Installation instructions for GPU-PRISM
Before following the standard instructions for building and installing PRISM from source code,
some additional steps must be performed first:
- Install the latest version of CUDA for your platform.
You must install the latest device driver, the toolkit, and the SDK.
- Unpack the GPU-PRISM archive, and edit the Makefile located in src/sparse. This file includes definitions INCLUDES
and LIBRARIES, and these need to include paths to the relevant folders of the CUDA toolkit and the SDK. Please change
-I/Developer/GPU\ Computing/C/common/inc to the appropriate subfolder of your SDK installation, and
-I/usr/local/cuda/include and -L/usr/local/cuda/lib to the appropriate subfolders of your toolkit installation.
Besides the functionality of standard PRISM, GPU-PRISM includes the command-line option -gpu, which invokes the
appropriate GPU method whenever the given model yields a sparse matrix. Apart from SOR (Gauss-Seidel), all sparse matrix
methods are supported.