[Vorbis-dev] OggVorbis encoder fixed point implementation

[parul]

Monty wrote:

>
>On Mon, Feb 20, 2006 at 07:37:40PM +0530, parul wrote:
>  
>
>>Hi,
>>
>>Have anybody worked on converting OggVorbis encoder floating point code 
>>to fixed point (32 bit or 24 bit). How do you handle _vp_noisemask in 
>>that case where precision goes upto 40 bits. Any help available?
>>    
>>
>
>If you're working on the encoder (ie, that wasn't a typo and you meant
>'decoder'), you need not worry about any features you don't plan to
>use yourself.  The full 40 bit range covers the possible range of many
>mutually exclusive techniques of using the floor and reside.  In any
>one given implementation, you are highly unlikely to need that kind of
>range.  
>
>A valid encoder need only output valid streams; a valid stream is one
>a valid decoder can successfully decode.  It makes no sense for an
>encoder to implement spec features it will never itself use.
>
>Monty
>
>
>  
>
Hi Monty,

Yes i am working on OggVorbis encoder only. And we are trying to convert 
"libvorbis" encoder to fixed point that too with 24 bit precision. So, 
you mean to say that we should change the implemetation such that it 
doesnt need 40 bit precision. I was thinking that but its not clear how 
this noise mask curve is generated (i am working on that to understand 
it), neither i was able to find any reference material related to that.
1.If you have any idea about what can be done to change the implemetation.?
2.In addition if we change the implemetation, what would be the basis of 
complaince. Is there any compliance testing for encoder? (or just that 
the stream generated should be valid and such that a valid decoder can 
decode as u said). As we are changing the code to 24 fixed point 
precision we dont knw how to do compliance testing for the same.
3. Is there any fixed point implemetation of encoder available as for 
decoder (Tremor)?

Thanks,
Parul  
Embedded Engineer
Einfochips Ltd