Towards Autotuning of OpenMP Applications on Multicore Architectures

摘要

In this paper we describe an autotuning tool for optimization of OpenMPapplications on highly multicore and multithreaded architectures. Our work wasmotivated by in-depth performance analysis of scientific applications andsynthetic benchmarks on IBM Power 775 architecture. The tool provides anautomatic code instrumentation of OpenMP parallel regions. Based on measurementof chosen hardware performance counters the tool decides on the number ofparallel threads that should be used for execution of chosen code fragments.