ARM on HPC: status of commonly used performance analysis tools

TitleARM on HPC: status of commonly used performance analysis tools
Publication TypeConference Paper
Year of Publication2016
AuthorsFialho, L
PublisherARM research summit 2016
Place PublishedCambridge, United Kingdom

The ARM processors have become a promising architecture for high performance computing (HPC) in the last few years. Whilst the main reason to use ARM processors is the need to reduce power consumption, improvements on the processor cache subsystem, number of cores, and vectorization capabilities also make the architecture very attractive in which concerns the expected performance of HPC applications.

As most of the software stack available for ARM is focused towards mobile computing (e.g., smartphones,  tablets), there is not too many libraries and tools focused on HPC available nowadays. However, this scenario is changing. As the ARM architecture enters the HPC scenario several well-stablished HPC-related research groups start to port their tools and libraries to the ARM architecture.

This talk presents the efforts from Bull S.A.S. to integrate into a consistent software stack the recent advances on performance analysis tools for ARM processors, which is part of its contribution to the Mont-Blanc 3 Horizon 2020 research project.

Contributions from the most active HPC research groups will be presented in this talk: Dyninst from The University of Wisconsin, HPC-Toolkit from the Rice University, PAPI from The University of Tennessee at Knoxville, Scalasca from Julich Supercomputing Centre, PerfExpert from the Texas Advanced Computing Center (The University of Texas at Austin), and Extrae from the Barcelona Supercomputing Center (Universidad Politecnica de Catalunya), the perf Linux kernel module from Google Inc.

Citation Key13255