Fabian Hoppe

@ClaudiaComito @mtar hopefully, this resolves the previous problems with QR @JuanPedroGHM cb/report is still "Pending" although the benchmarks have been run successfully in gitlab and also report success has been...

@JuanPedroGHM thx :+1:

I have addressed all except for the last comment; the last one I didnt understand because I dont see the inconsistency that is not there (?).

Interestingly, all failing tests involve calls to NumPy (?)

I have tried the tests on the GPU-partition of Terrabyte with CUDA 11.5.0, OpenMPI 4.1.2, and got: ``` test_rand (heat.core.tests.test_random.TestRandom) ... [hpdar01c05s02:31610:0:31610] Caught signal 11 (Segmentation fault: invalid permissions for...

Some updates: * the bug can already be reproduced with `ht.ones(...)` and `.numpy()`: ``` import heat as ht import numpy as np shape = (1825,10,10) split=2 X = ht.ones(shape,dtype=ht.float32,device="gpu", split=split)...

The actual error seems to happen inside `Allgatherv`: * `.numpy()` consists of a `.resplit_(axis=None)` and application of `.cpu().numpy()` to the resulting `larrays`'s of the unsplit tensor * `.resplit_(axis=None)` essentially performs...

my idea for a start: use the existing implementation of parallel FFT in Heat to implement roughly the functionality of `scipy.signal.fftconvolve` (also using the same API)

What I have done: * resolved merge conflict with main in manipulations in favour of the new file * added old concatenate-docstring and changed a lambda-function into a proper def-construction...

Now, it looks like that there are multiple errors for a single process (non-split arrays)