Possible to run inference purely sequentially and not have onnx pin a thread

[brettfazio]

If you want to run onnx cpu on a single thread using SEQUENTIAL - is it possible to do so without onnx pinning a new thread?

Presently I looked at the issue here and did the same solution of options = ort.SessionOptions() options.intra_op_num_threads = 1 options.inter_op_num_threads = 1 - but that pins a new thread?
Is it possible to just run inference in the same thread that onnx gets called from and have onnx pin no threads? In the case where one is doing external cpu pinning.

[snnn]

but that pins a new thread?

I don't think so. Did you see a new thread got started? If you set both of values to 1, I think ONNX Runtime should not create any thread pool.

[snnn]

The code is there:
https://github.com/microsoft/onnxruntime/blob/master/onnxruntime/core/util/thread_utils.cc#L18

If the size is 1, then the function directly returns.

[brettfazio]

Got it, thanks