[libcxx-commits] [libcxx] [libcxx] Adjust inline assembly constraints for the AMDGPU target (PR #101747)

[Mark de Wever via libcxx-commits]

mordante wrote:

> Okay, so I think the problematic cases for this are caused by the fact that the AMDGPU backend can't handle `char` as a valid register target https://godbolt.org/z/E9Esc5bf5. I wonder if there's a good way to handle that, it only causes two or so to fail.

Maybe by adding a specializations for 'char' something like https://godbolt.org/z/xGxYjaKKa for AMDGPU only with comments to explain the issue.

https://github.com/llvm/llvm-project/pull/101747