乱序处理器指令吞吐量平稳化的动态调节方法研究

发布时间：2018-04-05 14:38

本文选题：嵌入式实时系统　切入点：乱序处理器　出处：《华中科技大学》2012年硕士论文

【摘要】：嵌入式多媒体流计算具有软实时特性，对其进行自适应调节要求计算过程具有可预测性。乱序处理器通过在一个时钟周期内发射多条指令以乱序方式执行，导致计算时间难以预测，限制了其在嵌入式实时系统中的应用。因此迫切需要发展对乱序处理器指令吞吐量进行动态平稳化的方法，以提高程序执行时间的可预测性。在分析固定系数比例积分微分（PID）动态平稳化乱序处理器指令吞吐量局限性的基础上，提出了基于变速积分PID的乱序处理器指令吞吐量动态调节方法。通过分析乱序处理器的执行机制及其指令吞吐量变化特征，，发现乱序处理器在提高处理器指令执行效率和资源利用率的同时，也使得其执行指令过程变得复杂和不可预测。其指令吞吐量具有高度可变性，指令执行具有很大的不确定性。指令吞吐量最大值最高可达到最小值的4倍之多。而且，在程序实际运行时，指令吞吐量发生突变的情形相当普遍。基于固定系数PID反馈的调节方法虽然可以平稳化程序缓变时的指令吞吐量，但是，当程序发生突变时，由于固定系数PID反馈控制器的积分系数对系统误差的累积太大，导致系统在目标值附近振荡而不收敛。针对基于固定系数PID控制器的局限性，提出了变速积分PID控制算法。其能根据系统偏差的值动态调节积分系数，从而得到不同的PID控制量，通过动态电压频率调节（DVFS）机制，来平稳化非平稳变化程序的指令吞吐量。实验结果表明，变速积分PID控制算法不仅能调节程序平稳变化的指令吞吐量，对非平稳化变化的指令吞吐量，在经过一段时间的调节后，也能较快地收敛于目标吞吐量，从而实现指令吞吐量的平稳化，具有较好的调节效果。通过平稳化乱序处理器的指令吞吐量，可以更好地预测应用程序的执行时间。这无论是对于保证系统的实时性要求，还是提高系统自适应调节的有效性，都具有重要的潜在应用价值。
[Abstract]:Embedded multimedia stream computing has the characteristic of soft real time, and the calculation process is required to be predictable when it is adjusted adaptively.In a single clock cycle, the scrambled processor transmits multiple instructions to execute in a scrambled manner, which makes the computing time difficult to predict and limits its application in embedded real-time systems.Therefore, there is an urgent need to develop a method of dynamically stabilizing the instruction throughput of scrambled processors in order to improve the predictability of program execution time.On the basis of analyzing the limitation of instruction throughput of fixed coefficient proportional integral differential (PD) processor, a dynamic adjustment method of instruction throughput based on variable speed integral PID is proposed.By analyzing the execution mechanism of the scrambled processor and the characteristics of its instruction throughput, it is found that the scrambled processor not only improves the efficiency of instruction execution and resource utilization, but also makes the instruction execution process complex and unpredictable.Its instruction throughput is highly variable and instruction execution is uncertain.The maximum instruction throughput can be as many as four times the minimum.Moreover, when the program actually runs, the instruction throughput mutation situation is quite common.Although the regulation method based on fixed coefficient PID feedback can stabilize the instruction throughput when the program changes slowly, but when the program changes suddenly, because the integral coefficient of the fixed coefficient PID feedback controller accumulates too much system error,The system oscillates and does not converge near the target value.Aiming at the limitation of PID controller based on fixed coefficient, a variable speed integral PID control algorithm is proposed.It can dynamically adjust the integral coefficient according to the value of the system deviation, and obtain different PID control quantities. By adjusting the dynamic voltage frequency to adjust the VFS mechanism, the instruction throughput of the non-stationary program can be stabilized.The experimental results show that the variable-speed integral PID control algorithm can not only adjust the variable instruction throughput of the program, but also converge to the target throughput after a period of time adjustment.Thus, the instruction throughput is smooth and has good adjustment effect.The execution time of the application can be predicted better by stabilizing the instruction throughput of the scrambled processor.It has important potential application value both to guarantee the real-time requirement of the system and to improve the effectiveness of the adaptive regulation of the system.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP332

【参考文献】