LIBACCEL_WSMP
LIBACCEL_WSMP is a library specifically tailored to accelerate the Watson Sparse Matrix Package (WSMP) using NVIDIA GPUs. That is, by augmenting the standard WSMP distribution (i.e. CPU only code) with libaccel_wsmp.so, significant performance improvements (up to 2x or more) relative to the CPU-only performance can be obtained.
Usage
Simply prefix the launch of the program calling WSMP with
LD_PRELOAD=.../libaccel_wsmp.soto invoke GPU acceleration in WSMP. For example, using sample driver program provided below, assuming all the materials are downloaded to your current directory, and that you have obtained a sample matrix, Serena.mtx, from the Flordia Sparse Matrix Collection (at http://www.cise.ufl.edu/research/sparse/matrices/) simply use
LD_PRELOAD=./libaccel_wsmp.so ./wssmp_ex1 Serena.mtxto factorize the Serena matrix using GPU acceleration.
Please see the documentation for further information.
Downloadable materials
ACCEL_WSMP is currently distributed under the NVIDIA REGISTERED DEVELOPER PROGRAM AGREEMENT
- Documentation
- Sample Driver   A version of the wssmp_ex1.f code, distributed with WSMP, to read .mtx matrix files.
- libaccel_wsmp.so
- ACCEL_WSMP Libarary Download(updated 08/26/15)
Please direct any questions to accel_wsmp@nvidia.com