| 引用本文: | 管茂林,杨乾明,张春元,等.面向能耗有效高性能嵌入式微处理器的VLIW调度.[J].国防科技大学学报,2012,34(6):26-33.[点击复制] |
| GUAN Maolin,YANG Qianming,ZHANG Chunyuan,et al.VLIW scheduling for high performance embedded energy-efficient processor[J].Journal of National University of Defense Technology,2012,34(6):26-33[点击复制] |
|
| |
|
|
| 本文已被：浏览 7687次 下载 6335次 |
| 面向能耗有效高性能嵌入式微处理器的VLIW调度 |
| 管茂林, 杨乾明, 张春元, 文梅 |
|
(国防科技大学 计算机学院,长沙 410073)
|
| 摘要: |
| 为了降低功耗，目前能耗有效的嵌入式微处理器一般都采用分布式与层次化的寄存器文件结构。第一层的超小寄存器文件（TORF）的极小容量使得很多数据必须存放到第二层的通用寄存器文件（GRF）中，这给编译器带来了新的挑战。通过分析程序特征，提出了新的VLIW调度算法，通过在编译时对变量进行检测，在恰当的时机插入虚拟的copy操作并进行指令与通信调度，为对寄存器需求较大的全局变量与软流水变量构建了新的包含GRF的数据传输路由，将对TORF的压力转移到GRF中。实验结果表明，新的VLIW调度算法符合处理器的设计初衷。与不使用GRF相比，在程序性能只降低约8%的情况下，降低了约51%的寄存器访问能耗，43%的处理器能耗。最关键的是避免了程序员手工分配优化的难题。 |
| 关键词: 能耗有效 分布式与层次化寄存器文件 VLIW调度 |
| DOI： |
| 投稿日期：2011-06-20 |
| 基金项目:国家自然科学基金资助项目(61033008, 60903041, 61103080)；国家部委资助项目；教育部博士点基金资助项目(20104307110002)；湖南省研究生科研创新项目(CX2010B028)；国防科学技术大学优秀研究生创新资助项目(B100603, B120605) |
|
| VLIW scheduling for high performance embedded energy-efficient processor |
| GUAN Maolin, YANG Qianming, ZHANG Chunyuan, WEN Mei |
|
(College of Computer, National University of Defense Technology, Changsha 410073, China)
|
| Abstract: |
| To reduce the power, the energy-efficient embedded microprocessor always adopts the distributed and hierarchical register file structure (DHRF). Many data need to be stored in the second level general register file (GRF) because of the small capacity of the tiny operand register file (TORF), and this challenges the design of compiler. A new VLIW scheduling algorithm is proposed to solve the problem through analyzing the program characteristics. The variables are detected while the compiling and virtual copy operations are inserted at the appropriate time. Through instruction scheduling and communication scheduling for the copy operations and constructing new data transfer route including GRF for the global variables and software pipelining variables, which have great demand for the register, the pressure on TORF is transferred to GRF. The experimental results show that the VLIW scheduling algorithm is consistent with the starting point of the energy-efficient microprocessor. On the condition of 8% program performance decline, the energy consumption on register accessing is reduced by about 51%, and the energy consumption of the processor is reduced by about 43%. At the same time, the burden of programmer is avoided. |
| Keywords: energy-efficient distributed and hierarchical register file VLIW scheduling |
|
|
|
|
|
