Luckily, I'm soon about to get a new PowerBook G4 1GHz. In fact, it should be delivered this week. When I get that computer, I will probably want a project to exercise the AltiVec powers of the computer, as well as my (currently limited) skills in using it.
There are several algorithms in LAME that are optimized for other vector units (the SSX, for example). One approach would be to optimize those algorithms for the AltiVec as well. Another approach is to do what you suggest, and use Apple's DSP library, but I'm not sure that is allowed, LAME being LGPL (or is it GPL?). While the DSP library is open source, I'm pretty sure the license is not GPL compatible.
/ Regards, David Remahl
On Thursday, January 9, 2003, at 08:47 AM, Wai Liu wrote:
Apple ships Altivec powered DSP libraries for OSX. If you've ever seen iTunes encode Mp3s, its fast! I am quite interested in doing this (I've posted here before). The simplest way would be to replace FFT routines with the Apple ones etc. This is quite obvious, but real proof, just do some basic profiling of Lame.
However (and this is purely from memory) , it's not that simple as data structures and inputs are not "apples to apples". It would help if there was some documentation regarding function call parameters, general layout of data structures, and overall data flow.
You can read the source code, you can grab books on FFTs, Mp3 frame structures etc... LAME is great, but I suspect that not that many people are looking to Altivec because it requires a lot of effort to get up to speed, while good alternatives already exist e.g. iTunes, or they could work on newer codecs e.g. Vorbis.
_______________________________________________ mp3encoder mailing list [EMAIL PROTECTED] http://minnie.tuhs.org/mailman/listinfo/mp3encoder
