Presentation is loading. Please wait.

Presentation is loading. Please wait.

Pointer and Shape Analysis Seminar Context-sensitive points-to analysis: is it worth it? Article by Ondřej Lhoták & Laurie Hendren from McGill University.

Similar presentations


Presentation on theme: "Pointer and Shape Analysis Seminar Context-sensitive points-to analysis: is it worth it? Article by Ondřej Lhoták & Laurie Hendren from McGill University."— Presentation transcript:

1 Pointer and Shape Analysis Seminar Context-sensitive points-to analysis: is it worth it? Article by Ondřej Lhoták & Laurie Hendren from McGill University Presentation by Roza Pogalnikova

2 Pointer and Shape Analysis Seminar 28/02/082 Abstract Evaluate precision of subset-based points-to analysis Compare different context-sensitivity approaches:  call site strings  object sensitivity  algorithm by Zhu and Calman, Whaley and Lam (ZCWL)‏

3 Pointer and Shape Analysis Seminar 28/02/083 Subset-based PTA Finding allocation sites that reach variable:  S: a = new A() // allocation statement  for variable x somewhere in the program: can it point to object allocated at S?

4 Pointer and Shape Analysis Seminar 28/02/084 Context Sensitivity Call site: by program statement of method invocation Object sensitivity: by receiving object of method invocation ZCWL: k-CFA, where k is call graph depth without SCCs Run context-insensitive algorithm on cloned context-sensitive call graph. S: this->call_method()‏

5 Pointer and Shape Analysis Seminar 28/02/085 Parameters Include:  specialize only pointer variables  use heap abstraction as well Different lengths of context strings

6 Pointer and Shape Analysis Seminar 28/02/086 Measurements Measure to guide implementation:  number of contexts  number of distinct contexts  number of distinct point-to sets Measure to evaluate:  size of the call graph (methods/edges)‏  devirtualizable call sites  casts statically provable to be safe

7 Pointer and Shape Analysis Seminar 28/02/087 Results Object sensitivity is the best and most scalable Heap abstraction improves precision of analysis Reduced analysis precision when no context sensitivity call graph in cycles

8 Pointer and Shape Analysis Seminar 28/02/088 What Compare three kinds of context-sensitive points-to analysis:  call sites as context abstraction  object-sensitive analysis  ZCWL algorithm

9 Pointer and Shape Analysis Seminar 28/02/089 How Implemented with JEDD system:  language extension of Java  abstraction of work with Binary Decision Diagrams (BDDs)‏  Soot framework written in JEDD: points-to analysis call graph construction side-effect analysis in BDDs virtual call resolution

10 Pointer and Shape Analysis Seminar 28/02/0810 BDDs Binary decision tree and truth table for the function f(x1, x2, x3) = -x1 * -x2 * -x3 + x1 * x2 + x2 * x3 BDD for the function f * credit: http://en.wikipedia.org/wiki/Binary_decision_diagram

11 Pointer and Shape Analysis Seminar 28/02/0811 PTA using BDDs Program: A: a = new O() B: b = new O() C: c = new O() a = b b = a c = b Points-to: (a, A) (b, B) (c, C) (a, B) (b, A) (c, A), (c, B)

12 Pointer and Shape Analysis Seminar 28/02/0812 PTA using BDDs Binary representation:  a & A as 00  b & B as 01  c & C as 10 Points-to representation: (a, A) as 0000 (a, B) as 0001 (b, A) as 0100 (b, B) as 0101 (c, A) as 1000 (c, B) as 1001 (c, C) as 1010

13 Pointer and Shape Analysis Seminar 28/02/0813 PTA using BDDs Compact way to represent points-to relations: * credit: [2] Points-to Analysis using BDDs

14 Pointer and Shape Analysis Seminar 28/02/0814 Determine How many contexts generalized? How number of contexts relates to precision of analysis? How likely scalable solution to be feasible?

15 Pointer and Shape Analysis Seminar 28/02/0815 Background O - pointer targets (objects)‏ P – pointers I – method invocation p may point to o: O(o) pt(P(p))‏

16 Pointer and Shape Analysis Seminar 28/02/0816 Background O as – program statement where object was allocated P var - pointer to local variable [O(o), f] - field f of object o P fs (o.f) – pointer to a field f of object o

17 Pointer and Shape Analysis Seminar 28/02/0817 Background Compare 2 families of invocation abstraction:  call site I cs (i) (program statement of metacall)‏  receiver object I ro (i) = O(o) (object on which method was invoked)

18 Pointer and Shape Analysis Seminar 28/02/0818 Background String of contexts given base abstraction I base : I string (i) = [I base (i), I base (i 2 ), I base (i 3 ),...] i j is a j'th topmost invocation on stack during i (i = i 1 )‏ Two approaches to make it finite:  define limit k to length of context string  ZCWL: exclude cycle edges from call graph

19 Pointer and Shape Analysis Seminar 28/02/0819 Background Another choice: which pointers/objects to model context-sensitively? Given context-insensitive P ci and context I model run-time pointer p:  context-sensitively by P(p) = [I(i p ), P ci (p)] (i p method invocation with p)‏  context-insensitively by P(p) = P ci (p)‏

20 Pointer and Shape Analysis Seminar 28/02/0820 Background Given allocation site abstraction O as, and context I model object o:  context-sensitively by O(o) = [I(i o ), O as (o)] (i o method invocation where o was allocated)‏  context insensitively by O(o) = O as (o)‏

21 Pointer and Shape Analysis Seminar 28/02/0821 Benchmarks The study was performed on:  SpecJVM 98 benchmark suite  DaCapo benchmark suite (ver. beta050224)‏  Ashes benchmark suite  Polyglot extensible Java front-end SUN standard library 1.3.1_01

22 Pointer and Shape Analysis Seminar 28/02/0822 Benchmarks

23 Pointer and Shape Analysis Seminar 28/02/0823 Contexts Number Considered intractable:  propagate context from call site to called method  context strings number grows exponentially in the length of call chains

24 Pointer and Shape Analysis Seminar 28/02/0824 Contexts Number Clarify next issues:  how many of these contexts improve analysis results?  why BDDs can represent such number, and is there hope to represent it with traditional techniques?

25 Pointer and Shape Analysis Seminar 28/02/0825 Total contexts number Count method-context pairs Empty spots – analysis not completed with available memory BDD lib. could allocate 41 million BDD nodes (~820 MB)‏

26 Pointer and Shape Analysis Seminar 28/02/0826 Total contexts number

27 Pointer and Shape Analysis Seminar 28/02/0827 Total contexts number Explicit context representation not scaling good Contexts number grows slowly in object- sensitive (this pointer method invocations)‏ ZCWL  k is max call depth in the call graph after merging SCCs  big variations because k different for each benchmark

28 Pointer and Shape Analysis Seminar 28/02/0828 Equivalent contexts Method-context pairs (m 1, c 1 ) and (m 2, c 2 ) are equivalent if:  m 1 = m 2  ∀ local pointer p in the method, pt(P(p)) is the same for c 1 and c 2 Equivalence classes reflect precision improvement due to context sensitivity

29 Pointer and Shape Analysis Seminar 28/02/0829 Equivalent contexts

30 Pointer and Shape Analysis Seminar 28/02/0830 Equivalent contexts BDD “automatically” merges equal points-to relations, i. e. is effective Object-sensitive vs. call sites – more precise Context string length does not have great impact Surprisingly ZCWL is less precise due to context-insensitivity in SCCs

31 Pointer and Shape Analysis Seminar 28/02/0831 Distinct points-to sets Measures analysis cost Approximates space requirements in “traditional”representation, like shared bit- vectors Similar results for all context-sensitive variations Increase in distinct point-to sets with context- sensitive heap abstraction

32 Pointer and Shape Analysis Seminar 28/02/0832 Distinct points-to sets

33 Pointer and Shape Analysis Seminar 28/02/0833 Call Graph Compare context-insensitive projection of context-sensitive call graphs  each node is method (and not method-context pair)  reachable methods preserved  ZCWL excluded (same as input context-insensitive graph)‏

34 Pointer and Shape Analysis Seminar 28/02/0834 Reachable methods

35 Pointer and Shape Analysis Seminar 28/02/0835 Reachable methods Context-sensitivity discovers more unreachable methods (bloat)‏ Context-sensitivity for heap objects:  In object-sensitive adds precision (sablecc-j)‏  In call site no impact

36 Pointer and Shape Analysis Seminar 28/02/0836 Call edges

37 Pointer and Shape Analysis Seminar 28/02/0837 Call edges Compare size of call graph in call edges The same with exception of large difference in sablecc-j (specific code pattern)‏

38 Pointer and Shape Analysis Seminar 28/02/0838 Virtual call resolution Number of virtual calls with more then one implementation Object-sensitive analysis has clear advantage over call site.  heap objects add precision (sablecc-j)‏

39 Pointer and Shape Analysis Seminar 28/02/0839 Virtual call resolution

40 Pointer and Shape Analysis Seminar 28/02/0840 Cast safety Cast cannot fail if pointer can point-to only to object of “right” type (sub-type of the type in cast)‏ Count non-provable casts Object-sensitivity, especially with heap objects is the best (polyglot, javac)

41 Pointer and Shape Analysis Seminar 28/02/0841 Cast safety

42 Pointer and Shape Analysis Seminar 28/02/0842 Conclusions Context-sensitive variations:  object-sensitive analysis  call sites as context abstraction  ZCWL algorithm Evaluated effects:  generated contexts  distinct point-to sets  precision of call graph construction  virtual call resolution  cast safety analysis

43 Pointer and Shape Analysis Seminar 28/02/0843 Conclusions Context-sensitivity improvements:  small: call graph precision  medium: virtual call resolution  major: cast safety analysis Object-sensitive analysis was the best:  analysis precision  potential scalability

44 Pointer and Shape Analysis Seminar 28/02/0844 Conclusions Object-sensitive variations improvements:  small: length of context strings  significant: heap objects with context  implementable with other existing techniques

45 Pointer and Shape Analysis Seminar 28/02/0845 Conclusions ZCWL algorithm:  disappointing results  caused by context-insensitive treatment of calls within SCCs of the initial graph  large proportion of edges in SCC


Download ppt "Pointer and Shape Analysis Seminar Context-sensitive points-to analysis: is it worth it? Article by Ondřej Lhoták & Laurie Hendren from McGill University."

Similar presentations


Ads by Google