UNIT VI : CODE OPTIMIZATION AND CODE

GENERATION
COURSE : LANGUAGE PROCESSOR AND COMPILER CONSTRUCTION

Manisha Mali
[email protected]

BRACT’S, Vishwakarma Institute of Information Technology, Pune-48

(An Autonomous Institute affiliated to Savitribai Phule Pune University)
(NBA and NAAC accredited, ISO 9001:2015 certified)
Department of Computer Engineering
Unit VI
Code optimization
 THE PHASES OF A COMPILER
Source program

Lexical Analyzer

26-Oct-20
Syntax Analyzer

VIIT
Manisha Mali Computer Engg. Dept.,
Semantic Analyzer
Symbol Error
table handler
manager
Intermediate code generator

Code optimizer

Code generator
3

Target
program
CODE OPTIMIZATION

26-Oct-20
 Code optimization organization
 Optimization

VIIT
Manisha Mali Computer Engg. Dept.,
 Machine dependent/independent

 Criteria for Transformations

 Principal sources of optimization

 Peephole optimization

Manisha Mali
 Manisha Mali Computer Engg. Dept.,
26-Oct-20
VIIT
5
CODE OPTIMIZATION AND PHASES

Manisha Mali
OPTIMIZATION

 Code produced by standard algorithms can

26-Oct-20
often be made to run faster, take less space or
both

VIIT
Manisha Mali Computer Engg. Dept.,
 Theseimprovements are achieved through
transformations called optimizations

 Compilers that apply these transformations

are called optimizing compilers

Manisha Mali
OPTIMIZING INTERMEDIATE CODE
 This phase is generally only included in

26-Oct-20
optimizing compilers
 Offers the following advantages:

VIIT
Manisha Mali Computer Engg. Dept.,
– Operations needed to implement high-level
constructs are made explicit (i.e., address
calculations)
– Intermediate code is independent of target
machine; code generator can be replaced for
different machine
 We are assuming intermediate code uses three
address instructions
7

Manisha Mali
FACTORS INFLUENCING OPTIMIZATION

26-Oct-20
 The target machine: machine dependent
factors can be parameterized to compiler
for fine tuning

VIIT
Manisha Mali Computer Engg. Dept.,
 Architecture of Target CPU:
 Number of CPU registers
 RISC vs CISC
 Pipeline Architecture
 Number of functional units
 Machine Architecture
 Cache Size and type
 Cache/Memory transfer rate
8
MACHINE DEPENDENT/INDEPENDENT
 Machine Independent

26-Oct-20
 Common Subexpression elimination
 removing of loop invariants
Reduction in strengths.

VIIT
Manisha Mali Computer Engg. Dept.,

 dead code elimination,
 code motion
 Machine dependent Issues
 Assignment and use of registers
 latency hiding; managing bounded machine resources like
registers, functional units, caches.
 Rearrangement of Quadruples for code
optimization
9

Manisha Mali
CRITERIA FOR TRANSFORMATIONS

26-Oct-20
 Criteria for Code-Improving Transformation:
 Meaning must be preserved (correctness)
 Speedup must occur on average.

VIIT
Manisha Mali Computer Engg. Dept.,
 Work done must be worth the effort
 Cannot change the output produced for any
 Can not introduce an error

 Opportunities:
 Programmer (algorithm, directives)
 Intermediate code
 Target code
10
BEYOND OPTIMIZING COMPILERS
 Improvements can be made at various phases
 Source code:

26-Oct-20
– Algorithmic transformations can produce spectacular
improvements
– Profiling can be helpful to focus a programmer's

VIIT
Manisha Mali Computer Engg. Dept.,
attention on important code
 Intermediate code:
– Compiler can improve loops, procedure calls, and
address
calculations
– Typically only optimizing compilers include this phase
 Target code:
– Compilers can use registers efficiently
– Peephole transformation can be applied
11

Manisha Mali
THEMES BEHIND OPTIMIZATION
TECHNIQUES

26-Oct-20
 Avoid redundancy: something already
computed need not be computed again
 Smaller code: less work for CPU, cache, and

VIIT
Manisha Mali Computer Engg. Dept.,
memory!
 Less jumps: jumps interfere with code pre-fetch
 Code locality: codes executed close together in
time is generated close together in memory –
increase locality of reference
 Extract more information about code: More
info – better code generation

12
PRINCIPAL SOURCES OF
OPTIMIZATION

26-Oct-20
 Function-preserving transformations
 Common subexpressions

VIIT
Manisha Mali Computer Engg. Dept.,
 Copy propogation
 Dead-code elimination

 Loop optimization
 Code motion
 Induction variables
 Reduction in strength

13

Manisha Mali
OPTIMIZING TRANSFORMATIONS

26-Oct-20
 Compile time evaluation
 Constant folding

VIIT
Manisha Mali Computer Engg. Dept.,
 Constant propagation
 Common sub-expression elimination
 Code motion
 Strength Reduction
 Dead code elimination
 Copy propagation
 Loop optimization
 Induction variables and strength reduction 14
COMPILE-TIME EVALUATION

26-Oct-20
 Expressions whose values can be pre-computed
at the compilation time

VIIT
Manisha Mali Computer Engg. Dept.,
 Two ways:
 Constant folding
 Constant propagation

15
COMPILE-TIME EVALUATION

26-Oct-20
 Constant folding: Evaluation of an expression
with constant operands to replace the expression
with single value

VIIT
Manisha Mali Computer Engg. Dept.,
 Example:
area := (22.0/7.0) * r ** 2

area := 3.14286 * r ** 2

16
 CONSTANT PROPAGATION
 What does it mean?
 Given an assignment x = c, where c is a constant,

26-Oct-20
replace later uses of x with uses of c, provided there are
no intervening assignments to x.
Similar to copy propagation

VIIT
Manisha Mali Computer Engg. Dept.,


 Extra feature: It can analyze constant-value conditionals to

determine whether a branch should be executed or not.
 When is it performed?
 Early in the optimization process.
 What is the result?
 Smaller code
 Fewer registers

17
COMPILE-TIME EVALUATION

26-Oct-20
 Constant Propagation: Replace a variable with
constant which has been assigned to it earlier.

VIIT
Manisha Mali Computer Engg. Dept.,
 Example:
pi := 3.14286
area = pi * r ** 2

area = 3.14286 * r ** 2

18
CLASSIFICATIONS OF
OPTIMIZATION TECHNIQUES

26-Oct-20
 Peephole optimization
 Local optimizations

VIIT
Manisha Mali Computer Engg. Dept.,
 Global Optimizations
 Inter-procedural
 Intra-procedural

 Loop optimization

19
PEEPHOLE OPTIMIZATIONS

1. A Simple but effective technique for locally improving the

26-Oct-20
target code (can also be applied to intermediate code) is
peephole optimization,
2. A method for trying to improve the performance of the target

VIIT
Manisha Mali Computer Engg. Dept.,
program
3. The peephole is a small, moving window on the target program
4. By examining a short sequence of target instructions and
replacing these instructions by a shorter or faster sequence
whenever possible.
5. Each improvement may create opportunities for additional
improvements
6. Repeated passes may be necessary
20
PEEPHOLE OPTIMIZATIONS

Characteristics of peephole optimization

26-Oct-20
1. Redundant instruction elimination
2. Flow of control information

VIIT
Manisha Mali Computer Engg. Dept.,
3. Algebraic Simplification
4. Use of machine Idioms

21

Manisha Mali
SOME PEEPHOLE OPTIMIZATIONS
 A few algebraic identities that occur frequently (such as
x:= x + 0 or x := x * 1) can be eliminated

26-Oct-20
 Reduction in strength replaces expensive operations
with cheaper ones
– Calculating x * x is likely much cheaper than x^2 using

VIIT
Manisha Mali Computer Engg. Dept.,
an exponentiation routine
– It may be cheaper to implement x * 5 as x * 4 + x
 Some machines may have hardware instructions to
implement certain specific operations efficiently
– For example, auto-increment may be cheaper than a
straightforward x := x + 1
– Auto-increment and auto-decrement are also useful
when pushing into or popping off of a stack

22

Manisha Mali
ALGEBRAIC IDENTITIES

26-Oct-20
 Worth recognizing single instructions with a constant
operand:

VIIT
Manisha Mali Computer Engg. Dept.,
A * 1 = A
A * 0 = 0
A / 1 = A
A * 2 = A + A
More delicate with floating-point
 Strength reduction:
A ^ 2 = A * A

23
PEEPHOLE OPTIMIZATIONS (CONT’D)
REDUNDANT-INSTRUCTION ELIMINATION

26-Oct-20
 Redundant loads and stores
1. lw $t0, a

VIIT
Manisha Mali Computer Engg. Dept.,
2. sw $t0, a
 Unreachable code
#define debug 0
if (debug) {
/* print debugging information */
}

24

Manisha Mali
PEEPHOLE OPTIMIZATIONS
Constant Folding

26-Oct-20

x := 32 becomes x := 64
x := x + 32

VIIT
Manisha Mali Computer Engg. Dept.,
 Unreachable Code
goto L2
x := x + 1  No need
 Flow of control optimizations
goto L1 becomes goto L2
…
L1: goto L2  No needed if no other L1 branch

25
PEEPHOLE OPTIMIZATIONS

26-Oct-20
 Algebraic Simplification
x := x + 0  No needed
Dead code

VIIT
Manisha Mali Computer Engg. Dept.,

x := 32  where x not used after statement
y := x + y  y := y + 32
 Reduction in strength
x := x * 2  x := x + x
 x := x << 2

26
Common expression
can be eliminated

26-Oct-20
SIMPLE EXAMPLE: A[I+1] = B[I+1]

VIIT
Manisha Mali Computer Engg. Dept.,
t1 = i+1 t1 =i+1
t2 = b[t1] t2 = b[t1]
t3 = i + 1  no longer live
t3 = i + 1
a[t1] = t2
a[t3] = t2

27
Now, suppose i is a constant:

26-Oct-20
i =4 i =4 i=4
t1 = i+1 t1 =5 t1 = 5

VIIT
Manisha Mali Computer Engg. Dept.,
t2 = b[t1] t2 = b[t1] t2 = b[5]
a[t1] = t2 a[t1] = t2 a[5] = t2

i=4
Final Code:
t2 = b[5]
a[5] = t2

28
OPTIMIZATIONS ON CFG

26-Oct-20
 Must take control flow into account
 Common Sub-expression Elimination

VIIT
Manisha Mali Computer Engg. Dept.,
 Constant Propagation
 Dead Code Elimination
 Partial redundancy Elimination
 …
 Applying one optimization may raise
opportunities for other optimizations.

29
SIMPLE LOOP OPTIMIZATIONS

26-Oct-20
 Code Motion
Move invariants out of the loop.

VIIT
Manisha Mali Computer Engg. Dept.,
Example:
while (i <= limit - 2)
becomes
t := limit - 2
while (i <= t)

30
PRINCIPAL SOURCES OF
OPTIMIZATION

26-Oct-20
 Function-preserving transformations
 Common subexpressions

VIIT
Manisha Mali Computer Engg. Dept.,
 Copy propogation
 Dead-code elimination

 Loop optimization
 Code motion
 Induction variables
 Reduction in strength

31

Manisha Mali
BASIC BLOCK

26-Oct-20
 BB is a sequence of consecutive statements in
which the flow control enters at the beginning
and leaves at the end w/o halt or possible

VIIT
Manisha Mali Computer Engg. Dept.,
branching except at the end

 Flow Graph
 Once an intermediate code program is
partitioned into basic blocks, we represent the
flow of control between them by a flow graph

32
E X.

26-Oct-20
fact(x)
{

VIIT
Manisha Mali Computer Engg. Dept.,
int f=1;
for(i=2;i<=x;i++)
f=f*i;
return(f);
}

33
 QUICKSORT IN C
void quicksort(m,n)
int m, n;
{

26-Oct-20
int i,j;
int v,x;
if ( n <= m ) return;
/* fragment begins here */

VIIT
Manisha Mali Computer Engg. Dept.,
i = m-1; j = n; v = a[n];
While(1)
{
do i = i+1; while ( a[i] < v );
do j = j-1; while ( a[j] > v );
if (i >= j ) break;
x = a[i]; a[i] = a[j]; a[j] = x;
}
x = a[i]; a[i] = a[n]; a[n] = x; /* fragment ends here
*/
quicksort (m, j); quicksort (i+1 ,n);
}
34

Manisha Mali
 Manisha Mali Computer Engg. Dept.,
26-Oct-20
VIIT
35
BASIC BLOCK
THREE ADDRESS CODE OF QUICK SORT

1 i=m-1 16 t7 = 4 * i
2 j=n 17 t8 = 4 * j

26-Oct-20
3 t1 =4 * n 18 t9 = a[t8]
4 v = a[t1] 19 a[t7] = t9

VIIT
Manisha Mali Computer Engg. Dept.,
5 i=i +1 20 t10 = 4 * j
6 t2 = 4 * i 21 a[t10] = x
7 t3 = a[t2] 22 goto (5)
8 if t3 < v goto (5) 23 t11 = 4 * i
9 j=j–1 24 x = a[t11]
10 t4 = 4 * j 25 t12 = 4 * i
11 t5 = a[t4] 26 t13 = 4 * n
12 if t5 > v goto (9) 27 t14 = a[t13]
13 if i >= j goto (23) 28 a[t12] = t14
14 t6 = 4 * i 29 t15 = 4 * n
36
15 x = a[t6] 30 a[t15] = x
FIND THE BASIC BLOCK

1 i=m-1 16 t7 = 4 * I
2 j=n 17 t8 = 4 * j

26-Oct-20
3 t1 =4 * n 18 t9 = a[t8]
4 v = a[t1] 19 a[t7] = t9

VIIT
Manisha Mali Computer Engg. Dept.,
5 i=i +1 20 t10 = 4 * j
6 t2 = 4 * i 21 a[t10] = x
7 t3 = a[t2] 22 goto (5)
8 if t3 < v goto (5) 23 t11 = 4 * i
9 j=j–1 24 x = a[t11]
10 t4 = 4 * j 25 t12 = 4 * i
11 t5 = a[t4] 26 t13 = 4 * n
12 if t5 > v goto (9) 27 t14 = a[t13]
13 if i >= j goto (23) 28 a[t12] = t14
14 t6 = 4 * i 29 t15 = 4 * n
37
15 x = a[t6] 30 a[t15] = x
B1 FLOW GRAPH
i=m-1
j=n
t1 =4 * n

26-Oct-20
v = a[t1]
B5 B6
B2 t6 = 4 * i t11 = 4 * i

VIIT
Manisha Mali Computer Engg. Dept.,
i=i +1 x = a[t6] x = a[t11]
t2 = 4 * i t7 = 4 * i t12 = 4 * i
t3 = a[t2] t8 = 4 * j t13 = 4 * n
if t3 < v goto B2 t9 = a[t8] t14 = a[t13]
B3 a[t7] = t9 a[t12] = t14
j=j–1 t10 = 4 * j t15 = 4 * n
t4 = 4 * j
a[t10] = x a[t15] = x
t5 = a[t4]
goto B2
if t5 > v goto B3

B4
38
if i >= j goto B6
LOCAL VS. GLOBAL
TRANSFORMATIONS

26-Oct-20
 Optimizations within a basic block is called local
optimization.
 Optimizations across basic blocks is called global

VIIT
Manisha Mali Computer Engg. Dept.,
optimization.
 Local transformations involve statements
within a single basic block
 All other transformations are called global
transformations
 Local transformations are generally
performed first
 Many types of transformations can be
performed either locally or globally 39

Manisha Mali
COMMON SUBEXPRESSIONS

26-Oct-20
 E is a common subexpression if:
– E was previously computed

VIIT
Manisha Mali Computer Engg. Dept.,
– Variables in E have not changed since previous
computation
 Can avoid recomputing E if previously
computed value is still available
 DAGs are useful to detect common
subexpressions

40

Manisha Mali
B1 COMMON SUBEXPRESSION ELIMINATION
i=m-1
j=n
t1 =4 * n

26-Oct-20
v = a[t1]
B5 B6
B2 t6 = 4 * i t11 = 4 * i

VIIT
Manisha Mali Computer Engg. Dept.,
i=i +1 x = a[t6] x = a[t11]
t2 = 4 * i t7 = 4 * i t12 = 4 * i
t3 = a[t2] t8 = 4 * j t13 = 4 * n
if t3 < v goto B2 t9 = a[t8] t14 = a[t13]
B3 a[t7] = t9 a[t12] = t14
j=j–1 t10 = 4 * j t15 = 4 * n
t4 = 4 * j
a[t10] = x a[t15] = x
t5 = a[t4]
goto B2
if t5 > v goto B3

B4
41
if i >= j goto B6
B1 COMMON SUBEXPRESSION ELIMINATION
i=m-1
j=n
t1 =4 * n

26-Oct-20
v = a[t1]
B5 B6
B2 t6 = 4 * i t11 = 4 * i

VIIT
Manisha Mali Computer Engg. Dept.,
i=i +1 x = a[t6] x = a[t11]
t2 = 4 * i t8 = 4 * j t12 = 4 * i
t3 = a[t2] t9 = a[t8] t13 = 4 * n
if t3 < v goto B2 a[t6] = t9 t14 = a[t13]
B3 t10 = 4 * j a[t12] = t14
j=j–1 a[t10] = x t15 = 4 * n
t4 = 4 * j
goto B2 a[t15] = x
t5 = a[t4]
if t5 > v goto B3

B4
42
if i >= j goto B6
B1 COMMON SUBEXPRESSION ELIMINATION
i=m-1
j=n
t1 =4 * n

26-Oct-20
v = a[t1]
B5 B6
B2 t6 = 4 * i t11 = 4 *i

VIIT
Manisha Mali Computer Engg. Dept.,
i=i +1 x = a[t6] x = a[t11]
t2 = 4 * i t8 = 4 * j t12 = 4 * i
t3 = a[t2] t9 = a[t8] t13 = 4 * n
if t3 < v goto B2 a[t6] = t9 t14 = a[t13]
B3 a[t8] = x a[t12] = t14
j=j–1 goto B2 t15 = 4 * n
t4 = 4 * j
a[t15] = x
t5 = a[t4]
if t5 > v goto B3

B4
43
if i >= j goto B6
B1 COMMON SUBEXPRESSION ELIMINATION
i=m-1
j=n
t1 =4 * n

26-Oct-20
v = a[t1]
B5 B6
B2 t6 = 4 * i t11 = 4 * i

VIIT
Manisha Mali Computer Engg. Dept.,
i=i +1 x = a[t6] x = a[t11]
t2 = 4 * i t8 = 4 * j t12 = 4 * i
t3 = a[t2] t9 = a[t8] t13 = 4 * n
if t3 < v goto B2 a[t6] = t9 t14 = a[t13]
B3 a[t8] = x a[t12] = t14
j=j–1 goto B2 t15 = 4 * n
t4 = 4 * j
a[t15] = x
t5 = a[t4]
if t5 > v goto B3

B4
44
if i >= j goto B6
B1 COMMON SUBEXPRESSION ELIMINATION
i=m-1
j=n
t1 =4 * n

26-Oct-20
v = a[t1]
B5 B6
B2 t6 = 4 * i t11 = 4 * i

VIIT
Manisha Mali Computer Engg. Dept.,
i=i +1 x = a[t6] x = a[t11]
t2 = 4 * i t8 = 4 * j t13 = 4 * n
t3 = a[t2] t9 = a[t8] t14 = a[t13]
if t3 < v goto B2 a[t6] = t9 a[t11] = t14
B3 a[t8] = x t15 = 4 * n
j=j–1 goto B2 a[t15] = x
t4 = 4 * j
t5 = a[t4]
if t5 > v goto B3

B4
45
if i >= j goto B6
B1 COMMON SUBEXPRESSION ELIMINATION
i=m-1
j=n
t1 =4 * n

26-Oct-20
v = a[t1]
B5 B6
B2 t6 = 4 * i
t11 = 4 * i

VIIT
Manisha Mali Computer Engg. Dept.,
i=i +1 x = a[t6]
x = a[t11]
t2 = 4 * i t8 = 4 * j
t13 = 4 * n
t3 = a[t2] t9 = a[t8]
t14 = a[t13]
if t3 < v goto B2 a[t6] = t9
a[t11] = t14
B3 a[t8] = x
a[t13] = x
j=j–1 goto B2
t4 = 4 * j
t5 = a[t4]
if t5 > v goto B3

B4
46
if i >= j goto B6
B1 COMMON SUBEXPRESSION ELIMINATION
i=m-1
j=n
t1 =4 * n

26-Oct-20
v = a[t1]
B5 B6
B2 t6 = 4 * i
t11 = 4 * i

VIIT
Manisha Mali Computer Engg. Dept.,
i=i +1 x = a[t6]
x = a[t11]
t2 = 4 * i t8 = 4 * j
t13 = 4 * n
t3 = a[t2] t9 = a[t8]
t14 = a[t13]
if t3 < v goto B2 a[t6] = t9
a[t11] = t14
B3 a[t8] = x
a[t13] = x
j=j–1 goto B2
t4 = 4 * j
t5 = a[t4]
if t5 > v goto B3

B4
47
if i >= j goto B6
B1 COMMON SUBEXPRESSION ELIMINATION
i=m-1
j=n
t1 =4 * n

26-Oct-20
v = a[t1]
B5 B6
B2 x = a[t2]
t11 = 4 * i

VIIT
Manisha Mali Computer Engg. Dept.,
i=i +1 t8 = 4 * j
x = a[t11]
t2 = 4 * i t9 = a[t8]
t13 = 4 * n
t3 = a[t2] a[t2] = t9
t14 = a[t13]
if t3 < v goto B2
a[t8] = x
a[t11] = t14
B3 goto B2
a[t13] = x
j=j–1
t4 = 4 * j
t5 = a[t4]
if t5 > v goto B3

B4
48
if i >= j goto B6
B1 COMMON SUBEXPRESSION ELIMINATION
i=m-1
j=n
t1 =4 * n

26-Oct-20
v = a[t1]
B5 B6
B2 x = t3
t11 = 4 * i

VIIT
Manisha Mali Computer Engg. Dept.,
i=i +1 t8 = 4 * j
x = a[t11]
t2 = 4 * i t9 = a[t8]
t13 = 4 * n
t3 = a[t2] a[t2] = t9
t14 = a[t13]
if t3 < v goto B2
a[t8] = x
a[t11] = t14
B3 goto B2
a[t13] = x
j=j–1
t4 = 4 * j
t5 = a[t4]
if t5 > v goto B3

B4
49
if i >= j goto B6
B1 COMMON SUBEXPRESSION ELIMINATION
i=m-1
j=n
t1 =4 * n

26-Oct-20
v = a[t1]
B5 B6
B2 x = t3
t11 = 4 * i

VIIT
Manisha Mali Computer Engg. Dept.,
i=i +1 t9 = a[t4]
x = a[t11]
t2 = 4 * i a[t2] = t9
t13 = 4 * n
t3 = a[t2] a[t4] = x
t14 = a[t13]
if t3 < v goto B2
goto B2
a[t11] = t14
B3
a[t13] = x
j=j–1
t4 = 4 * j
t5 = a[t4]
if t5 > v goto B3

B4
50
if i >= j goto B6
B1 COMMON SUBEXPRESSION ELIMINATION
i=m-1
j=n
t1 =4 * n

26-Oct-20
v = a[t1]
B5 B6
B2 x = t3
t11 = 4 * i

VIIT
Manisha Mali Computer Engg. Dept.,
i=i +1 a[t2] = t5
x = a[t11]
t2 = 4 * i a[t4] = x
t13 = 4 * n
t3 = a[t2] goto B2
t14 = a[t13]
if t3 < v goto B2
a[t11] = t14
B3
a[t13] = x
j=j–1
t4 = 4 * j
t5 = a[t4]
if t5 > v goto B3

B4
51
if i >= j goto B6
B1 COMMON SUBEXPRESSION ELIMINATION
i=m-1
j=n
t1 =4 * n

26-Oct-20
v = a[t1]
B5 B6
B2 x = t3
x = t3

VIIT
Manisha Mali Computer Engg. Dept.,
i=i +1 a[t2] = t5
t14 = a[t1]
t2 = 4 * i a[t4] = x
a[t2] = t14
t3 = a[t2] goto B2
a[t1] = x
if t3 < v goto B2

B3
j=j–1
Similarly for B6
t4 = 4 * j
t5 = a[t4]
if t5 > v goto B3

B4
52
if i >= j goto B6
COPY PROPAGATION
 Assignments of the form f := g are called copy

26-Oct-20
statements (or copies)
 The idea behind copy propagation is to use g
for f whenever possible after such a statement

VIIT
Manisha Mali Computer Engg. Dept.,
 For example, applied to block B5 of the
previous flow graph, we obtain:
x := t3
a[t2] := t5
a[t4] := t3
goto B2
 Copy propagation often turns the copy
statement into "dead code" 53

Manisha Mali
COPY PROPAGATION

26-Oct-20
f := g are called copy statements or copies
 Use of g for f, whenever possible after copy

VIIT
Manisha Mali Computer Engg. Dept.,
statement

Example:
x[i] = a; x[i] = a;
sum = x[i] + a; sum = a + a;

 Maynot appear to be code improvement, but

opens up scope for other optimizations.
54
COPY PROPAGATION (CONT’D)

26-Oct-20
 Copies introduced during common subexpression
elimination

VIIT
Manisha Mali Computer Engg. Dept.,
55

Manisha Mali
DEAD CODE ELIMINATION
A variable is live at a point in a program if its
value can be used subsequently; otherwise, it is

26-Oct-20
dead at that point.
 Dead code includes code that can never be

VIIT
Manisha Mali Computer Engg. Dept.,
reached and
code that computes a value that never gets used
 Deducing at compile time that the value of an
expression is a constant and using the constant
instead is known as constant folding
 Data flow analysis

 E.g. debug = false;

if (debug) printf(“Welcome”); 56

Manisha Mali
B1 DEAD CODE ELIMINATION
i=m-1
j=n
t1 =4 * n

26-Oct-20
v = a[t1]
B5 B6
B2 x = t3
x = t3

VIIT
Manisha Mali Computer Engg. Dept.,
i=i +1 a[t2] = t5
t14 = a[t1]
t2 = 4 * i a[t4] = x
a[t2] = t14
t3 = a[t2] goto B2
a[t1] = x
if t3 < v goto B2

B3
j=j–1
t4 = 4 * j
t5 = a[t4]
if t5 > v goto B3

B4
57
if i >= j goto B6
B1 DEAD CODE ELIMINATION
i=m-1
j=n
t1 =4 * n

26-Oct-20
v = a[t1]
B5 B6
B2 a[t2] = t5 t14 = a[t1]

VIIT
Manisha Mali Computer Engg. Dept.,
a[t4] = t3 a[t2] = t14
i=i +1
t2 = 4 * i goto B2 a[t1] = t3
t3 = a[t2]
if t3 < v goto B2

B3
j=j–1
t4 = 4 * j
t5 = a[t4]
if t5 > v goto B3

B4
58
if i >= j goto B6
 LOOP OPTIMIZATIONS
 Induction variables: variables that remain in
"lock-step"
– For example, in block B3 of previous flow graph, j and

26-Oct-20
t4 are induction variables
– Induction-variable elimination can sometimes
eliminate all but one of a set of induction variables

VIIT
Manisha Mali Computer Engg. Dept.,
 Reduction in strength replaces a more expensive
operation with a less expensive one
– For example, in block B3, t4 decreases by four with
every iteration
– If initialized correctly, can replace multiplication with
subtraction
– Often application of reduction in strength leads to
induction-variable elimination
 Methods exist to recognize induction variables and apply
appropriate transformations automatically 59

Manisha Mali
 CODE MOTION
 An important modification that decreases the
amount of code in a loop

26-Oct-20
 The running time of a program may be improved if we do
both of the following:
– Decrease the number of statements in an inner loop

VIIT
Manisha Mali Computer Engg. Dept.,
– Increase the number of statements in the outer loop
 Code motion moves code outside of a loop

 E.g while ( i <= limit – 2 ) /* stmt does not change limit */

Code motion will result in the equivalent result of
t = limit – 2;
While (i <= t) /* stmt does not change limit or t */
60

Manisha Mali
STRENGTH REDUCTION

26-Oct-20
 Replacement of an operator with a less costly one.
Example:

VIIT
Manisha Mali Computer Engg. Dept.,
temp = 5;
for i=1 to 10 do for i=1 to 10 do
… …
x=i*5 x = temp
… …
temp = temp + 5
end end
• Typical cases of strength reduction occurs in address
calculation of array references.
• Applies to integer expressions involving induction
variables (loop optimization)
61
B1 REDUCTION IN STRENGTH
i=m-1
j=n
t1 =4 * n

26-Oct-20
v = a[t1]
B5 B6
B2 a[t2] = t5 t14 = a[t1]

VIIT
Manisha Mali Computer Engg. Dept.,
a[t4] = t3 a[t2] = t14
i=i +1
t2 = 4 * i goto B2 a[t1] = t3
t3 = a[t2]
if t3 < v goto B2

B3
j=j–1
t4 = 4 * j
t5 = a[t4]
if t5 > v goto B3

B4
62
if i >= j goto B6
B1 REDUCTION IN STRENGTH
i=m-1
j=n
t1 =4 * n

26-Oct-20
v = a[t1]
B5 B6
t2 = 4 * i
B2 t4 = 4 * j
a[t2] = t5 t14 = a[t1]

VIIT
Manisha Mali Computer Engg. Dept.,
a[t4] = t3 a[t2] = t14
goto B2 a[t1] = t3
t2 = t2 + 4
t3 = a[t2]
if t3 < v goto B2
B3

t4 = t4 - 4
t5 = a[t4]
if t5 > v goto B3

B4
63
if i >= j goto B6
EXERCISE

26-Oct-20
Read following piece of code: [4 Mks]
B1: j = j-1

VIIT
Manisha Mali Computer Engg. Dept.,
t4 = 4*j
t5 = a[t4]
if t5>j go to B1
which optimization technique can be applied to
this code? Explain it in detail.

64
EXERCISE 3
Consider the following code that produces a 10*10 identity
matrix.[12 Mks]
for (i=1; i=10; i++)
{

26-Oct-20
for (j=1; j=10; j++)
{
A[i, j] = 0;

VIIT
Manisha Mali Computer Engg. Dept.,
}}
for (i=1; i=10; i++)
{
A[i, i]=1;
}
Construct the
i) Three address code.
ii) Basic blocks.
iii) Flow graph.

 What kind of code optimization can you suggest for the above 65
code apply the same on the given code? [10 Mks]
 EXERCISE 4
Perform the following the code fragment given below. [10 mks]
for(i=1;i=n;i++)
{
for(j=1;j=n;j++)

26-Oct-20
{
C[i,j]=0;

VIIT
Manisha Mali Computer Engg. Dept.,
}}
for(i=1;i=n;i++)
{
for(j=1;j=n;j++)
{
For(k=1;k=n;k++)
{
C[i,j]=C[i,j] + A[i, k]*B[k,j];
}}}

i) Partition into basic block

66
ii) Find loops in basic block
iii) Move the loop invariants outside the loop.
 REFERENCES
 D. M. Dhamdhere, Systems Programming and Operating
Systems, Tata McGraw-Hill, ISBN 13:978-0-07-463579-7,
Second Revised Edition

Engineering, VIIT , Pune-48

Manisha Mali, Department of Computer
 Alfred V. Aho, Ravi Sethi, Jeffrey D. Ullman, Compilers
Principles, Techniques and Tools, Addison Wesley,
ISBN:981–235–885 - 4, Low Price Edition
 J. J. Donovan, Systems Programming, McGraw-Hill, ISBN
13:978-0-07-460482-3, Indian Edition

67
 26-Oct-20
Manisha Mali, Dept. Computer Engg., VIIT
68