[mlir][Vector] Handle 0-rank case in fold instead of RewriterPattern #130168
Conversation
This file contains 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
Groverkss
requested review from
banach-space,
dcaballe and
nicolasvasilache
as code owners
|
@llvm/pr-subscribers-mlir Author: Kunwar Grover (Groverkss) ChangesFor vector.extract, the folder always canonicalizes to a vector.extract operation, while the rewrite pattern canonicalizes to a vector.broadcast except in the case of 0-rank vectors. Remove this special casing, and instead handle the 0-rank vector case in the folder. Full diff: https://github.com/llvm/llvm-project/pull/130168.diff 2 Files Affected:
diff --git a/mlir/lib/Dialect/Vector/IR/VectorOps.cpp b/mlir/lib/Dialect/Vector/IR/VectorOps.cpp
index 8e0e723cf4ed3..31f1e82ff1174 100644
--- a/mlir/lib/Dialect/Vector/IR/VectorOps.cpp
+++ b/mlir/lib/Dialect/Vector/IR/VectorOps.cpp
@@ -1678,7 +1678,7 @@ static Value foldExtractFromBroadcast(ExtractOp extractOp) {
return source;
unsigned extractResultRank = getRank(extractOp.getType());
- if (extractResultRank >= broadcastSrcRank)
+ if (extractResultRank > broadcastSrcRank)
return Value();
// Check that the dimension of the result haven't been broadcasted.
auto extractVecType = llvm::dyn_cast<VectorType>(extractOp.getType());
@@ -2159,13 +2159,11 @@ class ExtractOpFromBroadcast final : public OpRewritePattern<ExtractOp> {
// folding patterns.
if (extractResultRank < broadcastSrcRank)
return failure();
+ // For scalar result, the input can only be a zero-dim vector, which will
+ // be handled by the folder.
+ if (extractResultRank == 0)
+ return failure();
- // Special case if broadcast src is a 0D vector.
- if (extractResultRank == 0) {
- assert(broadcastSrcRank == 0 && llvm::isa<VectorType>(source.getType()));
- rewriter.replaceOpWithNewOp<vector::ExtractElementOp>(extractOp, source);
- return success();
- }
rewriter.replaceOpWithNewOp<vector::BroadcastOp>(
extractOp, extractOp.getType(), source);
return success();
diff --git a/mlir/test/Dialect/Vector/canonicalize.mlir b/mlir/test/Dialect/Vector/canonicalize.mlir
index bf755b466c7eb..8a9204f042ff6 100644
--- a/mlir/test/Dialect/Vector/canonicalize.mlir
+++ b/mlir/test/Dialect/Vector/canonicalize.mlir
@@ -736,7 +736,7 @@ func.func @fold_extract_broadcast_same_input_output_vec(%a : vector<4xf32>,
// CHECK-LABEL: fold_extract_broadcast_0dvec_input_scalar_output
// CHECK-SAME: %[[A:.*]]: vector<f32>
-// CHECK: %[[B:.+]] = vector.extractelement %[[A]][] : vector<f32>
+// CHECK: %[[B:.+]] = vector.extract %[[A]][] : f32 from vector<f32>
// CHECK: return %[[B]] : f32
func.func @fold_extract_broadcast_0dvec_input_scalar_output(%a : vector<f32>,
%idx0 : index, %idx1 : index, %idx2: index) -> f32 {
@@ -2834,7 +2834,7 @@ func.func @extract_from_0d_splat_broadcast_regression(%a: f32, %b: vector<f32>,
%3 = vector.extract %2[] : f32 from vector<f32>
// Broadcast 0D to 3D and extract scalar.
- // CHECK: %[[extract1:.*]] = vector.extractelement %[[b]][] : vector<f32>
+ // CHECK: %[[extract1:.*]] = vector.extract %[[b]][] : f32 from vector<f32>
%4 = vector.broadcast %b : vector<f32> to vector<1x2x4xf32>
%5 = vector.extract %4[0, 0, 1] : f32 from vector<1x2x4xf32>
|
|
@llvm/pr-subscribers-mlir-vector Author: Kunwar Grover (Groverkss) ChangesFor vector.extract, the folder always canonicalizes to a vector.extract operation, while the rewrite pattern canonicalizes to a vector.broadcast except in the case of 0-rank vectors. Remove this special casing, and instead handle the 0-rank vector case in the folder. Full diff: https://github.com/llvm/llvm-project/pull/130168.diff 2 Files Affected:
diff --git a/mlir/lib/Dialect/Vector/IR/VectorOps.cpp b/mlir/lib/Dialect/Vector/IR/VectorOps.cpp
index 8e0e723cf4ed3..31f1e82ff1174 100644
--- a/mlir/lib/Dialect/Vector/IR/VectorOps.cpp
+++ b/mlir/lib/Dialect/Vector/IR/VectorOps.cpp
@@ -1678,7 +1678,7 @@ static Value foldExtractFromBroadcast(ExtractOp extractOp) {
return source;
unsigned extractResultRank = getRank(extractOp.getType());
- if (extractResultRank >= broadcastSrcRank)
+ if (extractResultRank > broadcastSrcRank)
return Value();
// Check that the dimension of the result haven't been broadcasted.
auto extractVecType = llvm::dyn_cast<VectorType>(extractOp.getType());
@@ -2159,13 +2159,11 @@ class ExtractOpFromBroadcast final : public OpRewritePattern<ExtractOp> {
// folding patterns.
if (extractResultRank < broadcastSrcRank)
return failure();
+ // For scalar result, the input can only be a zero-dim vector, which will
+ // be handled by the folder.
+ if (extractResultRank == 0)
+ return failure();
- // Special case if broadcast src is a 0D vector.
- if (extractResultRank == 0) {
- assert(broadcastSrcRank == 0 && llvm::isa<VectorType>(source.getType()));
- rewriter.replaceOpWithNewOp<vector::ExtractElementOp>(extractOp, source);
- return success();
- }
rewriter.replaceOpWithNewOp<vector::BroadcastOp>(
extractOp, extractOp.getType(), source);
return success();
diff --git a/mlir/test/Dialect/Vector/canonicalize.mlir b/mlir/test/Dialect/Vector/canonicalize.mlir
index bf755b466c7eb..8a9204f042ff6 100644
--- a/mlir/test/Dialect/Vector/canonicalize.mlir
+++ b/mlir/test/Dialect/Vector/canonicalize.mlir
@@ -736,7 +736,7 @@ func.func @fold_extract_broadcast_same_input_output_vec(%a : vector<4xf32>,
// CHECK-LABEL: fold_extract_broadcast_0dvec_input_scalar_output
// CHECK-SAME: %[[A:.*]]: vector<f32>
-// CHECK: %[[B:.+]] = vector.extractelement %[[A]][] : vector<f32>
+// CHECK: %[[B:.+]] = vector.extract %[[A]][] : f32 from vector<f32>
// CHECK: return %[[B]] : f32
func.func @fold_extract_broadcast_0dvec_input_scalar_output(%a : vector<f32>,
%idx0 : index, %idx1 : index, %idx2: index) -> f32 {
@@ -2834,7 +2834,7 @@ func.func @extract_from_0d_splat_broadcast_regression(%a: f32, %b: vector<f32>,
%3 = vector.extract %2[] : f32 from vector<f32>
// Broadcast 0D to 3D and extract scalar.
- // CHECK: %[[extract1:.*]] = vector.extractelement %[[b]][] : vector<f32>
+ // CHECK: %[[extract1:.*]] = vector.extract %[[b]][] : f32 from vector<f32>
%4 = vector.broadcast %b : vector<f32> to vector<1x2x4xf32>
%5 = vector.extract %4[0, 0, 1] : f32 from vector<1x2x4xf32>
|
banach-space
approved these changes
Mar 10, 2025
dcaballe
approved these changes
Mar 12, 2025
|
No regressions in IREE: iree-org/iree#20268 |
Groverkss
force-pushed
the
handle-0rank-case-in-fold
branch
from
18f90d0 to
412fb21
Compare
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.
For vector.extract, the folder always canonicalizes to a vector.extract operation, while the rewrite pattern canonicalizes to a vector.broadcast except in the case of 0-rank vectors.
Remove this special casing, and instead handle the 0-rank vector case in the folder.