[llvm] [AMDGPU][SIInsertWaitCnts] Gfx12.5 - Refactor xcnt optimization (PR #164357)

[Jay Foad via llvm-commits]

jayfoad wrote:

> Refactor the XCnt optimization checks so that they can be checked when applying a pre-existing waitcnt.

Hi @RyanRio, it looks to me like all the optimizations done in applyXcnt (called from applyWaitcnt) should instead be done in simplifyWaitcnt, which is already called on both paths (generating new waitcnts and applying pre-existing waitcnts). Can you try that please?

(There are also some Xcnt optimizations in generateWaitcnt that I don't fully understand, but perhaps they could also be subsumed by simplifyWaitcnt?)


https://github.com/llvm/llvm-project/pull/164357