llvm-project

12e8e0b1 - [AMDGPU] Correctly use the auxiliary toolchain to include libc++ (#109366)

[AMDGPU] Correctly use the auxiliary toolchain to include libc++ (#109366) Summary: Now that we have a functional build for `libc++` on the GPU, it will now find the target specific headers in `include/amdgcn-amd-amdhsa`. This is a problem for offloading via OpenMP because we need the CPU and GPU headers to match exactly. All the other toolchains forward this correctly except the AMDGPU OpenMP one, fix this by overriding it to use the host toolchain instead of the device one, so the triple is not returned as `amdgcn-amd-amdhsa`.