GPU-FPGA Heterogeneous Computing with OpenCL-Enabled Direct Memory Access

Abstract

Field-programmable gate arrays (FPGAs) have garnered significant interest in research on high-performance computing because their computation and communication capabilities have drastically improved in recent years due to advances in semiconductor integration technologies that rely on Moore's Law. In addition to improving FPGA performance, toolchains for the development of FPGAs in OpenCL have been developed and offered by FPGA vendors that reduce the programming effort required. These improvements reveal the possibility of implementing a concept to enable on-the-fly offloading computation at which CPUs/GPUs perform poorly to FPGAs while performing low-latency data movement. We think that this concept is key to improving the performance of heterogeneous supercomputers using accelerators such as the GPU. In this paper, we propose an OpenCL-enabled data movement method to directly access the global memory of the GPU and show how to implement cooperative GPU-FPGA computation using it. The results of experiments show that our proposed method can achieve a latency of 0.59 μs and a data transfer rate as high as 7.0 GB/s between the GPU and the FPGA, thus confirming that it is effective at realizing high-performance cooperative GPU-FPGA computation.