Analysis of Optimization Opportunities for Intel Xeon Phi and Intel Xeon Scalable Processors Environments of a Numerical Method for the Biphasic Flow of Fluids in Porous Media

This paper presents optimizations of a numerical method for the biphasic flow of fluids in porous media, for Intel^® Xeon Phi ^TM and Intel^® Xeon Scalable Processors^TM. Intel^® Parallel Studios XE suite tools were used in the study of three method implementations (naive, EWS-Sync and RC). The EWS-Sync implementation consists of replacing the OpenMP barriers by an explicit thread synchronization mechanism. By avoiding barriers, by explicitly synchronizing threads waiting only for their adjacent ones in execution, the spin time was reduced. These implementations obtained a Speedups of 27x, compared to a serial execution, in multi-core and manycore architectures.