diff options
Diffstat (limited to 'clang/lib/CodeGen')
| -rw-r--r-- | clang/lib/CodeGen/CGAtomic.cpp | 12 | ||||
| -rw-r--r-- | clang/lib/CodeGen/CGOpenMPRuntime.cpp | 5 | ||||
| -rw-r--r-- | clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp | 11 |
3 files changed, 22 insertions, 6 deletions
diff --git a/clang/lib/CodeGen/CGAtomic.cpp b/clang/lib/CodeGen/CGAtomic.cpp index 4a3446a..d95dab3 100644 --- a/clang/lib/CodeGen/CGAtomic.cpp +++ b/clang/lib/CodeGen/CGAtomic.cpp @@ -507,6 +507,18 @@ static llvm::Value *EmitPostAtomicMinMax(CGBuilderTy &Builder, bool IsSigned, llvm::Value *OldVal, llvm::Value *RHS) { + const bool IsFP = OldVal->getType()->isFloatingPointTy(); + + if (IsFP) { + llvm::Intrinsic::ID IID = (Op == AtomicExpr::AO__atomic_max_fetch || + Op == AtomicExpr::AO__scoped_atomic_max_fetch) + ? llvm::Intrinsic::maxnum + : llvm::Intrinsic::minnum; + + return Builder.CreateBinaryIntrinsic(IID, OldVal, RHS, llvm::FMFSource(), + "newval"); + } + llvm::CmpInst::Predicate Pred; switch (Op) { default: diff --git a/clang/lib/CodeGen/CGOpenMPRuntime.cpp b/clang/lib/CodeGen/CGOpenMPRuntime.cpp index fa94692..1ff2be7 100644 --- a/clang/lib/CodeGen/CGOpenMPRuntime.cpp +++ b/clang/lib/CodeGen/CGOpenMPRuntime.cpp @@ -1762,8 +1762,11 @@ void CGOpenMPRuntime::emitDeclareTargetFunction(const FunctionDecl *FD, // access its value. llvm::GlobalValue *Addr = GV; if (CGM.getLangOpts().OpenMPIsTargetDevice) { + llvm::PointerType *FnPtrTy = llvm::PointerType::get( + CGM.getLLVMContext(), + CGM.getModule().getDataLayout().getProgramAddressSpace()); Addr = new llvm::GlobalVariable( - CGM.getModule(), CGM.VoidPtrTy, + CGM.getModule(), FnPtrTy, /*isConstant=*/true, llvm::GlobalValue::ExternalLinkage, GV, Name, nullptr, llvm::GlobalValue::NotThreadLocal, CGM.getModule().getDataLayout().getDefaultGlobalsAddressSpace()); diff --git a/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp b/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp index 3613b6a..fddeba9 100644 --- a/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp +++ b/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp @@ -7,7 +7,7 @@ //===----------------------------------------------------------------------===// // // This provides a generalized class for OpenMP runtime code generation -// specialized by GPU targets NVPTX and AMDGCN. +// specialized by GPU targets NVPTX, AMDGCN and SPIR-V. // //===----------------------------------------------------------------------===// @@ -1242,12 +1242,13 @@ void CGOpenMPRuntimeGPU::emitParallelCall( CGBuilderTy &Bld = CGF.Builder; llvm::Value *NumThreadsVal = NumThreads; llvm::Function *WFn = WrapperFunctionsMap[OutlinedFn]; - llvm::Value *ID = llvm::ConstantPointerNull::get(CGM.Int8PtrTy); - if (WFn) - ID = Bld.CreateBitOrPointerCast(WFn, CGM.Int8PtrTy); - llvm::Type *FnPtrTy = llvm::PointerType::get( + llvm::PointerType *FnPtrTy = llvm::PointerType::get( CGF.getLLVMContext(), CGM.getDataLayout().getProgramAddressSpace()); + llvm::Value *ID = llvm::ConstantPointerNull::get(FnPtrTy); + if (WFn) + ID = Bld.CreateBitOrPointerCast(WFn, FnPtrTy); + llvm::Value *FnPtr = Bld.CreateBitOrPointerCast(OutlinedFn, FnPtrTy); // Create a private scope that will globalize the arguments |