Pymunk multi-core collision resolution?

Question

I have a very basic space set up with a large number (10000) of circles falling under gravity. It seems that only one of my cores is active at all at any point in time, but as far as I know collision resolution is extremely parallelizable with space partitioning, which the Chipmunk engine does indeed implement.

It seems that even when I create the space with threaded=True on Linux, and space.use_spatial_hash(), the issue persists. I really need the extra performance parallel threads could provide, what are my options? Should I move to pybullet instead with planar constraints to emulate 2d?

Answer 1

You need to set the number of thread to use as well, for example like your_space.threads = 2

In the docs its written like this (Space.init):

If you set threaded=True the step function will run in threaded mode which might give a speedup. Note that even when you set threaded=True you still have to set Space.threads=2 to actually use more than one thread.

You can verify that it works by running the threaded_space example python -m pymunk.examples.threaded_space. If I do that and watch CPU usage in htop I can easily see when it goes from 1 to 2 threads.

I should also add that unfortunately the multi-threading does not improve performance much in most cases. It can only help if there's a lot of collisions going on at the same time, but even then its not much (Maybe on ARM it gives bigger improvement, but that I cannot try).