[feature] cuSolver potrf() and potri() interfaces

[toxa81]

Warning! This requires cuda>=11.4

TODO:

• add cuda-11.4 container for the ci/cd runs on Daint

[toxa81]

@simonpintarelli @mtaillefumier Question: do you thinkg cuda>=11.4 dependency is a difficult requirement?

[mtaillefumier]

I do not think so. 11.4 is quite reasonable as a munimum required. It covers Kepler .. A100 .. grace hopper.

[simonpintarelli]

This should be fine. CUDA toolkit 11.x requires driver version >=450.80.2 (https://docs.nvidia.com/deploy/cuda-compatibility/index.html#minor-version-compatibility), on daint the driver is 470.57. I'll add a version constraint on cuda in the spack package after this is merged.

[toxa81]

@mtaillefumier @simonpintarelli
again, I'm doing something wrong. I rebased the branch on my local workstation, pushed the changes and now it looks like a big sum of many commits. In reality there are only three files changed.

[toxa81]

That should be it: a few changes on top of previous commits. All right there.

[simonpintarelli]

That should be it: a few changes on top of previous commits. All right there.

Let me cherry pick these commits.

[toxa81]

I think git push --force should do the correct job. https://stackoverflow.com/questions/43567577/what-is-the-difference-between-force-push-and-normal-push-in-git

[simonpintarelli]

I think git push --force should do the correct job. https://stackoverflow.com/questions/43567577/what-is-the-difference-between-force-push-and-normal-push-in-git

Agree, but I can't see the "force-pushed" tag in the PR history. Perhaps it's just too many commits to spot it.

[toxa81]

I think my mistake is not using git push --force, that is the source of history confusion

[simonpintarelli]

I might have been too quick, what was in the rebase commit? I forgot to pull apparently before I did the rebase.

[toxa81]

I rebased to latest develop 2f32150

[toxa81]

Summary of PR:
added (but not yet used) the potrf() and potri() functions form cuSolver library.

[toxa81]

@simonpintarelli is it good to be merged?

[simonpintarelli]

lgtm