C++: Make AST GVN a wrapper for IR-based GVN

config/identical-files.json

@@ -222,7 +222,12 @@
     "cpp/ql/src/semmle/code/cpp/ir/implementation/aliased_ssa/internal/PrintSSA.qll",
     "csharp/ql/src/semmle/code/csharp/ir/implementation/unaliased_ssa/internal/PrintSSA.qll"
   ],
-  "IR ValueNumber": [
+  "C++ IR ValueNumberInternal": [
+    "cpp/ql/src/semmle/code/cpp/ir/implementation/raw/gvn/internal/ValueNumberingInternal.qll",
+    "cpp/ql/src/semmle/code/cpp/ir/implementation/unaliased_ssa/gvn/internal/ValueNumberingInternal.qll",
+    "cpp/ql/src/semmle/code/cpp/ir/implementation/aliased_ssa/gvn/internal/ValueNumberingInternal.qll"
+  ],
+  "C++ IR ValueNumber": [
     "cpp/ql/src/semmle/code/cpp/ir/implementation/raw/gvn/ValueNumbering.qll",
     "cpp/ql/src/semmle/code/cpp/ir/implementation/unaliased_ssa/gvn/ValueNumbering.qll",
     "cpp/ql/src/semmle/code/cpp/ir/implementation/aliased_ssa/gvn/ValueNumbering.qll",

cpp/ql/src/semmle/code/cpp/ir/implementation/aliased_ssa/gvn/ValueNumbering.qll

@@ -1,6 +1,5 @@
 private import internal.ValueNumberingInternal
 private import internal.ValueNumberingImports
-private import IR
 
 /**
  * Provides additional information about value numbering in IR dumps.
@@ -10,62 +9,38 @@ class ValueNumberPropertyProvider extends IRPropertyProvider {
     exists(ValueNumber vn |
       vn = valueNumber(instr) and
       key = "valnum" and
-      if strictcount(vn.getAnInstruction()) > 1 then result = vn.toString() else result = "unique"
+      if strictcount(vn.getAnInstruction()) > 1
+      then result = vn.getDebugString()
+      else result = "unique"
     )
   }
 }
 
-newtype TValueNumber =
-  TVariableAddressValueNumber(IRFunction irFunc, IRVariable var) {
-    variableAddressValueNumber(_, irFunc, var)
-  } or
-  TInitializeParameterValueNumber(IRFunction irFunc, IRVariable var) {
-    initializeParameterValueNumber(_, irFunc, var)
-  } or
-  TInitializeThisValueNumber(IRFunction irFunc) { initializeThisValueNumber(_, irFunc) } or
-  TConstantValueNumber(IRFunction irFunc, IRType type, string value) {
-    constantValueNumber(_, irFunc, type, value)
-  } or
-  TStringConstantValueNumber(IRFunction irFunc, IRType type, string value) {
-    stringConstantValueNumber(_, irFunc, type, value)
-  } or
-  TFieldAddressValueNumber(IRFunction irFunc, Language::Field field, ValueNumber objectAddress) {
-    fieldAddressValueNumber(_, irFunc, field, objectAddress)
-  } or
-  TBinaryValueNumber(
-    IRFunction irFunc, Opcode opcode, IRType type, ValueNumber leftOperand, ValueNumber rightOperand
-  ) {
-    binaryValueNumber(_, irFunc, opcode, type, leftOperand, rightOperand)
-  } or
-  TPointerArithmeticValueNumber(
-    IRFunction irFunc, Opcode opcode, IRType type, int elementSize, ValueNumber leftOperand,
-    ValueNumber rightOperand
-  ) {
-    pointerArithmeticValueNumber(_, irFunc, opcode, type, elementSize, leftOperand, rightOperand)
-  } or
-  TUnaryValueNumber(IRFunction irFunc, Opcode opcode, IRType type, ValueNumber operand) {
-    unaryValueNumber(_, irFunc, opcode, type, operand)
-  } or
-  TInheritanceConversionValueNumber(
-    IRFunction irFunc, Opcode opcode, Language::Class baseClass, Language::Class derivedClass,
-    ValueNumber operand
-  ) {
-    inheritanceConversionValueNumber(_, irFunc, opcode, baseClass, derivedClass, operand)
-  } or
-  TLoadTotalOverlapValueNumber(
-    IRFunction irFunc, IRType type, ValueNumber memOperand, ValueNumber operand
-  ) {
-    loadTotalOverlapValueNumber(_, irFunc, type, memOperand, operand)
-  } or
-  TUniqueValueNumber(IRFunction irFunc, Instruction instr) { uniqueValueNumber(instr, irFunc) }
-
 /**
  * The value number assigned to a particular set of instructions that produce equivalent results.
  */
 class ValueNumber extends TValueNumber {
-  final string toString() { result = getExampleInstruction().getResultId() }
+  final string toString() { result = "GVN" }
+
+  final string getDebugString() { result = strictconcat(getAnInstruction().getResultId(), ", ") }
 
-  final Language::Location getLocation() { result = getExampleInstruction().getLocation() }
+  final Language::Location getLocation() {
+    if
+      exists(Instruction i |
+        i = getAnInstruction() and not i.getLocation() instanceof UnknownLocation
+      )
+    then
+      result =
+        min(Language::Location l |
+          l = getAnInstruction().getLocation() and not l instanceof UnknownLocation
+        |
+          l
+          order by
+            l.getFile().getAbsolutePath(), l.getStartLine(), l.getStartColumn(), l.getEndLine(),
+            l.getEndColumn()
+        )
+    else result instanceof UnknownDefaultLocation
+  }
 
   /**
    * Gets the instructions that have been assigned this value number. This will always produce at
@@ -90,260 +65,39 @@ class ValueNumber extends TValueNumber {
    * Gets an `Operand` whose definition is exact and has this value number.
    */
   final Operand getAUse() { this = valueNumber(result.getDef()) }
-}
 
-/**
- * A `CopyInstruction` whose source operand's value is congruent to the definition of that source
- * operand.
- * For example:
- * ```
- * Point p = { 1, 2 };
- * Point q = p;
- * int a = p.x;
- * ```
- * The use of `p` on line 2 is linked to the definition of `p` on line 1, and is congruent to that
- * definition because it accesses the exact same memory.
- * The use of `p.x` on line 3 is linked to the definition of `p` on line 1 as well, but is not
- * congruent to that definition because `p.x` accesses only a subset of the memory defined by `p`.
- */
-class CongruentCopyInstruction extends CopyInstruction {
-  CongruentCopyInstruction() {
-    this.getSourceValueOperand().getDefinitionOverlap() instanceof MustExactlyOverlap
+  final string getKind() {
+    this instanceof TVariableAddressValueNumber and result = "VariableAddress"
+    or
+    this instanceof TInitializeParameterValueNumber and result = "InitializeParameter"
+    or
+    this instanceof TInitializeThisValueNumber and result = "InitializeThis"
+    or
+    this instanceof TStringConstantValueNumber and result = "StringConstant"
+    or
+    this instanceof TFieldAddressValueNumber and result = "FieldAddress"
+    or
+    this instanceof TBinaryValueNumber and result = "Binary"
+    or
+    this instanceof TPointerArithmeticValueNumber and result = "PointerArithmetic"
+    or
+    this instanceof TUnaryValueNumber and result = "Unary"
+    or
+    this instanceof TInheritanceConversionValueNumber and result = "InheritanceConversion"
+    or
+    this instanceof TLoadTotalOverlapValueNumber and result = "LoadTotalOverlap"
+    or
+    this instanceof TUniqueValueNumber and result = "Unique"
   }
 }
 
-class LoadTotalOverlapInstruction extends LoadInstruction {
-  LoadTotalOverlapInstruction() {
-    this.getSourceValueOperand().getDefinitionOverlap() instanceof MustTotallyOverlap
-  }
-}
-
-/**
- * Holds if this library knows how to assign a value number to the specified instruction, other than
- * a `unique` value number that is never shared by multiple instructions.
- */
-private predicate numberableInstruction(Instruction instr) {
-  instr instanceof VariableAddressInstruction
-  or
-  instr instanceof InitializeParameterInstruction
-  or
-  instr instanceof InitializeThisInstruction
-  or
-  instr instanceof ConstantInstruction
-  or
-  instr instanceof StringConstantInstruction
-  or
-  instr instanceof FieldAddressInstruction
-  or
-  instr instanceof BinaryInstruction
-  or
-  instr instanceof UnaryInstruction and not instr instanceof CopyInstruction
-  or
-  instr instanceof PointerArithmeticInstruction
-  or
-  instr instanceof CongruentCopyInstruction
-  or
-  instr instanceof LoadTotalOverlapInstruction
-}
-
-private predicate variableAddressValueNumber(
-  VariableAddressInstruction instr, IRFunction irFunc, IRVariable var
-) {
-  instr.getEnclosingIRFunction() = irFunc and
-  instr.getIRVariable() = var
-}
-
-private predicate initializeParameterValueNumber(
-  InitializeParameterInstruction instr, IRFunction irFunc, IRVariable var
-) {
-  instr.getEnclosingIRFunction() = irFunc and
-  instr.getIRVariable() = var
-}
-
-private predicate initializeThisValueNumber(InitializeThisInstruction instr, IRFunction irFunc) {
-  instr.getEnclosingIRFunction() = irFunc
-}
-
-private predicate constantValueNumber(
-  ConstantInstruction instr, IRFunction irFunc, IRType type, string value
-) {
-  instr.getEnclosingIRFunction() = irFunc and
-  instr.getResultIRType() = type and
-  instr.getValue() = value
-}
-
-private predicate stringConstantValueNumber(
-  StringConstantInstruction instr, IRFunction irFunc, IRType type, string value
-) {
-  instr.getEnclosingIRFunction() = irFunc and
-  instr.getResultIRType() = type and
-  instr.getValue().getValue() = value
-}
-
-private predicate fieldAddressValueNumber(
-  FieldAddressInstruction instr, IRFunction irFunc, Language::Field field, ValueNumber objectAddress
-) {
-  instr.getEnclosingIRFunction() = irFunc and
-  instr.getField() = field and
-  valueNumber(instr.getObjectAddress()) = objectAddress
-}
-
-private predicate binaryValueNumber(
-  BinaryInstruction instr, IRFunction irFunc, Opcode opcode, IRType type, ValueNumber leftOperand,
-  ValueNumber rightOperand
-) {
-  instr.getEnclosingIRFunction() = irFunc and
-  not instr instanceof PointerArithmeticInstruction and
-  instr.getOpcode() = opcode and
-  instr.getResultIRType() = type and
-  valueNumber(instr.getLeft()) = leftOperand and
-  valueNumber(instr.getRight()) = rightOperand
-}
-
-private predicate pointerArithmeticValueNumber(
-  PointerArithmeticInstruction instr, IRFunction irFunc, Opcode opcode, IRType type,
-  int elementSize, ValueNumber leftOperand, ValueNumber rightOperand
-) {
-  instr.getEnclosingIRFunction() = irFunc and
-  instr.getOpcode() = opcode and
-  instr.getResultIRType() = type and
-  instr.getElementSize() = elementSize and
-  valueNumber(instr.getLeft()) = leftOperand and
-  valueNumber(instr.getRight()) = rightOperand
-}
-
-private predicate unaryValueNumber(
-  UnaryInstruction instr, IRFunction irFunc, Opcode opcode, IRType type, ValueNumber operand
-) {
-  instr.getEnclosingIRFunction() = irFunc and
-  not instr instanceof InheritanceConversionInstruction and
-  not instr instanceof CopyInstruction and
-  not instr instanceof FieldAddressInstruction and
-  instr.getOpcode() = opcode and
-  instr.getResultIRType() = type and
-  valueNumber(instr.getUnary()) = operand
-}
-
-private predicate inheritanceConversionValueNumber(
-  InheritanceConversionInstruction instr, IRFunction irFunc, Opcode opcode,
-  Language::Class baseClass, Language::Class derivedClass, ValueNumber operand
-) {
-  instr.getEnclosingIRFunction() = irFunc and
-  instr.getOpcode() = opcode and
-  instr.getBaseClass() = baseClass and
-  instr.getDerivedClass() = derivedClass and
-  valueNumber(instr.getUnary()) = operand
-}
-
-private predicate loadTotalOverlapValueNumber(
-  LoadTotalOverlapInstruction instr, IRFunction irFunc, IRType type, ValueNumber memOperand,
-  ValueNumber operand
-) {
-  instr.getEnclosingIRFunction() = irFunc and
-  instr.getResultIRType() = type and
-  valueNumber(instr.getAnOperand().(MemoryOperand).getAnyDef()) = memOperand and
-  valueNumberOfOperand(instr.getAnOperand().(AddressOperand)) = operand
-}
-
-/**
- * Holds if `instr` should be assigned a unique value number because this library does not know how
- * to determine if two instances of that instruction are equivalent.
- */
-private predicate uniqueValueNumber(Instruction instr, IRFunction irFunc) {
-  instr.getEnclosingIRFunction() = irFunc and
-  not instr.getResultIRType() instanceof IRVoidType and
-  not numberableInstruction(instr)
-}
-
 /**
  * Gets the value number assigned to `instr`, if any. Returns at most one result.
  */
-cached
-ValueNumber valueNumber(Instruction instr) {
-  result = nonUniqueValueNumber(instr)
-  or
-  exists(IRFunction irFunc |
-    uniqueValueNumber(instr, irFunc) and
-    result = TUniqueValueNumber(irFunc, instr)
-  )
-}
+ValueNumber valueNumber(Instruction instr) { result = tvalueNumber(instr) }
 
 /**
  * Gets the value number assigned to the exact definition of `op`, if any.
  * Returns at most one result.
  */
-ValueNumber valueNumberOfOperand(Operand op) { result = valueNumber(op.getDef()) }
-
-/**
- * Gets the value number assigned to `instr`, if any, unless that instruction is assigned a unique
- * value number.
- */
-private ValueNumber nonUniqueValueNumber(Instruction instr) {
-  exists(IRFunction irFunc |
-    irFunc = instr.getEnclosingIRFunction() and
-    (
-      exists(IRVariable var |
-        variableAddressValueNumber(instr, irFunc, var) and
-        result = TVariableAddressValueNumber(irFunc, var)
-      )
-      or
-      exists(IRVariable var |
-        initializeParameterValueNumber(instr, irFunc, var) and
-        result = TInitializeParameterValueNumber(irFunc, var)
-      )
-      or
-      initializeThisValueNumber(instr, irFunc) and
-      result = TInitializeThisValueNumber(irFunc)
-      or
-      exists(IRType type, string value |
-        constantValueNumber(instr, irFunc, type, value) and
-        result = TConstantValueNumber(irFunc, type, value)
-      )
-      or
-      exists(IRType type, string value |
-        stringConstantValueNumber(instr, irFunc, type, value) and
-        result = TStringConstantValueNumber(irFunc, type, value)
-      )
-      or
-      exists(Language::Field field, ValueNumber objectAddress |
-        fieldAddressValueNumber(instr, irFunc, field, objectAddress) and
-        result = TFieldAddressValueNumber(irFunc, field, objectAddress)
-      )
-      or
-      exists(Opcode opcode, IRType type, ValueNumber leftOperand, ValueNumber rightOperand |
-        binaryValueNumber(instr, irFunc, opcode, type, leftOperand, rightOperand) and
-        result = TBinaryValueNumber(irFunc, opcode, type, leftOperand, rightOperand)
-      )
-      or
-      exists(Opcode opcode, IRType type, ValueNumber operand |
-        unaryValueNumber(instr, irFunc, opcode, type, operand) and
-        result = TUnaryValueNumber(irFunc, opcode, type, operand)
-      )
-      or
-      exists(
-        Opcode opcode, Language::Class baseClass, Language::Class derivedClass, ValueNumber operand
-      |
-        inheritanceConversionValueNumber(instr, irFunc, opcode, baseClass, derivedClass, operand) and
-        result = TInheritanceConversionValueNumber(irFunc, opcode, baseClass, derivedClass, operand)
-      )
-      or
-      exists(
-        Opcode opcode, IRType type, int elementSize, ValueNumber leftOperand,
-        ValueNumber rightOperand
-      |
-        pointerArithmeticValueNumber(instr, irFunc, opcode, type, elementSize, leftOperand,
-          rightOperand) and
-        result =
-          TPointerArithmeticValueNumber(irFunc, opcode, type, elementSize, leftOperand, rightOperand)
-      )
-      or
-      exists(IRType type, ValueNumber memOperand, ValueNumber operand |
-        loadTotalOverlapValueNumber(instr, irFunc, type, memOperand, operand) and
-        result = TLoadTotalOverlapValueNumber(irFunc, type, memOperand, operand)
-      )
-      or
-      // The value number of a copy is just the value number of its source value.
-      result = valueNumber(instr.(CongruentCopyInstruction).getSourceValue())
-    )
-  )
-}
+ValueNumber valueNumberOfOperand(Operand op) { result = tvalueNumberOfOperand(op) }

cpp/ql/src/semmle/code/cpp/ir/implementation/aliased_ssa/gvn/internal/ValueNumberingImports.qll

@@ -1,2 +1,4 @@
+import semmle.code.cpp.ir.implementation.aliased_ssa.IR
 import semmle.code.cpp.ir.internal.Overlap
 import semmle.code.cpp.ir.internal.IRCppLanguage as Language
+import semmle.code.cpp.Location