https://github.com/jhuber6 created https://github.com/llvm/llvm-project/pull/191311
Summary: Previously we used the auto-forwarding mechanism to handle options like forwarding --cuda-path. The problem is that this went over the toolchain options and that meant if someone used just bare `--offload-link` there would be no CUDA or ROCm toolchain to figure out if we should forward it. Just do this unconditionally for all toolchains, there's no harm in setting it if it's unused. Fixes: https://github.com/llvm/llvm-project/issues/190979 >From d49c89e4d74ba19c19327eaa92cb62622cc79b75 Mon Sep 17 00:00:00 2001 From: Joseph Huber <[email protected]> Date: Thu, 9 Apr 2026 17:07:59 -0500 Subject: [PATCH] [Clang] Pass toolchain paths unconditionally in linker wrapper Summary: Previously we used the auto-forwarding mechanism to handle options like forwarding --cuda-path. The problem is that this went over the toolchain options and that meant if someone used just bare `--offload-link` there would be no CUDA or ROCm toolchain to figure out if we should forward it. Just do this unconditionally for all toolchains, there's no harm in setting it if it's unused. Fixes: https://github.com/llvm/llvm-project/issues/190979 --- clang/lib/Driver/ToolChains/Clang.cpp | 11 ++++++++--- clang/test/Driver/openmp-offload-gpu.c | 9 +++++++++ 2 files changed, 17 insertions(+), 3 deletions(-) diff --git a/clang/lib/Driver/ToolChains/Clang.cpp b/clang/lib/Driver/ToolChains/Clang.cpp index 0d5722cd536f2..701c344ac047a 100644 --- a/clang/lib/Driver/ToolChains/Clang.cpp +++ b/clang/lib/Driver/ToolChains/Clang.cpp @@ -9391,8 +9391,6 @@ void LinkerWrapper::ConstructJob(Compilation &C, const JobAction &JA, // compilation job. const llvm::DenseSet<unsigned> CompilerOptions{ OPT_v, - OPT_cuda_path_EQ, - OPT_rocm_path_EQ, OPT_hip_path_EQ, OPT_O_Group, OPT_g_Group, @@ -9549,9 +9547,16 @@ void LinkerWrapper::ConstructJob(Compilation &C, const JobAction &JA, // also `ld.lld`. if (Args.hasArg(options::OPT_v) && JA.getType() != types::TY_HIP_FATBIN) CmdArgs.push_back("--wrapper-verbose"); - if (Arg *A = Args.getLastArg(options::OPT_cuda_path_EQ)) + if (Arg *A = Args.getLastArg(options::OPT_cuda_path_EQ)) { CmdArgs.push_back( Args.MakeArgString(Twine("--cuda-path=") + A->getValue())); + CmdArgs.push_back(Args.MakeArgString( + Twine("--device-compiler=--cuda-path=") + A->getValue())); + } + if (Arg *A = Args.getLastArg(options::OPT_rocm_path_EQ)) { + CmdArgs.push_back(Args.MakeArgString( + Twine("--device-compiler=--rocm-path=") + A->getValue())); + } // Construct the link job so we can wrap around it. Linker->ConstructJob(C, JA, Output, Inputs, Args, LinkingOutput); diff --git a/clang/test/Driver/openmp-offload-gpu.c b/clang/test/Driver/openmp-offload-gpu.c index e057959d62044..368e4dc213ef7 100644 --- a/clang/test/Driver/openmp-offload-gpu.c +++ b/clang/test/Driver/openmp-offload-gpu.c @@ -428,3 +428,12 @@ // RUN: | FileCheck --check-prefix=NO-PROFILE %s // // NO-PROFILE-NOT: --device-compiler=amdgcn-amd-amdhsa=-fprofile-generate + +// +// Check that --cuda-path and --rocm-path are forwarded unconditionally +// +// RUN: %clang -### --target=x86_64-unknown-linux-gnu -fopenmp=libomp \ +// RUN: --cuda-path=%S/Inputs/CUDA_102/usr/local/cuda --rocm-path=%S/Inputs/rocm \ +// RUN: --offload-arch=sm_89 --offload-arch=gfx906 -nogpulib -nogpuinc %s 2>&1 \ +// RUN: | FileCheck --check-prefix=CUDA-PATH %s +// CUDA-PATH: clang-linker-wrapper{{.*}} "--device-compiler=--cuda-path={{.*}}"{{.*}}"--device-compiler=--rocm-path={{.*}}" _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
