Skip to content

A rule to remove operands in ConcatOp whose dimension size at axis is zero #3286

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Sep 30, 2025
Merged

A rule to remove operands in ConcatOp whose dimension size at axis is zero #3286

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
de3f770
A rule to remove opearands in ConcatOp whose dimension size at axis i…
tungld Sep 25, 2025
26116ac
Merge branch 'main' into concat-dim-0
tungld Sep 25, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
36 changes: 36 additions & 0 deletions src/Dialect/ONNX/ONNXOps/Canonicalize.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1630,6 +1630,41 @@ struct RecomposeConcatPattern : public OpRewritePattern<ONNXConcatOp> {
}
};

struct RemoveDimZeroInputInConcatPattern
: public OpRewritePattern<ONNXConcatOp> {
using OpRewritePattern<ONNXConcatOp>::OpRewritePattern;

LogicalResult matchAndRewrite(
ONNXConcatOp concatOp, PatternRewriter &rewriter) const final {
ValueRange inputs = concatOp.getOperands();
int64_t axis = concatOp.getAxis();

// Collect indices of inputs whose dim size at axis is zero.
SmallVector<int64_t> indices;
for (unsigned int i = 0; i < inputs.size(); ++i) {
Value inp = inputs[i];
if (!hasShapeAndRank(inp))
continue;
ArrayRef<int64_t> shape = getShape(inp.getType());
// Scalar with rank 0. Dim size is one (not zero).
if (shape.size() == 0)
continue;
if (shape[axis] == 0)
indices.emplace_back(i);
}
if (indices.empty())
return rewriter.notifyMatchFailure(
concatOp, "No operand whose dim at axis is zero");

// Rewrite: remove operands whose dim at axis is zero.
rewriter.modifyOpInPlace(concatOp, [&]() {
for (int64_t idx : indices)
concatOp.getOperation()->eraseOperand(idx);
});
return success();
}
};

// =============================================================================
// Rewrite pattern LayerNormalization
// =============================================================================
Expand Down Expand Up @@ -2193,6 +2228,7 @@ void ONNXCastOp::getCanonicalizationPatterns(
void ONNXConcatOp::getCanonicalizationPatterns(
RewritePatternSet &results, MLIRContext *context) {
results.insert<RecomposeConcatPattern>(context);
results.insert<RemoveDimZeroInputInConcatPattern>(context);
}

/// on the ONNXClipOp.
Expand Down
46 changes: 46 additions & 0 deletions test/mlir/onnx/onnx_canonicalization.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2130,6 +2130,8 @@ func.func @test_remove_where_equal_4(%arg0: tensor<?x?xi64>) -> tensor<2xi64> {

// -----

// COM: Canonicalize ConcatOp.

func.func @test_recompose_concat(%arg0: tensor<1x3x4xf32>, %arg1: tensor<1x3x4xf32> ) -> tensor<1x12x4xf32> {
%0 = "onnx.Concat"(%arg0, %arg1) {axis = 1 : si64, onnx_node_name = "onnx.Concat_0"} : (tensor<1x3x4xf32>, tensor<1x3x4xf32>) -> tensor<1x6x4xf32>
%1 = "onnx.Concat"(%0, %arg0) {axis = 1 : si64, onnx_node_name = "onnx.Concat_1"} : (tensor<1x6x4xf32>, tensor<1x3x4xf32>) -> tensor<1x9x4xf32>
Expand All @@ -2145,6 +2147,45 @@ return %2 : tensor<1x12x4xf32>
}

// -----

func.func @test_concat_remove_dim_0_operand_2_args(%arg0: tensor<?x?x512xf32>, %arg1: tensor<?x0x512xf32>) -> tensor<?x?x512xf32> {
%0 = "onnx.Concat"(%arg0, %arg1) {axis = 1 : si64 }: (tensor<?x?x512xf32>, tensor<?x0x512xf32>) -> tensor<?x?x512xf32>
onnx.Return %0 : tensor<?x?x512xf32>

// CHECK-LABEL: func.func @test_concat_remove_dim_0_operand_2_args
// CHECK-SAME: ([[PARAM_0_:%.+]]: tensor<?x?x512xf32>, [[PARAM_1_:%.+]]: tensor<?x0x512xf32>) -> tensor<?x?x512xf32> {
// CHECK: onnx.Return [[PARAM_0_]] : tensor<?x?x512xf32>
// CHECK: }
}

// -----

func.func @test_concat_remove_dim_0_operand_3_args(%arg0: tensor<?x?x512xf32>, %arg1: tensor<?x0x512xf32>, %arg2: tensor<?x?x512xf32>) -> tensor<?x?x512xf32> {
%0 = "onnx.Concat"(%arg0, %arg1, %arg2) {axis = 1 : si64 }: (tensor<?x?x512xf32>, tensor<?x0x512xf32>, tensor<?x?x512xf32>) -> tensor<?x?x512xf32>
onnx.Return %0 : tensor<?x?x512xf32>

// CHECK-LABEL: func.func @test_concat_remove_dim_0_operand_3_args
// CHECK-SAME: ([[PARAM_0_:%.+]]: tensor<?x?x512xf32>, [[PARAM_1_:%.+]]: tensor<?x0x512xf32>, [[PARAM_2_:%.+]]: tensor<?x?x512xf32>) -> tensor<?x?x512xf32> {
// CHECK: [[VAR_0_:%.+]] = "onnx.Concat"([[PARAM_0_]], [[PARAM_2_]]) {axis = 1 : si64} : (tensor<?x?x512xf32>, tensor<?x?x512xf32>) -> tensor<?x?x512xf32>
// CHECK: onnx.Return [[VAR_0_]] : tensor<?x?x512xf32>
// CHECK: }
}

// -----

func.func @test_concat_donot_remove_operand(%arg0: tensor<?x?x512xf32>, %arg1: tensor<?x1x512xf32>, %arg2: tensor<?x?x512xf32>) -> tensor<?x?x512xf32> {
%0 = "onnx.Concat"(%arg0, %arg1, %arg2) {axis = 1 : si64 }: (tensor<?x?x512xf32>, tensor<?x1x512xf32>, tensor<?x?x512xf32>) -> tensor<?x?x512xf32>
onnx.Return %0 : tensor<?x?x512xf32>

// CHECK-LABEL: func.func @test_concat_donot_remove_operand
// CHECK-SAME: ([[PARAM_0_:%.+]]: tensor<?x?x512xf32>, [[PARAM_1_:%.+]]: tensor<?x1x512xf32>, [[PARAM_2_:%.+]]: tensor<?x?x512xf32>) -> tensor<?x?x512xf32> {
// CHECK: [[VAR_0_:%.+]] = "onnx.Concat"([[PARAM_0_]], [[PARAM_1_]], [[PARAM_2_]]) {axis = 1 : si64} : (tensor<?x?x512xf32>, tensor<?x1x512xf32>, tensor<?x?x512xf32>) -> tensor<?x?x512xf32>
// CHECK: onnx.Return [[VAR_0_]] : tensor<?x?x512xf32>
// CHECK: }
}

// -----

func.func @test_split_relu_movement(%arg0: tensor<1x8x2xf32>) -> (tensor<1x2x2xf32>, tensor<1x3x2xf32>, tensor<1x3x2xf32>) {
%cst = onnx.Constant dense<[2, 3, 3]> : tensor<3xi64>
%0:3 = "onnx.Split"(%arg0, %cst) {axis = 1 : si64} : (tensor<1x8x2xf32>, tensor<3xi64>) -> (tensor<1x2x2xf32>, tensor<1x3x2xf32>, tensor<1x3x2xf32>)
Expand All @@ -2162,6 +2203,7 @@ func.func @test_split_relu_movement(%arg0: tensor<1x8x2xf32>) -> (tensor<1x2x2xf
// CHECK: }

// -----

func.func @test_split_relu_movement_not_all_equal(%arg0: tensor<1x8x2xf32>) -> (tensor<1x2x2xf32>, tensor<1x3x2xf32>, tensor<1x3x2xf32>) {
%cst = onnx.Constant dense<[2, 3, 3]> : tensor<3xi64>
%0:3 = "onnx.Split"(%arg0, %cst) {axis = 1 : si64} : (tensor<1x8x2xf32>, tensor<3xi64>) -> (tensor<1x2x2xf32>, tensor<1x3x2xf32>, tensor<1x3x2xf32>)
Expand All @@ -2181,6 +2223,9 @@ func.func @test_split_relu_movement_not_all_equal(%arg0: tensor<1x8x2xf32>) -> (
// CHECK: }

// -----

// -----

func.func @test_split_leakyrelu_movement(%arg0: tensor<1x8x2xf32>) -> (tensor<1x2x2xf32>, tensor<1x3x2xf32>, tensor<1x3x2xf32>) {
%cst = onnx.Constant dense<[2, 3, 3]> : tensor<3xi64>
%0:3 = "onnx.Split"(%arg0, %cst) {axis = 1 : si64} : (tensor<1x8x2xf32>, tensor<3xi64>) -> (tensor<1x2x2xf32>, tensor<1x3x2xf32>, tensor<1x3x2xf32>)
Expand All @@ -2198,6 +2243,7 @@ func.func @test_split_leakyrelu_movement(%arg0: tensor<1x8x2xf32>) -> (tensor<1x
// CHECK: }

// -----

func.func @test_split_leakyrelu_movement_different_alpha(%arg0: tensor<1x8x2xf32>) -> (tensor<1x2x2xf32>, tensor<1x3x2xf32>, tensor<1x3x2xf32>) {
%cst = onnx.Constant dense<[2, 3, 3]> : tensor<3xi64>
%0:3 = "onnx.Split"(%arg0, %cst) {axis = 1 : si64} : (tensor<1x8x2xf32>, tensor<3xi64>) -> (tensor<1x2x2xf32>, tensor<1x3x2xf32>, tensor<1x3x2xf32>)
Expand Down
Loading