[PATCH] D96258: [AMDGPU] Introduce Strict WQM mode

[Piotr Sobczak via Phabricator via llvm-commits]

piotr added inline comments.


================
Comment at: llvm/test/CodeGen/AMDGPU/wqm.ll:422
+
+; Check that Strict WQM writes aren't coalesced with non-strict writes, since
+; the Strict WQM write could clobber disabled channels in the non-strict one.
----------------
piotr wrote:
> critson wrote:
> > It's not clear to me how this test proves the "coalesced" part?
> I patterned this test on the existing one for WWM (test_strictwwm4), which follows the same structure. It seems the comment is wrong.
I looked closely at the original test and I believe the check for "endif" was missing. These two cases make sure that a WWM feeding a phi node is handled properly and the mov/add happens in the same block as WWM computations.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D96258/new/

https://reviews.llvm.org/D96258