delorie.com/archives/browse.cgi   search  
Mail Archives: djgpp/2000/01/12/05:56:41

From: Nate Eldredge <neldredge AT hmc DOT edu>
Newsgroups: comp.os.msdos.djgpp
Subject: Re: Is there any point in MMX?
Date: 12 Jan 2000 01:08:42 -0800
Organization: InterWorld Communications
Lines: 25
Message-ID: <83d7r7y7vp.fsf@mercury.st.hmc.edu>
References: <8599nt$1uo3r$1 AT reader2 DOT wxs DOT nl>
NNTP-Posting-Host: mercury.st.hmc.edu
X-Trace: nntp1.interworld.net 947668081 54024 134.173.45.219 (12 Jan 2000 09:08:01 GMT)
X-Complaints-To: usenet AT nntp1 DOT interworld DOT net
NNTP-Posting-Date: 12 Jan 2000 09:08:01 GMT
X-Newsreader: Gnus v5.7/Emacs 20.4
To: djgpp AT delorie DOT com
DJ-Gateway: from newsgroup comp.os.msdos.djgpp
Reply-To: djgpp AT delorie DOT com

"Bas Hamstra" <bas DOT hamstraNOSPAM AT wxs DOT nl> writes:

> Thanks for the answers about Win & RSXNT. I can report it works fine so far.
> 
> About MMX: why does one hear so little about it? Could there be any
> speedgain for my chess app using MMX? Aren't there a couple of extra
> registers, which the x86 needs so badly?

As I understand it, no.  There aren't extra registers.  They overload
the floating-point registers into "vector" registers, and add some
vector arithmetic.  This means you could parallelize some of your
arithmetic (i.e. do two or four or eight additions with one
instruction).  But you'll probably have to do it in assembly; it's a
difficult thing for a compiler to handle, and GCC currently doesn't
(though I believe there are some special patches).

The other aspect of MMX is that MMX chips tend to come with more cache
than their non-MMX counterparts.  Clever strategy for Intel; MMX makes
things faster right away!  Who cares that the instructions are mostly
useless?

-- 

Nate Eldredge
neldredge AT hmc DOT edu

- Raw text -


  webmaster     delorie software   privacy  
  Copyright © 2019   by DJ Delorie     Updated Jul 2019