To MMX or not to MMX [was: Re: tlug: Sony vaio 505]

[Manuel Chakravarty <chak@example.com>]

> I was just wondering, can Linux make use of the mmx specific features?

I'd say, the whole MMX business is mainly an overhyped
marketing plot by Intel, as the actual impact of the `MMX
technology' for the average user is probably close to zero
(independent of the operating system).

The reasons:

* The MMX instructions extend the instruction set of the
  processor.  The new instructions allow to do calculations
  that typically occur in video and audio applications
  somewhat faster.  For other applications, they are not
  very interesting.

* As we are talking about completely new instructions (and
  not the improvement of an existing feature), `old'
  binearies cannot benefit from the extension (in other
  words, you need binaries specifically compiled for an MMX
  cpu).

* So far, there actually seems to be no consensus on how to
  generate code containing the new instructions anyway.  Of
  course, you can tediously code it by hand (in assembler
  code), but compiler support is difficult.  The language C,
  for example, does not contain any programming construct
  that corresponds to the MMX instructions (i.e., which can
  be naturally translated to these instructions).  So,
  either a language extension is needed or the compiler has
  to perform some program analysis (which is actually
  similar to what is needed to generate code for vector
  computers and employed in vectorizing Fortran compilers).

Overall, these are very much operating system independent
problems.  A version of gcc that generates MMX instructions
could benefit applications like MPEG players (although, it
may be necessary that some kernel routines of the
applications have to be rewritten to allow the compiler to
spot opportunities for using these instructions).

Cheers,

Manuel

P.S.: There seem to be some additional features in MMX cpus
      (first-level cache was mentioned in a posting) that
      don't have much to do with the actually MMX
      instructions and are also independent of the OS.

DISCLAIMER: The expressed opinion is my personal one and I
	    explicitly want to emphasize that it does not
	    necessarily reflect my affiliations point of
	    view.
---------------------------------------------------------------
Next TLUG Nomikai: 11 March Wed 1998 Tengu TokyoEkiMae 19:30
Chuo-ku, Kyobashi 1-1-6, EchiZenYa Bld. B1/B2 03-3275-3691
Next TLUG Meeting: 11 April 1998 Saturday, Tokyo Station
Featuring Tague Griffith of Netscape i18n talking on source code
---------------------------------------------------------------
a word from the sponsor:
TWICS - Japan's First Public-Access Internet System
www.twics.com  info@example.com  Tel:03-3351-5977  Fax:03-3353-6096