Telcon: 2023 10 25
Peter Scheibel edited this page Oct 25, 2023
·
22 revisions
Wednesday October 25th, 9am PT (UTC -7:00)
- Peter Scheibel (host)
- Wileam Phan
- Jack Morrison
- Mark Krentel
- Nick Romero
- Richarda Butler
- Brian Van Essen
- Yang Liu
This meeting is for Q&A: there are no pre-planned general topics.
- Wileam: requesting testers for amd-gcn (best if done on an AMD GPU)
- nvptx has been broken since
cuda@12:
- There was a patch for
cuda@11
- sm35 must be bumped to sm53
- https://github.com/spack/spack/issues/36172
- The patch no longer applies (affected line doesn't appear to be anywhere)
- https://github.com/spack/spack/blob/develop/var/spack/repos/builtin/packages/gcc/package.py#L672-L687
- There was a patch for
- Brian:
- Two tiered system:
- one spack used as an upstream (this is older, from August 10th)
- another set of local spack instances using the upstream
- index db json was removed from the upstream
- tried to reindex the upstream (using the upstream)
- concretization failures in the downstream spack instances
- Tested: created a spack instance advanced by one week
- can we "bisect" this?
- Peter: try isolating downstream instance:
export SPACK_USER_CACHE_PATH=`pwd`/misc-cache
export SPACK_DISABLE_LOCAL_CONFIG=true
- Brian: was already doing this
- This upstream uses externals
- is the reindex failing to account for those?
- including external Python (installed by the upstream, used by downstreams)
- Two tiered system:
- Nick: issue with existing environment affecting Spack installation
- Had gcc software stack + mpich loaded
- e.g. with
spack load mpich
,spack load gcc
- e.g. with
- No active environment
- Started to compile a clang-based software stack
mpich %clang
- Got unexpected error: some package was unhappy with the curl installation
- Unloaded entire gcc sw stack, and tried to rebuild
- e.g.
spack unload gcc
- e.g.
- Had gcc software stack + mpich loaded
- Jack: checking in on https://github.com/spack/spack/pull/39445
- TODO: Peter: will check on path refactoring efforts (which could subsume this)
- Yang: question about
spack module refresh
- different spack instances generating different module files
- module in one case is missing some path modifications
- Peter: on further inspection, the prefix for that package is missing (although Spack thinks it's installed)