Abstract : This paper presents a review of algorithmic transforms called High Level Transforms for IBM, Intel and ARM SIMD multi-core pro-cessors to accelerate the implementation of low level image pro-cessing algorithms. We show that these optimizations provide a significant acceleration. A first evaluation of 512-bit SIMD Xeon-Phi is also presented. We focus on the point that the combination of optimizations leading to the best execution time cannot be pre-dicted, and thus, systematic benchmarking is mandatory. Once the best configuration is found for each architecture, a comparison of these performances is presented. The Harris points detection opera-tor is selected as being representative of low level image processing and computer vision algorithms. Being composed of five convolu-tions, it is more complex than a simple filter and enables more op-portunities to combine optimizations. The presented work can scale across a wide range of codes using 2D stencils and convolutions.
https://hal.inria.fr/hal-01094906
Contributeur : Lionel Lacassagne
<>
Soumis le : mardi 6 janvier 2015 - 16:41:42
Dernière modification le : jeudi 5 avril 2018 - 12:30:23
Document(s) archivé(s) le : samedi 15 avril 2017 - 08:30:12
Lionel Lacassagne, Daniel Etiemble, Hassan Zahraee, Alain Dominguez, Pascal Vezolle. High Level Transforms for SIMD and Low-Level Computer Vision Algorithms. Symposium on Principles and Practice of Parallel Programming / WPMVP, Feb 2014, Orlando, Florida, United States. pp.8, 〈https://sites.google.com/site/ppopp2014/〉. 〈10.1145/2568058.2568067〉. 〈hal-01094906〉
