Department of Computer Science at UH

University of Houston

Department of Computer Science

In Partial Fulfillment of the Requirements for the Degree of
Doctor of Philosophy

Swaroop S. Pophale

Will defend her dissertation

Static Analysis Tool for Synchronization Analysis,
Representation, and Optimizations for Applications Using OpenSHMEM


Programming models provide application developers abstraction from the underlying hardware. OpenSHMEM library is an example of the Partitioned Global Ad- dress Space programming model, which is characterized by local and global views of data. The OpenSHMEM library API provides synchronization primitives that require participation of some or all OpenSHMEM processes executing the application (collective). Since most distributed parallel applications spend 30-40% of their execution time performing synchronization, it is a constant struggle for most application programmers to relax the memory consistency constraints while guaranteeing reproducible and correct results. From our experience we determined that generally programmers tend to over-synchronize when in doubt, and the best approach towards creating correct, scalable and performance driven applications is to help programmers leverage optimizations based on the semantics of the OpenSHMEM library. Unfortunately, most application developers are not well acquainted with all the nuances of the targeted programming libraries and spend most of their development time focused on the correctness aspect alone. This leads to a need for a framework to provide programmers better understanding of the applications and provide useful feed back making it easier for the application developer to incorporate basic and advanced optimizations into their applications with ease. For this we collaborated with the Oak Ridge National Laboratory (ORNL) to build a compiler based tool called the OpenSHMEM analyzer (OSA), which makes the OpenUH compiler aware of the OpenSHMEM library semantics. Along with basic semantic checks, the analyzer provides useful feedback at compile time, leading to faster turn around time and lesser wastage of resources in terms of debugging time or failed execution runs.


Date: Wednesday, April 9, 2014
Time: 1:30 PM
Place: PGH 550

Faculty, students, and the general public are invited.
Advisor: Prof. Barbara Chapman