Re-commit r139643.

Make clang use Acquire loads and Release stores where necessary.



git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@139650 91177308-0d34-0410-b5e6-96231b3b80d8

lib/CodeGen/CGBuiltin.cpp

@@ -900,13 +900,14 @@ RValue CodeGenFunction::EmitBuiltinExpr(const FunctionDecl *FD,
   case Builtin::BI__sync_lock_release_8:
   case Builtin::BI__sync_lock_release_16: {
     Value *Ptr = EmitScalarExpr(E->getArg(0));
-    llvm::Type *ElTy =
+    llvm::Type *ElLLVMTy =
       cast<llvm::PointerType>(Ptr->getType())->getElementType();
     llvm::StoreInst *Store = 
-      Builder.CreateStore(llvm::Constant::getNullValue(ElTy), Ptr);
-    // FIXME: This is completely, utterly wrong; it only even sort-of works
-    // on x86.
-    Store->setVolatile(true);
+      Builder.CreateStore(llvm::Constant::getNullValue(ElLLVMTy), Ptr);
+    QualType ElTy = E->getArg(0)->getType()->getPointeeType();
+    CharUnits StoreSize = getContext().getTypeSizeInChars(ElTy);
+    Store->setAlignment(StoreSize.getQuantity());
+    Store->setAtomic(llvm::Release);
     return RValue::get(0);
   }
 

lib/CodeGen/ItaniumCXXABI.cpp

@@ -1130,20 +1130,27 @@ void ItaniumCXXABI::EmitGuardedInit(CodeGenFunction &CGF,
   } else {
     // Load the first byte of the guard variable.
     llvm::Type *PtrTy = Builder.getInt8PtrTy();
-    llvm::Value *V = 
+    llvm::LoadInst *LI = 
       Builder.CreateLoad(Builder.CreateBitCast(GuardVariable, PtrTy), "tmp");
-
-    IsInitialized = Builder.CreateIsNull(V, "guard.uninitialized");
+    LI->setAlignment(1);
+
+    // Itanium ABI:
+    //   An implementation supporting thread-safety on multiprocessor
+    //   systems must also guarantee that references to the initialized
+    //   object do not occur before the load of the initialization flag.
+    //
+    // In LLVM, we do this by marking the load Acquire.
+    if (threadsafe)
+      LI->setAtomic(llvm::Acquire);
+
+    IsInitialized = Builder.CreateIsNull(LI, "guard.uninitialized");
   }
 
   llvm::BasicBlock *InitCheckBlock = CGF.createBasicBlock("init.check");
   llvm::BasicBlock *EndBlock = CGF.createBasicBlock("init.end");
 
-  llvm::BasicBlock *NoCheckBlock = EndBlock;
-  if (threadsafe) NoCheckBlock = CGF.createBasicBlock("init.barrier");
-
   // Check if the first byte of the guard variable is zero.
-  Builder.CreateCondBr(IsInitialized, InitCheckBlock, NoCheckBlock);
+  Builder.CreateCondBr(IsInitialized, InitCheckBlock, EndBlock);
 
   CGF.EmitBlock(InitCheckBlock);
 
@@ -1177,23 +1184,5 @@ void ItaniumCXXABI::EmitGuardedInit(CodeGenFunction &CGF,
     Builder.CreateStore(llvm::ConstantInt::get(GuardTy, 1), GuardVariable);
   }
 
-  // Emit an acquire memory barrier if using thread-safe statics:
-  // Itanium ABI:
-  //   An implementation supporting thread-safety on multiprocessor
-  //   systems must also guarantee that references to the initialized
-  //   object do not occur before the load of the initialization flag.
-  if (threadsafe) {
-    Builder.CreateBr(EndBlock);
-    CGF.EmitBlock(NoCheckBlock);
-
-    llvm::Value *_false = Builder.getFalse();
-    llvm::Value *_true = Builder.getTrue();
-
-    Builder.CreateCall5(CGM.getIntrinsic(llvm::Intrinsic::memory_barrier),
-                        /* load-load, load-store */ _true, _true,
-                        /* store-load, store-store */ _false, _false,
-                        /* device or I/O */ _false);
-  }
-
   CGF.EmitBlock(EndBlock);
 }

test/CodeGen/Atomics.c

@@ -192,13 +192,12 @@ void test_lock (void)
 
   __sync_synchronize (); // CHECK: fence seq_cst
 
-  // FIXME: These are wrong!
-  __sync_lock_release (&sc); // CHECK: store volatile
-  __sync_lock_release (&uc); // CHECK: store volatile
-  __sync_lock_release (&ss); // CHECK: store volatile
-  __sync_lock_release (&us); // CHECK: store volatile
-  __sync_lock_release (&si); // CHECK: store volatile
-  __sync_lock_release (&ui); // CHECK: store volatile
-  __sync_lock_release (&sll); // CHECK: store volatile
-  __sync_lock_release (&ull); // CHECK: store volatile
+  __sync_lock_release (&sc); // CHECK: store atomic {{.*}} release, align 1
+  __sync_lock_release (&uc); // CHECK: store atomic {{.*}} release, align 1
+  __sync_lock_release (&ss); // CHECK: store atomic {{.*}} release, align 2
+  __sync_lock_release (&us); /// CHECK: store atomic {{.*}} release, align 2
+  __sync_lock_release (&si); // CHECK: store atomic {{.*}} release, align 4
+  __sync_lock_release (&ui); // CHECK: store atomic {{.*}} release, align 4
+  __sync_lock_release (&sll); // CHECK: store atomic {{.*}} release, align 8
+  __sync_lock_release (&ull); // CHECK: store atomic {{.*}} release, align 8
 }

test/CodeGen/atomic.c

@@ -75,8 +75,7 @@ int atomic(void) {
   // CHECK: cmpxchg i32* null, i32 0, i32 0 seq_cst
   
   __sync_lock_release(&val);
-  // FIXME: WRONG!
-  // CHECK: store volatile i32 0, i32* 
+  // CHECK: store atomic {{.*}} release, align 4
   
   __sync_synchronize ();
   // CHECK: fence seq_cst
@@ -88,8 +87,7 @@ int atomic(void) {
 void release_return(int *lock) {
   // Ensure this is actually returning void all the way through.
   return __sync_lock_release(lock);
-  // FIXME: WRONG!
-  // CHECK: store volatile i32 0, i32* 
+  // CHECK: store atomic {{.*}} release, align 4
 }
 
 

test/CodeGenCXX/static-init.cpp

@@ -12,13 +12,11 @@ struct A {
 };
 
 void f() {
+  // CHECK: load atomic i8* bitcast (i64* @_ZGVZ1fvE1a to i8*) acquire, align 1
   // CHECK: call i32 @__cxa_guard_acquire
   // CHECK: call void @_ZN1AC1Ev
   // CHECK: call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.A*)* @_ZN1AD1Ev to void (i8*)*), i8* getelementptr inbounds (%struct.A* @_ZZ1fvE1a, i32 0, i32 0), i8* bitcast (i8** @__dso_handle to i8*))
   // CHECK: call void @__cxa_guard_release
-
-  // rdar://problem/9496726
-  // CHECK: call void @llvm.memory.barrier(i1 true, i1 true, i1 false, i1 false, i1 false)
   static A a;
 }