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

Skip output lines that have UTF8 decoding error #441

Closed benrichard-amd closed 3 weeks ago

benrichard-amd commented 3 weeks ago

Omniperf assumes the output from rocprof is always UTF-8. But there is no guarantee that the user's code will always output UTF-8.

A customer was running Tensorflow and for whatever reason it output a non-UTF-8 character and crashed Omniperf.

Wrap existing code in a try/except block to catch UnicodeDecodeError so the line is skipped and profiling can continue.

benrichard-amd commented 3 weeks ago

LGTM -- just need to format Python code using black: https://github.com/psf/black?tab=readme-ov-file#usage

Done

feizheng10 commented 3 weeks ago

Should we force the omniperf running env with utf-8 first?

coleramos425 commented 3 weeks ago

Should we force the omniperf running env with utf-8 first?

We do set project locale here at Omniperf object instantiation. Is this what you're referring to?