Speculation Control Execution for Energy Efficiency

现代处理器通过推测来提高指令级并行性。数据和控制决策的结果被预测，操作被推测性地执行，只有在原始预测正确的情况下才被提交。还有许多其他方法可以使用处理器资源，例如线程化或紧急执行。随着投机使用的增加，更多的处理器将需要某种形式的投机控制，以平衡投机的好处和其他可能的活动。这项研究考察了一组可以从推测控制中受益的微体系结构特征，包括轨迹形成、指令获取管理、执行资源管理和内存访问优先级。这项研究还探索了无序处理器设计如何使用多个电压电平来降低能量，以及处理器如何向操作系统提供反馈，从而通过动态电压调整进一步控制能源使用。这项工作是使用支持多线程和多处理器模拟的微体系结构模型来完成的。这项工作是一项更广泛工作的一部分，其中包括对现有计算机进行实验，为提高能效设计和修改操作系统，以及采用旨在比较节能处理器设计的微体系结构模型。

Modern processors improve instruction level parallelism by speculation. The outcome of data and control decisions is predicted, and the operations are speculatively executed and only committed if the original predictions were correct. There are a number of other ways that processor resources could be used, such as threading or eager execution. As the use of speculation increases, more processors will need some form of speculation control to balance the benefits of speculation against other possible activities. This research examines a set of microarchitectural features that can benefit from speculation control, including trace formation; instruction fetch management, execution resource management and memory access prioritization. The research also explores how out-of-order processor designs can use multi-voltage levels to reduce energy, and how the processor can provide feedback to an operating system that further controls energy use via dynamic voltage scaling. This work is done using a microarchitectural model that supports multithreading and multiprocessor simulations. This work is part of a broader effort that includes experimentation with existing computers, design and modification of operating systems for energy efficiency and the adoption of microarchitectural models designed for comparing energy efficient processor designs.