Author: Bruce Moreland
Date: 09:28:34 10/05/01
Go up one level in this thread
On October 04, 2001 at 20:14:30, Robert Hyatt wrote: >On October 04, 2001 at 18:41:14, Bruce Moreland wrote: >>Two units of work is more than twice as fast as six units, so this particular >>case is super-linear, barely. You can make it more so by making the arrays >>longer. > > >Nope. I modify my serial algorithm to match your parallel algorithm. Then it >is not super-linear at all. I find it after two compares, one on the first >element, one on the last. The parallel algorithm finds it after two compares >also, but they are done in parallel in one cycle. Speedup = 2.0... Bob, I would be an idiot if I said that you couldn't get the same speedup by modifying the serial algorithm. I'm not saying that. I'm saying that you might be able to see improbable speedups when modifying a serial algorithm by making it run in parallel. I'd be a fool to argue that a parallel algorithm will beat a serial algorithm plus a guy with a compiler and some time to modify code. If you are now saying that a super-linear speedup could be possible, but that a guy with a compiler could fix the serial algorithm to make it run better, we aren't arguing about anything. bruce
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.