https://hgpu.org/?p=3493
High throughput multiple-precision GCD on the CUDA architecture