PCG Solver for Dense Matrix?

Accelerated Computing CUDA CUDA Programming and Performance
1

I noticed that cuSparse has a PCG solver for sparse matrices. I am looking for a PCG solver for dense matrices. I haven’t been able to find an implementation of one. I’d just like to ask if there is one around? If not, is there some reason for this? I’d like to know before I look into implementing one on my own?

Thanks!

2

The keyword ‘sparse’ is littered a few times on this paper, you might want to verify that they’re actually talking about dense matrices for their algorithm:
[url]http://www.vis.uni-stuttgart.de/~amentmo/docs/ament-pcgip-PDP-2010.pdf[/url]

Here’s another implementation, talking about a ‘deflation’ step that enables parallelization:
[url]http://repository.tudelft.nl/assets/uuid:26d31751-b0e2-436a-99db-27da61a66e11/10.20.Gupta.pdf[/url]

Here’s a presentation/talk, from the NVIDIA GPU tech conference last year:
[url]http://nvidia.fullviewmedia.com/gtc2012/0517-N-S0063.html[/url]

3

At first glance it seems that they assume that the matrix is usually sparse, but I will read more thoroughly and verify. I think that this will give me a great starting point. Thanks!