System Information

Tried with code before and after #24483 and also latest.

Detailed description

I am working on the subgraph fusion for BiasGelu. Model is attached here bias_gelu.onnx.zip. Its graph pattern looks like the following,

[Input] -> Add(bias) -> Gelu_pattern -> [Output]

The ONNX simplifier in dnn can already handle the Gelu pattern.

I added the following test case in test_graph_simplifier.cpp to check layer types after fusion,

TEST_F(Test_Graph_Simplifier, BiasGeluSubGraph) {
    test("bias_gelu", "Gelu");
}

and found that:

1. Before the merge of Commutative rules for DNN subgraphs fusion #24483 (Cummutative rules), layer types after simplification are correct, which are { "Gelu", "NaryEltwise" }.
2. After the merge of Commutative rules for DNN subgraphs fusion #24483 (Cummutative rules), layer types are incorrect, which are { "Erf", "NaryEltwise" }.
3. Also tried latest code, and it is the same result as the above 2.

Steps to reproduce

Download model and add the test case.

Issue submission checklist

• I report the issue, it's not a question
• I checked the problem with documentation, FAQ, open issues, forum.opencv.org, Stack Overflow, etc and have not found any solution
• I updated to the latest OpenCV version and the issue is still there
• There is reproducer code and related data files (videos, images, onnx, etc)