Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: 3 optimization questions

Author: Anthony Cozzie

Date: 07:49:42 06/19/03

Go up one level in this thread


Bo has it mainly right, but here are a few more tidbits:

MOVD is the instruction that transfers data (32bits at a time) between MMX and
regular registers.  It takes 3 cycles to move to an MMX reg and 5 cycles from.
If you look at the diagrams of the Athlon you will see that the floating point
and integer pipelines are more or less separate.  Also note that most MMX
instructions take 2 cycles to execute.

I don't think there are any control instructions based on MMX registers [cmpXX].

The Intel compiler will vectorize things for you, but I don't think it's smart
enough to do the kind of thing you want it to do.

anthony



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.