To reduce the compiled binary size of ONNX Runtime at x86_64 linux with "create_reduced_build_config.py", but got a Failed to find kernel for com.microsoft.nchwc.Conv(1)

[zcb794697510]

Describe the issue

Building onnx from souce code under Ubuntu !

First, I had got a ONNX model/s like this: gtcrn.onnx https://github.com/William1617/gtcrn_c/tree/main/model

And then, Try to create a reduced build config file from ONNX format model/s: python tools/python/create_reduced_build_config.py -f ONNX ./model/gtcrn.onnx ./model/ , Then, I get config file like this: required_operators.config

I was very curious and opened this config file: vim required_operators.config

#Generated from ONNX model/s:
#- /mnt/workspace1/samba/rambo/07.DeepLearnFramework/onnxruntime-1.20.1/model/gtcrn.onnx
ai.onnx;11;Add,BatchNormalization,Cast,Concat,ConvInteger,ConvTranspose,Div,DynamicQuantizeLinear,GRU,Gather,MatMulInteger,Mul,PRelu,Pad,Pow,ReduceMean,Reshape,ScatterND,Sigmoid,Slice,Sqrt,Squeeze,Sub,Tanh,Transpose,Unsqueeze

I am very happy and then I started Building onnx from souce code, build commnd : ./build.sh --config Release --build_shared_lib --parallel --compile_no_warning_as_error --skip_submodule_sync --include_ops_by_config ./model/required_operators.config --skip_tests

No surprise, I succeeded: ll build/Linux/Release/libonnxruntime.so*

lrwxrwxrwx 1 rambo rambo 19 12月 5 09:48 build/Linux/Release/libonnxruntime.so -> libonnxruntime.so.1*
lrwxrwxrwx 1 rambo rambo 24 12月 5 09:48 build/Linux/Release/libonnxruntime.so.1 -> libonnxruntime.so.1.20.1*
-rwxrwxr-x 1 rambo rambo 13452552 12月 5 09:48 build/Linux/Release/libonnxruntime.so.1.20.1*

I can't wait to try my gains, Run my executable!

However，I have rejoiced too soon！I was happy for only one minute.

Failed to find kernel for com.microsoft.nchwc.Conv(1) (node:'/encoder/en_convs.1/conv/Conv_output_0_nchwc' ep:'CPUExecutionProvider'). Kernel not foundat ans_gtcrn_create():52
ans_gtcrn: /mnt/workspace1/samba/rambo/09.quecans-lite/quecans-lite/examples/example_ans_gtcrn.c:62: denoise_gtcrn1: Assertion `err == QUEC_ANS_OK' failed.

My complete dynamic library like this:

-rwxrw-r-- 1 rambo rambo 24552784 12月 3 20:54 libonnxruntime.so*
-rwxrw-r-- 1 rambo rambo 24552784 12月 3 20:54 libonnxruntime.so.1*
-rwxrw-r-- 1 rambo rambo 24552784 12月 3 20:54 libonnxruntime.so.1.20.1*

I try the complete dynamic library, it is right!

I want to reduce the compiled binary size,what shoud I do?

To reproduce

step1: Get ONNX model/s : git clone https://github.com/William1617/gtcrn_c.git
step2: Get config file : python tools/python/create_reduced_build_config.py -f ONNX ./model/gtcrn.onnx ./model/
step3: Build ONNXRUNTIME: ./build.sh --config Release --build_shared_lib --parallel --compile_no_warning_as_error --skip_submodule_sync --include_ops_by_config ./model/required_operators.config --skip_tests
step4: Make libonnxruntime.so
step5: Use libonnxruntime.so，run my code
step6: Got Failed

Urgency

No response

Platform

Linux

OS Version

Linux quectel 5.19.0-32-generic #33~22.04.1-Ubuntu SMP PREEMPT_DYNAMIC Mon Jan 30 17:03:34 UTC 2 x86_64 x86_64 x86_64 GNU/Linux

ONNX Runtime Installation

Built from Source

ONNX Runtime Version or Commit ID

1.20.1

ONNX Runtime API

C++

Architecture

X64

Execution Provider

Default CPU

Execution Provider Library Version

NULL

[tianleiwu]

@zcb794697510,
Could you try change the graph optimization level to ORT_ENABLE_EXTENDED?
See https://onnxruntime.ai/docs/performance/model-optimizations/graph-optimizations.html.

[zcb794697510]

@zcb794697510, Could you try change the graph optimization level to ORT_ENABLE_EXTENDED? See https://onnxruntime.ai/docs/performance/model-optimizations/graph-optimizations.html.

nice，it works