[mlir] Fix ComposeExpandOfCollapseOp for dynamic case

[IanWood1]

Changes findCollapsingReassociation to return nullopt in all cases where source shape has >=2 dynamic dims. expand(collapse) can reshape to in any valid output shape but a collapse can only collapse contiguous dimensions. When there are >=2 dynamic dimensions it is impossible to determine if it can be simplified to a collapse or if it is preforming a more advanced reassociation.

This problem was uncovered by #137963

[llvmbot]

@llvm/pr-subscribers-mlir

@llvm/pr-subscribers-mlir-tensor

Author: Ian Wood (IanWood1)

Changes

Changes findCollapsingReassociation to return nullopt in all cases where source shape has >=2 dynamic dims. expand(collapse) can reshape to in any valid output shape but a single collapse can only collapse contiguous dimensions. When there are >=2 dynamic dimensions it is impossible to determine if it can be simplified to a collapse or if it is preforming a more advanced reassociation.

---

Full diff: https://github.com/llvm/llvm-project/pull/142663.diff

2 Files Affected:

• (modified) mlir/include/mlir/Dialect/Utils/ReshapeOpsUtils.h (+5-3)
• (modified) mlir/test/Dialect/Tensor/canonicalize.mlir (+14)

diff --git a/mlir/include/mlir/Dialect/Utils/ReshapeOpsUtils.h b/mlir/include/mlir/Dialect/Utils/ReshapeOpsUtils.h
index af575e10acc8e..c9f29ab8f15d5 100644
--- a/mlir/include/mlir/Dialect/Utils/ReshapeOpsUtils.h
+++ b/mlir/include/mlir/Dialect/Utils/ReshapeOpsUtils.h
@@ -387,11 +387,13 @@ struct ComposeExpandOfCollapseOp : public OpRewritePattern<ExpandOpTy> {
       auto resultSubShape =
           resultShape.slice(resultIndices.front(), resultIndices.size());
 
+      if (llvm::count_if(srcSubShape, ShapedType::isDynamic) >= 2)
+        return std::nullopt;
+
       if (srcSubShape.size() == resultSubShape.size()) {
-        if (srcSubShape != resultSubShape ||
-            llvm::count_if(srcSubShape, ShapedType::isDynamic) >= 2) {
+        if (srcSubShape != resultSubShape)
           return std::nullopt;
-        }
+
         for (auto index : llvm::seq<int64_t>(0, srcSubShape.size())) {
           composedReassociation.emplace_back(1, srcIndices.front() + index);
         }
diff --git a/mlir/test/Dialect/Tensor/canonicalize.mlir b/mlir/test/Dialect/Tensor/canonicalize.mlir
index 646b2197d9aa6..7e5e57423f44d 100644
--- a/mlir/test/Dialect/Tensor/canonicalize.mlir
+++ b/mlir/test/Dialect/Tensor/canonicalize.mlir
@@ -1319,6 +1319,20 @@ func.func @compose_expand_of_collapse_dynamic(%arg0 : tensor<4x?x10x64x2xf16>, %
 
 // -----
 
+func.func @no_compose_collapse_of_expand_dynamic(%arg0 : tensor<?x8x128x?xf16>, %arg1: index) -> tensor<?x128x?xf16> {
+  %collapse = tensor.collapse_shape %arg0 [[0, 1, 2, 3]] : tensor<?x8x128x?xf16> into tensor<?xf16>
+  %expanded_19 = tensor.expand_shape %collapse [[0, 1, 2]] output_shape [%arg1, 8, %arg1] : tensor<?xf16> into tensor<?x128x?xf16>
+  return %expanded_19 : tensor<?x128x?xf16>
+}
+// CHECK-LABEL: func @no_compose_collapse_of_expand_dynamic
+//  CHECK-SAME:   %[[ARG0:.+]]: tensor
+//  CHECK-SAME:   %[[ARG1:.+]]: index
+//       CHECK:   %[[COLLAPSE:.+]] = tensor.collapse_shape %[[ARG0]]
+//       CHECK:   %[[EXPAND:.+]] = tensor.expand_shape %[[COLLAPSE]]
+//       CHECK:   return %[[EXPAND]]
+
+// -----
+
 // CHECK-LABEL: func @zero_rank_reshape_multi
 func.func @zero_rank_reshape_multi(%arg0: tensor<f32>) -> tensor<f32> {
   // CHECK: return %arg0

[AGindinson]

LGTM after the previous thread, can't approve due to commit access still pending