Thanks for elaborating and thanks for the compliment:) I just started and your site is a constant source of inspiration.

I am just waiting for an idea to come along that would justify using PB, I think I got something and was planning on writing it up in the weekend, but these things has a way of suddenly taking longer that expected.
I would like to utilize sound and have PB write up a bump map/flow field from the fast Fourier transform, either having PB do both the FFT and the bump map or just the bump map, I would think that having flash doing the sound manipulation for a low res sample was just as fast as having the data sent back and forth… to the lab.

I’ll ping this post if I get something written up.

Cheers

Cheers for the comment. I’ve been really enjoying browsing your site! You have some great stuff going on there :)

Re. your comment – I didn’t find much by way of explanation either, so it was the Flex SDK docs that helped the most in the end. Perhaps I should write up the process and put up a new post on the subject?

This particular experiment doesn’t gain a huge amount of performance from using PixelBender – it was simply that I saw it as a good opportunity to implement the theory.

You’re pretty much spot on with your description though – each tentacle is made up of 150 to 300 segments, so what I’m doing is pre-computing the position of each segment using an additive sine / cosine based algorithm, but instead of doing this in Flash, I’m setting up a ShaderJob which will compute all of these values for me and write them to a Vector, which is passed back to Flash and the values are simply read from the Vector as the tentacle draws itself.

There’s some debate as to the correct and incorrect times to use this process, as writing the initial ByteArray or Vector, passing it to PB and then retrieving the results and reading them is a costly process in itself (as opposed to just performing the calculations inline with Actionscript – optimised AS can be just as good sometimes). I think that generally its a question of testing the performance of both approaches on a project-by-project basis. I can see why a 3D engine for example would benefit from the PixelBender approach – if the calculations for your meshes / cameras / textures are all bundled into a single job that can run in another thread asynchronously, then Flash can use its resources for rendering.

I’ll try and write up a quick example anyway. Generally it’s a pretty straight forward process, though there are a few pitfalls when dealing with PB!