__AVX__)
     a = _mm_extract_epi64(x, 0);
     b = _mm_extract_epi64(x, 1);
-#else
+    #else
     a = _mm_cvtsi128_si64(x);
     b = _mm_cvtsi128_si64(_mm_unpackhi_epi64(x, x));
-#endif
-#else
-#if defined(__SSE4_1__) || defined(__AVX__)
+    #endif
+  #else
+    #if defined(__SSE4_1__) || defined(__AVX__)
     a = (uint32_t)_mm_extract_epi32(x, 0) | (uint64_t)_mm_extract_epi32(x, 1)
                                                 << 32;
     b = (uint32_t)_mm_extract_epi32(x, 2) | (uint64_t)_mm_extract_epi32(x, 3)
                                                 << 32;
-#else
+    #else
     a = (uint32_t)_mm_cvtsi128_si32(x);
     a |= (uint64_t)_mm_cvtsi128_si32(_mm_shuffle_epi32(x, 1)) << 32;
     x = _mm_unpackhi_epi64(x, x);
     b = (uint32_t)_mm_cvtsi128_si32(x);
     b |= (uint64_t)_mm_cvtsi128_si32(_mm_shuffle_epi32(x, 1)) << 32;
-#endif
-#endif
-#ifdef __AVX__
+    #endif
+  #endif
+  #ifdef __AVX__
     _mm256_zeroupper();
-#elif !(defined(_X86_64_) || defined(__x86_64__) || defined(_M_X64) ||         \
-        defined(__e2k__))
+  #elif !(defined(_X86_64_) || defined(__x86_64__) || defined(_M_X64) || \
+          defined(__e2k__))
     _mm_empty();
-#endif
+  #endif
+
   }
 
   const uint64_t *v = (const uint64_t *)data;
   switch (len) {
-  default:
-    mixup64(&a, &b, fetch64_le_unaligned(v++), prime_4);
-  /* fall through */
-  case 24:
-  case 23:
-  case 22:
-  case 21:
-  case 20:
-  case 19:
-  case 18:
-  case 17:
-    mixup64(&b, &a, fetch64_le_unaligned(v++), prime_3);
-  /* fall through */
-  case 16:
-  case 15:
-  case 14:
-  case 13:
-  case 12:
-  case 11:
-  case 10:
-  case 9:
-    mixup64(&a, &b, fetch64_le_unaligned(v++), prime_2);
-  /* fall through */
-  case 8:
-  case 7:
-  case 6:
-  case 5:
-  case 4:
-  case 3:
-  case 2:
-  case 1:
-    mixup64(&b, &a, tail64_le_unaligned(v, len), prime_1);
-  /* fall through */
-  case 0:
-    return final64(a, b);
+
+    default:
+      mixup64(&a, &b, fetch64_le_unaligned(v++), prime_4);
+    /* fall through */
+    case 24:
+    case 23:
+    case 22:
+    case 21:
+    case 20:
+    case 19:
+    case 18:
+    case 17:
+      mixup64(&b, &a, fetch64_le_unaligned(v++), prime_3);
+    /* fall through */
+    case 16:
+    case 15:
+    case 14:
+    case 13:
+    case 12:
+    case 11:
+    case 10:
+    case 9:
+      mixup64(&a, &b, fetch64_le_unaligned(v++), prime_2);
+    /* fall through */
+    case 8:
+    case 7:
+    case 6:
+    case 5:
+    case 4:
+    case 3:
+    case 2:
+    case 1:
+      mixup64(&b, &a, tail64_le_unaligned(v, len), prime_1);
+    /* fall through */
+    case 0:
+      return final64(a, b);
+
   }
+
 }
 
-#endif /* T1HA0_AESNI_AVAILABLE */
+#endif                                             /* T1HA0_AESNI_AVAILABLE */
 #undef T1HA_IA32AES_NAME
+
diff --git a/include/t1ha_bits.h b/include/t1ha_bits.h
index 539369aa..e7a8d53c 100644
--- a/include/t1ha_bits.h
+++ b/include/t1ha_bits.h
@@ -44,30 +44,30 @@
 #pragma once
 
 #if defined(_MSC_VER)
-#pragma warning(disable : 4201) /* nameless struct/union */
-#if _MSC_VER > 1800
-#pragma warning(disable : 4464) /* relative include path contains '..' */
-#endif                          /* 1800 */
-#endif                          /* MSVC */
+  #pragma warning(disable : 4201)                  /* nameless struct/union */
+  #if _MSC_VER > 1800
+    #pragma warning(disable : 4464)  /* relative include path contains '..' */
+  #endif                                                            /* 1800 */
+#endif                                                              /* MSVC */
 #include "t1ha.h"
 
 #ifndef T1HA_USE_FAST_ONESHOT_READ
-/* Define it to 1 for little bit faster code.
- * Unfortunately this may triggering a false-positive alarms from Valgrind,
- * AddressSanitizer and other similar tool.
- * So, define it to 0 for calmness if doubt. */
-#define T1HA_USE_FAST_ONESHOT_READ 1
-#endif /* T1HA_USE_FAST_ONESHOT_READ */
+  /* Define it to 1 for little bit faster code.
+   * Unfortunately this may triggering a false-positive alarms from Valgrind,
+   * AddressSanitizer and other similar tool.
+   * So, define it to 0 for calmness if doubt. */
+  #define T1HA_USE_FAST_ONESHOT_READ 1
+#endif                                        /* T1HA_USE_FAST_ONESHOT_READ */
 
 /*****************************************************************************/
 
-#include <assert.h>  /* for assert() */
-#include <stdbool.h> /* for bool */
-#include <string.h>  /* for memcpy() */
+#include <assert.h>                                         /* for assert() */
+#include <stdbool.h>                                            /* for bool */
+#include <string.h>                                         /* for memcpy() */
 
-#if __BYTE_ORDER__ != __ORDER_LITTLE_ENDIAN__ &&                               \
+#if __BYTE_ORDER__ != __ORDER_LITTLE_ENDIAN__ && \
     __BYTE_ORDER__ != __ORDER_BIG_ENDIAN__
-#error Unsupported byte order.
+  #error Unsupported byte order.
 #endif
 
 #define T1HA_UNALIGNED_ACCESS__UNABLE 0
@@ -75,534 +75,600 @@
 #define T1HA_UNALIGNED_ACCESS__EFFICIENT 2
 
 #ifndef T1HA_SYS_UNALIGNED_ACCESS
-#if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)
-#define T1HA_SYS_UNALIGNED_ACCESS T1HA_UNALIGNED_ACCESS__EFFICIENT
-#elif defined(__ia32__)
-#define T1HA_SYS_UNALIGNED_ACCESS T1HA_UNALIGNED_ACCESS__EFFICIENT
-#elif defined(__e2k__)
-#define T1HA_SYS_UNALIGNED_ACCESS T1HA_UNALIGNED_ACCESS__SLOW
-#elif defined(__ARM_FEATURE_UNALIGNED)
-#define T1HA_SYS_UNALIGNED_ACCESS T1HA_UNALIGNED_ACCESS__EFFICIENT
-#else
-#define T1HA_SYS_UNALIGNED_ACCESS T1HA_UNALIGNED_ACCESS__UNABLE
-#endif
-#endif /* T1HA_SYS_UNALIGNED_ACCESS */
+  #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)
+    #define T1HA_SYS_UNALIGNED_ACCESS T1HA_UNALIGNED_ACCESS__EFFICIENT
+  #elif defined(__ia32__)
+    #define T1HA_SYS_UNALIGNED_ACCESS T1HA_UNALIGNED_ACCESS__EFFICIENT
+  #elif defined(__e2k__)
+    #define T1HA_SYS_UNALIGNED_ACCESS T1HA_UNALIGNED_ACCESS__SLOW
+  #elif defined(__ARM_FEATURE_UNALIGNED)
+    #define T1HA_SYS_UNALIGNED_ACCESS T1HA_UNALIGNED_ACCESS__EFFICIENT
+  #else
+    #define T1HA_SYS_UNALIGNED_ACCESS T1HA_UNALIGNED_ACCESS__UNABLE
+  #endif
+#endif                                         /* T1HA_SYS_UNALIGNED_ACCESS */
 
 #define ALIGNMENT_16 2
 #define ALIGNMENT_32 4
 #if UINTPTR_MAX > 0xffffFFFFul || ULONG_MAX > 0xffffFFFFul
-#define ALIGNMENT_64 8
+  #define ALIGNMENT_64 8
 #else
-#define ALIGNMENT_64 4
+  #define ALIGNMENT_64 4
 #endif
 
 #ifndef PAGESIZE
-#define PAGESIZE 4096
-#endif /* PAGESIZE */
+  #define PAGESIZE 4096
+#endif                                                          /* PAGESIZE */
 
 /***************************************************************************/
 
 #ifndef __has_builtin
-#define __has_builtin(x) (0)
+  #define __has_builtin(x) (0)
 #endif
 
 #ifndef __has_warning
-#define __has_warning(x) (0)
+  #define __has_warning(x) (0)
 #endif
 
 #ifndef __has_feature
-#define __has_feature(x) (0)
+  #define __has_feature(x) (0)
 #endif
 
 #ifndef __has_extension
-#define __has_extension(x) (0)
+  #define __has_extension(x) (0)
 #endif
 
 #if __has_feature(address_sanitizer)
-#define __SANITIZE_ADDRESS__ 1
+  #define __SANITIZE_ADDRESS__ 1
 #endif
 
 #ifndef __optimize
-#if defined(__clang__) && !__has_attribute(__optimize__)
-#define __optimize(ops)
-#elif defined(__GNUC__) || __has_attribute(__optimize__)
-#define __optimize(ops) __attribute__((__optimize__(ops)))
-#else
-#define __optimize(ops)
-#endif
-#endif /* __optimize */
+  #if defined(__clang__) && !__has_attribute(__optimize__)
+    #define __optimize(ops)
+  #elif defined(__GNUC__) || __has_attribute(__optimize__)
+    #define __optimize(ops) __attribute__((__optimize__(ops)))
+  #else
+    #define __optimize(ops)
+  #endif
+#endif                                                        /* __optimize */
 
 #ifndef __cold
-#if defined(__OPTIMIZE__)
-#if defined(__e2k__)
-#define __cold __optimize(1) __attribute__((__cold__))
-#elif defined(__clang__) && !__has_attribute(__cold__) &&                      \
-    __has_attribute(__section__)
-/* just put infrequently used functions in separate section */
-#define __cold __attribute__((__section__("text.unlikely"))) __optimize("Os")
-#elif defined(__GNUC__) || __has_attribute(__cold__)
-#define __cold __attribute__((__cold__)) __optimize("Os")
-#else
-#define __cold __optimize("Os")
-#endif
-#else
-#define __cold
-#endif
-#endif /* __cold */
+  #if defined(__OPTIMIZE__)
+    #if defined(__e2k__)
+      #define __cold __optimize(1) __attribute__((__cold__))
+    #elif defined(__clang__) && !__has_attribute(__cold__) && \
+        __has_attribute(__section__)
+    /* just put infrequently used functions in separate section */
+      #define __cold \
+        __attribute__((__section__("text.unlikely"))) __optimize("Os")
+    #elif defined(__GNUC__) || __has_attribute(__cold__)
+      #define __cold __attribute__((__cold__)) __optimize("Os")
+    #else
+      #define __cold __optimize("Os")
+    #endif
+  #else
+    #define __cold
+  #endif
+#endif                                                            /* __cold */
 
 #if __GNUC_PREREQ(4, 4) || defined(__clang__)
 
-#if defined(__ia32__) || defined(__e2k__)
-#include <x86intrin.h>
-#endif
+  #if defined(__ia32__) || defined(__e2k__)
+    #include <x86intrin.h>
+  #endif
 
-#if defined(__ia32__) && !defined(__cpuid_count)
-#include <cpuid.h>
-#endif
+  #if defined(__ia32__) && !defined(__cpuid_count)
+    #include <cpuid.h>
+  #endif
 
-#if defined(__e2k__)
-#include <e2kbuiltin.h>
-#endif
+  #if defined(__e2k__)
+    #include <e2kbuiltin.h>
+  #endif
 
-#ifndef likely
-#define likely(cond) __builtin_expect(!!(cond), 1)
-#endif
+  #ifndef likely
+    #define likely(cond) __builtin_expect(!!(cond), 1)
+  #endif
 
-#ifndef unlikely
-#define unlikely(cond) __builtin_expect(!!(cond), 0)
-#endif
+  #ifndef unlikely
+    #define unlikely(cond) __builtin_expect(!!(cond), 0)
+  #endif
 
-#if __GNUC_PREREQ(4, 5) || __has_builtin(__builtin_unreachable)
-#define unreachable() __builtin_unreachable()
-#endif
+  #if __GNUC_PREREQ(4, 5) || __has_builtin(__builtin_unreachable)
+    #define unreachable() __builtin_unreachable()
+  #endif
 
-#define bswap64(v) __builtin_bswap64(v)
-#define bswap32(v) __builtin_bswap32(v)
-#if __GNUC_PREREQ(4, 8) || __has_builtin(__builtin_bswap16)
-#define bswap16(v) __builtin_bswap16(v)
-#endif
+  #define bswap64(v) __builtin_bswap64(v)
+  #define bswap32(v) __builtin_bswap32(v)
+  #if __GNUC_PREREQ(4, 8) || __has_builtin(__builtin_bswap16)
+    #define bswap16(v) __builtin_bswap16(v)
+  #endif
 
-#if !defined(__maybe_unused) &&                                                \
-    (__GNUC_PREREQ(4, 3) || __has_attribute(__unused__))
-#define __maybe_unused __attribute__((__unused__))
-#endif
+  #if !defined(__maybe_unused) && \
+      (__GNUC_PREREQ(4, 3) || __has_attribute(__unused__))
+    #define __maybe_unused __attribute__((__unused__))
+  #endif
 
-#if !defined(__always_inline) &&                                               \
-    (__GNUC_PREREQ(3, 2) || __has_attribute(__always_inline__))
-#define __always_inline __inline __attribute__((__always_inline__))
-#endif
+  #if !defined(__always_inline) && \
+      (__GNUC_PREREQ(3, 2) || __has_attribute(__always_inline__))
+    #define __always_inline __inline __attribute__((__always_inline__))
+  #endif
 
-#if defined(__e2k__)
+  #if defined(__e2k__)
 
-#if __iset__ >= 3
-#define mul_64x64_high(a, b) __builtin_e2k_umulhd(a, b)
-#endif /* __iset__ >= 3 */
+    #if __iset__ >= 3
+      #define mul_64x64_high(a, b) __builtin_e2k_umulhd(a, b)
+    #endif                                                 /* __iset__ >= 3 */
+
+    #if __iset__ >= 5
+static __maybe_unused __always_inline unsigned e2k_add64carry_first(
+    uint64_t base, uint64_t addend, uint64_t *sum) {
 
-#if __iset__ >= 5
-static __maybe_unused __always_inline unsigned
-e2k_add64carry_first(uint64_t base, uint64_t addend, uint64_t *sum) {
   *sum = base + addend;
   return (unsigned)__builtin_e2k_addcd_c(base, addend, 0);
+
 }
-#define add64carry_first(base, addend, sum)                                    \
-  e2k_add64carry_first(base, addend, sum)
+\
+      #define add64carry_first(base, addend, sum) \
+        e2k_add64carry_first(base, addend, sum)
+
+static __maybe_unused __always_inline unsigned e2k_add64carry_next(
+    unsigned carry, uint64_t base, uint64_t addend, uint64_t *sum) {
 
-static __maybe_unused __always_inline unsigned
-e2k_add64carry_next(unsigned carry, uint64_t base, uint64_t addend,
-                    uint64_t *sum) {
   *sum = __builtin_e2k_addcd(base, addend, carry);
   return (unsigned)__builtin_e2k_addcd_c(base, addend, carry);
+
 }
-#define add64carry_next(carry, base, addend, sum)                              \
-  e2k_add64carry_next(carry, base, addend, sum)
+\
+      #define add64carry_next(carry, base, addend, sum) \
+        e2k_add64carry_next(carry, base, addend, sum)
 
-static __maybe_unused __always_inline void e2k_add64carry_last(unsigned carry,
-                                                               uint64_t base,
-                                                               uint64_t addend,
+static __maybe_unused __always_inline void e2k_add64carry_last(unsigned  carry,
+                                                               uint64_t  base,
+                                                               uint64_t  addend,
                                                                uint64_t *sum) {
+
   *sum = __builtin_e2k_addcd(base, addend, carry);
+
 }
-#define add64carry_last(carry, base, addend, sum)                              \
-  e2k_add64carry_last(carry, base, addend, sum)
-#endif /* __iset__ >= 5 */
+\
+      #define add64carry_last(carry, base, addend, sum) \
+        e2k_add64carry_last(carry, base, addend, sum)
+    #endif                                                 /* __iset__ >= 5 */
 
-#define fetch64_be_aligned(ptr) ((uint64_t)__builtin_e2k_ld_64s_be(ptr))
-#define fetch32_be_aligned(ptr) ((uint32_t)__builtin_e2k_ld_32u_be(ptr))
+    #define fetch64_be_aligned(ptr) ((uint64_t)__builtin_e2k_ld_64s_be(ptr))
+    #define fetch32_be_aligned(ptr) ((uint32_t)__builtin_e2k_ld_32u_be(ptr))
 
-#endif /* __e2k__ Elbrus */
+  #endif                                                  /* __e2k__ Elbrus */
 
 #elif defined(_MSC_VER)
 
-#if _MSC_FULL_VER < 190024234 && defined(_M_IX86)
-#pragma message(                                                               \
-    "For AES-NI at least \"Microsoft C/C++ Compiler\" version 19.00.24234 (Visual Studio 2015 Update 3) is required.")
-#endif
-#if _MSC_FULL_VER < 191526730
-#pragma message(                                                               \
-    "It is recommended to use \"Microsoft C/C++ Compiler\" version 19.15.26730 (Visual Studio 2017 15.8) or newer.")
-#endif
-#if _MSC_FULL_VER < 180040629
-#error At least "Microsoft C/C++ Compiler" version 18.00.40629 (Visual Studio 2013 Update 5) is required.
-#endif
-
-#pragma warning(push, 1)
-
-#include <intrin.h>
-#include <stdlib.h>
-#define likely(cond) (cond)
-#define unlikely(cond) (cond)
-#define unreachable() __assume(0)
-#define bswap64(v) _byteswap_uint64(v)
-#define bswap32(v) _byteswap_ulong(v)
-#define bswap16(v) _byteswap_ushort(v)
-#define rot64(v, s) _rotr64(v, s)
-#define rot32(v, s) _rotr(v, s)
-#define __always_inline __forceinline
-
-#if defined(_M_X64) || defined(_M_IA64)
-#pragma intrinsic(_umul128)
-#define mul_64x64_128(a, b, ph) _umul128(a, b, ph)
-#pragma intrinsic(_addcarry_u64)
-#define add64carry_first(base, addend, sum) _addcarry_u64(0, base, addend, sum)
-#define add64carry_next(carry, base, addend, sum)                              \
-  _addcarry_u64(carry, base, addend, sum)
-#define add64carry_last(carry, base, addend, sum)                              \
-  (void)_addcarry_u64(carry, base, addend, sum)
-#endif
-
-#if defined(_M_ARM64) || defined(_M_X64) || defined(_M_IA64)
-#pragma intrinsic(__umulh)
-#define mul_64x64_high(a, b) __umulh(a, b)
-#endif
-
-#if defined(_M_IX86)
-#pragma intrinsic(__emulu)
-#define mul_32x32_64(a, b) __emulu(a, b)
+  #if _MSC_FULL_VER < 190024234 && defined(_M_IX86)
+    #pragma message( \
+        "For AES-NI at least \"Microsoft C/C++ Compiler\" version 19.00.24234 (Visual Studio 2015 Update 3) is required.")
+  #endif
+  #if _MSC_FULL_VER < 191526730
+    #pragma message( \
+        "It is recommended to use \"Microsoft C/C++ Compiler\" version 19.15.26730 (Visual Studio 2017 15.8) or newer.")
+  #endif
+  #if _MSC_FULL_VER < 180040629
+    #error At least "Microsoft C/C++ Compiler" version 18.00.40629 (Visual Studio 2013 Update 5) is required.
+  #endif
+
+  #pragma warning(push, 1)
+
+  #include <intrin.h>
+  #include <stdlib.h>
+  #define likely(cond) (cond)
+  #define unlikely(cond) (cond)
+  #define unreachable() __assume(0)
+  #define bswap64(v) _byteswap_uint64(v)
+  #define bswap32(v) _byteswap_ulong(v)
+  #define bswap16(v) _byteswap_ushort(v)
+  #define rot64(v, s) _rotr64(v, s)
+  #define rot32(v, s) _rotr(v, s)
+  #define __always_inline __forceinline
+
+  #if defined(_M_X64) || defined(_M_IA64)
+    #pragma intrinsic(_umul128)
+    #define mul_64x64_128(a, b, ph) _umul128(a, b, ph)
+    #pragma intrinsic(_addcarry_u64)
+    #define add64carry_first(base, addend, sum) \
+      _addcarry_u64(0, base, addend, sum)
+    #define add64carry_next(carry, base, addend, sum) \
+      _addcarry_u64(carry, base, addend, sum)
+    #define add64carry_last(carry, base, addend, sum) \
+      (void)_addcarry_u64(carry, base, addend, sum)
+  #endif
+
+  #if defined(_M_ARM64) || defined(_M_X64) || defined(_M_IA64)
+    #pragma intrinsic(__umulh)
+    #define mul_64x64_high(a, b) __umulh(a, b)
+  #endif
+
+  #if defined(_M_IX86)
+    #pragma intrinsic(__emulu)
+    #define mul_32x32_64(a, b) __emulu(a, b)
+
+    #if _MSC_VER >= 1915            /* LY: workaround for SSA-optimizer bug */
+      #pragma intrinsic(_addcarry_u32)
+      #define add32carry_first(base, addend, sum) \
+        _addcarry_u32(0, base, addend, sum)
+      #define add32carry_next(carry, base, addend, sum) \
+        _addcarry_u32(carry, base, addend, sum)
+      #define add32carry_last(carry, base, addend, sum) \
+        (void)_addcarry_u32(carry, base, addend, sum)
+
+static __forceinline char msvc32_add64carry_first(uint64_t  base,
+                                                  uint64_t  addend,
+                                                  uint64_t *sum) {
 
-#if _MSC_VER >= 1915 /* LY: workaround for SSA-optimizer bug */
-#pragma intrinsic(_addcarry_u32)
-#define add32carry_first(base, addend, sum) _addcarry_u32(0, base, addend, sum)
-#define add32carry_next(carry, base, addend, sum)                              \
-  _addcarry_u32(carry, base, addend, sum)
-#define add32carry_last(carry, base, addend, sum)                              \
-  (void)_addcarry_u32(carry, base, addend, sum)
-
-static __forceinline char
-msvc32_add64carry_first(uint64_t base, uint64_t addend, uint64_t *sum) {
   uint32_t *const sum32 = (uint32_t *)sum;
-  const uint32_t base_32l = (uint32_t)base;
-  const uint32_t base_32h = (uint32_t)(base >> 32);
-  const uint32_t addend_32l = (uint32_t)addend;
-  const uint32_t addend_32h = (uint32_t)(addend >> 32);
+  const uint32_t  base_32l = (uint32_t)base;
+  const uint32_t  base_32h = (uint32_t)(base >> 32);
+  const uint32_t  addend_32l = (uint32_t)addend;
+  const uint32_t  addend_32h = (uint32_t)(addend >> 32);
   return add32carry_next(add32carry_first(base_32l, addend_32l, sum32),
                          base_32h, addend_32h, sum32 + 1);
+
 }
-#define add64carry_first(base, addend, sum)                                    \
-  msvc32_add64carry_first(base, addend, sum)
+\
+      #define add64carry_first(base, addend, sum) \
+        msvc32_add64carry_first(base, addend, sum)
 
 static __forceinline char msvc32_add64carry_next(char carry, uint64_t base,
-                                                 uint64_t addend,
+                                                 uint64_t  addend,
                                                  uint64_t *sum) {
+
   uint32_t *const sum32 = (uint32_t *)sum;
-  const uint32_t base_32l = (uint32_t)base;
-  const uint32_t base_32h = (uint32_t)(base >> 32);
-  const uint32_t addend_32l = (uint32_t)addend;
-  const uint32_t addend_32h = (uint32_t)(addend >> 32);
+  const uint32_t  base_32l = (uint32_t)base;
+  const uint32_t  base_32h = (uint32_t)(base >> 32);
+  const uint32_t  addend_32l = (uint32_t)addend;
+  const uint32_t  addend_32h = (uint32_t)(addend >> 32);
   return add32carry_next(add32carry_next(carry, base_32l, addend_32l, sum32),
                          base_32h, addend_32h, sum32 + 1);
+
 }
-#define add64carry_next(carry, base, addend, sum)                              \
-  msvc32_add64carry_next(carry, base, addend, sum)
+\
+      #define add64carry_next(carry, base, addend, sum) \
+        msvc32_add64carry_next(carry, base, addend, sum)
 
 static __forceinline void msvc32_add64carry_last(char carry, uint64_t base,
-                                                 uint64_t addend,
+                                                 uint64_t  addend,
                                                  uint64_t *sum) {
+
   uint32_t *const sum32 = (uint32_t *)sum;
-  const uint32_t base_32l = (uint32_t)base;
-  const uint32_t base_32h = (uint32_t)(base >> 32);
-  const uint32_t addend_32l = (uint32_t)addend;
-  const uint32_t addend_32h = (uint32_t)(addend >> 32);
+  const uint32_t  base_32l = (uint32_t)base;
+  const uint32_t  base_32h = (uint32_t)(base >> 32);
+  const uint32_t  addend_32l = (uint32_t)addend;
+  const uint32_t  addend_32h = (uint32_t)(addend >> 32);
   add32carry_last(add32carry_next(carry, base_32l, addend_32l, sum32), base_32h,
                   addend_32h, sum32 + 1);
-}
-#define add64carry_last(carry, base, addend, sum)                              \
-  msvc32_add64carry_last(carry, base, addend, sum)
-#endif /* _MSC_FULL_VER >= 190024231 */
-
-#elif defined(_M_ARM)
-#define mul_32x32_64(a, b) _arm_umull(a, b)
-#endif
 
-#pragma warning(pop)
-#pragma warning(disable : 4514) /* 'xyz': unreferenced inline function         \
-                                   has been removed */
-#pragma warning(disable : 4710) /* 'xyz': function not inlined */
-#pragma warning(disable : 4711) /* function 'xyz' selected for                 \
-                                   automatic inline expansion */
-#pragma warning(disable : 4127) /* conditional expression is constant */
-#pragma warning(disable : 4702) /* unreachable code */
-#endif                          /* Compiler */
+}
+\
+      #define add64carry_last(carry, base, addend, sum) \
+        msvc32_add64carry_last(carry, base, addend, sum)
+    #endif                                    /* _MSC_FULL_VER >= 190024231 */
+
+  #elif defined(_M_ARM)
+    #define mul_32x32_64(a, b) _arm_umull(a, b)
+  #endif
+
+  #pragma warning(pop)
+  #pragma warning(disable : 4514) /* 'xyz': unreferenced inline function \
+                                     has been removed */
+  #pragma warning(disable : 4710)            /* 'xyz': function not inlined */
+  #pragma warning(disable : 4711) /* function 'xyz' selected for \
+                                     automatic inline expansion */
+  #pragma warning(disable : 4127)     /* conditional expression is constant */
+  #pragma warning(disable : 4702)                       /* unreachable code */
+#endif                                                          /* Compiler */
 
 #ifndef likely
-#define likely(cond) (cond)
+  #define likely(cond) (cond)
 #endif
 #ifndef unlikely
-#define unlikely(cond) (cond)
+  #define unlikely(cond) (cond)
 #endif
 #ifndef __maybe_unused
-#define __maybe_unused
+  #define __maybe_unused
 #endif
 #ifndef __always_inline
-#define __always_inline __inline
+  #define __always_inline __inline
 #endif
 #ifndef unreachable
-#define unreachable()                                                          \
-  do {                                                                         \
-  } while (1)
+  #define unreachable() \
+    do {                \
+                        \
+    } while (1)
 #endif
 
 #ifndef bswap64
-#if defined(bswap_64)
-#define bswap64 bswap_64
-#elif defined(__bswap_64)
-#define bswap64 __bswap_64
-#else
+  #if defined(bswap_64)
+    #define bswap64 bswap_64
+  #elif defined(__bswap_64)
+    #define bswap64 __bswap_64
+  #else
 static __always_inline uint64_t bswap64(uint64_t v) {
+
   return v << 56 | v >> 56 | ((v << 40) & UINT64_C(0x00ff000000000000)) |
          ((v << 24) & UINT64_C(0x0000ff0000000000)) |
          ((v << 8) & UINT64_C(0x000000ff00000000)) |
          ((v >> 8) & UINT64_C(0x00000000ff000000)) |
          ((v >> 24) & UINT64_C(0x0000000000ff0000)) |
          ((v >> 40) & UINT64_C(0x000000000000ff00));
+
 }
-#endif
-#endif /* bswap64 */
+
+  #endif
+#endif                                                           /* bswap64 */
 
 #ifndef bswap32
-#if defined(bswap_32)
-#define bswap32 bswap_32
-#elif defined(__bswap_32)
-#define bswap32 __bswap_32
-#else
+  #if defined(bswap_32)
+    #define bswap32 bswap_32
+  #elif defined(__bswap_32)
+    #define bswap32 __bswap_32
+  #else
 static __always_inline uint32_t bswap32(uint32_t v) {
+
   return v << 24 | v >> 24 | ((v << 8) & UINT32_C(0x00ff0000)) |
          ((v >> 8) & UINT32_C(0x0000ff00));
+
 }
-#endif
-#endif /* bswap32 */
+
+  #endif
+#endif                                                           /* bswap32 */
 
 #ifndef bswap16
-#if defined(bswap_16)
-#define bswap16 bswap_16
-#elif defined(__bswap_16)
-#define bswap16 __bswap_16
-#else
-static __always_inline uint16_t bswap16(uint16_t v) { return v << 8 | v >> 8; }
-#endif
-#endif /* bswap16 */
+  #if defined(bswap_16)
+    #define bswap16 bswap_16
+  #elif defined(__bswap_16)
+    #define bswap16 __bswap_16
+  #else
+static __always_inline uint16_t bswap16(uint16_t v) {
+
+  return v << 8 | v >> 8;
+
+}
+
+  #endif
+#endif                                                           /* bswap16 */
 
-#if defined(__ia32__) ||                                                       \
+#if defined(__ia32__) || \
     T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__EFFICIENT
-/* The __builtin_assume_aligned() leads gcc/clang to load values into the
- * registers, even when it is possible to directly use an operand from memory.
- * This can lead to a shortage of registers and a significant slowdown.
- * Therefore avoid unnecessary use of  __builtin_assume_aligned() for x86. */
-#define read_unaligned(ptr, bits) (*(const uint##bits##_t *__restrict)(ptr))
-#define read_aligned(ptr, bits) (*(const uint##bits##_t *__restrict)(ptr))
-#endif /* __ia32__ */
+  /* The __builtin_assume_aligned() leads gcc/clang to load values into the
+   * registers, even when it is possible to directly use an operand from memory.
+   * This can lead to a shortage of registers and a significant slowdown.
+   * Therefore avoid unnecessary use of  __builtin_assume_aligned() for x86. */
+  #define read_unaligned(ptr, bits) (*(const uint##bits##_t *__restrict)(ptr))
+  #define read_aligned(ptr, bits) (*(const uint##bits##_t *__restrict)(ptr))
+#endif                                                          /* __ia32__ */
 
 #ifndef read_unaligned
-#if defined(__GNUC__) || __has_attribute(__packed__)
+  #if defined(__GNUC__) || __has_attribute(__packed__)
 typedef struct {
-  uint8_t unaligned_8;
+
+  uint8_t  unaligned_8;
   uint16_t unaligned_16;
   uint32_t unaligned_32;
   uint64_t unaligned_64;
+
 } __attribute__((__packed__)) t1ha_unaligned_proxy;
-#define read_unaligned(ptr, bits)                                              \
-  (((const t1ha_unaligned_proxy *)((const uint8_t *)(ptr)-offsetof(            \
-        t1ha_unaligned_proxy, unaligned_##bits)))                              \
-       ->unaligned_##bits)
-#elif defined(_MSC_VER)
-#pragma warning(                                                               \
-    disable : 4235) /* nonstandard extension used: '__unaligned'               \
-                     * keyword not supported on this architecture */
-#define read_unaligned(ptr, bits) (*(const __unaligned uint##bits##_t *)(ptr))
-#else
-#pragma pack(push, 1)
+\
+    #define read_unaligned(ptr, bits)                                   \
+      (((const t1ha_unaligned_proxy *)((const uint8_t *)(ptr)-offsetof( \
+            t1ha_unaligned_proxy, unaligned_##bits)))                   \
+           ->unaligned_##bits)
+  #elif defined(_MSC_VER)
+    #pragma warning(                                                 \
+        disable : 4235) /* nonstandard extension used: '__unaligned' \
+                         * keyword not supported on this architecture */
+    #define read_unaligned(ptr, bits) \
+      (*(const __unaligned uint##bits##_t *)(ptr))
+  #else
+    #pragma pack(push, 1)
 typedef struct {
-  uint8_t unaligned_8;
+
+  uint8_t  unaligned_8;
   uint16_t unaligned_16;
   uint32_t unaligned_32;
   uint64_t unaligned_64;
+
 } t1ha_unaligned_proxy;
-#pragma pack(pop)
-#define read_unaligned(ptr, bits)                                              \
-  (((const t1ha_unaligned_proxy *)((const uint8_t *)(ptr)-offsetof(            \
-        t1ha_unaligned_proxy, unaligned_##bits)))                              \
-       ->unaligned_##bits)
-#endif
-#endif /* read_unaligned */
+
+    #pragma pack(pop)
+    #define read_unaligned(ptr, bits)                                   \
+      (((const t1ha_unaligned_proxy *)((const uint8_t *)(ptr)-offsetof( \
+            t1ha_unaligned_proxy, unaligned_##bits)))                   \
+           ->unaligned_##bits)
+  #endif
+#endif                                                    /* read_unaligned */
 
 #ifndef read_aligned
-#if __GNUC_PREREQ(4, 8) || __has_builtin(__builtin_assume_aligned)
-#define read_aligned(ptr, bits)                                                \
-  (*(const uint##bits##_t *)__builtin_assume_aligned(ptr, ALIGNMENT_##bits))
-#elif (__GNUC_PREREQ(3, 3) || __has_attribute(__aligned__)) &&                 \
-    !defined(__clang__)
-#define read_aligned(ptr, bits)                                                \
-  (*(const uint##bits##_t                                                      \
-     __attribute__((__aligned__(ALIGNMENT_##bits))) *)(ptr))
-#elif __has_attribute(__assume_aligned__)
-
-static __always_inline const
-    uint16_t *__attribute__((__assume_aligned__(ALIGNMENT_16)))
-    cast_aligned_16(const void *ptr) {
+  #if __GNUC_PREREQ(4, 8) || __has_builtin(__builtin_assume_aligned)
+    #define read_aligned(ptr, bits) \
+      (*(const uint##bits##_t *)__builtin_assume_aligned(ptr, ALIGNMENT_##bits))
+  #elif (__GNUC_PREREQ(3, 3) || __has_attribute(__aligned__)) && \
+      !defined(__clang__)
+    #define read_aligned(ptr, bits) \
+      (*(const uint##bits##_t       \
+         __attribute__((__aligned__(ALIGNMENT_##bits))) *)(ptr))
+  #elif __has_attribute(__assume_aligned__)
+
+static __always_inline const uint16_t *__attribute__((
+    __assume_aligned__(ALIGNMENT_16))) cast_aligned_16(const void *ptr) {
+
   return (const uint16_t *)ptr;
+
 }
-static __always_inline const
-    uint32_t *__attribute__((__assume_aligned__(ALIGNMENT_32)))
-    cast_aligned_32(const void *ptr) {
+
+static __always_inline const uint32_t *__attribute__((
+    __assume_aligned__(ALIGNMENT_32))) cast_aligned_32(const void *ptr) {
+
   return (const uint32_t *)ptr;
+
 }
-static __always_inline const
-    uint64_t *__attribute__((__assume_aligned__(ALIGNMENT_64)))
-    cast_aligned_64(const void *ptr) {
+
+static __always_inline const uint64_t *__attribute__((
+    __assume_aligned__(ALIGNMENT_64))) cast_aligned_64(const void *ptr) {
+
   return (const uint64_t *)ptr;
+
 }
 
-#define read_aligned(ptr, bits) (*cast_aligned_##bits(ptr))
+    #define read_aligned(ptr, bits) (*cast_aligned_##bits(ptr))
 
-#elif defined(_MSC_VER)
-#define read_aligned(ptr, bits)                                                \
-  (*(const __declspec(align(ALIGNMENT_##bits)) uint##bits##_t *)(ptr))
-#else
-#define read_aligned(ptr, bits) (*(const uint##bits##_t *)(ptr))
-#endif
-#endif /* read_aligned */
+  #elif defined(_MSC_VER)
+    #define read_aligned(ptr, bits) \
+      (*(const __declspec(align(ALIGNMENT_##bits)) uint##bits##_t *)(ptr))
+  #else
+    #define read_aligned(ptr, bits) (*(const uint##bits##_t *)(ptr))
+  #endif
+#endif                                                      /* read_aligned */
 
 #ifndef prefetch
-#if (__GNUC_PREREQ(4, 0) || __has_builtin(__builtin_prefetch)) &&              \
-    !defined(__ia32__)
-#define prefetch(ptr) __builtin_prefetch(ptr)
-#elif defined(_M_ARM64) || defined(_M_ARM)
-#define prefetch(ptr) __prefetch(ptr)
-#else
-#define prefetch(ptr)                                                          \
-  do {                                                                         \
-    (void)(ptr);                                                               \
-  } while (0)
-#endif
-#endif /* prefetch */
+  #if (__GNUC_PREREQ(4, 0) || __has_builtin(__builtin_prefetch)) && \
+      !defined(__ia32__)
+    #define prefetch(ptr) __builtin_prefetch(ptr)
+  #elif defined(_M_ARM64) || defined(_M_ARM)
+    #define prefetch(ptr) __prefetch(ptr)
+  #else
+    #define prefetch(ptr) \
+      do {                \
+                          \
+        (void)(ptr);      \
+                          \
+      } while (0)
+  #endif
+#endif                                                          /* prefetch */
 
 #if __has_warning("-Wconstant-logical-operand")
-#if defined(__clang__)
-#pragma clang diagnostic ignored "-Wconstant-logical-operand"
-#elif defined(__GNUC__)
-#pragma GCC diagnostic ignored "-Wconstant-logical-operand"
-#else
-#pragma warning disable "constant-logical-operand"
-#endif
-#endif /* -Wconstant-logical-operand */
+  #if defined(__clang__)
+    #pragma clang diagnostic ignored "-Wconstant-logical-operand"
+  #elif defined(__GNUC__)
+    #pragma GCC diagnostic ignored "-Wconstant-logical-operand"
+  #else
+    #pragma warning disable "constant-logical-operand"
+  #endif
+#endif                                        /* -Wconstant-logical-operand */
 
 #if __has_warning("-Wtautological-pointer-compare")
-#if defined(__clang__)
-#pragma clang diagnostic ignored "-Wtautological-pointer-compare"
-#elif defined(__GNUC__)
-#pragma GCC diagnostic ignored "-Wtautological-pointer-compare"
-#else
-#pragma warning disable "tautological-pointer-compare"
-#endif
-#endif /* -Wtautological-pointer-compare */
+  #if defined(__clang__)
+    #pragma clang diagnostic ignored "-Wtautological-pointer-compare"
+  #elif defined(__GNUC__)
+    #pragma GCC diagnostic ignored "-Wtautological-pointer-compare"
+  #else
+    #pragma warning disable "tautological-pointer-compare"
+  #endif
+#endif                                    /* -Wtautological-pointer-compare */
 
 /***************************************************************************/
 
 #if __GNUC_PREREQ(4, 0)
-#pragma GCC visibility push(hidden)
-#endif /* __GNUC_PREREQ(4,0) */
+  #pragma GCC visibility push(hidden)
+#endif                                                /* __GNUC_PREREQ(4,0) */
 
 /*---------------------------------------------------------- Little Endian */
 
 #ifndef fetch16_le_aligned
 static __maybe_unused __always_inline uint16_t
 fetch16_le_aligned(const void *v) {
+
   assert(((uintptr_t)v) % ALIGNMENT_16 == 0);
-#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+  #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
   return read_aligned(v, 16);
-#else
+  #else
   return bswap16(read_aligned(v, 16));
-#endif
+  #endif
+
 }
-#endif /* fetch16_le_aligned */
+
+#endif                                                /* fetch16_le_aligned */
 
 #ifndef fetch16_le_unaligned
 static __maybe_unused __always_inline uint16_t
 fetch16_le_unaligned(const void *v) {
-#if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__UNABLE
+
+  #if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__UNABLE
   const uint8_t *p = (const uint8_t *)v;
   return p[0] | (uint16_t)p[1] << 8;
-#elif __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+  #elif __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
   return read_unaligned(v, 16);
-#else
+  #else
   return bswap16(read_unaligned(v, 16));
-#endif
+  #endif
+
 }
-#endif /* fetch16_le_unaligned */
+
+#endif                                              /* fetch16_le_unaligned */
 
 #ifndef fetch32_le_aligned
 static __maybe_unused __always_inline uint32_t
 fetch32_le_aligned(const void *v) {
+
   assert(((uintptr_t)v) % ALIGNMENT_32 == 0);
-#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+  #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
   return read_aligned(v, 32);
-#else
+  #else
   return bswap32(read_aligned(v, 32));
-#endif
+  #endif
+
 }
-#endif /* fetch32_le_aligned */
+
+#endif                                                /* fetch32_le_aligned */
 
 #ifndef fetch32_le_unaligned
 static __maybe_unused __always_inline uint32_t
 fetch32_le_unaligned(const void *v) {
-#if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__UNABLE
+
+  #if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__UNABLE
   return fetch16_le_unaligned(v) |
          (uint32_t)fetch16_le_unaligned((const uint8_t *)v + 2) << 16;
-#elif __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+  #elif __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
   return read_unaligned(v, 32);
-#else
+  #else
   return bswap32(read_unaligned(v, 32));
-#endif
+  #endif
+
 }
-#endif /* fetch32_le_unaligned */
+
+#endif                                              /* fetch32_le_unaligned */
 
 #ifndef fetch64_le_aligned
 static __maybe_unused __always_inline uint64_t
 fetch64_le_aligned(const void *v) {
+
   assert(((uintptr_t)v) % ALIGNMENT_64 == 0);
-#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+  #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
   return read_aligned(v, 64);
-#else
+  #else
   return bswap64(read_aligned(v, 64));
-#endif
+  #endif
+
 }
-#endif /* fetch64_le_aligned */
+
+#endif                                                /* fetch64_le_aligned */
 
 #ifndef fetch64_le_unaligned
 static __maybe_unused __always_inline uint64_t
 fetch64_le_unaligned(const void *v) {
-#if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__UNABLE
+
+  #if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__UNABLE
   return fetch32_le_unaligned(v) |
          (uint64_t)fetch32_le_unaligned((const uint8_t *)v + 4) << 32;
-#elif __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+  #elif __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
   return read_unaligned(v, 64);
-#else
+  #else
   return bswap64(read_unaligned(v, 64));
-#endif
+  #endif
+
 }
-#endif /* fetch64_le_unaligned */
+
+#endif                                              /* fetch64_le_unaligned */
 
 static __maybe_unused __always_inline uint64_t tail64_le_aligned(const void *v,
                                                                  size_t tail) {
+
   const uint8_t *const p = (const uint8_t *)v;
 #if T1HA_USE_FAST_ONESHOT_READ && !defined(__SANITIZE_ADDRESS__)
   /* We can perform a 'oneshot' read, which is little bit faster. */
@@ -611,79 +677,84 @@ static __maybe_unused __always_inline uint64_t tail64_le_aligned(const void *v,
 #else
   uint64_t r = 0;
   switch (tail & 7) {
-  default:
-    unreachable();
-/* fall through */
-#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
-  /* For most CPUs this code is better when not needed byte reordering. */
-  case 0:
-    return fetch64_le_aligned(p);
-  case 7:
-    r = (uint64_t)p[6] << 8;
-  /* fall through */
-  case 6:
-    r += p[5];
-    r <<= 8;
-  /* fall through */
-  case 5:
-    r += p[4];
-    r <<= 32;
-  /* fall through */
-  case 4:
-    return r + fetch32_le_aligned(p);
-  case 3:
-    r = (uint64_t)p[2] << 16;
-  /* fall through */
-  case 2:
-    return r + fetch16_le_aligned(p);
-  case 1:
-    return p[0];
-#else
-  case 0:
-    r = p[7] << 8;
-  /* fall through */
-  case 7:
-    r += p[6];
-    r <<= 8;
-  /* fall through */
-  case 6:
-    r += p[5];
-    r <<= 8;
-  /* fall through */
-  case 5:
-    r += p[4];
-    r <<= 8;
-  /* fall through */
-  case 4:
-    r += p[3];
-    r <<= 8;
-  /* fall through */
-  case 3:
-    r += p[2];
-    r <<= 8;
-  /* fall through */
-  case 2:
-    r += p[1];
-    r <<= 8;
+
+    default:
+      unreachable();
   /* fall through */
-  case 1:
-    return r + p[0];
-#endif
+  #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+    /* For most CPUs this code is better when not needed byte reordering. */
+    case 0:
+      return fetch64_le_aligned(p);
+    case 7:
+      r = (uint64_t)p[6] << 8;
+    /* fall through */
+    case 6:
+      r += p[5];
+      r <<= 8;
+    /* fall through */
+    case 5:
+      r += p[4];
+      r <<= 32;
+    /* fall through */
+    case 4:
+      return r + fetch32_le_aligned(p);
+    case 3:
+      r = (uint64_t)p[2] << 16;
+    /* fall through */
+    case 2:
+      return r + fetch16_le_aligned(p);
+    case 1:
+      return p[0];
+  #else
+    case 0:
+      r = p[7] << 8;
+    /* fall through */
+    case 7:
+      r += p[6];
+      r <<= 8;
+    /* fall through */
+    case 6:
+      r += p[5];
+      r <<= 8;
+    /* fall through */
+    case 5:
+      r += p[4];
+      r <<= 8;
+    /* fall through */
+    case 4:
+      r += p[3];
+      r <<= 8;
+    /* fall through */
+    case 3:
+      r += p[2];
+      r <<= 8;
+    /* fall through */
+    case 2:
+      r += p[1];
+      r <<= 8;
+    /* fall through */
+    case 1:
+      return r + p[0];
+  #endif
+
   }
-#endif /* T1HA_USE_FAST_ONESHOT_READ */
+
+#endif                                        /* T1HA_USE_FAST_ONESHOT_READ */
+
 }
 
-#if T1HA_USE_FAST_ONESHOT_READ &&                                              \
-    T1HA_SYS_UNALIGNED_ACCESS != T1HA_UNALIGNED_ACCESS__UNABLE &&              \
+#if T1HA_USE_FAST_ONESHOT_READ &&                                 \
+    T1HA_SYS_UNALIGNED_ACCESS != T1HA_UNALIGNED_ACCESS__UNABLE && \
     defined(PAGESIZE) && PAGESIZE > 42 && !defined(__SANITIZE_ADDRESS__)
-#define can_read_underside(ptr, size)                                          \
-  (((PAGESIZE - (size)) & (uintptr_t)(ptr)) != 0)
-#endif /* T1HA_USE_FAST_ONESHOT_READ */
+  #define can_read_underside(ptr, size) \
+    (((PAGESIZE - (size)) & (uintptr_t)(ptr)) != 0)
+#endif                                        /* T1HA_USE_FAST_ONESHOT_READ */
 
 static __maybe_unused __always_inline uint64_t
 tail64_le_unaligned(const void *v, size_t tail) {
+
   const uint8_t *p = (const uint8_t *)v;
-#if defined(can_read_underside) &&                                             \
+#if defined(can_read_underside) && \
     (UINTPTR_MAX > 0xffffFFFFul || ULONG_MAX > 0xffffFFFFul)
   /* On some systems (e.g. x86_64) we can perform a 'oneshot' read, which
    * is little bit faster. Thanks Marcin Żukowski <marcin.zukowski@gmail.com>
@@ -691,77 +762,84 @@ tail64_le_unaligned(const void *v, size_t tail) {
   const unsigned offset = (8 - tail) & 7;
   const unsigned shift = offset << 3;
   if (likely(can_read_underside(p, 8))) {
+
     p -= offset;
     return fetch64_le_unaligned(p) >> shift;
+
   }
+
   return fetch64_le_unaligned(p) & ((~UINT64_C(0)) >> shift);
 #else
   uint64_t r = 0;
   switch (tail & 7) {
-  default:
-    unreachable();
-/* fall through */
-#if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__EFFICIENT &&           \
-    __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
-  /* For most CPUs this code is better when not needed
-   * copying for alignment or byte reordering. */
-  case 0:
-    return fetch64_le_unaligned(p);
-  case 7:
-    r = (uint64_t)p[6] << 8;
-  /* fall through */
-  case 6:
-    r += p[5];
-    r <<= 8;
-  /* fall through */
-  case 5:
-    r += p[4];
-    r <<= 32;
-  /* fall through */
-  case 4:
-    return r + fetch32_le_unaligned(p);
-  case 3:
-    r = (uint64_t)p[2] << 16;
-  /* fall through */
-  case 2:
-    return r + fetch16_le_unaligned(p);
-  case 1:
-    return p[0];
-#else
-  /* For most CPUs this code is better than a
-   * copying for alignment and/or byte reordering. */
-  case 0:
-    r = p[7] << 8;
-  /* fall through */
-  case 7:
-    r += p[6];
-    r <<= 8;
-  /* fall through */
-  case 6:
-    r += p[5];
-    r <<= 8;
-  /* fall through */
-  case 5:
-    r += p[4];
-    r <<= 8;
-  /* fall through */
-  case 4:
-    r += p[3];
-    r <<= 8;
-  /* fall through */
-  case 3:
-    r += p[2];
-    r <<= 8;
-  /* fall through */
-  case 2:
-    r += p[1];
-    r <<= 8;
+
+    default:
+      unreachable();
   /* fall through */
-  case 1:
-    return r + p[0];
-#endif
+  #if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__EFFICIENT && \
+      __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+    /* For most CPUs this code is better when not needed
+     * copying for alignment or byte reordering. */
+    case 0:
+      return fetch64_le_unaligned(p);
+    case 7:
+      r = (uint64_t)p[6] << 8;
+    /* fall through */
+    case 6:
+      r += p[5];
+      r <<= 8;
+    /* fall through */
+    case 5:
+      r += p[4];
+      r <<= 32;
+    /* fall through */
+    case 4:
+      return r + fetch32_le_unaligned(p);
+    case 3:
+      r = (uint64_t)p[2] << 16;
+    /* fall through */
+    case 2:
+      return r + fetch16_le_unaligned(p);
+    case 1:
+      return p[0];
+  #else
+    /* For most CPUs this code is better than a
+     * copying for alignment and/or byte reordering. */
+    case 0:
+      r = p[7] << 8;
+    /* fall through */
+    case 7:
+      r += p[6];
+      r <<= 8;
+    /* fall through */
+    case 6:
+      r += p[5];
+      r <<= 8;
+    /* fall through */
+    case 5:
+      r += p[4];
+      r <<= 8;
+    /* fall through */
+    case 4:
+      r += p[3];
+      r <<= 8;
+    /* fall through */
+    case 3:
+      r += p[2];
+      r <<= 8;
+    /* fall through */
+    case 2:
+      r += p[1];
+      r <<= 8;
+    /* fall through */
+    case 1:
+      return r + p[0];
+  #endif
+
   }
-#endif /* can_read_underside */
+
+#endif                                                /* can_read_underside */
+
 }
 
 /*------------------------------------------------------------- Big Endian */
@@ -769,83 +847,102 @@ tail64_le_unaligned(const void *v, size_t tail) {
 #ifndef fetch16_be_aligned
 static __maybe_unused __always_inline uint16_t
 fetch16_be_aligned(const void *v) {
+
   assert(((uintptr_t)v) % ALIGNMENT_16 == 0);
-#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+  #if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
   return read_aligned(v, 16);
-#else
+  #else
   return bswap16(read_aligned(v, 16));
-#endif
+  #endif
+
 }
-#endif /* fetch16_be_aligned */
+
+#endif                                                /* fetch16_be_aligned */
 
 #ifndef fetch16_be_unaligned
 static __maybe_unused __always_inline uint16_t
 fetch16_be_unaligned(const void *v) {
-#if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__UNABLE
+
+  #if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__UNABLE
   const uint8_t *p = (const uint8_t *)v;
   return (uint16_t)p[0] << 8 | p[1];
-#elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+  #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
   return read_unaligned(v, 16);
-#else
+  #else
   return bswap16(read_unaligned(v, 16));
-#endif
+  #endif
+
 }
-#endif /* fetch16_be_unaligned */
+
+#endif                                              /* fetch16_be_unaligned */
 
 #ifndef fetch32_be_aligned
 static __maybe_unused __always_inline uint32_t
 fetch32_be_aligned(const void *v) {
+
   assert(((uintptr_t)v) % ALIGNMENT_32 == 0);
-#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+  #if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
   return read_aligned(v, 32);
-#else
+  #else
   return bswap32(read_aligned(v, 32));
-#endif
+  #endif
+
 }
-#endif /* fetch32_be_aligned */
+
+#endif                                                /* fetch32_be_aligned */
 
 #ifndef fetch32_be_unaligned
 static __maybe_unused __always_inline uint32_t
 fetch32_be_unaligned(const void *v) {
-#if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__UNABLE
+
+  #if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__UNABLE
   return (uint32_t)fetch16_be_unaligned(v) << 16 |
          fetch16_be_unaligned((const uint8_t *)v + 2);
-#elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+  #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
   return read_unaligned(v, 32);
-#else
+  #else
   return bswap32(read_unaligned(v, 32));
-#endif
+  #endif
+
 }
-#endif /* fetch32_be_unaligned */
+
+#endif                                              /* fetch32_be_unaligned */
 
 #ifndef fetch64_be_aligned
 static __maybe_unused __always_inline uint64_t
 fetch64_be_aligned(const void *v) {
+
   assert(((uintptr_t)v) % ALIGNMENT_64 == 0);
-#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+  #if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
   return read_aligned(v, 64);
-#else
+  #else
   return bswap64(read_aligned(v, 64));
-#endif
+  #endif
+
 }
-#endif /* fetch64_be_aligned */
+
+#endif                                                /* fetch64_be_aligned */
 
 #ifndef fetch64_be_unaligned
 static __maybe_unused __always_inline uint64_t
 fetch64_be_unaligned(const void *v) {
-#if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__UNABLE
+
+  #if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__UNABLE
   return (uint64_t)fetch32_be_unaligned(v) << 32 |
          fetch32_be_unaligned((const uint8_t *)v + 4);
-#elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+  #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
   return read_unaligned(v, 64);
-#else
+  #else
   return bswap64(read_unaligned(v, 64));
-#endif
+  #endif
+
 }
-#endif /* fetch64_be_unaligned */
+
+#endif                                              /* fetch64_be_unaligned */
 
 static __maybe_unused __always_inline uint64_t tail64_be_aligned(const void *v,
                                                                  size_t tail) {
+
   const uint8_t *const p = (const uint8_t *)v;
 #if T1HA_USE_FAST_ONESHOT_READ && !defined(__SANITIZE_ADDRESS__)
   /* We can perform a 'oneshot' read, which is little bit faster. */
@@ -853,61 +950,66 @@ static __maybe_unused __always_inline uint64_t tail64_be_aligned(const void *v,
   return fetch64_be_aligned(p) >> shift;
 #else
   switch (tail & 7) {
-  default:
-    unreachable();
-/* fall through */
-#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
-  /* For most CPUs this code is better when not byte reordering. */
-  case 1:
-    return p[0];
-  case 2:
-    return fetch16_be_aligned(p);
-  case 3:
-    return (uint32_t)fetch16_be_aligned(p) << 8 | p[2];
-  case 4:
-    return fetch32_be_aligned(p);
-  case 5:
-    return (uint64_t)fetch32_be_aligned(p) << 8 | p[4];
-  case 6:
-    return (uint64_t)fetch32_be_aligned(p) << 16 | fetch16_be_aligned(p + 4);
-  case 7:
-    return (uint64_t)fetch32_be_aligned(p) << 24 |
-           (uint32_t)fetch16_be_aligned(p + 4) << 8 | p[6];
-  case 0:
-    return fetch64_be_aligned(p);
-#else
-  case 1:
-    return p[0];
-  case 2:
-    return p[1] | (uint32_t)p[0] << 8;
-  case 3:
-    return p[2] | (uint32_t)p[1] << 8 | (uint32_t)p[0] << 16;
-  case 4:
-    return p[3] | (uint32_t)p[2] << 8 | (uint32_t)p[1] << 16 |
-           (uint32_t)p[0] << 24;
-  case 5:
-    return p[4] | (uint32_t)p[3] << 8 | (uint32_t)p[2] << 16 |
-           (uint32_t)p[1] << 24 | (uint64_t)p[0] << 32;
-  case 6:
-    return p[5] | (uint32_t)p[4] << 8 | (uint32_t)p[3] << 16 |
-           (uint32_t)p[2] << 24 | (uint64_t)p[1] << 32 | (uint64_t)p[0] << 40;
-  case 7:
-    return p[6] | (uint32_t)p[5] << 8 | (uint32_t)p[4] << 16 |
-           (uint32_t)p[3] << 24 | (uint64_t)p[2] << 32 | (uint64_t)p[1] << 40 |
-           (uint64_t)p[0] << 48;
-  case 0:
-    return p[7] | (uint32_t)p[6] << 8 | (uint32_t)p[5] << 16 |
-           (uint32_t)p[4] << 24 | (uint64_t)p[3] << 32 | (uint64_t)p[2] << 40 |
-           (uint64_t)p[1] << 48 | (uint64_t)p[0] << 56;
-#endif
+
+    default:
+      unreachable();
+  /* fall through */
+  #if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+    /* For most CPUs this code is better when not byte reordering. */
+    case 1:
+      return p[0];
+    case 2:
+      return fetch16_be_aligned(p);
+    case 3:
+      return (uint32_t)fetch16_be_aligned(p) << 8 | p[2];
+    case 4:
+      return fetch32_be_aligned(p);
+    case 5:
+      return (uint64_t)fetch32_be_aligned(p) << 8 | p[4];
+    case 6:
+      return (uint64_t)fetch32_be_aligned(p) << 16 | fetch16_be_aligned(p + 4);
+    case 7:
+      return (uint64_t)fetch32_be_aligned(p) << 24 |
+             (uint32_t)fetch16_be_aligned(p + 4) << 8 | p[6];
+    case 0:
+      return fetch64_be_aligned(p);
+  #else
+    case 1:
+      return p[0];
+    case 2:
+      return p[1] | (uint32_t)p[0] << 8;
+    case 3:
+      return p[2] | (uint32_t)p[1] << 8 | (uint32_t)p[0] << 16;
+    case 4:
+      return p[3] | (uint32_t)p[2] << 8 | (uint32_t)p[1] << 16 |
+             (uint32_t)p[0] << 24;
+    case 5:
+      return p[4] | (uint32_t)p[3] << 8 | (uint32_t)p[2] << 16 |
+             (uint32_t)p[1] << 24 | (uint64_t)p[0] << 32;
+    case 6:
+      return p[5] | (uint32_t)p[4] << 8 | (uint32_t)p[3] << 16 |
+             (uint32_t)p[2] << 24 | (uint64_t)p[1] << 32 | (uint64_t)p[0] << 40;
+    case 7:
+      return p[6] | (uint32_t)p[5] << 8 | (uint32_t)p[4] << 16 |
+             (uint32_t)p[3] << 24 | (uint64_t)p[2] << 32 |
+             (uint64_t)p[1] << 40 | (uint64_t)p[0] << 48;
+    case 0:
+      return p[7] | (uint32_t)p[6] << 8 | (uint32_t)p[5] << 16 |
+             (uint32_t)p[4] << 24 | (uint64_t)p[3] << 32 |
+             (uint64_t)p[2] << 40 | (uint64_t)p[1] << 48 | (uint64_t)p[0] << 56;
+  #endif
+
   }
-#endif /* T1HA_USE_FAST_ONESHOT_READ */
+
+#endif                                        /* T1HA_USE_FAST_ONESHOT_READ */
+
 }
 
 static __maybe_unused __always_inline uint64_t
 tail64_be_unaligned(const void *v, size_t tail) {
+
   const uint8_t *p = (const uint8_t *)v;
-#if defined(can_read_underside) &&                                             \
+#if defined(can_read_underside) && \
     (UINTPTR_MAX > 0xffffFFFFul || ULONG_MAX > 0xffffFFFFul)
   /* On some systems (e.g. x86_64) we can perform a 'oneshot' read, which
    * is little bit faster. Thanks Marcin Żukowski <marcin.zukowski@gmail.com>
@@ -915,139 +1017,167 @@ tail64_be_unaligned(const void *v, size_t tail) {
   const unsigned offset = (8 - tail) & 7;
   const unsigned shift = offset << 3;
   if (likely(can_read_underside(p, 8))) {
+
     p -= offset;
     return fetch64_be_unaligned(p) & ((~UINT64_C(0)) >> shift);
+
   }
+
   return fetch64_be_unaligned(p) >> shift;
 #else
   switch (tail & 7) {
-  default:
-    unreachable();
-/* fall through */
-#if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__EFFICIENT &&           \
-    __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
-  /* For most CPUs this code is better when not needed
-   * copying for alignment or byte reordering. */
-  case 1:
-    return p[0];
-  case 2:
-    return fetch16_be_unaligned(p);
-  case 3:
-    return (uint32_t)fetch16_be_unaligned(p) << 8 | p[2];
-  case 4:
-    return fetch32_be(p);
-  case 5:
-    return (uint64_t)fetch32_be_unaligned(p) << 8 | p[4];
-  case 6:
-    return (uint64_t)fetch32_be_unaligned(p) << 16 |
-           fetch16_be_unaligned(p + 4);
-  case 7:
-    return (uint64_t)fetch32_be_unaligned(p) << 24 |
-           (uint32_t)fetch16_be_unaligned(p + 4) << 8 | p[6];
-  case 0:
-    return fetch64_be_unaligned(p);
-#else
-  /* For most CPUs this code is better than a
-   * copying for alignment and/or byte reordering. */
-  case 1:
-    return p[0];
-  case 2:
-    return p[1] | (uint32_t)p[0] << 8;
-  case 3:
-    return p[2] | (uint32_t)p[1] << 8 | (uint32_t)p[0] << 16;
-  case 4:
-    return p[3] | (uint32_t)p[2] << 8 | (uint32_t)p[1] << 16 |
-           (uint32_t)p[0] << 24;
-  case 5:
-    return p[4] | (uint32_t)p[3] << 8 | (uint32_t)p[2] << 16 |
-           (uint32_t)p[1] << 24 | (uint64_t)p[0] << 32;
-  case 6:
-    return p[5] | (uint32_t)p[4] << 8 | (uint32_t)p[3] << 16 |
-           (uint32_t)p[2] << 24 | (uint64_t)p[1] << 32 | (uint64_t)p[0] << 40;
-  case 7:
-    return p[6] | (uint32_t)p[5] << 8 | (uint32_t)p[4] << 16 |
-           (uint32_t)p[3] << 24 | (uint64_t)p[2] << 32 | (uint64_t)p[1] << 40 |
-           (uint64_t)p[0] << 48;
-  case 0:
-    return p[7] | (uint32_t)p[6] << 8 | (uint32_t)p[5] << 16 |
-           (uint32_t)p[4] << 24 | (uint64_t)p[3] << 32 | (uint64_t)p[2] << 40 |
-           (uint64_t)p[1] << 48 | (uint64_t)p[0] << 56;
-#endif
+
+    default:
+      unreachable();
+  /* fall through */
+  #if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__EFFICIENT && \
+      __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+    /* For most CPUs this code is better when not needed
+     * copying for alignment or byte reordering. */
+    case 1:
+      return p[0];
+    case 2:
+      return fetch16_be_unaligned(p);
+    case 3:
+      return (uint32_t)fetch16_be_unaligned(p) << 8 | p[2];
+    case 4:
+      return fetch32_be(p);
+    case 5:
+      return (uint64_t)fetch32_be_unaligned(p) << 8 | p[4];
+    case 6:
+      return (uint64_t)fetch32_be_unaligned(p) << 16 |
+             fetch16_be_unaligned(p + 4);
+    case 7:
+      return (uint64_t)fetch32_be_unaligned(p) << 24 |
+             (uint32_t)fetch16_be_unaligned(p + 4) << 8 | p[6];
+    case 0:
+      return fetch64_be_unaligned(p);
+  #else
+    /* For most CPUs this code is better than a
+     * copying for alignment and/or byte reordering. */
+    case 1:
+      return p[0];
+    case 2:
+      return p[1] | (uint32_t)p[0] << 8;
+    case 3:
+      return p[2] | (uint32_t)p[1] << 8 | (uint32_t)p[0] << 16;
+    case 4:
+      return p[3] | (uint32_t)p[2] << 8 | (uint32_t)p[1] << 16 |
+             (uint32_t)p[0] << 24;
+    case 5:
+      return p[4] | (uint32_t)p[3] << 8 | (uint32_t)p[2] << 16 |
+             (uint32_t)p[1] << 24 | (uint64_t)p[0] << 32;
+    case 6:
+      return p[5] | (uint32_t)p[4] << 8 | (uint32_t)p[3] << 16 |
+             (uint32_t)p[2] << 24 | (uint64_t)p[1] << 32 | (uint64_t)p[0] << 40;
+    case 7:
+      return p[6] | (uint32_t)p[5] << 8 | (uint32_t)p[4] << 16 |
+             (uint32_t)p[3] << 24 | (uint64_t)p[2] << 32 |
+             (uint64_t)p[1] << 40 | (uint64_t)p[0] << 48;
+    case 0:
+      return p[7] | (uint32_t)p[6] << 8 | (uint32_t)p[5] << 16 |
+             (uint32_t)p[4] << 24 | (uint64_t)p[3] << 32 |
+             (uint64_t)p[2] << 40 | (uint64_t)p[1] << 48 | (uint64_t)p[0] << 56;
+  #endif
+
   }
-#endif /* can_read_underside */
+
+#endif                                                /* can_read_underside */
+
 }
 
 /***************************************************************************/
 
 #ifndef rot64
 static __maybe_unused __always_inline uint64_t rot64(uint64_t v, unsigned s) {
+
   return (v >> s) | (v << (64 - s));
+
 }
-#endif /* rot64 */
+
+#endif                                                             /* rot64 */
 
 #ifndef mul_32x32_64
 static __maybe_unused __always_inline uint64_t mul_32x32_64(uint32_t a,
                                                             uint32_t b) {
+
   return a * (uint64_t)b;
+
 }
-#endif /* mul_32x32_64 */
+
+#endif                                                      /* mul_32x32_64 */
 
 #ifndef add64carry_first
-static __maybe_unused __always_inline unsigned
-add64carry_first(uint64_t base, uint64_t addend, uint64_t *sum) {
-#if __has_builtin(__builtin_addcll)
+static __maybe_unused __always_inline unsigned add64carry_first(uint64_t base,
+                                                                uint64_t addend,
+                                                                uint64_t *sum) {
+
+  #if __has_builtin(__builtin_addcll)
   unsigned long long carryout;
   *sum = __builtin_addcll(base, addend, 0, &carryout);
   return (unsigned)carryout;
-#else
+  #else
   *sum = base + addend;
   return *sum < addend;
-#endif /* __has_builtin(__builtin_addcll) */
+  #endif                                 /* __has_builtin(__builtin_addcll) */
+
 }
-#endif /* add64carry_fist */
+
+#endif                                                   /* add64carry_fist */
 
 #ifndef add64carry_next
-static __maybe_unused __always_inline unsigned
-add64carry_next(unsigned carry, uint64_t base, uint64_t addend, uint64_t *sum) {
-#if __has_builtin(__builtin_addcll)
+static __maybe_unused __always_inline unsigned add64carry_next(unsigned  carry,
+                                                               uint64_t  base,
+                                                               uint64_t  addend,
+                                                               uint64_t *sum) {
+
+  #if __has_builtin(__builtin_addcll)
   unsigned long long carryout;
   *sum = __builtin_addcll(base, addend, carry, &carryout);
   return (unsigned)carryout;
-#else
+  #else
   *sum = base + addend + carry;
   return *sum < addend || (carry && *sum == addend);
-#endif /* __has_builtin(__builtin_addcll) */
+  #endif                                 /* __has_builtin(__builtin_addcll) */
+
 }
-#endif /* add64carry_next */
+
+#endif                                                   /* add64carry_next */
 
 #ifndef add64carry_last
-static __maybe_unused __always_inline void
-add64carry_last(unsigned carry, uint64_t base, uint64_t addend, uint64_t *sum) {
-#if __has_builtin(__builtin_addcll)
+static __maybe_unused __always_inline void add64carry_last(unsigned  carry,
+                                                           uint64_t  base,
+                                                           uint64_t  addend,
+                                                           uint64_t *sum) {
+
+  #if __has_builtin(__builtin_addcll)
   unsigned long long carryout;
   *sum = __builtin_addcll(base, addend, carry, &carryout);
   (void)carryout;
-#else
+  #else
   *sum = base + addend + carry;
-#endif /* __has_builtin(__builtin_addcll) */
+  #endif                                 /* __has_builtin(__builtin_addcll) */
+
 }
-#endif /* add64carry_last */
+
+#endif                                                   /* add64carry_last */
 
 #ifndef mul_64x64_128
-static __maybe_unused __always_inline uint64_t mul_64x64_128(uint64_t a,
-                                                             uint64_t b,
+static __maybe_unused __always_inline uint64_t mul_64x64_128(uint64_t  a,
+                                                             uint64_t  b,
                                                              uint64_t *h) {
-#if (defined(__SIZEOF_INT128__) ||                                             \
-     (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)) &&            \
-    (!defined(__LCC__) || __LCC__ != 124)
+
+  #if (defined(__SIZEOF_INT128__) ||                                  \
+       (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)) && \
+      (!defined(__LCC__) || __LCC__ != 124)
   __uint128_t r = (__uint128_t)a * (__uint128_t)b;
   /* modern GCC could nicely optimize this */
   *h = (uint64_t)(r >> 64);
   return (uint64_t)r;
-#elif defined(mul_64x64_high)
+  #elif defined(mul_64x64_high)
   *h = mul_64x64_high(a, b);
   return a * b;
-#else
+  #else
   /* performs 64x64 to 128 bit multiplication */
   const uint64_t ll = mul_32x32_64((uint32_t)a, (uint32_t)b);
   const uint64_t lh = mul_32x32_64(a >> 32, (uint32_t)b);
@@ -1062,18 +1192,23 @@ static __maybe_unused __always_inline uint64_t mul_64x64_128(uint64_t a,
   add64carry_last(add64carry_first(ll, lh << 32, &l), hh, lh >> 32, h);
   add64carry_last(add64carry_first(l, hl << 32, &l), *h, hl >> 32, h);
   return l;
-#endif
+  #endif
+
 }
-#endif /* mul_64x64_128() */
+
+#endif                                                   /* mul_64x64_128() */
 
 #ifndef mul_64x64_high
 static __maybe_unused __always_inline uint64_t mul_64x64_high(uint64_t a,
                                                               uint64_t b) {
+
   uint64_t h;
   mul_64x64_128(a, b, &h);
   return h;
+
 }
-#endif /* mul_64x64_high */
+
+#endif                                                    /* mul_64x64_high */
 
 /***************************************************************************/
 
@@ -1089,45 +1224,56 @@ static const uint64_t prime_6 = UINT64_C(0xCB5AF53AE3AAAC31);
 /* xor high and low parts of full 128-bit product */
 static __maybe_unused __always_inline uint64_t mux64(uint64_t v,
                                                      uint64_t prime) {
+
   uint64_t l, h;
   l = mul_64x64_128(v, prime, &h);
   return l ^ h;
+
 }
 
 static __maybe_unused __always_inline uint64_t final64(uint64_t a, uint64_t b) {
+
   uint64_t x = (a + rot64(b, 41)) * prime_0;
   uint64_t y = (rot64(a, 23) + b) * prime_6;
   return mux64(x ^ y, prime_5);
+
 }
 
 static __maybe_unused __always_inline void mixup64(uint64_t *__restrict a,
                                                    uint64_t *__restrict b,
                                                    uint64_t v, uint64_t prime) {
+
   uint64_t h;
   *a ^= mul_64x64_128(*b + v, prime, &h);
   *b += h;
+
 }
 
 /***************************************************************************/
 
 typedef union t1ha_uint128 {
-#if defined(__SIZEOF_INT128__) ||                                              \
+
+#if defined(__SIZEOF_INT128__) || \
     (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
   __uint128_t v;
 #endif
   struct {
+
 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
     uint64_t l, h;
 #else
     uint64_t h, l;
 #endif
+
   };
+
 } t1ha_uint128_t;
 
 static __maybe_unused __always_inline t1ha_uint128_t
 not128(const t1ha_uint128_t v) {
+
   t1ha_uint128_t r;
-#if defined(__SIZEOF_INT128__) ||                                              \
+#if defined(__SIZEOF_INT128__) || \
     (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
   r.v = ~v.v;
 #else
@@ -1135,13 +1281,15 @@ not128(const t1ha_uint128_t v) {
   r.h = ~v.h;
 #endif
   return r;
+
 }
 
 static __maybe_unused __always_inline t1ha_uint128_t
 left128(const t1ha_uint128_t v, unsigned s) {
+
   t1ha_uint128_t r;
   assert(s < 128);
-#if defined(__SIZEOF_INT128__) ||                                              \
+#if defined(__SIZEOF_INT128__) || \
     (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
   r.v = v.v << s;
 #else
@@ -1149,13 +1297,15 @@ left128(const t1ha_uint128_t v, unsigned s) {
   r.h = (s < 64) ? (v.h << s) | (s ? v.l >> (64 - s) : 0) : v.l << (s - 64);
 #endif
   return r;
+
 }
 
 static __maybe_unused __always_inline t1ha_uint128_t
 right128(const t1ha_uint128_t v, unsigned s) {
+
   t1ha_uint128_t r;
   assert(s < 128);
-#if defined(__SIZEOF_INT128__) ||                                              \
+#if defined(__SIZEOF_INT128__) || \
     (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
   r.v = v.v >> s;
 #else
@@ -1163,12 +1313,14 @@ right128(const t1ha_uint128_t v, unsigned s) {
   r.h = (s < 64) ? v.h >> s : 0;
 #endif
   return r;
+
 }
 
 static __maybe_unused __always_inline t1ha_uint128_t or128(t1ha_uint128_t x,
                                                            t1ha_uint128_t y) {
+
   t1ha_uint128_t r;
-#if defined(__SIZEOF_INT128__) ||                                              \
+#if defined(__SIZEOF_INT128__) || \
     (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
   r.v = x.v | y.v;
 #else
@@ -1176,12 +1328,14 @@ static __maybe_unused __always_inline t1ha_uint128_t or128(t1ha_uint128_t x,
   r.h = x.h | y.h;
 #endif
   return r;
+
 }
 
 static __maybe_unused __always_inline t1ha_uint128_t xor128(t1ha_uint128_t x,
                                                             t1ha_uint128_t y) {
+
   t1ha_uint128_t r;
-#if defined(__SIZEOF_INT128__) ||                                              \
+#if defined(__SIZEOF_INT128__) || \
     (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
   r.v = x.v ^ y.v;
 #else
@@ -1189,36 +1343,42 @@ static __maybe_unused __always_inline t1ha_uint128_t xor128(t1ha_uint128_t x,
   r.h = x.h ^ y.h;
 #endif
   return r;
+
 }
 
 static __maybe_unused __always_inline t1ha_uint128_t rot128(t1ha_uint128_t v,
-                                                            unsigned s) {
+                                                            unsigned       s) {
+
   s &= 127;
-#if defined(__SIZEOF_INT128__) ||                                              \
+#if defined(__SIZEOF_INT128__) || \
     (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
   v.v = (v.v << (128 - s)) | (v.v >> s);
   return v;
 #else
   return s ? or128(left128(v, 128 - s), right128(v, s)) : v;
 #endif
+
 }
 
 static __maybe_unused __always_inline t1ha_uint128_t add128(t1ha_uint128_t x,
                                                             t1ha_uint128_t y) {
+
   t1ha_uint128_t r;
-#if defined(__SIZEOF_INT128__) ||                                              \
+#if defined(__SIZEOF_INT128__) || \
     (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
   r.v = x.v + y.v;
 #else
   add64carry_last(add64carry_first(x.l, y.l, &r.l), x.h, y.h, &r.h);
 #endif
   return r;
+
 }
 
 static __maybe_unused __always_inline t1ha_uint128_t mul128(t1ha_uint128_t x,
                                                             t1ha_uint128_t y) {
+
   t1ha_uint128_t r;
-#if defined(__SIZEOF_INT128__) ||                                              \
+#if defined(__SIZEOF_INT128__) || \
     (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
   r.v = x.v * y.v;
 #else
@@ -1226,6 +1386,7 @@ static __maybe_unused __always_inline t1ha_uint128_t mul128(t1ha_uint128_t x,
   r.h += x.l * y.h + y.l * x.h;
 #endif
   return r;
+
 }
 
 /***************************************************************************/
@@ -1233,22 +1394,29 @@ static __maybe_unused __always_inline t1ha_uint128_t mul128(t1ha_uint128_t x,
 #if T1HA0_AESNI_AVAILABLE && defined(__ia32__)
 uint64_t t1ha_ia32cpu_features(void);
 
-static __maybe_unused __always_inline bool
-t1ha_ia32_AESNI_avail(uint64_t ia32cpu_features) {
+static __maybe_unused __always_inline bool t1ha_ia32_AESNI_avail(
+    uint64_t ia32cpu_features) {
+
   /* check for AES-NI */
   return (ia32cpu_features & UINT32_C(0x02000000)) != 0;
+
 }
 
-static __maybe_unused __always_inline bool
-t1ha_ia32_AVX_avail(uint64_t ia32cpu_features) {
+static __maybe_unused __always_inline bool t1ha_ia32_AVX_avail(
+    uint64_t ia32cpu_features) {
+
   /* check for any AVX */
   return (ia32cpu_features & UINT32_C(0x1A000000)) == UINT32_C(0x1A000000);
+
 }
 
-static __maybe_unused __always_inline bool
-t1ha_ia32_AVX2_avail(uint64_t ia32cpu_features) {
+static __maybe_unused __always_inline bool t1ha_ia32_AVX2_avail(
+    uint64_t ia32cpu_features) {
+
   /* check for 'Advanced Vector Extensions 2' */
   return ((ia32cpu_features >> 32) & 32) != 0;
+
 }
 
-#endif /* T1HA0_AESNI_AVAILABLE && __ia32__ */
+#endif                                 /* T1HA0_AESNI_AVAILABLE && __ia32__ */
+
diff --git a/include/t1ha_selfcheck.h b/include/t1ha_selfcheck.h
index ff7c589c..65343bfe 100644
--- a/include/t1ha_selfcheck.h
+++ b/include/t1ha_selfcheck.h
@@ -43,8 +43,8 @@
 
 #pragma once
 #if defined(_MSC_VER) && _MSC_VER > 1800
-#pragma warning(disable : 4464) /* relative include path contains '..' */
-#endif                          /* MSVC */
+  #pragma warning(disable : 4464)    /* relative include path contains '..' */
+#endif                                                              /* MSVC */
 #include "t1ha.h"
 
 /***************************************************************************/
@@ -59,18 +59,19 @@ extern const uint64_t t1ha_refval_2atonce[81];
 extern const uint64_t t1ha_refval_2atonce128[81];
 extern const uint64_t t1ha_refval_2stream[81];
 extern const uint64_t t1ha_refval_2stream128[81];
-#endif /* T1HA2_DISABLED */
+#endif                                                    /* T1HA2_DISABLED */
 
 #ifndef T1HA1_DISABLED
 extern const uint64_t t1ha_refval_64le[81];
 extern const uint64_t t1ha_refval_64be[81];
-#endif /* T1HA1_DISABLED */
+#endif                                                    /* T1HA1_DISABLED */
 
 #ifndef T1HA0_DISABLED
 extern const uint64_t t1ha_refval_32le[81];
 extern const uint64_t t1ha_refval_32be[81];
-#if T1HA0_AESNI_AVAILABLE
+  #if T1HA0_AESNI_AVAILABLE
 extern const uint64_t t1ha_refval_ia32aes_a[81];
 extern const uint64_t t1ha_refval_ia32aes_b[81];
-#endif /* T1HA0_AESNI_AVAILABLE */
-#endif /* T1HA0_DISABLED */
+  #endif                                           /* T1HA0_AESNI_AVAILABLE */
+#endif                                                    /* T1HA0_DISABLED */
+
diff --git a/include/xxhash.h b/include/xxhash.h
index d11f0f63..7697d0f2 100644
--- a/include/xxhash.h
+++ b/include/xxhash.h
@@ -36,8 +36,8 @@
 /*!
  * @mainpage xxHash
  *
- * xxHash is an extremely fast non-cryptographic hash algorithm, working at RAM speed
- * limits.
+ * xxHash is an extremely fast non-cryptographic hash algorithm, working at RAM
+ * speed limits.
  *
  * It is proposed in four flavors, in three families:
  * 1. @ref XXH32_family
@@ -54,44 +54,46 @@
  * Benchmarks
  * ---
  * The reference system uses an Intel i7-9700K CPU, and runs Ubuntu x64 20.04.
- * The open source benchmark program is compiled with clang v10.0 using -O3 flag.
- *
- * | Hash Name            | ISA ext | Width | Large Data Speed | Small Data Velocity |
- * | -------------------- | ------- | ----: | ---------------: | ------------------: |
- * | XXH3_64bits()        | @b AVX2 |    64 |        59.4 GB/s |               133.1 |
- * | MeowHash             | AES-NI  |   128 |        58.2 GB/s |                52.5 |
- * | XXH3_128bits()       | @b AVX2 |   128 |        57.9 GB/s |               118.1 |
- * | CLHash               | PCLMUL  |    64 |        37.1 GB/s |                58.1 |
- * | XXH3_64bits()        | @b SSE2 |    64 |        31.5 GB/s |               133.1 |
- * | XXH3_128bits()       | @b SSE2 |   128 |        29.6 GB/s |               118.1 |
- * | RAM sequential read  |         |   N/A |        28.0 GB/s |                 N/A |
- * | ahash                | AES-NI  |    64 |        22.5 GB/s |               107.2 |
- * | City64               |         |    64 |        22.0 GB/s |                76.6 |
- * | T1ha2                |         |    64 |        22.0 GB/s |                99.0 |
- * | City128              |         |   128 |        21.7 GB/s |                57.7 |
- * | FarmHash             | AES-NI  |    64 |        21.3 GB/s |                71.9 |
- * | XXH64()              |         |    64 |        19.4 GB/s |                71.0 |
- * | SpookyHash           |         |    64 |        19.3 GB/s |                53.2 |
- * | Mum                  |         |    64 |        18.0 GB/s |                67.0 |
- * | CRC32C               | SSE4.2  |    32 |        13.0 GB/s |                57.9 |
- * | XXH32()              |         |    32 |         9.7 GB/s |                71.9 |
- * | City32               |         |    32 |         9.1 GB/s |                66.0 |
- * | Blake3*              | @b AVX2 |   256 |         4.4 GB/s |                 8.1 |
- * | Murmur3              |         |    32 |         3.9 GB/s |                56.1 |
- * | SipHash*             |         |    64 |         3.0 GB/s |                43.2 |
- * | Blake3*              | @b SSE2 |   256 |         2.4 GB/s |                 8.1 |
- * | HighwayHash          |         |    64 |         1.4 GB/s |                 6.0 |
- * | FNV64                |         |    64 |         1.2 GB/s |                62.7 |
- * | Blake2*              |         |   256 |         1.1 GB/s |                 5.1 |
- * | SHA1*                |         |   160 |         0.8 GB/s |                 5.6 |
- * | MD5*                 |         |   128 |         0.6 GB/s |                 7.8 |
+ * The open source benchmark program is compiled with clang v10.0 using -O3
+ * flag.
+ *
+ * | Hash Name            | ISA ext | Width | Large Data Speed | Small Data
+ * Velocity | | -------------------- | ------- | ----: | ---------------: |
+ * ------------------: | | XXH3_64bits()        | @b AVX2 |    64 |        59.4
+ * GB/s |               133.1 | | MeowHash             | AES-NI  |   128 | 58.2
+ * GB/s |                52.5 | | XXH3_128bits()       | @b AVX2 |   128 | 57.9
+ * GB/s |               118.1 | | CLHash               | PCLMUL  |    64 | 37.1
+ * GB/s |                58.1 | | XXH3_64bits()        | @b SSE2 |    64 | 31.5
+ * GB/s |               133.1 | | XXH3_128bits()       | @b SSE2 |   128 | 29.6
+ * GB/s |               118.1 | | RAM sequential read  |         |   N/A | 28.0
+ * GB/s |                 N/A | | ahash                | AES-NI  |    64 | 22.5
+ * GB/s |               107.2 | | City64               |         |    64 | 22.0
+ * GB/s |                76.6 | | T1ha2                |         |    64 | 22.0
+ * GB/s |                99.0 | | City128              |         |   128 | 21.7
+ * GB/s |                57.7 | | FarmHash             | AES-NI  |    64 | 21.3
+ * GB/s |                71.9 | | XXH64()              |         |    64 | 19.4
+ * GB/s |                71.0 | | SpookyHash           |         |    64 | 19.3
+ * GB/s |                53.2 | | Mum                  |         |    64 | 18.0
+ * GB/s |                67.0 | | CRC32C               | SSE4.2  |    32 | 13.0
+ * GB/s |                57.9 | | XXH32()              |         |    32 | 9.7
+ * GB/s |                71.9 | | City32               |         |    32 | 9.1
+ * GB/s |                66.0 | | Blake3*              | @b AVX2 |   256 | 4.4
+ * GB/s |                 8.1 | | Murmur3              |         |    32 | 3.9
+ * GB/s |                56.1 | | SipHash*             |         |    64 | 3.0
+ * GB/s |                43.2 | | Blake3*              | @b SSE2 |   256 | 2.4
+ * GB/s |                 8.1 | | HighwayHash          |         |    64 | 1.4
+ * GB/s |                 6.0 | | FNV64                |         |    64 | 1.2
+ * GB/s |                62.7 | | Blake2*              |         |   256 | 1.1
+ * GB/s |                 5.1 | | SHA1*                |         |   160 | 0.8
+ * GB/s |                 5.6 | | MD5*                 |         |   128 | 0.6
+ * GB/s |                 7.8 |
  * @note
- *   - Hashes which require a specific ISA extension are noted. SSE2 is also noted,
- *     even though it is mandatory on x64.
- *   - Hashes with an asterisk are cryptographic. Note that MD5 is non-cryptographic
- *     by modern standards.
- *   - Small data velocity is a rough average of algorithm's efficiency for small
- *     data. For more accurate information, see the wiki.
+ *   - Hashes which require a specific ISA extension are noted. SSE2 is also
+ * noted, even though it is mandatory on x64.
+ *   - Hashes with an asterisk are cryptographic. Note that MD5 is
+ * non-cryptographic by modern standards.
+ *   - Small data velocity is a rough average of algorithm's efficiency for
+ * small data. For more accurate information, see the wiki.
  *   - More benchmarks and strength tests are found on the wiki:
  *         https://github.com/Cyan4973/xxHash/wiki
  *
@@ -106,14 +108,15 @@
  *    - The range from [`input`, `input + length`) is valid, readable memory.
  *      - The only exception is if the `length` is `0`, `input` may be `NULL`.
  *    - For C++, the objects must have the *TriviallyCopyable* property, as the
- *      functions access bytes directly as if it was an array of `unsigned char`.
+ *      functions access bytes directly as if it was an array of `unsigned
+ * char`.
  *
  * @anchor single_shot_example
  * **Single Shot**
  *
- * These functions are stateless functions which hash a contiguous block of memory,
- * immediately returning the result. They are the easiest and usually the fastest
- * option.
+ * These functions are stateless functions which hash a contiguous block of
+ * memory, immediately returning the result. They are the easiest and usually
+ * the fastest option.
  *
  * XXH32(), XXH64(), XXH3_64bits(), XXH3_128bits()
  *
@@ -121,9 +124,10 @@
  *   #include <string.h>
  *   #include "xxhash.h"
  *
- *   // Example for a function which hashes a null terminated string with XXH32().
- *   XXH32_hash_t hash_string(const char* string, XXH32_hash_t seed)
+ *   // Example for a function which hashes a null terminated string with
+ * XXH32(). XXH32_hash_t hash_string(const char* string, XXH32_hash_t seed)
  *   {
+
  *       // NULL pointers are only valid if the length is zero
  *       size_t length = (string == NULL) ? 0 : strlen(string);
  *       return XXH32(string, length, seed);
@@ -143,9 +147,10 @@
  *   #include <stdio.h>
  *   #include <assert.h>
  *   #include "xxhash.h"
- *   // Example for a function which hashes a FILE incrementally with XXH3_64bits().
- *   XXH64_hash_t hashFile(FILE* f)
+ *   // Example for a function which hashes a FILE incrementally with
+ * XXH3_64bits(). XXH64_hash_t hashFile(FILE* f)
  *   {
+
  *       // Allocate a state struct. Do not just use malloc() or new.
  *       XXH3_state_t* state = XXH3_createState();
  *       assert(state != NULL && "Out of memory!");
@@ -155,6 +160,7 @@
  *       size_t count;
  *       // Read the file in chunks
  *       while ((count = fread(buffer, 1, sizeof(buffer), f)) != 0) {
+
  *           // Run update() as many times as necessary to process the data
  *           XXH3_64bits_update(state, buffer, count);
  *       }
@@ -174,7 +180,8 @@
  *
  * Start a new hash by initializing the state with a seed using `XXH*_reset()`.
  *
- * Then, feed the hash state by calling `XXH*_update()` as many times as necessary.
+ * Then, feed the hash state by calling `XXH*_update()` as many times as
+ * necessary.
  *
  * The function returns an error code, with 0 meaning OK, and any other value
  * meaning there is an error.
@@ -195,11 +202,13 @@
  * integers.
  * This the simplest and fastest format for further post-processing.
  *
- * However, this leaves open the question of what is the order on the byte level,
- * since little and big endian conventions will store the same number differently.
+ * However, this leaves open the question of what is the order on the byte
+ * level, since little and big endian conventions will store the same number
+ * differently.
  *
  * The canonical representation settles this issue by mandating big-endian
- * convention, the same convention as human-readable numbers (large digits first).
+ * convention, the same convention as human-readable numbers (large digits
+ * first).
  *
  * When writing hash values to storage, sending them over a network, or printing
  * them, it's highly recommended to use the canonical representation to ensure
@@ -216,13 +225,15 @@
  *   #include <stdio.h>
  *   #include "xxhash.h"
  *
- *   // Example for a function which prints XXH32_hash_t in human readable format
- *   void printXxh32(XXH32_hash_t hash)
+ *   // Example for a function which prints XXH32_hash_t in human readable
+ * format void printXxh32(XXH32_hash_t hash)
  *   {
+
  *       XXH32_canonical_t cano;
  *       XXH32_canonicalFromHash(&cano, hash);
  *       size_t i;
  *       for(i = 0; i < sizeof(cano.digest); ++i) {
+
  *           printf("%02x", cano.digest[i]);
  *       }
  *       printf("\n");
@@ -231,6 +242,7 @@
  *   // Example for a function which converts XXH32_canonical_t to XXH32_hash_t
  *   XXH32_hash_t convertCanonicalToXxh32(XXH32_canonical_t cano)
  *   {
+
  *       XXH32_hash_t hash = XXH32_hashFromCanonical(&cano);
  *       return hash;
  *   }
@@ -241,8 +253,9 @@
  * xxHash prototypes and implementation
  */
 
-#if defined (__cplusplus)
+#if defined(__cplusplus)
 extern "C" {
+
 #endif
 
 /* ****************************
@@ -252,304 +265,328 @@ extern "C" {
  * @defgroup public Public API
  * Contains details on the public xxHash functions.
  * @{
- */
-#ifdef XXH_DOXYGEN
-/*!
- * @brief Gives access to internal state declaration, required for static allocation.
- *
- * Incompatible with dynamic linking, due to risks of ABI changes.
- *
- * Usage:
- * @code{.c}
- *     #define XXH_STATIC_LINKING_ONLY
- *     #include "xxhash.h"
- * @endcode
- */
-#  define XXH_STATIC_LINKING_ONLY
-/* Do not undef XXH_STATIC_LINKING_ONLY for Doxygen */
-
-/*!
- * @brief Gives access to internal definitions.
- *
- * Usage:
- * @code{.c}
- *     #define XXH_STATIC_LINKING_ONLY
- *     #define XXH_IMPLEMENTATION
- *     #include "xxhash.h"
- * @endcode
- */
-#  define XXH_IMPLEMENTATION
-/* Do not undef XXH_IMPLEMENTATION for Doxygen */
 
-/*!
- * @brief Exposes the implementation and marks all functions as `inline`.
- *
- * Use these build macros to inline xxhash into the target unit.
- * Inlining improves performance on small inputs, especially when the length is
- * expressed as a compile-time constant:
- *
- *  https://fastcompression.blogspot.com/2018/03/xxhash-for-small-keys-impressive-power.html
- *
- * It also keeps xxHash symbols private to the unit, so they are not exported.
- *
- * Usage:
- * @code{.c}
- *     #define XXH_INLINE_ALL
- *     #include "xxhash.h"
- * @endcode
- * Do not compile and link xxhash.o as a separate object, as it is not useful.
- */
-#  define XXH_INLINE_ALL
-#  undef XXH_INLINE_ALL
-/*!
- * @brief Exposes the implementation without marking functions as inline.
- */
-#  define XXH_PRIVATE_API
-#  undef XXH_PRIVATE_API
-/*!
- * @brief Emulate a namespace by transparently prefixing all symbols.
- *
- * If you want to include _and expose_ xxHash functions from within your own
- * library, but also want to avoid symbol collisions with other libraries which
- * may also include xxHash, you can use @ref XXH_NAMESPACE to automatically prefix
- * any public symbol from xxhash library with the value of @ref XXH_NAMESPACE
- * (therefore, avoid empty or numeric values).
- *
- * Note that no change is required within the calling program as long as it
- * includes `xxhash.h`: Regular symbol names will be automatically translated
- * by this header.
  */
-#  define XXH_NAMESPACE /* YOUR NAME HERE */
-#  undef XXH_NAMESPACE
+#ifdef XXH_DOXYGEN
+  /*!
+   * @brief Gives access to internal state declaration, required for static
+   * allocation.
+   *
+   * Incompatible with dynamic linking, due to risks of ABI changes.
+   *
+   * Usage:
+   * @code{.c}
+   *     #define XXH_STATIC_LINKING_ONLY
+   *     #include "xxhash.h"
+   * @endcode
+   */
+  #define XXH_STATIC_LINKING_ONLY
+  /* Do not undef XXH_STATIC_LINKING_ONLY for Doxygen */
+
+  /*!
+   * @brief Gives access to internal definitions.
+   *
+   * Usage:
+   * @code{.c}
+   *     #define XXH_STATIC_LINKING_ONLY
+   *     #define XXH_IMPLEMENTATION
+   *     #include "xxhash.h"
+   * @endcode
+   */
+  #define XXH_IMPLEMENTATION
+  /* Do not undef XXH_IMPLEMENTATION for Doxygen */
+
+  /*!
+   * @brief Exposes the implementation and marks all functions as `inline`.
+   *
+   * Use these build macros to inline xxhash into the target unit.
+   * Inlining improves performance on small inputs, especially when the length
+   * is expressed as a compile-time constant:
+   *
+   *  https://fastcompression.blogspot.com/2018/03/xxhash-for-small-keys-impressive-power.html
+   *
+   * It also keeps xxHash symbols private to the unit, so they are not exported.
+   *
+   * Usage:
+   * @code{.c}
+   *     #define XXH_INLINE_ALL
+   *     #include "xxhash.h"
+   * @endcode
+   * Do not compile and link xxhash.o as a separate object, as it is not useful.
+   */
+  #define XXH_INLINE_ALL
+  #undef XXH_INLINE_ALL
+  /*!
+   * @brief Exposes the implementation without marking functions as inline.
+   */
+  #define XXH_PRIVATE_API
+  #undef XXH_PRIVATE_API
+  /*!
+   * @brief Emulate a namespace by transparently prefixing all symbols.
+   *
+   * If you want to include _and expose_ xxHash functions from within your own
+   * library, but also want to avoid symbol collisions with other libraries
+   * which may also include xxHash, you can use @ref XXH_NAMESPACE to
+   * automatically prefix any public symbol from xxhash library with the value
+   * of @ref XXH_NAMESPACE (therefore, avoid empty or numeric values).
+   *
+   * Note that no change is required within the calling program as long as it
+   * includes `xxhash.h`: Regular symbol names will be automatically translated
+   * by this header.
+   */
+  #define XXH_NAMESPACE                                   /* YOUR NAME HERE */
+  #undef XXH_NAMESPACE
 #endif
 
-#if (defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API)) \
-    && !defined(XXH_INLINE_ALL_31684351384)
-   /* this section should be traversed only once */
-#  define XXH_INLINE_ALL_31684351384
-   /* give access to the advanced API, required to compile implementations */
-#  undef XXH_STATIC_LINKING_ONLY   /* avoid macro redef */
-#  define XXH_STATIC_LINKING_ONLY
-   /* make all functions private */
-#  undef XXH_PUBLIC_API
-#  if defined(__GNUC__)
-#    define XXH_PUBLIC_API static __inline __attribute__((unused))
-#  elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
-#    define XXH_PUBLIC_API static inline
-#  elif defined(_MSC_VER)
-#    define XXH_PUBLIC_API static __inline
-#  else
-     /* note: this version may generate warnings for unused static functions */
-#    define XXH_PUBLIC_API static
-#  endif
-
-   /*
-    * This part deals with the special case where a unit wants to inline xxHash,
-    * but "xxhash.h" has previously been included without XXH_INLINE_ALL,
-    * such as part of some previously included *.h header file.
-    * Without further action, the new include would just be ignored,
-    * and functions would effectively _not_ be inlined (silent failure).
-    * The following macros solve this situation by prefixing all inlined names,
-    * avoiding naming collision with previous inclusions.
-    */
-   /* Before that, we unconditionally #undef all symbols,
-    * in case they were already defined with XXH_NAMESPACE.
-    * They will then be redefined for XXH_INLINE_ALL
-    */
-#  undef XXH_versionNumber
-    /* XXH32 */
-#  undef XXH32
-#  undef XXH32_createState
-#  undef XXH32_freeState
-#  undef XXH32_reset
-#  undef XXH32_update
-#  undef XXH32_digest
-#  undef XXH32_copyState
-#  undef XXH32_canonicalFromHash
-#  undef XXH32_hashFromCanonical
-    /* XXH64 */
-#  undef XXH64
-#  undef XXH64_createState
-#  undef XXH64_freeState
-#  undef XXH64_reset
-#  undef XXH64_update
-#  undef XXH64_digest
-#  undef XXH64_copyState
-#  undef XXH64_canonicalFromHash
-#  undef XXH64_hashFromCanonical
-    /* XXH3_64bits */
-#  undef XXH3_64bits
-#  undef XXH3_64bits_withSecret
-#  undef XXH3_64bits_withSeed
-#  undef XXH3_64bits_withSecretandSeed
-#  undef XXH3_createState
-#  undef XXH3_freeState
-#  undef XXH3_copyState
-#  undef XXH3_64bits_reset
-#  undef XXH3_64bits_reset_withSeed
-#  undef XXH3_64bits_reset_withSecret
-#  undef XXH3_64bits_update
-#  undef XXH3_64bits_digest
-#  undef XXH3_generateSecret
-    /* XXH3_128bits */
-#  undef XXH128
-#  undef XXH3_128bits
-#  undef XXH3_128bits_withSeed
-#  undef XXH3_128bits_withSecret
-#  undef XXH3_128bits_reset
-#  undef XXH3_128bits_reset_withSeed
-#  undef XXH3_128bits_reset_withSecret
-#  undef XXH3_128bits_reset_withSecretandSeed
-#  undef XXH3_128bits_update
-#  undef XXH3_128bits_digest
-#  undef XXH128_isEqual
-#  undef XXH128_cmp
-#  undef XXH128_canonicalFromHash
-#  undef XXH128_hashFromCanonical
-    /* Finally, free the namespace itself */
-#  undef XXH_NAMESPACE
-
-    /* employ the namespace for XXH_INLINE_ALL */
-#  define XXH_NAMESPACE XXH_INLINE_
-   /*
-    * Some identifiers (enums, type names) are not symbols,
-    * but they must nonetheless be renamed to avoid redeclaration.
-    * Alternative solution: do not redeclare them.
-    * However, this requires some #ifdefs, and has a more dispersed impact.
-    * Meanwhile, renaming can be achieved in a single place.
-    */
-#  define XXH_IPREF(Id)   XXH_NAMESPACE ## Id
-#  define XXH_OK XXH_IPREF(XXH_OK)
-#  define XXH_ERROR XXH_IPREF(XXH_ERROR)
-#  define XXH_errorcode XXH_IPREF(XXH_errorcode)
-#  define XXH32_canonical_t  XXH_IPREF(XXH32_canonical_t)
-#  define XXH64_canonical_t  XXH_IPREF(XXH64_canonical_t)
-#  define XXH128_canonical_t XXH_IPREF(XXH128_canonical_t)
-#  define XXH32_state_s XXH_IPREF(XXH32_state_s)
-#  define XXH32_state_t XXH_IPREF(XXH32_state_t)
-#  define XXH64_state_s XXH_IPREF(XXH64_state_s)
-#  define XXH64_state_t XXH_IPREF(XXH64_state_t)
-#  define XXH3_state_s  XXH_IPREF(XXH3_state_s)
-#  define XXH3_state_t  XXH_IPREF(XXH3_state_t)
-#  define XXH128_hash_t XXH_IPREF(XXH128_hash_t)
-   /* Ensure the header is parsed again, even if it was previously included */
-#  undef XXHASH_H_5627135585666179
-#  undef XXHASH_H_STATIC_13879238742
-#endif /* XXH_INLINE_ALL || XXH_PRIVATE_API */
-
-/* ****************************************************************
- *  Stable API
- *****************************************************************/
-#ifndef XXHASH_H_5627135585666179
-#define XXHASH_H_5627135585666179 1
-
-/*! @brief Marks a global symbol. */
-#if !defined(XXH_INLINE_ALL) && !defined(XXH_PRIVATE_API)
-#  if defined(WIN32) && defined(_MSC_VER) && (defined(XXH_IMPORT) || defined(XXH_EXPORT))
-#    ifdef XXH_EXPORT
-#      define XXH_PUBLIC_API __declspec(dllexport)
-#    elif XXH_IMPORT
-#      define XXH_PUBLIC_API __declspec(dllimport)
-#    endif
-#  else
-#    define XXH_PUBLIC_API   /* do nothing */
-#  endif
-#endif
+#if (defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API)) && \
+    !defined(XXH_INLINE_ALL_31684351384)
+/* this section should be traversed only once */
+  #define XXH_INLINE_ALL_31684351384
+/* give access to the advanced API, required to compile implementations */
+  #undef XXH_STATIC_LINKING_ONLY                       /* avoid macro redef */
+  #define XXH_STATIC_LINKING_ONLY
+/* make all functions private */
+  #undef XXH_PUBLIC_API
+  #if defined(__GNUC__)
+    #define XXH_PUBLIC_API static __inline __attribute__((unused))
+  #elif defined(__cplusplus) || \
+      (defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+    #define XXH_PUBLIC_API static inline
+  #elif defined(_MSC_VER)
+    #define XXH_PUBLIC_API static __inline
+  #else
+  /* note: this version may generate warnings for unused static functions */
+    #define XXH_PUBLIC_API static
+  #endif
 
-#ifdef XXH_NAMESPACE
-#  define XXH_CAT(A,B) A##B
-#  define XXH_NAME2(A,B) XXH_CAT(A,B)
-#  define XXH_versionNumber XXH_NAME2(XXH_NAMESPACE, XXH_versionNumber)
+/*
+ * This part deals with the special case where a unit wants to inline xxHash,
+ * but "xxhash.h" has previously been included without XXH_INLINE_ALL,
+ * such as part of some previously included *.h header file.
+ * Without further action, the new include would just be ignored,
+ * and functions would effectively _not_ be inlined (silent failure).
+ * The following macros solve this situation by prefixing all inlined names,
+ * avoiding naming collision with previous inclusions.
+ */
+/* Before that, we unconditionally #undef all symbols,
+ * in case they were already defined with XXH_NAMESPACE.
+ * They will then be redefined for XXH_INLINE_ALL
+ */
+  #undef XXH_versionNumber
 /* XXH32 */
-#  define XXH32 XXH_NAME2(XXH_NAMESPACE, XXH32)
-#  define XXH32_createState XXH_NAME2(XXH_NAMESPACE, XXH32_createState)
-#  define XXH32_freeState XXH_NAME2(XXH_NAMESPACE, XXH32_freeState)
-#  define XXH32_reset XXH_NAME2(XXH_NAMESPACE, XXH32_reset)
-#  define XXH32_update XXH_NAME2(XXH_NAMESPACE, XXH32_update)
-#  define XXH32_digest XXH_NAME2(XXH_NAMESPACE, XXH32_digest)
-#  define XXH32_copyState XXH_NAME2(XXH_NAMESPACE, XXH32_copyState)
-#  define XXH32_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH32_canonicalFromHash)
-#  define XXH32_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH32_hashFromCanonical)
+  #undef XXH32
+  #undef XXH32_createState
+  #undef XXH32_freeState
+  #undef XXH32_reset
+  #undef XXH32_update
+  #undef XXH32_digest
+  #undef XXH32_copyState
+  #undef XXH32_canonicalFromHash
+  #undef XXH32_hashFromCanonical
 /* XXH64 */
-#  define XXH64 XXH_NAME2(XXH_NAMESPACE, XXH64)
-#  define XXH64_createState XXH_NAME2(XXH_NAMESPACE, XXH64_createState)
-#  define XXH64_freeState XXH_NAME2(XXH_NAMESPACE, XXH64_freeState)
-#  define XXH64_reset XXH_NAME2(XXH_NAMESPACE, XXH64_reset)
-#  define XXH64_update XXH_NAME2(XXH_NAMESPACE, XXH64_update)
-#  define XXH64_digest XXH_NAME2(XXH_NAMESPACE, XXH64_digest)
-#  define XXH64_copyState XXH_NAME2(XXH_NAMESPACE, XXH64_copyState)
-#  define XXH64_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH64_canonicalFromHash)
-#  define XXH64_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH64_hashFromCanonical)
+  #undef XXH64
+  #undef XXH64_createState
+  #undef XXH64_freeState
+  #undef XXH64_reset
+  #undef XXH64_update
+  #undef XXH64_digest
+  #undef XXH64_copyState
+  #undef XXH64_canonicalFromHash
+  #undef XXH64_hashFromCanonical
 /* XXH3_64bits */
-#  define XXH3_64bits XXH_NAME2(XXH_NAMESPACE, XXH3_64bits)
-#  define XXH3_64bits_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSecret)
-#  define XXH3_64bits_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSeed)
-#  define XXH3_64bits_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSecretandSeed)
-#  define XXH3_createState XXH_NAME2(XXH_NAMESPACE, XXH3_createState)
-#  define XXH3_freeState XXH_NAME2(XXH_NAMESPACE, XXH3_freeState)
-#  define XXH3_copyState XXH_NAME2(XXH_NAMESPACE, XXH3_copyState)
-#  define XXH3_64bits_reset XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset)
-#  define XXH3_64bits_reset_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSeed)
-#  define XXH3_64bits_reset_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSecret)
-#  define XXH3_64bits_reset_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSecretandSeed)
-#  define XXH3_64bits_update XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_update)
-#  define XXH3_64bits_digest XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_digest)
-#  define XXH3_generateSecret XXH_NAME2(XXH_NAMESPACE, XXH3_generateSecret)
-#  define XXH3_generateSecret_fromSeed XXH_NAME2(XXH_NAMESPACE, XXH3_generateSecret_fromSeed)
+  #undef XXH3_64bits
+  #undef XXH3_64bits_withSecret
+  #undef XXH3_64bits_withSeed
+  #undef XXH3_64bits_withSecretandSeed
+  #undef XXH3_createState
+  #undef XXH3_freeState
+  #undef XXH3_copyState
+  #undef XXH3_64bits_reset
+  #undef XXH3_64bits_reset_withSeed
+  #undef XXH3_64bits_reset_withSecret
+  #undef XXH3_64bits_update
+  #undef XXH3_64bits_digest
+  #undef XXH3_generateSecret
 /* XXH3_128bits */
-#  define XXH128 XXH_NAME2(XXH_NAMESPACE, XXH128)
-#  define XXH3_128bits XXH_NAME2(XXH_NAMESPACE, XXH3_128bits)
-#  define XXH3_128bits_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSeed)
-#  define XXH3_128bits_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSecret)
-#  define XXH3_128bits_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSecretandSeed)
-#  define XXH3_128bits_reset XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset)
-#  define XXH3_128bits_reset_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSeed)
-#  define XXH3_128bits_reset_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSecret)
-#  define XXH3_128bits_reset_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSecretandSeed)
-#  define XXH3_128bits_update XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_update)
-#  define XXH3_128bits_digest XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_digest)
-#  define XXH128_isEqual XXH_NAME2(XXH_NAMESPACE, XXH128_isEqual)
-#  define XXH128_cmp     XXH_NAME2(XXH_NAMESPACE, XXH128_cmp)
-#  define XXH128_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH128_canonicalFromHash)
-#  define XXH128_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH128_hashFromCanonical)
-#endif
-
-
-/* *************************************
-*  Compiler specifics
-***************************************/
-
-/* specific declaration modes for Windows */
-#if !defined(XXH_INLINE_ALL) && !defined(XXH_PRIVATE_API)
-#  if defined(WIN32) && defined(_MSC_VER) && (defined(XXH_IMPORT) || defined(XXH_EXPORT))
-#    ifdef XXH_EXPORT
-#      define XXH_PUBLIC_API __declspec(dllexport)
-#    elif XXH_IMPORT
-#      define XXH_PUBLIC_API __declspec(dllimport)
-#    endif
-#  else
-#    define XXH_PUBLIC_API   /* do nothing */
-#  endif
-#endif
-
-#if defined (__GNUC__)
-# define XXH_CONSTF  __attribute__((const))
-# define XXH_PUREF   __attribute__((pure))
-# define XXH_MALLOCF __attribute__((malloc))
-#else
-# define XXH_CONSTF  /* disable */
-# define XXH_PUREF
-# define XXH_MALLOCF
-#endif
+  #undef XXH128
+  #undef XXH3_128bits
+  #undef XXH3_128bits_withSeed
+  #undef XXH3_128bits_withSecret
+  #undef XXH3_128bits_reset
+  #undef XXH3_128bits_reset_withSeed
+  #undef XXH3_128bits_reset_withSecret
+  #undef XXH3_128bits_reset_withSecretandSeed
+  #undef XXH3_128bits_update
+  #undef XXH3_128bits_digest
+  #undef XXH128_isEqual
+  #undef XXH128_cmp
+  #undef XXH128_canonicalFromHash
+  #undef XXH128_hashFromCanonical
+/* Finally, free the namespace itself */
+  #undef XXH_NAMESPACE
+
+/* employ the namespace for XXH_INLINE_ALL */
+  #define XXH_NAMESPACE XXH_INLINE_
+/*
+ * Some identifiers (enums, type names) are not symbols,
+ * but they must nonetheless be renamed to avoid redeclaration.
+ * Alternative solution: do not redeclare them.
+ * However, this requires some #ifdefs, and has a more dispersed impact.
+ * Meanwhile, renaming can be achieved in a single place.
+ */
+  #define XXH_IPREF(Id) XXH_NAMESPACE##Id
+  #define XXH_OK XXH_IPREF(XXH_OK)
+  #define XXH_ERROR XXH_IPREF(XXH_ERROR)
+  #define XXH_errorcode XXH_IPREF(XXH_errorcode)
+  #define XXH32_canonical_t XXH_IPREF(XXH32_canonical_t)
+  #define XXH64_canonical_t XXH_IPREF(XXH64_canonical_t)
+  #define XXH128_canonical_t XXH_IPREF(XXH128_canonical_t)
+  #define XXH32_state_s XXH_IPREF(XXH32_state_s)
+  #define XXH32_state_t XXH_IPREF(XXH32_state_t)
+  #define XXH64_state_s XXH_IPREF(XXH64_state_s)
+  #define XXH64_state_t XXH_IPREF(XXH64_state_t)
+  #define XXH3_state_s XXH_IPREF(XXH3_state_s)
+  #define XXH3_state_t XXH_IPREF(XXH3_state_t)
+  #define XXH128_hash_t XXH_IPREF(XXH128_hash_t)
+/* Ensure the header is parsed again, even if it was previously included */
+  #undef XXHASH_H_5627135585666179
+  #undef XXHASH_H_STATIC_13879238742
+#endif                                 /* XXH_INLINE_ALL || XXH_PRIVATE_API */
 
-/* *************************************
-*  Version
-***************************************/
-#define XXH_VERSION_MAJOR    0
-#define XXH_VERSION_MINOR    8
-#define XXH_VERSION_RELEASE  2
-/*! @brief Version number, encoded as two digits each */
-#define XXH_VERSION_NUMBER  (XXH_VERSION_MAJOR *100*100 + XXH_VERSION_MINOR *100 + XXH_VERSION_RELEASE)
+/* ****************************************************************
+ *  Stable API
+ *****************************************************************/
+#ifndef XXHASH_H_5627135585666179
+  #define XXHASH_H_5627135585666179 1
+
+  /*! @brief Marks a global symbol. */
+  #if !defined(XXH_INLINE_ALL) && !defined(XXH_PRIVATE_API)
+    #if defined(WIN32) && defined(_MSC_VER) && \
+        (defined(XXH_IMPORT) || defined(XXH_EXPORT))
+      #ifdef XXH_EXPORT
+        #define XXH_PUBLIC_API __declspec(dllexport)
+      #elif XXH_IMPORT
+        #define XXH_PUBLIC_API __declspec(dllimport)
+      #endif
+    #else
+      #define XXH_PUBLIC_API                                  /* do nothing */
+    #endif
+  #endif
+
+  #ifdef XXH_NAMESPACE
+    #define XXH_CAT(A, B) A##B
+    #define XXH_NAME2(A, B) XXH_CAT(A, B)
+    #define XXH_versionNumber XXH_NAME2(XXH_NAMESPACE, XXH_versionNumber)
+    /* XXH32 */
+    #define XXH32 XXH_NAME2(XXH_NAMESPACE, XXH32)
+    #define XXH32_createState XXH_NAME2(XXH_NAMESPACE, XXH32_createState)
+    #define XXH32_freeState XXH_NAME2(XXH_NAMESPACE, XXH32_freeState)
+    #define XXH32_reset XXH_NAME2(XXH_NAMESPACE, XXH32_reset)
+    #define XXH32_update XXH_NAME2(XXH_NAMESPACE, XXH32_update)
+    #define XXH32_digest XXH_NAME2(XXH_NAMESPACE, XXH32_digest)
+    #define XXH32_copyState XXH_NAME2(XXH_NAMESPACE, XXH32_copyState)
+    #define XXH32_canonicalFromHash \
+      XXH_NAME2(XXH_NAMESPACE, XXH32_canonicalFromHash)
+    #define XXH32_hashFromCanonical \
+      XXH_NAME2(XXH_NAMESPACE, XXH32_hashFromCanonical)
+    /* XXH64 */
+    #define XXH64 XXH_NAME2(XXH_NAMESPACE, XXH64)
+    #define XXH64_createState XXH_NAME2(XXH_NAMESPACE, XXH64_createState)
+    #define XXH64_freeState XXH_NAME2(XXH_NAMESPACE, XXH64_freeState)
+    #define XXH64_reset XXH_NAME2(XXH_NAMESPACE, XXH64_reset)
+    #define XXH64_update XXH_NAME2(XXH_NAMESPACE, XXH64_update)
+    #define XXH64_digest XXH_NAME2(XXH_NAMESPACE, XXH64_digest)
+    #define XXH64_copyState XXH_NAME2(XXH_NAMESPACE, XXH64_copyState)
+    #define XXH64_canonicalFromHash \
+      XXH_NAME2(XXH_NAMESPACE, XXH64_canonicalFromHash)
+    #define XXH64_hashFromCanonical \
+      XXH_NAME2(XXH_NAMESPACE, XXH64_hashFromCanonical)
+    /* XXH3_64bits */
+    #define XXH3_64bits XXH_NAME2(XXH_NAMESPACE, XXH3_64bits)
+    #define XXH3_64bits_withSecret \
+      XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSecret)
+    #define XXH3_64bits_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSeed)
+    #define XXH3_64bits_withSecretandSeed \
+      XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSecretandSeed)
+    #define XXH3_createState XXH_NAME2(XXH_NAMESPACE, XXH3_createState)
+    #define XXH3_freeState XXH_NAME2(XXH_NAMESPACE, XXH3_freeState)
+    #define XXH3_copyState XXH_NAME2(XXH_NAMESPACE, XXH3_copyState)
+    #define XXH3_64bits_reset XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset)
+    #define XXH3_64bits_reset_withSeed \
+      XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSeed)
+    #define XXH3_64bits_reset_withSecret \
+      XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSecret)
+    #define XXH3_64bits_reset_withSecretandSeed \
+      XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSecretandSeed)
+    #define XXH3_64bits_update XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_update)
+    #define XXH3_64bits_digest XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_digest)
+    #define XXH3_generateSecret XXH_NAME2(XXH_NAMESPACE, XXH3_generateSecret)
+    #define XXH3_generateSecret_fromSeed \
+      XXH_NAME2(XXH_NAMESPACE, XXH3_generateSecret_fromSeed)
+    /* XXH3_128bits */
+    #define XXH128 XXH_NAME2(XXH_NAMESPACE, XXH128)
+    #define XXH3_128bits XXH_NAME2(XXH_NAMESPACE, XXH3_128bits)
+    #define XXH3_128bits_withSeed \
+      XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSeed)
+    #define XXH3_128bits_withSecret \
+      XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSecret)
+    #define XXH3_128bits_withSecretandSeed \
+      XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSecretandSeed)
+    #define XXH3_128bits_reset XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset)
+    #define XXH3_128bits_reset_withSeed \
+      XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSeed)
+    #define XXH3_128bits_reset_withSecret \
+      XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSecret)
+    #define XXH3_128bits_reset_withSecretandSeed \
+      XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSecretandSeed)
+    #define XXH3_128bits_update XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_update)
+    #define XXH3_128bits_digest XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_digest)
+    #define XXH128_isEqual XXH_NAME2(XXH_NAMESPACE, XXH128_isEqual)
+    #define XXH128_cmp XXH_NAME2(XXH_NAMESPACE, XXH128_cmp)
+    #define XXH128_canonicalFromHash \
+      XXH_NAME2(XXH_NAMESPACE, XXH128_canonicalFromHash)
+    #define XXH128_hashFromCanonical \
+      XXH_NAME2(XXH_NAMESPACE, XXH128_hashFromCanonical)
+  #endif
+
+  /* *************************************
+   *  Compiler specifics
+   ***************************************/
+
+  /* specific declaration modes for Windows */
+  #if !defined(XXH_INLINE_ALL) && !defined(XXH_PRIVATE_API)
+    #if defined(WIN32) && defined(_MSC_VER) && \
+        (defined(XXH_IMPORT) || defined(XXH_EXPORT))
+      #ifdef XXH_EXPORT
+        #define XXH_PUBLIC_API __declspec(dllexport)
+      #elif XXH_IMPORT
+        #define XXH_PUBLIC_API __declspec(dllimport)
+      #endif
+    #else
+      #define XXH_PUBLIC_API                                  /* do nothing */
+    #endif
+  #endif
+
+  #if defined(__GNUC__)
+    #define XXH_CONSTF __attribute__((const))
+    #define XXH_PUREF __attribute__((pure))
+    #define XXH_MALLOCF __attribute__((malloc))
+  #else
+    #define XXH_CONSTF                                           /* disable */
+    #define XXH_PUREF
+    #define XXH_MALLOCF
+  #endif
+
+  /* *************************************
+   *  Version
+   ***************************************/
+  #define XXH_VERSION_MAJOR 0
+  #define XXH_VERSION_MINOR 8
+  #define XXH_VERSION_RELEASE 2
+  /*! @brief Version number, encoded as two digits each */
+  #define XXH_VERSION_NUMBER                                   \
+    (XXH_VERSION_MAJOR * 100 * 100 + XXH_VERSION_MINOR * 100 + \
+     XXH_VERSION_RELEASE)
 
 /*!
  * @brief Obtains the xxHash version.
@@ -559,26 +596,26 @@ extern "C" {
  *
  * @return @ref XXH_VERSION_NUMBER of the invoked library.
  */
-XXH_PUBLIC_API XXH_CONSTF unsigned XXH_versionNumber (void);
-
+XXH_PUBLIC_API XXH_CONSTF unsigned XXH_versionNumber(void);
 
-/* ****************************
-*  Common basic types
-******************************/
-#include <stddef.h>   /* size_t */
+  /* ****************************
+   *  Common basic types
+   ******************************/
+  #include <stddef.h>                                             /* size_t */
 /*!
  * @brief Exit code for the streaming API.
  */
 typedef enum {
-    XXH_OK = 0, /*!< OK */
-    XXH_ERROR   /*!< Error */
-} XXH_errorcode;
 
+  XXH_OK = 0,                                                       /*!< OK */
+  XXH_ERROR                                                      /*!< Error */
 
-/*-**********************************************************************
-*  32-bit hash
-************************************************************************/
-#if defined(XXH_DOXYGEN) /* Don't show <stdint.h> include */
+} XXH_errorcode;
+
+  /*-**********************************************************************
+   *  32-bit hash
+   ************************************************************************/
+  #if defined(XXH_DOXYGEN)                 /* Don't show <stdint.h> include */
 /*!
  * @brief An unsigned 32-bit integer.
  *
@@ -586,22 +623,22 @@ typedef enum {
  */
 typedef uint32_t XXH32_hash_t;
 
-#elif !defined (__VMS) \
-  && (defined (__cplusplus) \
-  || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
-#   include <stdint.h>
-    typedef uint32_t XXH32_hash_t;
-
-#else
-#   include <limits.h>
-#   if UINT_MAX == 0xFFFFFFFFUL
-      typedef unsigned int XXH32_hash_t;
-#   elif ULONG_MAX == 0xFFFFFFFFUL
-      typedef unsigned long XXH32_hash_t;
-#   else
-#     error "unsupported platform: need a 32-bit type"
-#   endif
-#endif
+  #elif !defined(__VMS) &&     \
+      (defined(__cplusplus) || \
+       (defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */))
+    #include <stdint.h>
+typedef uint32_t XXH32_hash_t;
+
+  #else
+    #include <limits.h>
+    #if UINT_MAX == 0xFFFFFFFFUL
+typedef unsigned int XXH32_hash_t;
+    #elif ULONG_MAX == 0xFFFFFFFFUL
+typedef unsigned long XXH32_hash_t;
+    #else
+      #error "unsupported platform: need a 32-bit type"
+    #endif
+  #endif
 
 /*!
  * @}
@@ -618,12 +655,14 @@ typedef uint32_t XXH32_hash_t;
  * @see @ref XXH64_family, @ref XXH3_family : Other xxHash families
  * @see @ref XXH32_impl for implementation details
  * @{
+
  */
 
 /*!
  * @brief Calculates the 32-bit hash of @p input using xxHash32.
  *
- * @param input The block of data to be hashed, at least @p length bytes in size.
+ * @param input The block of data to be hashed, at least @p length bytes in
+ * size.
  * @param length The length of @p input, in bytes.
  * @param seed The 32-bit seed to alter the hash's output predictably.
  *
@@ -636,9 +675,10 @@ typedef uint32_t XXH32_hash_t;
  *
  * @see @ref single_shot_example "Single Shot Example" for an example.
  */
-XXH_PUBLIC_API XXH_PUREF XXH32_hash_t XXH32 (const void* input, size_t length, XXH32_hash_t seed);
+XXH_PUBLIC_API XXH_PUREF XXH32_hash_t XXH32(const void *input, size_t length,
+                                            XXH32_hash_t seed);
 
-#ifndef XXH_NO_STREAM
+  #ifndef XXH_NO_STREAM
 /*!
  * @typedef struct XXH32_state_s XXH32_state_t
  * @brief The opaque state struct for the XXH32 streaming API.
@@ -658,11 +698,12 @@ typedef struct XXH32_state_s XXH32_state_t;
  *
  * @see @ref streaming_example "Streaming Example"
  */
-XXH_PUBLIC_API XXH_MALLOCF XXH32_state_t* XXH32_createState(void);
+XXH_PUBLIC_API XXH_MALLOCF XXH32_state_t *XXH32_createState(void);
 /*!
  * @brief Frees an @ref XXH32_state_t.
  *
- * @param statePtr A pointer to an @ref XXH32_state_t allocated with @ref XXH32_createState().
+ * @param statePtr A pointer to an @ref XXH32_state_t allocated with @ref
+ * XXH32_createState().
  *
  * @return @ref XXH_OK.
  *
@@ -671,7 +712,7 @@ XXH_PUBLIC_API XXH_MALLOCF XXH32_state_t* XXH32_createState(void);
  * @see @ref streaming_example "Streaming Example"
  *
  */
-XXH_PUBLIC_API XXH_errorcode  XXH32_freeState(XXH32_state_t* statePtr);
+XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t *statePtr);
 /*!
  * @brief Copies one @ref XXH32_state_t to another.
  *
@@ -680,7 +721,8 @@ XXH_PUBLIC_API XXH_errorcode  XXH32_freeState(XXH32_state_t* statePtr);
  * @pre
  *   @p dst_state and @p src_state must not be `NULL` and must not overlap.
  */
-XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dst_state, const XXH32_state_t* src_state);
+XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t       *dst_state,
+                                    const XXH32_state_t *src_state);
 
 /*!
  * @brief Resets an @ref XXH32_state_t to begin a new hash.
@@ -694,17 +736,20 @@ XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dst_state, const XXH32_state_
  * @return @ref XXH_OK on success.
  * @return @ref XXH_ERROR on failure.
  *
- * @note This function resets and seeds a state. Call it before @ref XXH32_update().
+ * @note This function resets and seeds a state. Call it before @ref
+ * XXH32_update().
  *
  * @see @ref streaming_example "Streaming Example"
  */
-XXH_PUBLIC_API XXH_errorcode XXH32_reset  (XXH32_state_t* statePtr, XXH32_hash_t seed);
+XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t *statePtr,
+                                         XXH32_hash_t   seed);
 
 /*!
  * @brief Consumes a block of @p input to an @ref XXH32_state_t.
  *
  * @param statePtr The state struct to update.
- * @param input The block of data to be hashed, at least @p length bytes in size.
+ * @param input The block of data to be hashed, at least @p length bytes in
+ * size.
  * @param length The length of @p input, in bytes.
  *
  * @pre
@@ -721,7 +766,8 @@ XXH_PUBLIC_API XXH_errorcode XXH32_reset  (XXH32_state_t* statePtr, XXH32_hash_t
  *
  * @see @ref streaming_example "Streaming Example"
  */
-XXH_PUBLIC_API XXH_errorcode XXH32_update (XXH32_state_t* statePtr, const void* input, size_t length);
+XXH_PUBLIC_API XXH_errorcode XXH32_update(XXH32_state_t *statePtr,
+                                          const void *input, size_t length);
 
 /*!
  * @brief Returns the calculated hash value from an @ref XXH32_state_t.
@@ -739,8 +785,9 @@ XXH_PUBLIC_API XXH_errorcode XXH32_update (XXH32_state_t* statePtr, const void*
  *
  * @see @ref streaming_example "Streaming Example"
  */
-XXH_PUBLIC_API XXH_PUREF XXH32_hash_t XXH32_digest (const XXH32_state_t* statePtr);
-#endif /* !XXH_NO_STREAM */
+XXH_PUBLIC_API XXH_PUREF XXH32_hash_t
+XXH32_digest(const XXH32_state_t *statePtr);
+  #endif                                                  /* !XXH_NO_STREAM */
 
 /*******   Canonical representation   *******/
 
@@ -748,7 +795,9 @@ XXH_PUBLIC_API XXH_PUREF XXH32_hash_t XXH32_digest (const XXH32_state_t* statePt
  * @brief Canonical (big endian) representation of @ref XXH32_hash_t.
  */
 typedef struct {
-    unsigned char digest[4]; /*!< Hash bytes, big endian */
+
+  unsigned char digest[4];                      /*!< Hash bytes, big endian */
+
 } XXH32_canonical_t;
 
 /*!
@@ -762,7 +811,8 @@ typedef struct {
  *
  * @see @ref canonical_representation_example "Canonical Representation Example"
  */
-XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash);
+XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t *dst,
+                                            XXH32_hash_t       hash);
 
 /*!
  * @brief Converts an @ref XXH32_canonical_t to a native @ref XXH32_hash_t.
@@ -776,105 +826,106 @@ XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t
  *
  * @see @ref canonical_representation_example "Canonical Representation Example"
  */
-XXH_PUBLIC_API XXH_PUREF XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src);
-
-
-/*! @cond Doxygen ignores this part */
-#ifdef __has_attribute
-# define XXH_HAS_ATTRIBUTE(x) __has_attribute(x)
-#else
-# define XXH_HAS_ATTRIBUTE(x) 0
-#endif
-/*! @endcond */
-
-/*! @cond Doxygen ignores this part */
-/*
- * C23 __STDC_VERSION__ number hasn't been specified yet. For now
- * leave as `201711L` (C17 + 1).
- * TODO: Update to correct value when its been specified.
- */
-#define XXH_C23_VN 201711L
-/*! @endcond */
-
-/*! @cond Doxygen ignores this part */
-/* C-language Attributes are added in C23. */
-#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= XXH_C23_VN) && defined(__has_c_attribute)
-# define XXH_HAS_C_ATTRIBUTE(x) __has_c_attribute(x)
-#else
-# define XXH_HAS_C_ATTRIBUTE(x) 0
-#endif
-/*! @endcond */
-
-/*! @cond Doxygen ignores this part */
-#if defined(__cplusplus) && defined(__has_cpp_attribute)
-# define XXH_HAS_CPP_ATTRIBUTE(x) __has_cpp_attribute(x)
-#else
-# define XXH_HAS_CPP_ATTRIBUTE(x) 0
-#endif
-/*! @endcond */
-
-/*! @cond Doxygen ignores this part */
-/*
- * Define XXH_FALLTHROUGH macro for annotating switch case with the 'fallthrough' attribute
- * introduced in CPP17 and C23.
- * CPP17 : https://en.cppreference.com/w/cpp/language/attributes/fallthrough
- * C23   : https://en.cppreference.com/w/c/language/attributes/fallthrough
- */
-#if XXH_HAS_C_ATTRIBUTE(fallthrough) || XXH_HAS_CPP_ATTRIBUTE(fallthrough)
-# define XXH_FALLTHROUGH [[fallthrough]]
-#elif XXH_HAS_ATTRIBUTE(__fallthrough__)
-# define XXH_FALLTHROUGH __attribute__ ((__fallthrough__))
-#else
-# define XXH_FALLTHROUGH /* fallthrough */
-#endif
-/*! @endcond */
-
-/*! @cond Doxygen ignores this part */
-/*
- * Define XXH_NOESCAPE for annotated pointers in public API.
- * https://clang.llvm.org/docs/AttributeReference.html#noescape
- * As of writing this, only supported by clang.
- */
-#if XXH_HAS_ATTRIBUTE(noescape)
-# define XXH_NOESCAPE __attribute__((noescape))
-#else
-# define XXH_NOESCAPE
-#endif
+XXH_PUBLIC_API XXH_PUREF XXH32_hash_t
+XXH32_hashFromCanonical(const XXH32_canonical_t *src);
+
+  /*! @cond Doxygen ignores this part */
+  #ifdef __has_attribute
+    #define XXH_HAS_ATTRIBUTE(x) __has_attribute(x)
+  #else
+    #define XXH_HAS_ATTRIBUTE(x) 0
+  #endif
+  /*! @endcond */
+
+  /*! @cond Doxygen ignores this part */
+  /*
+   * C23 __STDC_VERSION__ number hasn't been specified yet. For now
+   * leave as `201711L` (C17 + 1).
+   * TODO: Update to correct value when its been specified.
+   */
+  #define XXH_C23_VN 201711L
+  /*! @endcond */
+
+  /*! @cond Doxygen ignores this part */
+  /* C-language Attributes are added in C23. */
+  #if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= XXH_C23_VN) && \
+      defined(__has_c_attribute)
+    #define XXH_HAS_C_ATTRIBUTE(x) __has_c_attribute(x)
+  #else
+    #define XXH_HAS_C_ATTRIBUTE(x) 0
+  #endif
+  /*! @endcond */
+
+  /*! @cond Doxygen ignores this part */
+  #if defined(__cplusplus) && defined(__has_cpp_attribute)
+    #define XXH_HAS_CPP_ATTRIBUTE(x) __has_cpp_attribute(x)
+  #else
+    #define XXH_HAS_CPP_ATTRIBUTE(x) 0
+  #endif
+  /*! @endcond */
+
+  /*! @cond Doxygen ignores this part */
+  /*
+   * Define XXH_FALLTHROUGH macro for annotating switch case with the
+   * 'fallthrough' attribute introduced in CPP17 and C23. CPP17 :
+   * https://en.cppreference.com/w/cpp/language/attributes/fallthrough C23   :
+   * https://en.cppreference.com/w/c/language/attributes/fallthrough
+   */
+  #if XXH_HAS_C_ATTRIBUTE(fallthrough) || XXH_HAS_CPP_ATTRIBUTE(fallthrough)
+    #define XXH_FALLTHROUGH [[fallthrough]]
+  #elif XXH_HAS_ATTRIBUTE(__fallthrough__)
+    #define XXH_FALLTHROUGH __attribute__((__fallthrough__))
+  #else
+    #define XXH_FALLTHROUGH                                  /* fallthrough */
+  #endif
+  /*! @endcond */
+
+  /*! @cond Doxygen ignores this part */
+  /*
+   * Define XXH_NOESCAPE for annotated pointers in public API.
+   * https://clang.llvm.org/docs/AttributeReference.html#noescape
+   * As of writing this, only supported by clang.
+   */
+  #if XXH_HAS_ATTRIBUTE(noescape)
+    #define XXH_NOESCAPE __attribute__((noescape))
+  #else
+    #define XXH_NOESCAPE
+  #endif
 /*! @endcond */
 
-
 /*!
  * @}
  * @ingroup public
  * @{
+
  */
 
-#ifndef XXH_NO_LONG_LONG
-/*-**********************************************************************
-*  64-bit hash
-************************************************************************/
-#if defined(XXH_DOXYGEN) /* don't include <stdint.h> */
+  #ifndef XXH_NO_LONG_LONG
+    /*-**********************************************************************
+     *  64-bit hash
+     ************************************************************************/
+    #if defined(XXH_DOXYGEN)                    /* don't include <stdint.h> */
 /*!
  * @brief An unsigned 64-bit integer.
  *
  * Not necessarily defined to `uint64_t` but functionally equivalent.
  */
 typedef uint64_t XXH64_hash_t;
-#elif !defined (__VMS) \
-  && (defined (__cplusplus) \
-  || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
-#  include <stdint.h>
-   typedef uint64_t XXH64_hash_t;
-#else
-#  include <limits.h>
-#  if defined(__LP64__) && ULONG_MAX == 0xFFFFFFFFFFFFFFFFULL
-     /* LP64 ABI says uint64_t is unsigned long */
-     typedef unsigned long XXH64_hash_t;
-#  else
-     /* the following type must have a width of 64-bit */
-     typedef unsigned long long XXH64_hash_t;
-#  endif
-#endif
+    #elif !defined(__VMS) &&                                   \
+        (defined(__cplusplus) || (defined(__STDC_VERSION__) && \
+                                  (__STDC_VERSION__ >= 199901L) /* C99 */))
+      #include <stdint.h>
+typedef uint64_t XXH64_hash_t;
+    #else
+      #include <limits.h>
+      #if defined(__LP64__) && ULONG_MAX == 0xFFFFFFFFFFFFFFFFULL
+/* LP64 ABI says uint64_t is unsigned long */
+typedef unsigned long XXH64_hash_t;
+      #else
+/* the following type must have a width of 64-bit */
+typedef unsigned long long XXH64_hash_t;
+      #endif
+    #endif
 
 /*!
  * @}
@@ -882,6 +933,7 @@ typedef uint64_t XXH64_hash_t;
  * @defgroup XXH64_family XXH64 family
  * @ingroup public
  * @{
+
  * Contains functions used in the classic 64-bit xxHash algorithm.
  *
  * @note
@@ -893,7 +945,8 @@ typedef uint64_t XXH64_hash_t;
 /*!
  * @brief Calculates the 64-bit hash of @p input using xxHash64.
  *
- * @param input The block of data to be hashed, at least @p length bytes in size.
+ * @param input The block of data to be hashed, at least @p length bytes in
+ * size.
  * @param length The length of @p input, in bytes.
  * @param seed The 64-bit seed to alter the hash's output predictably.
  *
@@ -906,17 +959,18 @@ typedef uint64_t XXH64_hash_t;
  *
  * @see @ref single_shot_example "Single Shot Example" for an example.
  */
-XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH64(XXH_NOESCAPE const void* input, size_t length, XXH64_hash_t seed);
+XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH64(XXH_NOESCAPE const void *input,
+                                            size_t length, XXH64_hash_t seed);
 
-/*******   Streaming   *******/
-#ifndef XXH_NO_STREAM
+    /*******   Streaming   *******/
+    #ifndef XXH_NO_STREAM
 /*!
  * @brief The opaque state struct for the XXH64 streaming API.
  *
  * @see XXH64_state_s for details.
  * @see @ref streaming_example "Streaming Example"
  */
-typedef struct XXH64_state_s XXH64_state_t;   /* incomplete type */
+typedef struct XXH64_state_s XXH64_state_t;              /* incomplete type */
 
 /*!
  * @brief Allocates an @ref XXH64_state_t.
@@ -928,12 +982,13 @@ typedef struct XXH64_state_s XXH64_state_t;   /* incomplete type */
  *
  * @see @ref streaming_example "Streaming Example"
  */
-XXH_PUBLIC_API XXH_MALLOCF XXH64_state_t* XXH64_createState(void);
+XXH_PUBLIC_API XXH_MALLOCF XXH64_state_t *XXH64_createState(void);
 
 /*!
  * @brief Frees an @ref XXH64_state_t.
  *
- * @param statePtr A pointer to an @ref XXH64_state_t allocated with @ref XXH64_createState().
+ * @param statePtr A pointer to an @ref XXH64_state_t allocated with @ref
+ * XXH64_createState().
  *
  * @return @ref XXH_OK.
  *
@@ -941,7 +996,7 @@ XXH_PUBLIC_API XXH_MALLOCF XXH64_state_t* XXH64_createState(void);
  *
  * @see @ref streaming_example "Streaming Example"
  */
-XXH_PUBLIC_API XXH_errorcode  XXH64_freeState(XXH64_state_t* statePtr);
+XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t *statePtr);
 
 /*!
  * @brief Copies one @ref XXH64_state_t to another.
@@ -951,7 +1006,8 @@ XXH_PUBLIC_API XXH_errorcode  XXH64_freeState(XXH64_state_t* statePtr);
  * @pre
  *   @p dst_state and @p src_state must not be `NULL` and must not overlap.
  */
-XXH_PUBLIC_API void XXH64_copyState(XXH_NOESCAPE XXH64_state_t* dst_state, const XXH64_state_t* src_state);
+XXH_PUBLIC_API void XXH64_copyState(XXH_NOESCAPE XXH64_state_t *dst_state,
+                                    const XXH64_state_t        *src_state);
 
 /*!
  * @brief Resets an @ref XXH64_state_t to begin a new hash.
@@ -965,17 +1021,20 @@ XXH_PUBLIC_API void XXH64_copyState(XXH_NOESCAPE XXH64_state_t* dst_state, const
  * @return @ref XXH_OK on success.
  * @return @ref XXH_ERROR on failure.
  *
- * @note This function resets and seeds a state. Call it before @ref XXH64_update().
+ * @note This function resets and seeds a state. Call it before @ref
+ * XXH64_update().
  *
  * @see @ref streaming_example "Streaming Example"
  */
-XXH_PUBLIC_API XXH_errorcode XXH64_reset  (XXH_NOESCAPE XXH64_state_t* statePtr, XXH64_hash_t seed);
+XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH_NOESCAPE XXH64_state_t *statePtr,
+                                         XXH64_hash_t                seed);
 
 /*!
  * @brief Consumes a block of @p input to an @ref XXH64_state_t.
  *
  * @param statePtr The state struct to update.
- * @param input The block of data to be hashed, at least @p length bytes in size.
+ * @param input The block of data to be hashed, at least @p length bytes in
+ * size.
  * @param length The length of @p input, in bytes.
  *
  * @pre
@@ -992,7 +1051,9 @@ XXH_PUBLIC_API XXH_errorcode XXH64_reset  (XXH_NOESCAPE XXH64_state_t* statePtr,
  *
  * @see @ref streaming_example "Streaming Example"
  */
-XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH_NOESCAPE XXH64_state_t* statePtr, XXH_NOESCAPE const void* input, size_t length);
+XXH_PUBLIC_API XXH_errorcode XXH64_update(XXH_NOESCAPE XXH64_state_t *statePtr,
+                                          XXH_NOESCAPE const void    *input,
+                                          size_t                      length);
 
 /*!
  * @brief Returns the calculated hash value from an @ref XXH64_state_t.
@@ -1010,14 +1071,19 @@ XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH_NOESCAPE XXH64_state_t* statePtr,
  *
  * @see @ref streaming_example "Streaming Example"
  */
-XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH64_digest (XXH_NOESCAPE const XXH64_state_t* statePtr);
-#endif /* !XXH_NO_STREAM */
+XXH_PUBLIC_API XXH_PUREF XXH64_hash_t
+XXH64_digest(XXH_NOESCAPE const XXH64_state_t *statePtr);
+    #endif                                                /* !XXH_NO_STREAM */
 /*******   Canonical representation   *******/
 
 /*!
  * @brief Canonical (big endian) representation of @ref XXH64_hash_t.
  */
-typedef struct { unsigned char digest[sizeof(XXH64_hash_t)]; } XXH64_canonical_t;
+typedef struct {
+
+  unsigned char digest[sizeof(XXH64_hash_t)];
+
+} XXH64_canonical_t;
 
 /*!
  * @brief Converts an @ref XXH64_hash_t to a big endian @ref XXH64_canonical_t.
@@ -1030,7 +1096,8 @@ typedef struct { unsigned char digest[sizeof(XXH64_hash_t)]; } XXH64_canonical_t
  *
  * @see @ref canonical_representation_example "Canonical Representation Example"
  */
-XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH_NOESCAPE XXH64_canonical_t* dst, XXH64_hash_t hash);
+XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH_NOESCAPE XXH64_canonical_t *dst,
+                                            XXH64_hash_t hash);
 
 /*!
  * @brief Converts an @ref XXH64_canonical_t to a native @ref XXH64_hash_t.
@@ -1044,9 +1111,10 @@ XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH_NOESCAPE XXH64_canonical_t* dst,
  *
  * @see @ref canonical_representation_example "Canonical Representation Example"
  */
-XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH64_hashFromCanonical(XXH_NOESCAPE const XXH64_canonical_t* src);
+XXH_PUBLIC_API XXH_PUREF XXH64_hash_t
+XXH64_hashFromCanonical(XXH_NOESCAPE const XXH64_canonical_t *src);
 
-#ifndef XXH_NO_XXH3
+    #ifndef XXH_NO_XXH3
 
 /*!
  * @}
@@ -1054,6 +1122,7 @@ XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH64_hashFromCanonical(XXH_NOESCAPE const
  * @defgroup XXH3_family XXH3 family
  * @ingroup public
  * @{
+
  *
  * XXH3 is a more recent hash algorithm featuring:
  *  - Improved speed for both small and large inputs
@@ -1085,8 +1154,9 @@ XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH64_hashFromCanonical(XXH_NOESCAPE const
  *   - POWER8 VSX
  *   - s390x ZVector
  * This can be controlled via the @ref XXH_VECTOR macro, but it automatically
- * selects the best version according to predefined macros. For the x86 family, an
- * automatic runtime dispatcher is included separately in @ref xxh_x86dispatch.c.
+ * selects the best version according to predefined macros. For the x86 family,
+ * an automatic runtime dispatcher is included separately in @ref
+ * xxh_x86dispatch.c.
  *
  * XXH3 implementation is portable:
  * it has a generic C90 formulation that can be compiled on any platform,
@@ -1103,13 +1173,14 @@ XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH64_hashFromCanonical(XXH_NOESCAPE const
  * The API supports one-shot hashing, streaming mode, and custom secrets.
  */
 /*-**********************************************************************
-*  XXH3 64-bit variant
-************************************************************************/
+ *  XXH3 64-bit variant
+ ************************************************************************/
 
 /*!
  * @brief Calculates 64-bit unseeded variant of XXH3 hash of @p input.
  *
- * @param input  The block of data to be hashed, at least @p length bytes in size.
+ * @param input  The block of data to be hashed, at least @p length bytes in
+ * size.
  * @param length The length of @p input, in bytes.
  *
  * @pre
@@ -1120,20 +1191,22 @@ XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH64_hashFromCanonical(XXH_NOESCAPE const
  * @return The calculated 64-bit XXH3 hash value.
  *
  * @note
- *   This is equivalent to @ref XXH3_64bits_withSeed() with a seed of `0`, however
- *   it may have slightly better performance due to constant propagation of the
- *   defaults.
+ *   This is equivalent to @ref XXH3_64bits_withSeed() with a seed of `0`,
+ * however it may have slightly better performance due to constant propagation
+ * of the defaults.
  *
  * @see
  *    XXH3_64bits_withSeed(), XXH3_64bits_withSecret(): other seeding variants
  * @see @ref single_shot_example "Single Shot Example" for an example.
  */
-XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits(XXH_NOESCAPE const void* input, size_t length);
+XXH_PUBLIC_API XXH_PUREF XXH64_hash_t
+XXH3_64bits(XXH_NOESCAPE const void *input, size_t length);
 
 /*!
  * @brief Calculates 64-bit seeded variant of XXH3 hash of @p input.
  *
- * @param input  The block of data to be hashed, at least @p length bytes in size.
+ * @param input  The block of data to be hashed, at least @p length bytes in
+ * size.
  * @param length The length of @p input, in bytes.
  * @param seed   The 64-bit seed to alter the hash result predictably.
  *
@@ -1154,21 +1227,23 @@ XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits(XXH_NOESCAPE const void* input
  *
  * @see @ref single_shot_example "Single Shot Example" for an example.
  */
-XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits_withSeed(XXH_NOESCAPE const void* input, size_t length, XXH64_hash_t seed);
+XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits_withSeed(
+    XXH_NOESCAPE const void *input, size_t length, XXH64_hash_t seed);
 
-/*!
- * The bare minimum size for a custom secret.
- *
- * @see
- *  XXH3_64bits_withSecret(), XXH3_64bits_reset_withSecret(),
- *  XXH3_128bits_withSecret(), XXH3_128bits_reset_withSecret().
- */
-#define XXH3_SECRET_SIZE_MIN 136
+      /*!
+       * The bare minimum size for a custom secret.
+       *
+       * @see
+       *  XXH3_64bits_withSecret(), XXH3_64bits_reset_withSecret(),
+       *  XXH3_128bits_withSecret(), XXH3_128bits_reset_withSecret().
+       */
+      #define XXH3_SECRET_SIZE_MIN 136
 
 /*!
  * @brief Calculates 64-bit variant of XXH3 with a custom "secret".
  *
- * @param data       The block of data to be hashed, at least @p len bytes in size.
+ * @param data       The block of data to be hashed, at least @p len bytes in
+ * size.
  * @param len        The length of @p data, in bytes.
  * @param secret     The secret data.
  * @param secretSize The length of @p secret, in bytes.
@@ -1180,28 +1255,29 @@ XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits_withSeed(XXH_NOESCAPE const vo
  *   readable, contiguous memory. However, if @p length is `0`, @p data may be
  *   `NULL`. In C++, this also must be *TriviallyCopyable*.
  *
- * It's possible to provide any blob of bytes as a "secret" to generate the hash.
- * This makes it more difficult for an external actor to prepare an intentional collision.
- * The main condition is that @p secretSize *must* be large enough (>= @ref XXH3_SECRET_SIZE_MIN).
- * However, the quality of the secret impacts the dispersion of the hash algorithm.
- * Therefore, the secret _must_ look like a bunch of random bytes.
- * Avoid "trivial" or structured data such as repeated sequences or a text document.
- * Whenever in doubt about the "randomness" of the blob of bytes,
- * consider employing @ref XXH3_generateSecret() instead (see below).
- * It will generate a proper high entropy secret derived from the blob of bytes.
- * Another advantage of using XXH3_generateSecret() is that
- * it guarantees that all bits within the initial blob of bytes
- * will impact every bit of the output.
- * This is not necessarily the case when using the blob of bytes directly
- * because, when hashing _small_ inputs, only a portion of the secret is employed.
+ * It's possible to provide any blob of bytes as a "secret" to generate the
+ * hash. This makes it more difficult for an external actor to prepare an
+ * intentional collision. The main condition is that @p secretSize *must* be
+ * large enough (>= @ref XXH3_SECRET_SIZE_MIN). However, the quality of the
+ * secret impacts the dispersion of the hash algorithm. Therefore, the secret
+ * _must_ look like a bunch of random bytes. Avoid "trivial" or structured data
+ * such as repeated sequences or a text document. Whenever in doubt about the
+ * "randomness" of the blob of bytes, consider employing @ref
+ * XXH3_generateSecret() instead (see below). It will generate a proper high
+ * entropy secret derived from the blob of bytes. Another advantage of using
+ * XXH3_generateSecret() is that it guarantees that all bits within the initial
+ * blob of bytes will impact every bit of the output. This is not necessarily
+ * the case when using the blob of bytes directly because, when hashing _small_
+ * inputs, only a portion of the secret is employed.
  *
  * @see @ref single_shot_example "Single Shot Example" for an example.
  */
-XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits_withSecret(XXH_NOESCAPE const void* data, size_t len, XXH_NOESCAPE const void* secret, size_t secretSize);
-
+XXH_PUBLIC_API XXH_PUREF XXH64_hash_t
+XXH3_64bits_withSecret(XXH_NOESCAPE const void *data, size_t len,
+                       XXH_NOESCAPE const void *secret, size_t secretSize);
 
-/*******   Streaming   *******/
-#ifndef XXH_NO_STREAM
+      /*******   Streaming   *******/
+      #ifndef XXH_NO_STREAM
 /*
  * Streaming requires state maintenance.
  * This operation costs memory and CPU.
@@ -1215,9 +1291,9 @@ XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits_withSecret(XXH_NOESCAPE const
  * @see XXH3_state_s for details.
  * @see @ref streaming_example "Streaming Example"
  */
-typedef struct XXH3_state_s XXH3_state_t;
-XXH_PUBLIC_API XXH_MALLOCF XXH3_state_t* XXH3_createState(void);
-XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t* statePtr);
+typedef struct XXH3_state_s              XXH3_state_t;
+XXH_PUBLIC_API XXH_MALLOCF XXH3_state_t *XXH3_createState(void);
+XXH_PUBLIC_API XXH_errorcode             XXH3_freeState(XXH3_state_t *statePtr);
 
 /*!
  * @brief Copies one @ref XXH3_state_t to another.
@@ -1227,7 +1303,8 @@ XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t* statePtr);
  * @pre
  *   @p dst_state and @p src_state must not be `NULL` and must not overlap.
  */
-XXH_PUBLIC_API void XXH3_copyState(XXH_NOESCAPE XXH3_state_t* dst_state, XXH_NOESCAPE const XXH3_state_t* src_state);
+XXH_PUBLIC_API void XXH3_copyState(XXH_NOESCAPE XXH3_state_t       *dst_state,
+                                   XXH_NOESCAPE const XXH3_state_t *src_state);
 
 /*!
  * @brief Resets an @ref XXH3_state_t to begin a new hash.
@@ -1241,14 +1318,16 @@ XXH_PUBLIC_API void XXH3_copyState(XXH_NOESCAPE XXH3_state_t* dst_state, XXH_NOE
  * @return @ref XXH_ERROR on failure.
  *
  * @note
- *   - This function resets `statePtr` and generate a secret with default parameters.
+ *   - This function resets `statePtr` and generate a secret with default
+ * parameters.
  *   - Call this function before @ref XXH3_64bits_update().
  *   - Digest will be equivalent to `XXH3_64bits()`.
  *
  * @see @ref streaming_example "Streaming Example"
  *
  */
-XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset(XXH_NOESCAPE XXH3_state_t* statePtr);
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_reset(XXH_NOESCAPE XXH3_state_t *statePtr);
 
 /*!
  * @brief Resets an @ref XXH3_state_t with 64-bit seed to begin a new hash.
@@ -1270,7 +1349,8 @@ XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset(XXH_NOESCAPE XXH3_state_t* stateP
  * @see @ref streaming_example "Streaming Example"
  *
  */
-XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH64_hash_t seed);
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSeed(
+    XXH_NOESCAPE XXH3_state_t *statePtr, XXH64_hash_t seed);
 
 /*!
  * @brief Resets an @ref XXH3_state_t with secret data to begin a new hash.
@@ -1296,13 +1376,16 @@ XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSeed(XXH_NOESCAPE XXH3_state_
  *
  * @see @ref streaming_example "Streaming Example"
  */
-XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSecret(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize);
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSecret(
+    XXH_NOESCAPE XXH3_state_t *statePtr, XXH_NOESCAPE const void *secret,
+    size_t secretSize);
 
 /*!
  * @brief Consumes a block of @p input to an @ref XXH3_state_t.
  *
  * @param statePtr The state struct to update.
- * @param input The block of data to be hashed, at least @p length bytes in size.
+ * @param input The block of data to be hashed, at least @p length bytes in
+ * size.
  * @param length The length of @p input, in bytes.
  *
  * @pre
@@ -1319,10 +1402,13 @@ XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSecret(XXH_NOESCAPE XXH3_stat
  *
  * @see @ref streaming_example "Streaming Example"
  */
-XXH_PUBLIC_API XXH_errorcode XXH3_64bits_update (XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* input, size_t length);
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_update(XXH_NOESCAPE XXH3_state_t *statePtr,
+                   XXH_NOESCAPE const void *input, size_t length);
 
 /*!
- * @brief Returns the calculated XXH3 64-bit hash value from an @ref XXH3_state_t.
+ * @brief Returns the calculated XXH3 64-bit hash value from an @ref
+ * XXH3_state_t.
  *
  * @param statePtr The state struct to calculate the hash from.
  *
@@ -1332,21 +1418,21 @@ XXH_PUBLIC_API XXH_errorcode XXH3_64bits_update (XXH_NOESCAPE XXH3_state_t* stat
  * @return The calculated XXH3 64-bit hash value from that state.
  *
  * @note
- *   Calling XXH3_64bits_digest() will not affect @p statePtr, so you can update,
- *   digest, and update again.
+ *   Calling XXH3_64bits_digest() will not affect @p statePtr, so you can
+ * update, digest, and update again.
  *
  * @see @ref streaming_example "Streaming Example"
  */
-XXH_PUBLIC_API XXH_PUREF XXH64_hash_t  XXH3_64bits_digest (XXH_NOESCAPE const XXH3_state_t* statePtr);
-#endif /* !XXH_NO_STREAM */
+XXH_PUBLIC_API XXH_PUREF XXH64_hash_t
+XXH3_64bits_digest(XXH_NOESCAPE const XXH3_state_t *statePtr);
+      #endif                                              /* !XXH_NO_STREAM */
 
 /* note : canonical representation of XXH3 is the same as XXH64
  * since they both produce XXH64_hash_t values */
 
-
 /*-**********************************************************************
-*  XXH3 128-bit variant
-************************************************************************/
+ *  XXH3 128-bit variant
+ ************************************************************************/
 
 /*!
  * @brief The return value from 128-bit hashes.
@@ -1355,8 +1441,10 @@ XXH_PUBLIC_API XXH_PUREF XXH64_hash_t  XXH3_64bits_digest (XXH_NOESCAPE const XX
  * endianness.
  */
 typedef struct {
-    XXH64_hash_t low64;   /*!< `value & 0xFFFFFFFFFFFFFFFF` */
-    XXH64_hash_t high64;  /*!< `value >> 64` */
+
+  XXH64_hash_t low64;                     /*!< `value & 0xFFFFFFFFFFFFFFFF` */
+  XXH64_hash_t high64;                                   /*!< `value >> 64` */
+
 } XXH128_hash_t;
 
 /*!
@@ -1370,14 +1458,16 @@ typedef struct {
  * The 128-bit variant of XXH3 has more strength, but it has a bit of overhead
  * for shorter inputs.
  *
- * This is equivalent to @ref XXH3_128bits_withSeed() with a seed of `0`, however
- * it may have slightly better performance due to constant propagation of the
- * defaults.
+ * This is equivalent to @ref XXH3_128bits_withSeed() with a seed of `0`,
+ * however it may have slightly better performance due to constant propagation
+ * of the defaults.
  *
- * @see XXH3_128bits_withSeed(), XXH3_128bits_withSecret(): other seeding variants
+ * @see XXH3_128bits_withSeed(), XXH3_128bits_withSecret(): other seeding
+ * variants
  * @see @ref single_shot_example "Single Shot Example" for an example.
  */
-XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits(XXH_NOESCAPE const void* data, size_t len);
+XXH_PUBLIC_API XXH_PUREF XXH128_hash_t
+XXH3_128bits(XXH_NOESCAPE const void *data, size_t len);
 /*! @brief Calculates 128-bit seeded variant of XXH3 hash of @p data.
  *
  * @param data The block of data to be hashed, at least @p length bytes in size.
@@ -1397,38 +1487,42 @@ XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits(XXH_NOESCAPE const void* dat
  * @see XXH3_128bits(), XXH3_128bits_withSecret(): other seeding variants
  * @see @ref single_shot_example "Single Shot Example" for an example.
  */
-XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits_withSeed(XXH_NOESCAPE const void* data, size_t len, XXH64_hash_t seed);
+XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits_withSeed(
+    XXH_NOESCAPE const void *data, size_t len, XXH64_hash_t seed);
 /*!
  * @brief Calculates 128-bit variant of XXH3 with a custom "secret".
  *
- * @param data       The block of data to be hashed, at least @p len bytes in size.
+ * @param data       The block of data to be hashed, at least @p len bytes in
+ * size.
  * @param len        The length of @p data, in bytes.
  * @param secret     The secret data.
  * @param secretSize The length of @p secret, in bytes.
  *
  * @return The calculated 128-bit variant of XXH3 value.
  *
- * It's possible to provide any blob of bytes as a "secret" to generate the hash.
- * This makes it more difficult for an external actor to prepare an intentional collision.
- * The main condition is that @p secretSize *must* be large enough (>= @ref XXH3_SECRET_SIZE_MIN).
- * However, the quality of the secret impacts the dispersion of the hash algorithm.
- * Therefore, the secret _must_ look like a bunch of random bytes.
- * Avoid "trivial" or structured data such as repeated sequences or a text document.
- * Whenever in doubt about the "randomness" of the blob of bytes,
- * consider employing @ref XXH3_generateSecret() instead (see below).
- * It will generate a proper high entropy secret derived from the blob of bytes.
- * Another advantage of using XXH3_generateSecret() is that
- * it guarantees that all bits within the initial blob of bytes
- * will impact every bit of the output.
- * This is not necessarily the case when using the blob of bytes directly
- * because, when hashing _small_ inputs, only a portion of the secret is employed.
+ * It's possible to provide any blob of bytes as a "secret" to generate the
+ * hash. This makes it more difficult for an external actor to prepare an
+ * intentional collision. The main condition is that @p secretSize *must* be
+ * large enough (>= @ref XXH3_SECRET_SIZE_MIN). However, the quality of the
+ * secret impacts the dispersion of the hash algorithm. Therefore, the secret
+ * _must_ look like a bunch of random bytes. Avoid "trivial" or structured data
+ * such as repeated sequences or a text document. Whenever in doubt about the
+ * "randomness" of the blob of bytes, consider employing @ref
+ * XXH3_generateSecret() instead (see below). It will generate a proper high
+ * entropy secret derived from the blob of bytes. Another advantage of using
+ * XXH3_generateSecret() is that it guarantees that all bits within the initial
+ * blob of bytes will impact every bit of the output. This is not necessarily
+ * the case when using the blob of bytes directly because, when hashing _small_
+ * inputs, only a portion of the secret is employed.
  *
  * @see @ref single_shot_example "Single Shot Example" for an example.
  */
-XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits_withSecret(XXH_NOESCAPE const void* data, size_t len, XXH_NOESCAPE const void* secret, size_t secretSize);
+XXH_PUBLIC_API XXH_PUREF XXH128_hash_t
+XXH3_128bits_withSecret(XXH_NOESCAPE const void *data, size_t len,
+                        XXH_NOESCAPE const void *secret, size_t secretSize);
 
-/*******   Streaming   *******/
-#ifndef XXH_NO_STREAM
+      /*******   Streaming   *******/
+      #ifndef XXH_NO_STREAM
 /*
  * Streaming requires state maintenance.
  * This operation costs memory and CPU.
@@ -1438,7 +1532,8 @@ XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits_withSecret(XXH_NOESCAPE cons
  * XXH3_128bits uses the same XXH3_state_t as XXH3_64bits().
  * Use already declared XXH3_createState() and XXH3_freeState().
  *
- * All reset and streaming functions have same meaning as their 64-bit counterpart.
+ * All reset and streaming functions have same meaning as their 64-bit
+ * counterpart.
  */
 
 /*!
@@ -1453,13 +1548,15 @@ XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits_withSecret(XXH_NOESCAPE cons
  * @return @ref XXH_ERROR on failure.
  *
  * @note
- *   - This function resets `statePtr` and generate a secret with default parameters.
+ *   - This function resets `statePtr` and generate a secret with default
+ * parameters.
  *   - Call it before @ref XXH3_128bits_update().
  *   - Digest will be equivalent to `XXH3_128bits()`.
  *
  * @see @ref streaming_example "Streaming Example"
  */
-XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset(XXH_NOESCAPE XXH3_state_t* statePtr);
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_reset(XXH_NOESCAPE XXH3_state_t *statePtr);
 
 /*!
  * @brief Resets an @ref XXH3_state_t with 64-bit seed to begin a new hash.
@@ -1480,7 +1577,8 @@ XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset(XXH_NOESCAPE XXH3_state_t* state
  *
  * @see @ref streaming_example "Streaming Example"
  */
-XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH64_hash_t seed);
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSeed(
+    XXH_NOESCAPE XXH3_state_t *statePtr, XXH64_hash_t seed);
 /*!
  * @brief Resets an @ref XXH3_state_t with secret data to begin a new hash.
  *
@@ -1503,7 +1601,9 @@ XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSeed(XXH_NOESCAPE XXH3_state
  *
  * @see @ref streaming_example "Streaming Example"
  */
-XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSecret(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize);
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSecret(
+    XXH_NOESCAPE XXH3_state_t *statePtr, XXH_NOESCAPE const void *secret,
+    size_t secretSize);
 
 /*!
  * @brief Consumes a block of @p input to an @ref XXH3_state_t.
@@ -1511,7 +1611,8 @@ XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSecret(XXH_NOESCAPE XXH3_sta
  * Call this to incrementally consume blocks of data.
  *
  * @param statePtr The state struct to update.
- * @param input The block of data to be hashed, at least @p length bytes in size.
+ * @param input The block of data to be hashed, at least @p length bytes in
+ * size.
  * @param length The length of @p input, in bytes.
  *
  * @pre
@@ -1526,10 +1627,13 @@ XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSecret(XXH_NOESCAPE XXH3_sta
  *   `NULL`. In C++, this also must be *TriviallyCopyable*.
  *
  */
-XXH_PUBLIC_API XXH_errorcode XXH3_128bits_update (XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* input, size_t length);
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_update(XXH_NOESCAPE XXH3_state_t *statePtr,
+                    XXH_NOESCAPE const void *input, size_t length);
 
 /*!
- * @brief Returns the calculated XXH3 128-bit hash value from an @ref XXH3_state_t.
+ * @brief Returns the calculated XXH3 128-bit hash value from an @ref
+ * XXH3_state_t.
  *
  * @param statePtr The state struct to calculate the hash from.
  *
@@ -1539,16 +1643,18 @@ XXH_PUBLIC_API XXH_errorcode XXH3_128bits_update (XXH_NOESCAPE XXH3_state_t* sta
  * @return The calculated XXH3 128-bit hash value from that state.
  *
  * @note
- *   Calling XXH3_128bits_digest() will not affect @p statePtr, so you can update,
- *   digest, and update again.
+ *   Calling XXH3_128bits_digest() will not affect @p statePtr, so you can
+ * update, digest, and update again.
  *
  */
-XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits_digest (XXH_NOESCAPE const XXH3_state_t* statePtr);
-#endif /* !XXH_NO_STREAM */
+XXH_PUBLIC_API XXH_PUREF XXH128_hash_t
+XXH3_128bits_digest(XXH_NOESCAPE const XXH3_state_t *statePtr);
+      #endif                                              /* !XXH_NO_STREAM */
 
 /* Following helper functions make it possible to compare XXH128_hast_t values.
- * Since XXH128_hash_t is a structure, this capability is not offered by the language.
- * Note: For better performance, these functions can be inlined using XXH_INLINE_ALL */
+ * Since XXH128_hash_t is a structure, this capability is not offered by the
+ * language. Note: For better performance, these functions can be inlined using
+ * XXH_INLINE_ALL */
 
 /*!
  * @brief Check equality of two XXH128_hash_t values
@@ -1573,15 +1679,19 @@ XXH_PUBLIC_API XXH_PUREF int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2);
  * @return =0 if @p h128_1 == @p h128_2
  * @return <0 if @p h128_1  < @p h128_2
  */
-XXH_PUBLIC_API XXH_PUREF int XXH128_cmp(XXH_NOESCAPE const void* h128_1, XXH_NOESCAPE const void* h128_2);
-
+XXH_PUBLIC_API XXH_PUREF int XXH128_cmp(XXH_NOESCAPE const void *h128_1,
+                                        XXH_NOESCAPE const void *h128_2);
 
 /*******   Canonical representation   *******/
-typedef struct { unsigned char digest[sizeof(XXH128_hash_t)]; } XXH128_canonical_t;
+typedef struct {
 
+  unsigned char digest[sizeof(XXH128_hash_t)];
+
+} XXH128_canonical_t;
 
 /*!
- * @brief Converts an @ref XXH128_hash_t to a big endian @ref XXH128_canonical_t.
+ * @brief Converts an @ref XXH128_hash_t to a big endian @ref
+ * XXH128_canonical_t.
  *
  * @param dst  The @ref XXH128_canonical_t pointer to be stored to.
  * @param hash The @ref XXH128_hash_t to be converted.
@@ -1590,7 +1700,8 @@ typedef struct { unsigned char digest[sizeof(XXH128_hash_t)]; } XXH128_canonical
  *   @p dst must not be `NULL`.
  * @see @ref canonical_representation_example "Canonical Representation Example"
  */
-XXH_PUBLIC_API void XXH128_canonicalFromHash(XXH_NOESCAPE XXH128_canonical_t* dst, XXH128_hash_t hash);
+XXH_PUBLIC_API void XXH128_canonicalFromHash(
+    XXH_NOESCAPE XXH128_canonical_t *dst, XXH128_hash_t hash);
 
 /*!
  * @brief Converts an @ref XXH128_canonical_t to a native @ref XXH128_hash_t.
@@ -1603,28 +1714,27 @@ XXH_PUBLIC_API void XXH128_canonicalFromHash(XXH_NOESCAPE XXH128_canonical_t* ds
  * @return The converted hash.
  * @see @ref canonical_representation_example "Canonical Representation Example"
  */
-XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH128_hashFromCanonical(XXH_NOESCAPE const XXH128_canonical_t* src);
-
+XXH_PUBLIC_API XXH_PUREF XXH128_hash_t
+XXH128_hashFromCanonical(XXH_NOESCAPE const XXH128_canonical_t *src);
 
-#endif  /* !XXH_NO_XXH3 */
-#endif  /* XXH_NO_LONG_LONG */
+    #endif                                                  /* !XXH_NO_XXH3 */
+  #endif                                                /* XXH_NO_LONG_LONG */
 
 /*!
  * @}
  */
-#endif /* XXHASH_H_5627135585666179 */
-
-
+#endif                                         /* XXHASH_H_5627135585666179 */
 
 #if defined(XXH_STATIC_LINKING_ONLY) && !defined(XXHASH_H_STATIC_13879238742)
-#define XXHASH_H_STATIC_13879238742
+  #define XXHASH_H_STATIC_13879238742
 /* ****************************************************************************
  * This section contains declarations which are not guaranteed to remain stable.
  * They may change in future versions, becoming incompatible with a different
  * version of the library.
  * These declarations should only be used with static linking.
  * Never use them in association with dynamic linking!
- ***************************************************************************** */
+ *****************************************************************************
+*/
 
 /*
  * These definitions are only present to allow static allocation
@@ -1645,16 +1755,19 @@ XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH128_hashFromCanonical(XXH_NOESCAPE con
  * @see XXH64_state_s, XXH3_state_s
  */
 struct XXH32_state_s {
-   XXH32_hash_t total_len_32; /*!< Total length hashed, modulo 2^32 */
-   XXH32_hash_t large_len;    /*!< Whether the hash is >= 16 (handles @ref total_len_32 overflow) */
-   XXH32_hash_t v[4];         /*!< Accumulator lanes */
-   XXH32_hash_t mem32[4];     /*!< Internal buffer for partial reads. Treated as unsigned char[16]. */
-   XXH32_hash_t memsize;      /*!< Amount of data in @ref mem32 */
-   XXH32_hash_t reserved;     /*!< Reserved field. Do not read nor write to it. */
-};   /* typedef'd to XXH32_state_t */
 
+  XXH32_hash_t total_len_32;          /*!< Total length hashed, modulo 2^32 */
+  XXH32_hash_t large_len;    /*!< Whether the hash is >= 16 (handles @ref
+                                total_len_32 overflow) */
+  XXH32_hash_t v[4];                                 /*!< Accumulator lanes */
+  XXH32_hash_t mem32[4];     /*!< Internal buffer for partial reads. Treated as
+                                unsigned char[16]. */
+  XXH32_hash_t memsize;                   /*!< Amount of data in @ref mem32 */
+  XXH32_hash_t reserved;  /*!< Reserved field. Do not read nor write to it. */
+
+};                                            /* typedef'd to XXH32_state_t */
 
-#ifndef XXH_NO_LONG_LONG  /* defined when there is no 64-bit support */
+  #ifndef XXH_NO_LONG_LONG       /* defined when there is no 64-bit support */
 
 /*!
  * @internal
@@ -1669,57 +1782,64 @@ struct XXH32_state_s {
  * @see XXH32_state_s, XXH3_state_s
  */
 struct XXH64_state_s {
-   XXH64_hash_t total_len;    /*!< Total length hashed. This is always 64-bit. */
-   XXH64_hash_t v[4];         /*!< Accumulator lanes */
-   XXH64_hash_t mem64[4];     /*!< Internal buffer for partial reads. Treated as unsigned char[32]. */
-   XXH32_hash_t memsize;      /*!< Amount of data in @ref mem64 */
-   XXH32_hash_t reserved32;   /*!< Reserved field, needed for padding anyways*/
-   XXH64_hash_t reserved64;   /*!< Reserved field. Do not read or write to it. */
-};   /* typedef'd to XXH64_state_t */
-
-#ifndef XXH_NO_XXH3
-
-#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L) /* >= C11 */
-#  include <stdalign.h>
-#  define XXH_ALIGN(n)      alignas(n)
-#elif defined(__cplusplus) && (__cplusplus >= 201103L) /* >= C++11 */
-/* In C++ alignas() is a keyword */
-#  define XXH_ALIGN(n)      alignas(n)
-#elif defined(__GNUC__)
-#  define XXH_ALIGN(n)      __attribute__ ((aligned(n)))
-#elif defined(_MSC_VER)
-#  define XXH_ALIGN(n)      __declspec(align(n))
-#else
-#  define XXH_ALIGN(n)   /* disabled */
-#endif
-
-/* Old GCC versions only accept the attribute after the type in structures. */
-#if !(defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L))   /* C11+ */ \
-    && ! (defined(__cplusplus) && (__cplusplus >= 201103L)) /* >= C++11 */ \
-    && defined(__GNUC__)
-#   define XXH_ALIGN_MEMBER(align, type) type XXH_ALIGN(align)
-#else
-#   define XXH_ALIGN_MEMBER(align, type) XXH_ALIGN(align) type
-#endif
-
-/*!
- * @brief The size of the internal XXH3 buffer.
- *
- * This is the optimal update size for incremental hashing.
- *
- * @see XXH3_64b_update(), XXH3_128b_update().
- */
-#define XXH3_INTERNALBUFFER_SIZE 256
 
-/*!
- * @internal
- * @brief Default size of the secret buffer (and @ref XXH3_kSecret).
- *
- * This is the size used in @ref XXH3_kSecret and the seeded functions.
- *
- * Not to be confused with @ref XXH3_SECRET_SIZE_MIN.
- */
-#define XXH3_SECRET_DEFAULT_SIZE 192
+  XXH64_hash_t total_len;  /*!< Total length hashed. This is always 64-bit. */
+  XXH64_hash_t v[4];                                 /*!< Accumulator lanes */
+  XXH64_hash_t mem64[4];   /*!< Internal buffer for partial reads. Treated as
+                              unsigned char[32]. */
+  XXH32_hash_t memsize;                   /*!< Amount of data in @ref mem64 */
+  XXH32_hash_t reserved32;   /*!< Reserved field, needed for padding anyways*/
+  XXH64_hash_t reserved64; /*!< Reserved field. Do not read or write to it. */
+
+};                                            /* typedef'd to XXH64_state_t */
+
+    #ifndef XXH_NO_XXH3
+
+      #if defined(__STDC_VERSION__) && \
+          (__STDC_VERSION__ >= 201112L)                           /* >= C11 */
+        #include <stdalign.h>
+        #define XXH_ALIGN(n) alignas(n)
+      #elif defined(__cplusplus) && (__cplusplus >= 201103L)    /* >= C++11 */
+      /* In C++ alignas() is a keyword */
+        #define XXH_ALIGN(n) alignas(n)
+      #elif defined(__GNUC__)
+        #define XXH_ALIGN(n) __attribute__((aligned(n)))
+      #elif defined(_MSC_VER)
+        #define XXH_ALIGN(n) __declspec(align(n))
+      #else
+        #define XXH_ALIGN(n)                                    /* disabled */
+      #endif
+
+      /* Old GCC versions only accept the attribute after the type in
+       * structures. */
+      #if !(defined(__STDC_VERSION__) &&                                     \
+            (__STDC_VERSION__ >= 201112L)) /* C11+ */                        \
+          &&                                                                 \
+          !(defined(__cplusplus) && (__cplusplus >= 201103L)) /* >= C++11 */ \
+          && defined(__GNUC__)
+        #define XXH_ALIGN_MEMBER(align, type) type XXH_ALIGN(align)
+      #else
+        #define XXH_ALIGN_MEMBER(align, type) XXH_ALIGN(align) type
+      #endif
+
+      /*!
+       * @brief The size of the internal XXH3 buffer.
+       *
+       * This is the optimal update size for incremental hashing.
+       *
+       * @see XXH3_64b_update(), XXH3_128b_update().
+       */
+      #define XXH3_INTERNALBUFFER_SIZE 256
+
+      /*!
+       * @internal
+       * @brief Default size of the secret buffer (and @ref XXH3_kSecret).
+       *
+       * This is the size used in @ref XXH3_kSecret and the seeded functions.
+       *
+       * Not to be confused with @ref XXH3_SECRET_SIZE_MIN.
+       */
+      #define XXH3_SECRET_DEFAULT_SIZE 192
 
 /*!
  * @internal
@@ -1744,54 +1864,60 @@ struct XXH64_state_s {
  * @see XXH32_state_s, XXH64_state_s
  */
 struct XXH3_state_s {
-   XXH_ALIGN_MEMBER(64, XXH64_hash_t acc[8]);
-       /*!< The 8 accumulators. See @ref XXH32_state_s::v and @ref XXH64_state_s::v */
-   XXH_ALIGN_MEMBER(64, unsigned char customSecret[XXH3_SECRET_DEFAULT_SIZE]);
-       /*!< Used to store a custom secret generated from a seed. */
-   XXH_ALIGN_MEMBER(64, unsigned char buffer[XXH3_INTERNALBUFFER_SIZE]);
-       /*!< The internal buffer. @see XXH32_state_s::mem32 */
-   XXH32_hash_t bufferedSize;
-       /*!< The amount of memory in @ref buffer, @see XXH32_state_s::memsize */
-   XXH32_hash_t useSeed;
-       /*!< Reserved field. Needed for padding on 64-bit. */
-   size_t nbStripesSoFar;
-       /*!< Number or stripes processed. */
-   XXH64_hash_t totalLen;
-       /*!< Total length hashed. 64-bit even on 32-bit targets. */
-   size_t nbStripesPerBlock;
-       /*!< Number of stripes per block. */
-   size_t secretLimit;
-       /*!< Size of @ref customSecret or @ref extSecret */
-   XXH64_hash_t seed;
-       /*!< Seed for _withSeed variants. Must be zero otherwise, @see XXH3_INITSTATE() */
-   XXH64_hash_t reserved64;
-       /*!< Reserved field. */
-   const unsigned char* extSecret;
-       /*!< Reference to an external secret for the _withSecret variants, NULL
-        *   for other variants. */
-   /* note: there may be some padding at the end due to alignment on 64 bytes */
-}; /* typedef'd to XXH3_state_t */
-
-#undef XXH_ALIGN_MEMBER
-
-/*!
- * @brief Initializes a stack-allocated `XXH3_state_s`.
- *
- * When the @ref XXH3_state_t structure is merely emplaced on stack,
- * it should be initialized with XXH3_INITSTATE() or a memset()
- * in case its first reset uses XXH3_NNbits_reset_withSeed().
- * This init can be omitted if the first reset uses default or _withSecret mode.
- * This operation isn't necessary when the state is created with XXH3_createState().
- * Note that this doesn't prepare the state for a streaming operation,
- * it's still necessary to use XXH3_NNbits_reset*() afterwards.
- */
-#define XXH3_INITSTATE(XXH3_state_ptr)                       \
-    do {                                                     \
-        XXH3_state_t* tmp_xxh3_state_ptr = (XXH3_state_ptr); \
-        tmp_xxh3_state_ptr->seed = 0;                        \
-        tmp_xxh3_state_ptr->extSecret = NULL;                \
-    } while(0)
 
+  XXH_ALIGN_MEMBER(64, XXH64_hash_t acc[8]);
+  /*!< The 8 accumulators. See @ref XXH32_state_s::v and @ref XXH64_state_s::v
+   */
+  XXH_ALIGN_MEMBER(64, unsigned char customSecret[XXH3_SECRET_DEFAULT_SIZE]);
+  /*!< Used to store a custom secret generated from a seed. */
+  XXH_ALIGN_MEMBER(64, unsigned char buffer[XXH3_INTERNALBUFFER_SIZE]);
+  /*!< The internal buffer. @see XXH32_state_s::mem32 */
+  XXH32_hash_t bufferedSize;
+  /*!< The amount of memory in @ref buffer, @see XXH32_state_s::memsize */
+  XXH32_hash_t useSeed;
+  /*!< Reserved field. Needed for padding on 64-bit. */
+  size_t nbStripesSoFar;
+  /*!< Number or stripes processed. */
+  XXH64_hash_t totalLen;
+  /*!< Total length hashed. 64-bit even on 32-bit targets. */
+  size_t nbStripesPerBlock;
+  /*!< Number of stripes per block. */
+  size_t secretLimit;
+  /*!< Size of @ref customSecret or @ref extSecret */
+  XXH64_hash_t seed;
+  /*!< Seed for _withSeed variants. Must be zero otherwise, @see
+   * XXH3_INITSTATE() */
+  XXH64_hash_t reserved64;
+  /*!< Reserved field. */
+  const unsigned char *extSecret;
+  /*!< Reference to an external secret for the _withSecret variants, NULL
+   *   for other variants. */
+  /* note: there may be some padding at the end due to alignment on 64 bytes */
+
+};                                             /* typedef'd to XXH3_state_t */
+
+      #undef XXH_ALIGN_MEMBER
+
+      /*!
+       * @brief Initializes a stack-allocated `XXH3_state_s`.
+       *
+       * When the @ref XXH3_state_t structure is merely emplaced on stack,
+       * it should be initialized with XXH3_INITSTATE() or a memset()
+       * in case its first reset uses XXH3_NNbits_reset_withSeed().
+       * This init can be omitted if the first reset uses default or _withSecret
+       * mode. This operation isn't necessary when the state is created with
+       * XXH3_createState(). Note that this doesn't prepare the state for a
+       * streaming operation, it's still necessary to use XXH3_NNbits_reset*()
+       * afterwards.
+       */
+      #define XXH3_INITSTATE(XXH3_state_ptr)                   \
+        do {                                                   \
+                                                               \
+          XXH3_state_t *tmp_xxh3_state_ptr = (XXH3_state_ptr); \
+          tmp_xxh3_state_ptr->seed = 0;                        \
+          tmp_xxh3_state_ptr->extSecret = NULL;                \
+                                                               \
+        } while (0)
 
 /*!
  * @brief Calculates the 128-bit hash of @p data using XXH3.
@@ -1809,27 +1935,31 @@ struct XXH3_state_s {
  *
  * @see @ref single_shot_example "Single Shot Example" for an example.
  */
-XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH128(XXH_NOESCAPE const void* data, size_t len, XXH64_hash_t seed);
-
+XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH128(XXH_NOESCAPE const void *data,
+                                              size_t len, XXH64_hash_t seed);
 
 /* ===   Experimental API   === */
-/* Symbols defined below must be considered tied to a specific library version. */
+/* Symbols defined below must be considered tied to a specific library version.
+ */
 
 /*!
- * @brief Derive a high-entropy secret from any user-defined content, named customSeed.
+ * @brief Derive a high-entropy secret from any user-defined content, named
+ * customSeed.
  *
- * @param secretBuffer    A writable buffer for derived high-entropy secret data.
- * @param secretSize      Size of secretBuffer, in bytes.  Must be >= XXH3_SECRET_DEFAULT_SIZE.
+ * @param secretBuffer    A writable buffer for derived high-entropy secret
+ * data.
+ * @param secretSize      Size of secretBuffer, in bytes.  Must be >=
+ * XXH3_SECRET_DEFAULT_SIZE.
  * @param customSeed      A user-defined content.
  * @param customSeedSize  Size of customSeed, in bytes.
  *
  * @return @ref XXH_OK on success.
  * @return @ref XXH_ERROR on failure.
  *
- * The generated secret can be used in combination with `*_withSecret()` functions.
- * The `_withSecret()` variants are useful to provide a higher level of protection
- * than 64-bit seed, as it becomes much more difficult for an external actor to
- * guess how to impact the calculation logic.
+ * The generated secret can be used in combination with `*_withSecret()`
+ * functions. The `_withSecret()` variants are useful to provide a higher level
+ * of protection than 64-bit seed, as it becomes much more difficult for an
+ * external actor to guess how to impact the calculation logic.
  *
  * The function accepts as input a custom seed of any length and any content,
  * and derives from it a high-entropy secret of length @p secretSize into an
@@ -1839,18 +1969,20 @@ XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH128(XXH_NOESCAPE const void* data, siz
  * The functions @ref XXH3_128bits_withSecret(), @ref XXH3_64bits_withSecret(),
  * @ref XXH3_128bits_reset_withSecret() and @ref XXH3_64bits_reset_withSecret()
  * are part of this list. They all accept a `secret` parameter
- * which must be large enough for implementation reasons (>= @ref XXH3_SECRET_SIZE_MIN)
- * _and_ feature very high entropy (consist of random-looking bytes).
- * These conditions can be a high bar to meet, so @ref XXH3_generateSecret() can
- * be employed to ensure proper quality.
+ * which must be large enough for implementation reasons (>= @ref
+ * XXH3_SECRET_SIZE_MIN) _and_ feature very high entropy (consist of
+ * random-looking bytes). These conditions can be a high bar to meet, so @ref
+ * XXH3_generateSecret() can be employed to ensure proper quality.
  *
  * @p customSeed can be anything. It can have any size, even small ones,
  * and its content can be anything, even "poor entropy" sources such as a bunch
- * of zeroes. The resulting `secret` will nonetheless provide all required qualities.
+ * of zeroes. The resulting `secret` will nonetheless provide all required
+ * qualities.
  *
  * @pre
  *   - @p secretSize must be >= @ref XXH3_SECRET_SIZE_MIN
- *   - When @p customSeedSize > 0, supplying NULL as customSeed is undefined behavior.
+ *   - When @p customSeedSize > 0, supplying NULL as customSeed is undefined
+ * behavior.
  *
  * Example code:
  * @code{.c}
@@ -1862,6 +1994,7 @@ XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH128(XXH_NOESCAPE const void* data, siz
  *    // Hashes argv[2] using the entropy from argv[1].
  *    int main(int argc, char* argv[])
  *    {
+
  *        char secret[XXH3_SECRET_SIZE_MIN];
  *        if (argv != 3) { return 1; }
  *        XXH3_generateSecret(secret, sizeof(secret), argv[1], strlen(argv[1]));
@@ -1873,7 +2006,9 @@ XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH128(XXH_NOESCAPE const void* data, siz
  *    }
  * @endcode
  */
-XXH_PUBLIC_API XXH_errorcode XXH3_generateSecret(XXH_NOESCAPE void* secretBuffer, size_t secretSize, XXH_NOESCAPE const void* customSeed, size_t customSeedSize);
+XXH_PUBLIC_API XXH_errorcode
+XXH3_generateSecret(XXH_NOESCAPE void *secretBuffer, size_t secretSize,
+                    XXH_NOESCAPE const void *customSeed, size_t customSeedSize);
 
 /*!
  * @brief Generate the same secret as the _withSeed() variants.
@@ -1891,34 +2026,43 @@ XXH_PUBLIC_API XXH_errorcode XXH3_generateSecret(XXH_NOESCAPE void* secretBuffer
  *    #include "xxhash.h"
  *    // Slow, seeds each time
  *    class HashSlow {
+
  *        XXH64_hash_t seed;
  *    public:
  *        HashSlow(XXH64_hash_t s) : seed{s} {}
  *        size_t operator()(const std::string& x) const {
+
  *            return size_t{XXH3_64bits_withSeed(x.c_str(), x.length(), seed)};
  *        }
  *    };
  *    // Fast, caches the seeded secret for future uses.
  *    class HashFast {
+
  *        unsigned char secret[XXH3_SECRET_SIZE_MIN];
  *    public:
  *        HashFast(XXH64_hash_t s) {
+
  *            XXH3_generateSecret_fromSeed(secret, seed);
  *        }
  *        size_t operator()(const std::string& x) const {
+
  *            return size_t{
- *                XXH3_64bits_withSecret(x.c_str(), x.length(), secret, sizeof(secret))
+
+ *                XXH3_64bits_withSecret(x.c_str(), x.length(), secret,
+ *sizeof(secret))
  *            };
  *        }
  *    };
  * @endcode
  */
-XXH_PUBLIC_API void XXH3_generateSecret_fromSeed(XXH_NOESCAPE void* secretBuffer, XXH64_hash_t seed);
+XXH_PUBLIC_API void XXH3_generateSecret_fromSeed(
+    XXH_NOESCAPE void *secretBuffer, XXH64_hash_t seed);
 
 /*!
  * @brief Calculates 64/128-bit seeded variant of XXH3 hash of @p data.
  *
- * @param data       The block of data to be hashed, at least @p len bytes in size.
+ * @param data       The block of data to be hashed, at least @p len bytes in
+ * size.
  * @param len        The length of @p data, in bytes.
  * @param secret     The secret data.
  * @param secretSize The length of @p secret, in bytes.
@@ -1946,17 +2090,17 @@ XXH_PUBLIC_API void XXH3_generateSecret_fromSeed(XXH_NOESCAPE void* secretBuffer
  * On top of speed, an added benefit is that each bit in the secret
  * has a 50% chance to swap each bit in the output, via its impact to the seed.
  *
- * This is not guaranteed when using the secret directly in "small data" scenarios,
- * because only portions of the secret are employed for small data.
+ * This is not guaranteed when using the secret directly in "small data"
+ * scenarios, because only portions of the secret are employed for small data.
  */
-XXH_PUBLIC_API XXH_PUREF XXH64_hash_t
-XXH3_64bits_withSecretandSeed(XXH_NOESCAPE const void* data, size_t len,
-                              XXH_NOESCAPE const void* secret, size_t secretSize,
-                              XXH64_hash_t seed);
+XXH_PUBLIC_API XXH_PUREF XXH64_hash_t XXH3_64bits_withSecretandSeed(
+    XXH_NOESCAPE const void *data, size_t len, XXH_NOESCAPE const void *secret,
+    size_t secretSize, XXH64_hash_t seed);
 /*!
  * @brief Calculates 128-bit seeded variant of XXH3 hash of @p data.
  *
- * @param input      The block of data to be hashed, at least @p len bytes in size.
+ * @param input      The block of data to be hashed, at least @p len bytes in
+ * size.
  * @param length     The length of @p data, in bytes.
  * @param secret     The secret data.
  * @param secretSize The length of @p secret, in bytes.
@@ -1967,15 +2111,15 @@ XXH3_64bits_withSecretandSeed(XXH_NOESCAPE const void* data, size_t len,
  *
  * @see XXH3_64bits_withSecretandSeed()
  */
-XXH_PUBLIC_API XXH_PUREF XXH128_hash_t
-XXH3_128bits_withSecretandSeed(XXH_NOESCAPE const void* input, size_t length,
-                               XXH_NOESCAPE const void* secret, size_t secretSize,
-                               XXH64_hash_t seed64);
-#ifndef XXH_NO_STREAM
+XXH_PUBLIC_API XXH_PUREF XXH128_hash_t XXH3_128bits_withSecretandSeed(
+    XXH_NOESCAPE const void *input, size_t length,
+    XXH_NOESCAPE const void *secret, size_t secretSize, XXH64_hash_t seed64);
+      #ifndef XXH_NO_STREAM
 /*!
  * @brief Resets an @ref XXH3_state_t with secret data to begin a new hash.
  *
- * @param statePtr   A pointer to an @ref XXH3_state_t allocated with @ref XXH3_createState().
+ * @param statePtr   A pointer to an @ref XXH3_state_t allocated with @ref
+ * XXH3_createState().
  * @param secret     The secret data.
  * @param secretSize The length of @p secret, in bytes.
  * @param seed64     The 64-bit seed to alter the hash result predictably.
@@ -1985,14 +2129,14 @@ XXH3_128bits_withSecretandSeed(XXH_NOESCAPE const void* input, size_t length,
  *
  * @see XXH3_64bits_withSecretandSeed()
  */
-XXH_PUBLIC_API XXH_errorcode
-XXH3_64bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr,
-                                    XXH_NOESCAPE const void* secret, size_t secretSize,
-                                    XXH64_hash_t seed64);
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSecretandSeed(
+    XXH_NOESCAPE XXH3_state_t *statePtr, XXH_NOESCAPE const void *secret,
+    size_t secretSize, XXH64_hash_t seed64);
 /*!
  * @brief Resets an @ref XXH3_state_t with secret data to begin a new hash.
  *
- * @param statePtr   A pointer to an @ref XXH3_state_t allocated with @ref XXH3_createState().
+ * @param statePtr   A pointer to an @ref XXH3_state_t allocated with @ref
+ * XXH3_createState().
  * @param secret     The secret data.
  * @param secretSize The length of @p secret, in bytes.
  * @param seed64     The 64-bit seed to alter the hash result predictably.
@@ -2002,26 +2146,24 @@ XXH3_64bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr,
  *
  * @see XXH3_64bits_withSecretandSeed()
  */
-XXH_PUBLIC_API XXH_errorcode
-XXH3_128bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr,
-                                     XXH_NOESCAPE const void* secret, size_t secretSize,
-                                     XXH64_hash_t seed64);
-#endif /* !XXH_NO_STREAM */
-
-#endif  /* !XXH_NO_XXH3 */
-#endif  /* XXH_NO_LONG_LONG */
-#if defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API)
-#  define XXH_IMPLEMENTATION
-#endif
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSecretandSeed(
+    XXH_NOESCAPE XXH3_state_t *statePtr, XXH_NOESCAPE const void *secret,
+    size_t secretSize, XXH64_hash_t seed64);
+      #endif                                              /* !XXH_NO_STREAM */
 
-#endif  /* defined(XXH_STATIC_LINKING_ONLY) && !defined(XXHASH_H_STATIC_13879238742) */
+    #endif                                                  /* !XXH_NO_XXH3 */
+  #endif                                                /* XXH_NO_LONG_LONG */
+  #if defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API)
+    #define XXH_IMPLEMENTATION
+  #endif
 
+#endif /* defined(XXH_STATIC_LINKING_ONLY) && \
+          !defined(XXHASH_H_STATIC_13879238742) */
 
 /* ======================================================================== */
 /* ======================================================================== */
 /* ======================================================================== */
 
-
 /*-**********************************************************************
  * xxHash implementation
  *-**********************************************************************
@@ -2044,277 +2186,290 @@ XXH3_128bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr,
  * which can then be linked into the final binary.
  ************************************************************************/
 
-#if ( defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API) \
-   || defined(XXH_IMPLEMENTATION) ) && !defined(XXH_IMPLEM_13a8737387)
-#  define XXH_IMPLEM_13a8737387
+#if (defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API) || \
+     defined(XXH_IMPLEMENTATION)) &&                        \
+    !defined(XXH_IMPLEM_13a8737387)
+  #define XXH_IMPLEM_13a8737387
+
+  /* *************************************
+   *  Tuning parameters
+   ***************************************/
+
+  /*!
+   * @defgroup tuning Tuning parameters
+   * @{
+
+   *
+   * Various macros to control xxHash's behavior.
+   */
+  #ifdef XXH_DOXYGEN
+    /*!
+     * @brief Define this to disable 64-bit code.
+     *
+     * Useful if only using the @ref XXH32_family and you have a strict C90
+     * compiler.
+     */
+    #define XXH_NO_LONG_LONG
+    #undef XXH_NO_LONG_LONG                               /* don't actually */
+    /*!
+     * @brief Controls how unaligned memory is accessed.
+     *
+     * By default, access to unaligned memory is controlled by `memcpy()`, which
+     * is safe and portable.
+     *
+     * Unfortunately, on some target/compiler combinations, the generated
+     * assembly is sub-optimal.
+     *
+     * The below switch allow selection of a different access method
+     * in the search for improved performance.
+     *
+     * @par Possible options:
+     *
+     *  - `XXH_FORCE_MEMORY_ACCESS=0` (default): `memcpy`
+     *   @par
+     *     Use `memcpy()`. Safe and portable. Note that most modern compilers
+     * will eliminate the function call and treat it as an unaligned access.
+     *
+     *  - `XXH_FORCE_MEMORY_ACCESS=1`: `__attribute__((aligned(1)))`
+     *   @par
+     *     Depends on compiler extensions and is therefore not portable.
+     *     This method is safe _if_ your compiler supports it,
+     *     and *generally* as fast or faster than `memcpy`.
+     *
+     *  - `XXH_FORCE_MEMORY_ACCESS=2`: Direct cast
+     *  @par
+     *     Casts directly and dereferences. This method doesn't depend on the
+     *     compiler, but it violates the C standard as it directly dereferences
+     * an unaligned pointer. It can generate buggy code on targets which do not
+     *     support unaligned memory accesses, but in some circumstances, it's
+     * the only known way to get the most performance.
+     *
+     *  - `XXH_FORCE_MEMORY_ACCESS=3`: Byteshift
+     *  @par
+     *     Also portable. This can generate the best code on old compilers which
+     * don't inline small `memcpy()` calls, and it might also be faster on
+     * big-endian systems which lack a native byteswap instruction. However,
+     * some compilers will emit literal byteshifts even if the target supports
+     * unaligned access.
+     *
+     *
+     * @warning
+     *   Methods 1 and 2 rely on implementation-defined behavior. Use these with
+     *   care, as what works on one compiler/platform/optimization level may
+     * cause another to read garbage data or even crash.
+     *
+     * See
+     * https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html
+     * for details.
+     *
+     * Prefer these methods in priority order (0 > 3 > 1 > 2)
+     */
+    #define XXH_FORCE_MEMORY_ACCESS 0
 
-/* *************************************
-*  Tuning parameters
-***************************************/
+    /*!
+     * @def XXH_SIZE_OPT
+     * @brief Controls how much xxHash optimizes for size.
+     *
+     * xxHash, when compiled, tends to result in a rather large binary size.
+     * This is mostly due to heavy usage to forced inlining and constant folding
+     * of the
+     * @ref XXH3_family to increase performance.
+     *
+     * However, some developers prefer size over speed. This option can
+     * significantly reduce the size of the generated code. When using the `-Os`
+     * or `-Oz` options on GCC or Clang, this is defined to 1 by default,
+     * otherwise it is defined to 0.
+     *
+     * Most of these size optimizations can be controlled manually.
+     *
+     * This is a number from 0-2.
+     *  - `XXH_SIZE_OPT` == 0: Default. xxHash makes no size optimizations.
+     * Speed comes first.
+     *  - `XXH_SIZE_OPT` == 1: Default for `-Os` and `-Oz`. xxHash is more
+     *    conservative and disables hacks that increase code size. It implies
+     * the options @ref XXH_NO_INLINE_HINTS == 1, @ref XXH_FORCE_ALIGN_CHECK ==
+     * 0, and @ref XXH3_NEON_LANES == 8 if they are not already defined.
+     *  - `XXH_SIZE_OPT` == 2: xxHash tries to make itself as small as possible.
+     *    Performance may cry. For example, the single shot functions just use
+     * the streaming API.
+     */
+    #define XXH_SIZE_OPT 0
 
-/*!
- * @defgroup tuning Tuning parameters
- * @{
- *
- * Various macros to control xxHash's behavior.
- */
-#ifdef XXH_DOXYGEN
-/*!
- * @brief Define this to disable 64-bit code.
- *
- * Useful if only using the @ref XXH32_family and you have a strict C90 compiler.
- */
-#  define XXH_NO_LONG_LONG
-#  undef XXH_NO_LONG_LONG /* don't actually */
-/*!
- * @brief Controls how unaligned memory is accessed.
- *
- * By default, access to unaligned memory is controlled by `memcpy()`, which is
- * safe and portable.
- *
- * Unfortunately, on some target/compiler combinations, the generated assembly
- * is sub-optimal.
- *
- * The below switch allow selection of a different access method
- * in the search for improved performance.
- *
- * @par Possible options:
- *
- *  - `XXH_FORCE_MEMORY_ACCESS=0` (default): `memcpy`
- *   @par
- *     Use `memcpy()`. Safe and portable. Note that most modern compilers will
- *     eliminate the function call and treat it as an unaligned access.
- *
- *  - `XXH_FORCE_MEMORY_ACCESS=1`: `__attribute__((aligned(1)))`
- *   @par
- *     Depends on compiler extensions and is therefore not portable.
- *     This method is safe _if_ your compiler supports it,
- *     and *generally* as fast or faster than `memcpy`.
- *
- *  - `XXH_FORCE_MEMORY_ACCESS=2`: Direct cast
- *  @par
- *     Casts directly and dereferences. This method doesn't depend on the
- *     compiler, but it violates the C standard as it directly dereferences an
- *     unaligned pointer. It can generate buggy code on targets which do not
- *     support unaligned memory accesses, but in some circumstances, it's the
- *     only known way to get the most performance.
- *
- *  - `XXH_FORCE_MEMORY_ACCESS=3`: Byteshift
- *  @par
- *     Also portable. This can generate the best code on old compilers which don't
- *     inline small `memcpy()` calls, and it might also be faster on big-endian
- *     systems which lack a native byteswap instruction. However, some compilers
- *     will emit literal byteshifts even if the target supports unaligned access.
- *
- *
- * @warning
- *   Methods 1 and 2 rely on implementation-defined behavior. Use these with
- *   care, as what works on one compiler/platform/optimization level may cause
- *   another to read garbage data or even crash.
- *
- * See https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html for details.
- *
- * Prefer these methods in priority order (0 > 3 > 1 > 2)
- */
-#  define XXH_FORCE_MEMORY_ACCESS 0
+    /*!
+     * @def XXH_FORCE_ALIGN_CHECK
+     * @brief If defined to non-zero, adds a special path for aligned inputs
+     * (XXH32() and XXH64() only).
+     *
+     * This is an important performance trick for architectures without decent
+     * unaligned memory access performance.
+     *
+     * It checks for input alignment, and when conditions are met, uses a "fast
+     * path" employing direct 32-bit/64-bit reads, resulting in _dramatically
+     * faster_ read speed.
+     *
+     * The check costs one initial branch per hash, which is generally
+     * negligible, but not zero.
+     *
+     * Moreover, it's not useful to generate an additional code path if memory
+     * access uses the same instruction for both aligned and unaligned
+     * addresses (e.g. x86 and aarch64).
+     *
+     * In these cases, the alignment check can be removed by setting this macro
+     * to 0. Then the code will always use unaligned memory access. Align check
+     * is automatically disabled on x86, x64, ARM64, and some ARM chips which
+     * are platforms known to offer good unaligned memory accesses performance.
+     *
+     * It is also disabled by default when @ref XXH_SIZE_OPT >= 1.
+     *
+     * This option does not affect XXH3 (only XXH32 and XXH64).
+     */
+    #define XXH_FORCE_ALIGN_CHECK 0
 
-/*!
- * @def XXH_SIZE_OPT
- * @brief Controls how much xxHash optimizes for size.
- *
- * xxHash, when compiled, tends to result in a rather large binary size. This
- * is mostly due to heavy usage to forced inlining and constant folding of the
- * @ref XXH3_family to increase performance.
- *
- * However, some developers prefer size over speed. This option can
- * significantly reduce the size of the generated code. When using the `-Os`
- * or `-Oz` options on GCC or Clang, this is defined to 1 by default,
- * otherwise it is defined to 0.
- *
- * Most of these size optimizations can be controlled manually.
- *
- * This is a number from 0-2.
- *  - `XXH_SIZE_OPT` == 0: Default. xxHash makes no size optimizations. Speed
- *    comes first.
- *  - `XXH_SIZE_OPT` == 1: Default for `-Os` and `-Oz`. xxHash is more
- *    conservative and disables hacks that increase code size. It implies the
- *    options @ref XXH_NO_INLINE_HINTS == 1, @ref XXH_FORCE_ALIGN_CHECK == 0,
- *    and @ref XXH3_NEON_LANES == 8 if they are not already defined.
- *  - `XXH_SIZE_OPT` == 2: xxHash tries to make itself as small as possible.
- *    Performance may cry. For example, the single shot functions just use the
- *    streaming API.
- */
-#  define XXH_SIZE_OPT 0
+    /*!
+     * @def XXH_NO_INLINE_HINTS
+     * @brief When non-zero, sets all functions to `static`.
+     *
+     * By default, xxHash tries to force the compiler to inline almost all
+     * internal functions.
+     *
+     * This can usually improve performance due to reduced jumping and improved
+     * constant folding, but significantly increases the size of the binary
+     * which might not be favorable.
+     *
+     * Additionally, sometimes the forced inlining can be detrimental to
+     * performance, depending on the architecture.
+     *
+     * XXH_NO_INLINE_HINTS marks all internal functions as static, giving the
+     * compiler full control on whether to inline or not.
+     *
+     * When not optimizing (-O0), using `-fno-inline` with GCC or Clang, or if
+     * @ref XXH_SIZE_OPT >= 1, this will automatically be defined.
+     */
+    #define XXH_NO_INLINE_HINTS 0
+
+    /*!
+     * @def XXH3_INLINE_SECRET
+     * @brief Determines whether to inline the XXH3 withSecret code.
+     *
+     * When the secret size is known, the compiler can improve the performance
+     * of XXH3_64bits_withSecret() and XXH3_128bits_withSecret().
+     *
+     * However, if the secret size is not known, it doesn't have any benefit.
+     * This happens when xxHash is compiled into a global symbol. Therefore, if
+     * @ref XXH_INLINE_ALL is *not* defined, this will be defined to 0.
+     *
+     * Additionally, this defaults to 0 on GCC 12+, which has an issue with
+     * function pointers that are *sometimes* force inline on -Og, and it is
+     * impossible to automatically detect this optimization level.
+     */
+    #define XXH3_INLINE_SECRET 0
+
+    /*!
+     * @def XXH32_ENDJMP
+     * @brief Whether to use a jump for `XXH32_finalize`.
+     *
+     * For performance, `XXH32_finalize` uses multiple branches in the
+     * finalizer. This is generally preferable for performance, but depending on
+     * exact architecture, a jmp may be preferable.
+     *
+     * This setting is only possibly making a difference for very small inputs.
+     */
+    #define XXH32_ENDJMP 0
+
+    /*!
+     * @internal
+     * @brief Redefines old internal names.
+     *
+     * For compatibility with code that uses xxHash's internals before the names
+     * were changed to improve namespacing. There is no other reason to use
+     * this.
+     */
+    #define XXH_OLD_NAMES
+    #undef XXH_OLD_NAMES                 /* don't actually use, it is ugly. */
 
+    /*!
+     * @def XXH_NO_STREAM
+     * @brief Disables the streaming API.
+     *
+     * When xxHash is not inlined and the streaming functions are not used,
+     * disabling the streaming functions can improve code size significantly,
+     * especially with the @ref XXH3_family which tends to make constant folded
+     * copies of itself.
+     */
+    #define XXH_NO_STREAM
+    #undef XXH_NO_STREAM                                  /* don't actually */
+  #endif                                                     /* XXH_DOXYGEN */
 /*!
- * @def XXH_FORCE_ALIGN_CHECK
- * @brief If defined to non-zero, adds a special path for aligned inputs (XXH32()
- * and XXH64() only).
- *
- * This is an important performance trick for architectures without decent
- * unaligned memory access performance.
- *
- * It checks for input alignment, and when conditions are met, uses a "fast
- * path" employing direct 32-bit/64-bit reads, resulting in _dramatically
- * faster_ read speed.
- *
- * The check costs one initial branch per hash, which is generally negligible,
- * but not zero.
- *
- * Moreover, it's not useful to generate an additional code path if memory
- * access uses the same instruction for both aligned and unaligned
- * addresses (e.g. x86 and aarch64).
- *
- * In these cases, the alignment check can be removed by setting this macro to 0.
- * Then the code will always use unaligned memory access.
- * Align check is automatically disabled on x86, x64, ARM64, and some ARM chips
- * which are platforms known to offer good unaligned memory accesses performance.
- *
- * It is also disabled by default when @ref XXH_SIZE_OPT >= 1.
- *
- * This option does not affect XXH3 (only XXH32 and XXH64).
+ * @}
  */
-#  define XXH_FORCE_ALIGN_CHECK 0
 
-/*!
- * @def XXH_NO_INLINE_HINTS
- * @brief When non-zero, sets all functions to `static`.
- *
- * By default, xxHash tries to force the compiler to inline almost all internal
- * functions.
- *
- * This can usually improve performance due to reduced jumping and improved
- * constant folding, but significantly increases the size of the binary which
- * might not be favorable.
- *
- * Additionally, sometimes the forced inlining can be detrimental to performance,
- * depending on the architecture.
- *
- * XXH_NO_INLINE_HINTS marks all internal functions as static, giving the
- * compiler full control on whether to inline or not.
- *
- * When not optimizing (-O0), using `-fno-inline` with GCC or Clang, or if
- * @ref XXH_SIZE_OPT >= 1, this will automatically be defined.
- */
-#  define XXH_NO_INLINE_HINTS 0
-
-/*!
- * @def XXH3_INLINE_SECRET
- * @brief Determines whether to inline the XXH3 withSecret code.
- *
- * When the secret size is known, the compiler can improve the performance
- * of XXH3_64bits_withSecret() and XXH3_128bits_withSecret().
- *
- * However, if the secret size is not known, it doesn't have any benefit. This
- * happens when xxHash is compiled into a global symbol. Therefore, if
- * @ref XXH_INLINE_ALL is *not* defined, this will be defined to 0.
- *
- * Additionally, this defaults to 0 on GCC 12+, which has an issue with function pointers
- * that are *sometimes* force inline on -Og, and it is impossible to automatically
- * detect this optimization level.
- */
-#  define XXH3_INLINE_SECRET 0
-
-/*!
- * @def XXH32_ENDJMP
- * @brief Whether to use a jump for `XXH32_finalize`.
- *
- * For performance, `XXH32_finalize` uses multiple branches in the finalizer.
- * This is generally preferable for performance,
- * but depending on exact architecture, a jmp may be preferable.
- *
- * This setting is only possibly making a difference for very small inputs.
- */
-#  define XXH32_ENDJMP 0
-
-/*!
- * @internal
- * @brief Redefines old internal names.
- *
- * For compatibility with code that uses xxHash's internals before the names
- * were changed to improve namespacing. There is no other reason to use this.
- */
-#  define XXH_OLD_NAMES
-#  undef XXH_OLD_NAMES /* don't actually use, it is ugly. */
-
-/*!
- * @def XXH_NO_STREAM
- * @brief Disables the streaming API.
- *
- * When xxHash is not inlined and the streaming functions are not used, disabling
- * the streaming functions can improve code size significantly, especially with
- * the @ref XXH3_family which tends to make constant folded copies of itself.
- */
-#  define XXH_NO_STREAM
-#  undef XXH_NO_STREAM /* don't actually */
-#endif /* XXH_DOXYGEN */
-/*!
- * @}
- */
-
-#ifndef XXH_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
-   /* prefer __packed__ structures (method 1) for GCC
-    * < ARMv7 with unaligned access (e.g. Raspbian armhf) still uses byte shifting, so we use memcpy
-    * which for some reason does unaligned loads. */
-#  if defined(__GNUC__) && !(defined(__ARM_ARCH) && __ARM_ARCH < 7 && defined(__ARM_FEATURE_UNALIGNED))
-#    define XXH_FORCE_MEMORY_ACCESS 1
-#  endif
-#endif
-
-#ifndef XXH_SIZE_OPT
-   /* default to 1 for -Os or -Oz */
-#  if (defined(__GNUC__) || defined(__clang__)) && defined(__OPTIMIZE_SIZE__)
-#    define XXH_SIZE_OPT 1
-#  else
-#    define XXH_SIZE_OPT 0
-#  endif
-#endif
-
-#ifndef XXH_FORCE_ALIGN_CHECK  /* can be defined externally */
-   /* don't check on sizeopt, x86, aarch64, or arm when unaligned access is available */
-#  if XXH_SIZE_OPT >= 1 || \
-      defined(__i386)  || defined(__x86_64__) || defined(__aarch64__) || defined(__ARM_FEATURE_UNALIGNED) \
-   || defined(_M_IX86) || defined(_M_X64)     || defined(_M_ARM64)    || defined(_M_ARM) /* visual */
-#    define XXH_FORCE_ALIGN_CHECK 0
-#  else
-#    define XXH_FORCE_ALIGN_CHECK 1
-#  endif
-#endif
-
-#ifndef XXH_NO_INLINE_HINTS
-#  if XXH_SIZE_OPT >= 1 || defined(__NO_INLINE__)  /* -O0, -fno-inline */
-#    define XXH_NO_INLINE_HINTS 1
-#  else
-#    define XXH_NO_INLINE_HINTS 0
-#  endif
-#endif
-
-#ifndef XXH3_INLINE_SECRET
-#  if (defined(__GNUC__) && !defined(__clang__) && __GNUC__ >= 12) \
-     || !defined(XXH_INLINE_ALL)
-#    define XXH3_INLINE_SECRET 0
-#  else
-#    define XXH3_INLINE_SECRET 1
-#  endif
-#endif
-
-#ifndef XXH32_ENDJMP
-/* generally preferable for performance */
-#  define XXH32_ENDJMP 0
-#endif
-
-/*!
- * @defgroup impl Implementation
- * @{
- */
-
-
-/* *************************************
-*  Includes & Memory related functions
-***************************************/
-#if defined(XXH_NO_STREAM)
-/* nothing */
-#elif defined(XXH_NO_STDLIB)
+  #ifndef XXH_FORCE_MEMORY_ACCESS /* can be defined externally, on command \
+                                     line for example */
+  /* prefer __packed__ structures (method 1) for GCC
+   * < ARMv7 with unaligned access (e.g. Raspbian armhf) still uses byte
+   * shifting, so we use memcpy which for some reason does unaligned loads. */
+    #if defined(__GNUC__) && !(defined(__ARM_ARCH) && __ARM_ARCH < 7 && \
+                               defined(__ARM_FEATURE_UNALIGNED))
+      #define XXH_FORCE_MEMORY_ACCESS 1
+    #endif
+  #endif
+
+  #ifndef XXH_SIZE_OPT
+  /* default to 1 for -Os or -Oz */
+    #if (defined(__GNUC__) || defined(__clang__)) && defined(__OPTIMIZE_SIZE__)
+      #define XXH_SIZE_OPT 1
+    #else
+      #define XXH_SIZE_OPT 0
+    #endif
+  #endif
+
+  #ifndef XXH_FORCE_ALIGN_CHECK                /* can be defined externally */
+  /* don't check on sizeopt, x86, aarch64, or arm when unaligned access is
+   * available */
+    #if XXH_SIZE_OPT >= 1 || defined(__i386) || defined(__x86_64__) || \
+        defined(__aarch64__) || defined(__ARM_FEATURE_UNALIGNED) ||    \
+        defined(_M_IX86) || defined(_M_X64) || defined(_M_ARM64) ||    \
+        defined(_M_ARM)                                           /* visual */
+      #define XXH_FORCE_ALIGN_CHECK 0
+    #else
+      #define XXH_FORCE_ALIGN_CHECK 1
+    #endif
+  #endif
+
+  #ifndef XXH_NO_INLINE_HINTS
+    #if XXH_SIZE_OPT >= 1 || defined(__NO_INLINE__)     /* -O0, -fno-inline */
+      #define XXH_NO_INLINE_HINTS 1
+    #else
+      #define XXH_NO_INLINE_HINTS 0
+    #endif
+  #endif
+
+  #ifndef XXH3_INLINE_SECRET
+    #if (defined(__GNUC__) && !defined(__clang__) && __GNUC__ >= 12) || \
+        !defined(XXH_INLINE_ALL)
+      #define XXH3_INLINE_SECRET 0
+    #else
+      #define XXH3_INLINE_SECRET 1
+    #endif
+  #endif
+
+  #ifndef XXH32_ENDJMP
+    /* generally preferable for performance */
+    #define XXH32_ENDJMP 0
+  #endif
+
+  /*!
+   * @defgroup impl Implementation
+   * @{
+
+   */
+
+  /* *************************************
+   *  Includes & Memory related functions
+   ***************************************/
+  #if defined(XXH_NO_STREAM)
+  /* nothing */
+  #elif defined(XXH_NO_STDLIB)
 
 /* When requesting to disable any mention of stdlib,
  * the library loses the ability to invoked malloc / free.
@@ -2325,173 +2480,212 @@ XXH3_128bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr,
  * without access to dynamic allocation.
  */
 
-static XXH_CONSTF void* XXH_malloc(size_t s) { (void)s; return NULL; }
-static void XXH_free(void* p) { (void)p; }
+static XXH_CONSTF void *XXH_malloc(size_t s) {
 
-#else
+  (void)s;
+  return NULL;
 
-/*
- * Modify the local functions below should you wish to use
- * different memory routines for malloc() and free()
- */
-#include <stdlib.h>
+}
 
-/*!
- * @internal
- * @brief Modify this function to use a different routine than malloc().
- */
-static XXH_MALLOCF void* XXH_malloc(size_t s) { return malloc(s); }
+static void XXH_free(void *p) {
 
-/*!
- * @internal
- * @brief Modify this function to use a different routine than free().
- */
-static void XXH_free(void* p) { free(p); }
+  (void)p;
 
-#endif  /* XXH_NO_STDLIB */
+}
+
+  #else
 
-#include <string.h>
+  /*
+   * Modify the local functions below should you wish to use
+   * different memory routines for malloc() and free()
+   */
+    #include <stdlib.h>
 
 /*!
  * @internal
- * @brief Modify this function to use a different routine than memcpy().
+ * @brief Modify this function to use a different routine than malloc().
  */
-static void* XXH_memcpy(void* dest, const void* src, size_t size)
-{
-    return memcpy(dest,src,size);
-}
-
-#include <limits.h>   /* ULLONG_MAX */
-
+static XXH_MALLOCF void *XXH_malloc(size_t s) {
 
-/* *************************************
-*  Compiler Specific Options
-***************************************/
-#ifdef _MSC_VER /* Visual Studio warning fix */
-#  pragma warning(disable : 4127) /* disable: C4127: conditional expression is constant */
-#endif
+  return malloc(s);
 
-#if XXH_NO_INLINE_HINTS  /* disable inlining hints */
-#  if defined(__GNUC__) || defined(__clang__)
-#    define XXH_FORCE_INLINE static __attribute__((unused))
-#  else
-#    define XXH_FORCE_INLINE static
-#  endif
-#  define XXH_NO_INLINE static
-/* enable inlining hints */
-#elif defined(__GNUC__) || defined(__clang__)
-#  define XXH_FORCE_INLINE static __inline__ __attribute__((always_inline, unused))
-#  define XXH_NO_INLINE static __attribute__((noinline))
-#elif defined(_MSC_VER)  /* Visual Studio */
-#  define XXH_FORCE_INLINE static __forceinline
-#  define XXH_NO_INLINE static __declspec(noinline)
-#elif defined (__cplusplus) \
-  || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L))   /* C99 */
-#  define XXH_FORCE_INLINE static inline
-#  define XXH_NO_INLINE static
-#else
-#  define XXH_FORCE_INLINE static
-#  define XXH_NO_INLINE static
-#endif
+}
 
-#if XXH3_INLINE_SECRET
-#  define XXH3_WITH_SECRET_INLINE XXH_FORCE_INLINE
-#else
-#  define XXH3_WITH_SECRET_INLINE XXH_NO_INLINE
-#endif
+/*!
+ * @internal
+ * @brief Modify this function to use a different routine than free().
+ */
+static void XXH_free(void *p) {
 
+  free(p);
 
-/* *************************************
-*  Debug
-***************************************/
-/*!
- * @ingroup tuning
- * @def XXH_DEBUGLEVEL
- * @brief Sets the debugging level.
- *
- * XXH_DEBUGLEVEL is expected to be defined externally, typically via the
- * compiler's command line options. The value must be a number.
- */
-#ifndef XXH_DEBUGLEVEL
-#  ifdef DEBUGLEVEL /* backwards compat */
-#    define XXH_DEBUGLEVEL DEBUGLEVEL
-#  else
-#    define XXH_DEBUGLEVEL 0
-#  endif
-#endif
+}
 
-#if (XXH_DEBUGLEVEL>=1)
-#  include <assert.h>   /* note: can still be disabled with NDEBUG */
-#  define XXH_ASSERT(c)   assert(c)
-#else
-#  if defined(__INTEL_COMPILER)
-#    define XXH_ASSERT(c)   XXH_ASSUME((unsigned char) (c))
-#  else
-#    define XXH_ASSERT(c)   XXH_ASSUME(c)
-#  endif
-#endif
+  #endif                                                   /* XXH_NO_STDLIB */
 
-/* note: use after variable declarations */
-#ifndef XXH_STATIC_ASSERT
-#  if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L)    /* C11 */
-#    define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { _Static_assert((c),m); } while(0)
-#  elif defined(__cplusplus) && (__cplusplus >= 201103L)            /* C++11 */
-#    define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { static_assert((c),m); } while(0)
-#  else
-#    define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { struct xxh_sa { char x[(c) ? 1 : -1]; }; } while(0)
-#  endif
-#  define XXH_STATIC_ASSERT(c) XXH_STATIC_ASSERT_WITH_MESSAGE((c),#c)
-#endif
+  #include <string.h>
 
 /*!
  * @internal
- * @def XXH_COMPILER_GUARD(var)
- * @brief Used to prevent unwanted optimizations for @p var.
- *
- * It uses an empty GCC inline assembly statement with a register constraint
- * which forces @p var into a general purpose register (eg eax, ebx, ecx
- * on x86) and marks it as modified.
- *
- * This is used in a few places to avoid unwanted autovectorization (e.g.
- * XXH32_round()). All vectorization we want is explicit via intrinsics,
- * and _usually_ isn't wanted elsewhere.
- *
- * We also use it to prevent unwanted constant folding for AArch64 in
- * XXH3_initCustomSecret_scalar().
+ * @brief Modify this function to use a different routine than memcpy().
  */
-#if defined(__GNUC__) || defined(__clang__)
-#  define XXH_COMPILER_GUARD(var) __asm__("" : "+r" (var))
-#else
-#  define XXH_COMPILER_GUARD(var) ((void)0)
-#endif
-
-/* Specifically for NEON vectors which use the "w" constraint, on
- * Clang. */
-#if defined(__clang__) && defined(__ARM_ARCH) && !defined(__wasm__)
-#  define XXH_COMPILER_GUARD_CLANG_NEON(var) __asm__("" : "+w" (var))
-#else
-#  define XXH_COMPILER_GUARD_CLANG_NEON(var) ((void)0)
-#endif
-
-/* *************************************
-*  Basic Types
-***************************************/
-#if !defined (__VMS) \
- && (defined (__cplusplus) \
- || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
-# include <stdint.h>
-  typedef uint8_t xxh_u8;
-#else
-  typedef unsigned char xxh_u8;
-#endif
+static void *XXH_memcpy(void *dest, const void *src, size_t size) {
+
+  return memcpy(dest, src, size);
+
+}
+
+  #include <limits.h>                                         /* ULLONG_MAX */
+
+  /* *************************************
+   *  Compiler Specific Options
+   ***************************************/
+  #ifdef _MSC_VER                              /* Visual Studio warning fix */
+    #pragma warning(disable : 4127) /* disable: C4127: conditional expression \
+                                       is constant */
+  #endif
+
+  #if XXH_NO_INLINE_HINTS                         /* disable inlining hints */
+    #if defined(__GNUC__) || defined(__clang__)
+      #define XXH_FORCE_INLINE static __attribute__((unused))
+    #else
+      #define XXH_FORCE_INLINE static
+    #endif
+    #define XXH_NO_INLINE static
+  /* enable inlining hints */
+  #elif defined(__GNUC__) || defined(__clang__)
+    #define XXH_FORCE_INLINE \
+      static __inline__ __attribute__((always_inline, unused))
+    #define XXH_NO_INLINE static __attribute__((noinline))
+  #elif defined(_MSC_VER)                                  /* Visual Studio */
+    #define XXH_FORCE_INLINE static __forceinline
+    #define XXH_NO_INLINE static __declspec(noinline)
+  #elif defined(__cplusplus) || \
+      (defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L))   /* C99 */
+    #define XXH_FORCE_INLINE static inline
+    #define XXH_NO_INLINE static
+  #else
+    #define XXH_FORCE_INLINE static
+    #define XXH_NO_INLINE static
+  #endif
+
+  #if XXH3_INLINE_SECRET
+    #define XXH3_WITH_SECRET_INLINE XXH_FORCE_INLINE
+  #else
+    #define XXH3_WITH_SECRET_INLINE XXH_NO_INLINE
+  #endif
+
+  /* *************************************
+   *  Debug
+   ***************************************/
+  /*!
+   * @ingroup tuning
+   * @def XXH_DEBUGLEVEL
+   * @brief Sets the debugging level.
+   *
+   * XXH_DEBUGLEVEL is expected to be defined externally, typically via the
+   * compiler's command line options. The value must be a number.
+   */
+  #ifndef XXH_DEBUGLEVEL
+    #ifdef DEBUGLEVEL                                   /* backwards compat */
+      #define XXH_DEBUGLEVEL DEBUGLEVEL
+    #else
+      #define XXH_DEBUGLEVEL 0
+    #endif
+  #endif
+
+  #if (XXH_DEBUGLEVEL >= 1)
+    #include <assert.h>          /* note: can still be disabled with NDEBUG */
+    #define XXH_ASSERT(c) assert(c)
+  #else
+    #if defined(__INTEL_COMPILER)
+      #define XXH_ASSERT(c) XXH_ASSUME((unsigned char)(c))
+    #else
+      #define XXH_ASSERT(c) XXH_ASSUME(c)
+    #endif
+  #endif
+
+  /* note: use after variable declarations */
+  #ifndef XXH_STATIC_ASSERT
+    #if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L)   /* C11 */
+      #define XXH_STATIC_ASSERT_WITH_MESSAGE(c, m) \
+        do {                                       \
+                                                   \
+          _Static_assert((c), m);                  \
+                                                   \
+        } while (0)
+    #elif defined(__cplusplus) && (__cplusplus >= 201103L)         /* C++11 */
+      #define XXH_STATIC_ASSERT_WITH_MESSAGE(c, m) \
+        do {                                       \
+                                                   \
+          static_assert((c), m);                   \
+                                                   \
+        } while (0)
+    #else
+      #define XXH_STATIC_ASSERT_WITH_MESSAGE(c, m) \
+        do {                                       \
+                                                   \
+          struct xxh_sa {                          \
+                                                   \
+            char x[(c) ? 1 : -1];                  \
+                                                   \
+          };                                       \
+                                                   \
+        } while (0)
+    #endif
+    #define XXH_STATIC_ASSERT(c) XXH_STATIC_ASSERT_WITH_MESSAGE((c), #c)
+  #endif
+
+  /*!
+   * @internal
+   * @def XXH_COMPILER_GUARD(var)
+   * @brief Used to prevent unwanted optimizations for @p var.
+   *
+   * It uses an empty GCC inline assembly statement with a register constraint
+   * which forces @p var into a general purpose register (eg eax, ebx, ecx
+   * on x86) and marks it as modified.
+   *
+   * This is used in a few places to avoid unwanted autovectorization (e.g.
+   * XXH32_round()). All vectorization we want is explicit via intrinsics,
+   * and _usually_ isn't wanted elsewhere.
+   *
+   * We also use it to prevent unwanted constant folding for AArch64 in
+   * XXH3_initCustomSecret_scalar().
+   */
+  #if defined(__GNUC__) || defined(__clang__)
+    #define XXH_COMPILER_GUARD(var) __asm__("" : "+r"(var))
+  #else
+    #define XXH_COMPILER_GUARD(var) ((void)0)
+  #endif
+
+  /* Specifically for NEON vectors which use the "w" constraint, on
+   * Clang. */
+  #if defined(__clang__) && defined(__ARM_ARCH) && !defined(__wasm__)
+    #define XXH_COMPILER_GUARD_CLANG_NEON(var) __asm__("" : "+w"(var))
+  #else
+    #define XXH_COMPILER_GUARD_CLANG_NEON(var) ((void)0)
+  #endif
+
+  /* *************************************
+   *  Basic Types
+   ***************************************/
+  #if !defined(__VMS) &&       \
+      (defined(__cplusplus) || \
+       (defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */))
+    #include <stdint.h>
+typedef uint8_t xxh_u8;
+  #else
+typedef unsigned char xxh_u8;
+  #endif
 typedef XXH32_hash_t xxh_u32;
 
-#ifdef XXH_OLD_NAMES
-#  warning "XXH_OLD_NAMES is planned to be removed starting v0.9. If the program depends on it, consider moving away from it by employing newer type names directly"
-#  define BYTE xxh_u8
-#  define U8   xxh_u8
-#  define U32  xxh_u32
-#endif
+  #ifdef XXH_OLD_NAMES
+    #warning \
+        "XXH_OLD_NAMES is planned to be removed starting v0.9. If the program depends on it, consider moving away from it by employing newer type names directly"
+    #define BYTE xxh_u8
+    #define U8 xxh_u8
+    #define U32 xxh_u32
+  #endif
 
 /* ***   Memory access   *** */
 
@@ -2545,118 +2739,132 @@ typedef XXH32_hash_t xxh_u32;
  * @return The 32-bit little endian integer from the bytes at @p ptr.
  */
 
-#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
-/*
- * Manual byteshift. Best for old compilers which don't inline memcpy.
- * We actually directly use XXH_readLE32 and XXH_readBE32.
- */
-#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))
+  #if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS == 3))
+  /*
+   * Manual byteshift. Best for old compilers which don't inline memcpy.
+   * We actually directly use XXH_readLE32 and XXH_readBE32.
+   */
+  #elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS == 2))
 
 /*
  * Force direct memory access. Only works on CPU which support unaligned memory
  * access in hardware.
  */
-static xxh_u32 XXH_read32(const void* memPtr) { return *(const xxh_u32*) memPtr; }
+static xxh_u32 XXH_read32(const void *memPtr) {
 
-#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))
+  return *(const xxh_u32 *)memPtr;
 
-/*
- * __attribute__((aligned(1))) is supported by gcc and clang. Originally the
- * documentation claimed that it only increased the alignment, but actually it
- * can decrease it on gcc, clang, and icc:
- * https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69502,
- * https://gcc.godbolt.org/z/xYez1j67Y.
- */
-#ifdef XXH_OLD_NAMES
-typedef union { xxh_u32 u32; } __attribute__((packed)) unalign;
-#endif
-static xxh_u32 XXH_read32(const void* ptr)
-{
-    typedef __attribute__((aligned(1))) xxh_u32 xxh_unalign32;
-    return *((const xxh_unalign32*)ptr);
 }
 
-#else
+  #elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS == 1))
 
-/*
- * Portable and safe solution. Generally efficient.
- * see: https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html
- */
-static xxh_u32 XXH_read32(const void* memPtr)
-{
-    xxh_u32 val;
-    XXH_memcpy(&val, memPtr, sizeof(val));
-    return val;
-}
+    /*
+     * __attribute__((aligned(1))) is supported by gcc and clang. Originally the
+     * documentation claimed that it only increased the alignment, but actually
+     * it can decrease it on gcc, clang, and icc:
+     * https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69502,
+     * https://gcc.godbolt.org/z/xYez1j67Y.
+     */
+    #ifdef XXH_OLD_NAMES
+typedef union {
 
-#endif   /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
+  xxh_u32 u32;
 
+} __attribute__((packed)) unalign;
 
-/* ***   Endianness   *** */
+    #endif
+static xxh_u32 XXH_read32(const void *ptr) {
+
+  typedef __attribute__((aligned(1))) xxh_u32 xxh_unalign32;
+  return *((const xxh_unalign32 *)ptr);
+
+}
+
+  #else
 
-/*!
- * @ingroup tuning
- * @def XXH_CPU_LITTLE_ENDIAN
- * @brief Whether the target is little endian.
- *
- * Defined to 1 if the target is little endian, or 0 if it is big endian.
- * It can be defined externally, for example on the compiler command line.
- *
- * If it is not defined,
- * a runtime check (which is usually constant folded) is used instead.
- *
- * @note
- *   This is not necessarily defined to an integer constant.
- *
- * @see XXH_isLittleEndian() for the runtime check.
- */
-#ifndef XXH_CPU_LITTLE_ENDIAN
 /*
- * Try to detect endianness automatically, to avoid the nonstandard behavior
- * in `XXH_isLittleEndian()`
- */
-#  if defined(_WIN32) /* Windows is always little endian */ \
-     || defined(__LITTLE_ENDIAN__) \
-     || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
-#    define XXH_CPU_LITTLE_ENDIAN 1
-#  elif defined(__BIG_ENDIAN__) \
-     || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
-#    define XXH_CPU_LITTLE_ENDIAN 0
-#  else
+ * Portable and safe solution. Generally efficient.
+ * see:
+ * https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html
+ */
+static xxh_u32 XXH_read32(const void *memPtr) {
+
+  xxh_u32 val;
+  XXH_memcpy(&val, memPtr, sizeof(val));
+  return val;
+
+}
+
+  #endif                                  /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
+
+  /* ***   Endianness   *** */
+
+  /*!
+   * @ingroup tuning
+   * @def XXH_CPU_LITTLE_ENDIAN
+   * @brief Whether the target is little endian.
+   *
+   * Defined to 1 if the target is little endian, or 0 if it is big endian.
+   * It can be defined externally, for example on the compiler command line.
+   *
+   * If it is not defined,
+   * a runtime check (which is usually constant folded) is used instead.
+   *
+   * @note
+   *   This is not necessarily defined to an integer constant.
+   *
+   * @see XXH_isLittleEndian() for the runtime check.
+   */
+  #ifndef XXH_CPU_LITTLE_ENDIAN
+    /*
+     * Try to detect endianness automatically, to avoid the nonstandard behavior
+     * in `XXH_isLittleEndian()`
+     */
+    #if defined(_WIN32) /* Windows is always little endian */ \
+        || defined(__LITTLE_ENDIAN__) ||                      \
+        (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
+      #define XXH_CPU_LITTLE_ENDIAN 1
+    #elif defined(__BIG_ENDIAN__) || \
+        (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
+      #define XXH_CPU_LITTLE_ENDIAN 0
+    #else
 /*!
  * @internal
  * @brief Runtime check for @ref XXH_CPU_LITTLE_ENDIAN.
  *
  * Most compilers will constant fold this.
  */
-static int XXH_isLittleEndian(void)
-{
-    /*
-     * Portable and well-defined behavior.
-     * Don't use static: it is detrimental to performance.
-     */
-    const union { xxh_u32 u; xxh_u8 c[4]; } one = { 1 };
-    return one.c[0];
-}
-#   define XXH_CPU_LITTLE_ENDIAN   XXH_isLittleEndian()
-#  endif
-#endif
+static int XXH_isLittleEndian(void) {
 
+  /*
+   * Portable and well-defined behavior.
+   * Don't use static: it is detrimental to performance.
+   */
+  const union {
 
+    xxh_u32 u;
+    xxh_u8  c[4];
 
+  } one = {1};
 
-/* ****************************************
-*  Compiler-specific Functions and Macros
-******************************************/
-#define XXH_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
+  return one.c[0];
 
-#ifdef __has_builtin
-#  define XXH_HAS_BUILTIN(x) __has_builtin(x)
-#else
-#  define XXH_HAS_BUILTIN(x) 0
-#endif
+}
+\
+      #define XXH_CPU_LITTLE_ENDIAN XXH_isLittleEndian()
+    #endif
+  #endif
 
+  /* ****************************************
+   *  Compiler-specific Functions and Macros
+   ******************************************/
+  #define XXH_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
 
+  #ifdef __has_builtin
+    #define XXH_HAS_BUILTIN(x) __has_builtin(x)
+  #else
+    #define XXH_HAS_BUILTIN(x) 0
+  #endif
 
 /*
  * C23 and future versions have standard "unreachable()".
@@ -2685,142 +2893,154 @@ static int XXH_isLittleEndian(void)
  * doesn't work on GCC12
  */
 
-#if XXH_HAS_BUILTIN(__builtin_unreachable)
-#  define XXH_UNREACHABLE() __builtin_unreachable()
-
-#elif defined(_MSC_VER)
-#  define XXH_UNREACHABLE() __assume(0)
-
-#else
-#  define XXH_UNREACHABLE()
-#endif
-
-#if XXH_HAS_BUILTIN(__builtin_assume)
-#  define XXH_ASSUME(c) __builtin_assume(c)
-#else
-#  define XXH_ASSUME(c) if (!(c)) { XXH_UNREACHABLE(); }
-#endif
-
-/*!
- * @internal
- * @def XXH_rotl32(x,r)
- * @brief 32-bit rotate left.
- *
- * @param x The 32-bit integer to be rotated.
- * @param r The number of bits to rotate.
- * @pre
- *   @p r > 0 && @p r < 32
- * @note
- *   @p x and @p r may be evaluated multiple times.
- * @return The rotated result.
- */
-#if !defined(NO_CLANG_BUILTIN) && XXH_HAS_BUILTIN(__builtin_rotateleft32) \
-                               && XXH_HAS_BUILTIN(__builtin_rotateleft64)
-#  define XXH_rotl32 __builtin_rotateleft32
-#  define XXH_rotl64 __builtin_rotateleft64
-/* Note: although _rotl exists for minGW (GCC under windows), performance seems poor */
-#elif defined(_MSC_VER)
-#  define XXH_rotl32(x,r) _rotl(x,r)
-#  define XXH_rotl64(x,r) _rotl64(x,r)
-#else
-#  define XXH_rotl32(x,r) (((x) << (r)) | ((x) >> (32 - (r))))
-#  define XXH_rotl64(x,r) (((x) << (r)) | ((x) >> (64 - (r))))
-#endif
-
-/*!
- * @internal
- * @fn xxh_u32 XXH_swap32(xxh_u32 x)
- * @brief A 32-bit byteswap.
- *
- * @param x The 32-bit integer to byteswap.
- * @return @p x, byteswapped.
- */
-#if defined(_MSC_VER)     /* Visual Studio */
-#  define XXH_swap32 _byteswap_ulong
-#elif XXH_GCC_VERSION >= 403
-#  define XXH_swap32 __builtin_bswap32
-#else
-static xxh_u32 XXH_swap32 (xxh_u32 x)
-{
-    return  ((x << 24) & 0xff000000 ) |
-            ((x <<  8) & 0x00ff0000 ) |
-            ((x >>  8) & 0x0000ff00 ) |
-            ((x >> 24) & 0x000000ff );
-}
-#endif
-
+  #if XXH_HAS_BUILTIN(__builtin_unreachable)
+    #define XXH_UNREACHABLE() __builtin_unreachable()
+
+  #elif defined(_MSC_VER)
+    #define XXH_UNREACHABLE() __assume(0)
+
+  #else
+    #define XXH_UNREACHABLE()
+  #endif
+
+  #if XXH_HAS_BUILTIN(__builtin_assume)
+    #define XXH_ASSUME(c) __builtin_assume(c)
+  #else
+    #define XXH_ASSUME(c) \
+      if (!(c)) { XXH_UNREACHABLE(); }
+  #endif
+
+  /*!
+   * @internal
+   * @def XXH_rotl32(x,r)
+   * @brief 32-bit rotate left.
+   *
+   * @param x The 32-bit integer to be rotated.
+   * @param r The number of bits to rotate.
+   * @pre
+   *   @p r > 0 && @p r < 32
+   * @note
+   *   @p x and @p r may be evaluated multiple times.
+   * @return The rotated result.
+   */
+  #if !defined(NO_CLANG_BUILTIN) && XXH_HAS_BUILTIN(__builtin_rotateleft32) && \
+      XXH_HAS_BUILTIN(__builtin_rotateleft64)
+    #define XXH_rotl32 __builtin_rotateleft32
+    #define XXH_rotl64 __builtin_rotateleft64
+  /* Note: although _rotl exists for minGW (GCC under windows), performance
+   * seems poor */
+  #elif defined(_MSC_VER)
+    #define XXH_rotl32(x, r) _rotl(x, r)
+    #define XXH_rotl64(x, r) _rotl64(x, r)
+  #else
+    #define XXH_rotl32(x, r) (((x) << (r)) | ((x) >> (32 - (r))))
+    #define XXH_rotl64(x, r) (((x) << (r)) | ((x) >> (64 - (r))))
+  #endif
+
+  /*!
+   * @internal
+   * @fn xxh_u32 XXH_swap32(xxh_u32 x)
+   * @brief A 32-bit byteswap.
+   *
+   * @param x The 32-bit integer to byteswap.
+   * @return @p x, byteswapped.
+   */
+  #if defined(_MSC_VER)                                    /* Visual Studio */
+    #define XXH_swap32 _byteswap_ulong
+  #elif XXH_GCC_VERSION >= 403
+    #define XXH_swap32 __builtin_bswap32
+  #else
+static xxh_u32 XXH_swap32(xxh_u32 x) {
+
+  return ((x << 24) & 0xff000000) | ((x << 8) & 0x00ff0000) |
+         ((x >> 8) & 0x0000ff00) | ((x >> 24) & 0x000000ff);
+
+}
+
+  #endif
 
 /* ***************************
-*  Memory reads
-*****************************/
+ *  Memory reads
+ *****************************/
 
 /*!
  * @internal
  * @brief Enum to indicate whether a pointer is aligned.
  */
 typedef enum {
-    XXH_aligned,  /*!< Aligned */
-    XXH_unaligned /*!< Possibly unaligned */
+
+  XXH_aligned,                                                 /*!< Aligned */
+  XXH_unaligned                                     /*!< Possibly unaligned */
+
 } XXH_alignment;
 
-/*
- * XXH_FORCE_MEMORY_ACCESS==3 is an endian-independent byteshift load.
- *
- * This is ideal for older compilers which don't inline memcpy.
- */
-#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
+  /*
+   * XXH_FORCE_MEMORY_ACCESS==3 is an endian-independent byteshift load.
+   *
+   * This is ideal for older compilers which don't inline memcpy.
+   */
+  #if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS == 3))
 
-XXH_FORCE_INLINE xxh_u32 XXH_readLE32(const void* memPtr)
-{
-    const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
-    return bytePtr[0]
-         | ((xxh_u32)bytePtr[1] << 8)
-         | ((xxh_u32)bytePtr[2] << 16)
-         | ((xxh_u32)bytePtr[3] << 24);
-}
+XXH_FORCE_INLINE xxh_u32 XXH_readLE32(const void *memPtr) {
+
+  const xxh_u8 *bytePtr = (const xxh_u8 *)memPtr;
+  return bytePtr[0] | ((xxh_u32)bytePtr[1] << 8) | ((xxh_u32)bytePtr[2] << 16) |
+         ((xxh_u32)bytePtr[3] << 24);
 
-XXH_FORCE_INLINE xxh_u32 XXH_readBE32(const void* memPtr)
-{
-    const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
-    return bytePtr[3]
-         | ((xxh_u32)bytePtr[2] << 8)
-         | ((xxh_u32)bytePtr[1] << 16)
-         | ((xxh_u32)bytePtr[0] << 24);
 }
 
-#else
-XXH_FORCE_INLINE xxh_u32 XXH_readLE32(const void* ptr)
-{
-    return XXH_CPU_LITTLE_ENDIAN ? XXH_read32(ptr) : XXH_swap32(XXH_read32(ptr));
+XXH_FORCE_INLINE xxh_u32 XXH_readBE32(const void *memPtr) {
+
+  const xxh_u8 *bytePtr = (const xxh_u8 *)memPtr;
+  return bytePtr[3] | ((xxh_u32)bytePtr[2] << 8) | ((xxh_u32)bytePtr[1] << 16) |
+         ((xxh_u32)bytePtr[0] << 24);
+
 }
 
-static xxh_u32 XXH_readBE32(const void* ptr)
-{
-    return XXH_CPU_LITTLE_ENDIAN ? XXH_swap32(XXH_read32(ptr)) : XXH_read32(ptr);
+  #else
+XXH_FORCE_INLINE xxh_u32 XXH_readLE32(const void *ptr) {
+
+  return XXH_CPU_LITTLE_ENDIAN ? XXH_read32(ptr) : XXH_swap32(XXH_read32(ptr));
+
 }
-#endif
 
-XXH_FORCE_INLINE xxh_u32
-XXH_readLE32_align(const void* ptr, XXH_alignment align)
-{
-    if (align==XXH_unaligned) {
-        return XXH_readLE32(ptr);
-    } else {
-        return XXH_CPU_LITTLE_ENDIAN ? *(const xxh_u32*)ptr : XXH_swap32(*(const xxh_u32*)ptr);
-    }
+static xxh_u32 XXH_readBE32(const void *ptr) {
+
+  return XXH_CPU_LITTLE_ENDIAN ? XXH_swap32(XXH_read32(ptr)) : XXH_read32(ptr);
+
 }
 
+  #endif
+
+XXH_FORCE_INLINE xxh_u32 XXH_readLE32_align(const void   *ptr,
+                                            XXH_alignment align) {
+
+  if (align == XXH_unaligned) {
+
+    return XXH_readLE32(ptr);
+
+  } else {
+
+    return XXH_CPU_LITTLE_ENDIAN ? *(const xxh_u32 *)ptr
+                                 : XXH_swap32(*(const xxh_u32 *)ptr);
+
+  }
+
+}
 
 /* *************************************
-*  Misc
-***************************************/
+ *  Misc
+ ***************************************/
 /*! @ingroup public */
-XXH_PUBLIC_API unsigned XXH_versionNumber (void) { return XXH_VERSION_NUMBER; }
+XXH_PUBLIC_API unsigned XXH_versionNumber(void) {
+
+  return XXH_VERSION_NUMBER;
 
+}
 
 /* *******************************************************************
-*  32-bit hash functions
-*********************************************************************/
+ *  32-bit hash functions
+ *********************************************************************/
 /*!
  * @}
  * @defgroup XXH32_impl XXH32 implementation
@@ -2828,21 +3048,22 @@ XXH_PUBLIC_API unsigned XXH_versionNumber (void) { return XXH_VERSION_NUMBER; }
  *
  * Details on the XXH32 implementation.
  * @{
+
  */
- /* #define instead of static const, to be used as initializers */
-#define XXH_PRIME32_1  0x9E3779B1U  /*!< 0b10011110001101110111100110110001 */
-#define XXH_PRIME32_2  0x85EBCA77U  /*!< 0b10000101111010111100101001110111 */
-#define XXH_PRIME32_3  0xC2B2AE3DU  /*!< 0b11000010101100101010111000111101 */
-#define XXH_PRIME32_4  0x27D4EB2FU  /*!< 0b00100111110101001110101100101111 */
-#define XXH_PRIME32_5  0x165667B1U  /*!< 0b00010110010101100110011110110001 */
-
-#ifdef XXH_OLD_NAMES
-#  define PRIME32_1 XXH_PRIME32_1
-#  define PRIME32_2 XXH_PRIME32_2
-#  define PRIME32_3 XXH_PRIME32_3
-#  define PRIME32_4 XXH_PRIME32_4
-#  define PRIME32_5 XXH_PRIME32_5
-#endif
+/* #define instead of static const, to be used as initializers */
+  #define XXH_PRIME32_1 0x9E3779B1U /*!< 0b10011110001101110111100110110001 */
+  #define XXH_PRIME32_2 0x85EBCA77U /*!< 0b10000101111010111100101001110111 */
+  #define XXH_PRIME32_3 0xC2B2AE3DU /*!< 0b11000010101100101010111000111101 */
+  #define XXH_PRIME32_4 0x27D4EB2FU /*!< 0b00100111110101001110101100101111 */
+  #define XXH_PRIME32_5 0x165667B1U /*!< 0b00010110010101100110011110110001 */
+
+  #ifdef XXH_OLD_NAMES
+    #define PRIME32_1 XXH_PRIME32_1
+    #define PRIME32_2 XXH_PRIME32_2
+    #define PRIME32_3 XXH_PRIME32_3
+    #define PRIME32_4 XXH_PRIME32_4
+    #define PRIME32_5 XXH_PRIME32_5
+  #endif
 
 /*!
  * @internal
@@ -2855,51 +3076,54 @@ XXH_PUBLIC_API unsigned XXH_versionNumber (void) { return XXH_VERSION_NUMBER; }
  * @param input The stripe of input to mix.
  * @return The mixed accumulator lane.
  */
-static xxh_u32 XXH32_round(xxh_u32 acc, xxh_u32 input)
-{
-    acc += input * XXH_PRIME32_2;
-    acc  = XXH_rotl32(acc, 13);
-    acc *= XXH_PRIME32_1;
-#if (defined(__SSE4_1__) || defined(__aarch64__) || defined(__wasm_simd128__)) && !defined(XXH_ENABLE_AUTOVECTORIZE)
-    /*
-     * UGLY HACK:
-     * A compiler fence is the only thing that prevents GCC and Clang from
-     * autovectorizing the XXH32 loop (pragmas and attributes don't work for some
-     * reason) without globally disabling SSE4.1.
-     *
-     * The reason we want to avoid vectorization is because despite working on
-     * 4 integers at a time, there are multiple factors slowing XXH32 down on
-     * SSE4:
-     * - There's a ridiculous amount of lag from pmulld (10 cycles of latency on
-     *   newer chips!) making it slightly slower to multiply four integers at
-     *   once compared to four integers independently. Even when pmulld was
-     *   fastest, Sandy/Ivy Bridge, it is still not worth it to go into SSE
-     *   just to multiply unless doing a long operation.
-     *
-     * - Four instructions are required to rotate,
-     *      movqda tmp,  v // not required with VEX encoding
-     *      pslld  tmp, 13 // tmp <<= 13
-     *      psrld  v,   19 // x >>= 19
-     *      por    v,  tmp // x |= tmp
-     *   compared to one for scalar:
-     *      roll   v, 13    // reliably fast across the board
-     *      shldl  v, v, 13 // Sandy Bridge and later prefer this for some reason
-     *
-     * - Instruction level parallelism is actually more beneficial here because
-     *   the SIMD actually serializes this operation: While v1 is rotating, v2
-     *   can load data, while v3 can multiply. SSE forces them to operate
-     *   together.
-     *
-     * This is also enabled on AArch64, as Clang is *very aggressive* in vectorizing
-     * the loop. NEON is only faster on the A53, and with the newer cores, it is less
-     * than half the speed.
-     *
-     * Additionally, this is used on WASM SIMD128 because it JITs to the same
-     * SIMD instructions and has the same issue.
-     */
-    XXH_COMPILER_GUARD(acc);
-#endif
-    return acc;
+static xxh_u32 XXH32_round(xxh_u32 acc, xxh_u32 input) {
+
+  acc += input * XXH_PRIME32_2;
+  acc = XXH_rotl32(acc, 13);
+  acc *= XXH_PRIME32_1;
+  #if (defined(__SSE4_1__) || defined(__aarch64__) || \
+       defined(__wasm_simd128__)) &&                  \
+      !defined(XXH_ENABLE_AUTOVECTORIZE)
+  /*
+   * UGLY HACK:
+   * A compiler fence is the only thing that prevents GCC and Clang from
+   * autovectorizing the XXH32 loop (pragmas and attributes don't work for some
+   * reason) without globally disabling SSE4.1.
+   *
+   * The reason we want to avoid vectorization is because despite working on
+   * 4 integers at a time, there are multiple factors slowing XXH32 down on
+   * SSE4:
+   * - There's a ridiculous amount of lag from pmulld (10 cycles of latency on
+   *   newer chips!) making it slightly slower to multiply four integers at
+   *   once compared to four integers independently. Even when pmulld was
+   *   fastest, Sandy/Ivy Bridge, it is still not worth it to go into SSE
+   *   just to multiply unless doing a long operation.
+   *
+   * - Four instructions are required to rotate,
+   *      movqda tmp,  v // not required with VEX encoding
+   *      pslld  tmp, 13 // tmp <<= 13
+   *      psrld  v,   19 // x >>= 19
+   *      por    v,  tmp // x |= tmp
+   *   compared to one for scalar:
+   *      roll   v, 13    // reliably fast across the board
+   *      shldl  v, v, 13 // Sandy Bridge and later prefer this for some reason
+   *
+   * - Instruction level parallelism is actually more beneficial here because
+   *   the SIMD actually serializes this operation: While v1 is rotating, v2
+   *   can load data, while v3 can multiply. SSE forces them to operate
+   *   together.
+   *
+   * This is also enabled on AArch64, as Clang is *very aggressive* in
+   * vectorizing the loop. NEON is only faster on the A53, and with the newer
+   * cores, it is less than half the speed.
+   *
+   * Additionally, this is used on WASM SIMD128 because it JITs to the same
+   * SIMD instructions and has the same issue.
+   */
+  XXH_COMPILER_GUARD(acc);
+  #endif
+  return acc;
+
 }
 
 /*!
@@ -2912,17 +3136,18 @@ static xxh_u32 XXH32_round(xxh_u32 acc, xxh_u32 input)
  * @param hash The hash to avalanche.
  * @return The avalanched hash.
  */
-static xxh_u32 XXH32_avalanche(xxh_u32 hash)
-{
-    hash ^= hash >> 15;
-    hash *= XXH_PRIME32_2;
-    hash ^= hash >> 13;
-    hash *= XXH_PRIME32_3;
-    hash ^= hash >> 16;
-    return hash;
+static xxh_u32 XXH32_avalanche(xxh_u32 hash) {
+
+  hash ^= hash >> 15;
+  hash *= XXH_PRIME32_2;
+  hash ^= hash >> 13;
+  hash *= XXH_PRIME32_3;
+  hash ^= hash >> 16;
+  return hash;
+
 }
 
-#define XXH_get32bits(p) XXH_readLE32_align(p, align)
+  #define XXH_get32bits(p) XXH_readLE32_align(p, align)
 
 /*!
  * @internal
@@ -2939,86 +3164,122 @@ static xxh_u32 XXH32_avalanche(xxh_u32 hash)
  * @return The finalized hash.
  * @see XXH64_finalize().
  */
-static XXH_PUREF xxh_u32
-XXH32_finalize(xxh_u32 hash, const xxh_u8* ptr, size_t len, XXH_alignment align)
-{
-#define XXH_PROCESS1 do {                             \
-    hash += (*ptr++) * XXH_PRIME32_5;                 \
-    hash = XXH_rotl32(hash, 11) * XXH_PRIME32_1;      \
-} while (0)
-
-#define XXH_PROCESS4 do {                             \
-    hash += XXH_get32bits(ptr) * XXH_PRIME32_3;       \
-    ptr += 4;                                         \
-    hash  = XXH_rotl32(hash, 17) * XXH_PRIME32_4;     \
-} while (0)
-
-    if (ptr==NULL) XXH_ASSERT(len == 0);
-
-    /* Compact rerolled version; generally faster */
-    if (!XXH32_ENDJMP) {
-        len &= 15;
-        while (len >= 4) {
-            XXH_PROCESS4;
-            len -= 4;
-        }
-        while (len > 0) {
-            XXH_PROCESS1;
-            --len;
-        }
+static XXH_PUREF xxh_u32 XXH32_finalize(xxh_u32 hash, const xxh_u8 *ptr,
+                                        size_t len, XXH_alignment align) {
+\
+  #define XXH_PROCESS1                             \
+    do {                                           \
+                                                   \
+      hash += (*ptr++) * XXH_PRIME32_5;            \
+      hash = XXH_rotl32(hash, 11) * XXH_PRIME32_1; \
+                                                   \
+    } while (0)
+
+  #define XXH_PROCESS4                             \
+    do {                                           \
+                                                   \
+      hash += XXH_get32bits(ptr) * XXH_PRIME32_3;  \
+      ptr += 4;                                    \
+      hash = XXH_rotl32(hash, 17) * XXH_PRIME32_4; \
+                                                   \
+    } while (0)
+
+  if (ptr == NULL) XXH_ASSERT(len == 0);
+
+  /* Compact rerolled version; generally faster */
+  if (!XXH32_ENDJMP) {
+
+    len &= 15;
+    while (len >= 4) {
+
+      XXH_PROCESS4;
+      len -= 4;
+
+    }
+
+    while (len > 0) {
+
+      XXH_PROCESS1;
+      --len;
+
+    }
+
+    return XXH32_avalanche(hash);
+
+  } else {
+
+    switch (len & 15) /* or switch(bEnd - p) */ {
+
+      case 12:
+        XXH_PROCESS4;
+        XXH_FALLTHROUGH;                                     /* fallthrough */
+      case 8:
+        XXH_PROCESS4;
+        XXH_FALLTHROUGH;                                     /* fallthrough */
+      case 4:
+        XXH_PROCESS4;
         return XXH32_avalanche(hash);
-    } else {
-         switch(len&15) /* or switch(bEnd - p) */ {
-           case 12:      XXH_PROCESS4;
-                         XXH_FALLTHROUGH;  /* fallthrough */
-           case 8:       XXH_PROCESS4;
-                         XXH_FALLTHROUGH;  /* fallthrough */
-           case 4:       XXH_PROCESS4;
-                         return XXH32_avalanche(hash);
-
-           case 13:      XXH_PROCESS4;
-                         XXH_FALLTHROUGH;  /* fallthrough */
-           case 9:       XXH_PROCESS4;
-                         XXH_FALLTHROUGH;  /* fallthrough */
-           case 5:       XXH_PROCESS4;
-                         XXH_PROCESS1;
-                         return XXH32_avalanche(hash);
-
-           case 14:      XXH_PROCESS4;
-                         XXH_FALLTHROUGH;  /* fallthrough */
-           case 10:      XXH_PROCESS4;
-                         XXH_FALLTHROUGH;  /* fallthrough */
-           case 6:       XXH_PROCESS4;
-                         XXH_PROCESS1;
-                         XXH_PROCESS1;
-                         return XXH32_avalanche(hash);
-
-           case 15:      XXH_PROCESS4;
-                         XXH_FALLTHROUGH;  /* fallthrough */
-           case 11:      XXH_PROCESS4;
-                         XXH_FALLTHROUGH;  /* fallthrough */
-           case 7:       XXH_PROCESS4;
-                         XXH_FALLTHROUGH;  /* fallthrough */
-           case 3:       XXH_PROCESS1;
-                         XXH_FALLTHROUGH;  /* fallthrough */
-           case 2:       XXH_PROCESS1;
-                         XXH_FALLTHROUGH;  /* fallthrough */
-           case 1:       XXH_PROCESS1;
-                         XXH_FALLTHROUGH;  /* fallthrough */
-           case 0:       return XXH32_avalanche(hash);
-        }
-        XXH_ASSERT(0);
-        return hash;   /* reaching this point is deemed impossible */
+
+      case 13:
+        XXH_PROCESS4;
+        XXH_FALLTHROUGH;                                     /* fallthrough */
+      case 9:
+        XXH_PROCESS4;
+        XXH_FALLTHROUGH;                                     /* fallthrough */
+      case 5:
+        XXH_PROCESS4;
+        XXH_PROCESS1;
+        return XXH32_avalanche(hash);
+
+      case 14:
+        XXH_PROCESS4;
+        XXH_FALLTHROUGH;                                     /* fallthrough */
+      case 10:
+        XXH_PROCESS4;
+        XXH_FALLTHROUGH;                                     /* fallthrough */
+      case 6:
+        XXH_PROCESS4;
+        XXH_PROCESS1;
+        XXH_PROCESS1;
+        return XXH32_avalanche(hash);
+
+      case 15:
+        XXH_PROCESS4;
+        XXH_FALLTHROUGH;                                     /* fallthrough */
+      case 11:
+        XXH_PROCESS4;
+        XXH_FALLTHROUGH;                                     /* fallthrough */
+      case 7:
+        XXH_PROCESS4;
+        XXH_FALLTHROUGH;                                     /* fallthrough */
+      case 3:
+        XXH_PROCESS1;
+        XXH_FALLTHROUGH;                                     /* fallthrough */
+      case 2:
+        XXH_PROCESS1;
+        XXH_FALLTHROUGH;                                     /* fallthrough */
+      case 1:
+        XXH_PROCESS1;
+        XXH_FALLTHROUGH;                                     /* fallthrough */
+      case 0:
+        return XXH32_avalanche(hash);
+
     }
+
+    XXH_ASSERT(0);
+    return hash;                /* reaching this point is deemed impossible */
+
+  }
+
 }
 
-#ifdef XXH_OLD_NAMES
-#  define PROCESS1 XXH_PROCESS1
-#  define PROCESS4 XXH_PROCESS4
-#else
-#  undef XXH_PROCESS1
-#  undef XXH_PROCESS4
-#endif
+  #ifdef XXH_OLD_NAMES
+    #define PROCESS1 XXH_PROCESS1
+    #define PROCESS4 XXH_PROCESS4
+  #else
+    #undef XXH_PROCESS1
+    #undef XXH_PROCESS4
+  #endif
 
 /*!
  * @internal
@@ -3028,372 +3289,459 @@ XXH32_finalize(xxh_u32 hash, const xxh_u8* ptr, size_t len, XXH_alignment align)
  * @param align Whether @p input is aligned.
  * @return The calculated hash.
  */
-XXH_FORCE_INLINE XXH_PUREF xxh_u32
-XXH32_endian_align(const xxh_u8* input, size_t len, xxh_u32 seed, XXH_alignment align)
-{
-    xxh_u32 h32;
-
-    if (input==NULL) XXH_ASSERT(len == 0);
-
-    if (len>=16) {
-        const xxh_u8* const bEnd = input + len;
-        const xxh_u8* const limit = bEnd - 15;
-        xxh_u32 v1 = seed + XXH_PRIME32_1 + XXH_PRIME32_2;
-        xxh_u32 v2 = seed + XXH_PRIME32_2;
-        xxh_u32 v3 = seed + 0;
-        xxh_u32 v4 = seed - XXH_PRIME32_1;
-
-        do {
-            v1 = XXH32_round(v1, XXH_get32bits(input)); input += 4;
-            v2 = XXH32_round(v2, XXH_get32bits(input)); input += 4;
-            v3 = XXH32_round(v3, XXH_get32bits(input)); input += 4;
-            v4 = XXH32_round(v4, XXH_get32bits(input)); input += 4;
-        } while (input < limit);
-
-        h32 = XXH_rotl32(v1, 1)  + XXH_rotl32(v2, 7)
-            + XXH_rotl32(v3, 12) + XXH_rotl32(v4, 18);
-    } else {
-        h32  = seed + XXH_PRIME32_5;
-    }
+XXH_FORCE_INLINE XXH_PUREF xxh_u32 XXH32_endian_align(const xxh_u8 *input,
+                                                      size_t len, xxh_u32 seed,
+                                                      XXH_alignment align) {
 
-    h32 += (xxh_u32)len;
+  xxh_u32 h32;
 
-    return XXH32_finalize(h32, input, len&15, align);
-}
+  if (input == NULL) XXH_ASSERT(len == 0);
 
-/*! @ingroup XXH32_family */
-XXH_PUBLIC_API XXH32_hash_t XXH32 (const void* input, size_t len, XXH32_hash_t seed)
-{
-#if !defined(XXH_NO_STREAM) && XXH_SIZE_OPT >= 2
-    /* Simple version, good for code maintenance, but unfortunately slow for small inputs */
-    XXH32_state_t state;
-    XXH32_reset(&state, seed);
-    XXH32_update(&state, (const xxh_u8*)input, len);
-    return XXH32_digest(&state);
-#else
-    if (XXH_FORCE_ALIGN_CHECK) {
-        if ((((size_t)input) & 3) == 0) {   /* Input is 4-bytes aligned, leverage the speed benefit */
-            return XXH32_endian_align((const xxh_u8*)input, len, seed, XXH_aligned);
-    }   }
-
-    return XXH32_endian_align((const xxh_u8*)input, len, seed, XXH_unaligned);
-#endif
-}
+  if (len >= 16) {
 
+    const xxh_u8 *const bEnd = input + len;
+    const xxh_u8 *const limit = bEnd - 15;
+    xxh_u32             v1 = seed + XXH_PRIME32_1 + XXH_PRIME32_2;
+    xxh_u32             v2 = seed + XXH_PRIME32_2;
+    xxh_u32             v3 = seed + 0;
+    xxh_u32             v4 = seed - XXH_PRIME32_1;
 
+    do {
 
-/*******   Hash streaming   *******/
-#ifndef XXH_NO_STREAM
-/*! @ingroup XXH32_family */
-XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void)
-{
-    return (XXH32_state_t*)XXH_malloc(sizeof(XXH32_state_t));
-}
-/*! @ingroup XXH32_family */
-XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr)
-{
-    XXH_free(statePtr);
-    return XXH_OK;
-}
+      v1 = XXH32_round(v1, XXH_get32bits(input));
+      input += 4;
+      v2 = XXH32_round(v2, XXH_get32bits(input));
+      input += 4;
+      v3 = XXH32_round(v3, XXH_get32bits(input));
+      input += 4;
+      v4 = XXH32_round(v4, XXH_get32bits(input));
+      input += 4;
 
-/*! @ingroup XXH32_family */
-XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dstState, const XXH32_state_t* srcState)
-{
-    XXH_memcpy(dstState, srcState, sizeof(*dstState));
-}
+    } while (input < limit);
 
-/*! @ingroup XXH32_family */
-XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t* statePtr, XXH32_hash_t seed)
-{
-    XXH_ASSERT(statePtr != NULL);
-    memset(statePtr, 0, sizeof(*statePtr));
-    statePtr->v[0] = seed + XXH_PRIME32_1 + XXH_PRIME32_2;
-    statePtr->v[1] = seed + XXH_PRIME32_2;
-    statePtr->v[2] = seed + 0;
-    statePtr->v[3] = seed - XXH_PRIME32_1;
-    return XXH_OK;
-}
+    h32 = XXH_rotl32(v1, 1) + XXH_rotl32(v2, 7) + XXH_rotl32(v3, 12) +
+          XXH_rotl32(v4, 18);
 
+  } else {
 
-/*! @ingroup XXH32_family */
-XXH_PUBLIC_API XXH_errorcode
-XXH32_update(XXH32_state_t* state, const void* input, size_t len)
-{
-    if (input==NULL) {
-        XXH_ASSERT(len == 0);
-        return XXH_OK;
-    }
+    h32 = seed + XXH_PRIME32_5;
 
-    {   const xxh_u8* p = (const xxh_u8*)input;
-        const xxh_u8* const bEnd = p + len;
+  }
 
-        state->total_len_32 += (XXH32_hash_t)len;
-        state->large_len |= (XXH32_hash_t)((len>=16) | (state->total_len_32>=16));
+  h32 += (xxh_u32)len;
 
-        if (state->memsize + len < 16)  {   /* fill in tmp buffer */
-            XXH_memcpy((xxh_u8*)(state->mem32) + state->memsize, input, len);
-            state->memsize += (XXH32_hash_t)len;
-            return XXH_OK;
-        }
+  return XXH32_finalize(h32, input, len & 15, align);
 
-        if (state->memsize) {   /* some data left from previous update */
-            XXH_memcpy((xxh_u8*)(state->mem32) + state->memsize, input, 16-state->memsize);
-            {   const xxh_u32* p32 = state->mem32;
-                state->v[0] = XXH32_round(state->v[0], XXH_readLE32(p32)); p32++;
-                state->v[1] = XXH32_round(state->v[1], XXH_readLE32(p32)); p32++;
-                state->v[2] = XXH32_round(state->v[2], XXH_readLE32(p32)); p32++;
-                state->v[3] = XXH32_round(state->v[3], XXH_readLE32(p32));
-            }
-            p += 16-state->memsize;
-            state->memsize = 0;
-        }
+}
 
-        if (p <= bEnd-16) {
-            const xxh_u8* const limit = bEnd - 16;
+/*! @ingroup XXH32_family */
+XXH_PUBLIC_API XXH32_hash_t XXH32(const void *input, size_t len,
+                                  XXH32_hash_t seed) {
+
+  #if !defined(XXH_NO_STREAM) && XXH_SIZE_OPT >= 2
+  /* Simple version, good for code maintenance, but unfortunately slow for small
+   * inputs */
+  XXH32_state_t state;
+  XXH32_reset(&state, seed);
+  XXH32_update(&state, (const xxh_u8 *)input, len);
+  return XXH32_digest(&state);
+  #else
+  if (XXH_FORCE_ALIGN_CHECK) {
+
+    if ((((size_t)input) & 3) ==
+        0) {        /* Input is 4-bytes aligned, leverage the speed benefit */
+      return XXH32_endian_align((const xxh_u8 *)input, len, seed, XXH_aligned);
 
-            do {
-                state->v[0] = XXH32_round(state->v[0], XXH_readLE32(p)); p+=4;
-                state->v[1] = XXH32_round(state->v[1], XXH_readLE32(p)); p+=4;
-                state->v[2] = XXH32_round(state->v[2], XXH_readLE32(p)); p+=4;
-                state->v[3] = XXH32_round(state->v[3], XXH_readLE32(p)); p+=4;
-            } while (p<=limit);
+    }
 
-        }
+  }
 
-        if (p < bEnd) {
-            XXH_memcpy(state->mem32, p, (size_t)(bEnd-p));
-            state->memsize = (unsigned)(bEnd-p);
-        }
-    }
+  return XXH32_endian_align((const xxh_u8 *)input, len, seed, XXH_unaligned);
+  #endif
 
-    return XXH_OK;
 }
 
+  /*******   Hash streaming   *******/
+  #ifndef XXH_NO_STREAM
+/*! @ingroup XXH32_family */
+XXH_PUBLIC_API XXH32_state_t *XXH32_createState(void) {
+
+  return (XXH32_state_t *)XXH_malloc(sizeof(XXH32_state_t));
+
+}
 
 /*! @ingroup XXH32_family */
-XXH_PUBLIC_API XXH32_hash_t XXH32_digest(const XXH32_state_t* state)
-{
-    xxh_u32 h32;
+XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t *statePtr) {
+
+  XXH_free(statePtr);
+  return XXH_OK;
+
+}
+
+/*! @ingroup XXH32_family */
+XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t       *dstState,
+                                    const XXH32_state_t *srcState) {
+
+  XXH_memcpy(dstState, srcState, sizeof(*dstState));
+
+}
+
+/*! @ingroup XXH32_family */
+XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t *statePtr,
+                                         XXH32_hash_t   seed) {
+
+  XXH_ASSERT(statePtr != NULL);
+  memset(statePtr, 0, sizeof(*statePtr));
+  statePtr->v[0] = seed + XXH_PRIME32_1 + XXH_PRIME32_2;
+  statePtr->v[1] = seed + XXH_PRIME32_2;
+  statePtr->v[2] = seed + 0;
+  statePtr->v[3] = seed - XXH_PRIME32_1;
+  return XXH_OK;
+
+}
+
+/*! @ingroup XXH32_family */
+XXH_PUBLIC_API XXH_errorcode XXH32_update(XXH32_state_t *state,
+                                          const void *input, size_t len) {
+
+  if (input == NULL) {
+
+    XXH_ASSERT(len == 0);
+    return XXH_OK;
+
+  }
+
+  {
+
+    const xxh_u8       *p = (const xxh_u8 *)input;
+    const xxh_u8 *const bEnd = p + len;
+
+    state->total_len_32 += (XXH32_hash_t)len;
+    state->large_len |=
+        (XXH32_hash_t)((len >= 16) | (state->total_len_32 >= 16));
+
+    if (state->memsize + len < 16) {                  /* fill in tmp buffer */
+      XXH_memcpy((xxh_u8 *)(state->mem32) + state->memsize, input, len);
+      state->memsize += (XXH32_hash_t)len;
+      return XXH_OK;
+
+    }
+
+    if (state->memsize) {            /* some data left from previous update */
+      XXH_memcpy((xxh_u8 *)(state->mem32) + state->memsize, input,
+                 16 - state->memsize);
+      {
+
+        const xxh_u32 *p32 = state->mem32;
+        state->v[0] = XXH32_round(state->v[0], XXH_readLE32(p32));
+        p32++;
+        state->v[1] = XXH32_round(state->v[1], XXH_readLE32(p32));
+        p32++;
+        state->v[2] = XXH32_round(state->v[2], XXH_readLE32(p32));
+        p32++;
+        state->v[3] = XXH32_round(state->v[3], XXH_readLE32(p32));
+
+      }
+
+      p += 16 - state->memsize;
+      state->memsize = 0;
 
-    if (state->large_len) {
-        h32 = XXH_rotl32(state->v[0], 1)
-            + XXH_rotl32(state->v[1], 7)
-            + XXH_rotl32(state->v[2], 12)
-            + XXH_rotl32(state->v[3], 18);
-    } else {
-        h32 = state->v[2] /* == seed */ + XXH_PRIME32_5;
     }
 
-    h32 += state->total_len_32;
+    if (p <= bEnd - 16) {
+
+      const xxh_u8 *const limit = bEnd - 16;
+
+      do {
+
+        state->v[0] = XXH32_round(state->v[0], XXH_readLE32(p));
+        p += 4;
+        state->v[1] = XXH32_round(state->v[1], XXH_readLE32(p));
+        p += 4;
+        state->v[2] = XXH32_round(state->v[2], XXH_readLE32(p));
+        p += 4;
+        state->v[3] = XXH32_round(state->v[3], XXH_readLE32(p));
+        p += 4;
+
+      } while (p <= limit);
+
+    }
+
+    if (p < bEnd) {
+
+      XXH_memcpy(state->mem32, p, (size_t)(bEnd - p));
+      state->memsize = (unsigned)(bEnd - p);
+
+    }
+
+  }
+
+  return XXH_OK;
+
+}
+
+/*! @ingroup XXH32_family */
+XXH_PUBLIC_API XXH32_hash_t XXH32_digest(const XXH32_state_t *state) {
+
+  xxh_u32 h32;
+
+  if (state->large_len) {
+
+    h32 = XXH_rotl32(state->v[0], 1) + XXH_rotl32(state->v[1], 7) +
+          XXH_rotl32(state->v[2], 12) + XXH_rotl32(state->v[3], 18);
+
+  } else {
+
+    h32 = state->v[2] /* == seed */ + XXH_PRIME32_5;
+
+  }
+
+  h32 += state->total_len_32;
+
+  return XXH32_finalize(h32, (const xxh_u8 *)state->mem32, state->memsize,
+                        XXH_aligned);
 
-    return XXH32_finalize(h32, (const xxh_u8*)state->mem32, state->memsize, XXH_aligned);
 }
-#endif /* !XXH_NO_STREAM */
+
+  #endif                                                  /* !XXH_NO_STREAM */
 
 /*******   Canonical representation   *******/
 
 /*! @ingroup XXH32_family */
-XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash)
-{
-    XXH_STATIC_ASSERT(sizeof(XXH32_canonical_t) == sizeof(XXH32_hash_t));
-    if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap32(hash);
-    XXH_memcpy(dst, &hash, sizeof(*dst));
+XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t *dst,
+                                            XXH32_hash_t       hash) {
+
+  XXH_STATIC_ASSERT(sizeof(XXH32_canonical_t) == sizeof(XXH32_hash_t));
+  if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap32(hash);
+  XXH_memcpy(dst, &hash, sizeof(*dst));
+
 }
+
 /*! @ingroup XXH32_family */
-XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src)
-{
-    return XXH_readBE32(src);
-}
+XXH_PUBLIC_API XXH32_hash_t
+XXH32_hashFromCanonical(const XXH32_canonical_t *src) {
+
+  return XXH_readBE32(src);
 
+}
 
-#ifndef XXH_NO_LONG_LONG
+  #ifndef XXH_NO_LONG_LONG
 
 /* *******************************************************************
-*  64-bit hash functions
-*********************************************************************/
+ *  64-bit hash functions
+ *********************************************************************/
 /*!
  * @}
  * @ingroup impl
  * @{
+
  */
 /*******   Memory access   *******/
 
 typedef XXH64_hash_t xxh_u64;
 
-#ifdef XXH_OLD_NAMES
-#  define U64 xxh_u64
-#endif
+    #ifdef XXH_OLD_NAMES
+      #define U64 xxh_u64
+    #endif
 
-#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
-/*
- * Manual byteshift. Best for old compilers which don't inline memcpy.
- * We actually directly use XXH_readLE64 and XXH_readBE64.
- */
-#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))
+    #if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS == 3))
+    /*
+     * Manual byteshift. Best for old compilers which don't inline memcpy.
+     * We actually directly use XXH_readLE64 and XXH_readBE64.
+     */
+    #elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS == 2))
+
+/* Force direct memory access. Only works on CPU which support unaligned memory
+ * access in hardware */
+static xxh_u64 XXH_read64(const void *memPtr) {
+
+  return *(const xxh_u64 *)memPtr;
 
-/* Force direct memory access. Only works on CPU which support unaligned memory access in hardware */
-static xxh_u64 XXH_read64(const void* memPtr)
-{
-    return *(const xxh_u64*) memPtr;
 }
 
-#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))
+    #elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS == 1))
+
+      /*
+       * __attribute__((aligned(1))) is supported by gcc and clang. Originally
+       * the documentation claimed that it only increased the alignment, but
+       * actually it can decrease it on gcc, clang, and icc:
+       * https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69502,
+       * https://gcc.godbolt.org/z/xYez1j67Y.
+       */
+      #ifdef XXH_OLD_NAMES
+typedef union {
+
+  xxh_u32 u32;
+  xxh_u64 u64;
+
+} __attribute__((packed)) unalign64;
+
+      #endif
+static xxh_u64 XXH_read64(const void *ptr) {
+
+  typedef __attribute__((aligned(1))) xxh_u64 xxh_unalign64;
+  return *((const xxh_unalign64 *)ptr);
 
-/*
- * __attribute__((aligned(1))) is supported by gcc and clang. Originally the
- * documentation claimed that it only increased the alignment, but actually it
- * can decrease it on gcc, clang, and icc:
- * https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69502,
- * https://gcc.godbolt.org/z/xYez1j67Y.
- */
-#ifdef XXH_OLD_NAMES
-typedef union { xxh_u32 u32; xxh_u64 u64; } __attribute__((packed)) unalign64;
-#endif
-static xxh_u64 XXH_read64(const void* ptr)
-{
-    typedef __attribute__((aligned(1))) xxh_u64 xxh_unalign64;
-    return *((const xxh_unalign64*)ptr);
 }
 
-#else
+    #else
 
 /*
  * Portable and safe solution. Generally efficient.
- * see: https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html
+ * see:
+ * https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html
  */
-static xxh_u64 XXH_read64(const void* memPtr)
-{
-    xxh_u64 val;
-    XXH_memcpy(&val, memPtr, sizeof(val));
-    return val;
+static xxh_u64 XXH_read64(const void *memPtr) {
+
+  xxh_u64 val;
+  XXH_memcpy(&val, memPtr, sizeof(val));
+  return val;
+
 }
 
-#endif   /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
+    #endif                                /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
+
+    #if defined(_MSC_VER)                                  /* Visual Studio */
+      #define XXH_swap64 _byteswap_uint64
+    #elif XXH_GCC_VERSION >= 403
+      #define XXH_swap64 __builtin_bswap64
+    #else
+static xxh_u64 XXH_swap64(xxh_u64 x) {
+
+  return ((x << 56) & 0xff00000000000000ULL) |
+         ((x << 40) & 0x00ff000000000000ULL) |
+         ((x << 24) & 0x0000ff0000000000ULL) |
+         ((x << 8) & 0x000000ff00000000ULL) |
+         ((x >> 8) & 0x00000000ff000000ULL) |
+         ((x >> 24) & 0x0000000000ff0000ULL) |
+         ((x >> 40) & 0x000000000000ff00ULL) |
+         ((x >> 56) & 0x00000000000000ffULL);
 
-#if defined(_MSC_VER)     /* Visual Studio */
-#  define XXH_swap64 _byteswap_uint64
-#elif XXH_GCC_VERSION >= 403
-#  define XXH_swap64 __builtin_bswap64
-#else
-static xxh_u64 XXH_swap64(xxh_u64 x)
-{
-    return  ((x << 56) & 0xff00000000000000ULL) |
-            ((x << 40) & 0x00ff000000000000ULL) |
-            ((x << 24) & 0x0000ff0000000000ULL) |
-            ((x << 8)  & 0x000000ff00000000ULL) |
-            ((x >> 8)  & 0x00000000ff000000ULL) |
-            ((x >> 24) & 0x0000000000ff0000ULL) |
-            ((x >> 40) & 0x000000000000ff00ULL) |
-            ((x >> 56) & 0x00000000000000ffULL);
 }
-#endif
 
+    #endif
 
-/* XXH_FORCE_MEMORY_ACCESS==3 is an endian-independent byteshift load. */
-#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
+    /* XXH_FORCE_MEMORY_ACCESS==3 is an endian-independent byteshift load. */
+    #if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS == 3))
+
+XXH_FORCE_INLINE xxh_u64 XXH_readLE64(const void *memPtr) {
+
+  const xxh_u8 *bytePtr = (const xxh_u8 *)memPtr;
+  return bytePtr[0] | ((xxh_u64)bytePtr[1] << 8) | ((xxh_u64)bytePtr[2] << 16) |
+         ((xxh_u64)bytePtr[3] << 24) | ((xxh_u64)bytePtr[4] << 32) |
+         ((xxh_u64)bytePtr[5] << 40) | ((xxh_u64)bytePtr[6] << 48) |
+         ((xxh_u64)bytePtr[7] << 56);
 
-XXH_FORCE_INLINE xxh_u64 XXH_readLE64(const void* memPtr)
-{
-    const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
-    return bytePtr[0]
-         | ((xxh_u64)bytePtr[1] << 8)
-         | ((xxh_u64)bytePtr[2] << 16)
-         | ((xxh_u64)bytePtr[3] << 24)
-         | ((xxh_u64)bytePtr[4] << 32)
-         | ((xxh_u64)bytePtr[5] << 40)
-         | ((xxh_u64)bytePtr[6] << 48)
-         | ((xxh_u64)bytePtr[7] << 56);
-}
-
-XXH_FORCE_INLINE xxh_u64 XXH_readBE64(const void* memPtr)
-{
-    const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
-    return bytePtr[7]
-         | ((xxh_u64)bytePtr[6] << 8)
-         | ((xxh_u64)bytePtr[5] << 16)
-         | ((xxh_u64)bytePtr[4] << 24)
-         | ((xxh_u64)bytePtr[3] << 32)
-         | ((xxh_u64)bytePtr[2] << 40)
-         | ((xxh_u64)bytePtr[1] << 48)
-         | ((xxh_u64)bytePtr[0] << 56);
-}
-
-#else
-XXH_FORCE_INLINE xxh_u64 XXH_readLE64(const void* ptr)
-{
-    return XXH_CPU_LITTLE_ENDIAN ? XXH_read64(ptr) : XXH_swap64(XXH_read64(ptr));
 }
 
-static xxh_u64 XXH_readBE64(const void* ptr)
-{
-    return XXH_CPU_LITTLE_ENDIAN ? XXH_swap64(XXH_read64(ptr)) : XXH_read64(ptr);
+XXH_FORCE_INLINE xxh_u64 XXH_readBE64(const void *memPtr) {
+
+  const xxh_u8 *bytePtr = (const xxh_u8 *)memPtr;
+  return bytePtr[7] | ((xxh_u64)bytePtr[6] << 8) | ((xxh_u64)bytePtr[5] << 16) |
+         ((xxh_u64)bytePtr[4] << 24) | ((xxh_u64)bytePtr[3] << 32) |
+         ((xxh_u64)bytePtr[2] << 40) | ((xxh_u64)bytePtr[1] << 48) |
+         ((xxh_u64)bytePtr[0] << 56);
+
 }
-#endif
 
-XXH_FORCE_INLINE xxh_u64
-XXH_readLE64_align(const void* ptr, XXH_alignment align)
-{
-    if (align==XXH_unaligned)
-        return XXH_readLE64(ptr);
-    else
-        return XXH_CPU_LITTLE_ENDIAN ? *(const xxh_u64*)ptr : XXH_swap64(*(const xxh_u64*)ptr);
+    #else
+XXH_FORCE_INLINE xxh_u64 XXH_readLE64(const void *ptr) {
+
+  return XXH_CPU_LITTLE_ENDIAN ? XXH_read64(ptr) : XXH_swap64(XXH_read64(ptr));
+
 }
 
+static xxh_u64 XXH_readBE64(const void *ptr) {
 
-/*******   xxh64   *******/
-/*!
- * @}
- * @defgroup XXH64_impl XXH64 implementation
- * @ingroup impl
- *
- * Details on the XXH64 implementation.
- * @{
- */
-/* #define rather that static const, to be used as initializers */
-#define XXH_PRIME64_1  0x9E3779B185EBCA87ULL  /*!< 0b1001111000110111011110011011000110000101111010111100101010000111 */
-#define XXH_PRIME64_2  0xC2B2AE3D27D4EB4FULL  /*!< 0b1100001010110010101011100011110100100111110101001110101101001111 */
-#define XXH_PRIME64_3  0x165667B19E3779F9ULL  /*!< 0b0001011001010110011001111011000110011110001101110111100111111001 */
-#define XXH_PRIME64_4  0x85EBCA77C2B2AE63ULL  /*!< 0b1000010111101011110010100111011111000010101100101010111001100011 */
-#define XXH_PRIME64_5  0x27D4EB2F165667C5ULL  /*!< 0b0010011111010100111010110010111100010110010101100110011111000101 */
-
-#ifdef XXH_OLD_NAMES
-#  define PRIME64_1 XXH_PRIME64_1
-#  define PRIME64_2 XXH_PRIME64_2
-#  define PRIME64_3 XXH_PRIME64_3
-#  define PRIME64_4 XXH_PRIME64_4
-#  define PRIME64_5 XXH_PRIME64_5
-#endif
+  return XXH_CPU_LITTLE_ENDIAN ? XXH_swap64(XXH_read64(ptr)) : XXH_read64(ptr);
+
+}
+
+    #endif
+
+XXH_FORCE_INLINE xxh_u64 XXH_readLE64_align(const void   *ptr,
+                                            XXH_alignment align) {
+
+  if (align == XXH_unaligned)
+    return XXH_readLE64(ptr);
+  else
+    return XXH_CPU_LITTLE_ENDIAN ? *(const xxh_u64 *)ptr
+                                 : XXH_swap64(*(const xxh_u64 *)ptr);
+
+}
+
+    /*******   xxh64   *******/
+    /*!
+     * @}
+     * @defgroup XXH64_impl XXH64 implementation
+     * @ingroup impl
+     *
+     * Details on the XXH64 implementation.
+     * @{
+
+     */
+    /* #define rather that static const, to be used as initializers */
+    #define XXH_PRIME64_1                                                                         \
+      0x9E3779B185EBCA87ULL /*!<                                                                  \
+                               0b1001111000110111011110011011000110000101111010111100101010000111 \
+                             */
+    #define XXH_PRIME64_2                                                                         \
+      0xC2B2AE3D27D4EB4FULL /*!<                                                                  \
+                               0b1100001010110010101011100011110100100111110101001110101101001111 \
+                             */
+    #define XXH_PRIME64_3                                                                         \
+      0x165667B19E3779F9ULL /*!<                                                                  \
+                               0b0001011001010110011001111011000110011110001101110111100111111001 \
+                             */
+    #define XXH_PRIME64_4                                                                         \
+      0x85EBCA77C2B2AE63ULL /*!<                                                                  \
+                               0b1000010111101011110010100111011111000010101100101010111001100011 \
+                             */
+    #define XXH_PRIME64_5                                                                         \
+      0x27D4EB2F165667C5ULL /*!<                                                                  \
+                               0b0010011111010100111010110010111100010110010101100110011111000101 \
+                             */
+
+    #ifdef XXH_OLD_NAMES
+      #define PRIME64_1 XXH_PRIME64_1
+      #define PRIME64_2 XXH_PRIME64_2
+      #define PRIME64_3 XXH_PRIME64_3
+      #define PRIME64_4 XXH_PRIME64_4
+      #define PRIME64_5 XXH_PRIME64_5
+    #endif
 
 /*! @copydoc XXH32_round */
-static xxh_u64 XXH64_round(xxh_u64 acc, xxh_u64 input)
-{
-    acc += input * XXH_PRIME64_2;
-    acc  = XXH_rotl64(acc, 31);
-    acc *= XXH_PRIME64_1;
-    return acc;
+static xxh_u64 XXH64_round(xxh_u64 acc, xxh_u64 input) {
+
+  acc += input * XXH_PRIME64_2;
+  acc = XXH_rotl64(acc, 31);
+  acc *= XXH_PRIME64_1;
+  return acc;
+
 }
 
-static xxh_u64 XXH64_mergeRound(xxh_u64 acc, xxh_u64 val)
-{
-    val  = XXH64_round(0, val);
-    acc ^= val;
-    acc  = acc * XXH_PRIME64_1 + XXH_PRIME64_4;
-    return acc;
+static xxh_u64 XXH64_mergeRound(xxh_u64 acc, xxh_u64 val) {
+
+  val = XXH64_round(0, val);
+  acc ^= val;
+  acc = acc * XXH_PRIME64_1 + XXH_PRIME64_4;
+  return acc;
+
 }
 
 /*! @copydoc XXH32_avalanche */
-static xxh_u64 XXH64_avalanche(xxh_u64 hash)
-{
-    hash ^= hash >> 33;
-    hash *= XXH_PRIME64_2;
-    hash ^= hash >> 29;
-    hash *= XXH_PRIME64_3;
-    hash ^= hash >> 32;
-    return hash;
-}
+static xxh_u64 XXH64_avalanche(xxh_u64 hash) {
+
+  hash ^= hash >> 33;
+  hash *= XXH_PRIME64_2;
+  hash ^= hash >> 29;
+  hash *= XXH_PRIME64_3;
+  hash ^= hash >> 32;
+  return hash;
 
+}
 
-#define XXH_get64bits(p) XXH_readLE64_align(p, align)
+    #define XXH_get64bits(p) XXH_readLE64_align(p, align)
 
 /*!
  * @internal
@@ -3410,41 +3758,51 @@ static xxh_u64 XXH64_avalanche(xxh_u64 hash)
  * @return The finalized hash
  * @see XXH32_finalize().
  */
-static XXH_PUREF xxh_u64
-XXH64_finalize(xxh_u64 hash, const xxh_u8* ptr, size_t len, XXH_alignment align)
-{
-    if (ptr==NULL) XXH_ASSERT(len == 0);
-    len &= 31;
-    while (len >= 8) {
-        xxh_u64 const k1 = XXH64_round(0, XXH_get64bits(ptr));
-        ptr += 8;
-        hash ^= k1;
-        hash  = XXH_rotl64(hash,27) * XXH_PRIME64_1 + XXH_PRIME64_4;
-        len -= 8;
-    }
-    if (len >= 4) {
-        hash ^= (xxh_u64)(XXH_get32bits(ptr)) * XXH_PRIME64_1;
-        ptr += 4;
-        hash = XXH_rotl64(hash, 23) * XXH_PRIME64_2 + XXH_PRIME64_3;
-        len -= 4;
-    }
-    while (len > 0) {
-        hash ^= (*ptr++) * XXH_PRIME64_5;
-        hash = XXH_rotl64(hash, 11) * XXH_PRIME64_1;
-        --len;
-    }
-    return  XXH64_avalanche(hash);
+static XXH_PUREF xxh_u64 XXH64_finalize(xxh_u64 hash, const xxh_u8 *ptr,
+                                        size_t len, XXH_alignment align) {
+
+  if (ptr == NULL) XXH_ASSERT(len == 0);
+  len &= 31;
+  while (len >= 8) {
+
+    xxh_u64 const k1 = XXH64_round(0, XXH_get64bits(ptr));
+    ptr += 8;
+    hash ^= k1;
+    hash = XXH_rotl64(hash, 27) * XXH_PRIME64_1 + XXH_PRIME64_4;
+    len -= 8;
+
+  }
+
+  if (len >= 4) {
+
+    hash ^= (xxh_u64)(XXH_get32bits(ptr)) * XXH_PRIME64_1;
+    ptr += 4;
+    hash = XXH_rotl64(hash, 23) * XXH_PRIME64_2 + XXH_PRIME64_3;
+    len -= 4;
+
+  }
+
+  while (len > 0) {
+
+    hash ^= (*ptr++) * XXH_PRIME64_5;
+    hash = XXH_rotl64(hash, 11) * XXH_PRIME64_1;
+    --len;
+
+  }
+
+  return XXH64_avalanche(hash);
+
 }
 
-#ifdef XXH_OLD_NAMES
-#  define PROCESS1_64 XXH_PROCESS1_64
-#  define PROCESS4_64 XXH_PROCESS4_64
-#  define PROCESS8_64 XXH_PROCESS8_64
-#else
-#  undef XXH_PROCESS1_64
-#  undef XXH_PROCESS4_64
-#  undef XXH_PROCESS8_64
-#endif
+    #ifdef XXH_OLD_NAMES
+      #define PROCESS1_64 XXH_PROCESS1_64
+      #define PROCESS4_64 XXH_PROCESS4_64
+      #define PROCESS8_64 XXH_PROCESS8_64
+    #else
+      #undef XXH_PROCESS1_64
+      #undef XXH_PROCESS4_64
+      #undef XXH_PROCESS8_64
+    #endif
 
 /*!
  * @internal
@@ -3454,349 +3812,416 @@ XXH64_finalize(xxh_u64 hash, const xxh_u8* ptr, size_t len, XXH_alignment align)
  * @param align Whether @p input is aligned.
  * @return The calculated hash.
  */
-XXH_FORCE_INLINE XXH_PUREF xxh_u64
-XXH64_endian_align(const xxh_u8* input, size_t len, xxh_u64 seed, XXH_alignment align)
-{
-    xxh_u64 h64;
-    if (input==NULL) XXH_ASSERT(len == 0);
-
-    if (len>=32) {
-        const xxh_u8* const bEnd = input + len;
-        const xxh_u8* const limit = bEnd - 31;
-        xxh_u64 v1 = seed + XXH_PRIME64_1 + XXH_PRIME64_2;
-        xxh_u64 v2 = seed + XXH_PRIME64_2;
-        xxh_u64 v3 = seed + 0;
-        xxh_u64 v4 = seed - XXH_PRIME64_1;
-
-        do {
-            v1 = XXH64_round(v1, XXH_get64bits(input)); input+=8;
-            v2 = XXH64_round(v2, XXH_get64bits(input)); input+=8;
-            v3 = XXH64_round(v3, XXH_get64bits(input)); input+=8;
-            v4 = XXH64_round(v4, XXH_get64bits(input)); input+=8;
-        } while (input<limit);
-
-        h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) + XXH_rotl64(v4, 18);
-        h64 = XXH64_mergeRound(h64, v1);
-        h64 = XXH64_mergeRound(h64, v2);
-        h64 = XXH64_mergeRound(h64, v3);
-        h64 = XXH64_mergeRound(h64, v4);
+XXH_FORCE_INLINE XXH_PUREF xxh_u64 XXH64_endian_align(const xxh_u8 *input,
+                                                      size_t len, xxh_u64 seed,
+                                                      XXH_alignment align) {
 
-    } else {
-        h64  = seed + XXH_PRIME64_5;
-    }
+  xxh_u64 h64;
+  if (input == NULL) XXH_ASSERT(len == 0);
 
-    h64 += (xxh_u64) len;
+  if (len >= 32) {
 
-    return XXH64_finalize(h64, input, len, align);
-}
+    const xxh_u8 *const bEnd = input + len;
+    const xxh_u8 *const limit = bEnd - 31;
+    xxh_u64             v1 = seed + XXH_PRIME64_1 + XXH_PRIME64_2;
+    xxh_u64             v2 = seed + XXH_PRIME64_2;
+    xxh_u64             v3 = seed + 0;
+    xxh_u64             v4 = seed - XXH_PRIME64_1;
+
+    do {
+
+      v1 = XXH64_round(v1, XXH_get64bits(input));
+      input += 8;
+      v2 = XXH64_round(v2, XXH_get64bits(input));
+      input += 8;
+      v3 = XXH64_round(v3, XXH_get64bits(input));
+      input += 8;
+      v4 = XXH64_round(v4, XXH_get64bits(input));
+      input += 8;
+
+    } while (input < limit);
+
+    h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) +
+          XXH_rotl64(v4, 18);
+    h64 = XXH64_mergeRound(h64, v1);
+    h64 = XXH64_mergeRound(h64, v2);
+    h64 = XXH64_mergeRound(h64, v3);
+    h64 = XXH64_mergeRound(h64, v4);
+
+  } else {
+
+    h64 = seed + XXH_PRIME64_5;
+
+  }
+
+  h64 += (xxh_u64)len;
 
+  return XXH64_finalize(h64, input, len, align);
+
+}
 
 /*! @ingroup XXH64_family */
-XXH_PUBLIC_API XXH64_hash_t XXH64 (XXH_NOESCAPE const void* input, size_t len, XXH64_hash_t seed)
-{
-#if !defined(XXH_NO_STREAM) && XXH_SIZE_OPT >= 2
-    /* Simple version, good for code maintenance, but unfortunately slow for small inputs */
-    XXH64_state_t state;
-    XXH64_reset(&state, seed);
-    XXH64_update(&state, (const xxh_u8*)input, len);
-    return XXH64_digest(&state);
-#else
-    if (XXH_FORCE_ALIGN_CHECK) {
-        if ((((size_t)input) & 7)==0) {  /* Input is aligned, let's leverage the speed advantage */
-            return XXH64_endian_align((const xxh_u8*)input, len, seed, XXH_aligned);
-    }   }
-
-    return XXH64_endian_align((const xxh_u8*)input, len, seed, XXH_unaligned);
+XXH_PUBLIC_API XXH64_hash_t XXH64(XXH_NOESCAPE const void *input, size_t len,
+                                  XXH64_hash_t seed) {
+
+    #if !defined(XXH_NO_STREAM) && XXH_SIZE_OPT >= 2
+  /* Simple version, good for code maintenance, but unfortunately slow for small
+   * inputs */
+  XXH64_state_t state;
+  XXH64_reset(&state, seed);
+  XXH64_update(&state, (const xxh_u8 *)input, len);
+  return XXH64_digest(&state);
+    #else
+  if (XXH_FORCE_ALIGN_CHECK) {
+
+    if ((((size_t)input) & 7) ==
+        0) {        /* Input is aligned, let's leverage the speed advantage */
+      return XXH64_endian_align((const xxh_u8 *)input, len, seed, XXH_aligned);
+
+    }
+
+  }
+
+  return XXH64_endian_align((const xxh_u8 *)input, len, seed, XXH_unaligned);
+
+    #endif
 
-#endif
 }
 
-/*******   Hash Streaming   *******/
-#ifndef XXH_NO_STREAM
+    /*******   Hash Streaming   *******/
+    #ifndef XXH_NO_STREAM
 /*! @ingroup XXH64_family*/
-XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void)
-{
-    return (XXH64_state_t*)XXH_malloc(sizeof(XXH64_state_t));
+XXH_PUBLIC_API XXH64_state_t *XXH64_createState(void) {
+
+  return (XXH64_state_t *)XXH_malloc(sizeof(XXH64_state_t));
+
 }
+
 /*! @ingroup XXH64_family */
-XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr)
-{
-    XXH_free(statePtr);
-    return XXH_OK;
+XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t *statePtr) {
+
+  XXH_free(statePtr);
+  return XXH_OK;
+
 }
 
 /*! @ingroup XXH64_family */
-XXH_PUBLIC_API void XXH64_copyState(XXH_NOESCAPE XXH64_state_t* dstState, const XXH64_state_t* srcState)
-{
-    XXH_memcpy(dstState, srcState, sizeof(*dstState));
+XXH_PUBLIC_API void XXH64_copyState(XXH_NOESCAPE XXH64_state_t *dstState,
+                                    const XXH64_state_t        *srcState) {
+
+  XXH_memcpy(dstState, srcState, sizeof(*dstState));
+
 }
 
 /*! @ingroup XXH64_family */
-XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH_NOESCAPE XXH64_state_t* statePtr, XXH64_hash_t seed)
-{
-    XXH_ASSERT(statePtr != NULL);
-    memset(statePtr, 0, sizeof(*statePtr));
-    statePtr->v[0] = seed + XXH_PRIME64_1 + XXH_PRIME64_2;
-    statePtr->v[1] = seed + XXH_PRIME64_2;
-    statePtr->v[2] = seed + 0;
-    statePtr->v[3] = seed - XXH_PRIME64_1;
-    return XXH_OK;
+XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH_NOESCAPE XXH64_state_t *statePtr,
+                                         XXH64_hash_t                seed) {
+
+  XXH_ASSERT(statePtr != NULL);
+  memset(statePtr, 0, sizeof(*statePtr));
+  statePtr->v[0] = seed + XXH_PRIME64_1 + XXH_PRIME64_2;
+  statePtr->v[1] = seed + XXH_PRIME64_2;
+  statePtr->v[2] = seed + 0;
+  statePtr->v[3] = seed - XXH_PRIME64_1;
+  return XXH_OK;
+
 }
 
 /*! @ingroup XXH64_family */
-XXH_PUBLIC_API XXH_errorcode
-XXH64_update (XXH_NOESCAPE XXH64_state_t* state, XXH_NOESCAPE const void* input, size_t len)
-{
-    if (input==NULL) {
-        XXH_ASSERT(len == 0);
-        return XXH_OK;
-    }
+XXH_PUBLIC_API XXH_errorcode XXH64_update(XXH_NOESCAPE XXH64_state_t *state,
+                                          XXH_NOESCAPE const void    *input,
+                                          size_t                      len) {
 
-    {   const xxh_u8* p = (const xxh_u8*)input;
-        const xxh_u8* const bEnd = p + len;
+  if (input == NULL) {
 
-        state->total_len += len;
+    XXH_ASSERT(len == 0);
+    return XXH_OK;
 
-        if (state->memsize + len < 32) {  /* fill in tmp buffer */
-            XXH_memcpy(((xxh_u8*)state->mem64) + state->memsize, input, len);
-            state->memsize += (xxh_u32)len;
-            return XXH_OK;
-        }
+  }
 
-        if (state->memsize) {   /* tmp buffer is full */
-            XXH_memcpy(((xxh_u8*)state->mem64) + state->memsize, input, 32-state->memsize);
-            state->v[0] = XXH64_round(state->v[0], XXH_readLE64(state->mem64+0));
-            state->v[1] = XXH64_round(state->v[1], XXH_readLE64(state->mem64+1));
-            state->v[2] = XXH64_round(state->v[2], XXH_readLE64(state->mem64+2));
-            state->v[3] = XXH64_round(state->v[3], XXH_readLE64(state->mem64+3));
-            p += 32 - state->memsize;
-            state->memsize = 0;
-        }
+  {
 
-        if (p+32 <= bEnd) {
-            const xxh_u8* const limit = bEnd - 32;
+    const xxh_u8       *p = (const xxh_u8 *)input;
+    const xxh_u8 *const bEnd = p + len;
 
-            do {
-                state->v[0] = XXH64_round(state->v[0], XXH_readLE64(p)); p+=8;
-                state->v[1] = XXH64_round(state->v[1], XXH_readLE64(p)); p+=8;
-                state->v[2] = XXH64_round(state->v[2], XXH_readLE64(p)); p+=8;
-                state->v[3] = XXH64_round(state->v[3], XXH_readLE64(p)); p+=8;
-            } while (p<=limit);
+    state->total_len += len;
 
-        }
+    if (state->memsize + len < 32) {                  /* fill in tmp buffer */
+      XXH_memcpy(((xxh_u8 *)state->mem64) + state->memsize, input, len);
+      state->memsize += (xxh_u32)len;
+      return XXH_OK;
 
-        if (p < bEnd) {
-            XXH_memcpy(state->mem64, p, (size_t)(bEnd-p));
-            state->memsize = (unsigned)(bEnd-p);
-        }
     }
 
-    return XXH_OK;
-}
+    if (state->memsize) {                             /* tmp buffer is full */
+      XXH_memcpy(((xxh_u8 *)state->mem64) + state->memsize, input,
+                 32 - state->memsize);
+      state->v[0] = XXH64_round(state->v[0], XXH_readLE64(state->mem64 + 0));
+      state->v[1] = XXH64_round(state->v[1], XXH_readLE64(state->mem64 + 1));
+      state->v[2] = XXH64_round(state->v[2], XXH_readLE64(state->mem64 + 2));
+      state->v[3] = XXH64_round(state->v[3], XXH_readLE64(state->mem64 + 3));
+      p += 32 - state->memsize;
+      state->memsize = 0;
 
+    }
+
+    if (p + 32 <= bEnd) {
+
+      const xxh_u8 *const limit = bEnd - 32;
+
+      do {
+
+        state->v[0] = XXH64_round(state->v[0], XXH_readLE64(p));
+        p += 8;
+        state->v[1] = XXH64_round(state->v[1], XXH_readLE64(p));
+        p += 8;
+        state->v[2] = XXH64_round(state->v[2], XXH_readLE64(p));
+        p += 8;
+        state->v[3] = XXH64_round(state->v[3], XXH_readLE64(p));
+        p += 8;
+
+      } while (p <= limit);
 
-/*! @ingroup XXH64_family */
-XXH_PUBLIC_API XXH64_hash_t XXH64_digest(XXH_NOESCAPE const XXH64_state_t* state)
-{
-    xxh_u64 h64;
-
-    if (state->total_len >= 32) {
-        h64 = XXH_rotl64(state->v[0], 1) + XXH_rotl64(state->v[1], 7) + XXH_rotl64(state->v[2], 12) + XXH_rotl64(state->v[3], 18);
-        h64 = XXH64_mergeRound(h64, state->v[0]);
-        h64 = XXH64_mergeRound(h64, state->v[1]);
-        h64 = XXH64_mergeRound(h64, state->v[2]);
-        h64 = XXH64_mergeRound(h64, state->v[3]);
-    } else {
-        h64  = state->v[2] /*seed*/ + XXH_PRIME64_5;
     }
 
-    h64 += (xxh_u64) state->total_len;
+    if (p < bEnd) {
 
-    return XXH64_finalize(h64, (const xxh_u8*)state->mem64, (size_t)state->total_len, XXH_aligned);
-}
-#endif /* !XXH_NO_STREAM */
+      XXH_memcpy(state->mem64, p, (size_t)(bEnd - p));
+      state->memsize = (unsigned)(bEnd - p);
 
-/******* Canonical representation   *******/
+    }
+
+  }
+
+  return XXH_OK;
 
-/*! @ingroup XXH64_family */
-XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH_NOESCAPE XXH64_canonical_t* dst, XXH64_hash_t hash)
-{
-    XXH_STATIC_ASSERT(sizeof(XXH64_canonical_t) == sizeof(XXH64_hash_t));
-    if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap64(hash);
-    XXH_memcpy(dst, &hash, sizeof(*dst));
 }
 
 /*! @ingroup XXH64_family */
-XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(XXH_NOESCAPE const XXH64_canonical_t* src)
-{
-    return XXH_readBE64(src);
+XXH_PUBLIC_API XXH64_hash_t
+XXH64_digest(XXH_NOESCAPE const XXH64_state_t *state) {
+
+  xxh_u64 h64;
+
+  if (state->total_len >= 32) {
+
+    h64 = XXH_rotl64(state->v[0], 1) + XXH_rotl64(state->v[1], 7) +
+          XXH_rotl64(state->v[2], 12) + XXH_rotl64(state->v[3], 18);
+    h64 = XXH64_mergeRound(h64, state->v[0]);
+    h64 = XXH64_mergeRound(h64, state->v[1]);
+    h64 = XXH64_mergeRound(h64, state->v[2]);
+    h64 = XXH64_mergeRound(h64, state->v[3]);
+
+  } else {
+
+    h64 = state->v[2] /*seed*/ + XXH_PRIME64_5;
+
+  }
+
+  h64 += (xxh_u64)state->total_len;
+
+  return XXH64_finalize(h64, (const xxh_u8 *)state->mem64,
+                        (size_t)state->total_len, XXH_aligned);
+
 }
 
-#ifndef XXH_NO_XXH3
+    #endif                                                /* !XXH_NO_STREAM */
 
-/* *********************************************************************
-*  XXH3
-*  New generation hash designed for speed on small keys and vectorization
-************************************************************************ */
-/*!
- * @}
- * @defgroup XXH3_impl XXH3 implementation
- * @ingroup impl
- * @{
- */
+/******* Canonical representation   *******/
 
-/* ===   Compiler specifics   === */
+/*! @ingroup XXH64_family */
+XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH_NOESCAPE XXH64_canonical_t *dst,
+                                            XXH64_hash_t hash) {
 
-#if ((defined(sun) || defined(__sun)) && __cplusplus) /* Solaris includes __STDC_VERSION__ with C++. Tested with GCC 5.5 */
-#  define XXH_RESTRICT   /* disable */
-#elif defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* >= C99 */
-#  define XXH_RESTRICT   restrict
-#elif (defined (__GNUC__) && ((__GNUC__ > 3) || (__GNUC__ == 3 && __GNUC_MINOR__ >= 1))) \
-   || (defined (__clang__)) \
-   || (defined (_MSC_VER) && (_MSC_VER >= 1400)) \
-   || (defined (__INTEL_COMPILER) && (__INTEL_COMPILER >= 1300))
-/*
- * There are a LOT more compilers that recognize __restrict but this
- * covers the major ones.
- */
-#  define XXH_RESTRICT   __restrict
-#else
-#  define XXH_RESTRICT   /* disable */
-#endif
+  XXH_STATIC_ASSERT(sizeof(XXH64_canonical_t) == sizeof(XXH64_hash_t));
+  if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap64(hash);
+  XXH_memcpy(dst, &hash, sizeof(*dst));
 
-#if (defined(__GNUC__) && (__GNUC__ >= 3))  \
-  || (defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 800)) \
-  || defined(__clang__)
-#    define XXH_likely(x) __builtin_expect(x, 1)
-#    define XXH_unlikely(x) __builtin_expect(x, 0)
-#else
-#    define XXH_likely(x) (x)
-#    define XXH_unlikely(x) (x)
-#endif
+}
 
-#ifndef XXH_HAS_INCLUDE
-#  ifdef __has_include
-/*
- * Not defined as XXH_HAS_INCLUDE(x) (function-like) because
- * this causes segfaults in Apple Clang 4.2 (on Mac OS X 10.7 Lion)
- */
-#    define XXH_HAS_INCLUDE __has_include
-#  else
-#    define XXH_HAS_INCLUDE(x) 0
-#  endif
-#endif
+/*! @ingroup XXH64_family */
+XXH_PUBLIC_API XXH64_hash_t
+XXH64_hashFromCanonical(XXH_NOESCAPE const XXH64_canonical_t *src) {
 
-#if defined(__GNUC__) || defined(__clang__)
-#  if defined(__ARM_FEATURE_SVE)
-#    include <arm_sve.h>
-#  endif
-#  if defined(__ARM_NEON__) || defined(__ARM_NEON) \
-   || (defined(_M_ARM) && _M_ARM >= 7) \
-   || defined(_M_ARM64) || defined(_M_ARM64EC) \
-   || (defined(__wasm_simd128__) && XXH_HAS_INCLUDE(<arm_neon.h>)) /* WASM SIMD128 via SIMDe */
-#    define inline __inline__  /* circumvent a clang bug */
-#    include <arm_neon.h>
-#    undef inline
-#  elif defined(__AVX2__)
-#    include <immintrin.h>
-#  elif defined(__SSE2__)
-#    include <emmintrin.h>
-#  endif
-#endif
+  return XXH_readBE64(src);
 
-#if defined(_MSC_VER)
-#  include <intrin.h>
-#endif
+}
 
-/*
- * One goal of XXH3 is to make it fast on both 32-bit and 64-bit, while
- * remaining a true 64-bit/128-bit hash function.
- *
- * This is done by prioritizing a subset of 64-bit operations that can be
- * emulated without too many steps on the average 32-bit machine.
- *
- * For example, these two lines seem similar, and run equally fast on 64-bit:
- *
- *   xxh_u64 x;
- *   x ^= (x >> 47); // good
- *   x ^= (x >> 13); // bad
- *
- * However, to a 32-bit machine, there is a major difference.
- *
- * x ^= (x >> 47) looks like this:
- *
- *   x.lo ^= (x.hi >> (47 - 32));
- *
- * while x ^= (x >> 13) looks like this:
- *
- *   // note: funnel shifts are not usually cheap.
- *   x.lo ^= (x.lo >> 13) | (x.hi << (32 - 13));
- *   x.hi ^= (x.hi >> 13);
- *
- * The first one is significantly faster than the second, simply because the
- * shift is larger than 32. This means:
- *  - All the bits we need are in the upper 32 bits, so we can ignore the lower
- *    32 bits in the shift.
- *  - The shift result will always fit in the lower 32 bits, and therefore,
- *    we can ignore the upper 32 bits in the xor.
- *
- * Thanks to this optimization, XXH3 only requires these features to be efficient:
- *
- *  - Usable unaligned access
- *  - A 32-bit or 64-bit ALU
- *      - If 32-bit, a decent ADC instruction
- *  - A 32 or 64-bit multiply with a 64-bit result
- *  - For the 128-bit variant, a decent byteswap helps short inputs.
- *
- * The first two are already required by XXH32, and almost all 32-bit and 64-bit
- * platforms which can run XXH32 can run XXH3 efficiently.
- *
- * Thumb-1, the classic 16-bit only subset of ARM's instruction set, is one
- * notable exception.
- *
- * First of all, Thumb-1 lacks support for the UMULL instruction which
- * performs the important long multiply. This means numerous __aeabi_lmul
- * calls.
- *
- * Second of all, the 8 functional registers are just not enough.
- * Setup for __aeabi_lmul, byteshift loads, pointers, and all arithmetic need
- * Lo registers, and this shuffling results in thousands more MOVs than A32.
- *
- * A32 and T32 don't have this limitation. They can access all 14 registers,
- * do a 32->64 multiply with UMULL, and the flexible operand allowing free
- * shifts is helpful, too.
- *
- * Therefore, we do a quick sanity check.
- *
- * If compiling Thumb-1 for a target which supports ARM instructions, we will
- * emit a warning, as it is not a "sane" platform to compile for.
- *
- * Usually, if this happens, it is because of an accident and you probably need
- * to specify -march, as you likely meant to compile for a newer architecture.
- *
- * Credit: large sections of the vectorial and asm source code paths
- *         have been contributed by @easyaspi314
- */
-#if defined(__thumb__) && !defined(__thumb2__) && defined(__ARM_ARCH_ISA_ARM)
-#   warning "XXH3 is highly inefficient without ARM or Thumb-2."
-#endif
+    #ifndef XXH_NO_XXH3
 
-/* ==========================================
- * Vectorization detection
- * ========================================== */
+    /* *********************************************************************
+     *  XXH3
+     *  New generation hash designed for speed on small keys and vectorization
+     ************************************************************************ */
+    /*!
+     * @}
+     * @defgroup XXH3_impl XXH3 implementation
+     * @ingroup impl
+     * @{
 
-#ifdef XXH_DOXYGEN
-/*!
- * @ingroup tuning
- * @brief Overrides the vectorization implementation chosen for XXH3.
- *
- * Can be defined to 0 to disable SIMD or any of the values mentioned in
- * @ref XXH_VECTOR_TYPE.
- *
- * If this is not defined, it uses predefined macros to determine the best
- * implementation.
- */
-#  define XXH_VECTOR XXH_SCALAR
+     */
+
+    /* ===   Compiler specifics   === */
+
+      #if ((defined(sun) || defined(__sun)) &&                                \
+           __cplusplus) /* Solaris includes __STDC_VERSION__ with C++. Tested \
+                           with GCC 5.5 */
+        #define XXH_RESTRICT                                     /* disable */
+      #elif defined(__STDC_VERSION__) && \
+          __STDC_VERSION__ >= 199901L                             /* >= C99 */
+        #define XXH_RESTRICT restrict
+      #elif (defined(__GNUC__) &&                                              \
+             ((__GNUC__ > 3) || (__GNUC__ == 3 && __GNUC_MINOR__ >= 1))) ||    \
+          (defined(__clang__)) || (defined(_MSC_VER) && (_MSC_VER >= 1400)) || \
+          (defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 1300))
+      /*
+       * There are a LOT more compilers that recognize __restrict but this
+       * covers the major ones.
+       */
+        #define XXH_RESTRICT __restrict
+      #else
+        #define XXH_RESTRICT                                     /* disable */
+      #endif
+
+      #if (defined(__GNUC__) && (__GNUC__ >= 3)) ||                   \
+          (defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 800)) || \
+          defined(__clang__)
+        #define XXH_likely(x) __builtin_expect(x, 1)
+        #define XXH_unlikely(x) __builtin_expect(x, 0)
+      #else
+        #define XXH_likely(x) (x)
+        #define XXH_unlikely(x) (x)
+      #endif
+
+      #ifndef XXH_HAS_INCLUDE
+        #ifdef __has_include
+          /*
+           * Not defined as XXH_HAS_INCLUDE(x) (function-like) because
+           * this causes segfaults in Apple Clang 4.2 (on Mac OS X 10.7 Lion)
+           */
+          #define XXH_HAS_INCLUDE __has_include
+        #else
+          #define XXH_HAS_INCLUDE(x) 0
+        #endif
+      #endif
+
+      #if defined(__GNUC__) || defined(__clang__)
+        #if defined(__ARM_FEATURE_SVE)
+          #include <arm_sve.h>
+        #endif
+        #if defined(__ARM_NEON__) || defined(__ARM_NEON) ||          \
+            (defined(_M_ARM) && _M_ARM >= 7) || defined(_M_ARM64) || \
+            defined(_M_ARM64EC) ||                                   \
+            (defined(__wasm_simd128__) &&                            \
+             XXH_HAS_INCLUDE(<arm_neon.h>))       /* WASM SIMD128 via SIMDe */
+          #define inline __inline__               /* circumvent a clang bug */
+          #include <arm_neon.h>
+          #undef inline
+        #elif defined(__AVX2__)
+          #include <immintrin.h>
+        #elif defined(__SSE2__)
+          #include <emmintrin.h>
+        #endif
+      #endif
+
+      #if defined(_MSC_VER)
+        #include <intrin.h>
+      #endif
+
+      /*
+       * One goal of XXH3 is to make it fast on both 32-bit and 64-bit, while
+       * remaining a true 64-bit/128-bit hash function.
+       *
+       * This is done by prioritizing a subset of 64-bit operations that can be
+       * emulated without too many steps on the average 32-bit machine.
+       *
+       * For example, these two lines seem similar, and run equally fast on
+       * 64-bit:
+       *
+       *   xxh_u64 x;
+       *   x ^= (x >> 47); // good
+       *   x ^= (x >> 13); // bad
+       *
+       * However, to a 32-bit machine, there is a major difference.
+       *
+       * x ^= (x >> 47) looks like this:
+       *
+       *   x.lo ^= (x.hi >> (47 - 32));
+       *
+       * while x ^= (x >> 13) looks like this:
+       *
+       *   // note: funnel shifts are not usually cheap.
+       *   x.lo ^= (x.lo >> 13) | (x.hi << (32 - 13));
+       *   x.hi ^= (x.hi >> 13);
+       *
+       * The first one is significantly faster than the second, simply because
+       * the shift is larger than 32. This means:
+       *  - All the bits we need are in the upper 32 bits, so we can ignore the
+       * lower 32 bits in the shift.
+       *  - The shift result will always fit in the lower 32 bits, and
+       * therefore, we can ignore the upper 32 bits in the xor.
+       *
+       * Thanks to this optimization, XXH3 only requires these features to be
+       * efficient:
+       *
+       *  - Usable unaligned access
+       *  - A 32-bit or 64-bit ALU
+       *      - If 32-bit, a decent ADC instruction
+       *  - A 32 or 64-bit multiply with a 64-bit result
+       *  - For the 128-bit variant, a decent byteswap helps short inputs.
+       *
+       * The first two are already required by XXH32, and almost all 32-bit and
+       * 64-bit platforms which can run XXH32 can run XXH3 efficiently.
+       *
+       * Thumb-1, the classic 16-bit only subset of ARM's instruction set, is
+       * one notable exception.
+       *
+       * First of all, Thumb-1 lacks support for the UMULL instruction which
+       * performs the important long multiply. This means numerous __aeabi_lmul
+       * calls.
+       *
+       * Second of all, the 8 functional registers are just not enough.
+       * Setup for __aeabi_lmul, byteshift loads, pointers, and all arithmetic
+       * need Lo registers, and this shuffling results in thousands more MOVs
+       * than A32.
+       *
+       * A32 and T32 don't have this limitation. They can access all 14
+       * registers, do a 32->64 multiply with UMULL, and the flexible operand
+       * allowing free shifts is helpful, too.
+       *
+       * Therefore, we do a quick sanity check.
+       *
+       * If compiling Thumb-1 for a target which supports ARM instructions, we
+       * will emit a warning, as it is not a "sane" platform to compile for.
+       *
+       * Usually, if this happens, it is because of an accident and you probably
+       * need to specify -march, as you likely meant to compile for a newer
+       * architecture.
+       *
+       * Credit: large sections of the vectorial and asm source code paths
+       *         have been contributed by @easyaspi314
+       */
+      #if defined(__thumb__) && !defined(__thumb2__) && \
+          defined(__ARM_ARCH_ISA_ARM)
+        #warning "XXH3 is highly inefficient without ARM or Thumb-2."
+      #endif
+
+    /* ==========================================
+     * Vectorization detection
+     * ========================================== */
+
+      #ifdef XXH_DOXYGEN
+        /*!
+         * @ingroup tuning
+         * @brief Overrides the vectorization implementation chosen for XXH3.
+         *
+         * Can be defined to 0 to disable SIMD or any of the values mentioned in
+         * @ref XXH_VECTOR_TYPE.
+         *
+         * If this is not defined, it uses predefined macros to determine the
+         * best implementation.
+         */
+        #define XXH_VECTOR XXH_SCALAR
 /*!
  * @ingroup tuning
  * @brief Possible values for @ref XXH_VECTOR.
@@ -3807,491 +4232,560 @@ XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(XXH_NOESCAPE const XXH64_can
  * internal macro XXH_X86DISPATCH overrides this.
  */
 enum XXH_VECTOR_TYPE /* fake enum */ {
-    XXH_SCALAR = 0,  /*!< Portable scalar version */
-    XXH_SSE2   = 1,  /*!<
-                      * SSE2 for Pentium 4, Opteron, all x86_64.
-                      *
-                      * @note SSE2 is also guaranteed on Windows 10, macOS, and
-                      * Android x86.
-                      */
-    XXH_AVX2   = 2,  /*!< AVX2 for Haswell and Bulldozer */
-    XXH_AVX512 = 3,  /*!< AVX512 for Skylake and Icelake */
-    XXH_NEON   = 4,  /*!<
-                       * NEON for most ARMv7-A, all AArch64, and WASM SIMD128
-                       * via the SIMDeverywhere polyfill provided with the
-                       * Emscripten SDK.
-                       */
-    XXH_VSX    = 5,  /*!< VSX and ZVector for POWER8/z13 (64-bit) */
-    XXH_SVE    = 6,  /*!< SVE for some ARMv8-A and ARMv9-A */
-};
-/*!
- * @ingroup tuning
- * @brief Selects the minimum alignment for XXH3's accumulators.
- *
- * When using SIMD, this should match the alignment required for said vector
- * type, so, for example, 32 for AVX2.
- *
- * Default: Auto detected.
- */
-#  define XXH_ACC_ALIGN 8
-#endif
-
-/* Actual definition */
-#ifndef XXH_DOXYGEN
-#  define XXH_SCALAR 0
-#  define XXH_SSE2   1
-#  define XXH_AVX2   2
-#  define XXH_AVX512 3
-#  define XXH_NEON   4
-#  define XXH_VSX    5
-#  define XXH_SVE    6
-#endif
-
-#ifndef XXH_VECTOR    /* can be defined on command line */
-#  if defined(__ARM_FEATURE_SVE)
-#    define XXH_VECTOR XXH_SVE
-#  elif ( \
-        defined(__ARM_NEON__) || defined(__ARM_NEON) /* gcc */ \
-     || defined(_M_ARM) || defined(_M_ARM64) || defined(_M_ARM64EC) /* msvc */ \
-     || (defined(__wasm_simd128__) && XXH_HAS_INCLUDE(<arm_neon.h>)) /* wasm simd128 via SIMDe */ \
-   ) && ( \
-        defined(_WIN32) || defined(__LITTLE_ENDIAN__) /* little endian only */ \
-    || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
-   )
-#    define XXH_VECTOR XXH_NEON
-#  elif defined(__AVX512F__)
-#    define XXH_VECTOR XXH_AVX512
-#  elif defined(__AVX2__)
-#    define XXH_VECTOR XXH_AVX2
-#  elif defined(__SSE2__) || defined(_M_AMD64) || defined(_M_X64) || (defined(_M_IX86_FP) && (_M_IX86_FP == 2))
-#    define XXH_VECTOR XXH_SSE2
-#  elif (defined(__PPC64__) && defined(__POWER8_VECTOR__)) \
-     || (defined(__s390x__) && defined(__VEC__)) \
-     && defined(__GNUC__) /* TODO: IBM XL */
-#    define XXH_VECTOR XXH_VSX
-#  else
-#    define XXH_VECTOR XXH_SCALAR
-#  endif
-#endif
-
-/* __ARM_FEATURE_SVE is only supported by GCC & Clang. */
-#if (XXH_VECTOR == XXH_SVE) && !defined(__ARM_FEATURE_SVE)
-#  ifdef _MSC_VER
-#    pragma warning(once : 4606)
-#  else
-#    warning "__ARM_FEATURE_SVE isn't supported. Use SCALAR instead."
-#  endif
-#  undef XXH_VECTOR
-#  define XXH_VECTOR XXH_SCALAR
-#endif
-
-/*
- * Controls the alignment of the accumulator,
- * for compatibility with aligned vector loads, which are usually faster.
- */
-#ifndef XXH_ACC_ALIGN
-#  if defined(XXH_X86DISPATCH)
-#     define XXH_ACC_ALIGN 64  /* for compatibility with avx512 */
-#  elif XXH_VECTOR == XXH_SCALAR  /* scalar */
-#     define XXH_ACC_ALIGN 8
-#  elif XXH_VECTOR == XXH_SSE2  /* sse2 */
-#     define XXH_ACC_ALIGN 16
-#  elif XXH_VECTOR == XXH_AVX2  /* avx2 */
-#     define XXH_ACC_ALIGN 32
-#  elif XXH_VECTOR == XXH_NEON  /* neon */
-#     define XXH_ACC_ALIGN 16
-#  elif XXH_VECTOR == XXH_VSX   /* vsx */
-#     define XXH_ACC_ALIGN 16
-#  elif XXH_VECTOR == XXH_AVX512  /* avx512 */
-#     define XXH_ACC_ALIGN 64
-#  elif XXH_VECTOR == XXH_SVE   /* sve */
-#     define XXH_ACC_ALIGN 64
-#  endif
-#endif
 
-#if defined(XXH_X86DISPATCH) || XXH_VECTOR == XXH_SSE2 \
-    || XXH_VECTOR == XXH_AVX2 || XXH_VECTOR == XXH_AVX512
-#  define XXH_SEC_ALIGN XXH_ACC_ALIGN
-#elif XXH_VECTOR == XXH_SVE
-#  define XXH_SEC_ALIGN XXH_ACC_ALIGN
-#else
-#  define XXH_SEC_ALIGN 8
-#endif
-
-#if defined(__GNUC__) || defined(__clang__)
-#  define XXH_ALIASING __attribute__((may_alias))
-#else
-#  define XXH_ALIASING /* nothing */
-#endif
+  XXH_SCALAR = 0,                              /*!< Portable scalar version */
+  XXH_SSE2 = 1,   /*!<
+                   * SSE2 for Pentium 4, Opteron, all x86_64.
+                   *
+                   * @note SSE2 is also guaranteed on Windows 10, macOS, and
+                   * Android x86.
+                   */
+  XXH_AVX2 = 2,                         /*!< AVX2 for Haswell and Bulldozer */
+  XXH_AVX512 = 3,                       /*!< AVX512 for Skylake and Icelake */
+  XXH_NEON = 4,   /*!<
+                   * NEON for most ARMv7-A, all AArch64, and WASM SIMD128
+                   * via the SIMDeverywhere polyfill provided with the
+                   * Emscripten SDK.
+                   */
+  XXH_VSX = 5,                 /*!< VSX and ZVector for POWER8/z13 (64-bit) */
+  XXH_SVE = 6,                        /*!< SVE for some ARMv8-A and ARMv9-A */
 
-/*
- * UGLY HACK:
- * GCC usually generates the best code with -O3 for xxHash.
- *
- * However, when targeting AVX2, it is overzealous in its unrolling resulting
- * in code roughly 3/4 the speed of Clang.
- *
- * There are other issues, such as GCC splitting _mm256_loadu_si256 into
- * _mm_loadu_si128 + _mm256_inserti128_si256. This is an optimization which
- * only applies to Sandy and Ivy Bridge... which don't even support AVX2.
- *
- * That is why when compiling the AVX2 version, it is recommended to use either
- *   -O2 -mavx2 -march=haswell
- * or
- *   -O2 -mavx2 -mno-avx256-split-unaligned-load
- * for decent performance, or to use Clang instead.
- *
- * Fortunately, we can control the first one with a pragma that forces GCC into
- * -O2, but the other one we can't control without "failed to inline always
- * inline function due to target mismatch" warnings.
- */
-#if XXH_VECTOR == XXH_AVX2 /* AVX2 */ \
-  && defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
-  && defined(__OPTIMIZE__) && XXH_SIZE_OPT <= 0 /* respect -O0 and -Os */
-#  pragma GCC push_options
-#  pragma GCC optimize("-O2")
-#endif
+};
 
-#if XXH_VECTOR == XXH_NEON
+        /*!
+         * @ingroup tuning
+         * @brief Selects the minimum alignment for XXH3's accumulators.
+         *
+         * When using SIMD, this should match the alignment required for said
+         * vector type, so, for example, 32 for AVX2.
+         *
+         * Default: Auto detected.
+         */
+        #define XXH_ACC_ALIGN 8
+      #endif
+
+      /* Actual definition */
+      #ifndef XXH_DOXYGEN
+        #define XXH_SCALAR 0
+        #define XXH_SSE2 1
+        #define XXH_AVX2 2
+        #define XXH_AVX512 3
+        #define XXH_NEON 4
+        #define XXH_VSX 5
+        #define XXH_SVE 6
+      #endif
+
+      #ifndef XXH_VECTOR                  /* can be defined on command line */
+        #if defined(__ARM_FEATURE_SVE)
+          #define XXH_VECTOR XXH_SVE
+        #elif (defined(__ARM_NEON__) || defined(__ARM_NEON) /* gcc */          \
+               || defined(_M_ARM) || defined(_M_ARM64) ||                      \
+               defined(_M_ARM64EC) /* msvc */                                  \
+               || (defined(__wasm_simd128__) &&                                \
+                   XXH_HAS_INCLUDE(<arm_neon.h>)) /* wasm simd128 via SIMDe */ \
+               ) &&                                                            \
+            (defined(_WIN32) ||                                                \
+             defined(__LITTLE_ENDIAN__) /* little endian only */               \
+             || (defined(__BYTE_ORDER__) &&                                    \
+                 __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__))
+          #define XXH_VECTOR XXH_NEON
+        #elif defined(__AVX512F__)
+          #define XXH_VECTOR XXH_AVX512
+        #elif defined(__AVX2__)
+          #define XXH_VECTOR XXH_AVX2
+        #elif defined(__SSE2__) || defined(_M_AMD64) || defined(_M_X64) || \
+            (defined(_M_IX86_FP) && (_M_IX86_FP == 2))
+          #define XXH_VECTOR XXH_SSE2
+        #elif (defined(__PPC64__) && defined(__POWER8_VECTOR__)) || \
+            (defined(__s390x__) && defined(__VEC__)) &&             \
+                defined(__GNUC__)                           /* TODO: IBM XL */
+          #define XXH_VECTOR XXH_VSX
+        #else
+          #define XXH_VECTOR XXH_SCALAR
+        #endif
+      #endif
+
+      /* __ARM_FEATURE_SVE is only supported by GCC & Clang. */
+      #if (XXH_VECTOR == XXH_SVE) && !defined(__ARM_FEATURE_SVE)
+        #ifdef _MSC_VER
+          #pragma warning(once : 4606)
+        #else
+          #warning "__ARM_FEATURE_SVE isn't supported. Use SCALAR instead."
+        #endif
+        #undef XXH_VECTOR
+        #define XXH_VECTOR XXH_SCALAR
+      #endif
+
+      /*
+       * Controls the alignment of the accumulator,
+       * for compatibility with aligned vector loads, which are usually faster.
+       */
+      #ifndef XXH_ACC_ALIGN
+        #if defined(XXH_X86DISPATCH)
+          #define XXH_ACC_ALIGN 64         /* for compatibility with avx512 */
+        #elif XXH_VECTOR == XXH_SCALAR                            /* scalar */
+          #define XXH_ACC_ALIGN 8
+        #elif XXH_VECTOR == XXH_SSE2                                /* sse2 */
+          #define XXH_ACC_ALIGN 16
+        #elif XXH_VECTOR == XXH_AVX2                                /* avx2 */
+          #define XXH_ACC_ALIGN 32
+        #elif XXH_VECTOR == XXH_NEON                                /* neon */
+          #define XXH_ACC_ALIGN 16
+        #elif XXH_VECTOR == XXH_VSX                                  /* vsx */
+          #define XXH_ACC_ALIGN 16
+        #elif XXH_VECTOR == XXH_AVX512                            /* avx512 */
+          #define XXH_ACC_ALIGN 64
+        #elif XXH_VECTOR == XXH_SVE                                  /* sve */
+          #define XXH_ACC_ALIGN 64
+        #endif
+      #endif
+
+      #if defined(XXH_X86DISPATCH) || XXH_VECTOR == XXH_SSE2 || \
+          XXH_VECTOR == XXH_AVX2 || XXH_VECTOR == XXH_AVX512
+        #define XXH_SEC_ALIGN XXH_ACC_ALIGN
+      #elif XXH_VECTOR == XXH_SVE
+        #define XXH_SEC_ALIGN XXH_ACC_ALIGN
+      #else
+        #define XXH_SEC_ALIGN 8
+      #endif
+
+      #if defined(__GNUC__) || defined(__clang__)
+        #define XXH_ALIASING __attribute__((may_alias))
+      #else
+        #define XXH_ALIASING                                     /* nothing */
+      #endif
+
+      /*
+       * UGLY HACK:
+       * GCC usually generates the best code with -O3 for xxHash.
+       *
+       * However, when targeting AVX2, it is overzealous in its unrolling
+       * resulting in code roughly 3/4 the speed of Clang.
+       *
+       * There are other issues, such as GCC splitting _mm256_loadu_si256 into
+       * _mm_loadu_si128 + _mm256_inserti128_si256. This is an optimization
+       * which only applies to Sandy and Ivy Bridge... which don't even support
+       * AVX2.
+       *
+       * That is why when compiling the AVX2 version, it is recommended to use
+       * either -O2 -mavx2 -march=haswell or -O2 -mavx2
+       * -mno-avx256-split-unaligned-load for decent performance, or to use
+       * Clang instead.
+       *
+       * Fortunately, we can control the first one with a pragma that forces GCC
+       * into -O2, but the other one we can't control without "failed to inline
+       * always inline function due to target mismatch" warnings.
+       */
+      #if XXH_VECTOR == XXH_AVX2                      /* AVX2 */           \
+          && defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
+          && defined(__OPTIMIZE__) &&                                      \
+          XXH_SIZE_OPT <= 0                          /* respect -O0 and -Os */
+        #pragma GCC push_options
+        #pragma GCC optimize("-O2")
+      #endif
+
+      #if XXH_VECTOR == XXH_NEON
 
 /*
- * UGLY HACK: While AArch64 GCC on Linux does not seem to care, on macOS, GCC -O3
- * optimizes out the entire hashLong loop because of the aliasing violation.
+ * UGLY HACK: While AArch64 GCC on Linux does not seem to care, on macOS, GCC
+ * -O3 optimizes out the entire hashLong loop because of the aliasing violation.
  *
  * However, GCC is also inefficient at load-store optimization with vld1q/vst1q,
  * so the only option is to mark it as aliasing.
  */
 typedef uint64x2_t xxh_aliasing_uint64x2_t XXH_ALIASING;
 
-/*!
- * @internal
- * @brief `vld1q_u64` but faster and alignment-safe.
- *
- * On AArch64, unaligned access is always safe, but on ARMv7-a, it is only
- * *conditionally* safe (`vld1` has an alignment bit like `movdq[ua]` in x86).
- *
- * GCC for AArch64 sees `vld1q_u8` as an intrinsic instead of a load, so it
- * prohibits load-store optimizations. Therefore, a direct dereference is used.
- *
- * Otherwise, `vld1q_u8` is used with `vreinterpretq_u8_u64` to do a safe
- * unaligned load.
- */
-#if defined(__aarch64__) && defined(__GNUC__) && !defined(__clang__)
-XXH_FORCE_INLINE uint64x2_t XXH_vld1q_u64(void const* ptr) /* silence -Wcast-align */
+        /*!
+         * @internal
+         * @brief `vld1q_u64` but faster and alignment-safe.
+         *
+         * On AArch64, unaligned access is always safe, but on ARMv7-a, it is
+         * only *conditionally* safe (`vld1` has an alignment bit like
+         * `movdq[ua]` in x86).
+         *
+         * GCC for AArch64 sees `vld1q_u8` as an intrinsic instead of a load, so
+         * it prohibits load-store optimizations. Therefore, a direct
+         * dereference is used.
+         *
+         * Otherwise, `vld1q_u8` is used with `vreinterpretq_u8_u64` to do a
+         * safe unaligned load.
+         */
+        #if defined(__aarch64__) && defined(__GNUC__) && !defined(__clang__)
+XXH_FORCE_INLINE uint64x2_t
+XXH_vld1q_u64(void const *ptr)                      /* silence -Wcast-align */
 {
-    return *(xxh_aliasing_uint64x2_t const *)ptr;
+
+  return *(xxh_aliasing_uint64x2_t const *)ptr;
+
 }
-#else
-XXH_FORCE_INLINE uint64x2_t XXH_vld1q_u64(void const* ptr)
-{
-    return vreinterpretq_u64_u8(vld1q_u8((uint8_t const*)ptr));
+
+        #else
+XXH_FORCE_INLINE uint64x2_t XXH_vld1q_u64(void const *ptr) {
+
+  return vreinterpretq_u64_u8(vld1q_u8((uint8_t const *)ptr));
+
 }
-#endif
 
-/*!
- * @internal
- * @brief `vmlal_u32` on low and high halves of a vector.
- *
- * This is a workaround for AArch64 GCC < 11 which implemented arm_neon.h with
- * inline assembly and were therefore incapable of merging the `vget_{low, high}_u32`
- * with `vmlal_u32`.
- */
-#if defined(__aarch64__) && defined(__GNUC__) && !defined(__clang__) && __GNUC__ < 11
-XXH_FORCE_INLINE uint64x2_t
-XXH_vmlal_low_u32(uint64x2_t acc, uint32x4_t lhs, uint32x4_t rhs)
-{
-    /* Inline assembly is the only way */
-    __asm__("umlal   %0.2d, %1.2s, %2.2s" : "+w" (acc) : "w" (lhs), "w" (rhs));
-    return acc;
+        #endif
+
+        /*!
+         * @internal
+         * @brief `vmlal_u32` on low and high halves of a vector.
+         *
+         * This is a workaround for AArch64 GCC < 11 which implemented
+         * arm_neon.h with inline assembly and were therefore incapable of
+         * merging the `vget_{low, high}_u32` with `vmlal_u32`.
+         */
+        #if defined(__aarch64__) && defined(__GNUC__) && \
+            !defined(__clang__) && __GNUC__ < 11
+XXH_FORCE_INLINE uint64x2_t XXH_vmlal_low_u32(uint64x2_t acc, uint32x4_t lhs,
+                                              uint32x4_t rhs) {
+
+  /* Inline assembly is the only way */
+  __asm__("umlal   %0.2d, %1.2s, %2.2s" : "+w"(acc) : "w"(lhs), "w"(rhs));
+  return acc;
+
 }
-XXH_FORCE_INLINE uint64x2_t
-XXH_vmlal_high_u32(uint64x2_t acc, uint32x4_t lhs, uint32x4_t rhs)
-{
-    /* This intrinsic works as expected */
-    return vmlal_high_u32(acc, lhs, rhs);
+
+XXH_FORCE_INLINE uint64x2_t XXH_vmlal_high_u32(uint64x2_t acc, uint32x4_t lhs,
+                                               uint32x4_t rhs) {
+
+  /* This intrinsic works as expected */
+  return vmlal_high_u32(acc, lhs, rhs);
+
 }
-#else
+
+        #else
 /* Portable intrinsic versions */
-XXH_FORCE_INLINE uint64x2_t
-XXH_vmlal_low_u32(uint64x2_t acc, uint32x4_t lhs, uint32x4_t rhs)
-{
-    return vmlal_u32(acc, vget_low_u32(lhs), vget_low_u32(rhs));
+XXH_FORCE_INLINE uint64x2_t XXH_vmlal_low_u32(uint64x2_t acc, uint32x4_t lhs,
+                                              uint32x4_t rhs) {
+
+  return vmlal_u32(acc, vget_low_u32(lhs), vget_low_u32(rhs));
+
 }
+
 /*! @copydoc XXH_vmlal_low_u32
  * Assume the compiler converts this to vmlal_high_u32 on aarch64 */
-XXH_FORCE_INLINE uint64x2_t
-XXH_vmlal_high_u32(uint64x2_t acc, uint32x4_t lhs, uint32x4_t rhs)
-{
-    return vmlal_u32(acc, vget_high_u32(lhs), vget_high_u32(rhs));
-}
-#endif
+XXH_FORCE_INLINE uint64x2_t XXH_vmlal_high_u32(uint64x2_t acc, uint32x4_t lhs,
+                                               uint32x4_t rhs) {
 
-/*!
- * @ingroup tuning
- * @brief Controls the NEON to scalar ratio for XXH3
- *
- * This can be set to 2, 4, 6, or 8.
- *
- * ARM Cortex CPUs are _very_ sensitive to how their pipelines are used.
- *
- * For example, the Cortex-A73 can dispatch 3 micro-ops per cycle, but only 2 of those
- * can be NEON. If you are only using NEON instructions, you are only using 2/3 of the CPU
- * bandwidth.
- *
- * This is even more noticeable on the more advanced cores like the Cortex-A76 which
- * can dispatch 8 micro-ops per cycle, but still only 2 NEON micro-ops at once.
- *
- * Therefore, to make the most out of the pipeline, it is beneficial to run 6 NEON lanes
- * and 2 scalar lanes, which is chosen by default.
- *
- * This does not apply to Apple processors or 32-bit processors, which run better with
- * full NEON. These will default to 8. Additionally, size-optimized builds run 8 lanes.
- *
- * This change benefits CPUs with large micro-op buffers without negatively affecting
- * most other CPUs:
- *
- *  | Chipset               | Dispatch type       | NEON only | 6:2 hybrid | Diff. |
- *  |:----------------------|:--------------------|----------:|-----------:|------:|
- *  | Snapdragon 730 (A76)  | 2 NEON/8 micro-ops  |  8.8 GB/s |  10.1 GB/s |  ~16% |
- *  | Snapdragon 835 (A73)  | 2 NEON/3 micro-ops  |  5.1 GB/s |   5.3 GB/s |   ~5% |
- *  | Marvell PXA1928 (A53) | In-order dual-issue |  1.9 GB/s |   1.9 GB/s |    0% |
- *  | Apple M1              | 4 NEON/8 micro-ops  | 37.3 GB/s |  36.1 GB/s |  ~-3% |
- *
- * It also seems to fix some bad codegen on GCC, making it almost as fast as clang.
- *
- * When using WASM SIMD128, if this is 2 or 6, SIMDe will scalarize 2 of the lanes meaning
- * it effectively becomes worse 4.
- *
- * @see XXH3_accumulate_512_neon()
- */
-# ifndef XXH3_NEON_LANES
-#  if (defined(__aarch64__) || defined(__arm64__) || defined(_M_ARM64) || defined(_M_ARM64EC)) \
-   && !defined(__APPLE__) && XXH_SIZE_OPT <= 0
-#   define XXH3_NEON_LANES 6
-#  else
-#   define XXH3_NEON_LANES XXH_ACC_NB
-#  endif
-# endif
-#endif  /* XXH_VECTOR == XXH_NEON */
+  return vmlal_u32(acc, vget_high_u32(lhs), vget_high_u32(rhs));
 
-/*
- * VSX and Z Vector helpers.
- *
- * This is very messy, and any pull requests to clean this up are welcome.
- *
- * There are a lot of problems with supporting VSX and s390x, due to
- * inconsistent intrinsics, spotty coverage, and multiple endiannesses.
- */
-#if XXH_VECTOR == XXH_VSX
-/* Annoyingly, these headers _may_ define three macros: `bool`, `vector`,
- * and `pixel`. This is a problem for obvious reasons.
- *
- * These keywords are unnecessary; the spec literally says they are
- * equivalent to `__bool`, `__vector`, and `__pixel` and may be undef'd
- * after including the header.
- *
- * We use pragma push_macro/pop_macro to keep the namespace clean. */
-#  pragma push_macro("bool")
-#  pragma push_macro("vector")
-#  pragma push_macro("pixel")
-/* silence potential macro redefined warnings */
-#  undef bool
-#  undef vector
-#  undef pixel
+}
 
-#  if defined(__s390x__)
-#    include <s390intrin.h>
-#  else
-#    include <altivec.h>
-#  endif
+        #endif
 
-/* Restore the original macro values, if applicable. */
-#  pragma pop_macro("pixel")
-#  pragma pop_macro("vector")
-#  pragma pop_macro("bool")
+        /*!
+         * @ingroup tuning
+         * @brief Controls the NEON to scalar ratio for XXH3
+         *
+         * This can be set to 2, 4, 6, or 8.
+         *
+         * ARM Cortex CPUs are _very_ sensitive to how their pipelines are used.
+         *
+         * For example, the Cortex-A73 can dispatch 3 micro-ops per cycle, but
+         * only 2 of those can be NEON. If you are only using NEON instructions,
+         * you are only using 2/3 of the CPU bandwidth.
+         *
+         * This is even more noticeable on the more advanced cores like the
+         * Cortex-A76 which can dispatch 8 micro-ops per cycle, but still only 2
+         * NEON micro-ops at once.
+         *
+         * Therefore, to make the most out of the pipeline, it is beneficial to
+         * run 6 NEON lanes and 2 scalar lanes, which is chosen by default.
+         *
+         * This does not apply to Apple processors or 32-bit processors, which
+         * run better with full NEON. These will default to 8. Additionally,
+         * size-optimized builds run 8 lanes.
+         *
+         * This change benefits CPUs with large micro-op buffers without
+         * negatively affecting most other CPUs:
+         *
+         *  | Chipset               | Dispatch type       | NEON only | 6:2
+         * hybrid | Diff. |
+         *  |:----------------------|:--------------------|----------:|-----------:|------:|
+         *  | Snapdragon 730 (A76)  | 2 NEON/8 micro-ops  |  8.8 GB/s |  10.1
+         * GB/s |  ~16% | | Snapdragon 835 (A73)  | 2 NEON/3 micro-ops  |  5.1
+         * GB/s |   5.3 GB/s |   ~5% | | Marvell PXA1928 (A53) | In-order
+         * dual-issue |  1.9 GB/s |   1.9 GB/s |    0% | | Apple M1 | 4 NEON/8
+         * micro-ops  | 37.3 GB/s |  36.1 GB/s |  ~-3% |
+         *
+         * It also seems to fix some bad codegen on GCC, making it almost as
+         * fast as clang.
+         *
+         * When using WASM SIMD128, if this is 2 or 6, SIMDe will scalarize 2 of
+         * the lanes meaning it effectively becomes worse 4.
+         *
+         * @see XXH3_accumulate_512_neon()
+         */
+        #ifndef XXH3_NEON_LANES
+          #if (defined(__aarch64__) || defined(__arm64__) || \
+               defined(_M_ARM64) || defined(_M_ARM64EC)) &&  \
+              !defined(__APPLE__) && XXH_SIZE_OPT <= 0
+            #define XXH3_NEON_LANES 6
+          #else
+            #define XXH3_NEON_LANES XXH_ACC_NB
+          #endif
+        #endif
+      #endif                                      /* XXH_VECTOR == XXH_NEON */
+
+      /*
+       * VSX and Z Vector helpers.
+       *
+       * This is very messy, and any pull requests to clean this up are welcome.
+       *
+       * There are a lot of problems with supporting VSX and s390x, due to
+       * inconsistent intrinsics, spotty coverage, and multiple endiannesses.
+       */
+      #if XXH_VECTOR == XXH_VSX
+        /* Annoyingly, these headers _may_ define three macros: `bool`,
+         * `vector`, and `pixel`. This is a problem for obvious reasons.
+         *
+         * These keywords are unnecessary; the spec literally says they are
+         * equivalent to `__bool`, `__vector`, and `__pixel` and may be undef'd
+         * after including the header.
+         *
+         * We use pragma push_macro/pop_macro to keep the namespace clean. */
+        #pragma push_macro("bool")
+        #pragma push_macro("vector")
+        #pragma push_macro("pixel")
+        /* silence potential macro redefined warnings */
+        #undef bool
+        #undef vector
+        #undef pixel
+
+        #if defined(__s390x__)
+          #include <s390intrin.h>
+        #else
+          #include <altivec.h>
+        #endif
+
+        /* Restore the original macro values, if applicable. */
+        #pragma pop_macro("pixel")
+        #pragma pop_macro("vector")
+        #pragma pop_macro("bool")
 
 typedef __vector unsigned long long xxh_u64x2;
-typedef __vector unsigned char xxh_u8x16;
-typedef __vector unsigned xxh_u32x4;
+typedef __vector unsigned char      xxh_u8x16;
+typedef __vector unsigned           xxh_u32x4;
 
 /*
- * UGLY HACK: Similar to aarch64 macOS GCC, s390x GCC has the same aliasing issue.
+ * UGLY HACK: Similar to aarch64 macOS GCC, s390x GCC has the same aliasing
+ * issue.
  */
 typedef xxh_u64x2 xxh_aliasing_u64x2 XXH_ALIASING;
 
-# ifndef XXH_VSX_BE
-#  if defined(__BIG_ENDIAN__) \
-  || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
-#    define XXH_VSX_BE 1
-#  elif defined(__VEC_ELEMENT_REG_ORDER__) && __VEC_ELEMENT_REG_ORDER__ == __ORDER_BIG_ENDIAN__
-#    warning "-maltivec=be is not recommended. Please use native endianness."
-#    define XXH_VSX_BE 1
-#  else
-#    define XXH_VSX_BE 0
-#  endif
-# endif /* !defined(XXH_VSX_BE) */
-
-# if XXH_VSX_BE
-#  if defined(__POWER9_VECTOR__) || (defined(__clang__) && defined(__s390x__))
-#    define XXH_vec_revb vec_revb
-#  else
+        #ifndef XXH_VSX_BE
+          #if defined(__BIG_ENDIAN__) ||  \
+              (defined(__BYTE_ORDER__) && \
+               __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
+            #define XXH_VSX_BE 1
+          #elif defined(__VEC_ELEMENT_REG_ORDER__) && \
+              __VEC_ELEMENT_REG_ORDER__ == __ORDER_BIG_ENDIAN__
+            #warning \
+                "-maltivec=be is not recommended. Please use native endianness."
+            #define XXH_VSX_BE 1
+          #else
+            #define XXH_VSX_BE 0
+          #endif
+        #endif                                      /* !defined(XXH_VSX_BE) */
+
+        #if XXH_VSX_BE
+          #if defined(__POWER9_VECTOR__) || \
+              (defined(__clang__) && defined(__s390x__))
+            #define XXH_vec_revb vec_revb
+          #else
 /*!
  * A polyfill for POWER9's vec_revb().
  */
-XXH_FORCE_INLINE xxh_u64x2 XXH_vec_revb(xxh_u64x2 val)
-{
-    xxh_u8x16 const vByteSwap = { 0x07, 0x06, 0x05, 0x04, 0x03, 0x02, 0x01, 0x00,
-                                  0x0F, 0x0E, 0x0D, 0x0C, 0x0B, 0x0A, 0x09, 0x08 };
-    return vec_perm(val, val, vByteSwap);
+XXH_FORCE_INLINE xxh_u64x2 XXH_vec_revb(xxh_u64x2 val) {
+
+  xxh_u8x16 const vByteSwap = {0x07, 0x06, 0x05, 0x04, 0x03, 0x02, 0x01, 0x00,
+                               0x0F, 0x0E, 0x0D, 0x0C, 0x0B, 0x0A, 0x09, 0x08};
+  return vec_perm(val, val, vByteSwap);
+
 }
-#  endif
-# endif /* XXH_VSX_BE */
+
+          #endif
+        #endif                                                /* XXH_VSX_BE */
 
 /*!
  * Performs an unaligned vector load and byte swaps it on big endian.
  */
-XXH_FORCE_INLINE xxh_u64x2 XXH_vec_loadu(const void *ptr)
-{
-    xxh_u64x2 ret;
-    XXH_memcpy(&ret, ptr, sizeof(xxh_u64x2));
-# if XXH_VSX_BE
-    ret = XXH_vec_revb(ret);
-# endif
-    return ret;
-}
+XXH_FORCE_INLINE xxh_u64x2 XXH_vec_loadu(const void *ptr) {
 
-/*
- * vec_mulo and vec_mule are very problematic intrinsics on PowerPC
- *
- * These intrinsics weren't added until GCC 8, despite existing for a while,
- * and they are endian dependent. Also, their meaning swap depending on version.
- * */
-# if defined(__s390x__)
- /* s390x is always big endian, no issue on this platform */
-#  define XXH_vec_mulo vec_mulo
-#  define XXH_vec_mule vec_mule
-# elif defined(__clang__) && XXH_HAS_BUILTIN(__builtin_altivec_vmuleuw) && !defined(__ibmxl__)
-/* Clang has a better way to control this, we can just use the builtin which doesn't swap. */
- /* The IBM XL Compiler (which defined __clang__) only implements the vec_* operations */
-#  define XXH_vec_mulo __builtin_altivec_vmulouw
-#  define XXH_vec_mule __builtin_altivec_vmuleuw
-# else
-/* gcc needs inline assembly */
-/* Adapted from https://github.com/google/highwayhash/blob/master/highwayhash/hh_vsx.h. */
-XXH_FORCE_INLINE xxh_u64x2 XXH_vec_mulo(xxh_u32x4 a, xxh_u32x4 b)
-{
-    xxh_u64x2 result;
-    __asm__("vmulouw %0, %1, %2" : "=v" (result) : "v" (a), "v" (b));
-    return result;
-}
-XXH_FORCE_INLINE xxh_u64x2 XXH_vec_mule(xxh_u32x4 a, xxh_u32x4 b)
-{
-    xxh_u64x2 result;
-    __asm__("vmuleuw %0, %1, %2" : "=v" (result) : "v" (a), "v" (b));
-    return result;
-}
-# endif /* XXH_vec_mulo, XXH_vec_mule */
-#endif /* XXH_VECTOR == XXH_VSX */
-
-#if XXH_VECTOR == XXH_SVE
-#define ACCRND(acc, offset) \
-do { \
-    svuint64_t input_vec = svld1_u64(mask, xinput + offset);         \
-    svuint64_t secret_vec = svld1_u64(mask, xsecret + offset);       \
-    svuint64_t mixed = sveor_u64_x(mask, secret_vec, input_vec);     \
-    svuint64_t swapped = svtbl_u64(input_vec, kSwap);                \
-    svuint64_t mixed_lo = svextw_u64_x(mask, mixed);                 \
-    svuint64_t mixed_hi = svlsr_n_u64_x(mask, mixed, 32);            \
-    svuint64_t mul = svmad_u64_x(mask, mixed_lo, mixed_hi, swapped); \
-    acc = svadd_u64_x(mask, acc, mul);                               \
-} while (0)
-#endif /* XXH_VECTOR == XXH_SVE */
-
-/* prefetch
- * can be disabled, by declaring XXH_NO_PREFETCH build macro */
-#if defined(XXH_NO_PREFETCH)
-#  define XXH_PREFETCH(ptr)  (void)(ptr)  /* disabled */
-#else
-#  if XXH_SIZE_OPT >= 1
-#    define XXH_PREFETCH(ptr) (void)(ptr)
-#  elif defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86))  /* _mm_prefetch() not defined outside of x86/x64 */
-#    include <mmintrin.h>   /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
-#    define XXH_PREFETCH(ptr)  _mm_prefetch((const char*)(ptr), _MM_HINT_T0)
-#  elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) )
-#    define XXH_PREFETCH(ptr)  __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */)
-#  else
-#    define XXH_PREFETCH(ptr) (void)(ptr)  /* disabled */
-#  endif
-#endif  /* XXH_NO_PREFETCH */
+  xxh_u64x2 ret;
+  XXH_memcpy(&ret, ptr, sizeof(xxh_u64x2));
+        #if XXH_VSX_BE
+  ret = XXH_vec_revb(ret);
+        #endif
+  return ret;
 
+}
 
-/* ==========================================
- * XXH3 default settings
- * ========================================== */
-
-#define XXH_SECRET_DEFAULT_SIZE 192   /* minimum XXH3_SECRET_SIZE_MIN */
-
-#if (XXH_SECRET_DEFAULT_SIZE < XXH3_SECRET_SIZE_MIN)
-#  error "default keyset is not large enough"
-#endif
+        /*
+         * vec_mulo and vec_mule are very problematic intrinsics on PowerPC
+         *
+         * These intrinsics weren't added until GCC 8, despite existing for a
+         * while, and they are endian dependent. Also, their meaning swap
+         * depending on version.
+         * */
+        #if defined(__s390x__)
+        /* s390x is always big endian, no issue on this platform */
+          #define XXH_vec_mulo vec_mulo
+          #define XXH_vec_mule vec_mule
+        #elif defined(__clang__) && \
+            XXH_HAS_BUILTIN(__builtin_altivec_vmuleuw) && !defined(__ibmxl__)
+        /* Clang has a better way to control this, we can just use the builtin
+         * which doesn't swap. */
+        /* The IBM XL Compiler (which defined __clang__) only implements the
+         * vec_* operations */
+          #define XXH_vec_mulo __builtin_altivec_vmulouw
+          #define XXH_vec_mule __builtin_altivec_vmuleuw
+        #else
+/* gcc needs inline assembly */
+/* Adapted from
+ * https://github.com/google/highwayhash/blob/master/highwayhash/hh_vsx.h. */
+XXH_FORCE_INLINE xxh_u64x2 XXH_vec_mulo(xxh_u32x4 a, xxh_u32x4 b) {
+
+  xxh_u64x2 result;
+  __asm__("vmulouw %0, %1, %2" : "=v"(result) : "v"(a), "v"(b));
+  return result;
+
+}
+
+XXH_FORCE_INLINE xxh_u64x2 XXH_vec_mule(xxh_u32x4 a, xxh_u32x4 b) {
+
+  xxh_u64x2 result;
+  __asm__("vmuleuw %0, %1, %2" : "=v"(result) : "v"(a), "v"(b));
+  return result;
+
+}
+
+        #endif                                /* XXH_vec_mulo, XXH_vec_mule */
+      #endif                                       /* XXH_VECTOR == XXH_VSX */
+
+      #if XXH_VECTOR == XXH_SVE
+        #define ACCRND(acc, offset)                                          \
+          do {                                                               \
+                                                                             \
+            svuint64_t input_vec = svld1_u64(mask, xinput + offset);         \
+            svuint64_t secret_vec = svld1_u64(mask, xsecret + offset);       \
+            svuint64_t mixed = sveor_u64_x(mask, secret_vec, input_vec);     \
+            svuint64_t swapped = svtbl_u64(input_vec, kSwap);                \
+            svuint64_t mixed_lo = svextw_u64_x(mask, mixed);                 \
+            svuint64_t mixed_hi = svlsr_n_u64_x(mask, mixed, 32);            \
+            svuint64_t mul = svmad_u64_x(mask, mixed_lo, mixed_hi, swapped); \
+            acc = svadd_u64_x(mask, acc, mul);                               \
+                                                                             \
+          } while (0)
+      #endif                                       /* XXH_VECTOR == XXH_SVE */
+
+      /* prefetch
+       * can be disabled, by declaring XXH_NO_PREFETCH build macro */
+      #if defined(XXH_NO_PREFETCH)
+        #define XXH_PREFETCH(ptr) (void)(ptr)                   /* disabled */
+      #else
+        #if XXH_SIZE_OPT >= 1
+          #define XXH_PREFETCH(ptr) (void)(ptr)
+        #elif defined(_MSC_VER) && \
+            (defined(_M_X64) ||    \
+             defined(              \
+                 _M_IX86)) /* _mm_prefetch() not defined outside of x86/x64 */
+          #include <mmintrin.h> /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
+          #define XXH_PREFETCH(ptr) \
+            _mm_prefetch((const char *)(ptr), _MM_HINT_T0)
+        #elif defined(__GNUC__) && \
+            ((__GNUC__ >= 4) || ((__GNUC__ == 3) && (__GNUC_MINOR__ >= 1)))
+          #define XXH_PREFETCH(ptr) \
+            __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */)
+        #else
+          #define XXH_PREFETCH(ptr) (void)(ptr)                 /* disabled */
+        #endif
+      #endif                                             /* XXH_NO_PREFETCH */
+
+    /* ==========================================
+     * XXH3 default settings
+     * ========================================== */
+
+      #define XXH_SECRET_DEFAULT_SIZE 192   /* minimum XXH3_SECRET_SIZE_MIN */
+
+      #if (XXH_SECRET_DEFAULT_SIZE < XXH3_SECRET_SIZE_MIN)
+        #error "default keyset is not large enough"
+      #endif
 
 /*! Pseudorandom secret taken directly from FARSH. */
-XXH_ALIGN(64) static const xxh_u8 XXH3_kSecret[XXH_SECRET_DEFAULT_SIZE] = {
-    0xb8, 0xfe, 0x6c, 0x39, 0x23, 0xa4, 0x4b, 0xbe, 0x7c, 0x01, 0x81, 0x2c, 0xf7, 0x21, 0xad, 0x1c,
-    0xde, 0xd4, 0x6d, 0xe9, 0x83, 0x90, 0x97, 0xdb, 0x72, 0x40, 0xa4, 0xa4, 0xb7, 0xb3, 0x67, 0x1f,
-    0xcb, 0x79, 0xe6, 0x4e, 0xcc, 0xc0, 0xe5, 0x78, 0x82, 0x5a, 0xd0, 0x7d, 0xcc, 0xff, 0x72, 0x21,
-    0xb8, 0x08, 0x46, 0x74, 0xf7, 0x43, 0x24, 0x8e, 0xe0, 0x35, 0x90, 0xe6, 0x81, 0x3a, 0x26, 0x4c,
-    0x3c, 0x28, 0x52, 0xbb, 0x91, 0xc3, 0x00, 0xcb, 0x88, 0xd0, 0x65, 0x8b, 0x1b, 0x53, 0x2e, 0xa3,
-    0x71, 0x64, 0x48, 0x97, 0xa2, 0x0d, 0xf9, 0x4e, 0x38, 0x19, 0xef, 0x46, 0xa9, 0xde, 0xac, 0xd8,
-    0xa8, 0xfa, 0x76, 0x3f, 0xe3, 0x9c, 0x34, 0x3f, 0xf9, 0xdc, 0xbb, 0xc7, 0xc7, 0x0b, 0x4f, 0x1d,
-    0x8a, 0x51, 0xe0, 0x4b, 0xcd, 0xb4, 0x59, 0x31, 0xc8, 0x9f, 0x7e, 0xc9, 0xd9, 0x78, 0x73, 0x64,
-    0xea, 0xc5, 0xac, 0x83, 0x34, 0xd3, 0xeb, 0xc3, 0xc5, 0x81, 0xa0, 0xff, 0xfa, 0x13, 0x63, 0xeb,
-    0x17, 0x0d, 0xdd, 0x51, 0xb7, 0xf0, 0xda, 0x49, 0xd3, 0x16, 0x55, 0x26, 0x29, 0xd4, 0x68, 0x9e,
-    0x2b, 0x16, 0xbe, 0x58, 0x7d, 0x47, 0xa1, 0xfc, 0x8f, 0xf8, 0xb8, 0xd1, 0x7a, 0xd0, 0x31, 0xce,
-    0x45, 0xcb, 0x3a, 0x8f, 0x95, 0x16, 0x04, 0x28, 0xaf, 0xd7, 0xfb, 0xca, 0xbb, 0x4b, 0x40, 0x7e,
+XXH_ALIGN(64)
+static const xxh_u8 XXH3_kSecret[XXH_SECRET_DEFAULT_SIZE] = {
+
+    0xb8, 0xfe, 0x6c, 0x39, 0x23, 0xa4, 0x4b, 0xbe, 0x7c, 0x01, 0x81, 0x2c,
+    0xf7, 0x21, 0xad, 0x1c, 0xde, 0xd4, 0x6d, 0xe9, 0x83, 0x90, 0x97, 0xdb,
+    0x72, 0x40, 0xa4, 0xa4, 0xb7, 0xb3, 0x67, 0x1f, 0xcb, 0x79, 0xe6, 0x4e,
+    0xcc, 0xc0, 0xe5, 0x78, 0x82, 0x5a, 0xd0, 0x7d, 0xcc, 0xff, 0x72, 0x21,
+    0xb8, 0x08, 0x46, 0x74, 0xf7, 0x43, 0x24, 0x8e, 0xe0, 0x35, 0x90, 0xe6,
+    0x81, 0x3a, 0x26, 0x4c, 0x3c, 0x28, 0x52, 0xbb, 0x91, 0xc3, 0x00, 0xcb,
+    0x88, 0xd0, 0x65, 0x8b, 0x1b, 0x53, 0x2e, 0xa3, 0x71, 0x64, 0x48, 0x97,
+    0xa2, 0x0d, 0xf9, 0x4e, 0x38, 0x19, 0xef, 0x46, 0xa9, 0xde, 0xac, 0xd8,
+    0xa8, 0xfa, 0x76, 0x3f, 0xe3, 0x9c, 0x34, 0x3f, 0xf9, 0xdc, 0xbb, 0xc7,
+    0xc7, 0x0b, 0x4f, 0x1d, 0x8a, 0x51, 0xe0, 0x4b, 0xcd, 0xb4, 0x59, 0x31,
+    0xc8, 0x9f, 0x7e, 0xc9, 0xd9, 0x78, 0x73, 0x64, 0xea, 0xc5, 0xac, 0x83,
+    0x34, 0xd3, 0xeb, 0xc3, 0xc5, 0x81, 0xa0, 0xff, 0xfa, 0x13, 0x63, 0xeb,
+    0x17, 0x0d, 0xdd, 0x51, 0xb7, 0xf0, 0xda, 0x49, 0xd3, 0x16, 0x55, 0x26,
+    0x29, 0xd4, 0x68, 0x9e, 0x2b, 0x16, 0xbe, 0x58, 0x7d, 0x47, 0xa1, 0xfc,
+    0x8f, 0xf8, 0xb8, 0xd1, 0x7a, 0xd0, 0x31, 0xce, 0x45, 0xcb, 0x3a, 0x8f,
+    0x95, 0x16, 0x04, 0x28, 0xaf, 0xd7, 0xfb, 0xca, 0xbb, 0x4b, 0x40, 0x7e,
+
 };
 
-static const xxh_u64 PRIME_MX1 = 0x165667919E3779F9ULL;  /*!< 0b0001011001010110011001111001000110011110001101110111100111111001 */
-static const xxh_u64 PRIME_MX2 = 0x9FB21C651E98DF25ULL;  /*!< 0b1001111110110010000111000110010100011110100110001101111100100101 */
+static const xxh_u64 PRIME_MX1 = 0x165667919E3779F9ULL; /*!<
+                                                           0b0001011001010110011001111001000110011110001101110111100111111001
+                                                         */
+static const xxh_u64 PRIME_MX2 = 0x9FB21C651E98DF25ULL; /*!<
+                                                           0b1001111110110010000111000110010100011110100110001101111100100101
+                                                         */
 
-#ifdef XXH_OLD_NAMES
-#  define kSecret XXH3_kSecret
-#endif
+      #ifdef XXH_OLD_NAMES
+        #define kSecret XXH3_kSecret
+      #endif
 
-#ifdef XXH_DOXYGEN
+      #ifdef XXH_DOXYGEN
 /*!
  * @brief Calculates a 32-bit to 64-bit long multiply.
  *
  * Implemented as a macro.
  *
- * Wraps `__emulu` on MSVC x86 because it tends to call `__allmul` when it doesn't
- * need to (but it shouldn't need to anyways, it is about 7 instructions to do
- * a 64x64 multiply...). Since we know that this will _always_ emit `MULL`, we
- * use that instead of the normal method.
+ * Wraps `__emulu` on MSVC x86 because it tends to call `__allmul` when it
+ * doesn't need to (but it shouldn't need to anyways, it is about 7 instructions
+ * to do a 64x64 multiply...). Since we know that this will _always_ emit
+ * `MULL`, we use that instead of the normal method.
  *
- * If you are compiling for platforms like Thumb-1 and don't have a better option,
- * you may also want to write your own long multiply routine here.
+ * If you are compiling for platforms like Thumb-1 and don't have a better
+ * option, you may also want to write your own long multiply routine here.
  *
  * @param x, y Numbers to be multiplied
  * @return 64-bit product of the low 32 bits of @p x and @p y.
  */
-XXH_FORCE_INLINE xxh_u64
-XXH_mult32to64(xxh_u64 x, xxh_u64 y)
-{
-   return (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF);
+XXH_FORCE_INLINE xxh_u64 XXH_mult32to64(xxh_u64 x, xxh_u64 y) {
+
+  return (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF);
+
 }
-#elif defined(_MSC_VER) && defined(_M_IX86)
-#    define XXH_mult32to64(x, y) __emulu((unsigned)(x), (unsigned)(y))
-#else
-/*
- * Downcast + upcast is usually better than masking on older compilers like
- * GCC 4.2 (especially 32-bit ones), all without affecting newer compilers.
- *
- * The other method, (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF), will AND both operands
- * and perform a full 64x64 multiply -- entirely redundant on 32-bit.
- */
-#    define XXH_mult32to64(x, y) ((xxh_u64)(xxh_u32)(x) * (xxh_u64)(xxh_u32)(y))
-#endif
+
+      #elif defined(_MSC_VER) && defined(_M_IX86)
+        #define XXH_mult32to64(x, y) __emulu((unsigned)(x), (unsigned)(y))
+      #else
+      /*
+       * Downcast + upcast is usually better than masking on older compilers
+       * like GCC 4.2 (especially 32-bit ones), all without affecting newer
+       * compilers.
+       *
+       * The other method, (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF), will AND both
+       * operands and perform a full 64x64 multiply -- entirely redundant on
+       * 32-bit.
+       */
+        #define XXH_mult32to64(x, y) \
+          ((xxh_u64)(xxh_u32)(x) * (xxh_u64)(xxh_u32)(y))
+      #endif
 
 /*!
  * @brief Calculates a 64->128-bit long multiply.
@@ -4302,164 +4796,170 @@ XXH_mult32to64(xxh_u64 x, xxh_u64 y)
  * @param lhs , rhs The 64-bit integers to be multiplied
  * @return The 128-bit result represented in an @ref XXH128_hash_t.
  */
-static XXH128_hash_t
-XXH_mult64to128(xxh_u64 lhs, xxh_u64 rhs)
-{
-    /*
-     * GCC/Clang __uint128_t method.
-     *
-     * On most 64-bit targets, GCC and Clang define a __uint128_t type.
-     * This is usually the best way as it usually uses a native long 64-bit
-     * multiply, such as MULQ on x86_64 or MUL + UMULH on aarch64.
-     *
-     * Usually.
-     *
-     * Despite being a 32-bit platform, Clang (and emscripten) define this type
-     * despite not having the arithmetic for it. This results in a laggy
-     * compiler builtin call which calculates a full 128-bit multiply.
-     * In that case it is best to use the portable one.
-     * https://github.com/Cyan4973/xxHash/issues/211#issuecomment-515575677
-     */
-#if (defined(__GNUC__) || defined(__clang__)) && !defined(__wasm__) \
-    && defined(__SIZEOF_INT128__) \
-    || (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
+static XXH128_hash_t XXH_mult64to128(xxh_u64 lhs, xxh_u64 rhs) {
+
+      /*
+       * GCC/Clang __uint128_t method.
+       *
+       * On most 64-bit targets, GCC and Clang define a __uint128_t type.
+       * This is usually the best way as it usually uses a native long 64-bit
+       * multiply, such as MULQ on x86_64 or MUL + UMULH on aarch64.
+       *
+       * Usually.
+       *
+       * Despite being a 32-bit platform, Clang (and emscripten) define this
+       * type despite not having the arithmetic for it. This results in a laggy
+       * compiler builtin call which calculates a full 128-bit multiply.
+       * In that case it is best to use the portable one.
+       * https://github.com/Cyan4973/xxHash/issues/211#issuecomment-515575677
+       */
+      #if (defined(__GNUC__) || defined(__clang__)) && !defined(__wasm__) && \
+              defined(__SIZEOF_INT128__) ||                                  \
+          (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
+
+  __uint128_t const product = (__uint128_t)lhs * (__uint128_t)rhs;
+  XXH128_hash_t     r128;
+  r128.low64 = (xxh_u64)(product);
+  r128.high64 = (xxh_u64)(product >> 64);
+  return r128;
 
-    __uint128_t const product = (__uint128_t)lhs * (__uint128_t)rhs;
-    XXH128_hash_t r128;
-    r128.low64  = (xxh_u64)(product);
-    r128.high64 = (xxh_u64)(product >> 64);
-    return r128;
-
-    /*
-     * MSVC for x64's _umul128 method.
-     *
-     * xxh_u64 _umul128(xxh_u64 Multiplier, xxh_u64 Multiplicand, xxh_u64 *HighProduct);
-     *
-     * This compiles to single operand MUL on x64.
-     */
-#elif (defined(_M_X64) || defined(_M_IA64)) && !defined(_M_ARM64EC)
-
-#ifndef _MSC_VER
-#   pragma intrinsic(_umul128)
-#endif
-    xxh_u64 product_high;
-    xxh_u64 const product_low = _umul128(lhs, rhs, &product_high);
-    XXH128_hash_t r128;
-    r128.low64  = product_low;
-    r128.high64 = product_high;
-    return r128;
-
-    /*
-     * MSVC for ARM64's __umulh method.
-     *
-     * This compiles to the same MUL + UMULH as GCC/Clang's __uint128_t method.
-     */
-#elif defined(_M_ARM64) || defined(_M_ARM64EC)
-
-#ifndef _MSC_VER
-#   pragma intrinsic(__umulh)
-#endif
-    XXH128_hash_t r128;
-    r128.low64  = lhs * rhs;
-    r128.high64 = __umulh(lhs, rhs);
-    return r128;
+        /*
+         * MSVC for x64's _umul128 method.
+         *
+         * xxh_u64 _umul128(xxh_u64 Multiplier, xxh_u64 Multiplicand, xxh_u64
+         * *HighProduct);
+         *
+         * This compiles to single operand MUL on x64.
+         */
+      #elif (defined(_M_X64) || defined(_M_IA64)) && !defined(_M_ARM64EC)
+
+        #ifndef _MSC_VER
+          #pragma intrinsic(_umul128)
+        #endif
+  xxh_u64       product_high;
+  xxh_u64 const product_low = _umul128(lhs, rhs, &product_high);
+  XXH128_hash_t r128;
+  r128.low64 = product_low;
+  r128.high64 = product_high;
+  return r128;
 
-#else
-    /*
-     * Portable scalar method. Optimized for 32-bit and 64-bit ALUs.
-     *
-     * This is a fast and simple grade school multiply, which is shown below
-     * with base 10 arithmetic instead of base 0x100000000.
-     *
-     *           9 3 // D2 lhs = 93
-     *         x 7 5 // D2 rhs = 75
-     *     ----------
-     *           1 5 // D2 lo_lo = (93 % 10) * (75 % 10) = 15
-     *         4 5 | // D2 hi_lo = (93 / 10) * (75 % 10) = 45
-     *         2 1 | // D2 lo_hi = (93 % 10) * (75 / 10) = 21
-     *     + 6 3 | | // D2 hi_hi = (93 / 10) * (75 / 10) = 63
-     *     ---------
-     *         2 7 | // D2 cross = (15 / 10) + (45 % 10) + 21 = 27
-     *     + 6 7 | | // D2 upper = (27 / 10) + (45 / 10) + 63 = 67
-     *     ---------
-     *       6 9 7 5 // D4 res = (27 * 10) + (15 % 10) + (67 * 100) = 6975
-     *
-     * The reasons for adding the products like this are:
-     *  1. It avoids manual carry tracking. Just like how
-     *     (9 * 9) + 9 + 9 = 99, the same applies with this for UINT64_MAX.
-     *     This avoids a lot of complexity.
-     *
-     *  2. It hints for, and on Clang, compiles to, the powerful UMAAL
-     *     instruction available in ARM's Digital Signal Processing extension
-     *     in 32-bit ARMv6 and later, which is shown below:
-     *
-     *         void UMAAL(xxh_u32 *RdLo, xxh_u32 *RdHi, xxh_u32 Rn, xxh_u32 Rm)
-     *         {
-     *             xxh_u64 product = (xxh_u64)*RdLo * (xxh_u64)*RdHi + Rn + Rm;
-     *             *RdLo = (xxh_u32)(product & 0xFFFFFFFF);
-     *             *RdHi = (xxh_u32)(product >> 32);
-     *         }
-     *
-     *     This instruction was designed for efficient long multiplication, and
-     *     allows this to be calculated in only 4 instructions at speeds
-     *     comparable to some 64-bit ALUs.
-     *
-     *  3. It isn't terrible on other platforms. Usually this will be a couple
-     *     of 32-bit ADD/ADCs.
-     */
+        /*
+         * MSVC for ARM64's __umulh method.
+         *
+         * This compiles to the same MUL + UMULH as GCC/Clang's __uint128_t
+         * method.
+         */
+      #elif defined(_M_ARM64) || defined(_M_ARM64EC)
+
+        #ifndef _MSC_VER
+          #pragma intrinsic(__umulh)
+        #endif
+  XXH128_hash_t r128;
+  r128.low64 = lhs * rhs;
+  r128.high64 = __umulh(lhs, rhs);
+  return r128;
+
+      #else
+  /*
+   * Portable scalar method. Optimized for 32-bit and 64-bit ALUs.
+   *
+   * This is a fast and simple grade school multiply, which is shown below
+   * with base 10 arithmetic instead of base 0x100000000.
+   *
+   *           9 3 // D2 lhs = 93
+   *         x 7 5 // D2 rhs = 75
+   *     ----------
+   *           1 5 // D2 lo_lo = (93 % 10) * (75 % 10) = 15
+   *         4 5 | // D2 hi_lo = (93 / 10) * (75 % 10) = 45
+   *         2 1 | // D2 lo_hi = (93 % 10) * (75 / 10) = 21
+   *     + 6 3 | | // D2 hi_hi = (93 / 10) * (75 / 10) = 63
+   *     ---------
+   *         2 7 | // D2 cross = (15 / 10) + (45 % 10) + 21 = 27
+   *     + 6 7 | | // D2 upper = (27 / 10) + (45 / 10) + 63 = 67
+   *     ---------
+   *       6 9 7 5 // D4 res = (27 * 10) + (15 % 10) + (67 * 100) = 6975
+   *
+   * The reasons for adding the products like this are:
+   *  1. It avoids manual carry tracking. Just like how
+   *     (9 * 9) + 9 + 9 = 99, the same applies with this for UINT64_MAX.
+   *     This avoids a lot of complexity.
+   *
+   *  2. It hints for, and on Clang, compiles to, the powerful UMAAL
+   *     instruction available in ARM's Digital Signal Processing extension
+   *     in 32-bit ARMv6 and later, which is shown below:
+   *
+   *         void UMAAL(xxh_u32 *RdLo, xxh_u32 *RdHi, xxh_u32 Rn, xxh_u32 Rm)
+   *         {
+
+   *             xxh_u64 product = (xxh_u64)*RdLo * (xxh_u64)*RdHi + Rn + Rm;
+   *             *RdLo = (xxh_u32)(product & 0xFFFFFFFF);
+   *             *RdHi = (xxh_u32)(product >> 32);
+   *         }
+   *
+   *     This instruction was designed for efficient long multiplication, and
+   *     allows this to be calculated in only 4 instructions at speeds
+   *     comparable to some 64-bit ALUs.
+   *
+   *  3. It isn't terrible on other platforms. Usually this will be a couple
+   *     of 32-bit ADD/ADCs.
+   */
+
+  /* First calculate all of the cross products. */
+  xxh_u64 const lo_lo = XXH_mult32to64(lhs & 0xFFFFFFFF, rhs & 0xFFFFFFFF);
+  xxh_u64 const hi_lo = XXH_mult32to64(lhs >> 32, rhs & 0xFFFFFFFF);
+  xxh_u64 const lo_hi = XXH_mult32to64(lhs & 0xFFFFFFFF, rhs >> 32);
+  xxh_u64 const hi_hi = XXH_mult32to64(lhs >> 32, rhs >> 32);
+
+  /* Now add the products together. These will never overflow. */
+  xxh_u64 const cross = (lo_lo >> 32) + (hi_lo & 0xFFFFFFFF) + lo_hi;
+  xxh_u64 const upper = (hi_lo >> 32) + (cross >> 32) + hi_hi;
+  xxh_u64 const lower = (cross << 32) | (lo_lo & 0xFFFFFFFF);
+
+  XXH128_hash_t r128;
+  r128.low64 = lower;
+  r128.high64 = upper;
+  return r128;
+      #endif
 
-    /* First calculate all of the cross products. */
-    xxh_u64 const lo_lo = XXH_mult32to64(lhs & 0xFFFFFFFF, rhs & 0xFFFFFFFF);
-    xxh_u64 const hi_lo = XXH_mult32to64(lhs >> 32,        rhs & 0xFFFFFFFF);
-    xxh_u64 const lo_hi = XXH_mult32to64(lhs & 0xFFFFFFFF, rhs >> 32);
-    xxh_u64 const hi_hi = XXH_mult32to64(lhs >> 32,        rhs >> 32);
-
-    /* Now add the products together. These will never overflow. */
-    xxh_u64 const cross = (lo_lo >> 32) + (hi_lo & 0xFFFFFFFF) + lo_hi;
-    xxh_u64 const upper = (hi_lo >> 32) + (cross >> 32)        + hi_hi;
-    xxh_u64 const lower = (cross << 32) | (lo_lo & 0xFFFFFFFF);
-
-    XXH128_hash_t r128;
-    r128.low64  = lower;
-    r128.high64 = upper;
-    return r128;
-#endif
 }
 
 /*!
  * @brief Calculates a 64-bit to 128-bit multiply, then XOR folds it.
  *
  * The reason for the separate function is to prevent passing too many structs
- * around by value. This will hopefully inline the multiply, but we don't force it.
+ * around by value. This will hopefully inline the multiply, but we don't force
+ * it.
  *
  * @param lhs , rhs The 64-bit integers to multiply
  * @return The low 64 bits of the product XOR'd by the high 64 bits.
  * @see XXH_mult64to128()
  */
-static xxh_u64
-XXH3_mul128_fold64(xxh_u64 lhs, xxh_u64 rhs)
-{
-    XXH128_hash_t product = XXH_mult64to128(lhs, rhs);
-    return product.low64 ^ product.high64;
+static xxh_u64 XXH3_mul128_fold64(xxh_u64 lhs, xxh_u64 rhs) {
+
+  XXH128_hash_t product = XXH_mult64to128(lhs, rhs);
+  return product.low64 ^ product.high64;
+
 }
 
 /*! Seems to produce slightly better code on GCC for some reason. */
-XXH_FORCE_INLINE XXH_CONSTF xxh_u64 XXH_xorshift64(xxh_u64 v64, int shift)
-{
-    XXH_ASSERT(0 <= shift && shift < 64);
-    return v64 ^ (v64 >> shift);
+XXH_FORCE_INLINE XXH_CONSTF xxh_u64 XXH_xorshift64(xxh_u64 v64, int shift) {
+
+  XXH_ASSERT(0 <= shift && shift < 64);
+  return v64 ^ (v64 >> shift);
+
 }
 
 /*
  * This is a fast avalanche stage,
  * suitable when input bits are already partially mixed
  */
-static XXH64_hash_t XXH3_avalanche(xxh_u64 h64)
-{
-    h64 = XXH_xorshift64(h64, 37);
-    h64 *= PRIME_MX1;
-    h64 = XXH_xorshift64(h64, 32);
-    return h64;
+static XXH64_hash_t XXH3_avalanche(xxh_u64 h64) {
+
+  h64 = XXH_xorshift64(h64, 37);
+  h64 *= PRIME_MX1;
+  h64 = XXH_xorshift64(h64, 32);
+  return h64;
+
 }
 
 /*
@@ -4467,16 +4967,16 @@ static XXH64_hash_t XXH3_avalanche(xxh_u64 h64)
  * inspired by Pelle Evensen's rrmxmx
  * preferable when input has not been previously mixed
  */
-static XXH64_hash_t XXH3_rrmxmx(xxh_u64 h64, xxh_u64 len)
-{
-    /* this mix is inspired by Pelle Evensen's rrmxmx */
-    h64 ^= XXH_rotl64(h64, 49) ^ XXH_rotl64(h64, 24);
-    h64 *= PRIME_MX2;
-    h64 ^= (h64 >> 35) + len ;
-    h64 *= PRIME_MX2;
-    return XXH_xorshift64(h64, 28);
-}
+static XXH64_hash_t XXH3_rrmxmx(xxh_u64 h64, xxh_u64 len) {
 
+  /* this mix is inspired by Pelle Evensen's rrmxmx */
+  h64 ^= XXH_rotl64(h64, 49) ^ XXH_rotl64(h64, 24);
+  h64 *= PRIME_MX2;
+  h64 ^= (h64 >> 35) + len;
+  h64 *= PRIME_MX2;
+  return XXH_xorshift64(h64, 28);
+
+}
 
 /* ==========================================
  * Short keys
@@ -4486,7 +4986,8 @@ static XXH64_hash_t XXH3_rrmxmx(xxh_u64 h64, xxh_u64 len)
  * favored lengths that were a multiple of 4 or 8.
  *
  * Instead of iterating over individual inputs, we use a set of single shot
- * functions which piece together a range of lengths and operate in constant time.
+ * functions which piece together a range of lengths and operate in constant
+ * time.
  *
  * Additionally, the number of multiplies has been significantly reduced. This
  * reduces latency, especially when emulating 64-bit multiplies on 32-bit.
@@ -4511,70 +5012,100 @@ static XXH64_hash_t XXH3_rrmxmx(xxh_u64 h64, xxh_u64 len)
  *
  * This adds an extra layer of strength for custom secrets.
  */
-XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t
-XXH3_len_1to3_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
-{
-    XXH_ASSERT(input != NULL);
-    XXH_ASSERT(1 <= len && len <= 3);
-    XXH_ASSERT(secret != NULL);
-    /*
-     * len = 1: combined = { input[0], 0x01, input[0], input[0] }
-     * len = 2: combined = { input[1], 0x02, input[0], input[1] }
-     * len = 3: combined = { input[2], 0x03, input[0], input[1] }
-     */
-    {   xxh_u8  const c1 = input[0];
-        xxh_u8  const c2 = input[len >> 1];
-        xxh_u8  const c3 = input[len - 1];
-        xxh_u32 const combined = ((xxh_u32)c1 << 16) | ((xxh_u32)c2  << 24)
-                               | ((xxh_u32)c3 <<  0) | ((xxh_u32)len << 8);
-        xxh_u64 const bitflip = (XXH_readLE32(secret) ^ XXH_readLE32(secret+4)) + seed;
-        xxh_u64 const keyed = (xxh_u64)combined ^ bitflip;
-        return XXH64_avalanche(keyed);
-    }
+XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t XXH3_len_1to3_64b(const xxh_u8 *input,
+                                                          size_t        len,
+                                                          const xxh_u8 *secret,
+                                                          XXH64_hash_t  seed) {
+
+  XXH_ASSERT(input != NULL);
+  XXH_ASSERT(1 <= len && len <= 3);
+  XXH_ASSERT(secret != NULL);
+  /*
+   * len = 1: combined = { input[0], 0x01, input[0], input[0] }
+   * len = 2: combined = { input[1], 0x02, input[0], input[1] }
+   * len = 3: combined = { input[2], 0x03, input[0], input[1] }
+   */
+  {
+
+    xxh_u8 const  c1 = input[0];
+    xxh_u8 const  c2 = input[len >> 1];
+    xxh_u8 const  c3 = input[len - 1];
+    xxh_u32 const combined = ((xxh_u32)c1 << 16) | ((xxh_u32)c2 << 24) |
+                             ((xxh_u32)c3 << 0) | ((xxh_u32)len << 8);
+    xxh_u64 const bitflip =
+        (XXH_readLE32(secret) ^ XXH_readLE32(secret + 4)) + seed;
+    xxh_u64 const keyed = (xxh_u64)combined ^ bitflip;
+    return XXH64_avalanche(keyed);
+
+  }
+
 }
 
-XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t
-XXH3_len_4to8_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
-{
-    XXH_ASSERT(input != NULL);
-    XXH_ASSERT(secret != NULL);
-    XXH_ASSERT(4 <= len && len <= 8);
-    seed ^= (xxh_u64)XXH_swap32((xxh_u32)seed) << 32;
-    {   xxh_u32 const input1 = XXH_readLE32(input);
-        xxh_u32 const input2 = XXH_readLE32(input + len - 4);
-        xxh_u64 const bitflip = (XXH_readLE64(secret+8) ^ XXH_readLE64(secret+16)) - seed;
-        xxh_u64 const input64 = input2 + (((xxh_u64)input1) << 32);
-        xxh_u64 const keyed = input64 ^ bitflip;
-        return XXH3_rrmxmx(keyed, len);
-    }
+XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t XXH3_len_4to8_64b(const xxh_u8 *input,
+                                                          size_t        len,
+                                                          const xxh_u8 *secret,
+                                                          XXH64_hash_t  seed) {
+
+  XXH_ASSERT(input != NULL);
+  XXH_ASSERT(secret != NULL);
+  XXH_ASSERT(4 <= len && len <= 8);
+  seed ^= (xxh_u64)XXH_swap32((xxh_u32)seed) << 32;
+  {
+
+    xxh_u32 const input1 = XXH_readLE32(input);
+    xxh_u32 const input2 = XXH_readLE32(input + len - 4);
+    xxh_u64 const bitflip =
+        (XXH_readLE64(secret + 8) ^ XXH_readLE64(secret + 16)) - seed;
+    xxh_u64 const input64 = input2 + (((xxh_u64)input1) << 32);
+    xxh_u64 const keyed = input64 ^ bitflip;
+    return XXH3_rrmxmx(keyed, len);
+
+  }
+
 }
 
-XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t
-XXH3_len_9to16_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
-{
-    XXH_ASSERT(input != NULL);
-    XXH_ASSERT(secret != NULL);
-    XXH_ASSERT(9 <= len && len <= 16);
-    {   xxh_u64 const bitflip1 = (XXH_readLE64(secret+24) ^ XXH_readLE64(secret+32)) + seed;
-        xxh_u64 const bitflip2 = (XXH_readLE64(secret+40) ^ XXH_readLE64(secret+48)) - seed;
-        xxh_u64 const input_lo = XXH_readLE64(input)           ^ bitflip1;
-        xxh_u64 const input_hi = XXH_readLE64(input + len - 8) ^ bitflip2;
-        xxh_u64 const acc = len
-                          + XXH_swap64(input_lo) + input_hi
-                          + XXH3_mul128_fold64(input_lo, input_hi);
-        return XXH3_avalanche(acc);
-    }
+XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t XXH3_len_9to16_64b(const xxh_u8 *input,
+                                                           size_t        len,
+                                                           const xxh_u8 *secret,
+                                                           XXH64_hash_t  seed) {
+
+  XXH_ASSERT(input != NULL);
+  XXH_ASSERT(secret != NULL);
+  XXH_ASSERT(9 <= len && len <= 16);
+  {
+
+    xxh_u64 const bitflip1 =
+        (XXH_readLE64(secret + 24) ^ XXH_readLE64(secret + 32)) + seed;
+    xxh_u64 const bitflip2 =
+        (XXH_readLE64(secret + 40) ^ XXH_readLE64(secret + 48)) - seed;
+    xxh_u64 const input_lo = XXH_readLE64(input) ^ bitflip1;
+    xxh_u64 const input_hi = XXH_readLE64(input + len - 8) ^ bitflip2;
+    xxh_u64 const acc = len + XXH_swap64(input_lo) + input_hi +
+                        XXH3_mul128_fold64(input_lo, input_hi);
+    return XXH3_avalanche(acc);
+
+  }
+
 }
 
-XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t
-XXH3_len_0to16_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
-{
-    XXH_ASSERT(len <= 16);
-    {   if (XXH_likely(len >  8)) return XXH3_len_9to16_64b(input, len, secret, seed);
-        if (XXH_likely(len >= 4)) return XXH3_len_4to8_64b(input, len, secret, seed);
-        if (len) return XXH3_len_1to3_64b(input, len, secret, seed);
-        return XXH64_avalanche(seed ^ (XXH_readLE64(secret+56) ^ XXH_readLE64(secret+64)));
-    }
+XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t XXH3_len_0to16_64b(const xxh_u8 *input,
+                                                           size_t        len,
+                                                           const xxh_u8 *secret,
+                                                           XXH64_hash_t  seed) {
+
+  XXH_ASSERT(len <= 16);
+  {
+
+    if (XXH_likely(len > 8))
+      return XXH3_len_9to16_64b(input, len, secret, seed);
+    if (XXH_likely(len >= 4))
+      return XXH3_len_4to8_64b(input, len, secret, seed);
+    if (len) return XXH3_len_1to3_64b(input, len, secret, seed);
+    return XXH64_avalanche(
+        seed ^ (XXH_readLE64(secret + 56) ^ XXH_readLE64(secret + 64)));
+
+  }
+
 }
 
 /*
@@ -4603,106 +5134,134 @@ XXH3_len_0to16_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_
  * by this, although it is always a good idea to use a proper seed if you care
  * about strength.
  */
-XXH_FORCE_INLINE xxh_u64 XXH3_mix16B(const xxh_u8* XXH_RESTRICT input,
-                                     const xxh_u8* XXH_RESTRICT secret, xxh_u64 seed64)
-{
-#if defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
-  && defined(__i386__) && defined(__SSE2__)  /* x86 + SSE2 */ \
-  && !defined(XXH_ENABLE_AUTOVECTORIZE)      /* Define to disable like XXH32 hack */
-    /*
-     * UGLY HACK:
-     * GCC for x86 tends to autovectorize the 128-bit multiply, resulting in
-     * slower code.
-     *
-     * By forcing seed64 into a register, we disrupt the cost model and
-     * cause it to scalarize. See `XXH32_round()`
-     *
-     * FIXME: Clang's output is still _much_ faster -- On an AMD Ryzen 3600,
-     * XXH3_64bits @ len=240 runs at 4.6 GB/s with Clang 9, but 3.3 GB/s on
-     * GCC 9.2, despite both emitting scalar code.
-     *
-     * GCC generates much better scalar code than Clang for the rest of XXH3,
-     * which is why finding a more optimal codepath is an interest.
-     */
-    XXH_COMPILER_GUARD(seed64);
-#endif
-    {   xxh_u64 const input_lo = XXH_readLE64(input);
-        xxh_u64 const input_hi = XXH_readLE64(input+8);
-        return XXH3_mul128_fold64(
-            input_lo ^ (XXH_readLE64(secret)   + seed64),
-            input_hi ^ (XXH_readLE64(secret+8) - seed64)
-        );
-    }
+XXH_FORCE_INLINE xxh_u64 XXH3_mix16B(const xxh_u8 *XXH_RESTRICT input,
+                                     const xxh_u8 *XXH_RESTRICT secret,
+                                     xxh_u64                    seed64) {
+
+      #if defined(__GNUC__) && !defined(__clang__)  /* GCC, not Clang */      \
+          && defined(__i386__) && defined(__SSE2__) /* x86 + SSE2 */          \
+          && !defined(XXH_ENABLE_AUTOVECTORIZE)     /* Define to disable like \
+                                                       XXH32 hack */
+  /*
+   * UGLY HACK:
+   * GCC for x86 tends to autovectorize the 128-bit multiply, resulting in
+   * slower code.
+   *
+   * By forcing seed64 into a register, we disrupt the cost model and
+   * cause it to scalarize. See `XXH32_round()`
+   *
+   * FIXME: Clang's output is still _much_ faster -- On an AMD Ryzen 3600,
+   * XXH3_64bits @ len=240 runs at 4.6 GB/s with Clang 9, but 3.3 GB/s on
+   * GCC 9.2, despite both emitting scalar code.
+   *
+   * GCC generates much better scalar code than Clang for the rest of XXH3,
+   * which is why finding a more optimal codepath is an interest.
+   */
+  XXH_COMPILER_GUARD(seed64);
+      #endif
+  {
+
+    xxh_u64 const input_lo = XXH_readLE64(input);
+    xxh_u64 const input_hi = XXH_readLE64(input + 8);
+    return XXH3_mul128_fold64(input_lo ^ (XXH_readLE64(secret) + seed64),
+                              input_hi ^ (XXH_readLE64(secret + 8) - seed64));
+
+  }
+
 }
 
 /* For mid range keys, XXH3 uses a Mum-hash variant. */
-XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t
-XXH3_len_17to128_64b(const xxh_u8* XXH_RESTRICT input, size_t len,
-                     const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
-                     XXH64_hash_t seed)
-{
-    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
-    XXH_ASSERT(16 < len && len <= 128);
-
-    {   xxh_u64 acc = len * XXH_PRIME64_1;
-#if XXH_SIZE_OPT >= 1
-        /* Smaller and cleaner, but slightly slower. */
-        unsigned int i = (unsigned int)(len - 1) / 32;
-        do {
-            acc += XXH3_mix16B(input+16 * i, secret+32*i, seed);
-            acc += XXH3_mix16B(input+len-16*(i+1), secret+32*i+16, seed);
-        } while (i-- != 0);
-#else
-        if (len > 32) {
-            if (len > 64) {
-                if (len > 96) {
-                    acc += XXH3_mix16B(input+48, secret+96, seed);
-                    acc += XXH3_mix16B(input+len-64, secret+112, seed);
-                }
-                acc += XXH3_mix16B(input+32, secret+64, seed);
-                acc += XXH3_mix16B(input+len-48, secret+80, seed);
-            }
-            acc += XXH3_mix16B(input+16, secret+32, seed);
-            acc += XXH3_mix16B(input+len-32, secret+48, seed);
+XXH_FORCE_INLINE XXH_PUREF XXH64_hash_t XXH3_len_17to128_64b(
+    const xxh_u8 *XXH_RESTRICT input, size_t len,
+    const xxh_u8 *XXH_RESTRICT secret, size_t secretSize, XXH64_hash_t seed) {
+
+  XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
+  (void)secretSize;
+  XXH_ASSERT(16 < len && len <= 128);
+
+  {
+
+    xxh_u64 acc = len * XXH_PRIME64_1;
+      #if XXH_SIZE_OPT >= 1
+    /* Smaller and cleaner, but slightly slower. */
+    unsigned int i = (unsigned int)(len - 1) / 32;
+    do {
+
+      acc += XXH3_mix16B(input + 16 * i, secret + 32 * i, seed);
+      acc +=
+          XXH3_mix16B(input + len - 16 * (i + 1), secret + 32 * i + 16, seed);
+
+    } while (i-- != 0);
+
+      #else
+    if (len > 32) {
+
+      if (len > 64) {
+
+        if (len > 96) {
+
+          acc += XXH3_mix16B(input + 48, secret + 96, seed);
+          acc += XXH3_mix16B(input + len - 64, secret + 112, seed);
+
         }
-        acc += XXH3_mix16B(input+0, secret+0, seed);
-        acc += XXH3_mix16B(input+len-16, secret+16, seed);
-#endif
-        return XXH3_avalanche(acc);
+
+        acc += XXH3_mix16B(input + 32, secret + 64, seed);
+        acc += XXH3_mix16B(input + len - 48, secret + 80, seed);
+
+      }
+
+      acc += XXH3_mix16B(input + 16, secret + 32, seed);
+      acc += XXH3_mix16B(input + len - 32, secret + 48, seed);
+
     }
+
+    acc += XXH3_mix16B(input + 0, secret + 0, seed);
+    acc += XXH3_mix16B(input + len - 16, secret + 16, seed);
+      #endif
+    return XXH3_avalanche(acc);
+
+  }
+
 }
 
-/*!
- * @brief Maximum size of "short" key in bytes.
- */
-#define XXH3_MIDSIZE_MAX 240
+      /*!
+       * @brief Maximum size of "short" key in bytes.
+       */
+      #define XXH3_MIDSIZE_MAX 240
 
-XXH_NO_INLINE XXH_PUREF XXH64_hash_t
-XXH3_len_129to240_64b(const xxh_u8* XXH_RESTRICT input, size_t len,
-                      const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
-                      XXH64_hash_t seed)
-{
-    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
+XXH_NO_INLINE XXH_PUREF XXH64_hash_t XXH3_len_129to240_64b(
+    const xxh_u8 *XXH_RESTRICT input, size_t len,
+    const xxh_u8 *XXH_RESTRICT secret, size_t secretSize, XXH64_hash_t seed) {
+
+  XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
+  (void)secretSize;
+  XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX);
+
+      #define XXH3_MIDSIZE_STARTOFFSET 3
+      #define XXH3_MIDSIZE_LASTOFFSET 17
+
+  {
+
+    xxh_u64            acc = len * XXH_PRIME64_1;
+    xxh_u64            acc_end;
+    unsigned int const nbRounds = (unsigned int)len / 16;
+    unsigned int       i;
     XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX);
+    for (i = 0; i < 8; i++) {
 
-    #define XXH3_MIDSIZE_STARTOFFSET 3
-    #define XXH3_MIDSIZE_LASTOFFSET  17
+      acc += XXH3_mix16B(input + (16 * i), secret + (16 * i), seed);
 
-    {   xxh_u64 acc = len * XXH_PRIME64_1;
-        xxh_u64 acc_end;
-        unsigned int const nbRounds = (unsigned int)len / 16;
-        unsigned int i;
-        XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX);
-        for (i=0; i<8; i++) {
-            acc += XXH3_mix16B(input+(16*i), secret+(16*i), seed);
-        }
-        /* last bytes */
-        acc_end = XXH3_mix16B(input + len - 16, secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET, seed);
-        XXH_ASSERT(nbRounds >= 8);
-        acc = XXH3_avalanche(acc);
-#if defined(__clang__)                                /* Clang */ \
-    && (defined(__ARM_NEON) || defined(__ARM_NEON__)) /* NEON */ \
-    && !defined(XXH_ENABLE_AUTOVECTORIZE)             /* Define to disable */
+    }
+
+    /* last bytes */
+    acc_end = XXH3_mix16B(
+        input + len - 16,
+        secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET, seed);
+    XXH_ASSERT(nbRounds >= 8);
+    acc = XXH3_avalanche(acc);
+      #if defined(__clang__)                                /* Clang */ \
+          && (defined(__ARM_NEON) || defined(__ARM_NEON__)) /* NEON */  \
+          && !defined(XXH_ENABLE_AUTOVECTORIZE)        /* Define to disable */
         /*
          * UGLY HACK:
          * Clang for ARMv7-A tries to vectorize this loop, similar to GCC x86.
@@ -4724,441 +5283,522 @@ XXH3_len_129to240_64b(const xxh_u8* XXH_RESTRICT input, size_t len,
          * SLP vectorization.
          */
         #pragma clang loop vectorize(disable)
-#endif
-        for (i=8 ; i < nbRounds; i++) {
-            /*
-             * Prevents clang for unrolling the acc loop and interleaving with this one.
-             */
-            XXH_COMPILER_GUARD(acc);
-            acc_end += XXH3_mix16B(input+(16*i), secret+(16*(i-8)) + XXH3_MIDSIZE_STARTOFFSET, seed);
-        }
-        return XXH3_avalanche(acc + acc_end);
+      #endif
+    for (i = 8; i < nbRounds; i++) {
+
+      /*
+       * Prevents clang for unrolling the acc loop and interleaving with this
+       * one.
+       */
+      XXH_COMPILER_GUARD(acc);
+      acc_end +=
+          XXH3_mix16B(input + (16 * i),
+                      secret + (16 * (i - 8)) + XXH3_MIDSIZE_STARTOFFSET, seed);
+
     }
-}
 
+    return XXH3_avalanche(acc + acc_end);
+
+  }
+
+}
+
+    /* =======     Long Keys     ======= */
+
+      #define XXH_STRIPE_LEN 64
+      #define XXH_SECRET_CONSUME_RATE \
+        8               /* nb of secret bytes consumed at each accumulation */
+      #define XXH_ACC_NB (XXH_STRIPE_LEN / sizeof(xxh_u64))
+
+      #ifdef XXH_OLD_NAMES
+        #define STRIPE_LEN XXH_STRIPE_LEN
+        #define ACC_NB XXH_ACC_NB
+      #endif
+
+      #ifndef XXH_PREFETCH_DIST
+        #ifdef __clang__
+          #define XXH_PREFETCH_DIST 320
+        #else
+          #if (XXH_VECTOR == XXH_AVX512)
+            #define XXH_PREFETCH_DIST 512
+          #else
+            #define XXH_PREFETCH_DIST 384
+          #endif
+        #endif                                                 /* __clang__ */
+      #endif                                           /* XXH_PREFETCH_DIST */
+
+      /*
+       * These macros are to generate an XXH3_accumulate() function.
+       * The two arguments select the name suffix and target attribute.
+       *
+       * The name of this symbol is XXH3_accumulate_<name>() and it calls
+       * XXH3_accumulate_512_<name>().
+       *
+       * It may be useful to hand implement this function if the compiler fails
+       * to optimize the inline function.
+       */
+      #define XXH3_ACCUMULATE_TEMPLATE(name)                                  \
+        void XXH3_accumulate_##name(                                          \
+            xxh_u64 *XXH_RESTRICT acc, const xxh_u8 *XXH_RESTRICT input,      \
+            const xxh_u8 *XXH_RESTRICT secret, size_t nbStripes) {            \
+                                                                              \
+          size_t n;                                                           \
+          for (n = 0; n < nbStripes; n++) {                                   \
+                                                                              \
+            const xxh_u8 *const in = input + n * XXH_STRIPE_LEN;              \
+            XXH_PREFETCH(in + XXH_PREFETCH_DIST);                             \
+            XXH3_accumulate_512_##name(acc, in,                               \
+                                       secret + n * XXH_SECRET_CONSUME_RATE); \
+                                                                              \
+          }                                                                   \
+                                                                              \
+        }
 
-/* =======     Long Keys     ======= */
+XXH_FORCE_INLINE void XXH_writeLE64(void *dst, xxh_u64 v64) {
 
-#define XXH_STRIPE_LEN 64
-#define XXH_SECRET_CONSUME_RATE 8   /* nb of secret bytes consumed at each accumulation */
-#define XXH_ACC_NB (XXH_STRIPE_LEN / sizeof(xxh_u64))
+  if (!XXH_CPU_LITTLE_ENDIAN) v64 = XXH_swap64(v64);
+  XXH_memcpy(dst, &v64, sizeof(v64));
 
-#ifdef XXH_OLD_NAMES
-#  define STRIPE_LEN XXH_STRIPE_LEN
-#  define ACC_NB XXH_ACC_NB
-#endif
+}
 
-#ifndef XXH_PREFETCH_DIST
-#  ifdef __clang__
-#    define XXH_PREFETCH_DIST 320
-#  else
-#    if (XXH_VECTOR == XXH_AVX512)
-#      define XXH_PREFETCH_DIST 512
-#    else
-#      define XXH_PREFETCH_DIST 384
-#    endif
-#  endif  /* __clang__ */
-#endif  /* XXH_PREFETCH_DIST */
+      /* Several intrinsic functions below are supposed to accept __int64 as
+       * argument, as documented in
+       * https://software.intel.com/sites/landingpage/IntrinsicsGuide/ .
+       * However, several environments do not define __int64 type,
+       * requiring a workaround.
+       */
+      #if !defined(__VMS) &&                                     \
+          (defined(__cplusplus) || (defined(__STDC_VERSION__) && \
+                                    (__STDC_VERSION__ >= 199901L) /* C99 */))
+typedef int64_t xxh_i64;
+      #else
+/* the following type must have a width of 64-bit */
+typedef long long xxh_i64;
+      #endif
 
-/*
- * These macros are to generate an XXH3_accumulate() function.
- * The two arguments select the name suffix and target attribute.
- *
- * The name of this symbol is XXH3_accumulate_<name>() and it calls
- * XXH3_accumulate_512_<name>().
- *
- * It may be useful to hand implement this function if the compiler fails to
- * optimize the inline function.
- */
-#define XXH3_ACCUMULATE_TEMPLATE(name)                      \
-void                                                        \
-XXH3_accumulate_##name(xxh_u64* XXH_RESTRICT acc,           \
-                       const xxh_u8* XXH_RESTRICT input,    \
-                       const xxh_u8* XXH_RESTRICT secret,   \
-                       size_t nbStripes)                    \
-{                                                           \
-    size_t n;                                               \
-    for (n = 0; n < nbStripes; n++ ) {                      \
-        const xxh_u8* const in = input + n*XXH_STRIPE_LEN;  \
-        XXH_PREFETCH(in + XXH_PREFETCH_DIST);               \
-        XXH3_accumulate_512_##name(                         \
-                 acc,                                       \
-                 in,                                        \
-                 secret + n*XXH_SECRET_CONSUME_RATE);       \
-    }                                                       \
-}
-
-
-XXH_FORCE_INLINE void XXH_writeLE64(void* dst, xxh_u64 v64)
-{
-    if (!XXH_CPU_LITTLE_ENDIAN) v64 = XXH_swap64(v64);
-    XXH_memcpy(dst, &v64, sizeof(v64));
-}
-
-/* Several intrinsic functions below are supposed to accept __int64 as argument,
- * as documented in https://software.intel.com/sites/landingpage/IntrinsicsGuide/ .
- * However, several environments do not define __int64 type,
- * requiring a workaround.
- */
-#if !defined (__VMS) \
-  && (defined (__cplusplus) \
-  || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
-    typedef int64_t xxh_i64;
-#else
-    /* the following type must have a width of 64-bit */
-    typedef long long xxh_i64;
-#endif
+    /*
+     * XXH3_accumulate_512 is the tightest loop for long inputs, and it is the
+     * most optimized.
+     *
+     * It is a hardened version of UMAC, based off of FARSH's implementation.
+     *
+     * This was chosen because it adapts quite well to 32-bit, 64-bit, and SIMD
+     * implementations, and it is ridiculously fast.
+     *
+     * We harden it by mixing the original input to the accumulators as well as
+     * the product.
+     *
+     * This means that in the (relatively likely) case of a multiply by zero,
+     * the original input is preserved.
+     *
+     * On 128-bit inputs, we swap 64-bit pairs when we add the input to improve
+     * cross-pollination, as otherwise the upper and lower halves would be
+     * essentially independent.
+     *
+     * This doesn't matter on 64-bit hashes since they all get merged together
+     * in the end, so we skip the extra step.
+     *
+     * Both XXH3_64bits and XXH3_128bits use this subroutine.
+     */
 
+      #if (XXH_VECTOR == XXH_AVX512) || \
+          (defined(XXH_DISPATCH_AVX512) && XXH_DISPATCH_AVX512 != 0)
 
-/*
- * XXH3_accumulate_512 is the tightest loop for long inputs, and it is the most optimized.
- *
- * It is a hardened version of UMAC, based off of FARSH's implementation.
- *
- * This was chosen because it adapts quite well to 32-bit, 64-bit, and SIMD
- * implementations, and it is ridiculously fast.
- *
- * We harden it by mixing the original input to the accumulators as well as the product.
- *
- * This means that in the (relatively likely) case of a multiply by zero, the
- * original input is preserved.
- *
- * On 128-bit inputs, we swap 64-bit pairs when we add the input to improve
- * cross-pollination, as otherwise the upper and lower halves would be
- * essentially independent.
- *
- * This doesn't matter on 64-bit hashes since they all get merged together in
- * the end, so we skip the extra step.
- *
- * Both XXH3_64bits and XXH3_128bits use this subroutine.
- */
+        #ifndef XXH_TARGET_AVX512
+          #define XXH_TARGET_AVX512             /* disable attribute target */
+        #endif
 
-#if (XXH_VECTOR == XXH_AVX512) \
-     || (defined(XXH_DISPATCH_AVX512) && XXH_DISPATCH_AVX512 != 0)
+XXH_FORCE_INLINE XXH_TARGET_AVX512 void XXH3_accumulate_512_avx512(
+    void *XXH_RESTRICT acc, const void *XXH_RESTRICT input,
+    const void *XXH_RESTRICT secret) {
 
-#ifndef XXH_TARGET_AVX512
-# define XXH_TARGET_AVX512  /* disable attribute target */
-#endif
+  __m512i *const xacc = (__m512i *)acc;
+  XXH_ASSERT((((size_t)acc) & 63) == 0);
+  XXH_STATIC_ASSERT(XXH_STRIPE_LEN == sizeof(__m512i));
 
-XXH_FORCE_INLINE XXH_TARGET_AVX512 void
-XXH3_accumulate_512_avx512(void* XXH_RESTRICT acc,
-                     const void* XXH_RESTRICT input,
-                     const void* XXH_RESTRICT secret)
-{
-    __m512i* const xacc = (__m512i *) acc;
-    XXH_ASSERT((((size_t)acc) & 63) == 0);
-    XXH_STATIC_ASSERT(XXH_STRIPE_LEN == sizeof(__m512i));
+  {
+
+    /* data_vec    = input[0]; */
+    __m512i const data_vec = _mm512_loadu_si512(input);
+    /* key_vec     = secret[0]; */
+    __m512i const key_vec = _mm512_loadu_si512(secret);
+    /* data_key    = data_vec ^ key_vec; */
+    __m512i const data_key = _mm512_xor_si512(data_vec, key_vec);
+    /* data_key_lo = data_key >> 32; */
+    __m512i const data_key_lo = _mm512_srli_epi64(data_key, 32);
+    /* product     = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
+    __m512i const product = _mm512_mul_epu32(data_key, data_key_lo);
+    /* xacc[0] += swap(data_vec); */
+    __m512i const data_swap =
+        _mm512_shuffle_epi32(data_vec, (_MM_PERM_ENUM)_MM_SHUFFLE(1, 0, 3, 2));
+    __m512i const sum = _mm512_add_epi64(*xacc, data_swap);
+    /* xacc[0] += product; */
+    *xacc = _mm512_add_epi64(product, sum);
+
+  }
 
-    {
-        /* data_vec    = input[0]; */
-        __m512i const data_vec    = _mm512_loadu_si512   (input);
-        /* key_vec     = secret[0]; */
-        __m512i const key_vec     = _mm512_loadu_si512   (secret);
-        /* data_key    = data_vec ^ key_vec; */
-        __m512i const data_key    = _mm512_xor_si512     (data_vec, key_vec);
-        /* data_key_lo = data_key >> 32; */
-        __m512i const data_key_lo = _mm512_srli_epi64 (data_key, 32);
-        /* product     = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
-        __m512i const product     = _mm512_mul_epu32     (data_key, data_key_lo);
-        /* xacc[0] += swap(data_vec); */
-        __m512i const data_swap = _mm512_shuffle_epi32(data_vec, (_MM_PERM_ENUM)_MM_SHUFFLE(1, 0, 3, 2));
-        __m512i const sum       = _mm512_add_epi64(*xacc, data_swap);
-        /* xacc[0] += product; */
-        *xacc = _mm512_add_epi64(product, sum);
-    }
 }
+
 XXH_FORCE_INLINE XXH_TARGET_AVX512 XXH3_ACCUMULATE_TEMPLATE(avx512)
 
-/*
- * XXH3_scrambleAcc: Scrambles the accumulators to improve mixing.
- *
- * Multiplication isn't perfect, as explained by Google in HighwayHash:
- *
- *  // Multiplication mixes/scrambles bytes 0-7 of the 64-bit result to
- *  // varying degrees. In descending order of goodness, bytes
- *  // 3 4 2 5 1 6 0 7 have quality 228 224 164 160 100 96 36 32.
- *  // As expected, the upper and lower bytes are much worse.
- *
- * Source: https://github.com/google/highwayhash/blob/0aaf66b/highwayhash/hh_avx2.h#L291
- *
- * Since our algorithm uses a pseudorandom secret to add some variance into the
- * mix, we don't need to (or want to) mix as often or as much as HighwayHash does.
- *
- * This isn't as tight as XXH3_accumulate, but still written in SIMD to avoid
- * extraction.
- *
- * Both XXH3_64bits and XXH3_128bits use this subroutine.
- */
+    /*
+     * XXH3_scrambleAcc: Scrambles the accumulators to improve mixing.
+     *
+     * Multiplication isn't perfect, as explained by Google in HighwayHash:
+     *
+     *  // Multiplication mixes/scrambles bytes 0-7 of the 64-bit result to
+     *  // varying degrees. In descending order of goodness, bytes
+     *  // 3 4 2 5 1 6 0 7 have quality 228 224 164 160 100 96 36 32.
+     *  // As expected, the upper and lower bytes are much worse.
+     *
+     * Source:
+     * https://github.com/google/highwayhash/blob/0aaf66b/highwayhash/hh_avx2.h#L291
+     *
+     * Since our algorithm uses a pseudorandom secret to add some variance into
+     * the mix, we don't need to (or want to) mix as often or as much as
+     * HighwayHash does.
+     *
+     * This isn't as tight as XXH3_accumulate, but still written in SIMD to
+     * avoid extraction.
+     *
+     * Both XXH3_64bits and XXH3_128bits use this subroutine.
+     */
+
+    XXH_FORCE_INLINE XXH_TARGET_AVX512
+    void XXH3_scrambleAcc_avx512(void *XXH_RESTRICT       acc,
+                                 const void *XXH_RESTRICT secret) {
+
+  XXH_ASSERT((((size_t)acc) & 63) == 0);
+  XXH_STATIC_ASSERT(XXH_STRIPE_LEN == sizeof(__m512i));
+  {
+
+    __m512i *const xacc = (__m512i *)acc;
+    const __m512i  prime32 = _mm512_set1_epi32((int)XXH_PRIME32_1);
+
+    /* xacc[0] ^= (xacc[0] >> 47) */
+    __m512i const acc_vec = *xacc;
+    __m512i const shifted = _mm512_srli_epi64(acc_vec, 47);
+    /* xacc[0] ^= secret; */
+    __m512i const key_vec = _mm512_loadu_si512(secret);
+    __m512i const data_key = _mm512_ternarylogic_epi32(
+        key_vec, acc_vec, shifted, 0x96 /* key_vec ^ acc_vec ^ shifted */);
+
+    /* xacc[0] *= XXH_PRIME32_1; */
+    __m512i const data_key_hi = _mm512_srli_epi64(data_key, 32);
+    __m512i const prod_lo = _mm512_mul_epu32(data_key, prime32);
+    __m512i const prod_hi = _mm512_mul_epu32(data_key_hi, prime32);
+    *xacc = _mm512_add_epi64(prod_lo, _mm512_slli_epi64(prod_hi, 32));
+
+  }
 
-XXH_FORCE_INLINE XXH_TARGET_AVX512 void
-XXH3_scrambleAcc_avx512(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
-{
-    XXH_ASSERT((((size_t)acc) & 63) == 0);
-    XXH_STATIC_ASSERT(XXH_STRIPE_LEN == sizeof(__m512i));
-    {   __m512i* const xacc = (__m512i*) acc;
-        const __m512i prime32 = _mm512_set1_epi32((int)XXH_PRIME32_1);
-
-        /* xacc[0] ^= (xacc[0] >> 47) */
-        __m512i const acc_vec     = *xacc;
-        __m512i const shifted     = _mm512_srli_epi64    (acc_vec, 47);
-        /* xacc[0] ^= secret; */
-        __m512i const key_vec     = _mm512_loadu_si512   (secret);
-        __m512i const data_key    = _mm512_ternarylogic_epi32(key_vec, acc_vec, shifted, 0x96 /* key_vec ^ acc_vec ^ shifted */);
-
-        /* xacc[0] *= XXH_PRIME32_1; */
-        __m512i const data_key_hi = _mm512_srli_epi64 (data_key, 32);
-        __m512i const prod_lo     = _mm512_mul_epu32     (data_key, prime32);
-        __m512i const prod_hi     = _mm512_mul_epu32     (data_key_hi, prime32);
-        *xacc = _mm512_add_epi64(prod_lo, _mm512_slli_epi64(prod_hi, 32));
-    }
 }
 
-XXH_FORCE_INLINE XXH_TARGET_AVX512 void
-XXH3_initCustomSecret_avx512(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
-{
-    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 63) == 0);
-    XXH_STATIC_ASSERT(XXH_SEC_ALIGN == 64);
-    XXH_ASSERT(((size_t)customSecret & 63) == 0);
-    (void)(&XXH_writeLE64);
-    {   int const nbRounds = XXH_SECRET_DEFAULT_SIZE / sizeof(__m512i);
-        __m512i const seed_pos = _mm512_set1_epi64((xxh_i64)seed64);
-        __m512i const seed     = _mm512_mask_sub_epi64(seed_pos, 0xAA, _mm512_set1_epi8(0), seed_pos);
-
-        const __m512i* const src  = (const __m512i*) ((const void*) XXH3_kSecret);
-              __m512i* const dest = (      __m512i*) customSecret;
-        int i;
-        XXH_ASSERT(((size_t)src & 63) == 0); /* control alignment */
-        XXH_ASSERT(((size_t)dest & 63) == 0);
-        for (i=0; i < nbRounds; ++i) {
-            dest[i] = _mm512_add_epi64(_mm512_load_si512(src + i), seed);
-    }   }
+XXH_FORCE_INLINE XXH_TARGET_AVX512 void XXH3_initCustomSecret_avx512(
+    void *XXH_RESTRICT customSecret, xxh_u64 seed64) {
+
+  XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 63) == 0);
+  XXH_STATIC_ASSERT(XXH_SEC_ALIGN == 64);
+  XXH_ASSERT(((size_t)customSecret & 63) == 0);
+  (void)(&XXH_writeLE64);
+  {
+
+    int const     nbRounds = XXH_SECRET_DEFAULT_SIZE / sizeof(__m512i);
+    __m512i const seed_pos = _mm512_set1_epi64((xxh_i64)seed64);
+    __m512i const seed =
+        _mm512_mask_sub_epi64(seed_pos, 0xAA, _mm512_set1_epi8(0), seed_pos);
+
+    const __m512i *const src = (const __m512i *)((const void *)XXH3_kSecret);
+    __m512i *const       dest = (__m512i *)customSecret;
+    int                  i;
+    XXH_ASSERT(((size_t)src & 63) == 0);               /* control alignment */
+    XXH_ASSERT(((size_t)dest & 63) == 0);
+    for (i = 0; i < nbRounds; ++i) {
+
+      dest[i] = _mm512_add_epi64(_mm512_load_si512(src + i), seed);
+
+    }
+
+  }
+
 }
 
-#endif
+      #endif
 
-#if (XXH_VECTOR == XXH_AVX2) \
-    || (defined(XXH_DISPATCH_AVX2) && XXH_DISPATCH_AVX2 != 0)
+      #if (XXH_VECTOR == XXH_AVX2) || \
+          (defined(XXH_DISPATCH_AVX2) && XXH_DISPATCH_AVX2 != 0)
 
-#ifndef XXH_TARGET_AVX2
-# define XXH_TARGET_AVX2  /* disable attribute target */
-#endif
+        #ifndef XXH_TARGET_AVX2
+          #define XXH_TARGET_AVX2               /* disable attribute target */
+        #endif
+
+XXH_FORCE_INLINE XXH_TARGET_AVX2 void XXH3_accumulate_512_avx2(
+    void *XXH_RESTRICT acc, const void *XXH_RESTRICT input,
+    const void *XXH_RESTRICT secret) {
+
+  XXH_ASSERT((((size_t)acc) & 31) == 0);
+  {
+
+    __m256i *const xacc = (__m256i *)acc;
+    /* Unaligned. This is mainly for pointer arithmetic, and because
+     * _mm256_loadu_si256 requires  a const __m256i * pointer for some reason.
+     */
+    const __m256i *const xinput = (const __m256i *)input;
+    /* Unaligned. This is mainly for pointer arithmetic, and because
+     * _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */
+    const __m256i *const xsecret = (const __m256i *)secret;
+
+    size_t i;
+    for (i = 0; i < XXH_STRIPE_LEN / sizeof(__m256i); i++) {
+
+      /* data_vec    = xinput[i]; */
+      __m256i const data_vec = _mm256_loadu_si256(xinput + i);
+      /* key_vec     = xsecret[i]; */
+      __m256i const key_vec = _mm256_loadu_si256(xsecret + i);
+      /* data_key    = data_vec ^ key_vec; */
+      __m256i const data_key = _mm256_xor_si256(data_vec, key_vec);
+      /* data_key_lo = data_key >> 32; */
+      __m256i const data_key_lo = _mm256_srli_epi64(data_key, 32);
+      /* product     = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
+      __m256i const product = _mm256_mul_epu32(data_key, data_key_lo);
+      /* xacc[i] += swap(data_vec); */
+      __m256i const data_swap =
+          _mm256_shuffle_epi32(data_vec, _MM_SHUFFLE(1, 0, 3, 2));
+      __m256i const sum = _mm256_add_epi64(xacc[i], data_swap);
+      /* xacc[i] += product; */
+      xacc[i] = _mm256_add_epi64(product, sum);
+
+    }
+
+  }
 
-XXH_FORCE_INLINE XXH_TARGET_AVX2 void
-XXH3_accumulate_512_avx2( void* XXH_RESTRICT acc,
-                    const void* XXH_RESTRICT input,
-                    const void* XXH_RESTRICT secret)
-{
-    XXH_ASSERT((((size_t)acc) & 31) == 0);
-    {   __m256i* const xacc    =       (__m256i *) acc;
-        /* Unaligned. This is mainly for pointer arithmetic, and because
-         * _mm256_loadu_si256 requires  a const __m256i * pointer for some reason. */
-        const         __m256i* const xinput  = (const __m256i *) input;
-        /* Unaligned. This is mainly for pointer arithmetic, and because
-         * _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */
-        const         __m256i* const xsecret = (const __m256i *) secret;
-
-        size_t i;
-        for (i=0; i < XXH_STRIPE_LEN/sizeof(__m256i); i++) {
-            /* data_vec    = xinput[i]; */
-            __m256i const data_vec    = _mm256_loadu_si256    (xinput+i);
-            /* key_vec     = xsecret[i]; */
-            __m256i const key_vec     = _mm256_loadu_si256   (xsecret+i);
-            /* data_key    = data_vec ^ key_vec; */
-            __m256i const data_key    = _mm256_xor_si256     (data_vec, key_vec);
-            /* data_key_lo = data_key >> 32; */
-            __m256i const data_key_lo = _mm256_srli_epi64 (data_key, 32);
-            /* product     = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
-            __m256i const product     = _mm256_mul_epu32     (data_key, data_key_lo);
-            /* xacc[i] += swap(data_vec); */
-            __m256i const data_swap = _mm256_shuffle_epi32(data_vec, _MM_SHUFFLE(1, 0, 3, 2));
-            __m256i const sum       = _mm256_add_epi64(xacc[i], data_swap);
-            /* xacc[i] += product; */
-            xacc[i] = _mm256_add_epi64(product, sum);
-    }   }
 }
+
 XXH_FORCE_INLINE XXH_TARGET_AVX2 XXH3_ACCUMULATE_TEMPLATE(avx2)
 
-XXH_FORCE_INLINE XXH_TARGET_AVX2 void
-XXH3_scrambleAcc_avx2(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
-{
-    XXH_ASSERT((((size_t)acc) & 31) == 0);
-    {   __m256i* const xacc = (__m256i*) acc;
-        /* Unaligned. This is mainly for pointer arithmetic, and because
-         * _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */
-        const         __m256i* const xsecret = (const __m256i *) secret;
-        const __m256i prime32 = _mm256_set1_epi32((int)XXH_PRIME32_1);
-
-        size_t i;
-        for (i=0; i < XXH_STRIPE_LEN/sizeof(__m256i); i++) {
-            /* xacc[i] ^= (xacc[i] >> 47) */
-            __m256i const acc_vec     = xacc[i];
-            __m256i const shifted     = _mm256_srli_epi64    (acc_vec, 47);
-            __m256i const data_vec    = _mm256_xor_si256     (acc_vec, shifted);
-            /* xacc[i] ^= xsecret; */
-            __m256i const key_vec     = _mm256_loadu_si256   (xsecret+i);
-            __m256i const data_key    = _mm256_xor_si256     (data_vec, key_vec);
-
-            /* xacc[i] *= XXH_PRIME32_1; */
-            __m256i const data_key_hi = _mm256_srli_epi64 (data_key, 32);
-            __m256i const prod_lo     = _mm256_mul_epu32     (data_key, prime32);
-            __m256i const prod_hi     = _mm256_mul_epu32     (data_key_hi, prime32);
-            xacc[i] = _mm256_add_epi64(prod_lo, _mm256_slli_epi64(prod_hi, 32));
-        }
+    XXH_FORCE_INLINE XXH_TARGET_AVX2
+    void XXH3_scrambleAcc_avx2(void *XXH_RESTRICT       acc,
+                               const void *XXH_RESTRICT secret) {
+
+  XXH_ASSERT((((size_t)acc) & 31) == 0);
+  {
+
+    __m256i *const xacc = (__m256i *)acc;
+    /* Unaligned. This is mainly for pointer arithmetic, and because
+     * _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */
+    const __m256i *const xsecret = (const __m256i *)secret;
+    const __m256i        prime32 = _mm256_set1_epi32((int)XXH_PRIME32_1);
+
+    size_t i;
+    for (i = 0; i < XXH_STRIPE_LEN / sizeof(__m256i); i++) {
+
+      /* xacc[i] ^= (xacc[i] >> 47) */
+      __m256i const acc_vec = xacc[i];
+      __m256i const shifted = _mm256_srli_epi64(acc_vec, 47);
+      __m256i const data_vec = _mm256_xor_si256(acc_vec, shifted);
+      /* xacc[i] ^= xsecret; */
+      __m256i const key_vec = _mm256_loadu_si256(xsecret + i);
+      __m256i const data_key = _mm256_xor_si256(data_vec, key_vec);
+
+      /* xacc[i] *= XXH_PRIME32_1; */
+      __m256i const data_key_hi = _mm256_srli_epi64(data_key, 32);
+      __m256i const prod_lo = _mm256_mul_epu32(data_key, prime32);
+      __m256i const prod_hi = _mm256_mul_epu32(data_key_hi, prime32);
+      xacc[i] = _mm256_add_epi64(prod_lo, _mm256_slli_epi64(prod_hi, 32));
+
     }
+
+  }
+
 }
 
-XXH_FORCE_INLINE XXH_TARGET_AVX2 void XXH3_initCustomSecret_avx2(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
-{
-    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 31) == 0);
-    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE / sizeof(__m256i)) == 6);
-    XXH_STATIC_ASSERT(XXH_SEC_ALIGN <= 64);
-    (void)(&XXH_writeLE64);
-    XXH_PREFETCH(customSecret);
-    {   __m256i const seed = _mm256_set_epi64x((xxh_i64)(0U - seed64), (xxh_i64)seed64, (xxh_i64)(0U - seed64), (xxh_i64)seed64);
+XXH_FORCE_INLINE XXH_TARGET_AVX2 void XXH3_initCustomSecret_avx2(
+    void *XXH_RESTRICT customSecret, xxh_u64 seed64) {
 
-        const __m256i* const src  = (const __m256i*) ((const void*) XXH3_kSecret);
-              __m256i*       dest = (      __m256i*) customSecret;
+  XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 31) == 0);
+  XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE / sizeof(__m256i)) == 6);
+  XXH_STATIC_ASSERT(XXH_SEC_ALIGN <= 64);
+  (void)(&XXH_writeLE64);
+  XXH_PREFETCH(customSecret);
+  {
+
+    __m256i const seed =
+        _mm256_set_epi64x((xxh_i64)(0U - seed64), (xxh_i64)seed64,
+                          (xxh_i64)(0U - seed64), (xxh_i64)seed64);
+
+    const __m256i *const src = (const __m256i *)((const void *)XXH3_kSecret);
+    __m256i             *dest = (__m256i *)customSecret;
+
+        #if defined(__GNUC__) || defined(__clang__)
+    /*
+     * On GCC & Clang, marking 'dest' as modified will cause the compiler:
+     *   - do not extract the secret from sse registers in the internal loop
+     *   - use less common registers, and avoid pushing these reg into stack
+     */
+    XXH_COMPILER_GUARD(dest);
+        #endif
+    XXH_ASSERT(((size_t)src & 31) == 0);               /* control alignment */
+    XXH_ASSERT(((size_t)dest & 31) == 0);
+
+    /* GCC -O2 need unroll loop manually */
+    dest[0] = _mm256_add_epi64(_mm256_load_si256(src + 0), seed);
+    dest[1] = _mm256_add_epi64(_mm256_load_si256(src + 1), seed);
+    dest[2] = _mm256_add_epi64(_mm256_load_si256(src + 2), seed);
+    dest[3] = _mm256_add_epi64(_mm256_load_si256(src + 3), seed);
+    dest[4] = _mm256_add_epi64(_mm256_load_si256(src + 4), seed);
+    dest[5] = _mm256_add_epi64(_mm256_load_si256(src + 5), seed);
+
+  }
 
-#       if defined(__GNUC__) || defined(__clang__)
-        /*
-         * On GCC & Clang, marking 'dest' as modified will cause the compiler:
-         *   - do not extract the secret from sse registers in the internal loop
-         *   - use less common registers, and avoid pushing these reg into stack
-         */
-        XXH_COMPILER_GUARD(dest);
-#       endif
-        XXH_ASSERT(((size_t)src & 31) == 0); /* control alignment */
-        XXH_ASSERT(((size_t)dest & 31) == 0);
-
-        /* GCC -O2 need unroll loop manually */
-        dest[0] = _mm256_add_epi64(_mm256_load_si256(src+0), seed);
-        dest[1] = _mm256_add_epi64(_mm256_load_si256(src+1), seed);
-        dest[2] = _mm256_add_epi64(_mm256_load_si256(src+2), seed);
-        dest[3] = _mm256_add_epi64(_mm256_load_si256(src+3), seed);
-        dest[4] = _mm256_add_epi64(_mm256_load_si256(src+4), seed);
-        dest[5] = _mm256_add_epi64(_mm256_load_si256(src+5), seed);
-    }
 }
 
-#endif
+      #endif
 
-/* x86dispatch always generates SSE2 */
-#if (XXH_VECTOR == XXH_SSE2) || defined(XXH_X86DISPATCH)
+      /* x86dispatch always generates SSE2 */
+      #if (XXH_VECTOR == XXH_SSE2) || defined(XXH_X86DISPATCH)
 
-#ifndef XXH_TARGET_SSE2
-# define XXH_TARGET_SSE2  /* disable attribute target */
-#endif
+        #ifndef XXH_TARGET_SSE2
+          #define XXH_TARGET_SSE2               /* disable attribute target */
+        #endif
+
+XXH_FORCE_INLINE XXH_TARGET_SSE2 void XXH3_accumulate_512_sse2(
+    void *XXH_RESTRICT acc, const void *XXH_RESTRICT input,
+    const void *XXH_RESTRICT secret) {
+
+  /* SSE2 is just a half-scale version of the AVX2 version. */
+  XXH_ASSERT((((size_t)acc) & 15) == 0);
+  {
+
+    __m128i *const xacc = (__m128i *)acc;
+    /* Unaligned. This is mainly for pointer arithmetic, and because
+     * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
+    const __m128i *const xinput = (const __m128i *)input;
+    /* Unaligned. This is mainly for pointer arithmetic, and because
+     * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
+    const __m128i *const xsecret = (const __m128i *)secret;
+
+    size_t i;
+    for (i = 0; i < XXH_STRIPE_LEN / sizeof(__m128i); i++) {
+
+      /* data_vec    = xinput[i]; */
+      __m128i const data_vec = _mm_loadu_si128(xinput + i);
+      /* key_vec     = xsecret[i]; */
+      __m128i const key_vec = _mm_loadu_si128(xsecret + i);
+      /* data_key    = data_vec ^ key_vec; */
+      __m128i const data_key = _mm_xor_si128(data_vec, key_vec);
+      /* data_key_lo = data_key >> 32; */
+      __m128i const data_key_lo =
+          _mm_shuffle_epi32(data_key, _MM_SHUFFLE(0, 3, 0, 1));
+      /* product     = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
+      __m128i const product = _mm_mul_epu32(data_key, data_key_lo);
+      /* xacc[i] += swap(data_vec); */
+      __m128i const data_swap =
+          _mm_shuffle_epi32(data_vec, _MM_SHUFFLE(1, 0, 3, 2));
+      __m128i const sum = _mm_add_epi64(xacc[i], data_swap);
+      /* xacc[i] += product; */
+      xacc[i] = _mm_add_epi64(product, sum);
+
+    }
+
+  }
 
-XXH_FORCE_INLINE XXH_TARGET_SSE2 void
-XXH3_accumulate_512_sse2( void* XXH_RESTRICT acc,
-                    const void* XXH_RESTRICT input,
-                    const void* XXH_RESTRICT secret)
-{
-    /* SSE2 is just a half-scale version of the AVX2 version. */
-    XXH_ASSERT((((size_t)acc) & 15) == 0);
-    {   __m128i* const xacc    =       (__m128i *) acc;
-        /* Unaligned. This is mainly for pointer arithmetic, and because
-         * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
-        const         __m128i* const xinput  = (const __m128i *) input;
-        /* Unaligned. This is mainly for pointer arithmetic, and because
-         * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
-        const         __m128i* const xsecret = (const __m128i *) secret;
-
-        size_t i;
-        for (i=0; i < XXH_STRIPE_LEN/sizeof(__m128i); i++) {
-            /* data_vec    = xinput[i]; */
-            __m128i const data_vec    = _mm_loadu_si128   (xinput+i);
-            /* key_vec     = xsecret[i]; */
-            __m128i const key_vec     = _mm_loadu_si128   (xsecret+i);
-            /* data_key    = data_vec ^ key_vec; */
-            __m128i const data_key    = _mm_xor_si128     (data_vec, key_vec);
-            /* data_key_lo = data_key >> 32; */
-            __m128i const data_key_lo = _mm_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1));
-            /* product     = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
-            __m128i const product     = _mm_mul_epu32     (data_key, data_key_lo);
-            /* xacc[i] += swap(data_vec); */
-            __m128i const data_swap = _mm_shuffle_epi32(data_vec, _MM_SHUFFLE(1,0,3,2));
-            __m128i const sum       = _mm_add_epi64(xacc[i], data_swap);
-            /* xacc[i] += product; */
-            xacc[i] = _mm_add_epi64(product, sum);
-    }   }
 }
+
 XXH_FORCE_INLINE XXH_TARGET_SSE2 XXH3_ACCUMULATE_TEMPLATE(sse2)
 
-XXH_FORCE_INLINE XXH_TARGET_SSE2 void
-XXH3_scrambleAcc_sse2(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
-{
-    XXH_ASSERT((((size_t)acc) & 15) == 0);
-    {   __m128i* const xacc = (__m128i*) acc;
-        /* Unaligned. This is mainly for pointer arithmetic, and because
-         * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
-        const         __m128i* const xsecret = (const __m128i *) secret;
-        const __m128i prime32 = _mm_set1_epi32((int)XXH_PRIME32_1);
-
-        size_t i;
-        for (i=0; i < XXH_STRIPE_LEN/sizeof(__m128i); i++) {
-            /* xacc[i] ^= (xacc[i] >> 47) */
-            __m128i const acc_vec     = xacc[i];
-            __m128i const shifted     = _mm_srli_epi64    (acc_vec, 47);
-            __m128i const data_vec    = _mm_xor_si128     (acc_vec, shifted);
-            /* xacc[i] ^= xsecret[i]; */
-            __m128i const key_vec     = _mm_loadu_si128   (xsecret+i);
-            __m128i const data_key    = _mm_xor_si128     (data_vec, key_vec);
-
-            /* xacc[i] *= XXH_PRIME32_1; */
-            __m128i const data_key_hi = _mm_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1));
-            __m128i const prod_lo     = _mm_mul_epu32     (data_key, prime32);
-            __m128i const prod_hi     = _mm_mul_epu32     (data_key_hi, prime32);
-            xacc[i] = _mm_add_epi64(prod_lo, _mm_slli_epi64(prod_hi, 32));
-        }
+    XXH_FORCE_INLINE XXH_TARGET_SSE2
+    void XXH3_scrambleAcc_sse2(void *XXH_RESTRICT       acc,
+                               const void *XXH_RESTRICT secret) {
+
+  XXH_ASSERT((((size_t)acc) & 15) == 0);
+  {
+
+    __m128i *const xacc = (__m128i *)acc;
+    /* Unaligned. This is mainly for pointer arithmetic, and because
+     * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
+    const __m128i *const xsecret = (const __m128i *)secret;
+    const __m128i        prime32 = _mm_set1_epi32((int)XXH_PRIME32_1);
+
+    size_t i;
+    for (i = 0; i < XXH_STRIPE_LEN / sizeof(__m128i); i++) {
+
+      /* xacc[i] ^= (xacc[i] >> 47) */
+      __m128i const acc_vec = xacc[i];
+      __m128i const shifted = _mm_srli_epi64(acc_vec, 47);
+      __m128i const data_vec = _mm_xor_si128(acc_vec, shifted);
+      /* xacc[i] ^= xsecret[i]; */
+      __m128i const key_vec = _mm_loadu_si128(xsecret + i);
+      __m128i const data_key = _mm_xor_si128(data_vec, key_vec);
+
+      /* xacc[i] *= XXH_PRIME32_1; */
+      __m128i const data_key_hi =
+          _mm_shuffle_epi32(data_key, _MM_SHUFFLE(0, 3, 0, 1));
+      __m128i const prod_lo = _mm_mul_epu32(data_key, prime32);
+      __m128i const prod_hi = _mm_mul_epu32(data_key_hi, prime32);
+      xacc[i] = _mm_add_epi64(prod_lo, _mm_slli_epi64(prod_hi, 32));
+
     }
+
+  }
+
 }
 
-XXH_FORCE_INLINE XXH_TARGET_SSE2 void XXH3_initCustomSecret_sse2(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
-{
-    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 15) == 0);
-    (void)(&XXH_writeLE64);
-    {   int const nbRounds = XXH_SECRET_DEFAULT_SIZE / sizeof(__m128i);
-
-#       if defined(_MSC_VER) && defined(_M_IX86) && _MSC_VER < 1900
-        /* MSVC 32bit mode does not support _mm_set_epi64x before 2015 */
-        XXH_ALIGN(16) const xxh_i64 seed64x2[2] = { (xxh_i64)seed64, (xxh_i64)(0U - seed64) };
-        __m128i const seed = _mm_load_si128((__m128i const*)seed64x2);
-#       else
-        __m128i const seed = _mm_set_epi64x((xxh_i64)(0U - seed64), (xxh_i64)seed64);
-#       endif
-        int i;
-
-        const void* const src16 = XXH3_kSecret;
-        __m128i* dst16 = (__m128i*) customSecret;
-#       if defined(__GNUC__) || defined(__clang__)
-        /*
-         * On GCC & Clang, marking 'dest' as modified will cause the compiler:
-         *   - do not extract the secret from sse registers in the internal loop
-         *   - use less common registers, and avoid pushing these reg into stack
-         */
-        XXH_COMPILER_GUARD(dst16);
-#       endif
-        XXH_ASSERT(((size_t)src16 & 15) == 0); /* control alignment */
-        XXH_ASSERT(((size_t)dst16 & 15) == 0);
+XXH_FORCE_INLINE XXH_TARGET_SSE2 void XXH3_initCustomSecret_sse2(
+    void *XXH_RESTRICT customSecret, xxh_u64 seed64) {
+
+  XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 15) == 0);
+  (void)(&XXH_writeLE64);
+  {
+
+    int const nbRounds = XXH_SECRET_DEFAULT_SIZE / sizeof(__m128i);
+
+        #if defined(_MSC_VER) && defined(_M_IX86) && _MSC_VER < 1900
+    /* MSVC 32bit mode does not support _mm_set_epi64x before 2015 */
+    XXH_ALIGN(16)
+    const xxh_i64 seed64x2[2] = {(xxh_i64)seed64, (xxh_i64)(0U - seed64)};
+    __m128i const seed = _mm_load_si128((__m128i const *)seed64x2);
+        #else
+    __m128i const seed =
+        _mm_set_epi64x((xxh_i64)(0U - seed64), (xxh_i64)seed64);
+        #endif
+    int i;
+
+    const void *const src16 = XXH3_kSecret;
+    __m128i          *dst16 = (__m128i *)customSecret;
+        #if defined(__GNUC__) || defined(__clang__)
+    /*
+     * On GCC & Clang, marking 'dest' as modified will cause the compiler:
+     *   - do not extract the secret from sse registers in the internal loop
+     *   - use less common registers, and avoid pushing these reg into stack
+     */
+    XXH_COMPILER_GUARD(dst16);
+        #endif
+    XXH_ASSERT(((size_t)src16 & 15) == 0);             /* control alignment */
+    XXH_ASSERT(((size_t)dst16 & 15) == 0);
+
+    for (i = 0; i < nbRounds; ++i) {
+
+      dst16[i] =
+          _mm_add_epi64(_mm_load_si128((const __m128i *)src16 + i), seed);
+
+    }
+
+  }
 
-        for (i=0; i < nbRounds; ++i) {
-            dst16[i] = _mm_add_epi64(_mm_load_si128((const __m128i *)src16+i), seed);
-    }   }
 }
 
-#endif
+      #endif
 
-#if (XXH_VECTOR == XXH_NEON)
+      #if (XXH_VECTOR == XXH_NEON)
 
 /* forward declarations for the scalar routines */
-XXH_FORCE_INLINE void
-XXH3_scalarRound(void* XXH_RESTRICT acc, void const* XXH_RESTRICT input,
-                 void const* XXH_RESTRICT secret, size_t lane);
+XXH_FORCE_INLINE void XXH3_scalarRound(void *XXH_RESTRICT       acc,
+                                       void const *XXH_RESTRICT input,
+                                       void const *XXH_RESTRICT secret,
+                                       size_t                   lane);
 
-XXH_FORCE_INLINE void
-XXH3_scalarScrambleRound(void* XXH_RESTRICT acc,
-                         void const* XXH_RESTRICT secret, size_t lane);
+XXH_FORCE_INLINE void XXH3_scalarScrambleRound(void *XXH_RESTRICT       acc,
+                                               void const *XXH_RESTRICT secret,
+                                               size_t                   lane);
 
 /*!
  * @internal
@@ -5168,7 +5808,8 @@ XXH3_scalarScrambleRound(void* XXH_RESTRICT acc,
  * is to optimize the pipelining and can have up to 15% speedup depending on the
  * CPU, and it also mitigates some GCC codegen issues.
  *
- * @see XXH3_NEON_LANES for configuring this and details about this optimization.
+ * @see XXH3_NEON_LANES for configuring this and details about this
+ * optimization.
  *
  * NEON's 32-bit to 64-bit long multiply takes a half vector of 32-bit
  * integers instead of the other platforms which mask full 64-bit vectors,
@@ -5180,740 +5821,866 @@ XXH3_scalarScrambleRound(void* XXH_RESTRICT acc,
  * there needs to be *three* versions of the accumulate operation used
  * for the remaining 2 lanes.
  *
- * WASM's SIMD128 uses SIMDe's arm_neon.h polyfill because the intrinsics overlap
- * nearly perfectly.
+ * WASM's SIMD128 uses SIMDe's arm_neon.h polyfill because the intrinsics
+ * overlap nearly perfectly.
  */
 
-XXH_FORCE_INLINE void
-XXH3_accumulate_512_neon( void* XXH_RESTRICT acc,
-                    const void* XXH_RESTRICT input,
-                    const void* XXH_RESTRICT secret)
-{
-    XXH_ASSERT((((size_t)acc) & 15) == 0);
-    XXH_STATIC_ASSERT(XXH3_NEON_LANES > 0 && XXH3_NEON_LANES <= XXH_ACC_NB && XXH3_NEON_LANES % 2 == 0);
-    {   /* GCC for darwin arm64 does not like aliasing here */
-        xxh_aliasing_uint64x2_t* const xacc = (xxh_aliasing_uint64x2_t*) acc;
-        /* We don't use a uint32x4_t pointer because it causes bus errors on ARMv7. */
-        uint8_t const* xinput = (const uint8_t *) input;
-        uint8_t const* xsecret  = (const uint8_t *) secret;
-
-        size_t i;
-#ifdef __wasm_simd128__
-        /*
-         * On WASM SIMD128, Clang emits direct address loads when XXH3_kSecret
-         * is constant propagated, which results in it converting it to this
-         * inside the loop:
-         *
-         *    a = v128.load(XXH3_kSecret +  0 + $secret_offset, offset = 0)
-         *    b = v128.load(XXH3_kSecret + 16 + $secret_offset, offset = 0)
-         *    ...
-         *
-         * This requires a full 32-bit address immediate (and therefore a 6 byte
-         * instruction) as well as an add for each offset.
-         *
-         * Putting an asm guard prevents it from folding (at the cost of losing
-         * the alignment hint), and uses the free offset in `v128.load` instead
-         * of adding secret_offset each time which overall reduces code size by
-         * about a kilobyte and improves performance.
-         */
-        XXH_COMPILER_GUARD(xsecret);
-#endif
-        /* Scalar lanes use the normal scalarRound routine */
-        for (i = XXH3_NEON_LANES; i < XXH_ACC_NB; i++) {
-            XXH3_scalarRound(acc, input, secret, i);
-        }
-        i = 0;
-        /* 4 NEON lanes at a time. */
-        for (; i+1 < XXH3_NEON_LANES / 2; i+=2) {
-            /* data_vec = xinput[i]; */
-            uint64x2_t data_vec_1 = XXH_vld1q_u64(xinput  + (i * 16));
-            uint64x2_t data_vec_2 = XXH_vld1q_u64(xinput  + ((i+1) * 16));
-            /* key_vec  = xsecret[i];  */
-            uint64x2_t key_vec_1  = XXH_vld1q_u64(xsecret + (i * 16));
-            uint64x2_t key_vec_2  = XXH_vld1q_u64(xsecret + ((i+1) * 16));
-            /* data_swap = swap(data_vec) */
-            uint64x2_t data_swap_1 = vextq_u64(data_vec_1, data_vec_1, 1);
-            uint64x2_t data_swap_2 = vextq_u64(data_vec_2, data_vec_2, 1);
-            /* data_key = data_vec ^ key_vec; */
-            uint64x2_t data_key_1 = veorq_u64(data_vec_1, key_vec_1);
-            uint64x2_t data_key_2 = veorq_u64(data_vec_2, key_vec_2);
-
-            /*
-             * If we reinterpret the 64x2 vectors as 32x4 vectors, we can use a
-             * de-interleave operation for 4 lanes in 1 step with `vuzpq_u32` to
-             * get one vector with the low 32 bits of each lane, and one vector
-             * with the high 32 bits of each lane.
-             *
-             * The intrinsic returns a double vector because the original ARMv7-a
-             * instruction modified both arguments in place. AArch64 and SIMD128 emit
-             * two instructions from this intrinsic.
-             *
-             *  [ dk11L | dk11H | dk12L | dk12H ] -> [ dk11L | dk12L | dk21L | dk22L ]
-             *  [ dk21L | dk21H | dk22L | dk22H ] -> [ dk11H | dk12H | dk21H | dk22H ]
-             */
-            uint32x4x2_t unzipped = vuzpq_u32(
-                vreinterpretq_u32_u64(data_key_1),
-                vreinterpretq_u32_u64(data_key_2)
-            );
-            /* data_key_lo = data_key & 0xFFFFFFFF */
-            uint32x4_t data_key_lo = unzipped.val[0];
-            /* data_key_hi = data_key >> 32 */
-            uint32x4_t data_key_hi = unzipped.val[1];
-            /*
-             * Then, we can split the vectors horizontally and multiply which, as for most
-             * widening intrinsics, have a variant that works on both high half vectors
-             * for free on AArch64. A similar instruction is available on SIMD128.
-             *
-             * sum = data_swap + (u64x2) data_key_lo * (u64x2) data_key_hi
-             */
-            uint64x2_t sum_1 = XXH_vmlal_low_u32(data_swap_1, data_key_lo, data_key_hi);
-            uint64x2_t sum_2 = XXH_vmlal_high_u32(data_swap_2, data_key_lo, data_key_hi);
-            /*
-             * Clang reorders
-             *    a += b * c;     // umlal   swap.2d, dkl.2s, dkh.2s
-             *    c += a;         // add     acc.2d, acc.2d, swap.2d
-             * to
-             *    c += a;         // add     acc.2d, acc.2d, swap.2d
-             *    c += b * c;     // umlal   acc.2d, dkl.2s, dkh.2s
-             *
-             * While it would make sense in theory since the addition is faster,
-             * for reasons likely related to umlal being limited to certain NEON
-             * pipelines, this is worse. A compiler guard fixes this.
-             */
-            XXH_COMPILER_GUARD_CLANG_NEON(sum_1);
-            XXH_COMPILER_GUARD_CLANG_NEON(sum_2);
-            /* xacc[i] = acc_vec + sum; */
-            xacc[i]   = vaddq_u64(xacc[i], sum_1);
-            xacc[i+1] = vaddq_u64(xacc[i+1], sum_2);
-        }
-        /* Operate on the remaining NEON lanes 2 at a time. */
-        for (; i < XXH3_NEON_LANES / 2; i++) {
-            /* data_vec = xinput[i]; */
-            uint64x2_t data_vec = XXH_vld1q_u64(xinput  + (i * 16));
-            /* key_vec  = xsecret[i];  */
-            uint64x2_t key_vec  = XXH_vld1q_u64(xsecret + (i * 16));
-            /* acc_vec_2 = swap(data_vec) */
-            uint64x2_t data_swap = vextq_u64(data_vec, data_vec, 1);
-            /* data_key = data_vec ^ key_vec; */
-            uint64x2_t data_key = veorq_u64(data_vec, key_vec);
-            /* For two lanes, just use VMOVN and VSHRN. */
-            /* data_key_lo = data_key & 0xFFFFFFFF; */
-            uint32x2_t data_key_lo = vmovn_u64(data_key);
-            /* data_key_hi = data_key >> 32; */
-            uint32x2_t data_key_hi = vshrn_n_u64(data_key, 32);
-            /* sum = data_swap + (u64x2) data_key_lo * (u64x2) data_key_hi; */
-            uint64x2_t sum = vmlal_u32(data_swap, data_key_lo, data_key_hi);
-            /* Same Clang workaround as before */
-            XXH_COMPILER_GUARD_CLANG_NEON(sum);
-            /* xacc[i] = acc_vec + sum; */
-            xacc[i] = vaddq_u64 (xacc[i], sum);
-        }
+XXH_FORCE_INLINE void XXH3_accumulate_512_neon(
+    void *XXH_RESTRICT acc, const void *XXH_RESTRICT input,
+    const void *XXH_RESTRICT secret) {
+
+  XXH_ASSERT((((size_t)acc) & 15) == 0);
+  XXH_STATIC_ASSERT(XXH3_NEON_LANES > 0 && XXH3_NEON_LANES <= XXH_ACC_NB &&
+                    XXH3_NEON_LANES % 2 == 0);
+  {                     /* GCC for darwin arm64 does not like aliasing here */
+    xxh_aliasing_uint64x2_t *const xacc = (xxh_aliasing_uint64x2_t *)acc;
+    /* We don't use a uint32x4_t pointer because it causes bus errors on ARMv7.
+     */
+    uint8_t const *xinput = (const uint8_t *)input;
+    uint8_t const *xsecret = (const uint8_t *)secret;
+
+    size_t i;
+        #ifdef __wasm_simd128__
+    /*
+     * On WASM SIMD128, Clang emits direct address loads when XXH3_kSecret
+     * is constant propagated, which results in it converting it to this
+     * inside the loop:
+     *
+     *    a = v128.load(XXH3_kSecret +  0 + $secret_offset, offset = 0)
+     *    b = v128.load(XXH3_kSecret + 16 + $secret_offset, offset = 0)
+     *    ...
+     *
+     * This requires a full 32-bit address immediate (and therefore a 6 byte
+     * instruction) as well as an add for each offset.
+     *
+     * Putting an asm guard prevents it from folding (at the cost of losing
+     * the alignment hint), and uses the free offset in `v128.load` instead
+     * of adding secret_offset each time which overall reduces code size by
+     * about a kilobyte and improves performance.
+     */
+    XXH_COMPILER_GUARD(xsecret);
+        #endif
+    /* Scalar lanes use the normal scalarRound routine */
+    for (i = XXH3_NEON_LANES; i < XXH_ACC_NB; i++) {
+
+      XXH3_scalarRound(acc, input, secret, i);
+
+    }
+
+    i = 0;
+    /* 4 NEON lanes at a time. */
+    for (; i + 1 < XXH3_NEON_LANES / 2; i += 2) {
+
+      /* data_vec = xinput[i]; */
+      uint64x2_t data_vec_1 = XXH_vld1q_u64(xinput + (i * 16));
+      uint64x2_t data_vec_2 = XXH_vld1q_u64(xinput + ((i + 1) * 16));
+      /* key_vec  = xsecret[i];  */
+      uint64x2_t key_vec_1 = XXH_vld1q_u64(xsecret + (i * 16));
+      uint64x2_t key_vec_2 = XXH_vld1q_u64(xsecret + ((i + 1) * 16));
+      /* data_swap = swap(data_vec) */
+      uint64x2_t data_swap_1 = vextq_u64(data_vec_1, data_vec_1, 1);
+      uint64x2_t data_swap_2 = vextq_u64(data_vec_2, data_vec_2, 1);
+      /* data_key = data_vec ^ key_vec; */
+      uint64x2_t data_key_1 = veorq_u64(data_vec_1, key_vec_1);
+      uint64x2_t data_key_2 = veorq_u64(data_vec_2, key_vec_2);
+
+      /*
+       * If we reinterpret the 64x2 vectors as 32x4 vectors, we can use a
+       * de-interleave operation for 4 lanes in 1 step with `vuzpq_u32` to
+       * get one vector with the low 32 bits of each lane, and one vector
+       * with the high 32 bits of each lane.
+       *
+       * The intrinsic returns a double vector because the original ARMv7-a
+       * instruction modified both arguments in place. AArch64 and SIMD128 emit
+       * two instructions from this intrinsic.
+       *
+       *  [ dk11L | dk11H | dk12L | dk12H ] -> [ dk11L | dk12L | dk21L | dk22L ]
+       *  [ dk21L | dk21H | dk22L | dk22H ] -> [ dk11H | dk12H | dk21H | dk22H ]
+       */
+      uint32x4x2_t unzipped = vuzpq_u32(vreinterpretq_u32_u64(data_key_1),
+                                        vreinterpretq_u32_u64(data_key_2));
+      /* data_key_lo = data_key & 0xFFFFFFFF */
+      uint32x4_t data_key_lo = unzipped.val[0];
+      /* data_key_hi = data_key >> 32 */
+      uint32x4_t data_key_hi = unzipped.val[1];
+      /*
+       * Then, we can split the vectors horizontally and multiply which, as for
+       * most widening intrinsics, have a variant that works on both high half
+       * vectors for free on AArch64. A similar instruction is available on
+       * SIMD128.
+       *
+       * sum = data_swap + (u64x2) data_key_lo * (u64x2) data_key_hi
+       */
+      uint64x2_t sum_1 =
+          XXH_vmlal_low_u32(data_swap_1, data_key_lo, data_key_hi);
+      uint64x2_t sum_2 =
+          XXH_vmlal_high_u32(data_swap_2, data_key_lo, data_key_hi);
+      /*
+       * Clang reorders
+       *    a += b * c;     // umlal   swap.2d, dkl.2s, dkh.2s
+       *    c += a;         // add     acc.2d, acc.2d, swap.2d
+       * to
+       *    c += a;         // add     acc.2d, acc.2d, swap.2d
+       *    c += b * c;     // umlal   acc.2d, dkl.2s, dkh.2s
+       *
+       * While it would make sense in theory since the addition is faster,
+       * for reasons likely related to umlal being limited to certain NEON
+       * pipelines, this is worse. A compiler guard fixes this.
+       */
+      XXH_COMPILER_GUARD_CLANG_NEON(sum_1);
+      XXH_COMPILER_GUARD_CLANG_NEON(sum_2);
+      /* xacc[i] = acc_vec + sum; */
+      xacc[i] = vaddq_u64(xacc[i], sum_1);
+      xacc[i + 1] = vaddq_u64(xacc[i + 1], sum_2);
+
+    }
+
+    /* Operate on the remaining NEON lanes 2 at a time. */
+    for (; i < XXH3_NEON_LANES / 2; i++) {
+
+      /* data_vec = xinput[i]; */
+      uint64x2_t data_vec = XXH_vld1q_u64(xinput + (i * 16));
+      /* key_vec  = xsecret[i];  */
+      uint64x2_t key_vec = XXH_vld1q_u64(xsecret + (i * 16));
+      /* acc_vec_2 = swap(data_vec) */
+      uint64x2_t data_swap = vextq_u64(data_vec, data_vec, 1);
+      /* data_key = data_vec ^ key_vec; */
+      uint64x2_t data_key = veorq_u64(data_vec, key_vec);
+      /* For two lanes, just use VMOVN and VSHRN. */
+      /* data_key_lo = data_key & 0xFFFFFFFF; */
+      uint32x2_t data_key_lo = vmovn_u64(data_key);
+      /* data_key_hi = data_key >> 32; */
+      uint32x2_t data_key_hi = vshrn_n_u64(data_key, 32);
+      /* sum = data_swap + (u64x2) data_key_lo * (u64x2) data_key_hi; */
+      uint64x2_t sum = vmlal_u32(data_swap, data_key_lo, data_key_hi);
+      /* Same Clang workaround as before */
+      XXH_COMPILER_GUARD_CLANG_NEON(sum);
+      /* xacc[i] = acc_vec + sum; */
+      xacc[i] = vaddq_u64(xacc[i], sum);
+
     }
+
+  }
+
 }
+
 XXH_FORCE_INLINE XXH3_ACCUMULATE_TEMPLATE(neon)
 
-XXH_FORCE_INLINE void
-XXH3_scrambleAcc_neon(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
-{
-    XXH_ASSERT((((size_t)acc) & 15) == 0);
-
-    {   xxh_aliasing_uint64x2_t* xacc       = (xxh_aliasing_uint64x2_t*) acc;
-        uint8_t const* xsecret = (uint8_t const*) secret;
-
-        size_t i;
-        /* WASM uses operator overloads and doesn't need these. */
-#ifndef __wasm_simd128__
-        /* { prime32_1, prime32_1 } */
-        uint32x2_t const kPrimeLo = vdup_n_u32(XXH_PRIME32_1);
-        /* { 0, prime32_1, 0, prime32_1 } */
-        uint32x4_t const kPrimeHi = vreinterpretq_u32_u64(vdupq_n_u64((xxh_u64)XXH_PRIME32_1 << 32));
-#endif
+    XXH_FORCE_INLINE
+    void XXH3_scrambleAcc_neon(void *XXH_RESTRICT       acc,
+                               const void *XXH_RESTRICT secret) {
 
-        /* AArch64 uses both scalar and neon at the same time */
-        for (i = XXH3_NEON_LANES; i < XXH_ACC_NB; i++) {
-            XXH3_scalarScrambleRound(acc, secret, i);
-        }
-        for (i=0; i < XXH3_NEON_LANES / 2; i++) {
-            /* xacc[i] ^= (xacc[i] >> 47); */
-            uint64x2_t acc_vec  = xacc[i];
-            uint64x2_t shifted  = vshrq_n_u64(acc_vec, 47);
-            uint64x2_t data_vec = veorq_u64(acc_vec, shifted);
-
-            /* xacc[i] ^= xsecret[i]; */
-            uint64x2_t key_vec  = XXH_vld1q_u64(xsecret + (i * 16));
-            uint64x2_t data_key = veorq_u64(data_vec, key_vec);
+  XXH_ASSERT((((size_t)acc) & 15) == 0);
+
+  {
+
+    xxh_aliasing_uint64x2_t *xacc = (xxh_aliasing_uint64x2_t *)acc;
+    uint8_t const           *xsecret = (uint8_t const *)secret;
+
+    size_t i;
+          /* WASM uses operator overloads and doesn't need these. */
+        #ifndef __wasm_simd128__
+    /* { prime32_1, prime32_1 } */
+    uint32x2_t const kPrimeLo = vdup_n_u32(XXH_PRIME32_1);
+    /* { 0, prime32_1, 0, prime32_1 } */
+    uint32x4_t const kPrimeHi =
+        vreinterpretq_u32_u64(vdupq_n_u64((xxh_u64)XXH_PRIME32_1 << 32));
+        #endif
+
+    /* AArch64 uses both scalar and neon at the same time */
+    for (i = XXH3_NEON_LANES; i < XXH_ACC_NB; i++) {
+
+      XXH3_scalarScrambleRound(acc, secret, i);
+
+    }
+
+    for (i = 0; i < XXH3_NEON_LANES / 2; i++) {
+
+      /* xacc[i] ^= (xacc[i] >> 47); */
+      uint64x2_t acc_vec = xacc[i];
+      uint64x2_t shifted = vshrq_n_u64(acc_vec, 47);
+      uint64x2_t data_vec = veorq_u64(acc_vec, shifted);
+
+      /* xacc[i] ^= xsecret[i]; */
+      uint64x2_t key_vec = XXH_vld1q_u64(xsecret + (i * 16));
+      uint64x2_t data_key = veorq_u64(data_vec, key_vec);
             /* xacc[i] *= XXH_PRIME32_1 */
-#ifdef __wasm_simd128__
-            /* SIMD128 has multiply by u64x2, use it instead of expanding and scalarizing */
-            xacc[i] = data_key * XXH_PRIME32_1;
-#else
-            /*
-             * Expanded version with portable NEON intrinsics
-             *
-             *    lo(x) * lo(y) + (hi(x) * lo(y) << 32)
-             *
-             * prod_hi = hi(data_key) * lo(prime) << 32
-             *
-             * Since we only need 32 bits of this multiply a trick can be used, reinterpreting the vector
-             * as a uint32x4_t and multiplying by { 0, prime, 0, prime } to cancel out the unwanted bits
-             * and avoid the shift.
-             */
-            uint32x4_t prod_hi = vmulq_u32 (vreinterpretq_u32_u64(data_key), kPrimeHi);
-            /* Extract low bits for vmlal_u32  */
-            uint32x2_t data_key_lo = vmovn_u64(data_key);
-            /* xacc[i] = prod_hi + lo(data_key) * XXH_PRIME32_1; */
-            xacc[i] = vmlal_u32(vreinterpretq_u64_u32(prod_hi), data_key_lo, kPrimeLo);
-#endif
-        }
+        #ifdef __wasm_simd128__
+      /* SIMD128 has multiply by u64x2, use it instead of expanding and
+       * scalarizing */
+      xacc[i] = data_key * XXH_PRIME32_1;
+        #else
+      /*
+       * Expanded version with portable NEON intrinsics
+       *
+       *    lo(x) * lo(y) + (hi(x) * lo(y) << 32)
+       *
+       * prod_hi = hi(data_key) * lo(prime) << 32
+       *
+       * Since we only need 32 bits of this multiply a trick can be used,
+       * reinterpreting the vector as a uint32x4_t and multiplying by { 0,
+       * prime, 0, prime } to cancel out the unwanted bits and avoid the shift.
+       */
+      uint32x4_t prod_hi = vmulq_u32(vreinterpretq_u32_u64(data_key), kPrimeHi);
+      /* Extract low bits for vmlal_u32  */
+      uint32x2_t data_key_lo = vmovn_u64(data_key);
+      /* xacc[i] = prod_hi + lo(data_key) * XXH_PRIME32_1; */
+      xacc[i] =
+          vmlal_u32(vreinterpretq_u64_u32(prod_hi), data_key_lo, kPrimeLo);
+        #endif
+
     }
+
+  }
+
 }
-#endif
 
-#if (XXH_VECTOR == XXH_VSX)
+      #endif
+
+      #if (XXH_VECTOR == XXH_VSX)
+
+XXH_FORCE_INLINE void XXH3_accumulate_512_vsx(void *XXH_RESTRICT       acc,
+                                              const void *XXH_RESTRICT input,
+                                              const void *XXH_RESTRICT secret) {
+
+  /* presumed aligned */
+  xxh_aliasing_u64x2 *const xacc = (xxh_aliasing_u64x2 *)acc;
+  xxh_u8 const *const       xinput =
+      (xxh_u8 const *)input;                    /* no alignment restriction */
+  xxh_u8 const *const xsecret =
+      (xxh_u8 const *)secret;                   /* no alignment restriction */
+  xxh_u64x2 const v32 = {32, 32};
+  size_t          i;
+  for (i = 0; i < XXH_STRIPE_LEN / sizeof(xxh_u64x2); i++) {
+
+    /* data_vec = xinput[i]; */
+    xxh_u64x2 const data_vec = XXH_vec_loadu(xinput + 16 * i);
+    /* key_vec = xsecret[i]; */
+    xxh_u64x2 const key_vec = XXH_vec_loadu(xsecret + 16 * i);
+    xxh_u64x2 const data_key = data_vec ^ key_vec;
+    /* shuffled = (data_key << 32) | (data_key >> 32); */
+    xxh_u32x4 const shuffled = (xxh_u32x4)vec_rl(data_key, v32);
+    /* product = ((xxh_u64x2)data_key & 0xFFFFFFFF) * ((xxh_u64x2)shuffled &
+     * 0xFFFFFFFF); */
+    xxh_u64x2 const product = XXH_vec_mulo((xxh_u32x4)data_key, shuffled);
+    /* acc_vec = xacc[i]; */
+    xxh_u64x2 acc_vec = xacc[i];
+    acc_vec += product;
+
+          /* swap high and low halves */
+        #ifdef __s390x__
+    acc_vec += vec_permi(data_vec, data_vec, 2);
+        #else
+    acc_vec += vec_xxpermdi(data_vec, data_vec, 2);
+        #endif
+    xacc[i] = acc_vec;
+
+  }
+
+}
+
+XXH_FORCE_INLINE XXH3_ACCUMULATE_TEMPLATE(vsx)
+
+    XXH_FORCE_INLINE
+    void XXH3_scrambleAcc_vsx(void *XXH_RESTRICT       acc,
+                              const void *XXH_RESTRICT secret) {
+
+  XXH_ASSERT((((size_t)acc) & 15) == 0);
+
+  {
 
-XXH_FORCE_INLINE void
-XXH3_accumulate_512_vsx(  void* XXH_RESTRICT acc,
-                    const void* XXH_RESTRICT input,
-                    const void* XXH_RESTRICT secret)
-{
-    /* presumed aligned */
-    xxh_aliasing_u64x2* const xacc = (xxh_aliasing_u64x2*) acc;
-    xxh_u8 const* const xinput   = (xxh_u8 const*) input;   /* no alignment restriction */
-    xxh_u8 const* const xsecret  = (xxh_u8 const*) secret;    /* no alignment restriction */
-    xxh_u64x2 const v32 = { 32, 32 };
-    size_t i;
+    xxh_aliasing_u64x2 *const xacc = (xxh_aliasing_u64x2 *)acc;
+    const xxh_u8 *const       xsecret = (const xxh_u8 *)secret;
+    /* constants */
+    xxh_u64x2 const v32 = {32, 32};
+    xxh_u64x2 const v47 = {47, 47};
+    xxh_u32x4 const prime = {XXH_PRIME32_1, XXH_PRIME32_1, XXH_PRIME32_1,
+                             XXH_PRIME32_1};
+    size_t          i;
     for (i = 0; i < XXH_STRIPE_LEN / sizeof(xxh_u64x2); i++) {
-        /* data_vec = xinput[i]; */
-        xxh_u64x2 const data_vec = XXH_vec_loadu(xinput + 16*i);
-        /* key_vec = xsecret[i]; */
-        xxh_u64x2 const key_vec  = XXH_vec_loadu(xsecret + 16*i);
-        xxh_u64x2 const data_key = data_vec ^ key_vec;
-        /* shuffled = (data_key << 32) | (data_key >> 32); */
-        xxh_u32x4 const shuffled = (xxh_u32x4)vec_rl(data_key, v32);
-        /* product = ((xxh_u64x2)data_key & 0xFFFFFFFF) * ((xxh_u64x2)shuffled & 0xFFFFFFFF); */
-        xxh_u64x2 const product  = XXH_vec_mulo((xxh_u32x4)data_key, shuffled);
-        /* acc_vec = xacc[i]; */
-        xxh_u64x2 acc_vec        = xacc[i];
-        acc_vec += product;
-
-        /* swap high and low halves */
-#ifdef __s390x__
-        acc_vec += vec_permi(data_vec, data_vec, 2);
-#else
-        acc_vec += vec_xxpermdi(data_vec, data_vec, 2);
-#endif
-        xacc[i] = acc_vec;
+
+      /* xacc[i] ^= (xacc[i] >> 47); */
+      xxh_u64x2 const acc_vec = xacc[i];
+      xxh_u64x2 const data_vec = acc_vec ^ (acc_vec >> v47);
+
+      /* xacc[i] ^= xsecret[i]; */
+      xxh_u64x2 const key_vec = XXH_vec_loadu(xsecret + 16 * i);
+      xxh_u64x2 const data_key = data_vec ^ key_vec;
+
+      /* xacc[i] *= XXH_PRIME32_1 */
+      /* prod_lo = ((xxh_u64x2)data_key & 0xFFFFFFFF) * ((xxh_u64x2)prime &
+       * 0xFFFFFFFF);  */
+      xxh_u64x2 const prod_even = XXH_vec_mule((xxh_u32x4)data_key, prime);
+      /* prod_hi = ((xxh_u64x2)data_key >> 32) * ((xxh_u64x2)prime >> 32);  */
+      xxh_u64x2 const prod_odd = XXH_vec_mulo((xxh_u32x4)data_key, prime);
+      xacc[i] = prod_odd + (prod_even << v32);
+
     }
+
+  }
+
 }
-XXH_FORCE_INLINE XXH3_ACCUMULATE_TEMPLATE(vsx)
 
-XXH_FORCE_INLINE void
-XXH3_scrambleAcc_vsx(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
-{
-    XXH_ASSERT((((size_t)acc) & 15) == 0);
-
-    {   xxh_aliasing_u64x2* const xacc = (xxh_aliasing_u64x2*) acc;
-        const xxh_u8* const xsecret = (const xxh_u8*) secret;
-        /* constants */
-        xxh_u64x2 const v32  = { 32, 32 };
-        xxh_u64x2 const v47 = { 47, 47 };
-        xxh_u32x4 const prime = { XXH_PRIME32_1, XXH_PRIME32_1, XXH_PRIME32_1, XXH_PRIME32_1 };
-        size_t i;
-        for (i = 0; i < XXH_STRIPE_LEN / sizeof(xxh_u64x2); i++) {
-            /* xacc[i] ^= (xacc[i] >> 47); */
-            xxh_u64x2 const acc_vec  = xacc[i];
-            xxh_u64x2 const data_vec = acc_vec ^ (acc_vec >> v47);
-
-            /* xacc[i] ^= xsecret[i]; */
-            xxh_u64x2 const key_vec  = XXH_vec_loadu(xsecret + 16*i);
-            xxh_u64x2 const data_key = data_vec ^ key_vec;
+      #endif
+
+      #if (XXH_VECTOR == XXH_SVE)
+
+XXH_FORCE_INLINE void XXH3_accumulate_512_sve(void *XXH_RESTRICT       acc,
+                                              const void *XXH_RESTRICT input,
+                                              const void *XXH_RESTRICT secret) {
+
+  uint64_t       *xacc = (uint64_t *)acc;
+  const uint64_t *xinput = (const uint64_t *)(const void *)input;
+  const uint64_t *xsecret = (const uint64_t *)(const void *)secret;
+  svuint64_t      kSwap = sveor_n_u64_z(svptrue_b64(), svindex_u64(0, 1), 1);
+  uint64_t        element_count = svcntd();
+  if (element_count >= 8) {
+
+    svbool_t   mask = svptrue_pat_b64(SV_VL8);
+    svuint64_t vacc = svld1_u64(mask, xacc);
+    ACCRND(vacc, 0);
+    svst1_u64(mask, xacc, vacc);
+
+  } else if (element_count == 2) {                                /* sve128 */
+
+    svbool_t   mask = svptrue_pat_b64(SV_VL2);
+    svuint64_t acc0 = svld1_u64(mask, xacc + 0);
+    svuint64_t acc1 = svld1_u64(mask, xacc + 2);
+    svuint64_t acc2 = svld1_u64(mask, xacc + 4);
+    svuint64_t acc3 = svld1_u64(mask, xacc + 6);
+    ACCRND(acc0, 0);
+    ACCRND(acc1, 2);
+    ACCRND(acc2, 4);
+    ACCRND(acc3, 6);
+    svst1_u64(mask, xacc + 0, acc0);
+    svst1_u64(mask, xacc + 2, acc1);
+    svst1_u64(mask, xacc + 4, acc2);
+    svst1_u64(mask, xacc + 6, acc3);
+
+  } else {
+
+    svbool_t   mask = svptrue_pat_b64(SV_VL4);
+    svuint64_t acc0 = svld1_u64(mask, xacc + 0);
+    svuint64_t acc1 = svld1_u64(mask, xacc + 4);
+    ACCRND(acc0, 0);
+    ACCRND(acc1, 4);
+    svst1_u64(mask, xacc + 0, acc0);
+    svst1_u64(mask, xacc + 4, acc1);
+
+  }
 
-            /* xacc[i] *= XXH_PRIME32_1 */
-            /* prod_lo = ((xxh_u64x2)data_key & 0xFFFFFFFF) * ((xxh_u64x2)prime & 0xFFFFFFFF);  */
-            xxh_u64x2 const prod_even  = XXH_vec_mule((xxh_u32x4)data_key, prime);
-            /* prod_hi = ((xxh_u64x2)data_key >> 32) * ((xxh_u64x2)prime >> 32);  */
-            xxh_u64x2 const prod_odd  = XXH_vec_mulo((xxh_u32x4)data_key, prime);
-            xacc[i] = prod_odd + (prod_even << v32);
-    }   }
 }
 
-#endif
+XXH_FORCE_INLINE void XXH3_accumulate_sve(xxh_u64 *XXH_RESTRICT      acc,
+                                          const xxh_u8 *XXH_RESTRICT input,
+                                          const xxh_u8 *XXH_RESTRICT secret,
+                                          size_t nbStripes) {
 
-#if (XXH_VECTOR == XXH_SVE)
+  if (nbStripes != 0) {
 
-XXH_FORCE_INLINE void
-XXH3_accumulate_512_sve( void* XXH_RESTRICT acc,
-                   const void* XXH_RESTRICT input,
-                   const void* XXH_RESTRICT secret)
-{
-    uint64_t *xacc = (uint64_t *)acc;
+    uint64_t       *xacc = (uint64_t *)acc;
     const uint64_t *xinput = (const uint64_t *)(const void *)input;
     const uint64_t *xsecret = (const uint64_t *)(const void *)secret;
-    svuint64_t kSwap = sveor_n_u64_z(svptrue_b64(), svindex_u64(0, 1), 1);
-    uint64_t element_count = svcntd();
+    svuint64_t      kSwap = sveor_n_u64_z(svptrue_b64(), svindex_u64(0, 1), 1);
+    uint64_t        element_count = svcntd();
     if (element_count >= 8) {
-        svbool_t mask = svptrue_pat_b64(SV_VL8);
-        svuint64_t vacc = svld1_u64(mask, xacc);
+
+      svbool_t   mask = svptrue_pat_b64(SV_VL8);
+      svuint64_t vacc = svld1_u64(mask, xacc + 0);
+      do {
+
+        /* svprfd(svbool_t, void *, enum svfprop); */
+        svprfd(mask, xinput + 128, SV_PLDL1STRM);
         ACCRND(vacc, 0);
-        svst1_u64(mask, xacc, vacc);
-    } else if (element_count == 2) {   /* sve128 */
-        svbool_t mask = svptrue_pat_b64(SV_VL2);
-        svuint64_t acc0 = svld1_u64(mask, xacc + 0);
-        svuint64_t acc1 = svld1_u64(mask, xacc + 2);
-        svuint64_t acc2 = svld1_u64(mask, xacc + 4);
-        svuint64_t acc3 = svld1_u64(mask, xacc + 6);
+        xinput += 8;
+        xsecret += 1;
+        nbStripes--;
+
+      } while (nbStripes != 0);
+
+      svst1_u64(mask, xacc + 0, vacc);
+
+    } else if (element_count == 2) {                              /* sve128 */
+
+      svbool_t   mask = svptrue_pat_b64(SV_VL2);
+      svuint64_t acc0 = svld1_u64(mask, xacc + 0);
+      svuint64_t acc1 = svld1_u64(mask, xacc + 2);
+      svuint64_t acc2 = svld1_u64(mask, xacc + 4);
+      svuint64_t acc3 = svld1_u64(mask, xacc + 6);
+      do {
+
+        svprfd(mask, xinput + 128, SV_PLDL1STRM);
         ACCRND(acc0, 0);
         ACCRND(acc1, 2);
         ACCRND(acc2, 4);
         ACCRND(acc3, 6);
-        svst1_u64(mask, xacc + 0, acc0);
-        svst1_u64(mask, xacc + 2, acc1);
-        svst1_u64(mask, xacc + 4, acc2);
-        svst1_u64(mask, xacc + 6, acc3);
+        xinput += 8;
+        xsecret += 1;
+        nbStripes--;
+
+      } while (nbStripes != 0);
+
+      svst1_u64(mask, xacc + 0, acc0);
+      svst1_u64(mask, xacc + 2, acc1);
+      svst1_u64(mask, xacc + 4, acc2);
+      svst1_u64(mask, xacc + 6, acc3);
+
     } else {
-        svbool_t mask = svptrue_pat_b64(SV_VL4);
-        svuint64_t acc0 = svld1_u64(mask, xacc + 0);
-        svuint64_t acc1 = svld1_u64(mask, xacc + 4);
+
+      svbool_t   mask = svptrue_pat_b64(SV_VL4);
+      svuint64_t acc0 = svld1_u64(mask, xacc + 0);
+      svuint64_t acc1 = svld1_u64(mask, xacc + 4);
+      do {
+
+        svprfd(mask, xinput + 128, SV_PLDL1STRM);
         ACCRND(acc0, 0);
         ACCRND(acc1, 4);
-        svst1_u64(mask, xacc + 0, acc0);
-        svst1_u64(mask, xacc + 4, acc1);
-    }
-}
+        xinput += 8;
+        xsecret += 1;
+        nbStripes--;
+
+      } while (nbStripes != 0);
+
+      svst1_u64(mask, xacc + 0, acc0);
+      svst1_u64(mask, xacc + 4, acc1);
 
-XXH_FORCE_INLINE void
-XXH3_accumulate_sve(xxh_u64* XXH_RESTRICT acc,
-               const xxh_u8* XXH_RESTRICT input,
-               const xxh_u8* XXH_RESTRICT secret,
-               size_t nbStripes)
-{
-    if (nbStripes != 0) {
-        uint64_t *xacc = (uint64_t *)acc;
-        const uint64_t *xinput = (const uint64_t *)(const void *)input;
-        const uint64_t *xsecret = (const uint64_t *)(const void *)secret;
-        svuint64_t kSwap = sveor_n_u64_z(svptrue_b64(), svindex_u64(0, 1), 1);
-        uint64_t element_count = svcntd();
-        if (element_count >= 8) {
-            svbool_t mask = svptrue_pat_b64(SV_VL8);
-            svuint64_t vacc = svld1_u64(mask, xacc + 0);
-            do {
-                /* svprfd(svbool_t, void *, enum svfprop); */
-                svprfd(mask, xinput + 128, SV_PLDL1STRM);
-                ACCRND(vacc, 0);
-                xinput += 8;
-                xsecret += 1;
-                nbStripes--;
-           } while (nbStripes != 0);
-
-           svst1_u64(mask, xacc + 0, vacc);
-        } else if (element_count == 2) { /* sve128 */
-            svbool_t mask = svptrue_pat_b64(SV_VL2);
-            svuint64_t acc0 = svld1_u64(mask, xacc + 0);
-            svuint64_t acc1 = svld1_u64(mask, xacc + 2);
-            svuint64_t acc2 = svld1_u64(mask, xacc + 4);
-            svuint64_t acc3 = svld1_u64(mask, xacc + 6);
-            do {
-                svprfd(mask, xinput + 128, SV_PLDL1STRM);
-                ACCRND(acc0, 0);
-                ACCRND(acc1, 2);
-                ACCRND(acc2, 4);
-                ACCRND(acc3, 6);
-                xinput += 8;
-                xsecret += 1;
-                nbStripes--;
-           } while (nbStripes != 0);
-
-           svst1_u64(mask, xacc + 0, acc0);
-           svst1_u64(mask, xacc + 2, acc1);
-           svst1_u64(mask, xacc + 4, acc2);
-           svst1_u64(mask, xacc + 6, acc3);
-        } else {
-            svbool_t mask = svptrue_pat_b64(SV_VL4);
-            svuint64_t acc0 = svld1_u64(mask, xacc + 0);
-            svuint64_t acc1 = svld1_u64(mask, xacc + 4);
-            do {
-                svprfd(mask, xinput + 128, SV_PLDL1STRM);
-                ACCRND(acc0, 0);
-                ACCRND(acc1, 4);
-                xinput += 8;
-                xsecret += 1;
-                nbStripes--;
-           } while (nbStripes != 0);
-
-           svst1_u64(mask, xacc + 0, acc0);
-           svst1_u64(mask, xacc + 4, acc1);
-       }
     }
+
+  }
+
 }
 
-#endif
+      #endif
 
-/* scalar variants - universal */
+    /* scalar variants - universal */
 
-#if defined(__aarch64__) && (defined(__GNUC__) || defined(__clang__))
+      #if defined(__aarch64__) && (defined(__GNUC__) || defined(__clang__))
 /*
  * In XXH3_scalarRound(), GCC and Clang have a similar codegen issue, where they
  * emit an excess mask and a full 64-bit multiply-add (MADD X-form).
  *
- * While this might not seem like much, as AArch64 is a 64-bit architecture, only
- * big Cortex designs have a full 64-bit multiplier.
+ * While this might not seem like much, as AArch64 is a 64-bit architecture,
+ * only big Cortex designs have a full 64-bit multiplier.
  *
  * On the little cores, the smaller 32-bit multiplier is used, and full 64-bit
  * multiplies expand to 2-3 multiplies in microcode. This has a major penalty
  * of up to 4 latency cycles and 2 stall cycles in the multiply pipeline.
  *
- * Thankfully, AArch64 still provides the 32-bit long multiply-add (UMADDL) which does
- * not have this penalty and does the mask automatically.
+ * Thankfully, AArch64 still provides the 32-bit long multiply-add (UMADDL)
+ * which does not have this penalty and does the mask automatically.
  */
-XXH_FORCE_INLINE xxh_u64
-XXH_mult32to64_add64(xxh_u64 lhs, xxh_u64 rhs, xxh_u64 acc)
-{
-    xxh_u64 ret;
-    /* note: %x = 64-bit register, %w = 32-bit register */
-    __asm__("umaddl %x0, %w1, %w2, %x3" : "=r" (ret) : "r" (lhs), "r" (rhs), "r" (acc));
-    return ret;
-}
-#else
-XXH_FORCE_INLINE xxh_u64
-XXH_mult32to64_add64(xxh_u64 lhs, xxh_u64 rhs, xxh_u64 acc)
-{
-    return XXH_mult32to64((xxh_u32)lhs, (xxh_u32)rhs) + acc;
+XXH_FORCE_INLINE xxh_u64 XXH_mult32to64_add64(xxh_u64 lhs, xxh_u64 rhs,
+                                              xxh_u64 acc) {
+
+  xxh_u64 ret;
+  /* note: %x = 64-bit register, %w = 32-bit register */
+  __asm__("umaddl %x0, %w1, %w2, %x3"
+          : "=r"(ret)
+          : "r"(lhs), "r"(rhs), "r"(acc));
+  return ret;
+
 }
-#endif
+
+      #else
+XXH_FORCE_INLINE xxh_u64 XXH_mult32to64_add64(xxh_u64 lhs, xxh_u64 rhs,
+                                              xxh_u64 acc) {
+
+  return XXH_mult32to64((xxh_u32)lhs, (xxh_u32)rhs) + acc;
+
+}
+
+      #endif
 
 /*!
  * @internal
  * @brief Scalar round for @ref XXH3_accumulate_512_scalar().
  *
- * This is extracted to its own function because the NEON path uses a combination
- * of NEON and scalar.
+ * This is extracted to its own function because the NEON path uses a
+ * combination of NEON and scalar.
  */
-XXH_FORCE_INLINE void
-XXH3_scalarRound(void* XXH_RESTRICT acc,
-                 void const* XXH_RESTRICT input,
-                 void const* XXH_RESTRICT secret,
-                 size_t lane)
-{
-    xxh_u64* xacc = (xxh_u64*) acc;
-    xxh_u8 const* xinput  = (xxh_u8 const*) input;
-    xxh_u8 const* xsecret = (xxh_u8 const*) secret;
-    XXH_ASSERT(lane < XXH_ACC_NB);
-    XXH_ASSERT(((size_t)acc & (XXH_ACC_ALIGN-1)) == 0);
-    {
-        xxh_u64 const data_val = XXH_readLE64(xinput + lane * 8);
-        xxh_u64 const data_key = data_val ^ XXH_readLE64(xsecret + lane * 8);
-        xacc[lane ^ 1] += data_val; /* swap adjacent lanes */
-        xacc[lane] = XXH_mult32to64_add64(data_key /* & 0xFFFFFFFF */, data_key >> 32, xacc[lane]);
-    }
+XXH_FORCE_INLINE void XXH3_scalarRound(void *XXH_RESTRICT       acc,
+                                       void const *XXH_RESTRICT input,
+                                       void const *XXH_RESTRICT secret,
+                                       size_t                   lane) {
+
+  xxh_u64      *xacc = (xxh_u64 *)acc;
+  xxh_u8 const *xinput = (xxh_u8 const *)input;
+  xxh_u8 const *xsecret = (xxh_u8 const *)secret;
+  XXH_ASSERT(lane < XXH_ACC_NB);
+  XXH_ASSERT(((size_t)acc & (XXH_ACC_ALIGN - 1)) == 0);
+  {
+
+    xxh_u64 const data_val = XXH_readLE64(xinput + lane * 8);
+    xxh_u64 const data_key = data_val ^ XXH_readLE64(xsecret + lane * 8);
+    xacc[lane ^ 1] += data_val;                      /* swap adjacent lanes */
+    xacc[lane] = XXH_mult32to64_add64(data_key /* & 0xFFFFFFFF */,
+                                      data_key >> 32, xacc[lane]);
+
+  }
+
 }
 
 /*!
  * @internal
  * @brief Processes a 64 byte block of data using the scalar path.
  */
-XXH_FORCE_INLINE void
-XXH3_accumulate_512_scalar(void* XXH_RESTRICT acc,
-                     const void* XXH_RESTRICT input,
-                     const void* XXH_RESTRICT secret)
-{
-    size_t i;
-    /* ARM GCC refuses to unroll this loop, resulting in a 24% slowdown on ARMv6. */
-#if defined(__GNUC__) && !defined(__clang__) \
-  && (defined(__arm__) || defined(__thumb2__)) \
-  && defined(__ARM_FEATURE_UNALIGNED) /* no unaligned access just wastes bytes */ \
-  && XXH_SIZE_OPT <= 0
-#  pragma GCC unroll 8
-#endif
-    for (i=0; i < XXH_ACC_NB; i++) {
-        XXH3_scalarRound(acc, input, secret, i);
-    }
+XXH_FORCE_INLINE void XXH3_accumulate_512_scalar(
+    void *XXH_RESTRICT acc, const void *XXH_RESTRICT input,
+    const void *XXH_RESTRICT secret) {
+
+  size_t i;
+      /* ARM GCC refuses to unroll this loop, resulting in a 24% slowdown on
+       * ARMv6. */
+      #if defined(__GNUC__) && !defined(__clang__) &&                         \
+          (defined(__arm__) || defined(__thumb2__)) &&                        \
+          defined(__ARM_FEATURE_UNALIGNED) /* no unaligned access just wastes \
+                                              bytes */                        \
+          && XXH_SIZE_OPT <= 0
+        #pragma GCC unroll 8
+      #endif
+  for (i = 0; i < XXH_ACC_NB; i++) {
+
+    XXH3_scalarRound(acc, input, secret, i);
+
+  }
+
 }
+
 XXH_FORCE_INLINE XXH3_ACCUMULATE_TEMPLATE(scalar)
 
-/*!
- * @internal
- * @brief Scalar scramble step for @ref XXH3_scrambleAcc_scalar().
- *
- * This is extracted to its own function because the NEON path uses a combination
- * of NEON and scalar.
- */
-XXH_FORCE_INLINE void
-XXH3_scalarScrambleRound(void* XXH_RESTRICT acc,
-                         void const* XXH_RESTRICT secret,
-                         size_t lane)
-{
-    xxh_u64* const xacc = (xxh_u64*) acc;   /* presumed aligned */
-    const xxh_u8* const xsecret = (const xxh_u8*) secret;   /* no alignment restriction */
-    XXH_ASSERT((((size_t)acc) & (XXH_ACC_ALIGN-1)) == 0);
-    XXH_ASSERT(lane < XXH_ACC_NB);
-    {
-        xxh_u64 const key64 = XXH_readLE64(xsecret + lane * 8);
-        xxh_u64 acc64 = xacc[lane];
-        acc64 = XXH_xorshift64(acc64, 47);
-        acc64 ^= key64;
-        acc64 *= XXH_PRIME32_1;
-        xacc[lane] = acc64;
-    }
+    /*!
+     * @internal
+     * @brief Scalar scramble step for @ref XXH3_scrambleAcc_scalar().
+     *
+     * This is extracted to its own function because the NEON path uses a
+     * combination of NEON and scalar.
+     */
+    XXH_FORCE_INLINE
+    void XXH3_scalarScrambleRound(void *XXH_RESTRICT       acc,
+                                  void const *XXH_RESTRICT secret,
+                                  size_t                   lane) {
+
+  xxh_u64 *const      xacc = (xxh_u64 *)acc;            /* presumed aligned */
+  const xxh_u8 *const xsecret =
+      (const xxh_u8 *)secret;                   /* no alignment restriction */
+  XXH_ASSERT((((size_t)acc) & (XXH_ACC_ALIGN - 1)) == 0);
+  XXH_ASSERT(lane < XXH_ACC_NB);
+  {
+
+    xxh_u64 const key64 = XXH_readLE64(xsecret + lane * 8);
+    xxh_u64       acc64 = xacc[lane];
+    acc64 = XXH_xorshift64(acc64, 47);
+    acc64 ^= key64;
+    acc64 *= XXH_PRIME32_1;
+    xacc[lane] = acc64;
+
+  }
+
 }
 
 /*!
  * @internal
  * @brief Scrambles the accumulators after a large chunk has been read
  */
-XXH_FORCE_INLINE void
-XXH3_scrambleAcc_scalar(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
-{
-    size_t i;
-    for (i=0; i < XXH_ACC_NB; i++) {
-        XXH3_scalarScrambleRound(acc, secret, i);
+XXH_FORCE_INLINE void XXH3_scrambleAcc_scalar(void *XXH_RESTRICT       acc,
+                                              const void *XXH_RESTRICT secret) {
+
+  size_t i;
+  for (i = 0; i < XXH_ACC_NB; i++) {
+
+    XXH3_scalarScrambleRound(acc, secret, i);
+
+  }
+
+}
+
+XXH_FORCE_INLINE void XXH3_initCustomSecret_scalar(
+    void *XXH_RESTRICT customSecret, xxh_u64 seed64) {
+
+  /*
+   * We need a separate pointer for the hack below,
+   * which requires a non-const pointer.
+   * Any decent compiler will optimize this out otherwise.
+   */
+  const xxh_u8 *kSecretPtr = XXH3_kSecret;
+  XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 15) == 0);
+
+      #if defined(__GNUC__) && defined(__aarch64__)
+  /*
+   * UGLY HACK:
+   * GCC and Clang generate a bunch of MOV/MOVK pairs for aarch64, and they are
+   * placed sequentially, in order, at the top of the unrolled loop.
+   *
+   * While MOVK is great for generating constants (2 cycles for a 64-bit
+   * constant compared to 4 cycles for LDR), it fights for bandwidth with
+   * the arithmetic instructions.
+   *
+   *   I   L   S
+   * MOVK
+   * MOVK
+   * MOVK
+   * MOVK
+   * ADD
+   * SUB      STR
+   *          STR
+   * By forcing loads from memory (as the asm line causes the compiler to assume
+   * that XXH3_kSecretPtr has been changed), the pipelines are used more
+   * efficiently:
+   *   I   L   S
+   *      LDR
+   *  ADD LDR
+   *  SUB     STR
+   *          STR
+   *
+   * See XXH3_NEON_LANES for details on the pipsline.
+   *
+   * XXH3_64bits_withSeed, len == 256, Snapdragon 835
+   *   without hack: 2654.4 MB/s
+   *   with hack:    3202.9 MB/s
+   */
+  XXH_COMPILER_GUARD(kSecretPtr);
+      #endif
+  {
+
+    int const nbRounds = XXH_SECRET_DEFAULT_SIZE / 16;
+    int       i;
+    for (i = 0; i < nbRounds; i++) {
+
+      /*
+       * The asm hack causes the compiler to assume that kSecretPtr aliases with
+       * customSecret, and on aarch64, this prevented LDP from merging two
+       * loads together for free. Putting the loads together before the stores
+       * properly generates LDP.
+       */
+      xxh_u64 lo = XXH_readLE64(kSecretPtr + 16 * i) + seed64;
+      xxh_u64 hi = XXH_readLE64(kSecretPtr + 16 * i + 8) - seed64;
+      XXH_writeLE64((xxh_u8 *)customSecret + 16 * i, lo);
+      XXH_writeLE64((xxh_u8 *)customSecret + 16 * i + 8, hi);
+
     }
-}
 
-XXH_FORCE_INLINE void
-XXH3_initCustomSecret_scalar(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
-{
-    /*
-     * We need a separate pointer for the hack below,
-     * which requires a non-const pointer.
-     * Any decent compiler will optimize this out otherwise.
-     */
-    const xxh_u8* kSecretPtr = XXH3_kSecret;
-    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 15) == 0);
+  }
 
-#if defined(__GNUC__) && defined(__aarch64__)
-    /*
-     * UGLY HACK:
-     * GCC and Clang generate a bunch of MOV/MOVK pairs for aarch64, and they are
-     * placed sequentially, in order, at the top of the unrolled loop.
-     *
-     * While MOVK is great for generating constants (2 cycles for a 64-bit
-     * constant compared to 4 cycles for LDR), it fights for bandwidth with
-     * the arithmetic instructions.
-     *
-     *   I   L   S
-     * MOVK
-     * MOVK
-     * MOVK
-     * MOVK
-     * ADD
-     * SUB      STR
-     *          STR
-     * By forcing loads from memory (as the asm line causes the compiler to assume
-     * that XXH3_kSecretPtr has been changed), the pipelines are used more
-     * efficiently:
-     *   I   L   S
-     *      LDR
-     *  ADD LDR
-     *  SUB     STR
-     *          STR
-     *
-     * See XXH3_NEON_LANES for details on the pipsline.
-     *
-     * XXH3_64bits_withSeed, len == 256, Snapdragon 835
-     *   without hack: 2654.4 MB/s
-     *   with hack:    3202.9 MB/s
-     */
-    XXH_COMPILER_GUARD(kSecretPtr);
-#endif
-    {   int const nbRounds = XXH_SECRET_DEFAULT_SIZE / 16;
-        int i;
-        for (i=0; i < nbRounds; i++) {
-            /*
-             * The asm hack causes the compiler to assume that kSecretPtr aliases with
-             * customSecret, and on aarch64, this prevented LDP from merging two
-             * loads together for free. Putting the loads together before the stores
-             * properly generates LDP.
-             */
-            xxh_u64 lo = XXH_readLE64(kSecretPtr + 16*i)     + seed64;
-            xxh_u64 hi = XXH_readLE64(kSecretPtr + 16*i + 8) - seed64;
-            XXH_writeLE64((xxh_u8*)customSecret + 16*i,     lo);
-            XXH_writeLE64((xxh_u8*)customSecret + 16*i + 8, hi);
-    }   }
 }
 
+typedef void (*XXH3_f_accumulate)(xxh_u64      *XXH_RESTRICT,
+                                  const xxh_u8 *XXH_RESTRICT,
+                                  const xxh_u8 *XXH_RESTRICT, size_t);
+typedef void (*XXH3_f_scrambleAcc)(void *XXH_RESTRICT, const void *);
+typedef void (*XXH3_f_initCustomSecret)(void *XXH_RESTRICT, xxh_u64);
 
-typedef void (*XXH3_f_accumulate)(xxh_u64* XXH_RESTRICT, const xxh_u8* XXH_RESTRICT, const xxh_u8* XXH_RESTRICT, size_t);
-typedef void (*XXH3_f_scrambleAcc)(void* XXH_RESTRICT, const void*);
-typedef void (*XXH3_f_initCustomSecret)(void* XXH_RESTRICT, xxh_u64);
+      #if (XXH_VECTOR == XXH_AVX512)
 
+        #define XXH3_accumulate_512 XXH3_accumulate_512_avx512
+        #define XXH3_accumulate XXH3_accumulate_avx512
+        #define XXH3_scrambleAcc XXH3_scrambleAcc_avx512
+        #define XXH3_initCustomSecret XXH3_initCustomSecret_avx512
 
-#if (XXH_VECTOR == XXH_AVX512)
+      #elif (XXH_VECTOR == XXH_AVX2)
 
-#define XXH3_accumulate_512 XXH3_accumulate_512_avx512
-#define XXH3_accumulate     XXH3_accumulate_avx512
-#define XXH3_scrambleAcc    XXH3_scrambleAcc_avx512
-#define XXH3_initCustomSecret XXH3_initCustomSecret_avx512
+        #define XXH3_accumulate_512 XXH3_accumulate_512_avx2
+        #define XXH3_accumulate XXH3_accumulate_avx2
+        #define XXH3_scrambleAcc XXH3_scrambleAcc_avx2
+        #define XXH3_initCustomSecret XXH3_initCustomSecret_avx2
 
-#elif (XXH_VECTOR == XXH_AVX2)
+      #elif (XXH_VECTOR == XXH_SSE2)
 
-#define XXH3_accumulate_512 XXH3_accumulate_512_avx2
-#define XXH3_accumulate     XXH3_accumulate_avx2
-#define XXH3_scrambleAcc    XXH3_scrambleAcc_avx2
-#define XXH3_initCustomSecret XXH3_initCustomSecret_avx2
+        #define XXH3_accumulate_512 XXH3_accumulate_512_sse2
+        #define XXH3_accumulate XXH3_accumulate_sse2
+        #define XXH3_scrambleAcc XXH3_scrambleAcc_sse2
+        #define XXH3_initCustomSecret XXH3_initCustomSecret_sse2
 
-#elif (XXH_VECTOR == XXH_SSE2)
+      #elif (XXH_VECTOR == XXH_NEON)
 
-#define XXH3_accumulate_512 XXH3_accumulate_512_sse2
-#define XXH3_accumulate     XXH3_accumulate_sse2
-#define XXH3_scrambleAcc    XXH3_scrambleAcc_sse2
-#define XXH3_initCustomSecret XXH3_initCustomSecret_sse2
+        #define XXH3_accumulate_512 XXH3_accumulate_512_neon
+        #define XXH3_accumulate XXH3_accumulate_neon
+        #define XXH3_scrambleAcc XXH3_scrambleAcc_neon
+        #define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
 
-#elif (XXH_VECTOR == XXH_NEON)
+      #elif (XXH_VECTOR == XXH_VSX)
 
-#define XXH3_accumulate_512 XXH3_accumulate_512_neon
-#define XXH3_accumulate     XXH3_accumulate_neon
-#define XXH3_scrambleAcc    XXH3_scrambleAcc_neon
-#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
+        #define XXH3_accumulate_512 XXH3_accumulate_512_vsx
+        #define XXH3_accumulate XXH3_accumulate_vsx
+        #define XXH3_scrambleAcc XXH3_scrambleAcc_vsx
+        #define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
 
-#elif (XXH_VECTOR == XXH_VSX)
+      #elif (XXH_VECTOR == XXH_SVE)
+        #define XXH3_accumulate_512 XXH3_accumulate_512_sve
+        #define XXH3_accumulate XXH3_accumulate_sve
+        #define XXH3_scrambleAcc XXH3_scrambleAcc_scalar
+        #define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
 
-#define XXH3_accumulate_512 XXH3_accumulate_512_vsx
-#define XXH3_accumulate     XXH3_accumulate_vsx
-#define XXH3_scrambleAcc    XXH3_scrambleAcc_vsx
-#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
+      #else                                                       /* scalar */
 
-#elif (XXH_VECTOR == XXH_SVE)
-#define XXH3_accumulate_512 XXH3_accumulate_512_sve
-#define XXH3_accumulate     XXH3_accumulate_sve
-#define XXH3_scrambleAcc    XXH3_scrambleAcc_scalar
-#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
+        #define XXH3_accumulate_512 XXH3_accumulate_512_scalar
+        #define XXH3_accumulate XXH3_accumulate_scalar
+        #define XXH3_scrambleAcc XXH3_scrambleAcc_scalar
+        #define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
 
-#else /* scalar */
+      #endif
 
-#define XXH3_accumulate_512 XXH3_accumulate_512_scalar
-#define XXH3_accumulate     XXH3_accumulate_scalar
-#define XXH3_scrambleAcc    XXH3_scrambleAcc_scalar
-#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
+      #if XXH_SIZE_OPT >= 1             /* don't do SIMD for initialization */
+        #undef XXH3_initCustomSecret
+        #define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
+      #endif
 
-#endif
+XXH_FORCE_INLINE void XXH3_hashLong_internal_loop(
+    xxh_u64 *XXH_RESTRICT acc, const xxh_u8 *XXH_RESTRICT input, size_t len,
+    const xxh_u8 *XXH_RESTRICT secret, size_t secretSize,
+    XXH3_f_accumulate f_acc, XXH3_f_scrambleAcc f_scramble) {
 
-#if XXH_SIZE_OPT >= 1 /* don't do SIMD for initialization */
-#  undef XXH3_initCustomSecret
-#  define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
-#endif
+  size_t const nbStripesPerBlock =
+      (secretSize - XXH_STRIPE_LEN) / XXH_SECRET_CONSUME_RATE;
+  size_t const block_len = XXH_STRIPE_LEN * nbStripesPerBlock;
+  size_t const nb_blocks = (len - 1) / block_len;
 
-XXH_FORCE_INLINE void
-XXH3_hashLong_internal_loop(xxh_u64* XXH_RESTRICT acc,
-                      const xxh_u8* XXH_RESTRICT input, size_t len,
-                      const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
-                            XXH3_f_accumulate f_acc,
-                            XXH3_f_scrambleAcc f_scramble)
-{
-    size_t const nbStripesPerBlock = (secretSize - XXH_STRIPE_LEN) / XXH_SECRET_CONSUME_RATE;
-    size_t const block_len = XXH_STRIPE_LEN * nbStripesPerBlock;
-    size_t const nb_blocks = (len - 1) / block_len;
+  size_t n;
+
+  XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
+
+  for (n = 0; n < nb_blocks; n++) {
+
+    f_acc(acc, input + n * block_len, secret, nbStripesPerBlock);
+    f_scramble(acc, secret + secretSize - XXH_STRIPE_LEN);
+
+  }
 
-    size_t n;
+  /* last partial block */
+  XXH_ASSERT(len > XXH_STRIPE_LEN);
+  {
 
-    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
+    size_t const nbStripes =
+        ((len - 1) - (block_len * nb_blocks)) / XXH_STRIPE_LEN;
+    XXH_ASSERT(nbStripes <= (secretSize / XXH_SECRET_CONSUME_RATE));
+    f_acc(acc, input + nb_blocks * block_len, secret, nbStripes);
+
+    /* last stripe */
+    {
+
+      const xxh_u8 *const p = input + len - XXH_STRIPE_LEN;
+      #define XXH_SECRET_LASTACC_START                                       \
+        7 /* not aligned on 8, last secret is different from acc & scrambler \
+           */
+      XXH3_accumulate_512(
+          acc, p,
+          secret + secretSize - XXH_STRIPE_LEN - XXH_SECRET_LASTACC_START);
 
-    for (n = 0; n < nb_blocks; n++) {
-        f_acc(acc, input + n*block_len, secret, nbStripesPerBlock);
-        f_scramble(acc, secret + secretSize - XXH_STRIPE_LEN);
     }
 
-    /* last partial block */
-    XXH_ASSERT(len > XXH_STRIPE_LEN);
-    {   size_t const nbStripes = ((len - 1) - (block_len * nb_blocks)) / XXH_STRIPE_LEN;
-        XXH_ASSERT(nbStripes <= (secretSize / XXH_SECRET_CONSUME_RATE));
-        f_acc(acc, input + nb_blocks*block_len, secret, nbStripes);
+  }
 
-        /* last stripe */
-        {   const xxh_u8* const p = input + len - XXH_STRIPE_LEN;
-#define XXH_SECRET_LASTACC_START 7  /* not aligned on 8, last secret is different from acc & scrambler */
-            XXH3_accumulate_512(acc, p, secret + secretSize - XXH_STRIPE_LEN - XXH_SECRET_LASTACC_START);
-    }   }
 }
 
-XXH_FORCE_INLINE xxh_u64
-XXH3_mix2Accs(const xxh_u64* XXH_RESTRICT acc, const xxh_u8* XXH_RESTRICT secret)
-{
-    return XXH3_mul128_fold64(
-               acc[0] ^ XXH_readLE64(secret),
-               acc[1] ^ XXH_readLE64(secret+8) );
+XXH_FORCE_INLINE xxh_u64 XXH3_mix2Accs(const xxh_u64 *XXH_RESTRICT acc,
+                                       const xxh_u8 *XXH_RESTRICT  secret) {
+
+  return XXH3_mul128_fold64(acc[0] ^ XXH_readLE64(secret),
+                            acc[1] ^ XXH_readLE64(secret + 8));
+
 }
 
-static XXH64_hash_t
-XXH3_mergeAccs(const xxh_u64* XXH_RESTRICT acc, const xxh_u8* XXH_RESTRICT secret, xxh_u64 start)
-{
-    xxh_u64 result64 = start;
-    size_t i = 0;
-
-    for (i = 0; i < 4; i++) {
-        result64 += XXH3_mix2Accs(acc+2*i, secret + 16*i);
-#if defined(__clang__)                                /* Clang */ \
-    && (defined(__arm__) || defined(__thumb__))       /* ARMv7 */ \
-    && (defined(__ARM_NEON) || defined(__ARM_NEON__)) /* NEON */  \
-    && !defined(XXH_ENABLE_AUTOVECTORIZE)             /* Define to disable */
-        /*
-         * UGLY HACK:
-         * Prevent autovectorization on Clang ARMv7-a. Exact same problem as
-         * the one in XXH3_len_129to240_64b. Speeds up shorter keys > 240b.
-         * XXH3_64bits, len == 256, Snapdragon 835:
-         *   without hack: 2063.7 MB/s
-         *   with hack:    2560.7 MB/s
-         */
-        XXH_COMPILER_GUARD(result64);
-#endif
-    }
+static XXH64_hash_t XXH3_mergeAccs(const xxh_u64 *XXH_RESTRICT acc,
+                                   const xxh_u8 *XXH_RESTRICT  secret,
+                                   xxh_u64                     start) {
+
+  xxh_u64 result64 = start;
+  size_t  i = 0;
+
+  for (i = 0; i < 4; i++) {
+
+    result64 += XXH3_mix2Accs(acc + 2 * i, secret + 16 * i);
+      #if defined(__clang__)                                /* Clang */ \
+          && (defined(__arm__) || defined(__thumb__))       /* ARMv7 */ \
+          && (defined(__ARM_NEON) || defined(__ARM_NEON__)) /* NEON */  \
+          && !defined(XXH_ENABLE_AUTOVECTORIZE)        /* Define to disable */
+    /*
+     * UGLY HACK:
+     * Prevent autovectorization on Clang ARMv7-a. Exact same problem as
+     * the one in XXH3_len_129to240_64b. Speeds up shorter keys > 240b.
+     * XXH3_64bits, len == 256, Snapdragon 835:
+     *   without hack: 2063.7 MB/s
+     *   with hack:    2560.7 MB/s
+     */
+    XXH_COMPILER_GUARD(result64);
+      #endif
+
+  }
+
+  return XXH3_avalanche(result64);
 
-    return XXH3_avalanche(result64);
 }
 
-#define XXH3_INIT_ACC { XXH_PRIME32_3, XXH_PRIME64_1, XXH_PRIME64_2, XXH_PRIME64_3, \
-                        XXH_PRIME64_4, XXH_PRIME32_2, XXH_PRIME64_5, XXH_PRIME32_1 }
+      #define XXH3_INIT_ACC                                              \
+        {                                                                \
+                                                                         \
+          XXH_PRIME32_3, XXH_PRIME64_1, XXH_PRIME64_2, XXH_PRIME64_3,    \
+              XXH_PRIME64_4, XXH_PRIME32_2, XXH_PRIME64_5, XXH_PRIME32_1 \
+                                                                         \
+        }
 
-XXH_FORCE_INLINE XXH64_hash_t
-XXH3_hashLong_64b_internal(const void* XXH_RESTRICT input, size_t len,
-                           const void* XXH_RESTRICT secret, size_t secretSize,
-                           XXH3_f_accumulate f_acc,
-                           XXH3_f_scrambleAcc f_scramble)
-{
-    XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[XXH_ACC_NB] = XXH3_INIT_ACC;
+XXH_FORCE_INLINE XXH64_hash_t XXH3_hashLong_64b_internal(
+    const void *XXH_RESTRICT input, size_t len, const void *XXH_RESTRICT secret,
+    size_t secretSize, XXH3_f_accumulate f_acc, XXH3_f_scrambleAcc f_scramble) {
+
+  XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[XXH_ACC_NB] = XXH3_INIT_ACC;
 
-    XXH3_hashLong_internal_loop(acc, (const xxh_u8*)input, len, (const xxh_u8*)secret, secretSize, f_acc, f_scramble);
+  XXH3_hashLong_internal_loop(acc, (const xxh_u8 *)input, len,
+                              (const xxh_u8 *)secret, secretSize, f_acc,
+                              f_scramble);
+
+  /* converge into final hash */
+  XXH_STATIC_ASSERT(sizeof(acc) == 64);
+      /* do not align on 8, so that the secret is different from the accumulator
+       */
+      #define XXH_SECRET_MERGEACCS_START 11
+  XXH_ASSERT(secretSize >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
+  return XXH3_mergeAccs(acc,
+                        (const xxh_u8 *)secret + XXH_SECRET_MERGEACCS_START,
+                        (xxh_u64)len * XXH_PRIME64_1);
 
-    /* converge into final hash */
-    XXH_STATIC_ASSERT(sizeof(acc) == 64);
-    /* do not align on 8, so that the secret is different from the accumulator */
-#define XXH_SECRET_MERGEACCS_START 11
-    XXH_ASSERT(secretSize >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
-    return XXH3_mergeAccs(acc, (const xxh_u8*)secret + XXH_SECRET_MERGEACCS_START, (xxh_u64)len * XXH_PRIME64_1);
 }
 
 /*
  * It's important for performance to transmit secret's size (when it's static)
  * so that the compiler can properly optimize the vectorized loop.
- * This makes a big performance difference for "medium" keys (<1 KB) when using AVX instruction set.
- * When the secret size is unknown, or on GCC 12 where the mix of NO_INLINE and FORCE_INLINE
- * breaks -Og, this is XXH_NO_INLINE.
+ * This makes a big performance difference for "medium" keys (<1 KB) when using
+ * AVX instruction set. When the secret size is unknown, or on GCC 12 where the
+ * mix of NO_INLINE and FORCE_INLINE breaks -Og, this is XXH_NO_INLINE.
  */
-XXH3_WITH_SECRET_INLINE XXH64_hash_t
-XXH3_hashLong_64b_withSecret(const void* XXH_RESTRICT input, size_t len,
-                             XXH64_hash_t seed64, const xxh_u8* XXH_RESTRICT secret, size_t secretLen)
-{
-    (void)seed64;
-    return XXH3_hashLong_64b_internal(input, len, secret, secretLen, XXH3_accumulate, XXH3_scrambleAcc);
+XXH3_WITH_SECRET_INLINE XXH64_hash_t XXH3_hashLong_64b_withSecret(
+    const void *XXH_RESTRICT input, size_t len, XXH64_hash_t seed64,
+    const xxh_u8 *XXH_RESTRICT secret, size_t secretLen) {
+
+  (void)seed64;
+  return XXH3_hashLong_64b_internal(input, len, secret, secretLen,
+                                    XXH3_accumulate, XXH3_scrambleAcc);
+
 }
 
 /*
  * It's preferable for performance that XXH3_hashLong is not inlined,
- * as it results in a smaller function for small data, easier to the instruction cache.
- * Note that inside this no_inline function, we do inline the internal loop,
- * and provide a statically defined secret size to allow optimization of vector loop.
+ * as it results in a smaller function for small data, easier to the instruction
+ * cache. Note that inside this no_inline function, we do inline the internal
+ * loop, and provide a statically defined secret size to allow optimization of
+ * vector loop.
  */
-XXH_NO_INLINE XXH_PUREF XXH64_hash_t
-XXH3_hashLong_64b_default(const void* XXH_RESTRICT input, size_t len,
-                          XXH64_hash_t seed64, const xxh_u8* XXH_RESTRICT secret, size_t secretLen)
-{
-    (void)seed64; (void)secret; (void)secretLen;
-    return XXH3_hashLong_64b_internal(input, len, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_accumulate, XXH3_scrambleAcc);
+XXH_NO_INLINE XXH_PUREF XXH64_hash_t XXH3_hashLong_64b_default(
+    const void *XXH_RESTRICT input, size_t len, XXH64_hash_t seed64,
+    const xxh_u8 *XXH_RESTRICT secret, size_t secretLen) {
+
+  (void)seed64;
+  (void)secret;
+  (void)secretLen;
+  return XXH3_hashLong_64b_internal(input, len, XXH3_kSecret,
+                                    sizeof(XXH3_kSecret), XXH3_accumulate,
+                                    XXH3_scrambleAcc);
+
 }
 
 /*
  * XXH3_hashLong_64b_withSeed():
- * Generate a custom key based on alteration of default XXH3_kSecret with the seed,
- * and then use this key for long mode hashing.
+ * Generate a custom key based on alteration of default XXH3_kSecret with the
+ * seed, and then use this key for long mode hashing.
  *
  * This operation is decently fast but nonetheless costs a little bit of time.
  * Try to avoid it whenever possible (typically when seed==0).
@@ -5921,98 +6688,116 @@ XXH3_hashLong_64b_default(const void* XXH_RESTRICT input, size_t len,
  * It's important for performance that XXH3_hashLong is not inlined. Not sure
  * why (uop cache maybe?), but the difference is large and easily measurable.
  */
-XXH_FORCE_INLINE XXH64_hash_t
-XXH3_hashLong_64b_withSeed_internal(const void* input, size_t len,
-                                    XXH64_hash_t seed,
-                                    XXH3_f_accumulate f_acc,
-                                    XXH3_f_scrambleAcc f_scramble,
-                                    XXH3_f_initCustomSecret f_initSec)
-{
-#if XXH_SIZE_OPT <= 0
-    if (seed == 0)
-        return XXH3_hashLong_64b_internal(input, len,
-                                          XXH3_kSecret, sizeof(XXH3_kSecret),
-                                          f_acc, f_scramble);
-#endif
-    {   XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
-        f_initSec(secret, seed);
-        return XXH3_hashLong_64b_internal(input, len, secret, sizeof(secret),
-                                          f_acc, f_scramble);
-    }
+XXH_FORCE_INLINE XXH64_hash_t XXH3_hashLong_64b_withSeed_internal(
+    const void *input, size_t len, XXH64_hash_t seed, XXH3_f_accumulate f_acc,
+    XXH3_f_scrambleAcc f_scramble, XXH3_f_initCustomSecret f_initSec) {
+
+      #if XXH_SIZE_OPT <= 0
+  if (seed == 0)
+    return XXH3_hashLong_64b_internal(input, len, XXH3_kSecret,
+                                      sizeof(XXH3_kSecret), f_acc, f_scramble);
+      #endif
+  {
+
+    XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
+    f_initSec(secret, seed);
+    return XXH3_hashLong_64b_internal(input, len, secret, sizeof(secret), f_acc,
+                                      f_scramble);
+
+  }
+
 }
 
 /*
  * It's important for performance that XXH3_hashLong is not inlined.
  */
-XXH_NO_INLINE XXH64_hash_t
-XXH3_hashLong_64b_withSeed(const void* XXH_RESTRICT input, size_t len,
-                           XXH64_hash_t seed, const xxh_u8* XXH_RESTRICT secret, size_t secretLen)
-{
-    (void)secret; (void)secretLen;
-    return XXH3_hashLong_64b_withSeed_internal(input, len, seed,
-                XXH3_accumulate, XXH3_scrambleAcc, XXH3_initCustomSecret);
-}
+XXH_NO_INLINE XXH64_hash_t XXH3_hashLong_64b_withSeed(
+    const void *XXH_RESTRICT input, size_t len, XXH64_hash_t seed,
+    const xxh_u8 *XXH_RESTRICT secret, size_t secretLen) {
 
+  (void)secret;
+  (void)secretLen;
+  return XXH3_hashLong_64b_withSeed_internal(input, len, seed, XXH3_accumulate,
+                                             XXH3_scrambleAcc,
+                                             XXH3_initCustomSecret);
+
+}
 
-typedef XXH64_hash_t (*XXH3_hashLong64_f)(const void* XXH_RESTRICT, size_t,
-                                          XXH64_hash_t, const xxh_u8* XXH_RESTRICT, size_t);
+typedef XXH64_hash_t (*XXH3_hashLong64_f)(const void *XXH_RESTRICT, size_t,
+                                          XXH64_hash_t,
+                                          const xxh_u8 *XXH_RESTRICT, size_t);
 
 XXH_FORCE_INLINE XXH64_hash_t
-XXH3_64bits_internal(const void* XXH_RESTRICT input, size_t len,
-                     XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen,
-                     XXH3_hashLong64_f f_hashLong)
-{
-    XXH_ASSERT(secretLen >= XXH3_SECRET_SIZE_MIN);
-    /*
-     * If an action is to be taken if `secretLen` condition is not respected,
-     * it should be done here.
-     * For now, it's a contract pre-condition.
-     * Adding a check and a branch here would cost performance at every hash.
-     * Also, note that function signature doesn't offer room to return an error.
-     */
-    if (len <= 16)
-        return XXH3_len_0to16_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, seed64);
-    if (len <= 128)
-        return XXH3_len_17to128_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
-    if (len <= XXH3_MIDSIZE_MAX)
-        return XXH3_len_129to240_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
-    return f_hashLong(input, len, seed64, (const xxh_u8*)secret, secretLen);
-}
+XXH3_64bits_internal(const void *XXH_RESTRICT input, size_t len,
+                     XXH64_hash_t seed64, const void *XXH_RESTRICT secret,
+                     size_t secretLen, XXH3_hashLong64_f f_hashLong) {
+
+  XXH_ASSERT(secretLen >= XXH3_SECRET_SIZE_MIN);
+  /*
+   * If an action is to be taken if `secretLen` condition is not respected,
+   * it should be done here.
+   * For now, it's a contract pre-condition.
+   * Adding a check and a branch here would cost performance at every hash.
+   * Also, note that function signature doesn't offer room to return an error.
+   */
+  if (len <= 16)
+    return XXH3_len_0to16_64b((const xxh_u8 *)input, len,
+                              (const xxh_u8 *)secret, seed64);
+  if (len <= 128)
+    return XXH3_len_17to128_64b((const xxh_u8 *)input, len,
+                                (const xxh_u8 *)secret, secretLen, seed64);
+  if (len <= XXH3_MIDSIZE_MAX)
+    return XXH3_len_129to240_64b((const xxh_u8 *)input, len,
+                                 (const xxh_u8 *)secret, secretLen, seed64);
+  return f_hashLong(input, len, seed64, (const xxh_u8 *)secret, secretLen);
 
+}
 
 /* ===   Public entry point   === */
 
 /*! @ingroup XXH3_family */
-XXH_PUBLIC_API XXH64_hash_t XXH3_64bits(XXH_NOESCAPE const void* input, size_t length)
-{
-    return XXH3_64bits_internal(input, length, 0, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_default);
+XXH_PUBLIC_API XXH64_hash_t XXH3_64bits(XXH_NOESCAPE const void *input,
+                                        size_t                   length) {
+
+  return XXH3_64bits_internal(input, length, 0, XXH3_kSecret,
+                              sizeof(XXH3_kSecret), XXH3_hashLong_64b_default);
+
 }
 
 /*! @ingroup XXH3_family */
 XXH_PUBLIC_API XXH64_hash_t
-XXH3_64bits_withSecret(XXH_NOESCAPE const void* input, size_t length, XXH_NOESCAPE const void* secret, size_t secretSize)
-{
-    return XXH3_64bits_internal(input, length, 0, secret, secretSize, XXH3_hashLong_64b_withSecret);
+XXH3_64bits_withSecret(XXH_NOESCAPE const void *input, size_t length,
+                       XXH_NOESCAPE const void *secret, size_t secretSize) {
+
+  return XXH3_64bits_internal(input, length, 0, secret, secretSize,
+                              XXH3_hashLong_64b_withSecret);
+
 }
 
 /*! @ingroup XXH3_family */
-XXH_PUBLIC_API XXH64_hash_t
-XXH3_64bits_withSeed(XXH_NOESCAPE const void* input, size_t length, XXH64_hash_t seed)
-{
-    return XXH3_64bits_internal(input, length, seed, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_withSeed);
-}
+XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_withSeed(XXH_NOESCAPE const void *input,
+                                                 size_t       length,
+                                                 XXH64_hash_t seed) {
+
+  return XXH3_64bits_internal(input, length, seed, XXH3_kSecret,
+                              sizeof(XXH3_kSecret), XXH3_hashLong_64b_withSeed);
 
-XXH_PUBLIC_API XXH64_hash_t
-XXH3_64bits_withSecretandSeed(XXH_NOESCAPE const void* input, size_t length, XXH_NOESCAPE const void* secret, size_t secretSize, XXH64_hash_t seed)
-{
-    if (length <= XXH3_MIDSIZE_MAX)
-        return XXH3_64bits_internal(input, length, seed, XXH3_kSecret, sizeof(XXH3_kSecret), NULL);
-    return XXH3_hashLong_64b_withSecret(input, length, seed, (const xxh_u8*)secret, secretSize);
 }
 
+XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_withSecretandSeed(
+    XXH_NOESCAPE const void *input, size_t length,
+    XXH_NOESCAPE const void *secret, size_t secretSize, XXH64_hash_t seed) {
+
+  if (length <= XXH3_MIDSIZE_MAX)
+    return XXH3_64bits_internal(input, length, seed, XXH3_kSecret,
+                                sizeof(XXH3_kSecret), NULL);
+  return XXH3_hashLong_64b_withSecret(input, length, seed,
+                                      (const xxh_u8 *)secret, secretSize);
+
+}
 
-/* ===   XXH3 streaming   === */
-#ifndef XXH_NO_STREAM
+      /* ===   XXH3 streaming   === */
+      #ifndef XXH_NO_STREAM
 /*
  * Malloc's a pointer that is always aligned to align.
  *
@@ -6036,48 +6821,58 @@ XXH3_64bits_withSecretandSeed(XXH_NOESCAPE const void* input, size_t length, XXH
  *
  * Align must be a power of 2 and 8 <= align <= 128.
  */
-static XXH_MALLOCF void* XXH_alignedMalloc(size_t s, size_t align)
-{
-    XXH_ASSERT(align <= 128 && align >= 8); /* range check */
-    XXH_ASSERT((align & (align-1)) == 0);   /* power of 2 */
-    XXH_ASSERT(s != 0 && s < (s + align));  /* empty/overflow */
-    {   /* Overallocate to make room for manual realignment and an offset byte */
-        xxh_u8* base = (xxh_u8*)XXH_malloc(s + align);
-        if (base != NULL) {
-            /*
-             * Get the offset needed to align this pointer.
-             *
-             * Even if the returned pointer is aligned, there will always be
-             * at least one byte to store the offset to the original pointer.
-             */
-            size_t offset = align - ((size_t)base & (align - 1)); /* base % align */
-            /* Add the offset for the now-aligned pointer */
-            xxh_u8* ptr = base + offset;
-
-            XXH_ASSERT((size_t)ptr % align == 0);
-
-            /* Store the offset immediately before the returned pointer. */
-            ptr[-1] = (xxh_u8)offset;
-            return ptr;
-        }
-        return NULL;
+static XXH_MALLOCF void *XXH_alignedMalloc(size_t s, size_t align) {
+
+  XXH_ASSERT(align <= 128 && align >= 8);                    /* range check */
+  XXH_ASSERT((align & (align - 1)) == 0);                     /* power of 2 */
+  XXH_ASSERT(s != 0 && s < (s + align));                  /* empty/overflow */
+  {  /* Overallocate to make room for manual realignment and an offset byte */
+    xxh_u8 *base = (xxh_u8 *)XXH_malloc(s + align);
+    if (base != NULL) {
+
+      /*
+       * Get the offset needed to align this pointer.
+       *
+       * Even if the returned pointer is aligned, there will always be
+       * at least one byte to store the offset to the original pointer.
+       */
+      size_t offset = align - ((size_t)base & (align - 1)); /* base % align */
+      /* Add the offset for the now-aligned pointer */
+      xxh_u8 *ptr = base + offset;
+
+      XXH_ASSERT((size_t)ptr % align == 0);
+
+      /* Store the offset immediately before the returned pointer. */
+      ptr[-1] = (xxh_u8)offset;
+      return ptr;
+
     }
+
+    return NULL;
+
+  }
+
 }
+
 /*
  * Frees an aligned pointer allocated by XXH_alignedMalloc(). Don't pass
  * normal malloc'd pointers, XXH_alignedMalloc has a specific data layout.
  */
-static void XXH_alignedFree(void* p)
-{
-    if (p != NULL) {
-        xxh_u8* ptr = (xxh_u8*)p;
-        /* Get the offset byte we added in XXH_malloc. */
-        xxh_u8 offset = ptr[-1];
-        /* Free the original malloc'd pointer */
-        xxh_u8* base = ptr - offset;
-        XXH_free(base);
-    }
+static void XXH_alignedFree(void *p) {
+
+  if (p != NULL) {
+
+    xxh_u8 *ptr = (xxh_u8 *)p;
+    /* Get the offset byte we added in XXH_malloc. */
+    xxh_u8 offset = ptr[-1];
+    /* Free the original malloc'd pointer */
+    xxh_u8 *base = ptr - offset;
+    XXH_free(base);
+
+  }
+
 }
+
 /*! @ingroup XXH3_family */
 /*!
  * @brief Allocate an @ref XXH3_state_t.
@@ -6089,19 +6884,22 @@ static void XXH_alignedFree(void* p)
  *
  * @see @ref streaming_example "Streaming Example"
  */
-XXH_PUBLIC_API XXH3_state_t* XXH3_createState(void)
-{
-    XXH3_state_t* const state = (XXH3_state_t*)XXH_alignedMalloc(sizeof(XXH3_state_t), 64);
-    if (state==NULL) return NULL;
-    XXH3_INITSTATE(state);
-    return state;
+XXH_PUBLIC_API XXH3_state_t *XXH3_createState(void) {
+
+  XXH3_state_t *const state =
+      (XXH3_state_t *)XXH_alignedMalloc(sizeof(XXH3_state_t), 64);
+  if (state == NULL) return NULL;
+  XXH3_INITSTATE(state);
+  return state;
+
 }
 
 /*! @ingroup XXH3_family */
 /*!
  * @brief Frees an @ref XXH3_state_t.
  *
- * @param statePtr A pointer to an @ref XXH3_state_t allocated with @ref XXH3_createState().
+ * @param statePtr A pointer to an @ref XXH3_state_t allocated with @ref
+ * XXH3_createState().
  *
  * @return @ref XXH_OK.
  *
@@ -6109,98 +6907,108 @@ XXH_PUBLIC_API XXH3_state_t* XXH3_createState(void)
  *
  * @see @ref streaming_example "Streaming Example"
  */
-XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t* statePtr)
-{
-    XXH_alignedFree(statePtr);
-    return XXH_OK;
+XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t *statePtr) {
+
+  XXH_alignedFree(statePtr);
+  return XXH_OK;
+
 }
 
 /*! @ingroup XXH3_family */
-XXH_PUBLIC_API void
-XXH3_copyState(XXH_NOESCAPE XXH3_state_t* dst_state, XXH_NOESCAPE const XXH3_state_t* src_state)
-{
-    XXH_memcpy(dst_state, src_state, sizeof(*dst_state));
+XXH_PUBLIC_API void XXH3_copyState(XXH_NOESCAPE XXH3_state_t       *dst_state,
+                                   XXH_NOESCAPE const XXH3_state_t *src_state) {
+
+  XXH_memcpy(dst_state, src_state, sizeof(*dst_state));
+
 }
 
-static void
-XXH3_reset_internal(XXH3_state_t* statePtr,
-                    XXH64_hash_t seed,
-                    const void* secret, size_t secretSize)
-{
-    size_t const initStart = offsetof(XXH3_state_t, bufferedSize);
-    size_t const initLength = offsetof(XXH3_state_t, nbStripesPerBlock) - initStart;
-    XXH_ASSERT(offsetof(XXH3_state_t, nbStripesPerBlock) > initStart);
-    XXH_ASSERT(statePtr != NULL);
-    /* set members from bufferedSize to nbStripesPerBlock (excluded) to 0 */
-    memset((char*)statePtr + initStart, 0, initLength);
-    statePtr->acc[0] = XXH_PRIME32_3;
-    statePtr->acc[1] = XXH_PRIME64_1;
-    statePtr->acc[2] = XXH_PRIME64_2;
-    statePtr->acc[3] = XXH_PRIME64_3;
-    statePtr->acc[4] = XXH_PRIME64_4;
-    statePtr->acc[5] = XXH_PRIME32_2;
-    statePtr->acc[6] = XXH_PRIME64_5;
-    statePtr->acc[7] = XXH_PRIME32_1;
-    statePtr->seed = seed;
-    statePtr->useSeed = (seed != 0);
-    statePtr->extSecret = (const unsigned char*)secret;
-    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
-    statePtr->secretLimit = secretSize - XXH_STRIPE_LEN;
-    statePtr->nbStripesPerBlock = statePtr->secretLimit / XXH_SECRET_CONSUME_RATE;
+static void XXH3_reset_internal(XXH3_state_t *statePtr, XXH64_hash_t seed,
+                                const void *secret, size_t secretSize) {
+
+  size_t const initStart = offsetof(XXH3_state_t, bufferedSize);
+  size_t const initLength =
+      offsetof(XXH3_state_t, nbStripesPerBlock) - initStart;
+  XXH_ASSERT(offsetof(XXH3_state_t, nbStripesPerBlock) > initStart);
+  XXH_ASSERT(statePtr != NULL);
+  /* set members from bufferedSize to nbStripesPerBlock (excluded) to 0 */
+  memset((char *)statePtr + initStart, 0, initLength);
+  statePtr->acc[0] = XXH_PRIME32_3;
+  statePtr->acc[1] = XXH_PRIME64_1;
+  statePtr->acc[2] = XXH_PRIME64_2;
+  statePtr->acc[3] = XXH_PRIME64_3;
+  statePtr->acc[4] = XXH_PRIME64_4;
+  statePtr->acc[5] = XXH_PRIME32_2;
+  statePtr->acc[6] = XXH_PRIME64_5;
+  statePtr->acc[7] = XXH_PRIME32_1;
+  statePtr->seed = seed;
+  statePtr->useSeed = (seed != 0);
+  statePtr->extSecret = (const unsigned char *)secret;
+  XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
+  statePtr->secretLimit = secretSize - XXH_STRIPE_LEN;
+  statePtr->nbStripesPerBlock = statePtr->secretLimit / XXH_SECRET_CONSUME_RATE;
+
 }
 
 /*! @ingroup XXH3_family */
 XXH_PUBLIC_API XXH_errorcode
-XXH3_64bits_reset(XXH_NOESCAPE XXH3_state_t* statePtr)
-{
-    if (statePtr == NULL) return XXH_ERROR;
-    XXH3_reset_internal(statePtr, 0, XXH3_kSecret, XXH_SECRET_DEFAULT_SIZE);
-    return XXH_OK;
+XXH3_64bits_reset(XXH_NOESCAPE XXH3_state_t *statePtr) {
+
+  if (statePtr == NULL) return XXH_ERROR;
+  XXH3_reset_internal(statePtr, 0, XXH3_kSecret, XXH_SECRET_DEFAULT_SIZE);
+  return XXH_OK;
+
 }
 
 /*! @ingroup XXH3_family */
-XXH_PUBLIC_API XXH_errorcode
-XXH3_64bits_reset_withSecret(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize)
-{
-    if (statePtr == NULL) return XXH_ERROR;
-    XXH3_reset_internal(statePtr, 0, secret, secretSize);
-    if (secret == NULL) return XXH_ERROR;
-    if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
-    return XXH_OK;
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSecret(
+    XXH_NOESCAPE XXH3_state_t *statePtr, XXH_NOESCAPE const void *secret,
+    size_t secretSize) {
+
+  if (statePtr == NULL) return XXH_ERROR;
+  XXH3_reset_internal(statePtr, 0, secret, secretSize);
+  if (secret == NULL) return XXH_ERROR;
+  if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
+  return XXH_OK;
+
 }
 
 /*! @ingroup XXH3_family */
-XXH_PUBLIC_API XXH_errorcode
-XXH3_64bits_reset_withSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH64_hash_t seed)
-{
-    if (statePtr == NULL) return XXH_ERROR;
-    if (seed==0) return XXH3_64bits_reset(statePtr);
-    if ((seed != statePtr->seed) || (statePtr->extSecret != NULL))
-        XXH3_initCustomSecret(statePtr->customSecret, seed);
-    XXH3_reset_internal(statePtr, seed, NULL, XXH_SECRET_DEFAULT_SIZE);
-    return XXH_OK;
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSeed(
+    XXH_NOESCAPE XXH3_state_t *statePtr, XXH64_hash_t seed) {
+
+  if (statePtr == NULL) return XXH_ERROR;
+  if (seed == 0) return XXH3_64bits_reset(statePtr);
+  if ((seed != statePtr->seed) || (statePtr->extSecret != NULL))
+    XXH3_initCustomSecret(statePtr->customSecret, seed);
+  XXH3_reset_internal(statePtr, seed, NULL, XXH_SECRET_DEFAULT_SIZE);
+  return XXH_OK;
+
 }
 
 /*! @ingroup XXH3_family */
-XXH_PUBLIC_API XXH_errorcode
-XXH3_64bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize, XXH64_hash_t seed64)
-{
-    if (statePtr == NULL) return XXH_ERROR;
-    if (secret == NULL) return XXH_ERROR;
-    if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
-    XXH3_reset_internal(statePtr, seed64, secret, secretSize);
-    statePtr->useSeed = 1; /* always, even if seed64==0 */
-    return XXH_OK;
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSecretandSeed(
+    XXH_NOESCAPE XXH3_state_t *statePtr, XXH_NOESCAPE const void *secret,
+    size_t secretSize, XXH64_hash_t seed64) {
+
+  if (statePtr == NULL) return XXH_ERROR;
+  if (secret == NULL) return XXH_ERROR;
+  if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
+  XXH3_reset_internal(statePtr, seed64, secret, secretSize);
+  statePtr->useSeed = 1;                       /* always, even if seed64==0 */
+  return XXH_OK;
+
 }
 
 /*!
  * @internal
  * @brief Processes a large input for XXH3_update() and XXH3_digest_long().
  *
- * Unlike XXH3_hashLong_internal_loop(), this can process data that overlaps a block.
+ * Unlike XXH3_hashLong_internal_loop(), this can process data that overlaps a
+ * block.
  *
  * @param acc                Pointer to the 8 accumulator lanes
- * @param nbStripesSoFarPtr  In/out pointer to the number of leftover stripes in the block*
+ * @param nbStripesSoFarPtr  In/out pointer to the number of leftover stripes in
+ * the block*
  * @param nbStripesPerBlock  Number of stripes in a block
  * @param input              Input pointer
  * @param nbStripes          Number of stripes to process
@@ -6210,200 +7018,233 @@ XXH3_64bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOE
  * @param f_scramble         Pointer to an XXH3_scrambleAcc implementation
  * @return                   Pointer past the end of @p input after processing
  */
-XXH_FORCE_INLINE const xxh_u8 *
-XXH3_consumeStripes(xxh_u64* XXH_RESTRICT acc,
-                    size_t* XXH_RESTRICT nbStripesSoFarPtr, size_t nbStripesPerBlock,
-                    const xxh_u8* XXH_RESTRICT input, size_t nbStripes,
-                    const xxh_u8* XXH_RESTRICT secret, size_t secretLimit,
-                    XXH3_f_accumulate f_acc,
-                    XXH3_f_scrambleAcc f_scramble)
-{
-    const xxh_u8* initialSecret = secret + *nbStripesSoFarPtr * XXH_SECRET_CONSUME_RATE;
-    /* Process full blocks */
-    if (nbStripes >= (nbStripesPerBlock - *nbStripesSoFarPtr)) {
-        /* Process the initial partial block... */
-        size_t nbStripesThisIter = nbStripesPerBlock - *nbStripesSoFarPtr;
-
-        do {
-            /* Accumulate and scramble */
-            f_acc(acc, input, initialSecret, nbStripesThisIter);
-            f_scramble(acc, secret + secretLimit);
-            input += nbStripesThisIter * XXH_STRIPE_LEN;
-            nbStripes -= nbStripesThisIter;
-            /* Then continue the loop with the full block size */
-            nbStripesThisIter = nbStripesPerBlock;
-            initialSecret = secret;
-        } while (nbStripes >= nbStripesPerBlock);
-        *nbStripesSoFarPtr = 0;
-    }
-    /* Process a partial block */
-    if (nbStripes > 0) {
-        f_acc(acc, input, initialSecret, nbStripes);
-        input += nbStripes * XXH_STRIPE_LEN;
-        *nbStripesSoFarPtr += nbStripes;
-    }
-    /* Return end pointer */
-    return input;
+XXH_FORCE_INLINE const xxh_u8 *XXH3_consumeStripes(
+    xxh_u64 *XXH_RESTRICT acc, size_t *XXH_RESTRICT nbStripesSoFarPtr,
+    size_t nbStripesPerBlock, const xxh_u8 *XXH_RESTRICT input,
+    size_t nbStripes, const xxh_u8 *XXH_RESTRICT secret, size_t secretLimit,
+    XXH3_f_accumulate f_acc, XXH3_f_scrambleAcc f_scramble) {
+
+  const xxh_u8 *initialSecret =
+      secret + *nbStripesSoFarPtr * XXH_SECRET_CONSUME_RATE;
+  /* Process full blocks */
+  if (nbStripes >= (nbStripesPerBlock - *nbStripesSoFarPtr)) {
+
+    /* Process the initial partial block... */
+    size_t nbStripesThisIter = nbStripesPerBlock - *nbStripesSoFarPtr;
+
+    do {
+
+      /* Accumulate and scramble */
+      f_acc(acc, input, initialSecret, nbStripesThisIter);
+      f_scramble(acc, secret + secretLimit);
+      input += nbStripesThisIter * XXH_STRIPE_LEN;
+      nbStripes -= nbStripesThisIter;
+      /* Then continue the loop with the full block size */
+      nbStripesThisIter = nbStripesPerBlock;
+      initialSecret = secret;
+
+    } while (nbStripes >= nbStripesPerBlock);
+
+    *nbStripesSoFarPtr = 0;
+
+  }
+
+  /* Process a partial block */
+  if (nbStripes > 0) {
+
+    f_acc(acc, input, initialSecret, nbStripes);
+    input += nbStripes * XXH_STRIPE_LEN;
+    *nbStripesSoFarPtr += nbStripes;
+
+  }
+
+  /* Return end pointer */
+  return input;
+
 }
 
-#ifndef XXH3_STREAM_USE_STACK
-# if XXH_SIZE_OPT <= 0 && !defined(__clang__) /* clang doesn't need additional stack space */
-#   define XXH3_STREAM_USE_STACK 1
-# endif
-#endif
+        #ifndef XXH3_STREAM_USE_STACK
+          #if XXH_SIZE_OPT <= 0 && \
+              !defined(            \
+                  __clang__)   /* clang doesn't need additional stack space */
+            #define XXH3_STREAM_USE_STACK 1
+          #endif
+        #endif
 /*
  * Both XXH3_64bits_update and XXH3_128bits_update use this routine.
  */
-XXH_FORCE_INLINE XXH_errorcode
-XXH3_update(XXH3_state_t* XXH_RESTRICT const state,
-            const xxh_u8* XXH_RESTRICT input, size_t len,
-            XXH3_f_accumulate f_acc,
-            XXH3_f_scrambleAcc f_scramble)
-{
-    if (input==NULL) {
-        XXH_ASSERT(len == 0);
-        return XXH_OK;
-    }
+XXH_FORCE_INLINE XXH_errorcode XXH3_update(
+    XXH3_state_t *XXH_RESTRICT const state, const xxh_u8 *XXH_RESTRICT input,
+    size_t len, XXH3_f_accumulate f_acc, XXH3_f_scrambleAcc f_scramble) {
 
-    XXH_ASSERT(state != NULL);
-    {   const xxh_u8* const bEnd = input + len;
-        const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret;
-#if defined(XXH3_STREAM_USE_STACK) && XXH3_STREAM_USE_STACK >= 1
-        /* For some reason, gcc and MSVC seem to suffer greatly
-         * when operating accumulators directly into state.
-         * Operating into stack space seems to enable proper optimization.
-         * clang, on the other hand, doesn't seem to need this trick */
-        XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[8];
-        XXH_memcpy(acc, state->acc, sizeof(acc));
-#else
-        xxh_u64* XXH_RESTRICT const acc = state->acc;
-#endif
-        state->totalLen += len;
-        XXH_ASSERT(state->bufferedSize <= XXH3_INTERNALBUFFER_SIZE);
-
-        /* small input : just fill in tmp buffer */
-        if (len <= XXH3_INTERNALBUFFER_SIZE - state->bufferedSize) {
-            XXH_memcpy(state->buffer + state->bufferedSize, input, len);
-            state->bufferedSize += (XXH32_hash_t)len;
-            return XXH_OK;
-        }
+  if (input == NULL) {
+
+    XXH_ASSERT(len == 0);
+    return XXH_OK;
+
+  }
+
+  XXH_ASSERT(state != NULL);
+  {
+
+    const xxh_u8 *const        bEnd = input + len;
+    const unsigned char *const secret =
+        (state->extSecret == NULL) ? state->customSecret : state->extSecret;
+        #if defined(XXH3_STREAM_USE_STACK) && XXH3_STREAM_USE_STACK >= 1
+    /* For some reason, gcc and MSVC seem to suffer greatly
+     * when operating accumulators directly into state.
+     * Operating into stack space seems to enable proper optimization.
+     * clang, on the other hand, doesn't seem to need this trick */
+    XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[8];
+    XXH_memcpy(acc, state->acc, sizeof(acc));
+        #else
+    xxh_u64 *XXH_RESTRICT const acc = state->acc;
+        #endif
+    state->totalLen += len;
+    XXH_ASSERT(state->bufferedSize <= XXH3_INTERNALBUFFER_SIZE);
+
+    /* small input : just fill in tmp buffer */
+    if (len <= XXH3_INTERNALBUFFER_SIZE - state->bufferedSize) {
+
+      XXH_memcpy(state->buffer + state->bufferedSize, input, len);
+      state->bufferedSize += (XXH32_hash_t)len;
+      return XXH_OK;
+
+    }
 
         /* total input is now > XXH3_INTERNALBUFFER_SIZE */
-        #define XXH3_INTERNALBUFFER_STRIPES (XXH3_INTERNALBUFFER_SIZE / XXH_STRIPE_LEN)
-        XXH_STATIC_ASSERT(XXH3_INTERNALBUFFER_SIZE % XXH_STRIPE_LEN == 0);   /* clean multiple */
+        #define XXH3_INTERNALBUFFER_STRIPES \
+          (XXH3_INTERNALBUFFER_SIZE / XXH_STRIPE_LEN)
+    XXH_STATIC_ASSERT(XXH3_INTERNALBUFFER_SIZE % XXH_STRIPE_LEN ==
+                      0);                                 /* clean multiple */
 
-        /*
-         * Internal buffer is partially filled (always, except at beginning)
-         * Complete it, then consume it.
-         */
-        if (state->bufferedSize) {
-            size_t const loadSize = XXH3_INTERNALBUFFER_SIZE - state->bufferedSize;
-            XXH_memcpy(state->buffer + state->bufferedSize, input, loadSize);
-            input += loadSize;
-            XXH3_consumeStripes(acc,
-                               &state->nbStripesSoFar, state->nbStripesPerBlock,
-                                state->buffer, XXH3_INTERNALBUFFER_STRIPES,
-                                secret, state->secretLimit,
-                                f_acc, f_scramble);
-            state->bufferedSize = 0;
-        }
-        XXH_ASSERT(input < bEnd);
-        if (bEnd - input > XXH3_INTERNALBUFFER_SIZE) {
-            size_t nbStripes = (size_t)(bEnd - 1 - input) / XXH_STRIPE_LEN;
-            input = XXH3_consumeStripes(acc,
-                                       &state->nbStripesSoFar, state->nbStripesPerBlock,
-                                       input, nbStripes,
-                                       secret, state->secretLimit,
-                                       f_acc, f_scramble);
-            XXH_memcpy(state->buffer + sizeof(state->buffer) - XXH_STRIPE_LEN, input - XXH_STRIPE_LEN, XXH_STRIPE_LEN);
+    /*
+     * Internal buffer is partially filled (always, except at beginning)
+     * Complete it, then consume it.
+     */
+    if (state->bufferedSize) {
+
+      size_t const loadSize = XXH3_INTERNALBUFFER_SIZE - state->bufferedSize;
+      XXH_memcpy(state->buffer + state->bufferedSize, input, loadSize);
+      input += loadSize;
+      XXH3_consumeStripes(acc, &state->nbStripesSoFar, state->nbStripesPerBlock,
+                          state->buffer, XXH3_INTERNALBUFFER_STRIPES, secret,
+                          state->secretLimit, f_acc, f_scramble);
+      state->bufferedSize = 0;
+
+    }
+
+    XXH_ASSERT(input < bEnd);
+    if (bEnd - input > XXH3_INTERNALBUFFER_SIZE) {
+
+      size_t nbStripes = (size_t)(bEnd - 1 - input) / XXH_STRIPE_LEN;
+      input = XXH3_consumeStripes(
+          acc, &state->nbStripesSoFar, state->nbStripesPerBlock, input,
+          nbStripes, secret, state->secretLimit, f_acc, f_scramble);
+      XXH_memcpy(state->buffer + sizeof(state->buffer) - XXH_STRIPE_LEN,
+                 input - XXH_STRIPE_LEN, XXH_STRIPE_LEN);
 
-        }
-        /* Some remaining input (always) : buffer it */
-        XXH_ASSERT(input < bEnd);
-        XXH_ASSERT(bEnd - input <= XXH3_INTERNALBUFFER_SIZE);
-        XXH_ASSERT(state->bufferedSize == 0);
-        XXH_memcpy(state->buffer, input, (size_t)(bEnd-input));
-        state->bufferedSize = (XXH32_hash_t)(bEnd-input);
-#if defined(XXH3_STREAM_USE_STACK) && XXH3_STREAM_USE_STACK >= 1
-        /* save stack accumulators into state */
-        XXH_memcpy(state->acc, acc, sizeof(acc));
-#endif
     }
 
-    return XXH_OK;
+    /* Some remaining input (always) : buffer it */
+    XXH_ASSERT(input < bEnd);
+    XXH_ASSERT(bEnd - input <= XXH3_INTERNALBUFFER_SIZE);
+    XXH_ASSERT(state->bufferedSize == 0);
+    XXH_memcpy(state->buffer, input, (size_t)(bEnd - input));
+    state->bufferedSize = (XXH32_hash_t)(bEnd - input);
+        #if defined(XXH3_STREAM_USE_STACK) && XXH3_STREAM_USE_STACK >= 1
+    /* save stack accumulators into state */
+    XXH_memcpy(state->acc, acc, sizeof(acc));
+        #endif
+
+  }
+
+  return XXH_OK;
+
 }
 
 /*! @ingroup XXH3_family */
 XXH_PUBLIC_API XXH_errorcode
-XXH3_64bits_update(XXH_NOESCAPE XXH3_state_t* state, XXH_NOESCAPE const void* input, size_t len)
-{
-    return XXH3_update(state, (const xxh_u8*)input, len,
-                       XXH3_accumulate, XXH3_scrambleAcc);
+XXH3_64bits_update(XXH_NOESCAPE XXH3_state_t *state,
+                   XXH_NOESCAPE const void *input, size_t len) {
+
+  return XXH3_update(state, (const xxh_u8 *)input, len, XXH3_accumulate,
+                     XXH3_scrambleAcc);
+
 }
 
+XXH_FORCE_INLINE void XXH3_digest_long(XXH64_hash_t        *acc,
+                                       const XXH3_state_t  *state,
+                                       const unsigned char *secret) {
 
-XXH_FORCE_INLINE void
-XXH3_digest_long (XXH64_hash_t* acc,
-                  const XXH3_state_t* state,
-                  const unsigned char* secret)
-{
-    xxh_u8 lastStripe[XXH_STRIPE_LEN];
-    const xxh_u8* lastStripePtr;
+  xxh_u8        lastStripe[XXH_STRIPE_LEN];
+  const xxh_u8 *lastStripePtr;
+
+  /*
+   * Digest on a local copy. This way, the state remains unaltered, and it can
+   * continue ingesting more input afterwards.
+   */
+  XXH_memcpy(acc, state->acc, sizeof(state->acc));
+  if (state->bufferedSize >= XXH_STRIPE_LEN) {
+
+    /* Consume remaining stripes then point to remaining data in buffer */
+    size_t const nbStripes = (state->bufferedSize - 1) / XXH_STRIPE_LEN;
+    size_t       nbStripesSoFar = state->nbStripesSoFar;
+    XXH3_consumeStripes(acc, &nbStripesSoFar, state->nbStripesPerBlock,
+                        state->buffer, nbStripes, secret, state->secretLimit,
+                        XXH3_accumulate, XXH3_scrambleAcc);
+    lastStripePtr = state->buffer + state->bufferedSize - XXH_STRIPE_LEN;
+
+  } else {                                 /* bufferedSize < XXH_STRIPE_LEN */
+
+    /* Copy to temp buffer */
+    size_t const catchupSize = XXH_STRIPE_LEN - state->bufferedSize;
+    XXH_ASSERT(state->bufferedSize >
+               0);                   /* there is always some input buffered */
+    XXH_memcpy(lastStripe, state->buffer + sizeof(state->buffer) - catchupSize,
+               catchupSize);
+    XXH_memcpy(lastStripe + catchupSize, state->buffer, state->bufferedSize);
+    lastStripePtr = lastStripe;
+
+  }
+
+  /* Last stripe */
+  XXH3_accumulate_512(acc, lastStripePtr,
+                      secret + state->secretLimit - XXH_SECRET_LASTACC_START);
 
-    /*
-     * Digest on a local copy. This way, the state remains unaltered, and it can
-     * continue ingesting more input afterwards.
-     */
-    XXH_memcpy(acc, state->acc, sizeof(state->acc));
-    if (state->bufferedSize >= XXH_STRIPE_LEN) {
-        /* Consume remaining stripes then point to remaining data in buffer */
-        size_t const nbStripes = (state->bufferedSize - 1) / XXH_STRIPE_LEN;
-        size_t nbStripesSoFar = state->nbStripesSoFar;
-        XXH3_consumeStripes(acc,
-                           &nbStripesSoFar, state->nbStripesPerBlock,
-                            state->buffer, nbStripes,
-                            secret, state->secretLimit,
-                            XXH3_accumulate, XXH3_scrambleAcc);
-        lastStripePtr = state->buffer + state->bufferedSize - XXH_STRIPE_LEN;
-    } else {  /* bufferedSize < XXH_STRIPE_LEN */
-        /* Copy to temp buffer */
-        size_t const catchupSize = XXH_STRIPE_LEN - state->bufferedSize;
-        XXH_ASSERT(state->bufferedSize > 0);  /* there is always some input buffered */
-        XXH_memcpy(lastStripe, state->buffer + sizeof(state->buffer) - catchupSize, catchupSize);
-        XXH_memcpy(lastStripe + catchupSize, state->buffer, state->bufferedSize);
-        lastStripePtr = lastStripe;
-    }
-    /* Last stripe */
-    XXH3_accumulate_512(acc,
-                        lastStripePtr,
-                        secret + state->secretLimit - XXH_SECRET_LASTACC_START);
 }
 
 /*! @ingroup XXH3_family */
-XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_digest (XXH_NOESCAPE const XXH3_state_t* state)
-{
-    const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret;
-    if (state->totalLen > XXH3_MIDSIZE_MAX) {
-        XXH_ALIGN(XXH_ACC_ALIGN) XXH64_hash_t acc[XXH_ACC_NB];
-        XXH3_digest_long(acc, state, secret);
-        return XXH3_mergeAccs(acc,
-                              secret + XXH_SECRET_MERGEACCS_START,
-                              (xxh_u64)state->totalLen * XXH_PRIME64_1);
-    }
-    /* totalLen <= XXH3_MIDSIZE_MAX: digesting a short input */
-    if (state->useSeed)
-        return XXH3_64bits_withSeed(state->buffer, (size_t)state->totalLen, state->seed);
-    return XXH3_64bits_withSecret(state->buffer, (size_t)(state->totalLen),
-                                  secret, state->secretLimit + XXH_STRIPE_LEN);
+XXH_PUBLIC_API XXH64_hash_t
+XXH3_64bits_digest(XXH_NOESCAPE const XXH3_state_t *state) {
+
+  const unsigned char *const secret =
+      (state->extSecret == NULL) ? state->customSecret : state->extSecret;
+  if (state->totalLen > XXH3_MIDSIZE_MAX) {
+
+    XXH_ALIGN(XXH_ACC_ALIGN) XXH64_hash_t acc[XXH_ACC_NB];
+    XXH3_digest_long(acc, state, secret);
+    return XXH3_mergeAccs(acc, secret + XXH_SECRET_MERGEACCS_START,
+                          (xxh_u64)state->totalLen * XXH_PRIME64_1);
+
+  }
+
+  /* totalLen <= XXH3_MIDSIZE_MAX: digesting a short input */
+  if (state->useSeed)
+    return XXH3_64bits_withSeed(state->buffer, (size_t)state->totalLen,
+                                state->seed);
+  return XXH3_64bits_withSecret(state->buffer, (size_t)(state->totalLen),
+                                secret, state->secretLimit + XXH_STRIPE_LEN);
+
 }
-#endif /* !XXH_NO_STREAM */
 
+      #endif                                              /* !XXH_NO_STREAM */
 
 /* ==========================================
  * XXH3 128 bits (a.k.a XXH128)
  * ==========================================
- * XXH3's 128-bit variant has better mixing and strength than the 64-bit variant,
- * even without counting the significantly larger output size.
+ * XXH3's 128-bit variant has better mixing and strength than the 64-bit
+ * variant, even without counting the significantly larger output size.
  *
  * For example, extra steps are taken to avoid the seed-dependent collisions
  * in 17-240 byte inputs (See XXH3_mix16B and XXH128_mix32B).
@@ -6416,503 +7257,614 @@ XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_digest (XXH_NOESCAPE const XXH3_state_t*
  * fast for a _128-bit_ hash on 32-bit (it usually clears XXH64).
  */
 
-XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t
-XXH3_len_1to3_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
-{
-    /* A doubled version of 1to3_64b with different constants. */
-    XXH_ASSERT(input != NULL);
-    XXH_ASSERT(1 <= len && len <= 3);
-    XXH_ASSERT(secret != NULL);
-    /*
-     * len = 1: combinedl = { input[0], 0x01, input[0], input[0] }
-     * len = 2: combinedl = { input[1], 0x02, input[0], input[1] }
-     * len = 3: combinedl = { input[2], 0x03, input[0], input[1] }
+XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t XXH3_len_1to3_128b(
+    const xxh_u8 *input, size_t len, const xxh_u8 *secret, XXH64_hash_t seed) {
+
+  /* A doubled version of 1to3_64b with different constants. */
+  XXH_ASSERT(input != NULL);
+  XXH_ASSERT(1 <= len && len <= 3);
+  XXH_ASSERT(secret != NULL);
+  /*
+   * len = 1: combinedl = { input[0], 0x01, input[0], input[0] }
+   * len = 2: combinedl = { input[1], 0x02, input[0], input[1] }
+   * len = 3: combinedl = { input[2], 0x03, input[0], input[1] }
+   */
+  {
+
+    xxh_u8 const  c1 = input[0];
+    xxh_u8 const  c2 = input[len >> 1];
+    xxh_u8 const  c3 = input[len - 1];
+    xxh_u32 const combinedl = ((xxh_u32)c1 << 16) | ((xxh_u32)c2 << 24) |
+                              ((xxh_u32)c3 << 0) | ((xxh_u32)len << 8);
+    xxh_u32 const combinedh = XXH_rotl32(XXH_swap32(combinedl), 13);
+    xxh_u64 const bitflipl =
+        (XXH_readLE32(secret) ^ XXH_readLE32(secret + 4)) + seed;
+    xxh_u64 const bitfliph =
+        (XXH_readLE32(secret + 8) ^ XXH_readLE32(secret + 12)) - seed;
+    xxh_u64 const keyed_lo = (xxh_u64)combinedl ^ bitflipl;
+    xxh_u64 const keyed_hi = (xxh_u64)combinedh ^ bitfliph;
+    XXH128_hash_t h128;
+    h128.low64 = XXH64_avalanche(keyed_lo);
+    h128.high64 = XXH64_avalanche(keyed_hi);
+    return h128;
+
+  }
+
+}
+
+XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t XXH3_len_4to8_128b(
+    const xxh_u8 *input, size_t len, const xxh_u8 *secret, XXH64_hash_t seed) {
+
+  XXH_ASSERT(input != NULL);
+  XXH_ASSERT(secret != NULL);
+  XXH_ASSERT(4 <= len && len <= 8);
+  seed ^= (xxh_u64)XXH_swap32((xxh_u32)seed) << 32;
+  {
+
+    xxh_u32 const input_lo = XXH_readLE32(input);
+    xxh_u32 const input_hi = XXH_readLE32(input + len - 4);
+    xxh_u64 const input_64 = input_lo + ((xxh_u64)input_hi << 32);
+    xxh_u64 const bitflip =
+        (XXH_readLE64(secret + 16) ^ XXH_readLE64(secret + 24)) + seed;
+    xxh_u64 const keyed = input_64 ^ bitflip;
+
+    /* Shift len to the left to ensure it is even, this avoids even multiplies.
      */
-    {   xxh_u8 const c1 = input[0];
-        xxh_u8 const c2 = input[len >> 1];
-        xxh_u8 const c3 = input[len - 1];
-        xxh_u32 const combinedl = ((xxh_u32)c1 <<16) | ((xxh_u32)c2 << 24)
-                                | ((xxh_u32)c3 << 0) | ((xxh_u32)len << 8);
-        xxh_u32 const combinedh = XXH_rotl32(XXH_swap32(combinedl), 13);
-        xxh_u64 const bitflipl = (XXH_readLE32(secret) ^ XXH_readLE32(secret+4)) + seed;
-        xxh_u64 const bitfliph = (XXH_readLE32(secret+8) ^ XXH_readLE32(secret+12)) - seed;
-        xxh_u64 const keyed_lo = (xxh_u64)combinedl ^ bitflipl;
-        xxh_u64 const keyed_hi = (xxh_u64)combinedh ^ bitfliph;
-        XXH128_hash_t h128;
-        h128.low64  = XXH64_avalanche(keyed_lo);
-        h128.high64 = XXH64_avalanche(keyed_hi);
-        return h128;
-    }
+    XXH128_hash_t m128 = XXH_mult64to128(keyed, XXH_PRIME64_1 + (len << 2));
+
+    m128.high64 += (m128.low64 << 1);
+    m128.low64 ^= (m128.high64 >> 3);
+
+    m128.low64 = XXH_xorshift64(m128.low64, 35);
+    m128.low64 *= PRIME_MX2;
+    m128.low64 = XXH_xorshift64(m128.low64, 28);
+    m128.high64 = XXH3_avalanche(m128.high64);
+    return m128;
+
+  }
+
 }
 
-XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t
-XXH3_len_4to8_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
-{
-    XXH_ASSERT(input != NULL);
-    XXH_ASSERT(secret != NULL);
-    XXH_ASSERT(4 <= len && len <= 8);
-    seed ^= (xxh_u64)XXH_swap32((xxh_u32)seed) << 32;
-    {   xxh_u32 const input_lo = XXH_readLE32(input);
-        xxh_u32 const input_hi = XXH_readLE32(input + len - 4);
-        xxh_u64 const input_64 = input_lo + ((xxh_u64)input_hi << 32);
-        xxh_u64 const bitflip = (XXH_readLE64(secret+16) ^ XXH_readLE64(secret+24)) + seed;
-        xxh_u64 const keyed = input_64 ^ bitflip;
-
-        /* Shift len to the left to ensure it is even, this avoids even multiplies. */
-        XXH128_hash_t m128 = XXH_mult64to128(keyed, XXH_PRIME64_1 + (len << 2));
-
-        m128.high64 += (m128.low64 << 1);
-        m128.low64  ^= (m128.high64 >> 3);
-
-        m128.low64   = XXH_xorshift64(m128.low64, 35);
-        m128.low64  *= PRIME_MX2;
-        m128.low64   = XXH_xorshift64(m128.low64, 28);
-        m128.high64  = XXH3_avalanche(m128.high64);
-        return m128;
+XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t XXH3_len_9to16_128b(
+    const xxh_u8 *input, size_t len, const xxh_u8 *secret, XXH64_hash_t seed) {
+
+  XXH_ASSERT(input != NULL);
+  XXH_ASSERT(secret != NULL);
+  XXH_ASSERT(9 <= len && len <= 16);
+  {
+
+    xxh_u64 const bitflipl =
+        (XXH_readLE64(secret + 32) ^ XXH_readLE64(secret + 40)) - seed;
+    xxh_u64 const bitfliph =
+        (XXH_readLE64(secret + 48) ^ XXH_readLE64(secret + 56)) + seed;
+    xxh_u64 const input_lo = XXH_readLE64(input);
+    xxh_u64       input_hi = XXH_readLE64(input + len - 8);
+    XXH128_hash_t m128 =
+        XXH_mult64to128(input_lo ^ input_hi ^ bitflipl, XXH_PRIME64_1);
+    /*
+     * Put len in the middle of m128 to ensure that the length gets mixed to
+     * both the low and high bits in the 128x64 multiply below.
+     */
+    m128.low64 += (xxh_u64)(len - 1) << 54;
+    input_hi ^= bitfliph;
+    /*
+     * Add the high 32 bits of input_hi to the high 32 bits of m128, then
+     * add the long product of the low 32 bits of input_hi and XXH_PRIME32_2 to
+     * the high 64 bits of m128.
+     *
+     * The best approach to this operation is different on 32-bit and 64-bit.
+     */
+    if (sizeof(void *) < sizeof(xxh_u64)) {                       /* 32-bit */
+      /*
+       * 32-bit optimized version, which is more readable.
+       *
+       * On 32-bit, it removes an ADC and delays a dependency between the two
+       * halves of m128.high64, but it generates an extra mask on 64-bit.
+       */
+      m128.high64 += (input_hi & 0xFFFFFFFF00000000ULL) +
+                     XXH_mult32to64((xxh_u32)input_hi, XXH_PRIME32_2);
+
+    } else {
+
+      /*
+       * 64-bit optimized (albeit more confusing) version.
+       *
+       * Uses some properties of addition and multiplication to remove the mask:
+       *
+       * Let:
+       *    a = input_hi.lo = (input_hi & 0x00000000FFFFFFFF)
+       *    b = input_hi.hi = (input_hi & 0xFFFFFFFF00000000)
+       *    c = XXH_PRIME32_2
+       *
+       *    a + (b * c)
+       * Inverse Property: x + y - x == y
+       *    a + (b * (1 + c - 1))
+       * Distributive Property: x * (y + z) == (x * y) + (x * z)
+       *    a + (b * 1) + (b * (c - 1))
+       * Identity Property: x * 1 == x
+       *    a + b + (b * (c - 1))
+       *
+       * Substitute a, b, and c:
+       *    input_hi.hi + input_hi.lo + ((xxh_u64)input_hi.lo * (XXH_PRIME32_2 -
+       * 1))
+       *
+       * Since input_hi.hi + input_hi.lo == input_hi, we get this:
+       *    input_hi + ((xxh_u64)input_hi.lo * (XXH_PRIME32_2 - 1))
+       */
+      m128.high64 +=
+          input_hi + XXH_mult32to64((xxh_u32)input_hi, XXH_PRIME32_2 - 1);
+
     }
-}
 
-XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t
-XXH3_len_9to16_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
-{
-    XXH_ASSERT(input != NULL);
-    XXH_ASSERT(secret != NULL);
-    XXH_ASSERT(9 <= len && len <= 16);
-    {   xxh_u64 const bitflipl = (XXH_readLE64(secret+32) ^ XXH_readLE64(secret+40)) - seed;
-        xxh_u64 const bitfliph = (XXH_readLE64(secret+48) ^ XXH_readLE64(secret+56)) + seed;
-        xxh_u64 const input_lo = XXH_readLE64(input);
-        xxh_u64       input_hi = XXH_readLE64(input + len - 8);
-        XXH128_hash_t m128 = XXH_mult64to128(input_lo ^ input_hi ^ bitflipl, XXH_PRIME64_1);
-        /*
-         * Put len in the middle of m128 to ensure that the length gets mixed to
-         * both the low and high bits in the 128x64 multiply below.
-         */
-        m128.low64 += (xxh_u64)(len - 1) << 54;
-        input_hi   ^= bitfliph;
-        /*
-         * Add the high 32 bits of input_hi to the high 32 bits of m128, then
-         * add the long product of the low 32 bits of input_hi and XXH_PRIME32_2 to
-         * the high 64 bits of m128.
-         *
-         * The best approach to this operation is different on 32-bit and 64-bit.
-         */
-        if (sizeof(void *) < sizeof(xxh_u64)) { /* 32-bit */
-            /*
-             * 32-bit optimized version, which is more readable.
-             *
-             * On 32-bit, it removes an ADC and delays a dependency between the two
-             * halves of m128.high64, but it generates an extra mask on 64-bit.
-             */
-            m128.high64 += (input_hi & 0xFFFFFFFF00000000ULL) + XXH_mult32to64((xxh_u32)input_hi, XXH_PRIME32_2);
-        } else {
-            /*
-             * 64-bit optimized (albeit more confusing) version.
-             *
-             * Uses some properties of addition and multiplication to remove the mask:
-             *
-             * Let:
-             *    a = input_hi.lo = (input_hi & 0x00000000FFFFFFFF)
-             *    b = input_hi.hi = (input_hi & 0xFFFFFFFF00000000)
-             *    c = XXH_PRIME32_2
-             *
-             *    a + (b * c)
-             * Inverse Property: x + y - x == y
-             *    a + (b * (1 + c - 1))
-             * Distributive Property: x * (y + z) == (x * y) + (x * z)
-             *    a + (b * 1) + (b * (c - 1))
-             * Identity Property: x * 1 == x
-             *    a + b + (b * (c - 1))
-             *
-             * Substitute a, b, and c:
-             *    input_hi.hi + input_hi.lo + ((xxh_u64)input_hi.lo * (XXH_PRIME32_2 - 1))
-             *
-             * Since input_hi.hi + input_hi.lo == input_hi, we get this:
-             *    input_hi + ((xxh_u64)input_hi.lo * (XXH_PRIME32_2 - 1))
-             */
-            m128.high64 += input_hi + XXH_mult32to64((xxh_u32)input_hi, XXH_PRIME32_2 - 1);
-        }
-        /* m128 ^= XXH_swap64(m128 >> 64); */
-        m128.low64  ^= XXH_swap64(m128.high64);
+    /* m128 ^= XXH_swap64(m128 >> 64); */
+    m128.low64 ^= XXH_swap64(m128.high64);
+
+    {                      /* 128x64 multiply: h128 = m128 * XXH_PRIME64_2; */
+      XXH128_hash_t h128 = XXH_mult64to128(m128.low64, XXH_PRIME64_2);
+      h128.high64 += m128.high64 * XXH_PRIME64_2;
 
-        {   /* 128x64 multiply: h128 = m128 * XXH_PRIME64_2; */
-            XXH128_hash_t h128 = XXH_mult64to128(m128.low64, XXH_PRIME64_2);
-            h128.high64 += m128.high64 * XXH_PRIME64_2;
+      h128.low64 = XXH3_avalanche(h128.low64);
+      h128.high64 = XXH3_avalanche(h128.high64);
+      return h128;
+
+    }
+
+  }
 
-            h128.low64   = XXH3_avalanche(h128.low64);
-            h128.high64  = XXH3_avalanche(h128.high64);
-            return h128;
-    }   }
 }
 
 /*
  * Assumption: `secret` size is >= XXH3_SECRET_SIZE_MIN
  */
-XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t
-XXH3_len_0to16_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
-{
-    XXH_ASSERT(len <= 16);
-    {   if (len > 8) return XXH3_len_9to16_128b(input, len, secret, seed);
-        if (len >= 4) return XXH3_len_4to8_128b(input, len, secret, seed);
-        if (len) return XXH3_len_1to3_128b(input, len, secret, seed);
-        {   XXH128_hash_t h128;
-            xxh_u64 const bitflipl = XXH_readLE64(secret+64) ^ XXH_readLE64(secret+72);
-            xxh_u64 const bitfliph = XXH_readLE64(secret+80) ^ XXH_readLE64(secret+88);
-            h128.low64 = XXH64_avalanche(seed ^ bitflipl);
-            h128.high64 = XXH64_avalanche( seed ^ bitfliph);
-            return h128;
-    }   }
+XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t XXH3_len_0to16_128b(
+    const xxh_u8 *input, size_t len, const xxh_u8 *secret, XXH64_hash_t seed) {
+
+  XXH_ASSERT(len <= 16);
+  {
+
+    if (len > 8) return XXH3_len_9to16_128b(input, len, secret, seed);
+    if (len >= 4) return XXH3_len_4to8_128b(input, len, secret, seed);
+    if (len) return XXH3_len_1to3_128b(input, len, secret, seed);
+    {
+
+      XXH128_hash_t h128;
+      xxh_u64 const bitflipl =
+          XXH_readLE64(secret + 64) ^ XXH_readLE64(secret + 72);
+      xxh_u64 const bitfliph =
+          XXH_readLE64(secret + 80) ^ XXH_readLE64(secret + 88);
+      h128.low64 = XXH64_avalanche(seed ^ bitflipl);
+      h128.high64 = XXH64_avalanche(seed ^ bitfliph);
+      return h128;
+
+    }
+
+  }
+
 }
 
 /*
  * A bit slower than XXH3_mix16B, but handles multiply by zero better.
  */
-XXH_FORCE_INLINE XXH128_hash_t
-XXH128_mix32B(XXH128_hash_t acc, const xxh_u8* input_1, const xxh_u8* input_2,
-              const xxh_u8* secret, XXH64_hash_t seed)
-{
-    acc.low64  += XXH3_mix16B (input_1, secret+0, seed);
-    acc.low64  ^= XXH_readLE64(input_2) + XXH_readLE64(input_2 + 8);
-    acc.high64 += XXH3_mix16B (input_2, secret+16, seed);
-    acc.high64 ^= XXH_readLE64(input_1) + XXH_readLE64(input_1 + 8);
-    return acc;
+XXH_FORCE_INLINE XXH128_hash_t XXH128_mix32B(XXH128_hash_t acc,
+                                             const xxh_u8 *input_1,
+                                             const xxh_u8 *input_2,
+                                             const xxh_u8 *secret,
+                                             XXH64_hash_t  seed) {
+
+  acc.low64 += XXH3_mix16B(input_1, secret + 0, seed);
+  acc.low64 ^= XXH_readLE64(input_2) + XXH_readLE64(input_2 + 8);
+  acc.high64 += XXH3_mix16B(input_2, secret + 16, seed);
+  acc.high64 ^= XXH_readLE64(input_1) + XXH_readLE64(input_1 + 8);
+  return acc;
+
 }
 
+XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t XXH3_len_17to128_128b(
+    const xxh_u8 *XXH_RESTRICT input, size_t len,
+    const xxh_u8 *XXH_RESTRICT secret, size_t secretSize, XXH64_hash_t seed) {
+
+  XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
+  (void)secretSize;
+  XXH_ASSERT(16 < len && len <= 128);
+
+  {
+
+    XXH128_hash_t acc;
+    acc.low64 = len * XXH_PRIME64_1;
+    acc.high64 = 0;
+
+      #if XXH_SIZE_OPT >= 1
+    {
+
+      /* Smaller, but slightly slower. */
+      unsigned int i = (unsigned int)(len - 1) / 32;
+      do {
+
+        acc = XXH128_mix32B(acc, input + 16 * i, input + len - 16 * (i + 1),
+                            secret + 32 * i, seed);
+
+      } while (i-- != 0);
 
-XXH_FORCE_INLINE XXH_PUREF XXH128_hash_t
-XXH3_len_17to128_128b(const xxh_u8* XXH_RESTRICT input, size_t len,
-                      const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
-                      XXH64_hash_t seed)
-{
-    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
-    XXH_ASSERT(16 < len && len <= 128);
-
-    {   XXH128_hash_t acc;
-        acc.low64 = len * XXH_PRIME64_1;
-        acc.high64 = 0;
-
-#if XXH_SIZE_OPT >= 1
-        {
-            /* Smaller, but slightly slower. */
-            unsigned int i = (unsigned int)(len - 1) / 32;
-            do {
-                acc = XXH128_mix32B(acc, input+16*i, input+len-16*(i+1), secret+32*i, seed);
-            } while (i-- != 0);
-        }
-#else
-        if (len > 32) {
-            if (len > 64) {
-                if (len > 96) {
-                    acc = XXH128_mix32B(acc, input+48, input+len-64, secret+96, seed);
-                }
-                acc = XXH128_mix32B(acc, input+32, input+len-48, secret+64, seed);
-            }
-            acc = XXH128_mix32B(acc, input+16, input+len-32, secret+32, seed);
-        }
-        acc = XXH128_mix32B(acc, input, input+len-16, secret, seed);
-#endif
-        {   XXH128_hash_t h128;
-            h128.low64  = acc.low64 + acc.high64;
-            h128.high64 = (acc.low64    * XXH_PRIME64_1)
-                        + (acc.high64   * XXH_PRIME64_4)
-                        + ((len - seed) * XXH_PRIME64_2);
-            h128.low64  = XXH3_avalanche(h128.low64);
-            h128.high64 = (XXH64_hash_t)0 - XXH3_avalanche(h128.high64);
-            return h128;
-        }
     }
-}
 
-XXH_NO_INLINE XXH_PUREF XXH128_hash_t
-XXH3_len_129to240_128b(const xxh_u8* XXH_RESTRICT input, size_t len,
-                       const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
-                       XXH64_hash_t seed)
-{
-    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
-    XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX);
+      #else
+    if (len > 32) {
+
+      if (len > 64) {
+
+        if (len > 96) {
+
+          acc = XXH128_mix32B(acc, input + 48, input + len - 64, secret + 96,
+                              seed);
 
-    {   XXH128_hash_t acc;
-        unsigned i;
-        acc.low64 = len * XXH_PRIME64_1;
-        acc.high64 = 0;
-        /*
-         *  We set as `i` as offset + 32. We do this so that unchanged
-         * `len` can be used as upper bound. This reaches a sweet spot
-         * where both x86 and aarch64 get simple agen and good codegen
-         * for the loop.
-         */
-        for (i = 32; i < 160; i += 32) {
-            acc = XXH128_mix32B(acc,
-                                input  + i - 32,
-                                input  + i - 16,
-                                secret + i - 32,
-                                seed);
-        }
-        acc.low64 = XXH3_avalanche(acc.low64);
-        acc.high64 = XXH3_avalanche(acc.high64);
-        /*
-         * NB: `i <= len` will duplicate the last 32-bytes if
-         * len % 32 was zero. This is an unfortunate necessity to keep
-         * the hash result stable.
-         */
-        for (i=160; i <= len; i += 32) {
-            acc = XXH128_mix32B(acc,
-                                input + i - 32,
-                                input + i - 16,
-                                secret + XXH3_MIDSIZE_STARTOFFSET + i - 160,
-                                seed);
-        }
-        /* last bytes */
-        acc = XXH128_mix32B(acc,
-                            input + len - 16,
-                            input + len - 32,
-                            secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET - 16,
-                            (XXH64_hash_t)0 - seed);
-
-        {   XXH128_hash_t h128;
-            h128.low64  = acc.low64 + acc.high64;
-            h128.high64 = (acc.low64    * XXH_PRIME64_1)
-                        + (acc.high64   * XXH_PRIME64_4)
-                        + ((len - seed) * XXH_PRIME64_2);
-            h128.low64  = XXH3_avalanche(h128.low64);
-            h128.high64 = (XXH64_hash_t)0 - XXH3_avalanche(h128.high64);
-            return h128;
         }
+
+        acc =
+            XXH128_mix32B(acc, input + 32, input + len - 48, secret + 64, seed);
+
+      }
+
+      acc = XXH128_mix32B(acc, input + 16, input + len - 32, secret + 32, seed);
+
+    }
+
+    acc = XXH128_mix32B(acc, input, input + len - 16, secret, seed);
+      #endif
+    {
+
+      XXH128_hash_t h128;
+      h128.low64 = acc.low64 + acc.high64;
+      h128.high64 = (acc.low64 * XXH_PRIME64_1) + (acc.high64 * XXH_PRIME64_4) +
+                    ((len - seed) * XXH_PRIME64_2);
+      h128.low64 = XXH3_avalanche(h128.low64);
+      h128.high64 = (XXH64_hash_t)0 - XXH3_avalanche(h128.high64);
+      return h128;
+
     }
+
+  }
+
 }
 
-XXH_FORCE_INLINE XXH128_hash_t
-XXH3_hashLong_128b_internal(const void* XXH_RESTRICT input, size_t len,
-                            const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
-                            XXH3_f_accumulate f_acc,
-                            XXH3_f_scrambleAcc f_scramble)
-{
-    XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[XXH_ACC_NB] = XXH3_INIT_ACC;
-
-    XXH3_hashLong_internal_loop(acc, (const xxh_u8*)input, len, secret, secretSize, f_acc, f_scramble);
-
-    /* converge into final hash */
-    XXH_STATIC_ASSERT(sizeof(acc) == 64);
-    XXH_ASSERT(secretSize >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
-    {   XXH128_hash_t h128;
-        h128.low64  = XXH3_mergeAccs(acc,
-                                     secret + XXH_SECRET_MERGEACCS_START,
-                                     (xxh_u64)len * XXH_PRIME64_1);
-        h128.high64 = XXH3_mergeAccs(acc,
-                                     secret + secretSize
-                                            - sizeof(acc) - XXH_SECRET_MERGEACCS_START,
-                                     ~((xxh_u64)len * XXH_PRIME64_2));
-        return h128;
+XXH_NO_INLINE XXH_PUREF XXH128_hash_t XXH3_len_129to240_128b(
+    const xxh_u8 *XXH_RESTRICT input, size_t len,
+    const xxh_u8 *XXH_RESTRICT secret, size_t secretSize, XXH64_hash_t seed) {
+
+  XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
+  (void)secretSize;
+  XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX);
+
+  {
+
+    XXH128_hash_t acc;
+    unsigned      i;
+    acc.low64 = len * XXH_PRIME64_1;
+    acc.high64 = 0;
+    /*
+     *  We set as `i` as offset + 32. We do this so that unchanged
+     * `len` can be used as upper bound. This reaches a sweet spot
+     * where both x86 and aarch64 get simple agen and good codegen
+     * for the loop.
+     */
+    for (i = 32; i < 160; i += 32) {
+
+      acc = XXH128_mix32B(acc, input + i - 32, input + i - 16, secret + i - 32,
+                          seed);
+
+    }
+
+    acc.low64 = XXH3_avalanche(acc.low64);
+    acc.high64 = XXH3_avalanche(acc.high64);
+    /*
+     * NB: `i <= len` will duplicate the last 32-bytes if
+     * len % 32 was zero. This is an unfortunate necessity to keep
+     * the hash result stable.
+     */
+    for (i = 160; i <= len; i += 32) {
+
+      acc = XXH128_mix32B(acc, input + i - 32, input + i - 16,
+                          secret + XXH3_MIDSIZE_STARTOFFSET + i - 160, seed);
+
+    }
+
+    /* last bytes */
+    acc = XXH128_mix32B(
+        acc, input + len - 16, input + len - 32,
+        secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET - 16,
+        (XXH64_hash_t)0 - seed);
+
+    {
+
+      XXH128_hash_t h128;
+      h128.low64 = acc.low64 + acc.high64;
+      h128.high64 = (acc.low64 * XXH_PRIME64_1) + (acc.high64 * XXH_PRIME64_4) +
+                    ((len - seed) * XXH_PRIME64_2);
+      h128.low64 = XXH3_avalanche(h128.low64);
+      h128.high64 = (XXH64_hash_t)0 - XXH3_avalanche(h128.high64);
+      return h128;
+
     }
+
+  }
+
+}
+
+XXH_FORCE_INLINE XXH128_hash_t XXH3_hashLong_128b_internal(
+    const void *XXH_RESTRICT input, size_t len,
+    const xxh_u8 *XXH_RESTRICT secret, size_t secretSize,
+    XXH3_f_accumulate f_acc, XXH3_f_scrambleAcc f_scramble) {
+
+  XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[XXH_ACC_NB] = XXH3_INIT_ACC;
+
+  XXH3_hashLong_internal_loop(acc, (const xxh_u8 *)input, len, secret,
+                              secretSize, f_acc, f_scramble);
+
+  /* converge into final hash */
+  XXH_STATIC_ASSERT(sizeof(acc) == 64);
+  XXH_ASSERT(secretSize >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
+  {
+
+    XXH128_hash_t h128;
+    h128.low64 = XXH3_mergeAccs(acc, secret + XXH_SECRET_MERGEACCS_START,
+                                (xxh_u64)len * XXH_PRIME64_1);
+    h128.high64 = XXH3_mergeAccs(
+        acc, secret + secretSize - sizeof(acc) - XXH_SECRET_MERGEACCS_START,
+        ~((xxh_u64)len * XXH_PRIME64_2));
+    return h128;
+
+  }
+
 }
 
 /*
  * It's important for performance that XXH3_hashLong() is not inlined.
  */
-XXH_NO_INLINE XXH_PUREF XXH128_hash_t
-XXH3_hashLong_128b_default(const void* XXH_RESTRICT input, size_t len,
-                           XXH64_hash_t seed64,
-                           const void* XXH_RESTRICT secret, size_t secretLen)
-{
-    (void)seed64; (void)secret; (void)secretLen;
-    return XXH3_hashLong_128b_internal(input, len, XXH3_kSecret, sizeof(XXH3_kSecret),
-                                       XXH3_accumulate, XXH3_scrambleAcc);
+XXH_NO_INLINE XXH_PUREF XXH128_hash_t XXH3_hashLong_128b_default(
+    const void *XXH_RESTRICT input, size_t len, XXH64_hash_t seed64,
+    const void *XXH_RESTRICT secret, size_t secretLen) {
+
+  (void)seed64;
+  (void)secret;
+  (void)secretLen;
+  return XXH3_hashLong_128b_internal(input, len, XXH3_kSecret,
+                                     sizeof(XXH3_kSecret), XXH3_accumulate,
+                                     XXH3_scrambleAcc);
+
 }
 
 /*
  * It's important for performance to pass @p secretLen (when it's static)
  * to the compiler, so that it can properly optimize the vectorized loop.
  *
- * When the secret size is unknown, or on GCC 12 where the mix of NO_INLINE and FORCE_INLINE
- * breaks -Og, this is XXH_NO_INLINE.
+ * When the secret size is unknown, or on GCC 12 where the mix of NO_INLINE and
+ * FORCE_INLINE breaks -Og, this is XXH_NO_INLINE.
  */
-XXH3_WITH_SECRET_INLINE XXH128_hash_t
-XXH3_hashLong_128b_withSecret(const void* XXH_RESTRICT input, size_t len,
-                              XXH64_hash_t seed64,
-                              const void* XXH_RESTRICT secret, size_t secretLen)
-{
-    (void)seed64;
-    return XXH3_hashLong_128b_internal(input, len, (const xxh_u8*)secret, secretLen,
-                                       XXH3_accumulate, XXH3_scrambleAcc);
+XXH3_WITH_SECRET_INLINE XXH128_hash_t XXH3_hashLong_128b_withSecret(
+    const void *XXH_RESTRICT input, size_t len, XXH64_hash_t seed64,
+    const void *XXH_RESTRICT secret, size_t secretLen) {
+
+  (void)seed64;
+  return XXH3_hashLong_128b_internal(input, len, (const xxh_u8 *)secret,
+                                     secretLen, XXH3_accumulate,
+                                     XXH3_scrambleAcc);
+
 }
 
-XXH_FORCE_INLINE XXH128_hash_t
-XXH3_hashLong_128b_withSeed_internal(const void* XXH_RESTRICT input, size_t len,
-                                XXH64_hash_t seed64,
-                                XXH3_f_accumulate f_acc,
-                                XXH3_f_scrambleAcc f_scramble,
-                                XXH3_f_initCustomSecret f_initSec)
-{
-    if (seed64 == 0)
-        return XXH3_hashLong_128b_internal(input, len,
-                                           XXH3_kSecret, sizeof(XXH3_kSecret),
-                                           f_acc, f_scramble);
-    {   XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
-        f_initSec(secret, seed64);
-        return XXH3_hashLong_128b_internal(input, len, (const xxh_u8*)secret, sizeof(secret),
-                                           f_acc, f_scramble);
-    }
+XXH_FORCE_INLINE XXH128_hash_t XXH3_hashLong_128b_withSeed_internal(
+    const void *XXH_RESTRICT input, size_t len, XXH64_hash_t seed64,
+    XXH3_f_accumulate f_acc, XXH3_f_scrambleAcc f_scramble,
+    XXH3_f_initCustomSecret f_initSec) {
+
+  if (seed64 == 0)
+    return XXH3_hashLong_128b_internal(input, len, XXH3_kSecret,
+                                       sizeof(XXH3_kSecret), f_acc, f_scramble);
+  {
+
+    XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
+    f_initSec(secret, seed64);
+    return XXH3_hashLong_128b_internal(input, len, (const xxh_u8 *)secret,
+                                       sizeof(secret), f_acc, f_scramble);
+
+  }
+
 }
 
 /*
  * It's important for performance that XXH3_hashLong is not inlined.
  */
 XXH_NO_INLINE XXH128_hash_t
-XXH3_hashLong_128b_withSeed(const void* input, size_t len,
-                            XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen)
-{
-    (void)secret; (void)secretLen;
-    return XXH3_hashLong_128b_withSeed_internal(input, len, seed64,
-                XXH3_accumulate, XXH3_scrambleAcc, XXH3_initCustomSecret);
+XXH3_hashLong_128b_withSeed(const void *input, size_t len, XXH64_hash_t seed64,
+                            const void *XXH_RESTRICT secret, size_t secretLen) {
+
+  (void)secret;
+  (void)secretLen;
+  return XXH3_hashLong_128b_withSeed_internal(input, len, seed64,
+                                              XXH3_accumulate, XXH3_scrambleAcc,
+                                              XXH3_initCustomSecret);
+
 }
 
-typedef XXH128_hash_t (*XXH3_hashLong128_f)(const void* XXH_RESTRICT, size_t,
-                                            XXH64_hash_t, const void* XXH_RESTRICT, size_t);
+typedef XXH128_hash_t (*XXH3_hashLong128_f)(const void *XXH_RESTRICT, size_t,
+                                            XXH64_hash_t,
+                                            const void *XXH_RESTRICT, size_t);
 
 XXH_FORCE_INLINE XXH128_hash_t
-XXH3_128bits_internal(const void* input, size_t len,
-                      XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen,
-                      XXH3_hashLong128_f f_hl128)
-{
-    XXH_ASSERT(secretLen >= XXH3_SECRET_SIZE_MIN);
-    /*
-     * If an action is to be taken if `secret` conditions are not respected,
-     * it should be done here.
-     * For now, it's a contract pre-condition.
-     * Adding a check and a branch here would cost performance at every hash.
-     */
-    if (len <= 16)
-        return XXH3_len_0to16_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, seed64);
-    if (len <= 128)
-        return XXH3_len_17to128_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
-    if (len <= XXH3_MIDSIZE_MAX)
-        return XXH3_len_129to240_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
-    return f_hl128(input, len, seed64, secret, secretLen);
-}
+XXH3_128bits_internal(const void *input, size_t len, XXH64_hash_t seed64,
+                      const void *XXH_RESTRICT secret, size_t secretLen,
+                      XXH3_hashLong128_f f_hl128) {
+
+  XXH_ASSERT(secretLen >= XXH3_SECRET_SIZE_MIN);
+  /*
+   * If an action is to be taken if `secret` conditions are not respected,
+   * it should be done here.
+   * For now, it's a contract pre-condition.
+   * Adding a check and a branch here would cost performance at every hash.
+   */
+  if (len <= 16)
+    return XXH3_len_0to16_128b((const xxh_u8 *)input, len,
+                               (const xxh_u8 *)secret, seed64);
+  if (len <= 128)
+    return XXH3_len_17to128_128b((const xxh_u8 *)input, len,
+                                 (const xxh_u8 *)secret, secretLen, seed64);
+  if (len <= XXH3_MIDSIZE_MAX)
+    return XXH3_len_129to240_128b((const xxh_u8 *)input, len,
+                                  (const xxh_u8 *)secret, secretLen, seed64);
+  return f_hl128(input, len, seed64, secret, secretLen);
 
+}
 
 /* ===   Public XXH128 API   === */
 
 /*! @ingroup XXH3_family */
-XXH_PUBLIC_API XXH128_hash_t XXH3_128bits(XXH_NOESCAPE const void* input, size_t len)
-{
-    return XXH3_128bits_internal(input, len, 0,
-                                 XXH3_kSecret, sizeof(XXH3_kSecret),
-                                 XXH3_hashLong_128b_default);
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits(XXH_NOESCAPE const void *input,
+                                          size_t                   len) {
+
+  return XXH3_128bits_internal(input, len, 0, XXH3_kSecret,
+                               sizeof(XXH3_kSecret),
+                               XXH3_hashLong_128b_default);
+
 }
 
 /*! @ingroup XXH3_family */
 XXH_PUBLIC_API XXH128_hash_t
-XXH3_128bits_withSecret(XXH_NOESCAPE const void* input, size_t len, XXH_NOESCAPE const void* secret, size_t secretSize)
-{
-    return XXH3_128bits_internal(input, len, 0,
-                                 (const xxh_u8*)secret, secretSize,
-                                 XXH3_hashLong_128b_withSecret);
+XXH3_128bits_withSecret(XXH_NOESCAPE const void *input, size_t len,
+                        XXH_NOESCAPE const void *secret, size_t secretSize) {
+
+  return XXH3_128bits_internal(input, len, 0, (const xxh_u8 *)secret,
+                               secretSize, XXH3_hashLong_128b_withSecret);
+
 }
 
 /*! @ingroup XXH3_family */
-XXH_PUBLIC_API XXH128_hash_t
-XXH3_128bits_withSeed(XXH_NOESCAPE const void* input, size_t len, XXH64_hash_t seed)
-{
-    return XXH3_128bits_internal(input, len, seed,
-                                 XXH3_kSecret, sizeof(XXH3_kSecret),
-                                 XXH3_hashLong_128b_withSeed);
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_withSeed(
+    XXH_NOESCAPE const void *input, size_t len, XXH64_hash_t seed) {
+
+  return XXH3_128bits_internal(input, len, seed, XXH3_kSecret,
+                               sizeof(XXH3_kSecret),
+                               XXH3_hashLong_128b_withSeed);
+
 }
 
 /*! @ingroup XXH3_family */
-XXH_PUBLIC_API XXH128_hash_t
-XXH3_128bits_withSecretandSeed(XXH_NOESCAPE const void* input, size_t len, XXH_NOESCAPE const void* secret, size_t secretSize, XXH64_hash_t seed)
-{
-    if (len <= XXH3_MIDSIZE_MAX)
-        return XXH3_128bits_internal(input, len, seed, XXH3_kSecret, sizeof(XXH3_kSecret), NULL);
-    return XXH3_hashLong_128b_withSecret(input, len, seed, secret, secretSize);
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_withSecretandSeed(
+    XXH_NOESCAPE const void *input, size_t len, XXH_NOESCAPE const void *secret,
+    size_t secretSize, XXH64_hash_t seed) {
+
+  if (len <= XXH3_MIDSIZE_MAX)
+    return XXH3_128bits_internal(input, len, seed, XXH3_kSecret,
+                                 sizeof(XXH3_kSecret), NULL);
+  return XXH3_hashLong_128b_withSecret(input, len, seed, secret, secretSize);
+
 }
 
 /*! @ingroup XXH3_family */
-XXH_PUBLIC_API XXH128_hash_t
-XXH128(XXH_NOESCAPE const void* input, size_t len, XXH64_hash_t seed)
-{
-    return XXH3_128bits_withSeed(input, len, seed);
-}
+XXH_PUBLIC_API XXH128_hash_t XXH128(XXH_NOESCAPE const void *input, size_t len,
+                                    XXH64_hash_t seed) {
 
+  return XXH3_128bits_withSeed(input, len, seed);
+
+}
 
-/* ===   XXH3 128-bit streaming   === */
-#ifndef XXH_NO_STREAM
+      /* ===   XXH3 128-bit streaming   === */
+      #ifndef XXH_NO_STREAM
 /*
- * All initialization and update functions are identical to 64-bit streaming variant.
- * The only difference is the finalization routine.
+ * All initialization and update functions are identical to 64-bit streaming
+ * variant. The only difference is the finalization routine.
  */
 
 /*! @ingroup XXH3_family */
 XXH_PUBLIC_API XXH_errorcode
-XXH3_128bits_reset(XXH_NOESCAPE XXH3_state_t* statePtr)
-{
-    return XXH3_64bits_reset(statePtr);
+XXH3_128bits_reset(XXH_NOESCAPE XXH3_state_t *statePtr) {
+
+  return XXH3_64bits_reset(statePtr);
+
 }
 
 /*! @ingroup XXH3_family */
-XXH_PUBLIC_API XXH_errorcode
-XXH3_128bits_reset_withSecret(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize)
-{
-    return XXH3_64bits_reset_withSecret(statePtr, secret, secretSize);
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSecret(
+    XXH_NOESCAPE XXH3_state_t *statePtr, XXH_NOESCAPE const void *secret,
+    size_t secretSize) {
+
+  return XXH3_64bits_reset_withSecret(statePtr, secret, secretSize);
+
 }
 
 /*! @ingroup XXH3_family */
-XXH_PUBLIC_API XXH_errorcode
-XXH3_128bits_reset_withSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH64_hash_t seed)
-{
-    return XXH3_64bits_reset_withSeed(statePtr, seed);
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSeed(
+    XXH_NOESCAPE XXH3_state_t *statePtr, XXH64_hash_t seed) {
+
+  return XXH3_64bits_reset_withSeed(statePtr, seed);
+
 }
 
 /*! @ingroup XXH3_family */
-XXH_PUBLIC_API XXH_errorcode
-XXH3_128bits_reset_withSecretandSeed(XXH_NOESCAPE XXH3_state_t* statePtr, XXH_NOESCAPE const void* secret, size_t secretSize, XXH64_hash_t seed)
-{
-    return XXH3_64bits_reset_withSecretandSeed(statePtr, secret, secretSize, seed);
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSecretandSeed(
+    XXH_NOESCAPE XXH3_state_t *statePtr, XXH_NOESCAPE const void *secret,
+    size_t secretSize, XXH64_hash_t seed) {
+
+  return XXH3_64bits_reset_withSecretandSeed(statePtr, secret, secretSize,
+                                             seed);
+
 }
 
 /*! @ingroup XXH3_family */
 XXH_PUBLIC_API XXH_errorcode
-XXH3_128bits_update(XXH_NOESCAPE XXH3_state_t* state, XXH_NOESCAPE const void* input, size_t len)
-{
-    return XXH3_64bits_update(state, input, len);
+XXH3_128bits_update(XXH_NOESCAPE XXH3_state_t *state,
+                    XXH_NOESCAPE const void *input, size_t len) {
+
+  return XXH3_64bits_update(state, input, len);
+
 }
 
 /*! @ingroup XXH3_family */
-XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_digest (XXH_NOESCAPE const XXH3_state_t* state)
-{
-    const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret;
-    if (state->totalLen > XXH3_MIDSIZE_MAX) {
-        XXH_ALIGN(XXH_ACC_ALIGN) XXH64_hash_t acc[XXH_ACC_NB];
-        XXH3_digest_long(acc, state, secret);
-        XXH_ASSERT(state->secretLimit + XXH_STRIPE_LEN >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
-        {   XXH128_hash_t h128;
-            h128.low64  = XXH3_mergeAccs(acc,
-                                         secret + XXH_SECRET_MERGEACCS_START,
-                                         (xxh_u64)state->totalLen * XXH_PRIME64_1);
-            h128.high64 = XXH3_mergeAccs(acc,
-                                         secret + state->secretLimit + XXH_STRIPE_LEN
-                                                - sizeof(acc) - XXH_SECRET_MERGEACCS_START,
-                                         ~((xxh_u64)state->totalLen * XXH_PRIME64_2));
-            return h128;
-        }
+XXH_PUBLIC_API XXH128_hash_t
+XXH3_128bits_digest(XXH_NOESCAPE const XXH3_state_t *state) {
+
+  const unsigned char *const secret =
+      (state->extSecret == NULL) ? state->customSecret : state->extSecret;
+  if (state->totalLen > XXH3_MIDSIZE_MAX) {
+
+    XXH_ALIGN(XXH_ACC_ALIGN) XXH64_hash_t acc[XXH_ACC_NB];
+    XXH3_digest_long(acc, state, secret);
+    XXH_ASSERT(state->secretLimit + XXH_STRIPE_LEN >=
+               sizeof(acc) + XXH_SECRET_MERGEACCS_START);
+    {
+
+      XXH128_hash_t h128;
+      h128.low64 = XXH3_mergeAccs(acc, secret + XXH_SECRET_MERGEACCS_START,
+                                  (xxh_u64)state->totalLen * XXH_PRIME64_1);
+      h128.high64 =
+          XXH3_mergeAccs(acc,
+                         secret + state->secretLimit + XXH_STRIPE_LEN -
+                             sizeof(acc) - XXH_SECRET_MERGEACCS_START,
+                         ~((xxh_u64)state->totalLen * XXH_PRIME64_2));
+      return h128;
+
     }
-    /* len <= XXH3_MIDSIZE_MAX : short code */
-    if (state->seed)
-        return XXH3_128bits_withSeed(state->buffer, (size_t)state->totalLen, state->seed);
-    return XXH3_128bits_withSecret(state->buffer, (size_t)(state->totalLen),
-                                   secret, state->secretLimit + XXH_STRIPE_LEN);
+
+  }
+
+  /* len <= XXH3_MIDSIZE_MAX : short code */
+  if (state->seed)
+    return XXH3_128bits_withSeed(state->buffer, (size_t)state->totalLen,
+                                 state->seed);
+  return XXH3_128bits_withSecret(state->buffer, (size_t)(state->totalLen),
+                                 secret, state->secretLimit + XXH_STRIPE_LEN);
+
 }
-#endif /* !XXH_NO_STREAM */
-/* 128-bit utility functions */
 
-#include <string.h>   /* memcmp, memcpy */
+      #endif                                              /* !XXH_NO_STREAM */
+    /* 128-bit utility functions */
+
+      #include <string.h>                                 /* memcmp, memcpy */
 
 /* return : 1 is equal, 0 if different */
 /*! @ingroup XXH3_family */
-XXH_PUBLIC_API int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2)
-{
-    /* note : XXH128_hash_t is compact, it has no padding byte */
-    return !(memcmp(&h1, &h2, sizeof(h1)));
+XXH_PUBLIC_API int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2) {
+
+  /* note : XXH128_hash_t is compact, it has no padding byte */
+  return !(memcmp(&h1, &h2, sizeof(h1)));
+
 }
 
 /* This prototype is compatible with stdlib's qsort().
@@ -6920,129 +7872,156 @@ XXH_PUBLIC_API int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2)
  *           <0 if *h128_1  < *h128_2
  *           =0 if *h128_1 == *h128_2  */
 /*! @ingroup XXH3_family */
-XXH_PUBLIC_API int XXH128_cmp(XXH_NOESCAPE const void* h128_1, XXH_NOESCAPE const void* h128_2)
-{
-    XXH128_hash_t const h1 = *(const XXH128_hash_t*)h128_1;
-    XXH128_hash_t const h2 = *(const XXH128_hash_t*)h128_2;
-    int const hcmp = (h1.high64 > h2.high64) - (h2.high64 > h1.high64);
-    /* note : bets that, in most cases, hash values are different */
-    if (hcmp) return hcmp;
-    return (h1.low64 > h2.low64) - (h2.low64 > h1.low64);
-}
+XXH_PUBLIC_API int XXH128_cmp(XXH_NOESCAPE const void *h128_1,
+                              XXH_NOESCAPE const void *h128_2) {
+
+  XXH128_hash_t const h1 = *(const XXH128_hash_t *)h128_1;
+  XXH128_hash_t const h2 = *(const XXH128_hash_t *)h128_2;
+  int const           hcmp = (h1.high64 > h2.high64) - (h2.high64 > h1.high64);
+  /* note : bets that, in most cases, hash values are different */
+  if (hcmp) return hcmp;
+  return (h1.low64 > h2.low64) - (h2.low64 > h1.low64);
 
+}
 
 /*======   Canonical representation   ======*/
 /*! @ingroup XXH3_family */
-XXH_PUBLIC_API void
-XXH128_canonicalFromHash(XXH_NOESCAPE XXH128_canonical_t* dst, XXH128_hash_t hash)
-{
-    XXH_STATIC_ASSERT(sizeof(XXH128_canonical_t) == sizeof(XXH128_hash_t));
-    if (XXH_CPU_LITTLE_ENDIAN) {
-        hash.high64 = XXH_swap64(hash.high64);
-        hash.low64  = XXH_swap64(hash.low64);
-    }
-    XXH_memcpy(dst, &hash.high64, sizeof(hash.high64));
-    XXH_memcpy((char*)dst + sizeof(hash.high64), &hash.low64, sizeof(hash.low64));
+XXH_PUBLIC_API void XXH128_canonicalFromHash(
+    XXH_NOESCAPE XXH128_canonical_t *dst, XXH128_hash_t hash) {
+
+  XXH_STATIC_ASSERT(sizeof(XXH128_canonical_t) == sizeof(XXH128_hash_t));
+  if (XXH_CPU_LITTLE_ENDIAN) {
+
+    hash.high64 = XXH_swap64(hash.high64);
+    hash.low64 = XXH_swap64(hash.low64);
+
+  }
+
+  XXH_memcpy(dst, &hash.high64, sizeof(hash.high64));
+  XXH_memcpy((char *)dst + sizeof(hash.high64), &hash.low64,
+             sizeof(hash.low64));
+
 }
 
 /*! @ingroup XXH3_family */
 XXH_PUBLIC_API XXH128_hash_t
-XXH128_hashFromCanonical(XXH_NOESCAPE const XXH128_canonical_t* src)
-{
-    XXH128_hash_t h;
-    h.high64 = XXH_readBE64(src);
-    h.low64  = XXH_readBE64(src->digest + 8);
-    return h;
+XXH128_hashFromCanonical(XXH_NOESCAPE const XXH128_canonical_t *src) {
+
+  XXH128_hash_t h;
+  h.high64 = XXH_readBE64(src);
+  h.low64 = XXH_readBE64(src->digest + 8);
+  return h;
+
 }
 
+      /* ==========================================
+       * Secret generators
+       * ==========================================
+       */
+      #define XXH_MIN(x, y) (((x) > (y)) ? (y) : (x))
 
+XXH_FORCE_INLINE void XXH3_combine16(void *dst, XXH128_hash_t h128) {
 
-/* ==========================================
- * Secret generators
- * ==========================================
- */
-#define XXH_MIN(x, y) (((x) > (y)) ? (y) : (x))
+  XXH_writeLE64(dst, XXH_readLE64(dst) ^ h128.low64);
+  XXH_writeLE64((char *)dst + 8, XXH_readLE64((char *)dst + 8) ^ h128.high64);
 
-XXH_FORCE_INLINE void XXH3_combine16(void* dst, XXH128_hash_t h128)
-{
-    XXH_writeLE64( dst, XXH_readLE64(dst) ^ h128.low64 );
-    XXH_writeLE64( (char*)dst+8, XXH_readLE64((char*)dst+8) ^ h128.high64 );
 }
 
 /*! @ingroup XXH3_family */
-XXH_PUBLIC_API XXH_errorcode
-XXH3_generateSecret(XXH_NOESCAPE void* secretBuffer, size_t secretSize, XXH_NOESCAPE const void* customSeed, size_t customSeedSize)
-{
-#if (XXH_DEBUGLEVEL >= 1)
-    XXH_ASSERT(secretBuffer != NULL);
-    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
-#else
-    /* production mode, assert() are disabled */
-    if (secretBuffer == NULL) return XXH_ERROR;
-    if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
-#endif
+XXH_PUBLIC_API XXH_errorcode XXH3_generateSecret(
+    XXH_NOESCAPE void *secretBuffer, size_t secretSize,
+    XXH_NOESCAPE const void *customSeed, size_t customSeedSize) {
+
+      #if (XXH_DEBUGLEVEL >= 1)
+  XXH_ASSERT(secretBuffer != NULL);
+  XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
+      #else
+  /* production mode, assert() are disabled */
+  if (secretBuffer == NULL) return XXH_ERROR;
+  if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
+      #endif
+
+  if (customSeedSize == 0) {
+
+    customSeed = XXH3_kSecret;
+    customSeedSize = XXH_SECRET_DEFAULT_SIZE;
+
+  }
+
+      #if (XXH_DEBUGLEVEL >= 1)
+  XXH_ASSERT(customSeed != NULL);
+      #else
+  if (customSeed == NULL) return XXH_ERROR;
+      #endif
+
+  /* Fill secretBuffer with a copy of customSeed - repeat as needed */
+  {
+
+    size_t pos = 0;
+    while (pos < secretSize) {
+
+      size_t const toCopy = XXH_MIN((secretSize - pos), customSeedSize);
+      memcpy((char *)secretBuffer + pos, customSeed, toCopy);
+      pos += toCopy;
 
-    if (customSeedSize == 0) {
-        customSeed = XXH3_kSecret;
-        customSeedSize = XXH_SECRET_DEFAULT_SIZE;
     }
-#if (XXH_DEBUGLEVEL >= 1)
-    XXH_ASSERT(customSeed != NULL);
-#else
-    if (customSeed == NULL) return XXH_ERROR;
-#endif
 
-    /* Fill secretBuffer with a copy of customSeed - repeat as needed */
-    {   size_t pos = 0;
-        while (pos < secretSize) {
-            size_t const toCopy = XXH_MIN((secretSize - pos), customSeedSize);
-            memcpy((char*)secretBuffer + pos, customSeed, toCopy);
-            pos += toCopy;
-    }   }
-
-    {   size_t const nbSeg16 = secretSize / 16;
-        size_t n;
-        XXH128_canonical_t scrambler;
-        XXH128_canonicalFromHash(&scrambler, XXH128(customSeed, customSeedSize, 0));
-        for (n=0; n<nbSeg16; n++) {
-            XXH128_hash_t const h128 = XXH128(&scrambler, sizeof(scrambler), n);
-            XXH3_combine16((char*)secretBuffer + n*16, h128);
-        }
-        /* last segment */
-        XXH3_combine16((char*)secretBuffer + secretSize - 16, XXH128_hashFromCanonical(&scrambler));
+  }
+
+  {
+
+    size_t const       nbSeg16 = secretSize / 16;
+    size_t             n;
+    XXH128_canonical_t scrambler;
+    XXH128_canonicalFromHash(&scrambler, XXH128(customSeed, customSeedSize, 0));
+    for (n = 0; n < nbSeg16; n++) {
+
+      XXH128_hash_t const h128 = XXH128(&scrambler, sizeof(scrambler), n);
+      XXH3_combine16((char *)secretBuffer + n * 16, h128);
+
     }
-    return XXH_OK;
+
+    /* last segment */
+    XXH3_combine16((char *)secretBuffer + secretSize - 16,
+                   XXH128_hashFromCanonical(&scrambler));
+
+  }
+
+  return XXH_OK;
+
 }
 
 /*! @ingroup XXH3_family */
-XXH_PUBLIC_API void
-XXH3_generateSecret_fromSeed(XXH_NOESCAPE void* secretBuffer, XXH64_hash_t seed)
-{
-    XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
-    XXH3_initCustomSecret(secret, seed);
-    XXH_ASSERT(secretBuffer != NULL);
-    memcpy(secretBuffer, secret, XXH_SECRET_DEFAULT_SIZE);
-}
+XXH_PUBLIC_API void XXH3_generateSecret_fromSeed(
+    XXH_NOESCAPE void *secretBuffer, XXH64_hash_t seed) {
 
+  XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
+  XXH3_initCustomSecret(secret, seed);
+  XXH_ASSERT(secretBuffer != NULL);
+  memcpy(secretBuffer, secret, XXH_SECRET_DEFAULT_SIZE);
 
+}
 
-/* Pop our optimization override from above */
-#if XXH_VECTOR == XXH_AVX2 /* AVX2 */ \
-  && defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
-  && defined(__OPTIMIZE__) && XXH_SIZE_OPT <= 0 /* respect -O0 and -Os */
-#  pragma GCC pop_options
-#endif
+      /* Pop our optimization override from above */
+      #if XXH_VECTOR == XXH_AVX2                      /* AVX2 */           \
+          && defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
+          && defined(__OPTIMIZE__) &&                                      \
+          XXH_SIZE_OPT <= 0                          /* respect -O0 and -Os */
+        #pragma GCC pop_options
+      #endif
 
-#endif  /* XXH_NO_LONG_LONG */
+    #endif                                              /* XXH_NO_LONG_LONG */
 
-#endif  /* XXH_NO_XXH3 */
+  #endif                                                     /* XXH_NO_XXH3 */
 
 /*!
  * @}
  */
-#endif  /* XXH_IMPLEMENTATION */
+#endif                                                /* XXH_IMPLEMENTATION */
 
+#if defined(__cplusplus)
+
+}                                                             /* extern "C" */
 
-#if defined (__cplusplus)
-} /* extern "C" */
 #endif
+
diff --git a/src/afl-fuzz-redqueen.c b/src/afl-fuzz-redqueen.c
index 3342445a..be41d6c4 100644
--- a/src/afl-fuzz-redqueen.c
+++ b/src/afl-fuzz-redqueen.c
@@ -1967,6 +1967,7 @@ static u8 cmp_fuzz(afl_state_t *afl, u32 key, u8 *orig_buf, u8 *buf, u8 *cbuf,
       continue;
 
     }
+
 #endif
 
 #ifdef _DEBUG
@@ -2789,6 +2790,7 @@ static u8 rtn_fuzz(afl_state_t *afl, u32 key, u8 *orig_buf, u8 *buf, u8 *cbuf,
       continue;
 
     }
+
 #endif
 
     t = taint;
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index 7c47f060..9c89b2a1 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -402,7 +402,9 @@ static void usage(u8 *argv0, int more_help) {
 #endif
 
 #ifdef _AFL_SPECIAL_PERFORMANCE
-  SAYF("Compiled with special performance options for this specific system, it might not work on other platforms!\n");
+  SAYF(
+      "Compiled with special performance options for this specific system, it "
+      "might not work on other platforms!\n");
 #endif
 
   SAYF("For additional help please consult %s/README.md :)\n\n", doc_path);
diff --git a/src/afl-performance.c b/src/afl-performance.c
index 22cf4dec..f730ca53 100644
--- a/src/afl-performance.c
+++ b/src/afl-performance.c
@@ -3,18 +3,17 @@
 #include "types.h"
 
 #ifdef _HAVE_AVX2
-#define T1HA0_AESNI_AVAILABLE 1
-#define T1HA_USE_FAST_ONESHOT_READ 1
-#define T1HA_USE_INDIRECT_FUNCTIONS 1
-#define T1HA_IA32AES_NAME XXH3_64bits
-#include "t1ha0_ia32aes_b.h"
+  #define T1HA0_AESNI_AVAILABLE 1
+  #define T1HA_USE_FAST_ONESHOT_READ 1
+  #define T1HA_USE_INDIRECT_FUNCTIONS 1
+  #define T1HA_IA32AES_NAME XXH3_64bits
+  #include "t1ha0_ia32aes_b.h"
 #else
-#define XXH_INLINE_ALL
-#include "xxhash.h"
-#undef XXH_INLINE_ALL
+  #define XXH_INLINE_ALL
+  #include "xxhash.h"
+  #undef XXH_INLINE_ALL
 #endif
 
-
 void rand_set_seed(afl_state_t *afl, s64 init_seed) {
 
   afl->init_seed = init_seed;
diff --git a/utils/bench/hash.c b/utils/bench/hash.c
index 013a5321..d4be0ab4 100644
--- a/utils/bench/hash.c
+++ b/utils/bench/hash.c
@@ -13,30 +13,41 @@
 #undef XXH_INLINE_ALL
 
 int main() {
-  char *data = malloc(4097);
+
+  char           *data = malloc(4097);
   struct timespec start, end;
-  long long duration;
-  int i;
-  uint64_t res;
+  long long       duration;
+  int             i;
+  uint64_t        res;
 
   clock_gettime(CLOCK_MONOTONIC, &start);
   for (i = 0; i < 100000000; ++i) {
-     res = XXH3_64bits(data, 4097);
-     memcpy(data + 16, (char*)&res, 8);
+
+    res = XXH3_64bits(data, 4097);
+    memcpy(data + 16, (char *)&res, 8);
+
   }
+
   clock_gettime(CLOCK_MONOTONIC, &end);
-  duration = (end.tv_sec - start.tv_sec) * 1000000000LL + (end.tv_nsec - start.tv_nsec);
+  duration = (end.tv_sec - start.tv_sec) * 1000000000LL +
+             (end.tv_nsec - start.tv_nsec);
   printf("xxh3 duration:          %lld ns\n", duration);
 
   memset(data, 0, 4097);
   clock_gettime(CLOCK_MONOTONIC, &start);
   for (i = 0; i < 100000000; ++i) {
-     res = t1ha0_ia32aes(data, 4097);
-     memcpy(data + 16, (char*)&res, 8);
+
+    res = t1ha0_ia32aes(data, 4097);
+    memcpy(data + 16, (char *)&res, 8);
+
   }
+
   clock_gettime(CLOCK_MONOTONIC, &end);
-  duration = (end.tv_sec - start.tv_sec) * 1000000000LL + (end.tv_nsec - start.tv_nsec);
+  duration = (end.tv_sec - start.tv_sec) * 1000000000LL +
+             (end.tv_nsec - start.tv_nsec);
   printf("t1ha0_ia32aes duration: %lld ns\n", duration);
 
   return 0;
+
 }
+
-- 
cgit 1.4.1


From 036a79268b48a0e3e061d5e3387711f69bed8d56 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Thu, 29 Feb 2024 09:10:22 +0100
Subject: gcc cmplog fix

---
 docs/Changelog.md                         |  3 ++-
 include/t1ha_bits.h                       | 15 ++++++++-------
 include/xxhash.h                          | 30 ++++++++++++++++++------------
 instrumentation/afl-gcc-cmptrs-pass.so.cc | 15 +++++++++------
 src/afl-cc.c                              |  3 ++-
 src/afl-fuzz.c                            | 14 ++++++--------
 6 files changed, 45 insertions(+), 35 deletions(-)

(limited to 'include')

diff --git a/docs/Changelog.md b/docs/Changelog.md
index ead015eb..da4b3a20 100644
--- a/docs/Changelog.md
+++ b/docs/Changelog.md
@@ -27,8 +27,9 @@
     - added collision free caller instrumentation to LTO mode. activate with
       `AFL_LLVM_LTO_CALLER=1`. You can set a max depth to go through single
       block functions with `AFL_LLVM_LTO_CALLER_DEPTH` (default 0)
+    - fix for GCC_PLUGIN cmplog that broke on std::strings
   - afl-whatsup:
-    - Now also displays current average speed
+    - now also displays current average speed
     - small bugfixes
   - Minor edits to afl-persistent-config
   - Prevent temporary files being left behind on aborted afl-whatsup
diff --git a/include/t1ha_bits.h b/include/t1ha_bits.h
index e7a8d53c..0b9bbda5 100644
--- a/include/t1ha_bits.h
+++ b/include/t1ha_bits.h
@@ -207,7 +207,7 @@ static __maybe_unused __always_inline unsigned e2k_add64carry_first(
   return (unsigned)__builtin_e2k_addcd_c(base, addend, 0);
 
 }
-\
+
       #define add64carry_first(base, addend, sum) \
         e2k_add64carry_first(base, addend, sum)
 
@@ -218,7 +218,7 @@ static __maybe_unused __always_inline unsigned e2k_add64carry_next(
   return (unsigned)__builtin_e2k_addcd_c(base, addend, carry);
 
 }
-\
+
       #define add64carry_next(carry, base, addend, sum) \
         e2k_add64carry_next(carry, base, addend, sum)
 
@@ -230,7 +230,7 @@ static __maybe_unused __always_inline void e2k_add64carry_last(unsigned  carry,
   *sum = __builtin_e2k_addcd(base, addend, carry);
 
 }
-\
+
       #define add64carry_last(carry, base, addend, sum) \
         e2k_add64carry_last(carry, base, addend, sum)
     #endif                                                 /* __iset__ >= 5 */
@@ -311,7 +311,7 @@ static __forceinline char msvc32_add64carry_first(uint64_t  base,
                          base_32h, addend_32h, sum32 + 1);
 
 }
-\
+
       #define add64carry_first(base, addend, sum) \
         msvc32_add64carry_first(base, addend, sum)
 
@@ -328,7 +328,7 @@ static __forceinline char msvc32_add64carry_next(char carry, uint64_t base,
                          base_32h, addend_32h, sum32 + 1);
 
 }
-\
+
       #define add64carry_next(carry, base, addend, sum) \
         msvc32_add64carry_next(carry, base, addend, sum)
 
@@ -345,7 +345,7 @@ static __forceinline void msvc32_add64carry_last(char carry, uint64_t base,
                   addend_32h, sum32 + 1);
 
 }
-\
+
       #define add64carry_last(carry, base, addend, sum) \
         msvc32_add64carry_last(carry, base, addend, sum)
     #endif                                    /* _MSC_FULL_VER >= 190024231 */
@@ -454,7 +454,7 @@ typedef struct {
   uint64_t unaligned_64;
 
 } __attribute__((__packed__)) t1ha_unaligned_proxy;
-\
+
     #define read_unaligned(ptr, bits)                                   \
       (((const t1ha_unaligned_proxy *)((const uint8_t *)(ptr)-offsetof( \
             t1ha_unaligned_proxy, unaligned_##bits)))                   \
@@ -539,6 +539,7 @@ static __always_inline const uint64_t *__attribute__((
         (void)(ptr);      \
                           \
       } while (0)
+
   #endif
 #endif                                                          /* prefetch */
 
diff --git a/include/xxhash.h b/include/xxhash.h
index 7697d0f2..991a8f1e 100644
--- a/include/xxhash.h
+++ b/include/xxhash.h
@@ -1734,7 +1734,7 @@ XXH128_hashFromCanonical(XXH_NOESCAPE const XXH128_canonical_t *src);
  * These declarations should only be used with static linking.
  * Never use them in association with dynamic linking!
  *****************************************************************************
-*/
+ */
 
 /*
  * These definitions are only present to allow static allocation
@@ -2399,9 +2399,9 @@ XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSecretandSeed(
     #define XXH_NO_STREAM
     #undef XXH_NO_STREAM                                  /* don't actually */
   #endif                                                     /* XXH_DOXYGEN */
-/*!
- * @}
- */
+                         /*!
+                          * @}
+                          */
 
   #ifndef XXH_FORCE_MEMORY_ACCESS /* can be defined externally, on command \
                                      line for example */
@@ -2614,6 +2614,7 @@ static void *XXH_memcpy(void *dest, const void *src, size_t size) {
           _Static_assert((c), m);                  \
                                                    \
         } while (0)
+
     #elif defined(__cplusplus) && (__cplusplus >= 201103L)         /* C++11 */
       #define XXH_STATIC_ASSERT_WITH_MESSAGE(c, m) \
         do {                                       \
@@ -2621,6 +2622,7 @@ static void *XXH_memcpy(void *dest, const void *src, size_t size) {
           static_assert((c), m);                   \
                                                    \
         } while (0)
+
     #else
       #define XXH_STATIC_ASSERT_WITH_MESSAGE(c, m) \
         do {                                       \
@@ -2632,6 +2634,7 @@ static void *XXH_memcpy(void *dest, const void *src, size_t size) {
           };                                       \
                                                    \
         } while (0)
+
     #endif
     #define XXH_STATIC_ASSERT(c) XXH_STATIC_ASSERT_WITH_MESSAGE((c), #c)
   #endif
@@ -2850,7 +2853,7 @@ static int XXH_isLittleEndian(void) {
   return one.c[0];
 
 }
-\
+
       #define XXH_CPU_LITTLE_ENDIAN XXH_isLittleEndian()
     #endif
   #endif
@@ -4679,6 +4682,7 @@ XXH_FORCE_INLINE xxh_u64x2 XXH_vec_mule(xxh_u32x4 a, xxh_u32x4 b) {
             acc = svadd_u64_x(mask, acc, mul);                               \
                                                                              \
           } while (0)
+
       #endif                                       /* XXH_VECTOR == XXH_SVE */
 
       /* prefetch
@@ -4737,12 +4741,14 @@ static const xxh_u8 XXH3_kSecret[XXH_SECRET_DEFAULT_SIZE] = {
 
 };
 
-static const xxh_u64 PRIME_MX1 = 0x165667919E3779F9ULL; /*!<
-                                                           0b0001011001010110011001111001000110011110001101110111100111111001
-                                                         */
-static const xxh_u64 PRIME_MX2 = 0x9FB21C651E98DF25ULL; /*!<
-                                                           0b1001111110110010000111000110010100011110100110001101111100100101
-                                                         */
+static const xxh_u64 PRIME_MX1 =
+    0x165667919E3779F9ULL; /*!<
+                              0b0001011001010110011001111001000110011110001101110111100111111001
+                            */
+static const xxh_u64 PRIME_MX2 =
+    0x9FB21C651E98DF25ULL; /*!<
+                              0b1001111110110010000111000110010100011110100110001101111100100101
+                            */
 
       #ifdef XXH_OLD_NAMES
         #define kSecret XXH3_kSecret
@@ -7854,7 +7860,7 @@ XXH3_128bits_digest(XXH_NOESCAPE const XXH3_state_t *state) {
 }
 
       #endif                                              /* !XXH_NO_STREAM */
-    /* 128-bit utility functions */
+             /* 128-bit utility functions */
 
       #include <string.h>                                 /* memcmp, memcpy */
 
diff --git a/instrumentation/afl-gcc-cmptrs-pass.so.cc b/instrumentation/afl-gcc-cmptrs-pass.so.cc
index 929a9d7a..96bd5ba8 100644
--- a/instrumentation/afl-gcc-cmptrs-pass.so.cc
+++ b/instrumentation/afl-gcc-cmptrs-pass.so.cc
@@ -180,19 +180,19 @@ struct afl_cmptrs_pass : afl_base_pass {
     c = DECL_CONTEXT(c);
     if (c && TREE_CODE(c) != TRANSLATION_UNIT_DECL) return false;
 
-    /* Check that the first nonstatic data member of the record type
+    /* Check that the first nonstatic named data member of the record type
        is named _M_dataplus.  */
     for (c = TYPE_FIELDS(t); c; c = DECL_CHAIN(c))
-      if (TREE_CODE(c) == FIELD_DECL) break;
+      if (TREE_CODE(c) == FIELD_DECL && DECL_NAME(c)) break;
     if (!c || !integer_zerop(DECL_FIELD_BIT_OFFSET(c)) ||
         strcmp(IDENTIFIER_POINTER(DECL_NAME(c)), "_M_dataplus") != 0)
       return false;
 
-    /* Check that the second nonstatic data member of the record type
+    /* Check that the second nonstatic named data member of the record type
        is named _M_string_length.  */
     tree f2;
     for (f2 = DECL_CHAIN(c); f2; f2 = DECL_CHAIN(f2))
-      if (TREE_CODE(f2) == FIELD_DECL) break;
+      if (TREE_CODE(f2) == FIELD_DECL && DECL_NAME(f2)) break;
     if (!f2                       /* No need to check this field's offset.  */
         || strcmp(IDENTIFIER_POINTER(DECL_NAME(f2)), "_M_string_length") != 0)
       return false;
@@ -208,9 +208,12 @@ struct afl_cmptrs_pass : afl_base_pass {
         strcmp(IDENTIFIER_POINTER(TYPE_IDENTIFIER(c)), "_Alloc_hider") != 0)
       return false;
 
-    /* And its first data member is named _M_p.  */
+    /* And its first nonstatic named data member should be named _M_p.
+       There may be (unnamed) subobjects from empty base classes.  We
+       skip the subobjects, then check the offset of the first data
+       member. */
     for (c = TYPE_FIELDS(c); c; c = DECL_CHAIN(c))
-      if (TREE_CODE(c) == FIELD_DECL) break;
+      if (TREE_CODE(c) == FIELD_DECL && DECL_NAME(c)) break;
     if (!c || !integer_zerop(DECL_FIELD_BIT_OFFSET(c)) ||
         strcmp(IDENTIFIER_POINTER(DECL_NAME(c)), "_M_p") != 0)
       return false;
diff --git a/src/afl-cc.c b/src/afl-cc.c
index 6aa0da6a..faa46103 100644
--- a/src/afl-cc.c
+++ b/src/afl-cc.c
@@ -828,7 +828,8 @@ static void instrument_mode_old_environ(aflcc_state_t *aflcc) {
   }
 
   if (getenv("AFL_LLVM_CTX")) aflcc->instrument_opt_mode |= INSTRUMENT_OPT_CTX;
-  if (getenv("AFL_LLVM_CALLER") || getenv("AFL_LLVM_LTO_CALLER") || getenv("AFL_LLVM_LTO_CTX"))
+  if (getenv("AFL_LLVM_CALLER") || getenv("AFL_LLVM_LTO_CALLER") ||
+      getenv("AFL_LLVM_LTO_CTX"))
     aflcc->instrument_opt_mode |= INSTRUMENT_OPT_CALLER;
 
   if (getenv("AFL_LLVM_NGRAM_SIZE")) {
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index 08f716fa..443d93b0 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -2493,17 +2493,15 @@ int main(int argc, char **argv_orig, char **envp) {
 
       for (entry = 0; entry < afl->queued_items; ++entry)
         if (!afl->queue_buf[entry]->disabled)
-          if ((afl->queue_buf[entry]->exec_us/1000) > max_ms)
-            max_ms = afl->queue_buf[entry]->exec_us/1000;
-      
+          if ((afl->queue_buf[entry]->exec_us / 1000) > max_ms)
+            max_ms = afl->queue_buf[entry]->exec_us / 1000;
+
       // Add 20% as a safety margin, capped to exec_tmout given in -t option
       max_ms *= 1.2;
-      if(max_ms > afl->fsrv.exec_tmout)
-        max_ms = afl->fsrv.exec_tmout;
-      
+      if (max_ms > afl->fsrv.exec_tmout) max_ms = afl->fsrv.exec_tmout;
+
       // Ensure that there is a sensible timeout even for very fast binaries
-      if(max_ms < 5)
-        max_ms = 5;
+      if (max_ms < 5) max_ms = 5;
 
       afl->fsrv.exec_tmout = max_ms;
       afl->timeout_given = 1;
-- 
cgit 1.4.1


From 52e19d35fac636f9ea4679d402b5eaabaa74aa0a Mon Sep 17 00:00:00 2001
From: "Christian Holler (:decoder)" <choller@mozilla.com>
Date: Wed, 6 Mar 2024 10:55:53 +0100
Subject: Add optional handling of Nyx InvalidWriteToPayload event

---
 include/envs.h       | 40 ++++++++++++++++++++--------------------
 src/afl-forkserver.c |  2 ++
 2 files changed, 22 insertions(+), 20 deletions(-)

(limited to 'include')

diff --git a/include/envs.h b/include/envs.h
index d32e2f92..56a4916c 100644
--- a/include/envs.h
+++ b/include/envs.h
@@ -95,26 +95,26 @@ static char *afl_environment_variables[] = {
     "AFL_MAX_DET_EXTRAS",
     "AFL_NO_X86",  // not really an env but we dont want to warn on it
     "AFL_NOOPT", "AFL_NYX_AUX_SIZE", "AFL_NYX_DISABLE_SNAPSHOT_MODE",
-    "AFL_NYX_LOG", "AFL_NYX_REUSE_SNAPSHOT", "AFL_PASSTHROUGH", "AFL_PATH",
-    "AFL_PERFORMANCE_FILE", "AFL_PERSISTENT_RECORD",
-    "AFL_POST_PROCESS_KEEP_ORIGINAL", "AFL_PRELOAD", "AFL_TARGET_ENV",
-    "AFL_PYTHON_MODULE", "AFL_QEMU_CUSTOM_BIN", "AFL_QEMU_COMPCOV",
-    "AFL_QEMU_COMPCOV_DEBUG", "AFL_QEMU_DEBUG_MAPS", "AFL_QEMU_DISABLE_CACHE",
-    "AFL_QEMU_DRIVER_NO_HOOK", "AFL_QEMU_FORCE_DFL", "AFL_QEMU_PERSISTENT_ADDR",
-    "AFL_QEMU_PERSISTENT_CNT", "AFL_QEMU_PERSISTENT_GPR",
-    "AFL_QEMU_PERSISTENT_HOOK", "AFL_QEMU_PERSISTENT_MEM",
-    "AFL_QEMU_PERSISTENT_RET", "AFL_QEMU_PERSISTENT_RETADDR_OFFSET",
-    "AFL_QEMU_PERSISTENT_EXITS", "AFL_QEMU_INST_RANGES",
-    "AFL_QEMU_EXCLUDE_RANGES", "AFL_QEMU_SNAPSHOT", "AFL_QEMU_TRACK_UNSTABLE",
-    "AFL_QUIET", "AFL_RANDOM_ALLOC_CANARY", "AFL_REAL_PATH",
-    "AFL_SHUFFLE_QUEUE", "AFL_SKIP_BIN_CHECK", "AFL_SKIP_CPUFREQ",
-    "AFL_SKIP_CRASHES", "AFL_SKIP_OSSFUZZ", "AFL_STATSD", "AFL_STATSD_HOST",
-    "AFL_STATSD_PORT", "AFL_STATSD_TAGS_FLAVOR", "AFL_SYNC_TIME",
-    "AFL_TESTCACHE_SIZE", "AFL_TESTCACHE_ENTRIES", "AFL_TMIN_EXACT",
-    "AFL_TMPDIR", "AFL_TOKEN_FILE", "AFL_TRACE_PC", "AFL_USE_ASAN",
-    "AFL_USE_MSAN", "AFL_USE_TRACE_PC", "AFL_USE_UBSAN", "AFL_USE_TSAN",
-    "AFL_USE_CFISAN", "AFL_USE_LSAN", "AFL_WINE_PATH", "AFL_NO_SNAPSHOT",
-    "AFL_EXPAND_HAVOC_NOW", "AFL_USE_FASAN", "AFL_USE_QASAN",
+    "AFL_NYX_HANDLE_INVALID_WRITE", "AFL_NYX_LOG", "AFL_NYX_REUSE_SNAPSHOT",
+    "AFL_PASSTHROUGH", "AFL_PATH", "AFL_PERFORMANCE_FILE",
+    "AFL_PERSISTENT_RECORD", "AFL_POST_PROCESS_KEEP_ORIGINAL", "AFL_PRELOAD",
+    "AFL_TARGET_ENV", "AFL_PYTHON_MODULE", "AFL_QEMU_CUSTOM_BIN",
+    "AFL_QEMU_COMPCOV", "AFL_QEMU_COMPCOV_DEBUG", "AFL_QEMU_DEBUG_MAPS",
+    "AFL_QEMU_DISABLE_CACHE", "AFL_QEMU_DRIVER_NO_HOOK", "AFL_QEMU_FORCE_DFL",
+    "AFL_QEMU_PERSISTENT_ADDR", "AFL_QEMU_PERSISTENT_CNT",
+    "AFL_QEMU_PERSISTENT_GPR", "AFL_QEMU_PERSISTENT_HOOK",
+    "AFL_QEMU_PERSISTENT_MEM", "AFL_QEMU_PERSISTENT_RET",
+    "AFL_QEMU_PERSISTENT_RETADDR_OFFSET", "AFL_QEMU_PERSISTENT_EXITS",
+    "AFL_QEMU_INST_RANGES", "AFL_QEMU_EXCLUDE_RANGES", "AFL_QEMU_SNAPSHOT",
+    "AFL_QEMU_TRACK_UNSTABLE", "AFL_QUIET", "AFL_RANDOM_ALLOC_CANARY",
+    "AFL_REAL_PATH", "AFL_SHUFFLE_QUEUE", "AFL_SKIP_BIN_CHECK",
+    "AFL_SKIP_CPUFREQ", "AFL_SKIP_CRASHES", "AFL_SKIP_OSSFUZZ", "AFL_STATSD",
+    "AFL_STATSD_HOST", "AFL_STATSD_PORT", "AFL_STATSD_TAGS_FLAVOR",
+    "AFL_SYNC_TIME", "AFL_TESTCACHE_SIZE", "AFL_TESTCACHE_ENTRIES",
+    "AFL_TMIN_EXACT", "AFL_TMPDIR", "AFL_TOKEN_FILE", "AFL_TRACE_PC",
+    "AFL_USE_ASAN", "AFL_USE_MSAN", "AFL_USE_TRACE_PC", "AFL_USE_UBSAN",
+    "AFL_USE_TSAN", "AFL_USE_CFISAN", "AFL_USE_LSAN", "AFL_WINE_PATH",
+    "AFL_NO_SNAPSHOT", "AFL_EXPAND_HAVOC_NOW", "AFL_USE_FASAN", "AFL_USE_QASAN",
     "AFL_PRINT_FILENAMES", "AFL_PIZZA_MODE", NULL
 
 };
diff --git a/src/afl-forkserver.c b/src/afl-forkserver.c
index 158651af..cf5c511e 100644
--- a/src/afl-forkserver.c
+++ b/src/afl-forkserver.c
@@ -1840,6 +1840,8 @@ afl_fsrv_run_target(afl_forkserver_t *fsrv, u32 timeout,
       case Timeout:
         return FSRV_RUN_TMOUT;
       case InvalidWriteToPayload:
+        if (!!getenv("AFL_NYX_HANDLE_INVALID_WRITE")) { return FSRV_RUN_CRASH; }
+
         /* ??? */
         FATAL("FixMe: Nyx InvalidWriteToPayload handler is missing");
         break;
-- 
cgit 1.4.1


From 2ed2ac80bc632f39fa95e1d51c4f14ecde48355a Mon Sep 17 00:00:00 2001
From: = <=>
Date: Mon, 11 Mar 2024 10:30:35 +0000
Subject: fix record compat loop to replay correct number of inputs, and at
 least one input

---
 include/afl-record-compat.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

(limited to 'include')

diff --git a/include/afl-record-compat.h b/include/afl-record-compat.h
index 2c79595d..677d47b0 100644
--- a/include/afl-record-compat.h
+++ b/include/afl-record-compat.h
@@ -24,7 +24,7 @@ unsigned char fuzz_buf[FUZZ_BUF_SIZE];
 
 int __afl_persistent_loop(unsigned int max_cnt) {
 
-  static unsigned int       cycle_cnt = 1;
+  static unsigned int       cycle_cnt = 2;
   static unsigned short int inited = 0;
   char                      tcase[PATH_MAX];
 
@@ -32,7 +32,7 @@ int __afl_persistent_loop(unsigned int max_cnt) {
 
     if (!inited) {
 
-      cycle_cnt = replay_record_cnt;
+      cycle_cnt = replay_record_cnt+1;
       inited = 1;
 
     }
-- 
cgit 1.4.1


From 08f6d59f505436f4aec70ddd341d27f1d7bd7ccf Mon Sep 17 00:00:00 2001
From: = <=>
Date: Mon, 11 Mar 2024 12:01:06 +0000
Subject: correct fix

---
 include/afl-record-compat.h | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

(limited to 'include')

diff --git a/include/afl-record-compat.h b/include/afl-record-compat.h
index 677d47b0..3e5d60e3 100644
--- a/include/afl-record-compat.h
+++ b/include/afl-record-compat.h
@@ -24,15 +24,15 @@ unsigned char fuzz_buf[FUZZ_BUF_SIZE];
 
 int __afl_persistent_loop(unsigned int max_cnt) {
 
-  static unsigned int       cycle_cnt = 2;
+  static unsigned int       cycle_cnt = 1;
   static unsigned short int inited = 0;
   char                      tcase[PATH_MAX];
 
-  if (is_replay_record) {
+  if (is_replay_record && cycle_cnt) {
 
     if (!inited) {
 
-      cycle_cnt = replay_record_cnt+1;
+      cycle_cnt = replay_record_cnt;
       inited = 1;
 
     }
@@ -59,7 +59,7 @@ int __afl_persistent_loop(unsigned int max_cnt) {
 
   }
 
-  return --cycle_cnt;
+  return cycle_cnt--;
 
 }
 
-- 
cgit 1.4.1


From f7ea0f569fa57e22548c1dc8eaba2903213e496e Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Fri, 5 Apr 2024 14:52:53 +0200
Subject: fix aflpp custom mutator + standalone tool

---
 custom_mutators/aflpp/aflpp.c                       | 1 +
 custom_mutators/aflpp/standalone/aflpp-standalone.c | 7 ++-----
 docs/Changelog.md                                   | 1 +
 include/afl-mutations.h                             | 5 ++++-
 src/afl-fuzz-state.c                                | 4 ----
 5 files changed, 8 insertions(+), 10 deletions(-)

(limited to 'include')

diff --git a/custom_mutators/aflpp/aflpp.c b/custom_mutators/aflpp/aflpp.c
index e15d0391..0b236f76 100644
--- a/custom_mutators/aflpp/aflpp.c
+++ b/custom_mutators/aflpp/aflpp.c
@@ -1,3 +1,4 @@
+#include "afl-fuzz.h"
 #include "afl-mutations.h"
 
 typedef struct my_mutator {
diff --git a/custom_mutators/aflpp/standalone/aflpp-standalone.c b/custom_mutators/aflpp/standalone/aflpp-standalone.c
index 361feaba..3a2cbc2f 100644
--- a/custom_mutators/aflpp/standalone/aflpp-standalone.c
+++ b/custom_mutators/aflpp/standalone/aflpp-standalone.c
@@ -1,9 +1,6 @@
+#include "afl-fuzz.h"
 #include "afl-mutations.h"
 
-s8  interesting_8[] = {INTERESTING_8};
-s16 interesting_16[] = {INTERESTING_8, INTERESTING_16};
-s32 interesting_32[] = {INTERESTING_8, INTERESTING_16, INTERESTING_32};
-
 typedef struct my_mutator {
 
   afl_state_t *afl;
@@ -155,7 +152,7 @@ int main(int argc, char *argv[]) {
     return -1;
   }
 
-  if (verbose) fprintf(stderr, "Mutation output length: %zu\n", outlen);
+  if (verbose) fprintf(stderr, "Mutation output length: %u\n", outlen);
 
   if (fwrite(outbuf, 1, outlen, out) != outlen) {
     fprintf(stderr, "Warning: incomplete write.\n");
diff --git a/docs/Changelog.md b/docs/Changelog.md
index 94ea5fca..70f4e375 100644
--- a/docs/Changelog.md
+++ b/docs/Changelog.md
@@ -33,6 +33,7 @@
   - afl-whatsup:
     - now also displays current average speed
     - small bugfixes
+  - Fixes for aflpp custom mutator and standalone tool
   - Minor edits to afl-persistent-config
   - Prevent temporary files being left behind on aborted afl-whatsup
   - More CPU benchmarks added to benchmark/
diff --git a/include/afl-mutations.h b/include/afl-mutations.h
index 75e66484..79cf7c6a 100644
--- a/include/afl-mutations.h
+++ b/include/afl-mutations.h
@@ -30,10 +30,13 @@
 
 #include <stdbool.h>
 #include <inttypes.h>
-#include "afl-fuzz.h"
 
 #define MUT_STRATEGY_ARRAY_SIZE 256
 
+s8  interesting_8[] = {INTERESTING_8};
+s16 interesting_16[] = {INTERESTING_8, INTERESTING_16};
+s32 interesting_32[] = {INTERESTING_8, INTERESTING_16, INTERESTING_32};
+
 enum {
 
   /* 00 */ MUT_FLIPBIT,
diff --git a/src/afl-fuzz-state.c b/src/afl-fuzz-state.c
index ae327117..c61f00bd 100644
--- a/src/afl-fuzz-state.c
+++ b/src/afl-fuzz-state.c
@@ -28,10 +28,6 @@
 #include "afl-fuzz.h"
 #include "envs.h"
 
-s8  interesting_8[] = {INTERESTING_8};
-s16 interesting_16[] = {INTERESTING_8, INTERESTING_16};
-s32 interesting_32[] = {INTERESTING_8, INTERESTING_16, INTERESTING_32};
-
 char *power_names[POWER_SCHEDULES_NUM] = {"explore", "mmopt", "exploit",
                                           "fast",    "coe",   "lin",
                                           "quad",    "rare",  "seek"};
-- 
cgit 1.4.1


From 48a862c503483f64db713fd6a0392148b5584ca4 Mon Sep 17 00:00:00 2001
From: Cornelius Aschermann <eqv@fb.com>
Date: Wed, 13 Mar 2024 11:43:58 -0700
Subject: :Adds stats tracking time spend in calibration/trim/sync

This currently does not affect statsd nor the UI. Only the fuzzer_stats file is updated
---
 include/afl-fuzz.h   |  10 +++-
 src/afl-fuzz-run.c   |  26 ++++++++---
 src/afl-fuzz-stats.c | 128 +++++++++++++++++++++++++++++----------------------
 src/afl-fuzz.c       |   1 -
 4 files changed, 100 insertions(+), 65 deletions(-)

(limited to 'include')

diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index be86910e..91eb6887 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -648,7 +648,10 @@ typedef struct afl_state {
       longest_find_time,                /* Longest time taken for a find    */
       exit_on_time,                     /* Delay to exit if no new paths    */
       sync_time,                        /* Sync time (ms)                   */
-      switch_fuzz_mode;                 /* auto or fixed fuzz mode          */
+      switch_fuzz_mode,                 /* auto or fixed fuzz mode          */
+      calibration_time_us,              /* Time spend on calibration        */
+      sync_time_us,                     /* Time spend on sync               */
+      trim_time_us;                     /* Time spend on trimming           */
 
   u32 slowest_exec_ms,                  /* Slowest testcase non hang in ms  */
       subseq_tmouts;                    /* Number of timeouts in a row      */
@@ -1215,6 +1218,10 @@ void show_stats_normal(afl_state_t *);
 void show_stats_pizza(afl_state_t *);
 void show_init_stats(afl_state_t *);
 
+void update_calibration_time(afl_state_t *afl, u64* time);
+void update_trim_time(afl_state_t *afl, u64* time);
+void update_sync_time(afl_state_t *afl, u64* time);
+
 /* StatsD */
 
 void statsd_setup_format(afl_state_t *afl);
@@ -1402,4 +1409,3 @@ void queue_testcase_store_mem(afl_state_t *afl, struct queue_entry *q, u8 *mem);
 #endif
 
 #endif
-
diff --git a/src/afl-fuzz-run.c b/src/afl-fuzz-run.c
index d764952c..82cdeb81 100644
--- a/src/afl-fuzz-run.c
+++ b/src/afl-fuzz-run.c
@@ -409,6 +409,7 @@ u8 calibrate_case(afl_state_t *afl, struct queue_entry *q, u8 *use_mem,
   u32 use_tmout = afl->fsrv.exec_tmout;
   u8 *old_sn = afl->stage_name;
 
+  u64 calibration_start_us = get_cur_time_us();
   if (unlikely(afl->shm.cmplog_mode)) { q->exec_cksum = 0; }
 
   /* Be a bit more generous about timeouts when resuming sessions, or when
@@ -504,6 +505,9 @@ u8 calibrate_case(afl_state_t *afl, struct queue_entry *q, u8 *use_mem,
 
     fault = fuzz_run_target(afl, &afl->fsrv, use_tmout);
 
+    // update the time spend in calibration after each execution, as those may be slow
+    update_calibration_time(afl, &calibration_start_us);
+
     /* afl->stop_soon is set by the handler for Ctrl+C. When it's pressed,
        we want to bail out quickly. */
 
@@ -650,6 +654,7 @@ abort_calibration:
 
   if (!first_run) { show_stats(afl); }
 
+  update_calibration_time(afl, &calibration_start_us);
   return fault;
 
 }
@@ -669,11 +674,15 @@ void sync_fuzzers(afl_state_t *afl) {
   afl->stage_max = afl->stage_cur = 0;
   afl->cur_depth = 0;
 
+  u64 sync_start_us = get_cur_time_us();
   /* Look at the entries created for every other fuzzer in the sync directory.
    */
 
   while ((sd_ent = readdir(sd))) {
 
+    // since sync can take substantial amounts of time, update time spend every iteration
+    update_sync_time(afl, &sync_start_us);
+
     u8  qd_synced_path[PATH_MAX], qd_path[PATH_MAX];
     u32 min_accept = 0, next_min_accept = 0;
 
@@ -861,6 +870,9 @@ void sync_fuzzers(afl_state_t *afl) {
 
   if (afl->foreign_sync_cnt) read_foreign_testcases(afl, 0);
 
+  //add time in sync one last time
+  update_sync_time(afl, &sync_start_us);
+
   afl->last_sync_time = get_cur_time();
   afl->last_sync_cycle = afl->queue_cycle;
 
@@ -872,8 +884,9 @@ void sync_fuzzers(afl_state_t *afl) {
 
 u8 trim_case(afl_state_t *afl, struct queue_entry *q, u8 *in_buf) {
 
+  u8  needs_write = 0, fault = 0;
   u32 orig_len = q->len;
-
+  u64 trim_start_us = get_cur_time_us();
   /* Custom mutator trimmer */
   if (afl->custom_mutators_count) {
 
@@ -897,11 +910,10 @@ u8 trim_case(afl_state_t *afl, struct queue_entry *q, u8 *in_buf) {
 
     }
 
-    if (custom_trimmed) return trimmed_case;
+    if (custom_trimmed) { fault = trimmed_case; goto abort_trimming; }
 
   }
 
-  u8  needs_write = 0, fault = 0;
   u32 trim_exec = 0;
   u32 remove_len;
   u32 len_p2;
@@ -912,7 +924,7 @@ u8 trim_case(afl_state_t *afl, struct queue_entry *q, u8 *in_buf) {
      detected, it will still work to some extent, so we don't check for
      this. */
 
-  if (unlikely(q->len < 5)) { return 0; }
+  if (unlikely(q->len < 5)) { fault = 0; goto abort_trimming; }
 
   afl->stage_name = afl->stage_name_buf;
   afl->bytes_trim_in += q->len;
@@ -946,6 +958,8 @@ u8 trim_case(afl_state_t *afl, struct queue_entry *q, u8 *in_buf) {
 
       fault = fuzz_run_target(afl, &afl->fsrv, afl->fsrv.exec_tmout);
 
+      update_trim_time(afl, &trim_start_us);
+
       if (afl->stop_soon || fault == FSRV_RUN_ERROR) { goto abort_trimming; }
 
       /* Note that we don't keep track of crashes or hangs here; maybe TODO?
@@ -1039,8 +1053,9 @@ u8 trim_case(afl_state_t *afl, struct queue_entry *q, u8 *in_buf) {
   }
 
 abort_trimming:
-
   afl->bytes_trim_out += q->len;
+  update_trim_time(afl, &trim_start_us);
+
   return fault;
 
 }
@@ -1104,4 +1119,3 @@ common_fuzz_stuff(afl_state_t *afl, u8 *out_buf, u32 len) {
   return 0;
 
 }
-
diff --git a/src/afl-fuzz-stats.c b/src/afl-fuzz-stats.c
index 4f398863..b39c8299 100644
--- a/src/afl-fuzz-stats.c
+++ b/src/afl-fuzz-stats.c
@@ -133,6 +133,10 @@ void write_setup_file(afl_state_t *afl, u32 argc, char **argv) {
 
 }
 
+static bool starts_with(char* key, char* line) {
+  return strncmp(key, line, strlen(key)) == 0;
+}
+
 /* load some of the existing stats file when resuming.*/
 void load_stats_file(afl_state_t *afl) {
 
@@ -175,65 +179,54 @@ void load_stats_file(afl_state_t *afl) {
       strcpy(keystring, lstartptr);
       lptr++;
       char *nptr;
-      switch (lineno) {
-
-        case 3:
-          if (!strcmp(keystring, "run_time          "))
-            afl->prev_run_time = 1000 * strtoull(lptr, &nptr, 10);
-          break;
-        case 5:
-          if (!strcmp(keystring, "cycles_done       "))
-            afl->queue_cycle =
-                strtoull(lptr, &nptr, 10) ? strtoull(lptr, &nptr, 10) + 1 : 0;
-          break;
-        case 7:
-          if (!strcmp(keystring, "execs_done        "))
-            afl->fsrv.total_execs = strtoull(lptr, &nptr, 10);
-          break;
-        case 10:
-          if (!strcmp(keystring, "corpus_count      ")) {
-
-            u32 corpus_count = strtoul(lptr, &nptr, 10);
-            if (corpus_count != afl->queued_items) {
-
-              WARNF(
-                  "queue/ has been modified -- things might not work, you're "
-                  "on your own!");
-
-            }
-
-          }
-
-          break;
-        case 12:
-          if (!strcmp(keystring, "corpus_found      "))
-            afl->queued_discovered = strtoul(lptr, &nptr, 10);
-          break;
-        case 13:
-          if (!strcmp(keystring, "corpus_imported   "))
-            afl->queued_imported = strtoul(lptr, &nptr, 10);
-          break;
-        case 14:
-          if (!strcmp(keystring, "max_depth         "))
-            afl->max_depth = strtoul(lptr, &nptr, 10);
-          break;
-        case 21:
-          if (!strcmp(keystring, "saved_crashes    "))
-            afl->saved_crashes = strtoull(lptr, &nptr, 10);
-          break;
-        case 22:
-          if (!strcmp(keystring, "saved_hangs      "))
-            afl->saved_hangs = strtoull(lptr, &nptr, 10);
-          break;
-        default:
-          break;
-
+      if (starts_with("run_time", keystring)){
+        afl->prev_run_time = 1000 * strtoull(lptr, &nptr, 10);
+      }
+      if (starts_with("cycles_done", keystring)){
+        afl->queue_cycle =
+            strtoull(lptr, &nptr, 10) ? strtoull(lptr, &nptr, 10) + 1 : 0;
+      }
+      if (starts_with("calibration_time", keystring)){
+        afl->calibration_time_us = strtoull(lptr, &nptr, 10) * 1000000;
+      }
+      if (starts_with("sync_time", keystring)){
+        afl->sync_time_us = strtoull(lptr, &nptr, 10) * 1000000;
+      }
+      if (starts_with("trim_time", keystring)){
+        afl->trim_time_us = strtoull(lptr, &nptr, 10) * 1000000;
       }
+      if (starts_with("execs_done", keystring)){
+        afl->fsrv.total_execs = strtoull(lptr, &nptr, 10);
+      }
+      if (starts_with("corpus_count", keystring)) {
 
-    }
+        u32 corpus_count = strtoul(lptr, &nptr, 10);
+        if (corpus_count != afl->queued_items) {
 
-  }
+          WARNF(
+              "queue/ has been modified -- things might not work, you're "
+              "on your own!");
 
+        }
+
+      }
+      if (starts_with("corpus_found", keystring)){
+        afl->queued_discovered = strtoul(lptr, &nptr, 10);
+      }
+      if (starts_with("corpus_imported", keystring)){
+        afl->queued_imported = strtoul(lptr, &nptr, 10);
+      }
+      if (starts_with("max_depth", keystring)) {
+        afl->max_depth = strtoul(lptr, &nptr, 10);
+      }
+      if (starts_with("saved_crashes", keystring)) {
+        afl->saved_crashes = strtoull(lptr, &nptr, 10);
+      }
+      if (starts_with("saved_hangs", keystring)) {
+        afl->saved_hangs = strtoull(lptr, &nptr, 10);
+      }
+    }
+  }
   if (afl->saved_crashes) { write_crash_readme(afl); }
 
   return;
@@ -300,6 +293,10 @@ void write_stats_file(afl_state_t *afl, u32 t_bytes, double bitmap_cvg,
       "cycles_done       : %llu\n"
       "cycles_wo_finds   : %llu\n"
       "time_wo_finds     : %llu\n"
+      "fuzz_time         : %llu\n"
+      "calibration_time  : %llu\n"
+      "sync_time         : %llu\n"
+      "trim_time         : %llu\n"
       "execs_done        : %llu\n"
       "execs_per_sec     : %0.02f\n"
       "execs_ps_last_min : %0.02f\n"
@@ -345,6 +342,10 @@ void write_stats_file(afl_state_t *afl, u32 t_bytes, double bitmap_cvg,
           : ((afl->start_time == 0 || afl->last_find_time == 0)
                  ? 0
                  : (cur_time - afl->last_find_time) / 1000),
+      (runtime - (afl->calibration_time_us + afl->sync_time_us + afl->trim_time_us) / 1000) / 1000,
+      afl->calibration_time_us / 1000000,
+      afl->sync_time_us / 1000000,
+      afl->trim_time_us / 1000000,
       afl->fsrv.total_execs, afl->fsrv.total_execs / ((double)(runtime) / 1000),
       afl->last_avg_execs_saved, afl->queued_items, afl->queued_favored,
       afl->queued_discovered, afl->queued_imported, afl->queued_variable,
@@ -414,7 +415,6 @@ void write_stats_file(afl_state_t *afl, u32 t_bytes, double bitmap_cvg,
 
   fclose(f);
   rename(fn_tmp, fn_final);
-
 }
 
 #ifdef INTROSPECTION
@@ -2438,4 +2438,20 @@ void show_init_stats(afl_state_t *afl) {
 #undef IB
 
 }
+void update_calibration_time(afl_state_t *afl, u64* time){
+  u64 cur =  get_cur_time_us();
+  afl->calibration_time_us += cur-*time;
+  *time = cur;
+}
+
+void update_trim_time(afl_state_t *afl, u64* time){
+  u64 cur =  get_cur_time_us();
+  afl->trim_time_us += cur-*time;
+  *time = cur;
+}
 
+void update_sync_time(afl_state_t *afl, u64* time){
+  u64 cur =  get_cur_time_us();
+  afl->sync_time_us += cur-*time;
+  *time = cur;
+}
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index 99491628..102809cd 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -3099,4 +3099,3 @@ stop_fuzzing:
 }
 
 #endif                                                          /* !AFL_LIB */
-
-- 
cgit 1.4.1


From 40adc344136c954cdc58e62acb46708816f5870a Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Tue, 9 Apr 2024 09:24:19 +0200
Subject: fix -V, code format

---
 docs/Changelog.md    |  2 ++
 include/afl-fuzz.h   | 13 +++----
 src/afl-fuzz-run.c   | 25 ++++++++++----
 src/afl-fuzz-stats.c | 98 +++++++++++++++++++++++++++++++++++++++-------------
 src/afl-fuzz.c       | 21 ++++++++---
 5 files changed, 117 insertions(+), 42 deletions(-)

(limited to 'include')

diff --git a/docs/Changelog.md b/docs/Changelog.md
index 72e20a18..116134ff 100644
--- a/docs/Changelog.md
+++ b/docs/Changelog.md
@@ -25,6 +25,8 @@
     - fix for `-t xxx+` feature
     - -e extension option now saves the queue items, crashes, etc. with the
       extension too
+    - fixes for trimmming, correct -V time and reading stats on resume by eqv
+      thanks a lot!
   - afl-cc:
     - added collision free caller instrumentation to LTO mode. activate with
       `AFL_LLVM_LTO_CALLER=1`. You can set a max depth to go through single
diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index 91eb6887..c813ae7e 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -5,9 +5,9 @@
    Originally written by Michal Zalewski
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                     Heiko Eißfeldt <heiko.eissfeldt@hexco.de>,
-                     Andrea Fioraldi <andreafioraldi@gmail.com>,
-                     Dominik Maier <mail@dmnk.co>
+                     Dominik Maier <mail@dmnk.co>,
+                     Andrea Fioraldi <andreafioraldi@gmail.com>, and
+                     Heiko Eissfeldt <heiko.eissfeldt@hexco.de>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
    Copyright 2019-2024 AFLplusplus Project. All rights reserved.
@@ -1218,9 +1218,9 @@ void show_stats_normal(afl_state_t *);
 void show_stats_pizza(afl_state_t *);
 void show_init_stats(afl_state_t *);
 
-void update_calibration_time(afl_state_t *afl, u64* time);
-void update_trim_time(afl_state_t *afl, u64* time);
-void update_sync_time(afl_state_t *afl, u64* time);
+void update_calibration_time(afl_state_t *afl, u64 *time);
+void update_trim_time(afl_state_t *afl, u64 *time);
+void update_sync_time(afl_state_t *afl, u64 *time);
 
 /* StatsD */
 
@@ -1409,3 +1409,4 @@ void queue_testcase_store_mem(afl_state_t *afl, struct queue_entry *q, u8 *mem);
 #endif
 
 #endif
+
diff --git a/src/afl-fuzz-run.c b/src/afl-fuzz-run.c
index 82cdeb81..1c6ce56a 100644
--- a/src/afl-fuzz-run.c
+++ b/src/afl-fuzz-run.c
@@ -505,7 +505,8 @@ u8 calibrate_case(afl_state_t *afl, struct queue_entry *q, u8 *use_mem,
 
     fault = fuzz_run_target(afl, &afl->fsrv, use_tmout);
 
-    // update the time spend in calibration after each execution, as those may be slow
+    // update the time spend in calibration after each execution, as those may
+    // be slow
     update_calibration_time(afl, &calibration_start_us);
 
     /* afl->stop_soon is set by the handler for Ctrl+C. When it's pressed,
@@ -680,7 +681,8 @@ void sync_fuzzers(afl_state_t *afl) {
 
   while ((sd_ent = readdir(sd))) {
 
-    // since sync can take substantial amounts of time, update time spend every iteration
+    // since sync can take substantial amounts of time, update time spend every
+    // iteration
     update_sync_time(afl, &sync_start_us);
 
     u8  qd_synced_path[PATH_MAX], qd_path[PATH_MAX];
@@ -870,7 +872,7 @@ void sync_fuzzers(afl_state_t *afl) {
 
   if (afl->foreign_sync_cnt) read_foreign_testcases(afl, 0);
 
-  //add time in sync one last time
+  // add time in sync one last time
   update_sync_time(afl, &sync_start_us);
 
   afl->last_sync_time = get_cur_time();
@@ -910,7 +912,12 @@ u8 trim_case(afl_state_t *afl, struct queue_entry *q, u8 *in_buf) {
 
     }
 
-    if (custom_trimmed) { fault = trimmed_case; goto abort_trimming; }
+    if (custom_trimmed) {
+
+      fault = trimmed_case;
+      goto abort_trimming;
+
+    }
 
   }
 
@@ -924,7 +931,12 @@ u8 trim_case(afl_state_t *afl, struct queue_entry *q, u8 *in_buf) {
      detected, it will still work to some extent, so we don't check for
      this. */
 
-  if (unlikely(q->len < 5)) { fault = 0; goto abort_trimming; }
+  if (unlikely(q->len < 5)) {
+
+    fault = 0;
+    goto abort_trimming;
+
+  }
 
   afl->stage_name = afl->stage_name_buf;
   afl->bytes_trim_in += q->len;
@@ -986,7 +998,6 @@ u8 trim_case(afl_state_t *afl, struct queue_entry *q, u8 *in_buf) {
 
         /* Let's save a clean trace, which will be needed by
            update_bitmap_score once we're done with the trimming stuff. */
-
         if (!needs_write) {
 
           needs_write = 1;
@@ -1001,7 +1012,6 @@ u8 trim_case(afl_state_t *afl, struct queue_entry *q, u8 *in_buf) {
       }
 
       /* Since this can be slow, update the screen every now and then. */
-
       if (!(trim_exec++ % afl->stats_update_freq)) { show_stats(afl); }
       ++afl->stage_cur;
 
@@ -1119,3 +1129,4 @@ common_fuzz_stuff(afl_state_t *afl, u8 *out_buf, u32 len) {
   return 0;
 
 }
+
diff --git a/src/afl-fuzz-stats.c b/src/afl-fuzz-stats.c
index b39c8299..7e1a3b92 100644
--- a/src/afl-fuzz-stats.c
+++ b/src/afl-fuzz-stats.c
@@ -133,8 +133,10 @@ void write_setup_file(afl_state_t *afl, u32 argc, char **argv) {
 
 }
 
-static bool starts_with(char* key, char* line) {
+static bool starts_with(char *key, char *line) {
+
   return strncmp(key, line, strlen(key)) == 0;
+
 }
 
 /* load some of the existing stats file when resuming.*/
@@ -179,25 +181,43 @@ void load_stats_file(afl_state_t *afl) {
       strcpy(keystring, lstartptr);
       lptr++;
       char *nptr;
-      if (starts_with("run_time", keystring)){
+      if (starts_with("run_time", keystring)) {
+
         afl->prev_run_time = 1000 * strtoull(lptr, &nptr, 10);
+
       }
-      if (starts_with("cycles_done", keystring)){
+
+      if (starts_with("cycles_done", keystring)) {
+
         afl->queue_cycle =
             strtoull(lptr, &nptr, 10) ? strtoull(lptr, &nptr, 10) + 1 : 0;
+
       }
-      if (starts_with("calibration_time", keystring)){
+
+      if (starts_with("calibration_time", keystring)) {
+
         afl->calibration_time_us = strtoull(lptr, &nptr, 10) * 1000000;
+
       }
-      if (starts_with("sync_time", keystring)){
+
+      if (starts_with("sync_time", keystring)) {
+
         afl->sync_time_us = strtoull(lptr, &nptr, 10) * 1000000;
+
       }
-      if (starts_with("trim_time", keystring)){
+
+      if (starts_with("trim_time", keystring)) {
+
         afl->trim_time_us = strtoull(lptr, &nptr, 10) * 1000000;
+
       }
-      if (starts_with("execs_done", keystring)){
+
+      if (starts_with("execs_done", keystring)) {
+
         afl->fsrv.total_execs = strtoull(lptr, &nptr, 10);
+
       }
+
       if (starts_with("corpus_count", keystring)) {
 
         u32 corpus_count = strtoul(lptr, &nptr, 10);
@@ -206,27 +226,46 @@ void load_stats_file(afl_state_t *afl) {
           WARNF(
               "queue/ has been modified -- things might not work, you're "
               "on your own!");
+          sleep(3);
 
         }
 
       }
-      if (starts_with("corpus_found", keystring)){
+
+      if (starts_with("corpus_found", keystring)) {
+
         afl->queued_discovered = strtoul(lptr, &nptr, 10);
+
       }
-      if (starts_with("corpus_imported", keystring)){
+
+      if (starts_with("corpus_imported", keystring)) {
+
         afl->queued_imported = strtoul(lptr, &nptr, 10);
+
       }
+
       if (starts_with("max_depth", keystring)) {
+
         afl->max_depth = strtoul(lptr, &nptr, 10);
+
       }
+
       if (starts_with("saved_crashes", keystring)) {
+
         afl->saved_crashes = strtoull(lptr, &nptr, 10);
+
       }
+
       if (starts_with("saved_hangs", keystring)) {
+
         afl->saved_hangs = strtoull(lptr, &nptr, 10);
+
       }
+
     }
+
   }
+
   if (afl->saved_crashes) { write_crash_readme(afl); }
 
   return;
@@ -334,7 +373,7 @@ void write_stats_file(afl_state_t *afl, u32 t_bytes, double bitmap_cvg,
       "\n"
       "target_mode       : %s%s%s%s%s%s%s%s%s%s\n"
       "command_line      : %s\n",
-      (afl->start_time - afl->prev_run_time) / 1000, cur_time / 1000,
+      (afl->start_time /*- afl->prev_run_time*/) / 1000, cur_time / 1000,
       runtime / 1000, (u32)getpid(),
       afl->queue_cycle ? (afl->queue_cycle - 1) : 0, afl->cycles_wo_finds,
       afl->longest_find_time > cur_time - afl->last_find_time
@@ -342,11 +381,13 @@ void write_stats_file(afl_state_t *afl, u32 t_bytes, double bitmap_cvg,
           : ((afl->start_time == 0 || afl->last_find_time == 0)
                  ? 0
                  : (cur_time - afl->last_find_time) / 1000),
-      (runtime - (afl->calibration_time_us + afl->sync_time_us + afl->trim_time_us) / 1000) / 1000,
-      afl->calibration_time_us / 1000000,
-      afl->sync_time_us / 1000000,
-      afl->trim_time_us / 1000000,
-      afl->fsrv.total_execs, afl->fsrv.total_execs / ((double)(runtime) / 1000),
+      (runtime -
+       (afl->calibration_time_us + afl->sync_time_us + afl->trim_time_us) /
+           1000) /
+          1000,
+      afl->calibration_time_us / 1000000, afl->sync_time_us / 1000000,
+      afl->trim_time_us / 1000000, afl->fsrv.total_execs,
+      afl->fsrv.total_execs / ((double)(runtime) / 1000),
       afl->last_avg_execs_saved, afl->queued_items, afl->queued_favored,
       afl->queued_discovered, afl->queued_imported, afl->queued_variable,
       afl->max_depth, afl->current_entry, afl->pending_favored,
@@ -415,6 +456,7 @@ void write_stats_file(afl_state_t *afl, u32 t_bytes, double bitmap_cvg,
 
   fclose(f);
   rename(fn_tmp, fn_final);
+
 }
 
 #ifdef INTROSPECTION
@@ -2438,20 +2480,28 @@ void show_init_stats(afl_state_t *afl) {
 #undef IB
 
 }
-void update_calibration_time(afl_state_t *afl, u64* time){
-  u64 cur =  get_cur_time_us();
-  afl->calibration_time_us += cur-*time;
+
+void update_calibration_time(afl_state_t *afl, u64 *time) {
+
+  u64 cur = get_cur_time_us();
+  afl->calibration_time_us += cur - *time;
   *time = cur;
+
 }
 
-void update_trim_time(afl_state_t *afl, u64* time){
-  u64 cur =  get_cur_time_us();
-  afl->trim_time_us += cur-*time;
+void update_trim_time(afl_state_t *afl, u64 *time) {
+
+  u64 cur = get_cur_time_us();
+  afl->trim_time_us += cur - *time;
   *time = cur;
+
 }
 
-void update_sync_time(afl_state_t *afl, u64* time){
-  u64 cur =  get_cur_time_us();
-  afl->sync_time_us += cur-*time;
+void update_sync_time(afl_state_t *afl, u64 *time) {
+
+  u64 cur = get_cur_time_us();
+  afl->sync_time_us += cur - *time;
   *time = cur;
+
 }
+
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index 102809cd..00d24ab1 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -5,8 +5,9 @@
    Originally written by Michal Zalewski
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                        Heiko Eißfeldt <heiko.eissfeldt@hexco.de> and
-                        Andrea Fioraldi <andreafioraldi@gmail.com>
+                     Dominik Meier <mail@dmnk.co>,
+                     Andrea Fioraldi <andreafioraldi@gmail.com>, and
+                     Heiko Eissfeldt <heiko.eissfeldt@hexco.de>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
    Copyright 2019-2024 AFLplusplus Project. All rights reserved.
@@ -199,7 +200,8 @@ static void usage(u8 *argv0, int more_help) {
 
       "Test settings:\n"
       "  -s seed       - use a fixed seed for the RNG\n"
-      "  -V seconds    - fuzz for a specified time then terminate\n"
+      "  -V seconds    - fuzz for a specified time then terminate (fuzz time "
+      "only!)\n"
       "  -E execs      - fuzz for an approx. no. of total executions then "
       "terminate\n"
       "                  Note: not precise and can have several more "
@@ -2543,8 +2545,6 @@ int main(int argc, char **argv_orig, char **envp) {
   }
 
   // (void)nice(-20);  // does not improve the speed
-  // real start time, we reset, so this works correctly with -V
-  afl->start_time = get_cur_time();
 
   #ifdef INTROSPECTION
   u32 prev_saved_crashes = 0, prev_saved_tmouts = 0;
@@ -2565,6 +2565,9 @@ int main(int argc, char **argv_orig, char **envp) {
   OKF("Writing mutation introspection to '%s'", ifn);
   #endif
 
+  // real start time, we reset, so this works correctly with -V
+  afl->start_time = get_cur_time();
+
   while (likely(!afl->stop_soon)) {
 
     cull_queue(afl);
@@ -2585,6 +2588,13 @@ int main(int argc, char **argv_orig, char **envp) {
 
         sync_fuzzers(afl);
 
+        if (!afl->queue_cycle && afl->afl_env.afl_import_first) {
+
+          // real start time, we reset, so this works correctly with -V
+          afl->start_time = get_cur_time();
+
+        }
+
       }
 
       ++afl->queue_cycle;
@@ -3099,3 +3109,4 @@ stop_fuzzing:
 }
 
 #endif                                                          /* !AFL_LIB */
+
-- 
cgit 1.4.1


From e01307a993387bfe842df1deb23ec7facffd4859 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Sat, 13 Apr 2024 11:39:19 +0200
Subject: v4.20c

---
 README.md         | 4 ++--
 docs/Changelog.md | 2 +-
 include/config.h  | 2 +-
 3 files changed, 4 insertions(+), 4 deletions(-)

(limited to 'include')

diff --git a/README.md b/README.md
index f15089c2..2583407e 100644
--- a/README.md
+++ b/README.md
@@ -2,9 +2,9 @@
 
 <img align="right" src="https://raw.githubusercontent.com/AFLplusplus/Website/main/static/aflpp_bg.svg" alt="AFL++ logo" width="250" heigh="250">
 
-Release version: [4.10c](https://github.com/AFLplusplus/AFLplusplus/releases)
+Release version: [4.20c](https://github.com/AFLplusplus/AFLplusplus/releases)
 
-GitHub version: 4.20a
+GitHub version: 4.20c
 
 Repository:
 [https://github.com/AFLplusplus/AFLplusplus](https://github.com/AFLplusplus/AFLplusplus)
diff --git a/docs/Changelog.md b/docs/Changelog.md
index 116134ff..2428d63f 100644
--- a/docs/Changelog.md
+++ b/docs/Changelog.md
@@ -3,7 +3,7 @@
   This is the list of all noteworthy changes made in every public
   release of the tool. See README.md for the general instruction manual.
 
-### Version ++4.20a (dev)
+### Version ++4.20c (release)
   ! A new forkserver communication model is now introduced. afl-fuzz is
     backward compatible to old compiled targets if they are not built
     for CMPLOG/Redqueen, but new compiled targets will not work with
diff --git a/include/config.h b/include/config.h
index 31d66b14..3ea059ff 100644
--- a/include/config.h
+++ b/include/config.h
@@ -26,7 +26,7 @@
 /* Version string: */
 
 // c = release, a = volatile github dev, e = experimental branch
-#define VERSION "++4.20a"
+#define VERSION "++4.20c"
 
 /******************************************************
  *                                                    *
-- 
cgit 1.4.1


From 6b049536f1614892df99f7c1ebd7710192b607a8 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Sat, 13 Apr 2024 11:54:02 +0200
Subject: v4.21 init

---
 .github/workflows/ci.yml | 1 -
 README.md                | 2 +-
 docs/Changelog.md        | 4 ++++
 include/config.h         | 2 +-
 4 files changed, 6 insertions(+), 3 deletions(-)

(limited to 'include')

diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index dd0d13e9..ed382fbb 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -5,7 +5,6 @@ on:
     branches:
       - stable
       - dev
-      - 420
   pull_request:
     branches:
       - dev # No need for stable-pull-request, as that equals dev-push
diff --git a/README.md b/README.md
index 2583407e..34d73890 100644
--- a/README.md
+++ b/README.md
@@ -4,7 +4,7 @@
 
 Release version: [4.20c](https://github.com/AFLplusplus/AFLplusplus/releases)
 
-GitHub version: 4.20c
+GitHub version: 4.21a
 
 Repository:
 [https://github.com/AFLplusplus/AFLplusplus](https://github.com/AFLplusplus/AFLplusplus)
diff --git a/docs/Changelog.md b/docs/Changelog.md
index 2428d63f..a7eb239b 100644
--- a/docs/Changelog.md
+++ b/docs/Changelog.md
@@ -3,6 +3,10 @@
   This is the list of all noteworthy changes made in every public
   release of the tool. See README.md for the general instruction manual.
 
+### Version ++4.21a (dev)
+  * your PR? :-)
+
+
 ### Version ++4.20c (release)
   ! A new forkserver communication model is now introduced. afl-fuzz is
     backward compatible to old compiled targets if they are not built
diff --git a/include/config.h b/include/config.h
index 3ea059ff..a2ff68ea 100644
--- a/include/config.h
+++ b/include/config.h
@@ -26,7 +26,7 @@
 /* Version string: */
 
 // c = release, a = volatile github dev, e = experimental branch
-#define VERSION "++4.20c"
+#define VERSION "++4.21a"
 
 /******************************************************
  *                                                    *
-- 
cgit 1.4.1


From 476aca5b67f2926f0cdc7c50e9669e68cad9a851 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Fri, 19 Apr 2024 15:45:00 +0200
Subject: nits

---
 include/afl-as.h                          | 2 +-
 include/afl-prealloc.h                    | 2 +-
 include/alloc-inl.h                       | 2 +-
 include/cmplog.h                          | 2 +-
 include/common.h                          | 2 +-
 include/debug.h                           | 2 +-
 include/forkserver.h                      | 2 +-
 include/list.h                            | 2 +-
 include/sharedmem.h                       | 2 +-
 include/snapshot-inl.h                    | 2 +-
 include/types.h                           | 2 +-
 instrumentation/split-compares-pass.so.cc | 4 ++--
 src/afl-analyze.c                         | 2 +-
 src/afl-as.c                              | 2 +-
 src/afl-cc.c                              | 8 ++++----
 src/afl-common.c                          | 2 +-
 src/afl-forkserver.c                      | 2 +-
 src/afl-fuzz-bitmap.c                     | 2 +-
 src/afl-fuzz-cmplog.c                     | 2 +-
 src/afl-fuzz-extras.c                     | 2 +-
 src/afl-fuzz-init.c                       | 2 +-
 src/afl-fuzz-mutators.c                   | 2 +-
 src/afl-fuzz-one.c                        | 2 +-
 src/afl-fuzz-python.c                     | 2 +-
 src/afl-fuzz-queue.c                      | 2 +-
 src/afl-fuzz-redqueen.c                   | 2 +-
 src/afl-fuzz-run.c                        | 5 ++---
 src/afl-fuzz-state.c                      | 2 +-
 src/afl-fuzz-stats.c                      | 9 +++++----
 src/afl-gotcpu.c                          | 2 +-
 src/afl-ld-lto.c                          | 2 +-
 src/afl-sharedmem.c                       | 2 +-
 src/afl-showmap.c                         | 2 +-
 src/afl-tmin.c                            | 2 +-
 34 files changed, 43 insertions(+), 43 deletions(-)

(limited to 'include')

diff --git a/include/afl-as.h b/include/afl-as.h
index 612f34f4..c005d43d 100644
--- a/include/afl-as.h
+++ b/include/afl-as.h
@@ -5,7 +5,7 @@
    Originally written by Michal Zalewski
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                     Heiko Eißfeldt <heiko.eissfeldt@hexco.de>,
+                     Heiko Eissfeldt <heiko.eissfeldt@hexco.de>,
                      Andrea Fioraldi <andreafioraldi@gmail.com>,
                      Dominik Maier <mail@dmnk.co>
 
diff --git a/include/afl-prealloc.h b/include/afl-prealloc.h
index 3c621d79..bcccb6b4 100644
--- a/include/afl-prealloc.h
+++ b/include/afl-prealloc.h
@@ -5,7 +5,7 @@
    Originally written by Michal Zalewski
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                     Heiko Eißfeldt <heiko.eissfeldt@hexco.de>,
+                     Heiko Eissfeldt <heiko.eissfeldt@hexco.de>,
                      Andrea Fioraldi <andreafioraldi@gmail.com>,
                      Dominik Maier <mail@dmnk.co>
 
diff --git a/include/alloc-inl.h b/include/alloc-inl.h
index 0aa417be..dad0652f 100644
--- a/include/alloc-inl.h
+++ b/include/alloc-inl.h
@@ -5,7 +5,7 @@
    Originally written by Michal Zalewski
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                     Heiko Eißfeldt <heiko.eissfeldt@hexco.de>,
+                     Heiko Eissfeldt <heiko.eissfeldt@hexco.de>,
                      Andrea Fioraldi <andreafioraldi@gmail.com>,
                      Dominik Maier <mail@dmnk.co>
 
diff --git a/include/cmplog.h b/include/cmplog.h
index a6162b59..a4449a60 100644
--- a/include/cmplog.h
+++ b/include/cmplog.h
@@ -7,7 +7,7 @@
    Forkserver design by Jann Horn <jannhorn@googlemail.com>
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                     Heiko Eißfeldt <heiko.eissfeldt@hexco.de>,
+                     Heiko Eissfeldt <heiko.eissfeldt@hexco.de>,
                      Andrea Fioraldi <andreafioraldi@gmail.com>,
                      Dominik Maier <mail@dmnk.co>
 
diff --git a/include/common.h b/include/common.h
index 0df07dee..a78dd60a 100644
--- a/include/common.h
+++ b/include/common.h
@@ -5,7 +5,7 @@
    Originally written by Michal Zalewski
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                     Heiko Eißfeldt <heiko.eissfeldt@hexco.de>,
+                     Heiko Eissfeldt <heiko.eissfeldt@hexco.de>,
                      Andrea Fioraldi <andreafioraldi@gmail.com>,
                      Dominik Maier <mail@dmnk.co>
 
diff --git a/include/debug.h b/include/debug.h
index 4b812f8e..5496135c 100644
--- a/include/debug.h
+++ b/include/debug.h
@@ -5,7 +5,7 @@
    Originally written by Michal Zalewski
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                     Heiko Eißfeldt <heiko.eissfeldt@hexco.de>,
+                     Heiko Eissfeldt <heiko.eissfeldt@hexco.de>,
                      Andrea Fioraldi <andreafioraldi@gmail.com>,
                      Dominik Maier <mail@dmnk.co>
 
diff --git a/include/forkserver.h b/include/forkserver.h
index be7f9e8d..68907376 100644
--- a/include/forkserver.h
+++ b/include/forkserver.h
@@ -7,7 +7,7 @@
    Forkserver design by Jann Horn <jannhorn@googlemail.com>
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                     Heiko Eißfeldt <heiko.eissfeldt@hexco.de>,
+                     Heiko Eissfeldt <heiko.eissfeldt@hexco.de>,
                      Andrea Fioraldi <andreafioraldi@gmail.com>,
                      Dominik Maier <mail@dmnk.co>>
 
diff --git a/include/list.h b/include/list.h
index 441eccd3..bec9abbc 100644
--- a/include/list.h
+++ b/include/list.h
@@ -5,7 +5,7 @@
    Originally written by Michal Zalewski
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                     Heiko Eißfeldt <heiko.eissfeldt@hexco.de>,
+                     Heiko Eissfeldt <heiko.eissfeldt@hexco.de>,
                      Andrea Fioraldi <andreafioraldi@gmail.com>,
                      Dominik Maier <mail@dmnk.co>
 
diff --git a/include/sharedmem.h b/include/sharedmem.h
index 4484066e..036fa560 100644
--- a/include/sharedmem.h
+++ b/include/sharedmem.h
@@ -7,7 +7,7 @@
    Forkserver design by Jann Horn <jannhorn@googlemail.com>
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                     Heiko Eißfeldt <heiko.eissfeldt@hexco.de>,
+                     Heiko Eissfeldt <heiko.eissfeldt@hexco.de>,
                      Andrea Fioraldi <andreafioraldi@gmail.com>,
                      Dominik Maier <mail@dmnk.co>
 
diff --git a/include/snapshot-inl.h b/include/snapshot-inl.h
index b2c81402..e577b013 100644
--- a/include/snapshot-inl.h
+++ b/include/snapshot-inl.h
@@ -7,7 +7,7 @@
    Forkserver design by Jann Horn <jannhorn@googlemail.com>
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                     Heiko Eißfeldt <heiko.eissfeldt@hexco.de>,
+                     Heiko Eissfeldt <heiko.eissfeldt@hexco.de>,
                      Andrea Fioraldi <andreafioraldi@gmail.com>,
                      Dominik Maier <mail@dmnk.co>
 
diff --git a/include/types.h b/include/types.h
index 18c5df91..cfb2f3d5 100644
--- a/include/types.h
+++ b/include/types.h
@@ -5,7 +5,7 @@
    Originally written by Michal Zalewski
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                     Heiko Eißfeldt <heiko.eissfeldt@hexco.de>,
+                     Heiko Eissfeldt <heiko.eissfeldt@hexco.de>,
                      Andrea Fioraldi <andreafioraldi@gmail.com>,
                      Dominik Maier <mail@dmnk.co>
 
diff --git a/instrumentation/split-compares-pass.so.cc b/instrumentation/split-compares-pass.so.cc
index 421a7c39..728ebc22 100644
--- a/instrumentation/split-compares-pass.so.cc
+++ b/instrumentation/split-compares-pass.so.cc
@@ -1,7 +1,7 @@
 /*
  * Copyright 2016 laf-intel
- * extended for floating point by Heiko Eißfeldt
- * adapted to new pass manager by Heiko Eißfeldt
+ * extended for floating point by Heiko Eissfeldt
+ * adapted to new pass manager by Heiko Eissfeldt
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/src/afl-analyze.c b/src/afl-analyze.c
index 95f32fee..d089cd08 100644
--- a/src/afl-analyze.c
+++ b/src/afl-analyze.c
@@ -5,7 +5,7 @@
    Originally written by Michal Zalewski
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                        Heiko Eißfeldt <heiko.eissfeldt@hexco.de> and
+                        Heiko Eissfeldt <heiko.eissfeldt@hexco.de> and
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
diff --git a/src/afl-as.c b/src/afl-as.c
index 09ba75bf..d4ddb94d 100644
--- a/src/afl-as.c
+++ b/src/afl-as.c
@@ -5,7 +5,7 @@
    Originally written by Michal Zalewski
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                        Heiko Eißfeldt <heiko.eissfeldt@hexco.de> and
+                        Heiko Eissfeldt <heiko.eissfeldt@hexco.de> and
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
diff --git a/src/afl-cc.c b/src/afl-cc.c
index 57089ae0..202e8145 100644
--- a/src/afl-cc.c
+++ b/src/afl-cc.c
@@ -2490,10 +2490,10 @@ void add_assembler(aflcc_state_t *aflcc) {
   u8 *slash = strrchr(afl_as, '/');
   if (slash) *slash = 0;
 
-  // Search for 'as' may be unreliable in some cases (see #2058)
-  // so use 'afl-as' instead, because 'as' is usually a symbolic link,
-  // or can be a renamed copy of 'afl-as' created in the same dir.
-  // Now we should verify if the compiler can find the 'as' we need.
+    // Search for 'as' may be unreliable in some cases (see #2058)
+    // so use 'afl-as' instead, because 'as' is usually a symbolic link,
+    // or can be a renamed copy of 'afl-as' created in the same dir.
+    // Now we should verify if the compiler can find the 'as' we need.
 
 #define AFL_AS_ERR "(should be a symlink or copy of 'afl-as')"
 
diff --git a/src/afl-common.c b/src/afl-common.c
index 6d915b00..d86b431b 100644
--- a/src/afl-common.c
+++ b/src/afl-common.c
@@ -5,7 +5,7 @@
    Originally written by Michal Zalewski
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                        Heiko Eißfeldt <heiko.eissfeldt@hexco.de> and
+                        Heiko Eissfeldt <heiko.eissfeldt@hexco.de> and
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
diff --git a/src/afl-forkserver.c b/src/afl-forkserver.c
index f28a2a64..149a973e 100644
--- a/src/afl-forkserver.c
+++ b/src/afl-forkserver.c
@@ -7,7 +7,7 @@
    Forkserver design by Jann Horn <jannhorn@googlemail.com>
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                        Heiko Eißfeldt <heiko.eissfeldt@hexco.de> and
+                        Heiko Eissfeldt <heiko.eissfeldt@hexco.de> and
                         Andrea Fioraldi <andreafioraldi@gmail.com> and
                         Dominik Maier <mail@dmnk.co>
 
diff --git a/src/afl-fuzz-bitmap.c b/src/afl-fuzz-bitmap.c
index d8561dde..5d4d80af 100644
--- a/src/afl-fuzz-bitmap.c
+++ b/src/afl-fuzz-bitmap.c
@@ -5,7 +5,7 @@
    Originally written by Michal Zalewski
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                        Heiko Eißfeldt <heiko.eissfeldt@hexco.de> and
+                        Heiko Eissfeldt <heiko.eissfeldt@hexco.de> and
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
diff --git a/src/afl-fuzz-cmplog.c b/src/afl-fuzz-cmplog.c
index 21f34e12..8c48eb49 100644
--- a/src/afl-fuzz-cmplog.c
+++ b/src/afl-fuzz-cmplog.c
@@ -7,7 +7,7 @@
    Forkserver design by Jann Horn <jannhorn@googlemail.com>
 
    Now maintained by by Marc Heuse <mh@mh-sec.de>,
-                        Heiko Eißfeldt <heiko.eissfeldt@hexco.de> and
+                        Heiko Eissfeldt <heiko.eissfeldt@hexco.de> and
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
diff --git a/src/afl-fuzz-extras.c b/src/afl-fuzz-extras.c
index c06896ef..55b6be04 100644
--- a/src/afl-fuzz-extras.c
+++ b/src/afl-fuzz-extras.c
@@ -5,7 +5,7 @@
    Originally written by Michal Zalewski
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                        Heiko Eißfeldt <heiko.eissfeldt@hexco.de> and
+                        Heiko Eissfeldt <heiko.eissfeldt@hexco.de> and
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
diff --git a/src/afl-fuzz-init.c b/src/afl-fuzz-init.c
index 21a8ba7e..2a8267cc 100644
--- a/src/afl-fuzz-init.c
+++ b/src/afl-fuzz-init.c
@@ -5,7 +5,7 @@
    Originally written by Michal Zalewski
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                        Heiko Eißfeldt <heiko.eissfeldt@hexco.de> and
+                        Heiko Eissfeldt <heiko.eissfeldt@hexco.de> and
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
diff --git a/src/afl-fuzz-mutators.c b/src/afl-fuzz-mutators.c
index ae4d6668..2f6af4bc 100644
--- a/src/afl-fuzz-mutators.c
+++ b/src/afl-fuzz-mutators.c
@@ -5,7 +5,7 @@
    Originally written by Shengtuo Hu
 
    Now maintained by  Marc Heuse <mh@mh-sec.de>,
-                        Heiko Eißfeldt <heiko.eissfeldt@hexco.de> and
+                        Heiko Eissfeldt <heiko.eissfeldt@hexco.de> and
                         Andrea Fioraldi <andreafioraldi@gmail.com>
                         Dominik Maier <mail@dmnk.co>
 
diff --git a/src/afl-fuzz-one.c b/src/afl-fuzz-one.c
index d9c074ec..74bb8cbc 100644
--- a/src/afl-fuzz-one.c
+++ b/src/afl-fuzz-one.c
@@ -5,7 +5,7 @@
    Originally written by Michal Zalewski
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                        Heiko Eißfeldt <heiko.eissfeldt@hexco.de> and
+                        Heiko Eissfeldt <heiko.eissfeldt@hexco.de> and
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
diff --git a/src/afl-fuzz-python.c b/src/afl-fuzz-python.c
index 16a398fd..873b25e2 100644
--- a/src/afl-fuzz-python.c
+++ b/src/afl-fuzz-python.c
@@ -5,7 +5,7 @@
    Originally written by Michal Zalewski
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                        Heiko Eißfeldt <heiko.eissfeldt@hexco.de> and
+                        Heiko Eissfeldt <heiko.eissfeldt@hexco.de> and
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
diff --git a/src/afl-fuzz-queue.c b/src/afl-fuzz-queue.c
index 1ea50418..df4e7d79 100644
--- a/src/afl-fuzz-queue.c
+++ b/src/afl-fuzz-queue.c
@@ -5,7 +5,7 @@
    Originally written by Michal Zalewski
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                        Heiko Eißfeldt <heiko.eissfeldt@hexco.de> and
+                        Heiko Eissfeldt <heiko.eissfeldt@hexco.de> and
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
diff --git a/src/afl-fuzz-redqueen.c b/src/afl-fuzz-redqueen.c
index be41d6c4..100b0dd6 100644
--- a/src/afl-fuzz-redqueen.c
+++ b/src/afl-fuzz-redqueen.c
@@ -7,7 +7,7 @@
    Forkserver design by Jann Horn <jannhorn@googlemail.com>
 
    Now maintained by by Marc Heuse <mh@mh-sec.de>,
-                        Heiko Eißfeldt <heiko.eissfeldt@hexco.de> and
+                        Heiko Eissfeldt <heiko.eissfeldt@hexco.de> and
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
diff --git a/src/afl-fuzz-run.c b/src/afl-fuzz-run.c
index edcddc8e..ab96c778 100644
--- a/src/afl-fuzz-run.c
+++ b/src/afl-fuzz-run.c
@@ -5,7 +5,7 @@
    Originally written by Michal Zalewski
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                        Heiko Eißfeldt <heiko.eissfeldt@hexco.de> and
+                        Heiko Eissfeldt <heiko.eissfeldt@hexco.de> and
                         Andrea Fioraldi <andreafioraldi@gmail.com> and
                         Dominik Maier <mail@dmnk.co>
 
@@ -829,8 +829,7 @@ void sync_fuzzers(afl_state_t *afl) {
         if (afl->stop_soon) { goto close_sync; }
 
         afl->syncing_party = sd_ent->d_name;
-        afl->queued_imported +=
-            save_if_interesting(afl, mem, new_len, fault);
+        afl->queued_imported += save_if_interesting(afl, mem, new_len, fault);
         afl->syncing_party = 0;
 
         munmap(mem, st.st_size);
diff --git a/src/afl-fuzz-state.c b/src/afl-fuzz-state.c
index c61f00bd..c21ae6be 100644
--- a/src/afl-fuzz-state.c
+++ b/src/afl-fuzz-state.c
@@ -5,7 +5,7 @@
    Originally written by Michal Zalewski
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                        Heiko Eißfeldt <heiko.eissfeldt@hexco.de> and
+                        Heiko Eissfeldt <heiko.eissfeldt@hexco.de> and
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
diff --git a/src/afl-fuzz-stats.c b/src/afl-fuzz-stats.c
index 7e1a3b92..755e1c50 100644
--- a/src/afl-fuzz-stats.c
+++ b/src/afl-fuzz-stats.c
@@ -5,8 +5,9 @@
    Originally written by Michal Zalewski
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                        Heiko Eißfeldt <heiko.eissfeldt@hexco.de> and
-                        Andrea Fioraldi <andreafioraldi@gmail.com>
+                     Dominik Meier <mail@dmnk.co>,
+                     Andrea Fioraldi <andreafioraldi@gmail.com>, and
+                     Heiko Eissfeldt <heiko.eissfeldt@hexco.de>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
    Copyright 2019-2024 AFLplusplus Project. All rights reserved.
@@ -382,8 +383,8 @@ void write_stats_file(afl_state_t *afl, u32 t_bytes, double bitmap_cvg,
                  ? 0
                  : (cur_time - afl->last_find_time) / 1000),
       (runtime -
-       (afl->calibration_time_us + afl->sync_time_us + afl->trim_time_us) /
-           1000) /
+       ((afl->calibration_time_us + afl->sync_time_us + afl->trim_time_us) /
+        1000)) /
           1000,
       afl->calibration_time_us / 1000000, afl->sync_time_us / 1000000,
       afl->trim_time_us / 1000000, afl->fsrv.total_execs,
diff --git a/src/afl-gotcpu.c b/src/afl-gotcpu.c
index 7aee2985..6a3bd037 100644
--- a/src/afl-gotcpu.c
+++ b/src/afl-gotcpu.c
@@ -5,7 +5,7 @@
    Originally written by Michal Zalewski
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                        Heiko Eißfeldt <heiko.eissfeldt@hexco.de> and
+                        Heiko Eissfeldt <heiko.eissfeldt@hexco.de> and
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
diff --git a/src/afl-ld-lto.c b/src/afl-ld-lto.c
index 513c1ae9..578552ba 100644
--- a/src/afl-ld-lto.c
+++ b/src/afl-ld-lto.c
@@ -5,7 +5,7 @@
   Written by Marc Heuse <mh@mh-sec.de> for AFL++
 
   Maintained by Marc Heuse <mh@mh-sec.de>,
-                Heiko Eißfeldt <heiko.eissfeldt@hexco.de>
+                Heiko Eissfeldt <heiko.eissfeldt@hexco.de>
                 Andrea Fioraldi <andreafioraldi@gmail.com>
                 Dominik Maier <domenukk@gmail.com>
 
diff --git a/src/afl-sharedmem.c b/src/afl-sharedmem.c
index daea8f46..8f685633 100644
--- a/src/afl-sharedmem.c
+++ b/src/afl-sharedmem.c
@@ -7,7 +7,7 @@
    Forkserver design by Jann Horn <jannhorn@googlemail.com>
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                        Heiko Eißfeldt <heiko.eissfeldt@hexco.de> and
+                        Heiko Eissfeldt <heiko.eissfeldt@hexco.de> and
                         Andrea Fioraldi <andreafioraldi@gmail.com>
 
    Copyright 2016, 2017 Google Inc. All rights reserved.
diff --git a/src/afl-showmap.c b/src/afl-showmap.c
index 20ba5a5e..07a4844a 100644
--- a/src/afl-showmap.c
+++ b/src/afl-showmap.c
@@ -7,7 +7,7 @@
    Forkserver design by Jann Horn <jannhorn@googlemail.com>
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                        Heiko Eißfeldt <heiko.eissfeldt@hexco.de> and
+                        Heiko Eissfeldt <heiko.eissfeldt@hexco.de> and
                         Andrea Fioraldi <andreafioraldi@gmail.com> and
                         Dominik Maier <mail@dmnk.co>
 
diff --git a/src/afl-tmin.c b/src/afl-tmin.c
index 994174ed..23e0ff13 100644
--- a/src/afl-tmin.c
+++ b/src/afl-tmin.c
@@ -7,7 +7,7 @@
    Forkserver design by Jann Horn <jannhorn@googlemail.com>
 
    Now maintained by Marc Heuse <mh@mh-sec.de>,
-                        Heiko Eißfeldt <heiko.eissfeldt@hexco.de> and
+                        Heiko Eissfeldt <heiko.eissfeldt@hexco.de> and
                         Andrea Fioraldi <andreafioraldi@gmail.com> and
                         Dominik Maier <mail@dmnk.co>
 
-- 
cgit 1.4.1


From 26eaf53a832be0b12dadbbd290b4a7e676818347 Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Thu, 2 May 2024 08:35:24 +0200
Subject: AFL_DISABLE_REDUNDANT

---
 docs/Changelog.md       | 2 ++
 docs/env_variables.md   | 3 +++
 include/afl-fuzz.h      | 2 +-
 include/envs.h          | 3 ++-
 src/afl-fuzz-init.c     | 7 +++++--
 src/afl-fuzz-queue.c    | 1 +
 src/afl-fuzz-redqueen.c | 9 +++++----
 src/afl-fuzz-state.c    | 7 +++++++
 src/afl-fuzz.c          | 1 +
 9 files changed, 27 insertions(+), 8 deletions(-)

(limited to 'include')

diff --git a/docs/Changelog.md b/docs/Changelog.md
index c1b2f62a..5cb6973a 100644
--- a/docs/Changelog.md
+++ b/docs/Changelog.md
@@ -5,6 +5,7 @@
 
 ### Version ++4.21a (dev)
   * afl-fuzz
+    - added AFL_DISABLE_REDUNDANT for huge queues
     - fix AFL_PERSISTENT_RECORD
     - prevent filenames in the queue that have spaces
     - minor fix for FAST schedules
@@ -13,6 +14,7 @@
     - ensure shared memory variables are visible in weird build setups
   * afl-cmin
     - work with input files that have a space
+  * enhanced the ASAN configuration
 
 
 ### Version ++4.20c (release)
diff --git a/docs/env_variables.md b/docs/env_variables.md
index 1e4fc7ba..01904aea 100644
--- a/docs/env_variables.md
+++ b/docs/env_variables.md
@@ -381,6 +381,9 @@ checks or alter some of the more exotic semantics of the tool:
   - Setting `AFL_DISABLE_TRIM` tells afl-fuzz not to trim test cases. This is
     usually a bad idea!
 
+  - Setting `AFL_DISABLE_REDUNDANT` disables any queue items that are redundant.
+    This can be useful with huge queues.
+
   - Setting `AFL_KEEP_TIMEOUTS` will keep longer running inputs if they reach
     new coverage
 
diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index c813ae7e..1a958006 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -452,7 +452,7 @@ typedef struct afl_env_vars {
       afl_keep_timeouts, afl_no_crash_readme, afl_ignore_timeouts,
       afl_no_startup_calibration, afl_no_warn_instability,
       afl_post_process_keep_original, afl_crashing_seeds_as_new_crash,
-      afl_final_sync, afl_ignore_seed_problems;
+      afl_final_sync, afl_ignore_seed_problems, afl_disable_redundant;
 
   u8 *afl_tmpdir, *afl_custom_mutator_library, *afl_python_module, *afl_path,
       *afl_hang_tmout, *afl_forksrv_init_tmout, *afl_preload,
diff --git a/include/envs.h b/include/envs.h
index 56a4916c..c895f726 100644
--- a/include/envs.h
+++ b/include/envs.h
@@ -26,7 +26,8 @@ static char *afl_environment_variables[] = {
     "AFL_CUSTOM_MUTATOR_ONLY", "AFL_CUSTOM_INFO_PROGRAM",
     "AFL_CUSTOM_INFO_PROGRAM_ARGV", "AFL_CUSTOM_INFO_PROGRAM_INPUT",
     "AFL_CUSTOM_INFO_OUT", "AFL_CXX", "AFL_CYCLE_SCHEDULES", "AFL_DEBUG",
-    "AFL_DEBUG_CHILD", "AFL_DEBUG_GDB", "AFL_DEBUG_UNICORN", "AFL_DISABLE_TRIM",
+    "AFL_DEBUG_CHILD", "AFL_DEBUG_GDB", "AFL_DEBUG_UNICORN",
+    "AFL_DISABLE_REDUNDANT", "AFL_DISABLE_TRIM",
     "AFL_DISABLE_LLVM_INSTRUMENTATION", "AFL_DONT_OPTIMIZE",
     "AFL_DRIVER_STDERR_DUPLICATE_FILENAME", "AFL_DUMB_FORKSRV",
     "AFL_EARLY_FORKSERVER", "AFL_ENTRYPOINT", "AFL_EXIT_WHEN_DONE",
diff --git a/src/afl-fuzz-init.c b/src/afl-fuzz-init.c
index 2d540eb1..b3fe9318 100644
--- a/src/afl-fuzz-init.c
+++ b/src/afl-fuzz-init.c
@@ -933,10 +933,13 @@ void perform_dry_run(afl_state_t *afl) {
     res = calibrate_case(afl, q, use_mem, 0, 1);
 
     /* For AFLFast schedules we update the queue entry */
-    if (unlikely(afl->schedule >= FAST && afl->schedule <= RARE) && likely(q->exec_cksum)) {
+    if (unlikely(afl->schedule >= FAST && afl->schedule <= RARE) &&
+        likely(q->exec_cksum)) {
+
       q->n_fuzz_entry = q->exec_cksum % N_FUZZ_SIZE;
+
     }
-     
+
     if (afl->stop_soon) { return; }
 
     if (res == afl->crash_mode || res == FSRV_RUN_NOBITS) {
diff --git a/src/afl-fuzz-queue.c b/src/afl-fuzz-queue.c
index df4e7d79..5987ad0c 100644
--- a/src/afl-fuzz-queue.c
+++ b/src/afl-fuzz-queue.c
@@ -370,6 +370,7 @@ void mark_as_redundant(afl_state_t *afl, struct queue_entry *q, u8 state) {
 
     s32 fd;
 
+    if (unlikely(afl->afl_env.afl_disable_redundant)) { q->disabled = 1; }
     fd = open(fn, O_WRONLY | O_CREAT | O_EXCL, DEFAULT_PERMISSION);
     if (fd < 0) { PFATAL("Unable to create '%s'", fn); }
     close(fd);
diff --git a/src/afl-fuzz-redqueen.c b/src/afl-fuzz-redqueen.c
index cfa57c1d..9316da71 100644
--- a/src/afl-fuzz-redqueen.c
+++ b/src/afl-fuzz-redqueen.c
@@ -2764,15 +2764,15 @@ static u8 rtn_fuzz(afl_state_t *afl, u32 key, u8 *orig_buf, u8 *buf, u8 *cbuf,
 #ifdef _DEBUG
     u32                j;
     struct cmp_header *hh = &afl->orig_cmp_map->headers[key];
-    fprintf(stderr, "RTN N hits=%u shape=%u attr=%u v0=", h->hits,
-            hshape, h->attribute);
+    fprintf(stderr, "RTN N hits=%u shape=%u attr=%u v0=", h->hits, hshape,
+            h->attribute);
     for (j = 0; j < 8; j++)
       fprintf(stderr, "%02x", o->v0[j]);
     fprintf(stderr, " v1=");
     for (j = 0; j < 8; j++)
       fprintf(stderr, "%02x", o->v1[j]);
-    fprintf(stderr, "\nRTN O hits=%u shape=%u attr=%u o0=", hh->hits,
-             hshape, hh->attribute);
+    fprintf(stderr, "\nRTN O hits=%u shape=%u attr=%u o0=", hh->hits, hshape,
+            hh->attribute);
     for (j = 0; j < 8; j++)
       fprintf(stderr, "%02x", orig_o->v0[j]);
     fprintf(stderr, " o1=");
@@ -3273,3 +3273,4 @@ exit_its:
   return r;
 
 }
+
diff --git a/src/afl-fuzz-state.c b/src/afl-fuzz-state.c
index c21ae6be..543fdc1c 100644
--- a/src/afl-fuzz-state.c
+++ b/src/afl-fuzz-state.c
@@ -293,6 +293,13 @@ void read_afl_environment(afl_state_t *afl, char **envp) {
             afl->afl_env.afl_cmplog_only_new =
                 get_afl_env(afl_environment_variables[i]) ? 1 : 0;
 
+          } else if (!strncmp(env, "AFL_DISABLE_REDUNDANT",
+
+                              afl_environment_variable_len)) {
+
+            afl->afl_env.afl_disable_redundant =
+                get_afl_env(afl_environment_variables[i]) ? 1 : 0;
+
           } else if (!strncmp(env, "AFL_NO_STARTUP_CALIBRATION",
 
                               afl_environment_variable_len)) {
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index 00d24ab1..329ce942 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -264,6 +264,7 @@ static void usage(u8 *argv0, int more_help) {
       "AFL_CYCLE_SCHEDULES: after completing a cycle, switch to a different -p schedule\n"
       "AFL_DEBUG: extra debugging output for Python mode trimming\n"
       "AFL_DEBUG_CHILD: do not suppress stdout/stderr from target\n"
+      "AFL_DISABLE_REDUNDANT: disable any queue item that is redundant\n"
       "AFL_DISABLE_TRIM: disable the trimming of test cases\n"
       "AFL_DUMB_FORKSRV: use fork server without feedback from target\n"
       "AFL_EXIT_WHEN_DONE: exit when all inputs are run and no new finds are found\n"
-- 
cgit 1.4.1


From c03f2897d081b2bf41e179a48d758f1f400b5929 Mon Sep 17 00:00:00 2001
From: Samuel Moelius <sam@moeli.us>
Date: Fri, 10 May 2024 16:55:32 -0400
Subject: Add `AFL_SHA1_FILENAMES` option

---
 docs/env_variables.md |   3 +
 include/afl-fuzz.h    |  29 ++++-
 include/envs.h        |  18 +--
 src/afl-fuzz-bitmap.c |  91 +++++++++++----
 src/afl-fuzz-init.c   |  78 +++++++++----
 src/afl-fuzz-queue.c  |   5 +-
 src/afl-fuzz-state.c  |   7 ++
 src/afl-performance.c | 310 ++++++++++++++++++++++++++++++++++++++++++++++++++
 8 files changed, 479 insertions(+), 62 deletions(-)

(limited to 'include')

diff --git a/docs/env_variables.md b/docs/env_variables.md
index 01904aea..b3519107 100644
--- a/docs/env_variables.md
+++ b/docs/env_variables.md
@@ -550,6 +550,9 @@ checks or alter some of the more exotic semantics of the tool:
     use a custom afl-qemu-trace or if you need to modify the afl-qemu-trace
     arguments.
 
+  - `AFL_SHA1_FILENAMES` causes AFL++ to generate files named by the SHA1 hash
+    of their contents, rather than use the standard `id:000000,...` names.
+
   - `AFL_SHUFFLE_QUEUE` randomly reorders the input queue on startup. Requested
     by some users for unorthodox parallelized fuzzing setups, but not advisable
     otherwise.
diff --git a/include/afl-fuzz.h b/include/afl-fuzz.h
index 1a958006..5efe5144 100644
--- a/include/afl-fuzz.h
+++ b/include/afl-fuzz.h
@@ -452,7 +452,8 @@ typedef struct afl_env_vars {
       afl_keep_timeouts, afl_no_crash_readme, afl_ignore_timeouts,
       afl_no_startup_calibration, afl_no_warn_instability,
       afl_post_process_keep_original, afl_crashing_seeds_as_new_crash,
-      afl_final_sync, afl_ignore_seed_problems, afl_disable_redundant;
+      afl_final_sync, afl_ignore_seed_problems, afl_disable_redundant,
+      afl_sha1_filenames;
 
   u8 *afl_tmpdir, *afl_custom_mutator_library, *afl_python_module, *afl_path,
       *afl_hang_tmout, *afl_forksrv_init_tmout, *afl_preload,
@@ -1404,6 +1405,32 @@ void queue_testcase_retake_mem(afl_state_t *afl, struct queue_entry *q, u8 *in,
 
 void queue_testcase_store_mem(afl_state_t *afl, struct queue_entry *q, u8 *mem);
 
+/* Compute the SHA1 hash of `data`, which is of `len` bytes, and return the
+ * result as a `\0`-terminated hex string, which the caller much `ck_free`. */
+char *sha1_hex(const u8 *data, size_t len);
+
+/* Apply `sha1_hex` to the first `len` bytes of data of the file at `fname`. */
+char *sha1_hex_for_file(const char *fname, u32 len);
+
+/* Create file `fn`, but allow it to already exist if `AFL_SHA1_FILENAMES` is
+ * enabled. */
+static inline int permissive_create(afl_state_t *afl, const char *fn) {
+
+  int fd = open(fn, O_WRONLY | O_CREAT | O_EXCL, DEFAULT_PERMISSION);
+  if (unlikely(fd < 0)) {
+
+    if (!(afl->afl_env.afl_sha1_filenames && errno == EEXIST)) {
+
+      PFATAL("Unable to create '%s'", fn);
+
+    }
+
+  }
+
+  return fd;
+
+}
+
 #if TESTCASE_CACHE == 1
   #error define of TESTCASE_CACHE must be zero or larger than 1
 #endif
diff --git a/include/envs.h b/include/envs.h
index c895f726..57f4d263 100644
--- a/include/envs.h
+++ b/include/envs.h
@@ -108,15 +108,15 @@ static char *afl_environment_variables[] = {
     "AFL_QEMU_PERSISTENT_RETADDR_OFFSET", "AFL_QEMU_PERSISTENT_EXITS",
     "AFL_QEMU_INST_RANGES", "AFL_QEMU_EXCLUDE_RANGES", "AFL_QEMU_SNAPSHOT",
     "AFL_QEMU_TRACK_UNSTABLE", "AFL_QUIET", "AFL_RANDOM_ALLOC_CANARY",
-    "AFL_REAL_PATH", "AFL_SHUFFLE_QUEUE", "AFL_SKIP_BIN_CHECK",
-    "AFL_SKIP_CPUFREQ", "AFL_SKIP_CRASHES", "AFL_SKIP_OSSFUZZ", "AFL_STATSD",
-    "AFL_STATSD_HOST", "AFL_STATSD_PORT", "AFL_STATSD_TAGS_FLAVOR",
-    "AFL_SYNC_TIME", "AFL_TESTCACHE_SIZE", "AFL_TESTCACHE_ENTRIES",
-    "AFL_TMIN_EXACT", "AFL_TMPDIR", "AFL_TOKEN_FILE", "AFL_TRACE_PC",
-    "AFL_USE_ASAN", "AFL_USE_MSAN", "AFL_USE_TRACE_PC", "AFL_USE_UBSAN",
-    "AFL_USE_TSAN", "AFL_USE_CFISAN", "AFL_USE_LSAN", "AFL_WINE_PATH",
-    "AFL_NO_SNAPSHOT", "AFL_EXPAND_HAVOC_NOW", "AFL_USE_FASAN", "AFL_USE_QASAN",
-    "AFL_PRINT_FILENAMES", "AFL_PIZZA_MODE", NULL
+    "AFL_REAL_PATH", "AFL_SHA1_FILENAMES", "AFL_SHUFFLE_QUEUE",
+    "AFL_SKIP_BIN_CHECK", "AFL_SKIP_CPUFREQ", "AFL_SKIP_CRASHES",
+    "AFL_SKIP_OSSFUZZ", "AFL_STATSD", "AFL_STATSD_HOST", "AFL_STATSD_PORT",
+    "AFL_STATSD_TAGS_FLAVOR", "AFL_SYNC_TIME", "AFL_TESTCACHE_SIZE",
+    "AFL_TESTCACHE_ENTRIES", "AFL_TMIN_EXACT", "AFL_TMPDIR", "AFL_TOKEN_FILE",
+    "AFL_TRACE_PC", "AFL_USE_ASAN", "AFL_USE_MSAN", "AFL_USE_TRACE_PC",
+    "AFL_USE_UBSAN", "AFL_USE_TSAN", "AFL_USE_CFISAN", "AFL_USE_LSAN",
+    "AFL_WINE_PATH", "AFL_NO_SNAPSHOT", "AFL_EXPAND_HAVOC_NOW", "AFL_USE_FASAN",
+    "AFL_USE_QASAN", "AFL_PRINT_FILENAMES", "AFL_PIZZA_MODE", NULL
 
 };
 
diff --git a/src/afl-fuzz-bitmap.c b/src/afl-fuzz-bitmap.c
index 5d4d80af..03bc5d6c 100644
--- a/src/afl-fuzz-bitmap.c
+++ b/src/afl-fuzz-bitmap.c
@@ -527,12 +527,24 @@ save_if_interesting(afl_state_t *afl, void *mem, u32 len, u8 fault) {
 
 #ifndef SIMPLE_FILES
 
-    queue_fn = alloc_printf(
-        "%s/queue/id:%06u,%s%s%s", afl->out_dir, afl->queued_items,
-        describe_op(afl, new_bits + is_timeout,
-                    NAME_MAX - strlen("id:000000,")),
-        afl->file_extension ? "." : "",
-        afl->file_extension ? (const char *)afl->file_extension : "");
+    if (!afl->afl_env.afl_sha1_filenames) {
+
+      queue_fn = alloc_printf(
+          "%s/queue/id:%06u,%s%s%s", afl->out_dir, afl->queued_items,
+          describe_op(afl, new_bits + is_timeout,
+                      NAME_MAX - strlen("id:000000,")),
+          afl->file_extension ? "." : "",
+          afl->file_extension ? (const char *)afl->file_extension : "");
+
+    } else {
+
+      const char *hex = sha1_hex(mem, len);
+      queue_fn = alloc_printf(
+          "%s/queue/%s%s%s", afl->out_dir, hex, afl->file_extension ? "." : "",
+          afl->file_extension ? (const char *)afl->file_extension : "");
+      ck_free((char *)hex);
+
+    }
 
 #else
 
@@ -542,10 +554,14 @@ save_if_interesting(afl_state_t *afl, void *mem, u32 len, u8 fault) {
         afl->file_extension ? (const char *)afl->file_extension : "");
 
 #endif                                                    /* ^!SIMPLE_FILES */
-    fd = open(queue_fn, O_WRONLY | O_CREAT | O_EXCL, DEFAULT_PERMISSION);
-    if (unlikely(fd < 0)) { PFATAL("Unable to create '%s'", queue_fn); }
-    ck_write(fd, mem, len, queue_fn);
-    close(fd);
+    fd = permissive_create(afl, queue_fn);
+    if (likely(fd >= 0)) {
+
+      ck_write(fd, mem, len, queue_fn);
+      close(fd);
+
+    }
+
     add_to_queue(afl, queue_fn, len, 0);
 
     if (unlikely(afl->fuzz_mode) &&
@@ -743,11 +759,23 @@ save_if_interesting(afl_state_t *afl, void *mem, u32 len, u8 fault) {
 
 #ifndef SIMPLE_FILES
 
-      snprintf(fn, PATH_MAX, "%s/hangs/id:%06llu,%s%s%s", afl->out_dir,
-               afl->saved_hangs,
-               describe_op(afl, 0, NAME_MAX - strlen("id:000000,")),
-               afl->file_extension ? "." : "",
-               afl->file_extension ? (const char *)afl->file_extension : "");
+      if (!afl->afl_env.afl_sha1_filenames) {
+
+        snprintf(fn, PATH_MAX, "%s/hangs/id:%06llu,%s%s%s", afl->out_dir,
+                 afl->saved_hangs,
+                 describe_op(afl, 0, NAME_MAX - strlen("id:000000,")),
+                 afl->file_extension ? "." : "",
+                 afl->file_extension ? (const char *)afl->file_extension : "");
+
+      } else {
+
+        const char *hex = sha1_hex(mem, len);
+        snprintf(fn, PATH_MAX, "%s/hangs/%s%s%s", afl->out_dir, hex,
+                 afl->file_extension ? "." : "",
+                 afl->file_extension ? (const char *)afl->file_extension : "");
+        ck_free((char *)hex);
+
+      }
 
 #else
 
@@ -799,11 +827,23 @@ save_if_interesting(afl_state_t *afl, void *mem, u32 len, u8 fault) {
 
 #ifndef SIMPLE_FILES
 
-      snprintf(fn, PATH_MAX, "%s/crashes/id:%06llu,sig:%02u,%s%s%s",
-               afl->out_dir, afl->saved_crashes, afl->fsrv.last_kill_signal,
-               describe_op(afl, 0, NAME_MAX - strlen("id:000000,sig:00,")),
-               afl->file_extension ? "." : "",
-               afl->file_extension ? (const char *)afl->file_extension : "");
+      if (!afl->afl_env.afl_sha1_filenames) {
+
+        snprintf(fn, PATH_MAX, "%s/crashes/id:%06llu,sig:%02u,%s%s%s",
+                 afl->out_dir, afl->saved_crashes, afl->fsrv.last_kill_signal,
+                 describe_op(afl, 0, NAME_MAX - strlen("id:000000,sig:00,")),
+                 afl->file_extension ? "." : "",
+                 afl->file_extension ? (const char *)afl->file_extension : "");
+
+      } else {
+
+        const char *hex = sha1_hex(mem, len);
+        snprintf(fn, PATH_MAX, "%s/crashes/%s%s%s", afl->out_dir, hex,
+                 afl->file_extension ? "." : "",
+                 afl->file_extension ? (const char *)afl->file_extension : "");
+        ck_free((char *)hex);
+
+      }
 
 #else
 
@@ -873,10 +913,13 @@ save_if_interesting(afl_state_t *afl, void *mem, u32 len, u8 fault) {
   /* If we're here, we apparently want to save the crash or hang
      test case, too. */
 
-  fd = open(fn, O_WRONLY | O_CREAT | O_EXCL, DEFAULT_PERMISSION);
-  if (unlikely(fd < 0)) { PFATAL("Unable to create '%s'", fn); }
-  ck_write(fd, mem, len, fn);
-  close(fd);
+  fd = permissive_create(afl, fn);
+  if (fd >= 0) {
+
+    ck_write(fd, mem, len, fn);
+    close(fd);
+
+  }
 
 #ifdef __linux__
   if (afl->fsrv.nyx_mode && fault == FSRV_RUN_CRASH) {
diff --git a/src/afl-fuzz-init.c b/src/afl-fuzz-init.c
index 01d0730d..7310e49f 100644
--- a/src/afl-fuzz-init.c
+++ b/src/afl-fuzz-init.c
@@ -1190,14 +1190,27 @@ void perform_dry_run(afl_state_t *afl) {
 
 #ifndef SIMPLE_FILES
 
-          snprintf(
-              crash_fn, PATH_MAX, "%s/crashes/id:%06llu,sig:%02u,%s%s%s%s",
-              afl->out_dir, afl->saved_crashes, afl->fsrv.last_kill_signal,
-              describe_op(
-                  afl, 0,
-                  NAME_MAX - strlen("id:000000,sig:00,") - strlen(use_name)),
-              use_name, afl->file_extension ? "." : "",
-              afl->file_extension ? (const char *)afl->file_extension : "");
+          if (!afl->afl_env.afl_sha1_filenames) {
+
+            snprintf(
+                crash_fn, PATH_MAX, "%s/crashes/id:%06llu,sig:%02u,%s%s%s%s",
+                afl->out_dir, afl->saved_crashes, afl->fsrv.last_kill_signal,
+                describe_op(
+                    afl, 0,
+                    NAME_MAX - strlen("id:000000,sig:00,") - strlen(use_name)),
+                use_name, afl->file_extension ? "." : "",
+                afl->file_extension ? (const char *)afl->file_extension : "");
+
+          } else {
+
+            const char *hex = sha1_hex(use_mem, read_len);
+            snprintf(
+                crash_fn, PATH_MAX, "%s/crashes/%s%s%s", afl->out_dir, hex,
+                afl->file_extension ? "." : "",
+                afl->file_extension ? (const char *)afl->file_extension : "");
+            ck_free((char *)hex);
+
+          }
 
 #else
 
@@ -1518,10 +1531,23 @@ void pivot_inputs(afl_state_t *afl) {
 
       }
 
-      nfn = alloc_printf(
-          "%s/queue/id:%06u,time:0,execs:%llu,orig:%s%s%s", afl->out_dir, id,
-          afl->fsrv.total_execs, use_name, afl->file_extension ? "." : "",
-          afl->file_extension ? (const char *)afl->file_extension : "");
+      if (!afl->afl_env.afl_sha1_filenames) {
+
+        nfn = alloc_printf(
+            "%s/queue/id:%06u,time:0,execs:%llu,orig:%s%s%s", afl->out_dir, id,
+            afl->fsrv.total_execs, use_name, afl->file_extension ? "." : "",
+            afl->file_extension ? (const char *)afl->file_extension : "");
+
+      } else {
+
+        const char *hex = sha1_hex_for_file(q->fname, q->len);
+        nfn = alloc_printf(
+            "%s/queue/%s%s%s", afl->out_dir, hex,
+            afl->file_extension ? "." : "",
+            afl->file_extension ? (const char *)afl->file_extension : "");
+        ck_free((char *)hex);
+
+      }
 
       u8 *pos = strrchr(nfn, '/');
       no_spaces(pos + 30);
@@ -1738,10 +1764,11 @@ double get_runnable_processes(void) {
 
 void nuke_resume_dir(afl_state_t *afl) {
 
-  u8 *fn;
+  u8 *const case_prefix = afl->afl_env.afl_sha1_filenames ? "" : CASE_PREFIX;
+  u8       *fn;
 
   fn = alloc_printf("%s/_resume/.state/deterministic_done", afl->out_dir);
-  if (delete_files(fn, CASE_PREFIX)) { goto dir_cleanup_failed; }
+  if (delete_files(fn, case_prefix)) { goto dir_cleanup_failed; }
   ck_free(fn);
 
   fn = alloc_printf("%s/_resume/.state/auto_extras", afl->out_dir);
@@ -1749,11 +1776,11 @@ void nuke_resume_dir(afl_state_t *afl) {
   ck_free(fn);
 
   fn = alloc_printf("%s/_resume/.state/redundant_edges", afl->out_dir);
-  if (delete_files(fn, CASE_PREFIX)) { goto dir_cleanup_failed; }
+  if (delete_files(fn, case_prefix)) { goto dir_cleanup_failed; }
   ck_free(fn);
 
   fn = alloc_printf("%s/_resume/.state/variable_behavior", afl->out_dir);
-  if (delete_files(fn, CASE_PREFIX)) { goto dir_cleanup_failed; }
+  if (delete_files(fn, case_prefix)) { goto dir_cleanup_failed; }
   ck_free(fn);
 
   fn = alloc_printf("%s/_resume/.state", afl->out_dir);
@@ -1761,7 +1788,7 @@ void nuke_resume_dir(afl_state_t *afl) {
   ck_free(fn);
 
   fn = alloc_printf("%s/_resume", afl->out_dir);
-  if (delete_files(fn, CASE_PREFIX)) { goto dir_cleanup_failed; }
+  if (delete_files(fn, case_prefix)) { goto dir_cleanup_failed; }
   ck_free(fn);
 
   return;
@@ -1778,8 +1805,9 @@ dir_cleanup_failed:
 
 static void handle_existing_out_dir(afl_state_t *afl) {
 
-  FILE *f;
-  u8   *fn = alloc_printf("%s/fuzzer_stats", afl->out_dir);
+  u8 *const case_prefix = afl->afl_env.afl_sha1_filenames ? "" : CASE_PREFIX;
+  FILE     *f;
+  u8       *fn = alloc_printf("%s/fuzzer_stats", afl->out_dir);
 
   /* See if the output directory is locked. If yes, bail out. If not,
      create a lock that will persist for the lifetime of the process
@@ -1901,7 +1929,7 @@ static void handle_existing_out_dir(afl_state_t *afl) {
   /* Next, we need to clean up <afl->out_dir>/queue/.state/ subdirectories: */
 
   fn = alloc_printf("%s/queue/.state/deterministic_done", afl->out_dir);
-  if (delete_files(fn, CASE_PREFIX)) { goto dir_cleanup_failed; }
+  if (delete_files(fn, case_prefix)) { goto dir_cleanup_failed; }
   ck_free(fn);
 
   fn = alloc_printf("%s/queue/.state/auto_extras", afl->out_dir);
@@ -1909,11 +1937,11 @@ static void handle_existing_out_dir(afl_state_t *afl) {
   ck_free(fn);
 
   fn = alloc_printf("%s/queue/.state/redundant_edges", afl->out_dir);
-  if (delete_files(fn, CASE_PREFIX)) { goto dir_cleanup_failed; }
+  if (delete_files(fn, case_prefix)) { goto dir_cleanup_failed; }
   ck_free(fn);
 
   fn = alloc_printf("%s/queue/.state/variable_behavior", afl->out_dir);
-  if (delete_files(fn, CASE_PREFIX)) { goto dir_cleanup_failed; }
+  if (delete_files(fn, case_prefix)) { goto dir_cleanup_failed; }
   ck_free(fn);
 
   /* Then, get rid of the .state subdirectory itself (should be empty by now)
@@ -1924,7 +1952,7 @@ static void handle_existing_out_dir(afl_state_t *afl) {
   ck_free(fn);
 
   fn = alloc_printf("%s/queue", afl->out_dir);
-  if (delete_files(fn, CASE_PREFIX)) { goto dir_cleanup_failed; }
+  if (delete_files(fn, case_prefix)) { goto dir_cleanup_failed; }
   ck_free(fn);
 
   /* All right, let's do <afl->out_dir>/crashes/id:* and
@@ -1971,7 +1999,7 @@ static void handle_existing_out_dir(afl_state_t *afl) {
 #ifdef AFL_PERSISTENT_RECORD
   delete_files(fn, RECORD_PREFIX);
 #endif
-  if (delete_files(fn, CASE_PREFIX)) { goto dir_cleanup_failed; }
+  if (delete_files(fn, case_prefix)) { goto dir_cleanup_failed; }
   ck_free(fn);
 
   fn = alloc_printf("%s/hangs", afl->out_dir);
@@ -2006,7 +2034,7 @@ static void handle_existing_out_dir(afl_state_t *afl) {
 #ifdef AFL_PERSISTENT_RECORD
   delete_files(fn, RECORD_PREFIX);
 #endif
-  if (delete_files(fn, CASE_PREFIX)) { goto dir_cleanup_failed; }
+  if (delete_files(fn, case_prefix)) { goto dir_cleanup_failed; }
   ck_free(fn);
 
   /* And now, for some finishing touches. */
diff --git a/src/afl-fuzz-queue.c b/src/afl-fuzz-queue.c
index 5987ad0c..2318df60 100644
--- a/src/afl-fuzz-queue.c
+++ b/src/afl-fuzz-queue.c
@@ -371,9 +371,8 @@ void mark_as_redundant(afl_state_t *afl, struct queue_entry *q, u8 state) {
     s32 fd;
 
     if (unlikely(afl->afl_env.afl_disable_redundant)) { q->disabled = 1; }
-    fd = open(fn, O_WRONLY | O_CREAT | O_EXCL, DEFAULT_PERMISSION);
-    if (fd < 0) { PFATAL("Unable to create '%s'", fn); }
-    close(fd);
+    fd = permissive_create(afl, fn);
+    if (fd >= 0) { close(fd); }
 
   } else {
 
diff --git a/src/afl-fuzz-state.c b/src/afl-fuzz-state.c
index 543fdc1c..74edaddf 100644
--- a/src/afl-fuzz-state.c
+++ b/src/afl-fuzz-state.c
@@ -626,6 +626,13 @@ void read_afl_environment(afl_state_t *afl, char **envp) {
 
             }
 
+          } else if (!strncmp(env, "AFL_SHA1_FILENAMES",
+
+                              afl_environment_variable_len)) {
+
+            afl->afl_env.afl_sha1_filenames =
+                get_afl_env(afl_environment_variables[i]) ? 1 : 0;
+
           }
 
         } else {
diff --git a/src/afl-performance.c b/src/afl-performance.c
index f730ca53..6c6e3c8b 100644
--- a/src/afl-performance.c
+++ b/src/afl-performance.c
@@ -95,3 +95,313 @@ inline u64 hash64(u8 *key, u32 len, u64 seed) {
 
 }
 
+// Public domain SHA1 implementation copied from:
+// https://github.com/x42/liboauth/blob/7001b8256cd654952ec2515b055d2c5b243be600/src/sha1.c
+
+/* This code is public-domain - it is based on libcrypt
+ * placed in the public domain by Wei Dai and other contributors.
+ */
+// gcc -Wall -DSHA1TEST -o sha1test sha1.c && ./sha1test
+
+#include <stdint.h>
+#include <string.h>
+
+#ifdef __BIG_ENDIAN__
+  #define SHA_BIG_ENDIAN
+#elif defined __LITTLE_ENDIAN__
+/* override */
+#elif defined __BYTE_ORDER
+  #if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+    #define SHA_BIG_ENDIAN
+  #endif
+#else                  // ! defined __LITTLE_ENDIAN__
+  #include <endian.h>  // machine/endian.h
+  #if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+    #define SHA_BIG_ENDIAN
+  #endif
+#endif
+
+/* header */
+
+#define HASH_LENGTH 20
+#define BLOCK_LENGTH 64
+
+typedef struct sha1nfo {
+
+  uint32_t buffer[BLOCK_LENGTH / 4];
+  uint32_t state[HASH_LENGTH / 4];
+  uint32_t byteCount;
+  uint8_t  bufferOffset;
+  uint8_t  keyBuffer[BLOCK_LENGTH];
+  uint8_t  innerHash[HASH_LENGTH];
+
+} sha1nfo;
+
+/* public API - prototypes - TODO: doxygen*/
+
+/**
+ */
+void sha1_init(sha1nfo *s);
+/**
+ */
+void sha1_writebyte(sha1nfo *s, uint8_t data);
+/**
+ */
+void sha1_write(sha1nfo *s, const char *data, size_t len);
+/**
+ */
+uint8_t *sha1_result(sha1nfo *s);
+/**
+ */
+void sha1_initHmac(sha1nfo *s, const uint8_t *key, int keyLength);
+/**
+ */
+uint8_t *sha1_resultHmac(sha1nfo *s);
+
+/* code */
+#define SHA1_K0 0x5a827999
+#define SHA1_K20 0x6ed9eba1
+#define SHA1_K40 0x8f1bbcdc
+#define SHA1_K60 0xca62c1d6
+
+void sha1_init(sha1nfo *s) {
+
+  s->state[0] = 0x67452301;
+  s->state[1] = 0xefcdab89;
+  s->state[2] = 0x98badcfe;
+  s->state[3] = 0x10325476;
+  s->state[4] = 0xc3d2e1f0;
+  s->byteCount = 0;
+  s->bufferOffset = 0;
+
+}
+
+uint32_t sha1_rol32(uint32_t number, uint8_t bits) {
+
+  return ((number << bits) | (number >> (32 - bits)));
+
+}
+
+void sha1_hashBlock(sha1nfo *s) {
+
+  uint8_t  i;
+  uint32_t a, b, c, d, e, t;
+
+  a = s->state[0];
+  b = s->state[1];
+  c = s->state[2];
+  d = s->state[3];
+  e = s->state[4];
+  for (i = 0; i < 80; i++) {
+
+    if (i >= 16) {
+
+      t = s->buffer[(i + 13) & 15] ^ s->buffer[(i + 8) & 15] ^
+          s->buffer[(i + 2) & 15] ^ s->buffer[i & 15];
+      s->buffer[i & 15] = sha1_rol32(t, 1);
+
+    }
+
+    if (i < 20) {
+
+      t = (d ^ (b & (c ^ d))) + SHA1_K0;
+
+    } else if (i < 40) {
+
+      t = (b ^ c ^ d) + SHA1_K20;
+
+    } else if (i < 60) {
+
+      t = ((b & c) | (d & (b | c))) + SHA1_K40;
+
+    } else {
+
+      t = (b ^ c ^ d) + SHA1_K60;
+
+    }
+
+    t += sha1_rol32(a, 5) + e + s->buffer[i & 15];
+    e = d;
+    d = c;
+    c = sha1_rol32(b, 30);
+    b = a;
+    a = t;
+
+  }
+
+  s->state[0] += a;
+  s->state[1] += b;
+  s->state[2] += c;
+  s->state[3] += d;
+  s->state[4] += e;
+
+}
+
+void sha1_addUncounted(sha1nfo *s, uint8_t data) {
+
+  uint8_t *const b = (uint8_t *)s->buffer;
+#ifdef SHA_BIG_ENDIAN
+  b[s->bufferOffset] = data;
+#else
+  b[s->bufferOffset ^ 3] = data;
+#endif
+  s->bufferOffset++;
+  if (s->bufferOffset == BLOCK_LENGTH) {
+
+    sha1_hashBlock(s);
+    s->bufferOffset = 0;
+
+  }
+
+}
+
+void sha1_writebyte(sha1nfo *s, uint8_t data) {
+
+  ++s->byteCount;
+  sha1_addUncounted(s, data);
+
+}
+
+void sha1_write(sha1nfo *s, const char *data, size_t len) {
+
+  for (; len--;)
+    sha1_writebyte(s, (uint8_t)*data++);
+
+}
+
+void sha1_pad(sha1nfo *s) {
+
+  // Implement SHA-1 padding (fips180-2 §5.1.1)
+
+  // Pad with 0x80 followed by 0x00 until the end of the block
+  sha1_addUncounted(s, 0x80);
+  while (s->bufferOffset != 56)
+    sha1_addUncounted(s, 0x00);
+
+  // Append length in the last 8 bytes
+  sha1_addUncounted(s, 0);  // We're only using 32 bit lengths
+  sha1_addUncounted(s, 0);  // But SHA-1 supports 64 bit lengths
+  sha1_addUncounted(s, 0);  // So zero pad the top bits
+  sha1_addUncounted(s, s->byteCount >> 29);  // Shifting to multiply by 8
+  sha1_addUncounted(
+      s, s->byteCount >> 21);  // as SHA-1 supports bitstreams as well as
+  sha1_addUncounted(s, s->byteCount >> 13);  // byte.
+  sha1_addUncounted(s, s->byteCount >> 5);
+  sha1_addUncounted(s, s->byteCount << 3);
+
+}
+
+uint8_t *sha1_result(sha1nfo *s) {
+
+  // Pad to complete the last block
+  sha1_pad(s);
+
+#ifndef SHA_BIG_ENDIAN
+  // Swap byte order back
+  int i;
+  for (i = 0; i < 5; i++) {
+
+    s->state[i] = (((s->state[i]) << 24) & 0xff000000) |
+                  (((s->state[i]) << 8) & 0x00ff0000) |
+                  (((s->state[i]) >> 8) & 0x0000ff00) |
+                  (((s->state[i]) >> 24) & 0x000000ff);
+
+  }
+
+#endif
+
+  // Return pointer to hash (20 characters)
+  return (uint8_t *)s->state;
+
+}
+
+#define HMAC_IPAD 0x36
+#define HMAC_OPAD 0x5c
+
+void sha1_initHmac(sha1nfo *s, const uint8_t *key, int keyLength) {
+
+  uint8_t i;
+  memset(s->keyBuffer, 0, BLOCK_LENGTH);
+  if (keyLength > BLOCK_LENGTH) {
+
+    // Hash long keys
+    sha1_init(s);
+    for (; keyLength--;)
+      sha1_writebyte(s, *key++);
+    memcpy(s->keyBuffer, sha1_result(s), HASH_LENGTH);
+
+  } else {
+
+    // Block length keys are used as is
+    memcpy(s->keyBuffer, key, keyLength);
+
+  }
+
+  // Start inner hash
+  sha1_init(s);
+  for (i = 0; i < BLOCK_LENGTH; i++) {
+
+    sha1_writebyte(s, s->keyBuffer[i] ^ HMAC_IPAD);
+
+  }
+
+}
+
+uint8_t *sha1_resultHmac(sha1nfo *s) {
+
+  uint8_t i;
+  // Complete inner hash
+  memcpy(s->innerHash, sha1_result(s), HASH_LENGTH);
+  // Calculate outer hash
+  sha1_init(s);
+  for (i = 0; i < BLOCK_LENGTH; i++)
+    sha1_writebyte(s, s->keyBuffer[i] ^ HMAC_OPAD);
+  for (i = 0; i < HASH_LENGTH; i++)
+    sha1_writebyte(s, s->innerHash[i]);
+  return sha1_result(s);
+
+}
+
+// End public domain SHA1 implementation
+
+void sha1(const u8 *data, size_t len, u8 *out) {
+
+  sha1nfo s;
+  sha1_init(&s);
+  sha1_write(&s, (const char *)data, len);
+  memcpy(out, sha1_result(&s), HASH_LENGTH);
+
+}
+
+char *sha1_hex(const u8 *data, size_t len) {
+
+  u8 digest[HASH_LENGTH];
+  sha1(data, len, digest);
+  u8 *hex = ck_alloc(HASH_LENGTH * 2 + 1);
+  for (size_t i = 0; i < HASH_LENGTH; ++i) {
+
+    sprintf((char *)(hex + i * 2), "%02x", digest[i]);
+
+  }
+
+  return hex;
+
+}
+
+char *sha1_hex_for_file(const char *fname, u32 len) {
+
+  int fd = open(fname, O_RDONLY);
+  if (fd < 0) { PFATAL("Unable to open '%s'", fname); }
+
+  u32 read_len = MIN(len, (u32)MAX_FILE);
+  u8 *tmp = ck_alloc(read_len);
+  ck_read(fd, tmp, read_len, fname);
+
+  close(fd);
+
+  char *hex = sha1_hex(tmp, read_len);
+  ck_free(tmp);
+  return hex;
+
+}
+
-- 
cgit 1.4.1


From 622474e9e45056c21abf4f8f39f2a6f7cc01053f Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Mon, 13 May 2024 19:51:38 +0200
Subject: disable -> no variants

---
 include/envs.h       | 4 ++--
 src/afl-fuzz-state.c | 3 +++
 src/afl-fuzz.c       | 6 +++++-
 3 files changed, 10 insertions(+), 3 deletions(-)

(limited to 'include')

diff --git a/include/envs.h b/include/envs.h
index 57f4d263..2f5b0d60 100644
--- a/include/envs.h
+++ b/include/envs.h
@@ -27,8 +27,8 @@ static char *afl_environment_variables[] = {
     "AFL_CUSTOM_INFO_PROGRAM_ARGV", "AFL_CUSTOM_INFO_PROGRAM_INPUT",
     "AFL_CUSTOM_INFO_OUT", "AFL_CXX", "AFL_CYCLE_SCHEDULES", "AFL_DEBUG",
     "AFL_DEBUG_CHILD", "AFL_DEBUG_GDB", "AFL_DEBUG_UNICORN",
-    "AFL_DISABLE_REDUNDANT", "AFL_DISABLE_TRIM",
-    "AFL_DISABLE_LLVM_INSTRUMENTATION", "AFL_DONT_OPTIMIZE",
+    "AFL_DISABLE_REDUNDANT", "AFL_NO_REDUNDANT", "AFL_DISABLE_TRIM",
+    "AFL_NO_TRIM", "AFL_DISABLE_LLVM_INSTRUMENTATION", "AFL_DONT_OPTIMIZE",
     "AFL_DRIVER_STDERR_DUPLICATE_FILENAME", "AFL_DUMB_FORKSRV",
     "AFL_EARLY_FORKSERVER", "AFL_ENTRYPOINT", "AFL_EXIT_WHEN_DONE",
     "AFL_EXIT_ON_TIME", "AFL_EXIT_ON_SEED_ISSUES", "AFL_FAST_CAL",
diff --git a/src/afl-fuzz-state.c b/src/afl-fuzz-state.c
index 74edaddf..333d57b2 100644
--- a/src/afl-fuzz-state.c
+++ b/src/afl-fuzz-state.c
@@ -295,6 +295,9 @@ void read_afl_environment(afl_state_t *afl, char **envp) {
 
           } else if (!strncmp(env, "AFL_DISABLE_REDUNDANT",
 
+                              afl_environment_variable_len) ||
+                     !strncmp(env, "AFL_NO_REDUNDANT",
+
                               afl_environment_variable_len)) {
 
             afl->afl_env.afl_disable_redundant =
diff --git a/src/afl-fuzz.c b/src/afl-fuzz.c
index 329ce942..bb9c270f 100644
--- a/src/afl-fuzz.c
+++ b/src/afl-fuzz.c
@@ -1565,7 +1565,11 @@ int main(int argc, char **argv_orig, char **envp) {
 
   setenv("__AFL_OUT_DIR", afl->out_dir, 1);
 
-  if (get_afl_env("AFL_DISABLE_TRIM")) { afl->disable_trim = 1; }
+  if (get_afl_env("AFL_DISABLE_TRIM") || get_afl_env("AFL_NO_TRIM")) {
+
+    afl->disable_trim = 1;
+
+  }
 
   if (getenv("AFL_NO_UI") && getenv("AFL_FORCE_UI")) {
 
-- 
cgit 1.4.1


From 831b8f35d5e01ef195d62c8377eb0458f2009d8f Mon Sep 17 00:00:00 2001
From: nj00001 <42004790+nj00001@users.noreply.github.com>
Date: Tue, 14 May 2024 18:23:53 +0800
Subject: Keep a backup of max_length in the afl_forkserver_t structure

---
 include/forkserver.h | 2 ++
 1 file changed, 2 insertions(+)

(limited to 'include')

diff --git a/include/forkserver.h b/include/forkserver.h
index 68907376..593e34a2 100644
--- a/include/forkserver.h
+++ b/include/forkserver.h
@@ -188,6 +188,8 @@ typedef struct afl_forkserver {
 
   u8 persistent_mode;
 
+  u32 max_length;
+
 #ifdef __linux__
   nyx_plugin_handler_t *nyx_handlers;
   char                 *out_dir_path;    /* path to the output directory     */
-- 
cgit 1.4.1


From 7d3530a22ea44eba0aa578041e41980b0d9c976a Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Tue, 14 May 2024 13:07:47 +0200
Subject: nit

---
 include/config.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'include')

diff --git a/include/config.h b/include/config.h
index a2ff68ea..3727dab1 100644
--- a/include/config.h
+++ b/include/config.h
@@ -464,7 +464,7 @@
 /* Do not change this unless you really know what you are doing. */
 
 #define MAP_SIZE (1U << MAP_SIZE_POW2)
-#if MAP_SIZE <= 65536
+#if MAP_SIZE <= 2097152
   #define MAP_INITIAL_SIZE (2 << 20)  // = 2097152
 #else
   #define MAP_INITIAL_SIZE MAP_SIZE
-- 
cgit 1.4.1


From 1db3b81d2eb855167dcf65734f8833a2329609da Mon Sep 17 00:00:00 2001
From: vanhauser-thc <vh@thc.org>
Date: Wed, 15 May 2024 14:16:44 +0200
Subject: dump cc

---
 TODO.md                                        |  1 +
 include/envs.h                                 | 24 +++++++-------
 instrumentation/SanitizerCoveragePCGUARD.so.cc |  6 +++-
 instrumentation/afl-llvm-common.cc             | 45 ++++++++++++++++++++++++++
 instrumentation/afl-llvm-common.h              |  1 +
 5 files changed, 64 insertions(+), 13 deletions(-)

(limited to 'include')

diff --git a/TODO.md b/TODO.md
index d47372b8..20f3425f 100644
--- a/TODO.md
+++ b/TODO.md
@@ -11,6 +11,7 @@
  - afl-showmap -f support
  - afl-fuzz multicore wrapper script
  - when trimming then perform crash detection
+ - cyclomatic complexity: 2 + calls + edges - blocks
 
 
 ## Should
diff --git a/include/envs.h b/include/envs.h
index 2f5b0d60..5b516905 100644
--- a/include/envs.h
+++ b/include/envs.h
@@ -21,18 +21,18 @@ static char *afl_environment_variables[] = {
     "AFL_BENCH_UNTIL_CRASH", "AFL_CAL_FAST", "AFL_CC", "AFL_CC_COMPILER",
     "AFL_CMIN_ALLOW_ANY", "AFL_CMIN_CRASHES_ONLY", "AFL_CMPLOG_ONLY_NEW",
     "AFL_CODE_END", "AFL_CODE_START", "AFL_COMPCOV_BINNAME",
-    "AFL_CMPLOG_MAX_LEN", "AFL_COMPCOV_LEVEL", "AFL_CRASH_EXITCODE",
-    "AFL_CRASHING_SEEDS_AS_NEW_CRASH", "AFL_CUSTOM_MUTATOR_LIBRARY",
-    "AFL_CUSTOM_MUTATOR_ONLY", "AFL_CUSTOM_INFO_PROGRAM",
-    "AFL_CUSTOM_INFO_PROGRAM_ARGV", "AFL_CUSTOM_INFO_PROGRAM_INPUT",
-    "AFL_CUSTOM_INFO_OUT", "AFL_CXX", "AFL_CYCLE_SCHEDULES", "AFL_DEBUG",
-    "AFL_DEBUG_CHILD", "AFL_DEBUG_GDB", "AFL_DEBUG_UNICORN",
-    "AFL_DISABLE_REDUNDANT", "AFL_NO_REDUNDANT", "AFL_DISABLE_TRIM",
-    "AFL_NO_TRIM", "AFL_DISABLE_LLVM_INSTRUMENTATION", "AFL_DONT_OPTIMIZE",
-    "AFL_DRIVER_STDERR_DUPLICATE_FILENAME", "AFL_DUMB_FORKSRV",
-    "AFL_EARLY_FORKSERVER", "AFL_ENTRYPOINT", "AFL_EXIT_WHEN_DONE",
-    "AFL_EXIT_ON_TIME", "AFL_EXIT_ON_SEED_ISSUES", "AFL_FAST_CAL",
-    "AFL_FINAL_SYNC", "AFL_FORCE_UI", "AFL_FRIDA_DEBUG_MAPS",
+    "AFL_DUMP_CYCLOMATIC_COMPLEXITY", "AFL_CMPLOG_MAX_LEN", "AFL_COMPCOV_LEVEL",
+    "AFL_CRASH_EXITCODE", "AFL_CRASHING_SEEDS_AS_NEW_CRASH",
+    "AFL_CUSTOM_MUTATOR_LIBRARY", "AFL_CUSTOM_MUTATOR_ONLY",
+    "AFL_CUSTOM_INFO_PROGRAM", "AFL_CUSTOM_INFO_PROGRAM_ARGV",
+    "AFL_CUSTOM_INFO_PROGRAM_INPUT", "AFL_CUSTOM_INFO_OUT", "AFL_CXX",
+    "AFL_CYCLE_SCHEDULES", "AFL_DEBUG", "AFL_DEBUG_CHILD", "AFL_DEBUG_GDB",
+    "AFL_DEBUG_UNICORN", "AFL_DISABLE_REDUNDANT", "AFL_NO_REDUNDANT",
+    "AFL_DISABLE_TRIM", "AFL_NO_TRIM", "AFL_DISABLE_LLVM_INSTRUMENTATION",
+    "AFL_DONT_OPTIMIZE", "AFL_DRIVER_STDERR_DUPLICATE_FILENAME",
+    "AFL_DUMB_FORKSRV", "AFL_EARLY_FORKSERVER", "AFL_ENTRYPOINT",
+    "AFL_EXIT_WHEN_DONE", "AFL_EXIT_ON_TIME", "AFL_EXIT_ON_SEED_ISSUES",
+    "AFL_FAST_CAL", "AFL_FINAL_SYNC", "AFL_FORCE_UI", "AFL_FRIDA_DEBUG_MAPS",
     "AFL_FRIDA_DRIVER_NO_HOOK", "AFL_FRIDA_EXCLUDE_RANGES",
     "AFL_FRIDA_INST_CACHE_SIZE", "AFL_FRIDA_INST_COVERAGE_ABSOLUTE",
     "AFL_FRIDA_INST_COVERAGE_FILE", "AFL_FRIDA_INST_DEBUG_FILE",
diff --git a/instrumentation/SanitizerCoveragePCGUARD.so.cc b/instrumentation/SanitizerCoveragePCGUARD.so.cc
index f88ce126..01881f28 100644
--- a/instrumentation/SanitizerCoveragePCGUARD.so.cc
+++ b/instrumentation/SanitizerCoveragePCGUARD.so.cc
@@ -195,7 +195,7 @@ class ModuleSanitizerCoverageAFL
 
   SanitizerCoverageOptions Options;
 
-  uint32_t        instr = 0, selects = 0, unhandled = 0;
+  uint32_t        instr = 0, selects = 0, unhandled = 0, dump_cc = 0;
   GlobalVariable *AFLMapPtr = NULL;
   ConstantInt    *One = NULL;
   ConstantInt    *Zero = NULL;
@@ -330,6 +330,8 @@ bool ModuleSanitizerCoverageAFL::instrumentModule(
 
   if (getenv("AFL_DEBUG")) { debug = 1; }
 
+  if (getenv("AFL_DUMP_CYCLOMATIC_COMPLEXITY")) { dump_cc = 1; }
+
   if ((isatty(2) && !getenv("AFL_QUIET")) || debug) {
 
     SAYF(cCYA "SanitizerCoveragePCGUARD" VERSION cRST "\n");
@@ -638,6 +640,8 @@ void ModuleSanitizerCoverageAFL::instrumentFunction(
   // InjectTraceForCmp(F, CmpTraceTargets);
   // InjectTraceForSwitch(F, SwitchTraceTargets);
 
+  if (dump_cc) { calcCyclomaticComplexity(&F); }
+
 }
 
 GlobalVariable *ModuleSanitizerCoverageAFL::CreateFunctionLocalArrayInSection(
diff --git a/instrumentation/afl-llvm-common.cc b/instrumentation/afl-llvm-common.cc
index 8e9e7800..ed9268dc 100644
--- a/instrumentation/afl-llvm-common.cc
+++ b/instrumentation/afl-llvm-common.cc
@@ -26,6 +26,51 @@ static std::list<std::string> allowListFunctions;
 static std::list<std::string> denyListFiles;
 static std::list<std::string> denyListFunctions;
 
+unsigned int calcCyclomaticComplexity(llvm::Function *F) {
+
+  unsigned int numBlocks = 0;
+  unsigned int numEdges = 0;
+  unsigned int numCalls = 0;
+
+  // Iterate through each basic block in the function
+  for (BasicBlock &BB : *F) {
+
+    // count all nodes == basic blocks
+    numBlocks++;
+    // Count the number of successors (outgoing edges)
+    for (BasicBlock *Succ : successors(&BB)) {
+
+      // count edges for CC
+      numEdges++;
+      (void)(Succ);
+
+    }
+
+    for (Instruction &I : BB) {
+
+      // every call is also an edge, so we need to count the calls too
+      if (isa<CallInst>(&I) || isa<InvokeInst>(&I)) { numCalls++; }
+
+    }
+
+  }
+
+  // Cyclomatic Complexity V(G) = E - N + 2P
+  // For a single function, P (number of connected components) is 1
+  // Calls are considered to be an edge
+  unsigned int CC = 2 + numCalls + numEdges - numBlocks;
+
+  // if (debug) {
+
+  fprintf(stderr, "CyclomaticComplexity for %s: %u\n",
+          F->getName().str().c_str(), CC);
+
+  //}
+
+  return CC;
+
+}
+
 char *getBBName(const llvm::BasicBlock *BB) {
 
   static char *name;
diff --git a/instrumentation/afl-llvm-common.h b/instrumentation/afl-llvm-common.h
index 23f67179..6b628d64 100644
--- a/instrumentation/afl-llvm-common.h
+++ b/instrumentation/afl-llvm-common.h
@@ -55,6 +55,7 @@ void  initInstrumentList();
 bool  isInInstrumentList(llvm::Function *F, std::string Filename);
 unsigned long long int calculateCollisions(uint32_t edges);
 void                   scanForDangerousFunctions(llvm::Module *M);
+unsigned int           calcCyclomaticComplexity(llvm::Function *F);
 
 #ifndef IS_EXTERN
   #define IS_EXTERN
-- 
cgit 1.4.1