Optimal loop unrolling for GPGPU programs (thesis)
Dept. of Comput. Sci. & Eng., Ohio State Univ., Columbus, OH, USA
Master of Science Thesis, Graduate School of The Ohio State University
@phdthesis{murthy2009optimal,
title={Optimal Loop Unrolling For GPGPU programs},
author={Murthy, G.S.},
year={2009},
school={The Ohio State University}
}
Graphics Processing Units (GPUs) are massively parallel, many-core processorswith tremendous computational power and very high memory bandwidth. GPUs areprimarily designed for accelerating 3D graphics applications on modern computersystems and are therefore, specialized for highly data parallel, compute intensiveproblems, unlike general-purpose CPUs. In recent times, there has been significantinterest in finding ways to accelerate general purpose (non-graphics), data parallelcomputations using the high processing power of GPUs. General-purpose Program-ming on GPUs (GPGPU) was earlier considered difficult because the only availabletechniques to program the GPUs were graphics-specific programming models suchas OpenGL and DirectX. However, with the advent of GPGPU programming mod-els such as NVIDIA
November 3, 2010 by hgpu
