Is local array memory in jit.gen possible?

Hey Diemo,

I think jit.gen is not the best option for the task.
The method Graham suggested works great for outputting a single value starting from a longer array, but it's not suitable for writing data at specific locations. The parallel nature of jit.gen/jit.gl.pix prevents arbitrary data storage and manipulation.

GLSL is a tricky option, as it's only possible to write data in buffers or textures at specific locations. It could be possible with image-type data instead of textures, but they're currently unsupported in jitter shaders.
A cumbersome and "abusive" way exists to push and pop elements to and from an array in GLSL, which is sometimes used in contexts such as ray tracing to traverse a data structure. Still, it's less than ideal and requires knowing the maximum length of the array a priori.

EDIT:
With GLSL, another option is a very exoteric method of writing data at arbitrary locations:
You render some points in a one-dimensional rendering window and use their color to bring the data you want to store. Then, you render such points using "add" as blending mode. This way, if you consider the rendering window your array, you can sum values at arbitrary indexes, as you can decide the position for each point. That's currently the only way to write data at arbitrary locations using the parallelized rendering pipeline. Still, I'd not suggest it, also because you would have somehow to keep track of the length of the resulting array by doing something even crazier, like abusing the depth buffer to know how many elements have already been written in the array.

I'd suggest implementing the process in javascript or with the good old standard Max programming (with matrices, getcell, and setcell).

I hope this helps; cheers!