Many resources are needed to download a project. Please understand that we have to compensate our server costs. Thank you in advance. Project price only 1 $
You can buy this project and download/modify it how often you want.
package soot.toolkits.graph.pdg;
/*-
* #%L
* Soot - a J*va Optimization Framework
* %%
* Copyright (C) 1999 - 2010 Hossein Sadat-Mohtasham
* %%
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU Lesser General Public License as
* published by the Free Software Foundation, either version 2.1 of the
* License, or (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Lesser Public License for more details.
*
* You should have received a copy of the GNU General Lesser Public
* License along with this program. If not, see
* .
* #L%
*/
import java.util.ArrayList;
import java.util.Enumeration;
import java.util.HashSet;
import java.util.Hashtable;
import java.util.Iterator;
import java.util.LinkedList;
import java.util.List;
import java.util.Queue;
import java.util.Set;
import soot.Body;
import soot.SootClass;
import soot.toolkits.graph.Block;
import soot.toolkits.graph.BlockGraph;
import soot.toolkits.graph.DominatorNode;
import soot.toolkits.graph.DominatorTree;
import soot.toolkits.graph.HashMutableEdgeLabelledDirectedGraph;
import soot.toolkits.graph.UnitGraph;
/**
*
* This class implements a Program Dependence Graph as defined in
*
* Ferrante, J., Ottenstein, K. J., and Warren, J. D. 1987. The program dependence graph and its use in optimization. ACM
* Trans. Program. Lang. Syst. 9, 3 (Jul. 1987), 319-349. DOI= http://doi.acm.org/10.1145/24039.24041
*
* Note: the implementation is not exactly as in the above paper. It first finds the regions of control dependence then uses
* part of the algorithm given in the above paper to build the graph.
*
* The constructor accepts a UnitGraph, which can be a BriefUnitGraph, an ExceptionalUnitGraph, or an EnhancedUnitGraph. At
* the absence of exception handling constructs in a method, all of these work the same. However, at the presence of
* exception handling constructs, BriefUnitGraph is multi-headed and potentially multi-tailed which makes the results of
* RegionAnalysis and PDG construction unreliable (It's not clear if it would be useful anyway); Also, ExceptionalGraph's
* usefulness when exception handling is present is not so clear since almost every unit can throw exception hence the
* dependency is affected. Currently, the PDG is based on a UnitGraph (BlockGraph) and does not care whether flow is
* exceptional or not.
*
* The nodes in a PDG are of type PDGNode and the edges can have three labels: "dependency", "dependency-back", and
* "controlflow"; however, the "controlflow" edges are auxiliary and the dependencies are represented by the labels beginning
* with "dependency". Other labels can be added later for application or domain-specific cases.
*
*
* To support methods that contain exception-handling and multiple-heads or tails, use EnhancedUnitGraph. It does not
* represent exceptional flow in the way ExceptionalUnitGraph does, but it integrates them in a concise way. Also, it adds
* START/STOP nodes to graph if necessary to make the graph single entry single exit.
*
*
* @author Hossein Sadat-Mohtasham Sep 2009
*/
public class HashMutablePDG extends HashMutableEdgeLabelledDirectedGraph implements ProgramDependenceGraph {
protected Body m_body = null;
protected SootClass m_class = null;
protected UnitGraph m_cfg = null;
protected BlockGraph m_blockCFG = null;
protected Hashtable m_obj2pdgNode = new Hashtable();
protected List m_weakRegions = null;
protected List m_strongRegions = null;
protected PDGNode m_startNode = null;
protected List m_pdgRegions = null;
private RegionAnalysis m_regionAnalysis = null;
private int m_strongRegionStartID;
public HashMutablePDG(UnitGraph cfg) {
this.m_body = cfg.getBody();
this.m_class = this.m_body.getMethod().getDeclaringClass();
this.m_cfg = cfg;
this.m_regionAnalysis = new RegionAnalysis(this.m_cfg, this.m_body.getMethod(), this.m_class);
/*
* Get the weak regions and save a copy. Note that the strong regions list is initially cloned from the weak region to be
* later modified.
*/
this.m_strongRegions = this.m_regionAnalysis.getRegions();
this.m_weakRegions = this.cloneRegions(this.m_strongRegions);
this.m_blockCFG = this.m_regionAnalysis.getBlockCFG();
// Construct the PDG
this.constructPDG();
this.m_pdgRegions = HashMutablePDG.computePDGRegions(this.m_startNode);
/*
* This is needed to convert the initially Region-typed inner node of the PDG's head to a PDGRegion-typed one after the
* whole graph is computed. The root PDGRegion is the one with no parent.
*/
IRegion r = this.m_pdgRegions.get(0);
while (r.getParent() != null) {
r = r.getParent();
}
this.m_startNode.setNode(r);
}
public BlockGraph getBlockGraph() {
return m_blockCFG;
}
/**
* This is the heart of the PDG contruction. It is huge and definitely needs some refactorings, but since it's been
* evlovong to cover some boundary cases it has become hard to rafactor.
*
* It uses the list of weak regions, along with the dominator and post-dominator trees to construct the PDG nodes.
*/
protected void constructPDG() {
Hashtable block2region = this.m_regionAnalysis.getBlock2RegionMap();
DominatorTree pdom = this.m_regionAnalysis.getPostDominatorTree();
DominatorTree dom = this.m_regionAnalysis.getDominatorTree();
List regions2process = new LinkedList();
Region topLevelRegion = this.m_regionAnalysis.getTopLevelRegion();
m_strongRegionStartID = m_weakRegions.size();
// This becomes the top-level region (or ENTRY region node)
PDGNode pdgnode = new PDGNode(topLevelRegion, PDGNode.Type.REGION);
this.addNode(pdgnode);
this.m_obj2pdgNode.put(topLevelRegion, pdgnode);
this.m_startNode = pdgnode;
topLevelRegion.setParent(null);
Set processedRegions = new HashSet();
regions2process.add(topLevelRegion);
// while there's a (weak) region to process
while (!regions2process.isEmpty()) {
Region r = regions2process.remove(0);
processedRegions.add(r);
// get the corresponding pdgnode
pdgnode = this.m_obj2pdgNode.get(r);
/*
* For all the CFG nodes in the region, create the corresponding PDG node and edges, and process them if they are in
* the dependence set of other regions, i.e. other regions depend on them.
*/
List blocks = r.getBlocks();
Hashtable> toBeRemoved = new Hashtable>();
PDGNode prevPDGNodeInRegion = null;
PDGNode curNodeInRegion = null;
for (Iterator itr = blocks.iterator(); itr.hasNext();) {
/*
* Add the PDG node corresponding to the CFG block node.
*/
Block a = itr.next();
PDGNode pdgNodeOfA = null;
if (!this.m_obj2pdgNode.containsKey(a)) {
pdgNodeOfA = new PDGNode(a, PDGNode.Type.CFGNODE);
this.addNode(pdgNodeOfA);
this.m_obj2pdgNode.put(a, pdgNodeOfA);
} else {
pdgNodeOfA = this.m_obj2pdgNode.get(a);
}
this.addEdge(pdgnode, pdgNodeOfA, "dependency");
pdgnode.addDependent(pdgNodeOfA);
//
curNodeInRegion = pdgNodeOfA;
/*
* For each successor B of A, if B does not post-dominate A, add all the nodes on the path from B to the L in the
* post-dominator tree, where L is the least common ancestor of A and B in the post-dominator tree (L will be either
* A itself or the parent of A.).
*/
List bs = this.m_blockCFG.getSuccsOf(a);
for (Iterator bItr = bs.iterator(); bItr.hasNext();) {
List dependents = new ArrayList();
Block b = bItr.next();
if (b.equals(a)) {
throw new RuntimeException("PDG construction: A and B are not supposed to be the same node!");
}
DominatorNode aDode = pdom.getDode(a);
DominatorNode bDode = pdom.getDode(b);
// If B post-dominates A, go to the next successor.
if (pdom.isDominatorOf(bDode, aDode)) {
continue;
}
// FIXME: what if the parent is null?!!
DominatorNode aParentDode = aDode.getParent();
DominatorNode dode = bDode;
while (dode != aParentDode) {
dependents.add((Block) dode.getGode());
// This occurs if the CFG is multi-tailed and therefore the pdom is a forest.
if (dode.getParent() == null) {
// throw new RuntimeException("parent dode in pdom is null: dode is " + aDode);
break;
}
dode = dode.getParent();
}
/*
* If node A is in the dependent list of A, then A is the header of a loop. Otherwise, A could still be the header
* of a loop or just a simple predicate.
*/
// first make A's pdg node be a conditional (predicate) pdgnode, if it's not already.
if (pdgNodeOfA.getAttrib() != PDGNode.Attribute.CONDHEADER) {
PDGNode oldA = pdgNodeOfA;
pdgNodeOfA = new ConditionalPDGNode(pdgNodeOfA);
this.replaceInGraph(pdgNodeOfA, oldA);
pdgnode.removeDependent(oldA);
this.m_obj2pdgNode.put(a, pdgNodeOfA);
pdgnode.addDependent(pdgNodeOfA);
pdgNodeOfA.setAttrib(PDGNode.Attribute.CONDHEADER);
curNodeInRegion = pdgNodeOfA;
}
List copyOfDependents = new ArrayList();
copyOfDependents.addAll(dependents);
// First, add the dependency for B and its corresponding region.
Region regionOfB = block2region.get(b);
PDGNode pdgnodeOfBRegion = null;
if (!this.m_obj2pdgNode.containsKey(regionOfB)) {
pdgnodeOfBRegion = new PDGNode(regionOfB, PDGNode.Type.REGION);
this.addNode(pdgnodeOfBRegion);
this.m_obj2pdgNode.put(regionOfB, pdgnodeOfBRegion);
} else {
pdgnodeOfBRegion = this.m_obj2pdgNode.get(regionOfB);
}
// set the region hierarchy
regionOfB.setParent(r);
r.addChildRegion(regionOfB);
// add the dependency edges
this.addEdge(pdgNodeOfA, pdgnodeOfBRegion, "dependency");
pdgNodeOfA.addDependent(pdgnodeOfBRegion);
if (!processedRegions.contains(regionOfB)) {
regions2process.add(regionOfB);
}
// now remove b and all the nodes in the same weak region from the list of dependents
copyOfDependents.remove(b);
copyOfDependents.removeAll(regionOfB.getBlocks());
/*
* What remains here in the dependence set needs to be processed separately. For each node X remained in the
* dependency set, find the corresponding PDG region node and add a dependency edge from the region of B to the
* region of X. If X's weak region contains other nodes not in the dependency set of A, create a new region for X
* and add the proper dependency edges (this actually happens if X is the header of a loop and B is a predicate
* guarding a break/continue.)
*
* Note: it seems the only case that there is a node remained in the dependents is when there is a path from b to
* the header of a loop.
*/
while (!copyOfDependents.isEmpty()) {
Block depB = copyOfDependents.remove(0);
Region rdepB = block2region.get(depB);
/*
* Actually, there are cases when depB is not the header of a loop and therefore would not dominate the current
* node (A) and therefore might not have been created yet. This has happened when an inner loop breaks out of the
* outer loop but could have other cases too.
*/
PDGNode depBPDGNode = this.m_obj2pdgNode.get(depB);
if (depBPDGNode == null) {
// First, add the dependency for depB and its corresponding region.
PDGNode pdgnodeOfdepBRegion = null;
if (!this.m_obj2pdgNode.containsKey(rdepB)) {
pdgnodeOfdepBRegion = new PDGNode(rdepB, PDGNode.Type.REGION);
this.addNode(pdgnodeOfdepBRegion);
this.m_obj2pdgNode.put(rdepB, pdgnodeOfdepBRegion);
} else {
pdgnodeOfdepBRegion = this.m_obj2pdgNode.get(rdepB);
}
// set the region hierarchy
rdepB.setParent(regionOfB);
regionOfB.addChildRegion(rdepB);
// add the dependency edges
this.addEdge(pdgnodeOfBRegion, pdgnodeOfdepBRegion, "dependency");
pdgnodeOfBRegion.addDependent(pdgnodeOfdepBRegion);
if (!processedRegions.contains(rdepB)) {
regions2process.add(rdepB);
}
// now remove all the nodes in the same weak region from the list of dependents
copyOfDependents.removeAll(rdepB.getBlocks());
continue;
}
/**
* If all the nodes in the weak region of depB are dependent on A, then add an edge from the region of B to the
* region of depB.
*
* else, a new region has to be created to contain the dependences of depB, if not already created.
*/
if (dependents.containsAll(rdepB.getBlocks())) {
/*
* Just add an edge to the pdg node of the existing depB region.
*
*/
// add the dependency edges
// First, add the dependency for depB and its corresponding region.
PDGNode pdgnodeOfdepBRegion = null;
if (!this.m_obj2pdgNode.containsKey(rdepB)) {
pdgnodeOfdepBRegion = new PDGNode(rdepB, PDGNode.Type.REGION);
this.addNode(pdgnodeOfdepBRegion);
this.m_obj2pdgNode.put(rdepB, pdgnodeOfdepBRegion);
} else {
pdgnodeOfdepBRegion = this.m_obj2pdgNode.get(rdepB);
}
// set the region hierarchy
// Do not set this because the region was created before so must have the
// proper parent already.
// rdepB.setParent(regionOfB);
// regionOfB.addChildRegion(rdepB);
this.addEdge(pdgnodeOfBRegion, pdgnodeOfdepBRegion, "dependency");
pdgnodeOfBRegion.addDependent(pdgnodeOfdepBRegion);
if (!processedRegions.contains(rdepB)) {
regions2process.add(rdepB);
}
// now remove all the nodes in the same weak region from the list of dependents
copyOfDependents.removeAll(rdepB.getBlocks());
continue;
} else {
PDGNode predPDGofdepB = (PDGNode) this.getPredsOf(depBPDGNode).get(0);
assert (this.m_obj2pdgNode.containsKey(rdepB));
PDGNode pdgnodeOfdepBRegion = this.m_obj2pdgNode.get(rdepB);
// If the loop header has not been separated from its weak region yet
if (predPDGofdepB == pdgnodeOfdepBRegion) {
/*
* Create a new region to represent the whole loop. In fact, this is a strong region as opposed to the weak
* regions that were created in the RegionAnalysis. This strong region only contains the header of the loop,
* A, and is dependent on it. Also, A is dependent on this strong region as well.
*/
Region newRegion = new Region(this.m_strongRegionStartID++, topLevelRegion.getSootMethod(),
topLevelRegion.getSootClass(), this.m_cfg);
newRegion.add(depB);
this.m_strongRegions.add(newRegion);
// toBeRemoved.add(depB);
List blocks2BRemoved;
if (toBeRemoved.contains(predPDGofdepB)) {
blocks2BRemoved = toBeRemoved.get(predPDGofdepB);
} else {
blocks2BRemoved = new ArrayList();
toBeRemoved.put(rdepB, blocks2BRemoved);
}
blocks2BRemoved.add(depB);
PDGNode newpdgnode = new LoopedPDGNode(newRegion, PDGNode.Type.REGION, depBPDGNode);
this.addNode(newpdgnode);
this.m_obj2pdgNode.put(newRegion, newpdgnode);
newpdgnode.setAttrib(PDGNode.Attribute.LOOPHEADER);
depBPDGNode.setAttrib(PDGNode.Attribute.LOOPHEADER);
this.removeEdge(pdgnodeOfdepBRegion, depBPDGNode, "dependency");
pdgnodeOfdepBRegion.removeDependent(depBPDGNode);
this.addEdge(pdgnodeOfdepBRegion, newpdgnode, "dependency");
this.addEdge(newpdgnode, depBPDGNode, "dependency");
pdgnodeOfdepBRegion.addDependent(newpdgnode);
newpdgnode.addDependent(depBPDGNode);
// If a is dependent on itself (simple loop)
if (depB == a) {
PDGNode loopBodyPDGNode = (PDGNode) this.getSuccsOf(depBPDGNode).get(0);
this.addEdge(depBPDGNode, newpdgnode, "dependency-back");
((LoopedPDGNode) newpdgnode).setBody(loopBodyPDGNode);
depBPDGNode.addBackDependent(newpdgnode);
// This is needed to correctly adjust the prev/next pointers for the new loop node. We should not need
// to adjust the old loop header node because the prev/next should not have been set previously for it.
curNodeInRegion = newpdgnode;
} else {
// this is a back-dependency
pdgnodeOfBRegion.addBackDependent(newpdgnode);
this.addEdge(pdgnodeOfBRegion, newpdgnode, "dependency-back");
// DEtermine which dependent of the loop header is actually the loop body region
PDGNode loopBodyPDGNode = null;
List successors = this.getSuccsOf(depBPDGNode);
Iterator succItr = successors.iterator();
while (succItr.hasNext()) {
PDGNode succRPDGNode = succItr.next();
assert (succRPDGNode.getType() == PDGNode.Type.REGION);
Region succR = (Region) succRPDGNode.getNode();
Block h = succR.getBlocks().get(0);
DominatorNode hdode = dom.getDode(h);
DominatorNode adode = dom.getDode(a);
if (dom.isDominatorOf(hdode, adode)) {
loopBodyPDGNode = succRPDGNode;
break;
}
}
assert (loopBodyPDGNode != null);
((LoopedPDGNode) newpdgnode).setBody(loopBodyPDGNode);
PDGNode prev = depBPDGNode.getPrev();
if (prev != null) {
this.removeEdge(prev, depBPDGNode, "controlflow");
this.addEdge(prev, newpdgnode, "controlflow");
prev.setNext(newpdgnode);
newpdgnode.setPrev(prev);
depBPDGNode.setPrev(null);
}
PDGNode next = depBPDGNode.getNext();
if (next != null) {
this.removeEdge(depBPDGNode, next, "controlflow");
this.addEdge(newpdgnode, next, "controlflow");
newpdgnode.setNext(next);
next.setPrev(newpdgnode);
depBPDGNode.setNext(null);
}
}
} else {
/*
* The strong region for the header has already been created and its corresponding PDGNode exist. Just add
* the dependency edge.
*/
this.addEdge(pdgnodeOfBRegion, predPDGofdepB, "dependency-back");
// this is a back-dependency
pdgnodeOfBRegion.addBackDependent(predPDGofdepB);
}
}
}
}
/*
* If there is a previous node in this region, add a control flow edge to indicate the the correct direction of
* control flow in the region.
*/
if (prevPDGNodeInRegion != null) {
this.addEdge(prevPDGNodeInRegion, curNodeInRegion, "controlflow");
prevPDGNodeInRegion.setNext(curNodeInRegion);
curNodeInRegion.setPrev(prevPDGNodeInRegion);
}
prevPDGNodeInRegion = curNodeInRegion;
}
// remove all the blocks marked to be removed from the region (to change a weak region
// to a strong region.)
Enumeration itr1 = toBeRemoved.keys();
while (itr1.hasMoreElements()) {
Region region = itr1.nextElement();
Iterator blockItr = toBeRemoved.get(region).iterator();
while (blockItr.hasNext()) {
region.remove(blockItr.next());
}
}
}
}
private List cloneRegions(List weak) {
List strong = new ArrayList();
Iterator itr = weak.iterator();
while (itr.hasNext()) {
Region r = itr.next();
strong.add((Region) r.clone());
}
return strong;
}
/**
*
* @return The Corresponding UnitGraph
*/
public UnitGraph getCFG() {
return this.m_cfg;
}
/**
* {@inheritDoc}
*/
public List getWeakRegions() {
return this.m_weakRegions;
}
/**
* {@inheritDoc}
*/
public List getStrongRegions() {
return this.m_strongRegions;
}
/**
* {@inheritDoc}
*/
public IRegion GetStartRegion() {
return (IRegion) GetStartNode().getNode();
}
/**
* {@inheritDoc}
*/
public PDGNode GetStartNode() {
return this.m_startNode;
}
public static List getPreorderRegionList(IRegion r) {
List list = new ArrayList();
Queue toProcess = new LinkedList();
toProcess.add(r);
while (!toProcess.isEmpty()) {
IRegion reg = toProcess.poll();
list.add(reg);
for (Iterator itr = reg.getChildRegions().iterator(); itr.hasNext();) {
toProcess.add((Region) itr.next());
}
}
return list;
}
public static List getPostorderRegionList(IRegion r) {
List list = new ArrayList();
postorder(r, list);
return list;
}
private static List postorder(IRegion r, List list) {
// If there are children, push the children to the stack
if (!r.getChildRegions().isEmpty()) {
for (Iterator itr = r.getChildRegions().iterator(); itr.hasNext();) {
postorder(itr.next(), list);
}
}
list.add(r);
return list;
}
/**
* {@inheritDoc}
*/
public List getPDGRegions() {
return this.m_pdgRegions;
}
/**
* This method returns a list of regions obtained by post-order traversal of the region hierarchy. This takes advantage of
* the hierarchical (parent/child) information encoded within the PDGNodes at construction time; it should be noted that,
* we have not counted the strong regions that represent the loop header as a separate region; instead, a PDGRegion that
* represents both the loop header and its body are counted.
*
* @param The
* root from which the traversal should begin.
* @return The list of regions obtained thru post-order traversal of the region hierarchy.
*/
public static List getPostorderPDGRegionList(PDGNode r) {
return computePDGRegions(r);
}
private static Hashtable node2Region = new Hashtable();
// compute the pdg region list with in post order
private static List computePDGRegions(PDGNode root) {
List regions = new ArrayList();
node2Region.clear();
pdgpostorder(root, regions);
return regions;
}
private static PDGRegion pdgpostorder(PDGNode node, List list) {
if (node.getVisited()) {
return null;
}
node.setVisited(true);
PDGRegion region = null;
if (!node2Region.containsKey(node)) {
region = new PDGRegion(node);
node2Region.put(node, region);
} else {
region = node2Region.get(node);
}
// If there are children, push the children to the stack
List dependents = node.getDependents();
if (!dependents.isEmpty()) {
for (Iterator itr = dependents.iterator(); itr.hasNext();) {
PDGNode curNode = (PDGNode) itr.next();
if (curNode.getVisited()) {
continue;
}
region.addPDGNode(curNode);
if (curNode instanceof LoopedPDGNode) {
PDGNode body = ((LoopedPDGNode) curNode).getBody();
PDGRegion kid = pdgpostorder(body, list);
if (kid != null) {
kid.setParent(region);
region.addChildRegion(kid);
// This sets the node from the old Region into a PDGRegion
body.setNode(kid);
}
} else if (curNode instanceof ConditionalPDGNode) {
List childs = curNode.getDependents();
Iterator condItr = childs.iterator();
while (condItr.hasNext()) {
PDGNode child = (PDGNode) condItr.next();
PDGRegion kid = pdgpostorder(child, list);
if (kid != null) {
kid.setParent(region);
region.addChildRegion(kid);
// This sets the node from the old Region into a PDGRegion
child.setNode(kid);
}
}
}
}
}
list.add(region);
return region;
}
/**
* {@inheritDoc}
*/
public boolean dependentOn(PDGNode node1, PDGNode node2) {
return node2.getDependents().contains(node1);
}
/**
* {@inheritDoc}
*/
public List getDependents(PDGNode node) {
List toReturn = new ArrayList();
for (PDGNode succ : this.getSuccsOf(node)) {
if (this.dependentOn(succ, node)) {
toReturn.add(succ);
}
}
return toReturn;
}
/**
* {@inheritDoc}
*/
public PDGNode getPDGNode(Object cfgNode) {
if (cfgNode != null && cfgNode instanceof Block) {
if (this.m_obj2pdgNode.containsKey(cfgNode)) {
return this.m_obj2pdgNode.get(cfgNode);
}
}
return null;
}
@SuppressWarnings("unchecked")
private void replaceInGraph(PDGNode newnode, PDGNode oldnode) {
this.addNode(newnode);
HashMutableEdgeLabelledDirectedGraph graph = (HashMutableEdgeLabelledDirectedGraph) this.clone();
List succs = graph.getSuccsOf(oldnode);
List preds = graph.getPredsOf(oldnode);
for (PDGNode succ : succs) {
List
