https://hgpu.org/?p=26502
Data transfer optimizations for heterogeneous managed runtime systems