Robert A. van de Geijn

Professor

Department of Computer Sciences
and
Texas Institute for Computational Engineering and Sciences (ICES)
The University of Texas at Austin
Austin, TX 78712
Office: GDC 5.704
Phone: (512) 471-9720
Fax: (512) 471-8885
e-mail: rvdg@cs.utexas.edu
URL: http://www.cs.utexas.edu/users/rvdg

ACM Digital Library Profile
Google Scholar Profile
ResearchGate Profile

B.S. in Mathematics and Computer Science (1981)
University of Wisconsin-Madison

Ph.D. in Applied Mathematics (1987)
University of Maryland College Park

Areas of Interest

Formal derivation of algorithms, mechanical generation of libraries, numerical analysis, parallel supercomputing, scientific computing, high-performance linear algebra libraries.

Summary of Research

The advent of architectures with multi-level memories, including caches, shared memories, and distributed memories, has forced a re-evaluation of how linear algebra libraries are developed. Surprisingly, we have found that the key to the development of high-performance linear algebra libraries for such architectures is the application of formal derivation methods from computer science. As part of the Formal Linear Algebra Methods Environment (FLAME), we have demonstrated that proof of correctness can be established hand-in-hand with the derivation of families of algorithms for a given linear algebra operation. By coding the resulting library using an API so that the code resembles the algorithm, the correctness of the algorithm implies the correctness of the implementation. The methodology has been demonstrated to apply to essentially all operations covered by the BLAS, LAPACK, and beyond. We believe the methodology is sufficiently systematic that is can be automated, including cost and stability analyses.


Curriculum Vitae


More Information about our Graduate Programs


Current Projects


Publications

For recent publications, see the FLAME Publication Webpage Or, see my Google Scholar Profile
(I suggest you click on "Year" to order them most recent first.)

Major Software Efforts


Current and Former Students