Skip to content

Commit 0234d90

Browse files
authored
[AMDGPU] Extend MFMA padding option to gfx90a+ (#86768)
It was shown experimentally that this may have some benefit on newer HW.
1 parent 8d8fff0 commit 0234d90

File tree

2 files changed

+507
-0
lines changed

2 files changed

+507
-0
lines changed

llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2357,6 +2357,9 @@ int GCNHazardRecognizer::checkMAIHazards90A(MachineInstr *MI) {
23572357
break;
23582358
}
23592359

2360+
// Pad neighboring MFMA with noops for better inter-wave performance.
2361+
WaitStatesNeeded = std::max(WaitStatesNeeded, checkMFMAPadding(MI));
2362+
23602363
return WaitStatesNeeded;
23612364
}
23622365

0 commit comments

Comments
 (0)