How to show PJRT Call Stack

[yuanfz98]

❓ Questions and Help

I wounder how to print PJRT Call Stack. Thanks

[ysiraichi]

Thank you for your question.
What exactly do you mean by that? What would you expect to see from such a log?
Could you provide an example?

[yuanfz98]

Thank you for your question. What exactly do you mean by that? What would you expect to see from such a log? Could you provide an example?

Thanks for your reply. I mean the API call stack like :

PJRT_Client_Create
PJRT_Client_Compile
PJRT_Buffer_CopyToMemory
PJRT_Buffer_CopyToMemory
PJRT_Buffer_ToHostBuffer
PJRT_Client_Destroy

So that I know what api is called from XLA frontend.

[ysiraichi]

I don't think there's currently a way to do that within PyTorch/XLA. Maybe OpenXLA exposes some log configuration? @tengyifei @ManfeiBai @zpcore

[tengyifei]

I also don't know of any way to print the call stack like that. @yuanfz98 could you elaborate what are you trying to achieve? Are you trying to capture memory related metrics?