Autotuning of worklaod distribution granularity for multi-core processors applied to reverse time migration
autotuning, parallel processing, seismic processing, coupled simulated annealing
Wave propagation methods are largely used for modeling seismic data in exploration geophysics. Due to its high computational costs, the use of parallelization methods is increasing, in the attempt of obtaining complex results in an affordable time. In order to obtain further time accelerations in the parallel programs, we may explore alternative ways of performing the load distributions. Along with this qualification manuscript one find appended two conference articles and one manuscript, submitted to a journal, witch present part of the research performed during the master degree. We used dynamic load distribution by computing the wave propagation method in parallel, in chunks of data with sizes allowing minimal execution time. We proposed automatic ways of finding these chunk sizes, with the use of the global optimization algorithms Coupled Simulated Annealing (CSA) and Nelder-Mead (NM). Then, we illustrate, in numerical experiments, that the optimal chunk size varies according to the architecture, compiler, and the number of threads used. We also illustrate, in our tests, that the use of the CSA method is quite promising for the obtainment of the optimal chunk size for these different computational setups, resulting in significant time savings. The last paper applied this methodology to Reverse time migration (RTM), which is an algorithm widely used in the oil and gas industry for seismic migration. RTM also employs wave propagation methods. In the final of the manuscript, we show ours proposes to improve our load balancing using many cores and selecting better parameters for CSA. Finally, we present the schedule of activities of this research.