Data parallel patterns on CPU/GPU mix

  We propose a model that uses a small set of quite simple parameters to devise a proper partitioning of the available data parallel tasks between CPU cores and GPU cores. The model takes into account both hardware and application dependent parameters. In particular, it eventually computes the percentage of tasks to be executed on CPU cores and GPU cores to achieve the better performance figures. Different experimental results on state-of-the-art CPU/GPU architectures are shown that assess the model properties.