diff options
Diffstat (limited to 'clang/lib/CodeGen/Targets/AMDGPU.cpp')
| -rw-r--r-- | clang/lib/CodeGen/Targets/AMDGPU.cpp | 20 |
1 files changed, 20 insertions, 0 deletions
diff --git a/clang/lib/CodeGen/Targets/AMDGPU.cpp b/clang/lib/CodeGen/Targets/AMDGPU.cpp index 0fcbf7e..16d5919 100644 --- a/clang/lib/CodeGen/Targets/AMDGPU.cpp +++ b/clang/lib/CodeGen/Targets/AMDGPU.cpp @@ -402,6 +402,26 @@ void AMDGPUTargetCodeGenInfo::setFunctionDeclAttributes( F->addFnAttr("amdgpu-max-num-workgroups", AttrVal.str()); } + + if (auto *Attr = FD->getAttr<CUDAClusterDimsAttr>()) { + auto GetExprVal = [&](const auto &E) { + return E ? E->EvaluateKnownConstInt(M.getContext()).getExtValue() : 1; + }; + unsigned X = GetExprVal(Attr->getX()); + unsigned Y = GetExprVal(Attr->getY()); + unsigned Z = GetExprVal(Attr->getZ()); + llvm::SmallString<32> AttrVal; + llvm::raw_svector_ostream OS(AttrVal); + OS << X << ',' << Y << ',' << Z; + F->addFnAttr("amdgpu-cluster-dims", AttrVal.str()); + } + + // OpenCL doesn't support cluster feature. + const TargetInfo &TTI = M.getContext().getTargetInfo(); + if ((IsOpenCLKernel && + TTI.hasFeatureEnabled(TTI.getTargetOpts().FeatureMap, "clusters")) || + FD->hasAttr<CUDANoClusterAttr>()) + F->addFnAttr("amdgpu-cluster-dims", "0,0,0"); } void AMDGPUTargetCodeGenInfo::setTargetAttributes( |