Author: Gian-Carlo Pascutto
Date: 11:26:52 05/25/02
Go up one level in this thread
On May 25, 2002 at 14:21:44, Russell Reagan wrote: >On May 25, 2002 at 14:11:01, Gian-Carlo Pascutto wrote: > >>What do you mean by 'stable and reliable?' Either it works or it does not... > >I was referring to your statement about how the shift operation was predictable >and the if statement was not. I assumed you meant that in an if-statement, the >contents of the expressing being evaluated could be anything, and therefore the >shift/AND sequence is more "stable" and "reliable" than the add/if-test >sequence. I mean more predicatble for the branch prediction unit on your CPU. The shift/and is just arithmetic, whereas the raytracing will be mispredicted at least once per direction, of which there are four per piece. On Athlon, that means about (not sure of exact numbers) 12*4 = 48 cycles wasted. You can do a lot of arithmetic in that time. Pentium IV is even worse (+80 cycles). Then again, a disadvantage of the bitboard method is that the precaclulated data takes more space in the data cache. -- GCP
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.