A source-to-source open ACC compiler for CUDA

Abstract

OpenACC is a new directive-based programming interface for accelerators such as GPGPU. OpenACC allows the programmer to express the offloading of data and computations to accelerators to simplify the porting process for legacy CPU-based applications. In this paper, we present the design and implementation of an open-source OpenACC compiler that translates C code with OpenACC directives to C code with the CUDA API, which is the most widely used GPU programming environment provided for NVIDIA GPU. We adopt a source-to-source approach using the Omni compiler infrastructure for source code analysis and translations. Our approach leaves detailed machine-specific code optimization to the mature NVIDIA CUDA compiler. An experimental evaluation of the implementation shows that some parallel benchmark codes compiled by our compiler achieve speeds up to 31 times greater than those of CPUs, and that it is competitive with commercial implementations. However, the results also indicate the optimization of OpenACC programs has several problems, such as assigning iterations to GPU threads. © 2014 Springer-Verlag Berlin Heidelberg.