Merge pull request #1750 from dave-bartolomeo/dave/ZooKeeper

C++: Minimal IR support for `GNUVectorType`

Author: Robert Marsh

cpp/ql/src/semmle/code/cpp/exprs/BuiltInOperations.qll

@@ -4,6 +4,7 @@ import semmle.code.cpp.exprs.Expr
  * A C/C++ builtin operation.
  */
 abstract class BuiltInOperation extends Expr {
+  override string getCanonicalQLClass() { result = "BuiltInOperation" }
 }
 
 /**

cpp/ql/src/semmle/code/cpp/exprs/Literal.qll

@@ -234,16 +234,33 @@ class ClassAggregateLiteral extends AggregateLiteral {
 }
 
 /**
- * A C/C++ aggregate literal that initializes an array
+ * A C/C++ aggregate literal that initializes an array or a GNU vector type.
  */
-class ArrayAggregateLiteral extends AggregateLiteral {
-  ArrayType arrayType;
+class ArrayOrVectorAggregateLiteral extends AggregateLiteral {
+  ArrayOrVectorAggregateLiteral() {
+    exists(DerivedType type |
+      type = this.getUnspecifiedType() and
+      (
+        type instanceof ArrayType or
+        type instanceof GNUVectorType
+      )
+    )
+  }
 
-  ArrayAggregateLiteral() {
-    arrayType = this.getUnspecifiedType()
+  /**
+   * Gets the number of elements initialized by this initializer list, either explicitly with an
+   * expression, or by implicit value initialization.
+   */
+  int getArraySize() {
+    none()
   }
 
-  override string getCanonicalQLClass() { result = "ArrayAggregateLiteral" }
+  /**
+   * Gets the type of the elements in the initializer list.
+   */
+  Type getElementType() {
+    none()
+  }
 
   /**
    * Gets the expression within the aggregate literal that is used to initialize
@@ -262,7 +279,7 @@ class ArrayAggregateLiteral extends AggregateLiteral {
   bindingset[elementIndex]
   predicate isInitialized(int elementIndex) {
     elementIndex >= 0 and
-    elementIndex < arrayType.getArraySize()
+    elementIndex < getArraySize()
   }
 
   /**
@@ -279,3 +296,45 @@ class ArrayAggregateLiteral extends AggregateLiteral {
     not exists(getElementExpr(elementIndex))
   }
 }
+
+/**
+ * A C/C++ aggregate literal that initializes an array
+ */
+class ArrayAggregateLiteral extends ArrayOrVectorAggregateLiteral {
+  ArrayType arrayType;
+
+  ArrayAggregateLiteral() {
+    arrayType = this.getUnspecifiedType()
+  }
+
+  override string getCanonicalQLClass() { result = "ArrayAggregateLiteral" }
+
+  override int getArraySize() {
+    result = arrayType.getArraySize()
+  }
+
+  override Type getElementType() {
+    result = arrayType.getBaseType()
+  }
+}
+
+/**
+ * A C/C++ aggregate literal that initializes a GNU vector type.
+ */
+class VectorAggregateLiteral extends ArrayOrVectorAggregateLiteral {
+  GNUVectorType vectorType;
+
+  VectorAggregateLiteral() {
+    vectorType = this.getUnspecifiedType()
+  }
+
+  override string getCanonicalQLClass() { result = "VectorAggregateLiteral" }
+
+  override int getArraySize() {
+    result = vectorType.getNumElements()
+  }
+
+  override Type getElementType() {
+    result = vectorType.getBaseType()
+  }
+}

cpp/ql/src/semmle/code/cpp/ir/implementation/Opcode.qll

@@ -57,6 +57,7 @@ private newtype TOpcode =
   TUnmodeledUse() or
   TAliasedDefinition() or
   TPhi() or
+  TBuiltIn() or
   TVarArgsStart() or
   TVarArgsEnd() or
   TVarArg() or
@@ -115,7 +116,7 @@ abstract class CatchOpcode extends Opcode {}
 
 abstract class OpcodeWithCondition extends Opcode {}
 
-abstract class BuiltInOpcode extends Opcode {}
+abstract class BuiltInOperationOpcode extends Opcode {}
 
 abstract class SideEffectOpcode extends Opcode {}
 
@@ -204,10 +205,11 @@ module Opcode {
   class UnmodeledUse extends Opcode, TUnmodeledUse { override final string toString() { result = "UnmodeledUse" } }
   class AliasedDefinition extends Opcode, TAliasedDefinition { override final string toString() { result = "AliasedDefinition" } }
   class Phi extends Opcode, TPhi { override final string toString() { result = "Phi" } }
-  class VarArgsStart extends BuiltInOpcode, TVarArgsStart { override final string toString() { result = "VarArgsStart" } }
-  class VarArgsEnd extends BuiltInOpcode, TVarArgsEnd { override final string toString() { result = "VarArgsEnd" } }
-  class VarArg extends BuiltInOpcode, TVarArg { override final string toString() { result = "VarArg" } }
-  class VarArgCopy extends BuiltInOpcode, TVarArgCopy { override final string toString() { result = "VarArgCopy" } }
+  class BuiltIn extends BuiltInOperationOpcode, TBuiltIn { override final string toString() { result = "BuiltIn" } }
+  class VarArgsStart extends BuiltInOperationOpcode, TVarArgsStart { override final string toString() { result = "VarArgsStart" } }
+  class VarArgsEnd extends BuiltInOperationOpcode, TVarArgsEnd { override final string toString() { result = "VarArgsEnd" } }
+  class VarArg extends BuiltInOperationOpcode, TVarArg { override final string toString() { result = "VarArg" } }
+  class VarArgCopy extends BuiltInOperationOpcode, TVarArgCopy { override final string toString() { result = "VarArgCopy" } }
   class CallSideEffect extends MayWriteSideEffectOpcode, TCallSideEffect { override final string toString() { result = "CallSideEffect" } }
   class CallReadSideEffect extends ReadSideEffectOpcode, TCallReadSideEffect { override final string toString() { result = "CallReadSideEffect" } }
   class IndirectReadSideEffect extends ReadSideEffectOpcode, MemoryAccessOpcode, TIndirectReadSideEffect { override final string toString() { result = "IndirectReadSideEffect" } }

cpp/ql/src/semmle/code/cpp/ir/implementation/aliased_ssa/Instruction.qll

@@ -71,7 +71,7 @@ module InstructionSanity {
       operand.getOperandTag() = tag) and
     not expectsOperand(instr, tag) and
     not (instr instanceof CallInstruction and tag instanceof ArgumentOperandTag) and
-    not (instr instanceof BuiltInInstruction and tag instanceof PositionalArgumentOperandTag) and
+    not (instr instanceof BuiltInOperationInstruction and tag instanceof PositionalArgumentOperandTag) and
     not (instr instanceof InlineAsmInstruction and tag instanceof AsmOperandTag)
   }
 
@@ -1831,8 +1831,29 @@ class UnreachedInstruction extends Instruction {
  * An instruction representing a built-in operation. This is used to represent
  * operations such as access to variable argument lists.
  */
-class BuiltInInstruction extends Instruction {
+class BuiltInOperationInstruction extends Instruction {
+  Language::BuiltInOperation operation;
+
+  BuiltInOperationInstruction() {
+    getOpcode() instanceof BuiltInOperationOpcode and
+    operation = Construction::getInstructionBuiltInOperation(this)
+  }
+
+  final Language::BuiltInOperation getBuiltInOperation() {
+    result = operation
+  }
+}
+
+/**
+ * An instruction representing a built-in operation that does not have a specific opcode. The
+ * actual operation is specified by the `getBuiltInOperation()` predicate.
+ */
+class BuiltInInstruction extends BuiltInOperationInstruction {
   BuiltInInstruction() {
-    getOpcode() instanceof BuiltInOpcode
+    getOpcode() instanceof Opcode::BuiltIn
+  }
+
+  override final string getImmediateString() {
+    result = getBuiltInOperation().toString()
   }
 }

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

@@ -325,6 +325,10 @@ cached private module Cached {
     result = getOldInstruction(instruction).(OldIR::StringConstantInstruction).getValue()
   }
 
+  cached BuiltInOperation getInstructionBuiltInOperation(Instruction instruction) {
+    result = getOldInstruction(instruction).(OldIR::BuiltInOperationInstruction).getBuiltInOperation()
+  }
+
   cached Type getInstructionExceptionType(Instruction instruction) {
     result = getOldInstruction(instruction).(OldIR::CatchByTypeInstruction).getExceptionType()
   }

cpp/ql/src/semmle/code/cpp/ir/implementation/raw/Instruction.qll

@@ -71,7 +71,7 @@ module InstructionSanity {
       operand.getOperandTag() = tag) and
     not expectsOperand(instr, tag) and
     not (instr instanceof CallInstruction and tag instanceof ArgumentOperandTag) and
-    not (instr instanceof BuiltInInstruction and tag instanceof PositionalArgumentOperandTag) and
+    not (instr instanceof BuiltInOperationInstruction and tag instanceof PositionalArgumentOperandTag) and
     not (instr instanceof InlineAsmInstruction and tag instanceof AsmOperandTag)
   }
 
@@ -1831,8 +1831,29 @@ class UnreachedInstruction extends Instruction {
  * An instruction representing a built-in operation. This is used to represent
  * operations such as access to variable argument lists.
  */
-class BuiltInInstruction extends Instruction {
+class BuiltInOperationInstruction extends Instruction {
+  Language::BuiltInOperation operation;
+
+  BuiltInOperationInstruction() {
+    getOpcode() instanceof BuiltInOperationOpcode and
+    operation = Construction::getInstructionBuiltInOperation(this)
+  }
+
+  final Language::BuiltInOperation getBuiltInOperation() {
+    result = operation
+  }
+}
+
+/**
+ * An instruction representing a built-in operation that does not have a specific opcode. The
+ * actual operation is specified by the `getBuiltInOperation()` predicate.
+ */
+class BuiltInInstruction extends BuiltInOperationInstruction {
   BuiltInInstruction() {
-    getOpcode() instanceof BuiltInOpcode
+    getOpcode() instanceof Opcode::BuiltIn
+  }
+
+  override final string getImmediateString() {
+    result = getBuiltInOperation().toString()
   }
 }

cpp/ql/src/semmle/code/cpp/ir/implementation/raw/internal/IRConstruction.qll

@@ -250,6 +250,12 @@ cached private module Cached {
         getInstructionTag(instruction))
   }
 
+  cached BuiltInOperation getInstructionBuiltInOperation(Instruction instruction) {
+    result =
+      getInstructionTranslatedElement(instruction).getInstructionBuiltInOperation(
+        getInstructionTag(instruction))
+  }
+
   cached Type getInstructionExceptionType(Instruction instruction) {
     result =
       getInstructionTranslatedElement(instruction).getInstructionExceptionType(

cpp/ql/src/semmle/code/cpp/ir/implementation/raw/internal/TranslatedElement.qll

@@ -274,7 +274,7 @@ newtype TTranslatedElement =
       exists(ClassAggregateLiteral initList |
         initList.getFieldExpr(_).getFullyConverted() = expr
       ) or
-      exists(ArrayAggregateLiteral initList |
+      exists(ArrayOrVectorAggregateLiteral initList |
         initList.getElementExpr(_).getFullyConverted() = expr
       ) or
       exists(ReturnStmt returnStmt |
@@ -320,13 +320,13 @@ newtype TTranslatedElement =
   } or
   // The initialization of an array element via a member of an initializer list.
   TTranslatedExplicitElementInitialization(
-      ArrayAggregateLiteral initList, int elementIndex) {
+      ArrayOrVectorAggregateLiteral initList, int elementIndex) {
     not ignoreExpr(initList) and
     exists(initList.getElementExpr(elementIndex))
   } or
   // The value initialization of a range of array elements that were omitted
   // from an initializer list.
-  TTranslatedElementValueInitialization(ArrayAggregateLiteral initList,
+  TTranslatedElementValueInitialization(ArrayOrVectorAggregateLiteral initList,
       int elementIndex, int elementCount) {
     not ignoreExpr(initList) and
     isFirstValueInitializedElementInRange(initList, elementIndex) and
@@ -412,12 +412,13 @@ newtype TTranslatedElement =
  * `initList`, the result is the total number of elements in the array being
  * initialized.
  */
-private int getEndOfValueInitializedRange(ArrayAggregateLiteral initList, int afterElementIndex) {
+private int getEndOfValueInitializedRange(ArrayOrVectorAggregateLiteral initList,
+    int afterElementIndex) {
   result = getNextExplicitlyInitializedElementAfter(initList, afterElementIndex)
   or
   isFirstValueInitializedElementInRange(initList, afterElementIndex) and
   not exists(getNextExplicitlyInitializedElementAfter(initList, afterElementIndex)) and
-  result = initList.getUnspecifiedType().(ArrayType).getArraySize()
+  result = initList.getArraySize()
 }
 
 /**
@@ -427,7 +428,7 @@ private int getEndOfValueInitializedRange(ArrayAggregateLiteral initList, int af
  * `initList`.
  */
 private int getNextExplicitlyInitializedElementAfter(
-    ArrayAggregateLiteral initList, int afterElementIndex) {
+    ArrayOrVectorAggregateLiteral initList, int afterElementIndex) {
   isFirstValueInitializedElementInRange(initList, afterElementIndex) and
   result = min(int i | exists(initList.getElementExpr(i)) and i > afterElementIndex)
 }
@@ -437,7 +438,7 @@ private int getNextExplicitlyInitializedElementAfter(
  * range of one or more consecutive value-initialized elements in `initList`.
  */
 private predicate isFirstValueInitializedElementInRange(
-  ArrayAggregateLiteral initList, int elementIndex) {
+  ArrayOrVectorAggregateLiteral initList, int elementIndex) {
   initList.isValueInitialized(elementIndex) and
   (
     elementIndex = 0 or
@@ -634,6 +635,13 @@ abstract class TranslatedElement extends TTranslatedElement {
     none()
   }
 
+  /**
+   * If the instruction specified by `tag` is a `BuiltInInstruction`, gets the built-in operation.
+   */
+  BuiltInOperation getInstructionBuiltInOperation(InstructionTag tag) {
+    none()
+  }
+
   /**
    * If the instruction specified by `tag` is a `CatchByTypeInstruction`,
    * gets the type of the exception to be caught.

cpp/ql/src/semmle/code/cpp/ir/implementation/raw/internal/TranslatedExpr.qll

@@ -612,10 +612,15 @@ class TranslatedPostfixCrementOperation extends TranslatedCrementOperation {
   }
 }
 
+/**
+ * IR translation of an array access expression (e.g. `a[i]`). The array being accessed will either
+ * be a prvalue of pointer type (possibly due to an implicit array-to-pointer conversion), or a
+ * glvalue of a GNU vector type.
+ */
 class TranslatedArrayExpr extends TranslatedNonConstantExpr {
   override ArrayExpr expr;
 
-  override Instruction getFirstInstruction() {
+  override final Instruction getFirstInstruction() {
     result = getBaseOperand().getFirstInstruction()
   }
 
@@ -650,8 +655,8 @@ class TranslatedArrayExpr extends TranslatedNonConstantExpr {
     Type resultType, boolean isGLValue) {
     tag = OnlyInstructionTag() and
     opcode instanceof Opcode::PointerAdd and
-    resultType = getBaseOperand().getResultType() and
-    isGLValue = false
+    resultType = getResultType() and
+    isGLValue = true
   }
 
   override Instruction getInstructionOperand(InstructionTag tag,
@@ -671,7 +676,7 @@ class TranslatedArrayExpr extends TranslatedNonConstantExpr {
 
   override int getInstructionElementSize(InstructionTag tag) {
     tag = OnlyInstructionTag() and
-    result = max(getBaseOperand().getResultType().(PointerType).getBaseType().getSize())
+    result = max(getResultType().getSize())
   }
 
   private TranslatedExpr getBaseOperand() {
@@ -2373,7 +2378,13 @@ class TranslatedReThrowExpr extends TranslatedThrowExpr {
  * The IR translation of a built-in operation (i.e. anything that extends
  * `BuiltInOperation`).
  */
-abstract class TranslatedBuiltInOperation extends TranslatedNonConstantExpr {
+class TranslatedBuiltInOperation extends TranslatedNonConstantExpr {
+  override BuiltInOperation expr;
+
+  TranslatedBuiltInOperation() {
+    not expr instanceof BuiltInOperationBuiltInAddressOf  // Handled specially
+  }
+
   override final Instruction getResult() {
     result = getInstruction(OnlyInstructionTag())
   }
@@ -2423,7 +2434,14 @@ abstract class TranslatedBuiltInOperation extends TranslatedNonConstantExpr {
     )
   }
 
-  abstract Opcode getOpcode();
+  Opcode getOpcode() {
+    result instanceof Opcode::BuiltIn
+  }
+
+  override final BuiltInOperation getInstructionBuiltInOperation(InstructionTag tag) {
+    tag = OnlyInstructionTag() and
+    result = expr
+  }
 }
 
 /**