Is the OpenCL GPU Solver in bullet3 deterministic? (I am assuming convex rigid body simulation) https://github.com/erwincoumans/bullet3 ... 354eacf09c would seem to suggest that it is, but there are multiple levels here:
1) Does the GPU solver give the same result for a simulation started the same way each time
2) Does the GPU solver give the same result as the CPU solver for two simulations started the same way
3) Is it possible yet for simulations not started at the same time to be deterministic? It was mentioned previously this would be a feature in bullet 3 but I'm not sure if this has been done yet, and whether it applies to the GPU solver (e.g. a simulation has been running for 100 frames, the simulation state is saved and then recreated elsewhere, will the two simulations give the same result from then on?)
If these are not the case, are there technical reasons why they cannot be or is it just not implemented yet?
Any insight would be greatly appreciated.