MLS-MPM vs. APIC, I guess… From me understanding these methods are the best alternatives I have, for real-time fluid simulation in a game that needs to run well. The game itself will be quite simple, the fluid simulation is what makes it special. Being the most important feature and at the same time the hardest feature to implement, this decision gets quite hard for me. Would a method like APIC make more sense to use for the case of fluid simulation? Performance is super important, but at the same time the fluid simulation needs to be able to react to collisions and to interact with objects.
So in other words, which would perform better for fluid simulations, and which would be better for this use case (collisions and interacting with other objects)?
In my opinion, SPH method can be the best for your need: no grid therefore naturally no boundary, extremely fast for small liquids, although not so accurate than APIC or MPM, it is used frequently in real-time games. And a simple hash implementation for neighbor lookup can reduce O(n2) to O(n), which i heard taichi sparse structures supports too.
In this case a boundary is what I want. I would like high accuracy and it being able to interact properly to rigidbodies and being mixed with fluids.
The game is going to be a puzzle game revolving around fluids. Players will be put in fantasy environments, many of which being inspired by John Lin’s upcoming sandbox game (“John Lin” on youtube).
Players can create and and remove liquid, and there will already be liquid spread over the worlds.
The boundary would then be used to prevent liquid from leaving the world.