[bugfix] correct Givens rotations after SIMD vectorization

For SIMD vectorization, the branching in the Givens rotations was rewritten using conditionals (which are vectorizable). Unfortunately, the branches for cs and sn were mixed up during this step. This patch corrects the mathematical behavior to the state before SIMD vectorization while retaining the vectorizable conditionals.

added 1 commit

• 2536345e - [bugfix] correct fix-up between real_type and field_type

Compare with previous version

I believe you, if you say that the vectorization patch was wrong, what strikes me is, that we had tests and they yielded the same results as before.

I'll have a look.

I was extremely surprised myself. But when the MINRESSolver failed to solve a small Stokes system without preconditioning, I was taken abash. Even more funny: CG was able solve it.

I'd be quite thankful for a second pair of eyes. This is really easy to mess up.

OK, I finally found the time to look through the changes. You are right and I somehow mixed up some lines... Please squash the two commits to improve readability and feel free to merge then.

added 4 commits

• 2536345e...99907ad2 - 3 commits from branch master
• e42354d9 - [bugfix] correct Givens rotations after SIMD vectorization

Compare with previous version

enabled an automatic merge when the pipeline for e42354d9 succeeds

mentioned in commit c0a02e3e

merged