[all-commits] [llvm/llvm-project] 30498c: [X86] SkylakeClientModel - conversion instructions...

[Simon Pilgrim via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 30498cf7c46f90db0f67ff01f0246860e55be0f2
      https://github.com/llvm/llvm-project/commit/30498cf7c46f90db0f67ff01f0246860e55be0f2
  Author: Simon Pilgrim <llvm-dev at redking.me.uk>
  Date:   2022-11-10 (Thu, 10 Nov 2022)

  Changed paths:
    M llvm/lib/Target/X86/X86SchedSkylakeClient.td
    M llvm/test/tools/llvm-mca/X86/SkylakeClient/resources-avx1.s
    M llvm/test/tools/llvm-mca/X86/SkylakeClient/resources-f16c.s
    M llvm/test/tools/llvm-mca/X86/SkylakeClient/resources-sse1.s
    M llvm/test/tools/llvm-mca/X86/SkylakeClient/resources-sse2.s

  Log Message:
  -----------
  [X86] SkylakeClientModel - conversion instructions don't use Port015

Fixes a lot of throughput mismatches - the more complicated conversion instructions use SKLPort5+SKLPort01, not SKLPort5+SKLPort015 (SKLPort015 is mainly used for basic Logic + blend ops)

Fixing this should allow us to remove a lot of unnecessary scheduler overrides from SkylakeClientModel

Confirmed by both Agner + uops.info