Author: Hristo
Date: 17:23:11 12/01/98
Go up one level in this thread
On December 01, 1998 at 18:02:47, Bruce Moreland wrote: > >On November 30, 1998 at 12:01:05, Tom Kerrigan wrote: > >>This is all fine and good for larger arrays that you don't access a lot, but if >>you're talking about your representation of the board, >> >>int chess_board[64]; >> >>will be a serious win over >> >>char chess_board[64]; > >I don't know why you are saying this in the face of at least two people saying >that they tried this exact thing and it was slow. > >On an Intel machine you have to do an extra instruction to read a char into EAX, >unless you use movzx or movsx, which suck. > >On an Alpha you have to deal with the alpha (21164) not even having an >instruction that will read a byte. > >These are reasons why I intuit that it might go slower, but if I used this data >structure I would surely test it and know for sure. > >I think that discussions of which implementation will go faster, that are based >upon predicted instruction timing or cache behavior, are almost always useless. > >bruce Bruce, the Alphas will kill anybody who uses memory not bound(packed) in 64 bits ... PII will simply wipe an Alpha chip ... now if one has everthung bound to 64 bit-size or testing floats then the results are in favor of the Alphas ... I did work for a while on a convertion of a comertial program(multitrack audio editor called SAW+) from WinNT-i386 to WinNT-Alpha the results were totaly devistating, since a much slower P133-166 would outperform Alpha 333 by a margin of three! Needless to say the convertion didn't go anywhere! Best regards. hristo
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.