install older version of nimcuda for arraymancer

[metagn]

Attempt to fix CI failure, refs #24495 (comment), alternative is to use a commit version like SciNim/nimcuda@bc65375

[Clonkk]

@Vindaar are you aware of the issue with latest nimcuda ?

[Vindaar]

Kind of? Recently I've seen weird installation issues across a few repositories (weave, arraymancer and maybe something else, don't remember). I assumed it was a nimble issue (the error messages were not very helpful in each case). I can investigate, +but here I also don't understand how the fix is a nimcuda issue.+

edit: misread the last part of the fix and that 0.2.1 is not the latest anymore, heh. I'll investigate

[Clonkk]

Could it be related to how Nimble resolves the commit hash used to compile dependencies?

I have had issue on atlas (if they use the same algo?) when not using MaxVer

[Vindaar]

Could it be related to how Nimble resolves the commit hash used to compile dependencies?

I have had issue on atlas (if they use the same algo?) when not using MaxVer

Had a short look. The issue here is indeed just an issue with a) how nimcuda now handles the wrapper (need to supply paths to the CUDA installation at CT due to SciNim/nimcuda#27) and b) the fact that arraymancer (still) always builds with CUDA support. The latter was never a problem when nimcuda only loaded the cuda libraries at runtime in case they were actually needed.

I'll likely make a -d:cuda switch mandatory for arraymancer to build with CUDA support in the future.

[metagn]

Merging for now

[github-actions]

Thanks for your hard work on this PR!
The lines below are statistics of the Nim compiler built from 33dc236

Hint: mm: orc; opt: speed; options: -d:release
177926 lines; 8.676s; 653.449MiB peakmem