Pipelining in computer architecture

Pipeline in computer organisation

1. Chapter 8. Pipelining
Ramakrishna Reddy bijjam
9966484777
Asst. Professor
Avanthi group of colleges

2. Overview
Pipelining is widely used in modern
processors.
Pipelining improves system performance in
terms of throughput.[No of work done at a
given time]
Pipelined organization requires sophisticated
compilation techniques.

3. Basic Concepts

4. Making the Execution of
Programs Faster
Use faster circuit technology to build the
processor and the main memory.
Arrange the hardware so that more than one
operation can be performed at the same time.
In the latter way, the number of operations
performed per second is increased even
though the elapsed time needed to perform
any one operation is not changed.

5. pipeline
It is technique of decomposing a sequential
process into suboperation, with each
suboperation completed in dedicated
segment.
Pipeline is commonly known as an assembly
line operation.
It is similar like assembly line of car
manufacturing.
First station in an assembly line set up a
chasis, next station is installing the engine,
another group of workers fitting the body.

6. Traditional Pipeline Concept
Laundry Example
Ann, Brian, Cathy, Dave
each have one load of clothes
to wash, dry, and fold
Washer takes 30 minutes
Dryer takes 40 minutes
“Folder” takes 20 minutes
ABCD

7. Traditional Pipeline Concept
Sequential laundry takes 6
hours for 4 loads
If they learned pipelining,
how long would laundry
take?
A
B
C
D
304020304020304020304020
6 PM 7 8 9 10 11 Midnight
Time

8. Traditional Pipeline Concept
Pipelined laundry takes
3.5 hours for 4 loads
A
B
C
D
6 PM 7 8 9 10 11 Midnight
T
a
s
k
O
r
d
e
r
Time
304040404020

9. Traditional Pipeline Concept
Pipelining doesn’t help
latency of single task, it
helps throughput of entire
workload
Pipeline rate limited by
slowest pipeline stage
Multiple tasks operating
simultaneously using
different resources
Potential speedup = Number
pipe stages
Unbalanced lengths of pipe
stages reduces speedup
Time to “fill” pipeline and
time to “drain” it reduces
speedup
Stall for Dependences
A
B
C
D
6 PM 7 8 9
T
a
s
k
O
r
d
e
r
Time
304040404020

10. Idea of pipelining in computer
The processor execute the program by
fetching and executing instructions. One after
the other.
Let F
i
and E
i
refer to the fetch and execute
steps for instruction I
i

11. Use the Idea of Pipelining in a
Computer
F
1
E
1
F
2
E
2
F
3
E
3
I
1
I
2
I
3
(a) Sequential execution
Instruction
fetch
unit
Execution
unit
Interstage buffer
B1
(b) Hardware organization
Time
F
1
E
1
F
2
E
2
F
3
E
3
I
1
I
2
I
3
Instruction
(c) Pipelined execution
Figure 8.1. Basic idea of instruction pipelining.
Clock cycle1 2 3 4
Time
Fetch + Execution

12. Contd.,
Computer that has two separate hardware
units, one for fetching and another for
executing them.
 the instruction fetched by the fetch unit is
deposited in an intermediate buffer B1.
This buffer needed to enable the execution
unit while fetch unit fetching the next
instruction.

13. 8.1(c)
The computer is controlled by a clock.
Any instruction fetch and execute steps
completed in one clock cycle.

14. Use the Idea of Pipelining in a
Computer
F
4I
4
F
1
F
2
F
3
I
1
I
2
I
3
D
1
D
2
D
3
D
4
E
1
E
2
E
3
E
4
W
1
W
2
W
3
W
4
Instruction
Figure 8.2.A 4­stage pipeline.
Clock cycle1 2 3 4 5 6 7
(a) Instruction execution divided into four steps
F : Fetch
instruction
D : Decode
instruction
and fetch
operands
E: Execute
operation
W : Write
results
Interstage buffers
(b) Hardware organization
B1 B2 B3
Time
Fetch + Decode
+ Execution + Write
Textbook page: 457

15. Fetch(F)- read the instruction from the memory
Decode(D)- Decode the instruction and fetch
the source operand
Execute(E)- perform the operation specified by
the instruction
Write(W)- store the result in the destination
location

16. Role of Cache Memory
Each pipeline stage is expected to complete in one
clock cycle.
The clock period should be long enough to let the
slowest pipeline stage to complete.
Faster stages can only wait for the slowest one to
complete.
Since main memory is very slow compared to the
execution, if each instruction needs to be fetched
from main memory, pipeline is almost useless.[ten
times greater than the time needed to perform
pipeline stage]
Fortunately, we have cache.

17. Pipeline Performance
The potential increase in performance
resulting from pipelining is proportional to the
number of pipeline stages.
However, this increase would be achieved
only if all pipeline stages require the same
time to complete, and there is no interruption
throughout program execution.
Unfortunately, this is not true.
Floating point may involve many clock cycle

18. Pipeline Performance
F
1
F
2
F
3
I
1
I
2
I
3
E
1
E
2
E
3
D
1
D
2
D
3
W
1
W
2
W
3
Instruction
F
4
D
4I
4
Clock cycle1 2 3 4 5 6 7 8 9
Figure 8.3.Effect of an execution operation taking more than one clock cycle.
E
4
F
5
I
5 D
5
Time
E
5
W
4

19. Pipeline Performance
The previous pipeline is said to have been stalled for two clock
cycles.
Any condition that causes a pipeline to stall is called a hazard.
Data hazard – any condition in which either the source or the
destination operands of an instruction are not available at the
time expected in the pipeline. So some operation has to be
delayed, and the pipeline stalls.
Instruction (control) hazard – a delay in the availability of an
instruction causes the pipeline to stall.[cache miss]
Structural hazard – the situation when two instructions require
the use of a given hardware resource at the same time.

20. Pipeline Performance
F
1
F
2
F
3
I
1
I
2
I
3
D
1
D
2
D
3
E
1
E
2
E
3
W
1
W
2
W
3
Instruction
Figure 8.4.Pipeline stall caused by a cache miss in F2.
1 2 3 4 5 6 7 8 9Clock cycle
(a) Instruction execution steps in successive clock cycles
1 2 3 4 5 6 7 8Clock cycle
Stage
F: Fetch
D: Decode
E: Execute
W: Write
F
1 F
2 F
3
D
1
D
2
D
3
idleidle idle
E
1 E
2 E
3idle idleidle
W
1
W
2
idleidleidle
(b) Function performed by each processor stage in successive clock cycles
9
W
3
F
2 F
2 F
2
Time
Time
Idle periods –
stalls (bubbles)
Instruction
hazard(Cache
miss)
Decode unit is idle
in cycles 3 through
5, Execute unit idle
in cycle 4 through 6
and write unit is idle
in cycle 5 through 7
such idle period is
called stalls.

21. Pipeline Performance
F
1
F
2
F
3
I
1
I
2 (Load)
I
3
E
1
M
2
D
1
D
2
D
3
W
1
W
2
Instruction
F
4
I
4
Clock cycle1 2 3 4 5 6 7
Figure 8.5.Effect of a Load instruction on pipeline timing.
F
5
I
5 D
5
Time
E
2
E
3 W
3
E
4D
4
Load X(R1), R2
Structural
hazard
The memory address, X+(R1) is computed in step E2in cycle4, then memory
access takes place in cycle5. the operand read from memory is written into
register R2 in cycle 6[Execution takes 2 cycles] it stalls pipeline to stall for one
cycle. Bcox both instruction I2 and I3 require access of register file in cycle 6.

22. Pipeline Performance
Again, pipelining does not result in individual
instructions being executed faster; rather, it is the
throughput that increases.
Throughput is measured by the rate at which
instruction execution is completed.
Pipeline stall causes degradation in pipeline
performance.
We need to identify all hazards that may cause the
pipeline to stall and to find ways to minimize their
impact.

23. Quiz
Four instructions, the I2 takes two clock
cycles for execution. Pls draw the figure for 4-
stage pipeline, and figure out the total cycles
needed for the four instructions to complete.

24. Data Hazards

25. Data Hazards
We must ensure that the results obtained when instructions are
executed in a pipelined processor are identical to those obtained
when the same instructions are executed sequentially.
Hazard occurs
A ← 3 + A
B ← 4 × A
No hazard
A ← 5 × C
B ← 20 + C
When two operations depend on each other, they must be
executed sequentially in the correct order.
Another example:
Mul R2, R3, R4
Add R5, R4, R6

26. Data Hazards
F
1
F
2
F
3
I
1
 (Mul)
I
2 (Add)
I
3
D
1
D
3
E
1
E
3
E
2
W
3
Instruction
Figure 8.6.Pipeline stalled by data dependency between D
2 and W
1.
1 2 3 4 5 6 7 8 9Clock cycle
W
1
D
2A W
2
F
4 D
4 E
4W
4I
4
D
2
Time
Figure 8.6. Pipeline stalled by data dependency between D
2
and W
1
.

27. Operand Forwarding
Instead of from the register file, the second
instruction can get data directly from the
output of ALU after the previous instruction is
completed.
A special arrangement needs to be made to
“forward” the output of ALU to the input of
ALU.

28. Register
file
SRC1 SRC2
RSLT
Destination
Source 1
Source 2
(a) Datapath
ALU
E: Execute
(ALU)
W: Write
(Register file)
SRC1,SRC2 RSLT
(b) Position of the source and result registers in the processor pipeline
Figure 8.7.Operand forwarding in a pipelined processor.
Forwarding path

29. Handling Data Hazards in
Software
Let the compiler detect and handle the
hazard:
I1: Mul R2, R3, R4
NOP
NOP
I2: Add R5, R4, R6
The compiler can reorder the instructions to
perform some useful work during the NOP
slots.

30. Side Effects
The previous example is explicit and easily detected.
Sometimes an instruction changes the contents of a register
other than the one named as the destination.
When a location other than one explicitly named in an instruction
as a destination operand is affected, the instruction is said to
have a side effect. (Example?)
Example: conditional code flags:
Add R1, R3
AddWithCarry R2, R4
Instructions designed for execution on pipelined hardware should
have few side effects.

31. Instruction Hazards

32. Overview
Whenever the stream of instructions supplied
by the instruction fetch unit is interrupted, the
pipeline stalls.
Cache miss
Branch

33. Unconditional Branches
F
2I
2 (Branch)
I
3
I
k
E
2
F
3
F
k E
k
F
k+1E
k+1I
k+1
Instruction
Figure 8.8.An idle cycle caused by a branch instruction.
Execution unit idle
1 2 3 4 5Clock cycle
Time
F
1
I
1
E
1
6
X

34. Unconditional Branches
The time lost as a result of a branch
instruction is referred to as the branch
penalty.
The previous example instruction I3 is
wrongly fetched and branch target address k
will discard the i3.
Reducing the branch penalty requires the
branch address to be computed earlier in the
pipeline.
Typically the Fetch unit has dedicated h/w
which will identify the branch target address
as quick as possible after an instruction is
fetched.

35. Branch Timing
X
Figure 8.9.Branch timing.
F
1D
1E
1W
1
I
2 (Branch)
I
1
1 2 3 4 5 6 7Clock cycle
F
2D
2
F
3 X
F
kD
kE
k
F
k+1D
k+1
I
3
I
k
I
k+1
W
k
E
k+1
(b) Branch address computed in Decode stage
F
1D
1E
1W
1
I
2 (Branch)
I
1
1 2 3 4 5 6 7Clock cycle
F
2D
2
F
3
F
kD
kE
k
F
k+1D
k+1
I
3
I
k
I
k+1
W
k
E
k+1
(a) Branch address computed in Execute stage
E
2
D
3
F
4 XI
4
8
Time
Time
- Branch penalty
- Reducing the penalty

36. Instruction Queue and Prefetching
Either cache (or) branch instruction stalls the
pipeline.
Many processor employs dedicated fetch unit
which will fetch the instruction and put them
into a queue.
It can store several instruction at a time.
A separate unit called dispatch unit, takes
instructions from the front of the queue and
send them to the execution unit.

37. Instruction Queue and
Prefetching
F : Fetch
instruction
E : Execute
instruction
W : Write
results
D : Dispatch/
Decode
Instruction queue
Instruction fetch unit
Figure 8.10. Use of an instruction queue in the hardware organization of Figure 8.2b.
unit

38. Conditional Braches
A conditional branch instruction introduces
the added hazard caused by the dependency
of the branch condition on the result of a
preceding instruction.
The decision to branch cannot be made until
the execution of that instruction has been
completed.
Branch instructions represent about 20% of
the dynamic instruction count of most
programs.

39. Delayed Branch
The instructions in the delay slots are always
fetched. Therefore, we would like to arrange
for them to be fully executed whether or not
the branch is taken.
The objective is to place useful instructions in
these slots.
The effectiveness of the delayed branch
approach depends on how often it is possible
to reorder instructions.

40. Delayed Branch
Add
LOOP Shift_left R1
Decrement
Branch=0
R2
LOOP
NEXT
(a) Original program loop
LOOP Decrement R2
Branch=0
Shift_left
LOOP
R1
NEXT
(b) Reordered instructions
Figure 8.12. Reordering of instructions for a delayed branch.
Add
R1,R3
R1,R3

41. Delayed Branch
F E
F E
F E
F E
F E
F E
F E
Instruction
Decrement
Branch
Shift (delay slot)
Figure 8.13.Execution timing showing the delay slot being filled
during the last two passes through the loop in Figure 8.12.
Decrement (Branch taken)
Branch
Shift (delay slot)
Add (Branch not taken)
1 2 3 4 5 6 7 8Clock cycle
Time

42. Branch Prediction
To predict whether or not a particular branch will be taken.
Simplest form: assume branch will not take place and continue to
fetch instructions in sequential address order.
Until the branch is evaluated, instruction execution along the
predicted path must be done on a speculative basis.
Speculative execution: instructions are executed before the
processor is certain that they are in the correct execution
sequence.
Need to be careful so that no processor registers or memory
locations are updated until it is confirmed that these instructions
should indeed be executed.

43. Incorrectly Predicted Branch
F
1
F
2
I
1
 (Compare)
I
2
 (Branch>0)
I
3
D
1
E
1
W
1
F
3
F
4
F
k
D
k
D
3
X
XI
4
I
k
Instruction
Figure 8.14.Timing when a branch decision has been incorrectly predicted
as not taken.
E
2
Clock cycle 1 2 3 4 5 6
D
2
/P
2
Time

44. Branch Prediction
Better performance can be achieved if we arrange
for some branch instructions to be predicted as
taken and others as not taken.
Use hardware to observe whether the target
address is lower or higher than that of the branch
instruction.
Let compiler include a branch prediction bit.
So far the branch prediction decision is always the
same every time a given instruction is executed –
static branch prediction.

45. Influence on
Instruction Sets

46. Overview
Some instructions are much better suited to
pipeline execution than others.
Addressing modes
Conditional code flags

47. Addressing Modes
Addressing modes include simple ones and
complex ones.
In choosing the addressing modes to be
implemented in a pipelined processor, we
must consider the effect of each addressing
mode on instruction flow in the pipeline:
Side effects
The extent to which complex addressing modes cause
the pipeline to stall
Whether a given mode is likely to be used by compilers

48. Recall
F
1
F
2
F
3
I
1
I
2 (Load)
I
3
E
1
M
2
D
1
D
2
D
3
W
1
W
2
Instruction
F
4
I
4
Clock cycle1 2 3 4 5 6 7
Figure 8.5.Effect of a Load instruction on pipeline timing.
F
5
I
5 D
5
Time
E
2
E
3 W
3
E
4D
4
Load X(R1), R2
Load (R1), R2

49. Complex Addressing Mode
F
F D
D E
X +[R1][X +[R1]][[X +[R1]]]Load
Next instruction
(a) Complex addressing mode
W
1 2 3 4 5 6 7Clock cycle
Time
W
Forward
Load (X(R1)), R2

50. Simple Addressing Mode
X +[R1]F D
F
F
F D
D
D
E
[X +[R1]]
[[X +[R1]]]
Add
Load
Load
Next instruction
(b) Simple addressing mode
W
W
W
W
Add #X, R1, R2
Load (R2), R2
Load (R2), R2

51. Addressing Modes
In a pipelined processor, complex addressing
modes do not necessarily lead to faster execution.
Advantage: reducing the number of instructions /
program space
Disadvantage: cause pipeline to stall / more
hardware to decode / not convenient for compiler to
work with
Conclusion: complex addressing modes are not
suitable for pipelined execution.

52. Addressing Modes
Good addressing modes should have:
Access to an operand does not require more than one
access to the memory
Only load and store instruction access memory operands
The addressing modes used do not have side effects
Register, register indirect, index

53. Conditional Codes
If an optimizing compiler attempts to reorder
instruction to avoid stalling the pipeline when
branches or data dependencies between
successive instructions occur, it must ensure
that reordering does not cause a change in
the outcome of a computation.
The dependency introduced by the condition-
code flags reduces the flexibility available for
the compiler to reorder instructions.

54. Conditional Codes
Add
Compare
Branch=0
R1,R2
R3,R4
. . .
Compare
Add
Branch=0
R3,R4
R1,R2
. . .
(a) A program fragment
(b) Instructions reordered
Figure 8.17. Instruction reordering.

55. Conditional Codes
Two conclusion:
To provide flexibility in reordering instructions, the
condition-code flags should be affected by as few
instruction as possible.
The compiler should be able to specify in which
instructions of a program the condition codes are
affected and in which they are not.

56. Datapath and Control
Considerations

57. Original Design
Memory bus
data lines
Figure 7.8.Three­bus organization of the datapath.
Bus ABus B Bus C
Instruction
decoder
PC
Register
file
Constant 4
ALU
MDR
A
B
R
M
U
X
Incrementer
Address
lines
MAR
IR

58. Instruction cache
Figure 8.18.Datapath modified for pipelined execution, with
B
u
s
 A
B
u
s
 B
Control signal pipeline
IMAR
PC
Register
file
ALU
Instruction
A
B
R
decoder
Incrementer
MDR/Write
Instruction
queue
B
u
s
 C
Data cache
Memory address
MDR/ReadDMAR
Memory address
(Instruction fetches)
(Data access)
interstage buffers at the input and output of the ALU.
Pipelined Design
- Separate instruction and data caches
- PC is connected to IMAR
- DMAR
- Separate MDR
- Buffers for ALU
- Instruction queue
- Instruction decoder output
- Reading an instruction from the instruction cache
- Incrementing the PC
- Decoding an instruction
- Reading from or writing into the data cache
- Reading the contents of up to two regs
- Writing into one register in the reg file
- Performing an ALU operation

59. Superscalar Operation

60. Pipeline architecture can executes several
instruction concurrently.
Many instructions are present in the pipeline at
the same time,but they are in different stages of
their execution.
While instruction being fetched at the same time
another instruction being decoded stage (or)
execution.
One instruction completes execution in each
clock cycle.

61. Overview
The maximum throughput of a pipelined processor
is one instruction per clock cycle.
If we equip the processor with multiple processing
units to handle several instructions in parallel in
each processing stage, several instructions start
execution in the same clock cycle – multiple-issue.
Processors are capable of achieving an instruction
execution throughput of more than one instruction
per cycle – superscalar processors.
Multiple-issue requires a wider path to the cache
and multiple execution units to keep the instruction
queue to be filled.

62. The superscalar operation have multiple
execution units.

63. Superscalar
W : Write
results
Dispatch
unit
Instruction queue
Floating­
point
unit
Integer
unit
Figure 8.19.A processor with two execution units.
F : Instruction
fetch unit

64. Architecture
The above fig. shows the superscalar processor
with two execution unit.
The Fetch unit capable of reading two instruction
at a time and store it in the queue.
The dispatch unit decodes upto two instruction
from the front of queue(one is integer and another
one is floating point) dispatched in the same clock
cycle.
Processor’s program control unit – capable of
fetching and decoding several instruction
concurrently. It can issue multiple instructions
simultaneously.

65. Timing
I
1
 (Fadd) D
1
D
2
D
3
D
4
E
1A
E
1B
E
1C
E
2
E
3
E
3
E
3
E
4
W
1
W
2
W
3
W
4
I
2
 (Add)
I
3
 (Fsub)
I
4
 (Sub)
Figure 8.20.An example of instruction execution flow in the processor of Figure 8.19,
assuming no hazards are encountered. 
1 2 3 4 5 6Clock cycle
Time
F
1
F
2
F
3
F
4
7

66. Assume Floating point takes 3 clock cycles
5.Assume floating point unit as a 3 stage pipeline. So
it can accept a new instruction for execution in each
clock cycle. During cycle 4 the execution of I1 in
progress but it will enter the different stages inside
the execution pipleline, so this unit accept for I3 for
execution.
6.The integer unit can accept new instruction for
execution because I2 has entered into the write
stage.

67. Out-of-Order Execution
Hazards
Exceptions
Imprecise exceptions
Precise exceptions
I
1
 (Fadd) D
1
D
2
D
3
D
4
E
1A
E
1B
E
1C
E
2
E
3A
E
3B
E
3C
E
4
W
1
W
2
W
3
W
4
I
2
 (Add)
I
3
 (Fsub)
I
4
 (Sub)
1 2 3 4 5 6Clock cycle
Time
(a) Delayed write
F
1
F
2
F
3
F
4
7

68. Execution Completion
It is desirable to used out-of-order execution, so that an
execution unit is freed to execute other instructions as soon as
possible.
At the same time, instructions must be completed in program
order to allow precise exceptions.
The use of temporary registers
Commitment unit
I
1
 (Fadd) D
1
D
2
D
3
D
4
E
1A
E
1B
E
1C
E
2
E
3A
E
3B
E
3C
E
4
W
1
W
2
W
3
W
4
I
2
 (Add)
I
3
 (Fsub)
I
4
 (Sub)
1 2 3 4 5 6Clock cycle
Time
(b) Using temporary registers
TW
2
TW
4
7
F
1
F
2
F
3
F
4

69. If I
2
depends on the result of I
1,
the execution of
I
2
will be delayed. These dependencies are handled
correctly,as long as the execution of the instruction
will not be delayed.
one more reason which delays the execution is,
Exception
Two causes:
Bus error
Illegal operation(divide by zero)

70. Two types of Exception
* Imprecise * Precise
Imprecise- the result of I2 written into RF during cycle
4. suppose I1 causes an exception, the processor
allows to complete execution of succeeding
instruction(I2) is said to have imprecise exception.
Precise-in imprecise exception, consistent state is not
guaranteed, when an exception occurs.(The
processor will not allow to write the succeeding
instruction)
In consistent state, the result of the instruction must
be written into the program order(ie. The I2 must be
allow to written till cycle 6.

71. The integer execution unit has to retain the
result until cycle 6 and it cannot accept
instruction I4 until cycle 6.
Thus, in this method the output is partially
executed (or) discarded.
I
1
 (Fadd) D
1
D
2
E
1A
E
1B
E
1C
E
2
W
1
W
2
I
2
 (Add)
1 2 3 4 5 6Clock cycle
Time
F
1
F
2
7

72. If an external interrupt is received, the dispatch
unit stops reading new instructions from instruction
queue, and the instruction remaining in the queue
are discarded.

73. Execution Completion
In precise exception, the results are temporarily stored into the
temp register and later they are transferred to the permanent
registers in correct program order.

Thus, two write operations TW and W are carried out.
The step W are called commitment step(temp reg to perm
register)
TW- write into a temp registers
W- Transfer the contents back to the permanent reg
I
1
 (Fadd) D
1
D
2
E
1A
E
1B
E
1C
E
2
W
1
W
2
I
2
 (Add)
1 2 3 4 5 6Clock cycle
TW
2
7
F
1
F
2

74. Register renaming
A temporary register assumes the role of the
permanent register whose data it is holding
and is given the same name.

75. Performance
Considerations

76. Overview
The execution time T of a program that has a
dynamic instruction count N is given by:
where S is the average number of clock cycles it
takes to fetch and execute one instruction, and
R is the clock rate.
Instruction throughput is defined as the number
of instructions executed per second.
R
SN
T
´
=
S
R
P
s
=

77. Overview
An n-stage pipeline has the potential to increase the
throughput by n times.
However, the only real measure of performance is
the total execution time of a program.
Higher instruction throughput will not necessarily
lead to higher performance.
Two questions regarding pipelining
How much of this potential increase in instruction throughput can be
realized in practice?
What is good value of n?

78. Number of Pipeline Stages
Since an n-stage pipeline has the potential to
increase the throughput by n times, how about we
use a 10,000-stage pipeline?
As the number of stages increase, the probability of
the pipeline being stalled increases.
The inherent delay in the basic operations
increases.
Hardware considerations (area, power, complexity,
…)