Applying software-managed caching and CPU/GPU task scheduling for accelerating dynamic workloads