Optimized primitives for collective multi-GPU communication - View it on GitHub