I managed to make a shader based heightmap fluid simulation that runs in shaders in LÖVE.
I added some basic shading as well as adding/removing fluid with LMB/RMB.
The algorithm is a slightly modified version of the "Pipe method" described in https://trepo.tuni.fi/handle/10024/115052
Basically it assumes that the hydrostatic pressure is a direct function of height and flow can only happen in vertical and horizontal directions to neighbor cells (therefore pixels can be mentally thought of as square basins connected with a grid of pipes). I also added a blurring step after every 3 simulation steps for stability (simulates viscosity)
BrotSagtMist wrote: ↑Wed Jul 17, 2024 7:16 pm
I mean yes, this looks great, cool light play and depth.
But i only get 1 picture every 10 seconds.
More of a sirup simulator.
Hello BrotSagtMist, what graphics card do you have? This simulation is run 100% on the GPU. I didn't consider the performance at first, because I am using an NVidia RTX 2080, but I now made a lighter version.
I am uploading a lighter version where I reduced the fluid calculation resolution to half (heightmap remains at full resolution). I also increased the timestep as much as I could without ruining the stability, then I reduced the calculations from 24 to 4 simulation steps per game tick, so this should now run 24 times faster.
The OP's version works at 5 FPS in my 10 year old desktop (nVIDIA), at 36 FPS in a mini-PC (Intel), and at 120 FPS in a gaming laptop (nVIDIA), all with vsync on. It seems very sensitive to something, not sure what; maybe the graphics driver abilities to optimize shader code.
I wonder if the performance drop could be because I used rgba32f canvases for the fluid sim. I reduced the bit depth to 16 bit and it still seems to run just fine.
That helped! Now it runs at 30 FPS on the desktop, at 60 FPS on the mini (i.e. it's maxed out), and still at 120 FPS on the laptop (which is also maxed out).
, thanks for your valuable testing.