4 files changed, 49 insertions, 9 deletions

diff --git a/lib/Core/Executor.cpp b/lib/Core/Executor.cpp
index d755403f..8b885527 100644
--- a/lib/Core/Executor.cpp
+++ b/lib/Core/Executor.cpp
@@ -2543,9 +2543,15 @@ void Executor::executeInstruction(ExecutionState &state, KInstruction *ki) {
     bindLocal(ki, state, result);
     break;
   }
- 
-    // Other instructions...
-    // Unhandled
+#if LLVM_VERSION_CODE >= LLVM_VERSION(3, 3)
+  case Instruction::Fence: {
+    // Ignore for now
+    break;
+  }
+#endif
+
+  // Other instructions...
+  // Unhandled
   case Instruction::ExtractElement:
   case Instruction::InsertElement:
   case Instruction::ShuffleVector:
diff --git a/lib/Module/Passes.h b/lib/Module/Passes.h
index accb64d0..4f1a1453 100644
--- a/lib/Module/Passes.h
+++ b/lib/Module/Passes.h
@@ -21,8 +21,9 @@
 #include "llvm/Instructions.h"
 #include "llvm/Module.h"
 #endif
-#include "llvm/Pass.h"
+#include "llvm/ADT/Triple.h"
 #include "llvm/CodeGen/IntrinsicLowering.h"
+#include "llvm/Pass.h"
 
 namespace llvm {
   class Function;
@@ -46,6 +47,8 @@ class RaiseAsmPass : public llvm::ModulePass {
 
   const llvm::TargetLowering *TLI;
 
+  llvm::Triple triple;
+
   llvm::Function *getIntrinsic(llvm::Module &M,
                                unsigned IID,
                                LLVM_TYPE_Q llvm::Type **Tys,
diff --git a/lib/Module/RaiseAsm.cpp b/lib/Module/RaiseAsm.cpp
index 12e5479b..7c0e6ccf 100644
--- a/lib/Module/RaiseAsm.cpp
+++ b/lib/Module/RaiseAsm.cpp
@@ -9,12 +9,16 @@
 
 #include "Passes.h"
 #include "klee/Config/Version.h"
+#include "klee/Internal/Support/ErrorHandling.h"
 #if LLVM_VERSION_CODE >= LLVM_VERSION(3, 3)
+#include "llvm/IR/IRBuilder.h"
 #include "llvm/IR/InlineAsm.h"
 #include "llvm/IR/LLVMContext.h"
+#include "llvm/IR/Instructions.h"
 #else
 #include "llvm/InlineAsm.h"
 #include "llvm/LLVMContext.h"
+#include "llvm/Support/IRBuilder.h"
 #endif
 
 #include "llvm/Support/raw_ostream.h"
@@ -46,10 +50,32 @@ Function *RaiseAsmPass::getIntrinsic(llvm::Module &M,
 // FIXME: This should just be implemented as a patch to
 // X86TargetAsmInfo.cpp, then everyone will benefit.
 bool RaiseAsmPass::runOnInstruction(Module &M, Instruction *I) {
-  if (CallInst *ci = dyn_cast<CallInst>(I)) {
-    if (InlineAsm *ia = dyn_cast<InlineAsm>(ci->getCalledValue())) {
-      (void) ia;
-      return TLI && TLI->ExpandInlineAsm(ci);
+  // We can just raise inline assembler using calls
+  CallInst *ci = dyn_cast<CallInst>(I);
+  if (!ci)
+    return false;
+
+  InlineAsm *ia = dyn_cast<InlineAsm>(ci->getCalledValue());
+  if (!ia)
+    return false;
+
+  // Try to use existing infrastructure
+  if (!TLI)
+    return false;
+
+  if (TLI->ExpandInlineAsm(ci))
+    return true;
+
+  if (triple.getArch() == llvm::Triple::x86_64 &&
+      triple.getOS() == llvm::Triple::Linux) {
+
+    if (ia->getAsmString() == "" && ia->hasSideEffects()) {
+#if LLVM_VERSION_CODE >= LLVM_VERSION(3, 3)
+      IRBuilder<> Builder(I);
+      Builder.CreateFence(llvm::SequentiallyConsistent);
+#endif
+      I->eraseFromParent();
+      return true;
     }
   }
 
@@ -66,9 +92,10 @@ bool RaiseAsmPass::runOnModule(Module &M) {
   std::string HostTriple = llvm::sys::getHostTriple();
 #endif
   const Target *NativeTarget = TargetRegistry::lookupTarget(HostTriple, Err);
+
   TargetMachine * TM = 0;
   if (NativeTarget == 0) {
-    llvm::errs() << "Warning: unable to select native target: " << Err << "\n";
+    klee_warning("Warning: unable to select native target: %s", Err.c_str());
     TLI = 0;
   } else {
 #if LLVM_VERSION_CODE >= LLVM_VERSION(3, 1)
@@ -80,6 +107,8 @@ bool RaiseAsmPass::runOnModule(Module &M) {
     TM = NativeTarget->createTargetMachine(HostTriple, "");
 #endif
     TLI = TM->getTargetLowering();
+
+    triple = llvm::Triple(HostTriple);
   }
   
   for (Module::iterator fi = M.begin(), fe = M.end(); fi != fe; ++fi) {
diff --git a/test/Feature/RaiseAsm.c b/test/Feature/RaiseAsm.c
index 6a4b7b3c..f8ba173a 100644
--- a/test/Feature/RaiseAsm.c
+++ b/test/Feature/RaiseAsm.c
@@ -36,5 +36,7 @@ int main() {
   assert(ui16 == byteswap_uint16(byteswap_uint16_asm(ui16)));
   assert(ui32 == byteswap_uint32(byteswap_uint32_asm(ui32)));
 
+  __asm__ __volatile__("" : : : "memory");
+
   return 0;
 }