Hi Mo,
And it seems that the roctracer is missing since it needs roctx.h
The roctx API is just used to add annotations that help to group kernels when you're profiling the code for performance. I believe you can patch out all calls to roctx functions and the only impact should be that performance profiling data will be less nicely organized.
On salsa roctracer was packaged but not yet uploaded:
https://salsa.debian.org/rocm-team/roctracer/-/blob/master/debian/changelog?ref_type=heads
Do you need help here?
Yes, roctracer is one of the items listed as "help needed" [1].
roctracer is a small library that will need to be packaged to enable GPU profiling. For this, it needs to be ported from the deprecated cppheaderparser to the newer cxxheaderparser, which is available in Debian as python3-cxxheaderparser.
It is a bit unfortunate that an otherwise simple library is stuck on this problem. However, the tracing APIs provided by the HIP runtime are not considered stable, so I've not been pushing on it too hard. The roctracer library is likely to break on HIP runtime updates as it depends on those unstable interfaces.
I'm also not entirely sure that AMD GPU profiling will work on Debian anyway, as the aqlprofile library used by the ROCm Runtime is non-free and has not been packaged.
This is all to say that anything relating to performance
profiling in the ROCm ecosystem is still quite immature. We may
wish to proceed without it.
Sincerely,
Cory Bloor
[1]:
https://salsa.debian.org/rocm-team/community/team-project/-/wikis/Help-needed