Designing Efficient MPI and UPC Runtime for Multicore Clusters with InfiniBand, Accelerators and Co-Processors