Sparse Matrix-Vector Multiplication Toolkit for Graphics Processing Units

• An x86 or x86-64 CPU which is connected to a "CUDA-enabled" GPU device
(e.g., GeForce 8800 GTX, 8800 GTS, GTX 280, GTX 260).
• Linux OS only
• Machine should support 32-bit ELF file format.
• CUDA driver and CUDA toolkit should have been installed. Please visit http://www.nvidia.com/object/cuda_get.html for more details on how to download and install the CUDA driver and toolkit for your machine.
• The toolkit has been tested on CUDA 2.0 and 2.1 on 8800 GTX and GTX 280 GPUs.

Installing the SpMV Toolkit

1. Download the package 'SpMV-src-package.tar.gz'.
2. Untar the package and invoke make
   $ tar zxvf SpMV-src-package.tar.gz
   $ make
   
   
3. Let the CUDA install path on the machine be . Include /lib in LD_LIBRARY_PATH environment variable, so that the executable can locate the 'cuda runtime library' while execution.
4. Run the executable $ ./SpMV <input matrix path>[ <input vector path> [<output vector path > ]] , with the following command line arguments:
   1. 'input sparse matrix' path - mandatory
   2. 'input vector' path - optional
   3. 'output vector' path - optional [if given, 'input vector' path should also be given]