Microarchitecture. Outline Architecture vs. Microarchitecture Components MIPS Datapath 1.

Microarchitecture

2

Outline

• Architecture vs. Microarchitecture • Components• MIPS Datapath

3

Architecture vs. Microarchitecture

• Architecture• What the CPU does

• Microarchitecture• How the CPU does it

4

Architecture

• What the CPU does

• CPU is a black box• Only care about what goes in / comes out

• Only need to know what instructions the CPU can execute• Instruction set architecture (ISA)

• Realm of computer scientists• Don’t really care how it works under the hood

5

Microarchitecture

• How the CPU does it

• Electrical mechanisms that allow CPU to execute instructions• Control paths, adder(s), multiplexors, memory units, etc.

• Many possible microarchitectures for single architecture

• Domain of computer and electrical engineers

6


• Analogy: factory that produces auto parts

7


• Architecture: don’t care about “how” factory runs, as long as the parts are good

8


• Microarchitecture: care about “how” the factory runs

9


• Microarchitecture: care about “how” the factory runs

10


Intel Pentium IV AMD Bobcat

x86 Architecture

11

Outline


12

Von Neumann Model (basic structure of computer)

MEMORY

INPUT OUTPUT

PROCESSING UNIT

CONTROL UNIT

PC IR

MAR MDR

TEMPALU

13

Components: Memory

MEMORY

INPUT OUTPUT

PROCESSING UNIT

CONTROL UNIT

PC IR

MAR MDR

TEMPALU

14

Components: Memory

• 2k × m array of stored bits• E.g., 24 × 8

• Address• Unique (k-bit) identifier of location

• Contents• m-bit value stored in location

• Basic operations• STORE: write a value to memory location• LOAD: read a value from memory location

MEMORY

MAR MDR

0000

0001

0010

0110 1111

0011

…

1101

1110

1111

0001 0110

15

• Result of “LOAD 0010”?

Components: Memory MEMORY

MAR MDR

0000

0001

0010

0110 1111

0011

…

1101

1110

1111

0001 0110

16

• Result of “LOAD 0010”? 0110 1111

Components: Memory

0000

0001

0010

0110 1111

0011

…

1101

1110

1111

0001 0110

MEMORY

MAR MDR

17

Components: Memory

• How to read data to/from memory?• MAR: Memory Address Register• MDR: Memory Data Register

• LOAD data from memory (address A)1. Write the address (A) into MAR2. Send a “read” signal to memory unit

(memory unit will go get bytes, put into MDR)

3. Read the data from MDR

MEMORY

MAR MDR

18

Components: Memory

• How to read data to/from memory?• MAR: Memory Address Register• MDR: Memory Data Register

• STORE data X in memory B1. Write the data (X) into MDR2. Write the address (B) into MAR3. Send a “write” signal to memory

(memory unit will do its thing to write X in location B)

MEMORY

MAR MDR

19

Components: Memory

• What is controlling Memory?• The CONTROL UNIT of course!!

• CONTROL UNIT sends signals • Instructs the other components

• How does it know what signals (instructions) to send?

MEMORY

CONTROL UNIT

PC IR

MAR MDR

20

Components

MEMORY

PROCESSING UNIT

CONTROL UNIT

PC IR

MAR MDR

TEMPALU

21

CONTROL UNITComponents: Control Unit

• How does CONTROL UNIT know what to do?• Well, it doesn’t “know”..

• Control Unit follows command in the Instruction Register (IR)

• Program Counter (PC) register holds memory address of the current instruction

• What types of instructions do you think are available?• LOAD? STORE? ADD? SUBTRACT? OTHERS?

PC IR

22


• Control Unit decodes instruction stored in IR

• Sends signals to components based on decoded instruction

• For example, perhaps the following instruction is in the IR

PC IR

0 0 0 1 1 0 0 1

23


• In this simplified example, the instruction is decoded as:

PC IR

0 0 0 1 1 0 0 1

first two bits:00 means

LOAD

next four bits:memory address

to read from (0110)

last two bits:01 refer to

MDR

24


• Based on this instruction, the Control Unit will activate the corresponding circuitry to signal the memory unit to get data• This circuitry goes beyond the scope of this course

• Consists of finite state machines of multiplexors, system clock, bus etc.

PC IR

0 0 0 1 1 0 0 1

first two bits:00 means

LOADnext four bits:

memory address to read from

(0110)

last two bits:01 refer to

MDR

25


• If the instruction in the IR contains logic or arithmetic, the Processing Unit gets involved

PC IR

26

Components

MEMORY

PROCESSING UNIT

CONTROL UNIT

PC IR

MAR MDR

TEMPALU

27

Components: Processing Unit

• Contains ALU• Arithmetic Logic Unit

• Also contains temporary registers (TEMP)• Used for arithmetic operands• Used to store intermediate results of arithmetic operations

• May also contain other functional units• Special purpose arithmetic: e.g., multiply, divide, etc.

PROCESSING UNIT

TEMPALU

28


• How does Processing Unit know what to do?

• Again, Control Unit sends electric signals that correspond to the appropriate instruction (e.g., ADD, SUBTRACT)

PROCESSING UNIT

TEMPALU

CONTROL UNIT

PC IR

29


• Processing Unit gets data from Memory Unit• Processing Unit puts data into Memory Unit

MEMORY

PROCESSING UNIT

MAR MDR

TEMPALU

30

Components: Processing Unit ALU

• ALU has its own inputs and output• A is m-bit binary (input)• B is m-bit binary (input)• control tells ALU what to do (input)

• E.g., add, subtract, divide, etc.• Usually control pins (high / low)

• C is m-bit result (output)

PROCESSING UNIT

TEMPALU

control

A B

C

31

Break Time!!

... wish I was there!!!

32

Outline


33

MIPS Datapath

• MIPS• Microprocessor without Interlocked Pipeline Stages• RISC Architecture: Reduced Instruction Set Computer

• Microarchitecture of single-cycle MIPS processor…

34

MIPS Single-Cycle Microarchitecture

• We are going to “build” a single-cycle MIPS CPU

35


• System clock• An oscillator that produces a square wave• E.g., 2.4 GHz processor- means 2.4 billion square waves per second• Components of CPU (memory, ALU, bus, etc.) synchronized to clock

• Actions occur on “rising” or “falling” edge

36


• Single-cycle CPU: • Executes an entire instruction in one clock cycle• Clock cycle time limited by slowest instruction

one cycle

37


• Overview of components used in diagram

38


• PC: Program Counter• 32-bit register• Input (PC’): next instruction• Output (PC): current instruction• CLK: system clock

39


• Instruction Memory• Single read port• Input: 32-bit instruction address (A)• Output: 32-bit data (i.e., instruction) (RD)

• Note: this is an oversimplification that assumes the instruction memory is ROM (read only memory)• In reality, the instruction memory is writable; the OS can load new

programs into memory

40


• Register File (32 registers, 32-bits each)• Inputs:

• A1: 5-bit address specifying a register (to read)• A2: 5-bit address specifying a register (to read)• A3: 5-bit address specifying a register (to write)• WD3: 32-bit data to write to register• CLK: system clock• WE3: write enable (single bit)

• Outputs:• RD1: 32-bit register value specified by 5-bit address A1• RD2: 32-bit register value specified by 5-bit address A2

41


• Data Memory• If WE (Write Enable) is 1,

• Writes data WD into address A on rising edge of clock (CLK)

• If WE is 0,• Reads address A onto RD on rising edge of clock (CLK)

42


• Start by building the datapath• First step, connect PC to Instruction Memory

• PC register contains address of instruction• Instruction Memory fetches instruction (instr) from memory

PC instr

43


44


• Next step, things start to get more complicated• The instruction (instr) gets decoded by the control unit

• The appropriates pins (e.g., write enable) get set• The 5-bit register addresses (in the Register File) get specified

• (Things get “set up” for execution)

45


46


• Next, execution occurs (based on instruction)

47


• Then, the result of execution interacts with memory (somehow)

48


• The process then starts over…

49


• Let’s look at the datapath for a concrete example• MIPS assembly instruction lw (stands for “Load Word”)

• lw loads 32-bits of data from a memory address to a register

• Please keep in mind that we will NOT use MIPS assembly, we will use x86• This is for demonstration purposes only, x86 datapath is more complex

50

MIPS Single-Cycle Microarchitecture: lw example

• Step 1: fetch the instruction

51


• Step 2: decode the instruction• In this case, read source operands from register file

• Source operands come from the instruction itself• Says what registers to get information from

52


• Step 3: sign-extend the immediate result (decode)• This has to do with the “offset” of the memory address• This is a mechanism to make the “offset” 32-bits long • Don’t worry about this..

53


• Step 4: Compute the memory address (execute)• ALUControl2:0 is from the control unit, tells ALU to add• SrcA is “base”• SrcB is “offset”

54


• Step 5: Read data from memory, write back to Register File• RegWrite is command from Control Unit• 20:16 is part of instruction: 5-bit register address to load word

55


• Step 6: Increment the Program Counter (PC)• PC will point to next instruction• 32-bit instruction width = 4 bytes (hence PC + 4)

56


• The datapath depends on the instruction…

57

Next Time…

• Execution Cycle• Big vs. Little Endian• CPU execution time..

Microarchitecture. Outline Architecture vs. Microarchitecture Components MIPS Datapath 1.

Documents

data tofrom memory

memory bwrite

memory units

memory locationload

memory address registermdr

memory unitmemory unit

memory data register

memory address awrite