Title :
BOAST: Bringing Optimization through Automatic Source-to-Source Transformations
Author :
Cronsioe, Johan ; Videau, Brice ; Marangozova-Martin, Vania
Author_Institution :
Grenoble Univ., Grenoble, France
Abstract :
In this paper we present an automatic source-to-source transformation which optimizes loop structures in order to find the best performance configuration for a given multi-core platform. We focus on convolution operators, optimize them using loop unrolling and show that our approach can achieve up to 67% performance gain. Our experiments have been done with the BigDFT scientific application executed on the Tibidado high-performance and low-energy-consumption machine.
Keywords :
multiprocessing systems; parallel processing; power aware computing; BOAST; BigDFT scientific application; Tibidado high-performance machine; automatic source-to-source transformations; convolution operators; loop structure optimization; loop unrolling; low-energy-consumption machine; multicore platform; performance configuration; performance gain; Benchmark testing; Computer architecture; Convolution; Hardware; Optimization; Program processors; Radiation detectors; Auto-tuning; High performance computing; Optimization; Source to Source transformations;
Conference_Titel :
Embedded Multicore Socs (MCSoC), 2013 IEEE 7th International Symposium on
Conference_Location :
Tokyo
DOI :
10.1109/MCSoC.2013.12
