[prev in list] [next in list] [prev in thread] [next in thread] 

List:       mp3encoder
Subject:    Re: [MP3 ENCODER] problem with lame
From:       Alexander Leidinger <Alexander () leidinger ! net>
Date:       2001-11-20 11:51:41
Message-ID: 200111201151.fAKBpgn02630 () Magelan ! Leidinger ! net
[Download RAW message or body]

On 19 Nov, Robert Hegemann wrote:

>> Download intels compiler for linux (you have to register, but for
>> personal use you get it for free and unlimited). I used it here on
>> FreeBSD to recompile libmp3lame -> 20% speed increase (compared against
>> gcc 2.95.3 on an AMD Duron, I modiefied the Makefile and libtool by
>> hand). Used compiler options for icc: "-rcd -xi -ipo".

> according to www.heise.de, "ipo" optimization is broken in current 
> intel compilers.

Yes, I know. But not in this particular case. I've linked only the
shared lib, not the complete executable (not possible with the linux
version if you use FreeBSD, it depends on glibc, which we didn't have in
this case). The "ipo" optimization only takes place if you link the
complete executable (some kind of inter object optimization), icc prints
a message when it does this kind of optimization.

The resulting lib doesn't produce a different mp3 if I use "-i" (or how
this is named, I can't look at the moment) instead of "-ipo", so I
decided to go with "-ipo", because I tried to get a native FreeBSD
executable. But I failed to find an option to use user specified
libs/objects instead of the default libs/objects of icc.

Using ld itself is an option, but in this case you lose the "ipo"
optimization, which may be fixed in the next release. Using ld also
implies to hack the LAME Makefiles on your own, configure itself isn't
able to handle this. I already thought about some kind of bridging lib,
which wraps FreeBSD libc calls withhin the used glibc symbols, but I
haven't wnough knowledge about the internals of the glibc.

Bye,
Alexander.

-- 
   If Bill Gates had a dime for every time a Windows box crashed...
                ...Oh, wait a minute, he already does.

http://www.Leidinger.net                       Alexander @ Leidinger.net
  GPG fingerprint = C518 BC70 E67F 143F BE91  3365 79E2 9C60 B006 3FE7


[prev in list] [next in list] [prev in thread] [next in thread]