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

numActiveCUs is per-XCD #248

Closed skyreflectedinmirrors closed 8 months ago

skyreflectedinmirrors commented 8 months ago

Describe the bug

The "numActiveCUs" field (used in 3.1.10, Active CUs) seems to be calculating per-XCD only, e.g.:

image

this impacts the "Wavefront Occupancy" shown in the same table. I need to think about what this thing is actually doing + how to fix it