Author: Robert Hyatt
Date: 14:24:55 09/18/04
Go up one level in this thread
On September 18, 2004 at 03:39:00, Gerd Isenberg wrote: >On September 17, 2004 at 14:33:41, Dann Corbit wrote: > >>http://www.imm.dtu.dk/pubdb/views/publication_details.php?id=3267 >> >>I got ~96% efficiency, going from 1 thread to 2 threads on a dual CPU machine. >>1172442 / 599317 = 1.9563 >> > >A cool paper, good explanation, very well structured and readable source code. >A realy great C++ repository. > >About the Issue "Parallel Chess Searching with Bitboards", i have the same >problem as others to get David's point, what makes the board representation or >it's additional size of 17 bitboards so unique while splitting nodes? Absolutely nothing. Cray Blitz wasn't bitmap. Crafty is. Has nothing to do whatsoever with the parallel implementation of the search, because the search is completenly independent of the representation of the chess board. > >The rotated bitboard approach is the "classical" as described by Bob Hyatt and >Ernst Heinz, with 256 occupied states: >http://www.cis.uab.edu/hyatt/bitmaps.html >http://supertech.lcs.mit.edu/~heinz/dt/node8.html > >What is the difference between David's lockless hashing approach and Bob >Hyatt's? Isn't it the same xor idea? >http://www.cis.uab.edu/hyatt/hashing.html > >Btw. is a 128-bit movdqa-instruction sufficent as an atomic read or write >operation? > No. Because the PIV can't really read 128 bits in parallel, so you have the same problem unless Intel guarantees that the two 64 bit reads are done on consecutive bus cycles, same for the two 64 bit writes... And I don't believe they guarantee that with out of order reads/writes everywhere... >Thanks, >Gerd
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.