[PATCH] D155284: [GlobalISel] Fix infinite loop in reassociation combine

[Amara Emerson via Phabricator via llvm-commits]

aemerson accepted this revision.
aemerson added inline comments.
This revision is now accepted and ready to land.


================
Comment at: llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp:4478-4479
 
-  if (isConstantOrConstantSplatVector(*MRI.getVRegDef(OpLHSRHS), MRI)) {
+  if (isConstantOrConstantSplatVector(*MRI.getVRegDef(OpLHSRHS), MRI) &&
+      !isConstantOrConstantSplatVector(*MRI.getVRegDef(OpLHSLHS), MRI)) {
     if (isConstantOrConstantSplatVector(*OpRHSDef, MRI)) {
----------------
Ca you add a comment here to explain why we do the second constant check.


================
Comment at: llvm/test/CodeGen/AMDGPU/GlobalISel/postlegalizer-combiner-reassoc.mir:21
+    %0:_(s32) = COPY $vgpr0
+    %7:_(s32) = G_CONSTANT i32 0
+    %13:_(s32) = G_CONSTANT i32 1
----------------
arsenm wrote:
> should run-pass=none to compact the register numbers, it's sometimes annoying when the register numbers actually tested don't match the ones in the file
TIL


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D155284/new/

https://reviews.llvm.org/D155284