https://hgpu.org/?p=10566
A streaming model for nested data parallelism