Thursday, May 3, 2001
Noon - 1 pm
Wean Hall 8220
University of Wisconsin
Towards Critical-Path Instruction Processing
Although some instructions hurt performance more than others, processors
apply scheduling and speculation as if each instruction was equally costly.
The natural method for understanding the costs of parallel instructions is the
critical path: if we could predict it, we could replace egalitarian policies with
cost-sensitive strategies that will grow increasingly effective as processors
become more sophisticated.
This talk will present our technology for critical-path instruction
processing. I will start by demystifying the critical path of a microexecution,
and follow by describing a predictor that computes the critical path very efficiently,
without actually building a dependence graph. I will then focus on the multiple
uses of the critical-path predictor in a modern processor, and show that it can improve
instruction scheduling by up to 20%, without requiring any major changes to the existing
A recent Pitt graduate, Ras Bodik is returning to Pittsburgh (the Miami of
the Midwest) after what must have been the longest winter in Wisconsin history.
When not shoveling snow, he is applying his compiler-optimization tricks to
microarchitecture, program profiling, and dynamic program optimization.
For Further Seminar Info: