This is an archive of the discontinued LLVM Phabricator instance.

Differential D9172

[NVPTX] do not run DCE after SLSR and SeparateConstOffsetFromGEP
ClosedPublic

Authored by jingyue on Apr 21 2015, 1:39 PM.

Details

Reviewers
jholewinski
meheff
Commits
rG66a161f05e71: [NVPTX] do not run DCE after SLSR and SeparateConstOffsetFromGEP
rL235415: [NVPTX] do not run DCE after SLSR and SeparateConstOffsetFromGEP
Summary

With D9096 and D9101, there's no need to run DCE after SLSR and
SeparateConstOffsetFromGEP.

Diff Detail

Event Timeline

jingyue updated this revision to Diff 24167.Apr 21 2015, 1:39 PM
jingyue retitled this revision from to [NVPTX] do not run DCE after SLSR and SeparateConstOffsetFromGEP.
jingyue updated this object.
jingyue edited the test plan for this revision. (Show Details)
jingyue added reviewers: meheff, jholewinski.
jingyue added a subscriber: Unknown Object (MLST).
Herald added a subscriber: jholewinski. · View Herald TranscriptApr 21 2015, 1:39 PM
meheff edited edge metadata.Apr 21 2015, 1:45 PM

LGTM.

Closed by commit rL235415: [NVPTX] do not run DCE after SLSR and SeparateConstOffsetFromGEP (authored by jingyue). · Explain WhyApr 21 2015, 1:50 PM
This revision was automatically updated to reflect the committed changes.

Revision Contents

PathSize
lib/
Target/
NVPTX/
14 lines

Diff 24167

lib/Target/NVPTX/NVPTXTargetMachine.cpp

Show First 20 Lines • Show All 158 Lines • ▼ Show 20 Linesvoid NVPTXPassConfig::addIRPasses() {
disablePass(&BranchFolderPassID); disablePass(&BranchFolderPassID);
disablePass(&TailDuplicateID); disablePass(&TailDuplicateID);
addPass(createNVPTXImageOptimizerPass()); addPass(createNVPTXImageOptimizerPass());
TargetPassConfig::addIRPasses(); TargetPassConfig::addIRPasses();
addPass(createNVPTXAssignValidGlobalNamesPass()); addPass(createNVPTXAssignValidGlobalNamesPass());
addPass(createGenericToNVVMPass()); addPass(createGenericToNVVMPass());
addPass(createNVPTXFavorNonGenericAddrSpacesPass()); addPass(createNVPTXFavorNonGenericAddrSpacesPass());
// FavorNonGenericAddrSpaces shortcuts unnecessary addrspacecasts, and leave
// them unused. We could remove dead code in an ad-hoc manner, but that
// requires manual work and might be error-prone.
addPass(createDeadCodeEliminationPass());
addPass(createStraightLineStrengthReducePass()); addPass(createStraightLineStrengthReducePass());
addPass(createSeparateConstOffsetFromGEPPass()); addPass(createSeparateConstOffsetFromGEPPass());
// The SeparateConstOffsetFromGEP pass creates variadic bases that can be used // The SeparateConstOffsetFromGEP pass creates variadic bases that can be used
// by multiple GEPs. Run GVN or EarlyCSE to really reuse them. GVN generates // by multiple GEPs. Run GVN or EarlyCSE to really reuse them. GVN generates
// significantly better code than EarlyCSE for some of our benchmarks. // significantly better code than EarlyCSE for some of our benchmarks.
if (getOptLevel() == CodeGenOpt::Aggressive) if (getOptLevel() == CodeGenOpt::Aggressive)
addPass(createGVNPass()); addPass(createGVNPass());
else else
addPass(createEarlyCSEPass()); addPass(createEarlyCSEPass());
// Both FavorNonGenericAddrSpaces and SeparateConstOffsetFromGEP may leave
// some dead code. We could remove dead code in an ad-hoc manner, but that
// requires manual work and might be error-prone.
//
// The FavorNonGenericAddrSpaces pass shortcuts unnecessary addrspacecasts,
// and leave them unused.
//
// SeparateConstOffsetFromGEP rebuilds a new index from the old index, and the
// old index and some of its intermediate results may become unused.
addPass(createDeadCodeEliminationPass());
} }
bool NVPTXPassConfig::addInstSelector() { bool NVPTXPassConfig::addInstSelector() {
const NVPTXSubtarget &ST = *getTM<NVPTXTargetMachine>().getSubtargetImpl(); const NVPTXSubtarget &ST = *getTM<NVPTXTargetMachine>().getSubtargetImpl();
addPass(createLowerAggrCopies()); addPass(createLowerAggrCopies());
addPass(createAllocaHoisting()); addPass(createAllocaHoisting());
addPass(createNVPTXISelDag(getNVPTXTargetMachine(), getOptLevel())); addPass(createNVPTXISelDag(getNVPTXTargetMachine(), getOptLevel()));
▲ Show 20 Lines • Show All 84 Lines • Show Last 20 Lines