[LV] Allow tryToCreateWidenRecipe to return a VPValue, use for blends.

Generalize the return value of tryToCreateWidenRecipe to return either a
newly create recipe or an existing VPValue. Use this to avoid creating
unnecessary VPBlendRecipes.

Fixes PR44800.

Author: fhahn

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

@@ -8249,15 +8249,24 @@ VPRecipeBuilder::tryToOptimizeInductionTruncate(TruncInst *I, VFRange &Range,
   return nullptr;
 }
 
-VPBlendRecipe *VPRecipeBuilder::tryToBlend(PHINode *Phi, VPlanPtr &Plan) {
+VPRecipeOrVPValueTy VPRecipeBuilder::tryToBlend(PHINode *Phi, VPlanPtr &Plan) {
+  // If all incoming values are equal, the incoming VPValue can be used directly
+  // instead of creating a new VPBlendRecipe.
+  Value *FirstIncoming = Phi->getIncomingValue(0);
+  if (all_of(Phi->incoming_values(), [FirstIncoming](const Value *Inc) {
+        return FirstIncoming == Inc;
+      })) {
+    return Plan->getOrAddVPValue(Phi->getIncomingValue(0));
+  }
+
   // We know that all PHIs in non-header blocks are converted into selects, so
   // we don't have to worry about the insertion order and we can just use the
   // builder. At this point we generate the predication tree. There may be
   // duplications since this is a simple recursive scan, but future
   // optimizations will clean it up.
-
   SmallVector<VPValue *, 2> Operands;
   unsigned NumIncoming = Phi->getNumIncomingValues();
+
   for (unsigned In = 0; In < NumIncoming; In++) {
     VPValue *EdgeMask =
       createEdgeMask(Phi->getIncomingBlock(In), Phi->getParent(), Plan);
@@ -8449,9 +8458,9 @@ VPRegionBlock *VPRecipeBuilder::createReplicateRegion(Instruction *Instr,
   return Region;
 }
 
-VPRecipeBase *VPRecipeBuilder::tryToCreateWidenRecipe(Instruction *Instr,
-                                                      VFRange &Range,
-                                                      VPlanPtr &Plan) {
+VPRecipeOrVPValueTy VPRecipeBuilder::tryToCreateWidenRecipe(Instruction *Instr,
+                                                            VFRange &Range,
+                                                            VPlanPtr &Plan) {
   // First, check for specific widening recipes that deal with calls, memory
   // operations, inductions and Phi nodes.
   if (auto *CI = dyn_cast<CallInst>(Instr))
@@ -8623,17 +8632,15 @@ VPlanPtr LoopVectorizationPlanner::buildVPlanWithVPRecipes(
       if (isa<BranchInst>(Instr) || DeadInstructions.count(Instr))
         continue;
 
-      if (auto Recipe =
+      if (auto RecipeOrValue =
               RecipeBuilder.tryToCreateWidenRecipe(Instr, Range, Plan)) {
-
-        // VPBlendRecipes with a single incoming (value, mask) pair are no-ops.
-        // Use the incoming value directly.
-        if (isa<VPBlendRecipe>(Recipe) && Recipe->getNumOperands() <= 2) {
-          Plan->removeVPValueFor(Instr);
-          Plan->addVPValue(Instr, Recipe->getOperand(0));
-          delete Recipe;
+        // If Instr can be simplified to an existing VPValue, use it.
+        if (RecipeOrValue.is<VPValue *>()) {
+          Plan->addVPValue(Instr, RecipeOrValue.get<VPValue *>());
           continue;
         }
+        // Otherwise, add the new recipe.
+        VPRecipeBase *Recipe = RecipeOrValue.get<VPRecipeBase *>();
         for (auto *Def : Recipe->definedValues()) {
           auto *UV = Def->getUnderlyingValue();
           Plan->addVPValue(UV, Def);

llvm/lib/Transforms/Vectorize/VPRecipeBuilder.h

@@ -12,6 +12,7 @@
 #include "LoopVectorizationPlanner.h"
 #include "VPlan.h"
 #include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/PointerUnion.h"
 #include "llvm/IR/IRBuilder.h"
 
 namespace llvm {
@@ -20,6 +21,8 @@ class LoopVectorizationLegality;
 class LoopVectorizationCostModel;
 class TargetLibraryInfo;
 
+using VPRecipeOrVPValueTy = PointerUnion<VPRecipeBase *, VPValue *>;
+
 /// Helper class to create VPRecipies from IR instructions.
 class VPRecipeBuilder {
   /// The loop that we evaluate.
@@ -75,10 +78,11 @@ class VPRecipeBuilder {
   tryToOptimizeInductionTruncate(TruncInst *I, VFRange &Range,
                                  VPlan &Plan) const;
 
-  /// Handle non-loop phi nodes. Currently all such phi nodes are turned into
-  /// a sequence of select instructions as the vectorizer currently performs
-  /// full if-conversion.
-  VPBlendRecipe *tryToBlend(PHINode *Phi, VPlanPtr &Plan);
+  /// Handle non-loop phi nodes. Return a VPValue, if all incoming values match
+  /// or a new VPBlendRecipe otherwise. Currently all such phi nodes are turned
+  /// into a sequence of select instructions as the vectorizer currently
+  /// performs full if-conversion.
+  VPRecipeOrVPValueTy tryToBlend(PHINode *Phi, VPlanPtr &Plan);
 
   /// Handle call instructions. If \p CI can be widened for \p Range.Start,
   /// return a new VPWidenCallRecipe. Range.End may be decreased to ensure same
@@ -99,10 +103,12 @@ class VPRecipeBuilder {
       : OrigLoop(OrigLoop), TLI(TLI), Legal(Legal), CM(CM), PSE(PSE),
         Builder(Builder) {}
 
-  /// Check if a recipe can be create for \p I withing the given VF \p Range.
-  /// If a recipe can be created, return it. Otherwise return nullptr.
-  VPRecipeBase *tryToCreateWidenRecipe(Instruction *Instr, VFRange &Range,
-                                       VPlanPtr &Plan);
+  /// Check if an existing VPValue can be used for \p Instr or a recipe can be
+  /// create for \p I withing the given VF \p Range. If an existing VPValue can
+  /// be used or if a recipe can be created, return it. Otherwise return a
+  /// VPRecipeOrVPValueTy with nullptr.
+  VPRecipeOrVPValueTy tryToCreateWidenRecipe(Instruction *Instr, VFRange &Range,
+                                             VPlanPtr &Plan);
 
   /// Set the recipe created for given ingredient. This operation is a no-op for
   /// ingredients that were not marked using a nullptr entry in the map.

llvm/test/Transforms/LoopVectorize/single-value-blend-phis.ll

@@ -382,3 +382,65 @@ loop.latch:
 exit:
   ret void
 }
+
+; Test case for PR44800.
+define void @duplicated_incoming_blocks_blend(i32 %x, i32* %ptr) {
+; CHECK-LABEL: @duplicated_incoming_blocks_blend(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; CHECK:       vector.ph:
+; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[X:%.*]], i32 0
+; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
+; CHECK:       vector.body:
+; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[TMP0:%.*]] = icmp ugt <2 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x i32> [[VEC_IND]], i32 0
+; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr i32, i32* [[PTR:%.*]], i32 [[TMP1]]
+; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr i32, i32* [[TMP2]], i32 0
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast i32* [[TMP3]] to <2 x i32>*
+; CHECK-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP4]], align 4
+; CHECK-NEXT:    [[INDEX_NEXT]] = add i32 [[INDEX]], 2
+; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp eq i32 [[INDEX_NEXT]], 1000
+; CHECK-NEXT:    br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP10:!llvm.loop !.*]]
+; CHECK:       middle.block:
+; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i32 1000, 1000
+; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
+; CHECK:       scalar.ph:
+; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ 1000, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-NEXT:    br label [[LOOP_HEADER:%.*]]
+; CHECK:       loop.header:
+; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[ADD_I:%.*]], [[LOOP_LATCH:%.*]] ]
+; CHECK-NEXT:    [[C_0:%.*]] = icmp ugt i32 [[IV]], [[X]]
+; CHECK-NEXT:    br i1 [[C_0]], label [[LOOP_LATCH]], label [[LOOP_LATCH]]
+; CHECK:       loop.latch:
+; CHECK-NEXT:    [[P:%.*]] = phi i32 [ [[IV]], [[LOOP_HEADER]] ], [ [[IV]], [[LOOP_HEADER]] ]
+; CHECK-NEXT:    [[GEP_PTR:%.*]] = getelementptr i32, i32* [[PTR]], i32 [[P]]
+; CHECK-NEXT:    store i32 [[P]], i32* [[GEP_PTR]], align 4
+; CHECK-NEXT:    [[ADD_I]] = add nsw i32 [[P]], 1
+; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[ADD_I]], 1000
+; CHECK-NEXT:    br i1 [[CMP]], label [[LOOP_HEADER]], label [[EXIT]], [[LOOP11:!llvm.loop !.*]]
+; CHECK:       exit:
+; CHECK-NEXT:    ret void
+;
+entry:
+  br label %loop.header
+
+loop.header:
+  %iv = phi i32 [ 0 , %entry ], [ %add.i, %loop.latch ]
+  %c.0 = icmp ugt i32 %iv, %x
+  br i1 %c.0, label %loop.latch, label %loop.latch
+
+loop.latch:
+  %p = phi i32 [ %iv, %loop.header ], [ %iv, %loop.header ]
+  %gep.ptr = getelementptr i32, i32* %ptr, i32 %p
+  store i32 %p, i32* %gep.ptr
+  %add.i = add nsw i32 %p, 1
+  %cmp = icmp slt i32 %add.i, 1000
+  br i1 %cmp, label %loop.header, label %exit
+
+exit:
+  ret void
+}