Implementing batched mldivide on the GPU

Question

Rodrigo 2014-3-19

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/122164-implementing-batched-mldivide-on-the-gpu

回答： Edric Ellis 2014-12-10

Hi, I need to solve A*x=B for size(A)~[10,10,1e5] and size(B)~[10,1,1e5], and it seems like something like pagefun would be the way to do it: except that mldivide is not an option. I understand that cuBLAS has had the ability to run batch LU decompositions for some time and the newer cublas also seems to have a batched solver.

Is there a reasonably simple way to implement this in MATLAB (preferably R2013b or earlier)?

Thanks

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Edric Ellis 2014-3-19

1
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/122164-implementing-batched-mldivide-on-the-gpu#answer_129065

Unfortunately, as you have found PAGEFUN does not yet support batched MLDIVIDE. You might be able to use the GPU Mex interface to call CUBLAS yourself.

2 个评论
显示无隐藏无

Rodrigo 2014-3-19

I was hoping I wouldn't have to do this, but I suppose there's no better alternative yet.

Adam 2014-4-29

Just chiming in: I also have a need for a pagefun 'mldivide'. If someone has done the MEX implementation already, please link. It would be greatly appreciated.

请先登录，再进行评论。