Static Program Analysis for Performance Modeling

Date and time: 
Monday, December 5, 2016 - 09:00
Location: 
220 Deschutes
Author(s):
Kewen Meng
University of Oregon
Host/Committee: 
  • Boyana Norris (Chair)
  • Allen Malony
  • Hank Childs

Abstract

As the development of new hardware and architectures progresses, the computing capability of high performance computing systems continues to increase dramatically. However along with the rise in computing capability, it is also true that a lot of applications didn’t use the full computing potential, and so considerable amount of computing power is being wasted. The inability to take fully utilize available computing resources or specific advantages of architectures during application development partially accounts for this. Hence, it is important to be able to understand and model program behavior in order to gain more information about its bottlenecks and performance potential. Program analysis provides insight on how exactly the instructions are executed in CPU and data are transferred in memory, which can be used for further optimization of a program. In my direct research project, we proposed a pure static approach for analyzing and modeling programs using primarily static analysis techniques.