Skip to content

Conversation

@vinayakdsci
Copy link
Collaborator

Initial draft for the integration of rocprofv3 into the data summarization pipeline.

Currently we generate pftrace files from rocprofv3 and then convert them to a reliably merged JSON files through the traceconv tool provided by perfetto, as other ways of merging JSON output files are not reliable.

[WIP]: Still requires wiring in TraceLens and final report generation.

@oyazdanb
Copy link
Collaborator

oyazdanb commented Dec 8, 2025

we could define the output format as json in the yaml input passed to rocprof. why do we need to convert to perfetto first? https://github.com/ROCm/aorta/blob/main/scripts/gemm_analysis/README.md#simple-yaml-based-tracing-recommended

@vinayakdsci
Copy link
Collaborator Author

vinayakdsci commented Dec 9, 2025

we could define the output format as json in the yaml input passed to rocprof. why do we need to convert to perfetto first? https://github.com/ROCm/aorta/blob/main/scripts/gemm_analysis/README.md#simple-yaml-based-tracing-recommended

As per the documentation at Rocprof Section, I and @amd-vivekag were under the impression that we need to generate a single merged JSON file from multiple rocprof traces. We tried merging json files, but it was not working as expected, so we used pftrace approach to merge these traces. Let us know if you see any issue in this approach.
Also, we wanted to know if we really need to merge the JSON files, because tracelens should be able to convert individual json files as well into html. So, we can avoid merging step if it is not required.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants