https://hgpu.org/?p=7100
GMP implementation on CUDA - A Backward Compatible Design With Performance Tuning