41–50 of 54 posts
This reminds me very strongly of the WireWorld cellular automaton: https://quinapalus.com/wires0.html (Also, how did you get a post with both text and a link? I didn’t know it was possible.)
Wired-Logic uses a GIF as output.
Interesting take on making AND and XOR gates, makes it a bit more colorful.
We’re also working on something like this, but we’re making a game around it https://charperbonaroo.github.io/bls/#0
Oh wow this is awesome. I got through the first few levels and would like to complete it later... Also, is this the same as the 2015 'bitmap logic simulator'? I found this while reworking reso awhile back: https://realhet.wordpress.com/2015/09/02/bitmap-logic-simula...
I’m rewriting it to Unity and there I can run the 8 bit cpu from that blog with about 14000 ticks per second.
We’re planning to create a game around it with few dozen of levels, including controlling robots, traffic lights, a bowling pin sorter and tanks/units in a tiny top-down shooter (where programs from players play against each other).
Edit: if you download the cpu image from the blog and upload it to the game, you can see it run. It’s visually a bit broken because the game in its current state isn’t really designed to have such a big canvas, but it runs perfectly fine.
Edit 2: the CPU running in unity web build here: https://charperbonaroo.github.io/microengineer-build/
We’re also working on something like this, but we’re making a game around it https://charperbonaroo.github.io/bls/#0
I haven't looked at the source, but I bet it's massively parallel, at the pixel level? (I do see you say it's "not a cellular automaton", though...)
If so, I wonder how difficult it would be to port this to GPU, either writing CUDA yourself, or maybe even using PyTorch.
If PyTorch, you might be able to make it differentiable (by changing ANDs to MULs and such). What would be achieved by making the simulation differentiable I don't know. But surely you'd come up with something interesting. Given how much effort has gone into developing architectures whose gradients don't explode, my guess is you'd run into difficulties, but maybe you could even learn programs that model various sequences.
The main point would just be to run it on the GPU, though.
Brilliant idea! That logo however is the most repulsive piece of artwork I've seen in a long time. (I am sorry for being blunt)
Super-cool. I haven't looked at the source, but I bet it's massively parallel, at the pixel level? (I do see you say it's "not a cellular automaton", though...) If so, I wonder how difficult it would be to port this to GPU, either writing CUDA yourself, or maybe even using PyTorch. If PyTorch, you might be able to make it differentiable (by changing ANDs to MULs and such). What would be achieved by making the simulat…
Regarding differentiation, the logic gates could look like a differential equation. Between iterations, input nodes amass a binary vector from adjacent wires, and pass them onto adjacent output nodes. So, mapping {0,1}^n --> {0, 1}, you could write them all in terms of some periodic function. (E.g. xor implemented with sin, or implemented with a sigmoid made from a quarter of a sin, etc.)
Either way, yeah! There is plenty of room to speed this up
Similar idea to Wired-Logic https://github.com/martinkirsche/wired-logic Wired-Logic uses a GIF as output. Interesting take on making AND and XOR gates, makes it a bit more colorful.