...
首页> 外文期刊>電子情報通信学会技術研究報告. VLSI設計技術. VLSI Design Technologies >A SIMD operation optimization algorithm in HW/SW partitioning for SIMD processor cores
【24h】

A SIMD operation optimization algorithm in HW/SW partitioning for SIMD processor cores

机译:SIMD处理器内核在硬件/软件分区中的SIMD操作优化算法

获取原文
获取原文并翻译 | 示例

摘要

A SIMD operation is composed of an arithmetic operation, a shift operation, and a saturation operation. They are called suboperations. By executing all the suboperations in a SIMD operation in one clock cycle, we can execute an application in a short time, but its corresponding functional unit in a processor core needs a complex configuration. By assigning all suboperations to their respective functional unit, we can simplify a configuration of a SIMD functional unit, but it takes much execution time. This paper proposes a SIMD operation optimization algorithm in HW/SW partitioning for SIMD processor cores. Given an application program and a timing constraint, we assume an initial processor core which executes all the suboperations in each SIMD operation in one clock cycle. Then, the algorithm divides a complex SIMD operation into two arithmetic operations with the bit extending operation and one bit extracting operation with the shift operation one by one. By repeating this process while a timing constraint is satisfied, a total cost of a processor core can be reduced. The experimental results show effectiveness of the algorithm.
机译:SIMD运算由算术运算,移位运算和饱和运算组成。它们称为子操作。通过在一个时钟周期内执行SIMD操作中的所有子操作,我们可以在很短的时间内执行一个应用程序,但是它在处理器内核中的对应功能单元需要复杂的配置。通过将所有子操作分配给它们各自的功能单元,我们可以简化SIMD功能单元的配置,但是会花费很多执行时间。本文针对SIMD处理器内核,提出了一种在硬件/软件分区中的SIMD操作优化算法。给定一个应用程序和一个时序约束,我们假设一个初始处理器内核在一个时钟周期内执行每个SIMD操作中的所有子操作。然后,该算法将一个复杂的SIMD运算分为两个算术运算,一个是位扩展运算,另一个是一个比特提取运算,一个移位运算。通过在满足时序约束的同时重复此过程,可以降低处理器内核的总成本。实验结果表明了该算法的有效性。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号