Keshav Pingali, professor of computer science at The University of Texas at Austin has received the Programming Languages Achievement Award by the Association for Computing Machinery’s Special Interest Group on Programming Languages (SIGPLAN).
Pingali, who holds the W. A. “Tex” Moncrief, Jr. Chair in Distributed and Grid Computing and is core faculty member of the Oden Institute for Computational Engineering and Sciences, works in programming languages and compiler technology for program understanding, optimization and parallelization. His current research interests are methodologies and tools for programming multicore processors, with a focus on irregular applications from domains like graphics, social networks and data mining.
As stated in the citation from SIGPLAN: Before Keshav Pingali, parallelism in regular dense matrix programs was well understood, but little was known about how to exploit parallelism systematically in algorithms that use data structures like trees, sparse matrices, and graphs. Pingali changed all that by showing us how to exploit scalable parallelism in such “irregular” algorithms.
The intellectual crown jewel of his research program was the “operator formulation of algorithms” that showed that these algorithms can be specified abstractly using atomic state updates and a schedule for performing the updates, and executed in parallel using a combination of compile-time and runtime techniques. He backed up these abstract ideas with the Galois system, which is the first high-level, general-purpose parallel programming system for irregular algorithms. Pingali used this system to perform the first measurements of the amount of parallelism in irregular algorithms.
Over his career, Pingali also invented a host of other techniques used widely in programming systems in academia and industry, including “data-centric compilation” and the “owner-computes rule” for generating code for distributed-memory machines, loop transformation techniques for enhancing cache utilization, and the “program structure tree” for representing programs in compilers. His “fractal symbolic analysis” is the only known technique for restructuring linear algebra codes with pivoting. He has also invented optimal algorithms for computing the strong and weak control dependence relations, and for phi-function placement in converting a program to SSA form.
The annual award recognizes significant and lasting contributions to the field of programming languages and includes a monetary award. It was presented at SIGPLAN’s Programming Language Design and Implementation conference in June in Copenhagen.
Last year, Pingali was announced as the winner of the Ken Kennedy Award for High Performance and Parallel Computing by the Association for Computer Machinery for substantial contributions to programmability and productivity in computing and substantial community service and the IEEE CS Charles Babbage Award. Pingali has mentored almost 40 Ph.D. students and post-docs, and spearheaded important initiatives to strenghten the parallelism community worldwide.