Raising the level of abstraction of GPU-programming

Abstract — General-purpose computing on GPUs (graphics processing units) has received much attention lately due to the benefits of stream processing to exploit limitations of parallel processing. However, programming GPUs has sev-eral challenges with respect to the amount of effort spent in combining the kernel functional code of an application with the parallel concerns offered by APIs from various GPUs. This paper introduces our approach for raising the level of abstaction for programming GPUs. We have implemented an abstract API that can be used with the Compute Unified De-vice Architecture (CUDA) and the Open Compute Language (OpenCL) frameworks, so that the mechanical steps involved in writing the GPU code are abstracted in separate modules. The approach involves static code analysis and generative programming techniques for automatically generating the host code required for CUDA and OpenCL frameworks from minimal specifications provided by the programmers. The generated code resembles the hand-written code with comparable performance