search

xmlyqing00 / Cholmod-Scikit-Sparse-Windows

Set up cholmod and scikit-sparse python package on Windows.
GNU General Public License v3.0
38 stars 11 forks source link

How to use GPU acceleration? #6

Closed Lzshanshan closed 2 years ago

Lzshanshan commented 2 years ago

Hello,

Could you pleaes give me some suggestions about GPU acceleration when using cholesky?

Thank you for your time!

xmlyqing00 commented 2 years ago

Thanks for your interest in my project. Sorry that I have no idea how to do that in GPU mode. I think the Sksparse package only exposes some functions from the code that is written in the CPU, which has been done for years. So maybe you can check other repositories or use PyTorch to implement the GPU Cholesky decomposition.

Lzshanshan commented 2 years ago

Thank you for reply!

Your suggestions are useful! I will try it. I need to process very large sparse matrix, and now it costs almost 25 minutes for each one.

xmlyqing00 commented 2 years ago

I'm familiar with using PyTorch to do some matrix arithmetic. But the basic class in PyTorch is Tensor, which is used for dense data. I'm not sure whether PyTorch has a sparse data structure to solve your problem. Good luck!