16 Feb
2011
16 Feb
'11
12:33 a.m.
The other day I tried somthing unusual. I use 96 frame buffers and so I
allocate two of those for each of in and out, and then read/write
(blocking) 96 frames at a time - makes sense, yes?
Well no, it turns out that - without doing any processing - looping over
read/writing only 16 frames at a time is more robust.
How come?
The same effect can be observed with 64 frames and 16 appears to be the
optimal value in the loop.
/j