Let memref.collapse_shape implement ReifyRankedShapedTypeOpInterface. #107752
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
|
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-memref Author: None (ddubov100) ChangesThis MR lets memref.collapse_shape implement ReifyRankedShapedTypeOpInterface.
Full diff: https://github.com/llvm/llvm-project/pull/107752.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 2ff9d612a5efa7..604611f2e8b5ea 100644
--- a/mlir/include/mlir/Dialect/MemRef/IR/MemRefOps.td
+++ b/mlir/include/mlir/Dialect/MemRef/IR/MemRefOps.td
@@ -1670,7 +1670,7 @@ def MemRef_ExpandShapeOp : MemRef_ReassociativeReshapeOp<"expand_shape", [
}]>,
// Builder that infers the result layout map. The result shape must be
- // specified. Otherwise, the op may be ambiguous. The output shape for
+ // specified. Otherwise, the op may be ambiguous. The output shape for
// the op will be inferred using the inferOutputShape() method.
OpBuilder<(ins "ArrayRef<int64_t>":$resultShape, "Value":$src,
"ArrayRef<ReassociationIndices>":$reassociation)>,
@@ -1699,7 +1699,9 @@ 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 779ffbfc23f4d7..48df018ee73d05 100644
--- a/mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp
+++ b/mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp
@@ -2497,6 +2497,42 @@ MemRefType CollapseShapeOp::computeCollapsedType(
srcType.getMemorySpace());
}
+static bool isDynamicInGroup(ReassociationIndices group,
+ ArrayRef<int64_t> sourceShape) {
+ return llvm::any_of(group, [sourceShape](int64_t dim) {
+ return ShapedType::isDynamic(sourceShape[dim]);
+ });
+}
+
+// This method supports following cases only:
+// - There is dynamic dimension in reassociation groups with single element.
+LogicalResult CollapseShapeOp::reifyResultShapes(
+ OpBuilder &builder, ReifiedRankedShapedTypeDims &reifiedResultShapes) {
+ SmallVector<ReassociationIndices, 4> reassociationArray =
+ getReassociationIndices();
+ Value source = getSrc();
+ auto sourceShape = cast<MemRefType>(source.getType()).getShape();
+ if (!ShapedType::isDynamicShape(sourceShape))
+ return failure();
+ for (auto group : enumerate(reassociationArray)) {
+ bool isDynInGroup = isDynamicInGroup(group.value(), sourceShape);
+ if (isDynInGroup && group.value().size() > 1)
+ return failure();
+ }
+ auto resultShape = cast<ShapedType>(getResultType()).getShape();
+
+ SmallVector<Value> dynamicValues;
+ for (int64_t i = 0; i < resultShape.size(); ++i) {
+ if (ShapedType::isDynamic(resultShape[i]))
+ dynamicValues.push_back(builder.create<DimOp>(
+ source.getLoc(), source,
+ builder.create<arith::ConstantIndexOp>(source.getLoc(), i)));
+ }
+ 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 85a4853972457c..912261515144d0 100644
--- a/mlir/test/Dialect/MemRef/resolve-dim-ops.mlir
+++ b/mlir/test/Dialect/MemRef/resolve-dim-ops.mlir
@@ -12,6 +12,71 @@ func.func @dim_out_of_bounds(%m : memref<7x8xf32>) -> index {
// -----
+// CHECK-LABEL: func.func @dyn_dim_of_memref_collapse_shape(
+// CHECK-SAME: %[[VAL_0:.*]]: memref<?x4x8x32xsi8>) -> index {
+// CHECK: %[[VAL_1:.*]] = arith.constant 0 : index
+// CHECK: %[[VAL_2:.*]] = memref.dim %[[VAL_0]], %[[VAL_1]] : memref<?x4x8x32xsi8>
+// CHECK: return %[[VAL_2]] : index
+// CHECK: }
+
+func.func @dyn_dim_of_memref_collapse_shape(%arg0: memref<?x4x8x32xsi8>)
+ -> index
+{
+ %c0 = arith.constant 0 : index
+ %dim_16 = memref.dim %arg0, %c0 : memref<?x4x8x32xsi8>
+ %alloc_17 = memref.alloc(%dim_16) {alignment = 32 : i64} : memref<?x32x4x8xsi8>
+ %collapse_shape = memref.collapse_shape %alloc_17 [[0], [1], [2, 3]] : memref<?x32x4x8xsi8> into memref<?x32x32xsi8>
+ %dim_18 = memref.dim %collapse_shape, %c0 : memref<?x32x32xsi8>
+ return %dim_18: index
+}
+
+// -----
+
+// CHECK-LABEL: func.func @resolve_when_collapse_after_collapse(
+// CHECK-SAME: %[[VAL_0:.*]]: memref<?x4x8x32xsi8>) -> index {
+// CHECK: %[[VAL_1:.*]] = arith.constant 0 : index
+// CHECK: %[[VAL_2:.*]] = memref.dim %[[VAL_0]], %[[VAL_1]] : memref<?x4x8x32xsi8>
+// CHECK: return %[[VAL_2]] : index
+// CHECK: }
+
+func.func @resolve_when_collapse_after_collapse(%arg0: memref<?x4x8x32xsi8>)
+ -> index
+{
+ %c0 = arith.constant 0 : index
+ %dim_16 = memref.dim %arg0, %c0 : memref<?x4x8x32xsi8>
+ %alloc_17 = memref.alloc(%dim_16) {alignment = 32 : i64} : memref<?x32x4x8xsi8>
+ %collapse_shape = memref.collapse_shape %alloc_17 [[0], [1], [2, 3]] : memref<?x32x4x8xsi8> into memref<?x32x32xsi8>
+ %collapse_shape_1 = memref.collapse_shape %collapse_shape [[0], [1, 2]] : memref<?x32x32xsi8> into memref<?x1024xsi8>
+ %dim_18 = memref.dim %collapse_shape_1, %c0 : memref<?x1024xsi8>
+ return %dim_18: index
+}
+
+// -----
+
+// CHECK-LABEL: func.func @unfoldable_memref_collapse_shape(
+// CHECK-SAME: %[[VAL_0:.*]]: memref<1x?x8x32xsi8>) -> index {
+// CHECK: %[[VAL_1:.*]] = arith.constant 1 : index
+// CHECK: %[[VAL_2:.*]] = arith.constant 2 : index
+// CHECK: %[[VAL_3:.*]] = memref.dim %[[VAL_0]], %[[VAL_1]] : memref<1x?x8x32xsi8>
+// CHECK: %[[VAL_4:.*]] = memref.alloc(%[[VAL_3]]) {alignment = 32 : i64} : memref<1x32x?x8xsi8>
+// CHECK: %[[VAL_5:.*]] = memref.collapse_shape %[[VAL_4]] {{\[\[}}0], [1], [2, 3]] : memref<1x32x?x8xsi8> into memref<1x32x?xsi8>
+// CHECK: %[[VAL_6:.*]] = memref.dim %[[VAL_5]], %[[VAL_2]] : memref<1x32x?xsi8>
+// CHECK: return %[[VAL_6]] : index
+// CHECK: }
+func.func @unfoldable_memref_collapse_shape(%arg0: memref<1x?x8x32xsi8>)
+ -> index
+{
+ %c1 = arith.constant 1 : index
+ %c2 = arith.constant 2 : index
+ %dim_1 = memref.dim %arg0, %c1 : memref<1x?x8x32xsi8>
+ %alloc_0 = memref.alloc(%dim_1) {alignment = 32 : i64} : memref<1x32x?x8xsi8>
+ %collapse_shape = memref.collapse_shape %alloc_0 [[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 @dim_out_of_bounds_2(
// CHECK-NEXT: arith.constant
// CHECK-NEXT: arith.constant
|
|
@llvm/pr-subscribers-mlir Author: None (ddubov100) ChangesThis MR lets memref.collapse_shape implement ReifyRankedShapedTypeOpInterface.
Full diff: https://github.com/llvm/llvm-project/pull/107752.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 2ff9d612a5efa7..604611f2e8b5ea 100644
--- a/mlir/include/mlir/Dialect/MemRef/IR/MemRefOps.td
+++ b/mlir/include/mlir/Dialect/MemRef/IR/MemRefOps.td
@@ -1670,7 +1670,7 @@ def MemRef_ExpandShapeOp : MemRef_ReassociativeReshapeOp<"expand_shape", [
}]>,
// Builder that infers the result layout map. The result shape must be
- // specified. Otherwise, the op may be ambiguous. The output shape for
+ // specified. Otherwise, the op may be ambiguous. The output shape for
// the op will be inferred using the inferOutputShape() method.
OpBuilder<(ins "ArrayRef<int64_t>":$resultShape, "Value":$src,
"ArrayRef<ReassociationIndices>":$reassociation)>,
@@ -1699,7 +1699,9 @@ 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 779ffbfc23f4d7..48df018ee73d05 100644
--- a/mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp
+++ b/mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp
@@ -2497,6 +2497,42 @@ MemRefType CollapseShapeOp::computeCollapsedType(
srcType.getMemorySpace());
}
+static bool isDynamicInGroup(ReassociationIndices group,
+ ArrayRef<int64_t> sourceShape) {
+ return llvm::any_of(group, [sourceShape](int64_t dim) {
+ return ShapedType::isDynamic(sourceShape[dim]);
+ });
+}
+
+// This method supports following cases only:
+// - There is dynamic dimension in reassociation groups with single element.
+LogicalResult CollapseShapeOp::reifyResultShapes(
+ OpBuilder &builder, ReifiedRankedShapedTypeDims &reifiedResultShapes) {
+ SmallVector<ReassociationIndices, 4> reassociationArray =
+ getReassociationIndices();
+ Value source = getSrc();
+ auto sourceShape = cast<MemRefType>(source.getType()).getShape();
+ if (!ShapedType::isDynamicShape(sourceShape))
+ return failure();
+ for (auto group : enumerate(reassociationArray)) {
+ bool isDynInGroup = isDynamicInGroup(group.value(), sourceShape);
+ if (isDynInGroup && group.value().size() > 1)
+ return failure();
+ }
+ auto resultShape = cast<ShapedType>(getResultType()).getShape();
+
+ SmallVector<Value> dynamicValues;
+ for (int64_t i = 0; i < resultShape.size(); ++i) {
+ if (ShapedType::isDynamic(resultShape[i]))
+ dynamicValues.push_back(builder.create<DimOp>(
+ source.getLoc(), source,
+ builder.create<arith::ConstantIndexOp>(source.getLoc(), i)));
+ }
+ 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 85a4853972457c..912261515144d0 100644
--- a/mlir/test/Dialect/MemRef/resolve-dim-ops.mlir
+++ b/mlir/test/Dialect/MemRef/resolve-dim-ops.mlir
@@ -12,6 +12,71 @@ func.func @dim_out_of_bounds(%m : memref<7x8xf32>) -> index {
// -----
+// CHECK-LABEL: func.func @dyn_dim_of_memref_collapse_shape(
+// CHECK-SAME: %[[VAL_0:.*]]: memref<?x4x8x32xsi8>) -> index {
+// CHECK: %[[VAL_1:.*]] = arith.constant 0 : index
+// CHECK: %[[VAL_2:.*]] = memref.dim %[[VAL_0]], %[[VAL_1]] : memref<?x4x8x32xsi8>
+// CHECK: return %[[VAL_2]] : index
+// CHECK: }
+
+func.func @dyn_dim_of_memref_collapse_shape(%arg0: memref<?x4x8x32xsi8>)
+ -> index
+{
+ %c0 = arith.constant 0 : index
+ %dim_16 = memref.dim %arg0, %c0 : memref<?x4x8x32xsi8>
+ %alloc_17 = memref.alloc(%dim_16) {alignment = 32 : i64} : memref<?x32x4x8xsi8>
+ %collapse_shape = memref.collapse_shape %alloc_17 [[0], [1], [2, 3]] : memref<?x32x4x8xsi8> into memref<?x32x32xsi8>
+ %dim_18 = memref.dim %collapse_shape, %c0 : memref<?x32x32xsi8>
+ return %dim_18: index
+}
+
+// -----
+
+// CHECK-LABEL: func.func @resolve_when_collapse_after_collapse(
+// CHECK-SAME: %[[VAL_0:.*]]: memref<?x4x8x32xsi8>) -> index {
+// CHECK: %[[VAL_1:.*]] = arith.constant 0 : index
+// CHECK: %[[VAL_2:.*]] = memref.dim %[[VAL_0]], %[[VAL_1]] : memref<?x4x8x32xsi8>
+// CHECK: return %[[VAL_2]] : index
+// CHECK: }
+
+func.func @resolve_when_collapse_after_collapse(%arg0: memref<?x4x8x32xsi8>)
+ -> index
+{
+ %c0 = arith.constant 0 : index
+ %dim_16 = memref.dim %arg0, %c0 : memref<?x4x8x32xsi8>
+ %alloc_17 = memref.alloc(%dim_16) {alignment = 32 : i64} : memref<?x32x4x8xsi8>
+ %collapse_shape = memref.collapse_shape %alloc_17 [[0], [1], [2, 3]] : memref<?x32x4x8xsi8> into memref<?x32x32xsi8>
+ %collapse_shape_1 = memref.collapse_shape %collapse_shape [[0], [1, 2]] : memref<?x32x32xsi8> into memref<?x1024xsi8>
+ %dim_18 = memref.dim %collapse_shape_1, %c0 : memref<?x1024xsi8>
+ return %dim_18: index
+}
+
+// -----
+
+// CHECK-LABEL: func.func @unfoldable_memref_collapse_shape(
+// CHECK-SAME: %[[VAL_0:.*]]: memref<1x?x8x32xsi8>) -> index {
+// CHECK: %[[VAL_1:.*]] = arith.constant 1 : index
+// CHECK: %[[VAL_2:.*]] = arith.constant 2 : index
+// CHECK: %[[VAL_3:.*]] = memref.dim %[[VAL_0]], %[[VAL_1]] : memref<1x?x8x32xsi8>
+// CHECK: %[[VAL_4:.*]] = memref.alloc(%[[VAL_3]]) {alignment = 32 : i64} : memref<1x32x?x8xsi8>
+// CHECK: %[[VAL_5:.*]] = memref.collapse_shape %[[VAL_4]] {{\[\[}}0], [1], [2, 3]] : memref<1x32x?x8xsi8> into memref<1x32x?xsi8>
+// CHECK: %[[VAL_6:.*]] = memref.dim %[[VAL_5]], %[[VAL_2]] : memref<1x32x?xsi8>
+// CHECK: return %[[VAL_6]] : index
+// CHECK: }
+func.func @unfoldable_memref_collapse_shape(%arg0: memref<1x?x8x32xsi8>)
+ -> index
+{
+ %c1 = arith.constant 1 : index
+ %c2 = arith.constant 2 : index
+ %dim_1 = memref.dim %arg0, %c1 : memref<1x?x8x32xsi8>
+ %alloc_0 = memref.alloc(%dim_1) {alignment = 32 : i64} : memref<1x32x?x8xsi8>
+ %collapse_shape = memref.collapse_shape %alloc_0 [[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 @dim_out_of_bounds_2(
// CHECK-NEXT: arith.constant
// CHECK-NEXT: arith.constant
|
CoTinker
requested review from
ftynse,
MaheshRavishankar,
matthias-springer and
asb
|
ping |
MaheshRavishankar
requested changes
Sep 19, 2024
There was a problem hiding this comment.
I dont see a reason to bail for dynamic dimensions. If the reassociation indices associated with a result is a `SmallVector<int64_t> reassociation;
AffineExpr s0, s1;
bindSymbols(rewriter.getContext(), s0, s1);
AffineExpr mulExpr = s0 * s1;
auto getDimAsOpFoldResult(Value v, int64_t dim) {
return getAsOpFoldResult(rewriter.create<tensor::DimOp>(loc, v, dim));
};
OpFoldResult resultDim = getDimAsOpFoldResult(source, reassociation.front());
for (auto dim : reassociation.drop_front()) {
resultDim = affine::makeComposedFoldedAffineApplyOp(rewriter, mulExpr, {resultDim,
getDimAsOpFoldResult(source, dim)});
}
| getReassociationIndices(); | ||
| Value source = getSrc(); | ||
| auto sourceShape = cast<MemRefType>(source.getType()).getShape(); | ||
| if (!ShapedType::isDynamicShape(sourceShape)) |
There was a problem hiding this comment.
It doesnt have to fail if the source shape is dynamic?
| return failure(); | ||
| for (auto group : enumerate(reassociationArray)) { | ||
| bool isDynInGroup = isDynamicInGroup(group.value(), sourceShape); | ||
| if (isDynInGroup && group.value().size() > 1) |
There was a problem hiding this comment.
I dont think we need this restriction either. You can always construct an expression for the result dimension in terms of source dimensions... So there seems to be little reason to bail
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 lets memref.collapse_shape implement ReifyRankedShapedTypeOpInterface.
To be on the safe side it adds support for following case only: