The Intel compiler attempts to do this. Unfortunately, detecting
vectorizable code is a very difficult problem. I tried the Intel
compiler and it didn't vectorize any of the floating point loops, and
only one rarely-called integer loop.
I have lots of problems getting VTune to work on lame to gather
adequate profiling information, but would be happy to write
some assembly for FFT, etc. It may be possible to speed up
the quantize_xrpow even more with 3DNow or KNI.
----- Original Message -----
From: Mathew Hendry <[EMAIL PROTECTED]>
To: <[EMAIL PROTECTED]>
Sent: Friday, 07 January, 2000 20:17
Subject: Re: [MP3 ENCODER] Windows Binaries
> > From: "Ross Levis" <[EMAIL PROTECTED]>
> >
> > Is there any chance of compiling a Windows version that takes advantage
> > of MMX or 3DNow?
>
> Can you name a compiler which will do this on normal code, i.e. without
the
> use of MMX/3DNow-specific functions and macros or [inline] assembler?
> Normally special-case code will be required to take advantage of these
> enhanced instruction sets.
>
> However, there are some customised versions of LAME around. Take a look at
> the GOGO-no-coda, which I believe has MMX and 3DNow! enhancements, amongst
> other things.
>
> http://www.kurims.kyoto-u.ac.jp/~shigeo/gogo_e.html
>
> Note that GOGO is based on an earlier version of LAME (3.29 IIRC).
>
> -- Mat.
>
>
> --
> MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
