FFTW grid sizes

Description

Do not allow FFTW grids that are not compatible with CUFFT ones. This avoids getting different results between GPUs and CPUs.

News snippet

Do not allow FFTW grids that are not compatible with CUFFT ones.

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 %13.0

added Core label

requested review from @martin.lueders

assigned to @martin.lueders

assigned to @nicolastd

added 1 commit

• 406512bd - Do not allow FFTW grids that are not compatible with CUFFT ones. This avoids...

Compare with previous version

added 40 commits

• 406512bd...5595245a - 39 commits from branch main
• d2413ac8 - Do not allow FFTW grids that are not compatible with CUFFT ones. This avoids...

Compare with previous version

resolved all threads

What was done before did actually not conform to the description at http://www.fftw.org/doc/Complex-DFTs.html:

FFTW is best at handling sizes of the form 2^a 3^b 5^c 7^d 11^e 13^f, where e+f is either 0 or 1

So I think it is a good idea to restrict to multiples of 2, 3, 5, and 7.

The problems we see on the buildbot seem to be due to a memory leek, which was introduced by the changes. If I run the tests locally, I see that the test finite_systems_3d/41-gpu-bug.test is running out of memory. Probably, due to the issue @sohlmann pointed out, the FFT is misbehaving and causing this. I guess, fixing this issue will also fix the memory leek.

added 1 commit

• d8c0bab5 - Do not allow FFTW grids that are not compatible with CUFFT ones. This avoids...

Compare with previous version

resolved all threads

It seems that the testsuite completes now without the issues as seen before.

added 1 commit

• cdf942c5 - Update some references after the change of FFTW grid sizes

Compare with previous version

I would suggest to fix the tolerances as indicated by hand and then let's discuss about the app again in the next developers meeting.