[llvm] [AMDGPU] Improve selection of ballot.i64 intrinsic in wave32 mode in SelectionDAG. (PR #71556)

[Valery Pykhtin via llvm-commits]

vpykhtin wrote:

> Is this just about code quality? This probably isn't necessary if you consider the pipeline as a whole. I believe InstCombine is already reducing i64 ballot to i32 based on the subtarget's wave size

Right, but InstCombine is a bit late. This is needed after https://github.com/llvm/llvm-project/pull/68714, otherwise I need special processing of i64 mask in wave32 mode there. I've implemented the TODO with https://github.com/llvm/llvm-project/pull/71556/commits/51578ea440a20a0785e0a081f30ce333a5db7fbf.

https://github.com/llvm/llvm-project/pull/71556