Computation–communication overlap and parameter auto-tuning for scalable parallel 3-D FFT

•We design a new method of parallel 3-D FFT based on 2-D decomposition of an input 3-D array.•We optimize the performance through computation–communication overlap and parameter auto-tuning.•Experimental results from two supercomputers confirm that our method is faster than three existing libraries....

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Journal of computational science Ročník 14; číslo C; s. 38 - 50
Hlavní autoři: Song, Sukhyun, Hollingsworth, Jeffrey K.
Médium: Journal Article
Jazyk:angličtina
Vydáno: Netherlands Elsevier B.V 01.05.2016
Elsevier
Témata:
3-D FFT
Auto-tuning
Computation–communication overlap
MPI
Non-blocking collective
Non-blocking collective
3-D FFT
Computation–communication overlap
Auto-tuning
MPI
ISSN:1877-7503, 1877-7511
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:•We design a new method of parallel 3-D FFT based on 2-D decomposition of an input 3-D array.•We optimize the performance through computation–communication overlap and parameter auto-tuning.•Experimental results from two supercomputers confirm that our method is faster than three existing libraries. Parallel 3-D FFT is widely used in scientific applications, therefore it is important to achieve high performance on large-scale systems with many thousands of computing cores. This paper describes a new method for scalable high-performance parallel 3-D FFT. We use a 2-D decomposition of 3-D arrays to increase scaling to a large number of cores. In order to achieve high performance, we use non-blocking MPI all-to-all operations and exploit computation-communication overlap. We also auto-tune our 3-D FFT code efficiently in a large parameter space and cope with the complex trade-off in optimizing our code in various system environments. According to experimental results from two systems, our method computes parallel 3-D FFT significantly faster than three existing libraries, and scales well to at least 32,768 compute cores.
Bibliografie:USDOE Office of Science (SC), Advanced Scientific Computing Research (ASCR)
ER25763; ER26054; AC02-05CH11231
ISSN:1877-7503
1877-7511
DOI:10.1016/j.jocs.2015.12.001