high performance computing on graphics processing units: hgpu.org

The growing demand of processing power is being satisfied mainly by an increase in the number of computing cores in a system. One of the main challenges to be addressed is efficient utilization of these architectures. This demands data-communication aware mapping of applications on these architectures. Appropriate tools are required to provide the detailed intra-application data-communication information and highlight memory-access patterns to port existing sequential applications efficiently to these architectures or to optimize existing parallel applications. In this work, we present an efficient data-communication profiler which provides a detailed data-communication profile for single and multi-threaded applications. In contrast to prior work, our tool reports such information with manageable overheads for realistic workloads. Experimental results show that on the average, the proposed profiler has at least an order of magnitude less overhead as compared to other state-of-the-art data-communication profilers for a wide range of benchmarks.