[libc-commits] [libc] [libc] Fix linking of AMDGPU device runtime control constants for math (PR #65676)

[Joseph Huber via libc-commits]

jhuber6 wrote:

> So technically ocml functions are now subtarget independent. It's ockl that's still problematic. They still could have the base wave size and other ABI incompatibilities. The ISA versions are only used for detecting if FMA is fast or not, which barely matters anymore. We could find a way to avoid that

Yeah, none of the `ocml` functions use the wavesize. We'd still need to set the math optimization flags though.

https://github.com/llvm/llvm-project/pull/65676