>
> >Idea for quantize_xrpow()
> > - calculate xrpow_max when initially calculating the xrpow[] array
> from
> > xr[]
> > - in quantize_xrpow[] before doing the 576 calculations, check that
> > xrpow_max * istep does not exceed 8191 + 14. If it does, exit and
> > choose a new step size.
> > - this basically eliminates the maximum value check in count_bits
> and
> > saves a chunk of time.
>
> Am I missing something or shouldnt this xrpow quantizer function be
> implemented
> in the calc_noise and in the decoder too?
>
> ..And isnt that calc_noise calculations just the reverse what it's
> doing in
> the quantize routine, just (re/reverse) calculating the double -> int
> round
> offs which could be stored for later use in the quantize function...
> even
> calculateing the noise at the same time.. I'll try to implement.
I thought about this after reading your message, but couldn't see how
to do this - did you figure it out? However, most of the pow()'s
in calc_noise are done with table lookup instead of floating point
calculations.
>
> > Idea for binsearch_stepsize()
> > - JHL's binsearch is good. It may be possible to *predict* the
> > stepsize rather than keep guessing it.
>
> Even better to predict it for each channel separately..
>
Actually, the old quantanf_init() was ISO's attempt to do this
predition. The problem was it would sometimes predict too large of a
stepsize. This was bad because the old binsearch_stepsize() could
only increase the starting stepsize during it's search. JHL's
binsearch will work correctly with any inital guess for stepsize.
Mark
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
