Francois Lanusse

@ASKabalan can you try to rerun your precommit locally to squash the formatting test error?

I did some more testing this this[ little script](https://gist.github.com/EiffL/645a765069e6d465514c6fc3d1e756fc) and depending on the pdims, the result of the FFT is in different order, but what's worse is that apparently, there...

It does actually matter, because when we want to perform filtering operations in Fourier space, we need to understand the order of the array, to know what Fourier frequency corresponds...

Hummm no, I don't think that's the issue. Since the slices all have the same shape in this case, we should get the right result up to a transpose at...

Right, so, I'm suspecting something is wrong with the striding. I added this debug line in the FourierExecutor::Initialize() function: ```c++ std::cout

ok ignore that comment, it's normal behaviour.... enforced at the time of creation of the grid descriptor.

oK, so, after a bunch of experimentation, I think we should start by getting transpose operations to work again. I'm running out of time for now, so just taking notes...

This trick also works with (4,1) (see the jaxpm example). I looked at the output of the pfft3 for different combinations of compute mesh shape. **Note**: the order of processes...