Lecture Notes for Computer Architecture II - St. Cloud State University

More documents

Recommendations

Info

Lecture 31- Pipelining Chapter 6 Performance improvement of single cycle processor implementation was discussed in chapter 5 Chapter 3 Performance improvement of ALU components Adders Multipliers Dividers Floating point adders Considered Datapath Control path Page | 188 Chapter 5 Datapath Control path For single cycle pipelined processor Chapter 6 learning objectives: Understand how pipelining can improve performance. Understand pipeline problems and design of Forwarding, Hazard Detection units Explain branch prediction’s use in enhancing the pipeline performance Understand how speculative execution can improve performance. Describe unfolding in superscalar architectures Understand the importance of program correctness when executing instructions out-of-order. Compare Superscalar, VLIW and EPIC architectures Pipelining Improves performance By increasing instruction throughput Not by decreasing execution time of an individual instruction as we learned in chapter 2 In chapter 2 performance = 1/CPU time No change in execution time per instruction In general Speed up = number of stages in pipeline Pipelining will improve performance if there are No data dependencies between two instructions No branch instructions No exceptions
Problems with pipelining Hazards Data Branch Data Hazards Solutions forwarding stalls Branch hazards Prediction Exceptions Page | 189 Superscalar and dynamic pipelining Pipelined data path Divide the instruction execution into 5 stages 5 stage pipelining for MIPS instruction set Introduce pipeline registers to store intermediate results in between stages 5 stages for instruction execution instruction fetch IF Instruction decode/Register Fetch ID Instruction execution EX Memory access MEM Write back WB IF PC INST Memory ID EX Register file ALU MEM Data Memory WB Register file
Page 1 and 2:
Page | 1 Lecture Notes for Computer
Page 3 and 4:
knowledge given in the book chapter
Page 5 and 6:
CS 320 SCHEDULE (tentative) Week Wh
Page 7 and 8:
6. Project reports - Presentation C
Page 9 and 10:
9. Course policies: 1) Satisfactory
Page 11 and 12:
Size - keep shrinking 1994 124mm 2
Page 13 and 14:
Families of different models - Inte
Page 15 and 16:
Super Computers CRAY - 1 st vector
Page 17 and 18:
Fastest super computers November 20
Page 19 and 20:
Class Activity - 2. Trace the conte
Page 21 and 22:
TUTOR 1.32:> PHYSICAL ADDRESS=00002
Page 23 and 24:
Page 25 and 26:
Class Activity 3 - Study of instruc
Page 27 and 28:
Class Activity 4 - Study of instruc
Page 29 and 30:
Page 31 and 32:
Page 33 and 34:
Performance Improvement at Instruct
Page 35 and 36:
5. Register indirect addressing mod
Page 37 and 38:
High Level Language Statements Ass
Page 39 and 40:
Lecture 7 Data transfer memory reg
Page 41 and 42:
Problem Translate HLL statement A[
Page 43 and 44:
More MIPS instructions jump j jr $S
Page 45 and 46:
Problem Translate the following Whi
Page 47 and 48:
Problem Translate the following exp
Page 49 and 50:
Problem Translate the following exp
Page 51 and 52:
Lecture 10 Translation of High Leve
Page 53 and 54:
Translation of Procedures Problem H
Page 55 and 56:
Example Write the MIPS code to comp
Page 57 and 58:
Program Counter -relative op rs rt
Page 59 and 60:
Example Opcode always 1 byte Memory
Page 61 and 62:
Lecture 13 Enhancing Performance at
Page 63 and 64:
Number Systems Numbers + ve numbers
Page 65 and 66:
Example: Design a one bit ALU with
Page 67 and 68:
Lecture 14 - Performance Improvemen
Page 69 and 70:
Design of Adders c 1 a n , ..., a 1
Page 71 and 72:
Carry lookahead adder Problem with
Page 73 and 74:
Another look at CLA xi + yi = Pi xi
Page 75 and 76:
4-bit CLA Page | 75 generates C3 ge
Page 77 and 78:
Lecture 15 Performance Improvement
Page 79 and 80:
Example 981 X 1234 0981 X 1234 mult
Page 81 and 82:
Example Consider the multiplication
Page 83 and 84:
2 nd version Reduce cost by reducin
Page 85 and 86:
4th step 4 bits registers 0010 4 bi
Page 87 and 88:
0010 ADDER Page | 87 0001 0010 ADDE
Page 89 and 90:
2's complement of m'cand 1111110101
Page 91 and 92:
0010 ADDER Page | 91 1111 0001 1 no
Page 93 and 94:
Lecture 16 - Performance Improvemen
Page 95 and 96:
Hardware units needed to perform th
Page 97 and 98:
Step 1 Rem = Rem - Divisor ( this t
Page 99 and 100:
5th step Rem = Rem - Divisor Now Re
Page 101 and 102:
2nd step Shift remainder left Rem =
Page 103 and 104:
3 rd version of Division algorithm
Page 105 and 106:
Step 3 Rem = Rem - Div Rem > 0 Shif
Page 107 and 108:
Normalized Form keeps one digit lef
Page 109 and 110:
Let us consider the addition of two
Page 111 and 112:
Control path Inputs Exponent differ
Page 113 and 114:
Lecture 19 Architectures for Crypto
Page 115 and 116:
1. Following pseudo code represents
Page 117 and 118:
Find 7 36 mod 11 using right to lef
Page 119 and 120:
Lecture 20 - Performance Chapter -
Page 121 and 122:
total amount of work done in a give
Page 123 and 124:
Question 4 Assuming the CPI for pro
Page 125 and 126:
Question 9 Yet another user has the
Page 127 and 128:
Question 12 Assuming the CPI values
Page 129 and 130:
Use C2 CPI on M1 = 4*0.3+6*0.2+8*0.
Page 131 and 132:
Question 15 We are interested in tw
Page 133 and 134:
Question 18 You are the lead design
Page 135 and 136:
Arithmetic Logic unit Register file
Page 137 and 138: Lecture 22 - Processor Design - Dat
Page 139 and 140: Example Design the data path to fet
Page 141 and 142: Now combine two functional units to
Page 143 and 144: Lecture 23 Processor Design - Data
Page 145 and 146: Example : - Design DataPath for SW
Page 147 and 148: Processor Design - Data path for R&
Page 149 and 150: Example : - Implement R+ I +J Forma
Page 151 and 152: 31 26 0 4 control signals EX/Addres
Page 153 and 154: Draw the logic circuit _ _ _ _ _ _
Page 155 and 156: Single cycle implementation of the
Page 157 and 158: from Aluop B from output PC Data Ad
Page 159 and 160: Lecture 27 High level view of the F
Page 161 and 162: 0 Page | 161 1 2 3 5 4 0 - MemRead
Page 163 and 164: FSM Controller for lw/sw 6 states 0
Page 165 and 166: Lecture 28 R-type instruction conti
Page 167 and 168: Branch Instruction Fetch Page | 167
Page 169 and 170: Jump Inst Fetch Page | 169 Inst dec
Page 171 and 172: Page | 171
Page 173 and 174: 0 1 Page | 173 2 6 8 9 3 5 7 4 op 0
Page 175 and 176: NS0 = 1 + 3 + 5 + 7 + 9 NS1 = 2 + 3
Page 177 and 178: Lecture 30 Microprogramming - FSM c
Page 179 and 180: - Microinstruction placed in a ROM
Page 181 and 182: - In multicycle design - assume tha
Page 183 and 184: Multi-cycle implementation of the M
Page 185 and 186: 2.5 Draw the functional units neces
Page 187: 2.9 The instruction, addi (add imme
Page 191 and 192: Ex: Represent the 5 stage pipeline
Page 193 and 194: Example Represent the 5 stage pipel
Page 195 and 196: Example: Find the size of each pipe
Page 197 and 198: Pipelined control path - Single cyc
Page 199 and 200: Control signals with instructions E
Page 201 and 202: Ex: Show the control signals genera
Page 203 and 204: Ex: Show the control signals genera
Page 205 and 206: Lecture 33 Forwarding Unit Design E
Page 207 and 208: Example: Illustrate the forwarding
Page 209 and 210: Design a Forwarding unit to integra
Page 211 and 212: Ex: Forwarding with 2 inst. (1) add
Page 213 and 214: Lecture 34 Data hazard and stalls F
Page 215 and 216: Tomasulo • Lasting Contributions
Page 217 and 218: Tomasulo Organization Page | 217
Page 219 and 220: 4. Show the pipeline execution of f
Page 221 and 222: 7. Illustrate the pipelined data pa
Page 223 and 224: 11. Reorder the following instructi
Page 225 and 226: 15. Design the forwarding unit for
Page 227 and 228: Temporal locality Low Data Accessin
Page 229 and 230: Associative mapping More flexible m
Page 231 and 232: Problems 1. Here is a series of add
Page 233 and 234: 6. What are the miss rate of the fo
Page 235 and 236: Installation and Usage of Maxplus I
Page 237 and 238: intend to read registers this instr
Page 239 and 240:
Tomasulo hardware algorithm From Wi
Page 241:
When all operands are available o I
show all

Lecture Notes for Computer Architecture II - St. Cloud State University

Create successful ePaper yourself

Delete template?

Save as template?