[minseobshin11]

Motivation

The 8 rocprofv3-test-rocpd-*-multiproc tests were disabled due to instability. The instability was caused by test-harness
misconfiguration, not by the feature under test. This PR fixes the root causes and re-enables the tests so RocPD multi-process profiling is actually covered in CI.

Technical Details

Changes in projects/rocprofiler-sdk/tests/rocprofv3/rocpd/CMakeLists.txt:

• Removed the blanket ROCPROFILER_DISABLE_UNSTABLE_CTESTS guard that unconditionally disabled the multiproc tests.
• Added the required ENVIRONMENT (MPI vars OMPI_ALLOW_RUN_AS_ROOT* + PYTHONPATH) to the mpiexec-launched test so OpenMPI can run under root and resolve the rocprofv3 Python module.
• Repointed the otf2 / perfetto / summary generation-multiproc converter tests to the multiproc fixture (was the single-proc fixture), fixing a race that produced missing input DBs under parallel ctest.

JIRA ID

Resolves AIROCVAL-46

Test Plan

Tests are MPI-gated (find_package(MPI)); when MPI is absent they are skipped. They will activate automatically once OpenMPI is provisioned in the CI image (tracked separately in #6671). Verified locally with MPI present: all 8 multiproc tests pass, serial and parallel.

• Built rocprofiler-sdk with rocpd + MPI and ran the 8 multiproc tests locally.
• Serial run, repeated parallel runs (10x), and direct mpiexec reproduction.
• Confirmed no interference with the single-proc rocpd tests.

Test Result

• Multiproc tests: 100% pass, serial and parallel, with no flakiness across repeated runs.
• A rare single-proc PMC GPU memory-fault flake was observed but reproduces with these tests disabled too — it is pre-existing and orthogonal to this change.

Submission Checklist

• [V] Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

[yhuiYH]

Just saw this. I have equivalent PR #4927 that I've been trying to get merged too. Would you be kind enough to give a 1st review?

[yhuiYH]

As mentioned. This is superseded by #4927 and this draft PR can probably be closed.

[minseobshin11]

As mentioned. This is superseded by #4927 and this draft PR can probably be closed.

Hi thanks for the heads up! I agree #4927 covers everything here. I'll close this draft in favor of your PR. I can also first review your PR if needed (I'm an intern), but I have let my team know of your PR. Thank you.