Author: Robert Hyatt
Date: 08:01:11 06/08/98
Go up one level in this thread
On June 08, 1998 at 10:34:23, Vincent Diepeveen wrote: >On June 08, 1998 at 07:37:44, Robert Hyatt wrote: > >>On June 08, 1998 at 07:17:08, Vincent Diepeveen wrote: >> >>> >>>On June 05, 1998 at 21:48:57, Robert Hyatt wrote: >>> >>>>On June 05, 1998 at 17:47:07, Bruce Moreland wrote: >>>> >>>>> >>>>>On June 05, 1998 at 08:53:53, Ernst A. Heinz wrote: >>>>> >>>>>>BTW, we have been waiting to get our hands on a Alpha-21264 for roughly >>>>>>a year because "DarkThought" will hopefully speed up as much on them >>>>>>as "Crafty". >>>>> >>>>>What percentage of execution time do you and Bob spend doing these >>>>>operations, do you think?netscape -install > >>>>> >>>>>bruce >>>> >>>>FindFirstOne() = 1-2% in crafty. popcnt is much less... so that won't >>>>help a whole lot. but the 64 bit stuff helps in general... >>> >>>Well that doesn't say anything concrete. >>> >>>More important is the overall delay. >>> >>>At a P133 intel the speed difference between crafty's move generator >>>and diep's move generator is somewhat more than 3 times. >>> >>>Crafty is more than 3 times slower, because of bitboards, which perform >>>badly at 32 bits. Note that the more possibilities, this 3 times will >>>become >>>4 times. >>> >>>For something taking 1 clockcycle in my program and which can >>>be even pipelined in U + V pipe, >>>crafty needs sometimes even a whole function. >>> >>>Greetings, >>>Vincent >> >> >>yes... but try to generate nothing but captures and compare that, which >>is a major part of the total tree search... > >Where do i need to generate captures for? > >But to answer your question: 2.5 times faster. Because in ordering alpha/beta moves, captures are searched first. And they cause most of the cutoffs. If you just produce captures, you don't have to continually wade through the non-captures picking out capture moves to try... and the best test to try is to set up a position in Crafty, and use the "perf" command... set up a position such as kopec22 which is a reasonable early middlegame position, and type "perf". On a single processor pentium pro, that produces the following: Black(1): perf generated 3300000 moves, time=1.38 seconds generated 2391304 moves per second generated/made/unmade 3300000 moves, time=4.94 seconds generated/made/unmade 668016 moves per second IE I can generate all moves from that position for white, at 2.4M moves per second, or I can generate all the moves, and then make each one, and do that at 670K moves per second... What's your speed? here is the position, diagram and FEN: Black(1): d +---+---+---+---+---+---+---+---+ 8 | | | *R| | | *R| *K| | +---+---+---+---+---+---+---+---+ 7 | | *B| *Q| *N| *B| *P| *P| | +---+---+---+---+---+---+---+---+ 6 | | *P| | *P| *P| *N| | *P| +---+---+---+---+---+---+---+---+ 5 | *P| P | | | | | | | +---+---+---+---+---+---+---+---+ 4 | N | | P | | P | | | | +---+---+---+---+---+---+---+---+ 3 | P | | | B | | N | | P | +---+---+---+---+---+---+---+---+ 2 | | B | | | Q | P | P | | +---+---+---+---+---+---+---+---+ 1 | R | | | R | | | K | | +---+---+---+---+---+---+---+---+ a b c d e f g h 2r2rk/1bqnbpp/1p1ppn1p/pP/N1P1P/P2B1N1P/1B2QPP/R2R2K b
This page took 0.01 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.