Add profiling option for wrapping ROCTx

[thomasgibson]

This PR introduces Fortran interfaces for wrapping ROCTx, allowing for instrumenting Fortran codes for trace visualization on AMD GPUs. I have tested this using the tra_adv.f90 example so far, but still experimenting with it.

Want to create this so others can try this out and review

[hiker]

Exciting - I've approved the workflows to run for you.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.95%. Comparing base (f37cec8) to head (987b1d1).

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #3317   +/-   ##
=======================================
  Coverage   99.95%   99.95%           
=======================================
  Files         381      381           
  Lines       53995    53995           
=======================================
  Hits        53973    53973           
  Misses         22       22           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[arporter]

Nice job @thomasgibson - thanks for the README in particular. I've built the wrapper library without problems but haven't tried using it yet (@LonelyCat124 may want to try it too). If you could update the user-guide (https://psyclone.readthedocs.io/en/latest/user_guide/profiling.html#interface-to-third-party-profiling-tools) as part of your PR that would be great (PSyclone/doc/user_guide/profiling.rst). In fact, if you grep for nvidia in the various rst files, there may be other places you'll want to update too.

[arporter]

Have just tried examples/gocean/eg2 and, after tweaking the Makefile a little, succeeded in generating a profile and loading the timeline into perfetto :-)

[thomasgibson]

If you could update the user-guide (https://psyclone.readthedocs.io/en/latest/user_guide/profiling.html#interface-to-third-party-profiling-tools) as part of your PR that would be great (PSyclone/doc/user_guide/profiling.rst). In fact, if you grep for nvidia in the various rst files, there may be other places you'll want to update too.

Thank you for pointing this out! I can certainly go through and make sure the existing documentation is up-to-date.

Have just tried examples/gocean/eg2 and, after tweaking the Makefile a little, succeeded in generating a profile and loading the timeline into perfetto :-)

Excellent! I did try this with the tracer advection benchmark and was able to see the marker injections. Also do you think the Makefile as currently written needs some adjustment? If so, I'm happy to update it. For example, I use amdflang by default but compiling with gfortran (which is more widely distributed) may be preferrable.

[arporter]

Excellent! I did try this with the tracer advection benchmark and was able to see the marker injections. Also do you think the Makefile as currently written needs some adjustment? If so, I'm happy to update it. For example, I use amdflang by default but compiling with gfortran (which is more widely distributed) may be preferrable.

I nearly suggested that but then thought it might be a bit cheeky. Feel free to generalise/update it so that it can work with amd too. (e.g. some of the variables could be re-named and need ?= rather than = to allow them to be overridden.)