Análise e comparação dos tipos escalonadores do OpenMP para propagação de onda

Abstract

The wave propagation phenomenon is important for several physical applications, such as fluid dynamics, elasticity, heat propagation, electrostatics, and seismic analysis. An example of use in seismic analysis is in imaging a region of interest, to discover possible features of a subsurface. This technique involves making a computer simulation of how a wave is propagated within an area. Different algorithms perform such a simulation, here we use the Sphere Algorithm. This algorithm is under development at the Laboratory of Parallel Architectures for Signal Processing (LAPPS) at the Federal University of Rio Grande do Norte (UFRN). To improve and optimize this seismic analysis method tests were carried out with different types of schedulers and size of task blocks (chunksize). This algorithm uses the finite difference method and parallelization in C++ using the OpenMP library. The research confirmed that OpenMP sets the static scheduler as the system default and showed that auto, for this code, had runtime values approximately equal to the static type values. The static scheduler with the default chunksize, which is the number of iterations divided by the number of threads, had results of running times between 95.56% to 194.47% faster than the others.