Abstraction of Graphics Hardware Through The Use of Modern Interfaces to Increase Performance of Linear Algebra Routines

摘要

General purpose computation on graphics processing units (GPGPU) is introduced through the application of modern interfaces that abstract graphics hardware. In order to provide an example of these techniques, implementation of an iterative matrix solving algorithm is detailed using two interfaces - Stanford's BrookGPU and Accelerator from Microsoft Research. Performance of the Accelerator implementation is then analyzed.