One should add parallel processing support to solver iteration steps.
The primary goal should be to use OpenMP for multiple threads at CPU. This should be simpler in terms of code adaptation, and give a good understanding at parallelism within the game.
However, the main goal is actually to add support for CUDA kernels when processing the data. In fact, we shall try to move the whole memory data and processing to GPU. Handling for CPU only the calls divergence, agents/connection handling and potential outputs.
One should add parallel processing support to solver iteration steps.
The primary goal should be to use OpenMP for multiple threads at CPU. This should be simpler in terms of code adaptation, and give a good understanding at parallelism within the game.
However, the main goal is actually to add support for CUDA kernels when processing the data. In fact, we shall try to move the whole memory data and processing to GPU. Handling for CPU only the calls divergence, agents/connection handling and potential outputs.