体系A4.doc

计算机体系结构 A4 叶振宇 座机电话号码33 计算机设计基础 Computer Architecture plays an important role in performance improvement. Pipeline, dynamic schedueing, ooo, branch prediction, speculation, superscalar, VLIW, prediction Instructions and so on. 微处理器四个阶段：Microprocessors- Quantitative Architecture-Instruction-Level Parallelism-Thread-level/Data-level parallelism 体系的三个分类：Instruction set architecture-Microarchitecture-System Design ISA七个方面：Class of ISA-Memory addressing-Addressing modes-Types and sizes of operands-Operations-Control flow instructions-Encoding an ISA Three Wall：ILP（指令级并行）Memory Power 计算机分类：根据指令和数据流 SISD,MISD 少 ,SIMD llliac-IV CM-2 ,MIMD SPARCCenter T3D MIMD的两种编程模式：SPMD 单程序 ,MPMD 桌面计算-Optimized price- performance 服务器-dependability可靠scalability-efficient throughput 嵌入式Embedded-Real time-Strict resource constraints Register machine vs Stack machine vs Accumulater machine RISC vs. CISC 技术趋势：Cost decrease rate ~ density increase rate Technology improves continuously, an impact of this improvements can be in discrete leaps. 电能趋势：（电压低更省电性能没降多少，多核降电压提性能） 挑战：distributing the power-removing the heat-preventing hot spot Cost趋势：因素：time volume commodification Component direct（加工） indirect（渠道）ASP list price（仓储利润） 可靠性The quality of the delivered service such that quality of the delivered service such that reliance can justifiably be placed on this service. reliability, maintainability, availability 量度标准：MTTF MTTR FIT 1/MTTF MTBF MTTF+MTTR Availability MTTF/MTBF 解决：time/resource redundancy 测度性能： Wall-clock-time（response/elapse）唯一性能通用指标 CPU time 不含IO user + system Throughout：Amount of work done in a given time（带宽可能重要于延迟） Response升throughout升 换CPU Throughout升response不一定升 多CPU 同指令集对比 SPEC The System Performance Evaluation Cooperative Maximizing performance means minimizing response execution time 等价权重计算 1/ ti*SUM 1/tj Geometric mean does NOT predict run time 定量原则：利用流水线（最重要）流水线CPU（指令级） 组关联cache 流水功能部件（操作级） Locality：时空 Focus on the common case：（重要 普遍pervasive） Simple Is fast Am