Years I and II: Automated Empirical Optimization of Software

This subproject involves a graduate student devoted 100% for two years. This student is Mr. Thomas Slabach, a graduate student at the department of computer science and engineering at the University of Notre Dame, who has a B.S. in computer engineering from the same university. The purpose of this subproject is to automatically adjust parameters to tune the application so that it runs more efficiently. These optimizations usually involve an extremely long and tedious process, requiring even months of work by highly skilled computer scientists to come up with the correct optimizations for one specific architecture, which often-times will actually cause the code to run slower on other architectures. The tedious nature of this work, and the fact that an optimization on one system may not be a wise optimization on another leads us to desire some sort of automatic optimization technique that is generic for all architectures. Compiler flags can take care of some of these problems, but that only optimizes low-level options, leaving algorithm choices to be made by the user. The recent trend in programming is to have an optimization algorithm at run time find the right parameters. This is called ``Automatic Empirical Optimization of Software'', or AEOS [96]. In particular, parameters for fast electrostatics in combination with MTS, and those of HMC, will be optimized. This will require monitoring modules in PROTOMOL for time, memory, and force computation accuracy, and the ability to ``plug-and-play'' modules at runtime. The componentized, object-oriented design of the program makes it easy to achieve this flexibility.