[VPlan] Add and use debug location for VPScalarCastRecipe.

Update the recipe it always take a debug location and set it.

Author: fhahn

llvm/lib/Transforms/Vectorize/LoopVectorizationPlanner.h

@@ -243,8 +243,9 @@ class VPBuilder {
   }
 
   VPScalarCastRecipe *createScalarCast(Instruction::CastOps Opcode, VPValue *Op,
-                                       Type *ResultTy) {
-    return tryInsertInstruction(new VPScalarCastRecipe(Opcode, Op, ResultTy));
+                                       Type *ResultTy, DebugLoc DL) {
+    return tryInsertInstruction(
+        new VPScalarCastRecipe(Opcode, Op, ResultTy, DL));
   }
 
   VPWidenCastRecipe *createWidenCast(Instruction::CastOps Opcode, VPValue *Op,

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

@@ -8903,7 +8903,8 @@ static VPValue *addResumePhiRecipeForInduction(VPWidenInductionRecipe *WideIV,
   Type *ScalarTypeOfWideIV = TypeInfo.inferScalarType(WideIV);
   if (ScalarTypeOfWideIV != TypeInfo.inferScalarType(EndValue)) {
     EndValue = VectorPHBuilder.createScalarCast(Instruction::Trunc, EndValue,
-                                                ScalarTypeOfWideIV);
+                                                ScalarTypeOfWideIV,
+                                                WideIV->getDebugLoc());
   }
 
   auto *ResumePhiRecipe =

llvm/lib/Transforms/Vectorize/VPlan.h

@@ -1580,14 +1580,16 @@ class VPScalarCastRecipe : public VPSingleDefRecipe {
   Value *generate(VPTransformState &State);
 
 public:
-  VPScalarCastRecipe(Instruction::CastOps Opcode, VPValue *Op, Type *ResultTy)
-      : VPSingleDefRecipe(VPDef::VPScalarCastSC, {Op}), Opcode(Opcode),
+  VPScalarCastRecipe(Instruction::CastOps Opcode, VPValue *Op, Type *ResultTy,
+                     DebugLoc DL)
+      : VPSingleDefRecipe(VPDef::VPScalarCastSC, {Op}, DL), Opcode(Opcode),
         ResultTy(ResultTy) {}
 
   ~VPScalarCastRecipe() override = default;
 
   VPScalarCastRecipe *clone() override {
-    return new VPScalarCastRecipe(Opcode, getOperand(0), ResultTy);
+    return new VPScalarCastRecipe(Opcode, getOperand(0), ResultTy,
+                                  getDebugLoc());
   }
 
   VP_CLASSOF_IMPL(VPDef::VPScalarCastSC)

llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp

@@ -2328,6 +2328,7 @@ void VPReplicateRecipe::print(raw_ostream &O, const Twine &Indent,
 #endif
 
 Value *VPScalarCastRecipe ::generate(VPTransformState &State) {
+  State.setDebugLocFrom(getDebugLoc());
   assert(vputils::onlyFirstLaneUsed(this) &&
          "Codegen only implemented for first lane.");
   switch (Opcode) {

llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp

@@ -527,7 +527,8 @@ static VPScalarIVStepsRecipe *
 createScalarIVSteps(VPlan &Plan, InductionDescriptor::InductionKind Kind,
                     Instruction::BinaryOps InductionOpcode,
                     FPMathOperator *FPBinOp, Instruction *TruncI,
-                    VPValue *StartV, VPValue *Step, VPBuilder &Builder) {
+                    VPValue *StartV, VPValue *Step, DebugLoc DL,
+                    VPBuilder &Builder) {
   VPBasicBlock *HeaderVPBB = Plan.getVectorLoopRegion()->getEntryBasicBlock();
   VPCanonicalIVPHIRecipe *CanonicalIV = Plan.getCanonicalIV();
   VPSingleDefRecipe *BaseIV = Builder.createDerivedIV(
@@ -542,7 +543,7 @@ createScalarIVSteps(VPlan &Plan, InductionDescriptor::InductionKind Kind,
     assert(ResultTy->getScalarSizeInBits() > TruncTy->getScalarSizeInBits() &&
            "Not truncating.");
     assert(ResultTy->isIntegerTy() && "Truncation requires an integer type");
-    BaseIV = Builder.createScalarCast(Instruction::Trunc, BaseIV, TruncTy);
+    BaseIV = Builder.createScalarCast(Instruction::Trunc, BaseIV, TruncTy, DL);
     ResultTy = TruncTy;
   }
 
@@ -556,7 +557,7 @@ createScalarIVSteps(VPlan &Plan, InductionDescriptor::InductionKind Kind,
         cast<VPBasicBlock>(HeaderVPBB->getSingleHierarchicalPredecessor());
     VPBuilder::InsertPointGuard Guard(Builder);
     Builder.setInsertPoint(VecPreheader);
-    Step = Builder.createScalarCast(Instruction::Trunc, Step, ResultTy);
+    Step = Builder.createScalarCast(Instruction::Trunc, Step, ResultTy, DL);
   }
   return Builder.createScalarIVSteps(InductionOpcode, FPBinOp, BaseIV, Step);
 }
@@ -588,7 +589,7 @@ static void legalizeAndOptimizeInductions(VPlan &Plan) {
       VPValue *StepV = PtrIV->getOperand(1);
       VPScalarIVStepsRecipe *Steps = createScalarIVSteps(
           Plan, InductionDescriptor::IK_IntInduction, Instruction::Add, nullptr,
-          nullptr, StartV, StepV, Builder);
+          nullptr, StartV, StepV, PtrIV->getDebugLoc(), Builder);
 
       VPValue *PtrAdd = Builder.createPtrAdd(PtrIV->getStartValue(), Steps,
                                              PtrIV->getDebugLoc(), "next.gep");
@@ -612,7 +613,7 @@ static void legalizeAndOptimizeInductions(VPlan &Plan) {
         Plan, ID.getKind(), ID.getInductionOpcode(),
         dyn_cast_or_null<FPMathOperator>(ID.getInductionBinOp()),
         WideIV->getTruncInst(), WideIV->getStartValue(), WideIV->getStepValue(),
-        Builder);
+        WideIV->getDebugLoc(), Builder);
 
     // Update scalar users of IV to use Step instead.
     if (!HasOnlyVectorVFs)
@@ -1689,9 +1690,9 @@ bool VPlanTransforms::tryAddExplicitVectorLength(
   VPSingleDefRecipe *OpVPEVL = VPEVL;
   if (unsigned IVSize = CanonicalIVPHI->getScalarType()->getScalarSizeInBits();
       IVSize != 32) {
-    OpVPEVL = new VPScalarCastRecipe(IVSize < 32 ? Instruction::Trunc
-                                                 : Instruction::ZExt,
-                                     OpVPEVL, CanonicalIVPHI->getScalarType());
+    OpVPEVL = new VPScalarCastRecipe(
+        IVSize < 32 ? Instruction::Trunc : Instruction::ZExt, OpVPEVL,
+        CanonicalIVPHI->getScalarType(), CanonicalIVIncrement->getDebugLoc());
     OpVPEVL->insertBefore(CanonicalIVIncrement);
   }
   auto *NextEVLIV =

llvm/test/Transforms/LoopVectorize/preserve-dbg-loc-and-loop-metadata.ll

@@ -91,7 +91,7 @@ define void @scalar_cast_dbg(ptr nocapture %a, i32 %start, i64 %k) {
 ; DEBUGLOC:   = trunc i64 %index to i32, !dbg [[CASTLOC:![0-9]+]]
 ;
 ; DEBUGLOC: loop:
-; DEBUGLOC-NOT:   %trunc.iv = trunc i64 %iv to i32, !dbg [[CASTLOC]]
+; DEBUGLOC:   %trunc.iv = trunc i64 %iv to i32, !dbg [[CASTLOC]]
 ;
 entry:
   br label %loop