Loop optimizations

• Optimize loops
– Loop invariant code motion [last time]
Loop Optimizations – Strength reduction of induction variables
and Pointer Analysis – Induction variable elimination

CS 412/413 Spring 2008 Introduction to Compilers 1 CS 412/413 Spring 2008 Introduction to Compilers 2

Strength Reduction Induction Variables

• Basic idea: replace expensive operations (multiplications) with • An induction variable is a variable in a loop,
cheaper ones (additions) in definitions of induction variables
whose value is a function of the loop iteration
s = 3*i+1; number v = f(i)
while (i<10) { while (i<10) {
j = 3*i+1; //<i,3,1> j = s;
• In compilers, this a linear function:
a[j] = a[j] –2; a[j] = a[j] –2;
i = i+2; i = i+2; f(i) = c*i + d
} s= s+6;
} • Observation: linear combinations of linear
• Benefit: cheaper to compute s = s+6 than j = 3*i functions are linear functions
– s = s+6 requires an addition – Consequence: linear combinations of induction
– j = 3*i requires a multiplication variables are induction variables

CS 412/413 Spring 2008 Introduction to Compilers 3 CS 412/413 Spring 2008 Introduction to Compilers 4

1
 Families of Induction Variables Representation
• Basic induction variable: a variable whose only definition in the • Representation of induction variables in family i by triples:
loop body is of the form
– Denote basic induction variable i by <i, 1, 0>
i=i+c
– Denote induction variable k=i*a+b by triple <i, a, b>
where c is a loop-invariant value

• Derived induction variables: Each basic induction variable i defines

a family of induction variables Family(i)
– i  Family(i)
– k  Family(i) if there is only one definition of k in the loop body , and it
has the form k = c*j or k=j+c, where
(a) j  Family(i)
(b) c is loop invariant
(c) The only definition of j that reaches the definition of k is in the loop
(d) There is no definition of i between the definitions of j and k

CS 412/413 Spring 2008 Introduction to Compilers 5 CS 412/413 Spring 2008 Introduction to Compilers 6

Finding Induction Variables Strength Reduction

Scan loop body to find all basic induction variables • Basic idea: replace expensive operations (multiplications) with
cheaper ones (additions) in definitions of induction variables
do
Scan loop to find all variables k with one assignment of form k = s = 3*i+1;
j*b, where j is an induction variable <i,c,d>, and make k an while (i<10) { while (i<10) {
induction variable with triple <i,c*b,d> j = …; // <i,3,1> j = s;
a[j] = a[j] –2; a[j] = a[j] –2;
Scan loop to find all variables k with one assignment of form k =
jb where j is an induction variable with triple <i,c,d>, and
i = i+2; i = i+2;
make k an induction variable with triple <i,c,bd> } s= s+6;
}
until no more induction variables found
• Benefit: cheaper to compute s = s+6 than j = 3*i
– s = s+6 requires an addition
– j = 3*i requires a multiplication

CS 412/413 Spring 2008 Introduction to Compilers 7 CS 412/413 Spring 2008 Introduction to Compilers 8

2
 General Algorithm Strength Reduction
• Algorithm: • Gives opportunities for copy propagation, dead code
elimination
For each induction variable j with triple <i,a,b>
whose definition involves multiplication: s = 3*i+1; s = 3*i+1;
1. create a new variable s while (i<10) { while (i<10) {
2. replace definition of j with j=s j = s;
3. immediately after i=i+c, insert s = s+a*c a[j] = a[j] –2; a[s] = a[s] –2;
(here a*c is constant) i = i+2; i = i+2;
s= s+6; s= s+6;
4. insert s = a*i+b into preheader
} }
• Correctness: transformation maintains invariant s = a*i+b

CS 412/413 Spring 2008 Introduction to Compilers 9 CS 412/413 Spring 2008 Introduction to Compilers 10

Induction Variable Elimination Induction Variable Elimination

• Idea: eliminate each basic induction variable whose only uses
are in loop test conditions and in their own definitions i = i+c • Rewrite test condition using derived induction variables
- rewrite loop test to eliminate induction variable • Remove definition of basic induction variables (if not used
after the loop)
s = 3*i+1;
while (i<10) {
a[s] = a[s] –2; s = 3*i+1; s = 3*i+1;
i = i+2; while (i<10) { while (s<31) {
s= s+6; a[s] = a[s] –2; a[s] = a[s] –2;
} i = i+2; s= s+6;
• When are induction variables used only in loop tests? s= s+6; }
– Usually, after strength reduction }
– Use algorithm from strength reduction even if definitions
of induction variables don’t involve multiplications

CS 412/413 Spring 2008 Introduction to Compilers 11 CS 412/413 Spring 2008 Introduction to Compilers 12

3
 Induction Variable Elimination Where We Are
For each basic induction variable i whose only uses are • Defined dataflow analysis framework
– The test condition i < u
• Used it for several analyses
– The definition of i: i = i + c – Live variables
– Available expressions
• Take a derived induction variable k in family i, with – Reaching definitions
triple <i,c,d> – Constant folding
• Replace test condition i < u with k < c*u+d
• Remove definition i = i+c if i is not live on loop exit • Loop transformations
– Loop invariant code motion
– Induction variables

• Next:
– Pointer alias analysis

CS 412/413 Spring 2008 Introduction to Compilers 13 CS 412/413 Spring 2008 Introduction to Compilers 14

Pointer Alias Analysis Pointer Alias Analysis

• Most languages use variables containing addresses • Worst case scenarios
– E.g. pointers (C,C++), references (Java), call-by- – *p = y may write any memory location
reference parameters (Pascal, C++, Fortran) – x = *p may read any memory location
• Such assumptions may affect the precision of other analyses
• Pointer aliases: multiple names for the same memory
• Example1: Live variables
location, which occur when dereferencing variables that hold
memory addresses before any instruction x = *p, all the variables may be live

• Problem: • Example 2: Constant folding

a = 1; b = 2;*p = 0; c = a+b;
– Don’t know what variables read and written by accesses
via pointer aliases (e.g. *p=y; x=*p; p->f=y; x=p->f; • c = 3 at the end of code only if *p is not an alias for a or b!
etc.)
• Conclusion: precision of result for all other analyses depends
– Need to know accessed variables to compute dataflow on the amount of alias information available
information after each instruction - hence, it is a fundamental analysis
CS 412/413 Spring 2008 Introduction to Compilers 15 CS 412/413 Spring 2008 Introduction to Compilers 16

4
 Alias Analysis Problem Simple Alias Analyses
• Address-taken analysis:
• Goal: for each variable v that may hold an address, – Consider AT = set of variables whose addresses are taken
compute the set Ptr(v) of possible targets of v – Then, Ptr(v) = AT, for each pointer variable v
– Ptr(v) is a set of variables (or objects) – Addresses of heap variables are always taken at allocation
– Ptr(v) includes stack- and heap-allocated variables (objects) sites (e.g., x = new int[2]; x=malloc(8); )
– Hence AT includes all heap variables
• Is a “may” analysis: if x  Ptr(v), then v may hold the
• Type-based alias analysis:
address of x in some execution of the program
– If v is a pointer (or reference) to type T, then Ptr(v) is the
set of all variables of type T
• No alias information: for each variable v, Ptr(v) = V, – Example: p->f and q->f can be aliases only if p and q are
where V is the set of all variables in the program references to objects of the same type
– Works only for strongly-typed languages

CS 412/413 Spring 2008 Introduction to Compilers 17 CS 412/413 Spring 2008 Introduction to Compilers 18

Dataflow Alias Analysis Dataflow Alias Analysis

• Dataflow analysis: for each variable v, compute points- • Dataflow Lattice: (2 V x V, ⊇ )
to set Ptr(v) at each program point – V x V represents “every variable may point to every var.”
– “may” analysis: top element is , meet operation is ⋃
• Dataflow information: set Ptr(v) for each variable v
• Transfer functions: use standard dataflow transfer functions:
– Can be represented as a graph G ⊆ 2 VxV
out[I] = (in[I]-kill[I]) U gen[I]
– Nodes = V (program variables)
p = addr q kill[I]={p} x V gen[I]={<p,q>}
– There is an edge vu if u  Ptr(v)
p=q kill[I]={p} x V gen[I]={p} x Ptr(q)
p = *q kill[I]={p} x V gen[I]={p} x Ptr(Ptr(q))
Ptr(x) = {y} z
x y *p = q kill[I]= … gen[I]=Ptr(p) x Ptr(q)
Ptr(y) = {z,t} For all other instruction, kill[I] = {}, gen[I] = {}
t
• Transfer functions are monotonic, but not distributive!

CS 412/413 Spring 2008 Introduction to Compilers 19 CS 412/413 Spring 2008 Introduction to Compilers 20

5
 Alias Analysis Example Alias Analysis Uses
• Once alias information is available, use it in other
Program Points-to dataflow analyses
CFG
Graph
x=&a (at the end of program) • Example: Live variable analysis
x=&a; y=&b Use alias information to compute use[I] and def[I] for
y=&b; c=&i x a
load and store statements:
if(i) i
c=&i; y b
if(i) x=y; x = *y use[I] = {y} ⋃ Ptr(y) def[I]={x}
x=y
*x=c; c
*x = y use[I] = {x,y} def[I]=Ptr(x)
*x=c

CS 412/413 Spring 2008 Introduction to Compilers 21 CS 412/413 Spring 2008 Introduction to Compilers 22