fixing bug in points-to info and OTF call-graph algorithm

Author: pdschubert

include/phasar/Controller/AnalysisController.h

@@ -33,7 +33,7 @@ class AnalysisController {
   std::vector<DataFlowAnalysisType> DataFlowAnalyses;
   std::vector<std::string> AnalysisConfigs;
   std::set<std::string> EntryPoints;
-  AnalysisStrategy Strategy;
+  [[maybe_unused]] AnalysisStrategy Strategy;
 
   void executeDemandDriven();
   void executeIncremental();

include/phasar/PhasarLLVM/ControlFlow/CFG.h

@@ -21,6 +21,8 @@
 #include <utility> // std::pair
 #include <vector>
 
+#include <json.hpp>
+
 namespace psr {
 
 template <typename N, typename M> class CFG {
@@ -52,6 +54,10 @@ template <typename N, typename M> class CFG {
   virtual std::string getStatementId(N stmt) const = 0;
 
   virtual std::string getFunctionName(M fun) const = 0;
+
+  virtual void print(M fun, std::ostream &OS) const = 0;
+
+  virtual nlohmann::json getAsJson(M fun) const = 0;
 };
 
 } // namespace psr

include/phasar/PhasarLLVM/ControlFlow/ICFG.h

@@ -17,7 +17,7 @@
 #ifndef PHASAR_PHASARLLVM_CONTROLFLOW_ICFG_H_
 #define PHASAR_PHASARLLVM_CONTROLFLOW_ICFG_H_
 
-#include <iosfwd>
+#include <iostream>
 #include <map>
 #include <set>
 #include <string>
@@ -68,6 +68,10 @@ template <typename N, typename M> class ICFG : public virtual CFG<N, M> {
 
   virtual std::set<N> getReturnSitesOfCallAt(N stmt) const = 0;
 
+  using CFG<N, M>::print; // tell the compiler we wish to have both prints
+  virtual void print(std::ostream &OS = std::cout) const = 0;
+
+  using CFG<N, M>::getAsJson; // tell the compiler we wish to have both prints
   virtual nlohmann::json getAsJson() const = 0;
 };
 

include/phasar/PhasarLLVM/ControlFlow/LLVMBasedBackwardCFG.h

@@ -73,6 +73,10 @@ class LLVMBasedBackwardCFG
   std::string getFunctionName(const llvm::Function *fun) const override;
 
   std::string getStatementId(const llvm::Instruction *stmt) const override;
+
+  void print(const llvm::Function *F, std::ostream &OS) const override;
+
+  nlohmann::json getAsJson(const llvm::Function *F) const override;
 };
 } // namespace psr
 

include/phasar/PhasarLLVM/ControlFlow/LLVMBasedBackwardICFG.h

@@ -92,12 +92,16 @@ class LLVMBasedBackwardsICFG
 
   bool isPrimitiveFunction(const std::string &name);
 
-  void print(std::ostream &OS);
+  using LLVMBasedBackwardCFG::print; // tell the compiler we wish to have both
+                                     // prints
+  void print(std::ostream &OS) const override;
 
-  void printAsDot(std::ostream &OS);
+  void printAsDot(std::ostream &OS) const;
 
   void printInternalPTGAsDot(const std::string &filename);
 
+  using LLVMBasedBackwardCFG::getAsJson; // tell the compiler we wish to have
+                                         // both prints
   nlohmann::json getAsJson() const override;
 
   unsigned getNumOfVertices();

include/phasar/PhasarLLVM/ControlFlow/LLVMBasedCFG.h

@@ -17,6 +17,7 @@
 #ifndef PHASAR_PHASARLLVM_CONTROLFLOW_LLVMBASEDCFG_H_
 #define PHASAR_PHASARLLVM_CONTROLFLOW_LLVMBASEDCFG_H_
 
+#include <iostream>
 #include <set>
 #include <string>
 #include <vector>
@@ -69,6 +70,11 @@ class LLVMBasedCFG
   std::string getStatementId(const llvm::Instruction *stmt) const override;
 
   std::string getFunctionName(const llvm::Function *fun) const override;
+
+  void print(const llvm::Function *F,
+             std::ostream &OS = std::cout) const override;
+
+  nlohmann::json getAsJson(const llvm::Function *F) const override;
 };
 
 } // namespace psr

include/phasar/PhasarLLVM/ControlFlow/LLVMBasedICFG.h

@@ -157,12 +157,15 @@ class LLVMBasedICFG
 
   bool isPrimitiveFunction(const std::string &name);
 
-  void print(std::ostream &OS = std::cout);
+  using LLVMBasedCFG::print; // tell the compiler we wish to have both prints
+  void print(std::ostream &OS = std::cout) const override;
 
-  void printAsDot(std::ostream &OS = std::cout);
+  void printAsDot(std::ostream &OS = std::cout) const;
 
   void printInternalPTGAsDot(const std::string &filename);
 
+  using LLVMBasedCFG::getAsJson; // tell the compiler we wish to have both
+                                 // prints
   nlohmann::json getAsJson() const override;
 
   unsigned getNumOfVertices();

include/phasar/PhasarLLVM/ControlFlow/Resolver/OTFResolver.h

@@ -57,6 +57,9 @@ class OTFResolver : public CHAResolver {
 
   std::set<const llvm::Function *>
   resolveVirtualCall(llvm::ImmutableCallSite CS) override;
+
+  std::set<const llvm::Function *>
+  resolveFunctionPointer(llvm::ImmutableCallSite CS) override;
 };
 } // namespace psr
 

include/phasar/PhasarLLVM/Pointer/LLVMPointsToGraph.h

@@ -18,6 +18,8 @@
 #define PHASAR_PHASARLLVM_POINTER_POINTSTOGRAPH_H_
 
 #include <iostream>
+#include <unordered_map>
+#include <unordered_set>
 #include <vector>
 
 #include <boost/graph/adjacency_list.hpp>
@@ -48,18 +50,6 @@ static inline bool isInterestingPointer(llvm::Value *V) {
          !llvm::isa<llvm::ConstantPointerNull>(V);
 }
 
-enum class PointerAnalysisType {
-#define ANALYSIS_SETUP_POINTER_TYPE(NAME, CMDFLAG, TYPE) TYPE,
-#include <phasar/PhasarLLVM/Utils/AnalysisSetups.def>
-  Invalid
-};
-
-std::string to_string(const PointerAnalysisType &PA);
-
-PointerAnalysisType to_PointerAnalysisType(const std::string &S);
-
-std::ostream &operator<<(std::ostream &os, const PointerAnalysisType &PA);
-
 // TODO: add a more high level description.
 /**
  * 	This class is a representation of a points-to graph. It is possible to
@@ -83,9 +73,9 @@ class PointsToGraph {
      * This might be an Instruction, an Operand of an Instruction, Global
      * Variable or a formal Argument.
      */
-    const llvm::Value *value = nullptr;
+    const llvm::Value *V = nullptr;
     /// Holds the llvm IR code for that vertex.
-    std::string ir_code;
+    std::string IR;
 
     VertexProperties() = default;
     VertexProperties(const llvm::Value *v);
@@ -96,9 +86,9 @@ class PointsToGraph {
    */
   struct EdgeProperties {
     /// This might be an Instruction, in particular a Call Instruction.
-    const llvm::Value *value = nullptr;
+    const llvm::Value *V = nullptr;
     /// Holds the llvm IR code for that edge.
-    std::string ir_code;
+    std::string IR;
 
     EdgeProperties() = default;
     EdgeProperties(const llvm::Value *v);
@@ -117,23 +107,22 @@ class PointsToGraph {
 
   /// The type for a vertex iterator.
   typedef boost::graph_traits<graph_t>::vertex_iterator vertex_iterator_t;
-  typedef boost::graph_traits<graph_t>::vertex_iterator vertex_iterator;
-  typedef boost::graph_traits<graph_t>::out_edge_iterator out_edge_iterator;
-  typedef boost::graph_traits<graph_t>::in_edge_iterator in_edge_iterator;
+  typedef boost::graph_traits<graph_t>::out_edge_iterator out_edge_iterator_t;
+  typedef boost::graph_traits<graph_t>::in_edge_iterator in_edge_iterator_t;
 
   /// Set of functions that allocate heap memory, e.g. new, new[], malloc.
   inline const static std::set<std::string> HeapAllocationFunctions = {
       "_Znwm", "_Znam", "malloc", "calloc", "realloc"};
 
 private:
-  struct allocation_site_dfs_visitor;
-  struct reachability_dfs_visitor;
+  struct AllocationSiteDFSVisitor;
+  struct ReachabilityDFSVisitor;
 
   /// The points to graph.
-  graph_t ptg;
-  std::map<const llvm::Value *, vertex_t> value_vertex_map;
+  graph_t PAG;
+  std::unordered_map<const llvm::Value *, vertex_t> ValueVertexMap;
   /// Keep track of what has already been merged into this points-to graph.
-  std::set<std::string> ContainedFunctions;
+  std::unordered_set<std::string> ContainedFunctions;
 
 public:
   /**
@@ -147,18 +136,7 @@ class PointsToGraph {
    *                              False, if May and Must Aliases should be
    * considered.
    */
-  PointsToGraph(llvm::Function *F, llvm::AAResults &AA,
-                bool onlyConsiderMustAlias = false);
-
-  /**
-   * It is used when a points-to graph is restored from the database.
-   *
-   * @brief This will create an empty points-to graph, except the functions
-   * names
-   * that are contained in the points-to graph.
-   * @param fnames Names of functions contained in the points-to graph.
-   */
-  PointsToGraph(std::vector<std::string> fnames);
+  PointsToGraph(llvm::Function *F, llvm::AAResults &AA);
 
   /**
    * @brief This will create an empty points-to graph. It is used when points-to

include/phasar/PhasarLLVM/Pointer/LLVMPointsToInfo.h

@@ -15,6 +15,8 @@
 #include <string>
 
 #include <llvm/Analysis/AliasAnalysis.h>
+#include <llvm/IR/PassManager.h>
+#include <llvm/Passes/PassBuilder.h>
 
 #include <json.hpp>
 
@@ -31,15 +33,30 @@ namespace psr {
 class ProjectIRDB;
 class PointsToGraph;
 
+enum class PointerAnalysisType {
+#define ANALYSIS_SETUP_POINTER_TYPE(NAME, CMDFLAG, TYPE) TYPE,
+#include <phasar/PhasarLLVM/Utils/AnalysisSetups.def>
+  Invalid
+};
+
+std::string to_string(const PointerAnalysisType &PA);
+
+PointerAnalysisType to_PointerAnalysisType(const std::string &S);
+
+std::ostream &operator<<(std::ostream &os, const PointerAnalysisType &PA);
+
 class LLVMPointsToInfo
     : public PointsToInfo<const llvm::Value *, const llvm::Instruction *> {
 private:
-  std::unordered_map<const llvm::Function *, llvm::AAResults> AAInfos;
+  llvm::PassBuilder PB;
+  llvm::FunctionAnalysisManager FAM;
+  mutable std::unordered_map<const llvm::Function *, llvm::AAResults *> AAInfos;
   std::map<const llvm::Function *, std::unique_ptr<PointsToGraph>>
       PointsToGraphs;
 
 public:
-  LLVMPointsToInfo(ProjectIRDB &IRDB);
+  LLVMPointsToInfo(ProjectIRDB &IRDB,
+                   PointerAnalysisType PAT = PointerAnalysisType::CFLAnders);
 
   ~LLVMPointsToInfo() override = default;
 
@@ -54,6 +71,8 @@ class LLVMPointsToInfo
 
   nlohmann::json getAsJson() const override;
 
+  llvm::AAResults *getAAResults(const llvm::Function *F) const;
+
   PointsToGraph *getPointsToGraph(const llvm::Function *F) const;
 };