TOAST: Automatic tiling for iterative stencil computations on GPUs

Abstract

Summary The stencil pattern is important in many scientific and engineering domains, spurring great interest from researchers and industry. In recent years, various optimizations have been proposed for parallel stencil applications running on graphics processing units (GPUs). In particular, tiling is a technique that can significantly enhance application performance by improving data locality and by reducing the volume of communication between host memory and GPU. In addition, tiling enables stencil applications to process inputs that are larger than the physical GPU memory. However, implementing tiling efficiently is complex, time‐consuming, and error‐prone. In this paper, we propose transparently optimized automatic stencil tiling (TOAST), an automatic tiling mechanism for iterative stencil computations running on GPUs; TOAST has 3 main benefits: (1) It incorporates an optimization model that seeks to maximize data reuse within tiles while respecting the amount of dynamically available GPU memory; (2) it offers a virtualized GPU memory for stencil computations, allowing for large input data; and (3) it performs optimal tiling transparently to the developer of the parallel stencil application. The current implementation of TOAST augments the PSkel framework with an internal solver based on genetic algorithms. Our experimental results show that TOAST improves the performance of iterative stencil applications by up to 13 × compared with their multithreaded (central processing unit–based) optimized versions and up to 48 × compared with a naive tiling approach on GPU. The TOAST mechanism is able to automatically achieve a low percentual overhead of data management compared with actual stencil computation.

GPU performance analysis for viscoacoustic wave equations using fast stencil computation from the symbolic specification

“…According to [22] tiling can improve parallel stencil applications in at least 3 ways. First, tiling partitions loop data and computations into tiles, thereby enabling the GPU to handle amounts of input data that exceed the capacity of its internal memory.…”

An automatic mapping technique for OpenACC kernel code based on deeply fused and heterogeneous many-core architecture

“…In addition, researchers such as Alyson Deives Pereira and Rodrigo Caetano Rocha did research on OpenACC's extending support for stencil calculations (Alyson et al 2017a;Alyson et al 2017aAlyson et al , 2015Alyson et al 2017c;Rodrigo et al 2017), and proposed an extension pragma stencil. They designed and developed a source-to-source compiler that can identify stencil pragma and performs corresponding code transformations to generate efficient PSkel code that can be compiled and executed on GPUaccelerated devices.…”

Execução Energeticamente Eficiente de Aplicações Estêncil com o Processador Manycore MPPA-256

“…Quando são feitas computações estêncil sobre o tile, dependências de vizinhança, inerentes ao padrão paralelo do estêncil, precisam ser consideradas durante o particionamento dos dados. Uma das principais soluções para satisfazer essas dependênciasé via blocos sobrepostos, resultando em dados redundantes e computação por tile [Meng and Skadron 2011, Holewinski et al 2012, Rocha et al 2017. Essa técnicaé muito importante em manycores de baixa potência como o MPPA-256, onde o sobrecusto de comunicação pode ser elevado.…”