Performance Optimization Of Fully Anisotropic Elastic Wave Propagation On 2nd Generation Intel (R) Xeon Phi (Tm) Processors

Improving the performance of stencil computations is a long-standing optimization challenge due to their inherent heavy memory-access patterns. This problem has been explored in many wave-propagation simulation engines. Moving towards implementations with elastic waves instead of acoustic ones (e.g., used in medical imaging) results in computationally more expensive processes along with increased memory usage. Despite the computational demand, the elevated cost of exploration combined the need for higher success rates is driving the oil & gas industry to adopt elastic anisotropic wave-propagation models as the core of many geophysical imaging mechanisms to extract subsurface features more accurately, increasing return on investment. To reduce time-to-solution, the more complex stencil codes must run efficiently on modern CPU architectures. The Intel Xeon Phi processors emerge as an energy-efficient solution that provides a good trade-off between market price and computing capability. In this paper, we study the effect of several optimization techniques using the YASK stencil-generation framework to implement and evaluate a 25-point stencil of an elastic-wave propagation engine for Intel Xeon Phi processors. The results showed improvements of up to 7x in computations and 8x in memory bandwidth with respect to the non-tuned version, reaching up to 75% of the attainable floating-point performance at the given operational intensity. We collected performance metrics for a set of the most representative optimizations and revealed the relation between each strategy and fundamental characteristics of both code and hardware.