yxsamliu wrote:

The refactoring looks good to me.

One concern: for the AMD SPIR-V path, the device cc1 triple is still 
`spirv64-amd-amdhsa`, so it uses the SPIR/SPIR-V address-space map. Should 
`LangAS::amdgpu_barrier` also be mapped to local AS 3 there, like 
`cuda_shared`, with a CodeGen test? Otherwise a named barrier global for the 
AMD SPIR-V path may fall back to the default address space.

https://github.com/llvm/llvm-project/pull/207687
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to