Evaluation and enhancement of memory efficiency targeting general-purpose computations on scalable data-parallel GPU architectures