search

ROCm / rocprofiler

ROC profiler library. Profiling with perf-counters and derived metrics.
https://rocm.docs.amd.com/projects/rocprofiler/en/latest/
MIT License
126 stars 46 forks source link

Fbar, the kernel's barriers limitation #99

Closed etiennemlb closed 1 week ago

etiennemlb commented 1 year ago

Hi, I would like to get more details about the fbar information reported by rocprof.

What does it mean. How can it be interpreted.

Is it related to : Max fbarriers/Workgrp: 32 Which is given by rocminfo.

Thanks

ppanchad-amd commented 1 month ago

@etiennemlb Apologies for the lack of response. Do you still need assistance with this ticket? If not, please close the ticket. Thanks!

etiennemlb commented 1 month ago

Yes I would like more/extensive documentation on the counter the AMD GPUs provide. Do you have a document I could rely on ?

Thanks!

sohaibnd commented 1 week ago

Hi @etiennemlb, thanks for you patience.

Here is a link to the documentation for rocprofv1. However, note that fbar is a deprecated metric from rocprofv1.

Moreover, rocprofv1 is no longer under development and there is a new rocprofv3 released in ROCm 6.2 as a beta, which is built on top of the new rocprofiler-sdk. I strongly suggest using rocprofv3 since it is very close to having feature parity, has a lower overhead than v1 and v2, and is significantly better tested.

Here is a link to the documentation for rocprofv3 (See the "Using rocprofv3" section). You can also refer to this link for information on relevant performance counters and metrics or try "rocprofv3 --list-metrics" for a list of basic HW counters on your system.

Please let me know if need assistance with anything else!

etiennemlb commented 1 week ago

Im taking a look at rocprofv3, and already I like it for a simple reason.. it seems to support this syntax: rocprofv3 <tracing_option> -- <app_relative_path>. So finally, we can split options given to rocprof from options given to the software under profiling.

sohaibnd commented 1 week ago

Great! And if you encounter any issues with rocprofv3, you can create another ticket under the rocprofiler-sdk repository.