论文部分内容阅读
为了提高软硬件划分方法的效率,针对已有遗传算法求解软硬件划分没有结合特定问题处理、不满足约束个体的不足,提出一种混合并行的两步调整遗传算法.采用两步调整策略将不满足约束的个体转换为可行个体,当提高方法的运行效率时,图形处理单元用于计算每个个体的硬件耗费、软件耗费和通信耗费,多核CPU(中央处理器)用于并行执行个体间的调整,流并发传输策略进一步减少CPU和GPU(图形处理器)之间的传输开销.在基准数据集上,与求解该问题的已有方法相比,运行时间和求解质量都有明显优势.实验结果验证了该方法的有效性和合理性.
In order to improve the efficiency of the hardware and software partitioning methods, a hybrid genetic algorithm with two-step adjustment is proposed to solve the problem that the existing hardware and software partition of the genetic algorithm does not solve the problem of dealing with specific problems and not satisfy the constraints of individuals. When the method satisfies the operational efficiency of the method, the GPU is used to calculate hardware consumption, software consumption and communication consumption of each individual, and the multi-core CPU (central processing unit) is used to execute in parallel between individuals Adjustments, streaming concurrency, and transmission strategies further reduce the overhead of transmission between the CPU and the GPU (graphics processor), which has obvious advantages over run-time and solution quality on benchmarks datasets compared to the existing methods for solving this problem. The result verifies the validity and rationality of this method.