[Openmp-commits] [lldb] [clang] [clang-tools-extra] [openmp] [lld] [libc] [libcxx] [mlir] [pstl] [compiler-rt] [llvm] [Driver] Test ignored target-specific options for AMDGPU/NVPTX (PR #79222)

[Fangrui Song via Openmp-commits]

================
@@ -0,0 +1,7 @@
+/// Some target-specific options are ignored for GPU, so %clang exits with code 0.
+// DEFINE: %{gpu_opts} = --cuda-gpu-arch=sm_60 --cuda-path=%S/Inputs/CUDA/usr/local/cuda --no-cuda-version-check
+// DEFINE: %{check} = %clang -### -c %{gpu_opts} -mcmodel=medium %s
+// RUN: %{check} -fbasic-block-sections=all
+
+// REDEFINE: %{gpu_opts} = -x hip --rocm-path=%S/Inputs/rocm -nogpulib
----------------
MaskRay wrote:

Added!

https://github.com/llvm/llvm-project/pull/79222