[PATCH] D138476: [AMDGPU][InsertWaits] No wait for WAW for global/scratch_load

[Jay Foad via Phabricator via llvm-commits]

foad accepted this revision.
foad added a comment.
This revision is now accepted and ready to land.

LGTM. @nhaehnle already explained offline that the basic idea is sound.



================
Comment at: llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp:144
 
+static bool accessVMEMOnly(const MachineInstr &Inst) {
+  return SIInstrInfo::isVMEM(Inst) || SIInstrInfo::isFLATGlobal(Inst) ||
----------------
I guess this is OK but it I really don't know what we mean by "VMEM" any more.


================
Comment at: llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp:151
+  assert(accessVMEMOnly(Inst));
+  if (!SIInstrInfo::isMIMG(Inst) || SIInstrInfo::isFLATGlobal(Inst) ||
+      SIInstrInfo::isFLATScratch(Inst))
----------------
I don't think you need to change this line. Global and scratch will already fail the isMIMG test.


================
Comment at: llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp:1187
             // results in order anyway.
-            if (Op.isUse() || !SIInstrInfo::isVMEM(MI) ||
+            if (Op.isUse() || !(accessVMEMOnly(MI)) ||
                 ScoreBrackets.hasOtherPendingVmemTypes(RegNo,
----------------
Don't need the outer parentheses here.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D138476/new/

https://reviews.llvm.org/D138476