https://hgpu.org/?p=10753
Efficient fine grained shared buffer management for multiple OpenCL devices