[jitter] pixel by pixel performance

Hello

I am analysing videos on a pixel by pixel basis.  What I am basically doing is addressing the pixels I need to extract from one  matrix and convert/dump them into another which is then sent to daemon (via jit.net.send) that does other stuff.  I have to do it fast (about 50 fps at a minimum) and dump the data just as fast.

I have been able to improve performance tenfold by rewriting my analysis in JS and it worked ver well on small matrices (256 pixels).  However, trying to move on to a larger matrix (1300 pixels) I hit 100% CPU (and in the end I need to operate on 4500 pixels).

so far I've been using getcell and setcell methods to transfer data from one matrix to another using pixel addresses stored in arrays and simple for() loops were doing the work.  However, for loops on large arrays in JS take a long time to compute (up to 180ms).

Does anyone have any odea how I could improve performance in such a scennario.  I've begun looking into slab, hoping that I can transfer the pixel munging onto the GPU and just grab my matrix back when it's done.  The GPU isn't used for other tasks, anyways.

I would like to know if anyone can confirm that I will actually gain anything from gl.slab performance-wise?

Otherwise, if anyone has any idea on improving performance in this context, I would appreciate.

m.