To jest strona Piotra.
Jest doktorantem, a wkrótce będzie doktorem. ma Żone i syna, Żona jest fajna a syn wiadomo.
On this page you may find information about a method of implementing subroutines for sparse symmetric matrix vector product on CUDA. The method is based on so called task scheduling approach.
Source files may be downloaded from here. Temporarily a matlab scripts for conversion from CRS to suitable format are enclosed to the package (test matrices may also be converted by me on demand), but will be updated with C++ matrix encoding as soon as possible.
The method used will be described in details in an article that is under preparation. The documentation will also be extented in the future, but still you may ask me privately.
Here is a comparation with NVidia standard library:
You may contact me by e-mail: piotr _dot_ mironowicz _at_ gmail.
Last update of this page: 31.X.2012.