[VPlan] Set correct flags when creating and cloning VPWidenCastRecipe.

Make sure that we set the correct wrap flags when creating new
VPWidenCastRecipes for truncs and preserve the flags from the recipe
directly when cloning, to make sure they are not dropped.

Fixes llvm#160396

Author: fhahn

llvm/lib/Transforms/Vectorize/VPlan.h

@@ -705,6 +705,9 @@ class VPIRFlags {
   VPIRFlags(WrapFlagsTy WrapFlags)
       : OpType(OperationType::OverflowingBinOp), WrapFlags(WrapFlags) {}
 
+  VPIRFlags(TruncFlagsTy TruncFlags)
+      : OpType(OperationType::Trunc), TruncFlags(TruncFlags) {}
+
   VPIRFlags(FastMathFlags FMFs) : OpType(OperationType::FPMathOp), FMFs(FMFs) {}
 
   VPIRFlags(DisjointFlagsTy DisjointFlags)
@@ -1494,21 +1497,22 @@ class VPWidenCastRecipe : public VPRecipeWithIRFlags, public VPIRMetadata {
 
   VPWidenCastRecipe(Instruction::CastOps Opcode, VPValue *Op, Type *ResultTy,
                     const VPIRFlags &Flags = {},
+                    const VPIRMetadata &Metadata = {},
                     DebugLoc DL = DebugLoc::getUnknown())
       : VPRecipeWithIRFlags(VPDef::VPWidenCastSC, Op, Flags, DL),
-        VPIRMetadata(), Opcode(Opcode), ResultTy(ResultTy) {
+        VPIRMetadata(Metadata), Opcode(Opcode), ResultTy(ResultTy) {
     assert(flagsValidForOpcode(Opcode) &&
            "Set flags not supported for the provided opcode");
   }
 
   ~VPWidenCastRecipe() override = default;
 
   VPWidenCastRecipe *clone() override {
+    auto *New = new VPWidenCastRecipe(Opcode, getOperand(0), ResultTy, *this,
+                                      *this, getDebugLoc());
     if (auto *UV = getUnderlyingValue())
-      return new VPWidenCastRecipe(Opcode, getOperand(0), ResultTy,
-                                   *cast<CastInst>(UV));
-
-    return new VPWidenCastRecipe(Opcode, getOperand(0), ResultTy);
+      New->setUnderlyingValue(UV);
+    return New;
   }
 
   VP_CLASSOF_IMPL(VPDef::VPWidenCastSC)

llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp

@@ -2016,13 +2016,13 @@ bool VPIRFlags::flagsValidForOpcode(unsigned Opcode) const {
     return Opcode == Instruction::FAdd || Opcode == Instruction::FMul ||
            Opcode == Instruction::FSub || Opcode == Instruction::FNeg ||
            Opcode == Instruction::FDiv || Opcode == Instruction::FRem ||
+           Opcode == Instruction::FPExt || Opcode == Instruction::FPTrunc ||
            Opcode == Instruction::FCmp || Opcode == Instruction::Select ||
            Opcode == VPInstruction::WideIVStep ||
            Opcode == VPInstruction::ReductionStartVector ||
            Opcode == VPInstruction::ComputeReductionResult;
   case OperationType::NonNegOp:
-    return Opcode == Instruction::ZExt;
-    break;
+    return Opcode == Instruction::ZExt || Opcode == Instruction::UIToFP;
   case OperationType::Cmp:
     return Opcode == Instruction::FCmp || Opcode == Instruction::ICmp;
   case OperationType::Other:

llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp

@@ -2195,7 +2195,8 @@ void VPlanTransforms::truncateToMinimalBitwidths(
         auto [ProcessedIter, IterIsEmpty] = ProcessedTruncs.try_emplace(Op);
         VPWidenCastRecipe *NewOp =
             IterIsEmpty
-                ? new VPWidenCastRecipe(Instruction::Trunc, Op, NewResTy)
+                ? new VPWidenCastRecipe(Instruction::Trunc, Op, NewResTy,
+                                        VPIRFlags::TruncFlagsTy(false, false))
                 : ProcessedIter->second;
         R.setOperand(Idx, NewOp);
         if (!IterIsEmpty)
@@ -3566,13 +3567,13 @@ tryToMatchAndCreateMulAccumulateReduction(VPReductionRecipe *Red,
                                    Mul, Ext0, Ext1, Ext)) {
       auto *NewExt0 = new VPWidenCastRecipe(
           Ext0->getOpcode(), Ext0->getOperand(0), Ext->getResultType(), *Ext0,
-          Ext0->getDebugLoc());
+          *Ext0, Ext0->getDebugLoc());
       NewExt0->insertBefore(Ext0);
 
       VPWidenCastRecipe *NewExt1 = NewExt0;
       if (Ext0 != Ext1) {
         NewExt1 = new VPWidenCastRecipe(Ext1->getOpcode(), Ext1->getOperand(0),
-                                        Ext->getResultType(), *Ext1,
+                                        Ext->getResultType(), *Ext1, *Ext1,
                                         Ext1->getDebugLoc());
         NewExt1->insertBefore(Ext1);
       }