Revert "[DebugInfo][RemoveDIs] Remove a swathe of debug-intrinsic code (llvm#144389)"

This reverts commit 9eb0020.

Author: ronlieb

llvm/include/llvm/Analysis/IRSimilarityIdentifier.h

@@ -545,6 +545,10 @@ struct IRInstructionMapper {
     // dependent.
     InstrType visitLandingPadInst(LandingPadInst &LPI) { return Illegal; }
     InstrType visitFuncletPadInst(FuncletPadInst &FPI) { return Illegal; }
+    // DebugInfo should be included in the regions, but should not be
+    // analyzed for similarity as it has no bearing on the outcome of the
+    // program.
+    InstrType visitDbgInfoIntrinsic(DbgInfoIntrinsic &DII) { return Invisible; }
     InstrType visitIntrinsicInst(IntrinsicInst &II) {
       // These are disabled due to complications in the CodeExtractor when
       // outlining these instructions.  For instance, It is unclear what we

llvm/include/llvm/Analysis/PtrUseVisitor.h

@@ -285,6 +285,7 @@ class PtrUseVisitor : protected InstVisitor<DerivedT>,
 
   // No-op intrinsics which we know don't escape the pointer to logic in
   // some other function.
+  void visitDbgInfoIntrinsic(DbgInfoIntrinsic &I) {}
   void visitMemIntrinsic(MemIntrinsic &I) {}
   void visitIntrinsicInst(IntrinsicInst &II) {
     switch (II.getIntrinsicID()) {

llvm/include/llvm/Transforms/Utils/Local.h

@@ -394,9 +394,12 @@ handleUnreachableTerminator(Instruction *I,
                             SmallVectorImpl<Value *> &PoisonedValues);
 
 /// Remove all instructions from a basic block other than its terminator
-/// and any present EH pad instructions. Returns the number of instructions
+/// and any present EH pad instructions. Returns a pair where the first element
+/// is the number of instructions (excluding debug info intrinsics) that have
+/// been removed, and the second element is the number of debug info intrinsics
 /// that have been removed.
-LLVM_ABI unsigned removeAllNonTerminatorAndEHPadInstructions(BasicBlock *BB);
+LLVM_ABI std::pair<unsigned, unsigned>
+removeAllNonTerminatorAndEHPadInstructions(BasicBlock *BB);
 
 /// Insert an unreachable instruction before the specified
 /// instruction, making it and the rest of the code in the block dead.

llvm/lib/Analysis/AliasSetTracker.cpp

@@ -343,6 +343,9 @@ void AliasSetTracker::add(AnyMemTransferInst *MTI) {
 }
 
 void AliasSetTracker::addUnknown(Instruction *Inst) {
+  if (isa<DbgInfoIntrinsic>(Inst))
+    return; // Ignore DbgInfo Intrinsics.
+
   if (auto *II = dyn_cast<IntrinsicInst>(Inst)) {
     // These intrinsics will show up as affecting memory, but they are just
     // markers.

llvm/lib/Analysis/CallGraph.cpp

@@ -34,7 +34,8 @@ CallGraph::CallGraph(Module &M)
       CallsExternalNode(std::make_unique<CallGraphNode>(this, nullptr)) {
   // Add every interesting function to the call graph.
   for (Function &F : M)
-    addToCallGraph(&F);
+    if (!isDbgInfoIntrinsic(F.getIntrinsicID()))
+      addToCallGraph(&F);
 }
 
 CallGraph::CallGraph(CallGraph &&Arg)
@@ -100,7 +101,7 @@ void CallGraph::populateCallGraphNode(CallGraphNode *Node) {
         const Function *Callee = Call->getCalledFunction();
         if (!Callee)
           Node->addCalledFunction(Call, CallsExternalNode.get());
-        else
+        else if (!isDbgInfoIntrinsic(Callee->getIntrinsicID()))
           Node->addCalledFunction(Call, getOrInsertFunction(Callee));
 
         // Add reference to callback functions.

llvm/lib/Analysis/DemandedBits.cpp

@@ -46,7 +46,8 @@ using namespace llvm::PatternMatch;
 #define DEBUG_TYPE "demanded-bits"
 
 static bool isAlwaysLive(Instruction *I) {
-  return I->isTerminator() || I->isEHPad() || I->mayHaveSideEffects();
+  return I->isTerminator() || isa<DbgInfoIntrinsic>(I) || I->isEHPad() ||
+         I->mayHaveSideEffects();
 }
 
 void DemandedBits::determineLiveOperandBits(

llvm/lib/Analysis/Loads.cpp

@@ -434,7 +434,7 @@ bool llvm::isSafeToLoadUnconditionally(Value *V, Align Alignment, const APInt &S
     // If we see a free or a call which may write to memory (i.e. which might do
     // a free) the pointer could be marked invalid.
     if (isa<CallInst>(BBI) && BBI->mayWriteToMemory() &&
-        !isa<LifetimeIntrinsic>(BBI))
+        !isa<LifetimeIntrinsic>(BBI) && !isa<DbgInfoIntrinsic>(BBI))
       return false;
 
     Value *AccessedPtr;

llvm/lib/Analysis/MemoryDependenceAnalysis.cpp

@@ -188,6 +188,9 @@ MemDepResult MemoryDependenceResults::getCallDependencyFrom(
   // Walk backwards through the block, looking for dependencies.
   while (ScanIt != BB->begin()) {
     Instruction *Inst = &*--ScanIt;
+    // Debug intrinsics don't cause dependences and should not affect Limit
+    if (isa<DbgInfoIntrinsic>(Inst))
+      continue;
 
     // Limit the amount of scanning we do so we don't end up with quadratic
     // running time on extreme testcases.
@@ -429,6 +432,11 @@ MemDepResult MemoryDependenceResults::getSimplePointerDependencyFrom(
   while (ScanIt != BB->begin()) {
     Instruction *Inst = &*--ScanIt;
 
+    if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(Inst))
+      // Debug intrinsics don't (and can't) cause dependencies.
+      if (isa<DbgInfoIntrinsic>(II))
+        continue;
+
     // Limit the amount of scanning we do so we don't end up with quadratic
     // running time on extreme testcases.
     --*Limit;

llvm/lib/Analysis/ValueTracking.cpp

@@ -7846,6 +7846,8 @@ bool llvm::isGuaranteedToTransferExecutionToSuccessor(
    iterator_range<BasicBlock::const_iterator> Range, unsigned ScanLimit) {
   assert(ScanLimit && "scan limit must be non-zero");
   for (const Instruction &I : Range) {
+    if (isa<DbgInfoIntrinsic>(I))
+        continue;
     if (--ScanLimit == 0)
       return false;
     if (!isGuaranteedToTransferExecutionToSuccessor(&I))
@@ -8048,6 +8050,8 @@ static bool programUndefinedIfUndefOrPoison(const Value *V,
     // well-defined operands.
 
     for (const auto &I : make_range(Begin, End)) {
+      if (isa<DbgInfoIntrinsic>(I))
+        continue;
       if (--ScanLimit == 0)
         break;
 
@@ -8072,6 +8076,8 @@ static bool programUndefinedIfUndefOrPoison(const Value *V,
 
   while (true) {
     for (const auto &I : make_range(Begin, End)) {
+      if (isa<DbgInfoIntrinsic>(I))
+        continue;
       if (--ScanLimit == 0)
         return false;
       if (mustTriggerUB(&I, YieldsPoison))

llvm/lib/CodeGen/CodeGenPrepare.cpp

@@ -896,7 +896,12 @@ BasicBlock *CodeGenPrepare::findDestBlockOfMergeableEmptyBlock(BasicBlock *BB) {
   BasicBlock::iterator BBI = BI->getIterator();
   if (BBI != BB->begin()) {
     --BBI;
-    if (!isa<PHINode>(BBI))
+    while (isa<DbgInfoIntrinsic>(BBI)) {
+      if (BBI == BB->begin())
+        break;
+      --BBI;
+    }
+    if (!isa<DbgInfoIntrinsic>(BBI) && !isa<PHINode>(BBI))
       return nullptr;
   }
 
@@ -2976,9 +2981,10 @@ bool CodeGenPrepare::dupRetToEnableTailCallOpts(BasicBlock *BB,
   // Make sure there are no instructions between the first instruction
   // and return.
   BasicBlock::const_iterator BI = BB->getFirstNonPHIIt();
-  // Skip over pseudo-probes and the bitcast.
-  while (&*BI == BCI || &*BI == EVI || isa<PseudoProbeInst>(BI) ||
-         isLifetimeEndOrBitCastFor(&*BI) || isFakeUse(&*BI))
+  // Skip over debug and the bitcast.
+  while (isa<DbgInfoIntrinsic>(BI) || &*BI == BCI || &*BI == EVI ||
+         isa<PseudoProbeInst>(BI) || isLifetimeEndOrBitCastFor(&*BI) ||
+         isFakeUse(&*BI))
     BI = std::next(BI);
   if (&*BI != RetI)
     return false;

llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

@@ -1320,7 +1320,10 @@ void SelectionDAGBuilder::visit(const Instruction &I) {
     HandlePHINodesInSuccessorBlocks(I.getParent());
   }
 
-  ++SDNodeOrder;
+  // Increase the SDNodeOrder if dealing with a non-debug instruction.
+  if (!isa<DbgInfoIntrinsic>(I))
+    ++SDNodeOrder;
+
   CurInst = &I;
 
   // Set inserted listener only if required.

llvm/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp

@@ -1507,6 +1507,7 @@ static bool isFoldedOrDeadInstruction(const Instruction *I,
                                       const FunctionLoweringInfo &FuncInfo) {
   return !I->mayWriteToMemory() && // Side-effecting instructions aren't folded.
          !I->isTerminator() &&     // Terminators aren't folded.
+         !isa<DbgInfoIntrinsic>(I) && // Debug instructions aren't folded.
          !I->isEHPad() &&             // EH pad instructions aren't folded.
          !FuncInfo.isExportedInst(I); // Exported instrs must be computed.
 }

llvm/lib/IR/DebugInfo.cpp

@@ -586,6 +586,11 @@ bool llvm::stripDebugInfo(Function &F) {
   DenseMap<MDNode *, MDNode *> LoopIDsMap;
   for (BasicBlock &BB : F) {
     for (Instruction &I : llvm::make_early_inc_range(BB)) {
+      if (isa<DbgInfoIntrinsic>(&I)) {
+        I.eraseFromParent();
+        Changed = true;
+        continue;
+      }
       if (I.getDebugLoc()) {
         Changed = true;
         I.setDebugLoc(DebugLoc());

llvm/lib/Target/AArch64/AArch64StackTagging.cpp

@@ -369,7 +369,8 @@ Instruction *AArch64StackTagging::collectInitializers(Instruction *StartInst,
 
   unsigned Count = 0;
   for (; Count < ClScanLimit && !BI->isTerminator(); ++BI) {
-    ++Count;
+    if (!isa<DbgInfoIntrinsic>(*BI))
+      ++Count;
 
     if (isNoModRef(AA->getModRefInfo(&*BI, AllocaLoc)))
       continue;

llvm/lib/Target/Hexagon/HexagonLoopIdiomRecognition.cpp

@@ -2318,7 +2318,7 @@ bool HexagonLoopIdiomRecognize::coverLoop(Loop *L,
   // instructions in it that are not involved in the original set Insts.
   for (auto *B : L->blocks()) {
     for (auto &In : *B) {
-      if (isa<BranchInst>(In))
+      if (isa<BranchInst>(In) || isa<DbgInfoIntrinsic>(In))
         continue;
       if (!Worklist.count(&In) && In.mayHaveSideEffects())
         return false;

llvm/lib/Target/PowerPC/PPCBoolRetToInt.cpp

@@ -117,7 +117,7 @@ class PPCBoolRetToInt : public FunctionPass {
 
   // A PHINode is Promotable if:
   // 1. Its type is i1 AND
-  // 2. All of its uses are ReturnInt, CallInst, or PHINode
+  // 2. All of its uses are ReturnInt, CallInst, PHINode, or DbgInfoIntrinsic
   // AND
   // 3. All of its operands are Constant or Argument or
   //    CallInst or PHINode AND
@@ -136,7 +136,8 @@ class PPCBoolRetToInt : public FunctionPass {
     for (const PHINode *P : Promotable) {
       // Condition 2 and 3
       auto IsValidUser = [] (const Value *V) -> bool {
-        return isa<ReturnInst>(V) || isa<CallInst>(V) || isa<PHINode>(V);
+        return isa<ReturnInst>(V) || isa<CallInst>(V) || isa<PHINode>(V) ||
+        isa<DbgInfoIntrinsic>(V);
       };
       auto IsValidOperand = [] (const Value *V) -> bool {
         return isa<Constant>(V) || isa<Argument>(V) || isa<CallInst>(V) ||

llvm/lib/Transforms/AggressiveInstCombine/AggressiveInstCombine.cpp

@@ -719,7 +719,9 @@ static bool foldLoadsRecursive(Value *V, LoadOps &LOps, const DataLayout &DL,
     if (Inst.mayWriteToMemory() && isModSet(AA.getModRefInfo(&Inst, Loc)))
       return false;
 
-    if (++NumScanned > MaxInstrsToScan)
+    // Ignore debug info so that's not counted against MaxInstrsToScan.
+    // Otherwise debug info could affect codegen.
+    if (!isa<DbgInfoIntrinsic>(Inst) && ++NumScanned > MaxInstrsToScan)
       return false;
   }
 

llvm/lib/Transforms/IPO/IROutliner.cpp

@@ -717,6 +717,8 @@ static void moveFunctionData(Function &Old, Function &New,
     if (ReturnInst *RI = dyn_cast<ReturnInst>(I))
       NewEnds.insert(std::make_pair(RI->getReturnValue(), &CurrBB));
 
+    std::vector<Instruction *> DebugInsts;
+
     for (Instruction &Val : CurrBB) {
       // Since debug-info originates from many different locations in the
       // program, it will cause incorrect reporting from a debugger if we keep
@@ -747,12 +749,21 @@ static void moveFunctionData(Function &Old, Function &New,
       // From this point we are only handling call instructions.
       CallInst *CI = cast<CallInst>(&Val);
 
+      // Collect debug intrinsics for later removal.
+      if (isa<DbgInfoIntrinsic>(CI)) {
+        DebugInsts.push_back(&Val);
+        continue;
+      }
+
       // Edit the scope of called functions inside of outlined functions.
       if (DISubprogram *SP = New.getSubprogram()) {
         DILocation *DI = DILocation::get(New.getContext(), 0, 0, SP);
         Val.setDebugLoc(DI);
       }
     }
+
+    for (Instruction *I : DebugInsts)
+      I->eraseFromParent();
   }
 }
 

llvm/lib/Transforms/IPO/SampleProfileProbe.cpp

@@ -385,7 +385,8 @@ void SampleProfileProber::instrumentOneFunc(Function &F, TargetMachine *TM) {
     // line number. Real instructions generated by optimizations may not come
     // with a line number either.
     auto HasValidDbgLine = [](Instruction *J) {
-      return !isa<PHINode>(J) && !J->isLifetimeStartOrEnd() && J->getDebugLoc();
+      return !isa<PHINode>(J) && !isa<DbgInfoIntrinsic>(J) &&
+             !J->isLifetimeStartOrEnd() && J->getDebugLoc();
     };
 
     Instruction *J = &*BB->getFirstInsertionPt();

llvm/lib/Transforms/InstCombine/InstructionCombining.cpp

@@ -4787,7 +4787,11 @@ bool InstCombinerImpl::freezeOtherUses(FreezeInst &FI) {
     MoveBefore = *MoveBeforeOpt;
   }
 
-  // Re-point iterator to come after any debug-info records.
+  // Don't move to the position of a debug intrinsic.
+  if (isa<DbgInfoIntrinsic>(MoveBefore))
+    MoveBefore = MoveBefore->getNextNonDebugInstruction()->getIterator();
+  // Re-point iterator to come after any debug-info records, if we're
+  // running in "RemoveDIs" mode
   MoveBefore.setHeadBit(false);
 
   bool Changed = false;
@@ -5578,9 +5582,11 @@ bool InstCombinerImpl::prepareWorklist(Function &F) {
       continue;
 
     unsigned NumDeadInstInBB;
-    NumDeadInstInBB = removeAllNonTerminatorAndEHPadInstructions(&BB);
+    unsigned NumDeadDbgInstInBB;
+    std::tie(NumDeadInstInBB, NumDeadDbgInstInBB) =
+        removeAllNonTerminatorAndEHPadInstructions(&BB);
 
-    MadeIRChange |= NumDeadInstInBB != 0;
+    MadeIRChange |= NumDeadInstInBB + NumDeadDbgInstInBB > 0;
     NumDeadInst += NumDeadInstInBB;
   }
 

llvm/lib/Transforms/Instrumentation/GCOVProfiling.cpp

@@ -583,6 +583,10 @@ static bool functionHasLines(const Function &F, unsigned &EndLine) {
   EndLine = 0;
   for (const auto &BB : F) {
     for (const auto &I : BB) {
+      // Debug intrinsic locations correspond to the location of the
+      // declaration, not necessarily any statements or expressions.
+      if (isa<DbgInfoIntrinsic>(&I)) continue;
+
       const DebugLoc &Loc = I.getDebugLoc();
       if (!Loc)
         continue;
@@ -870,6 +874,10 @@ bool GCOVProfiler::emitProfileNotes(
         }
 
         for (const auto &I : BB) {
+          // Debug intrinsic locations correspond to the location of the
+          // declaration, not necessarily any statements or expressions.
+          if (isa<DbgInfoIntrinsic>(&I)) continue;
+
           const DebugLoc &Loc = I.getDebugLoc();
           if (!Loc)
             continue;

llvm/lib/Transforms/Instrumentation/ThreadSanitizer.cpp

@@ -527,7 +527,8 @@ bool ThreadSanitizer::sanitizeFunction(Function &F,
         AtomicAccesses.push_back(&Inst);
       else if (isa<LoadInst>(Inst) || isa<StoreInst>(Inst))
         LocalLoadsAndStores.push_back(&Inst);
-      else if (isa<CallInst>(Inst) || isa<InvokeInst>(Inst)) {
+      else if ((isa<CallInst>(Inst) && !isa<DbgInfoIntrinsic>(Inst)) ||
+               isa<InvokeInst>(Inst)) {
         if (CallInst *CI = dyn_cast<CallInst>(&Inst))
           maybeMarkSanitizerLibraryCallNoBuiltin(CI, &TLI);
         if (isa<MemIntrinsic>(Inst))

llvm/lib/Transforms/Scalar/ADCE.cpp

@@ -562,7 +562,20 @@ ADCEChanged AggressiveDeadCodeElimination::removeDeadInstructions() {
     if (isLive(&I))
       continue;
 
-    Changed.ChangedNonDebugInstr = true;
+    if (auto *DII = dyn_cast<DbgInfoIntrinsic>(&I)) {
+      // Avoid removing a dbg.assign that is linked to instructions because it
+      // holds information about an existing store.
+      if (auto *DAI = dyn_cast<DbgAssignIntrinsic>(DII))
+        if (!at::getAssignmentInsts(DAI).empty())
+          continue;
+      // Check if the scope of this variable location is alive.
+      if (AliveScopes.count(DII->getDebugLoc()->getScope()))
+        continue;
+
+      // Fallthrough and drop the intrinsic.
+    } else {
+      Changed.ChangedNonDebugInstr = true;
+    }
 
     // Prepare to delete.
     Worklist.push_back(&I);

llvm/lib/Transforms/Scalar/GVN.cpp

@@ -2684,6 +2684,10 @@ bool GVNPass::propagateEquality(Value *LHS, Value *RHS,
 /// When calculating availability, handle an instruction
 /// by inserting it into the appropriate sets.
 bool GVNPass::processInstruction(Instruction *I) {
+  // Ignore dbg info intrinsics.
+  if (isa<DbgInfoIntrinsic>(I))
+    return false;
+
   // If the instruction can be easily simplified then do so now in preference
   // to value numbering it.  Value numbering often exposes redundancies, for
   // example if it determines that %y is equal to %x then the instruction
@@ -2970,7 +2974,8 @@ bool GVNPass::performScalarPREInsertion(Instruction *Instr, BasicBlock *Pred,
 bool GVNPass::performScalarPRE(Instruction *CurInst) {
   if (isa<AllocaInst>(CurInst) || CurInst->isTerminator() ||
       isa<PHINode>(CurInst) || CurInst->getType()->isVoidTy() ||
-      CurInst->mayReadFromMemory() || CurInst->mayHaveSideEffects())
+      CurInst->mayReadFromMemory() || CurInst->mayHaveSideEffects() ||
+      isa<DbgInfoIntrinsic>(CurInst))
     return false;
 
   // Don't do PRE on compares. The PHI would prevent CodeGenPrepare from

llvm/lib/Transforms/Scalar/GVNHoist.cpp

@@ -1166,7 +1166,8 @@ std::pair<unsigned, unsigned> GVNHoist::hoistExpressions(Function &F) {
         SI.insert(Store, VN);
       else if (auto *Call = dyn_cast<CallInst>(&I1)) {
         if (auto *Intr = dyn_cast<IntrinsicInst>(Call)) {
-          if (Intr->getIntrinsicID() == Intrinsic::assume ||
+          if (isa<DbgInfoIntrinsic>(Intr) ||
+              Intr->getIntrinsicID() == Intrinsic::assume ||
               Intr->getIntrinsicID() == Intrinsic::sideeffect)
             continue;
         }