|
Title:
|
ACCELERATE TWO-DIMENSIONAL CONTINUOUS DYNAMIC PROGRAMMING BY MEMORY REDUCTION AND PARALLEL PROCESSING |
|
Author(s):
|
Yukihiro Yoshida, Koushi Yamaguchi, Yuichi Yaguchi, Yuichi Okuyama, Ken-ichi Kuroda and Ryuichi Oka |
|
ISBN:
|
978-972-8939-30-4 |
|
Editors:
|
Hans Weghorn, Pedro Isaías and Radu Vasiu |
|
Year:
|
2010 |
|
Edition:
|
Single |
|
Keywords:
|
Image recognition, Pixel correspondence, Continuous DP, Parallelization |
|
Type:
|
Full Paper |
|
First Page:
|
61 |
|
Last Page:
|
68 |
|
Language:
|
English |
|
Cover:
|
|
|
Full Contents:
|
click to dowload 
|
|
Paper Abstract:
|
This paper contains a proposal for optimizing and accelerating the computation of two-dimensional continuous dynamic programming (2DCDP). 2DCDP processing is optimized by memory reduction and parallelization using OpenMP. We apply buffer resizing and utilize toggle-type buffers to reduce the required memory size. In addition, same-rank processes and pixel correspondence calculation are parallelized by OpenMP instructions to reduce the computation cost/time of 2DCDP. For accumulation, we also apply a realignment of buffering addresses for SIMD on multi-cores/multi-processors. The experimental results show that the computational time and the memory usage have reduced to about 1/4 and 1/5 of the original ones, respectively. Moreover, the concurrency of 2DCDP hot-spot is improved from 5.8 to 7.1 on a quad-core CPU with 8 threads. |
|
|
|
|
|
|
