System 3: An Automatic Performance Tracking System for Scientific Software Evolution


Shoichi Hirasawa

11:35:00 - 12:00:00

101 , Mathematics Research Center Building (ori. New Math. Bldg.)

We propose an automatic performance tracking system for analyzing execution performance changes and finding the source code modifications that cause the performance changes. The proposed system works together with an Integrated Developing Environment(IDE) in order to interactively support programmers to evolve their scientific computing applications while maintaining their performance portability across multiple target computing platforms. By evaluating the performances on every target platform whenever source code changes are committed, the system helps application developers to find source code modifications that degrade performance portability. Moreover, when a new target platform is chosen, the system retrieves past versions of the application from the version controlled repository and automatically executes them on the platform. As a result, application developers can analyze how the past source code modifications affect the performance on the new platform. Based on the analysis, application developers can review source code changes and improve the design of scientific computing applications.