[PD] very compressed chip sounds

[Mathieu Bouchard]

Le 2011-10-15 à 17:21:00, glerm soares a écrit :

> & sorry about the naive question - but can someone here try to explain 
> shortly what this line does? This ">>" means some kind of bitshifts ?

Yeah... but to be clear, it does convert to plain int (truncating), shifts 
the integer, and converts back to float, which is both slow and often less 
useful than ldexp. But there *is* a real use for a truncating bitshifter 
in pd.

About this, I have an anecdote : I went in a big IRC channel on FreeNode 
about C or C++, asking for a float equivalent of the << and >> operators. 
At least five people began by assuming that I wanted something weird that 
has no possible use, and they started flaming me for it (real angry). In 
the end, when I got them to figure out what a useful equivalent of << and 
>> could be, they didn't know. Then I probably scanned in <math.h> for all 
the functions I didn't know and finally found ldexp. I think that in pd, 
it's only available in [expr] and in GF.

so, the difference is that 31>>3 == 3 whereas ldexp(31,3)==3.875, with 
fractionary bits kept.

BTW, in floats, when bitshifting, the bits don't actually shift, and 
instead, the exponent field increases or decreases, because that's how the 
float format is multi-scale : it has a builtin concept of << and >> at its 
core (even though hardly anyone ever knows what ldexp is !)

> I think some curious question could be - if there is a an [>>~] operator 
> in pd what it does? Moves a bit for the next cycle of the of the sample 
> block ?

What happens with nearly all simple math operators that have a class in pd 
for floats, and another similarly-named class for signals, is that the 
latter does the job of the former on every float that is found inside each 
block. This happens without interactions between floats of different 
instants.

So, [>>~] does, for each time t inside each block, take input x[t], apply 
[>>], put into output y[t], that's all. It's all the same pattern as [+] 
vs [+~], and [*] vs [*~], etc.

The big exception is stuff like [cos~], that has a different scale factor 
because Miller said so.

> Can you indicate some patches to clarify this theory?

No... actually, I don't know much of a use for [>>~] in particular... 
whereas [expr~ ldexp($v1,...)] might be optimised way(s) to do certain 
cases of [*~]. I use >> a lot, but never in a signal context.

With people solving control-style problems using signal-style solutions 
(e.g. Barknecht's experiments), I can see more use of typically non-signal 
stuff being done with signal anyways, and this would explain the existence 
of [>>~].

But frankly, I think that the existence of [>>~] is simply for 
consistency, for completing the pattern of correspondence between float 
ops and signal ops. In a certain sense, pd is simpler when it is more 
complete, because there are less exceptions in its design. You know what I 
mean ?

  ______________________________________________________________________
| Mathieu BOUCHARD ----- téléphone : +1.514.383.3801 ----- Montréal, QC