Workload Characterization Methodology for OpenCL Kernels

Abstract

Graphics Processing Units (GPUs) have enjoyed a dramatic increase in programma bility as well as in computational power, which allowed them to be utilized as co-processors for general purpose applications. OpenCL, by the Khronos Group, is an open standard for parallel programming using CPUs, GPUs and other types of processors. The OpenCL standard o ers a common API for program execution on systems composed of di erent types of computational devices. While 3D Graphics workload characterization is a well developed area and standard benchmarks are available, comparatively little has been devoted to the analysis and characterization of GPGPU workloads to assist future work in micro-architecture design, application re-structuring and compiler optimizations. Design goal for ArchOCL software model is to characterize the workload for OpenCL. ArchOCL is a software model which runs any OpenCL application and produces interesting statistics that reveals the dynamic behavior of the application. Development of ArchOCL model includes implementation of OpenCL APIs, Compiler enhancement to support OpenCL kernel compilation and Statistics collection.