26
Computer Architecture

Architectural Developments

Embed Size (px)

Citation preview

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 1/26

Computer Architecture

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 2/26

Computer architecture

• design

• Refers to those attributes of thesystem visible to the programmer,that have a direct impact on thelogical execution of the program

• Hardware and software

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 3/26

attributes

• Instruction set

• Number of bits to represent datatypes

• I/O mechanisms

• Memory addressing

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 4/26

types

• Open architecture

 – Allows the system to easily beconnected to devices and programs

made by other manufacturer

• Closed architecture

 – Design is proprietary, make it difficult to

connect the system to other system

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 5/26

developments

• Clock speeds

 – 70’s - 1MHz

 – 80’s - 8  – 20 MHz

 – 90’s - 50  – 100 MHz

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 6/26

• Scalar processor

 – Achieve average execution of no morethan one instruction per clock cycle

• Superscalar processor

 – More instructions per clock cycle

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 7/26

Bus

 – Communication channel between thevarious parts of the system

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 8/26

parts

• Data lines

 – Carry instruction from memory toprocessor during each instruction fetch

cycle and data between processor andmemory or I/O.

 – Bi-directional

 – Specify size of data being processedinternally

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 9/26

• Address lines

 – Carry address of memory or I/Olocations to be accessed

8 bit - 16 lines

16 bit - 20 lines / 24 lines

32 bit - 32 lines

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 10/26

• Control lines

 – Carry signals to activate the data orinstruction transfer within the system

 – Indicate type of transfer

• Example:

 – Memory Read or write

 – I/O read or write

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 11/26

1. Pipelined processor design

pipelined – consist of a number of stages witheach stage performing one

sequential step of the overall task.

Architectural developments

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 12/26

Instruction fetch/execute overlap

- execution and fetching of the nextinstruction can be donesimultaneously.

3 steps can be overlapped: fetchdecode execute 

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 13/26

2. Processor  – memory interface

concerned with:

a. carrying information frommemory to

processor during fetch cycle.

b. carrying data from processor tomemory during execute cycle.

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 14/26

2a. Increase communication bandwidth

- increase data line- match the speed of memory toprocessor

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 15/26

2b. Prefetch buffers

- instruction and data is held in aFIFO buffer.

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 16/26

2c. Cache

- a high-speed storagemechanism.

- separate of part of the memory

unified cache – both instruction and data can be

stored.

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 17/26

2d. Memory hierarchy

inboard

outboard

offline

- as one goes down,1. Decrease cost per bit

2. Increase capacity

3. Increase access time (slower)

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 18/26

1. Use of memory and registers

- 32 registers chosen for storingintegers

- R0 to R31 where R0 = 0.

Instruction sets

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 19/26

2. Operand size

- use of size field in the instructiondetermine the size of operands.

00 - 8 bit

01 - 16 bit

10 - 32 bit11 - 64 bits

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 20/26

3. Size of memory locations

n bits allows 2n locations

4. Instructions bits

6 bits for opcode

5 bits per register in the instruction.

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 21/26

1. Register  – register format

opcode –

Rd –

Rs1 –

Rs2 –

unused

Example: add R4, R5, R2

Classes of instruction

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 22/26

2. Register  – constant format

opcode –

Rd –

Rs1 –

16-bit constant

Exmaple: add R5, R6, 32

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 23/26

3. Register  – memory format

- operand must be loaded first toregister from memory.

- need to specify address of memorylocation

LOAD  – memory to registerSTORE  – register to memory

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 24/26

Example:

LD R1, 100[R4]

ST R4, 200[R2}

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 25/26

LOAD:

opcode  – Rd  – add. Reg.  – 16-bitoffset

STORE:

opcode  – source  – add. Reg.  – 16-bitoffset

8/3/2019 Architectural Developments

http://slidepdf.com/reader/full/architectural-developments 26/26

LOAD:

opcode  – Rd  – add. Reg.  – unused  – 11-bitoffset

STORE:

opcode  – unused  – add. Reg.  – source  – 11-

bit offset