saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-07-30 20:18:08 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Fix MatmulTransposeFusion::ApplyImpl() setting of modified flag (#4775)

Browse source 
Update MatmulTransposeFusion::ApplyImpl() to set modified flag whenever a fusion is performed.
This commit is contained in:
edgchen1 2020-08-13 09:51:52 -07:00 committed by GitHub
parent 8fb743f767
commit 74b3b8448c
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
4 changed files with 59 additions and 4 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
onnxruntime/core/optimizer/matmul_transpose_fusion.cc
View file

@ -149,6 +149,8 @@ Status MatmulTransposeFusion::ApplyImpl(Graph& graph, bool& modified, int graph_
matmul_node.SetExecutionProviderType(node.GetExecutionProviderType());
graph_utils::FinalizeNodeFusion(graph, matmul_node, node);
modified = true;
}
// Have to remove node in reversed order for now to walk around the issue in RemoveNode
@ -156,10 +158,6 @@ Status MatmulTransposeFusion::ApplyImpl(Graph& graph, bool& modified, int graph_
graph.RemoveNode(removed_node);
}
if (!removed_nodes.empty()) {
modified = true;
}
return Status::OK();
}
} // namespace onnxruntime

19
onnxruntime/test/optimizer/graph_transform_test.cc
View file

@ -805,6 +805,25 @@ TEST_F(GraphTransformationTests, TransposeMatmulFusionFromTransposeScaleMatMul)
ASSERT_EQ(transpose_scale_matmul_node.GetAttributes().at("alpha").f(), expected_alpha);
}
TEST_F(GraphTransformationTests, TransposeMatmulFusionWithPreservedTranspose) {
auto model_uri = MODEL_FOLDER "fusion/transpose_matmul_2d_fusion_with_preserved_transpose.onnx";
std::shared_ptr<Model> p_model;
ASSERT_STATUS_OK(Model::Load(model_uri, p_model, nullptr, *logger_));
Graph& graph = p_model->MainGraph();
onnxruntime::GraphTransformerManager graph_transformation_mgr{5};
ASSERT_STATUS_OK(graph_transformation_mgr.Register(
onnxruntime::make_unique<MatmulTransposeFusion>(), TransformerLevel::Level1));
ASSERT_STATUS_OK(graph_transformation_mgr.ApplyTransformers(graph, TransformerLevel::Level1, *logger_));
std::map<std::string, int> op_to_count = CountOpsInGraph(graph);
ASSERT_EQ(op_to_count["Transpose"], 1);
ASSERT_EQ(op_to_count["MatMul"], 0);
ASSERT_EQ(op_to_count["TransposeScaleMatMul"], 1);
ASSERT_FALSE(graph.GraphResolveNeeded());
}
TEST_F(GraphTransformationTests, Gemm_LeakyRelu_Fusion) {
auto model_uri = MODEL_FOLDER "gemm_activation_fusion/gemm_activation_fusion.onnx";

BIN
onnxruntime/test/testdata/transform/fusion/transpose_matmul_2d_fusion_with_preserved_transpose.onnx vendored Normal file
View file

Binary file not shown.

38
onnxruntime/test/testdata/transform/fusion/transpose_matmul_gen.py vendored
View file

@ -12,6 +12,7 @@ msdomain.version = 1
msdomain.domain = "com.microsoft"
opsets = [onnxdomain, msdomain]
def save(model_path, nodes, inputs, outputs, initializers):
graph = helper.make_graph(
nodes,
@ -88,3 +89,40 @@ def gen_invalid_default_perm(model_path):
gen_invalid_default_perm(
"transpose_matmul_4d_fusion_invalid_default_perm.onnx")
def gen_with_preserved_transpose(model_path):
nodes = [
helper.make_node(
"Transpose",
["input_0"],
["transposed_input_0"]),
helper.make_node(
"MatMul",
["transposed_input_0", "input_1"],
["output_0"]),
helper.make_node(
"Identity",
["transposed_input_0"],
["output_1"])
]
inputs = [
helper.make_tensor_value_info(
"input_0", TensorProto.FLOAT, ['K', 'M']),
helper.make_tensor_value_info(
"input_1", TensorProto.FLOAT, ['K', 'N'])
]
outputs = [
helper.make_tensor_value_info(
"output_0", TensorProto.FLOAT, ['M', 'N']),
helper.make_tensor_value_info(
"output_1", TensorProto.FLOAT, ['M', 'K'])
]
save(model_path, nodes, inputs, outputs, [])
gen_with_preserved_transpose(
"transpose_matmul_2d_fusion_with_preserved_transpose.onnx")