https://hgpu.org/?p=2320
Revisiting sorting for GPGPU stream architectures