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

[Joseph Huber via cfe-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
----------------
jhuber6 wrote:

Offloading compilation for these single-source languages pretty much just combines one "host" compilation job with N "Device" compilation jobs. Doing `--offload-device-only` and `--offload-host-only` simply does one part of that. There's probably some flags that behave differently depending on which end you're compiling on, so maybe it would be useful for separating that behavior if needed.

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