forked from OSchip/llvm-project
Emit memmove, not memcpy, for structure copies; this is unfortunately
required for correctness in cases of copying a struct to itself or to an overlapping struct (itself for cases like *a = *a, and overlapping is possible with unions). Hopefully, this won't end up being a perf issue; LLVM *should* be able to optimize memmove to memcpy in a lot of cases, and for small copies the generated code *should* be mostly comparable. (In reality, LLVM is currently horrible at optimizing memmove, but that's a bug, not a fundamental issue.) gcc currently generates wrong code; that's http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32667. llvm-svn: 51566
This commit is contained in:
parent
d3f21d165f
commit
df649f3da5
|
@ -131,7 +131,7 @@ void AggExprEmitter::EmitAggregateCopy(llvm::Value *DestPtr,
|
|||
llvm::Value *SrcPtr, QualType Ty) {
|
||||
assert(!Ty->isAnyComplexType() && "Shouldn't happen for complex");
|
||||
|
||||
// Aggregate assignment turns into llvm.memcpy.
|
||||
// Aggregate assignment turns into llvm.memmove.
|
||||
const llvm::Type *BP = llvm::PointerType::getUnqual(llvm::Type::Int8Ty);
|
||||
if (DestPtr->getType() != BP)
|
||||
DestPtr = Builder.CreateBitCast(DestPtr, BP, "tmp");
|
||||
|
@ -144,14 +144,14 @@ void AggExprEmitter::EmitAggregateCopy(llvm::Value *DestPtr,
|
|||
// FIXME: Handle variable sized types.
|
||||
const llvm::Type *IntPtr = llvm::IntegerType::get(CGF.LLVMPointerWidth);
|
||||
|
||||
llvm::Value *MemCpyOps[4] = {
|
||||
llvm::Value *MemMoveOps[4] = {
|
||||
DestPtr, SrcPtr,
|
||||
// TypeInfo.first describes size in bits.
|
||||
llvm::ConstantInt::get(IntPtr, TypeInfo.first/8),
|
||||
llvm::ConstantInt::get(llvm::Type::Int32Ty, TypeInfo.second/8)
|
||||
};
|
||||
|
||||
Builder.CreateCall(CGF.CGM.getMemCpyFn(), MemCpyOps, MemCpyOps+4);
|
||||
Builder.CreateCall(CGF.CGM.getMemMoveFn(), MemMoveOps, MemMoveOps+4);
|
||||
}
|
||||
|
||||
|
||||
|
|
|
@ -35,7 +35,8 @@ CodeGenModule::CodeGenModule(ASTContext &C, const LangOptions &LO,
|
|||
llvm::Module &M, const llvm::TargetData &TD,
|
||||
Diagnostic &diags, bool GenerateDebugInfo)
|
||||
: Context(C), Features(LO), TheModule(M), TheTargetData(TD), Diags(diags),
|
||||
Types(C, M, TD), MemCpyFn(0), MemSetFn(0), CFConstantStringClassRef(0) {
|
||||
Types(C, M, TD), MemCpyFn(0), MemMoveFn(0), MemSetFn(0),
|
||||
CFConstantStringClassRef(0) {
|
||||
//TODO: Make this selectable at runtime
|
||||
Runtime = CreateObjCRuntime(M,
|
||||
getTypes().ConvertType(getContext().IntTy),
|
||||
|
@ -589,6 +590,17 @@ llvm::Function *CodeGenModule::getMemCpyFn() {
|
|||
return MemCpyFn = getIntrinsic(IID);
|
||||
}
|
||||
|
||||
llvm::Function *CodeGenModule::getMemMoveFn() {
|
||||
if (MemMoveFn) return MemMoveFn;
|
||||
llvm::Intrinsic::ID IID;
|
||||
switch (Context.Target.getPointerWidth(0)) {
|
||||
default: assert(0 && "Unknown ptr width");
|
||||
case 32: IID = llvm::Intrinsic::memmove_i32; break;
|
||||
case 64: IID = llvm::Intrinsic::memmove_i64; break;
|
||||
}
|
||||
return MemMoveFn = getIntrinsic(IID);
|
||||
}
|
||||
|
||||
llvm::Function *CodeGenModule::getMemSetFn() {
|
||||
if (MemSetFn) return MemSetFn;
|
||||
llvm::Intrinsic::ID IID;
|
||||
|
|
|
@ -59,6 +59,7 @@ class CodeGenModule {
|
|||
CGDebugInfo *DebugInfo;
|
||||
|
||||
llvm::Function *MemCpyFn;
|
||||
llvm::Function *MemMoveFn;
|
||||
llvm::Function *MemSetFn;
|
||||
llvm::DenseMap<const Decl*, llvm::Constant*> GlobalDeclMap;
|
||||
std::vector<const NamedDecl*> StaticDecls;
|
||||
|
@ -101,6 +102,7 @@ public:
|
|||
/// array containing the literal. The result is pointer to array type.
|
||||
llvm::Constant *GetAddrOfConstantString(const std::string& str);
|
||||
llvm::Function *getMemCpyFn();
|
||||
llvm::Function *getMemMoveFn();
|
||||
llvm::Function *getMemSetFn();
|
||||
llvm::Function *getIntrinsic(unsigned IID, const llvm::Type **Tys = 0,
|
||||
unsigned NumTys = 0);
|
||||
|
|
Loading…
Reference in New Issue