Skip to main content
Seminar | Mathematics and Computer Science Division

Building High-Performance Post-Processing Tools with the Polyhedral Model

Abstract: The polyhedral model represents iteration and data access patterns commonly found in scientific applications. However, the impact of polyhedral-enabled optimizations is limited due to the challenges of incorporating it into legacy applications. These applications run in a production environment and their developers are reluctant to use research software that may disrupt stable operations.

In this talk I propose rebuilding pre- and post-processing tools as a way to solidify optimization software and gain access to smaller application codes that exhibit the same patterns. This talk highlights the I/O performance gains and improvements in reproducibility for post-processing tools demonstrated with the Parflow application.