[DebugInfo][DwarfDebug] Separate creation and population of abstract subprogram DIEs (#159104)

With this change, construction of abstract subprogram DIEs is split in
two stages/functions:
creation of DIE (in DwarfCompileUnit::getOrCreateAbstractSubprogramDIE)
and its population with children (in
DwarfCompileUnit::constructAbstractSubprogramScopeDIE).
With that, abstract subprograms can be created/referenced from
DwarfDebug::beginModule, which should solve the issue with static local
variables DIE creation of inlined functons with optimized-out
definitions. It fixes #29985.

LexicalScopes class now stores mapping from DISubprograms to their
corresponding llvm::Function's. It is supposed to be built before
processing of each function (so, now LexicalScopes class has a method
for "module initialization" alongside the method for "function
initialization"). It is used by DwarfCompileUnit to determine whether a
DISubprogram needs an abstract DIE before DwarfDebug::beginFunction is
invoked.

DwarfCompileUnit::getOrCreateSubprogramDIE method is added, which can
create an abstract or a concrete DIE for a subprogram. It accepts
llvm::Function* argument to determine whether a concrete DIE must be
created.

This is a temporary fix for
#29985. Ideally, it will be
fixed by moving global variables and types emission to
DwarfDebug::endModule (https://reviews.llvm.org/D144007,
https://reviews.llvm.org/D144005).

Some code proposed by Ellis Hoag <[email protected]> in
#90523 was taken for this
commit.

Author: dzhidzhoev

llvm/include/llvm/CodeGen/DebugHandlerBase.h

@@ -144,6 +144,8 @@ class DebugHandlerBase : public AsmPrinterHandler {
   static bool isUnsignedDIType(const DIType *Ty);
 
   const InstructionOrdering &getInstOrdering() const { return InstOrdering; }
+
+  const LexicalScopes &getLexicalScopes() const { return LScopes; }
 };
 
 } // namespace llvm

llvm/include/llvm/CodeGen/LexicalScopes.h

@@ -141,12 +141,18 @@ class LexicalScopes {
 public:
   LexicalScopes() = default;
 
+  /// Scan module to build subprogram-to-function map.
+  LLVM_ABI void initialize(const Module &);
+
   /// Scan machine function and constuct lexical scope nest, resets
   /// the instance if necessary.
-  LLVM_ABI void initialize(const MachineFunction &);
+  LLVM_ABI void scanFunction(const MachineFunction &);
+
+  /// Reset the instance so that it's prepared for another module.
+  LLVM_ABI void resetModule();
 
-  /// Release memory.
-  LLVM_ABI void reset();
+  /// Reset the instance so that it's prepared for another function.
+  LLVM_ABI void resetFunction();
 
   /// Return true if there is any lexical scope information available.
   bool empty() { return CurrentFnLexicalScope == nullptr; }
@@ -196,6 +202,11 @@ class LexicalScopes {
   /// Find or create an abstract lexical scope.
   LLVM_ABI LexicalScope *getOrCreateAbstractScope(const DILocalScope *Scope);
 
+  /// Get function to which the given subprogram is attached, if exists.
+  const Function *getFunction(const DISubprogram *SP) const {
+    return FunctionMap.lookup(SP);
+  }
+
 private:
   /// Find lexical scope for the given Scope/IA. If not available
   /// then create new lexical scope.
@@ -225,6 +236,9 @@ class LexicalScopes {
 
   const MachineFunction *MF = nullptr;
 
+  /// Mapping between DISubprograms and IR functions.
+  DenseMap<const DISubprogram *, const Function *> FunctionMap;
+
   /// Tracks the scopes in the current function.
   // Use an unordered_map to ensure value pointer validity over insertion.
   std::unordered_map<const DILocalScope *, LexicalScope> LexicalScopeMap;

llvm/lib/CodeGen/AsmPrinter/DebugHandlerBase.cpp

@@ -105,6 +105,8 @@ DebugHandlerBase::~DebugHandlerBase() = default;
 void DebugHandlerBase::beginModule(Module *M) {
   if (M->debug_compile_units().empty())
     Asm = nullptr;
+  else
+    LScopes.initialize(*M);
 }
 
 // Each LexicalScope has first instruction and last instruction to mark
@@ -269,7 +271,7 @@ void DebugHandlerBase::beginFunction(const MachineFunction *MF) {
 
   // Grab the lexical scopes for the function, if we don't have any of those
   // then we're not going to be able to do anything.
-  LScopes.initialize(*MF);
+  LScopes.scanFunction(*MF);
   if (LScopes.empty()) {
     beginFunctionImpl(MF);
     return;

llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp

@@ -537,8 +537,9 @@ void DwarfCompileUnit::addWasmRelocBaseGlobal(DIELoc *Loc, StringRef GlobalName,
 // and DW_AT_high_pc attributes. If there are global variables in this
 // scope then create and insert DIEs for these variables.
 DIE &DwarfCompileUnit::updateSubprogramScopeDIE(const DISubprogram *SP,
+                                                const Function &F,
                                                 MCSymbol *LineTableSym) {
-  DIE *SPDie = getOrCreateSubprogramDIE(SP, includeMinimalInlineScopes());
+  DIE *SPDie = getOrCreateSubprogramDIE(SP, &F, includeMinimalInlineScopes());
   SmallVector<RangeSpan, 2> BB_List;
   // If basic block sections are on, ranges for each basic block section has
   // to be emitted separately.
@@ -1122,9 +1123,10 @@ sortLocalVars(SmallVectorImpl<DbgVariable *> &Input) {
 }
 
 DIE &DwarfCompileUnit::constructSubprogramScopeDIE(const DISubprogram *Sub,
+                                                   const Function &F,
                                                    LexicalScope *Scope,
                                                    MCSymbol *LineTableSym) {
-  DIE &ScopeDIE = updateSubprogramScopeDIE(Sub, LineTableSym);
+  DIE &ScopeDIE = updateSubprogramScopeDIE(Sub, F, LineTableSym);
 
   if (Scope) {
     assert(!Scope->getInlinedAt());
@@ -1198,32 +1200,17 @@ DIE *DwarfCompileUnit::createAndAddScopeChildren(LexicalScope *Scope,
   return ObjectPointer;
 }
 
-void DwarfCompileUnit::constructAbstractSubprogramScopeDIE(
-    LexicalScope *Scope) {
-  auto *SP = cast<DISubprogram>(Scope->getScopeNode());
-  if (getAbstractScopeDIEs().count(SP))
-    return;
+DIE &DwarfCompileUnit::getOrCreateAbstractSubprogramDIE(
+    const DISubprogram *SP) {
+  if (auto *AbsDef = getAbstractScopeDIEs().lookup(SP))
+    return *AbsDef;
 
-  DIE *ContextDIE;
-  DwarfCompileUnit *ContextCU = this;
-
-  if (includeMinimalInlineScopes())
-    ContextDIE = &getUnitDie();
-  // Some of this is duplicated from DwarfUnit::getOrCreateSubprogramDIE, with
-  // the important distinction that the debug node is not associated with the
-  // DIE (since the debug node will be associated with the concrete DIE, if
-  // any). It could be refactored to some common utility function.
-  else if (auto *SPDecl = SP->getDeclaration()) {
-    ContextDIE = &getUnitDie();
-    getOrCreateSubprogramDIE(SPDecl);
-  } else {
-    ContextDIE = getOrCreateContextDIE(SP->getScope());
-    // The scope may be shared with a subprogram that has already been
-    // constructed in another CU, in which case we need to construct this
-    // subprogram in the same CU.
-    ContextCU = DD->lookupCU(ContextDIE->getUnitDie());
-  }
+  auto [ContextDIE, ContextCU] = getOrCreateAbstractSubprogramContextDIE(SP);
+  return createAbstractSubprogramDIE(SP, ContextDIE, ContextCU);
+}
 
+DIE &DwarfCompileUnit::createAbstractSubprogramDIE(
+    const DISubprogram *SP, DIE *ContextDIE, DwarfCompileUnit *ContextCU) {
   // Passing null as the associated node because the abstract definition
   // shouldn't be found by lookup.
   DIE &AbsDef = ContextCU->createAndAddDIE(dwarf::DW_TAG_subprogram,
@@ -1237,8 +1224,45 @@ void DwarfCompileUnit::constructAbstractSubprogramScopeDIE(
                      DD->getDwarfVersion() <= 4 ? std::optional<dwarf::Form>()
                                                 : dwarf::DW_FORM_implicit_const,
                      dwarf::DW_INL_inlined);
-  if (DIE *ObjectPointer = ContextCU->createAndAddScopeChildren(Scope, AbsDef))
-    ContextCU->addDIEEntry(AbsDef, dwarf::DW_AT_object_pointer, *ObjectPointer);
+
+  return AbsDef;
+}
+
+std::pair<DIE *, DwarfCompileUnit *>
+DwarfCompileUnit::getOrCreateAbstractSubprogramContextDIE(
+    const DISubprogram *SP) {
+  bool Minimal = includeMinimalInlineScopes();
+  bool IgnoreScope = shouldPlaceInUnitDIE(SP, Minimal);
+  DIE *ContextDIE = getOrCreateSubprogramContextDIE(SP, IgnoreScope);
+
+  if (auto *SPDecl = SP->getDeclaration())
+    if (!Minimal)
+      getOrCreateSubprogramDIE(SPDecl, nullptr);
+
+  // The scope may be shared with a subprogram that has already been
+  // constructed in another CU, in which case we need to construct this
+  // subprogram in the same CU.
+  auto *ContextCU = IgnoreScope ? this : DD->lookupCU(ContextDIE->getUnitDie());
+
+  return std::make_pair(ContextDIE, ContextCU);
+}
+
+void DwarfCompileUnit::constructAbstractSubprogramScopeDIE(
+    LexicalScope *Scope) {
+  auto *SP = cast<DISubprogram>(Scope->getScopeNode());
+
+  // Populate subprogram DIE only once.
+  if (!getFinalizedAbstractSubprograms().insert(SP).second)
+    return;
+
+  auto [ContextDIE, ContextCU] = getOrCreateAbstractSubprogramContextDIE(SP);
+  DIE *AbsDef = getAbstractScopeDIEs().lookup(SP);
+  if (!AbsDef)
+    AbsDef = &createAbstractSubprogramDIE(SP, ContextDIE, ContextCU);
+
+  if (DIE *ObjectPointer = ContextCU->createAndAddScopeChildren(Scope, *AbsDef))
+    ContextCU->addDIEEntry(*AbsDef, dwarf::DW_AT_object_pointer,
+                           *ObjectPointer);
 }
 
 bool DwarfCompileUnit::useGNUAnalogForDwarf5Feature() const {
@@ -1293,9 +1317,9 @@ DwarfCompileUnit::getDwarf5OrGNULocationAtom(dwarf::LocationAtom Loc) const {
 }
 
 DIE &DwarfCompileUnit::constructCallSiteEntryDIE(
-    DIE &ScopeDIE, const DISubprogram *CalleeSP, bool IsTail,
-    const MCSymbol *PCAddr, const MCSymbol *CallAddr, unsigned CallReg,
-    DIType *AllocSiteTy) {
+    DIE &ScopeDIE, const DISubprogram *CalleeSP, const Function *CalleeF,
+    bool IsTail, const MCSymbol *PCAddr, const MCSymbol *CallAddr,
+    unsigned CallReg, DIType *AllocSiteTy) {
   // Insert a call site entry DIE within ScopeDIE.
   DIE &CallSiteDIE = createAndAddDIE(getDwarf5OrGNUTag(dwarf::DW_TAG_call_site),
                                      ScopeDIE, nullptr);
@@ -1305,7 +1329,7 @@ DIE &DwarfCompileUnit::constructCallSiteEntryDIE(
     addAddress(CallSiteDIE, getDwarf5OrGNUAttr(dwarf::DW_AT_call_target),
                MachineLocation(CallReg));
   } else if (CalleeSP) {
-    DIE *CalleeDIE = getOrCreateSubprogramDIE(CalleeSP);
+    DIE *CalleeDIE = getOrCreateSubprogramDIE(CalleeSP, CalleeF);
     assert(CalleeDIE && "Could not create DIE for call site entry origin");
     if (AddLinkageNamesToDeclCallOriginsForTuning(DD) &&
         !CalleeSP->isDefinition() &&
@@ -1396,7 +1420,7 @@ DIE *DwarfCompileUnit::constructImportedEntityDIE(
     if (auto *AbsSPDie = getAbstractScopeDIEs().lookup(SP))
       EntityDie = AbsSPDie;
     else
-      EntityDie = getOrCreateSubprogramDIE(SP);
+      EntityDie = getOrCreateSubprogramDIE(SP, nullptr);
   } else if (auto *T = dyn_cast<DIType>(Entity))
     EntityDie = getOrCreateTypeDIE(T);
   else if (auto *GV = dyn_cast<DIGlobalVariable>(Entity))
@@ -1805,3 +1829,16 @@ DIE *DwarfCompileUnit::getOrCreateContextDIE(const DIScope *Context) {
   }
   return DwarfUnit::getOrCreateContextDIE(Context);
 }
+
+DIE *DwarfCompileUnit::getOrCreateSubprogramDIE(const DISubprogram *SP,
+                                                const Function *F,
+                                                bool Minimal) {
+  if (!F && SP->isDefinition()) {
+    F = DD->getLexicalScopes().getFunction(SP);
+
+    if (!F)
+      return &getCU().getOrCreateAbstractSubprogramDIE(SP);
+  }
+
+  return DwarfUnit::getOrCreateSubprogramDIE(SP, F, Minimal);
+}

llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.h

@@ -81,6 +81,7 @@ class DwarfCompileUnit final : public DwarfUnit {
 
   // List of abstract local scopes (either DISubprogram or DILexicalBlock).
   DenseMap<const DILocalScope *, DIE *> AbstractLocalScopeDIEs;
+  SmallPtrSet<const DISubprogram *, 8> FinalizedAbstractSubprograms;
 
   // List of inlined lexical block scopes that belong to subprograms within this
   // CU.
@@ -137,12 +138,28 @@ class DwarfCompileUnit final : public DwarfUnit {
     return DU->getAbstractEntities();
   }
 
+  auto &getFinalizedAbstractSubprograms() {
+    if (isDwoUnit() && !DD->shareAcrossDWOCUs())
+      return FinalizedAbstractSubprograms;
+    return DU->getFinalizedAbstractSubprograms();
+  }
+
   void finishNonUnitTypeDIE(DIE& D, const DICompositeType *CTy) override;
 
   /// Add info for Wasm-global-based relocation.
   void addWasmRelocBaseGlobal(DIELoc *Loc, StringRef GlobalName,
                               uint64_t GlobalIndex);
 
+  /// Create context DIE for abstract subprogram.
+  /// \returns The context DIE and the compile unit where abstract
+  ///          DIE should be constructed.
+  std::pair<DIE *, DwarfCompileUnit *>
+  getOrCreateAbstractSubprogramContextDIE(const DISubprogram *SP);
+
+  /// Create new DIE for abstract subprogram.
+  DIE &createAbstractSubprogramDIE(const DISubprogram *SP, DIE *ContextDIE,
+                                   DwarfCompileUnit *ContextCU);
+
 public:
   DwarfCompileUnit(unsigned UID, const DICompileUnit *Node, AsmPrinter *A,
                    DwarfDebug *DW, DwarfFile *DWU,
@@ -216,7 +233,8 @@ class DwarfCompileUnit final : public DwarfUnit {
   /// DW_AT_low_pc, DW_AT_high_pc and DW_AT_LLVM_stmt_sequence attributes.
   /// If there are global variables in this scope then create and insert DIEs
   /// for these variables.
-  DIE &updateSubprogramScopeDIE(const DISubprogram *SP, MCSymbol *LineTableSym);
+  DIE &updateSubprogramScopeDIE(const DISubprogram *SP, const Function &F,
+                                MCSymbol *LineTableSym);
 
   void constructScopeDIE(LexicalScope *Scope, DIE &ParentScopeDIE);
 
@@ -259,12 +277,18 @@ class DwarfCompileUnit final : public DwarfUnit {
   /// This instance of 'getOrCreateContextDIE()' can handle DILocalScope.
   DIE *getOrCreateContextDIE(const DIScope *Ty) override;
 
+  DIE *getOrCreateSubprogramDIE(const DISubprogram *SP, const Function *F,
+                                bool Minimal = false) override;
+
   /// Construct a DIE for this subprogram scope.
-  DIE &constructSubprogramScopeDIE(const DISubprogram *Sub, LexicalScope *Scope,
-                                   MCSymbol *LineTableSym);
+  DIE &constructSubprogramScopeDIE(const DISubprogram *Sub, const Function &F,
+                                   LexicalScope *Scope, MCSymbol *LineTableSym);
 
   DIE *createAndAddScopeChildren(LexicalScope *Scope, DIE &ScopeDIE);
 
+  /// Create an abstract subprogram DIE, that should later be populated
+  /// by \ref constructAbstractSubprogramScopeDIE.
+  DIE &getOrCreateAbstractSubprogramDIE(const DISubprogram *SP);
   void constructAbstractSubprogramScopeDIE(LexicalScope *Scope);
 
   /// Whether to use the GNU analog for a DWARF5 tag, attribute, or location
@@ -281,14 +305,15 @@ class DwarfCompileUnit final : public DwarfUnit {
   dwarf::LocationAtom getDwarf5OrGNULocationAtom(dwarf::LocationAtom Loc) const;
 
   /// Construct a call site entry DIE describing a call within \p Scope to a
-  /// callee described by \p CalleeSP.
+  /// callee described by \p CalleeSP and \p CalleeF.
   /// \p IsTail specifies whether the call is a tail call.
   /// \p PCAddr points to the PC value after the call instruction.
   /// \p CallAddr points to the PC value at the call instruction (or is null).
   /// \p CallReg is a register location for an indirect call. For direct calls
   /// the \p CallReg is set to 0.
   DIE &constructCallSiteEntryDIE(DIE &ScopeDIE, const DISubprogram *CalleeSP,
-                                 bool IsTail, const MCSymbol *PCAddr,
+                                 const Function *CalleeF, bool IsTail,
+                                 const MCSymbol *PCAddr,
                                  const MCSymbol *CallAddr, unsigned CallReg,
                                  DIType *AllocSiteTy);
   /// Construct call site parameter DIEs for the \p CallSiteDIE. The \p Params

llvm/lib/CodeGen/AsmPrinter/DwarfDebug.cpp

@@ -1001,8 +1001,9 @@ void DwarfDebug::constructCallSiteEntryDIEs(const DISubprogram &SP,
                                                        ->getName(CallReg)))
                         << (IsTail ? " [IsTail]" : "") << "\n");
 
-      DIE &CallSiteDIE = CU.constructCallSiteEntryDIE(
-          ScopeDIE, CalleeSP, IsTail, PCAddr, CallAddr, CallReg, AllocSiteTy);
+      DIE &CallSiteDIE =
+          CU.constructCallSiteEntryDIE(ScopeDIE, CalleeSP, CalleeDecl, IsTail,
+                                       PCAddr, CallAddr, CallReg, AllocSiteTy);
 
       // Optionally emit call-site-param debug info.
       if (emitDebugEntryValues()) {
@@ -2711,7 +2712,8 @@ void DwarfDebug::skippedNonDebugFunction() {
 
 // Gather and emit post-function debug information.
 void DwarfDebug::endFunctionImpl(const MachineFunction *MF) {
-  const DISubprogram *SP = MF->getFunction().getSubprogram();
+  const Function &F = MF->getFunction();
+  const DISubprogram *SP = F.getSubprogram();
 
   assert(CurFn == MF &&
       "endFunction should be called with the same function as beginFunction");
@@ -2780,11 +2782,12 @@ void DwarfDebug::endFunctionImpl(const MachineFunction *MF) {
 
   ProcessedSPNodes.insert(SP);
   DIE &ScopeDIE =
-      TheCU.constructSubprogramScopeDIE(SP, FnScope, FunctionLineTableLabel);
+      TheCU.constructSubprogramScopeDIE(SP, F, FnScope, FunctionLineTableLabel);
   if (auto *SkelCU = TheCU.getSkeleton())
     if (!LScopes.getAbstractScopesList().empty() &&
         TheCU.getCUNode()->getSplitDebugInlining())
-      SkelCU->constructSubprogramScopeDIE(SP, FnScope, FunctionLineTableLabel);
+      SkelCU->constructSubprogramScopeDIE(SP, F, FnScope,
+                                          FunctionLineTableLabel);
 
   FunctionLineTableLabel = nullptr;
 

llvm/lib/CodeGen/AsmPrinter/DwarfFile.h

@@ -11,6 +11,7 @@
 
 #include "DwarfStringPool.h"
 #include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/SmallPtrSet.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/StringRef.h"
 #include "llvm/CodeGen/DIE.h"
@@ -27,6 +28,7 @@ class DbgVariable;
 class DbgLabel;
 class DINode;
 class DILocalScope;
+class DISubprogram;
 class DwarfCompileUnit;
 class DwarfUnit;
 class LexicalScope;
@@ -94,6 +96,9 @@ class DwarfFile {
   // Collection of abstract subprogram DIEs.
   DenseMap<const DILocalScope *, DIE *> AbstractLocalScopeDIEs;
   DenseMap<const DINode *, std::unique_ptr<DbgEntity>> AbstractEntities;
+  /// Keeps track of abstract subprograms to populate them only once.
+  // FIXME: merge creation and population of abstract scopes.
+  SmallPtrSet<const DISubprogram *, 8> FinalizedAbstractSubprograms;
 
   /// Maps MDNodes for type system with the corresponding DIEs. These DIEs can
   /// be shared across CUs, that is why we keep the map here instead
@@ -174,6 +179,10 @@ class DwarfFile {
     return AbstractEntities;
   }
 
+  auto &getFinalizedAbstractSubprograms() {
+    return FinalizedAbstractSubprograms;
+  }
+
   void insertDIE(const MDNode *TypeMD, DIE *Die) {
     DITypeNodeToDieMap.insert(std::make_pair(TypeMD, Die));
   }