https://hgpu.org/?p=14696
Accelerating Applications with Pattern-specific Optimizations on Accelerators and Coprocessors