Closed albestro closed 9 months ago
cscs-ci run
cscs-ci run
All modified and coverable lines are covered by tests :white_check_mark:
Comparison is base (
6ef1381
) 94.06% compared to head (bf3a17f
) 94.06%. Report is 5 commits behind head on master.
:exclamation: Your organization needs to install the Codecov GitHub app to enable full functionality.
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
cscs-ci run
Fixes #878 Close #626
This PR starts with introducing logic for using Intel OneAPI MKL (or oneMKL), which now provides an official CMake Config file. From discussions with @rasolca, for the moment, we opted for keeping also the "legacy" custom module.
DLAF_WITH_MKL
triggers the new logic using official oneMKL CMake config file;DLAF_WITH_MKL_LEGACY
triggers the "legacy" logic which uses our custom CMake module._note: currently,
DLAF_WITH_MKL_LEGACY
seems to work fine also for oneAPI MKL installation._Spack
The spack package has been updated the same way. The logic could be simplified by fully relying on spack (have a look here) and just use the "link-line" logic as any other BLAS/LAPACK library would do. This would require a change, since DLAF_WITH_MKL is used not just for linking purpose but also for implementation details like
mkl_set_num_threads
. Anyway, this is not needed now and we might evaluate in the future (e.g. when MKL legacy branch will be dropped).note:
intel-parallel-studio
is not supported (to my knowledge it has been deprecated too and merged into intel-oneapi)CI
In the CI nothing has changed yet. It still installs
intel-mkl
as external package, and then spack uses it. The package should™️ deal correctly with it by usingDLAF_WITH_MKL_LEGACY
. Switching tointel-oneapi-mkl
is outside the objectives of this PR (~I will open an issue to track this and decide what to do~ see #1074).TODO