Investigate signal diffusion acceleration tools

[alice-i-cecile]

PDE approximation with operator learning: https://www.quantamagazine.org/new-neural-networks-solve-hardest-equations-faster-than-ever-20210419/

[alice-i-cecile]

Use fluid simulator? https://www.youtube.com/watch?v=Yzr9va5UtiE&t=4s

[alice-i-cecile]

More ideas:

• GPU acceleration
• Infrequent ticks
• Async tasks that are picked up opportunistically when spare threads are idle, especially in combination with frame pacing

[TimJentzsch]

What about a simple blurring technique? We could store the signal strengths in an image texture and then just blur it in a shader every X seconds.
We'd have to see which blurring technique to use for it to make sense with hex coordinates, but I assume that could be performant enough.

See also this video where a technique like this was used for slime simulation.

[alice-i-cecile]

For blurring, I would be worried that the signals wouldn't respect walls or other obstructions, which leads to very poor pathfinding behavior.

[bzm3r]

What about a simple blurring technique? We could store the signal strengths in an image texture and then just blur it in a shader every X seconds. We'd have to see which blurring technique to use for it to make sense with hex coordinates, but I assume that could be performant enough.

See also this video where a technique like this was used for slime simulation.

How would we get the data back from the GPU? Typically, reading data back from the GPU is not cheap.

[TimJentzsch]

What about a simple blurring technique? We could store the signal strengths in an image texture and then just blur it in a shader every X seconds. We'd have to see which blurring technique to use for it to make sense with hex coordinates, but I assume that could be performant enough.
See also this video where a technique like this was used for slime simulation.

How would we get the data back from the GPU? Typically, reading data back from the GPU is not cheap.

We'd have to send the texture to the GPU, process it there and then send the modified texture back to the CPU.
I assume at a certain size this will be faster because of the faster processing time, but we'd need to benchmark it to know for sure

[RobWalt]

A friend of mine just showed me this and it kind of reminded me of the problem outlined here. Maybe it helps