sound

sinc sound synthesis
(based on http://slack.net/~ant/bl-synth)

The first difference of a band-limited step is a sinc function (or at least, is closely approximated by same).

So we first generate a bunch of values of the sinc function, at high (near-CPU) resolution.  Then, for each offset within one audio sample, we go along and make the table of summed groups, like this:

sinc (not):  0 0 0 1 1 1 0 1 1 1 0 0
offset 0/4:     1       3       2
offset 1/4:       2       3       1
offset 2/4:         3       3
offset 3/4:   0       3       3

Then we just have to keep live a record of all the currently active steps.

Shutdown sequence (needed because abuf is shared):
While running we have                                  [1, 1]
Core wants to exit, sets !C                         -> [0, 1]
Core spins on !A
Audio sees !C, fills audio with silence and sets !A -> [0, 0]
Core sees !A, exits (and abuf is destroyed)
Audio continues writing silence and thread eventually exits