Author: Walter Faxon
Date: 22:47:27 12/16/02
Go up one level in this thread
On December 16, 2002 at 15:05:56, Gerd Isenberg wrote: >Hi all, > >My conclusion with Athlons bit scan (and reset) routines so far: > >Inlined bsf is the fastest for IsiChess, because of shortest code! > >Walter Faxon's magic LSB_64 and bitSearchAndReset_MMX (both faster not inlined) >both produce a about 3-5% slower IsiChess. LSB_64 and bitSearchAndReset_MMX are >equal in IsiChess. In a dumb-loop test bitSearchAndReset_MMX is clearly the >fastest (14.9 seconds) for 10^8 * 10 bits resets. > <snip> It's understandable that even if bsf itself is relatively slow, its not needing so many registers to do its work may more than make up the difference. Still, I'm confused. If the other two routines are not inlined, there must be at least a little procedure call overhead, right? The registers they need will have to be saved/restored either way. How can _not_ inlining produce faster code? Matt, help! -- Walter
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.