Is the开发者_StackOverflow中文版re any effective implement of the solution for sparse matrix linear equation using CUDA?
With the new 3.2 Toolkit (I think it's being released on 2010-09-22, though the RC is currently available) NVIDIA is providing CUSPARSE, a Sparse Matrix solver library. This will be used in a fashion similar to CUFFT and other libraries from NVIDIA, and the full documentation is available on their webpage.
As well as the other answers, you could also check out cusp-library
NVidia published a technical paper on this topic a couple years back.
Update: CUSPARSE library is being released by NVIDIA CUDA Toolkit 3.2 shortly with a large number of sparse matrix / vector operations.
you can try openNL, there is also a commercial package called cula which has some basic stuff for free. there is no magic built for sparse matrix solvers, it greatly depends on the sparsity pattern of the matrix and the way you want to use it, do you have allot of solves ? dose the matrix remain the same ? In my case we ended up writing our own proprietary solver, because of the reasons mentioned above.
精彩评论