OrderN / CONQUEST-release

Full public release of large scale and linear scaling DFT code CONQUEST
http://www.order-n.org/
MIT License
96 stars 25 forks source link

Test all multiply kernels in CI #292

Closed tkoskela closed 4 months ago

tkoskela commented 5 months ago

Closes #270

Also contains bug fixes to multiply kernels discovered during testing.

By default the CI workflow only runs the default multiply kernel and excludes the rest, but you can dispatch the workflow manually and set multiply_kernel_test = true to run all the multiply kernels in the matrix. Unfortunately this is difficult to test until the change is in the default branch because of how the GitHub GUI works. You used to be able to manually dispatch from branches using the CLI, but the information seems to be gone from the page so it might have been removed.

tkoskela commented 5 months ago

I hope this makes more sense now that I've also removed the files of test_004

tkoskela commented 4 months ago

Following on from our meeting today, are we ok to merge this? @davidbowler

davidbowler commented 4 months ago

Yes, I think so - but why are we removing the OpenMP directives in the ompDoji kernel? Are they redundant?

tkoskela commented 4 months ago

Yes, OpenMP directives that were removed were redundant because the parallel region has been moved to multiply_module in #248. These should have been removed then, but I missed them (because the kernels weren't being tested :sweat_smile: ).