Porting an explicit time-domain volume-integral-equation solver on gpus with openacc [open problems in cem]

Abstract

Graphics processing units (GPUs) are gradually becoming mainstream in high-performance computing, as their capabilities for enhancing performance of a large spectrum of scientifi c applications to many fold when compared to multi-core CPUs have been clearly identifi ed and proven. In this paper, implementation and performance-tuning details for porting an explicit marching-on-in-time (MOT)-based time-domain volume-integral-equation (TDVIE) solver onto GPUs are described in detail. To this end, a high-level approach, utilizing the OpenACC directive-based parallel programming model, is used to minimize two often-faced challenges in GPU programming: developer productivity and code portability. The MOT-TDVIE solver code, originally developed for CPUs, is annotated with compiler directives to port it to GPUs in a fashion similar to how OpenMP targets multi-core CPUs. In contrast to CUDA and OpenCL, where signifi cant modifi cations to CPU-based codes are required, this high-level approach therefore requires minimal changes to the codes. In this work, we make use of two available OpenACC compilers, CAPS and PGI. Our experience reveals that different annotations of the code are required for each of the compilers, due to different interpretations

Parallel PWTD-Accelerated Explicit Solution of the Time-Domain Electric Field Volume Integral Equation

“…Moreover, CPU parallelized [23] and GPU accelerated [24] implementations are developed to further advance the capability of the solver. That said, the applicability of this solver to the real-life scattering problems is still limited by its high computational complexity, i.e.,…”

Auxílio a Portabilidade De Código Em Aplicações De Alto Desempenho

“…Abordagens como OpenMP 4.0 [52], OpenACC [53], OCCA [54] e HMPP [55], que utilizam um único código portável com anotações, ainda não atingem o desempenho obtido ao codificar diretamente com a API mais adequada a cada dispositivo [56,57,58,59]. Ou seja, para se obter o maior desempenho em uma determinada arquitetura é preciso utilizar a API que melhor se adapta àquela arquitetura.…”