C++: Fix IR generation for builtins and add flow through __builtin_bit_cast

cpp/ql/lib/semmle/code/cpp/ir/dataflow/internal/DataFlowUtil.qll

@@ -114,6 +114,13 @@ predicate conversionFlow(
       instrTo.(CheckedConvertOrNullInstruction).getUnaryOperand() = opFrom
       or
       instrTo.(InheritanceConversionInstruction).getUnaryOperand() = opFrom
+      or
+      exists(BuiltInInstruction builtIn |
+        builtIn = instrTo and
+        // __builtin_bit_cast
+        builtIn.getBuiltInOperation() instanceof BuiltInBitCast and
+        opFrom = builtIn.getAnOperand()
+      )
     )
     or
     additional = true and

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

@@ -3208,9 +3208,20 @@ class TranslatedBuiltInOperation extends TranslatedNonConstantExpr {
 
   final override Instruction getResult() { result = this.getInstruction(OnlyInstructionTag()) }
 
+  /**
+   * Gets the rnk'th (0-indexed) child for which a `TranslatedElement` exists.
+   *
+   * We use this predicate to filter out `TypeName` expressions that sometimes
+   * occur in builtin operations since the IR doesn't have an instruction to
+   * represent a reference to a type.
+   */
+  private TranslatedElement getRankedChild(int rnk) {
+    result = rank[rnk + 1](int id, TranslatedElement te | te = this.getChild(id) | te order by id)
+  }
+
   final override Instruction getFirstInstruction(EdgeKind kind) {
-    if exists(this.getChild(0))
-    then result = this.getChild(0).getFirstInstruction(kind)
+    if exists(this.getRankedChild(0))
+    then result = this.getRankedChild(0).getFirstInstruction(kind)
     else (
       kind instanceof GotoEdge and result = this.getInstruction(OnlyInstructionTag())
     )
@@ -3230,11 +3241,11 @@ class TranslatedBuiltInOperation extends TranslatedNonConstantExpr {
   }
 
   final override Instruction getChildSuccessorInternal(TranslatedElement child, EdgeKind kind) {
-    exists(int id | child = this.getChild(id) |
-      result = this.getChild(id + 1).getFirstInstruction(kind)
+    exists(int id | child = this.getRankedChild(id) |
+      result = this.getRankedChild(id + 1).getFirstInstruction(kind)
       or
       kind instanceof GotoEdge and
-      not exists(this.getChild(id + 1)) and
+      not exists(this.getRankedChild(id + 1)) and
       result = this.getInstruction(OnlyInstructionTag())
     )
   }
@@ -3249,7 +3260,7 @@ class TranslatedBuiltInOperation extends TranslatedNonConstantExpr {
     tag = OnlyInstructionTag() and
     exists(int index |
       operandTag = positionalArgumentOperand(index) and
-      result = this.getChild(index).(TranslatedExpr).getResult()
+      result = this.getRankedChild(index).(TranslatedExpr).getResult()
     )
   }
 

cpp/ql/test/library-tests/dataflow/dataflow-tests/clang.cpp

@@ -52,3 +52,9 @@ void following_pointers( // $ ast-def=sourceStruct1_ptr ir-def=*cleanArray1 ir-d
   sink(stackArray); // $ ast,ir
   indirect_sink(stackArray); // $ ast ir=50:25 ir=50:35 ir=51:19
 }
+
+void test_bitcast() {
+  unsigned long x = source();
+  double d = __builtin_bit_cast(double, x);
+  sink(d); // $ ir MISSING: ast
+}

cpp/ql/test/library-tests/dataflow/dataflow-tests/test-source-sink.expected

@@ -153,6 +153,7 @@ irFlow
 | clang.cpp:50:25:50:30 | call to source | clang.cpp:53:17:53:26 | *stackArray |
 | clang.cpp:50:35:50:40 | call to source | clang.cpp:53:17:53:26 | *stackArray |
 | clang.cpp:51:19:51:24 | call to source | clang.cpp:53:17:53:26 | *stackArray |
+| clang.cpp:57:21:57:28 | call to source | clang.cpp:59:8:59:8 | d |
 | dispatch.cpp:9:37:9:42 | call to source | dispatch.cpp:35:16:35:25 | call to notSource1 |
 | dispatch.cpp:9:37:9:42 | call to source | dispatch.cpp:43:15:43:24 | call to notSource1 |
 | dispatch.cpp:10:37:10:42 | call to source | dispatch.cpp:36:16:36:25 | call to notSource2 |

cpp/ql/test/library-tests/ir/ir/PrintAST.expected

@@ -22734,6 +22734,25 @@ ir.cpp:
 # 2552|             Type = [Class] ClassWithDestructor
 # 2552|             ValueCategory = xvalue
 # 2553|     getStmt(1): [ReturnStmt] return ...
+# 2555| [TopLevelFunction] void builtin_bitcast(unsigned long)
+# 2555|   <params>: 
+# 2555|     getParameter(0): [Parameter] ul
+# 2555|         Type = [LongType] unsigned long
+# 2555|   getEntryPoint(): [BlockStmt] { ... }
+# 2556|     getStmt(0): [DeclStmt] declaration
+# 2556|       getDeclarationEntry(0): [VariableDeclarationEntry] definition of d
+# 2556|           Type = [DoubleType] double
+# 2556|         getVariable().getInitializer(): [Initializer] initializer for d
+# 2556|           getExpr(): [BuiltInBitCast] __builtin_bit_cast
+# 2556|               Type = [DoubleType] double
+# 2556|               ValueCategory = prvalue
+# 2556|             getChild(0): [TypeName] double
+# 2556|                 Type = [DoubleType] double
+# 2556|                 ValueCategory = prvalue
+# 2556|             getChild(1): [VariableAccess] ul
+# 2556|                 Type = [LongType] unsigned long
+# 2556|                 ValueCategory = prvalue(load)
+# 2557|     getStmt(1): [ReturnStmt] return ...
 perf-regression.cpp:
 #    4| [CopyAssignmentOperator] Big& Big::operator=(Big const&)
 #    4|   <params>: 

cpp/ql/test/library-tests/ir/ir/aliased_ir.expected

@@ -18351,6 +18351,24 @@ ir.cpp:
 # 2550|   Block 2
 # 2550|     v2550_10(void) = Unreached : 
 
+# 2555| void builtin_bitcast(unsigned long)
+# 2555|   Block 0
+# 2555|     v2555_1(void)                 = EnterFunction               : 
+# 2555|     m2555_2(unknown)              = AliasedDefinition           : 
+# 2555|     m2555_3(unknown)              = InitializeNonLocal          : 
+# 2555|     m2555_4(unknown)              = Chi                         : total:m2555_2, partial:m2555_3
+# 2555|     r2555_5(glval<unsigned long>) = VariableAddress[ul]         : 
+# 2555|     m2555_6(unsigned long)        = InitializeParameter[ul]     : &:r2555_5
+# 2556|     r2556_1(glval<double>)        = VariableAddress[d]          : 
+# 2556|     r2556_2(glval<unsigned long>) = VariableAddress[ul]         : 
+# 2556|     r2556_3(unsigned long)        = Load[ul]                    : &:r2556_2, m2555_6
+# 2556|     r2556_4(double)               = BuiltIn[__builtin_bit_cast] : 0:r2556_3
+# 2556|     m2556_5(double)               = Store[d]                    : &:r2556_1, r2556_4
+# 2557|     v2557_1(void)                 = NoOp                        : 
+# 2555|     v2555_7(void)                 = ReturnVoid                  : 
+# 2555|     v2555_8(void)                 = AliasedUse                  : m2555_3
+# 2555|     v2555_9(void)                 = ExitFunction                : 
+
 perf-regression.cpp:
 #    6| void Big::Big()
 #    6|   Block 0

cpp/ql/test/library-tests/ir/ir/ir.cpp

@@ -2552,4 +2552,8 @@ void constexpr_inconsistency(bool b) {
     ;
 }
 
+void builtin_bitcast(unsigned long ul) {
+    double d = __builtin_bit_cast(double, ul);
+}
+
 // semmle-extractor-options: -std=c++20 --clang

cpp/ql/test/library-tests/ir/ir/raw_ir.expected

@@ -16693,6 +16693,23 @@ ir.cpp:
 # 2550|     v2550_7(void)                       = AliasedUse                            : ~m?
 # 2550|     v2550_8(void)                       = ExitFunction                          : 
 
+# 2555| void builtin_bitcast(unsigned long)
+# 2555|   Block 0
+# 2555|     v2555_1(void)                 = EnterFunction               : 
+# 2555|     mu2555_2(unknown)             = AliasedDefinition           : 
+# 2555|     mu2555_3(unknown)             = InitializeNonLocal          : 
+# 2555|     r2555_4(glval<unsigned long>) = VariableAddress[ul]         : 
+# 2555|     mu2555_5(unsigned long)       = InitializeParameter[ul]     : &:r2555_4
+# 2556|     r2556_1(glval<double>)        = VariableAddress[d]          : 
+# 2556|     r2556_2(glval<unsigned long>) = VariableAddress[ul]         : 
+# 2556|     r2556_3(unsigned long)        = Load[ul]                    : &:r2556_2, ~m?
+# 2556|     r2556_4(double)               = BuiltIn[__builtin_bit_cast] : 0:r2556_3
+# 2556|     mu2556_5(double)              = Store[d]                    : &:r2556_1, r2556_4
+# 2557|     v2557_1(void)                 = NoOp                        : 
+# 2555|     v2555_6(void)                 = ReturnVoid                  : 
+# 2555|     v2555_7(void)                 = AliasedUse                  : ~m?
+# 2555|     v2555_8(void)                 = ExitFunction                : 
+
 perf-regression.cpp:
 #    6| void Big::Big()
 #    6|   Block 0