Reapply "[clang][CodeGen] Zero init unspecified fields in initializers in C" (llvm#109898) (llvm#110051)

This reverts commit d50eaac. Also fixes
a bug calculating offsets for bit fields in the original patch.

Author: yabinc

clang/docs/LanguageExtensions.rst

@@ -5883,3 +5883,26 @@ specify the starting offset to begin embedding from. The resources is treated
 as being empty if the specified offset is larger than the number of bytes in
 the resource. The offset will be applied *before* any ``limit`` parameters are
 applied.
+
+Union and aggregate initialization in C
+=======================================
+
+In C23 (N2900), when an object is initialized from initializer ``= {}``, all
+elements of arrays, all members of structs, and the first members of unions are
+empty-initialized recursively. In addition, all padding bits are initialized to
+zero.
+
+Clang guarantees the following behaviors:
+
+* ``1:`` Clang supports initializer ``= {}`` mentioned above in all C
+  standards.
+
+* ``2:`` When unions are initialized from initializer ``= {}``, bytes outside
+  of the first members of unions are also initialized to zero.
+
+* ``3:`` When unions, structures and arrays are initialized from initializer
+  ``= { initializer-list }``, all members not explicitly initialized in
+  the initializer list are empty-initialized recursively. In addition, all
+  padding bits are initialized to zero.
+
+Currently, the above extension only applies to C source code, not C++.

clang/lib/CodeGen/CGExprAgg.cpp

@@ -12,6 +12,7 @@
 
 #include "CGCXXABI.h"
 #include "CGObjCRuntime.h"
+#include "CGRecordLayout.h"
 #include "CodeGenFunction.h"
 #include "CodeGenModule.h"
 #include "ConstantEmitter.h"
@@ -64,6 +65,9 @@ class AggExprEmitter : public StmtVisitor<AggExprEmitter> {
   void withReturnValueSlot(const Expr *E,
                            llvm::function_ref<RValue(ReturnValueSlot)> Fn);
 
+  void DoZeroInitPadding(uint64_t &PaddingStart, uint64_t PaddingEnd,
+                         const FieldDecl *NextField);
+
 public:
   AggExprEmitter(CodeGenFunction &cgf, AggValueSlot Dest, bool IsResultUnused)
     : CGF(cgf), Builder(CGF.Builder), Dest(Dest),
@@ -1698,6 +1702,9 @@ void AggExprEmitter::VisitCXXParenListOrInitListExpr(
   // Prepare a 'this' for CXXDefaultInitExprs.
   CodeGenFunction::FieldConstructionScope FCS(CGF, Dest.getAddress());
 
+  const bool ZeroInitPadding =
+      CGF.CGM.shouldZeroInitPadding() && !Dest.isZeroed();
+
   if (record->isUnion()) {
     // Only initialize one field of a union. The field itself is
     // specified by the initializer list.
@@ -1722,16 +1729,27 @@ void AggExprEmitter::VisitCXXParenListOrInitListExpr(
     if (NumInitElements) {
       // Store the initializer into the field
       EmitInitializationToLValue(InitExprs[0], FieldLoc);
+      if (ZeroInitPadding) {
+        uint64_t TotalSize = CGF.getContext().toBits(
+            Dest.getPreferredSize(CGF.getContext(), DestLV.getType()));
+        uint64_t FieldSize = CGF.getContext().getTypeSize(FieldLoc.getType());
+        DoZeroInitPadding(FieldSize, TotalSize, nullptr);
+      }
     } else {
       // Default-initialize to null.
-      EmitNullInitializationToLValue(FieldLoc);
+      if (ZeroInitPadding)
+        EmitNullInitializationToLValue(DestLV);
+      else
+        EmitNullInitializationToLValue(FieldLoc);
     }
-
     return;
   }
 
   // Here we iterate over the fields; this makes it simpler to both
   // default-initialize fields and skip over unnamed fields.
+  const ASTRecordLayout &Layout = CGF.getContext().getASTRecordLayout(record);
+  uint64_t PaddingStart = 0;
+
   for (const auto *field : record->fields()) {
     // We're done once we hit the flexible array member.
     if (field->getType()->isIncompleteArrayType())
@@ -1748,6 +1766,9 @@ void AggExprEmitter::VisitCXXParenListOrInitListExpr(
         CGF.getTypes().isZeroInitializable(ExprToVisit->getType()))
       break;
 
+    if (ZeroInitPadding)
+      DoZeroInitPadding(PaddingStart,
+                        Layout.getFieldOffset(field->getFieldIndex()), field);
 
     LValue LV = CGF.EmitLValueForFieldInitialization(DestLV, field);
     // We never generate write-barries for initialized fields.
@@ -1774,6 +1795,54 @@ void AggExprEmitter::VisitCXXParenListOrInitListExpr(
       }
     }
   }
+  if (ZeroInitPadding) {
+    uint64_t TotalSize = CGF.getContext().toBits(
+        Dest.getPreferredSize(CGF.getContext(), DestLV.getType()));
+    DoZeroInitPadding(PaddingStart, TotalSize, nullptr);
+  }
+}
+
+void AggExprEmitter::DoZeroInitPadding(uint64_t &PaddingStart,
+                                       uint64_t PaddingEnd,
+                                       const FieldDecl *NextField) {
+
+  auto InitBytes = [&](uint64_t StartBit, uint64_t EndBit) {
+    CharUnits Start = CGF.getContext().toCharUnitsFromBits(StartBit);
+    CharUnits End = CGF.getContext().toCharUnitsFromBits(EndBit);
+    Address Addr = Dest.getAddress().withElementType(CGF.CharTy);
+    if (!Start.isZero())
+      Addr = Builder.CreateConstGEP(Addr, Start.getQuantity());
+    llvm::Constant *SizeVal = Builder.getInt64((End - Start).getQuantity());
+    CGF.Builder.CreateMemSet(Addr, Builder.getInt8(0), SizeVal, false);
+  };
+
+  if (NextField != nullptr && NextField->isBitField()) {
+    // For bitfield, zero init StorageSize before storing the bits. So we don't
+    // need to handle big/little endian.
+    const CGRecordLayout &RL =
+        CGF.getTypes().getCGRecordLayout(NextField->getParent());
+    const CGBitFieldInfo &Info = RL.getBitFieldInfo(NextField);
+    uint64_t StorageStart = CGF.getContext().toBits(Info.StorageOffset);
+    if (StorageStart + Info.StorageSize > PaddingStart) {
+      if (StorageStart > PaddingStart)
+        InitBytes(PaddingStart, StorageStart);
+      Address Addr = Dest.getAddress();
+      if (!Info.StorageOffset.isZero())
+        Addr = Builder.CreateConstGEP(Addr.withElementType(CGF.CharTy),
+                                      Info.StorageOffset.getQuantity());
+      Addr = Addr.withElementType(
+          llvm::Type::getIntNTy(CGF.getLLVMContext(), Info.StorageSize));
+      Builder.CreateStore(Builder.getIntN(Info.StorageSize, 0), Addr);
+      PaddingStart = StorageStart + Info.StorageSize;
+    }
+    return;
+  }
+
+  if (PaddingStart < PaddingEnd)
+    InitBytes(PaddingStart, PaddingEnd);
+  if (NextField != nullptr)
+    PaddingStart =
+        PaddingEnd + CGF.getContext().getTypeSize(NextField->getType());
 }
 
 void AggExprEmitter::VisitArrayInitLoopExpr(const ArrayInitLoopExpr *E,

clang/lib/CodeGen/CGExprConstant.cpp

@@ -42,6 +42,16 @@ using namespace CodeGen;
 namespace {
 class ConstExprEmitter;
 
+llvm::Constant *getPadding(const CodeGenModule &CGM, CharUnits PadSize) {
+  llvm::Type *Ty = CGM.CharTy;
+  if (PadSize > CharUnits::One())
+    Ty = llvm::ArrayType::get(Ty, PadSize.getQuantity());
+  if (CGM.shouldZeroInitPadding()) {
+    return llvm::Constant::getNullValue(Ty);
+  }
+  return llvm::UndefValue::get(Ty);
+}
+
 struct ConstantAggregateBuilderUtils {
   CodeGenModule &CGM;
 
@@ -61,10 +71,7 @@ struct ConstantAggregateBuilderUtils {
   }
 
   llvm::Constant *getPadding(CharUnits PadSize) const {
-    llvm::Type *Ty = CGM.CharTy;
-    if (PadSize > CharUnits::One())
-      Ty = llvm::ArrayType::get(Ty, PadSize.getQuantity());
-    return llvm::UndefValue::get(Ty);
+    return ::getPadding(CGM, PadSize);
   }
 
   llvm::Constant *getZeroes(CharUnits ZeroSize) const {
@@ -591,6 +598,11 @@ class ConstStructBuilder {
   bool Build(const InitListExpr *ILE, bool AllowOverwrite);
   bool Build(const APValue &Val, const RecordDecl *RD, bool IsPrimaryBase,
              const CXXRecordDecl *VTableClass, CharUnits BaseOffset);
+  bool DoZeroInitPadding(const ASTRecordLayout &Layout, unsigned FieldNo,
+                         const FieldDecl &Field, bool AllowOverwrite,
+                         CharUnits &SizeSoFar, bool &ZeroFieldSize);
+  bool DoZeroInitPadding(const ASTRecordLayout &Layout, bool AllowOverwrite,
+                         CharUnits SizeSoFar);
   llvm::Constant *Finalize(QualType Ty);
 };
 
@@ -715,6 +727,10 @@ bool ConstStructBuilder::Build(const InitListExpr *ILE, bool AllowOverwrite) {
     if (CXXRD->getNumBases())
       return false;
 
+  const bool ZeroInitPadding = CGM.shouldZeroInitPadding();
+  bool ZeroFieldSize = false;
+  CharUnits SizeSoFar = CharUnits::Zero();
+
   for (FieldDecl *Field : RD->fields()) {
     ++FieldNo;
 
@@ -732,8 +748,13 @@ bool ConstStructBuilder::Build(const InitListExpr *ILE, bool AllowOverwrite) {
     const Expr *Init = nullptr;
     if (ElementNo < ILE->getNumInits())
       Init = ILE->getInit(ElementNo++);
-    if (isa_and_nonnull<NoInitExpr>(Init))
+    if (isa_and_nonnull<NoInitExpr>(Init)) {
+      if (ZeroInitPadding &&
+          !DoZeroInitPadding(Layout, FieldNo, *Field, AllowOverwrite, SizeSoFar,
+                             ZeroFieldSize))
+        return false;
       continue;
+    }
 
     // Zero-sized fields are not emitted, but their initializers may still
     // prevent emission of this struct as a constant.
@@ -743,6 +764,11 @@ bool ConstStructBuilder::Build(const InitListExpr *ILE, bool AllowOverwrite) {
       continue;
     }
 
+    if (ZeroInitPadding &&
+        !DoZeroInitPadding(Layout, FieldNo, *Field, AllowOverwrite, SizeSoFar,
+                           ZeroFieldSize))
+      return false;
+
     // When emitting a DesignatedInitUpdateExpr, a nested InitListExpr
     // represents additional overwriting of our current constant value, and not
     // a new constant to emit independently.
@@ -768,6 +794,10 @@ bool ConstStructBuilder::Build(const InitListExpr *ILE, bool AllowOverwrite) {
     if (!EltInit)
       return false;
 
+    if (ZeroInitPadding && ZeroFieldSize)
+      SizeSoFar += CharUnits::fromQuantity(
+          CGM.getDataLayout().getTypeAllocSize(EltInit->getType()));
+
     if (!Field->isBitField()) {
       // Handle non-bitfield members.
       if (!AppendField(Field, Layout.getFieldOffset(FieldNo), EltInit,
@@ -785,6 +815,9 @@ bool ConstStructBuilder::Build(const InitListExpr *ILE, bool AllowOverwrite) {
     }
   }
 
+  if (ZeroInitPadding && !DoZeroInitPadding(Layout, AllowOverwrite, SizeSoFar))
+    return false;
+
   return true;
 }
 
@@ -849,6 +882,9 @@ bool ConstStructBuilder::Build(const APValue &Val, const RecordDecl *RD,
 
   unsigned FieldNo = 0;
   uint64_t OffsetBits = CGM.getContext().toBits(Offset);
+  const bool ZeroInitPadding = CGM.shouldZeroInitPadding();
+  bool ZeroFieldSize = false;
+  CharUnits SizeSoFar = CharUnits::Zero();
 
   bool AllowOverwrite = false;
   for (RecordDecl::field_iterator Field = RD->field_begin(),
@@ -870,6 +906,15 @@ bool ConstStructBuilder::Build(const APValue &Val, const RecordDecl *RD,
     if (!EltInit)
       return false;
 
+    if (ZeroInitPadding) {
+      if (!DoZeroInitPadding(Layout, FieldNo, **Field, AllowOverwrite,
+                             SizeSoFar, ZeroFieldSize))
+        return false;
+      if (ZeroFieldSize)
+        SizeSoFar += CharUnits::fromQuantity(
+            CGM.getDataLayout().getTypeAllocSize(EltInit->getType()));
+    }
+
     if (!Field->isBitField()) {
       // Handle non-bitfield members.
       if (!AppendField(*Field, Layout.getFieldOffset(FieldNo) + OffsetBits,
@@ -886,7 +931,49 @@ bool ConstStructBuilder::Build(const APValue &Val, const RecordDecl *RD,
         return false;
     }
   }
+  if (ZeroInitPadding && !DoZeroInitPadding(Layout, AllowOverwrite, SizeSoFar))
+    return false;
+
+  return true;
+}
+
+bool ConstStructBuilder::DoZeroInitPadding(
+    const ASTRecordLayout &Layout, unsigned FieldNo, const FieldDecl &Field,
+    bool AllowOverwrite, CharUnits &SizeSoFar, bool &ZeroFieldSize) {
+  uint64_t StartBitOffset = Layout.getFieldOffset(FieldNo);
+  CharUnits StartOffset = CGM.getContext().toCharUnitsFromBits(StartBitOffset);
+  if (SizeSoFar < StartOffset)
+    if (!AppendBytes(SizeSoFar, getPadding(CGM, StartOffset - SizeSoFar),
+                     AllowOverwrite))
+      return false;
+
+  if (!Field.isBitField()) {
+    CharUnits FieldSize = CGM.getContext().getTypeSizeInChars(Field.getType());
+    SizeSoFar = StartOffset + FieldSize;
+    ZeroFieldSize = FieldSize.isZero();
+  } else {
+    const CGRecordLayout &RL =
+        CGM.getTypes().getCGRecordLayout(Field.getParent());
+    const CGBitFieldInfo &Info = RL.getBitFieldInfo(&Field);
+    uint64_t EndBitOffset = StartBitOffset + Info.Size;
+    SizeSoFar = CGM.getContext().toCharUnitsFromBits(EndBitOffset);
+    if (EndBitOffset % CGM.getContext().getCharWidth() != 0) {
+      SizeSoFar++;
+    }
+    ZeroFieldSize = Info.Size == 0;
+  }
+  return true;
+}
 
+bool ConstStructBuilder::DoZeroInitPadding(const ASTRecordLayout &Layout,
+                                           bool AllowOverwrite,
+                                           CharUnits SizeSoFar) {
+  CharUnits TotalSize = Layout.getSize();
+  if (SizeSoFar < TotalSize)
+    if (!AppendBytes(SizeSoFar, getPadding(CGM, TotalSize - SizeSoFar),
+                     AllowOverwrite))
+      return false;
+  SizeSoFar = TotalSize;
   return true;
 }
 
@@ -1127,12 +1214,10 @@ class ConstExprEmitter
 
       assert(CurSize <= TotalSize && "Union size mismatch!");
       if (unsigned NumPadBytes = TotalSize - CurSize) {
-        llvm::Type *Ty = CGM.CharTy;
-        if (NumPadBytes > 1)
-          Ty = llvm::ArrayType::get(Ty, NumPadBytes);
-
-        Elts.push_back(llvm::UndefValue::get(Ty));
-        Types.push_back(Ty);
+        llvm::Constant *Padding =
+            getPadding(CGM, CharUnits::fromQuantity(NumPadBytes));
+        Elts.push_back(Padding);
+        Types.push_back(Padding->getType());
       }
 
       llvm::StructType *STy = llvm::StructType::get(VMContext, Types, false);