Closed jlowe closed 2 months ago
Verified that with this change by default RMM NVTX ranges appear in profiles, but I can add -DRMM_USE_NVTX=OFF
to our libcudf build to remove RMM NVTX ranges while still preserving the other NVTX ranges in libcudf.
One final question before we merge: should this be RMM_NVTX
instead of RMM_USE_NVTX
? We have precedent for {LIB}_NVTX
(without "USE") in RAFT here: https://github.com/rapidsai/raft/blob/5bf6642b9ec4c82e97db21f1ce3bb057fc20db82/cpp/CMakeLists.txt#L68
Oh I didn't realize there was already precedence for a project-specific NVTX config. Makes sense to be consistent.
@robertmaynard is this ready to merge?
Can someone with permissions update the PR labels or grant me the permissions to do so?
I think this is ready to go. Would really like to see this get into the 24.08 release.
/merge
Description
Adds a new RMM_USE_NVTX cmake option that defaults to the prior USE_NVTX setting if not set. If RMM_USE_NVTX is OFF, it stubs out the RMM_FUNC_RANGE macro which is used for RMM's NVTX ranges, effectively disabling NVTX within RMM yet leaving it enabled for the rest of the project using RMM.
Closes #1595
Checklist