I finally got graph coloring working properly in the solver with good performance. Now large piles can be solved on multiple threads in deterministic order. It really helps with stability. Big thanks to Erin Catto pointing me in the right direction!

@erin_catto Looks amazing! Is there any paper for graph coloring usage in physics solvers? I assume the graph itself connects all colliding bodies with what they collide, so you can process independent collision pairs in parallel?

@erin_catto Right here is all you need to know! The post is about SIMD, but exactly the same approach can be used (and is used in Box2D) for multithreading.

Setting Process Goals for Long-Term Success: A Guide to Achieving Better Results #business #success #selfhelp

@erin_catto Deterministic multi-threading of PHYSICS?

@erin_catto Digging in a bin of legos for that one piece

@erin_catto Wow awesome!! Is this using your own engine? Any recommendations for beginners trying to get into this area?

@erin_catto Yes, custom engine. I'd recommend the Box2D blog:

@erin_catto sweet! is it stateless?

@erin_catto That depends on how you define it. There is obviously quite a lot of state in each body (velocity, position, etc), but it does not have any pair-wise internal state, as most other physics engines do.

@erin_catto Very cool, just wondering, is this a step forward to real building physics or close to something similar? where one voxel doesn't hold a building. Also, if that came to be, I guess it would very much impact the story missions right?

@erin_catto That remains to be seen. I think it could be a step in that direction, at least for certain types of buildings. With this it's realistic to make buildings as jointed parts. Not quite the same as brittle fracture though, so more for wood/metal, rather than concrete/brick