Improved Programming of GPU Architectures through Automated Data Allocation and Loop Restructuring