Intelligent Performance Analysis and Porting Tools for ARM-based HPC Systems

Statement of the Problem or Situation that is Being Addressed: The ARM microprocessor architecture, commonly used in cell phones and tablets, is poised to be a key technology for energy-efficient high performance computing. An easy-to-use and robust tool chain for understanding, porting and refactoring existing HPC codes is essential to unlocking the full potential of ARM in the HPC realm. General Statement of How this Problem is Being Addressed: EP Analytics will commercialize an easy-to-use set of performance analysis and profiling tools to enable high performance computing (HPC) applications on ARM processor-based systems. Unlike the traditional approach of relying on the non-expert user to sift through mountains of performance data, IPAT employs supervised machine learning to automatically mine and extract insights from the detailed performance data, and presents the results to the user via an intuitive graphical interface. IPAT provides a uniquely rich level of detail by leveraging a technique known as binary instrumentation (BI). As no BI capability yet exists for ARM, one aspect of our proposal is to build on a prior DOE-funded effort to complete development of the underlying BI capability. What is to be done in Phase I: During Phase I we propose to develop an end-to-end prototype of IPAT. To demonstrate the feasibility of our approach, this prototype will be fully operational, capable of providing intelligent performance analysis for certain classes of performance behavior (e.g., inter-process communication and floating point vectorization). Our Phase II efforts will focus on broadening the range of performance behaviors handled by the IPAT prototype developed during Phase I and hardening the tools for commercial release. Commercial Applications and Other Benefits: The ARM software development tools being created by EP Analytics have commercial applications across HPC, hyper-scale data centers (Internet), cloud and mobile computing. In the case of the latter, the potential exists to enable HPC apps on future tablets and smartphones, where today ARM is the predominant architecture. This research also has the potential to contribute to and advance the national leadership in Exascale supercomputing technology. Key Words: microprocessors, software development, energy-efficient computing, ARM, high performance computing, exascale, performance tools Summary for Members of Congress: EP Analytics, Inc., of San Diego, CA (52nd Congressional District), has been awarded a grant to create software development tools for the ARM computer processor. ARM, which is highly energy-efficient and extensively used in smartphones and tablet computers such as the iPad, is now being used for large-scale, commercial cloud computing datacenters, and in the supercomputers used to solve important science and national security problems. The software tools developed by EP Analytics will facilitate the transition of the ARM processor from mobile computing devices to large-scale, energy-efficient data centers and supercomputers, and help enable national leadership in Exascale computing.