A Tightly Coupled Heterogeneous Core with Highly Efficient Low-Power Mode

A Tightly Coupled Heterogeneous Core with Highly Efficient Low-Power Mode
复制标题

DOI:
10.1007/978-3-319-77610-1_16
复制
发表时间:
2018-04
期刊:
--
影响因子:
--
通讯作者:
Yasumasa Chidai;Kojiro Izuoka;Ryota Shioya;M. Goshima;H. Ando
Yasumasa Chidai;Kojiro Izuoka;Ryota Shioya;M. Goshima;H. Ando
中科院分区:
其他
文献类型:
--
作者:
Yasumasa Chidai;Kojiro Izuoka;Ryota Shioya;M. Goshima;H. Ando

文献摘要

相似文献

紧耦合异构核(TCHC)在核内具有不同特性的异构执行单元。复合核心(CC)和前端执行架构(FXA)是最先进的TCHC的示例。这些TCHC在核心中具有有序和无序执行单元。它们有选择地按顺序执行指令,与乱序执行相比,它提高了能效,而不会显着降低性能。然而,这些TCHC不能充分提高能量效率。CC具有大的执行单元切换代价,因此,CC不能充分地按顺序执行指令。当FXA按顺序执行指令时,它不能挂起消耗能量的乱序执行单元。在FXA的基础上提出了一种双模前端执行架构(DM-FXA)。DM-FXA具有我们提出的低功耗执行模式,该模式在有序执行时完全挂起乱序执行单元,因此,DM-FXA消耗的能量比FXA少。此外,DM-FXA具有比CC更小的切换代价。在我们的评估中,所提出的方法减少了34.7%的能耗相比,传统的无序处理器,性能下降是在3.2%。
A tightly coupled heterogeneous core (TCHC) has heterogeneous execution units with different characteristics inside the core. The composite core (CC) and the front-end execution architecture (FXA) are examples of state-of-the-art TCHCs. These TCHCs have in-order and out-of-order execution units in the core. They selectively execute instructions in-order and it improves the energy efficiency without significant performance degradation compared to out-of-order execution. However, these TCHCs cannot improve the energy efficiency sufficiently. CC has a large switching penalty of the execution units, and thus, CC cannot sufficiently execute instructions in-order. FXA cannot suspend energy consuming out-of-order execution units when it executes instructions in-order. We propose a dual-mode frontend execution architecture (DM-FXA), which is based on the FXA. DM-FXA has our proposed low-power execution mode, which completely suspends the out-of-order execution unit on in-order execution, and thus, DM-FXA consumes less energy than does the FXA. In addition, DM-FXA has a smaller switching penalty than CC. In our evaluation, the proposed methods reduce energy consumption by 34.7% compared with a conventional out-of-order processor, and performance degradation is within 3.2%.