Author: Gian-Carlo Pascutto
Date: 05:37:48 08/08/02
Go up one level in this thread
On August 08, 2002 at 08:11:13, Vincent Diepeveen wrote: >>This is very short, but in tests with my Athlon XP, it is no faster than the >>version with branches and checks for zeroed bitmaps. > >that's of course a dumb test to do. you should try each time a random >chosen bitmap. I think you misread what he wrote. He didn't test with zero bitmaps. The problem of the short routine is that bsf is very slow on the Athlon, and it can only be handled by the vector decoder. In a bitboard program, typical case is to have an empty half-bitboard, and avoiding the bsf is faster even if it's occasionally mispredicted. -- 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.