Domain decomposition 2D for GPUs and CPU - technology required MPI and CUDA

The program calculates the diffusion of liquids in 2D. It is written in CUDA and C . (here is repository: [login to view URL]) It is necessary to make a block decomposition, so that each block is calculated on a separate GPU (CUDA), and these in turn communicate with each other through the CPU (MPI - just two commands for exchange data, and sending the data back to gpu). Data should be included and transmitted in the created halo cells. Preferred system: Linux The program is easy to modify. It's so brief. I invite you to cooperation.