[MLIR][Tensor,MemRef] Fold expand_shape and collapse_shape if identity #80658

newling · 2024-02-05T10:50:40Z

Before: op verifiers failed if the input and output ranks were the same (i.e. no expansion or collapse). This behavior requires users of these shape ops to verify manually that they are not creating identity versions of these ops every time they build them -- problematic. This PR removes this strict verification, and introduces folders for the the identity cases.

The PR also removes the special case handling of rank-0 tensors for expand_shape and collapse_shape, there doesn't seem to be any reason to treat them differently.

github-actions · 2024-02-05T10:50:56Z

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 @ followed by their GitHub username.

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.

llvmbot · 2024-02-05T10:51:25Z

@llvm/pr-subscribers-mlir-tensor
@llvm/pr-subscribers-mlir-memref

@llvm/pr-subscribers-mlir

Author: James Newling (newling)

Changes

Before: op verifiers failed if the input and output ranks were the same (i.e. no expansion or collapse). This behaviour requires users of these shape ops to verify manually that they are not creating identity versions of these ops every time they build them -- problematic. This PR removes this strict verification, and introduces folders for the the identity cases.

Full diff: https://github.com/llvm/llvm-project/pull/80658.diff

7 Files Affected:

(modified) mlir/include/mlir/Dialect/Utils/ReshapeOpsUtils.h (-2)
(modified) mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp (+4-8)
(modified) mlir/lib/Dialect/Tensor/IR/TensorOps.cpp (+4-12)
(modified) mlir/test/Dialect/MemRef/canonicalize.mlir (+19)
(modified) mlir/test/Dialect/MemRef/invalid.mlir (-14)
(modified) mlir/test/Dialect/Tensor/canonicalize.mlir (+20-1)
(modified) mlir/test/Dialect/Tensor/invalid.mlir (-14)

diff --git a/mlir/include/mlir/Dialect/Utils/ReshapeOpsUtils.h b/mlir/include/mlir/Dialect/Utils/ReshapeOpsUtils.h
index 61c929dee0f27..3a672a1cc6060 100644
--- a/mlir/include/mlir/Dialect/Utils/ReshapeOpsUtils.h
+++ b/mlir/include/mlir/Dialect/Utils/ReshapeOpsUtils.h
@@ -111,8 +111,6 @@ static LogicalResult verifyReshapeLikeTypes(Op op, T expandedType,
            << " to have higher rank than the type = " << collapsedType;
   if (expandedRank == 0)
     return op.emitOpError("expected non-zero memref ranks");
-  if (expandedRank == collapsedRank)
-    return op.emitOpError("expected to collapse or expand dims");
 
   if (collapsedRank == 0) {
     // If collapsed rank is 0, then expanded type must be static shaped and of
diff --git a/mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp b/mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp
index b79ab8f3d671e..b79022d675bd6 100644
--- a/mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp
+++ b/mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp
@@ -2223,10 +2223,6 @@ LogicalResult ExpandShapeOp::verify() {
   MemRefType srcType = getSrcType();
   MemRefType resultType = getResultType();
 
-  if (srcType.getRank() >= resultType.getRank())
-    return emitOpError("expected rank expansion, but found source rank ")
-           << srcType.getRank() << " >= result rank " << resultType.getRank();
-
   // Verify result shape.
   if (failed(verifyCollapsedShape(getOperation(), srcType.getShape(),
                                   resultType.getShape(),
@@ -2377,10 +2373,6 @@ LogicalResult CollapseShapeOp::verify() {
   MemRefType srcType = getSrcType();
   MemRefType resultType = getResultType();
 
-  if (srcType.getRank() <= resultType.getRank())
-    return emitOpError("expected rank reduction, but found source rank ")
-           << srcType.getRank() << " <= result rank " << resultType.getRank();
-
   // Verify result shape.
   if (failed(verifyCollapsedShape(getOperation(), resultType.getShape(),
                                   srcType.getShape(), getReassociationIndices(),
@@ -2455,11 +2447,15 @@ void CollapseShapeOp::getCanonicalizationPatterns(RewritePatternSet &results,
 }
 
 OpFoldResult ExpandShapeOp::fold(FoldAdaptor adaptor) {
+  if (getSrcType() == getType())
+    return getSrc();
   return foldReshapeOp<ExpandShapeOp, CollapseShapeOp>(*this,
                                                        adaptor.getOperands());
 }
 
 OpFoldResult CollapseShapeOp::fold(FoldAdaptor adaptor) {
+  if (getSrcType() == getType())
+    return getSrc();
   return foldReshapeOp<CollapseShapeOp, ExpandShapeOp>(*this,
                                                        adaptor.getOperands());
 }
diff --git a/mlir/lib/Dialect/Tensor/IR/TensorOps.cpp b/mlir/lib/Dialect/Tensor/IR/TensorOps.cpp
index b2fe58099b2fb..591cbcde80084 100644
--- a/mlir/lib/Dialect/Tensor/IR/TensorOps.cpp
+++ b/mlir/lib/Dialect/Tensor/IR/TensorOps.cpp
@@ -1656,22 +1656,10 @@ static LogicalResult verifyTensorReshapeOp(TensorReshapeOp op,
 }
 
 LogicalResult ExpandShapeOp::verify() {
-  auto srcType = getSrcType();
-  auto resultType = getResultType();
-  if (srcType.getRank() >= resultType.getRank())
-    return emitOpError("expected rank expansion, but found source rank ")
-           << srcType.getRank() << " >= result rank " << resultType.getRank();
-
   return verifyTensorReshapeOp(*this, getResultType(), getSrcType());
 }
 
 LogicalResult CollapseShapeOp::verify() {
-  auto srcType = getSrcType();
-  auto resultType = getResultType();
-  if (srcType.getRank() <= resultType.getRank())
-    return emitOpError("expected rank reduction, but found source rank ")
-           << srcType.getRank() << " <= result rank " << resultType.getRank();
-
   return verifyTensorReshapeOp(*this, getSrcType(), getResultType());
 }
 
@@ -1872,11 +1860,15 @@ void CollapseShapeOp::getCanonicalizationPatterns(RewritePatternSet &results,
 }
 
 OpFoldResult ExpandShapeOp::fold(FoldAdaptor adaptor) {
+  if (getSrcType() == getType())
+    return getSrc();
   return foldReshapeOp<ExpandShapeOp, CollapseShapeOp>(*this,
                                                        adaptor.getOperands());
 }
 
 OpFoldResult CollapseShapeOp::fold(FoldAdaptor adaptor) {
+  if (getSrcType() == getType())
+    return getSrc();
   return foldReshapeOp<CollapseShapeOp, ExpandShapeOp>(*this,
                                                        adaptor.getOperands());
 }
diff --git a/mlir/test/Dialect/MemRef/canonicalize.mlir b/mlir/test/Dialect/MemRef/canonicalize.mlir
index eccfc485b2034..8c0b765c1b123 100644
--- a/mlir/test/Dialect/MemRef/canonicalize.mlir
+++ b/mlir/test/Dialect/MemRef/canonicalize.mlir
@@ -1,5 +1,24 @@
 // RUN: mlir-opt %s -canonicalize="test-convergence" --split-input-file -allow-unregistered-dialect | FileCheck %s
 
+
+// CHECK-LABEL: collapse_shape_identity_fold
+// CHECK-NEXT: return
+func.func @collapse_shape_identity_fold(%arg0 : memref<5xi8>) -> memref<5xi8> {
+  %0 = memref.collapse_shape %arg0 [[0]] : memref<5xi8> into memref<5xi8>
+  return %0 : memref<5xi8>
+}
+
+// -----
+
+// CHECK-LABEL: expand_shape_identity_fold
+// CHECK-NEXT: return
+func.func @expand_shape_identity_fold(%arg0 : memref<5x4xi8>) -> memref<5x4xi8> {
+  %0 = memref.expand_shape %arg0 [[0], [1]] : memref<5x4xi8> into memref<5x4xi8>
+  return %0 : memref<5x4xi8>
+}
+
+// -----
+
 // CHECK-LABEL: func @subview_of_size_memcast
 //  CHECK-SAME:   %[[ARG0:.[a-z0-9A-Z_]+]]: memref<4x6x16x32xi8>
 //       CHECK:   %[[S:.+]] = memref.subview %[[ARG0]][0, 1, 0, 0] [1, 1, 16, 32] [1, 1, 1, 1] : memref<4x6x16x32xi8> to memref<16x32xi8, strided{{.*}}>
diff --git a/mlir/test/Dialect/MemRef/invalid.mlir b/mlir/test/Dialect/MemRef/invalid.mlir
index 7bb7a2affcbd1..53841fa836da7 100644
--- a/mlir/test/Dialect/MemRef/invalid.mlir
+++ b/mlir/test/Dialect/MemRef/invalid.mlir
@@ -415,20 +415,6 @@ func.func @collapse_shape_out_of_bounds(%arg0: memref<?x?xf32>) {
 
 // -----
 
-func.func @expand_shape_invalid_ranks(%arg0: memref<?x?xf32>) {
-  // expected-error @+1 {{op expected rank expansion, but found source rank 2 >= result rank 2}}
-  %0 = memref.expand_shape %arg0 [[0], [1]] : memref<?x?xf32> into memref<?x?xf32>
-}
-
-// -----
-
-func.func @collapse_shape_invalid_ranks(%arg0: memref<?x?xf32>) {
-  // expected-error @+1 {{op expected rank reduction, but found source rank 2 <= result rank 2}}
-  %0 = memref.collapse_shape %arg0 [[0], [1]] : memref<?x?xf32> into memref<?x?xf32>
-}
-
-// -----
-
 func.func @expand_shape_out_of_bounds(%arg0: memref<?xf32>) {
   // expected-error @+1 {{op reassociation index 2 is out of bounds}}
   %0 = memref.expand_shape %arg0 [[0, 1, 2]] : memref<?xf32> into memref<4x?xf32>
diff --git a/mlir/test/Dialect/Tensor/canonicalize.mlir b/mlir/test/Dialect/Tensor/canonicalize.mlir
index ed964071358ac..9b31d858d63c5 100644
--- a/mlir/test/Dialect/Tensor/canonicalize.mlir
+++ b/mlir/test/Dialect/Tensor/canonicalize.mlir
@@ -1,5 +1,24 @@
 // RUN: mlir-opt %s -split-input-file -canonicalize="test-convergence" | FileCheck %s
 
+
+// CHECK-LABEL: expand_shape_identity_fold
+// CHECK-NEXT: return
+func.func @expand_shape_identity_fold(%arg0 : tensor<5xf32>) -> tensor<5xf32> {
+  %0 = tensor.expand_shape %arg0 [[0]] : tensor<5xf32> into tensor<5xf32>
+  return %0 : tensor<5xf32>
+}
+
+// -----
+
+// CHECK-LABEL: collapse_shape_identity_fold
+// CHECK-NEXT: return
+func.func @collapse_shape_identity_fold(%arg0 : tensor<5x4xf32>) -> tensor<5x4xf32> {
+  %0 = tensor.collapse_shape %arg0 [[0], [1]] : tensor<5x4xf32> into tensor<5x4xf32>
+  return %0 : tensor<5x4xf32>
+}
+
+// -----
+
 // CHECK-LABEL: @tensor_bitcast_chain_ok
 // CHECK-SAME: %[[IN:.*]]: tensor<2xi32>
 func.func @tensor_bitcast_chain_ok(%input: tensor<2xi32>) -> tensor<2xf32> {
@@ -1977,7 +1996,7 @@ func.func @unpack_pack(%t: tensor<128x128xf32>) -> tensor<128x128xf32> {
 
 // Chain: NC -> NCnc -> NCnc -> NC
 // CHECK: func.func @unpack_pack(
-// CHECK-SAME: %[[T:.+]]: tensor<128x128xf32>, 
+// CHECK-SAME: %[[T:.+]]: tensor<128x128xf32>,
 // CHECK: return %[[T]] : tensor<128x128xf32>
 func.func @unpack_pack(%t: tensor<128x128xf32>, %tile1: index, %tile2: index) -> tensor<128x128xf32> {
   %tensor_empty = tensor.empty(%tile1, %tile2) : tensor<16x16x?x?xf32>
diff --git a/mlir/test/Dialect/Tensor/invalid.mlir b/mlir/test/Dialect/Tensor/invalid.mlir
index 735e5146e9dbc..39cd3788cb081 100644
--- a/mlir/test/Dialect/Tensor/invalid.mlir
+++ b/mlir/test/Dialect/Tensor/invalid.mlir
@@ -343,20 +343,6 @@ func.func @illegal_collapsing_reshape_mixed_tensor_2(%arg0 : tensor<?x4x5xf32>)
 
 // -----
 
-func.func @expand_shape_invalid_ranks(%arg0: tensor<?x?xf32>) {
-  // expected-error @+1 {{op expected rank expansion, but found source rank 2 >= result rank 2}}
-  %0 = tensor.expand_shape %arg0 [[0], [1]] : tensor<?x?xf32> into tensor<?x?xf32>
-}
-
-// -----
-
-func.func @collapse_shape_invalid_ranks(%arg0: tensor<?x?xf32>) {
-  // expected-error @+1 {{op expected rank reduction, but found source rank 2 <= result rank 2}}
-  %0 = tensor.collapse_shape %arg0 [[0], [1]] : tensor<?x?xf32> into tensor<?x?xf32>
-}
-
-// -----
-
 func.func @rank(%0: f32) {
   // expected-error@+1 {{'tensor.rank' op operand #0 must be tensor of any type values}}
   "tensor.rank"(%0): (f32)->index

matthias-springer

Allowing identity expansions/collapsion (and folding them) makes sense to me. Interestingly, I have seen this pattern (disallowing "identity" ops) in other places. E.g., certain "cast" ops that do not change the type fail verification.

mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp

mlir/test/Dialect/Tensor/invalid.mlir

github-actions · 2024-02-05T13:36:06Z

✅ With the latest revision this PR passed the C/C++ code formatter.

newling · 2024-02-12T16:22:52Z

Ping

newling · 2024-02-16T19:12:05Z

@nicolasvasilache and @matthias-springer please let me know if you can review this, else I can ask someone else

matthias-springer

Sorry for the delay

mlir/test/Dialect/MemRef/invalid.mlir

…ase handling

newling · 2024-03-11T18:42:46Z

Finally made it through CI.

@matthias-springer I don't have write access to this repo. Would you mind merging this PR for me?

github-actions · 2024-03-12T01:12:15Z

@newling 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 recieve 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!

llvmbot added mlir mlir:tensor mlir:memref labels Feb 5, 2024

matthias-springer reviewed Feb 5, 2024

View reviewed changes

mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp Show resolved Hide resolved

matthias-springer requested a review from nicolasvasilache February 5, 2024 11:19

matthias-springer reviewed Feb 5, 2024

View reviewed changes

mlir/test/Dialect/Tensor/invalid.mlir Show resolved Hide resolved

newling requested a review from matthias-springer February 19, 2024 16:52

matthias-springer approved these changes Feb 19, 2024

View reviewed changes

mlir/test/Dialect/MemRef/invalid.mlir Outdated Show resolved Hide resolved

newling force-pushed the relax_expand_collapse_verifiers branch from 5da6330 to 1a09fd9 Compare February 24, 2024 21:48

newling added 5 commits February 26, 2024 13:34

let identity versions pass

847a899

test

7906997

updates. additional testing, doc corrections, remove rank-0 special c…

75ddea2

…ase handling

clang-format

3684e45

improved error message when rank change is not correct

e9088b0

newling force-pushed the relax_expand_collapse_verifiers branch from 1a09fd9 to e9088b0 Compare February 26, 2024 21:35

newling added 4 commits February 27, 2024 21:23

Merge branch 'main' into relax_expand_collapse_verifiers

a42b3e5

Merge branch 'main' into relax_expand_collapse_verifiers

689e7b1

Merge branch 'main' into relax_expand_collapse_verifiers

85eb5d0

Merge branch 'main' into relax_expand_collapse_verifiers

64d4798

matthias-springer merged commit 67ef4ae into llvm:main Mar 12, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[MLIR][Tensor,MemRef] Fold expand_shape and collapse_shape if identity #80658

[MLIR][Tensor,MemRef] Fold expand_shape and collapse_shape if identity #80658

Uh oh!

newling commented Feb 5, 2024 •

edited

Loading

Uh oh!

github-actions bot commented Feb 5, 2024

Uh oh!

llvmbot commented Feb 5, 2024 •

edited

Loading

Uh oh!

matthias-springer left a comment

Uh oh!

Uh oh!

Uh oh!

github-actions bot commented Feb 5, 2024 •

edited

Loading

Uh oh!

newling commented Feb 12, 2024

Uh oh!

newling commented Feb 16, 2024

Uh oh!

matthias-springer left a comment

Uh oh!

Uh oh!

newling commented Mar 11, 2024

Uh oh!

github-actions bot commented Mar 12, 2024

Uh oh!

Uh oh!

[MLIR][Tensor,MemRef] Fold expand_shape and collapse_shape if identity #80658

[MLIR][Tensor,MemRef] Fold expand_shape and collapse_shape if identity #80658

Uh oh!

Conversation

newling commented Feb 5, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented Feb 5, 2024

Uh oh!

llvmbot commented Feb 5, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

matthias-springer left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

github-actions bot commented Feb 5, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

newling commented Feb 12, 2024

Uh oh!

newling commented Feb 16, 2024

Uh oh!

matthias-springer left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

newling commented Mar 11, 2024

Uh oh!

github-actions bot commented Mar 12, 2024

Uh oh!

Uh oh!

newling commented Feb 5, 2024 •

edited

Loading

llvmbot commented Feb 5, 2024 •

edited

Loading

github-actions bot commented Feb 5, 2024 •

edited

Loading