https://hgpu.org/?p=13221
Compiler-Level Explicit Cache for a GPGPU Programming Framework