https://hgpu.org/?p=8512
Exploiting Coarse-grained Parallelism in B+ Tree Searches on an APU