Author: David Rasmussen
Date: 04:23:30 11/17/02
Go up one level in this thread
On November 16, 2002 at 22:31:23, Christophe Theron wrote: > > >Inferior? > >Chess Tiger 15.0 compiled with GCC is as fast as Chess Tiger 15.0 compiled with >MSVC 6 (with tons of patches). > >Maybe the number of bugs in GCC is INFERIOR to the number of bugs in MSVC, I >don't know. > It depends very much on the particular chess engine, whether one or the other is best. Specifically, gcc is bad at bitboards on x86 machines. I've discussed it with the x86 backend people of gcc, and they admit that because of the infrastructure of gcc, it doesn't handle these artificial 64-bit (double 32-bit) calculations well. As they explained to me, the problem is that one 64-bit number allocates registers as a whole, that is, two 32-bit registers are tied up for the 64-bit number (which is ok), but register scheduling can't, by design, separate them later on, move on of the 32-bit parts into another register etc. even if it would result in faster code. They are aware of this, and a rewrite of the backend to address this will come at some point, but not soon. And by the time 64-bit Intels and AMDs are in general use, the interest wont be as big anymore. /David
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.