The GPU support for the libc generates all its own headers. Since
these headers use the same names as the system headers we need to make
sure that they are separate. Currently, we either use the system headers
on the GPU or the GPU headers on the system. This patch makes them
explicitly separate. A follow-up patch will then make clang look in
this folder by default.
Details
Details
Diff Detail
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
Comment Actions
I prefer a name which is more in line with the triple formats listed here: https://clang.llvm.org/docs/CrossCompilation.html#target-triple
- gpu-none-llvm
- gpu-amd-none-llvm or gpu-nvidia-none-llvm
Comment Actions
These should be common between the gpu architecture, any target specific stuff should be hidden behind the __NVPTX__ and __AMDGPU__ macros. I'm in favor of gpu-llvm or llvm-gpu however. Since it makes it clear that these are GPU specific headers, separate from the system headers, and they were provided by and used in the LLVM toolchain.
Comment Actions
I think gpu-llvm is almost ideal: It is for the GPU target and contains LLVM related/specific bits. Since target triples normally contains three components, I am tempted to say this should be gpu-none-llvm. The none kind of also indicates it is vendor agnostic.