Study of Bandwidth Partitioning for Co-executing GPU Kernels