From 70da0c2e405102dc044cb4bed0f4f1e847c90d0b Mon Sep 17 00:00:00 2001 From: vanhauser-thc Date: Wed, 10 May 2023 16:09:18 +0200 Subject: better tritondse support --- include/envs.h | 4 ++++ 1 file changed, 4 insertions(+) (limited to 'include/envs.h') 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", -- cgit 1.4.1 From 1429c9724efb62e5ac90ec27d93a64c28632ba5d Mon Sep 17 00:00:00 2001 From: Junwha Date: Wed, 2 Aug 2023 02:59:07 +0900 Subject: Add option for treating crashing input as new crash Signed-off-by: Junwha Hong --- 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/envs.h') 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 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/envs.h') 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 08a6fd7c29489f5477f50b94d7a0e425f64fef34 Mon Sep 17 00:00:00 2001 From: Sergej Schumilo 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/envs.h') 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 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/envs.h') 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 8823f22a9c87123c1bfcc5bff10044de4c7a4a1f Mon Sep 17 00:00:00 2001 From: marc 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/envs.h') 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 549e5dd9269238ac43ff482d439f7f671946185c Mon Sep 17 00:00:00 2001 From: vanhauser-thc 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/envs.h') 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 48bff70cdd7fb7aa8333533e01a372205c670a4f Mon Sep 17 00:00:00 2001 From: vanhauser-thc 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/envs.h') 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 cf458a7d25dc3448b94ffe08d3d89531fc8d4818 Mon Sep 17 00:00:00 2001 From: Jesse Schwartzentruber 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/envs.h') 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 f290bdd83ba1b540396db2215d133e5a40570419 Mon Sep 17 00:00:00 2001 From: vanhauser-thc 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/envs.h') 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 1fc1b32db261b27cf14f0d1d7f77a06854b7376c Mon Sep 17 00:00:00 2001 From: vanhauser-thc 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/envs.h') 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" +"'\"><\""))) { 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 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/envs.h') 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 34a3060b0fee900331decce0bae55d4caefafc01 Mon Sep 17 00:00:00 2001 From: vanhauser-thc 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/envs.h') 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 369fce9c85bf3b850a7109e4604fee71f694d2cb Mon Sep 17 00:00:00 2001 From: vanhauser-thc 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/envs.h') 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 -#include -#include + #include + #include + #include #else -#include -#include -#include + #include + #include + #include #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() #include -#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 /* for assert() */ -#include /* for bool */ -#include /* for memcpy() */ +#include /* for assert() */ +#include /* for bool */ +#include /* 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 -#endif + #if defined(__ia32__) || defined(__e2k__) + #include + #endif -#if defined(__ia32__) && !defined(__cpuid_count) -#include -#endif + #if defined(__ia32__) && !defined(__cpuid_count) + #include + #endif -#if defined(__e2k__) -#include -#endif + #if defined(__e2k__) + #include + #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 -#include -#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 + #include + #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 ลปukowski @@ -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 ลปukowski @@ -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 * #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 * #include * #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 * #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 /* size_t */ + /* **************************** + * Common basic types + ******************************/ + #include /* 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 include */ +} XXH_errorcode; + + /*-********************************************************************** + * 32-bit hash + ************************************************************************/ + #if defined(XXH_DOXYGEN) /* Don't show 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 - typedef uint32_t XXH32_hash_t; - -#else -# include -# 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 +typedef uint32_t XXH32_hash_t; + + #else + #include + #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 */ + #ifndef XXH_NO_LONG_LONG + /*-********************************************************************** + * 64-bit hash + ************************************************************************/ + #if defined(XXH_DOXYGEN) /* don't include */ /*! * @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 - typedef uint64_t XXH64_hash_t; -#else -# include -# 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 +typedef uint64_t XXH64_hash_t; + #else + #include + #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 -# 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 + #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 +} -/*! - * @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 + /* + * Modify the local functions below should you wish to use + * different memory routines for malloc() and free() + */ + #include /*! * @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 /* 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 /* 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 /*! * @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 - 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 /* 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 /* 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 +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= 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 -# 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()) /* WASM SIMD128 via SIMDe */ -# define inline __inline__ /* circumvent a clang bug */ -# include -# undef inline -# elif defined(__AVX2__) -# include -# elif defined(__SSE2__) -# include -# endif -#endif + return XXH_readBE64(src); -#if defined(_MSC_VER) -# include -#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 + #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()) /* WASM SIMD128 via SIMDe */ + #define inline __inline__ /* circumvent a clang bug */ + #include + #undef inline + #elif defined(__AVX2__) + #include + #elif defined(__SSE2__) + #include + #endif + #endif + + #if defined(_MSC_VER) + #include + #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()) /* 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()) /* 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 -# else -# include -# 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 + #else + #include + #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 /* 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 /* 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_() and it calls + * XXH3_accumulate_512_(). + * + * 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_() and it calls - * XXH3_accumulate_512_(). - * - * 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 /* memcmp, memcpy */ + #endif /* !XXH_NO_STREAM */ + /* 128-bit utility functions */ + + #include /* 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; ninit_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 52e19d35fac636f9ea4679d402b5eaabaa74aa0a Mon Sep 17 00:00:00 2001 From: "Christian Holler (:decoder)" 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/envs.h') 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 26eaf53a832be0b12dadbbd290b4a7e676818347 Mon Sep 17 00:00:00 2001 From: vanhauser-thc 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/envs.h') 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 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/envs.h') 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 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 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 +#include + +#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 // 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 ยง5.1.1) + + // 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 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/envs.h') 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 1db3b81d2eb855167dcf65734f8833a2329609da Mon Sep 17 00:00:00 2001 From: vanhauser-thc 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/envs.h') 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 allowListFunctions; static std::list denyListFiles; static std::list 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(&I) || isa(&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