[Documentation Request]

[ookkeeeee]

[Enter feedback here]

Hello. I am a student who is learning the process of profiling a model through onnx runtime. I received the result as a json format file through the method, but I can't interpret it

In particular, there was no information that could be obtained even if Googling was performed on the lower label of "thread_scheduling_stats". I'd appreciate it if you could explain this in detail.

Below is the result of the profiling I received.

[ {"cat" : "Session","pid" :21716,"tid" :7292,"dur" :3623,"ts" :27,"ph" : "X","name" :"model_loading_uri","args" : {}}, {"cat" : "Session","pid" :21716,"tid" :7292,"dur" :4422,"ts" :3690,"ph" : "X","name" :"session_initialization","args" : {}}, {"cat" : "Node","pid" :21716,"tid" :7292,"dur" :2,"ts" :9484,"ph" : "X","name" :"/Conv_fence_before","args" : {"op_name" : "Conv"}}, {"cat" : "Node","pid" :21716,"tid" :7292,"dur" :1299,"ts" :9489,"ph" : "X","name" :"/Conv_kernel_time","args" : {"parameter_size" : "20040","provider" : "CPUExecutionProvider","op_name" : "Conv","input_type_shape" : [{"float":[10,100,100]},{"float":[10,100,5]},{"float":[10]}],"node_index" : "0","output_type_shape" : [{"float":[10,10,98]}],"activation_size" : "400000","output_size" : "39200","thread_scheduling_stats" : {"main_thread": {"thread_pool_name": "session-1-intra-op", "thread_id": "7292", "block_size": [1, 1, 1, 1, 1, 1, 1, 1, 1, 1], "core": 51, "Distribution": 16, "DistributionEnqueue": 0, "Run": 385, "Wait": 753, "WaitRevoke": 6}, "sub_threads": {"27448": {"num_run": 10, "core": 3},"24384": {"num_run": 10, "core": 5},"4984": {"num_run": 10, "core": 7},"25304": {"num_run": 4, "core": 8},"5896": {"num_run": 3, "core": 11},"17356": {"num_run": 10, "core": 12},"18088": {"num_run": 0, "core": -1},"17400": {"num_run": 0, "core": -1},"0": {"num_run": 0, "core": -1},"26228": {"num_run": 0, "core": -1},"12916": {"num_run": 0, "core": -1},"0": {"num_run": 0, "core": -1},"22788": {"num_run": 0, "core": -1},"20256": {"num_run": 0, "core": -1},"4804": {"num_run": 0, "core": -1},"25112": {"num_run": 0, "core": -1},"4944": {"num_run": 0, "core": -1},"22036": {"num_run": 0, "core": -1},"4908": {"num_run": 0, "core": -1},"2136": {"num_run": 0, "core": -1},"9520": {"num_run": 0, "core": -1},"19552": {"num_run": 0, "core": -1},"12744": {"num_run": 0, "core": -1},"4832": {"num_run": 0, "core": -1},"0": {"num_run": 0, "core": -1}}}}}, {"cat" : "Node","pid" :21716,"tid" :7292,"dur" :0,"ts" :10799,"ph" : "X","name" :"/Conv_fence_after","args" : {"op_name" : "Conv"}}, {"cat" : "Session","pid" :21716,"tid" :7292,"dur" :1329,"ts" :9478,"ph" : "X","name" :"SequentialExecutor::Execute","args" : {}}, {"cat" : "Session","pid" :21716,"tid" :7292,"dur" :1379,"ts" :9435,"ph" : "X","name" :"model_run","args" : {}} ]

This is the result of profiling the process of inputting data into one layer of conv1d and forwarding it.

thank you for read.

Document Details

• Title: ONNX Runtime
• Page: https://onnxruntime.ai/docs/
• Page Source: index.md

[mszhanyi]

1. save the json to one file, for example, aa.json.
2. open chrome, input chrome://traceing in the address bar and load the aa.json https://www.chromium.org/developers/how-tos/trace-event-profiling-tool/
3. Use a oneline json parser to read the json snippet. http://json.parser.online.fr/

[ookkeeeee]

Thank you for your reply.

Thanks to your reply, I got some hint on the profiling result.

But what I ultimately want to know is what labels under "thread_scheduling_stats" clearly mean, such as "block_size," "core," "Distribution" and "DistributionEnqueue."

Thank you again for your answer and I hope you have a good day.

[github-actions[bot]]

This issue has been automatically marked as stale due to inactivity and will be closed in 7 days if no further activity occurs. If further support is needed, please provide an update and/or more details.