[Vorbis-dev] port of vorbis encoder to non-fpu archs

xiphmont at xiph.org xiphmont at xiph.org
Thu Jan 17 09:49:15 PST 2008


On Jan 17, 2008 11:58 AM, Ian Malone <ibmalone at gmail.com> wrote:
> Hmm.  The barrier is typically:
> 1. Most embedded platforms lack floating point.  This means the
>   existing encoder would have to be ported or rewritten to use
>   integers.  It's possible, but you'd have to get acquainted with
>   the encoder and there's probably some thought required in places
>   too.

Yes.  The hard parts, in a sense, are already ported.  In some ways
the encoder is easier because it need only work for one strategy-- the
one you're specifically implementing-- whereas a decoder must be valid
for any strategy any encoder decides to use.  Vorbis is an odd format
in that it gives an encoder 'alot of rope'.

> 2. Memory.  Monty pointed out a while ago that this is a problem
>   for the decoder in hardware

Not terribly relevant for an encoder, again because a decoder needs to
support anything an encoder may choose to do whereas an encoder only
needs to implement any one strategy (and all the parts that a decoder
must treat as stream-defineable can be static in an encoder because
the encoder controls the stream).

Of course, someone needs to do the work as no one has (to my
knowledge) done serious work on an integer encoder.  I would say,
though, that it would not be fundamentally more difficult than the
fixed point decoder was.

Monty


More information about the Vorbis-dev mailing list