what algorithms are used in cusparse

Hi,

I wonder what algorithm is used in CUSPARSE for multiplying sparse matrix in CSR format with vector. Could I get source code of Cusparse or is this algorithm documented somewhere?

Thanks, Tomas.

It is similar to the one provided in CUSP library (which is open-source). You can get it at code.google.com/cusp-library

Hi, Cusp library uses mainly the Hybrid format combining Ellpack and COO format. This format is not as fast as Cusparse library.

Tomas.

Did you read this paper :
http://www.nvidia.com/object/nvidia_research_pub_001.html

Also, here are the GTC 2010 presentations about CUSPARSE and all other libraries (CUBLAS, CUFFT, etc.), which contains similar results:
http://www.nvidia.com/object/gtc2010-presentation-archive.html#session2070
http://www.nvidia.com/object/gtc2010-presentation-archive.html#session2216