gregtj / stable-fluids Goto Github PK

In the original paper a delta-time parameter is used, this allows duration to be expressed in seconds rather that frames, and also the viscosity and velocities and other vectors can have more sane units. In my humble opinion pixels/frame^2 is trickier to think about compared to pixels/s^2... 🤔 Furthermore, after studying the code it should be fairly easy to add a delta time parameter to both advect_diffuse() and project() functions. Possibly with default value of dt=1.

The original paper uses a nifty set_bnd function that was very easy to extend that allowed boundary conditions to be set, e.g. adding walls to the box, preventing densities (and velocities) from escaping. Furthermore, interesting effects could be achieved by adding internal boundaries of various shapes. In this very beautiful and awesome re-implementation of the paper the boundary conditions is not possible to specify. After reading the code, it seems very doable. Perhaps you prefer to have it completely outside the Fluid class? 🤔

I just cloned and run this, installing latest numpy and scipy. I'm getting the following warning:

SparseEfficiencyWarning: splu requires CSC matrix format

Perhaps this blazing fast simulation can be speed up even further if whatever function is emitting this warning is given a sparse csc matrix instead?