Skip to content

[mlir] [linalg] Fix bufferize error in tensor.parallel_insert_slice op #98312

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 1 commit into from
Jul 11, 2024
Merged

[mlir] [linalg] Fix bufferize error in tensor.parallel_insert_slice op #98312

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
16 changes: 12 additions & 4 deletions mlir/lib/Dialect/Tensor/Transforms/BufferizableOpInterfaceImpl.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -387,8 +387,8 @@ struct ExtractSliceOpInterface
if (failed(resultMemrefType))
return failure();
Value subView = rewriter.create<memref::SubViewOp>(
loc, llvm::cast<MemRefType>(*resultMemrefType), *srcMemref, mixedOffsets,
mixedSizes, mixedStrides);
loc, llvm::cast<MemRefType>(*resultMemrefType), *srcMemref,
mixedOffsets, mixedSizes, mixedStrides);

replaceOpWithBufferizedValues(rewriter, op, subView);
return success();
Expand All @@ -407,8 +407,9 @@ struct ExtractSliceOpInterface
SmallVector<OpFoldResult> mixedSizes = extractSliceOp.getMixedSizes();
SmallVector<OpFoldResult> mixedStrides = extractSliceOp.getMixedStrides();
return cast<BaseMemRefType>(memref::SubViewOp::inferRankReducedResultType(
extractSliceOp.getType().getShape(), llvm::cast<MemRefType>(*srcMemrefType),
mixedOffsets, mixedSizes, mixedStrides));
extractSliceOp.getType().getShape(),
llvm::cast<MemRefType>(*srcMemrefType), mixedOffsets, mixedSizes,
mixedStrides));
}
};

Expand Down Expand Up @@ -997,6 +998,13 @@ struct ParallelInsertSliceOpInterface
rewriter.eraseOp(op);
return success();
}

/// tensor.parallel_insert_slice op has implicit inplace behavior. We
/// shouldn't create copy to resolve conflict.
LogicalResult resolveConflicts(Operation *op, RewriterBase &rewriter,
const AnalysisState &state) const {
return success();
}
};

/// Bufferization of tensor.splat. Bufferizes to a new allocation that is filled
Expand Down
20 changes: 20 additions & 0 deletions mlir/test/Dialect/Tensor/bufferize.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -626,3 +626,23 @@ func.func @tensor.splat_dynamic(%f: f32, %m: index, %n: index) -> tensor<?x3x?xf
return %0 : tensor<?x3x?xf32>
}

// -----

// CHECK-LABEL: func.func @parallel_insert_slice_copy_before_write
func.func @parallel_insert_slice_copy_before_write(%in: tensor<4xf32>, %out: tensor<4xf32>) {
%c1 = arith.constant 1 : index
%num_threads = arith.constant 4 : index

// CHECK: scf.forall {{.*}} {
%result = scf.forall (%thread_idx) in (%num_threads) shared_outs (%o = %out) -> tensor<4xf32> {
%1 = tensor.extract_slice %in[%thread_idx][1][1] : tensor<4xf32> to tensor<1xf32>
scf.forall.in_parallel {
// CHECK: memref.subview %{{.*}}[%{{.*}}] [1] [1] : memref<4xf32> to memref<1xf32, strided<[1], offset: ?>>
// CHECK: memref.subview %{{.*}}[%{{.*}}] [1] [1] : memref<4xf32> to memref<1xf32, strided<[1], offset: ?>>
tensor.parallel_insert_slice %1 into %o[%thread_idx][1][1] :
tensor<1xf32> into tensor<4xf32>
}
}
// CHECK: }
return
}
Loading