[MLIR][Tensor] Enhance bufferization of tensor.expand_shape op #128871
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
arnab-polymage
requested review from
hanhanW and
nicolasvasilache
as code owners
|
@llvm/pr-subscribers-mlir-tensor @llvm/pr-subscribers-mlir Author: Arnab Dutta (arnab-polymage) ChangesInstead of inferring the output shape argument of Full diff: https://github.com/llvm/llvm-project/pull/128871.diff 2 Files Affected:
diff --git a/mlir/lib/Dialect/Tensor/Transforms/BufferizableOpInterfaceImpl.cpp b/mlir/lib/Dialect/Tensor/Transforms/BufferizableOpInterfaceImpl.cpp
index 81404fa664cd4..efbe09f4d2419 100644
--- a/mlir/lib/Dialect/Tensor/Transforms/BufferizableOpInterfaceImpl.cpp
+++ b/mlir/lib/Dialect/Tensor/Transforms/BufferizableOpInterfaceImpl.cpp
@@ -337,14 +337,27 @@ struct ExpandShapeOpInterface
if (failed(buffer))
return failure();
- // Memref result type is inferred by the builder based on reassociation
- // indices and result shape.
- // TODO: Instead of inferring the output shape argument of
- // memref.expand_shape op, use output_shape argument of tensor.expand_shape
- // op.
- replaceOpWithNewBufferizedOp<memref::ExpandShapeOp>(
- rewriter, op, tensorResultType.getShape(), *buffer,
- expandShapeOp.getReassociationIndices());
+ // Use output_shape argument of tensor.expand_shape op to get the result
+ // shapes of the memref.expand_shape op to be created.
+ SmallVector<OpFoldResult> outShape;
+ unsigned dynDimCount = 0;
+ for (unsigned i = 0, e = tensorResultType.getRank(); i < e; i++) {
+ if (tensorResultType.isDynamicDim(i))
+ outShape.push_back(expandShapeOp.getOutputShape()[dynDimCount++]);
+ }
+ auto memrefExpandShape = rewriter.create<memref::ExpandShapeOp>(
+ op->getLoc(), tensorResultType.getShape(), *buffer,
+ expandShapeOp.getReassociationIndices(), outShape);
+ SmallVector<int64_t> staticShape;
+ for (unsigned i = 0, e = tensorResultType.getRank(); i < e; i++) {
+ if (tensorResultType.isDynamicDim(i))
+ staticShape.push_back(ShapedType::kDynamic);
+ else
+ staticShape.push_back(tensorResultType.getDimSize(i));
+ }
+ memrefExpandShape.setStaticOutputShape(staticShape);
+ replaceOpWithBufferizedValues(rewriter, op,
+ memrefExpandShape->getResults());
return success();
}
};
diff --git a/mlir/test/Dialect/Tensor/bufferize.mlir b/mlir/test/Dialect/Tensor/bufferize.mlir
index 9ea0a15f31185..c1beed95f2006 100644
--- a/mlir/test/Dialect/Tensor/bufferize.mlir
+++ b/mlir/test/Dialect/Tensor/bufferize.mlir
@@ -366,14 +366,10 @@ func.func @tensor.insert(%t1: tensor<5xf32>, %idx1: index, %f: f32) -> tensor<5x
// -----
// CHECK-LABEL: func @tensor.expand_shape(
-// CHECK-SAME: %[[t1:.*]]: tensor<?x10xf32>
+// CHECK-SAME: %[[t1:.*]]: tensor<?x10xf32>, %[[sz0:.*]]: index
func.func @tensor.expand_shape(%t1: tensor<?x10xf32>, %sz0: index) -> tensor<2x?x10xf32> {
// CHECK: %[[m1:.*]] = bufferization.to_memref %[[t1]]
- // CHECK: %[[C0:.*]] = arith.constant 0 : index
- // CHECK: %[[DIM:.*]] = memref.dim %[[m1]], %[[C0]] : memref<?x10xf32>
- // CHECK: %[[C2:.*]] = arith.constant 2 : index
- // CHECK: %[[VAL_1:.*]] = arith.divsi %[[DIM]], %[[C2]] : index
- // CHECK: %[[expanded:.*]] = memref.expand_shape %[[m1]] {{\[\[}}0, 1], [2]] output_shape [2, %[[VAL_1]], 10] : memref<?x10xf32> into memref<2x?x10xf32>
+ // CHECK: %[[expanded:.*]] = memref.expand_shape %[[m1]] {{\[\[}}0, 1], [2]] output_shape [2, %[[sz0]], 10] : memref<?x10xf32> into memref<2x?x10xf32>
%0 = tensor.expand_shape %t1 [[0, 1], [2]] output_shape [2, %sz0, 10]
: tensor<?x10xf32> into tensor<2x?x10xf32>
@@ -385,23 +381,20 @@ func.func @tensor.expand_shape(%t1: tensor<?x10xf32>, %sz0: index) -> tensor<2x?
// -----
// CHECK-LABEL: func @tensor.expand_shape_of_slice(
-// CHECK-SAME: %[[t1:.*]]: tensor<?x20xf32>
+// CHECK-SAME: %[[t1:.*]]: tensor<?x20xf32>, %{{.*}}: index, %{{.*}}: index, %[[sz0:.*]]: index
func.func @tensor.expand_shape_of_slice(
%t1: tensor<?x20xf32>, %o1: index, %s1: index, %sz0: index) -> tensor<?x7x2x5xf32> {
// CHECK: %[[m1:.*]] = bufferization.to_memref %[[t1]] :
// CHECK: %[[subview:.*]] = memref.subview %[[m1]][%{{.*}}, 5] [%{{.*}}, 10] [1, 1] : memref<?x20xf32> to memref<?x10xf32, strided<[20, 1], offset: ?>>
%0 = tensor.extract_slice %t1[%o1, 5][%s1, 10][1, 1] :
tensor<?x20xf32> to tensor<?x10xf32>
- // CHECK: %[[C7:.*]] = arith.constant 7 : index
- // CHECK: %[[VAL_1:.*]] = arith.divsi %{{.*}}, %[[C7]] : index
- // CHECK: %[[expanded:.*]] = memref.expand_shape %[[subview]] {{\[\[}}0, 1], [2, 3]] output_shape [%[[VAL_1]], 7, 2, 5] : memref<?x10xf32, strided<[20, 1], offset: ?>> into memref<?x7x2x5xf32, strided<[140, 20, 5, 1], offset: ?>>
+ // CHECK: %[[expanded:.*]] = memref.expand_shape %[[subview]] {{\[\[}}0, 1], [2, 3]] output_shape [%[[sz0]], 7, 2, 5] : memref<?x10xf32, strided<[20, 1], offset: ?>> into memref<?x7x2x5xf32, strided<[140, 20, 5, 1], offset: ?>>
%1 = tensor.expand_shape %0 [[0, 1], [2, 3]] output_shape [%sz0, 7, 2, 5] :
tensor<?x10xf32> into tensor<?x7x2x5xf32>
// CHECK: %[[r:.*]] = bufferization.to_tensor %[[expanded]]
// CHECK: return %[[r]]
return %1 : tensor<?x7x2x5xf32>
}
-
// -----
// CHECK-LABEL: func @tensor.expand_shape_of_scalar_slice(
@@ -417,7 +410,20 @@ func.func @tensor.expand_shape_of_scalar_slice(
// CHECK: return %[[r]]
return %1 : tensor<1xf32>
}
+// -----
+// CHECK-LABEL: func @tensor.expand_shape_multiple_dynamic_indices(
+// CHECK-SAME: %[[t1:.*]]: tensor<?x256xf32>, %[[sz0:.*]]: index, %[[sz1:.*]]: index, %[[sz2:.*]]: index
+func.func @tensor.expand_shape_multiple_dynamic_indices(%t1: tensor<?x256xf32>, %sz0: index, %sz1: index, %sz2: index) -> tensor<?x?x?x256xf32> {
+ // CHECK: %[[m1:.*]] = bufferization.to_memref %[[t1]]
+ // CHECK: %[[expanded:.*]] = memref.expand_shape %[[m1]] {{\[\[}}0, 1, 2], [3]] output_shape [%[[sz0]], %[[sz1]], %[[sz2]], 256] : memref<?x256xf32> into memref<?x?x?x256xf32>
+ %0 = tensor.expand_shape %t1 [[0, 1, 2], [3]] output_shape [%sz0, %sz1, %sz2, 256]
+ : tensor<?x256xf32> into tensor<?x?x?x256xf32>
+
+ // CHECK: %[[r:.*]] = bufferization.to_tensor %[[expanded]]
+ // CHECK: return %[[r]]
+ return %0 : tensor<?x?x?x256xf32>
+}
// -----
// CHECK-LABEL: func @tensor.collapse_shape(
@@ -646,3 +652,6 @@ func.func @parallel_insert_slice_copy_before_write(%in: tensor<4xf32>, %out: ten
// CHECK: }
return
}
+
+// -----
+
|
|
@ramiro050 @bondhugula please review |
bondhugula
requested review from
matthias-springer and
MaheshRavishankar
and removed request for
nicolasvasilache
mlir/lib/Dialect/Tensor/Transforms/BufferizableOpInterfaceImpl.cpp
Outdated
Show resolved
Hide resolved
mlir/lib/Dialect/Tensor/Transforms/BufferizableOpInterfaceImpl.cpp
Outdated
Show resolved
Hide resolved
What is dd support? |
I read it as dynamic dimensions :) |
Instead of inferring the output shape argument of memref.expand_shape op, use output_shape argument of tensor.expand_shape op by adding dynamic dimension support for bufferization of tensor.expand_shape when there are more than one dynamic dim within a reassociation set.
arnab-polymage
force-pushed
the
ornib/tensor_expand_shape_bufferize_bug
branch
from
c710f83 to
d9b2521
Compare
MaheshRavishankar
approved these changes
Feb 27, 2025
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/52/builds/6432 Here is the relevant piece of the build log for the reference |
cheezeburglar
pushed a commit
to cheezeburglar/llvm-project
that referenced
this pull request
Feb 28, 2025
…128871) Instead of inferring the output shape argument of memref.expand_shape op, use output_shape argument of tensor.expand_shape op by adding dynamic dimension support for bufferization of tensor.expand_shape when there are more than one dynamic dim within a reassociation set.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Instead of inferring the output shape argument of
memref.expand_shape op, use output_shape argument of tensor.expand_shape op by adding dynamic dimension support for bufferization of tensor.expand_shape when there are more than one dynamic dim within a reassociation set.