This is an archive of the discontinued LLVM Phabricator instance.

Differential D100814

[mlir][spirv] Support conversion of extract op from vector<1xT> type
ClosedPublic

Authored by ThomasRaoux on Apr 19 2021, 9:29 PM.

Details

Reviewers
antiagainst
aartbik
mravishankar
Commits
rGb2e72cd38de8: [mlir][spirv] Support conversion of extract op from vector<1xT> type

Diff Detail

Event Timeline

ThomasRaoux created this revision.Apr 19 2021, 9:29 PM
Herald added a reviewer: aartbik. · View Herald TranscriptApr 19 2021, 9:29 PM
Herald added a reviewer: mravishankar. · View Herald Transcript
Herald added subscribers: dcaballe, cota, teijeong and 17 others. · View Herald Transcript
ThomasRaoux requested review of this revision.Apr 19 2021, 9:29 PM
Herald added a project: Restricted Project. · View Herald TranscriptApr 19 2021, 9:29 PM
Herald added subscribers: stephenneuendorffer, nicolasvasilache. · View Herald Transcript
Harbormaster completed remote builds in B99612: Diff 338708.Apr 19 2021, 10:12 PM
antiagainst requested changes to this revision.Apr 20 2021, 5:33 AM
antiagainst added inline comments.
mlir/test/Conversion/VectorToSPIRV/simple.mlir
38

Should we actually create another function to host this so that we can return the value to really CHECK against it?

This revision now requires changes to proceed.Apr 20 2021, 5:33 AM
ThomasRaoux updated this revision to Diff 338852.Apr 20 2021, 6:52 AM
ThomasRaoux added inline comments.
mlir/test/Conversion/VectorToSPIRV/simple.mlir
38

Moved it to a different function and check that the extract is replaced by the scalar source.

Harbormaster completed remote builds in B99708: Diff 338852.Apr 20 2021, 7:52 AM
antiagainst accepted this revision.Apr 20 2021, 9:10 AM
This revision is now accepted and ready to land.Apr 20 2021, 9:10 AM
Closed by commit rGb2e72cd38de8: [mlir][spirv] Support conversion of extract op from vector<1xT> type (authored by ThomasRaoux). · Explain WhyApr 20 2021, 9:12 AM
This revision was automatically updated to reflect the committed changes.
ThomasRaoux added a commit: rGb2e72cd38de8: [mlir][spirv] Support conversion of extract op from vector<1xT> type.

Revision Contents

PathSize
mlir/
lib/
Conversion/
VectorToSPIRV/
5 lines
test/
Conversion/
VectorToSPIRV/
1 line

Diff 338708

mlir/lib/Conversion/VectorToSPIRV/VectorToSPIRV.cpp

Show First 20 Lines • Show All 83 Lines • ▼ Show 20 Lines matchAndRewrite(vector::ExtractOp extractOp, ArrayRef<Value> operands,
if (resultVectorType && resultVectorType.getNumElements() > 1) if (resultVectorType && resultVectorType.getNumElements() > 1)
return failure(); return failure();
auto dstType = getTypeConverter()->convertType(extractOp.getType()); auto dstType = getTypeConverter()->convertType(extractOp.getType());
if (!dstType) if (!dstType)
return failure(); return failure();
vector::ExtractOp::Adaptor adaptor(operands); vector::ExtractOp::Adaptor adaptor(operands);
if (adaptor.vector().getType().isa<spirv::ScalarType>()) {
rewriter.replaceOp(extractOp, adaptor.vector());
return success();
}
int32_t id = getFirstIntValue(extractOp.position()); int32_t id = getFirstIntValue(extractOp.position());
rewriter.replaceOpWithNewOp<spirv::CompositeExtractOp>( rewriter.replaceOpWithNewOp<spirv::CompositeExtractOp>(
extractOp, adaptor.vector(), id); extractOp, adaptor.vector(), id);
return success(); return success();
} }
}; };
struct VectorExtractStridedSliceOpConvert final struct VectorExtractStridedSliceOpConvert final
▲ Show 20 Lines • Show All 153 Lines • Show Last 20 Lines

mlir/test/Conversion/VectorToSPIRV/simple.mlir

Show All 29 Lines
// CHECK-LABEL: func @extract // CHECK-LABEL: func @extract
// CHECK-SAME: %[[ARG:.+]]: vector<2xf32> // CHECK-SAME: %[[ARG:.+]]: vector<2xf32>
// CHECK: %{{.+}} = spv.CompositeExtract %[[ARG]][0 : i32] : vector<2xf32> // CHECK: %{{.+}} = spv.CompositeExtract %[[ARG]][0 : i32] : vector<2xf32>
// CHECK: %{{.+}} = spv.CompositeExtract %[[ARG]][1 : i32] : vector<2xf32> // CHECK: %{{.+}} = spv.CompositeExtract %[[ARG]][1 : i32] : vector<2xf32>
func @extract(%arg0 : vector<2xf32>) { func @extract(%arg0 : vector<2xf32>) {
%0 = "vector.extract"(%arg0) {position = [0]} : (vector<2xf32>) -> vector<1xf32> %0 = "vector.extract"(%arg0) {position = [0]} : (vector<2xf32>) -> vector<1xf32>
%1 = "vector.extract"(%arg0) {position = [1]} : (vector<2xf32>) -> f32 %1 = "vector.extract"(%arg0) {position = [1]} : (vector<2xf32>) -> f32
%2 = "vector.extract"(%0) {position = [0]} : (vector<1xf32>) -> f32
antiagainstUnsubmitted
Not Done
ReplyInline Actions

Should we actually create another function to host this so that we can return the value to really CHECK against it?

antiagainst: Should we actually create another function to host this so that we can return the value to…
ThomasRaouxAuthorUnsubmitted
Done
ReplyInline Actions

Moved it to a different function and check that the extract is replaced by the scalar source.

ThomasRaoux: Moved it to a different function and check that the extract is replaced by the scalar source.
spv.Return spv.Return
} }
// ----- // -----
// CHECK-LABEL: extract_insert // CHECK-LABEL: extract_insert
// CHECK-SAME: %[[V:.*]]: vector<4xf32> // CHECK-SAME: %[[V:.*]]: vector<4xf32>
// CHECK: %[[S:.*]] = spv.CompositeExtract %[[V]][1 : i32] : vector<4xf32> // CHECK: %[[S:.*]] = spv.CompositeExtract %[[V]][1 : i32] : vector<4xf32>
▲ Show 20 Lines • Show All 72 Lines • Show Last 20 Lines