Dynamic Block Size Adjustment and Workload Balancing Strategy Based on CPU-GPU Heterogeneous Platform

In recent years, with the development of processor architecture, heterogeneous processors including CPUs and GPUs have become mainstream processors. In order to take full advantage of the computing power of heterogeneous cores, it is necessary to maintain workload balance between heterogeneous cores while the system is executing applications. And it is also important to optimize the efficiency of the application execution on the GPU by improving the thread organization. In order to improve performance, we propose a block size adjustment strategy that adapts to the current application and GPU environment. Based on this, we propose a strategy for balancing CPU-GPU workload that preferentially protects the core. These strategies optimize the execution time of applications on CPU-GPU heterogeneous platforms. Finally, we tested the actual effect of the strategy by running four benchmarks in a CPU-GPU heterogeneous environment. The experimental results show that the performance can be significantly improved by block size adjustment and workload balancing strategy. These two strategies reduce application execution time by up to 26.21% and 58.01%, respectively, compared to GPU-only execution time.