"Autotuning for Petascale Systems and Beyond"

Abstract:

Given a program, its run-time inputs, and a machine, the goal of an "autotuner" is to search for the best possible implementation, using intelligent performance modeling and automated experiments (i.e., actually running the code). This past summer, autotuning researchers met in a U.S. Dept. of Energy-sponsored workshop, which I co-organized, to discuss the prospects for autotuners to produce scalable code for current petascale systems as well as future acquisitions. In the first half of this talk, I provide an introduction to autotuning and summarize some of what I found to be the most interesting results of this meeting. These results cut across a variety of areas in computer and computational science, including applications, architectures, compilers, and programming models. In the latter half of the talk, I'll describe some of the work of my group in this area, which includes autotuning for sparse linear algebra, new programming models, and techniques for the latest multicore and accelerator systems.

You are cordially invited to attend a reception that will follow the seminar to chat informally with faculty and students. Refreshments will be provided.

To receive future announcements, please sign up to the cse-seminar email list:

https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar

Hope to see you there!