GPU as a Parallel Machine: Sorting on the GPU
University of Pennsylvania
Lecture of University of Pennsylvania (2005)
@article{kider2005gpu,
title={GPU as a Parallel Machine: Sorting on the GPU},
author={Kider, J.T.},
journal={Lecture of University of Pennsylvania},
year={2005}
}
Sorting is a fundamental algorithmic building block. One of the most studied problems in computer science is ordering a list of items efficiently. Buck and Purcell showed how the parallel bitonic merge sort algorithm, could exploit many of the parallel features of the SIMD architecture of the GPU. Efficient sorting has practical importance to optimizing algorithms that require sorted lists to work correctly. Moreover, the cost of reading data back from the GPU to the CPU is incredibly inefficient; sorting the data directly on the GPU is preferable for many graphics algorithms.
February 1, 2011 by hgpu
