Cache blocking vnl bra

Description

Optimization: Cache blocking for the routine vnl_mat_bra. On a test calculation this lead to a boost of ~35%. This is confirmed by a significamt reduction of caches misses using cachegrind.

News snippet

Optimization of the application of the pseudopotential in some cases.

Checklist

• I have checked that my code follows the Octopus coding standards
• I have added tests for all the new features added in this request.

changed milestone to %9.0

added Optimization label

The pseudopotentials/13-U235.test is failing consistently (I relaunched the job to be sure) with this change on one of the builders (http://octopus-code.org/buildbot/#/builders/105/builds/247). Since this change does affect the pseudopotentials part, probably the test reference or tolerance needs to be adjusted. Can you check it?

There is a OpenMP bug I by change at least. I will fix it.

added 2 commits

• 0af6f2fb - Phase must be private here.
• f06de960 - Merge branch 'optim_proj_mat_scatter' into cache_blocking_vnl_bra

Compare with previous version

The failing build does not use OpenMP.

I know, but there was still a OpenMP problem.

What is the reason to merge optim_proj_mat_scatter into this branch, when there is a different merge request for it?

Because I stupidly created by mistake this branch from the branch optim_proj_mat_scatter....

Are these the commits that should be in this branch but not in optim_proj_mat_scatter: 77174f8f 0af6f2fb

I am confused now. Hoever, looking at the merge request for optim_proj_mat_scatter, there only the commits that should be there (i.e. modification of vnl_mat_ket).

So 0af6f2fb does not belong here, right?

Yes indeed

Okay, so I'm going to rebase this branch onto develop and delete any commits that should not be there. This will change the history, so make sure you delete your local copy of the branch before doing anything else:

 git branch -D cache_blocking_vnl_bra
 git fetch
 git checkout cache_blocking_vnl_bra

Thak you for taking care of that!

added 1 commit

• 16d3628a - Optimization: Cache blocking for the routine vnl_mat_bra. On a test calculation…

Compare with previous version

added 1 commit

• 1eeae72e - Changing the tolerance for the test pseudopotentials/13-U235.test.

Compare with previous version

merged

mentioned in commit 876b7091