[build] Adding fixes to allow full ASAN pass#2785
Conversation
HereThereBeDragons
left a comment
HereThereBeDragons
left a comment
There was a problem hiding this comment.
did you check if other third-party deps also need -DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE}?
| "LIB_NAMES" | ||
| ) | ||
|
|
||
| # Skip shared-library dlopen validation for sanitizer builds. |
There was a problem hiding this comment.
why do we skip it? because we do those testings on the normal release build or because something would be breaking? take too much time?
There was a problem hiding this comment.
these tests (as far as i understood) were not compatible with ASAN and caused some errors during my attempts to build
i tried an argument to add the correct path to find these via clang but it was not a fan
| # TODO(#2632): Re-enable RCCL for sanitizer builds once RCCL build time is acceptable | ||
| if(THEROCK_ENABLE_RCCL AND THEROCK_SANITIZER STREQUAL "") |
There was a problem hiding this comment.
With my previous comment I rather had in mind to disable building RCCL with ASAN not to disable building RCCL at all. This might be acceptable for now but only if we do not need to build anything that depends on RCCL.
There was a problem hiding this comment.
Ah this if statement indicates that RCCL will build for normal builds, but for any sanitizer build (asan right now), it will not build RCCL!
normal builds:
2026-01-08T19:16:45.1607958Z [566/633] Building sub-project rccl (in background)
2026-01-08T19:16:45.7777997Z [567/633] Stage installing sub-project rccl
2026-01-08T19:16:48.3538093Z [568/633] Configure sub-project rccl-tests (in background)
2026-01-08T19:16:51.1596643Z [569/633] Building sub-project rccl-tests (in background)
2026-01-08T19:16:51.7753864Z [570/633] Stage installing sub-project rccl-tests
2026-01-08T19:16:51.8910293Z [571/633] Populate artifact rccl
2026-01-08T19:16:51.9454460Z [572/633] Creating archive /__w/TheRock/TheRock/build/artifacts/rccl_dbg_gfx94X-dcgpu.tar.xz
2026-01-08T19:16:51.9472601Z [573/633] Creating archive /__w/TheRock/TheRock/build/artifacts/rccl_doc_gfx94X-dcgpu.tar.xz
2026-01-08T19:16:51.9511184Z [574/633] Creating archive /__w/TheRock/TheRock/build/artifacts/rccl_dev_gfx94X-dcgpu.tar.xz
2026-01-08T19:16:51.9640894Z [575/633] Creating archive /__w/TheRock/TheRock/build/artifacts/rccl_run_gfx94X-dcgpu.tar.xz
2026-01-08T19:16:52.3899141Z [576/633] Creating archive /__w/TheRock/TheRock/build/artifacts/rccl_test_gfx94X-dcgpu.tar.xz
2026-01-08T19:16:54.0375044Z [577/633] Creating archive /__w/TheRock/TheRock/build/artifacts/rccl_lib_gfx94X-dcgpu.tar.xz
while the ASAN build does not build RCCL
the other third-party dependencies don't fail at all! so I assume it is not needed |
3 participants
Motivation
For some time, ASAN builds have been failing due to variety of
buildandTest Packagingerrors. With these variety of fixes and conditionals, the CI ASAN pipeline will now pass properlyCloses #1990
Related to #2632
Technical Details
Test Plan
Testing via CI ASAN
Test Result
Build passes here: https://github.com/ROCm/TheRock/actions/runs/20722419413/job/59489435646
Build without RCCL running here: https://github.com/ROCm/TheRock/actions/runs/20827525680/job/59832753950
Submission Checklist