[WebNN] Add op support validation for decomposed WebNN ops #23370
Conversation
|
/azp run ONNX Runtime Web CI Pipeline,Windows GPU CI Pipeline,Linux Android Emulator QNN CI Pipeline |
|
/azp run Linux CPU CI Pipeline,Linux CPU Minimal Build E2E CI Pipeline,Linux GPU CI Pipeline,Linux GPU TensorRT CI Pipeline,Linux OpenVINO CI Pipeline,Linux QNN CI Pipeline,MacOS CI Pipeline,Windows ARM64 QNN CI Pipeline,Windows CPU CI Pipeline |
|
Azure Pipelines successfully started running 2 pipeline(s). |
|
/azp run Windows GPU TensorRT CI Pipeline,onnxruntime-binary-size-checks-ci-pipeline,orttraining-linux-ci-pipeline,orttraining-linux-gpu-ci-pipeline,orttraining-ortmodule-distributed,Windows x64 QNN CI Pipeline,Big Models |
|
/azp run Windows GPU CUDA CI Pipeline,Windows GPU DML CI Pipeline,Windows GPU Doc Gen CI Pipeline, Win_TRT_Minimal_CUDA_Test_CI |
|
Azure Pipelines successfully started running 4 pipeline(s). |
1 similar comment
|
Azure Pipelines successfully started running 4 pipeline(s). |
|
Azure Pipelines successfully started running 9 pipeline(s). |
- Some ONNX op are supported by decomposed WebNN ops, defines a decomposed_op_map map to specific decomposed WebNN ops list. - WebNN ops have various first input names such as 'a', 'input', 'inputs', etc. Defines a webnn_op_first_input_name_map map to record special names other than 'input', and a GetWebNNOpFirstInputName function to retrieve the first input name of a WebNN op. - Check if the input and output data types are supported by each decomposed WebNN op. - Remove the unnecessary CheckSingleOp function, WebNN's OpSupportLimits has already covered op supported check.
| @@ -154,7 +150,7 @@ bool AreInputDataTypesSame(const std::string& op_type, | |||
| return true; | |||
| } | |||
|
|
|||
| bool IsSupportedDataType(const int32_t onnx_data_type, const emscripten::val& webnn_supported_data_types) { | |||
| bool IsSupportedDataType(const int32_t& onnx_data_type, const emscripten::val& webnn_supported_data_types) { | |||
There was a problem hiding this comment.
Why is an immutable simple value being passed by reference? 🤔
| } | ||
| // WebNN op name to its first input name mapping, only record the name that is different from "input". | ||
| // This map is used to determine the first input name of a WebNN op and is utilized by OpSupportLimits. | ||
| static const InlinedHashMap<std::string, std::string> webnn_op_first_input_name_map = { |
There was a problem hiding this comment.
(consider) All static const initializers add to DLL load time (or rather WASM module initialization), even if never used - death by a thousand paper cuts. Consider populating the std::map on WebNN EP usage and declaring a simple constexpr array here.
Also can the map be std::string_view -> std::string_view since they are constants anyway? The .at() might not working directly if assigning const auto& op_type = node.OpType();, but it should if doing std::string_view op_type = node.OpType(); ... decomposed_op_map.at(op_type) ...
|
/azp run ONNX Runtime Web CI Pipeline,Windows GPU CI Pipeline,Linux Android Emulator QNN CI Pipeline |
|
/azp run Linux CPU CI Pipeline,Linux CPU Minimal Build E2E CI Pipeline,Linux GPU CI Pipeline,Linux GPU TensorRT CI Pipeline,Linux OpenVINO CI Pipeline,Linux QNN CI Pipeline,MacOS CI Pipeline,Windows ARM64 QNN CI Pipeline,Windows CPU CI Pipeline |
|
/azp run Windows GPU CUDA CI Pipeline,Windows GPU DML CI Pipeline,Windows GPU Doc Gen CI Pipeline,Win_TRT_Minimal_CUDA_Test_CI |
|
/azp run Windows GPU TensorRT CI Pipeline,onnxruntime-binary-size-checks-ci-pipeline,orttraining-linux-ci-pipeline,orttraining-linux-gpu-ci-pipeline,orttraining-ortmodule-distributed,Windows x64 QNN CI Pipeline,Big Models |
|
Azure Pipelines successfully started running 2 pipeline(s). |
|
Azure Pipelines successfully started running 4 pipeline(s). |
1 similar comment
|
Azure Pipelines successfully started running 4 pipeline(s). |
|
Azure Pipelines successfully started running 9 pipeline(s). |
decomposed_op_mapmap to specific decomposed WebNN ops list.webnn_op_first_input_name_mapmap to record special names other than 'input', and aGetWebNNOpFirstInputNamefunction to retrieve the first input name of a WebNN op.CheckSingleOpfunction, because WebNN'sOpSupportLimitshas already covered op supported check.