Research on the mechanism for handling non-uniform GPGPU applicationsby data-driven matching

针对GPU处理非均匀应用的特点，提出了基于任务的细粒度执行机制。它可以平衡GPU内和GPU间的处理负载。实验结果表明，即使在线程内核中并行执行单元的粒度分布不均匀的情况下，该方法也能提高分配给GPGPU块的负载的总体平衡性。

Task-based fine-grain execution mechanism is proposed in order to use GPGPUs for non-uniform applications. It can balance the processing load on intra GPU and inter GPUs. We can find out that this method can improve total balanceof load which is assigned to the block of GPGPU, even when the grains of parallel execution unit in a thread kernel are not uniformly balanced.