[PD] int float with pure data

Mathieu Bouchard matju at artengine.ca
Tue Dec 12 03:31:59 CET 2006


On Mon, 11 Dec 2006, marius schebella wrote:
> padawan12 at obiwannabe.co.uk wrote:
>> Indexing tables with very big sizes is one place I've seen
>> the precision limitations come up in Pd. If you index a huge
>> sound file with a slow moving phasor you will hear tiny jumps
>> as it gets further and further into the array.
>
> you're right, with my calculation (16777216) with sr=44.100 Hz that affects 
> every table greater than 6min20sec???

With [tabread] and [tabread~] only: with [tabread4] it's actually much 
worse, because above half of that value they can't do interpolation, and 
even for values lower than that they can't get to the precision that might 
be required for proper antialiasing.

IMHO this is fixable by introducing either 64-bit floats in place of 
32-bit floats, or by introducing a dual system in which 32-bit floats 
coexist (32 or high performance, 64 for high precision). The latter is 
more work, but it's better because you wouldn't need to have separate 
executables for high speed pd vs high precision pd (and all DSP externals 
need to be recompiled differently to be compatible with a 64-bit-only pd)

  _ _ __ ___ _____ ________ _____________ _____________________ ...
| Mathieu Bouchard - tél:+1.514.383.3801 - http://artengine.ca/matju
| Freelance Digital Arts Engineer, Montréal QC Canada


More information about the Pd-list mailing list