https://hgpu.org/?p=16873
Batched Shift Reduce Parsing with Lists of Vectors on CUDA