DocumentCode :
580515
Title :
Performance evaluation of total focusing method on GPP and GPU
Author :
Lambert, Jason ; Pédron, Antoine ; Gens, Guillaume ; Bimbard, Franck ; Lacassagne, Lionel ; Iakovleva, Ekaterina
Author_Institution :
LIST, CEA, Gif-sur-Yvette, France
fYear :
2012
fDate :
23-25 Oct. 2012
Firstpage :
1
Lastpage :
8
Abstract :
Ultrasonic imaging and reconstruction tools are commonly used to detect, identify and measure defects in different mechanical parts. Due to the complexity of the underlying physics and to the evergrowing quantity of acquired data, computation time is becoming a limitation to the optimal inspection of a mechanical part. This article presents the performances of several implementations of a computational heavy algorithm, named Total Focusing Method (TFM), on both Graphics Processing Units (GPU) and General Purpose Processors (GPP). Combination of algorithmic and architectural optimizations have been used. More specifically, on GPU, details are given on shared memory usage and manual thread handling to improve cache localty, whereas on GPP, benefits of SIM-Dization and multithreading are disucussed. With those optimizations, domain specific dimensions enabled the maximum usage of architectural capabilities resulting in a refined algorithm similar to the Berkley dwarf of MapReduce. Both GPU and GPP optimized implementations result in a memory-bound behavior where GPU outperforms the GPP of a factor ×4.
Keywords :
computational complexity; general purpose computers; graphics processing units; multi-threading; optimisation; performance evaluation; shared memory systems; Berkley dwarf; GPP; GPU; MapReduce; SIM-dization; TFM; architectural capabilities; cache localty improvement; computation time; domain specific dimensions; general purpose processors; graphics processing units; manual thread handling; memory-bound behavior; multithreading; performance evaluation; reconstruction tools; shared memory usage; total focusing method; ultrasonic imaging; Acoustics; Delay effects; Graphics processing units; Instruction sets; Receivers; Transceivers; Transmitters; general purpose processors; graphic processing units; non-destructive testing; parallelization; total focusing method; ultrasonic reconstruction;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Design and Architectures for Signal and Image Processing (DASIP), 2012 Conference on
Conference_Location :
Karlsruhe
Print_ISBN :
978-1-4673-2089-4
Electronic_ISBN :
978-2-9539987-4-0
Type :
conf
Filename :
6385392
Link To Document :
بازگشت