Let memref.collapse_shape implement ReifyRankedShapedTypeOpInterface. #138452
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
This MR implements ReifyRankedShapedTypeOpInterface for memref.collapse_shape and adds support in reifyResultShapes for memref.dim to operate directly on shaped values, eliminating reliance on collapse_shape. The new logic fully supports all collapse sizes and reifies dynamic dimensions, improving shape inference and lowering fidelity.
|
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-memref Author: None (davidlerner96) ChangesThis MR implements ReifyRankedShapedTypeOpInterface for memref.collapse_shape and adds support in reifyResultShapes for memref.dim to operate directly on shaped values, eliminating reliance on collapse_shape. The new logic fully supports all collapse sizes and reifies dynamic dimensions, improving shape inference and lowering fidelity. Full diff: https://github.com/llvm/llvm-project/pull/138452.diff 3 Files Affected:
diff --git a/mlir/include/mlir/Dialect/MemRef/IR/MemRefOps.td b/mlir/include/mlir/Dialect/MemRef/IR/MemRefOps.td
index d6d8161d3117b..e401e3e8a53ae 100644
--- a/mlir/include/mlir/Dialect/MemRef/IR/MemRefOps.td
+++ b/mlir/include/mlir/Dialect/MemRef/IR/MemRefOps.td
@@ -1761,7 +1761,8 @@ def MemRef_ExpandShapeOp : MemRef_ReassociativeReshapeOp<"expand_shape", [
}
def MemRef_CollapseShapeOp : MemRef_ReassociativeReshapeOp<"collapse_shape", [
- DeclareOpInterfaceMethods<OpAsmOpInterface, ["getAsmResultNames"]>]> {
+ DeclareOpInterfaceMethods<OpAsmOpInterface, ["getAsmResultNames"]>,
+ DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface>]>{
let summary = "operation to produce a memref with a smaller rank.";
let description = [{
The `memref.collapse_shape` op produces a new view with a smaller rank
diff --git a/mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp b/mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp
index 6f10a31c15626..177b4a69d256f 100644
--- a/mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp
+++ b/mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp
@@ -2482,6 +2482,37 @@ MemRefType CollapseShapeOp::computeCollapsedType(
srcType.getMemorySpace());
}
+// This method handles groups of dimensions where at least one dimension is dynamic.
+// For each such group, it computes the combined size by multiplying all the sizes
+// of the dimensions in that group. These computed sizes are then used to describe
+// the resulting shape after collapsing.
+LogicalResult CollapseShapeOp::reifyResultShapes(
+ OpBuilder &builder, ReifiedRankedShapedTypeDims &reifiedResultShapes) {
+ SmallVector<ReassociationIndices, 4> reassociationArray =
+ getReassociationIndices();
+ Value source = getSrc();
+ Location loc = getLoc();
+ SmallVector<Value> dynamicValues;
+ auto resultShape = cast<ShapedType>(getResultType()).getShape();
+ auto sourceShape = cast<MemRefType>(source.getType()).getShape();
+ for (auto group : reassociationArray) {
+ if (!llvm::any_of(group, [&](int64_t dim) {
+ return ShapedType::isDynamic(sourceShape[dim]);
+ }))
+ continue;
+ Value resultVal = builder.create<memref::DimOp>(loc, source, group[0]);
+ for (auto dim : llvm::drop_begin(group)) {
+ Value nextVal = builder.create<memref::DimOp>(loc, source, dim);
+ resultVal = builder.create<arith::MulIOp>(loc, resultVal, nextVal);
+ }
+
+ dynamicValues.push_back(resultVal);
+ }
+
+ reifiedResultShapes = {getMixedValues(resultShape, dynamicValues, builder)};
+ return success();
+}
+
void CollapseShapeOp::build(OpBuilder &b, OperationState &result, Value src,
ArrayRef<ReassociationIndices> reassociation,
ArrayRef<NamedAttribute> attrs) {
diff --git a/mlir/test/Dialect/MemRef/resolve-dim-ops.mlir b/mlir/test/Dialect/MemRef/resolve-dim-ops.mlir
index e354eb91d7557..f40b0ad849fa0 100644
--- a/mlir/test/Dialect/MemRef/resolve-dim-ops.mlir
+++ b/mlir/test/Dialect/MemRef/resolve-dim-ops.mlir
@@ -97,3 +97,94 @@ func.func @iter_to_init_arg_loop_like(
}
return %result : tensor<?x?xf32>
}
+
+// -----
+
+// CHECK-LABEL: func.func @collapse_dynamic_with_unit_dims(
+// CHECK-SAME: %[[arg0:.*]]: memref<1x32x?x1xsi8>) -> index {
+// CHECK: %[[c2:.*]] = arith.constant 2 : index
+// CHECK: %[[dim:.*]] = memref.dim %[[arg0]], %[[c2]] : memref<1x32x?x1xsi8>
+// CHECK: return %[[dim]] : index
+// CHECK: }
+func.func @collapse_dynamic_with_unit_dims (%arg0: memref<1x32x?x1xsi8>)
+ -> index {
+ %c2 = arith.constant 2 : index
+ %collapse_shape = memref.collapse_shape %arg0 [[0], [1], [2, 3]] : memref<1x32x?x1xsi8> into memref<1x32x?xsi8>
+ %dim_3 = memref.dim %collapse_shape, %c2 : memref<1x32x?xsi8>
+ return %dim_3: index
+}
+
+// -----
+
+// CHECK-LABEL: func.func @fold_dynamic_and_const_with_dynamic_on_right(
+// CHECK-SAME: %[[arg0:.*]]: memref<1x32x8x?xsi8>) -> index {
+// CHECK: %[[c8:.*]] = arith.constant 8 : index
+// CHECK: %[[c3:.*]] = arith.constant 3 : index
+// CHECK: %[[dim:.*]] = memref.dim %[[arg0]], %[[c3]] : memref<1x32x8x?xsi8>
+// CHECK: %[[res:.*]] = arith.muli %[[dim]], %[[c8]] : index
+// CHECK: return %[[res]] : index
+// CHECK: }
+func.func @fold_dynamic_and_const_with_dynamic_on_right(%arg0: memref<1x32x8x?xsi8>)
+ -> index {
+ %c2 = arith.constant 2 : index
+ %collapse_shape = memref.collapse_shape %arg0 [[0], [1], [2, 3]] : memref<1x32x8x?xsi8> into memref<1x32x?xsi8>
+ %dim_3 = memref.dim %collapse_shape, %c2 : memref<1x32x?xsi8>
+ return %dim_3: index
+}
+
+// -----
+
+// CHECK-LABEL: func.func @fold_dynamic_and_const_with_dynamic_on_left(
+// CHECK-SAME: %[[arg0:.*]]: memref<1x32x?x8xsi8>) -> index {
+// CHECK: %[[c8:.*]] = arith.constant 8 : index
+// CHECK: %[[c2:.*]] = arith.constant 2 : index
+// CHECK: %[[dim:.*]] = memref.dim %[[arg0]], %[[c2]] : memref<1x32x?x8xsi8>
+// CHECK: %[[res:.*]] = arith.muli %[[dim]], %[[c8]] : index
+// CHECK: return %[[res]] : index
+// CHECK: }
+func.func @fold_dynamic_and_const_with_dynamic_on_left(%arg0: memref<1x32x?x8xsi8>)
+ -> index {
+ %c2 = arith.constant 2 : index
+ %collapse_shape = memref.collapse_shape %arg0 [[0], [1], [2, 3]] : memref<1x32x?x8xsi8> into memref<1x32x?xsi8>
+ %dim_3 = memref.dim %collapse_shape, %c2 : memref<1x32x?xsi8>
+ return %dim_3: index
+}
+
+// -----
+
+// CHECK-LABEL: func.func @fold_more_than_two_elements_group(
+// CHECK-SAME: %[[arg0:.*]]: memref<2x32x?x8xsi8>) -> index {
+// CHECK: %[[c8:.*]] = arith.constant 8 : index
+// CHECK: %[[c64:.*]] = arith.constant 64 : index
+// CHECK: %[[c2:.*]] = arith.constant 2 : index
+// CHECK: %[[dim:.*]] = memref.dim %[[arg0]], %[[c2]] : memref<2x32x?x8xsi8>
+// CHECK: %[[res0:.*]] = arith.muli %[[dim]], %[[c64]] : index
+// CHECK: %[[res1:.*]] = arith.muli %[[res0]], %[[c8]] : index
+// CHECK: return %[[res1]] : index
+// CHECK: }
+func.func @fold_more_than_two_elements_group(%arg0: memref<2x32x?x8xsi8>)
+ -> index {
+ %c1 = arith.constant 0 : index
+ %collapse_shape = memref.collapse_shape %arg0 [[0, 1, 2, 3]] : memref<2x32x?x8xsi8> into memref<?xsi8>
+ %dim_3 = memref.dim %collapse_shape, %c1 : memref<?xsi8>
+ return %dim_3: index
+}
+
+// -----
+
+// CHECK-LABEL: func.func @fold_group_with_two_dynamic(
+// CHECK-SAME: %[[arg0:.*]]: memref<1x32x?x?xsi8>) -> index {
+// CHECK: %[[c3:.*]] = arith.constant 3 : index
+// CHECK: %[[c2:.*]] = arith.constant 2 : index
+// CHECK: %[[dim2:.*]] = memref.dim %[[arg0]], %[[c2]] : memref<1x32x?x?xsi8>
+// CHECK: %[[dim3:.*]] = memref.dim %[[arg0]], %[[c3]] : memref<1x32x?x?xsi8>
+// CHECK: %[[res:.*]] = arith.muli %[[dim2]], %[[dim3]] : index
+// CHECK: return %[[res]] : index
+// CHECK: }
+func.func @fold_group_with_two_dynamic(%arg0: memref<1x32x?x?xsi8>)
+ -> index {
+ %c2 = arith.constant 2 : index
+ %collapse_shape = memref.collapse_shape %arg0 [[0], [1], [2, 3]] : memref<1x32x?x?xsi8> into memref<1x32x?xsi8>
+ %dim_3 = memref.dim %collapse_shape, %c2 : memref<1x32x?xsi8>
+ return %dim_3: index
+}
|
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.
This MR implements ReifyRankedShapedTypeOpInterface for memref.collapse_shape and adds support in reifyResultShapes for memref.dim to operate directly on shaped values, eliminating reliance on collapse_shape. The new logic fully supports all collapse sizes and reifies dynamic dimensions, improving shape inference and lowering fidelity.