1 files changed, 82 insertions, 1 deletions

diff --git a/llvm/lib/Transforms/Utils/LoopRotationUtils.cpp b/llvm/lib/Transforms/Utils/LoopRotationUtils.cpp
index ae155ac..cbef27d 100644
--- a/llvm/lib/Transforms/Utils/LoopRotationUtils.cpp
+++ b/llvm/lib/Transforms/Utils/LoopRotationUtils.cpp
@@ -159,7 +159,8 @@ static void RewriteUsesOfClonedInstructions(BasicBlock *OrigHeader,
     // Replace MetadataAsValue(ValueAsMetadata(OrigHeaderVal)) uses in debug
     // intrinsics.
     SmallVector<DbgValueInst *, 1> DbgValues;
-    llvm::findDbgValues(DbgValues, OrigHeaderVal);
+    SmallVector<DPValue *, 1> DPValues;
+    llvm::findDbgValues(DbgValues, OrigHeaderVal, &DPValues);
     for (auto &DbgValue : DbgValues) {
       // The original users in the OrigHeader are already using the original
       // definitions.
@@ -180,6 +181,29 @@ static void RewriteUsesOfClonedInstructions(BasicBlock *OrigHeader,
         NewVal = UndefValue::get(OrigHeaderVal->getType());
       DbgValue->replaceVariableLocationOp(OrigHeaderVal, NewVal);
     }
+
+    // RemoveDIs: duplicate implementation for non-instruction debug-info
+    // storage in DPValues.
+    for (DPValue *DPV : DPValues) {
+      // The original users in the OrigHeader are already using the original
+      // definitions.
+      BasicBlock *UserBB = DPV->getMarker()->getParent();
+      if (UserBB == OrigHeader)
+        continue;
+
+      // Users in the OrigPreHeader need to use the value to which the
+      // original definitions are mapped and anything else can be handled by
+      // the SSAUpdater. To avoid adding PHINodes, check if the value is
+      // available in UserBB, if not substitute undef.
+      Value *NewVal;
+      if (UserBB == OrigPreheader)
+        NewVal = OrigPreHeaderVal;
+      else if (SSA.HasValueForBlock(UserBB))
+        NewVal = SSA.GetValueInMiddleOfBlock(UserBB);
+      else
+        NewVal = UndefValue::get(OrigHeaderVal->getType());
+      DPV->replaceVariableLocationOp(OrigHeaderVal, NewVal);
+    }
   }
 }
 
@@ -531,6 +555,18 @@ bool LoopRotate::rotateLoop(Loop *L, bool SimplifiedLatch) {
         break;
     }
 
+    // Duplicate implementation for DPValues, the non-instruction format of
+    // debug-info records in RemoveDIs.
+    auto makeHashDPV = [](const DPValue &D) -> DbgIntrinsicHash {
+      auto VarLocOps = D.location_ops();
+      return {{hash_combine_range(VarLocOps.begin(), VarLocOps.end()),
+               D.getVariable()},
+              D.getExpression()};
+    };
+    for (Instruction &I : llvm::drop_begin(llvm::reverse(*OrigPreheader)))
+      for (const DPValue &DPV : I.getDbgValueRange())
+        DbgIntrinsics.insert(makeHashDPV(DPV));
+
     // Remember the local noalias scope declarations in the header. After the
     // rotation, they must be duplicated and the scope must be cloned. This
     // avoids unwanted interaction across iterations.
@@ -539,6 +575,29 @@ bool LoopRotate::rotateLoop(Loop *L, bool SimplifiedLatch) {
       if (auto *Decl = dyn_cast<NoAliasScopeDeclInst>(&I))
         NoAliasDeclInstructions.push_back(Decl);
 
+    Module *M = OrigHeader->getModule();
+
+    // Track the next DPValue to clone. If we have a sequence where an
+    // instruction is hoisted instead of being cloned:
+    //    DPValue blah
+    //    %foo = add i32 0, 0
+    //    DPValue xyzzy
+    //    %bar = call i32 @foobar()
+    // where %foo is hoisted, then the DPValue "blah" will be seen twice, once
+    // attached to %foo, then when %foo his hoisted it will "fall down" onto the
+    // function call:
+    //    DPValue blah
+    //    DPValue xyzzy
+    //    %bar = call i32 @foobar()
+    // causing it to appear attached to the call too.
+    //
+    // To avoid this, cloneDebugInfoFrom takes an optional "start cloning from
+    // here" position to account for this behaviour. We point it at any DPValues
+    // on the next instruction, here labelled xyzzy, before we hoist %foo.
+    // Later, we only only clone DPValues from that position (xyzzy) onwards,
+    // which avoids cloning DPValue "blah" multiple times.
+    std::optional<DPValue::self_iterator> NextDbgInst = std::nullopt;
+
     while (I != E) {
       Instruction *Inst = &*I++;
 
@@ -551,7 +610,17 @@ bool LoopRotate::rotateLoop(Loop *L, bool SimplifiedLatch) {
       if (L->hasLoopInvariantOperands(Inst) && !Inst->mayReadFromMemory() &&
           !Inst->mayWriteToMemory() && !Inst->isTerminator() &&
           !isa<DbgInfoIntrinsic>(Inst) && !isa<AllocaInst>(Inst)) {
+
+        if (LoopEntryBranch->getParent()->IsNewDbgInfoFormat) {
+          auto DbgValueRange =
+              LoopEntryBranch->cloneDebugInfoFrom(Inst, NextDbgInst);
+          RemapDPValueRange(M, DbgValueRange, ValueMap,
+                            RF_NoModuleLevelChanges | RF_IgnoreMissingLocals);
+        }
+
+        NextDbgInst = I->getDbgValueRange().begin();
         Inst->moveBefore(LoopEntryBranch);
+
         ++NumInstrsHoisted;
         continue;
       }
@@ -562,6 +631,17 @@ bool LoopRotate::rotateLoop(Loop *L, bool SimplifiedLatch) {
 
       ++NumInstrsDuplicated;
 
+      if (LoopEntryBranch->getParent()->IsNewDbgInfoFormat) {
+        auto Range = C->cloneDebugInfoFrom(Inst, NextDbgInst);
+        // Erase anything we've seen before.
+        for (DPValue &DPV : make_early_inc_range(Range))
+          if (DbgIntrinsics.count(makeHashDPV(DPV)))
+            DPV.eraseFromParent();
+        RemapDPValueRange(M, Range, ValueMap,
+                          RF_NoModuleLevelChanges | RF_IgnoreMissingLocals);
+        NextDbgInst = std::nullopt;
+      }
+
       // Eagerly remap the operands of the instruction.
       RemapInstruction(C, ValueMap,
                        RF_NoModuleLevelChanges | RF_IgnoreMissingLocals);
@@ -676,6 +756,7 @@ bool LoopRotate::rotateLoop(Loop *L, bool SimplifiedLatch) {
     // OrigPreHeader's old terminator (the original branch into the loop), and
     // remove the corresponding incoming values from the PHI nodes in OrigHeader.
     LoopEntryBranch->eraseFromParent();
+    OrigPreheader->flushTerminatorDbgValues();
 
     // Update MemorySSA before the rewrite call below changes the 1:1
     // instruction:cloned_instruction_or_value mapping.