Inputs of aspect oriented programming for the profiling of C++ parallel applications on manycore platforms

High Performance Computing systems expect applications to leverage the most of their processing power. This need is even more present for applications such as Monte Carlo simulations that require noteworthy CPU time and memory footprint. Optimizing applications is one approach to reduce the consumpt...

Full description

Saved in:
Bibliographic Details
Published in2014 International Conference on High Performance Computing & Simulation (HPCS) pp. 793 - 802
Main Authors Schweitzer, Pierre, Mazel, Claude, Hill, David R. C., Carloganu, Cristina
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.07.2014
Subjects
Aspect Oriented Programming
C++
callgrind
Context
Kernel
Libraries
manycores
Multicore processing
parallel programming
profiling
Programming
Radiation detectors
Online AccessGet full text

Cover

More Information
Summary:High Performance Computing systems expect applications to leverage the most of their processing power. This need is even more present for applications such as Monte Carlo simulations that require noteworthy CPU time and memory footprint. Optimizing applications is one approach to reduce the consumption of these resources. Before optimizing, it is mandatory to profile the application in order to pinpoint bottlenecks and hot spots. In this paper, we propose an approach to applications profiling based on Aspect Oriented Programming (AOP). We introduce a profiling approach for C++ codes with the pthread library based on free open source software with low overhead, multicore awareness, multi-threading handling, ease of use and quality outputs compared to established profilers. We will present how our prototype, based on an AOP approach proved to be useful and efficient on a test application.
ISBN:9781479953127
1479953121
DOI:10.1109/HPCSim.2014.6903769