Computer Chess Club Archives


Search

Terms

Messages

Subject: A simple example of super-linear speedup on real machines

Author: J. Wesley Cleveland

Date: 09:40:53 10/04/01


Take an iterative function F(a), i.e. F1(a) = a, F2(a) = F(a), F3(a) = F(F(a)),
... Assume the function is complicated enough to use all the available registers
on your machine and requires the same number of machine cycles for each
iteration. Now find the first iteration for which F(b) > F(a). A serial
implementation will spend a significant amount of time saving and restoring
state as it goes between the two functions, while the two processor solution
will only have to spend a minimal amount of time syncronizing at the end of each
iteration (due to my second condition).



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.