Fix wrong pointer passed to Fortran MPI in FFTE

[luszczek]

This fixes the outstanding problem from #13 and shows how to call other functions from FFTE backend implementation.

[luszczek]

CI seems to be saying something different than my build. I've run the FFTE test with the same set of packages and got no issues. It seems that the build inside CI doesn't get the latest patch and runs the old code that segfaulted before. Here's my output:

mpirun -np 2 test3D_C2C -lib ffte -backend fftw -size 4 4 4 -pgrid 1 2
-----------------------------------------------
Benchmarking ffte library using fftw backend
-----------------------------------------------
[0] proc_grid 1x1x2 - [New MPI process 0] I am located at (0, 0).   0-0-0    3-3-1   local_size={32}
[ffte] successfully initialized.
[1] proc_grid 1x1x2 - [New MPI process 1] I am located at (0, 1).   0-0-2    3-3-3   local_size={32}
Benchmarking FFTE: Get Complex-to-Complex backend using ptest3d.f
Benchmarking FFTE: Get Complex-to-Complex backend using ptest3d.f
Time for FFT plan  = 2.3956e-05
Time for execution =    333.864
Benchmarking FFTE: Get Complex-to-Complex backend using ptest3d.f
Benchmarking FFTE: Get Complex-to-Complex backend using ptest3d.f
FFTE: |X - ifft(fft(X)) |_{max}: 3.051562e+01

[G-Ragghianti]

You are correct. The CI uninstalls the fiber package at the begining of the job, but doesn't touch the previously installed dependencies. I will update the CI so that it will uninstall all the FFT libraries as well.

CI seems to be saying something different than my build. I've run the FFTE test with the same set of packages and got no issues. It seems that the build inside CI doesn't get the latest patch and runs the old code that segfaulted before. Here's my output:

[G-Ragghianti]

Correction: The CI is supposed to be uninstalling the fiber and fft packages, but it doesn't appear to be doing so. I'm investigating the cause.

[G-Ragghianti]

OK, the CI is now able to uninstall fiber and the fft library before attempting to install and test. I will run the CI checks on this PR.

[af-ayala]

For uniformity, I would suggest to keep the same 3.10 version as in the main CMakeList?

[luszczek]

This is not for FIBER. This is for those who build FFTE with CMake. FFTE provides only Makefile's for their tests and nothing else. FFTE doesn't need a new CMake version than 3.8 that has features for libraries and for CUDA language.

[luszczek]

I pushed d5d4896 to make them uniform

[G-Ragghianti]

Can you merge the changes that I recently made to the CI into your branch in order for the test to run with the updated CI config? The changes are under .github/

[luszczek]

It took me forever to merge your CI branch. I hope it was worth it.

[luszczek]

I'd be okay to close this and start from scratch after the tests finish.

[luszczek]

Closing due to wrong rebase