[jitter] pixel by pixel performance

Without seeing exactly what type of calculations you are doing, it is 
difficult to say what you should do.  The only potential issues with 
using a shader is the limited functionality of float textures on some 
video cards, and the performance hit incurred when transferring to and 
from the GPU.  Nevertheless, I generally think that pixel-by-pixel 
calculations can be avoided altogether by using jit.op or jit.expr in 
clever ways.

Also, fwiw, I typically find that using jit.poke~/peek~ is way faster 
than using getcell/setcell.

Best,
Andrew B.