Skip to content

[MLIR][Vector] Implement TransferReadOfExtractSliceOp as MaskableOpRewritePattern #91960

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
May 16, 2024
Merged

[MLIR][Vector] Implement TransferReadOfExtractSliceOp as MaskableOpRewritePattern #91960

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
1b7e66a
[MLIR][Vector] Implement TransferReadOfExtractSliceOpFolder as Maskab…
nujaa May 10, 2024
37b7f95
Update mlir/test/Dialect/Tensor/fold-tensor-subset-ops-into-vector-tr…
nujaa May 14, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions mlir/lib/Dialect/Tensor/Transforms/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,5 +39,6 @@ add_mlir_dialect_library(MLIRTensorTransforms
MLIRTilingInterface
MLIRTransforms
MLIRVectorDialect
MLIRVectorUtils
MLIRValueBoundsOpInterface
)
29 changes: 18 additions & 11 deletions mlir/lib/Dialect/Tensor/Transforms/FoldTensorSubsetOps.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,6 +18,7 @@
#include "mlir/Dialect/Tensor/Transforms/Transforms.h"
#include "mlir/Dialect/Utils/IndexingUtils.h"
#include "mlir/Dialect/Vector/IR/VectorOps.h"
#include "mlir/Dialect/Vector/Utils/VectorUtils.h"
#include "mlir/IR/AffineMap.h"
#include "mlir/IR/BuiltinAttributes.h"
#include "mlir/Transforms/GreedyPatternRewriteDriver.h"
Expand Down Expand Up @@ -48,12 +49,14 @@ static Value getTensorOperand(tensor::InsertSliceOp op) {
namespace {
/// Merge extract_slice operation with load/transferRead operation.
class TransferReadOfExtractSliceOpFolder final
: public OpRewritePattern<vector::TransferReadOp> {
: public vector::MaskableOpRewritePattern<vector::TransferReadOp> {
public:
using OpRewritePattern<vector::TransferReadOp>::OpRewritePattern;
using MaskableOpRewritePattern::MaskableOpRewritePattern;

LogicalResult matchAndRewrite(vector::TransferReadOp readOp,
PatternRewriter &rewriter) const override;
FailureOr<mlir::Value>
matchAndRewriteMaskableOp(vector::TransferReadOp readOp,
vector::MaskingOpInterface maskOp,
PatternRewriter &rewriter) const override;
};

/// Merge insert_slice operation with store/transferWriteOp operation.
Expand Down Expand Up @@ -84,8 +87,10 @@ static LogicalResult preconditionsFoldExtractOrInsertWithTransferOp(
return success();
}

LogicalResult TransferReadOfExtractSliceOpFolder::matchAndRewrite(
vector::TransferReadOp readOp, PatternRewriter &rewriter) const {
FailureOr<mlir::Value>
TransferReadOfExtractSliceOpFolder::matchAndRewriteMaskableOp(
vector::TransferReadOp readOp, vector::MaskingOpInterface maskOp,
PatternRewriter &rewriter) const {
auto extractSliceOp =
getTensorOperand(readOp).getDefiningOp<tensor::ExtractSliceOp>();
if (!extractSliceOp)
Expand All @@ -95,7 +100,7 @@ LogicalResult TransferReadOfExtractSliceOpFolder::matchAndRewrite(
preconditionsFoldExtractOrInsertWithTransferOp(rewriter, readOp,
extractSliceOp);
if (failed(preconditionResult))
return preconditionResult;
return rewriter.notifyMatchFailure(readOp, "Failed preconditions");

SmallVector<Value> indices(readOp.getIndices().begin(),
readOp.getIndices().end());
Expand All @@ -105,15 +110,17 @@ LogicalResult TransferReadOfExtractSliceOpFolder::matchAndRewrite(
extractSliceOp.getMixedStrides(), extractSliceOp.getDroppedDims(),
indices, sourceIndices);

rewriter.replaceOpWithNewOp<vector::TransferReadOp>(
readOp, readOp.getVectorType(), extractSliceOp.getSource(), sourceIndices,
Operation *newOp = rewriter.create<vector::TransferReadOp>(
readOp.getLoc(), readOp.getVectorType(), extractSliceOp.getSource(),
sourceIndices,
AffineMapAttr::get(expandDimsToRank(
readOp.getPermutationMap(), extractSliceOp.getSourceType().getRank(),
extractSliceOp.getDroppedDims())),
readOp.getPadding(),
/*mask=*/Value(), readOp.getInBoundsAttr());

return success();
if (maskOp)
newOp = mlir::vector::maskOperation(rewriter, newOp, maskOp.getMask());
return newOp->getResults()[0];
}

LogicalResult InsertSliceOfTransferWriteOpFolder::matchAndRewrite(
Expand Down
15 changes: 15 additions & 0 deletions mlir/test/Dialect/Tensor/fold-tensor-subset-ops-into-vector-transfers.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -75,6 +75,21 @@ func.func @transfer_read_of_extract_slice_non_leading_rank_reduction(%t : tensor
return %1 : vector<5x6xf32>
}

// CHECK-LABEL: func @masked_transfer_read_of_extract_slice
// CHECK-SAME: %[[t:.*]]: tensor<?x?xf32>, %[[s1:.*]]: index, %[[s2:.*]]: index
// CHECK-DAG: %[[m:.*]] = vector.create_mask{{.*}} : vector<5x6xi1>
// CHECK-DAG: %[[a:.*]] = affine.apply {{.*}}[[s1]]
// CHECK: vector.mask %[[m]] { vector.transfer_read %[[t]]{{.*}}: tensor<?x?xf32>, vector<5x6xf32> } : vector<5x6xi1> -> vector<5x6xf32>
func.func @masked_transfer_read_of_extract_slice(%t : tensor<?x?xf32>, %s1 : index, %s2 : index) -> vector<5x6xf32> {
%c3 = arith.constant 3 : index
%c4 = arith.constant 4 : index
%cst = arith.constant 0.0 : f32
%0 = tensor.extract_slice %t[5, %s1] [10, %s2] [1, 1] : tensor<?x?xf32> to tensor<10x?xf32>
%mask = vector.create_mask %c3, %c4 : vector<5x6xi1>
%1 = vector.mask %mask {vector.transfer_read %0[%c3, %c4], %cst {in_bounds = [true, true]} : tensor<10x?xf32>, vector<5x6xf32>} : vector<5x6xi1> -> vector<5x6xf32>
return %1 : vector<5x6xf32>
}

// CHECK-LABEL: func @insert_slice_of_transfer_write(
// CHECK-SAME: %[[t1:.*]]: tensor<?x12xf32>, %[[v:.*]]: vector<5x6xf32>, %[[s:.*]]: index
// CHECK: %[[c3:.*]] = arith.constant 3 : index
Expand Down
Loading