FLINT

 

 

Overview

Programmability is a major problem in the field of High-Performance Comuting. Most of the code that run on massive computing clusters is so complex not many people can understand what the program is trying to do at first glance. The FLAME project of the University of Texas at Austin strives to solve the problem of programmability.

Unfortunately the FLAME code is coded at such a high level that performance takes a major hit during the smaller subproblem. FLINT is a tool that is being developed to solve this problem. Using the information that we chose to encode it is possible to either re-create the algorithm in the index free style of coding, generate the algorithm in the more traditional style of coding(i.e. loops with explicit indexing and direct BLAS calls), or potentially to any other form or language of choosing.

 

Issue tracking and enhancements are kept track using Google Project Hosting.

The FLINT project was originally known as FLAMES2S.

 

 

How it works

System.jpg

This is a graphical representation of the final product. The bold line is what is currently being developed but as you can see there is many outputs that this tool has the potential to accomplish.

 

 

Related Publications

 

Technical Reports

Richard M. Veras, Jonathan S. Monette, Enrique S. Quintana-Ortí, and Robert A. Van de Geijn. "FLAMES2S: From Abstraction to High Performance." FLAME Working Note #35. The University of Texas at Austin, Department of Computer Sciences. Technical Report TR-08-49. Dec. 14, 2008.

 

Posters

FLAMES2S poster that was presented at the Undergraduate Research Forum at the University of Texas at Austin. April 2009.