18 Jul
2011
18 Jul
'11
4:56 p.m.
On Mon, Jul 18, 2011 at 08:52:55PM +1000, Erik de Castro Lopo wrote:
For instance in Secret Rabbit Code (my code), all data
entering
and leaving the converter plus the actual filter coefficients are
stored as 32 bit floats. However, the inner loop which does the
multiply accumulate (similar to what is done when mixing) does:
double result = 0.0 ;
for ( ..... )
sum += coeff [k] * data [k] ;
Specifically all the inputs are 32 bit floats, but all intermediate
results are 64 bit.
In the case they would probably be even if the sum was declared
float - it's very likely that the intermediate result will be
stored in a FP register anyway. OTOH there are compiler options
that would force it to single precision even in that case.
Zita-resampler does this a little bit different: it calculates
the sum from the start and end of the filter (where the very small
coefficients are) and working towards the center.
Ciao,
--
FA