John Cavazos

Overview

Optimizing the Optimizer:

I was one of the first researchers to introduce the use of machine learning to optimize an optimizing compiler itself [MS Thesis,NIPS 97,PLDI 04]. Compilers typically contain many heuristics to solve hard problems approximately and efficiently. Finding heuristics that perform well on a broad range of applications and processors is one of the most complex tasks faced by compiler writers. My research involves using machine learning techniques to automatically construct compiler optimization heuristics. For example, I applied machine learning to construct an instruction scheduling heuristic, a heuristic that has been tuned for over 20 years with dozens of publications introducing small variations. Effectively, we removed a state-of-the-art heuristic and used machine learning to automatically generate a solution as good as the human-created solution [NIPS 1997]. We have shown that this technique can completely eliminate the human from heuristic design.

Selected Recent Publications (ALL):

Iterative Optimization in the Polyhedral Model: Part II, Multidimensional Time.
Louis-Noel Pouchet, Cedric Bastoul, Albert Cohen, and John Cavazos.
PLDI 2008 [Coming Soon]

Using Predictive Modeling for Cross-Program Design Space Exploration in Multicore Systems.
Salman Khan, Polychronis Xekalakis, John Cavazos, and Marcelo Cintra.
PACT 2007 [gzip'd PS][PDF]

Rapidly Selecting Good Compiler Optimizations using Performance Counters.
John Cavazos, Grigori Fursin, Felix Agakov, Edwin Bonilla, Michael F. P. O'Boyle, and Olivier Temam.
CGO 2007 [gzip'd PS][PDF]

MiDataSets: Creating The Conditions For A More Realistic Evaluation of Iterative Optimization.
Grigori Fursin, John Cavazos, Michael F.P. O'Boyle, and Olivier Temam.
HiPEAC 2007 [gzip'd PS][PDF]

Automatic Performance Model Construction for the Fast Software Exploration of New Hardware Designs.
John Cavazos, Christophe Dubach, Felix Agakov, Edwin Bonilla, Michael F. P. O'Boyle, Grigori Fursin, and Olivier Temam.
CASES 2006 [gzip'd PS][PDF] Finalist Best Paper Award

Method-Specific Dynamic Compilation using Logistic Regression.
John Cavazos and Michael F. P. O'Boyle.
OOPSLA 2006 [gzip'd PS][PDF]

---