Demand-Driven Context-Sensitive Alias Analysis for Java

Demand-Driven Context-Sensitive Alias
Analysis for Java

Dacong (Tony) Yan
Guoqing (Harry) Xu
Atanas Rountev
Ohio State University

PRESTO: Program Analyses and Software Tools Research Group, Ohio State University

Alias Analysis
• Many static analysis tools need highly precise and
efficiently-computed alias information
• Desirable properties
– Demand-driven: query “could x and y alias?”
– Client-driven: client-defined time budget for a query
• Typical solution: use points-to analysis
– Compute the objects that x may point to; same for y;
are there common objects?
• Goal: answer alias queries precisely and efficiently,
without computing the complete points-to sets

2

Redundancy in Points-to Analysis

3


m(a) {
b = a;
x = a.f;
y = b.f;
}

3


m(a) {
b = a;
x = a.f;
y = b.f;
alias?
x y
}

3


m(a) { O
b = a;
points-to? points-to?
x = a.f;
y = b.f;
alias?
x y
}

3


m(a) {
b = a;
x = a.f;
y = b.f;
f f
x a b y
}

3


m(a) {
b = a;
x = a.f;
y = b.f;
f alias? f
x a b y
}

3


m(a) {
b = a; O
x = a.f;
points-to? points-to?
y = b.f;
f alias? f
x a b y
}

3


m(a) {
b = a;
x = a.f;
y = b.f; alias
f f
x a b y
}

3


m(a) {
b = a;
x = a.f;
y = b.f; alias
f f
x a b y
}
alias
3

Our Approach
• Alias analysis
– Demand-driven and client-driven
– Field-sensitive and calling-context-sensitive
– Does not require complete points-to set computation
– Better performance through method summaries
• Symbolic Points-to Graph
– A specialized program representation that enables the
demand-driven alias analysis
– Facilitates computation and use of method summaries

4

Program Representation
• Intraprocedural Symbolic Points-To Graph
– Introduce a symbolic object node s for each
• formal parameter
• field read a.fld
• a call site that returns a reference-typed value
– Compute intraprocedural points-to relationships

m(a) {
c = new …; // o1 a sa o1 c
f
g
a.f = c;
return c.g; ret sg
}
5

Interprocedural Symbolic Points-To Graph
• Connect the intraprocedural graphs using entry
and exit edges

6

and exit edges
m(a) {
c = new …; // o1
a.f = c;
return c.g;
}

6

and exit edges
m(a) { f
a sa o1 c
c = new …; // o1 g
a.f = c;
ret sg
return c.g;
}

6

and exit edges
m(a) { f
a sa o1 c
c = new …; // o1 g
a.f = c;
ret sg
return c.g;
}

d = new …; // o2
b = m(d); // call m

6

and exit edges
m(a) { f
a sa o1 c
c = new …; // o1 g
a.f = c;
ret sg
return c.g;
}
o2 d
d = new …; // o2
b sm
b = m(d); // call m

6

and exit edges
m(a) { f
a sa o1 c
c = new …; // o1 g
a.f = c;
entrym ret sg
return c.g;
exitm
}
o2 d
d = new …; // o2
b sm
b = m(d); // call m

6

Alias Analysis Formulation
• Field-sensitivity and calling-context-sensitivity:
matched points-to edges and entry/exit edges
• Traverses the object nodes (including symbolic
ones) and the edges between them

7

a o1 s3 c
f f
b o2 s2 d
entry1 exit1
s1

7

a o1 s3 c
f f
b o2 s2 d
entry1 exit1
s1

7
a alias? c

a o1 s3 c
f f
b o2 s2 d
entry1 exit1
s1
may represent the
o1 s3
7 same concrete object?

a o1 s3 c
f f
b o2 s2 d
entry1 exit1
s1

o1 f o2 s3
7

a o1 s3 c
f f
b o2 s2 d
entry1 exit1
s1

o1 f o entry1 s s3
2 1
7

a o1 s3 c
f f
b o2 s2 d
entry1 exit1
s1

o1 f o entry1 s exit1 s s3
2 1 2
7

a o1 s3 c
f f
b o2 s2 d
entry1 exit1
s1
f o entry1, exit1
o1 2 s2 s3
7

a o1 s3 c
f f
b o2 s2 d
entry1 exit1
s1
f o entry1, exit1
o1 2 s2 f s3
7

a o1 s3 c
f f
b o2 s2 d
entry1 exit1
s1
f, entry1, exit1, f
o1 s3
7

Using Method Summaries
• Several reachability strings for a selected method
– Each string is a sequence of field points-to edges
between boundary objects of the method

8

m(a) { f
a sa o1 c
c = new …; // o1 g
a.f = c; sg
ret
return c.g;
}

8

m(a) { f
a sa o1 c
c = new …; // o1 g
a.f = c; sg
ret
return c.g;
}
f, g
summary(m): sa sg
8

• Selecting methods for summarization
– Compute a summary for a method only if it is invoked
from many different call sites

8

• Selecting methods for summarization
– Compute a summary for a method only if it is invoked
from many different call sites
– Summarization ratio for method m
• the number of incoming call graph edges of m,
divided by the average number of incoming call
graph edges for all methods

8

Experimental Evaluation
• 19 Java programs
– Number of methods in the whole-program call graph
ranging from 2344 to 8789
• Experiment 1: compare the precision with a field
and context-sensitive, demand-driven, client-
driven points-to analysis [PLDI’06]
– Under the same time budget per alias query
– Result 1: for 14 programs, the number of alias pairs is
lower when using our analysis
– Result 2: summarization leads to better precision
• Experiment 2: compare running times with and
without method summaries
9

Running Time Reduction When Using
Method Summaries

Running Time Reduction = (RTno-summ  RTsumm) / RTno-summ

10

Running Time Reduction When Using
Method Summaries

Running Time Reduction = (RTno-summ  RTsumm) / RTno-summ

24% average reduction with threshold T=2
10

Conclusions
• A demand-driven alias analysis
– Answers alias queries directly, without computing the
complete points-to sets
– Selects methods for online summarization to reduce
analysis running time
– Outperforms a highly precise state-of-the-art points-to
analysis

11

Demand-Driven Context-Sensitive Alias Analysis for Java

More Related Content

What's hot

Viewers also liked

Similar to Demand-Driven Context-Sensitive Alias Analysis for Java

More from Dacong (Tony) Yan

Recently uploaded

Demand-Driven Context-Sensitive Alias Analysis for Java