[PATCH] D139780: [AMDGPU] Unify divergent nodes if the PostDom tree has one root

Ruiling, Song via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Tue Jan 3 21:03:00 PST 2023


ruiling added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUUnifyDivergentExitNodes.cpp:109
   // This is a cluster of orthogonal Transforms
   AU.addPreservedID(LowerSwitchID);
   FunctionPass::getAnalysisUsage(AU);
----------------
gandhi21299 wrote:
> arsenm wrote:
> > We should have a required LowerSwitchID too
> I will have a seperate patch for that, it seems to be causing difficulties when the pass manager schedules UnifyDivergentExitNodes.
I think for function pass dependency or pass ordering, I still prefer they are managed by compiler developer. If I remember correctly, the new pass manager does not support dependency between function passes?


================
Comment at: llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll:657
+; GFX908-NEXT:    s_mov_b64 s[22:23], -1
 ; GFX908-NEXT:    s_branch .LBB3_4
 ;
----------------
ruiling wrote:
> Why there is no DummyReturnBlock for GFX908?
Did you try to get the answer for the question? It sounds strange we get different behavior for gfx908 and gfx90A here.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D139780/new/

https://reviews.llvm.org/D139780



More information about the llvm-commits mailing list