Author: Gerd Isenberg
Date: 05:47:42 08/08/03
Go up one level in this thread
On August 08, 2003 at 07:56:21, Vincent Diepeveen wrote: >On August 08, 2003 at 07:46:17, José Carlos wrote: > >>On August 08, 2003 at 07:40:14, Vincent Diepeveen wrote: >> >>>On August 08, 2003 at 02:50:01, José Carlos wrote: >>> >>>>On August 07, 2003 at 23:48:51, Vincent Diepeveen wrote: >>>> >>>> >>>> I've tried something like that in the past. My move generator is faster that >>>>this one. In my other program (with non rotated bitboards) the move generator >>>>alone is slower, but I can calculate many things very cheap. If I try to do in >>>>my 0x88 program all the stuff I do in my bitboard program, it gets twice slower >>>>instantly. >>> >>> >>> >>>> Feel free to say that this happens because I'm a very bad programmer, if you >>>>feel better that way. >>> >>>Yes you are a very poor programmer unless you tested at a 486. >>> >>>You hardly have branch mispredictions with this move generator in case you >>>didn't notice. So it kicks the hell out of anything you do in advance, >>>because 2 branch mispredictions is already more expensive than generating >>>1 move with this generator. >> >> Thanks for the (expected) compliment. >> I have zero branch mispredictions in my current move generator. >> >> José C. > >then you should really test this. > >bitboards branchless is a zillion penalties. just see what penalties and >register stalls Gerd measured. Sorry Vincent, can you explain. I have no idea what you are talking about. Hmm, may be the conditional bitscan, which was "surprisingly" faster than the branchless? But that was not a branch issue, rather than the expensive vector path instructions in a row ( bsf, bsf, btr). Even with sophisticated branch prediction heuristics, i prefere branchless code, no matter whether 32-bit or 64-bit data. Regards, Gerd <snip>
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.