[clang] [llvm] [AMDGPU] Enable sin/cos to sincos folding for HIP (PR #181774)

[Shilei Tian via cfe-commits]

https://github.com/shiltian commented:

I think the proposed approach is really hacky. Maybe we should try not to eliminate functions that are not used yet at the builtin link time, or just use a different approach to link them.

https://github.com/llvm/llvm-project/pull/181774