[PATCH] D157089: [AMDGPU] Fix dealing with register interval endpoints in SIInsertWaitcnts.

Ivan Kosarev via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Tue Aug 22 05:27:31 PDT 2023


kosarev added a comment.

> Yes it looks fine but I would not call it a "fix" since there is no bug here.

Right. There still seems to be a problem with testing against the SGPR range, however, with `SQ_MAX_PGM_SGPRS` being used as the end of the range where in fact it seems to be the size of the SGPR interval (and `Encoding.SGPRL` never used). Then changing it to `assert(Reg >= Encoding.SGPR0 && Reg <= Encoding.SGPRL);` doesn't work because `getAddressableNumSGPRs()` doesn't seem to include EXEC, which it seems was the expectation initially.

I'm going to update this once again to fix that, and then let's see if using RegUnits can be a better alternative to the intervals as you suggested elsewhere.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D157089/new/

https://reviews.llvm.org/D157089



More information about the llvm-commits mailing list