search

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

'Issue loading top kernels. Check pmc_kernel_top.csv' when using --no-roof on MI200 #313

Closed skyreflectedinmirrors closed 3 weeks ago

skyreflectedinmirrors commented 7 months ago

Describe the bug

omniperf profile -n test --no-roof -- ./whatever
omniperf analyze -p workloads/test/MI200
$ omniperf analyze -p workloads/stream/MI200/

  ___                  _                  __ 
 / _ \ _ __ ___  _ __ (_)_ __   ___ _ __ / _|
| | | | '_ ` _ \| '_ \| | '_ \ / _ \ '__| |_ 
| |_| | | | | | | | | | | |_) |  __/ |  |  _|
 \___/|_| |_| |_|_| |_|_| .__/ \___|_|  |_|  
                        |_|                  

Analysis mode = cli
[analysis] deriving Omniperf metrics...
Issue loading top kernels. Check pmc_kernel_top.csv

Digging a bit, it appears we're emitting this from here:

https://github.com/AMDResearch/omniperf/blob/bfef4200993f0110f45753f8f9c4a7810b83c6ff/src/utils/parser.py#L926

when the roofline csv doesn't exist. At the very least we shouldn't always say it's top_kernels.csv or w/e, and say "no roofline detected". Arguably, this shouldn't be a warning at all as --no-roof is a valid use-case?

ppanchad-amd commented 3 weeks ago

@skyreflectedinmirrors Issue is fixed. Closing ticket. Thanks!