[llvm] [AMDGPU] Use s_cmovk_i32 instead of s_cselect_b32 when applicable (PR #135232)

[Jay Foad via llvm-commits]

https://github.com/jayfoad commented:

You can't just commute the select since that changes the meaning of the instruction! You would have to find the cmp instruction the defines scc and invert the condition, which is probably not worth doing just to save a few bytes in the encoding.

https://github.com/llvm/llvm-project/pull/135232