This is an archive of the discontinued LLVM Phabricator instance.

Differential D54775

AMDGPU: Don't optimize exec masks at -O0
ClosedPublic

Authored by arsenm on Nov 20 2018, 2:54 PM.

Details

Reviewers
rampitec

Diff Detail

Event Timeline

arsenm created this revision.Nov 20 2018, 2:54 PM
Herald added subscribers: t-tye, tpr, dstuttard and 5 others. · View Herald TranscriptNov 20 2018, 2:54 PM
rampitec accepted this revision.Nov 20 2018, 5:07 PM

LGTM

This revision is now accepted and ready to land.Nov 20 2018, 5:07 PM
arsenm closed this revision.Nov 26 2018, 9:05 AM

r347573

Revision Contents

PathSize
lib/
Target/
AMDGPU/
3 lines

Diff 174850

lib/Target/AMDGPU/AMDGPUTargetMachine.cpp

Show First 20 Lines • Show All 874 Lines • ▼ Show 20 Linesvoid GCNPassConfig::addOptimizedRegAlloc(FunctionPass *RegAllocPass) {
// machine-level CFG, but before register allocation. // machine-level CFG, but before register allocation.
insertPass(&SILowerControlFlowID, &SIFixWWMLivenessID, false); insertPass(&SILowerControlFlowID, &SIFixWWMLivenessID, false);
TargetPassConfig::addOptimizedRegAlloc(RegAllocPass); TargetPassConfig::addOptimizedRegAlloc(RegAllocPass);
} }
void GCNPassConfig::addPostRegAlloc() { void GCNPassConfig::addPostRegAlloc() {
addPass(&SIFixVGPRCopiesID); addPass(&SIFixVGPRCopiesID);
if (getOptLevel() > CodeGenOpt::None)
addPass(&SIOptimizeExecMaskingID); addPass(&SIOptimizeExecMaskingID);
TargetPassConfig::addPostRegAlloc(); TargetPassConfig::addPostRegAlloc();
} }
void GCNPassConfig::addPreSched2() { void GCNPassConfig::addPreSched2() {
} }
void GCNPassConfig::addPreEmitPass() { void GCNPassConfig::addPreEmitPass() {
addPass(createSIMemoryLegalizerPass()); addPass(createSIMemoryLegalizerPass());
Show All 24 Lines