Emit global variables as S_CONSTANT records for codeview debug info.

[akhuang]

Summary

This emits S_CONSTANT records for global variables.
Currently this emits records for the global variables already being tracked in the
LLVM IR metadata, which are just constant global variables; we'll also want S_CONSTANTs
for static data members and enums.

Related to https://bugs.llvm.org/show_bug.cgi?id=41615

[rnk]

llvm/lib/CodeGen/AsmPrinter/CodeViewDebug.h

203

I think we might want to try to generalize this CVGlobalVariable type to handle constants by leaving the GV member null. Then we won't need a second vector. I think there may be some cases where we should emit the S_CONSTANT inside a function, something like:

int f(int x) {
  static const int CppIsAwesome = 42; // Is this an S_CONSTANT or an S_LOCAL?
  enum { CppIsAwesome = 42 }; // If its written this way, is this an S_CONSTANT? Where does it come out?
  return x + CppIsAwesome;
}

I guess eventually CVGlobalVariable may become a bad fit for enumerator values, but I could see generalizing it.

To carry the DIExpression which contains the constant value in CVGlobalVariable, you can use a PointerIntUnion to store either a DIExpression* for constants or GlobalVariable* for true globals. Then the global emission code can check which is active and emit S_CONSTANT or S_[GL]DATA32

[akhuang]

• Save constant global variables in the same vector as other global variables

[rnk]

I was going to ask for a test case that makes local constants, but I see clang generates pretty weird metadata for this test case:

int f(int x) {
  static const int MyConst = 42;
  return x + MyConst;
}

That's supposed to generate a local S_CONSTANT, but we don't populate the DIExpression properly. Hm.

llvm/lib/CodeGen/AsmPrinter/CodeViewDebug.cpp

2997–2998

I think it's impossible for S_CONSTANTs to be comdat because we find the comdat on the global variable, so I would suggest using .get instead of dyn_cast, so that we crash earlier if that invariant is violated.

3030–3031

I think it would be simpler to have this take a const CVGlobalVariable & rather than taking the two members individually.

3052–3053

Maybe do this, so that this code complains loudly if another PointerUnion case is added instead of silently emitting nothing:

} else {
  const DIExpression *DIE = GVInfo.get<const DIExpression *>();

3056

I think it would be better to check isConstant() earlier, before we set up the pointer union, so that no non-constant expressions make it here.

[akhuang]

• Small code changes

[akhuang]

Yeah, I noticed that the DIExpression is empty for local constants. I guess that's another thing to look at later?

[rnk]

lgtm with some nits

llvm/lib/CodeGen/AsmPrinter/CodeViewDebug.cpp

3056

nit: I guess we should assert isConstant() here. I wish there was some helper for pulling out the constant value. :(

3063

I think 10 bytes is all that's needed to write an encoded integer, so this might as well be 10 bytes long with a comment about it.

[akhuang]

nits

[akhuang]

whitespace

llvm/lib/CodeGen/AsmPrinter/CodeViewDebug.h

 
 #ifndef LLVM_LIB_CODEGEN_ASMPRINTER_CODEVIEWDEBUG_H
 #define LLVM_LIB_CODEGEN_ASMPRINTER_CODEVIEWDEBUG_H
 
 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/DenseSet.h"
 #include "llvm/ADT/MapVector.h"
+#include "llvm/ADT/PointerUnion.h"
 #include "llvm/ADT/SetVector.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/CodeGen/DbgEntityHistoryCalculator.h"
 #include "llvm/CodeGen/DebugHandlerBase.h"
 #include "llvm/DebugInfo/CodeView/CodeView.h"
 #include "llvm/DebugInfo/CodeView/GlobalTypeTableBuilder.h"
 #include "llvm/DebugInfo/CodeView/TypeIndex.h"
 #include "llvm/IR/DebugLoc.h"
class LLVM_LIBRARY_VISIBILITY CodeViewDebug : public DebugHandlerBase {
   struct LocalVariable {
     const DILocalVariable *DIVar = nullptr;
     SmallVector<LocalVarDefRange, 1> DefRanges;
     bool UseReferenceType = false;
   };
 
   struct CVGlobalVariable {
     const DIGlobalVariable *DIGV;
-    const GlobalVariable *GV;
+    PointerUnion<const GlobalVariable *, const DIExpression *> GVInfo;
   };
 
   struct InlineSite {
     SmallVector<LocalVariable, 1> InlinedLocals;
     SmallVector<const DILocation *, 1> ChildSites;
     const DISubprogram *Inlinee = nullptr;
 
     /// The ID of the inline site or function used with .cv_loc. Not a type
class LLVM_LIBRARY_VISIBILITY CodeViewDebug : public DebugHandlerBase {
   // belong in. A null local scope represents the global scope.
   typedef SmallVector<CVGlobalVariable, 1> GlobalVariableList;
   DenseMap<const DIScope*, std::unique_ptr<GlobalVariableList> > ScopeGlobals;
 
   // Array of global variables which  need to be emitted into a COMDAT section.
   SmallVector<CVGlobalVariable, 1> ComdatVariables;
 
   // Array of non-COMDAT global variables.
   SmallVector<CVGlobalVariable, 1> GlobalVariables;
 
   /// The set of comdat .debug$S sections that we've seen so far. Each section
   /// must start with a magic version number that must only be emitted once.
   /// This set tracks which sections we've already opened.
   DenseSet<MCSectionCOFF *> ComdatDebugSections;
 
   /// Switch to the appropriate .debug$S section for GVSym. If GVSym, the symbol
   /// of an emitted global value, is in a comdat COFF section, this will switch
class LLVM_LIBRARY_VISIBILITY CodeViewDebug : public DebugHandlerBase {
 
   void emitDebugInfoForRetainedTypes();
 
   void
   emitDebugInfoForUDTs(ArrayRef<std::pair<std::string, const DIType *>> UDTs);
 
   void emitDebugInfoForGlobals();
   void emitGlobalVariableList(ArrayRef<CVGlobalVariable> Globals);
-  void emitDebugInfoForGlobal(const DIGlobalVariable *DIGV,
-                              const GlobalVariable *GV, MCSymbol *GVSym);
+  void emitDebugInfoForGlobal(const CVGlobalVariable &CVGV);
 
   /// Opens a subsection of the given kind in a .debug$S codeview section.
   /// Returns an end label for use with endCVSubsection when the subsection is
   /// finished.
   MCSymbol *beginCVSubsection(codeview::DebugSubsectionKind Kind);
   void endCVSubsection(MCSymbol *EndLabel);
 
   /// Opens a symbol record of the given kind. Returns an end label for use with

llvm/lib/CodeGen/AsmPrinter/CodeViewDebug.cpp

 #include "DwarfExpression.h"
 #include "llvm/ADT/APSInt.h"
 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/DenseSet.h"
 #include "llvm/ADT/MapVector.h"
 #include "llvm/ADT/None.h"
 #include "llvm/ADT/Optional.h"
+#include "llvm/ADT/PointerUnion.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/SmallString.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/StringRef.h"
 #include "llvm/ADT/TinyPtrVector.h"
 #include "llvm/ADT/Triple.h"
 #include "llvm/ADT/Twine.h"
 #include "llvm/BinaryFormat/COFF.h"
 #include "llvm/BinaryFormat/Dwarf.h"
 #include "llvm/CodeGen/AsmPrinter.h"
 #include "llvm/CodeGen/LexicalScopes.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstr.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineOperand.h"
 #include "llvm/CodeGen/TargetFrameLowering.h"
 #include "llvm/CodeGen/TargetRegisterInfo.h"
 #include "llvm/CodeGen/TargetSubtargetInfo.h"
 #include "llvm/Config/llvm-config.h"
 #include "llvm/DebugInfo/CodeView/CVTypeVisitor.h"
 #include "llvm/DebugInfo/CodeView/CodeView.h"
+#include "llvm/DebugInfo/CodeView/CodeViewRecordIO.h"
 #include "llvm/DebugInfo/CodeView/ContinuationRecordBuilder.h"
 #include "llvm/DebugInfo/CodeView/DebugInlineeLinesSubsection.h"
 #include "llvm/DebugInfo/CodeView/EnumTables.h"
 #include "llvm/DebugInfo/CodeView/Line.h"
 #include "llvm/DebugInfo/CodeView/SymbolRecord.h"
 #include "llvm/DebugInfo/CodeView/TypeDumpVisitor.h"
 #include "llvm/DebugInfo/CodeView/TypeIndex.h"
 #include "llvm/DebugInfo/CodeView/TypeRecord.h"

 #include "llvm/IR/Module.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCSectionCOFF.h"
 #include "llvm/MC/MCStreamer.h"
 #include "llvm/MC/MCSymbol.h"
 #include "llvm/Support/BinaryByteStream.h"
 #include "llvm/Support/BinaryStreamReader.h"
+#include "llvm/Support/BinaryStreamWriter.h"
 #include "llvm/Support/Casting.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Compiler.h"
 #include "llvm/Support/Endian.h"
 #include "llvm/Support/Error.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/FormatVariadic.h"
 #include "llvm/Support/Path.h"
for (const GlobalVariable &GV : MMI->getModule()->globals()) {
     for (const auto *GVE : GVEs)
       GlobalMap[GVE] = &GV;
   }
 
   NamedMDNode *CUs = MMI->getModule()->getNamedMetadata("llvm.dbg.cu");
   for (const MDNode *Node : CUs->operands()) {
     const auto *CU = cast<DICompileUnit>(Node);
     for (const auto *GVE : CU->getGlobalVariables()) {
+      const DIGlobalVariable *DIGV = GVE->getVariable();
+      const DIExpression *DIE = GVE->getExpression();
+
+      // Emit constant global variables in a global symbol section.
+      if (GlobalMap.count(GVE) == 0 && DIE->isConstant()) {
+        CVGlobalVariable CVGV = {DIGV, DIE};
+        GlobalVariables.emplace_back(std::move(CVGV));
+      }
+
       const auto *GV = GlobalMap.lookup(GVE);
       if (!GV || GV->isDeclarationForLinker())
         continue;
-      const DIGlobalVariable *DIGV = GVE->getVariable();
+
       DIScope *Scope = DIGV->getScope();
       SmallVector<CVGlobalVariable, 1> *VariableList;
       if (Scope && isa<DILocalScope>(Scope)) {
         // Locate a global variable list for this scope, creating one if
         // necessary.
         auto Insertion = ScopeGlobals.insert(
             {Scope, std::unique_ptr<GlobalVariableList>()});
         if (Insertion.second)
           Insertion.first->second = llvm::make_unique<GlobalVariableList>();
         VariableList = Insertion.first->second.get();
       } else if (GV->hasComdat())
         // Emit this global variable into a COMDAT section.
         VariableList = &ComdatVariables;
       else
-        // Emit this globla variable in a single global symbol section.
+        // Emit this global variable in a single global symbol section.
         VariableList = &GlobalVariables;
       CVGlobalVariable CVGV = {DIGV, GV};
       VariableList->emplace_back(std::move(CVGV));
     }
   }
 }
 
 void CodeViewDebug::emitDebugInfoForGlobals() {
   // First, emit all globals that are not in a comdat in a single symbol
   // substream. MSVC doesn't like it if the substream is empty, so only open
   // it if we have at least one global to emit.
   switchToDebugSectionForSymbol(nullptr);
+
   if (!GlobalVariables.empty()) {
     OS.AddComment("Symbol subsection for globals");
     MCSymbol *EndLabel = beginCVSubsection(DebugSubsectionKind::Symbols);
     emitGlobalVariableList(GlobalVariables);
     endCVSubsection(EndLabel);
   }
 
   // Second, emit each global that is in a comdat into its own .debug$S
   // section along with its own symbol substream.
   for (const CVGlobalVariable &CVGV : ComdatVariables) {
-    MCSymbol *GVSym = Asm->getSymbol(CVGV.GV);
+    const GlobalVariable *GV = CVGV.GVInfo.get<const GlobalVariable *>();
+    MCSymbol *GVSym = Asm->getSymbol(GV);
     OS.AddComment("Symbol subsection for " +
-            Twine(GlobalValue::dropLLVMManglingEscape(CVGV.GV->getName())));
+                  Twine(GlobalValue::dropLLVMManglingEscape(GV->getName())));
     switchToDebugSectionForSymbol(GVSym);
     MCSymbol *EndLabel = beginCVSubsection(DebugSubsectionKind::Symbols);
     // FIXME: emitDebugInfoForGlobal() doesn't handle DIExpressions.
-    emitDebugInfoForGlobal(CVGV.DIGV, CVGV.GV, GVSym);
+    emitDebugInfoForGlobal(CVGV);
     endCVSubsection(EndLabel);
   }
 }
 
 void CodeViewDebug::emitDebugInfoForRetainedTypes() {
   NamedMDNode *CUs = MMI->getModule()->getNamedMetadata("llvm.dbg.cu");
   for (const MDNode *Node : CUs->operands()) {
     for (auto *Ty : cast<DICompileUnit>(Node)->getRetainedTypes()) {
       if (DIType *RT = dyn_cast<DIType>(Ty)) {
         getTypeIndex(RT);
         // FIXME: Add to global/local DTU list.
       }
     }
   }
 }
 
 // Emit each global variable in the specified array.
 void CodeViewDebug::emitGlobalVariableList(ArrayRef<CVGlobalVariable> Globals) {
   for (const CVGlobalVariable &CVGV : Globals) {
-    MCSymbol *GVSym = Asm->getSymbol(CVGV.GV);
     // FIXME: emitDebugInfoForGlobal() doesn't handle DIExpressions.
-    emitDebugInfoForGlobal(CVGV.DIGV, CVGV.GV, GVSym);
+    emitDebugInfoForGlobal(CVGV);
   }
 }
 
-void CodeViewDebug::emitDebugInfoForGlobal(const DIGlobalVariable *DIGV,
-                                           const GlobalVariable *GV,
-                                           MCSymbol *GVSym) {
+void CodeViewDebug::emitDebugInfoForGlobal(const CVGlobalVariable &CVGV) {
+  const DIGlobalVariable *DIGV = CVGV.DIGV;
+  if (const GlobalVariable *GV =
+          CVGV.GVInfo.dyn_cast<const GlobalVariable *>()) {
     // DataSym record, see SymbolRecord.h for more info. Thread local data
     // happens to have the same format as global data.
+    MCSymbol *GVSym = Asm->getSymbol(GV);
     SymbolKind DataSym = GV->isThreadLocal()
                              ? (DIGV->isLocalToUnit() ? SymbolKind::S_LTHREAD32
                                                       : SymbolKind::S_GTHREAD32)
                              : (DIGV->isLocalToUnit() ? SymbolKind::S_LDATA32
                                                       : SymbolKind::S_GDATA32);
     MCSymbol *DataEnd = beginSymbolRecord(DataSym);
     OS.AddComment("Type");
     OS.EmitIntValue(getCompleteTypeIndex(DIGV->getType()).getIndex(), 4);
     OS.AddComment("DataOffset");
     OS.EmitCOFFSecRel32(GVSym, /*Offset=*/0);
     OS.AddComment("Segment");
     OS.EmitCOFFSectionIndex(GVSym);
     OS.AddComment("Name");
     const unsigned LengthOfDataRecord = 12;
     emitNullTerminatedSymbolName(OS, DIGV->getName(), LengthOfDataRecord);
     endSymbolRecord(DataEnd);
+  } else {
+    // FIXME: Currently this only emits the global variables in the IR metadata.
+    // This should also emit enums and static data members.
+    const DIExpression *DIE = CVGV.GVInfo.get<const DIExpression *>();
+    uint64_t Val = DIE->getElement(1);
+
+    MCSymbol *SConstantEnd = beginSymbolRecord(SymbolKind::S_CONSTANT);
+    OS.AddComment("Type");
+    OS.EmitIntValue(getTypeIndex(DIGV->getType()).getIndex(), 4);
+    OS.AddComment("Value");
+
+    uint8_t data[64];
+    BinaryStreamWriter Writer(data, llvm::support::endianness::little);
+    CodeViewRecordIO IO(Writer);
+    cantFail(IO.mapEncodedInteger(Val));
+    StringRef SRef((char *)data, Writer.getOffset());
+    OS.EmitBinaryData(SRef);
+
+    OS.AddComment("Name");
+    emitNullTerminatedSymbolName(OS, DIGV->getDisplayName());
+    endSymbolRecord(SConstantEnd);
+  }
 }

llvm/test/DebugInfo/COFF/global-constants.ll

+; RUN: llc < %s | FileCheck %s --check-prefix=ASM
+; RUN: llc < %s -filetype=obj | llvm-readobj - --codeview | FileCheck %s --check-prefix=OBJ
+
+; C++ source to regenerate:
+; const int Test1 = 1;
+; int main() {
+;   return Test1;
+; }
+; $ clang t.cpp -S -emit-llvm -g -gcodeview -o t.ll
+
+; ASM-LABEL:  .long 241              # Symbol subsection for globals
+
+; ASM:        .short {{.*-.*}}       # Record length
+; ASM:        .short 4359            # Record kind: S_CONSTANT
+; ASM-NEXT:   .long 4099             # Type
+; ASM-NEXT:   .byte 0x01, 0x00       # Value
+; ASM-NEXT:   .asciz "Test1"         # Name
+
+; OBJ:        CodeViewDebugInfo [
+; OBJ:          Section: .debug$S
+; OBJ:          Magic: 0x4
+; OBJ:          Subsection [
+; OBJ:            SubSectionType: Symbols (0xF1)
+; OBJ:            ConstantSym {
+; OBJ-NEXT:         Kind: S_CONSTANT (0x1107)
+; OBJ-NEXT:         Type: const int (0x1003)
+; OBJ-NEXT:         Value: 1
+; OBJ-NEXT:         Name: Test1
+; OBJ-NEXT:       }
+
+; ModuleID = 't.cpp'
+source_filename = "t.cpp"
+target datalayout = "e-m:w-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-pc-windows-msvc"
+
+; Function Attrs: noinline norecurse nounwind optnone
+define dso_local i32 @main() #0 !dbg !13 {
+entry:
+  %retval = alloca i32, align 4
+  store i32 0, i32* %retval, align 4
+  ret i32 1, !dbg !16
+}
+
+!llvm.dbg.cu = !{!0}
+!llvm.module.flags = !{!9, !10, !11}
+!llvm.ident = !{!12}
+
+!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus, file: !1, producer: "clang version 9.0.0 (https://github.com/llvm/llvm-project.git 4a1902b6739e3087a03c0ac7ab85b640764e9335)", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, enums: !2, globals: !3, nameTableKind: None)
+!1 = !DIFile(filename: "<stdin>", directory: "C:\5Csrc\5Ctest", checksumkind: CSK_MD5, checksum: "0d5ef00bdd80bdb409a3deac9938f20d")
+!2 = !{}
+!3 = !{!4}
+!4 = !DIGlobalVariableExpression(var: !5, expr: !DIExpression(DW_OP_constu, 1, DW_OP_stack_value))
+!5 = distinct !DIGlobalVariable(name: "Test1", scope: !0, file: !6, line: 1, type: !7, isLocal: true, isDefinition: true)
+!6 = !DIFile(filename: "t.cpp", directory: "C:\5Csrc\5Ctest", checksumkind: CSK_MD5, checksum: "0d5ef00bdd80bdb409a3deac9938f20d")
+!7 = !DIDerivedType(tag: DW_TAG_const_type, baseType: !8)
+!8 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
+!9 = !{i32 2, !"CodeView", i32 1}
+!10 = !{i32 2, !"Debug Info Version", i32 3}
+!11 = !{i32 1, !"wchar_size", i32 2}
+!12 = !{!"clang version 9.0.0 (https://github.com/llvm/llvm-project.git 4a1902b6739e3087a03c0ac7ab85b640764e9335)"}
+!13 = distinct !DISubprogram(name: "main", scope: !6, file: !6, line: 3, type: !14, scopeLine: 3, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !2)
+!14 = !DISubroutineType(types: !15)
+!15 = !{!8}
+!16 = !DILocation(line: 4, scope: !13)