Author: Ed Schröder
Date: 04:59:09 02/03/02
Go up one level in this thread
On February 03, 2002 at 04:40:13, Wylie Garvin wrote: >On February 01, 2002 at 06:05:22, Steve Maughan wrote: > >>Ed, >> >>>It will be even slightly stronger because surprise surprise the Windows version >>>runs 11-12% faster than the DOS version due to the use of a new compiler. >> >>I thought Rebel's engine was 100% assembler, in which case the compiler would >>make little or no difference to the execution speed. What am I missing? >> >>Regards, >> >>Steve > >This is not surprising since Pentiums and up are primarily designed to execute >32-bit code (i.e. in a 32-bit code segment). When running DOS code under >Windows you are running in v86 mode from a 16-bit code segment, so the processor >does some things slower. If you use 32-bit instructions, then in certain cases >(e.g. instructions with immediates, I think) this will slow things down some >more. > >Hey Ed, > I found a web page today that I somehow never noticed before. It contains >some of the most detailed information I've seen to date about optimizing for all >kinds of Pentiums (PPlain thru PIII; P4 not included). I only noticed one thing >in it which *might* be incorrect, and more likely it is me who is incorrect. :) > >http://www.agner.org/assem/pentopt.zip I have the article already, I now have a second copy on my desktop. Thanks anyway. Ed >cheers, >wylie > >The thing I noticed was, it claims that on PPro/P2/P3 the xor reg,reg and sub >reg,reg patterns do NOT break dependency chains, while mov reg,0 does. The >intel documentation for P2/P3 does explicitly claim that xor reg,reg breaks >dependency chains. On P4 it most certainly does, along with sub reg,reg and >pxor mm,mm. Anyways.
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.