Vorbis dev - Mar 2005 - Query on Codewords length

Hi,



We are porting the Ogg Vorbis fixed point decoder to a 24 bit fixed
point precision processor. We are trying to reduce the number of cycles
for MIPS optimization. In this process we observed that most of the
songs do not have "maximum codeword length" greater than 24 bits.



We are planning to reduce most of the double word operations (in our
case 32 bit data is stored as double word) to single word, i.e. 24 bit.



I would like to know, what is the chance of getting a codeword of length
greater than 24 bits in a song or for songs of different qualities while
encoding? 



The maximum possible codeword length being 32 bits, we are currently
storing it in double word. If the likeliness of it occurring is very
less then we plan to use only single word ( 24 bits ) operations for the
Huffman decoding.



Regards,

Valmick













Confidentiality Notice

The information contained in this electronic message and any attachments to this
message are intended
for the exclusive use of the addressee(s) and may contain confidential or
privileged information. If
you are not the intended recipient, please notify the sender at Wipro or
Mailadmin@wipro.com immediately
and destroy all copies of this message and any attachments.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.xiph.org/pipermail/vorbis-dev/attachments/20050331/7686f4bc/attachment.html

On Thu, Mar 31, 2005 at 12:38:08PM +0530, valmick.guha@wipro.com
wrote:
> We are porting the Ogg Vorbis fixed point decoder to a 24 bit fixed
> point precision processor. We are trying to reduce the number of cycles
> for MIPS optimization. In this process we observed that most of the
> songs do not have "maximum codeword length" greater than 24 bits.
Correct; you will be able to tell this from the codebooks in use when
loading the headers.
> I would like to know, what is the chance of getting a codeword of length
> greater than 24 bits in a song or for songs of different qualities while
> encoding? 
It is always possible that an encoder may decide to use codewords over
24 bits; this is not governed by anything ther than the encoder's
whim, that is, there is no aspect of a given mode, bitrate, or sample
rate that would disallow use of greater than 24 bit codewords.

That said, the liklihood of them being employed is admittedly fairly
low.
> The maximum possible codeword length being 32 bits, we are currently
> storing it in double word. If the likeliness of it occurring is very
> less then we plan to use only single word ( 24 bits ) operations for the
> Huffman decoding.
If it occurred and your decoder could not handle it, it would render
the entire file undecodable.

I agree that the optimization is a good one (Tremor's low-memory
version does something similar).  Hoever, there should also be
facility for supporting the occasional codebook using > 24 bit
codewords.

Monty