Telcon: 2022 02 02
Peter Scheibel edited this page Feb 2, 2022
·
12 revisions
- Peter Scheibel
- Brian Van Essen
- Mark Krentel
- Wileam Phan
- Andrew Elble
- Nick Sly (LANL)
- Tom Payerle
- Amiya Maji
- Srinath Vadlamani
- Greg Becker
- Richarda Butler
- Massimilano Culpo
- Tammy Dahlgren
-
(Massimiliano) Sticky variants: some variants should not be aggressively tweaked by Spack: they can be set by the user but should not be messed with to improve the outcome of concretization WRT our optimization criteria.
- Example: variant which decides whether normally-forbidden CUDA/compiler combinations are allowed
- We think only the user should be able to set this
- But Spack will tweak the variant
- See: https://github.com/spack/spack/pull/19736#issuecomment-943165642
- See: https://github.com/spack/spack/pull/28630
- Example: variant which decides whether normally-forbidden CUDA/compiler combinations are allowed
-
(Mark) Could we add messages to the end of a concretization mentioning which preferences were not actually satisfied
- (Andrew) Could highlight parts of the DAG which differ
-
(Peter) Porting Spack-built binaries across systems: https://github.com/spack/spack/issues/28716
- (Todd) Try adding
os_compatible("rocky8", "centos8").
Toconcretize.lp
- We should probably create a file to record more OS compatibilities and then automatically update that in
asp.py
- We should probably create a file to record more OS compatibilities and then automatically update that in
- (Todd) Try adding
- Possible continuation from last week: vendored dependencies
- (Andrew) nvhpc installs CUDA, so which CUDA is being used if I install nvhpc with Spack?
- (Wileam) Should nvhpc be modularized like oneAPI? This would partially solve the embedded CUDA issue, I think
- (Peter) Externals cut off dependencies: https://github.com/spack/spack/issues/9149#issuecomment-1020740273