How does Omniperf profile a kernel launched in multiple streams?

ROCm / rocprofiler-compute

Advanced Profiling and Analytics for AMD Hardware

https://rocm.docs.amd.com/projects/omniperf/en/latest/

MIT License

138 stars 49 forks source link

How does Omniperf profile a kernel launched in multiple streams? #444

Open ashesh2512 opened 1 month ago

ashesh2512 commented 1 month ago

Hello, I have a kernel that is launched across four HIP streams. When profiling with Omniperf, how are streams handled?

coleramos425 commented 1 month ago

Copying @nartmada for triage and issue assignment

ppanchad-amd commented 1 month ago

@ashesh2512 @coleramos425 - Internal ticket has been created to assist with this issue. Thanks!

sohaibnd commented 3 weeks ago

Hi @ashesh2512, sorry for the delay. Currently, you can filter collection of counters by kernel name or dispatch id (see documentation) but not by stream. In your case, if you are launching a single kernel in each stream, I believe it should be sufficient to filter by dispatch id.