https://hgpu.org/?p=12945
The Distribution of OpenCL Kernel Execution Across Multiple Devices