Implementation of MPI SUMMA matrix mul

[astroC86]

No description provided.

[mrava87]

@astroC86 very well done, this code looks amazingly clear and (I hope) efficient as the SUMMA promises to be 😄

I think at this point we are left with the following:

• add a chaining of MPISummaMatrixMult with another operator like we did for MPIMatrixMult
• add unittests
• refractor MatrixMult.py to contain 2 private classes _MPIBlockMatrixMult and _MPISummaMatrixMult and a public class MPIMatrixMult which takes all the inputs required by both of the private classes and a kind parameter to choose between block and summa - see https://github.com/PyLops/pylops/blob/dev/pylops/signalprocessing/fft2d.py for a similar pattern

[mrava87SW]

@astroC86 good job!

I think the code reads well and the way you created the MPIMatrixMult operator. I noticed (even before I pushed some minor stylistic fixes) that one test was failing... but did not have time to look into it, I guess you can probably more quickly figure out why 😉

Once the tests are passing, @hongyx11 has also looked at the SUMMA implementation and we have discussed the point I raised above about upper vs lower case communication routines*, we should be ready to merge this PR!

• as I suspected, after my last commit, the kind=blocked works with CuPy arrays whilst kind=summa leads to a segmentation fault - likely due to the same cause of this #144

[hongyx11]

@mrava87 I've reviewed the code and think we can merge to the repo

[mrava87]

@astroC86 the code has a lot of flake8 issue (https://github.com/PyLops/pylops-mpi/actions/runs/17504191079/job/49724046622?pr=160). Please fix them, and then we can merge this 😄

Note that I pushed a commit to update the documentation, so make sure to pull