Parallel on GPU

[EnxEng]

Hi to all,
great code gstools.
I was working more by generating multiple fields, by changing correlation length and variance, in 2d or 3d and I noted that the process could be speed up by using CUDA.

Have also a look here:
https://wp.unil.ch/geocomputing/software/grfs/
https://wp.unil.ch/geocomputing/files/2019/07/R%C3%A4ss_etal_2019_random3D.pdf

A good idea should be to incorporate a new module with and option to process random fields with gpus.

All the best!!!

[MuellerSeb]

Hey there!
Thanks for your interest in GSTools.

That paper looks interesting. This is the exact same way, we calculate our random fields with the Randomization method as documented here:
https://doi.org/10.1016/j.envsoft.2014.01.013

We already use parallel calculations to generate the fields:

GSTools/gstools/field/summator.pyx

Line 35 in 57b94b1

for i in prange(X_len, nogil=True):

You can compile GSTools with OpenMP support. In the future there might also be GPU support directly from Cython:
cython/cython#3449

For parallel sampling we also thought about using schwimmbad with emcee: #35

What are your exact ideas on providing CUDA support?

Cheers,
Sebastian