Automatic Optimization of Thread Mapping for a GPGPU Programming Framework