Many-Thread Aware Prefetching Mechanisms for GPGPU Applications