GPGPU Usage Facilitator

Abstract

The Graphical Processing Unit [GPU] has become a major role player of high performance computing. During the last decade, the capabilities and performance of the GPU remarkably increased. The modern GPU architectures have begun to emphasize versatility and o er new ways to use re-con gured graphic pipelines to solve computationally demanding, complex problems. The great parallel architectures of GPUs have dawned a new era of parallel computing. GPU computing can position as a compelling alternative to traditional general purpose computing. In this thesis we explorer how to use GPU computing to reduce the work load of day-to-day computing. We propose a new system called Application Program Running Interface [APRI] as an initial step of a CPU-GPU hybrid operating system. Application Program Running Interface acts as a coordinator. Basically it migrates the parallel portion of a program to the graphic hardware and after the computation it pumps back the results. Initially we focus to reduce to the workload of the CPU and map general purpose computing with GPU computing. This proposed system can be extended to a CPU-GPU hybrid operating system to improved the performance. We also propose a C to CUDA source code translator to translate already written C program codes in to the CUDA compatible source codes as a part of the Application Program Running Interface. This translator is designed to generate optimized CUDA compatible source codes for the available device at compile time. Basically Application Program Running Interface is focused to create a bridge between CPU and GPU computing but at present it does not thoroughly consider about performance. Due to that reason in the evaluation we consider only about the capability to coordinate CPU and GPU computing. We use a simulated high level instruction set and relevant data to demonstrate and evaluate the Application Program Running Interface.