Performance portability in reverse time migration and seismic modelling via OpenACC

Abstract

Heterogeneity among the computational resources within a single machine has significantly increased in high performance computing to exploit the tremendous potential of graphics processing units (GPUs). Portability in terms of code development and performance has been a challenge due to major differences between GPU programming and memory models from one side and conventional central processing units (CPUs) from another side. Performance characteristics of compilers and processors also vary between machines. Emerging high-level directive-based programming models such as OpenACC has been proposed to target this challenge. In this work, we develop OpenACC implementations for both seismic modelling and reverse time migration algorithms that solve the isotropic, acoustic, and elastic wave equations. We employ OpenACC to take advantage of the computational power of two Nvidia GPU cards: (1) M2090 and (2) K40, residing in IBM and CRAY XC30 clusters respectively. We also explore the main aspects of hybridization seismic modelling and reverse time migration by implementing an Message Passing Interface (MPI)+OpenACC approach. We expose various mapping techniques to develop a portable code that maximizes performance regardless of compiler or platform. Depending on the intensity of the computations, different propagators exhibited different speedup behaviours against a full socket CPU MPI implementation. A performance enhancement of ~10× was obtained, when the acoustic model was ported to a single GPU, compared with a 1.7× speedup obtained using the isotropic model. Our MPI+OpenACC implementation of reverse time migration and seismic modelling shows promising scaling when multiple GPUs were used.

Porting and Evaluation of a Distributed Task-driven Stencil-based Application

“…On the programming models front, in 2012, Ghosh et al [17] analyzed the performance and programmability of three high-level directive-based GPU programming models (PGI, CAPS, and OpenACC) on an NVIDIA GPU on kernels for reverse time migration (RTM). In 2017, Qawasmeh et al [32] implemented an MPI + OpenACC approach for seismic modeling and RTM. In a separate Figure 2: A 9-point 2D stencil in shown, where green colored elements represent grid points within the computing domain, yellow elements are not computed but needed by the stencil (e.g.…”

Energy efficiency and portability of oil and gas simulations on multicore and graphics processing unit architectures

“…Some works focused on the energy efficiency of stencil‐based applications, while others focused on metrics to measure the portability of codes. Qawasmeh et al 36 developed an OpenACC implementation of seismic modeling and reverse time migration (RTM) algorithms for two NVIDIA GPUs. They reached performance improvements of up to 10×.…”

“…Qawasmeh et al 36 developed two OpenACC applications. However, they evaluated their performance only on GPU architectures, neither comparing with CPUs nor evaluating its energy efficiency.…”

Minimod: A Finite Difference solver for Seismic Modeling

“…In 2012, Ghosh et al [7] analyzed the performance and programmability of three high-level directive-based GPU programming models (PGI, CAPS, and OpenACC) on an NVIDIA GPU for kernels of the same type as described in previous sections and for Reverse Time Migration (RTM, [1]), widely used method in geophysics. In 2017, Qawasmeh et al [15] implemented an MPI plus OpenACC approach for seismic modeling and RTM. Domain-specific languages (DSLs) for stencil algorithms have also been proposed.…”