Lecture 4: CPU Performance. A Modern Processor Intel Core i7.

Lecture 4:Lecture 4:

CPU PerformanceCPU Performance

A Modern Processor

Intel Core i7

Processor Performance

Lower bounds that characterize the maximum performance:

Latency Bound

• Occurs when operations must be performed in strict sequence (e.g. data dependency)

• Minimum time to perform the operations sequentially

Throughput Bound

• Characterizes the raw computing capacity of the processor’s functional units.

• Maximum operations per cycle

Pipelining

s1 s2 s3

time

stages

s1

s2

s3

time

stages

s1

s2

s3

Without pipeline With pipeline

Pipelining

s – stagesn – taskst – time per stage

time

stages

s1

s2

s3

time

stages

s1

s2

s3

Without pipeline With pipeline

T1 = s . t . n Tp = s . t + (n-1).t

Speedup = T1 / Tp = s.n = s . s+(n-1) s/n +(1-1/n)

Speedup = s n

Throughput = n . Tp

Pipelining

Slowest stage determines the pipeline performance

s1 s2 s3

time

stages

s1

s2

s3

time

stages

s1

s2

s3

Without pipeline With pipeline

10 30 20

Computational Pipelines

Combinatorial logic R

egComb.log.

AR

Comb.log.B

RComb.log.

CR

clock

clock

Limitations of Pipelining Nonuniform partitioning

• Stage delays may be nonuniform

• Throughput is limited by the slowest stage

Deep pipelining

• Large number of stages

• Modern processors have deep pipelines (15 or more) to increase the clock rate.

Comb.log.A

RComb.log.

BR

Comb.log.C

R

clock

50ps 20ps 150ps 20ps 100ps 20ps

Comb.log.A

RComb.log.

BR

Comb.log.C

R

clock

50ps 20ps 50ps 20ps 50ps 20ps

…

Pipelined Parallel Addera1,b1a4,b4 a3,b3 a2,b2

Pipelined Parallel Adderc1,d1

a1+b1

c4,d4 c3,d3 c2,d2

a4,b4 a3,b3 a2,b2

Pipelined Parallel Adder

c1+d1

a1+b1

c4,d4 c3,d3 c2,d2

a4,b4 a3,b3 a2+b2

e1,f1e2,f2e3,f3e4,f4

Pipelined Parallel Adder

c1+d1

a1+b1

c4,d4 c3,d3 c2+d2

a4,b4 a3+b3 a2+b2

e1+f1e2,f2e3,f3e4,f4

g1,h1g2,h2g3,h3g4,h4

Pipelined Parallel Adder

c1+d1

a1+b1

c4,d4 c3+d3 c2+d2

a4+b4 a3+b3 a2+b2

e1+f1e2+f2e3,f3e4,f4

g1+h1g2,h2g3,h3g4,h4

Instruction Execution Pipeline

IF

1. Instruction Fetch Cycle (IF)

• Fetch current instruction from memory

• Increment PC

2. Instruction decode / register fetch cycle (ID)

• Decode instruction

• Compute possible branch target

• Read registers from the register file

3. Execution / effective address cycle (EX)

• Form the effective address

• ALU performs the operation specified by the opcode

4. Memory access (MEM)

• Memory read for load instruction

• Memory write for store instruction

5. Write-back cycle (WB)

• Write result into register file

WBMEMEXID

Instruction Execution Pipeline

time

stages

IF

IF WBMEMEXID

WBMEM

EX

ID

Pipeline Hazards

1. Structural hazards

2. Data Hazards

3. Control Hazards

Pipeline Hazards

Structural Hazards Arise from resource conflicts when the hardware cannot support all

possible combinations of instructions simultaneously in overlapped execution.

time

stages

IF

IF WBMEMEXID

WBMEM

EX

ID

stall (bubble)

Mem Reg ALU Mem Reg

Pipeline Hazards

Data Hazards Arise when an instruction depends on the results of a previous

instruction in a way that is exposed by the overlapping of instructions.

time

stages

IF

IF WBMEMEXID

WBMEM

EX

ID

Mem Reg ALU Mem Reg

ADD R1, R2, R3SUB R4, R1, R5AND R6, R1, R7OR R8, R1, R9XOR R10, R1, R11

Pipeline Hazards

Data Hazards Forwarding (by-passing)

IF WBMEMEXID

Mem Reg ALU Mem Reg

IF WBMEMEXID

IF WBMEMEXID

IF WBMEMEXID

Mem Reg ALU Mem Reg

Mem Reg ALU Mem Reg

Mem Reg ALU Mem Reg

Pipeline Hazards

Control (Branch) Hazards Arise from pipelining of instructions (e.g. branch) that change PC.

time

stages

IF

WBMEM

EX

ID

LOOP: LOAD 100,XADD 200,XSTORE 300,XDECXBNE LOOP...

for i=n to 1 ci = ai + bi

Pipeline Hazards

Control (Branch) Hazards Freeze (flush)

time

stages

IF

WBMEM

EX

ID

BRA L1...

L1: NEXTNEXTNEXTNEXT

Pipeline Hazards

Control (Branch) Hazards Predicted-not-taken

time

stages

IF

WBMEM

EX

ID

BNE L1NEXTNEXTNEXT...

L1: NEXTNEXTNEXT

Not taken Taken

Pipeline Hazards

Control (Branch) Hazards Predicted-taken

time

stages

IF

WBMEM

EX

ID

BNE L1NEXTNEXTNEXT...

L1: NEXTNEXTNEXT

Not taken Taken

Pipeline Hazards

Control (Branch) Hazards Delayed branch

time

stages

IF

WBMEM

EX

ID

ADD R1,R2,R3if (R2=0) branch L1delay slotNEXTNEXT...

L1: NEXTNEXTNEXT

Not taken Taken

branch instructionsequential successorBranch target if taken

if (R2=0) branch L1ADD R1,R2,R3NEXTNEXT...

L1: NEXTNEXTNEXT

Levels of Parallelism

Bit level parallelism

• Within arithmetic logic circuits Instruction level parallelism

• Multiple instructions execute per clock cycle Memory system parallelism

• Overlap of memory operations with computation Operating system parallelism

• More than one processor

• Multiple jobs run in parallel on SMP

• Loop level

• Procedure level