Maximizing Parallelism and GPU Utilization For Direct GPU Compilation Through Ensemble Execution