Sparse arrays

[jakirkham]

Given this is a general purpose GPU array library, was wondering if sparse arrays were being considered. Admittedly this is a pretty broad topic. Though am generally interested in thoughts on this topic.

At least for me initially, am interested in leveraging cuSPARSE and clSPARSE. Though having some object models, basic operations, and conversion back and forth between sparse and dense arrays in libgpuarray/pygpu would be really helpful.

[nouiz]

I think it also find here to wrap cusparse/clsparse. It is not in the short term plan. making sparse computation fast on GPU is pretty hard. contribution welcome.

…

On Mon, Jun 26, 2017 at 9:44 AM jakirkham ***@***.***> wrote: Given this is a general purpose GPU array library, was wondering if sparse arrays were being considered. Admittedly this is a pretty broad topic. Though am generally interested in thoughts on this topic. At least for me initially, am interested in leveraging cuSPARSE <http://docs.nvidia.com/cuda/cusparse/index.html> and clSPARSE <https://github.com/clMathLibraries/clSPARSE>. Though having some object models, basic operations, and conversion back and forth between sparse and dense arrays in libgpuarray/pygpu would be really helpful. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#464>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AALC-045bqsU0m5B-EQKbVpNVCyxhgkNks5sH7W2gaJpZM4OFVJy> .

[abergeron]

I would say this is not even on the mid-term plans. However, if someone where to do it in a way that fits the library, I would not say no.