[Mobile] How to print the xnnpack EP related information when created the sesson?

[kangli2025]

Describe the issue

I want to use libonnxruntime(1.22.0) and compare the acceleration performance of XNNPACK EP. However, I found that the logs from XNNPACK EP are not detailed enough. They only show the available EPs but do not indicate which EP is currently being used or how many nodes in the ONNX model are being accelerated by XNNPACK EP. The xnnpack delegate in tflite provide such information like

INFO: Created TensorFlow Lite XNNPACK delegate for CPU
VERBOSE: Replacing xxx out of xxx node(s) with delegate (TfLiteXNNPackDelegate) node, yielding xx partitions for subgraps

To reproduce

int main() {
    Ort::Env env(ORT_LOGGING_LEVEL_VERBOSE, "test");
    Ort::SessionOptions session_options;
    // Append EPs as desired
    session_options.AppendExecutionProvider("XNNPACK", { {"intra_op_num_threads", "4"} });
    session_options.SetGraphOptimizationLevel(GraphOptimizationLevel::ORT_ENABLE_BASIC);


  // Print available execution providers
    auto providers = Ort::GetAvailableProviders();
    std::cout << "Available EPs:\n";
    for (const auto& ep : providers) {
        std::cout << ep << std::endl;
    }

// Print available execution providers
    auto providers = Ort::GetAvailableProviders();
    std::cout << "Available EPs:\n";
    for (const auto& ep : providers) {
        std::cout << ep << std::endl;
    }

    // Load your model
    Ort::Session session(env, "your_model.onnx", session_options);
}

Urgency

No response

Platform

Android

OS Version

15

ONNX Runtime Installation

Built from Source

Compiler Version (if 'Built from Source')

bazel 6.5.0

Package Name (if 'Released Package')

None

ONNX Runtime Version or Commit ID

f217402

ONNX Runtime API

C++/C

Architecture

ARM64

Execution Provider

Other / Unknown

Execution Provider Library Version

No response

[skottmckay]

Look for "Node placements" in the output. That should be there if log severity is verbose, unless you have done a minimal build in which case that is excluded to reduce binary size.

[kangli2025]

@skottmckay Thanks for your reply.
I used --minimal_build extended, is this right?
Here is the full command for libonnxruntime building.
./build.sh --config Release --android --android_sdk_path /home/li.kang1/Android/Sdk/ --android_ndk_path /home/xxx/Android/Sdk/ndk/27.2.12479018/ --android_abi arm64-v8a --android_api 29 --minimal_build extended --use_xnnpack --disable_ml_ops --disable_exceptions --build_shared_lib --skip_tests