[PATCH] D105062: [AMDGPU] Reduce AGPR to AGPR copies with same source

[Stanislav Mekhanoshin via Phabricator via llvm-commits]

rampitec added a comment.

In D105062#2845855 <https://reviews.llvm.org/D105062#2845855>, @arsenm wrote:

> I think copyPhysReg should not have to do any nontrivial scanning. These situations where this matters should be resolvable before register allocation

I concur. This is untrivial and error prone. There are quite a number of foldings for agpr copy and init in different places, this looks like a candidate for yet another one, starting from a relevant testcase. This place is really a last resort.

Do you have a minimal ir test showing a problem you are solving?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D105062/new/

https://reviews.llvm.org/D105062