Computer hardware has been evolving...

so should the way you program.

Spring 2003 EE599/EE699:
Advanced Program Optimization & Parallelization

Time & Place:	MWF 9:00-9:50 AM, 263AH
Instructor:	Professor Hank Dietz
Instructor URL:	http://aggregate.org/hankd/
Course URL:	http://aggregate.org/hankd/POP/

Advanced Program Optimization & Parallelization is about how to make programs execute more efficiently on modern computer hardware ranging from simple microcontrollers to supercomputers. Toward this goal, we will explore a variety of performance-critical architectural features and, for each, discuss how compiler technology and manual recoding can use that feature to improve performance. Even apparently minor abuse of architectural features can have a huge impact on performance: for example, changing how a program interacts with the caches and TLBs of a modern PC can yield more than 100X speedup!

There are two numbers for this course: EE599 is for Undergraduate credit and EE699 is for Graduate credit. The courses will meet together, but differ as follows:

• EE599: Grading will be based on two in-class exams and speedups obtained using specific techniques to improve a small set of benchmark programs.
• EE699: The course will be heavily project oriented. Each student will be expected to take a non-trivial application program (assigned, or of their selection with approval of the instructor) and speed it up using some of the techniques learned. At the end of the course, each student will be required to write a brief paper, and give a short (~10 minute) presentation to the class, summarizing how they optimized their application. Grading will be based on two in-class exams, the methods used and speedup obtained, and their presentation.

Prerequisites

Aside from undergraduate vs. graduate standing, both course have the same prerequisites. Students must have a good understanding of basic computer architecture, minimally EE380. In addition, they are expected to be fluent in a high-level programming language, preferably C or Fortran.