[PATCH] D31161: [AMDGPU] New Waitcnt Insertion Pass

[Stanislav Mekhanoshin via Phabricator via llvm-commits]

rampitec added a comment.

Test with barrier surrounded with fences is needed. All relevant combinations of fences needs to be checked and pattern shall check that only needed counters are used, wait is produced, and that is only one wait. Please refer to OpenCL barrier() implementation for the code to check.



================
Comment at: lib/Target/AMDGPU/SIInsertWaitcnts.cpp:1005
+  // TODO: Tie force zero to a compiler triage option.
+  bool ForceZero = false;
+
----------------
Not needed anymore.


https://reviews.llvm.org/D31161