High Performance Communication Middleware With On-The-Fly Gpu-Based Compression For Hpc And Deep Learning Applications