discopatrick

I've written my own version of jit.alphablend in Gen. It differs slightly from jit.alphablend, in that my version preserves the transparency in the output, while jit.alphablend 'flattens' the output to be fully opaque.

https://en.wikipedia.org/wiki/Alpha_compositing#Alpha_blending

I've been following the great Gen Tutorials 2a and 2b, where 2a uses swiz to unpack RGB values, and then 2b encourages you to operate on all the values as a single vector.

Here are my two versions of the Gen patcher, correlating to the two tutorials:

However, I have a couple of thoughts about my v2. There is one point in the patch where I need to process the rgb values, but not the alpha channel. So, processing all channels as a single vector may have resulted in de-duplication in my code, but it seems that I'm also performing a wasted calculation.

I also still end up using swiz and vec towards the bottom of v2, to pack together values from two equations, ready for output.

* Is processing all channels in a single vector (and avoiding swiz and vec) simply a de-duplication/code readability issue, or will I get better performance out of it too?

* If I can indeed get better performance, how could I improve v2 to avoid the 'wasted calculations' I mentioned above, and use fewer swiz and vec objects?

Here's the equation I based it on: https://en.wikipedia.org/wiki/Alpha_compositing#Alpha_blending

However, I have a couple of thoughts about my v2. There is one point in the patch where I need to process the rgb values, but not the alpha channel. So, processing all channels as a  single vector may have resulted in de-duplication in my code, but it seems that I'm also performing a wasted calculation.

* Is processing all channels in a single vector (and avoiding swiz and vec) simply a de-duplication/code readability issue, or will I get better performance out of it too?
* If I can indeed get better performance, how could I improve v2 to avoid the 'wasted calculations' I mentioned above, and use fewer swiz and vec objects?


alphablend-in-gen-optimisation-and-de-duplication

Well I dont know, this looks very helpful thank you, and it could be very helpful for my video glitching with audio. But the problem with your question is that the gen version cannot use parallel processing for large arrays like the jit version cann, so the performance differnce in most cases wont be helpful to most people.

Don't know if it helps with performance, but you can use [swiz rgb] before the expression and [concat 3 1] to re-pack all together:

@ernest - you're welcome, I hope it comes to some use in your project!

On the subject of parallelism, I read this other post where Wesley Smith says that jit.pix does indeed do parallel processing in the same way that jit objects do: 

https://cycling74.com/forums/gen-and-parallel-computing/

On the subject of parallelism, I read this other post where Wesley Smith says that jit.pix does indeed do parallel processing in the same way that jit objects do: https://cycling74.com/forums/gen-and-parallel-computing/


@lska - that's great, I thought there must be something like that in Gen. And it would definitely help with readability, if not also performance.

alphablend in Gen - optimisation and de-duplication