Extending a Run-time Resource Management framework to support OpenCL and Heterogeneous Systems

Abstract

From Mobile to High-Performance Computing (HPC) systems, performance and energy efficiency are becoming always more challenging requirements. In this regard, heterogeneous systems, made by a general-purpose processor and one or more hardware accelerators, are emerging as affordable solutions. However, the effective exploitation of such platforms requires specific programming languages, like for instance OpenCL, and suitable run-time software layers. This work illustrates the extension of a run-time resource management (RTRM) framework, to support the execution of OpenCL applications on systems featuring a multi-core CPU and multiple GPUs. Early results show how this solution leads to benefits both for the applications, in terms of performance, and for the system, in terms of resource utilization, i.e. load balancing and thermal leveling over the computing devices.