ROCm / rocprofiler-compute

Advanced Profiling and Analytics for AMD Hardware
https://rocm.docs.amd.com/projects/omniperf/en/latest/
MIT License
135 stars 49 forks source link

Add local dispatch ID filtering #363

Open IanBogle opened 5 months ago

IanBogle commented 5 months ago

Is your feature request related to a problem? Please describe. Currently, filtering by dispatch ID requires global dispatch IDs across all kernel launches of an application, making it difficult to single out the second launch of a specific kernel. To do this currently requires multiple runs of omniperf, or to run rocprof to get the global dispatch ID. Describe the solution you'd like This will likely require discussion to turn into a full-fledged feature. The solution I'm looking for is some sort of flag that you can add to a -k 0 to get a "local" dispatch filtering to get any single instance of the kernel already selected by other filtering. Ideally it should be a different flag than -d. Describe alternatives you've considered Additional context This functionality has been mentioned in passing at some DoE hackathons, and internally in my team.

ashesh2512 commented 4 weeks ago

Curious if there has been any movement on this front?