This is an archive of the discontinued LLVM Phabricator instance.

AMDGPU: Run SIInsertWaits as pre-emit pass
ClosedPublic

Authored by arsenm on Jun 27 2015, 2:40 PM.

Reviewers

Summary

Running this after the scheduler enables scheduling
waits later so other ALU instructions can run while
this would be waiting.

When combined with enabling the post-RA scheduler, this
gives about a ~20% improvement on sgemm.

arsenm retitled this revision from to AMDGPU: Run SIInsertWaits as pre-emit pass.

arsenm updated this object.

arsenm edited the test plan for this revision. (Show Details)

arsenm added a subscriber: Unknown Object (MLST).

LGTM.

This revision is now accepted and ready to land.Jul 6 2015, 7:20 AM

r241473

Path

Size

lib/

Target/

AMDGPU/

2 lines

	Show First 20 Lines • Show All 274 Lines • ▼ Show 20 Lines
	}			}

	void GCNPassConfig::addPostRegAlloc() {			void GCNPassConfig::addPostRegAlloc() {
	addPass(createSIPrepareScratchRegs(), false);			addPass(createSIPrepareScratchRegs(), false);
	addPass(createSIShrinkInstructionsPass(), false);			addPass(createSIShrinkInstructionsPass(), false);
	}			}

	void GCNPassConfig::addPreSched2() {			void GCNPassConfig::addPreSched2() {
	addPass(createSIInsertWaits(*TM), false);
	}			}

	void GCNPassConfig::addPreEmitPass() {			void GCNPassConfig::addPreEmitPass() {
				addPass(createSIInsertWaits(*TM), false);
	addPass(createSILowerControlFlowPass(*TM), false);			addPass(createSILowerControlFlowPass(*TM), false);
	}			}

	TargetPassConfig *GCNTargetMachine::createPassConfig(PassManagerBase &PM) {			TargetPassConfig *GCNTargetMachine::createPassConfig(PassManagerBase &PM) {
	return new GCNPassConfig(this, PM);			return new GCNPassConfig(this, PM);
	}			}