Hi there,

I've run into a roadblock with my programming and I cannot continue until I've found a solution to the problem. The problem involves PhysX and executing custom code entirely on the GPU.

Here is a scenario similar to the one I'm faced with:

Imagine you wanted to create a simple particle simulation with particles bouncing off of in-game surfaces and such. For each physics cycle, you need the GPU to calculate the average position of all particles at the end of the cycle.

Is it possible to have something like this entirely performed on the GPU, even without using the CPU as a middle-man to instruct the GPU to calculate the average position?

My first idea would be to create a custom class for PhysX which integrates into the PhysX pipeline. This would likely require a DLL and a pointer to the PhysX interface used by 3DGS, through which I can customize the PhysX pipeline in C++.

But if this is not possible, my second thought was to perhaps write the customized code for CUDA, and get PhysX to cooperate with CUDA through a DLL.

I'm not sure. Have any pointers? Any help is -strongly- appreciated. Thank you.

-Steve