pfeatherstone

I think this is more a problem with gcc, NVCC and cmake versions. If I can get those from @davisking i can reproduce on my machine. It could just be...

I've been building with CUDA on multiple machines no problem, using CUDA versions 11.8 and 12.3. Looking online, CUDA 11.0 has the issue you were describing. The Matlab stuff definitely...

I can add a few more logs as well.

I think the newer Matlab libraries starting with libmw have all necessary BLAS and LAPACK symbols now. Maybe you had to do it previously because of an old Matlab installation

I can try a really old one if you want but I doubt anyone out there is using something older than R2018

@davisking shall we use https://github.com/marketplace/actions/cuda-toolkit to enable CUDA in one of the runners?

So i've split CMakelists.txt back into all the separate folders. I'm still not a fan, but i'm not dlib's benevolent dictator for life :) Most libraries i've seen have `XXX_BUILD_TESTS`,...

Yes please

The other option would be to use `torch.nested.nested_tensor()` but almost nothing supports that.

Oh yeah I could maybe ignore the zeros. However, if two inputs match exactly, then the output would be zero and we would be ignoring it incorrectly. Unlikely but can...