[MLIR] Fix canonicalization pattern for 'shape.shape_of' #134234
Conversation
|
Thank you for submitting a Pull Request (PR) to the LLVM Project! This PR will be automatically labeled and the relevant teams will be notified. If you wish to, you can add reviewers by using the "Reviewers" section on this page. If this is not working for you, it is probably because you do not have write permissions for the repository. In which case you can instead tag reviewers by name in a comment by using If you have received no comments on your PR for a week, you can request a review by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate is once a week. Please remember that you are asking for valuable time from other developers. If you have further questions, they may be answered by the LLVM GitHub User Guide. You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums. |
|
@llvm/pr-subscribers-mlir @llvm/pr-subscribers-mlir-shape Author: Alaa Ali (alaa-ali) ChangesThis PR will fix a bug in a canonicalization pattern (operation shape.shape_of: shape of reshape) See file canonicalize.mlir in the change list for an example. For the context, this bug was found while running a test on Keras 3, the canonicalizer errors out due to an invalid tensor.cast operation when the batch size is dynamic. @sjarus Full diff: https://github.com/llvm/llvm-project/pull/134234.diff 2 Files Affected:
diff --git a/mlir/lib/Dialect/Shape/IR/Shape.cpp b/mlir/lib/Dialect/Shape/IR/Shape.cpp
index 10ba808cd26c2..b8eac7c86797b 100644
--- a/mlir/lib/Dialect/Shape/IR/Shape.cpp
+++ b/mlir/lib/Dialect/Shape/IR/Shape.cpp
@@ -1734,10 +1734,22 @@ struct ShapeOfFromReshape : public OpRewritePattern<shape::ShapeOfOp> {
// Operand 'shape' of 'tensor.reshape' may now be used as the result of
// 'shape.shape_of'. While its type is guaranteed to be compatible in well-
// formed IR, it may not be identical (dynamically vs statically shaped),
- // in which case it needs to be cast first.
+ // in which case it needs to be cast first using 'tensor.cast'.
+ // Additionally, it may not have identical element type (i32 vs index)
+ // while it has identical shaped type (dynamic vs static), in which case it needs
+ // to be cast first using 'arith.index_cast'.
+ // Note: 'shape.shape_of' op result must be shape or extent tensor.
Value shape = tensorReshapeOp.getShape();
- if (op.getType() != shape.getType())
- shape = rewriter.create<tensor::CastOp>(op.getLoc(), op.getType(), shape);
+
+ auto opTensorType = llvm::dyn_cast<RankedTensorType>(op.getType());
+ auto shapeTensorType = llvm::dyn_cast<RankedTensorType>(shape.getType());
+
+ if (op.getType() != shape.getType()) {
+ if (opTensorType.getElementType() == shapeTensorType.getElementType())
+ shape = rewriter.create<tensor::CastOp>(op.getLoc(), op.getType(), shape);
+ else if (!isExtentTensorType(shape.getType()))
+ shape = rewriter.create<arith::IndexCastOp>(op.getLoc(), op.getType(), shape);
+ }
rewriter.replaceOp(op, shape);
return success();
diff --git a/mlir/test/Dialect/Shape/canonicalize.mlir b/mlir/test/Dialect/Shape/canonicalize.mlir
index cf439c9c1b854..9b25468b3ab1e 100644
--- a/mlir/test/Dialect/Shape/canonicalize.mlir
+++ b/mlir/test/Dialect/Shape/canonicalize.mlir
@@ -1389,10 +1389,25 @@ func.func @shape_of_from_reshape(%arg0: tensor<*xf32>, %arg1: tensor<?xindex>) -
// -----
-// CHECK-LABEL: func @shape_of_from_reshape_compatible_types
+// Check statically shaped types, with element types i32 to index.
+// CHECK-LABEL: func @shape_of_from_reshape_compatible_types1
+// CHECK-SAME: %[[INPUT:.*]]: tensor<?x1xf32>
+// CHECK-SAME: %[[SHAPE:.*]]: tensor<3xi32>
+func.func @shape_of_from_reshape_compatible_types1(%arg0: tensor<?x1xf32>, %arg1: tensor<3xi32>) -> tensor<3xindex> {
+ // CHECK: %[[CAST_SHAPE:.*]] = arith.index_cast %[[SHAPE]] : tensor<3xi32> to tensor<3xindex>
+ // CHECK: return %[[CAST_SHAPE]] : tensor<3xindex>
+ %0 = tensor.reshape %arg0(%arg1) : (tensor<?x1xf32>, tensor<3xi32>) -> tensor<?x1x1xf32>
+ %1 = shape.shape_of %0 : tensor<?x1x1xf32> -> tensor<3xindex>
+ return %1 : tensor<3xindex>
+}
+
+// -----
+
+// Check similar element types, with statically shaped to dynamically shaped.
+// CHECK-LABEL: func @shape_of_from_reshape_compatible_types2
// CHECK-SAME: %[[INPUT:.*]]: tensor<*xf32>
// CHECK-SAME: %[[SHAPE:.*]]: tensor<5xindex>
-func.func @shape_of_from_reshape_compatible_types(%arg0: tensor<*xf32>, %arg1: tensor<5xindex>) -> tensor<?xindex> {
+func.func @shape_of_from_reshape_compatible_types2(%arg0: tensor<*xf32>, %arg1: tensor<5xindex>) -> tensor<?xindex> {
// CHECK: %[[CAST_SHAPE:.*]] = tensor.cast %[[SHAPE]] : tensor<5xindex> to tensor<?xindex>
// CHECK: return %[[CAST_SHAPE]] : tensor<?xindex>
%0 = tensor.reshape %arg0(%arg1) : (tensor<*xf32>, tensor<5xindex>) -> tensor<*xf32>
@@ -1402,6 +1417,20 @@ func.func @shape_of_from_reshape_compatible_types(%arg0: tensor<*xf32>, %arg1: t
// -----
+// Check similar element types, with dynamically shaped to statically shaped.
+// CHECK-LABEL: func @shape_of_from_reshape_compatible_types3
+// CHECK-SAME: %[[INPUT:.*]]: tensor<*xf32>
+// CHECK-SAME: %[[SHAPE:.*]]: tensor<?xindex>
+func.func @shape_of_from_reshape_compatible_types3(%arg0: tensor<*xf32>, %arg1: tensor<?xindex>) -> tensor<5xindex> {
+ // CHECK: %[[CAST_SHAPE:.*]] = tensor.cast %[[SHAPE]] : tensor<?xindex> to tensor<5xindex>
+ // CHECK: return %[[CAST_SHAPE]] : tensor<5xindex>
+ %0 = tensor.reshape %arg0(%arg1) : (tensor<*xf32>, tensor<?xindex>) -> tensor<*xf32>
+ %1 = shape.shape_of %0 : tensor<*xf32> -> tensor<5xindex>
+ return %1 : tensor<5xindex>
+}
+
+// -----
+
// CHECK-LABEL: func @shape_of_from_reshape_nofold
// CHECK-SAME: %[[INPUT:.*]]: tensor<*xf32>
// CHECK-SAME: %[[SHAPE:.*]]: tensor<?xindex>
|
|
[like] Ivan Garcia Alsina reacted to your message:
________________________________
From: Alaa Ali ***@***.***>
Sent: Thursday, April 3, 2025 9:10:04 PM
To: llvm/llvm-project ***@***.***>
Cc: Ivan Garcia Alsina ***@***.***>; Mention ***@***.***>
Subject: Re: [llvm/llvm-project] [MLIR] Fix canonicalization pattern for 'shape.shape_of' (PR #134234)
@alaa-ali commented on this pull request.
________________________________
In mlir/lib/Dialect/Shape/IR/Shape.cpp<#134234 (comment)>:
Value shape = tensorReshapeOp.getShape();
- if (op.getType() != shape.getType())
- shape = rewriter.create<tensor::CastOp>(op.getLoc(), op.getType(), shape);
+
+ auto opTensorType = llvm::dyn_cast<RankedTensorType>(op.getType());
+ auto shapeTensorType = llvm::dyn_cast<RankedTensorType>(shape.getType());
I figured out that both tensors must be ranked tensors and cannot be unranked.
So, I should have used llvm::cast() instead of llvm::dyn_cast().
1. 'tensor.reshape' op operand #1<#1> 'shape' must be 1D tensor of signless integer or index values.
2. 'shape.shape_of' op result #0 must be shape or extent tensor.
I updated Shape.cpp
Thanks for your feedback.
—
Reply to this email directly, view it on GitHub<#134234 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AIXTYTIJVCWKU3Y3CGJQSMD2XWPSZAVCNFSM6AAAAAB2L7AQJOVHI2DSMVQWIX3LMV43YUDVNRWFEZLROVSXG5CSMV3GSZLXHMZDONBRGE4DAOJYHE>.
You are receiving this because you were mentioned.Message ID: ***@***.***>
|
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
Co-authored-by: Mehdi Amini <[email protected]>
Co-authored-by: Mehdi Amini <[email protected]>
|
@alaa-ali Congratulations on having your first Pull Request (PR) merged into the LLVM Project! Your changes will be combined with recent changes from other authors, then tested by our build bots. If there is a problem with a build, you may receive a report in an email or a comment on this PR. Please check whether problems have been caused by your change specifically, as the builds can include changes from many authors. It is not uncommon for your change to be included in a build that fails due to someone else's changes, or infrastructure issues. How to do this, and the rest of the post-merge process, is covered in detail here. If your change does cause a problem, it may be reverted, or you can revert it yourself. This is a normal part of LLVM development. You can fix your changes and open a new PR to merge them again. If you don't get any reports, no action is required from you. Your changes are working as expected, well done! |
This PR will fix a bug in a canonicalization pattern (operation shape.shape_of: shape of reshape)
See file canonicalize.mlir in the change list for an example.
For the context, this bug was found while running a test on Keras 3, the canonicalizer errors out due to an invalid tensor.cast operation when the batch size is dynamic.
The operands of the op are tensor<3xi32> cast to tensor<3xindex>.
This change is related to a previous PR: #98531