[VPlan] Update wide induction inc recipes to use same step as Wide IV.

Update wide induction increments to use the same step as the corresponding
wide induction. This enables detecting induction increments directly in
VPlan and removes redundant splats.

Author: fhahn

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

@@ -9311,6 +9311,20 @@ LoopVectorizationPlanner::tryToBuildVPlanWithVPRecipes(VFRange &Range) {
          "VPBasicBlock");
   RecipeBuilder.fixHeaderPhis();
 
+  // Update wide induction increments to use the same step as the corresponding
+  // wide induction. This enables detecting induction increments directly in
+  // VPlan and removes redundant splats.
+  for (const auto &[Phi, ID] : Legal->getInductionVars()) {
+    auto *IVInc = cast<Instruction>(
+        Phi->getIncomingValueForBlock(OrigLoop->getLoopLatch()));
+    if (IVInc->getOperand(0) != Phi || IVInc->getOpcode() != Instruction::Add)
+      continue;
+    VPWidenInductionRecipe *WideIV =
+        cast<VPWidenInductionRecipe>(RecipeBuilder.getRecipe(Phi));
+    VPRecipeBase *R = RecipeBuilder.getRecipe(IVInc);
+    R->setOperand(1, WideIV->getStepValue());
+  }
+
   if (auto *UncountableExitingBlock =
           Legal->getUncountableEarlyExitingBlock()) {
     VPlanTransforms::handleUncountableEarlyExit(

llvm/test/Transforms/LoopVectorize/X86/induction-step.ll

@@ -21,16 +21,14 @@ define i16 @wide_add_induction_step_live_in(ptr %dst, i64 %N, i16 %off) {
 ; CHECK-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <4 x i16> [[DOTSPLATINSERT]], <4 x i16> poison, <4 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP2:%.*]] = mul <4 x i16> <i16 0, i16 1, i16 2, i16 3>, [[DOTSPLAT]]
 ; CHECK-NEXT:    [[INDUCTION:%.*]] = add <4 x i16> zeroinitializer, [[TMP2]]
-; CHECK-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <4 x i16> poison, i16 [[O_1]], i64 0
-; CHECK-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <4 x i16> [[BROADCAST_SPLATINSERT1]], <4 x i16> poison, <4 x i32> zeroinitializer
 ; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
 ; CHECK:       vector.body:
 ; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
 ; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <4 x i16> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
 ; CHECK-NEXT:    [[STEP_ADD:%.*]] = add <4 x i16> [[VEC_IND]], [[TMP1]]
 ; CHECK-NEXT:    [[TMP3:%.*]] = add i64 [[INDEX]], 0
-; CHECK-NEXT:    [[TMP4:%.*]] = add <4 x i16> [[VEC_IND]], [[BROADCAST_SPLAT2]]
-; CHECK-NEXT:    [[TMP9:%.*]] = add <4 x i16> [[STEP_ADD]], [[BROADCAST_SPLAT2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = add <4 x i16> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT:    [[TMP9:%.*]] = add <4 x i16> [[STEP_ADD]], [[BROADCAST_SPLAT]]
 ; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i16, ptr [[DST:%.*]], i64 [[TMP3]]
 ; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i16, ptr [[TMP5]], i32 0
 ; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds i16, ptr [[TMP5]], i32 4