teak-llvm/clang/lib/StaticAnalyzer/Core/ExplodedGraph.cpp
Ben Craig 4067e35fae [Analyzer] Don't cache report generation ExplodedNodes
During the core analysis, ExplodedNodes are added to the
ExplodedGraph, and those nodes are cached for deduplication purposes.

After core analysis, reports are generated. Here, trimmed copies of
the ExplodedGraph are made. Since the ExplodedGraph has already been
deduplicated, there is no need to deduplicate again.

This change makes it possible to add ExplodedNodes to an
ExplodedGraph without the overhead of deduplication. "Uncached" nodes
also cannot be iterated over, but none of the report generation code
attempts to iterate over all nodes. This change reduces the analysis
time of a large .C file from 3m43.941s to 3m40.256s (~1.6% speedup).
It should slightly reduce memory consumption. Gains should be roughly
proportional to the number (and path length) of static analysis
warnings.

This patch enables future work that should remove the need for an
InterExplodedGraphMap inverse map. I plan on using the (now unused)
ExplodedNode link to connect new nodes to the original nodes.

http://reviews.llvm.org/D21229

llvm-svn: 273572
2016-06-23 15:47:12 +00:00

451 lines
14 KiB
C++

//=-- ExplodedGraph.cpp - Local, Path-Sens. "Exploded Graph" -*- C++ -*------=//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file defines the template classes ExplodedNode and ExplodedGraph,
// which represent a path-sensitive, intra-procedural "exploded graph."
//
//===----------------------------------------------------------------------===//
#include "clang/StaticAnalyzer/Core/PathSensitive/ExplodedGraph.h"
#include "clang/AST/ParentMap.h"
#include "clang/AST/Stmt.h"
#include "clang/StaticAnalyzer/Core/PathSensitive/CallEvent.h"
#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramState.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/DenseSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Statistic.h"
#include <vector>
using namespace clang;
using namespace ento;
//===----------------------------------------------------------------------===//
// Node auditing.
//===----------------------------------------------------------------------===//
// An out of line virtual method to provide a home for the class vtable.
ExplodedNode::Auditor::~Auditor() {}
#ifndef NDEBUG
static ExplodedNode::Auditor* NodeAuditor = nullptr;
#endif
void ExplodedNode::SetAuditor(ExplodedNode::Auditor* A) {
#ifndef NDEBUG
NodeAuditor = A;
#endif
}
//===----------------------------------------------------------------------===//
// Cleanup.
//===----------------------------------------------------------------------===//
ExplodedGraph::ExplodedGraph()
: NumNodes(0), ReclaimNodeInterval(0) {}
ExplodedGraph::~ExplodedGraph() {}
//===----------------------------------------------------------------------===//
// Node reclamation.
//===----------------------------------------------------------------------===//
bool ExplodedGraph::isInterestingLValueExpr(const Expr *Ex) {
if (!Ex->isLValue())
return false;
return isa<DeclRefExpr>(Ex) ||
isa<MemberExpr>(Ex) ||
isa<ObjCIvarRefExpr>(Ex);
}
bool ExplodedGraph::shouldCollect(const ExplodedNode *node) {
// First, we only consider nodes for reclamation of the following
// conditions apply:
//
// (1) 1 predecessor (that has one successor)
// (2) 1 successor (that has one predecessor)
//
// If a node has no successor it is on the "frontier", while a node
// with no predecessor is a root.
//
// After these prerequisites, we discard all "filler" nodes that
// are used only for intermediate processing, and are not essential
// for analyzer history:
//
// (a) PreStmtPurgeDeadSymbols
//
// We then discard all other nodes where *all* of the following conditions
// apply:
//
// (3) The ProgramPoint is for a PostStmt, but not a PostStore.
// (4) There is no 'tag' for the ProgramPoint.
// (5) The 'store' is the same as the predecessor.
// (6) The 'GDM' is the same as the predecessor.
// (7) The LocationContext is the same as the predecessor.
// (8) Expressions that are *not* lvalue expressions.
// (9) The PostStmt isn't for a non-consumed Stmt or Expr.
// (10) The successor is neither a CallExpr StmtPoint nor a CallEnter or
// PreImplicitCall (so that we would be able to find it when retrying a
// call with no inlining).
// FIXME: It may be safe to reclaim PreCall and PostCall nodes as well.
// Conditions 1 and 2.
if (node->pred_size() != 1 || node->succ_size() != 1)
return false;
const ExplodedNode *pred = *(node->pred_begin());
if (pred->succ_size() != 1)
return false;
const ExplodedNode *succ = *(node->succ_begin());
if (succ->pred_size() != 1)
return false;
// Now reclaim any nodes that are (by definition) not essential to
// analysis history and are not consulted by any client code.
ProgramPoint progPoint = node->getLocation();
if (progPoint.getAs<PreStmtPurgeDeadSymbols>())
return !progPoint.getTag();
// Condition 3.
if (!progPoint.getAs<PostStmt>() || progPoint.getAs<PostStore>())
return false;
// Condition 4.
if (progPoint.getTag())
return false;
// Conditions 5, 6, and 7.
ProgramStateRef state = node->getState();
ProgramStateRef pred_state = pred->getState();
if (state->store != pred_state->store || state->GDM != pred_state->GDM ||
progPoint.getLocationContext() != pred->getLocationContext())
return false;
// All further checks require expressions. As per #3, we know that we have
// a PostStmt.
const Expr *Ex = dyn_cast<Expr>(progPoint.castAs<PostStmt>().getStmt());
if (!Ex)
return false;
// Condition 8.
// Do not collect nodes for "interesting" lvalue expressions since they are
// used extensively for generating path diagnostics.
if (isInterestingLValueExpr(Ex))
return false;
// Condition 9.
// Do not collect nodes for non-consumed Stmt or Expr to ensure precise
// diagnostic generation; specifically, so that we could anchor arrows
// pointing to the beginning of statements (as written in code).
ParentMap &PM = progPoint.getLocationContext()->getParentMap();
if (!PM.isConsumedExpr(Ex))
return false;
// Condition 10.
const ProgramPoint SuccLoc = succ->getLocation();
if (Optional<StmtPoint> SP = SuccLoc.getAs<StmtPoint>())
if (CallEvent::isCallStmt(SP->getStmt()))
return false;
// Condition 10, continuation.
if (SuccLoc.getAs<CallEnter>() || SuccLoc.getAs<PreImplicitCall>())
return false;
return true;
}
void ExplodedGraph::collectNode(ExplodedNode *node) {
// Removing a node means:
// (a) changing the predecessors successor to the successor of this node
// (b) changing the successors predecessor to the predecessor of this node
// (c) Putting 'node' onto freeNodes.
assert(node->pred_size() == 1 || node->succ_size() == 1);
ExplodedNode *pred = *(node->pred_begin());
ExplodedNode *succ = *(node->succ_begin());
pred->replaceSuccessor(succ);
succ->replacePredecessor(pred);
FreeNodes.push_back(node);
Nodes.RemoveNode(node);
--NumNodes;
node->~ExplodedNode();
}
void ExplodedGraph::reclaimRecentlyAllocatedNodes() {
if (ChangedNodes.empty())
return;
// Only periodically reclaim nodes so that we can build up a set of
// nodes that meet the reclamation criteria. Freshly created nodes
// by definition have no successor, and thus cannot be reclaimed (see below).
assert(ReclaimCounter > 0);
if (--ReclaimCounter != 0)
return;
ReclaimCounter = ReclaimNodeInterval;
for (NodeVector::iterator it = ChangedNodes.begin(), et = ChangedNodes.end();
it != et; ++it) {
ExplodedNode *node = *it;
if (shouldCollect(node))
collectNode(node);
}
ChangedNodes.clear();
}
//===----------------------------------------------------------------------===//
// ExplodedNode.
//===----------------------------------------------------------------------===//
// An NodeGroup's storage type is actually very much like a TinyPtrVector:
// it can be either a pointer to a single ExplodedNode, or a pointer to a
// BumpVector allocated with the ExplodedGraph's allocator. This allows the
// common case of single-node NodeGroups to be implemented with no extra memory.
//
// Consequently, each of the NodeGroup methods have up to four cases to handle:
// 1. The flag is set and this group does not actually contain any nodes.
// 2. The group is empty, in which case the storage value is null.
// 3. The group contains a single node.
// 4. The group contains more than one node.
typedef BumpVector<ExplodedNode *> ExplodedNodeVector;
typedef llvm::PointerUnion<ExplodedNode *, ExplodedNodeVector *> GroupStorage;
void ExplodedNode::addPredecessor(ExplodedNode *V, ExplodedGraph &G) {
assert (!V->isSink());
Preds.addNode(V, G);
V->Succs.addNode(this, G);
#ifndef NDEBUG
if (NodeAuditor) NodeAuditor->AddEdge(V, this);
#endif
}
void ExplodedNode::NodeGroup::replaceNode(ExplodedNode *node) {
assert(!getFlag());
GroupStorage &Storage = reinterpret_cast<GroupStorage&>(P);
assert(Storage.is<ExplodedNode *>());
Storage = node;
assert(Storage.is<ExplodedNode *>());
}
void ExplodedNode::NodeGroup::addNode(ExplodedNode *N, ExplodedGraph &G) {
assert(!getFlag());
GroupStorage &Storage = reinterpret_cast<GroupStorage&>(P);
if (Storage.isNull()) {
Storage = N;
assert(Storage.is<ExplodedNode *>());
return;
}
ExplodedNodeVector *V = Storage.dyn_cast<ExplodedNodeVector *>();
if (!V) {
// Switch from single-node to multi-node representation.
ExplodedNode *Old = Storage.get<ExplodedNode *>();
BumpVectorContext &Ctx = G.getNodeAllocator();
V = G.getAllocator().Allocate<ExplodedNodeVector>();
new (V) ExplodedNodeVector(Ctx, 4);
V->push_back(Old, Ctx);
Storage = V;
assert(!getFlag());
assert(Storage.is<ExplodedNodeVector *>());
}
V->push_back(N, G.getNodeAllocator());
}
unsigned ExplodedNode::NodeGroup::size() const {
if (getFlag())
return 0;
const GroupStorage &Storage = reinterpret_cast<const GroupStorage &>(P);
if (Storage.isNull())
return 0;
if (ExplodedNodeVector *V = Storage.dyn_cast<ExplodedNodeVector *>())
return V->size();
return 1;
}
ExplodedNode * const *ExplodedNode::NodeGroup::begin() const {
if (getFlag())
return nullptr;
const GroupStorage &Storage = reinterpret_cast<const GroupStorage &>(P);
if (Storage.isNull())
return nullptr;
if (ExplodedNodeVector *V = Storage.dyn_cast<ExplodedNodeVector *>())
return V->begin();
return Storage.getAddrOfPtr1();
}
ExplodedNode * const *ExplodedNode::NodeGroup::end() const {
if (getFlag())
return nullptr;
const GroupStorage &Storage = reinterpret_cast<const GroupStorage &>(P);
if (Storage.isNull())
return nullptr;
if (ExplodedNodeVector *V = Storage.dyn_cast<ExplodedNodeVector *>())
return V->end();
return Storage.getAddrOfPtr1() + 1;
}
ExplodedNode *ExplodedGraph::getNode(const ProgramPoint &L,
ProgramStateRef State,
bool IsSink,
bool* IsNew) {
// Profile 'State' to determine if we already have an existing node.
llvm::FoldingSetNodeID profile;
void *InsertPos = nullptr;
NodeTy::Profile(profile, L, State, IsSink);
NodeTy* V = Nodes.FindNodeOrInsertPos(profile, InsertPos);
if (!V) {
if (!FreeNodes.empty()) {
V = FreeNodes.back();
FreeNodes.pop_back();
}
else {
// Allocate a new node.
V = (NodeTy*) getAllocator().Allocate<NodeTy>();
}
new (V) NodeTy(L, State, IsSink);
if (ReclaimNodeInterval)
ChangedNodes.push_back(V);
// Insert the node into the node set and return it.
Nodes.InsertNode(V, InsertPos);
++NumNodes;
if (IsNew) *IsNew = true;
}
else
if (IsNew) *IsNew = false;
return V;
}
ExplodedNode *ExplodedGraph::createUncachedNode(const ProgramPoint &L,
ProgramStateRef State,
bool IsSink) {
NodeTy *V = (NodeTy *) getAllocator().Allocate<NodeTy>();
new (V) NodeTy(L, State, IsSink);
return V;
}
std::unique_ptr<ExplodedGraph>
ExplodedGraph::trim(ArrayRef<const NodeTy *> Sinks,
InterExplodedGraphMap *ForwardMap,
InterExplodedGraphMap *InverseMap) const {
if (Nodes.empty())
return nullptr;
typedef llvm::DenseSet<const ExplodedNode*> Pass1Ty;
Pass1Ty Pass1;
typedef InterExplodedGraphMap Pass2Ty;
InterExplodedGraphMap Pass2Scratch;
Pass2Ty &Pass2 = ForwardMap ? *ForwardMap : Pass2Scratch;
SmallVector<const ExplodedNode*, 10> WL1, WL2;
// ===- Pass 1 (reverse DFS) -===
for (ArrayRef<const NodeTy *>::iterator I = Sinks.begin(), E = Sinks.end();
I != E; ++I) {
if (*I)
WL1.push_back(*I);
}
// Process the first worklist until it is empty.
while (!WL1.empty()) {
const ExplodedNode *N = WL1.pop_back_val();
// Have we already visited this node? If so, continue to the next one.
if (!Pass1.insert(N).second)
continue;
// If this is a root enqueue it to the second worklist.
if (N->Preds.empty()) {
WL2.push_back(N);
continue;
}
// Visit our predecessors and enqueue them.
WL1.append(N->Preds.begin(), N->Preds.end());
}
// We didn't hit a root? Return with a null pointer for the new graph.
if (WL2.empty())
return nullptr;
// Create an empty graph.
std::unique_ptr<ExplodedGraph> G = MakeEmptyGraph();
// ===- Pass 2 (forward DFS to construct the new graph) -===
while (!WL2.empty()) {
const ExplodedNode *N = WL2.pop_back_val();
// Skip this node if we have already processed it.
if (Pass2.find(N) != Pass2.end())
continue;
// Create the corresponding node in the new graph and record the mapping
// from the old node to the new node.
ExplodedNode *NewN = G->createUncachedNode(N->getLocation(), N->State, N->isSink());
Pass2[N] = NewN;
// Also record the reverse mapping from the new node to the old node.
if (InverseMap) (*InverseMap)[NewN] = N;
// If this node is a root, designate it as such in the graph.
if (N->Preds.empty())
G->addRoot(NewN);
// In the case that some of the intended predecessors of NewN have already
// been created, we should hook them up as predecessors.
// Walk through the predecessors of 'N' and hook up their corresponding
// nodes in the new graph (if any) to the freshly created node.
for (ExplodedNode::pred_iterator I = N->Preds.begin(), E = N->Preds.end();
I != E; ++I) {
Pass2Ty::iterator PI = Pass2.find(*I);
if (PI == Pass2.end())
continue;
NewN->addPredecessor(const_cast<ExplodedNode *>(PI->second), *G);
}
// In the case that some of the intended successors of NewN have already
// been created, we should hook them up as successors. Otherwise, enqueue
// the new nodes from the original graph that should have nodes created
// in the new graph.
for (ExplodedNode::succ_iterator I = N->Succs.begin(), E = N->Succs.end();
I != E; ++I) {
Pass2Ty::iterator PI = Pass2.find(*I);
if (PI != Pass2.end()) {
const_cast<ExplodedNode *>(PI->second)->addPredecessor(NewN, *G);
continue;
}
// Enqueue nodes to the worklist that were marked during pass 1.
if (Pass1.count(*I))
WL2.push_back(*I);
}
}
return G;
}