Author: Robert Hyatt
Date: 17:40:59 05/14/98
Go up one level in this thread
On May 14, 1998 at 19:15:24, Bruce Moreland wrote: > >On May 14, 1998 at 18:32:26, Ivan Tirado wrote: > >> In contemporary chess programs, what gets more use: >>floating point math or integer math? If integer math is preferred, how >>do different processors handle it ( ie perform )? Can a chess program be >>optimized to use MMX if integer math is used? In the case floating point >>is preferred, why this is so? > >There is probably some guy out there working furiously on a program, and >his big angle is that he does floating point math in just the right >spots, or so he thinks. > >On all the micros that I know of, you should use integers if you can. >Most chess programs are extreme examples of optimization, so they'll >tend to not use floating point at all. > >If there is a processor where floating point is faster, or some >combination of floating point and integer is fastest, there is some >lunatic out there right now writing floating point chess program code in >assembly language. > >And if the assembler doesn't do exactly what he wants, he'll input the >op-codes by hand into a hex editor. > >bruce I know such a lunatic. :) On the Cray floating point math is *fast*. because it was designed for floating point... so we did mostly fp stuff as it gave us another degree of parallelism in that on that machine, we can do 1 32bit integer op, 1 64bit integer op, and one 64bit fp op, all in one cycle (on one cpu)... not in crafty of course...
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.