Repository navigation
docs: KOKKOS_LIBM_ON_VS_OFF, the device libm option of a CUDA build - #194
Merged
Merged
Conversation
Goal eb10363f, from the owner's session. A decision record for the CMake option WW_KOKKOS_LIBM: ON (device copies of glibc 2.44, WFIP-0006, the library default) against OFF (libdevice, judged by tolerance per ADR-0014), which the CUDA image of PR #192 uses because Ubuntu's libmvec has no .symtab. It compares bits, verification, portability across CPUs, build requirements and speed, lists what is incomplete with ON, and evaluates three ways to make ON work in the image. The position: ON on the lab machine with the Nix toolchain, OFF in the distributed image, timings both ways. A check on this host shows Ubuntu's libc6-dbg file has the symbol and Ubuntu 2.43's libmvec tables equal Nix 2.44's byte for byte. GLOSSARY gains one row (stripped library / .symtab / debug file). Tested: vale --config=.vale.ini on docs/KOKKOS_LIBM_ON_VS_OFF.md and docs/GLOSSARY.md (the steps of scripts/vale.sh): 0 errors, 0 warnings; every relative link of the new page resolves (ls of each target); no em dash; readelf and a byte comparison of the two libmvec.so.1 files run as section 7 of the page says; nothing ran on a GPU Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
h0ffmann
force-pushed
the
goal/eb10363f-5307-kokkos-libm-on-vs-off
branch
from
October 18, 2026 16:37
317e3de to
7e0dece
Compare
6 tasks
h0ffmann
added a commit
that referenced
this pull request
Oct 18, 2026
Plan 003: #193 is merged; the CUDA Docker image of #192 builds with WW_KOKKOS_LIBM=OFF, so it calls CUDA's libm and stays out of the first GPU run; #194's KOKKOS_LIBM_ON_VS_OFF.md holds the owner's decision on that option, and A3 records the value the run's build has. Plan 004: the UFS NUOPC cap on production/GFS.v17 reads VA after W3WAVE, so its host uses are listed and synced, and one coupled case runs with the VA poison, as section 29 did for the ESMF cap. Tested: nix shell nixpkgs#vale -c scripts/vale.sh on both plans, rc 0 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Before. The CMake option
WW_KOKKOS_LIBMof a CUDA build has the default ON (device copies of glibc 2.44, WFIP-0006). The CUDA image of #192 sets it to OFF (libdevice, judged by tolerance per ADR-0014), because Ubuntu'slibmvec.so.1has no.symtabandgen_tablesexits 3. No document compared the two values or said when the image can use ON.After.
docs/KOKKOS_LIBM_ON_VS_OFF.mdcompares ON and OFF (bits on the CPU and the GPU, verification, portability across CPUs, build requirements, speed, what ADR-0014 and WFIP-0006 say), lists what is incomplete with ON, evaluates three ways to make ON work in the image, and gives a Position: ON on the lab machine with the pinned Nix toolchain, OFF in the distributed image, GPU timings recorded both ways. One new check (v, this host): Ubuntu'slibc6-dbg2.43 debug file has__svml_spow_data, and the libmvec tables of Ubuntu 2.43 and Nix 2.44-25 are byte-identical.docs/GLOSSARY.mdgains one row (stripped library /.symtab/ debug file).Goal
eb10363f, from the owner's session. Related: #192 (the CUDA image). Nothing ran on a GPU.vale --config=.vale.inion the two files (the steps ofscripts/vale.sh): 0 errors, 0 warnings; every relative link of the new page resolves; no em dash; thereadelfand byte comparison of section 7 run on this host🤖 Generated with Claude Code