[PATCH] D91636: [AMDGPU] Fix and extend vccz workarounds

[Jay Foad via Phabricator via llvm-commits]

foad created this revision.
foad added reviewers: arsenm, rampitec.
Herald added subscribers: llvm-commits, kerbowa, hiraditya, t-tye, tpr, dstuttard, yaxunl, nhaehnle, jvesely, kzhuravl.
Herald added a project: LLVM.
foad requested review of this revision.
Herald added a subscriber: wdng.

We have workarounds for two different cases where vccz can get out of
sync with the value in vcc. This fixes them in two ways:

1. Fix the case where the def of vcc was in a previous basic block, by

pessimistically assuming that vccz might be incorrect at a basic block
boundary.

2. Fix the handling of pre-existing waitcnt instructions by calling

generateWaitcntInstBefore before examining ScoreBrackets to determine
whether there's an outstanding smem read operation.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D91636

Files:
  llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp
  llvm/test/CodeGen/AMDGPU/infinite-loop.ll
  llvm/test/CodeGen/AMDGPU/nested-loop-conditions.ll
  llvm/test/CodeGen/AMDGPU/si-annotate-cf.ll
  llvm/test/CodeGen/AMDGPU/si-annotate-cfg-loop-assert.ll
  llvm/test/CodeGen/AMDGPU/vccz-corrupt-bug-workaround.mir

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D91636.305799.patch
Type: text/x-patch
Size: 8366 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20201117/33e520ac/attachment.bin>