This is an archive of the discontinued LLVM Phabricator instance.

Differential D147950

Fix invalid op result access in SparseTensorCodegen
ClosedPublic

Authored by frgossen on Apr 10 2023, 9:31 AM.

Details

Reviewers
aartbik
nicolasvasilache
Commits
rG4a33c6799505: Fix invalid op result access in SparseTensorCodegen
Summary

This trigges an assertion introduced in https://reviews.llvm.org/D147883

Diff Detail

Event Timeline

frgossen created this revision.Apr 10 2023, 9:31 AM
Herald added a reviewer: aartbik. · View Herald TranscriptApr 10 2023, 9:31 AM
Herald added a project: Restricted Project. · View Herald Transcript
Herald added subscribers: bviyer, hanchung, jsetoain and 25 others. · View Herald Transcript
frgossen requested review of this revision.Apr 10 2023, 9:31 AM
Herald added a reviewer: nicolasvasilache. · View Herald TranscriptApr 10 2023, 9:31 AM
Herald added a project: Restricted Project. · View Herald Transcript
Herald added subscribers: stephenneuendorffer, nicolasvasilache. · View Herald Transcript
aartbik accepted this revision.Apr 10 2023, 9:34 AM
This revision is now accepted and ready to land.Apr 10 2023, 9:34 AM
This revision was landed with ongoing or failed builds.Apr 10 2023, 9:36 AM
Closed by commit rG4a33c6799505: Fix invalid op result access in SparseTensorCodegen (authored by frgossen). · Explain Why
This revision was automatically updated to reflect the committed changes.
frgossen added a commit: rG4a33c6799505: Fix invalid op result access in SparseTensorCodegen.
Harbormaster completed remote builds in B224594: Diff 512176.Apr 10 2023, 10:13 AM

Revision Contents

PathSize
mlir/
lib/
Dialect/
SparseTensor/
Transforms/
3 lines

Diff 512179

mlir/lib/Dialect/SparseTensor/Transforms/SparseTensorCodegen.cpp

Show First 20 Lines • Show All 1,412 Lines • ▼ Show 20 Lines matchAndRewrite(NewOp op, OpAdaptor adaptor,
// the sparse tensor reader. // the sparse tensor reader.
SmallVector<Value> dynSizes; SmallVector<Value> dynSizes;
if (dstTp.hasDynamicDimShape()) { if (dstTp.hasDynamicDimShape()) {
// FIXME: call `getSparseTensorReaderDimSizes` instead, because // FIXME: call `getSparseTensorReaderDimSizes` instead, because
// `copySparseTensorReaderDimSizes` copies the memref over, // `copySparseTensorReaderDimSizes` copies the memref over,
// instead of just accessing the reader's memory directly. // instead of just accessing the reader's memory directly.
Value dimSizes = genAlloca(rewriter, loc, dimRank, indexTp); Value dimSizes = genAlloca(rewriter, loc, dimRank, indexTp);
createFuncCall(rewriter, loc, "copySparseTensorReaderDimSizes", {}, createFuncCall(rewriter, loc, "copySparseTensorReaderDimSizes", {},
{reader, dimSizes}, EmitCInterface::On) {reader, dimSizes}, EmitCInterface::On);
.getResult(0);
for (const auto &d : llvm::enumerate(dstTp.getDimShape())) for (const auto &d : llvm::enumerate(dstTp.getDimShape()))
if (ShapedType::isDynamic(d.value())) if (ShapedType::isDynamic(d.value()))
dynSizes.push_back(rewriter.create<memref::LoadOp>( dynSizes.push_back(rewriter.create<memref::LoadOp>(
loc, dimSizes, constantIndex(rewriter, loc, d.index()))); loc, dimSizes, constantIndex(rewriter, loc, d.index())));
} }
Value nse = createFuncCall(rewriter, loc, "getSparseTensorReaderNSE", Value nse = createFuncCall(rewriter, loc, "getSparseTensorReaderNSE",
{indexTp}, {reader}, EmitCInterface::Off) {indexTp}, {reader}, EmitCInterface::Off)
▲ Show 20 Lines • Show All 115 Lines • Show Last 20 Lines