[PATCH] D105104: [AMDGPU] PHI node cost should not be counted for the size and latency.

[Alexander via Phabricator via llvm-commits]

alex-t added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp:841
-  case Instruction::PHI:
-    // TODO: 1. A prediction phi won't be eliminated?
-    //       2. Estimate data copy instructions in this case.
----------------
dfukalov wrote:
> Nit: leave this or similar related todo comments somewhere, it wasn't done.
Such a prediction is unlikely possible. The number of copies that survived after the register coalescing too much dependent on the passes that run in between.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D105104/new/

https://reviews.llvm.org/D105104