[llvm] 6b19a54 - [LV] Bundle partial reductions inside VPExpressionRecipe (#147302)
via llvm-commits
llvm-commits at lists.llvm.org
Thu Oct 23 04:18:59 PDT 2025
Author: Sam Tebbs
Date: 2025-10-23T11:18:55Z
New Revision: 6b19a546aa8d341dfdfa2a3a0a37fb90ac786f92
URL: https://github.com/llvm/llvm-project/commit/6b19a546aa8d341dfdfa2a3a0a37fb90ac786f92
DIFF: https://github.com/llvm/llvm-project/commit/6b19a546aa8d341dfdfa2a3a0a37fb90ac786f92.diff
LOG: [LV] Bundle partial reductions inside VPExpressionRecipe (#147302)
This PR bundles partial reductions inside the VPExpressionRecipe class.
Stacked PRs:
1. https://github.com/llvm/llvm-project/pull/147026
2. https://github.com/llvm/llvm-project/pull/147255
3. https://github.com/llvm/llvm-project/pull/156976
4. https://github.com/llvm/llvm-project/pull/160154
5. -> https://github.com/llvm/llvm-project/pull/147302
6. https://github.com/llvm/llvm-project/pull/162503
7. https://github.com/llvm/llvm-project/pull/147513
Added:
llvm/test/Transforms/LoopVectorize/WebAssembly/partial-reduce-accumulate.ll
Modified:
llvm/include/llvm/Analysis/TargetTransformInfo.h
llvm/lib/Analysis/TargetTransformInfo.cpp
llvm/lib/Transforms/Vectorize/VPlan.h
llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
llvm/lib/Transforms/Vectorize/VPlanUtils.cpp
llvm/test/Transforms/LoopVectorize/AArch64/fully-unrolled-cost.ll
llvm/test/Transforms/LoopVectorize/AArch64/maxbandwidth-regpressure.ll
llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-chained.ll
llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-constant-ops.ll
llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-epilogue.ll
llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll
llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-neon.ll
llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product.ll
llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-interleave.ll
llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-sub.ll
llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce.ll
llvm/test/Transforms/LoopVectorize/AArch64/reg-usage.ll
llvm/test/Transforms/LoopVectorize/AArch64/vplan-printing.ll
llvm/test/Transforms/LoopVectorize/RISCV/partial-reduce-dot-product.ll
Removed:
################################################################################
diff --git a/llvm/include/llvm/Analysis/TargetTransformInfo.h b/llvm/include/llvm/Analysis/TargetTransformInfo.h
index 5d3b233ed6b6a..7b7dc1b46dd80 100644
--- a/llvm/include/llvm/Analysis/TargetTransformInfo.h
+++ b/llvm/include/llvm/Analysis/TargetTransformInfo.h
@@ -227,6 +227,9 @@ class TargetTransformInfo {
/// Get the kind of extension that an instruction represents.
LLVM_ABI static PartialReductionExtendKind
getPartialReductionExtendKind(Instruction *I);
+ /// Get the kind of extension that a cast opcode represents.
+ LLVM_ABI static PartialReductionExtendKind
+ getPartialReductionExtendKind(Instruction::CastOps CastOpc);
/// Construct a TTI object using a type implementing the \c Concept
/// API below.
diff --git a/llvm/lib/Analysis/TargetTransformInfo.cpp b/llvm/lib/Analysis/TargetTransformInfo.cpp
index bf62623099a97..c47a1c1b23a37 100644
--- a/llvm/lib/Analysis/TargetTransformInfo.cpp
+++ b/llvm/lib/Analysis/TargetTransformInfo.cpp
@@ -1001,13 +1001,25 @@ InstructionCost TargetTransformInfo::getShuffleCost(
TargetTransformInfo::PartialReductionExtendKind
TargetTransformInfo::getPartialReductionExtendKind(Instruction *I) {
- if (isa<SExtInst>(I))
- return PR_SignExtend;
- if (isa<ZExtInst>(I))
- return PR_ZeroExtend;
+ if (auto *Cast = dyn_cast<CastInst>(I))
+ return getPartialReductionExtendKind(Cast->getOpcode());
return PR_None;
}
+TargetTransformInfo::PartialReductionExtendKind
+TargetTransformInfo::getPartialReductionExtendKind(
+ Instruction::CastOps CastOpc) {
+ switch (CastOpc) {
+ case Instruction::CastOps::ZExt:
+ return PR_ZeroExtend;
+ case Instruction::CastOps::SExt:
+ return PR_SignExtend;
+ default:
+ return PR_None;
+ }
+ llvm_unreachable("Unhandled cast opcode");
+}
+
TTI::CastContextHint
TargetTransformInfo::getCastContextHint(const Instruction *I) {
if (!I)
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.h b/llvm/lib/Transforms/Vectorize/VPlan.h
index 06bea2fef87aa..a1ad2dbcba8be 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.h
+++ b/llvm/lib/Transforms/Vectorize/VPlan.h
@@ -2712,7 +2712,8 @@ class LLVM_ABI_FOR_TEST VPReductionRecipe : public VPRecipeWithIRFlags {
static inline bool classof(const VPRecipeBase *R) {
return R->getVPDefID() == VPRecipeBase::VPReductionSC ||
- R->getVPDefID() == VPRecipeBase::VPReductionEVLSC;
+ R->getVPDefID() == VPRecipeBase::VPReductionEVLSC ||
+ R->getVPDefID() == VPRecipeBase::VPPartialReductionSC;
}
static inline bool classof(const VPUser *U) {
@@ -2783,7 +2784,10 @@ class VPPartialReductionRecipe : public VPReductionRecipe {
Opcode(Opcode), VFScaleFactor(ScaleFactor) {
[[maybe_unused]] auto *AccumulatorRecipe =
getChainOp()->getDefiningRecipe();
- assert((isa<VPReductionPHIRecipe>(AccumulatorRecipe) ||
+ // When cloning as part of a VPExpressionRecipe the chain op could have
+ // replaced by a temporary VPValue, so it doesn't have a defining recipe.
+ assert((!AccumulatorRecipe ||
+ isa<VPReductionPHIRecipe>(AccumulatorRecipe) ||
isa<VPPartialReductionRecipe>(AccumulatorRecipe)) &&
"Unexpected operand order for partial reduction recipe");
}
@@ -3093,6 +3097,11 @@ class VPExpressionRecipe : public VPSingleDefRecipe {
/// removed before codegen.
void decompose();
+ unsigned getVFScaleFactor() const {
+ auto *PR = dyn_cast<VPPartialReductionRecipe>(ExpressionRecipes.back());
+ return PR ? PR->getVFScaleFactor() : 1;
+ }
+
/// Method for generating code, must not be called as this recipe is abstract.
void execute(VPTransformState &State) override {
llvm_unreachable("recipe must be removed before execute");
diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index 1f1b42bb9c19f..931a5b7582c4e 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -168,6 +168,7 @@ bool VPRecipeBase::mayHaveSideEffects() const {
return cast<VPWidenIntrinsicRecipe>(this)->mayHaveSideEffects();
case VPBlendSC:
case VPReductionEVLSC:
+ case VPPartialReductionSC:
case VPReductionSC:
case VPScalarIVStepsSC:
case VPVectorPointerSC:
@@ -300,14 +301,23 @@ InstructionCost
VPPartialReductionRecipe::computeCost(ElementCount VF,
VPCostContext &Ctx) const {
std::optional<unsigned> Opcode;
- VPValue *Op = getOperand(0);
- VPRecipeBase *OpR = Op->getDefiningRecipe();
-
- // If the partial reduction is predicated, a select will be operand 0
- if (match(getOperand(1), m_Select(m_VPValue(), m_VPValue(Op), m_VPValue()))) {
- OpR = Op->getDefiningRecipe();
+ VPValue *Op = getVecOp();
+ uint64_t MulConst;
+ // If the partial reduction is predicated, a select will be operand 1.
+ // If it isn't predicated and the mul isn't operating on a constant, then it
+ // should have been turned into a VPExpressionRecipe.
+ // FIXME: Replace the entire function with this once all partial reduction
+ // variants are bundled into VPExpressionRecipe.
+ if (!match(Op, m_Select(m_VPValue(), m_VPValue(Op), m_VPValue())) &&
+ !match(Op, m_Mul(m_VPValue(), m_ConstantInt(MulConst)))) {
+ auto *PhiType = Ctx.Types.inferScalarType(getChainOp());
+ auto *InputType = Ctx.Types.inferScalarType(getVecOp());
+ return Ctx.TTI.getPartialReductionCost(getOpcode(), InputType, InputType,
+ PhiType, VF, TTI::PR_None,
+ TTI::PR_None, {}, Ctx.CostKind);
}
+ VPRecipeBase *OpR = Op->getDefiningRecipe();
Type *InputTypeA = nullptr, *InputTypeB = nullptr;
TTI::PartialReductionExtendKind ExtAType = TTI::PR_None,
ExtBType = TTI::PR_None;
@@ -2856,11 +2866,19 @@ InstructionCost VPExpressionRecipe::computeCost(ElementCount VF,
cast<VPReductionRecipe>(ExpressionRecipes.back())->getRecurrenceKind());
switch (ExpressionType) {
case ExpressionTypes::ExtendedReduction: {
- return Ctx.TTI.getExtendedReductionCost(
- Opcode,
- cast<VPWidenCastRecipe>(ExpressionRecipes.front())->getOpcode() ==
- Instruction::ZExt,
- RedTy, SrcVecTy, std::nullopt, Ctx.CostKind);
+ unsigned Opcode = RecurrenceDescriptor::getOpcode(
+ cast<VPReductionRecipe>(ExpressionRecipes[1])->getRecurrenceKind());
+ auto *ExtR = cast<VPWidenCastRecipe>(ExpressionRecipes[0]);
+ return isa<VPPartialReductionRecipe>(ExpressionRecipes.back())
+ ? Ctx.TTI.getPartialReductionCost(
+ Opcode, Ctx.Types.inferScalarType(getOperand(0)), nullptr,
+ RedTy, VF,
+ TargetTransformInfo::getPartialReductionExtendKind(
+ ExtR->getOpcode()),
+ TargetTransformInfo::PR_None, std::nullopt, Ctx.CostKind)
+ : Ctx.TTI.getExtendedReductionCost(
+ Opcode, ExtR->getOpcode() == Instruction::ZExt, RedTy,
+ SrcVecTy, std::nullopt, Ctx.CostKind);
}
case ExpressionTypes::MulAccReduction:
return Ctx.TTI.getMulAccReductionCost(false, Opcode, RedTy, SrcVecTy,
@@ -2871,6 +2889,19 @@ InstructionCost VPExpressionRecipe::computeCost(ElementCount VF,
Opcode = Instruction::Sub;
[[fallthrough]];
case ExpressionTypes::ExtMulAccReduction: {
+ if (isa<VPPartialReductionRecipe>(ExpressionRecipes.back())) {
+ auto *Ext0R = cast<VPWidenCastRecipe>(ExpressionRecipes[0]);
+ auto *Ext1R = cast<VPWidenCastRecipe>(ExpressionRecipes[1]);
+ auto *Mul = cast<VPWidenRecipe>(ExpressionRecipes[2]);
+ return Ctx.TTI.getPartialReductionCost(
+ Opcode, Ctx.Types.inferScalarType(getOperand(0)),
+ Ctx.Types.inferScalarType(getOperand(1)), RedTy, VF,
+ TargetTransformInfo::getPartialReductionExtendKind(
+ Ext0R->getOpcode()),
+ TargetTransformInfo::getPartialReductionExtendKind(
+ Ext1R->getOpcode()),
+ Mul->getOpcode(), Ctx.CostKind);
+ }
return Ctx.TTI.getMulAccReductionCost(
cast<VPWidenCastRecipe>(ExpressionRecipes.front())->getOpcode() ==
Instruction::ZExt,
@@ -2910,12 +2941,13 @@ void VPExpressionRecipe::print(raw_ostream &O, const Twine &Indent,
O << " = ";
auto *Red = cast<VPReductionRecipe>(ExpressionRecipes.back());
unsigned Opcode = RecurrenceDescriptor::getOpcode(Red->getRecurrenceKind());
+ bool IsPartialReduction = isa<VPPartialReductionRecipe>(Red);
switch (ExpressionType) {
case ExpressionTypes::ExtendedReduction: {
getOperand(1)->printAsOperand(O, SlotTracker);
- O << " +";
- O << " reduce." << Instruction::getOpcodeName(Opcode) << " (";
+ O << " + " << (IsPartialReduction ? "partial." : "") << "reduce.";
+ O << Instruction::getOpcodeName(Opcode) << " (";
getOperand(0)->printAsOperand(O, SlotTracker);
Red->printFlags(O);
@@ -2931,8 +2963,8 @@ void VPExpressionRecipe::print(raw_ostream &O, const Twine &Indent,
}
case ExpressionTypes::ExtNegatedMulAccReduction: {
getOperand(getNumOperands() - 1)->printAsOperand(O, SlotTracker);
- O << " + reduce."
- << Instruction::getOpcodeName(
+ O << " + " << (IsPartialReduction ? "partial." : "") << "reduce.";
+ O << Instruction::getOpcodeName(
RecurrenceDescriptor::getOpcode(Red->getRecurrenceKind()))
<< " (sub (0, mul";
auto *Mul = cast<VPWidenRecipe>(ExpressionRecipes[2]);
@@ -2956,9 +2988,8 @@ void VPExpressionRecipe::print(raw_ostream &O, const Twine &Indent,
case ExpressionTypes::MulAccReduction:
case ExpressionTypes::ExtMulAccReduction: {
getOperand(getNumOperands() - 1)->printAsOperand(O, SlotTracker);
- O << " + ";
- O << "reduce."
- << Instruction::getOpcodeName(
+ O << " + " << (IsPartialReduction ? "partial." : "") << "reduce.";
+ O << Instruction::getOpcodeName(
RecurrenceDescriptor::getOpcode(Red->getRecurrenceKind()))
<< " (";
O << "mul";
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index f5a3af4e983f6..3e85e6ff514ef 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -3519,18 +3519,31 @@ tryToMatchAndCreateExtendedReduction(VPReductionRecipe *Red, VPCostContext &Ctx,
VPValue *VecOp = Red->getVecOp();
// Clamp the range if using extended-reduction is profitable.
- auto IsExtendedRedValidAndClampRange = [&](unsigned Opcode, bool isZExt,
- Type *SrcTy) -> bool {
+ auto IsExtendedRedValidAndClampRange =
+ [&](unsigned Opcode, Instruction::CastOps ExtOpc, Type *SrcTy) -> bool {
return LoopVectorizationPlanner::getDecisionAndClampRange(
[&](ElementCount VF) {
auto *SrcVecTy = cast<VectorType>(toVectorTy(SrcTy, VF));
TTI::TargetCostKind CostKind = TTI::TCK_RecipThroughput;
- InstructionCost ExtRedCost = Ctx.TTI.getExtendedReductionCost(
- Opcode, isZExt, RedTy, SrcVecTy, Red->getFastMathFlags(),
- CostKind);
+
+ InstructionCost ExtRedCost;
InstructionCost ExtCost =
cast<VPWidenCastRecipe>(VecOp)->computeCost(VF, Ctx);
InstructionCost RedCost = Red->computeCost(VF, Ctx);
+
+ if (isa<VPPartialReductionRecipe>(Red)) {
+ TargetTransformInfo::PartialReductionExtendKind ExtKind =
+ TargetTransformInfo::getPartialReductionExtendKind(ExtOpc);
+ // FIXME: Move partial reduction creation, costing and clamping
+ // here from LoopVectorize.cpp.
+ ExtRedCost = Ctx.TTI.getPartialReductionCost(
+ Opcode, SrcTy, nullptr, RedTy, VF, ExtKind,
+ llvm::TargetTransformInfo::PR_None, std::nullopt, Ctx.CostKind);
+ } else {
+ ExtRedCost = Ctx.TTI.getExtendedReductionCost(
+ Opcode, ExtOpc == Instruction::CastOps::ZExt, RedTy, SrcVecTy,
+ Red->getFastMathFlags(), CostKind);
+ }
return ExtRedCost.isValid() && ExtRedCost < ExtCost + RedCost;
},
Range);
@@ -3541,8 +3554,7 @@ tryToMatchAndCreateExtendedReduction(VPReductionRecipe *Red, VPCostContext &Ctx,
if (match(VecOp, m_ZExtOrSExt(m_VPValue(A))) &&
IsExtendedRedValidAndClampRange(
RecurrenceDescriptor::getOpcode(Red->getRecurrenceKind()),
- cast<VPWidenCastRecipe>(VecOp)->getOpcode() ==
- Instruction::CastOps::ZExt,
+ cast<VPWidenCastRecipe>(VecOp)->getOpcode(),
Ctx.Types.inferScalarType(A)))
return new VPExpressionRecipe(cast<VPWidenCastRecipe>(VecOp), Red);
@@ -3560,6 +3572,8 @@ tryToMatchAndCreateExtendedReduction(VPReductionRecipe *Red, VPCostContext &Ctx,
static VPExpressionRecipe *
tryToMatchAndCreateMulAccumulateReduction(VPReductionRecipe *Red,
VPCostContext &Ctx, VFRange &Range) {
+ bool IsPartialReduction = isa<VPPartialReductionRecipe>(Red);
+
unsigned Opcode = RecurrenceDescriptor::getOpcode(Red->getRecurrenceKind());
if (Opcode != Instruction::Add && Opcode != Instruction::Sub)
return nullptr;
@@ -3568,16 +3582,41 @@ tryToMatchAndCreateMulAccumulateReduction(VPReductionRecipe *Red,
// Clamp the range if using multiply-accumulate-reduction is profitable.
auto IsMulAccValidAndClampRange =
- [&](bool isZExt, VPWidenRecipe *Mul, VPWidenCastRecipe *Ext0,
- VPWidenCastRecipe *Ext1, VPWidenCastRecipe *OuterExt) -> bool {
+ [&](VPWidenRecipe *Mul, VPWidenCastRecipe *Ext0, VPWidenCastRecipe *Ext1,
+ VPWidenCastRecipe *OuterExt) -> bool {
return LoopVectorizationPlanner::getDecisionAndClampRange(
[&](ElementCount VF) {
TTI::TargetCostKind CostKind = TTI::TCK_RecipThroughput;
Type *SrcTy =
Ext0 ? Ctx.Types.inferScalarType(Ext0->getOperand(0)) : RedTy;
- auto *SrcVecTy = cast<VectorType>(toVectorTy(SrcTy, VF));
- InstructionCost MulAccCost = Ctx.TTI.getMulAccReductionCost(
- isZExt, Opcode, RedTy, SrcVecTy, CostKind);
+ InstructionCost MulAccCost;
+
+ if (IsPartialReduction) {
+ Type *SrcTy2 =
+ Ext1 ? Ctx.Types.inferScalarType(Ext1->getOperand(0)) : nullptr;
+ // FIXME: Move partial reduction creation, costing and clamping
+ // here from LoopVectorize.cpp.
+ MulAccCost = Ctx.TTI.getPartialReductionCost(
+ Opcode, SrcTy, SrcTy2, RedTy, VF,
+ Ext0 ? TargetTransformInfo::getPartialReductionExtendKind(
+ Ext0->getOpcode())
+ : TargetTransformInfo::PR_None,
+ Ext1 ? TargetTransformInfo::getPartialReductionExtendKind(
+ Ext1->getOpcode())
+ : TargetTransformInfo::PR_None,
+ Mul->getOpcode(), CostKind);
+ } else {
+ // Only partial reductions support mixed extends at the moment.
+ if (Ext0 && Ext1 && Ext0->getOpcode() != Ext1->getOpcode())
+ return false;
+
+ bool IsZExt =
+ !Ext0 || Ext0->getOpcode() == Instruction::CastOps::ZExt;
+ auto *SrcVecTy = cast<VectorType>(toVectorTy(SrcTy, VF));
+ MulAccCost = Ctx.TTI.getMulAccReductionCost(IsZExt, Opcode, RedTy,
+ SrcVecTy, CostKind);
+ }
+
InstructionCost MulCost = Mul->computeCost(VF, Ctx);
InstructionCost RedCost = Red->computeCost(VF, Ctx);
InstructionCost ExtCost = 0;
@@ -3611,14 +3650,10 @@ tryToMatchAndCreateMulAccumulateReduction(VPReductionRecipe *Red,
dyn_cast_if_present<VPWidenCastRecipe>(B->getDefiningRecipe());
auto *Mul = cast<VPWidenRecipe>(VecOp->getDefiningRecipe());
- // Match reduce.add(mul(ext, ext)).
- if (RecipeA && RecipeB &&
- (RecipeA->getOpcode() == RecipeB->getOpcode() || A == B) &&
- match(RecipeA, m_ZExtOrSExt(m_VPValue())) &&
+ // Match reduce.add/sub(mul(ext, ext)).
+ if (RecipeA && RecipeB && match(RecipeA, m_ZExtOrSExt(m_VPValue())) &&
match(RecipeB, m_ZExtOrSExt(m_VPValue())) &&
- IsMulAccValidAndClampRange(RecipeA->getOpcode() ==
- Instruction::CastOps::ZExt,
- Mul, RecipeA, RecipeB, nullptr)) {
+ IsMulAccValidAndClampRange(Mul, RecipeA, RecipeB, nullptr)) {
if (Sub)
return new VPExpressionRecipe(RecipeA, RecipeB, Mul,
cast<VPWidenRecipe>(Sub), Red);
@@ -3626,8 +3661,7 @@ tryToMatchAndCreateMulAccumulateReduction(VPReductionRecipe *Red,
}
// Match reduce.add(mul).
// TODO: Add an expression type for this variant with a negated mul
- if (!Sub &&
- IsMulAccValidAndClampRange(true, Mul, nullptr, nullptr, nullptr))
+ if (!Sub && IsMulAccValidAndClampRange(Mul, nullptr, nullptr, nullptr))
return new VPExpressionRecipe(Mul, Red);
}
// TODO: Add an expression type for negated versions of other expression
@@ -3647,9 +3681,7 @@ tryToMatchAndCreateMulAccumulateReduction(VPReductionRecipe *Red,
cast<VPWidenCastRecipe>(Mul->getOperand(1)->getDefiningRecipe());
if ((Ext->getOpcode() == Ext0->getOpcode() || Ext0 == Ext1) &&
Ext0->getOpcode() == Ext1->getOpcode() &&
- IsMulAccValidAndClampRange(Ext0->getOpcode() ==
- Instruction::CastOps::ZExt,
- Mul, Ext0, Ext1, Ext)) {
+ IsMulAccValidAndClampRange(Mul, Ext0, Ext1, Ext) && Mul->hasOneUse()) {
auto *NewExt0 = new VPWidenCastRecipe(
Ext0->getOpcode(), Ext0->getOperand(0), Ext->getResultType(), *Ext0,
*Ext0, Ext0->getDebugLoc());
diff --git a/llvm/lib/Transforms/Vectorize/VPlanUtils.cpp b/llvm/lib/Transforms/Vectorize/VPlanUtils.cpp
index 32e4b8872b1df..06c3d75e5708c 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanUtils.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanUtils.cpp
@@ -151,6 +151,8 @@ unsigned vputils::getVFScaleFactor(VPRecipeBase *R) {
return RR->getVFScaleFactor();
if (auto *RR = dyn_cast<VPPartialReductionRecipe>(R))
return RR->getVFScaleFactor();
+ if (auto *ER = dyn_cast<VPExpressionRecipe>(R))
+ return ER->getVFScaleFactor();
assert(
(!isa<VPInstruction>(R) || cast<VPInstruction>(R)->getOpcode() !=
VPInstruction::ReductionStartVector) &&
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/fully-unrolled-cost.ll b/llvm/test/Transforms/LoopVectorize/AArch64/fully-unrolled-cost.ll
index 27ca4143b5be5..199203a9f5cb0 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/fully-unrolled-cost.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/fully-unrolled-cost.ll
@@ -86,7 +86,8 @@ define i64 @test_two_ivs(ptr %a, ptr %b, i64 %start) #0 {
; CHECK-NEXT: Cost of 0 for VF 16: induction instruction %i.iv = phi i64 [ 0, %entry ], [ %i.iv.next, %for.body ]
; CHECK-NEXT: Cost of 0 for VF 16: induction instruction %j.iv = phi i64 [ %start, %entry ], [ %j.iv.next, %for.body ]
; CHECK-NEXT: Cost of 0 for VF 16: EMIT vp<{{.+}}> = CANONICAL-INDUCTION ir<0>, vp<%index.next>
-; CHECK: Cost for VF 16: 41
+; CHECK: Cost of 1 for VF 16: EXPRESSION vp<%11> = ir<%sum> + partial.reduce.add (mul nuw nsw (ir<%1> zext to i64), (ir<%0> zext to i64))
+; CHECK: Cost for VF 16: 3
; CHECK: LV: Selecting VF: 16
entry:
br label %for.body
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/maxbandwidth-regpressure.ll b/llvm/test/Transforms/LoopVectorize/AArch64/maxbandwidth-regpressure.ll
index 2d1543185098f..8109d0683fe71 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/maxbandwidth-regpressure.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/maxbandwidth-regpressure.ll
@@ -6,8 +6,7 @@ target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
target triple = "aarch64-none-unknown-elf"
define i32 @dotp(ptr %a, ptr %b) #0 {
-; CHECK-REGS-VP-NOT: LV(REG): Not considering vector loop of width vscale x 16 because it uses too many registers
-; CHECK-REGS-VP: LV: Selecting VF: vscale x 8.
+; CHECK-REGS-VP: LV: Selecting VF: vscale x 16.
;
; CHECK-NOREGS-VP: LV(REG): Not considering vector loop of width vscale x 8 because it uses too many registers
; CHECK-NOREGS-VP: LV(REG): Not considering vector loop of width vscale x 16 because it uses too many registers
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-chained.ll b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-chained.ll
index 3dfa6df3313a5..287226f14b753 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-chained.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-chained.ll
@@ -31,9 +31,9 @@ define i32 @chained_partial_reduce_add_sub(ptr %a, ptr %b, ptr %c, i32 %N) #0 {
; CHECK-NEON-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP3]], align 1
; CHECK-NEON-NEXT: [[TMP7:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-NEON-NEXT: [[TMP8:%.*]] = sext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
-; CHECK-NEON-NEXT: [[TMP9:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-NEON-NEXT: [[TMP10:%.*]] = mul nsw <16 x i32> [[TMP7]], [[TMP8]]
; CHECK-NEON-NEXT: [[PARTIAL_REDUCE:%.*]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP10]])
+; CHECK-NEON-NEXT: [[TMP9:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-NEON-NEXT: [[TMP12:%.*]] = mul nsw <16 x i32> [[TMP7]], [[TMP9]]
; CHECK-NEON-NEXT: [[TMP13:%.*]] = sub <16 x i32> zeroinitializer, [[TMP12]]
; CHECK-NEON-NEXT: [[PARTIAL_REDUCE3]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[PARTIAL_REDUCE]], <16 x i32> [[TMP13]])
@@ -52,37 +52,34 @@ define i32 @chained_partial_reduce_add_sub(ptr %a, ptr %b, ptr %c, i32 %N) #0 {
; CHECK-SVE-NEXT: [[CMP28_NOT:%.*]] = icmp ult i32 [[N]], 2
; CHECK-SVE-NEXT: [[DIV27:%.*]] = lshr i32 [[N]], 1
; CHECK-SVE-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext nneg i32 [[DIV27]] to i64
-; CHECK-SVE-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-SVE-NEXT: [[TMP1:%.*]] = shl nuw nsw i64 [[TMP0]], 2
-; CHECK-SVE-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[WIDE_TRIP_COUNT]], [[TMP1]]
+; CHECK-SVE-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[WIDE_TRIP_COUNT]], 16
; CHECK-SVE-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK-SVE: vector.ph:
-; CHECK-SVE-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-SVE-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 4
-; CHECK-SVE-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[WIDE_TRIP_COUNT]], [[TMP3]]
+; CHECK-SVE-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[WIDE_TRIP_COUNT]], 16
; CHECK-SVE-NEXT: [[N_VEC:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], [[N_MOD_VF]]
; CHECK-SVE-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-SVE: vector.body:
; CHECK-SVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-SVE-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP19:%.*]], [[VECTOR_BODY]] ]
+; CHECK-SVE-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE3:%.*]], [[VECTOR_BODY]] ]
; CHECK-SVE-NEXT: [[TMP7:%.*]] = getelementptr inbounds nuw i8, ptr [[A]], i64 [[INDEX]]
; CHECK-SVE-NEXT: [[TMP8:%.*]] = getelementptr inbounds nuw i8, ptr [[B]], i64 [[INDEX]]
; CHECK-SVE-NEXT: [[TMP9:%.*]] = getelementptr inbounds nuw i8, ptr [[C]], i64 [[INDEX]]
-; CHECK-SVE-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i8>, ptr [[TMP7]], align 1
-; CHECK-SVE-NEXT: [[WIDE_LOAD1:%.*]] = load <vscale x 4 x i8>, ptr [[TMP8]], align 1
-; CHECK-SVE-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 4 x i8>, ptr [[TMP9]], align 1
-; CHECK-SVE-NEXT: [[TMP13:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD]] to <vscale x 4 x i32>
-; CHECK-SVE-NEXT: [[TMP14:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD1]] to <vscale x 4 x i32>
-; CHECK-SVE-NEXT: [[TMP15:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD2]] to <vscale x 4 x i32>
-; CHECK-SVE-NEXT: [[TMP16:%.*]] = mul nsw <vscale x 4 x i32> [[TMP13]], [[TMP14]]
-; CHECK-SVE-NEXT: [[TMP17:%.*]] = add <vscale x 4 x i32> [[VEC_PHI]], [[TMP16]]
-; CHECK-SVE-NEXT: [[TMP18:%.*]] = mul nsw <vscale x 4 x i32> [[TMP13]], [[TMP15]]
-; CHECK-SVE-NEXT: [[TMP19]] = sub <vscale x 4 x i32> [[TMP17]], [[TMP18]]
-; CHECK-SVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
+; CHECK-SVE-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
+; CHECK-SVE-NEXT: [[WIDE_LOAD1:%.*]] = load <16 x i8>, ptr [[TMP8]], align 1
+; CHECK-SVE-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP9]], align 1
+; CHECK-SVE-NEXT: [[TMP3:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
+; CHECK-SVE-NEXT: [[TMP4:%.*]] = sext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
+; CHECK-SVE-NEXT: [[TMP5:%.*]] = mul nsw <16 x i32> [[TMP3]], [[TMP4]]
+; CHECK-SVE-NEXT: [[PARTIAL_REDUCE:%.*]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP5]])
+; CHECK-SVE-NEXT: [[TMP6:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
+; CHECK-SVE-NEXT: [[TMP11:%.*]] = mul nsw <16 x i32> [[TMP3]], [[TMP6]]
+; CHECK-SVE-NEXT: [[TMP12:%.*]] = sub <16 x i32> zeroinitializer, [[TMP11]]
+; CHECK-SVE-NEXT: [[PARTIAL_REDUCE3]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[PARTIAL_REDUCE]], <16 x i32> [[TMP12]])
+; CHECK-SVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-SVE-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-SVE-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK-SVE: middle.block:
-; CHECK-SVE-NEXT: [[TMP21:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[TMP19]])
+; CHECK-SVE-NEXT: [[TMP10:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE3]])
; CHECK-SVE-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC]]
; CHECK-SVE-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]]
; CHECK-SVE: scalar.ph:
@@ -114,10 +111,10 @@ define i32 @chained_partial_reduce_add_sub(ptr %a, ptr %b, ptr %c, i32 %N) #0 {
; CHECK-SVE-MAXBW-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP9]], align 1
; CHECK-SVE-MAXBW-NEXT: [[TMP13:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
; CHECK-SVE-MAXBW-NEXT: [[TMP14:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD1]] to <vscale x 8 x i32>
-; CHECK-SVE-MAXBW-NEXT: [[TMP15:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
; CHECK-SVE-MAXBW-NEXT: [[TMP16:%.*]] = mul nsw <vscale x 8 x i32> [[TMP13]], [[TMP14]]
; CHECK-SVE-MAXBW-NEXT: [[PARTIAL_REDUCE:%.*]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI]], <vscale x 8 x i32> [[TMP16]])
-; CHECK-SVE-MAXBW-NEXT: [[TMP17:%.*]] = mul nsw <vscale x 8 x i32> [[TMP13]], [[TMP15]]
+; CHECK-SVE-MAXBW-NEXT: [[TMP11:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
+; CHECK-SVE-MAXBW-NEXT: [[TMP17:%.*]] = mul nsw <vscale x 8 x i32> [[TMP13]], [[TMP11]]
; CHECK-SVE-MAXBW-NEXT: [[TMP18:%.*]] = sub <vscale x 8 x i32> zeroinitializer, [[TMP17]]
; CHECK-SVE-MAXBW-NEXT: [[PARTIAL_REDUCE3]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[PARTIAL_REDUCE]], <vscale x 8 x i32> [[TMP18]])
; CHECK-SVE-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
@@ -184,9 +181,9 @@ define i32 @chained_partial_reduce_add_add(ptr %a, ptr %b, ptr %c, i32 %N) #0 {
; CHECK-NEON-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP3]], align 1
; CHECK-NEON-NEXT: [[TMP7:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-NEON-NEXT: [[TMP8:%.*]] = sext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
-; CHECK-NEON-NEXT: [[TMP9:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-NEON-NEXT: [[TMP10:%.*]] = mul nsw <16 x i32> [[TMP7]], [[TMP8]]
; CHECK-NEON-NEXT: [[PARTIAL_REDUCE:%.*]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP10]])
+; CHECK-NEON-NEXT: [[TMP9:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-NEON-NEXT: [[TMP11:%.*]] = mul nsw <16 x i32> [[TMP7]], [[TMP9]]
; CHECK-NEON-NEXT: [[PARTIAL_REDUCE3]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[PARTIAL_REDUCE]], <16 x i32> [[TMP11]])
; CHECK-NEON-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
@@ -221,9 +218,9 @@ define i32 @chained_partial_reduce_add_add(ptr %a, ptr %b, ptr %c, i32 %N) #0 {
; CHECK-SVE-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP9]], align 1
; CHECK-SVE-NEXT: [[TMP3:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-SVE-NEXT: [[TMP4:%.*]] = sext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
-; CHECK-SVE-NEXT: [[TMP5:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-SVE-NEXT: [[TMP6:%.*]] = mul nsw <16 x i32> [[TMP3]], [[TMP4]]
; CHECK-SVE-NEXT: [[PARTIAL_REDUCE:%.*]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP6]])
+; CHECK-SVE-NEXT: [[TMP5:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-SVE-NEXT: [[TMP10:%.*]] = mul nsw <16 x i32> [[TMP3]], [[TMP5]]
; CHECK-SVE-NEXT: [[PARTIAL_REDUCE3]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[PARTIAL_REDUCE]], <16 x i32> [[TMP10]])
; CHECK-SVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
@@ -262,10 +259,10 @@ define i32 @chained_partial_reduce_add_add(ptr %a, ptr %b, ptr %c, i32 %N) #0 {
; CHECK-SVE-MAXBW-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP9]], align 1
; CHECK-SVE-MAXBW-NEXT: [[TMP13:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
; CHECK-SVE-MAXBW-NEXT: [[TMP14:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD1]] to <vscale x 8 x i32>
-; CHECK-SVE-MAXBW-NEXT: [[TMP15:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
; CHECK-SVE-MAXBW-NEXT: [[TMP16:%.*]] = mul nsw <vscale x 8 x i32> [[TMP13]], [[TMP14]]
; CHECK-SVE-MAXBW-NEXT: [[PARTIAL_REDUCE:%.*]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI]], <vscale x 8 x i32> [[TMP16]])
-; CHECK-SVE-MAXBW-NEXT: [[TMP17:%.*]] = mul nsw <vscale x 8 x i32> [[TMP13]], [[TMP15]]
+; CHECK-SVE-MAXBW-NEXT: [[TMP11:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
+; CHECK-SVE-MAXBW-NEXT: [[TMP17:%.*]] = mul nsw <vscale x 8 x i32> [[TMP13]], [[TMP11]]
; CHECK-SVE-MAXBW-NEXT: [[PARTIAL_REDUCE3]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[PARTIAL_REDUCE]], <vscale x 8 x i32> [[TMP17]])
; CHECK-SVE-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
; CHECK-SVE-MAXBW-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
@@ -331,11 +328,11 @@ define i32 @chained_partial_reduce_sub_add(ptr %a, ptr %b, ptr %c, i32 %N) #0 {
; CHECK-NEON-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP3]], align 1
; CHECK-NEON-NEXT: [[TMP7:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-NEON-NEXT: [[TMP8:%.*]] = sext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
-; CHECK-NEON-NEXT: [[TMP9:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-NEON-NEXT: [[TMP10:%.*]] = mul nsw <16 x i32> [[TMP7]], [[TMP8]]
; CHECK-NEON-NEXT: [[TMP11:%.*]] = sub nsw <16 x i32> zeroinitializer, [[TMP10]]
; CHECK-NEON-NEXT: [[PARTIAL_REDUCE:%.*]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP11]])
-; CHECK-NEON-NEXT: [[TMP12:%.*]] = mul nsw <16 x i32> [[TMP7]], [[TMP9]]
+; CHECK-NEON-NEXT: [[TMP13:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
+; CHECK-NEON-NEXT: [[TMP12:%.*]] = mul nsw <16 x i32> [[TMP7]], [[TMP13]]
; CHECK-NEON-NEXT: [[PARTIAL_REDUCE3]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[PARTIAL_REDUCE]], <16 x i32> [[TMP12]])
; CHECK-NEON-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-NEON-NEXT: [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
@@ -352,37 +349,34 @@ define i32 @chained_partial_reduce_sub_add(ptr %a, ptr %b, ptr %c, i32 %N) #0 {
; CHECK-SVE-NEXT: [[CMP28_NOT:%.*]] = icmp ult i32 [[N]], 2
; CHECK-SVE-NEXT: [[DIV27:%.*]] = lshr i32 [[N]], 1
; CHECK-SVE-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext nneg i32 [[DIV27]] to i64
-; CHECK-SVE-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-SVE-NEXT: [[TMP1:%.*]] = shl nuw nsw i64 [[TMP0]], 2
-; CHECK-SVE-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[WIDE_TRIP_COUNT]], [[TMP1]]
+; CHECK-SVE-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[WIDE_TRIP_COUNT]], 16
; CHECK-SVE-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK-SVE: vector.ph:
-; CHECK-SVE-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-SVE-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 4
-; CHECK-SVE-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[WIDE_TRIP_COUNT]], [[TMP3]]
+; CHECK-SVE-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[WIDE_TRIP_COUNT]], 16
; CHECK-SVE-NEXT: [[N_VEC:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], [[N_MOD_VF]]
; CHECK-SVE-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-SVE: vector.body:
; CHECK-SVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-SVE-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP19:%.*]], [[VECTOR_BODY]] ]
+; CHECK-SVE-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE3:%.*]], [[VECTOR_BODY]] ]
; CHECK-SVE-NEXT: [[TMP7:%.*]] = getelementptr inbounds nuw i8, ptr [[A]], i64 [[INDEX]]
; CHECK-SVE-NEXT: [[TMP8:%.*]] = getelementptr inbounds nuw i8, ptr [[B]], i64 [[INDEX]]
; CHECK-SVE-NEXT: [[TMP9:%.*]] = getelementptr inbounds nuw i8, ptr [[C]], i64 [[INDEX]]
-; CHECK-SVE-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i8>, ptr [[TMP7]], align 1
-; CHECK-SVE-NEXT: [[WIDE_LOAD1:%.*]] = load <vscale x 4 x i8>, ptr [[TMP8]], align 1
-; CHECK-SVE-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 4 x i8>, ptr [[TMP9]], align 1
-; CHECK-SVE-NEXT: [[TMP13:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD]] to <vscale x 4 x i32>
-; CHECK-SVE-NEXT: [[TMP14:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD1]] to <vscale x 4 x i32>
-; CHECK-SVE-NEXT: [[TMP15:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD2]] to <vscale x 4 x i32>
-; CHECK-SVE-NEXT: [[TMP16:%.*]] = mul nsw <vscale x 4 x i32> [[TMP13]], [[TMP14]]
-; CHECK-SVE-NEXT: [[TMP17:%.*]] = sub <vscale x 4 x i32> [[VEC_PHI]], [[TMP16]]
-; CHECK-SVE-NEXT: [[TMP18:%.*]] = mul nsw <vscale x 4 x i32> [[TMP13]], [[TMP15]]
-; CHECK-SVE-NEXT: [[TMP19]] = add <vscale x 4 x i32> [[TMP17]], [[TMP18]]
-; CHECK-SVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
+; CHECK-SVE-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
+; CHECK-SVE-NEXT: [[WIDE_LOAD1:%.*]] = load <16 x i8>, ptr [[TMP8]], align 1
+; CHECK-SVE-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP9]], align 1
+; CHECK-SVE-NEXT: [[TMP3:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
+; CHECK-SVE-NEXT: [[TMP4:%.*]] = sext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
+; CHECK-SVE-NEXT: [[TMP5:%.*]] = mul nsw <16 x i32> [[TMP3]], [[TMP4]]
+; CHECK-SVE-NEXT: [[TMP6:%.*]] = sub nsw <16 x i32> zeroinitializer, [[TMP5]]
+; CHECK-SVE-NEXT: [[PARTIAL_REDUCE:%.*]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP6]])
+; CHECK-SVE-NEXT: [[TMP11:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
+; CHECK-SVE-NEXT: [[TMP12:%.*]] = mul nsw <16 x i32> [[TMP3]], [[TMP11]]
+; CHECK-SVE-NEXT: [[PARTIAL_REDUCE3]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[PARTIAL_REDUCE]], <16 x i32> [[TMP12]])
+; CHECK-SVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-SVE-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-SVE-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
; CHECK-SVE: middle.block:
-; CHECK-SVE-NEXT: [[TMP21:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[TMP19]])
+; CHECK-SVE-NEXT: [[TMP10:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE3]])
; CHECK-SVE-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC]]
; CHECK-SVE-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]]
; CHECK-SVE: scalar.ph:
@@ -414,11 +408,11 @@ define i32 @chained_partial_reduce_sub_add(ptr %a, ptr %b, ptr %c, i32 %N) #0 {
; CHECK-SVE-MAXBW-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP9]], align 1
; CHECK-SVE-MAXBW-NEXT: [[TMP13:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
; CHECK-SVE-MAXBW-NEXT: [[TMP14:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD1]] to <vscale x 8 x i32>
-; CHECK-SVE-MAXBW-NEXT: [[TMP15:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
; CHECK-SVE-MAXBW-NEXT: [[TMP16:%.*]] = mul nsw <vscale x 8 x i32> [[TMP13]], [[TMP14]]
; CHECK-SVE-MAXBW-NEXT: [[TMP17:%.*]] = sub nsw <vscale x 8 x i32> zeroinitializer, [[TMP16]]
; CHECK-SVE-MAXBW-NEXT: [[PARTIAL_REDUCE:%.*]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI]], <vscale x 8 x i32> [[TMP17]])
-; CHECK-SVE-MAXBW-NEXT: [[TMP18:%.*]] = mul nsw <vscale x 8 x i32> [[TMP13]], [[TMP15]]
+; CHECK-SVE-MAXBW-NEXT: [[TMP12:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
+; CHECK-SVE-MAXBW-NEXT: [[TMP18:%.*]] = mul nsw <vscale x 8 x i32> [[TMP13]], [[TMP12]]
; CHECK-SVE-MAXBW-NEXT: [[PARTIAL_REDUCE3]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[PARTIAL_REDUCE]], <vscale x 8 x i32> [[TMP18]])
; CHECK-SVE-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
; CHECK-SVE-MAXBW-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
@@ -486,11 +480,11 @@ define i32 @chained_partial_reduce_sub_sub(ptr %a, ptr %b, ptr %c, i32 %N) #0 {
; CHECK-NEON-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP3]], align 1
; CHECK-NEON-NEXT: [[TMP7:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-NEON-NEXT: [[TMP8:%.*]] = sext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
-; CHECK-NEON-NEXT: [[TMP9:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-NEON-NEXT: [[TMP10:%.*]] = mul nsw <16 x i32> [[TMP7]], [[TMP8]]
; CHECK-NEON-NEXT: [[TMP11:%.*]] = sub nsw <16 x i32> zeroinitializer, [[TMP10]]
; CHECK-NEON-NEXT: [[PARTIAL_REDUCE:%.*]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP11]])
-; CHECK-NEON-NEXT: [[TMP12:%.*]] = mul nsw <16 x i32> [[TMP7]], [[TMP9]]
+; CHECK-NEON-NEXT: [[TMP16:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
+; CHECK-NEON-NEXT: [[TMP12:%.*]] = mul nsw <16 x i32> [[TMP7]], [[TMP16]]
; CHECK-NEON-NEXT: [[TMP13:%.*]] = sub <16 x i32> zeroinitializer, [[TMP12]]
; CHECK-NEON-NEXT: [[PARTIAL_REDUCE3]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[PARTIAL_REDUCE]], <16 x i32> [[TMP13]])
; CHECK-NEON-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
@@ -508,37 +502,35 @@ define i32 @chained_partial_reduce_sub_sub(ptr %a, ptr %b, ptr %c, i32 %N) #0 {
; CHECK-SVE-NEXT: [[CMP28_NOT:%.*]] = icmp ult i32 [[N]], 2
; CHECK-SVE-NEXT: [[DIV27:%.*]] = lshr i32 [[N]], 1
; CHECK-SVE-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext nneg i32 [[DIV27]] to i64
-; CHECK-SVE-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-SVE-NEXT: [[TMP1:%.*]] = shl nuw nsw i64 [[TMP0]], 2
-; CHECK-SVE-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[WIDE_TRIP_COUNT]], [[TMP1]]
+; CHECK-SVE-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[WIDE_TRIP_COUNT]], 16
; CHECK-SVE-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK-SVE: vector.ph:
-; CHECK-SVE-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-SVE-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 4
-; CHECK-SVE-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[WIDE_TRIP_COUNT]], [[TMP3]]
+; CHECK-SVE-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[WIDE_TRIP_COUNT]], 16
; CHECK-SVE-NEXT: [[N_VEC:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], [[N_MOD_VF]]
; CHECK-SVE-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-SVE: vector.body:
; CHECK-SVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-SVE-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP19:%.*]], [[VECTOR_BODY]] ]
+; CHECK-SVE-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE3:%.*]], [[VECTOR_BODY]] ]
; CHECK-SVE-NEXT: [[TMP7:%.*]] = getelementptr inbounds nuw i8, ptr [[A]], i64 [[INDEX]]
; CHECK-SVE-NEXT: [[TMP8:%.*]] = getelementptr inbounds nuw i8, ptr [[B]], i64 [[INDEX]]
; CHECK-SVE-NEXT: [[TMP9:%.*]] = getelementptr inbounds nuw i8, ptr [[C]], i64 [[INDEX]]
-; CHECK-SVE-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i8>, ptr [[TMP7]], align 1
-; CHECK-SVE-NEXT: [[WIDE_LOAD1:%.*]] = load <vscale x 4 x i8>, ptr [[TMP8]], align 1
-; CHECK-SVE-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 4 x i8>, ptr [[TMP9]], align 1
-; CHECK-SVE-NEXT: [[TMP13:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD]] to <vscale x 4 x i32>
-; CHECK-SVE-NEXT: [[TMP14:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD1]] to <vscale x 4 x i32>
-; CHECK-SVE-NEXT: [[TMP15:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD2]] to <vscale x 4 x i32>
-; CHECK-SVE-NEXT: [[TMP16:%.*]] = mul nsw <vscale x 4 x i32> [[TMP13]], [[TMP14]]
-; CHECK-SVE-NEXT: [[TMP17:%.*]] = sub <vscale x 4 x i32> [[VEC_PHI]], [[TMP16]]
-; CHECK-SVE-NEXT: [[TMP18:%.*]] = mul nsw <vscale x 4 x i32> [[TMP13]], [[TMP15]]
-; CHECK-SVE-NEXT: [[TMP19]] = sub <vscale x 4 x i32> [[TMP17]], [[TMP18]]
-; CHECK-SVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
+; CHECK-SVE-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
+; CHECK-SVE-NEXT: [[WIDE_LOAD1:%.*]] = load <16 x i8>, ptr [[TMP8]], align 1
+; CHECK-SVE-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP9]], align 1
+; CHECK-SVE-NEXT: [[TMP3:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
+; CHECK-SVE-NEXT: [[TMP4:%.*]] = sext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
+; CHECK-SVE-NEXT: [[TMP5:%.*]] = mul nsw <16 x i32> [[TMP3]], [[TMP4]]
+; CHECK-SVE-NEXT: [[TMP6:%.*]] = sub nsw <16 x i32> zeroinitializer, [[TMP5]]
+; CHECK-SVE-NEXT: [[PARTIAL_REDUCE:%.*]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP6]])
+; CHECK-SVE-NEXT: [[TMP10:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
+; CHECK-SVE-NEXT: [[TMP12:%.*]] = mul nsw <16 x i32> [[TMP3]], [[TMP10]]
+; CHECK-SVE-NEXT: [[TMP13:%.*]] = sub <16 x i32> zeroinitializer, [[TMP12]]
+; CHECK-SVE-NEXT: [[PARTIAL_REDUCE3]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[PARTIAL_REDUCE]], <16 x i32> [[TMP13]])
+; CHECK-SVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-SVE-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-SVE-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
; CHECK-SVE: middle.block:
-; CHECK-SVE-NEXT: [[TMP21:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[TMP19]])
+; CHECK-SVE-NEXT: [[TMP11:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE3]])
; CHECK-SVE-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC]]
; CHECK-SVE-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]]
; CHECK-SVE: scalar.ph:
@@ -570,11 +562,11 @@ define i32 @chained_partial_reduce_sub_sub(ptr %a, ptr %b, ptr %c, i32 %N) #0 {
; CHECK-SVE-MAXBW-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP9]], align 1
; CHECK-SVE-MAXBW-NEXT: [[TMP13:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
; CHECK-SVE-MAXBW-NEXT: [[TMP14:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD1]] to <vscale x 8 x i32>
-; CHECK-SVE-MAXBW-NEXT: [[TMP15:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
; CHECK-SVE-MAXBW-NEXT: [[TMP16:%.*]] = mul nsw <vscale x 8 x i32> [[TMP13]], [[TMP14]]
; CHECK-SVE-MAXBW-NEXT: [[TMP17:%.*]] = sub nsw <vscale x 8 x i32> zeroinitializer, [[TMP16]]
; CHECK-SVE-MAXBW-NEXT: [[PARTIAL_REDUCE:%.*]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI]], <vscale x 8 x i32> [[TMP17]])
-; CHECK-SVE-MAXBW-NEXT: [[TMP18:%.*]] = mul nsw <vscale x 8 x i32> [[TMP13]], [[TMP15]]
+; CHECK-SVE-MAXBW-NEXT: [[TMP12:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
+; CHECK-SVE-MAXBW-NEXT: [[TMP18:%.*]] = mul nsw <vscale x 8 x i32> [[TMP13]], [[TMP12]]
; CHECK-SVE-MAXBW-NEXT: [[TMP19:%.*]] = sub <vscale x 8 x i32> zeroinitializer, [[TMP18]]
; CHECK-SVE-MAXBW-NEXT: [[PARTIAL_REDUCE3]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[PARTIAL_REDUCE]], <vscale x 8 x i32> [[TMP19]])
; CHECK-SVE-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
@@ -644,12 +636,12 @@ define i32 @chained_partial_reduce_add_add_add(ptr %a, ptr %b, ptr %c, i32 %N) #
; CHECK-NEON-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP3]], align 1
; CHECK-NEON-NEXT: [[TMP7:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-NEON-NEXT: [[TMP8:%.*]] = sext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
-; CHECK-NEON-NEXT: [[TMP9:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-NEON-NEXT: [[TMP10:%.*]] = mul nsw <16 x i32> [[TMP7]], [[TMP8]]
; CHECK-NEON-NEXT: [[PARTIAL_REDUCE:%.*]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP10]])
-; CHECK-NEON-NEXT: [[TMP11:%.*]] = mul nsw <16 x i32> [[TMP7]], [[TMP9]]
+; CHECK-NEON-NEXT: [[TMP15:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
+; CHECK-NEON-NEXT: [[TMP11:%.*]] = mul nsw <16 x i32> [[TMP7]], [[TMP15]]
; CHECK-NEON-NEXT: [[PARTIAL_REDUCE3:%.*]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[PARTIAL_REDUCE]], <16 x i32> [[TMP11]])
-; CHECK-NEON-NEXT: [[TMP12:%.*]] = mul nsw <16 x i32> [[TMP8]], [[TMP9]]
+; CHECK-NEON-NEXT: [[TMP12:%.*]] = mul nsw <16 x i32> [[TMP8]], [[TMP15]]
; CHECK-NEON-NEXT: [[PARTIAL_REDUCE4]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[PARTIAL_REDUCE3]], <16 x i32> [[TMP12]])
; CHECK-NEON-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-NEON-NEXT: [[TMP13:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
@@ -683,9 +675,9 @@ define i32 @chained_partial_reduce_add_add_add(ptr %a, ptr %b, ptr %c, i32 %N) #
; CHECK-SVE-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP9]], align 1
; CHECK-SVE-NEXT: [[TMP3:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-SVE-NEXT: [[TMP4:%.*]] = sext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
-; CHECK-SVE-NEXT: [[TMP5:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-SVE-NEXT: [[TMP6:%.*]] = mul nsw <16 x i32> [[TMP3]], [[TMP4]]
; CHECK-SVE-NEXT: [[PARTIAL_REDUCE:%.*]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP6]])
+; CHECK-SVE-NEXT: [[TMP5:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-SVE-NEXT: [[TMP11:%.*]] = mul nsw <16 x i32> [[TMP3]], [[TMP5]]
; CHECK-SVE-NEXT: [[PARTIAL_REDUCE3:%.*]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[PARTIAL_REDUCE]], <16 x i32> [[TMP11]])
; CHECK-SVE-NEXT: [[TMP12:%.*]] = mul nsw <16 x i32> [[TMP4]], [[TMP5]]
@@ -726,12 +718,12 @@ define i32 @chained_partial_reduce_add_add_add(ptr %a, ptr %b, ptr %c, i32 %N) #
; CHECK-SVE-MAXBW-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP9]], align 1
; CHECK-SVE-MAXBW-NEXT: [[TMP13:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
; CHECK-SVE-MAXBW-NEXT: [[TMP14:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD1]] to <vscale x 8 x i32>
-; CHECK-SVE-MAXBW-NEXT: [[TMP15:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
; CHECK-SVE-MAXBW-NEXT: [[TMP16:%.*]] = mul nsw <vscale x 8 x i32> [[TMP13]], [[TMP14]]
; CHECK-SVE-MAXBW-NEXT: [[PARTIAL_REDUCE:%.*]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI]], <vscale x 8 x i32> [[TMP16]])
-; CHECK-SVE-MAXBW-NEXT: [[TMP17:%.*]] = mul nsw <vscale x 8 x i32> [[TMP13]], [[TMP15]]
+; CHECK-SVE-MAXBW-NEXT: [[TMP11:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
+; CHECK-SVE-MAXBW-NEXT: [[TMP17:%.*]] = mul nsw <vscale x 8 x i32> [[TMP13]], [[TMP11]]
; CHECK-SVE-MAXBW-NEXT: [[PARTIAL_REDUCE3:%.*]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[PARTIAL_REDUCE]], <vscale x 8 x i32> [[TMP17]])
-; CHECK-SVE-MAXBW-NEXT: [[TMP18:%.*]] = mul nsw <vscale x 8 x i32> [[TMP14]], [[TMP15]]
+; CHECK-SVE-MAXBW-NEXT: [[TMP18:%.*]] = mul nsw <vscale x 8 x i32> [[TMP14]], [[TMP11]]
; CHECK-SVE-MAXBW-NEXT: [[PARTIAL_REDUCE4]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[PARTIAL_REDUCE3]], <vscale x 8 x i32> [[TMP18]])
; CHECK-SVE-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
; CHECK-SVE-MAXBW-NEXT: [[TMP19:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
@@ -802,13 +794,13 @@ define i32 @chained_partial_reduce_sub_add_sub(ptr %a, ptr %b, ptr %c, i32 %N) #
; CHECK-NEON-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP3]], align 1
; CHECK-NEON-NEXT: [[TMP7:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-NEON-NEXT: [[TMP8:%.*]] = sext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
-; CHECK-NEON-NEXT: [[TMP9:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-NEON-NEXT: [[TMP10:%.*]] = mul nsw <16 x i32> [[TMP7]], [[TMP8]]
; CHECK-NEON-NEXT: [[TMP11:%.*]] = sub nsw <16 x i32> zeroinitializer, [[TMP10]]
; CHECK-NEON-NEXT: [[PARTIAL_REDUCE:%.*]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP11]])
-; CHECK-NEON-NEXT: [[TMP12:%.*]] = mul nsw <16 x i32> [[TMP7]], [[TMP9]]
+; CHECK-NEON-NEXT: [[TMP13:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
+; CHECK-NEON-NEXT: [[TMP12:%.*]] = mul nsw <16 x i32> [[TMP7]], [[TMP13]]
; CHECK-NEON-NEXT: [[PARTIAL_REDUCE3:%.*]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[PARTIAL_REDUCE]], <16 x i32> [[TMP12]])
-; CHECK-NEON-NEXT: [[TMP14:%.*]] = mul nsw <16 x i32> [[TMP8]], [[TMP9]]
+; CHECK-NEON-NEXT: [[TMP14:%.*]] = mul nsw <16 x i32> [[TMP8]], [[TMP13]]
; CHECK-NEON-NEXT: [[TMP15:%.*]] = sub <16 x i32> zeroinitializer, [[TMP14]]
; CHECK-NEON-NEXT: [[PARTIAL_REDUCE4]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[PARTIAL_REDUCE3]], <16 x i32> [[TMP15]])
; CHECK-NEON-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
@@ -826,39 +818,37 @@ define i32 @chained_partial_reduce_sub_add_sub(ptr %a, ptr %b, ptr %c, i32 %N) #
; CHECK-SVE-NEXT: [[CMP28_NOT:%.*]] = icmp ult i32 [[N]], 2
; CHECK-SVE-NEXT: [[DIV27:%.*]] = lshr i32 [[N]], 1
; CHECK-SVE-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext nneg i32 [[DIV27]] to i64
-; CHECK-SVE-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-SVE-NEXT: [[TMP1:%.*]] = shl nuw nsw i64 [[TMP0]], 2
-; CHECK-SVE-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[WIDE_TRIP_COUNT]], [[TMP1]]
+; CHECK-SVE-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[WIDE_TRIP_COUNT]], 16
; CHECK-SVE-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK-SVE: vector.ph:
-; CHECK-SVE-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-SVE-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 4
-; CHECK-SVE-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[WIDE_TRIP_COUNT]], [[TMP3]]
+; CHECK-SVE-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[WIDE_TRIP_COUNT]], 16
; CHECK-SVE-NEXT: [[N_VEC:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], [[N_MOD_VF]]
; CHECK-SVE-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-SVE: vector.body:
; CHECK-SVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-SVE-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP21:%.*]], [[VECTOR_BODY]] ]
+; CHECK-SVE-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE4:%.*]], [[VECTOR_BODY]] ]
; CHECK-SVE-NEXT: [[TMP7:%.*]] = getelementptr inbounds nuw i8, ptr [[A]], i64 [[INDEX]]
; CHECK-SVE-NEXT: [[TMP8:%.*]] = getelementptr inbounds nuw i8, ptr [[B]], i64 [[INDEX]]
; CHECK-SVE-NEXT: [[TMP9:%.*]] = getelementptr inbounds nuw i8, ptr [[C]], i64 [[INDEX]]
-; CHECK-SVE-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i8>, ptr [[TMP7]], align 1
-; CHECK-SVE-NEXT: [[WIDE_LOAD1:%.*]] = load <vscale x 4 x i8>, ptr [[TMP8]], align 1
-; CHECK-SVE-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 4 x i8>, ptr [[TMP9]], align 1
-; CHECK-SVE-NEXT: [[TMP13:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD]] to <vscale x 4 x i32>
-; CHECK-SVE-NEXT: [[TMP14:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD1]] to <vscale x 4 x i32>
-; CHECK-SVE-NEXT: [[TMP15:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD2]] to <vscale x 4 x i32>
-; CHECK-SVE-NEXT: [[TMP16:%.*]] = mul nsw <vscale x 4 x i32> [[TMP13]], [[TMP14]]
-; CHECK-SVE-NEXT: [[TMP17:%.*]] = sub <vscale x 4 x i32> [[VEC_PHI]], [[TMP16]]
-; CHECK-SVE-NEXT: [[TMP18:%.*]] = mul nsw <vscale x 4 x i32> [[TMP13]], [[TMP15]]
-; CHECK-SVE-NEXT: [[TMP19:%.*]] = add <vscale x 4 x i32> [[TMP17]], [[TMP18]]
-; CHECK-SVE-NEXT: [[TMP20:%.*]] = mul nsw <vscale x 4 x i32> [[TMP14]], [[TMP15]]
-; CHECK-SVE-NEXT: [[TMP21]] = sub <vscale x 4 x i32> [[TMP19]], [[TMP20]]
-; CHECK-SVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
+; CHECK-SVE-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
+; CHECK-SVE-NEXT: [[WIDE_LOAD1:%.*]] = load <16 x i8>, ptr [[TMP8]], align 1
+; CHECK-SVE-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP9]], align 1
+; CHECK-SVE-NEXT: [[TMP3:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
+; CHECK-SVE-NEXT: [[TMP4:%.*]] = sext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
+; CHECK-SVE-NEXT: [[TMP5:%.*]] = mul nsw <16 x i32> [[TMP3]], [[TMP4]]
+; CHECK-SVE-NEXT: [[TMP6:%.*]] = sub nsw <16 x i32> zeroinitializer, [[TMP5]]
+; CHECK-SVE-NEXT: [[PARTIAL_REDUCE:%.*]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP6]])
+; CHECK-SVE-NEXT: [[TMP11:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
+; CHECK-SVE-NEXT: [[TMP13:%.*]] = mul nsw <16 x i32> [[TMP3]], [[TMP11]]
+; CHECK-SVE-NEXT: [[PARTIAL_REDUCE3:%.*]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[PARTIAL_REDUCE]], <16 x i32> [[TMP13]])
+; CHECK-SVE-NEXT: [[TMP14:%.*]] = mul nsw <16 x i32> [[TMP4]], [[TMP11]]
+; CHECK-SVE-NEXT: [[TMP10:%.*]] = sub <16 x i32> zeroinitializer, [[TMP14]]
+; CHECK-SVE-NEXT: [[PARTIAL_REDUCE4]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[PARTIAL_REDUCE3]], <16 x i32> [[TMP10]])
+; CHECK-SVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-SVE-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-SVE-NEXT: br i1 [[TMP22]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP14:![0-9]+]]
; CHECK-SVE: middle.block:
-; CHECK-SVE-NEXT: [[TMP23:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[TMP21]])
+; CHECK-SVE-NEXT: [[TMP12:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE4]])
; CHECK-SVE-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC]]
; CHECK-SVE-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]]
; CHECK-SVE: scalar.ph:
@@ -890,13 +880,13 @@ define i32 @chained_partial_reduce_sub_add_sub(ptr %a, ptr %b, ptr %c, i32 %N) #
; CHECK-SVE-MAXBW-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP9]], align 1
; CHECK-SVE-MAXBW-NEXT: [[TMP13:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
; CHECK-SVE-MAXBW-NEXT: [[TMP14:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD1]] to <vscale x 8 x i32>
-; CHECK-SVE-MAXBW-NEXT: [[TMP15:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
; CHECK-SVE-MAXBW-NEXT: [[TMP16:%.*]] = mul nsw <vscale x 8 x i32> [[TMP13]], [[TMP14]]
; CHECK-SVE-MAXBW-NEXT: [[TMP17:%.*]] = sub nsw <vscale x 8 x i32> zeroinitializer, [[TMP16]]
; CHECK-SVE-MAXBW-NEXT: [[PARTIAL_REDUCE:%.*]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI]], <vscale x 8 x i32> [[TMP17]])
-; CHECK-SVE-MAXBW-NEXT: [[TMP18:%.*]] = mul nsw <vscale x 8 x i32> [[TMP13]], [[TMP15]]
+; CHECK-SVE-MAXBW-NEXT: [[TMP12:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
+; CHECK-SVE-MAXBW-NEXT: [[TMP18:%.*]] = mul nsw <vscale x 8 x i32> [[TMP13]], [[TMP12]]
; CHECK-SVE-MAXBW-NEXT: [[PARTIAL_REDUCE3:%.*]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[PARTIAL_REDUCE]], <vscale x 8 x i32> [[TMP18]])
-; CHECK-SVE-MAXBW-NEXT: [[TMP19:%.*]] = mul nsw <vscale x 8 x i32> [[TMP14]], [[TMP15]]
+; CHECK-SVE-MAXBW-NEXT: [[TMP19:%.*]] = mul nsw <vscale x 8 x i32> [[TMP14]], [[TMP12]]
; CHECK-SVE-MAXBW-NEXT: [[TMP20:%.*]] = sub <vscale x 8 x i32> zeroinitializer, [[TMP19]]
; CHECK-SVE-MAXBW-NEXT: [[PARTIAL_REDUCE4]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[PARTIAL_REDUCE3]], <vscale x 8 x i32> [[TMP20]])
; CHECK-SVE-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
@@ -969,9 +959,9 @@ define i32 @chained_partial_reduce_madd_extadd(ptr %a, ptr %b, ptr %c, i32 %N) #
; CHECK-NEON-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP3]], align 1
; CHECK-NEON-NEXT: [[TMP7:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-NEON-NEXT: [[TMP8:%.*]] = sext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
-; CHECK-NEON-NEXT: [[TMP9:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-NEON-NEXT: [[TMP10:%.*]] = mul nsw <16 x i32> [[TMP7]], [[TMP8]]
; CHECK-NEON-NEXT: [[PARTIAL_REDUCE:%.*]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP10]])
+; CHECK-NEON-NEXT: [[TMP9:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-NEON-NEXT: [[PARTIAL_REDUCE3]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[PARTIAL_REDUCE]], <16 x i32> [[TMP9]])
; CHECK-NEON-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-NEON-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
@@ -1005,9 +995,9 @@ define i32 @chained_partial_reduce_madd_extadd(ptr %a, ptr %b, ptr %c, i32 %N) #
; CHECK-SVE-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP9]], align 1
; CHECK-SVE-NEXT: [[TMP3:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-SVE-NEXT: [[TMP4:%.*]] = sext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
-; CHECK-SVE-NEXT: [[TMP5:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-SVE-NEXT: [[TMP6:%.*]] = mul nsw <16 x i32> [[TMP3]], [[TMP4]]
; CHECK-SVE-NEXT: [[PARTIAL_REDUCE:%.*]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP6]])
+; CHECK-SVE-NEXT: [[TMP5:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-SVE-NEXT: [[PARTIAL_REDUCE3]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[PARTIAL_REDUCE]], <16 x i32> [[TMP5]])
; CHECK-SVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-SVE-NEXT: [[TMP19:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
@@ -1045,9 +1035,9 @@ define i32 @chained_partial_reduce_madd_extadd(ptr %a, ptr %b, ptr %c, i32 %N) #
; CHECK-SVE-MAXBW-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP9]], align 1
; CHECK-SVE-MAXBW-NEXT: [[TMP13:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
; CHECK-SVE-MAXBW-NEXT: [[TMP14:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD1]] to <vscale x 8 x i32>
-; CHECK-SVE-MAXBW-NEXT: [[TMP15:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
; CHECK-SVE-MAXBW-NEXT: [[TMP16:%.*]] = mul nsw <vscale x 8 x i32> [[TMP13]], [[TMP14]]
; CHECK-SVE-MAXBW-NEXT: [[PARTIAL_REDUCE:%.*]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI]], <vscale x 8 x i32> [[TMP16]])
+; CHECK-SVE-MAXBW-NEXT: [[TMP15:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
; CHECK-SVE-MAXBW-NEXT: [[PARTIAL_REDUCE3]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[PARTIAL_REDUCE]], <vscale x 8 x i32> [[TMP15]])
; CHECK-SVE-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
; CHECK-SVE-MAXBW-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
@@ -1110,8 +1100,8 @@ define i32 @chained_partial_reduce_extadd_extadd(ptr %a, ptr %b, i32 %N) #0 {
; CHECK-NEON-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1
; CHECK-NEON-NEXT: [[WIDE_LOAD1:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1
; CHECK-NEON-NEXT: [[TMP5:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
-; CHECK-NEON-NEXT: [[TMP6:%.*]] = sext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
; CHECK-NEON-NEXT: [[PARTIAL_REDUCE:%.*]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP5]])
+; CHECK-NEON-NEXT: [[TMP6:%.*]] = sext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
; CHECK-NEON-NEXT: [[PARTIAL_REDUCE2]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[PARTIAL_REDUCE]], <16 x i32> [[TMP6]])
; CHECK-NEON-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-NEON-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
@@ -1142,8 +1132,8 @@ define i32 @chained_partial_reduce_extadd_extadd(ptr %a, ptr %b, i32 %N) #0 {
; CHECK-SVE-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
; CHECK-SVE-NEXT: [[WIDE_LOAD1:%.*]] = load <16 x i8>, ptr [[TMP8]], align 1
; CHECK-SVE-NEXT: [[TMP2:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
-; CHECK-SVE-NEXT: [[TMP3:%.*]] = sext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
; CHECK-SVE-NEXT: [[PARTIAL_REDUCE:%.*]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP2]])
+; CHECK-SVE-NEXT: [[TMP3:%.*]] = sext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
; CHECK-SVE-NEXT: [[PARTIAL_REDUCE2]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[PARTIAL_REDUCE]], <16 x i32> [[TMP3]])
; CHECK-SVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-SVE-NEXT: [[TMP15:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
@@ -1178,8 +1168,8 @@ define i32 @chained_partial_reduce_extadd_extadd(ptr %a, ptr %b, i32 %N) #0 {
; CHECK-SVE-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
; CHECK-SVE-MAXBW-NEXT: [[WIDE_LOAD1:%.*]] = load <vscale x 8 x i8>, ptr [[TMP8]], align 1
; CHECK-SVE-MAXBW-NEXT: [[TMP11:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
-; CHECK-SVE-MAXBW-NEXT: [[TMP12:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD1]] to <vscale x 8 x i32>
; CHECK-SVE-MAXBW-NEXT: [[PARTIAL_REDUCE:%.*]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI]], <vscale x 8 x i32> [[TMP11]])
+; CHECK-SVE-MAXBW-NEXT: [[TMP12:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD1]] to <vscale x 8 x i32>
; CHECK-SVE-MAXBW-NEXT: [[PARTIAL_REDUCE2]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[PARTIAL_REDUCE]], <vscale x 8 x i32> [[TMP12]])
; CHECK-SVE-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
; CHECK-SVE-MAXBW-NEXT: [[TMP15:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
@@ -1240,10 +1230,10 @@ define i32 @chained_partial_reduce_extadd_madd(ptr %a, ptr %b, ptr %c, i32 %N) #
; CHECK-NEON-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1
; CHECK-NEON-NEXT: [[WIDE_LOAD1:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1
; CHECK-NEON-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP3]], align 1
-; CHECK-NEON-NEXT: [[TMP7:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
-; CHECK-NEON-NEXT: [[TMP8:%.*]] = sext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
; CHECK-NEON-NEXT: [[TMP9:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-NEON-NEXT: [[PARTIAL_REDUCE:%.*]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP9]])
+; CHECK-NEON-NEXT: [[TMP7:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
+; CHECK-NEON-NEXT: [[TMP8:%.*]] = sext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
; CHECK-NEON-NEXT: [[TMP10:%.*]] = mul nsw <16 x i32> [[TMP7]], [[TMP8]]
; CHECK-NEON-NEXT: [[PARTIAL_REDUCE3]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[PARTIAL_REDUCE]], <16 x i32> [[TMP10]])
; CHECK-NEON-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
@@ -1276,10 +1266,10 @@ define i32 @chained_partial_reduce_extadd_madd(ptr %a, ptr %b, ptr %c, i32 %N) #
; CHECK-SVE-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
; CHECK-SVE-NEXT: [[WIDE_LOAD1:%.*]] = load <16 x i8>, ptr [[TMP8]], align 1
; CHECK-SVE-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP9]], align 1
-; CHECK-SVE-NEXT: [[TMP3:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
-; CHECK-SVE-NEXT: [[TMP4:%.*]] = sext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
; CHECK-SVE-NEXT: [[TMP5:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-SVE-NEXT: [[PARTIAL_REDUCE:%.*]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP5]])
+; CHECK-SVE-NEXT: [[TMP3:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
+; CHECK-SVE-NEXT: [[TMP4:%.*]] = sext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
; CHECK-SVE-NEXT: [[TMP6:%.*]] = mul nsw <16 x i32> [[TMP3]], [[TMP4]]
; CHECK-SVE-NEXT: [[PARTIAL_REDUCE3]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[PARTIAL_REDUCE]], <16 x i32> [[TMP6]])
; CHECK-SVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
@@ -1316,10 +1306,10 @@ define i32 @chained_partial_reduce_extadd_madd(ptr %a, ptr %b, ptr %c, i32 %N) #
; CHECK-SVE-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
; CHECK-SVE-MAXBW-NEXT: [[WIDE_LOAD1:%.*]] = load <vscale x 8 x i8>, ptr [[TMP8]], align 1
; CHECK-SVE-MAXBW-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP9]], align 1
-; CHECK-SVE-MAXBW-NEXT: [[TMP13:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
-; CHECK-SVE-MAXBW-NEXT: [[TMP14:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD1]] to <vscale x 8 x i32>
; CHECK-SVE-MAXBW-NEXT: [[TMP15:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
; CHECK-SVE-MAXBW-NEXT: [[PARTIAL_REDUCE:%.*]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI]], <vscale x 8 x i32> [[TMP15]])
+; CHECK-SVE-MAXBW-NEXT: [[TMP13:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
+; CHECK-SVE-MAXBW-NEXT: [[TMP14:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD1]] to <vscale x 8 x i32>
; CHECK-SVE-MAXBW-NEXT: [[TMP16:%.*]] = mul nsw <vscale x 8 x i32> [[TMP13]], [[TMP14]]
; CHECK-SVE-MAXBW-NEXT: [[PARTIAL_REDUCE3]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[PARTIAL_REDUCE]], <vscale x 8 x i32> [[TMP16]])
; CHECK-SVE-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-constant-ops.ll b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-constant-ops.ll
index b033f6051f812..b430efc9e5283 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-constant-ops.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-constant-ops.ll
@@ -467,3 +467,83 @@ loop:
exit:
ret i32 %red.next
}
+
+define i64 @partial_reduction_mul_two_users(i64 %n, ptr %a, i16 %b, i32 %c) {
+; CHECK-LABEL: define i64 @partial_reduction_mul_two_users(
+; CHECK-SAME: i64 [[N:%.*]], ptr [[A:%.*]], i16 [[B:%.*]], i32 [[C:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*]]:
+; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[N]], 1
+; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 8
+; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; CHECK: [[VECTOR_PH]]:
+; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 8
+; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <8 x i16> poison, i16 [[B]], i64 0
+; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <8 x i16> [[BROADCAST_SPLATINSERT]], <8 x i16> poison, <8 x i32> zeroinitializer
+; CHECK-NEXT: [[TMP1:%.*]] = sext <8 x i16> [[BROADCAST_SPLAT]] to <8 x i32>
+; CHECK-NEXT: [[TMP2:%.*]] = mul <8 x i32> [[TMP1]], [[TMP1]]
+; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
+; CHECK: [[VECTOR_BODY]]:
+; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i64> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP4:%.*]] = load i16, ptr [[A]], align 2
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <8 x i16> poison, i16 [[TMP4]], i64 0
+; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <8 x i16> [[BROADCAST_SPLATINSERT1]], <8 x i16> poison, <8 x i32> zeroinitializer
+; CHECK-NEXT: [[TMP3:%.*]] = zext <8 x i32> [[TMP2]] to <8 x i64>
+; CHECK-NEXT: [[PARTIAL_REDUCE]] = call <4 x i64> @llvm.vector.partial.reduce.add.v4i64.v8i64(<4 x i64> [[VEC_PHI]], <8 x i64> [[TMP3]])
+; CHECK-NEXT: [[TMP5:%.*]] = sext <8 x i16> [[BROADCAST_SPLAT2]] to <8 x i32>
+; CHECK-NEXT: [[TMP6:%.*]] = sext <8 x i32> [[TMP5]] to <8 x i64>
+; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
+; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP18:![0-9]+]]
+; CHECK: [[MIDDLE_BLOCK]]:
+; CHECK-NEXT: [[TMP8:%.*]] = call i64 @llvm.vector.reduce.add.v4i64(<4 x i64> [[PARTIAL_REDUCE]])
+; CHECK-NEXT: [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <8 x i64> [[TMP6]], i32 7
+; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
+; CHECK: [[SCALAR_PH]]:
+; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
+; CHECK-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi i64 [ [[VECTOR_RECUR_EXTRACT]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
+; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i64 [ [[TMP8]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
+; CHECK-NEXT: br label %[[LOOP:.*]]
+; CHECK: [[LOOP]]:
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
+; CHECK-NEXT: [[RES1:%.*]] = phi i64 [ [[SCALAR_RECUR_INIT]], %[[SCALAR_PH]] ], [ [[LOAD_EXT_EXT:%.*]], %[[LOOP]] ]
+; CHECK-NEXT: [[RES2:%.*]] = phi i64 [ [[BC_MERGE_RDX]], %[[SCALAR_PH]] ], [ [[ADD:%.*]], %[[LOOP]] ]
+; CHECK-NEXT: [[LOAD:%.*]] = load i16, ptr [[A]], align 2
+; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 1
+; CHECK-NEXT: [[CONV:%.*]] = sext i16 [[B]] to i32
+; CHECK-NEXT: [[MUL:%.*]] = mul i32 [[CONV]], [[CONV]]
+; CHECK-NEXT: [[MUL_EXT:%.*]] = zext i32 [[MUL]] to i64
+; CHECK-NEXT: [[ADD]] = add i64 [[RES2]], [[MUL_EXT]]
+; CHECK-NEXT: [[OR:%.*]] = or i32 [[MUL]], [[C]]
+; CHECK-NEXT: [[LOAD_EXT:%.*]] = sext i16 [[LOAD]] to i32
+; CHECK-NEXT: [[LOAD_EXT_EXT]] = sext i32 [[LOAD_EXT]] to i64
+; CHECK-NEXT: [[EXITCOND740_NOT:%.*]] = icmp eq i64 [[IV]], [[N]]
+; CHECK-NEXT: br i1 [[EXITCOND740_NOT]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP19:![0-9]+]]
+; CHECK: [[EXIT]]:
+; CHECK-NEXT: [[ADD_LCSSA:%.*]] = phi i64 [ [[ADD]], %[[LOOP]] ], [ [[TMP8]], %[[MIDDLE_BLOCK]] ]
+; CHECK-NEXT: ret i64 [[ADD_LCSSA]]
+;
+entry:
+ br label %loop
+
+loop:
+ %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
+ %res1 = phi i64 [ 0, %entry ], [ %load.ext.ext, %loop ]
+ %res2 = phi i64 [ 0, %entry ], [ %add, %loop ]
+ %load = load i16, ptr %a, align 2
+ %iv.next = add i64 %iv, 1
+ %conv = sext i16 %b to i32
+ %mul = mul i32 %conv, %conv
+ %mul.ext = zext i32 %mul to i64
+ %add = add i64 %res2, %mul.ext
+ %second_use = or i32 %mul, %c ; this value is otherwise unused, but that's sufficient for the test
+ %load.ext = sext i16 %load to i32
+ %load.ext.ext = sext i32 %load.ext to i64
+ %exitcond740.not = icmp eq i64 %iv, %n
+ br i1 %exitcond740.not, label %exit, label %loop
+
+exit:
+ ret i64 %add
+}
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-epilogue.ll b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-epilogue.ll
index 8ece59aef90a9..d8f1a86c9ebda 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-epilogue.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-epilogue.ll
@@ -16,10 +16,10 @@ define i32 @dotp(ptr %a, ptr %b) #0 {
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP3]], align 1
-; CHECK-NEXT: [[TMP5:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; CHECK-NEXT: [[WIDE_LOAD1:%.*]] = load <16 x i8>, ptr [[TMP6]], align 1
; CHECK-NEXT: [[TMP8:%.*]] = zext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
+; CHECK-NEXT: [[TMP5:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-NEXT: [[TMP9:%.*]] = mul <16 x i32> [[TMP8]], [[TMP5]]
; CHECK-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP9]])
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
@@ -68,7 +68,6 @@ define void @dotp_small_epilogue_vf(i64 %idx.neg, i8 %a) #1 {
; CHECK-NEXT: [[IV_NEXT:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i8> poison, i8 [[A]], i64 0
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <16 x i8> [[BROADCAST_SPLATINSERT]], <16 x i8> poison, <16 x i32> zeroinitializer
-; CHECK-NEXT: [[TMP1:%.*]] = sext <16 x i8> [[BROADCAST_SPLAT]] to <16 x i32>
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
@@ -77,11 +76,12 @@ define void @dotp_small_epilogue_vf(i64 %idx.neg, i8 %a) #1 {
; CHECK-NEXT: [[BROADCAST_SPLATINSERT2:%.*]] = insertelement <16 x i8> poison, i8 [[TMP2]], i64 0
; CHECK-NEXT: [[BROADCAST_SPLAT3:%.*]] = shufflevector <16 x i8> [[BROADCAST_SPLATINSERT2]], <16 x i8> poison, <16 x i32> zeroinitializer
; CHECK-NEXT: [[TMP3:%.*]] = sext <16 x i8> [[BROADCAST_SPLAT3]] to <16 x i32>
+; CHECK-NEXT: [[TMP1:%.*]] = sext <16 x i8> [[BROADCAST_SPLAT]] to <16 x i32>
; CHECK-NEXT: [[TMP4:%.*]] = mul <16 x i32> [[TMP3]], [[TMP1]]
; CHECK-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP4]])
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[IV_NEXT]]
-; CHECK-NEXT: br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
+; CHECK-NEXT: br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
; CHECK: middle.block:
; CHECK-NEXT: [[TMP6:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE]])
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[IV_NEXT]]
@@ -89,7 +89,7 @@ define void @dotp_small_epilogue_vf(i64 %idx.neg, i8 %a) #1 {
; CHECK: vec.epilog.iter.check:
; CHECK-NEXT: [[IND_END6:%.*]] = add i64 [[IDX_NEG]], [[IV_NEXT]]
; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], 4
-; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]
+; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF4:![0-9]+]]
; CHECK: vec.epilog.ph:
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT]], [[WHILE_BODY]] ], [ 0, [[ENTRY]] ]
; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP6]], [[WHILE_BODY]] ], [ 0, [[ENTRY]] ]
@@ -112,7 +112,7 @@ define void @dotp_small_epilogue_vf(i64 %idx.neg, i8 %a) #1 {
; CHECK-NEXT: [[TMP13]] = add <4 x i32> [[TMP14]], [[VEC_PHI9]]
; CHECK-NEXT: [[INDEX_NEXT14]] = add nuw i64 [[INDEX9]], 4
; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT14]], [[N_VEC5]]
-; CHECK-NEXT: br i1 [[TMP12]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
+; CHECK-NEXT: br i1 [[TMP12]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
; CHECK: vec.epilog.middle.block:
; CHECK-NEXT: [[TMP15:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP13]])
; CHECK-NEXT: [[CMP_N15:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC5]]
@@ -136,7 +136,7 @@ define void @dotp_small_epilogue_vf(i64 %idx.neg, i8 %a) #1 {
; CHECK-NEXT: [[CMP_IV_NEG:%.*]] = icmp ugt i64 [[IV_NEG]], 0
; CHECK-NEXT: [[CMP_IV:%.*]] = icmp ne i64 [[ACCUM1]], -1
; CHECK-NEXT: [[EXITCOND:%.*]] = and i1 [[CMP_IV_NEG]], [[CMP_IV]]
-; CHECK-NEXT: br i1 [[EXITCOND]], label [[WHILE_BODY1]], label [[WHILE_END_LOOPEXIT]], !llvm.loop [[LOOP5:![0-9]+]]
+; CHECK-NEXT: br i1 [[EXITCOND]], label [[WHILE_BODY1]], label [[WHILE_END_LOOPEXIT]], !llvm.loop [[LOOP6:![0-9]+]]
; CHECK: while.end.loopexit:
; CHECK-NEXT: [[RESULT:%.*]] = phi i32 [ [[ADD]], [[WHILE_BODY1]] ], [ [[TMP6]], [[MIDDLE_BLOCK]] ], [ [[TMP15]], [[VEC_EPILOG_MIDDLE_BLOCK]] ]
; CHECK-NEXT: ret void
@@ -495,7 +495,7 @@ define i32 @dotp_predicated(i64 %N, ptr %a, ptr %b) {
; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 16
; CHECK-NEXT: [[VEC_IND_NEXT]] = add <16 x i64> [[VEC_IND]], splat (i64 16)
; CHECK-NEXT: [[TMP181:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT: br i1 [[TMP181]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
+; CHECK-NEXT: br i1 [[TMP181]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]]
; CHECK: middle.block:
; CHECK-NEXT: [[TMP182:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE]])
; CHECK-NEXT: br label [[EXIT:%.*]]
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll
index 09b41fb551775..26e630f969ef3 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-mixed.ll
@@ -26,26 +26,26 @@ define i32 @sudot(ptr %a, ptr %b) #0 {
; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP9]]
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 16 x i8>, ptr [[TMP6]], align 1
; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 16 x i8>, ptr [[TMP10]], align 1
-; CHECK-NEXT: [[TMP11:%.*]] = zext <vscale x 16 x i8> [[WIDE_LOAD]] to <vscale x 16 x i32>
-; CHECK-NEXT: [[TMP7:%.*]] = zext <vscale x 16 x i8> [[WIDE_LOAD2]] to <vscale x 16 x i32>
; CHECK-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; CHECK-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP16:%.*]] = shl nuw i64 [[TMP15]], 4
; CHECK-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP13]], i64 [[TMP16]]
; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 16 x i8>, ptr [[TMP13]], align 1
; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 16 x i8>, ptr [[TMP17]], align 1
-; CHECK-NEXT: [[TMP12:%.*]] = sext <vscale x 16 x i8> [[WIDE_LOAD3]] to <vscale x 16 x i32>
-; CHECK-NEXT: [[TMP18:%.*]] = sext <vscale x 16 x i8> [[WIDE_LOAD4]] to <vscale x 16 x i32>
-; CHECK-NEXT: [[TMP14:%.*]] = mul <vscale x 16 x i32> [[TMP12]], [[TMP11]]
-; CHECK-NEXT: [[TMP19:%.*]] = mul <vscale x 16 x i32> [[TMP18]], [[TMP7]]
-; CHECK-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 4 x i32> @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32(<vscale x 4 x i32> [[VEC_PHI]], <vscale x 16 x i32> [[TMP14]])
-; CHECK-NEXT: [[PARTIAL_REDUCE5]] = call <vscale x 4 x i32> @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32(<vscale x 4 x i32> [[VEC_PHI1]], <vscale x 16 x i32> [[TMP19]])
+; CHECK-NEXT: [[TMP18:%.*]] = sext <vscale x 16 x i8> [[WIDE_LOAD3]] to <vscale x 16 x i32>
+; CHECK-NEXT: [[TMP11:%.*]] = zext <vscale x 16 x i8> [[WIDE_LOAD]] to <vscale x 16 x i32>
+; CHECK-NEXT: [[TMP12:%.*]] = mul <vscale x 16 x i32> [[TMP18]], [[TMP11]]
+; CHECK-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 4 x i32> @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32(<vscale x 4 x i32> [[VEC_PHI]], <vscale x 16 x i32> [[TMP12]])
+; CHECK-NEXT: [[TMP19:%.*]] = sext <vscale x 16 x i8> [[WIDE_LOAD4]] to <vscale x 16 x i32>
+; CHECK-NEXT: [[TMP14:%.*]] = zext <vscale x 16 x i8> [[WIDE_LOAD2]] to <vscale x 16 x i32>
+; CHECK-NEXT: [[TMP20:%.*]] = mul <vscale x 16 x i32> [[TMP19]], [[TMP14]]
+; CHECK-NEXT: [[PARTIAL_REDUCE5]] = call <vscale x 4 x i32> @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32(<vscale x 4 x i32> [[VEC_PHI1]], <vscale x 16 x i32> [[TMP20]])
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
; CHECK-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT: br i1 [[TMP22]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK: middle.block:
; CHECK-NEXT: [[BIN_RDX:%.*]] = add <vscale x 4 x i32> [[PARTIAL_REDUCE5]], [[PARTIAL_REDUCE]]
-; CHECK-NEXT: [[TMP20:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[BIN_RDX]])
+; CHECK-NEXT: [[TMP21:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[BIN_RDX]])
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH:%.*]]
; CHECK: scalar.ph:
@@ -62,8 +62,8 @@ define i32 @sudot(ptr %a, ptr %b) #0 {
; CHECK-NOI8MM-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-NOI8MM: vector.body:
; CHECK-NOI8MM-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 8 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP22:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 8 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP23:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 8 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP18:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 8 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP20:%.*]], [[VECTOR_BODY]] ]
; CHECK-NOI8MM-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; CHECK-NOI8MM-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NOI8MM-NEXT: [[TMP9:%.*]] = shl nuw i64 [[TMP8]], 3
@@ -71,25 +71,25 @@ define i32 @sudot(ptr %a, ptr %b) #0 {
; CHECK-NOI8MM-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP6]], align 1
; CHECK-NOI8MM-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP10]], align 1
; CHECK-NOI8MM-NEXT: [[TMP11:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP12:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP7:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
; CHECK-NOI8MM-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; CHECK-NOI8MM-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NOI8MM-NEXT: [[TMP16:%.*]] = shl nuw i64 [[TMP15]], 3
; CHECK-NOI8MM-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP13]], i64 [[TMP16]]
; CHECK-NOI8MM-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP13]], align 1
; CHECK-NOI8MM-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 8 x i8>, ptr [[TMP17]], align 1
-; CHECK-NOI8MM-NEXT: [[TMP18:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD3]] to <vscale x 8 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP19:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD4]] to <vscale x 8 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP20:%.*]] = mul <vscale x 8 x i32> [[TMP18]], [[TMP11]]
-; CHECK-NOI8MM-NEXT: [[TMP21:%.*]] = mul <vscale x 8 x i32> [[TMP19]], [[TMP12]]
-; CHECK-NOI8MM-NEXT: [[TMP22]] = add <vscale x 8 x i32> [[TMP20]], [[VEC_PHI]]
-; CHECK-NOI8MM-NEXT: [[TMP23]] = add <vscale x 8 x i32> [[TMP21]], [[VEC_PHI1]]
+; CHECK-NOI8MM-NEXT: [[TMP12:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD3]] to <vscale x 8 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP21:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD4]] to <vscale x 8 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP14:%.*]] = mul <vscale x 8 x i32> [[TMP12]], [[TMP11]]
+; CHECK-NOI8MM-NEXT: [[TMP22:%.*]] = mul <vscale x 8 x i32> [[TMP21]], [[TMP7]]
+; CHECK-NOI8MM-NEXT: [[TMP18]] = add <vscale x 8 x i32> [[TMP14]], [[VEC_PHI]]
+; CHECK-NOI8MM-NEXT: [[TMP20]] = add <vscale x 8 x i32> [[TMP22]], [[VEC_PHI1]]
; CHECK-NOI8MM-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
; CHECK-NOI8MM-NEXT: [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NOI8MM-NEXT: br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK-NOI8MM: middle.block:
-; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <vscale x 8 x i32> [[TMP23]], [[TMP22]]
-; CHECK-NOI8MM-NEXT: [[TMP25:%.*]] = call i32 @llvm.vector.reduce.add.nxv8i32(<vscale x 8 x i32> [[BIN_RDX]])
+; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <vscale x 8 x i32> [[TMP20]], [[TMP18]]
+; CHECK-NOI8MM-NEXT: [[TMP19:%.*]] = call i32 @llvm.vector.reduce.add.nxv8i32(<vscale x 8 x i32> [[BIN_RDX]])
; CHECK-NOI8MM-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
; CHECK-NOI8MM-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH:%.*]]
; CHECK-NOI8MM: scalar.ph:
@@ -137,26 +137,26 @@ define i32 @usdot(ptr %a, ptr %b) #0 {
; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP6]], i64 [[TMP9]]
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 16 x i8>, ptr [[TMP6]], align 1
; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 16 x i8>, ptr [[TMP10]], align 1
-; CHECK-NEXT: [[TMP11:%.*]] = sext <vscale x 16 x i8> [[WIDE_LOAD]] to <vscale x 16 x i32>
-; CHECK-NEXT: [[TMP7:%.*]] = sext <vscale x 16 x i8> [[WIDE_LOAD2]] to <vscale x 16 x i32>
; CHECK-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; CHECK-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP16:%.*]] = shl nuw i64 [[TMP15]], 4
; CHECK-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP13]], i64 [[TMP16]]
; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 16 x i8>, ptr [[TMP13]], align 1
; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 16 x i8>, ptr [[TMP17]], align 1
-; CHECK-NEXT: [[TMP12:%.*]] = zext <vscale x 16 x i8> [[WIDE_LOAD3]] to <vscale x 16 x i32>
-; CHECK-NEXT: [[TMP18:%.*]] = zext <vscale x 16 x i8> [[WIDE_LOAD4]] to <vscale x 16 x i32>
-; CHECK-NEXT: [[TMP14:%.*]] = mul <vscale x 16 x i32> [[TMP12]], [[TMP11]]
-; CHECK-NEXT: [[TMP19:%.*]] = mul <vscale x 16 x i32> [[TMP18]], [[TMP7]]
-; CHECK-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 4 x i32> @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32(<vscale x 4 x i32> [[VEC_PHI]], <vscale x 16 x i32> [[TMP14]])
-; CHECK-NEXT: [[PARTIAL_REDUCE5]] = call <vscale x 4 x i32> @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32(<vscale x 4 x i32> [[VEC_PHI1]], <vscale x 16 x i32> [[TMP19]])
+; CHECK-NEXT: [[TMP18:%.*]] = zext <vscale x 16 x i8> [[WIDE_LOAD3]] to <vscale x 16 x i32>
+; CHECK-NEXT: [[TMP11:%.*]] = sext <vscale x 16 x i8> [[WIDE_LOAD]] to <vscale x 16 x i32>
+; CHECK-NEXT: [[TMP12:%.*]] = mul <vscale x 16 x i32> [[TMP18]], [[TMP11]]
+; CHECK-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 4 x i32> @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32(<vscale x 4 x i32> [[VEC_PHI]], <vscale x 16 x i32> [[TMP12]])
+; CHECK-NEXT: [[TMP19:%.*]] = zext <vscale x 16 x i8> [[WIDE_LOAD4]] to <vscale x 16 x i32>
+; CHECK-NEXT: [[TMP14:%.*]] = sext <vscale x 16 x i8> [[WIDE_LOAD2]] to <vscale x 16 x i32>
+; CHECK-NEXT: [[TMP20:%.*]] = mul <vscale x 16 x i32> [[TMP19]], [[TMP14]]
+; CHECK-NEXT: [[PARTIAL_REDUCE5]] = call <vscale x 4 x i32> @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32(<vscale x 4 x i32> [[VEC_PHI1]], <vscale x 16 x i32> [[TMP20]])
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
; CHECK-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT: br i1 [[TMP22]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK: middle.block:
; CHECK-NEXT: [[BIN_RDX:%.*]] = add <vscale x 4 x i32> [[PARTIAL_REDUCE5]], [[PARTIAL_REDUCE]]
-; CHECK-NEXT: [[TMP20:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[BIN_RDX]])
+; CHECK-NEXT: [[TMP21:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[BIN_RDX]])
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH:%.*]]
; CHECK: scalar.ph:
@@ -173,8 +173,8 @@ define i32 @usdot(ptr %a, ptr %b) #0 {
; CHECK-NOI8MM-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-NOI8MM: vector.body:
; CHECK-NOI8MM-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 8 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP22:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 8 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP23:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 8 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP18:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NOI8MM-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 8 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP20:%.*]], [[VECTOR_BODY]] ]
; CHECK-NOI8MM-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; CHECK-NOI8MM-NEXT: [[TMP8:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NOI8MM-NEXT: [[TMP9:%.*]] = shl nuw i64 [[TMP8]], 3
@@ -182,25 +182,25 @@ define i32 @usdot(ptr %a, ptr %b) #0 {
; CHECK-NOI8MM-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP6]], align 1
; CHECK-NOI8MM-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP10]], align 1
; CHECK-NOI8MM-NEXT: [[TMP11:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP12:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP7:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
; CHECK-NOI8MM-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; CHECK-NOI8MM-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NOI8MM-NEXT: [[TMP16:%.*]] = shl nuw i64 [[TMP15]], 3
; CHECK-NOI8MM-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP13]], i64 [[TMP16]]
; CHECK-NOI8MM-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i8>, ptr [[TMP13]], align 1
; CHECK-NOI8MM-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 8 x i8>, ptr [[TMP17]], align 1
-; CHECK-NOI8MM-NEXT: [[TMP18:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD3]] to <vscale x 8 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP19:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD4]] to <vscale x 8 x i32>
-; CHECK-NOI8MM-NEXT: [[TMP20:%.*]] = mul <vscale x 8 x i32> [[TMP18]], [[TMP11]]
-; CHECK-NOI8MM-NEXT: [[TMP21:%.*]] = mul <vscale x 8 x i32> [[TMP19]], [[TMP12]]
-; CHECK-NOI8MM-NEXT: [[TMP22]] = add <vscale x 8 x i32> [[TMP20]], [[VEC_PHI]]
-; CHECK-NOI8MM-NEXT: [[TMP23]] = add <vscale x 8 x i32> [[TMP21]], [[VEC_PHI1]]
+; CHECK-NOI8MM-NEXT: [[TMP12:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD3]] to <vscale x 8 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP21:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD4]] to <vscale x 8 x i32>
+; CHECK-NOI8MM-NEXT: [[TMP14:%.*]] = mul <vscale x 8 x i32> [[TMP12]], [[TMP11]]
+; CHECK-NOI8MM-NEXT: [[TMP22:%.*]] = mul <vscale x 8 x i32> [[TMP21]], [[TMP7]]
+; CHECK-NOI8MM-NEXT: [[TMP18]] = add <vscale x 8 x i32> [[TMP14]], [[VEC_PHI]]
+; CHECK-NOI8MM-NEXT: [[TMP20]] = add <vscale x 8 x i32> [[TMP22]], [[VEC_PHI1]]
; CHECK-NOI8MM-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
; CHECK-NOI8MM-NEXT: [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NOI8MM-NEXT: br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK-NOI8MM: middle.block:
-; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <vscale x 8 x i32> [[TMP23]], [[TMP22]]
-; CHECK-NOI8MM-NEXT: [[TMP25:%.*]] = call i32 @llvm.vector.reduce.add.nxv8i32(<vscale x 8 x i32> [[BIN_RDX]])
+; CHECK-NOI8MM-NEXT: [[BIN_RDX:%.*]] = add <vscale x 8 x i32> [[TMP20]], [[TMP18]]
+; CHECK-NOI8MM-NEXT: [[TMP19:%.*]] = call i32 @llvm.vector.reduce.add.nxv8i32(<vscale x 8 x i32> [[BIN_RDX]])
; CHECK-NOI8MM-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
; CHECK-NOI8MM-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH:%.*]]
; CHECK-NOI8MM: scalar.ph:
@@ -242,18 +242,18 @@ define i32 @sudot_neon(ptr %a, ptr %b) #1 {
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP0]], i32 16
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP0]], align 1
; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1
-; CHECK-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
-; CHECK-NEXT: [[TMP4:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP5]], i32 16
; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP5]], align 1
; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
-; CHECK-NEXT: [[TMP8:%.*]] = sext <16 x i8> [[WIDE_LOAD3]] to <16 x i32>
-; CHECK-NEXT: [[TMP9:%.*]] = sext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>
-; CHECK-NEXT: [[TMP10:%.*]] = mul <16 x i32> [[TMP8]], [[TMP3]]
+; CHECK-NEXT: [[TMP9:%.*]] = sext <16 x i8> [[WIDE_LOAD3]] to <16 x i32>
+; CHECK-NEXT: [[TMP4:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-NEXT: [[TMP11:%.*]] = mul <16 x i32> [[TMP9]], [[TMP4]]
-; CHECK-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP10]])
-; CHECK-NEXT: [[PARTIAL_REDUCE5]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP11]])
+; CHECK-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP11]])
+; CHECK-NEXT: [[TMP10:%.*]] = sext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>
+; CHECK-NEXT: [[TMP8:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
+; CHECK-NEXT: [[TMP14:%.*]] = mul <16 x i32> [[TMP10]], [[TMP8]]
+; CHECK-NEXT: [[PARTIAL_REDUCE5]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP14]])
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
; CHECK-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
@@ -337,18 +337,18 @@ define i32 @usdot_neon(ptr %a, ptr %b) #1 {
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP0]], i32 16
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP0]], align 1
; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1
-; CHECK-NEXT: [[TMP3:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
-; CHECK-NEXT: [[TMP4:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP5]], i32 16
; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP5]], align 1
; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
-; CHECK-NEXT: [[TMP8:%.*]] = zext <16 x i8> [[WIDE_LOAD3]] to <16 x i32>
-; CHECK-NEXT: [[TMP9:%.*]] = zext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>
-; CHECK-NEXT: [[TMP10:%.*]] = mul <16 x i32> [[TMP8]], [[TMP3]]
+; CHECK-NEXT: [[TMP9:%.*]] = zext <16 x i8> [[WIDE_LOAD3]] to <16 x i32>
+; CHECK-NEXT: [[TMP4:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-NEXT: [[TMP11:%.*]] = mul <16 x i32> [[TMP9]], [[TMP4]]
-; CHECK-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP10]])
-; CHECK-NEXT: [[PARTIAL_REDUCE5]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP11]])
+; CHECK-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP11]])
+; CHECK-NEXT: [[TMP10:%.*]] = zext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>
+; CHECK-NEXT: [[TMP8:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
+; CHECK-NEXT: [[TMP14:%.*]] = mul <16 x i32> [[TMP10]], [[TMP8]]
+; CHECK-NEXT: [[PARTIAL_REDUCE5]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP14]])
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
; CHECK-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]]
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-neon.ll b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-neon.ll
index 801eb810d8625..b84763142b686 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-neon.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product-neon.ll
@@ -18,10 +18,10 @@ define i32 @dotp(ptr %a, ptr %b) {
; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
; CHECK-INTERLEAVE1-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1
-; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[TMP4:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD1:%.*]] = load <16 x i8>, ptr [[TMP4]], align 1
; CHECK-INTERLEAVE1-NEXT: [[TMP6:%.*]] = zext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
+; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[TMP7:%.*]] = mul <16 x i32> [[TMP6]], [[TMP3]]
; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP7]])
; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
@@ -47,17 +47,17 @@ define i32 @dotp(ptr %a, ptr %b) {
; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[TMP1]], i32 16
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP3]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[TMP6]], i32 16
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP6]], align 1
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP8]], align 1
; CHECK-INTERLEAVED-NEXT: [[TMP9:%.*]] = zext <16 x i8> [[WIDE_LOAD3]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = zext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = mul <16 x i32> [[TMP9]], [[TMP4]]
-; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = mul <16 x i32> [[TMP10]], [[TMP5]]
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP11]])
+; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = zext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = mul <16 x i32> [[TMP7]], [[TMP10]]
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE5]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP12]])
; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
@@ -80,10 +80,10 @@ define i32 @dotp(ptr %a, ptr %b) {
; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
; CHECK-MAXBW-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1
-; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-MAXBW-NEXT: [[TMP4:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; CHECK-MAXBW-NEXT: [[WIDE_LOAD1:%.*]] = load <16 x i8>, ptr [[TMP4]], align 1
; CHECK-MAXBW-NEXT: [[TMP6:%.*]] = zext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
+; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-MAXBW-NEXT: [[TMP7:%.*]] = mul <16 x i32> [[TMP6]], [[TMP3]]
; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP7]])
; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
@@ -720,26 +720,26 @@ define i32 @dotp_unrolled(i32 %num_out, i64 %num_in, ptr %a, ptr %b) {
; CHECK-INTERLEAVE1-NEXT: [[TMP10:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[TMP9]]
; CHECK-INTERLEAVE1-NEXT: [[TMP11:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[TMP9]]
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1
-; CHECK-INTERLEAVE1-NEXT: [[TMP13:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1
; CHECK-INTERLEAVE1-NEXT: [[TMP15:%.*]] = sext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>
+; CHECK-INTERLEAVE1-NEXT: [[TMP13:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[TMP16:%.*]] = mul nsw <16 x i32> [[TMP15]], [[TMP13]]
; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI3]], <16 x i32> [[TMP16]])
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD5:%.*]] = load <16 x i8>, ptr [[TMP4]], align 1
-; CHECK-INTERLEAVE1-NEXT: [[TMP18:%.*]] = sext <16 x i8> [[WIDE_LOAD5]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD6:%.*]] = load <16 x i8>, ptr [[TMP5]], align 1
+; CHECK-INTERLEAVE1-NEXT: [[TMP18:%.*]] = sext <16 x i8> [[WIDE_LOAD5]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[TMP20:%.*]] = sext <16 x i8> [[WIDE_LOAD6]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[TMP21:%.*]] = mul nsw <16 x i32> [[TMP18]], [[TMP20]]
; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE7]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI2]], <16 x i32> [[TMP21]])
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD8:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
-; CHECK-INTERLEAVE1-NEXT: [[TMP23:%.*]] = sext <16 x i8> [[WIDE_LOAD8]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD9:%.*]] = load <16 x i8>, ptr [[TMP8]], align 1
+; CHECK-INTERLEAVE1-NEXT: [[TMP23:%.*]] = sext <16 x i8> [[WIDE_LOAD8]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[TMP25:%.*]] = sext <16 x i8> [[WIDE_LOAD9]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[TMP26:%.*]] = mul nsw <16 x i32> [[TMP23]], [[TMP25]]
; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE10]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP26]])
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD11:%.*]] = load <16 x i8>, ptr [[TMP10]], align 1
-; CHECK-INTERLEAVE1-NEXT: [[TMP28:%.*]] = sext <16 x i8> [[WIDE_LOAD11]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD12:%.*]] = load <16 x i8>, ptr [[TMP11]], align 1
+; CHECK-INTERLEAVE1-NEXT: [[TMP28:%.*]] = sext <16 x i8> [[WIDE_LOAD11]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[TMP30:%.*]] = sext <16 x i8> [[WIDE_LOAD12]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[TMP31:%.*]] = mul nsw <16 x i32> [[TMP28]], [[TMP30]]
; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE13]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP31]])
@@ -788,59 +788,59 @@ define i32 @dotp_unrolled(i32 %num_out, i64 %num_in, ptr %a, ptr %b) {
; CHECK-INTERLEAVED-NEXT: [[TMP38:%.*]] = getelementptr inbounds i8, ptr [[TMP1]], i32 16
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD8:%.*]] = load <16 x i8>, ptr [[TMP38]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = sext <16 x i8> [[WIDE_LOAD8]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: [[TMP16:%.*]] = getelementptr inbounds i8, ptr [[TMP2]], i32 16
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD9:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD10:%.*]] = load <16 x i8>, ptr [[TMP16]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP17:%.*]] = sext <16 x i8> [[WIDE_LOAD9]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP18:%.*]] = sext <16 x i8> [[WIDE_LOAD10]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP19:%.*]] = mul nsw <16 x i32> [[TMP17]], [[TMP13]]
+; CHECK-INTERLEAVED-NEXT: [[TMP18:%.*]] = sext <16 x i8> [[WIDE_LOAD9]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: [[TMP20:%.*]] = mul nsw <16 x i32> [[TMP18]], [[TMP14]]
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE1]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI6]], <16 x i32> [[TMP19]])
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE11]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI7]], <16 x i32> [[TMP20]])
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE1]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI6]], <16 x i32> [[TMP20]])
+; CHECK-INTERLEAVED-NEXT: [[TMP19:%.*]] = sext <16 x i8> [[WIDE_LOAD10]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP17:%.*]] = sext <16 x i8> [[WIDE_LOAD8]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP21:%.*]] = mul nsw <16 x i32> [[TMP19]], [[TMP17]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE11]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI7]], <16 x i32> [[TMP21]])
; CHECK-INTERLEAVED-NEXT: [[TMP22:%.*]] = getelementptr inbounds i8, ptr [[TMP4]], i32 16
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD12:%.*]] = load <16 x i8>, ptr [[TMP4]], align 1
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD13:%.*]] = load <16 x i8>, ptr [[TMP22]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP23:%.*]] = sext <16 x i8> [[WIDE_LOAD12]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP24:%.*]] = sext <16 x i8> [[WIDE_LOAD13]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: [[TMP26:%.*]] = getelementptr inbounds i8, ptr [[TMP5]], i32 16
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD14:%.*]] = load <16 x i8>, ptr [[TMP5]], align 1
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD15:%.*]] = load <16 x i8>, ptr [[TMP26]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP27:%.*]] = sext <16 x i8> [[WIDE_LOAD14]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP28:%.*]] = sext <16 x i8> [[WIDE_LOAD15]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP29:%.*]] = mul nsw <16 x i32> [[TMP23]], [[TMP27]]
+; CHECK-INTERLEAVED-NEXT: [[TMP24:%.*]] = sext <16 x i8> [[WIDE_LOAD12]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP28:%.*]] = sext <16 x i8> [[WIDE_LOAD14]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: [[TMP30:%.*]] = mul nsw <16 x i32> [[TMP24]], [[TMP28]]
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE16]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI4]], <16 x i32> [[TMP29]])
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE17]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI5]], <16 x i32> [[TMP30]])
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE16]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI4]], <16 x i32> [[TMP30]])
+; CHECK-INTERLEAVED-NEXT: [[TMP27:%.*]] = sext <16 x i8> [[WIDE_LOAD13]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP25:%.*]] = sext <16 x i8> [[WIDE_LOAD15]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP29:%.*]] = mul nsw <16 x i32> [[TMP27]], [[TMP25]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE17]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI5]], <16 x i32> [[TMP29]])
; CHECK-INTERLEAVED-NEXT: [[TMP32:%.*]] = getelementptr inbounds i8, ptr [[TMP7]], i32 16
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD18:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD19:%.*]] = load <16 x i8>, ptr [[TMP32]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP33:%.*]] = sext <16 x i8> [[WIDE_LOAD18]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP34:%.*]] = sext <16 x i8> [[WIDE_LOAD19]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: [[TMP36:%.*]] = getelementptr inbounds i8, ptr [[TMP8]], i32 16
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD20:%.*]] = load <16 x i8>, ptr [[TMP8]], align 1
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD21:%.*]] = load <16 x i8>, ptr [[TMP36]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP37:%.*]] = sext <16 x i8> [[WIDE_LOAD20]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP56:%.*]] = sext <16 x i8> [[WIDE_LOAD21]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP39:%.*]] = mul nsw <16 x i32> [[TMP33]], [[TMP37]]
+; CHECK-INTERLEAVED-NEXT: [[TMP34:%.*]] = sext <16 x i8> [[WIDE_LOAD18]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP56:%.*]] = sext <16 x i8> [[WIDE_LOAD20]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: [[TMP40:%.*]] = mul nsw <16 x i32> [[TMP34]], [[TMP56]]
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE7]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI2]], <16 x i32> [[TMP39]])
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI3]], <16 x i32> [[TMP40]])
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE7]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI2]], <16 x i32> [[TMP40]])
+; CHECK-INTERLEAVED-NEXT: [[TMP35:%.*]] = sext <16 x i8> [[WIDE_LOAD19]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP33:%.*]] = sext <16 x i8> [[WIDE_LOAD21]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP37:%.*]] = mul nsw <16 x i32> [[TMP35]], [[TMP33]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI3]], <16 x i32> [[TMP37]])
; CHECK-INTERLEAVED-NEXT: [[TMP42:%.*]] = getelementptr inbounds i8, ptr [[TMP10]], i32 16
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD24:%.*]] = load <16 x i8>, ptr [[TMP10]], align 1
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD25:%.*]] = load <16 x i8>, ptr [[TMP42]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP43:%.*]] = sext <16 x i8> [[WIDE_LOAD24]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP44:%.*]] = sext <16 x i8> [[WIDE_LOAD25]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: [[TMP46:%.*]] = getelementptr inbounds i8, ptr [[TMP11]], i32 16
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD26:%.*]] = load <16 x i8>, ptr [[TMP11]], align 1
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD27:%.*]] = load <16 x i8>, ptr [[TMP46]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP47:%.*]] = sext <16 x i8> [[WIDE_LOAD26]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP48:%.*]] = sext <16 x i8> [[WIDE_LOAD27]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP49:%.*]] = mul nsw <16 x i32> [[TMP43]], [[TMP47]]
+; CHECK-INTERLEAVED-NEXT: [[TMP44:%.*]] = sext <16 x i8> [[WIDE_LOAD24]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP48:%.*]] = sext <16 x i8> [[WIDE_LOAD26]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: [[TMP50:%.*]] = mul nsw <16 x i32> [[TMP44]], [[TMP48]]
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE13]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP49]])
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE10]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP50]])
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE13]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP50]])
+; CHECK-INTERLEAVED-NEXT: [[TMP43:%.*]] = sext <16 x i8> [[WIDE_LOAD25]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP41:%.*]] = sext <16 x i8> [[WIDE_LOAD27]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP45:%.*]] = mul nsw <16 x i32> [[TMP43]], [[TMP41]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE10]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP45]])
; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
; CHECK-INTERLEAVED-NEXT: [[TMP51:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-INTERLEAVED-NEXT: br i1 [[TMP51]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
@@ -884,26 +884,26 @@ define i32 @dotp_unrolled(i32 %num_out, i64 %num_in, ptr %a, ptr %b) {
; CHECK-MAXBW-NEXT: [[TMP10:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[TMP9]]
; CHECK-MAXBW-NEXT: [[TMP11:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[TMP9]]
; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1
-; CHECK-MAXBW-NEXT: [[TMP13:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-MAXBW-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1
; CHECK-MAXBW-NEXT: [[TMP15:%.*]] = sext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>
+; CHECK-MAXBW-NEXT: [[TMP13:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-MAXBW-NEXT: [[TMP16:%.*]] = mul nsw <16 x i32> [[TMP15]], [[TMP13]]
; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI3]], <16 x i32> [[TMP16]])
; CHECK-MAXBW-NEXT: [[WIDE_LOAD5:%.*]] = load <16 x i8>, ptr [[TMP4]], align 1
-; CHECK-MAXBW-NEXT: [[TMP18:%.*]] = sext <16 x i8> [[WIDE_LOAD5]] to <16 x i32>
; CHECK-MAXBW-NEXT: [[WIDE_LOAD6:%.*]] = load <16 x i8>, ptr [[TMP5]], align 1
+; CHECK-MAXBW-NEXT: [[TMP18:%.*]] = sext <16 x i8> [[WIDE_LOAD5]] to <16 x i32>
; CHECK-MAXBW-NEXT: [[TMP20:%.*]] = sext <16 x i8> [[WIDE_LOAD6]] to <16 x i32>
; CHECK-MAXBW-NEXT: [[TMP21:%.*]] = mul nsw <16 x i32> [[TMP18]], [[TMP20]]
; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE7]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI2]], <16 x i32> [[TMP21]])
; CHECK-MAXBW-NEXT: [[WIDE_LOAD8:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
-; CHECK-MAXBW-NEXT: [[TMP23:%.*]] = sext <16 x i8> [[WIDE_LOAD8]] to <16 x i32>
; CHECK-MAXBW-NEXT: [[WIDE_LOAD9:%.*]] = load <16 x i8>, ptr [[TMP8]], align 1
+; CHECK-MAXBW-NEXT: [[TMP23:%.*]] = sext <16 x i8> [[WIDE_LOAD8]] to <16 x i32>
; CHECK-MAXBW-NEXT: [[TMP25:%.*]] = sext <16 x i8> [[WIDE_LOAD9]] to <16 x i32>
; CHECK-MAXBW-NEXT: [[TMP26:%.*]] = mul nsw <16 x i32> [[TMP23]], [[TMP25]]
; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE10]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP26]])
; CHECK-MAXBW-NEXT: [[WIDE_LOAD11:%.*]] = load <16 x i8>, ptr [[TMP10]], align 1
-; CHECK-MAXBW-NEXT: [[TMP28:%.*]] = sext <16 x i8> [[WIDE_LOAD11]] to <16 x i32>
; CHECK-MAXBW-NEXT: [[WIDE_LOAD12:%.*]] = load <16 x i8>, ptr [[TMP11]], align 1
+; CHECK-MAXBW-NEXT: [[TMP28:%.*]] = sext <16 x i8> [[WIDE_LOAD11]] to <16 x i32>
; CHECK-MAXBW-NEXT: [[TMP30:%.*]] = sext <16 x i8> [[WIDE_LOAD12]] to <16 x i32>
; CHECK-MAXBW-NEXT: [[TMP31:%.*]] = mul nsw <16 x i32> [[TMP28]], [[TMP30]]
; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE13]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP31]])
@@ -2025,7 +2025,7 @@ define i32 @not_dotp_extend_user(ptr %a, ptr %b) {
; CHECK-INTERLEAVE1-NEXT: [[TMP8]] = add <16 x i32> [[TMP7]], [[VEC_PHI]]
; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-INTERLEAVE1-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
-; CHECK-INTERLEAVE1-NEXT: br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
+; CHECK-INTERLEAVE1-NEXT: br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
; CHECK-INTERLEAVE1: middle.block:
; CHECK-INTERLEAVE1-NEXT: [[TMP10:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[TMP8]])
; CHECK-INTERLEAVE1-NEXT: [[TMP11:%.*]] = extractelement <16 x i32> [[TMP6]], i32 15
@@ -2062,7 +2062,7 @@ define i32 @not_dotp_extend_user(ptr %a, ptr %b) {
; CHECK-INTERLEAVED-NEXT: [[TMP14]] = add <16 x i32> [[TMP12]], [[VEC_PHI1]]
; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
; CHECK-INTERLEAVED-NEXT: [[TMP15:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
-; CHECK-INTERLEAVED-NEXT: br i1 [[TMP15]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
+; CHECK-INTERLEAVED-NEXT: br i1 [[TMP15]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
; CHECK-INTERLEAVED: middle.block:
; CHECK-INTERLEAVED-NEXT: [[BIN_RDX:%.*]] = add <16 x i32> [[TMP14]], [[TMP13]]
; CHECK-INTERLEAVED-NEXT: [[TMP16:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[BIN_RDX]])
@@ -2091,7 +2091,7 @@ define i32 @not_dotp_extend_user(ptr %a, ptr %b) {
; CHECK-MAXBW-NEXT: [[TMP8]] = add <16 x i32> [[TMP7]], [[VEC_PHI]]
; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-MAXBW-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
-; CHECK-MAXBW-NEXT: br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
+; CHECK-MAXBW-NEXT: br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
; CHECK-MAXBW: middle.block:
; CHECK-MAXBW-NEXT: [[TMP10:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[TMP8]])
; CHECK-MAXBW-NEXT: [[TMP11:%.*]] = extractelement <16 x i32> [[TMP6]], i32 15
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product.ll b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product.ll
index 1ace7d44125b9..4636c1b63da82 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-dot-product.ll
@@ -18,11 +18,11 @@ define i32 @dotp(ptr %a, ptr %b) #0 {
; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
; CHECK-INTERLEAVE1-NEXT: [[TMP16:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX1]]
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP16]], align 1
-; CHECK-INTERLEAVE1-NEXT: [[TMP1:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[TMP20:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX1]]
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD1:%.*]] = load <16 x i8>, ptr [[TMP20]], align 1
-; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
-; CHECK-INTERLEAVE1-NEXT: [[TMP4:%.*]] = mul <16 x i32> [[TMP3]], [[TMP1]]
+; CHECK-INTERLEAVE1-NEXT: [[TMP2:%.*]] = zext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
+; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
+; CHECK-INTERLEAVE1-NEXT: [[TMP4:%.*]] = mul <16 x i32> [[TMP2]], [[TMP3]]
; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP4]])
; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT1]] = add nuw i64 [[INDEX1]], 16
; CHECK-INTERLEAVE1-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT1]], 1024
@@ -47,17 +47,17 @@ define i32 @dotp(ptr %a, ptr %b) #0 {
; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[TMP20]], i32 16
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP20]], align 1
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP2:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: [[TMP28:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX1]]
-; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[TMP28]], i32 16
+; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[TMP28]], i32 16
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP28]], align 1
-; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP5]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = zext <16 x i8> [[WIDE_LOAD3]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP3]], align 1
+; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = zext <16 x i8> [[WIDE_LOAD3]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = mul <16 x i32> [[TMP4]], [[TMP5]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP6]])
; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = zext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = mul <16 x i32> [[TMP6]], [[TMP2]]
-; CHECK-INTERLEAVED-NEXT: [[TMP9:%.*]] = mul <16 x i32> [[TMP7]], [[TMP3]]
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP8]])
+; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP9:%.*]] = mul <16 x i32> [[TMP7]], [[TMP8]]
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE5]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP9]])
; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT1]] = add nuw i64 [[INDEX1]], 32
; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT1]], 1024
@@ -75,26 +75,26 @@ define i32 @dotp(ptr %a, ptr %b) #0 {
; CHECK-MAXBW-NEXT: br label [[VECTOR_PH:%.*]]
; CHECK-MAXBW: vector.ph:
; CHECK-MAXBW-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 8
+; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 16
; CHECK-MAXBW-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
; CHECK-MAXBW-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
; CHECK-MAXBW-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-MAXBW: vector.body:
; CHECK-MAXBW-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-MAXBW-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ]
+; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
; CHECK-MAXBW-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
-; CHECK-MAXBW-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
-; CHECK-MAXBW-NEXT: [[TMP13:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i32>
+; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 16 x i8>, ptr [[TMP7]], align 1
; CHECK-MAXBW-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
-; CHECK-MAXBW-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 8 x i8>, ptr [[TMP14]], align 1
-; CHECK-MAXBW-NEXT: [[TMP20:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD4]] to <vscale x 8 x i32>
-; CHECK-MAXBW-NEXT: [[TMP22:%.*]] = mul <vscale x 8 x i32> [[TMP20]], [[TMP13]]
-; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE5]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI1]], <vscale x 8 x i32> [[TMP22]])
+; CHECK-MAXBW-NEXT: [[WIDE_LOAD1:%.*]] = load <vscale x 16 x i8>, ptr [[TMP14]], align 1
+; CHECK-MAXBW-NEXT: [[TMP4:%.*]] = zext <vscale x 16 x i8> [[WIDE_LOAD1]] to <vscale x 16 x i32>
+; CHECK-MAXBW-NEXT: [[TMP5:%.*]] = zext <vscale x 16 x i8> [[WIDE_LOAD]] to <vscale x 16 x i32>
+; CHECK-MAXBW-NEXT: [[TMP6:%.*]] = mul <vscale x 16 x i32> [[TMP4]], [[TMP5]]
+; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 4 x i32> @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32(<vscale x 4 x i32> [[VEC_PHI]], <vscale x 16 x i32> [[TMP6]])
; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
; CHECK-MAXBW-NEXT: [[TMP23:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-MAXBW-NEXT: br i1 [[TMP23]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK-MAXBW: middle.block:
-; CHECK-MAXBW-NEXT: [[TMP16:%.*]] = call i32 @llvm.vector.reduce.add.nxv2i32(<vscale x 2 x i32> [[PARTIAL_REDUCE5]])
+; CHECK-MAXBW-NEXT: [[TMP8:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[PARTIAL_REDUCE]])
; CHECK-MAXBW-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
; CHECK-MAXBW-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH:%.*]]
; CHECK-MAXBW: scalar.ph:
@@ -134,10 +134,10 @@ define i64 @not_dotp_i8_to_i64_has_neon_dotprod(ptr readonly %a, ptr readonly %b
; CHECK-INTERLEAVE1-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; CHECK-INTERLEAVE1-NEXT: [[NEXT_GEP1:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[NEXT_GEP]], align 1
-; CHECK-INTERLEAVE1-NEXT: [[TMP0:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i64>
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[NEXT_GEP1]], align 1
-; CHECK-INTERLEAVE1-NEXT: [[TMP1:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i64>
-; CHECK-INTERLEAVE1-NEXT: [[TMP2:%.*]] = mul nuw nsw <16 x i64> [[TMP1]], [[TMP0]]
+; CHECK-INTERLEAVE1-NEXT: [[TMP0:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i64>
+; CHECK-INTERLEAVE1-NEXT: [[TMP1:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i64>
+; CHECK-INTERLEAVE1-NEXT: [[TMP2:%.*]] = mul nuw nsw <16 x i64> [[TMP0]], [[TMP1]]
; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v16i64(<2 x i64> [[VEC_PHI]], <16 x i64> [[TMP2]])
; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
@@ -157,54 +157,78 @@ define i64 @not_dotp_i8_to_i64_has_neon_dotprod(ptr readonly %a, ptr readonly %b
; CHECK-INTERLEAVED: vector.body:
; CHECK-INTERLEAVED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
+; CHECK-INTERLEAVED-NEXT: [[VEC_PHI1:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE6:%.*]], [[VECTOR_BODY]] ]
+; CHECK-INTERLEAVED-NEXT: [[VEC_PHI2:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE13:%.*]], [[VECTOR_BODY]] ]
+; CHECK-INTERLEAVED-NEXT: [[VEC_PHI3:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE14:%.*]], [[VECTOR_BODY]] ]
; CHECK-INTERLEAVED-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; CHECK-INTERLEAVED-NEXT: [[NEXT_GEP1:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
+; CHECK-INTERLEAVED-NEXT: [[TMP0:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i32 16
+; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i32 32
+; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i32 48
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[NEXT_GEP]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP0:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i64>
-; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[NEXT_GEP1]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i64>
-; CHECK-INTERLEAVED-NEXT: [[TMP2:%.*]] = mul nuw nsw <16 x i64> [[TMP1]], [[TMP0]]
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v16i64(<2 x i64> [[VEC_PHI]], <16 x i64> [[TMP2]])
-; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
-; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
-; CHECK-INTERLEAVED-NEXT: br i1 [[TMP3]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP0]], align 1
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD6:%.*]] = load <16 x i8>, ptr [[TMP10]], align 1
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD7:%.*]] = load <16 x i8>, ptr [[TMP11]], align 1
+; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[NEXT_GEP1]], i32 16
+; CHECK-INTERLEAVED-NEXT: [[TMP18:%.*]] = getelementptr i8, ptr [[NEXT_GEP1]], i32 32
+; CHECK-INTERLEAVED-NEXT: [[TMP19:%.*]] = getelementptr i8, ptr [[NEXT_GEP1]], i32 48
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[NEXT_GEP1]], align 1
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD5:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD10:%.*]] = load <16 x i8>, ptr [[TMP18]], align 1
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD11:%.*]] = load <16 x i8>, ptr [[TMP19]], align 1
+; CHECK-INTERLEAVED-NEXT: [[TMP2:%.*]] = zext <16 x i8> [[WIDE_LOAD4]] to <16 x i64>
+; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i64>
+; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = mul nuw nsw <16 x i64> [[TMP2]], [[TMP3]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v16i64(<2 x i64> [[VEC_PHI]], <16 x i64> [[TMP4]])
+; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = zext <16 x i8> [[WIDE_LOAD5]] to <16 x i64>
+; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = zext <16 x i8> [[WIDE_LOAD3]] to <16 x i64>
+; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = mul nuw nsw <16 x i64> [[TMP5]], [[TMP6]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE6]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v16i64(<2 x i64> [[VEC_PHI1]], <16 x i64> [[TMP7]])
+; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = zext <16 x i8> [[WIDE_LOAD10]] to <16 x i64>
+; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = zext <16 x i8> [[WIDE_LOAD6]] to <16 x i64>
+; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = mul nuw nsw <16 x i64> [[TMP12]], [[TMP13]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE13]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v16i64(<2 x i64> [[VEC_PHI2]], <16 x i64> [[TMP14]])
+; CHECK-INTERLEAVED-NEXT: [[TMP15:%.*]] = zext <16 x i8> [[WIDE_LOAD11]] to <16 x i64>
+; CHECK-INTERLEAVED-NEXT: [[TMP16:%.*]] = zext <16 x i8> [[WIDE_LOAD7]] to <16 x i64>
+; CHECK-INTERLEAVED-NEXT: [[TMP17:%.*]] = mul nuw nsw <16 x i64> [[TMP15]], [[TMP16]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE14]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v16i64(<2 x i64> [[VEC_PHI3]], <16 x i64> [[TMP17]])
+; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 64
+; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
+; CHECK-INTERLEAVED-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
; CHECK-INTERLEAVED: middle.block:
-; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[PARTIAL_REDUCE]])
+; CHECK-INTERLEAVED-NEXT: [[BIN_RDX:%.*]] = add <2 x i64> [[PARTIAL_REDUCE6]], [[PARTIAL_REDUCE]]
+; CHECK-INTERLEAVED-NEXT: [[BIN_RDX15:%.*]] = add <2 x i64> [[PARTIAL_REDUCE13]], [[BIN_RDX]]
+; CHECK-INTERLEAVED-NEXT: [[BIN_RDX16:%.*]] = add <2 x i64> [[PARTIAL_REDUCE14]], [[BIN_RDX15]]
+; CHECK-INTERLEAVED-NEXT: [[TMP9:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[BIN_RDX16]])
; CHECK-INTERLEAVED-NEXT: br label [[FOR_EXIT:%.*]]
; CHECK-INTERLEAVED: for.exit:
-; CHECK-INTERLEAVED-NEXT: ret i64 [[TMP4]]
+; CHECK-INTERLEAVED-NEXT: ret i64 [[TMP9]]
;
; CHECK-MAXBW-LABEL: define i64 @not_dotp_i8_to_i64_has_neon_dotprod(
; CHECK-MAXBW-SAME: ptr readonly [[A:%.*]], ptr readonly [[B:%.*]]) #[[ATTR1:[0-9]+]] {
; CHECK-MAXBW-NEXT: entry:
; CHECK-MAXBW-NEXT: br label [[VECTOR_PH:%.*]]
; CHECK-MAXBW: vector.ph:
-; CHECK-MAXBW-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 8
-; CHECK-MAXBW-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
-; CHECK-MAXBW-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
-; CHECK-MAXBW-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[A]], i64 [[N_VEC]]
-; CHECK-MAXBW-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[B]], i64 [[N_VEC]]
; CHECK-MAXBW-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-MAXBW: vector.body:
; CHECK-MAXBW-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 8 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP15:%.*]], [[VECTOR_BODY]] ]
+; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
; CHECK-MAXBW-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; CHECK-MAXBW-NEXT: [[NEXT_GEP1:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
-; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[NEXT_GEP]], align 1
-; CHECK-MAXBW-NEXT: [[TMP11:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i64>
-; CHECK-MAXBW-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[NEXT_GEP1]], align 1
-; CHECK-MAXBW-NEXT: [[TMP13:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i64>
-; CHECK-MAXBW-NEXT: [[TMP14:%.*]] = mul nuw nsw <vscale x 8 x i64> [[TMP13]], [[TMP11]]
-; CHECK-MAXBW-NEXT: [[TMP15]] = add <vscale x 8 x i64> [[TMP14]], [[VEC_PHI]]
-; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
-; CHECK-MAXBW-NEXT: [[TMP16:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-MAXBW-NEXT: br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
+; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[NEXT_GEP]], align 1
+; CHECK-MAXBW-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[NEXT_GEP1]], align 1
+; CHECK-MAXBW-NEXT: [[TMP0:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i64>
+; CHECK-MAXBW-NEXT: [[TMP1:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i64>
+; CHECK-MAXBW-NEXT: [[TMP2:%.*]] = mul nuw nsw <16 x i64> [[TMP0]], [[TMP1]]
+; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v16i64(<2 x i64> [[VEC_PHI]], <16 x i64> [[TMP2]])
+; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
+; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
+; CHECK-MAXBW-NEXT: br i1 [[TMP3]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK-MAXBW: middle.block:
-; CHECK-MAXBW-NEXT: [[TMP17:%.*]] = call i64 @llvm.vector.reduce.add.nxv8i64(<vscale x 8 x i64> [[TMP15]])
-; CHECK-MAXBW-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
-; CHECK-MAXBW-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH:%.*]]
-; CHECK-MAXBW: scalar.ph:
+; CHECK-MAXBW-NEXT: [[TMP4:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[PARTIAL_REDUCE]])
+; CHECK-MAXBW-NEXT: br label [[FOR_EXIT:%.*]]
+; CHECK-MAXBW: for.exit:
+; CHECK-MAXBW-NEXT: ret i64 [[TMP4]]
;
entry:
br label %for.body
@@ -245,10 +269,10 @@ define i64 @not_dotp_i16_to_i64_has_neon_dotprod(ptr readonly %a, ptr readonly %
; CHECK-INTERLEAVE1-NEXT: [[OFFSET_IDX1:%.*]] = mul i64 [[INDEX]], 2
; CHECK-INTERLEAVE1-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr [[B]], i64 [[OFFSET_IDX1]]
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i16>, ptr [[NEXT_GEP]], align 2
-; CHECK-INTERLEAVE1-NEXT: [[TMP0:%.*]] = zext <8 x i16> [[WIDE_LOAD]] to <8 x i64>
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD3:%.*]] = load <8 x i16>, ptr [[NEXT_GEP2]], align 2
-; CHECK-INTERLEAVE1-NEXT: [[TMP1:%.*]] = zext <8 x i16> [[WIDE_LOAD3]] to <8 x i64>
-; CHECK-INTERLEAVE1-NEXT: [[TMP2:%.*]] = mul nuw nsw <8 x i64> [[TMP1]], [[TMP0]]
+; CHECK-INTERLEAVE1-NEXT: [[TMP0:%.*]] = zext <8 x i16> [[WIDE_LOAD3]] to <8 x i64>
+; CHECK-INTERLEAVE1-NEXT: [[TMP1:%.*]] = zext <8 x i16> [[WIDE_LOAD]] to <8 x i64>
+; CHECK-INTERLEAVE1-NEXT: [[TMP2:%.*]] = mul nuw nsw <8 x i64> [[TMP0]], [[TMP1]]
; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v8i64(<2 x i64> [[VEC_PHI]], <8 x i64> [[TMP2]])
; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
@@ -269,30 +293,50 @@ define i64 @not_dotp_i16_to_i64_has_neon_dotprod(ptr readonly %a, ptr readonly %
; CHECK-INTERLEAVED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
; CHECK-INTERLEAVED-NEXT: [[VEC_PHI1:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE7:%.*]], [[VECTOR_BODY]] ]
+; CHECK-INTERLEAVED-NEXT: [[VEC_PHI2:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE14:%.*]], [[VECTOR_BODY]] ]
+; CHECK-INTERLEAVED-NEXT: [[VEC_PHI3:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE15:%.*]], [[VECTOR_BODY]] ]
; CHECK-INTERLEAVED-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 2
; CHECK-INTERLEAVED-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[A]], i64 [[OFFSET_IDX]]
; CHECK-INTERLEAVED-NEXT: [[OFFSET_IDX2:%.*]] = mul i64 [[INDEX]], 2
; CHECK-INTERLEAVED-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr [[B]], i64 [[OFFSET_IDX2]]
; CHECK-INTERLEAVED-NEXT: [[TMP0:%.*]] = getelementptr i16, ptr [[NEXT_GEP]], i32 8
+; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = getelementptr i16, ptr [[NEXT_GEP]], i32 16
+; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = getelementptr i16, ptr [[NEXT_GEP]], i32 24
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i16>, ptr [[NEXT_GEP]], align 2
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD4:%.*]] = load <8 x i16>, ptr [[TMP0]], align 2
-; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = zext <8 x i16> [[WIDE_LOAD]] to <8 x i64>
-; CHECK-INTERLEAVED-NEXT: [[TMP2:%.*]] = zext <8 x i16> [[WIDE_LOAD4]] to <8 x i64>
-; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = getelementptr i16, ptr [[NEXT_GEP3]], i32 8
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD7:%.*]] = load <8 x i16>, ptr [[TMP10]], align 2
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD8:%.*]] = load <8 x i16>, ptr [[TMP11]], align 2
+; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = getelementptr i16, ptr [[NEXT_GEP3]], i32 8
+; CHECK-INTERLEAVED-NEXT: [[TMP18:%.*]] = getelementptr i16, ptr [[NEXT_GEP3]], i32 16
+; CHECK-INTERLEAVED-NEXT: [[TMP19:%.*]] = getelementptr i16, ptr [[NEXT_GEP3]], i32 24
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD5:%.*]] = load <8 x i16>, ptr [[NEXT_GEP3]], align 2
-; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD6:%.*]] = load <8 x i16>, ptr [[TMP3]], align 2
-; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = zext <8 x i16> [[WIDE_LOAD5]] to <8 x i64>
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD6:%.*]] = load <8 x i16>, ptr [[TMP1]], align 2
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD11:%.*]] = load <8 x i16>, ptr [[TMP18]], align 2
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD12:%.*]] = load <8 x i16>, ptr [[TMP19]], align 2
+; CHECK-INTERLEAVED-NEXT: [[TMP2:%.*]] = zext <8 x i16> [[WIDE_LOAD5]] to <8 x i64>
+; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = zext <8 x i16> [[WIDE_LOAD]] to <8 x i64>
+; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = mul nuw nsw <8 x i64> [[TMP2]], [[TMP3]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v8i64(<2 x i64> [[VEC_PHI]], <8 x i64> [[TMP4]])
; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = zext <8 x i16> [[WIDE_LOAD6]] to <8 x i64>
-; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = mul nuw nsw <8 x i64> [[TMP4]], [[TMP1]]
-; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = mul nuw nsw <8 x i64> [[TMP5]], [[TMP2]]
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v8i64(<2 x i64> [[VEC_PHI]], <8 x i64> [[TMP6]])
+; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = zext <8 x i16> [[WIDE_LOAD4]] to <8 x i64>
+; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = mul nuw nsw <8 x i64> [[TMP5]], [[TMP6]]
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE7]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v8i64(<2 x i64> [[VEC_PHI1]], <8 x i64> [[TMP7]])
-; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
+; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = zext <8 x i16> [[WIDE_LOAD11]] to <8 x i64>
+; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = zext <8 x i16> [[WIDE_LOAD7]] to <8 x i64>
+; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = mul nuw nsw <8 x i64> [[TMP12]], [[TMP13]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE14]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v8i64(<2 x i64> [[VEC_PHI2]], <8 x i64> [[TMP14]])
+; CHECK-INTERLEAVED-NEXT: [[TMP15:%.*]] = zext <8 x i16> [[WIDE_LOAD12]] to <8 x i64>
+; CHECK-INTERLEAVED-NEXT: [[TMP16:%.*]] = zext <8 x i16> [[WIDE_LOAD8]] to <8 x i64>
+; CHECK-INTERLEAVED-NEXT: [[TMP17:%.*]] = mul nuw nsw <8 x i64> [[TMP15]], [[TMP16]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE15]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v8i64(<2 x i64> [[VEC_PHI3]], <8 x i64> [[TMP17]])
+; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
; CHECK-INTERLEAVED-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK-INTERLEAVED: middle.block:
; CHECK-INTERLEAVED-NEXT: [[BIN_RDX:%.*]] = add <2 x i64> [[PARTIAL_REDUCE7]], [[PARTIAL_REDUCE]]
-; CHECK-INTERLEAVED-NEXT: [[TMP9:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[BIN_RDX]])
+; CHECK-INTERLEAVED-NEXT: [[BIN_RDX16:%.*]] = add <2 x i64> [[PARTIAL_REDUCE14]], [[BIN_RDX]]
+; CHECK-INTERLEAVED-NEXT: [[BIN_RDX17:%.*]] = add <2 x i64> [[PARTIAL_REDUCE15]], [[BIN_RDX16]]
+; CHECK-INTERLEAVED-NEXT: [[TMP9:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[BIN_RDX17]])
; CHECK-INTERLEAVED-NEXT: br label [[FOR_EXIT:%.*]]
; CHECK-INTERLEAVED: for.exit:
; CHECK-INTERLEAVED-NEXT: ret i64 [[TMP9]]
@@ -302,36 +346,28 @@ define i64 @not_dotp_i16_to_i64_has_neon_dotprod(ptr readonly %a, ptr readonly %
; CHECK-MAXBW-NEXT: entry:
; CHECK-MAXBW-NEXT: br label [[VECTOR_PH:%.*]]
; CHECK-MAXBW: vector.ph:
-; CHECK-MAXBW-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 4
-; CHECK-MAXBW-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
-; CHECK-MAXBW-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
-; CHECK-MAXBW-NEXT: [[TMP6:%.*]] = mul i64 [[N_VEC]], 2
-; CHECK-MAXBW-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP6]]
-; CHECK-MAXBW-NEXT: [[TMP8:%.*]] = mul i64 [[N_VEC]], 2
-; CHECK-MAXBW-NEXT: [[TMP9:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP8]]
; CHECK-MAXBW-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-MAXBW: vector.body:
; CHECK-MAXBW-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 4 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP17:%.*]], [[VECTOR_BODY]] ]
+; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
; CHECK-MAXBW-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 2
; CHECK-MAXBW-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[A]], i64 [[OFFSET_IDX]]
; CHECK-MAXBW-NEXT: [[OFFSET_IDX1:%.*]] = mul i64 [[INDEX]], 2
; CHECK-MAXBW-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr [[B]], i64 [[OFFSET_IDX1]]
-; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i16>, ptr [[NEXT_GEP]], align 2
-; CHECK-MAXBW-NEXT: [[TMP13:%.*]] = zext <vscale x 4 x i16> [[WIDE_LOAD]] to <vscale x 4 x i64>
-; CHECK-MAXBW-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 4 x i16>, ptr [[NEXT_GEP2]], align 2
-; CHECK-MAXBW-NEXT: [[TMP15:%.*]] = zext <vscale x 4 x i16> [[WIDE_LOAD3]] to <vscale x 4 x i64>
-; CHECK-MAXBW-NEXT: [[TMP16:%.*]] = mul nuw nsw <vscale x 4 x i64> [[TMP15]], [[TMP13]]
-; CHECK-MAXBW-NEXT: [[TMP17]] = add <vscale x 4 x i64> [[TMP16]], [[VEC_PHI]]
-; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
-; CHECK-MAXBW-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-MAXBW-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
+; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i16>, ptr [[NEXT_GEP]], align 2
+; CHECK-MAXBW-NEXT: [[WIDE_LOAD3:%.*]] = load <8 x i16>, ptr [[NEXT_GEP2]], align 2
+; CHECK-MAXBW-NEXT: [[TMP0:%.*]] = zext <8 x i16> [[WIDE_LOAD3]] to <8 x i64>
+; CHECK-MAXBW-NEXT: [[TMP1:%.*]] = zext <8 x i16> [[WIDE_LOAD]] to <8 x i64>
+; CHECK-MAXBW-NEXT: [[TMP2:%.*]] = mul nuw nsw <8 x i64> [[TMP0]], [[TMP1]]
+; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v8i64(<2 x i64> [[VEC_PHI]], <8 x i64> [[TMP2]])
+; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
+; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
+; CHECK-MAXBW-NEXT: br i1 [[TMP3]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
; CHECK-MAXBW: middle.block:
-; CHECK-MAXBW-NEXT: [[TMP19:%.*]] = call i64 @llvm.vector.reduce.add.nxv4i64(<vscale x 4 x i64> [[TMP17]])
-; CHECK-MAXBW-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
-; CHECK-MAXBW-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH:%.*]]
-; CHECK-MAXBW: scalar.ph:
+; CHECK-MAXBW-NEXT: [[TMP4:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[PARTIAL_REDUCE]])
+; CHECK-MAXBW-NEXT: br label [[FOR_EXIT:%.*]]
+; CHECK-MAXBW: for.exit:
+; CHECK-MAXBW-NEXT: ret i64 [[TMP4]]
;
entry:
br label %for.body
@@ -687,7 +723,7 @@ define i32 @not_dotp_
diff erent_types(ptr %a, ptr %b) #0 {
; CHECK-MAXBW-NEXT: [[TMP138]] = add <16 x i32> [[TMP136]], [[VEC_PHI1]]
; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-MAXBW-NEXT: [[TMP70:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
-; CHECK-MAXBW-NEXT: br i1 [[TMP70]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
+; CHECK-MAXBW-NEXT: br i1 [[TMP70]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
; CHECK-MAXBW: middle.block:
; CHECK-MAXBW-NEXT: [[TMP71:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[TMP138]])
; CHECK-MAXBW-NEXT: br label [[FOR_EXIT:%.*]]
@@ -835,7 +871,7 @@ define i32 @not_dotp_not_loop_carried(ptr %a, ptr %b) #0 {
; CHECK-MAXBW-NEXT: [[TMP27:%.*]] = add <vscale x 8 x i32> [[TMP25]], [[TMP26]]
; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
; CHECK-MAXBW-NEXT: [[TMP28:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-MAXBW-NEXT: br i1 [[TMP28]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
+; CHECK-MAXBW-NEXT: br i1 [[TMP28]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]]
; CHECK-MAXBW: middle.block:
; CHECK-MAXBW-NEXT: [[TMP20:%.*]] = call i32 @llvm.vscale.i32()
; CHECK-MAXBW-NEXT: [[TMP21:%.*]] = mul nuw i32 [[TMP20]], 8
@@ -964,7 +1000,7 @@ define i32 @not_dotp_not_phi(ptr %a, ptr %b) #0 {
; CHECK-MAXBW-NEXT: [[TMP21:%.*]] = add <vscale x 8 x i32> [[TMP20]], [[TMP19]]
; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
; CHECK-MAXBW-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-MAXBW-NEXT: br i1 [[TMP22]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]]
+; CHECK-MAXBW-NEXT: br i1 [[TMP22]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
; CHECK-MAXBW: middle.block:
; CHECK-MAXBW-NEXT: [[TMP23:%.*]] = call i32 @llvm.vscale.i32()
; CHECK-MAXBW-NEXT: [[TMP24:%.*]] = mul nuw i32 [[TMP23]], 8
@@ -1024,26 +1060,26 @@ define i32 @dotp_unrolled(i32 %num_out, i64 %num_in, ptr %a, ptr %b) #0 {
; CHECK-INTERLEAVE1-NEXT: [[TMP10:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[TMP9]]
; CHECK-INTERLEAVE1-NEXT: [[TMP11:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[TMP9]]
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1
-; CHECK-INTERLEAVE1-NEXT: [[TMP23:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1
-; CHECK-INTERLEAVE1-NEXT: [[TMP12:%.*]] = sext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>
-; CHECK-INTERLEAVE1-NEXT: [[TMP13:%.*]] = mul nsw <16 x i32> [[TMP12]], [[TMP23]]
+; CHECK-INTERLEAVE1-NEXT: [[TMP23:%.*]] = sext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>
+; CHECK-INTERLEAVE1-NEXT: [[TMP12:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
+; CHECK-INTERLEAVE1-NEXT: [[TMP13:%.*]] = mul nsw <16 x i32> [[TMP23]], [[TMP12]]
; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI3]], <16 x i32> [[TMP13]])
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD5:%.*]] = load <16 x i8>, ptr [[TMP4]], align 1
-; CHECK-INTERLEAVE1-NEXT: [[TMP14:%.*]] = sext <16 x i8> [[WIDE_LOAD5]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD6:%.*]] = load <16 x i8>, ptr [[TMP5]], align 1
+; CHECK-INTERLEAVE1-NEXT: [[TMP14:%.*]] = sext <16 x i8> [[WIDE_LOAD5]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[TMP15:%.*]] = sext <16 x i8> [[WIDE_LOAD6]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[TMP16:%.*]] = mul nsw <16 x i32> [[TMP14]], [[TMP15]]
; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE7]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI2]], <16 x i32> [[TMP16]])
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD8:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
-; CHECK-INTERLEAVE1-NEXT: [[TMP17:%.*]] = sext <16 x i8> [[WIDE_LOAD8]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD9:%.*]] = load <16 x i8>, ptr [[TMP8]], align 1
+; CHECK-INTERLEAVE1-NEXT: [[TMP17:%.*]] = sext <16 x i8> [[WIDE_LOAD8]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[TMP18:%.*]] = sext <16 x i8> [[WIDE_LOAD9]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[TMP19:%.*]] = mul nsw <16 x i32> [[TMP17]], [[TMP18]]
; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE10]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP19]])
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD11:%.*]] = load <16 x i8>, ptr [[TMP10]], align 1
-; CHECK-INTERLEAVE1-NEXT: [[TMP20:%.*]] = sext <16 x i8> [[WIDE_LOAD11]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD12:%.*]] = load <16 x i8>, ptr [[TMP11]], align 1
+; CHECK-INTERLEAVE1-NEXT: [[TMP20:%.*]] = sext <16 x i8> [[WIDE_LOAD11]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[TMP21:%.*]] = sext <16 x i8> [[WIDE_LOAD12]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[TMP22:%.*]] = mul nsw <16 x i32> [[TMP20]], [[TMP21]]
; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE13]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP22]])
@@ -1092,58 +1128,58 @@ define i32 @dotp_unrolled(i32 %num_out, i64 %num_in, ptr %a, ptr %b) #0 {
; CHECK-INTERLEAVED-NEXT: [[TMP43:%.*]] = getelementptr inbounds i8, ptr [[TMP1]], i32 16
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD8:%.*]] = load <16 x i8>, ptr [[TMP43]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = sext <16 x i8> [[WIDE_LOAD8]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = getelementptr inbounds i8, ptr [[TMP2]], i32 16
+; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = getelementptr inbounds i8, ptr [[TMP2]], i32 16
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD9:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1
-; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD10:%.*]] = load <16 x i8>, ptr [[TMP14]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP15:%.*]] = sext <16 x i8> [[WIDE_LOAD9]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD10:%.*]] = load <16 x i8>, ptr [[TMP12]], align 1
+; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = sext <16 x i8> [[WIDE_LOAD9]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP15:%.*]] = mul nsw <16 x i32> [[TMP13]], [[TMP14]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI6]], <16 x i32> [[TMP15]])
; CHECK-INTERLEAVED-NEXT: [[TMP16:%.*]] = sext <16 x i8> [[WIDE_LOAD10]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP17:%.*]] = mul nsw <16 x i32> [[TMP15]], [[TMP12]]
-; CHECK-INTERLEAVED-NEXT: [[TMP18:%.*]] = mul nsw <16 x i32> [[TMP16]], [[TMP13]]
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI6]], <16 x i32> [[TMP17]])
+; CHECK-INTERLEAVED-NEXT: [[TMP17:%.*]] = sext <16 x i8> [[WIDE_LOAD8]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP18:%.*]] = mul nsw <16 x i32> [[TMP16]], [[TMP17]]
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE11]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI7]], <16 x i32> [[TMP18]])
; CHECK-INTERLEAVED-NEXT: [[TMP19:%.*]] = getelementptr inbounds i8, ptr [[TMP4]], i32 16
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD12:%.*]] = load <16 x i8>, ptr [[TMP4]], align 1
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD13:%.*]] = load <16 x i8>, ptr [[TMP19]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP20:%.*]] = sext <16 x i8> [[WIDE_LOAD12]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP21:%.*]] = sext <16 x i8> [[WIDE_LOAD13]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP22:%.*]] = getelementptr inbounds i8, ptr [[TMP5]], i32 16
+; CHECK-INTERLEAVED-NEXT: [[TMP20:%.*]] = getelementptr inbounds i8, ptr [[TMP5]], i32 16
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD14:%.*]] = load <16 x i8>, ptr [[TMP5]], align 1
-; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD15:%.*]] = load <16 x i8>, ptr [[TMP22]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP23:%.*]] = sext <16 x i8> [[WIDE_LOAD14]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP24:%.*]] = sext <16 x i8> [[WIDE_LOAD15]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP25:%.*]] = mul nsw <16 x i32> [[TMP20]], [[TMP23]]
-; CHECK-INTERLEAVED-NEXT: [[TMP26:%.*]] = mul nsw <16 x i32> [[TMP21]], [[TMP24]]
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE16]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI4]], <16 x i32> [[TMP25]])
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD15:%.*]] = load <16 x i8>, ptr [[TMP20]], align 1
+; CHECK-INTERLEAVED-NEXT: [[TMP21:%.*]] = sext <16 x i8> [[WIDE_LOAD12]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP22:%.*]] = sext <16 x i8> [[WIDE_LOAD14]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP23:%.*]] = mul nsw <16 x i32> [[TMP21]], [[TMP22]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE16]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI4]], <16 x i32> [[TMP23]])
+; CHECK-INTERLEAVED-NEXT: [[TMP24:%.*]] = sext <16 x i8> [[WIDE_LOAD13]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP25:%.*]] = sext <16 x i8> [[WIDE_LOAD15]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP26:%.*]] = mul nsw <16 x i32> [[TMP24]], [[TMP25]]
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE17]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI5]], <16 x i32> [[TMP26]])
; CHECK-INTERLEAVED-NEXT: [[TMP27:%.*]] = getelementptr inbounds i8, ptr [[TMP7]], i32 16
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD18:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD19:%.*]] = load <16 x i8>, ptr [[TMP27]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP28:%.*]] = sext <16 x i8> [[WIDE_LOAD18]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP29:%.*]] = sext <16 x i8> [[WIDE_LOAD19]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP30:%.*]] = getelementptr inbounds i8, ptr [[TMP8]], i32 16
+; CHECK-INTERLEAVED-NEXT: [[TMP28:%.*]] = getelementptr inbounds i8, ptr [[TMP8]], i32 16
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD20:%.*]] = load <16 x i8>, ptr [[TMP8]], align 1
-; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD21:%.*]] = load <16 x i8>, ptr [[TMP30]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP31:%.*]] = sext <16 x i8> [[WIDE_LOAD20]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP48:%.*]] = sext <16 x i8> [[WIDE_LOAD21]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP33:%.*]] = mul nsw <16 x i32> [[TMP28]], [[TMP31]]
-; CHECK-INTERLEAVED-NEXT: [[TMP34:%.*]] = mul nsw <16 x i32> [[TMP29]], [[TMP48]]
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE22]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI2]], <16 x i32> [[TMP33]])
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD21:%.*]] = load <16 x i8>, ptr [[TMP28]], align 1
+; CHECK-INTERLEAVED-NEXT: [[TMP29:%.*]] = sext <16 x i8> [[WIDE_LOAD18]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP30:%.*]] = sext <16 x i8> [[WIDE_LOAD20]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP31:%.*]] = mul nsw <16 x i32> [[TMP29]], [[TMP30]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE22]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI2]], <16 x i32> [[TMP31]])
+; CHECK-INTERLEAVED-NEXT: [[TMP48:%.*]] = sext <16 x i8> [[WIDE_LOAD19]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP33:%.*]] = sext <16 x i8> [[WIDE_LOAD21]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP34:%.*]] = mul nsw <16 x i32> [[TMP48]], [[TMP33]]
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE23]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI3]], <16 x i32> [[TMP34]])
; CHECK-INTERLEAVED-NEXT: [[TMP35:%.*]] = getelementptr inbounds i8, ptr [[TMP10]], i32 16
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD24:%.*]] = load <16 x i8>, ptr [[TMP10]], align 1
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD25:%.*]] = load <16 x i8>, ptr [[TMP35]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP36:%.*]] = sext <16 x i8> [[WIDE_LOAD24]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP37:%.*]] = sext <16 x i8> [[WIDE_LOAD25]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP38:%.*]] = getelementptr inbounds i8, ptr [[TMP11]], i32 16
+; CHECK-INTERLEAVED-NEXT: [[TMP36:%.*]] = getelementptr inbounds i8, ptr [[TMP11]], i32 16
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD26:%.*]] = load <16 x i8>, ptr [[TMP11]], align 1
-; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD27:%.*]] = load <16 x i8>, ptr [[TMP38]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP39:%.*]] = sext <16 x i8> [[WIDE_LOAD26]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP40:%.*]] = sext <16 x i8> [[WIDE_LOAD27]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP41:%.*]] = mul nsw <16 x i32> [[TMP36]], [[TMP39]]
-; CHECK-INTERLEAVED-NEXT: [[TMP42:%.*]] = mul nsw <16 x i32> [[TMP37]], [[TMP40]]
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE28]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP41]])
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD27:%.*]] = load <16 x i8>, ptr [[TMP36]], align 1
+; CHECK-INTERLEAVED-NEXT: [[TMP37:%.*]] = sext <16 x i8> [[WIDE_LOAD24]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP38:%.*]] = sext <16 x i8> [[WIDE_LOAD26]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP39:%.*]] = mul nsw <16 x i32> [[TMP37]], [[TMP38]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE28]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP39]])
+; CHECK-INTERLEAVED-NEXT: [[TMP40:%.*]] = sext <16 x i8> [[WIDE_LOAD25]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP41:%.*]] = sext <16 x i8> [[WIDE_LOAD27]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP42:%.*]] = mul nsw <16 x i32> [[TMP40]], [[TMP41]]
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE29]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP42]])
; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
; CHECK-INTERLEAVED-NEXT: [[TMP32:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
@@ -1165,21 +1201,21 @@ define i32 @dotp_unrolled(i32 %num_out, i64 %num_in, ptr %a, ptr %b) #0 {
; CHECK-MAXBW-SAME: i32 [[NUM_OUT:%.*]], i64 [[NUM_IN:%.*]], ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] {
; CHECK-MAXBW-NEXT: entry:
; CHECK-MAXBW-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-MAXBW-NEXT: [[TMP1:%.*]] = shl nuw nsw i64 [[TMP0]], 3
+; CHECK-MAXBW-NEXT: [[TMP1:%.*]] = shl nuw nsw i64 [[TMP0]], 4
; CHECK-MAXBW-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[NUM_IN]], [[TMP1]]
; CHECK-MAXBW-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK-MAXBW: vector.ph:
; CHECK-MAXBW-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 8
+; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 16
; CHECK-MAXBW-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[NUM_IN]], [[TMP3]]
; CHECK-MAXBW-NEXT: [[N_VEC:%.*]] = sub i64 [[NUM_IN]], [[N_MOD_VF]]
; CHECK-MAXBW-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-MAXBW: vector.body:
; CHECK-MAXBW-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-MAXBW-NEXT: [[VEC_PHI4:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE16:%.*]], [[VECTOR_BODY]] ]
-; CHECK-MAXBW-NEXT: [[VEC_PHI5:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE17:%.*]], [[VECTOR_BODY]] ]
-; CHECK-MAXBW-NEXT: [[VEC_PHI6:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
-; CHECK-MAXBW-NEXT: [[VEC_PHI7:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE11:%.*]], [[VECTOR_BODY]] ]
+; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE13:%.*]], [[VECTOR_BODY]] ]
+; CHECK-MAXBW-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE10:%.*]], [[VECTOR_BODY]] ]
+; CHECK-MAXBW-NEXT: [[VEC_PHI2:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE7:%.*]], [[VECTOR_BODY]] ]
+; CHECK-MAXBW-NEXT: [[VEC_PHI3:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
; CHECK-MAXBW-NEXT: [[TMP7:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[INDEX]]
; CHECK-MAXBW-NEXT: [[TMP8:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[INDEX]]
; CHECK-MAXBW-NEXT: [[TMP9:%.*]] = or disjoint i64 [[INDEX]], 1
@@ -1191,38 +1227,38 @@ define i32 @dotp_unrolled(i32 %num_out, i64 %num_in, ptr %a, ptr %b) #0 {
; CHECK-MAXBW-NEXT: [[TMP15:%.*]] = or disjoint i64 [[INDEX]], 3
; CHECK-MAXBW-NEXT: [[TMP16:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[TMP15]]
; CHECK-MAXBW-NEXT: [[TMP17:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[TMP15]]
-; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
-; CHECK-MAXBW-NEXT: [[TMP23:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
-; CHECK-MAXBW-NEXT: [[WIDE_LOAD9:%.*]] = load <vscale x 8 x i8>, ptr [[TMP8]], align 1
-; CHECK-MAXBW-NEXT: [[TMP29:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD9]] to <vscale x 8 x i32>
-; CHECK-MAXBW-NEXT: [[TMP31:%.*]] = mul nsw <vscale x 8 x i32> [[TMP29]], [[TMP23]]
-; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE11]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI7]], <vscale x 8 x i32> [[TMP31]])
-; CHECK-MAXBW-NEXT: [[WIDE_LOAD12:%.*]] = load <vscale x 8 x i8>, ptr [[TMP10]], align 1
-; CHECK-MAXBW-NEXT: [[TMP37:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD12]] to <vscale x 8 x i32>
-; CHECK-MAXBW-NEXT: [[WIDE_LOAD14:%.*]] = load <vscale x 8 x i8>, ptr [[TMP11]], align 1
-; CHECK-MAXBW-NEXT: [[TMP43:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD14]] to <vscale x 8 x i32>
-; CHECK-MAXBW-NEXT: [[TMP45:%.*]] = mul nsw <vscale x 8 x i32> [[TMP37]], [[TMP43]]
-; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI6]], <vscale x 8 x i32> [[TMP45]])
-; CHECK-MAXBW-NEXT: [[WIDE_LOAD18:%.*]] = load <vscale x 8 x i8>, ptr [[TMP13]], align 1
-; CHECK-MAXBW-NEXT: [[TMP51:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD18]] to <vscale x 8 x i32>
-; CHECK-MAXBW-NEXT: [[WIDE_LOAD20:%.*]] = load <vscale x 8 x i8>, ptr [[TMP14]], align 1
-; CHECK-MAXBW-NEXT: [[TMP57:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD20]] to <vscale x 8 x i32>
-; CHECK-MAXBW-NEXT: [[TMP59:%.*]] = mul nsw <vscale x 8 x i32> [[TMP51]], [[TMP57]]
-; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE17]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI5]], <vscale x 8 x i32> [[TMP59]])
-; CHECK-MAXBW-NEXT: [[WIDE_LOAD24:%.*]] = load <vscale x 8 x i8>, ptr [[TMP16]], align 1
-; CHECK-MAXBW-NEXT: [[TMP65:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD24]] to <vscale x 8 x i32>
-; CHECK-MAXBW-NEXT: [[WIDE_LOAD26:%.*]] = load <vscale x 8 x i8>, ptr [[TMP17]], align 1
-; CHECK-MAXBW-NEXT: [[TMP71:%.*]] = sext <vscale x 8 x i8> [[WIDE_LOAD26]] to <vscale x 8 x i32>
-; CHECK-MAXBW-NEXT: [[TMP73:%.*]] = mul nsw <vscale x 8 x i32> [[TMP65]], [[TMP71]]
-; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE16]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI4]], <vscale x 8 x i32> [[TMP73]])
+; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 16 x i8>, ptr [[TMP7]], align 1
+; CHECK-MAXBW-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 16 x i8>, ptr [[TMP8]], align 1
+; CHECK-MAXBW-NEXT: [[TMP27:%.*]] = sext <vscale x 16 x i8> [[WIDE_LOAD4]] to <vscale x 16 x i32>
+; CHECK-MAXBW-NEXT: [[TMP32:%.*]] = sext <vscale x 16 x i8> [[WIDE_LOAD]] to <vscale x 16 x i32>
+; CHECK-MAXBW-NEXT: [[TMP33:%.*]] = mul nsw <vscale x 16 x i32> [[TMP27]], [[TMP32]]
+; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 4 x i32> @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32(<vscale x 4 x i32> [[VEC_PHI3]], <vscale x 16 x i32> [[TMP33]])
+; CHECK-MAXBW-NEXT: [[WIDE_LOAD5:%.*]] = load <vscale x 16 x i8>, ptr [[TMP10]], align 1
+; CHECK-MAXBW-NEXT: [[WIDE_LOAD6:%.*]] = load <vscale x 16 x i8>, ptr [[TMP11]], align 1
+; CHECK-MAXBW-NEXT: [[TMP18:%.*]] = sext <vscale x 16 x i8> [[WIDE_LOAD5]] to <vscale x 16 x i32>
+; CHECK-MAXBW-NEXT: [[TMP19:%.*]] = sext <vscale x 16 x i8> [[WIDE_LOAD6]] to <vscale x 16 x i32>
+; CHECK-MAXBW-NEXT: [[TMP20:%.*]] = mul nsw <vscale x 16 x i32> [[TMP18]], [[TMP19]]
+; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE7]] = call <vscale x 4 x i32> @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32(<vscale x 4 x i32> [[VEC_PHI2]], <vscale x 16 x i32> [[TMP20]])
+; CHECK-MAXBW-NEXT: [[WIDE_LOAD8:%.*]] = load <vscale x 16 x i8>, ptr [[TMP13]], align 1
+; CHECK-MAXBW-NEXT: [[WIDE_LOAD9:%.*]] = load <vscale x 16 x i8>, ptr [[TMP14]], align 1
+; CHECK-MAXBW-NEXT: [[TMP21:%.*]] = sext <vscale x 16 x i8> [[WIDE_LOAD8]] to <vscale x 16 x i32>
+; CHECK-MAXBW-NEXT: [[TMP22:%.*]] = sext <vscale x 16 x i8> [[WIDE_LOAD9]] to <vscale x 16 x i32>
+; CHECK-MAXBW-NEXT: [[TMP23:%.*]] = mul nsw <vscale x 16 x i32> [[TMP21]], [[TMP22]]
+; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE10]] = call <vscale x 4 x i32> @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32(<vscale x 4 x i32> [[VEC_PHI1]], <vscale x 16 x i32> [[TMP23]])
+; CHECK-MAXBW-NEXT: [[WIDE_LOAD11:%.*]] = load <vscale x 16 x i8>, ptr [[TMP16]], align 1
+; CHECK-MAXBW-NEXT: [[WIDE_LOAD12:%.*]] = load <vscale x 16 x i8>, ptr [[TMP17]], align 1
+; CHECK-MAXBW-NEXT: [[TMP24:%.*]] = sext <vscale x 16 x i8> [[WIDE_LOAD11]] to <vscale x 16 x i32>
+; CHECK-MAXBW-NEXT: [[TMP25:%.*]] = sext <vscale x 16 x i8> [[WIDE_LOAD12]] to <vscale x 16 x i32>
+; CHECK-MAXBW-NEXT: [[TMP26:%.*]] = mul nsw <vscale x 16 x i32> [[TMP24]], [[TMP25]]
+; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE13]] = call <vscale x 4 x i32> @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32(<vscale x 4 x i32> [[VEC_PHI]], <vscale x 16 x i32> [[TMP26]])
; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
; CHECK-MAXBW-NEXT: [[TMP74:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-MAXBW-NEXT: br i1 [[TMP74]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]]
+; CHECK-MAXBW-NEXT: br i1 [[TMP74]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]]
; CHECK-MAXBW: middle.block:
-; CHECK-MAXBW-NEXT: [[TMP39:%.*]] = call i32 @llvm.vector.reduce.add.nxv2i32(<vscale x 2 x i32> [[PARTIAL_REDUCE16]])
-; CHECK-MAXBW-NEXT: [[TMP40:%.*]] = call i32 @llvm.vector.reduce.add.nxv2i32(<vscale x 2 x i32> [[PARTIAL_REDUCE17]])
-; CHECK-MAXBW-NEXT: [[TMP41:%.*]] = call i32 @llvm.vector.reduce.add.nxv2i32(<vscale x 2 x i32> [[PARTIAL_REDUCE]])
-; CHECK-MAXBW-NEXT: [[TMP42:%.*]] = call i32 @llvm.vector.reduce.add.nxv2i32(<vscale x 2 x i32> [[PARTIAL_REDUCE11]])
+; CHECK-MAXBW-NEXT: [[TMP28:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[PARTIAL_REDUCE13]])
+; CHECK-MAXBW-NEXT: [[TMP29:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[PARTIAL_REDUCE10]])
+; CHECK-MAXBW-NEXT: [[TMP30:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[PARTIAL_REDUCE7]])
+; CHECK-MAXBW-NEXT: [[TMP31:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[PARTIAL_REDUCE]])
; CHECK-MAXBW-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[NUM_IN]], [[N_VEC]]
; CHECK-MAXBW-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
; CHECK-MAXBW: scalar.ph:
@@ -1390,7 +1426,7 @@ define i32 @dotp_predicated(i64 %N, ptr %a, ptr %b) #0 {
; CHECK-MAXBW-NEXT: [[ACTIVE_LANE_MASK_NEXT]] = call <vscale x 16 x i1> @llvm.get.active.lane.mask.nxv16i1.i64(i64 [[INDEX]], i64 [[TMP9]])
; CHECK-MAXBW-NEXT: [[TMP19:%.*]] = extractelement <vscale x 16 x i1> [[ACTIVE_LANE_MASK_NEXT]], i32 0
; CHECK-MAXBW-NEXT: [[TMP20:%.*]] = xor i1 [[TMP19]], true
-; CHECK-MAXBW-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]]
+; CHECK-MAXBW-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]]
; CHECK-MAXBW: middle.block:
; CHECK-MAXBW-NEXT: [[TMP21:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[PARTIAL_REDUCE]])
; CHECK-MAXBW-NEXT: br label [[EXIT:%.*]]
@@ -1525,7 +1561,7 @@ define i32 @not_dotp_extend_user(ptr %a, ptr %b) #0 {
; CHECK-MAXBW-NEXT: [[TMP24]] = add <vscale x 8 x i32> [[TMP22]], [[VEC_PHI1]]
; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
; CHECK-MAXBW-NEXT: [[TMP25:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-MAXBW-NEXT: br i1 [[TMP25]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP17:![0-9]+]]
+; CHECK-MAXBW-NEXT: br i1 [[TMP25]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]]
; CHECK-MAXBW: middle.block:
; CHECK-MAXBW-NEXT: [[TMP16:%.*]] = call i32 @llvm.vector.reduce.add.nxv8i32(<vscale x 8 x i32> [[TMP24]])
; CHECK-MAXBW-NEXT: [[TMP17:%.*]] = call i32 @llvm.vscale.i32()
@@ -1565,110 +1601,93 @@ define i64 @dotp_cost_disagreement(ptr %a, ptr %b) #0 {
; CHECK-INTERLEAVE1-NEXT: entry:
; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_PH:%.*]]
; CHECK-INTERLEAVE1: vector.ph:
-; CHECK-INTERLEAVE1-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 2
-; CHECK-INTERLEAVE1-NEXT: [[N_MOD_VF:%.*]] = urem i64 41, [[TMP3]]
-; CHECK-INTERLEAVE1-NEXT: [[N_VEC:%.*]] = sub i64 41, [[N_MOD_VF]]
; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-INTERLEAVE1: vector.body:
; CHECK-INTERLEAVE1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP15:%.*]], [[VECTOR_BODY]] ]
+; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
; CHECK-INTERLEAVE1-NEXT: [[TMP7:%.*]] = getelementptr inbounds nuw i8, ptr [[A]], i64 [[INDEX]]
-; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 2 x i8>, ptr [[TMP7]], align 1
-; CHECK-INTERLEAVE1-NEXT: [[TMP9:%.*]] = zext <vscale x 2 x i8> [[WIDE_LOAD]] to <vscale x 2 x i64>
+; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
; CHECK-INTERLEAVE1-NEXT: [[TMP10:%.*]] = add nuw nsw i64 [[INDEX]], 1
; CHECK-INTERLEAVE1-NEXT: [[TMP11:%.*]] = getelementptr inbounds nuw i8, ptr [[B]], i64 [[TMP10]]
-; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD1:%.*]] = load <vscale x 2 x i8>, ptr [[TMP11]], align 1
-; CHECK-INTERLEAVE1-NEXT: [[TMP13:%.*]] = zext <vscale x 2 x i8> [[WIDE_LOAD1]] to <vscale x 2 x i64>
-; CHECK-INTERLEAVE1-NEXT: [[TMP14:%.*]] = mul nuw nsw <vscale x 2 x i64> [[TMP13]], [[TMP9]]
-; CHECK-INTERLEAVE1-NEXT: [[TMP15]] = add <vscale x 2 x i64> [[VEC_PHI]], [[TMP14]]
-; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
-; CHECK-INTERLEAVE1-NEXT: [[TMP16:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-INTERLEAVE1-NEXT: br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]
+; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD1:%.*]] = load <16 x i8>, ptr [[TMP11]], align 1
+; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[WIDE_LOAD1]] to <16 x i64>
+; CHECK-INTERLEAVE1-NEXT: [[TMP4:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i64>
+; CHECK-INTERLEAVE1-NEXT: [[TMP5:%.*]] = mul nuw nsw <16 x i64> [[TMP3]], [[TMP4]]
+; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v16i64(<2 x i64> [[VEC_PHI]], <16 x i64> [[TMP5]])
+; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
+; CHECK-INTERLEAVE1-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], 32
+; CHECK-INTERLEAVE1-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]
; CHECK-INTERLEAVE1: middle.block:
-; CHECK-INTERLEAVE1-NEXT: [[TMP17:%.*]] = call i64 @llvm.vector.reduce.add.nxv2i64(<vscale x 2 x i64> [[TMP15]])
-; CHECK-INTERLEAVE1-NEXT: [[CMP_N:%.*]] = icmp eq i64 41, [[N_VEC]]
-; CHECK-INTERLEAVE1-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH:%.*]]
+; CHECK-INTERLEAVE1-NEXT: [[TMP8:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[PARTIAL_REDUCE]])
+; CHECK-INTERLEAVE1-NEXT: br label [[SCALAR_PH:%.*]]
; CHECK-INTERLEAVE1: scalar.ph:
;
; CHECK-INTERLEAVED-LABEL: define i64 @dotp_cost_disagreement(
; CHECK-INTERLEAVED-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] {
; CHECK-INTERLEAVED-NEXT: entry:
-; CHECK-INTERLEAVED-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = shl nuw nsw i64 [[TMP0]], 2
-; CHECK-INTERLEAVED-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 41, [[TMP1]]
-; CHECK-INTERLEAVED-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_PH:%.*]]
; CHECK-INTERLEAVED: vector.ph:
-; CHECK-INTERLEAVED-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 4
-; CHECK-INTERLEAVED-NEXT: [[N_MOD_VF:%.*]] = urem i64 41, [[TMP3]]
-; CHECK-INTERLEAVED-NEXT: [[N_VEC:%.*]] = sub i64 41, [[N_MOD_VF]]
; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-INTERLEAVED: vector.body:
; CHECK-INTERLEAVED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP24:%.*]], [[VECTOR_BODY]] ]
-; CHECK-INTERLEAVED-NEXT: [[VEC_PHI1:%.*]] = phi <vscale x 2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP25:%.*]], [[VECTOR_BODY]] ]
+; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
+; CHECK-INTERLEAVED-NEXT: [[VEC_PHI1:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ]
; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = getelementptr inbounds nuw i8, ptr [[A]], i64 [[INDEX]]
-; CHECK-INTERLEAVED-NEXT: [[TMP9:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = shl nuw i64 [[TMP9]], 1
-; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP7]], i64 [[TMP10]]
-; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 2 x i8>, ptr [[TMP7]], align 1
-; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 2 x i8>, ptr [[TMP11]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = zext <vscale x 2 x i8> [[WIDE_LOAD]] to <vscale x 2 x i64>
-; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = zext <vscale x 2 x i8> [[WIDE_LOAD2]] to <vscale x 2 x i64>
+; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP7]], i32 16
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1
; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = add nuw nsw i64 [[INDEX]], 1
; CHECK-INTERLEAVED-NEXT: [[TMP15:%.*]] = getelementptr inbounds nuw i8, ptr [[B]], i64 [[TMP14]]
-; CHECK-INTERLEAVED-NEXT: [[TMP17:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-INTERLEAVED-NEXT: [[TMP18:%.*]] = shl nuw i64 [[TMP17]], 1
-; CHECK-INTERLEAVED-NEXT: [[TMP19:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP15]], i64 [[TMP18]]
-; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 2 x i8>, ptr [[TMP15]], align 1
-; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD4:%.*]] = load <vscale x 2 x i8>, ptr [[TMP19]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP20:%.*]] = zext <vscale x 2 x i8> [[WIDE_LOAD3]] to <vscale x 2 x i64>
-; CHECK-INTERLEAVED-NEXT: [[TMP21:%.*]] = zext <vscale x 2 x i8> [[WIDE_LOAD4]] to <vscale x 2 x i64>
-; CHECK-INTERLEAVED-NEXT: [[TMP22:%.*]] = mul nuw nsw <vscale x 2 x i64> [[TMP20]], [[TMP12]]
-; CHECK-INTERLEAVED-NEXT: [[TMP23:%.*]] = mul nuw nsw <vscale x 2 x i64> [[TMP21]], [[TMP13]]
-; CHECK-INTERLEAVED-NEXT: [[TMP24]] = add <vscale x 2 x i64> [[VEC_PHI]], [[TMP22]]
-; CHECK-INTERLEAVED-NEXT: [[TMP25]] = add <vscale x 2 x i64> [[VEC_PHI1]], [[TMP23]]
-; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
-; CHECK-INTERLEAVED-NEXT: [[TMP26:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-INTERLEAVED-NEXT: br i1 [[TMP26]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]
+; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP15]], i32 16
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP15]], align 1
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP4]], align 1
+; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = zext <16 x i8> [[WIDE_LOAD3]] to <16 x i64>
+; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i64>
+; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = mul nuw nsw <16 x i64> [[TMP5]], [[TMP6]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v16i64(<2 x i64> [[VEC_PHI]], <16 x i64> [[TMP13]])
+; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = zext <16 x i8> [[WIDE_LOAD4]] to <16 x i64>
+; CHECK-INTERLEAVED-NEXT: [[TMP9:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i64>
+; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = mul nuw nsw <16 x i64> [[TMP8]], [[TMP9]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE5]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v16i64(<2 x i64> [[VEC_PHI1]], <16 x i64> [[TMP10]])
+; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
+; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], 32
+; CHECK-INTERLEAVED-NEXT: br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]
; CHECK-INTERLEAVED: middle.block:
-; CHECK-INTERLEAVED-NEXT: [[BIN_RDX:%.*]] = add <vscale x 2 x i64> [[TMP25]], [[TMP24]]
-; CHECK-INTERLEAVED-NEXT: [[TMP27:%.*]] = call i64 @llvm.vector.reduce.add.nxv2i64(<vscale x 2 x i64> [[BIN_RDX]])
-; CHECK-INTERLEAVED-NEXT: [[CMP_N:%.*]] = icmp eq i64 41, [[N_VEC]]
-; CHECK-INTERLEAVED-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
+; CHECK-INTERLEAVED-NEXT: [[BIN_RDX:%.*]] = add <2 x i64> [[PARTIAL_REDUCE5]], [[PARTIAL_REDUCE]]
+; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[BIN_RDX]])
+; CHECK-INTERLEAVED-NEXT: br label [[SCALAR_PH:%.*]]
; CHECK-INTERLEAVED: scalar.ph:
;
; CHECK-MAXBW-LABEL: define i64 @dotp_cost_disagreement(
; CHECK-MAXBW-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] {
; CHECK-MAXBW-NEXT: entry:
; CHECK-MAXBW-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-MAXBW-NEXT: [[TMP1:%.*]] = shl nuw nsw i64 [[TMP0]], 3
+; CHECK-MAXBW-NEXT: [[TMP1:%.*]] = shl nuw nsw i64 [[TMP0]], 4
; CHECK-MAXBW-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 41, [[TMP1]]
; CHECK-MAXBW-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK-MAXBW: vector.ph:
; CHECK-MAXBW-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 8
+; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 16
; CHECK-MAXBW-NEXT: [[N_MOD_VF:%.*]] = urem i64 41, [[TMP3]]
; CHECK-MAXBW-NEXT: [[N_VEC:%.*]] = sub i64 41, [[N_MOD_VF]]
; CHECK-MAXBW-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-MAXBW: vector.body:
; CHECK-MAXBW-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 8 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP14:%.*]], [[VECTOR_BODY]] ]
+; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
; CHECK-MAXBW-NEXT: [[TMP7:%.*]] = getelementptr inbounds nuw i8, ptr [[A]], i64 [[INDEX]]
-; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
-; CHECK-MAXBW-NEXT: [[TMP9:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i64>
+; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 16 x i8>, ptr [[TMP7]], align 1
; CHECK-MAXBW-NEXT: [[TMP10:%.*]] = add nuw nsw i64 [[INDEX]], 1
; CHECK-MAXBW-NEXT: [[TMP11:%.*]] = getelementptr inbounds nuw i8, ptr [[B]], i64 [[TMP10]]
-; CHECK-MAXBW-NEXT: [[WIDE_LOAD1:%.*]] = load <vscale x 8 x i8>, ptr [[TMP11]], align 1
-; CHECK-MAXBW-NEXT: [[TMP8:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD1]] to <vscale x 8 x i64>
-; CHECK-MAXBW-NEXT: [[TMP13:%.*]] = mul nuw nsw <vscale x 8 x i64> [[TMP8]], [[TMP9]]
-; CHECK-MAXBW-NEXT: [[TMP14]] = add <vscale x 8 x i64> [[VEC_PHI]], [[TMP13]]
+; CHECK-MAXBW-NEXT: [[WIDE_LOAD1:%.*]] = load <vscale x 16 x i8>, ptr [[TMP11]], align 1
+; CHECK-MAXBW-NEXT: [[TMP12:%.*]] = zext <vscale x 16 x i8> [[WIDE_LOAD1]] to <vscale x 16 x i64>
+; CHECK-MAXBW-NEXT: [[TMP8:%.*]] = zext <vscale x 16 x i8> [[WIDE_LOAD]] to <vscale x 16 x i64>
+; CHECK-MAXBW-NEXT: [[TMP9:%.*]] = mul nuw nsw <vscale x 16 x i64> [[TMP12]], [[TMP8]]
+; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 2 x i64> @llvm.vector.partial.reduce.add.nxv2i64.nxv16i64(<vscale x 2 x i64> [[VEC_PHI]], <vscale x 16 x i64> [[TMP9]])
; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
; CHECK-MAXBW-NEXT: [[TMP15:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-MAXBW-NEXT: br i1 [[TMP15]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP19:![0-9]+]]
+; CHECK-MAXBW-NEXT: br i1 [[TMP15]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP17:![0-9]+]]
; CHECK-MAXBW: middle.block:
-; CHECK-MAXBW-NEXT: [[TMP12:%.*]] = call i64 @llvm.vector.reduce.add.nxv8i64(<vscale x 8 x i64> [[TMP14]])
+; CHECK-MAXBW-NEXT: [[TMP13:%.*]] = call i64 @llvm.vector.reduce.add.nxv2i64(<vscale x 2 x i64> [[PARTIAL_REDUCE]])
; CHECK-MAXBW-NEXT: [[CMP_N:%.*]] = icmp eq i64 41, [[N_VEC]]
; CHECK-MAXBW-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
; CHECK-MAXBW: scalar.ph:
@@ -1971,7 +1990,7 @@ define i64 @not_dotp_ext_outside_plan(ptr %a, i16 %b, i64 %n) #0 {
; CHECK-MAXBW-NEXT: [[TMP11]] = add <vscale x 4 x i64> [[TMP10]], [[VEC_PHI]]
; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
; CHECK-MAXBW-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-MAXBW-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP21:![0-9]+]]
+; CHECK-MAXBW-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP19:![0-9]+]]
; CHECK-MAXBW: middle.block:
; CHECK-MAXBW-NEXT: [[TMP13:%.*]] = call i64 @llvm.vector.reduce.add.nxv4i64(<vscale x 4 x i64> [[TMP11]])
; CHECK-MAXBW-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
@@ -2104,7 +2123,7 @@ define i64 @not_dotp_ext_outside_plan2(ptr %a, i16 %b, i64 %n) #0 {
; CHECK-MAXBW-NEXT: [[TMP11]] = add <vscale x 4 x i64> [[TMP10]], [[VEC_PHI]]
; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
; CHECK-MAXBW-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-MAXBW-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP23:![0-9]+]]
+; CHECK-MAXBW-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP21:![0-9]+]]
; CHECK-MAXBW: middle.block:
; CHECK-MAXBW-NEXT: [[TMP13:%.*]] = call i64 @llvm.vector.reduce.add.nxv4i64(<vscale x 4 x i64> [[TMP11]])
; CHECK-MAXBW-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
@@ -2160,10 +2179,10 @@ define dso_local i32 @not_dotp_vscale1(ptr %a, ptr %b, i32 %n, i64 %cost) #0 {
; CHECK-INTERLEAVE1-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; CHECK-INTERLEAVE1-NEXT: [[NEXT_GEP1:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[NEXT_GEP]], align 1
-; CHECK-INTERLEAVE1-NEXT: [[TMP5:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i64>
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[NEXT_GEP1]], align 1
-; CHECK-INTERLEAVE1-NEXT: [[TMP6:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i64>
-; CHECK-INTERLEAVE1-NEXT: [[TMP10:%.*]] = mul nuw nsw <16 x i64> [[TMP6]], [[TMP5]]
+; CHECK-INTERLEAVE1-NEXT: [[TMP5:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i64>
+; CHECK-INTERLEAVE1-NEXT: [[TMP6:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i64>
+; CHECK-INTERLEAVE1-NEXT: [[TMP10:%.*]] = mul nuw nsw <16 x i64> [[TMP5]], [[TMP6]]
; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v16i64(<2 x i64> [[VEC_PHI]], <16 x i64> [[TMP10]])
; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-INTERLEAVE1-NEXT: [[TMP19:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
@@ -2181,10 +2200,10 @@ define dso_local i32 @not_dotp_vscale1(ptr %a, ptr %b, i32 %n, i64 %cost) #0 {
; CHECK-INTERLEAVED-NEXT: br i1 [[CMP]], label [[FOR_BODY_PREHEADER:%.*]], label [[EXIT:%.*]]
; CHECK-INTERLEAVED: for.body.preheader:
; CHECK-INTERLEAVED-NEXT: [[TMP0:%.*]] = zext i32 [[N]] to i64
-; CHECK-INTERLEAVED-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 16
+; CHECK-INTERLEAVED-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 32
; CHECK-INTERLEAVED-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK-INTERLEAVED: vector.ph:
-; CHECK-INTERLEAVED-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 16
+; CHECK-INTERLEAVED-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 32
; CHECK-INTERLEAVED-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]
; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = trunc i64 [[N_VEC]] to i32
; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[A]], i64 [[N_VEC]]
@@ -2194,19 +2213,29 @@ define dso_local i32 @not_dotp_vscale1(ptr %a, ptr %b, i32 %n, i64 %cost) #0 {
; CHECK-INTERLEAVED: vector.body:
; CHECK-INTERLEAVED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi <2 x i64> [ [[TMP4]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
+; CHECK-INTERLEAVED-NEXT: [[VEC_PHI1:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE6:%.*]], [[VECTOR_BODY]] ]
; CHECK-INTERLEAVED-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; CHECK-INTERLEAVED-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
+; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i32 16
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[NEXT_GEP]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i64>
-; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[NEXT_GEP2]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i64>
-; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = mul nuw nsw <16 x i64> [[TMP6]], [[TMP5]]
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v16i64(<2 x i64> [[VEC_PHI]], <16 x i64> [[TMP10]])
-; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP5]], align 1
+; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[NEXT_GEP2]], i32 16
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[NEXT_GEP2]], align 1
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD5:%.*]] = load <16 x i8>, ptr [[TMP6]], align 1
+; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = zext <16 x i8> [[WIDE_LOAD4]] to <16 x i64>
+; CHECK-INTERLEAVED-NEXT: [[TMP15:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i64>
+; CHECK-INTERLEAVED-NEXT: [[TMP16:%.*]] = mul nuw nsw <16 x i64> [[TMP13]], [[TMP15]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v16i64(<2 x i64> [[VEC_PHI]], <16 x i64> [[TMP16]])
+; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = zext <16 x i8> [[WIDE_LOAD5]] to <16 x i64>
+; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = zext <16 x i8> [[WIDE_LOAD3]] to <16 x i64>
+; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = mul nuw nsw <16 x i64> [[TMP10]], [[TMP11]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE6]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v16i64(<2 x i64> [[VEC_PHI1]], <16 x i64> [[TMP12]])
+; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
; CHECK-INTERLEAVED-NEXT: [[TMP29:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-INTERLEAVED-NEXT: br i1 [[TMP29]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP24:![0-9]+]]
; CHECK-INTERLEAVED: middle.block:
-; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[PARTIAL_REDUCE]])
+; CHECK-INTERLEAVED-NEXT: [[BIN_RDX:%.*]] = add <2 x i64> [[PARTIAL_REDUCE6]], [[PARTIAL_REDUCE]]
+; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[BIN_RDX]])
; CHECK-INTERLEAVED-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]
; CHECK-INTERLEAVED-NEXT: br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]]
; CHECK-INTERLEAVED: scalar.ph:
@@ -2219,35 +2248,35 @@ define dso_local i32 @not_dotp_vscale1(ptr %a, ptr %b, i32 %n, i64 %cost) #0 {
; CHECK-MAXBW: for.body.preheader:
; CHECK-MAXBW-NEXT: [[TMP0:%.*]] = zext i32 [[N]] to i64
; CHECK-MAXBW-NEXT: [[TMP1:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-MAXBW-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 3
+; CHECK-MAXBW-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 4
; CHECK-MAXBW-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], [[TMP2]]
; CHECK-MAXBW-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK-MAXBW: vector.ph:
; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-MAXBW-NEXT: [[TMP4:%.*]] = mul nuw i64 [[TMP3]], 8
+; CHECK-MAXBW-NEXT: [[TMP4:%.*]] = mul nuw i64 [[TMP3]], 16
; CHECK-MAXBW-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], [[TMP4]]
; CHECK-MAXBW-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]
; CHECK-MAXBW-NEXT: [[TMP7:%.*]] = trunc i64 [[N_VEC]] to i32
; CHECK-MAXBW-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[A]], i64 [[N_VEC]]
; CHECK-MAXBW-NEXT: [[TMP9:%.*]] = getelementptr i8, ptr [[B]], i64 [[N_VEC]]
-; CHECK-MAXBW-NEXT: [[TMP10:%.*]] = insertelement <vscale x 8 x i64> zeroinitializer, i64 [[COST]], i32 0
+; CHECK-MAXBW-NEXT: [[TMP12:%.*]] = insertelement <vscale x 2 x i64> zeroinitializer, i64 [[COST]], i32 0
; CHECK-MAXBW-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-MAXBW: vector.body:
; CHECK-MAXBW-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 8 x i64> [ [[TMP10]], [[VECTOR_PH]] ], [ [[TMP20:%.*]], [[VECTOR_BODY]] ]
+; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 2 x i64> [ [[TMP12]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
; CHECK-MAXBW-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; CHECK-MAXBW-NEXT: [[NEXT_GEP1:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
-; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[NEXT_GEP]], align 1
-; CHECK-MAXBW-NEXT: [[TMP14:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i64>
-; CHECK-MAXBW-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 8 x i8>, ptr [[NEXT_GEP1]], align 1
-; CHECK-MAXBW-NEXT: [[TMP16:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD2]] to <vscale x 8 x i64>
-; CHECK-MAXBW-NEXT: [[TMP17:%.*]] = mul nuw nsw <vscale x 8 x i64> [[TMP16]], [[TMP14]]
-; CHECK-MAXBW-NEXT: [[TMP20]] = add <vscale x 8 x i64> [[TMP17]], [[VEC_PHI]]
+; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 16 x i8>, ptr [[NEXT_GEP]], align 1
+; CHECK-MAXBW-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 16 x i8>, ptr [[NEXT_GEP1]], align 1
+; CHECK-MAXBW-NEXT: [[TMP14:%.*]] = zext <vscale x 16 x i8> [[WIDE_LOAD2]] to <vscale x 16 x i64>
+; CHECK-MAXBW-NEXT: [[TMP10:%.*]] = zext <vscale x 16 x i8> [[WIDE_LOAD]] to <vscale x 16 x i64>
+; CHECK-MAXBW-NEXT: [[TMP11:%.*]] = mul nuw nsw <vscale x 16 x i64> [[TMP14]], [[TMP10]]
+; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 2 x i64> @llvm.vector.partial.reduce.add.nxv2i64.nxv16i64(<vscale x 2 x i64> [[VEC_PHI]], <vscale x 16 x i64> [[TMP11]])
; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP4]]
; CHECK-MAXBW-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-MAXBW-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP25:![0-9]+]]
+; CHECK-MAXBW-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP23:![0-9]+]]
; CHECK-MAXBW: middle.block:
-; CHECK-MAXBW-NEXT: [[TMP19:%.*]] = call i64 @llvm.vector.reduce.add.nxv8i64(<vscale x 8 x i64> [[TMP20]])
+; CHECK-MAXBW-NEXT: [[TMP13:%.*]] = call i64 @llvm.vector.reduce.add.nxv2i64(<vscale x 2 x i64> [[PARTIAL_REDUCE]])
; CHECK-MAXBW-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]
; CHECK-MAXBW-NEXT: br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]]
; CHECK-MAXBW: scalar.ph:
@@ -2319,17 +2348,16 @@ define dso_local void @not_dotp_high_register_pressure(ptr %a, ptr %b, ptr %sum,
; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-INTERLEAVE1: vector.body:
; CHECK-INTERLEAVE1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ [[TMP0]], [[VECTOR_PH]] ], [ [[TMP36:%.*]], [[VECTOR_BODY]] ]
-; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i32> [ [[TMP1]], [[VECTOR_PH]] ], [ [[TMP33:%.*]], [[VECTOR_BODY]] ]
-; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI2:%.*]] = phi <4 x i32> [ [[TMP2]], [[VECTOR_PH]] ], [ [[TMP30:%.*]], [[VECTOR_BODY]] ]
-; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI3:%.*]] = phi <4 x i32> [ [[TMP3]], [[VECTOR_PH]] ], [ [[TMP27:%.*]], [[VECTOR_BODY]] ]
-; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI4:%.*]] = phi <4 x i32> [ [[TMP4]], [[VECTOR_PH]] ], [ [[TMP24:%.*]], [[VECTOR_BODY]] ]
-; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI5:%.*]] = phi <4 x i32> [ [[TMP5]], [[VECTOR_PH]] ], [ [[TMP21:%.*]], [[VECTOR_BODY]] ]
-; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI6:%.*]] = phi <4 x i32> [ [[TMP6]], [[VECTOR_PH]] ], [ [[TMP18:%.*]], [[VECTOR_BODY]] ]
-; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI7:%.*]] = phi <4 x i32> [ [[TMP7]], [[VECTOR_PH]] ], [ [[TMP15:%.*]], [[VECTOR_BODY]] ]
+; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ [[TMP0]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE21:%.*]], [[VECTOR_BODY]] ]
+; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i32> [ [[TMP1]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE20:%.*]], [[VECTOR_BODY]] ]
+; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI2:%.*]] = phi <4 x i32> [ [[TMP2]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE19:%.*]], [[VECTOR_BODY]] ]
+; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI3:%.*]] = phi <4 x i32> [ [[TMP3]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE18:%.*]], [[VECTOR_BODY]] ]
+; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI4:%.*]] = phi <4 x i32> [ [[TMP4]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE17:%.*]], [[VECTOR_BODY]] ]
+; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI5:%.*]] = phi <4 x i32> [ [[TMP5]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE16:%.*]], [[VECTOR_BODY]] ]
+; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI6:%.*]] = phi <4 x i32> [ [[TMP6]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE15:%.*]], [[VECTOR_BODY]] ]
+; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI7:%.*]] = phi <4 x i32> [ [[TMP7]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
; CHECK-INTERLEAVE1-NEXT: [[TMP8:%.*]] = getelementptr inbounds nuw i8, ptr [[A]], i64 [[INDEX]]
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP8]], align 1
-; CHECK-INTERLEAVE1-NEXT: [[TMP10:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[TMP11:%.*]] = shl nsw i64 [[INDEX]], 3
; CHECK-INTERLEAVE1-NEXT: [[TMP12:%.*]] = getelementptr inbounds nuw i8, ptr [[B]], i64 [[TMP11]]
; CHECK-INTERLEAVE1-NEXT: [[WIDE_VEC:%.*]] = load <128 x i8>, ptr [[TMP12]], align 1
@@ -2341,42 +2369,43 @@ define dso_local void @not_dotp_high_register_pressure(ptr %a, ptr %b, ptr %sum,
; CHECK-INTERLEAVE1-NEXT: [[STRIDED_VEC12:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> <i32 5, i32 13, i32 21, i32 29, i32 37, i32 45, i32 53, i32 61, i32 69, i32 77, i32 85, i32 93, i32 101, i32 109, i32 117, i32 125>
; CHECK-INTERLEAVE1-NEXT: [[STRIDED_VEC13:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> <i32 6, i32 14, i32 22, i32 30, i32 38, i32 46, i32 54, i32 62, i32 70, i32 78, i32 86, i32 94, i32 102, i32 110, i32 118, i32 126>
; CHECK-INTERLEAVE1-NEXT: [[STRIDED_VEC14:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> <i32 7, i32 15, i32 23, i32 31, i32 39, i32 47, i32 55, i32 63, i32 71, i32 79, i32 87, i32 95, i32 103, i32 111, i32 119, i32 127>
-; CHECK-INTERLEAVE1-NEXT: [[TMP13:%.*]] = sext <16 x i8> [[STRIDED_VEC]] to <16 x i32>
-; CHECK-INTERLEAVE1-NEXT: [[TMP14:%.*]] = mul nsw <16 x i32> [[TMP13]], [[TMP10]]
-; CHECK-INTERLEAVE1-NEXT: [[TMP15]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI7]], <16 x i32> [[TMP14]])
-; CHECK-INTERLEAVE1-NEXT: [[TMP29:%.*]] = sext <16 x i8> [[STRIDED_VEC8]] to <16 x i32>
-; CHECK-INTERLEAVE1-NEXT: [[TMP16:%.*]] = mul nsw <16 x i32> [[TMP29]], [[TMP10]]
-; CHECK-INTERLEAVE1-NEXT: [[TMP18]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI6]], <16 x i32> [[TMP16]])
+; CHECK-INTERLEAVE1-NEXT: [[TMP35:%.*]] = sext <16 x i8> [[STRIDED_VEC]] to <16 x i32>
+; CHECK-INTERLEAVE1-NEXT: [[TMP44:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
+; CHECK-INTERLEAVE1-NEXT: [[TMP13:%.*]] = mul nsw <16 x i32> [[TMP35]], [[TMP44]]
+; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI7]], <16 x i32> [[TMP13]])
+; CHECK-INTERLEAVE1-NEXT: [[TMP14:%.*]] = sext <16 x i8> [[STRIDED_VEC8]] to <16 x i32>
+; CHECK-INTERLEAVE1-NEXT: [[TMP16:%.*]] = mul nsw <16 x i32> [[TMP14]], [[TMP44]]
+; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE15]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI6]], <16 x i32> [[TMP16]])
; CHECK-INTERLEAVE1-NEXT: [[TMP17:%.*]] = sext <16 x i8> [[STRIDED_VEC9]] to <16 x i32>
-; CHECK-INTERLEAVE1-NEXT: [[TMP31:%.*]] = mul nsw <16 x i32> [[TMP17]], [[TMP10]]
-; CHECK-INTERLEAVE1-NEXT: [[TMP21]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI5]], <16 x i32> [[TMP31]])
-; CHECK-INTERLEAVE1-NEXT: [[TMP19:%.*]] = sext <16 x i8> [[STRIDED_VEC10]] to <16 x i32>
-; CHECK-INTERLEAVE1-NEXT: [[TMP20:%.*]] = mul nsw <16 x i32> [[TMP19]], [[TMP10]]
-; CHECK-INTERLEAVE1-NEXT: [[TMP24]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI4]], <16 x i32> [[TMP20]])
-; CHECK-INTERLEAVE1-NEXT: [[TMP32:%.*]] = sext <16 x i8> [[STRIDED_VEC11]] to <16 x i32>
-; CHECK-INTERLEAVE1-NEXT: [[TMP22:%.*]] = mul nsw <16 x i32> [[TMP32]], [[TMP10]]
-; CHECK-INTERLEAVE1-NEXT: [[TMP27]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI3]], <16 x i32> [[TMP22]])
-; CHECK-INTERLEAVE1-NEXT: [[TMP23:%.*]] = sext <16 x i8> [[STRIDED_VEC12]] to <16 x i32>
-; CHECK-INTERLEAVE1-NEXT: [[TMP34:%.*]] = mul nsw <16 x i32> [[TMP23]], [[TMP10]]
-; CHECK-INTERLEAVE1-NEXT: [[TMP30]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI2]], <16 x i32> [[TMP34]])
-; CHECK-INTERLEAVE1-NEXT: [[TMP25:%.*]] = sext <16 x i8> [[STRIDED_VEC13]] to <16 x i32>
-; CHECK-INTERLEAVE1-NEXT: [[TMP26:%.*]] = mul nsw <16 x i32> [[TMP25]], [[TMP10]]
-; CHECK-INTERLEAVE1-NEXT: [[TMP33]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP26]])
-; CHECK-INTERLEAVE1-NEXT: [[TMP35:%.*]] = sext <16 x i8> [[STRIDED_VEC14]] to <16 x i32>
-; CHECK-INTERLEAVE1-NEXT: [[TMP28:%.*]] = mul nsw <16 x i32> [[TMP35]], [[TMP10]]
-; CHECK-INTERLEAVE1-NEXT: [[TMP36]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP28]])
+; CHECK-INTERLEAVE1-NEXT: [[TMP19:%.*]] = mul nsw <16 x i32> [[TMP17]], [[TMP44]]
+; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE16]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI5]], <16 x i32> [[TMP19]])
+; CHECK-INTERLEAVE1-NEXT: [[TMP20:%.*]] = sext <16 x i8> [[STRIDED_VEC10]] to <16 x i32>
+; CHECK-INTERLEAVE1-NEXT: [[TMP22:%.*]] = mul nsw <16 x i32> [[TMP20]], [[TMP44]]
+; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE17]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI4]], <16 x i32> [[TMP22]])
+; CHECK-INTERLEAVE1-NEXT: [[TMP23:%.*]] = sext <16 x i8> [[STRIDED_VEC11]] to <16 x i32>
+; CHECK-INTERLEAVE1-NEXT: [[TMP25:%.*]] = mul nsw <16 x i32> [[TMP23]], [[TMP44]]
+; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE18]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI3]], <16 x i32> [[TMP25]])
+; CHECK-INTERLEAVE1-NEXT: [[TMP26:%.*]] = sext <16 x i8> [[STRIDED_VEC12]] to <16 x i32>
+; CHECK-INTERLEAVE1-NEXT: [[TMP28:%.*]] = mul nsw <16 x i32> [[TMP26]], [[TMP44]]
+; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE19]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI2]], <16 x i32> [[TMP28]])
+; CHECK-INTERLEAVE1-NEXT: [[TMP29:%.*]] = sext <16 x i8> [[STRIDED_VEC13]] to <16 x i32>
+; CHECK-INTERLEAVE1-NEXT: [[TMP31:%.*]] = mul nsw <16 x i32> [[TMP29]], [[TMP44]]
+; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE20]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP31]])
+; CHECK-INTERLEAVE1-NEXT: [[TMP32:%.*]] = sext <16 x i8> [[STRIDED_VEC14]] to <16 x i32>
+; CHECK-INTERLEAVE1-NEXT: [[TMP34:%.*]] = mul nsw <16 x i32> [[TMP32]], [[TMP44]]
+; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE21]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP34]])
; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-INTERLEAVE1-NEXT: [[TMP37:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-INTERLEAVE1-NEXT: br i1 [[TMP37]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP24:![0-9]+]]
; CHECK-INTERLEAVE1: middle.block:
-; CHECK-INTERLEAVE1-NEXT: [[TMP38:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP36]])
-; CHECK-INTERLEAVE1-NEXT: [[TMP39:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP33]])
-; CHECK-INTERLEAVE1-NEXT: [[TMP40:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP30]])
-; CHECK-INTERLEAVE1-NEXT: [[TMP41:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP27]])
-; CHECK-INTERLEAVE1-NEXT: [[TMP42:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP24]])
-; CHECK-INTERLEAVE1-NEXT: [[TMP43:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP21]])
-; CHECK-INTERLEAVE1-NEXT: [[TMP44:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP18]])
-; CHECK-INTERLEAVE1-NEXT: [[TMP45:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP15]])
+; CHECK-INTERLEAVE1-NEXT: [[TMP36:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE21]])
+; CHECK-INTERLEAVE1-NEXT: [[TMP45:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE20]])
+; CHECK-INTERLEAVE1-NEXT: [[TMP38:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE19]])
+; CHECK-INTERLEAVE1-NEXT: [[TMP39:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE18]])
+; CHECK-INTERLEAVE1-NEXT: [[TMP40:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE17]])
+; CHECK-INTERLEAVE1-NEXT: [[TMP41:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE16]])
+; CHECK-INTERLEAVE1-NEXT: [[TMP42:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE15]])
+; CHECK-INTERLEAVE1-NEXT: [[TMP43:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE]])
; CHECK-INTERLEAVE1-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC]]
; CHECK-INTERLEAVE1-NEXT: br i1 [[CMP_N]], label [[FOR_COND_FOR_COND_CLEANUP_CRIT_EDGE:%.*]], label [[SCALAR_PH]]
; CHECK-INTERLEAVE1: scalar.ph:
@@ -2429,7 +2458,6 @@ define dso_local void @not_dotp_high_register_pressure(ptr %a, ptr %b, ptr %sum,
; CHECK-INTERLEAVED-NEXT: [[VEC_PHI7:%.*]] = phi <4 x i32> [ [[TMP7]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = getelementptr inbounds nuw i8, ptr [[A]], i64 [[INDEX]]
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP8]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = shl nsw i64 [[INDEX]], 3
; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = getelementptr inbounds nuw i8, ptr [[B]], i64 [[TMP11]]
; CHECK-INTERLEAVED-NEXT: [[WIDE_VEC:%.*]] = load <128 x i8>, ptr [[TMP12]], align 1
@@ -2441,42 +2469,43 @@ define dso_local void @not_dotp_high_register_pressure(ptr %a, ptr %b, ptr %sum,
; CHECK-INTERLEAVED-NEXT: [[STRIDED_VEC12:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> <i32 5, i32 13, i32 21, i32 29, i32 37, i32 45, i32 53, i32 61, i32 69, i32 77, i32 85, i32 93, i32 101, i32 109, i32 117, i32 125>
; CHECK-INTERLEAVED-NEXT: [[STRIDED_VEC13:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> <i32 6, i32 14, i32 22, i32 30, i32 38, i32 46, i32 54, i32 62, i32 70, i32 78, i32 86, i32 94, i32 102, i32 110, i32 118, i32 126>
; CHECK-INTERLEAVED-NEXT: [[STRIDED_VEC14:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> <i32 7, i32 15, i32 23, i32 31, i32 39, i32 47, i32 55, i32 63, i32 71, i32 79, i32 87, i32 95, i32 103, i32 111, i32 119, i32 127>
-; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = sext <16 x i8> [[STRIDED_VEC]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = mul nsw <16 x i32> [[TMP13]], [[TMP10]]
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI7]], <16 x i32> [[TMP14]])
-; CHECK-INTERLEAVED-NEXT: [[TMP15:%.*]] = sext <16 x i8> [[STRIDED_VEC8]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP16:%.*]] = mul nsw <16 x i32> [[TMP15]], [[TMP10]]
+; CHECK-INTERLEAVED-NEXT: [[TMP35:%.*]] = sext <16 x i8> [[STRIDED_VEC]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP44:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = mul nsw <16 x i32> [[TMP35]], [[TMP44]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI7]], <16 x i32> [[TMP13]])
+; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = sext <16 x i8> [[STRIDED_VEC8]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP16:%.*]] = mul nsw <16 x i32> [[TMP14]], [[TMP44]]
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE15]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI6]], <16 x i32> [[TMP16]])
; CHECK-INTERLEAVED-NEXT: [[TMP17:%.*]] = sext <16 x i8> [[STRIDED_VEC9]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP18:%.*]] = mul nsw <16 x i32> [[TMP17]], [[TMP10]]
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE16]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI5]], <16 x i32> [[TMP18]])
-; CHECK-INTERLEAVED-NEXT: [[TMP19:%.*]] = sext <16 x i8> [[STRIDED_VEC10]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP20:%.*]] = mul nsw <16 x i32> [[TMP19]], [[TMP10]]
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE17]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI4]], <16 x i32> [[TMP20]])
-; CHECK-INTERLEAVED-NEXT: [[TMP21:%.*]] = sext <16 x i8> [[STRIDED_VEC11]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP22:%.*]] = mul nsw <16 x i32> [[TMP21]], [[TMP10]]
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE18]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI3]], <16 x i32> [[TMP22]])
-; CHECK-INTERLEAVED-NEXT: [[TMP23:%.*]] = sext <16 x i8> [[STRIDED_VEC12]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP24:%.*]] = mul nsw <16 x i32> [[TMP23]], [[TMP10]]
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE19]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI2]], <16 x i32> [[TMP24]])
-; CHECK-INTERLEAVED-NEXT: [[TMP25:%.*]] = sext <16 x i8> [[STRIDED_VEC13]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP26:%.*]] = mul nsw <16 x i32> [[TMP25]], [[TMP10]]
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE20]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP26]])
-; CHECK-INTERLEAVED-NEXT: [[TMP27:%.*]] = sext <16 x i8> [[STRIDED_VEC14]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP28:%.*]] = mul nsw <16 x i32> [[TMP27]], [[TMP10]]
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE21]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP28]])
+; CHECK-INTERLEAVED-NEXT: [[TMP19:%.*]] = mul nsw <16 x i32> [[TMP17]], [[TMP44]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE16]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI5]], <16 x i32> [[TMP19]])
+; CHECK-INTERLEAVED-NEXT: [[TMP20:%.*]] = sext <16 x i8> [[STRIDED_VEC10]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP22:%.*]] = mul nsw <16 x i32> [[TMP20]], [[TMP44]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE17]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI4]], <16 x i32> [[TMP22]])
+; CHECK-INTERLEAVED-NEXT: [[TMP23:%.*]] = sext <16 x i8> [[STRIDED_VEC11]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP25:%.*]] = mul nsw <16 x i32> [[TMP23]], [[TMP44]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE18]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI3]], <16 x i32> [[TMP25]])
+; CHECK-INTERLEAVED-NEXT: [[TMP26:%.*]] = sext <16 x i8> [[STRIDED_VEC12]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP28:%.*]] = mul nsw <16 x i32> [[TMP26]], [[TMP44]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE19]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI2]], <16 x i32> [[TMP28]])
+; CHECK-INTERLEAVED-NEXT: [[TMP45:%.*]] = sext <16 x i8> [[STRIDED_VEC13]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP31:%.*]] = mul nsw <16 x i32> [[TMP45]], [[TMP44]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE20]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP31]])
+; CHECK-INTERLEAVED-NEXT: [[TMP32:%.*]] = sext <16 x i8> [[STRIDED_VEC14]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP34:%.*]] = mul nsw <16 x i32> [[TMP32]], [[TMP44]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE21]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP34]])
; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-INTERLEAVED-NEXT: [[TMP29:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-INTERLEAVED-NEXT: br i1 [[TMP29]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP26:![0-9]+]]
; CHECK-INTERLEAVED: middle.block:
-; CHECK-INTERLEAVED-NEXT: [[TMP30:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE21]])
-; CHECK-INTERLEAVED-NEXT: [[TMP31:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE20]])
-; CHECK-INTERLEAVED-NEXT: [[TMP32:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE19]])
-; CHECK-INTERLEAVED-NEXT: [[TMP33:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE18]])
-; CHECK-INTERLEAVED-NEXT: [[TMP34:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE17]])
-; CHECK-INTERLEAVED-NEXT: [[TMP35:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE16]])
-; CHECK-INTERLEAVED-NEXT: [[TMP36:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE15]])
-; CHECK-INTERLEAVED-NEXT: [[TMP37:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE]])
+; CHECK-INTERLEAVED-NEXT: [[TMP36:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE21]])
+; CHECK-INTERLEAVED-NEXT: [[TMP37:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE20]])
+; CHECK-INTERLEAVED-NEXT: [[TMP38:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE19]])
+; CHECK-INTERLEAVED-NEXT: [[TMP39:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE18]])
+; CHECK-INTERLEAVED-NEXT: [[TMP40:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE17]])
+; CHECK-INTERLEAVED-NEXT: [[TMP41:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE16]])
+; CHECK-INTERLEAVED-NEXT: [[TMP42:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE15]])
+; CHECK-INTERLEAVED-NEXT: [[TMP43:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE]])
; CHECK-INTERLEAVED-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC]]
; CHECK-INTERLEAVED-NEXT: br i1 [[CMP_N]], label [[FOR_COND_FOR_COND_CLEANUP_CRIT_EDGE:%.*]], label [[SCALAR_PH]]
; CHECK-INTERLEAVED: scalar.ph:
@@ -2529,7 +2558,6 @@ define dso_local void @not_dotp_high_register_pressure(ptr %a, ptr %b, ptr %sum,
; CHECK-MAXBW-NEXT: [[VEC_PHI7:%.*]] = phi <4 x i32> [ [[TMP7]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
; CHECK-MAXBW-NEXT: [[TMP8:%.*]] = getelementptr inbounds nuw i8, ptr [[A]], i64 [[INDEX]]
; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP8]], align 1
-; CHECK-MAXBW-NEXT: [[TMP10:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-MAXBW-NEXT: [[TMP11:%.*]] = shl nsw i64 [[INDEX]], 3
; CHECK-MAXBW-NEXT: [[TMP12:%.*]] = getelementptr inbounds nuw i8, ptr [[B]], i64 [[TMP11]]
; CHECK-MAXBW-NEXT: [[WIDE_VEC:%.*]] = load <128 x i8>, ptr [[TMP12]], align 1
@@ -2541,42 +2569,43 @@ define dso_local void @not_dotp_high_register_pressure(ptr %a, ptr %b, ptr %sum,
; CHECK-MAXBW-NEXT: [[STRIDED_VEC12:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> <i32 5, i32 13, i32 21, i32 29, i32 37, i32 45, i32 53, i32 61, i32 69, i32 77, i32 85, i32 93, i32 101, i32 109, i32 117, i32 125>
; CHECK-MAXBW-NEXT: [[STRIDED_VEC13:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> <i32 6, i32 14, i32 22, i32 30, i32 38, i32 46, i32 54, i32 62, i32 70, i32 78, i32 86, i32 94, i32 102, i32 110, i32 118, i32 126>
; CHECK-MAXBW-NEXT: [[STRIDED_VEC14:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> <i32 7, i32 15, i32 23, i32 31, i32 39, i32 47, i32 55, i32 63, i32 71, i32 79, i32 87, i32 95, i32 103, i32 111, i32 119, i32 127>
-; CHECK-MAXBW-NEXT: [[TMP13:%.*]] = sext <16 x i8> [[STRIDED_VEC]] to <16 x i32>
-; CHECK-MAXBW-NEXT: [[TMP14:%.*]] = mul nsw <16 x i32> [[TMP13]], [[TMP10]]
-; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI7]], <16 x i32> [[TMP14]])
-; CHECK-MAXBW-NEXT: [[TMP15:%.*]] = sext <16 x i8> [[STRIDED_VEC8]] to <16 x i32>
-; CHECK-MAXBW-NEXT: [[TMP16:%.*]] = mul nsw <16 x i32> [[TMP15]], [[TMP10]]
+; CHECK-MAXBW-NEXT: [[TMP35:%.*]] = sext <16 x i8> [[STRIDED_VEC]] to <16 x i32>
+; CHECK-MAXBW-NEXT: [[TMP44:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
+; CHECK-MAXBW-NEXT: [[TMP13:%.*]] = mul nsw <16 x i32> [[TMP35]], [[TMP44]]
+; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI7]], <16 x i32> [[TMP13]])
+; CHECK-MAXBW-NEXT: [[TMP14:%.*]] = sext <16 x i8> [[STRIDED_VEC8]] to <16 x i32>
+; CHECK-MAXBW-NEXT: [[TMP16:%.*]] = mul nsw <16 x i32> [[TMP14]], [[TMP44]]
; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE15]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI6]], <16 x i32> [[TMP16]])
; CHECK-MAXBW-NEXT: [[TMP17:%.*]] = sext <16 x i8> [[STRIDED_VEC9]] to <16 x i32>
-; CHECK-MAXBW-NEXT: [[TMP18:%.*]] = mul nsw <16 x i32> [[TMP17]], [[TMP10]]
-; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE16]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI5]], <16 x i32> [[TMP18]])
-; CHECK-MAXBW-NEXT: [[TMP19:%.*]] = sext <16 x i8> [[STRIDED_VEC10]] to <16 x i32>
-; CHECK-MAXBW-NEXT: [[TMP20:%.*]] = mul nsw <16 x i32> [[TMP19]], [[TMP10]]
-; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE17]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI4]], <16 x i32> [[TMP20]])
-; CHECK-MAXBW-NEXT: [[TMP21:%.*]] = sext <16 x i8> [[STRIDED_VEC11]] to <16 x i32>
-; CHECK-MAXBW-NEXT: [[TMP22:%.*]] = mul nsw <16 x i32> [[TMP21]], [[TMP10]]
-; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE18]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI3]], <16 x i32> [[TMP22]])
-; CHECK-MAXBW-NEXT: [[TMP23:%.*]] = sext <16 x i8> [[STRIDED_VEC12]] to <16 x i32>
-; CHECK-MAXBW-NEXT: [[TMP24:%.*]] = mul nsw <16 x i32> [[TMP23]], [[TMP10]]
-; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE19]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI2]], <16 x i32> [[TMP24]])
-; CHECK-MAXBW-NEXT: [[TMP25:%.*]] = sext <16 x i8> [[STRIDED_VEC13]] to <16 x i32>
-; CHECK-MAXBW-NEXT: [[TMP26:%.*]] = mul nsw <16 x i32> [[TMP25]], [[TMP10]]
-; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE20]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP26]])
-; CHECK-MAXBW-NEXT: [[TMP27:%.*]] = sext <16 x i8> [[STRIDED_VEC14]] to <16 x i32>
-; CHECK-MAXBW-NEXT: [[TMP28:%.*]] = mul nsw <16 x i32> [[TMP27]], [[TMP10]]
-; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE21]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP28]])
+; CHECK-MAXBW-NEXT: [[TMP19:%.*]] = mul nsw <16 x i32> [[TMP17]], [[TMP44]]
+; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE16]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI5]], <16 x i32> [[TMP19]])
+; CHECK-MAXBW-NEXT: [[TMP20:%.*]] = sext <16 x i8> [[STRIDED_VEC10]] to <16 x i32>
+; CHECK-MAXBW-NEXT: [[TMP22:%.*]] = mul nsw <16 x i32> [[TMP20]], [[TMP44]]
+; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE17]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI4]], <16 x i32> [[TMP22]])
+; CHECK-MAXBW-NEXT: [[TMP23:%.*]] = sext <16 x i8> [[STRIDED_VEC11]] to <16 x i32>
+; CHECK-MAXBW-NEXT: [[TMP25:%.*]] = mul nsw <16 x i32> [[TMP23]], [[TMP44]]
+; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE18]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI3]], <16 x i32> [[TMP25]])
+; CHECK-MAXBW-NEXT: [[TMP26:%.*]] = sext <16 x i8> [[STRIDED_VEC12]] to <16 x i32>
+; CHECK-MAXBW-NEXT: [[TMP28:%.*]] = mul nsw <16 x i32> [[TMP26]], [[TMP44]]
+; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE19]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI2]], <16 x i32> [[TMP28]])
+; CHECK-MAXBW-NEXT: [[TMP45:%.*]] = sext <16 x i8> [[STRIDED_VEC13]] to <16 x i32>
+; CHECK-MAXBW-NEXT: [[TMP31:%.*]] = mul nsw <16 x i32> [[TMP45]], [[TMP44]]
+; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE20]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP31]])
+; CHECK-MAXBW-NEXT: [[TMP32:%.*]] = sext <16 x i8> [[STRIDED_VEC14]] to <16 x i32>
+; CHECK-MAXBW-NEXT: [[TMP34:%.*]] = mul nsw <16 x i32> [[TMP32]], [[TMP44]]
+; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE21]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP34]])
; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-MAXBW-NEXT: [[TMP29:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-MAXBW-NEXT: br i1 [[TMP29]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP27:![0-9]+]]
+; CHECK-MAXBW-NEXT: br i1 [[TMP29]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP25:![0-9]+]]
; CHECK-MAXBW: middle.block:
-; CHECK-MAXBW-NEXT: [[TMP30:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE21]])
-; CHECK-MAXBW-NEXT: [[TMP31:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE20]])
-; CHECK-MAXBW-NEXT: [[TMP32:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE19]])
-; CHECK-MAXBW-NEXT: [[TMP33:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE18]])
-; CHECK-MAXBW-NEXT: [[TMP34:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE17]])
-; CHECK-MAXBW-NEXT: [[TMP35:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE16]])
-; CHECK-MAXBW-NEXT: [[TMP36:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE15]])
-; CHECK-MAXBW-NEXT: [[TMP37:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE]])
+; CHECK-MAXBW-NEXT: [[TMP36:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE21]])
+; CHECK-MAXBW-NEXT: [[TMP37:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE20]])
+; CHECK-MAXBW-NEXT: [[TMP38:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE19]])
+; CHECK-MAXBW-NEXT: [[TMP39:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE18]])
+; CHECK-MAXBW-NEXT: [[TMP40:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE17]])
+; CHECK-MAXBW-NEXT: [[TMP41:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE16]])
+; CHECK-MAXBW-NEXT: [[TMP42:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE15]])
+; CHECK-MAXBW-NEXT: [[TMP43:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE]])
; CHECK-MAXBW-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC]]
; CHECK-MAXBW-NEXT: br i1 [[CMP_N]], label [[FOR_COND_FOR_COND_CLEANUP_CRIT_EDGE:%.*]], label [[SCALAR_PH]]
; CHECK-MAXBW: scalar.ph:
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-interleave.ll b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-interleave.ll
index b308b925181b1..bd9fae6cd610b 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-interleave.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-interleave.ll
@@ -23,12 +23,12 @@ define i32 @partial_reduce_with_non_constant_start_value(ptr %src, i32 %rdx.star
; IC2-NEXT: [[TMP3:%.*]] = getelementptr inbounds i8, ptr [[TMP1]], i32 16
; IC2-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1
; IC2-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP3]], align 1
-; IC2-NEXT: [[TMP4:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
-; IC2-NEXT: [[TMP5:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
-; IC2-NEXT: [[TMP6:%.*]] = mul nuw nsw <16 x i32> [[TMP4]], [[TMP4]]
+; IC2-NEXT: [[TMP5:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; IC2-NEXT: [[TMP7:%.*]] = mul nuw nsw <16 x i32> [[TMP5]], [[TMP5]]
-; IC2-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP6]])
-; IC2-NEXT: [[PARTIAL_REDUCE3]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP7]])
+; IC2-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP7]])
+; IC2-NEXT: [[TMP8:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
+; IC2-NEXT: [[TMP6:%.*]] = mul nuw nsw <16 x i32> [[TMP8]], [[TMP8]]
+; IC2-NEXT: [[PARTIAL_REDUCE3]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP6]])
; IC2-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
; IC2-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; IC2-NEXT: br i1 [[TMP10]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
@@ -80,18 +80,18 @@ define i32 @partial_reduce_with_non_constant_start_value(ptr %src, i32 %rdx.star
; IC4-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP3]], align 1
; IC4-NEXT: [[WIDE_LOAD5:%.*]] = load <16 x i8>, ptr [[TMP4]], align 1
; IC4-NEXT: [[WIDE_LOAD6:%.*]] = load <16 x i8>, ptr [[TMP5]], align 1
-; IC4-NEXT: [[TMP6:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
+; IC4-NEXT: [[TMP9:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
+; IC4-NEXT: [[TMP13:%.*]] = mul nuw nsw <16 x i32> [[TMP9]], [[TMP9]]
+; IC4-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP13]])
; IC4-NEXT: [[TMP7:%.*]] = zext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>
-; IC4-NEXT: [[TMP8:%.*]] = zext <16 x i8> [[WIDE_LOAD5]] to <16 x i32>
-; IC4-NEXT: [[TMP9:%.*]] = zext <16 x i8> [[WIDE_LOAD6]] to <16 x i32>
-; IC4-NEXT: [[TMP10:%.*]] = mul nuw nsw <16 x i32> [[TMP6]], [[TMP6]]
; IC4-NEXT: [[TMP11:%.*]] = mul nuw nsw <16 x i32> [[TMP7]], [[TMP7]]
-; IC4-NEXT: [[TMP12:%.*]] = mul nuw nsw <16 x i32> [[TMP8]], [[TMP8]]
-; IC4-NEXT: [[TMP13:%.*]] = mul nuw nsw <16 x i32> [[TMP9]], [[TMP9]]
-; IC4-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP10]])
; IC4-NEXT: [[PARTIAL_REDUCE7]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP11]])
+; IC4-NEXT: [[TMP10:%.*]] = zext <16 x i8> [[WIDE_LOAD5]] to <16 x i32>
+; IC4-NEXT: [[TMP12:%.*]] = mul nuw nsw <16 x i32> [[TMP10]], [[TMP10]]
; IC4-NEXT: [[PARTIAL_REDUCE8]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI2]], <16 x i32> [[TMP12]])
-; IC4-NEXT: [[PARTIAL_REDUCE9]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI3]], <16 x i32> [[TMP13]])
+; IC4-NEXT: [[TMP14:%.*]] = zext <16 x i8> [[WIDE_LOAD6]] to <16 x i32>
+; IC4-NEXT: [[TMP16:%.*]] = mul nuw nsw <16 x i32> [[TMP14]], [[TMP14]]
+; IC4-NEXT: [[PARTIAL_REDUCE9]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI3]], <16 x i32> [[TMP16]])
; IC4-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 64
; IC4-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; IC4-NEXT: br i1 [[TMP18]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-sub.ll b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-sub.ll
index 7bb47157d7cf2..6dae09ef97e1c 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-sub.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-sub.ll
@@ -18,11 +18,11 @@ define i32 @dotp(ptr %a, ptr %b) #0 {
; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
; CHECK-INTERLEAVE1-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
-; CHECK-INTERLEAVE1-NEXT: [[TMP1:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD1:%.*]] = load <16 x i8>, ptr [[TMP10]], align 1
-; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
-; CHECK-INTERLEAVE1-NEXT: [[TMP4:%.*]] = mul <16 x i32> [[TMP3]], [[TMP1]]
+; CHECK-INTERLEAVE1-NEXT: [[TMP2:%.*]] = zext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
+; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
+; CHECK-INTERLEAVE1-NEXT: [[TMP4:%.*]] = mul <16 x i32> [[TMP2]], [[TMP3]]
; CHECK-INTERLEAVE1-NEXT: [[TMP5:%.*]] = sub <16 x i32> zeroinitializer, [[TMP4]]
; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP5]])
; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
@@ -48,19 +48,19 @@ define i32 @dotp(ptr %a, ptr %b) #0 {
; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[TMP7]], i32 16
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP7]], align 1
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP2:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
-; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[TMP14]], i32 16
+; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[TMP14]], i32 16
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP14]], align 1
-; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP5]], align 1
-; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = zext <16 x i8> [[WIDE_LOAD3]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP15:%.*]] = zext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = mul <16 x i32> [[TMP6]], [[TMP2]]
-; CHECK-INTERLEAVED-NEXT: [[TMP9:%.*]] = mul <16 x i32> [[TMP15]], [[TMP3]]
-; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = sub <16 x i32> zeroinitializer, [[TMP8]]
-; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = sub <16 x i32> zeroinitializer, [[TMP9]]
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP10]])
+; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP3]], align 1
+; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = zext <16 x i8> [[WIDE_LOAD3]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = mul <16 x i32> [[TMP4]], [[TMP5]]
+; CHECK-INTERLEAVED-NEXT: [[TMP15:%.*]] = sub <16 x i32> zeroinitializer, [[TMP6]]
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP15]])
+; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = zext <16 x i8> [[WIDE_LOAD4]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP9:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = mul <16 x i32> [[TMP8]], [[TMP9]]
+; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = sub <16 x i32> zeroinitializer, [[TMP10]]
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE5]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP11]])
; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
@@ -78,27 +78,27 @@ define i32 @dotp(ptr %a, ptr %b) #0 {
; CHECK-MAXBW-NEXT: br label [[ENTRY:%.*]]
; CHECK-MAXBW: vector.ph:
; CHECK-MAXBW-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 8
+; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 16
; CHECK-MAXBW-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]]
; CHECK-MAXBW-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]]
; CHECK-MAXBW-NEXT: br label [[FOR_BODY:%.*]]
; CHECK-MAXBW: vector.body:
; CHECK-MAXBW-NEXT: [[IV:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ]
-; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 2 x i32> [ zeroinitializer, [[ENTRY]] ], [ [[PARTIAL_REDUCE:%.*]], [[FOR_BODY]] ]
+; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, [[ENTRY]] ], [ [[PARTIAL_REDUCE:%.*]], [[FOR_BODY]] ]
; CHECK-MAXBW-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[A]], i64 [[IV]]
-; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i8>, ptr [[TMP7]], align 1
-; CHECK-MAXBW-NEXT: [[TMP9:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD]] to <vscale x 8 x i32>
+; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 16 x i8>, ptr [[TMP7]], align 1
; CHECK-MAXBW-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[B]], i64 [[IV]]
-; CHECK-MAXBW-NEXT: [[WIDE_LOAD1:%.*]] = load <vscale x 8 x i8>, ptr [[TMP10]], align 1
-; CHECK-MAXBW-NEXT: [[TMP12:%.*]] = zext <vscale x 8 x i8> [[WIDE_LOAD1]] to <vscale x 8 x i32>
-; CHECK-MAXBW-NEXT: [[TMP13:%.*]] = mul <vscale x 8 x i32> [[TMP12]], [[TMP9]]
-; CHECK-MAXBW-NEXT: [[TMP14:%.*]] = sub <vscale x 8 x i32> zeroinitializer, [[TMP13]]
-; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 2 x i32> @llvm.vector.partial.reduce.add.nxv2i32.nxv8i32(<vscale x 2 x i32> [[VEC_PHI]], <vscale x 8 x i32> [[TMP14]])
+; CHECK-MAXBW-NEXT: [[WIDE_LOAD1:%.*]] = load <vscale x 16 x i8>, ptr [[TMP10]], align 1
+; CHECK-MAXBW-NEXT: [[TMP4:%.*]] = zext <vscale x 16 x i8> [[WIDE_LOAD1]] to <vscale x 16 x i32>
+; CHECK-MAXBW-NEXT: [[TMP5:%.*]] = zext <vscale x 16 x i8> [[WIDE_LOAD]] to <vscale x 16 x i32>
+; CHECK-MAXBW-NEXT: [[TMP6:%.*]] = mul <vscale x 16 x i32> [[TMP4]], [[TMP5]]
+; CHECK-MAXBW-NEXT: [[TMP8:%.*]] = sub <vscale x 16 x i32> zeroinitializer, [[TMP6]]
+; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 4 x i32> @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32(<vscale x 4 x i32> [[VEC_PHI]], <vscale x 16 x i32> [[TMP8]])
; CHECK-MAXBW-NEXT: [[IV_NEXT]] = add nuw i64 [[IV]], [[TMP3]]
; CHECK-MAXBW-NEXT: [[TMP16:%.*]] = icmp eq i64 [[IV_NEXT]], [[N_VEC]]
; CHECK-MAXBW-NEXT: br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[FOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK-MAXBW: middle.block:
-; CHECK-MAXBW-NEXT: [[TMP17:%.*]] = call i32 @llvm.vector.reduce.add.nxv2i32(<vscale x 2 x i32> [[PARTIAL_REDUCE]])
+; CHECK-MAXBW-NEXT: [[TMP9:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[PARTIAL_REDUCE]])
; CHECK-MAXBW-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]]
; CHECK-MAXBW-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH:%.*]]
; CHECK-MAXBW: scalar.ph:
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce.ll b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce.ll
index 70532ad6586c7..46ec858d7455c 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce.ll
@@ -45,8 +45,8 @@ define i32 @zext_add_reduc_i8_i32_sve(ptr %a) #0 {
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP0]], align 1
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1
; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP3]])
+; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE3]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP4]])
; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
@@ -138,8 +138,8 @@ define i32 @zext_add_reduc_i8_i32_neon(ptr %a) #2 {
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP0]], align 1
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1
; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP3]])
+; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE3]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP4]])
; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
@@ -227,8 +227,8 @@ define i64 @zext_add_reduc_i8_i64(ptr %a) #0 {
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP3]], align 1
; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i64>
-; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i64>
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v16i64(<2 x i64> [[VEC_PHI]], <16 x i64> [[TMP4]])
+; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i64>
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE3]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v16i64(<2 x i64> [[VEC_PHI1]], <16 x i64> [[TMP5]])
; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
@@ -321,8 +321,8 @@ define i64 @zext_add_reduc_i16_i64(ptr %a) #0 {
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i16>, ptr [[TMP1]], align 2
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD2:%.*]] = load <8 x i16>, ptr [[TMP3]], align 2
; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = zext <8 x i16> [[WIDE_LOAD]] to <8 x i64>
-; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = zext <8 x i16> [[WIDE_LOAD2]] to <8 x i64>
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v8i64(<2 x i64> [[VEC_PHI]], <8 x i64> [[TMP4]])
+; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = zext <8 x i16> [[WIDE_LOAD2]] to <8 x i64>
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE3]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v8i64(<2 x i64> [[VEC_PHI1]], <8 x i64> [[TMP5]])
; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
@@ -421,12 +421,12 @@ define i32 @zext_add_reduc_i8_i32_has_neon_dotprod(ptr %a) #1 {
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD5:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD6:%.*]] = load <16 x i8>, ptr [[TMP8]], align 1
; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = zext <16 x i8> [[WIDE_LOAD5]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = zext <16 x i8> [[WIDE_LOAD6]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP4]])
+; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE3]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP5]])
+; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = zext <16 x i8> [[WIDE_LOAD5]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE8]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI2]], <16 x i32> [[TMP10]])
+; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = zext <16 x i8> [[WIDE_LOAD6]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE9]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI3]], <16 x i32> [[TMP7]])
; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 64
; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
@@ -811,8 +811,8 @@ define i32 @sext_add_reduc_i8_i32(ptr %a) #0 {
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP3]], align 1
; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
-; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP4]])
+; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = sext <16 x i8> [[WIDE_LOAD2]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE3]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP5]])
; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
@@ -1000,11 +1000,10 @@ define i32 @add_of_loop_invariant_zext(i32 %a, ptr %b, i8 %c, i32 %d) #0 {
; CHECK-INTERLEAVE1: vector.ph:
; CHECK-INTERLEAVE1-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[TMP0]], 16
; CHECK-INTERLEAVE1-NEXT: [[N_VEC:%.*]] = sub i32 [[TMP0]], [[N_MOD_VF]]
-; CHECK-INTERLEAVE1-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i8> poison, i8 [[C]], i64 0
-; CHECK-INTERLEAVE1-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <16 x i8> [[BROADCAST_SPLATINSERT]], <16 x i8> poison, <16 x i32> zeroinitializer
; CHECK-INTERLEAVE1-NEXT: [[TMP1:%.*]] = add i32 [[D]], [[N_VEC]]
; CHECK-INTERLEAVE1-NEXT: [[TMP2:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[A]], i32 0
-; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[BROADCAST_SPLAT]] to <16 x i32>
+; CHECK-INTERLEAVE1-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i8> poison, i8 [[C]], i64 0
+; CHECK-INTERLEAVE1-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <16 x i8> [[BROADCAST_SPLATINSERT]], <16 x i8> poison, <16 x i32> zeroinitializer
; CHECK-INTERLEAVE1-NEXT: br label [[FOR_BODY:%.*]]
; CHECK-INTERLEAVE1: vector.body:
; CHECK-INTERLEAVE1-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[FOR_BODY]] ]
@@ -1012,7 +1011,8 @@ define i32 @add_of_loop_invariant_zext(i32 %a, ptr %b, i8 %c, i32 %d) #0 {
; CHECK-INTERLEAVE1-NEXT: [[OFFSET_IDX:%.*]] = add i32 [[D]], [[INDEX]]
; CHECK-INTERLEAVE1-NEXT: [[TMP4:%.*]] = getelementptr inbounds i8, ptr [[B]], i32 [[OFFSET_IDX]]
; CHECK-INTERLEAVE1-NEXT: store <16 x i8> zeroinitializer, ptr [[TMP4]], align 1
-; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP3]])
+; CHECK-INTERLEAVE1-NEXT: [[TMP5:%.*]] = zext <16 x i8> [[BROADCAST_SPLAT]] to <16 x i32>
+; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP5]])
; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 16
; CHECK-INTERLEAVE1-NEXT: [[TMP7:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-INTERLEAVE1-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[FOR_BODY]], !llvm.loop [[LOOP19:![0-9]+]]
@@ -1031,23 +1031,23 @@ define i32 @add_of_loop_invariant_zext(i32 %a, ptr %b, i8 %c, i32 %d) #0 {
; CHECK-INTERLEAVED: vector.ph:
; CHECK-INTERLEAVED-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[TMP2]], 32
; CHECK-INTERLEAVED-NEXT: [[N_VEC:%.*]] = sub i32 [[TMP2]], [[N_MOD_VF]]
+; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = add i32 [[D]], [[N_VEC]]
+; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[A]], i32 0
; CHECK-INTERLEAVED-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i8> poison, i8 [[C]], i64 0
; CHECK-INTERLEAVED-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <16 x i8> [[BROADCAST_SPLATINSERT]], <16 x i8> poison, <16 x i32> zeroinitializer
-; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = add i32 [[D]], [[N_VEC]]
-; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[A]], i32 0
-; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[BROADCAST_SPLAT]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-INTERLEAVED: vector.body:
; CHECK-INTERLEAVED-NEXT: [[VEC_PHI1:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP22:%.*]], [[VECTOR_BODY]] ]
-; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ [[TMP5]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
+; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ [[TMP3]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
; CHECK-INTERLEAVED-NEXT: [[VEC_PHI2:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE2:%.*]], [[VECTOR_BODY]] ]
; CHECK-INTERLEAVED-NEXT: [[OFFSET_IDX:%.*]] = add i32 [[D]], [[VEC_PHI1]]
; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = getelementptr inbounds i8, ptr [[B]], i32 [[OFFSET_IDX]]
; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = getelementptr inbounds i8, ptr [[TMP4]], i32 16
; CHECK-INTERLEAVED-NEXT: store <16 x i8> zeroinitializer, ptr [[TMP4]], align 1
; CHECK-INTERLEAVED-NEXT: store <16 x i8> zeroinitializer, ptr [[TMP6]], align 1
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP3]])
-; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE2]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI2]], <16 x i32> [[TMP3]])
+; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = zext <16 x i8> [[BROADCAST_SPLAT]] to <16 x i32>
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP5]])
+; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE2]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI2]], <16 x i32> [[TMP5]])
; CHECK-INTERLEAVED-NEXT: [[TMP22]] = add nuw i32 [[VEC_PHI1]], 32
; CHECK-INTERLEAVED-NEXT: [[TMP9:%.*]] = icmp eq i32 [[TMP22]], [[N_VEC]]
; CHECK-INTERLEAVED-NEXT: br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP19:![0-9]+]]
@@ -1071,11 +1071,10 @@ define i32 @add_of_loop_invariant_zext(i32 %a, ptr %b, i8 %c, i32 %d) #0 {
; CHECK-MAXBW-NEXT: [[TMP4:%.*]] = mul nuw i32 [[TMP3]], 16
; CHECK-MAXBW-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[TMP0]], [[TMP4]]
; CHECK-MAXBW-NEXT: [[N_VEC:%.*]] = sub i32 [[TMP0]], [[N_MOD_VF]]
-; CHECK-MAXBW-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 16 x i8> poison, i8 [[C]], i64 0
-; CHECK-MAXBW-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 16 x i8> [[BROADCAST_SPLATINSERT]], <vscale x 16 x i8> poison, <vscale x 16 x i32> zeroinitializer
; CHECK-MAXBW-NEXT: [[TMP7:%.*]] = add i32 [[D]], [[N_VEC]]
; CHECK-MAXBW-NEXT: [[TMP8:%.*]] = insertelement <vscale x 4 x i32> zeroinitializer, i32 [[A]], i32 0
-; CHECK-MAXBW-NEXT: [[TMP9:%.*]] = zext <vscale x 16 x i8> [[BROADCAST_SPLAT]] to <vscale x 16 x i32>
+; CHECK-MAXBW-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 16 x i8> poison, i8 [[C]], i64 0
+; CHECK-MAXBW-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 16 x i8> [[BROADCAST_SPLATINSERT]], <vscale x 16 x i8> poison, <vscale x 16 x i32> zeroinitializer
; CHECK-MAXBW-NEXT: br label [[FOR_BODY:%.*]]
; CHECK-MAXBW: vector.body:
; CHECK-MAXBW-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[FOR_BODY]] ]
@@ -1083,6 +1082,7 @@ define i32 @add_of_loop_invariant_zext(i32 %a, ptr %b, i8 %c, i32 %d) #0 {
; CHECK-MAXBW-NEXT: [[OFFSET_IDX:%.*]] = add i32 [[D]], [[INDEX]]
; CHECK-MAXBW-NEXT: [[TMP10:%.*]] = getelementptr inbounds i8, ptr [[B]], i32 [[OFFSET_IDX]]
; CHECK-MAXBW-NEXT: store <vscale x 16 x i8> zeroinitializer, ptr [[TMP10]], align 1
+; CHECK-MAXBW-NEXT: [[TMP9:%.*]] = zext <vscale x 16 x i8> [[BROADCAST_SPLAT]] to <vscale x 16 x i32>
; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 4 x i32> @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32(<vscale x 4 x i32> [[VEC_PHI]], <vscale x 16 x i32> [[TMP9]])
; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], [[TMP4]]
; CHECK-MAXBW-NEXT: [[TMP12:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
@@ -1164,12 +1164,12 @@ define i64 @sext_reduction_i32_to_i64(ptr %arr, i64 %n) #1 {
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD5:%.*]] = load <4 x i32>, ptr [[TMP14]], align 4
; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD6:%.*]] = load <4 x i32>, ptr [[TMP3]], align 4
; CHECK-INTERLEAVED-NEXT: [[TMP15:%.*]] = sext <4 x i32> [[WIDE_LOAD]] to <4 x i64>
-; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = sext <4 x i32> [[WIDE_LOAD4]] to <4 x i64>
-; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = sext <4 x i32> [[WIDE_LOAD5]] to <4 x i64>
-; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = sext <4 x i32> [[WIDE_LOAD6]] to <4 x i64>
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v4i64(<2 x i64> [[VEC_PHI]], <4 x i64> [[TMP15]])
+; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = sext <4 x i32> [[WIDE_LOAD4]] to <4 x i64>
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE7]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v4i64(<2 x i64> [[VEC_PHI1]], <4 x i64> [[TMP5]])
+; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = sext <4 x i32> [[WIDE_LOAD5]] to <4 x i64>
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE8]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v4i64(<2 x i64> [[VEC_PHI2]], <4 x i64> [[TMP6]])
+; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = sext <4 x i32> [[WIDE_LOAD6]] to <4 x i64>
; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE9]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v4i64(<2 x i64> [[VEC_PHI3]], <4 x i64> [[TMP7]])
; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/reg-usage.ll b/llvm/test/Transforms/LoopVectorize/AArch64/reg-usage.ll
index 25ee10077fb47..70685c1c3fe12 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/reg-usage.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/reg-usage.ll
@@ -192,7 +192,7 @@ define i32 @dotp_unrolled(i32 %num_out, i64 %num_in, ptr %a, ptr %b) {
; CHECK: LV(REG): VF = 16
; CHECK-NEXT: LV(REG): Found max usage: 2 item
; CHECK-NEXT: LV(REG): RegisterClass: Generic::ScalarRC, 9 registers
-; CHECK-NEXT: LV(REG): RegisterClass: Generic::VectorRC, 12 registers
+; CHECK-NEXT: LV(REG): RegisterClass: Generic::VectorRC, 6 registers
; CHECK-NEXT: LV(REG): Found invariant usage: 1 item
entry:
br label %for.body
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/vplan-printing.ll b/llvm/test/Transforms/LoopVectorize/AArch64/vplan-printing.ll
index d4e5dea3d4aab..49f663f5703b6 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/vplan-printing.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/vplan-printing.ll
@@ -23,18 +23,15 @@ define i32 @print_partial_reduction(ptr %a, ptr %b) {
; CHECK-NEXT: <x1> vector loop: {
; CHECK-NEXT: vector.body:
; CHECK-NEXT: EMIT vp<[[CAN_IV:%.+]]> = CANONICAL-INDUCTION ir<0>, vp<[[CAN_IV_NEXT:%.+]]>
-; CHECK-NEXT: WIDEN-REDUCTION-PHI ir<[[ACC:%.+]]> = phi vp<[[RDX_START]]>, ir<[[REDUCE:%.+]]> (VF scaled by 1/4)
+; CHECK-NEXT: WIDEN-REDUCTION-PHI ir<[[ACC:%.+]]> = phi vp<[[RDX_START]]>, vp<[[REDUCE:%.+]]> (VF scaled by 1/4)
; CHECK-NEXT: vp<[[STEPS:%.+]]> = SCALAR-STEPS vp<[[CAN_IV]]>, ir<1>, vp<[[VF]]>
; CHECK-NEXT: CLONE ir<%gep.a> = getelementptr ir<%a>, vp<[[STEPS]]>
; CHECK-NEXT: vp<[[PTR_A:%.+]]> = vector-pointer ir<%gep.a>
; CHECK-NEXT: WIDEN ir<%load.a> = load vp<[[PTR_A]]>
-; CHECK-NEXT: WIDEN-CAST ir<%ext.a> = zext ir<%load.a> to i32
; CHECK-NEXT: CLONE ir<%gep.b> = getelementptr ir<%b>, vp<[[STEPS]]>
; CHECK-NEXT: vp<[[PTR_B:%.+]]> = vector-pointer ir<%gep.b>
; CHECK-NEXT: WIDEN ir<%load.b> = load vp<[[PTR_B]]>
-; CHECK-NEXT: WIDEN-CAST ir<%ext.b> = zext ir<%load.b> to i32
-; CHECK-NEXT: WIDEN ir<%mul> = mul ir<%ext.b>, ir<%ext.a>
-; CHECK-NEXT: PARTIAL-REDUCE ir<[[REDUCE]]> = add ir<[[ACC]]>, ir<%mul>
+; CHECK-NEXT: EXPRESSION vp<[[REDUCE]]> = ir<[[ACC]]> + partial.reduce.add (mul (ir<%load.b> zext to i32), (ir<%load.a> zext to i32))
; CHECK-NEXT: EMIT vp<[[CAN_IV_NEXT]]> = add nuw vp<[[CAN_IV]]>, vp<[[VFxUF]]>
; CHECK-NEXT: EMIT branch-on-count vp<[[CAN_IV_NEXT]]>, vp<[[VEC_TC]]>
; CHECK-NEXT: No successors
@@ -42,7 +39,7 @@ define i32 @print_partial_reduction(ptr %a, ptr %b) {
; CHECK-NEXT: Successor(s): middle.block
; CHECK-EMPTY:
; CHECK-NEXT: middle.block:
-; CHECK-NEXT: EMIT vp<[[RED_RESULT:%.+]]> = compute-reduction-result ir<[[ACC]]>, ir<[[REDUCE]]>
+; CHECK-NEXT: EMIT vp<[[RED_RESULT:%.+]]> = compute-reduction-result ir<[[ACC]]>, vp<[[REDUCE]]>
; CHECK-NEXT: EMIT vp<[[CMP:%.+]]> = icmp eq ir<1024>, vp<[[VEC_TC]]>
; CHECK-NEXT: EMIT branch-on-cond vp<[[CMP]]>
; CHECK-NEXT: Successor(s): ir-bb<exit>, scalar.ph
@@ -89,10 +86,10 @@ define i32 @print_partial_reduction(ptr %a, ptr %b) {
; CHECK-NEXT: WIDEN-REDUCTION-PHI ir<%accum> = phi vp<[[RDX_START]]>, ir<%add> (VF scaled by 1/4)
; CHECK-NEXT: CLONE ir<%gep.a> = getelementptr ir<%a>, vp<[[EP_IV]]>
; CHECK-NEXT: WIDEN ir<%load.a> = load ir<%gep.a>
-; CHECK-NEXT: WIDEN-CAST ir<%ext.a> = zext ir<%load.a> to i32
; CHECK-NEXT: CLONE ir<%gep.b> = getelementptr ir<%b>, vp<[[EP_IV]]>
; CHECK-NEXT: WIDEN ir<%load.b> = load ir<%gep.b>
; CHECK-NEXT: WIDEN-CAST ir<%ext.b> = zext ir<%load.b> to i32
+; CHECK-NEXT: WIDEN-CAST ir<%ext.a> = zext ir<%load.a> to i32
; CHECK-NEXT: WIDEN ir<%mul> = mul ir<%ext.b>, ir<%ext.a>
; CHECK-NEXT: PARTIAL-REDUCE ir<%add> = add ir<%accum>, ir<%mul>
; CHECK-NEXT: EMIT vp<[[EP_IV_NEXT:%.+]]> = add nuw vp<[[EP_IV]]>, ir<16>
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/partial-reduce-dot-product.ll b/llvm/test/Transforms/LoopVectorize/RISCV/partial-reduce-dot-product.ll
index 0a9b1e0af48bc..61e3a1848ceed 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/partial-reduce-dot-product.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/partial-reduce-dot-product.ll
@@ -60,10 +60,10 @@ define i32 @vqdot(ptr %a, ptr %b) #0 {
; ZVQDOTQ-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 1 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
; ZVQDOTQ-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; ZVQDOTQ-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i8>, ptr [[TMP6]], align 1
-; ZVQDOTQ-NEXT: [[TMP8:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD]] to <vscale x 4 x i32>
; ZVQDOTQ-NEXT: [[TMP9:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; ZVQDOTQ-NEXT: [[WIDE_LOAD1:%.*]] = load <vscale x 4 x i8>, ptr [[TMP9]], align 1
; ZVQDOTQ-NEXT: [[TMP11:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD1]] to <vscale x 4 x i32>
+; ZVQDOTQ-NEXT: [[TMP8:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD]] to <vscale x 4 x i32>
; ZVQDOTQ-NEXT: [[TMP12:%.*]] = mul <vscale x 4 x i32> [[TMP11]], [[TMP8]]
; ZVQDOTQ-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 1 x i32> @llvm.vector.partial.reduce.add.nxv1i32.nxv4i32(<vscale x 1 x i32> [[VEC_PHI]], <vscale x 4 x i32> [[TMP12]])
; ZVQDOTQ-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
@@ -125,17 +125,17 @@ define i32 @vqdot(ptr %a, ptr %b) #0 {
; FIXED-ZVQDOTQ-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP0]], i32 8
; FIXED-ZVQDOTQ-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i8>, ptr [[TMP0]], align 1
; FIXED-ZVQDOTQ-NEXT: [[WIDE_LOAD2:%.*]] = load <8 x i8>, ptr [[TMP2]], align 1
-; FIXED-ZVQDOTQ-NEXT: [[TMP3:%.*]] = sext <8 x i8> [[WIDE_LOAD]] to <8 x i32>
-; FIXED-ZVQDOTQ-NEXT: [[TMP4:%.*]] = sext <8 x i8> [[WIDE_LOAD2]] to <8 x i32>
; FIXED-ZVQDOTQ-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; FIXED-ZVQDOTQ-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP5]], i32 8
; FIXED-ZVQDOTQ-NEXT: [[WIDE_LOAD3:%.*]] = load <8 x i8>, ptr [[TMP5]], align 1
; FIXED-ZVQDOTQ-NEXT: [[WIDE_LOAD4:%.*]] = load <8 x i8>, ptr [[TMP7]], align 1
; FIXED-ZVQDOTQ-NEXT: [[TMP8:%.*]] = sext <8 x i8> [[WIDE_LOAD3]] to <8 x i32>
-; FIXED-ZVQDOTQ-NEXT: [[TMP9:%.*]] = sext <8 x i8> [[WIDE_LOAD4]] to <8 x i32>
+; FIXED-ZVQDOTQ-NEXT: [[TMP3:%.*]] = sext <8 x i8> [[WIDE_LOAD]] to <8 x i32>
; FIXED-ZVQDOTQ-NEXT: [[TMP10:%.*]] = mul <8 x i32> [[TMP8]], [[TMP3]]
-; FIXED-ZVQDOTQ-NEXT: [[TMP11:%.*]] = mul <8 x i32> [[TMP9]], [[TMP4]]
; FIXED-ZVQDOTQ-NEXT: [[PARTIAL_REDUCE]] = call <2 x i32> @llvm.vector.partial.reduce.add.v2i32.v8i32(<2 x i32> [[VEC_PHI]], <8 x i32> [[TMP10]])
+; FIXED-ZVQDOTQ-NEXT: [[TMP9:%.*]] = sext <8 x i8> [[WIDE_LOAD4]] to <8 x i32>
+; FIXED-ZVQDOTQ-NEXT: [[TMP12:%.*]] = sext <8 x i8> [[WIDE_LOAD2]] to <8 x i32>
+; FIXED-ZVQDOTQ-NEXT: [[TMP11:%.*]] = mul <8 x i32> [[TMP9]], [[TMP12]]
; FIXED-ZVQDOTQ-NEXT: [[PARTIAL_REDUCE5]] = call <2 x i32> @llvm.vector.partial.reduce.add.v2i32.v8i32(<2 x i32> [[VEC_PHI1]], <8 x i32> [[TMP11]])
; FIXED-ZVQDOTQ-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; FIXED-ZVQDOTQ-NEXT: [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
@@ -222,10 +222,10 @@ define i32 @vqdotu(ptr %a, ptr %b) #0 {
; ZVQDOTQ-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 1 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
; ZVQDOTQ-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; ZVQDOTQ-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i8>, ptr [[TMP6]], align 1
-; ZVQDOTQ-NEXT: [[TMP8:%.*]] = zext <vscale x 4 x i8> [[WIDE_LOAD]] to <vscale x 4 x i32>
; ZVQDOTQ-NEXT: [[TMP9:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; ZVQDOTQ-NEXT: [[WIDE_LOAD1:%.*]] = load <vscale x 4 x i8>, ptr [[TMP9]], align 1
; ZVQDOTQ-NEXT: [[TMP11:%.*]] = zext <vscale x 4 x i8> [[WIDE_LOAD1]] to <vscale x 4 x i32>
+; ZVQDOTQ-NEXT: [[TMP8:%.*]] = zext <vscale x 4 x i8> [[WIDE_LOAD]] to <vscale x 4 x i32>
; ZVQDOTQ-NEXT: [[TMP12:%.*]] = mul <vscale x 4 x i32> [[TMP11]], [[TMP8]]
; ZVQDOTQ-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 1 x i32> @llvm.vector.partial.reduce.add.nxv1i32.nxv4i32(<vscale x 1 x i32> [[VEC_PHI]], <vscale x 4 x i32> [[TMP12]])
; ZVQDOTQ-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
@@ -287,17 +287,17 @@ define i32 @vqdotu(ptr %a, ptr %b) #0 {
; FIXED-ZVQDOTQ-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP0]], i32 8
; FIXED-ZVQDOTQ-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i8>, ptr [[TMP0]], align 1
; FIXED-ZVQDOTQ-NEXT: [[WIDE_LOAD2:%.*]] = load <8 x i8>, ptr [[TMP2]], align 1
-; FIXED-ZVQDOTQ-NEXT: [[TMP3:%.*]] = zext <8 x i8> [[WIDE_LOAD]] to <8 x i32>
-; FIXED-ZVQDOTQ-NEXT: [[TMP4:%.*]] = zext <8 x i8> [[WIDE_LOAD2]] to <8 x i32>
; FIXED-ZVQDOTQ-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; FIXED-ZVQDOTQ-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP5]], i32 8
; FIXED-ZVQDOTQ-NEXT: [[WIDE_LOAD3:%.*]] = load <8 x i8>, ptr [[TMP5]], align 1
; FIXED-ZVQDOTQ-NEXT: [[WIDE_LOAD4:%.*]] = load <8 x i8>, ptr [[TMP7]], align 1
; FIXED-ZVQDOTQ-NEXT: [[TMP8:%.*]] = zext <8 x i8> [[WIDE_LOAD3]] to <8 x i32>
-; FIXED-ZVQDOTQ-NEXT: [[TMP9:%.*]] = zext <8 x i8> [[WIDE_LOAD4]] to <8 x i32>
+; FIXED-ZVQDOTQ-NEXT: [[TMP3:%.*]] = zext <8 x i8> [[WIDE_LOAD]] to <8 x i32>
; FIXED-ZVQDOTQ-NEXT: [[TMP10:%.*]] = mul <8 x i32> [[TMP8]], [[TMP3]]
-; FIXED-ZVQDOTQ-NEXT: [[TMP11:%.*]] = mul <8 x i32> [[TMP9]], [[TMP4]]
; FIXED-ZVQDOTQ-NEXT: [[PARTIAL_REDUCE]] = call <2 x i32> @llvm.vector.partial.reduce.add.v2i32.v8i32(<2 x i32> [[VEC_PHI]], <8 x i32> [[TMP10]])
+; FIXED-ZVQDOTQ-NEXT: [[TMP9:%.*]] = zext <8 x i8> [[WIDE_LOAD4]] to <8 x i32>
+; FIXED-ZVQDOTQ-NEXT: [[TMP12:%.*]] = zext <8 x i8> [[WIDE_LOAD2]] to <8 x i32>
+; FIXED-ZVQDOTQ-NEXT: [[TMP11:%.*]] = mul <8 x i32> [[TMP9]], [[TMP12]]
; FIXED-ZVQDOTQ-NEXT: [[PARTIAL_REDUCE5]] = call <2 x i32> @llvm.vector.partial.reduce.add.v2i32.v8i32(<2 x i32> [[VEC_PHI1]], <8 x i32> [[TMP11]])
; FIXED-ZVQDOTQ-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; FIXED-ZVQDOTQ-NEXT: [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
@@ -384,10 +384,10 @@ define i32 @vqdotsu(ptr %a, ptr %b) #0 {
; ZVQDOTQ-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 1 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
; ZVQDOTQ-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; ZVQDOTQ-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i8>, ptr [[TMP6]], align 1
-; ZVQDOTQ-NEXT: [[TMP8:%.*]] = zext <vscale x 4 x i8> [[WIDE_LOAD]] to <vscale x 4 x i32>
; ZVQDOTQ-NEXT: [[TMP9:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; ZVQDOTQ-NEXT: [[WIDE_LOAD1:%.*]] = load <vscale x 4 x i8>, ptr [[TMP9]], align 1
; ZVQDOTQ-NEXT: [[TMP11:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD1]] to <vscale x 4 x i32>
+; ZVQDOTQ-NEXT: [[TMP8:%.*]] = zext <vscale x 4 x i8> [[WIDE_LOAD]] to <vscale x 4 x i32>
; ZVQDOTQ-NEXT: [[TMP12:%.*]] = mul <vscale x 4 x i32> [[TMP11]], [[TMP8]]
; ZVQDOTQ-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 1 x i32> @llvm.vector.partial.reduce.add.nxv1i32.nxv4i32(<vscale x 1 x i32> [[VEC_PHI]], <vscale x 4 x i32> [[TMP12]])
; ZVQDOTQ-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
@@ -449,18 +449,18 @@ define i32 @vqdotsu(ptr %a, ptr %b) #0 {
; FIXED-ZVQDOTQ-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP0]], i32 8
; FIXED-ZVQDOTQ-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i8>, ptr [[TMP0]], align 1
; FIXED-ZVQDOTQ-NEXT: [[WIDE_LOAD2:%.*]] = load <8 x i8>, ptr [[TMP2]], align 1
-; FIXED-ZVQDOTQ-NEXT: [[TMP3:%.*]] = zext <8 x i8> [[WIDE_LOAD]] to <8 x i32>
-; FIXED-ZVQDOTQ-NEXT: [[TMP4:%.*]] = zext <8 x i8> [[WIDE_LOAD2]] to <8 x i32>
; FIXED-ZVQDOTQ-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; FIXED-ZVQDOTQ-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP5]], i32 8
; FIXED-ZVQDOTQ-NEXT: [[WIDE_LOAD3:%.*]] = load <8 x i8>, ptr [[TMP5]], align 1
; FIXED-ZVQDOTQ-NEXT: [[WIDE_LOAD4:%.*]] = load <8 x i8>, ptr [[TMP7]], align 1
-; FIXED-ZVQDOTQ-NEXT: [[TMP8:%.*]] = sext <8 x i8> [[WIDE_LOAD3]] to <8 x i32>
-; FIXED-ZVQDOTQ-NEXT: [[TMP9:%.*]] = sext <8 x i8> [[WIDE_LOAD4]] to <8 x i32>
-; FIXED-ZVQDOTQ-NEXT: [[TMP10:%.*]] = mul <8 x i32> [[TMP8]], [[TMP3]]
+; FIXED-ZVQDOTQ-NEXT: [[TMP9:%.*]] = sext <8 x i8> [[WIDE_LOAD3]] to <8 x i32>
+; FIXED-ZVQDOTQ-NEXT: [[TMP4:%.*]] = zext <8 x i8> [[WIDE_LOAD]] to <8 x i32>
; FIXED-ZVQDOTQ-NEXT: [[TMP11:%.*]] = mul <8 x i32> [[TMP9]], [[TMP4]]
-; FIXED-ZVQDOTQ-NEXT: [[PARTIAL_REDUCE]] = call <2 x i32> @llvm.vector.partial.reduce.add.v2i32.v8i32(<2 x i32> [[VEC_PHI]], <8 x i32> [[TMP10]])
-; FIXED-ZVQDOTQ-NEXT: [[PARTIAL_REDUCE5]] = call <2 x i32> @llvm.vector.partial.reduce.add.v2i32.v8i32(<2 x i32> [[VEC_PHI1]], <8 x i32> [[TMP11]])
+; FIXED-ZVQDOTQ-NEXT: [[PARTIAL_REDUCE]] = call <2 x i32> @llvm.vector.partial.reduce.add.v2i32.v8i32(<2 x i32> [[VEC_PHI]], <8 x i32> [[TMP11]])
+; FIXED-ZVQDOTQ-NEXT: [[TMP10:%.*]] = sext <8 x i8> [[WIDE_LOAD4]] to <8 x i32>
+; FIXED-ZVQDOTQ-NEXT: [[TMP8:%.*]] = zext <8 x i8> [[WIDE_LOAD2]] to <8 x i32>
+; FIXED-ZVQDOTQ-NEXT: [[TMP14:%.*]] = mul <8 x i32> [[TMP10]], [[TMP8]]
+; FIXED-ZVQDOTQ-NEXT: [[PARTIAL_REDUCE5]] = call <2 x i32> @llvm.vector.partial.reduce.add.v2i32.v8i32(<2 x i32> [[VEC_PHI1]], <8 x i32> [[TMP14]])
; FIXED-ZVQDOTQ-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; FIXED-ZVQDOTQ-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
; FIXED-ZVQDOTQ-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
@@ -545,10 +545,10 @@ define i32 @vqdotsu2(ptr %a, ptr %b) #0 {
; ZVQDOTQ-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 1 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ]
; ZVQDOTQ-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]]
; ZVQDOTQ-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i8>, ptr [[TMP6]], align 1
-; ZVQDOTQ-NEXT: [[TMP8:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD]] to <vscale x 4 x i32>
; ZVQDOTQ-NEXT: [[TMP9:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; ZVQDOTQ-NEXT: [[WIDE_LOAD1:%.*]] = load <vscale x 4 x i8>, ptr [[TMP9]], align 1
; ZVQDOTQ-NEXT: [[TMP11:%.*]] = zext <vscale x 4 x i8> [[WIDE_LOAD1]] to <vscale x 4 x i32>
+; ZVQDOTQ-NEXT: [[TMP8:%.*]] = sext <vscale x 4 x i8> [[WIDE_LOAD]] to <vscale x 4 x i32>
; ZVQDOTQ-NEXT: [[TMP12:%.*]] = mul <vscale x 4 x i32> [[TMP11]], [[TMP8]]
; ZVQDOTQ-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 1 x i32> @llvm.vector.partial.reduce.add.nxv1i32.nxv4i32(<vscale x 1 x i32> [[VEC_PHI]], <vscale x 4 x i32> [[TMP12]])
; ZVQDOTQ-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
@@ -610,18 +610,18 @@ define i32 @vqdotsu2(ptr %a, ptr %b) #0 {
; FIXED-ZVQDOTQ-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP0]], i32 8
; FIXED-ZVQDOTQ-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i8>, ptr [[TMP0]], align 1
; FIXED-ZVQDOTQ-NEXT: [[WIDE_LOAD2:%.*]] = load <8 x i8>, ptr [[TMP2]], align 1
-; FIXED-ZVQDOTQ-NEXT: [[TMP3:%.*]] = sext <8 x i8> [[WIDE_LOAD]] to <8 x i32>
-; FIXED-ZVQDOTQ-NEXT: [[TMP4:%.*]] = sext <8 x i8> [[WIDE_LOAD2]] to <8 x i32>
; FIXED-ZVQDOTQ-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]]
; FIXED-ZVQDOTQ-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP5]], i32 8
; FIXED-ZVQDOTQ-NEXT: [[WIDE_LOAD3:%.*]] = load <8 x i8>, ptr [[TMP5]], align 1
; FIXED-ZVQDOTQ-NEXT: [[WIDE_LOAD4:%.*]] = load <8 x i8>, ptr [[TMP7]], align 1
-; FIXED-ZVQDOTQ-NEXT: [[TMP8:%.*]] = zext <8 x i8> [[WIDE_LOAD3]] to <8 x i32>
-; FIXED-ZVQDOTQ-NEXT: [[TMP9:%.*]] = zext <8 x i8> [[WIDE_LOAD4]] to <8 x i32>
-; FIXED-ZVQDOTQ-NEXT: [[TMP10:%.*]] = mul <8 x i32> [[TMP8]], [[TMP3]]
+; FIXED-ZVQDOTQ-NEXT: [[TMP9:%.*]] = zext <8 x i8> [[WIDE_LOAD3]] to <8 x i32>
+; FIXED-ZVQDOTQ-NEXT: [[TMP4:%.*]] = sext <8 x i8> [[WIDE_LOAD]] to <8 x i32>
; FIXED-ZVQDOTQ-NEXT: [[TMP11:%.*]] = mul <8 x i32> [[TMP9]], [[TMP4]]
-; FIXED-ZVQDOTQ-NEXT: [[PARTIAL_REDUCE]] = call <2 x i32> @llvm.vector.partial.reduce.add.v2i32.v8i32(<2 x i32> [[VEC_PHI]], <8 x i32> [[TMP10]])
-; FIXED-ZVQDOTQ-NEXT: [[PARTIAL_REDUCE5]] = call <2 x i32> @llvm.vector.partial.reduce.add.v2i32.v8i32(<2 x i32> [[VEC_PHI1]], <8 x i32> [[TMP11]])
+; FIXED-ZVQDOTQ-NEXT: [[PARTIAL_REDUCE]] = call <2 x i32> @llvm.vector.partial.reduce.add.v2i32.v8i32(<2 x i32> [[VEC_PHI]], <8 x i32> [[TMP11]])
+; FIXED-ZVQDOTQ-NEXT: [[TMP10:%.*]] = zext <8 x i8> [[WIDE_LOAD4]] to <8 x i32>
+; FIXED-ZVQDOTQ-NEXT: [[TMP8:%.*]] = sext <8 x i8> [[WIDE_LOAD2]] to <8 x i32>
+; FIXED-ZVQDOTQ-NEXT: [[TMP14:%.*]] = mul <8 x i32> [[TMP10]], [[TMP8]]
+; FIXED-ZVQDOTQ-NEXT: [[PARTIAL_REDUCE5]] = call <2 x i32> @llvm.vector.partial.reduce.add.v2i32.v8i32(<2 x i32> [[VEC_PHI1]], <8 x i32> [[TMP14]])
; FIXED-ZVQDOTQ-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; FIXED-ZVQDOTQ-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
; FIXED-ZVQDOTQ-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
diff --git a/llvm/test/Transforms/LoopVectorize/WebAssembly/partial-reduce-accumulate.ll b/llvm/test/Transforms/LoopVectorize/WebAssembly/partial-reduce-accumulate.ll
new file mode 100644
index 0000000000000..2338da5339354
--- /dev/null
+++ b/llvm/test/Transforms/LoopVectorize/WebAssembly/partial-reduce-accumulate.ll
@@ -0,0 +1,126 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
+; RUN: opt -mattr=+simd128 -passes=loop-vectorize %s -S | FileCheck %s
+; RUN: opt -mattr=+simd128 -passes=loop-vectorize -vectorizer-maximize-bandwidth %s -S | FileCheck %s --check-prefix=CHECK-MAX-BANDWIDTH
+
+target triple = "wasm32"
+
+define hidden i32 @accumulate_add_u8_u8(ptr noundef readonly %a, ptr noundef readonly %b, i32 noundef %N) {
+; CHECK-LABEL: define hidden i32 @accumulate_add_u8_u8(
+; CHECK-SAME: ptr noundef readonly [[A:%.*]], ptr noundef readonly [[B:%.*]], i32 noundef [[N:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT: [[ENTRY:.*]]:
+; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[N]], 4
+; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; CHECK: [[VECTOR_PH]]:
+; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[N]], 4
+; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[N]], [[N_MOD_VF]]
+; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
+; CHECK: [[VECTOR_BODY]]:
+; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP8:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds nuw i8, ptr [[A]], i32 [[INDEX]]
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i8>, ptr [[TMP2]], align 1
+; CHECK-NEXT: [[TMP3:%.*]] = zext <4 x i8> [[WIDE_LOAD]] to <4 x i32>
+; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds nuw i8, ptr [[B]], i32 [[INDEX]]
+; CHECK-NEXT: [[WIDE_LOAD1:%.*]] = load <4 x i8>, ptr [[TMP5]], align 1
+; CHECK-NEXT: [[TMP6:%.*]] = zext <4 x i8> [[WIDE_LOAD1]] to <4 x i32>
+; CHECK-NEXT: [[TMP7:%.*]] = add <4 x i32> [[VEC_PHI]], [[TMP3]]
+; CHECK-NEXT: [[TMP8]] = add <4 x i32> [[TMP7]], [[TMP6]]
+; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
+; CHECK-NEXT: [[TMP9:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK: [[MIDDLE_BLOCK]]:
+; CHECK-NEXT: [[TMP10:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP8]])
+; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[N]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[CMP_N]], label %[[FOR_COND_CLEANUP:.*]], label %[[SCALAR_PH]]
+; CHECK: [[SCALAR_PH]]:
+; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
+; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP10]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_COND_CLEANUP]]:
+; CHECK-NEXT: [[RESULT_0_LCSSA:%.*]] = phi i32 [ [[ADD3:%.*]], %[[FOR_BODY]] ], [ [[TMP10]], %[[MIDDLE_BLOCK]] ]
+; CHECK-NEXT: ret i32 [[RESULT_0_LCSSA]]
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ]
+; CHECK-NEXT: [[RED:%.*]] = phi i32 [ [[ADD3]], %[[FOR_BODY]] ], [ [[BC_MERGE_RDX]], %[[SCALAR_PH]] ]
+; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds nuw i8, ptr [[A]], i32 [[IV]]
+; CHECK-NEXT: [[TMP11:%.*]] = load i8, ptr [[ARRAYIDX]], align 1
+; CHECK-NEXT: [[CONV:%.*]] = zext i8 [[TMP11]] to i32
+; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds nuw i8, ptr [[B]], i32 [[IV]]
+; CHECK-NEXT: [[TMP12:%.*]] = load i8, ptr [[ARRAYIDX1]], align 1
+; CHECK-NEXT: [[CONV2:%.*]] = zext i8 [[TMP12]] to i32
+; CHECK-NEXT: [[ADD:%.*]] = add i32 [[RED]], [[CONV]]
+; CHECK-NEXT: [[ADD3]] = add i32 [[ADD]], [[CONV2]]
+; CHECK-NEXT: [[INC]] = add nuw i32 [[IV]], 1
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
+;
+; CHECK-MAX-BANDWIDTH-LABEL: define hidden i32 @accumulate_add_u8_u8(
+; CHECK-MAX-BANDWIDTH-SAME: ptr noundef readonly [[A:%.*]], ptr noundef readonly [[B:%.*]], i32 noundef [[N:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-MAX-BANDWIDTH-NEXT: [[ENTRY:.*]]:
+; CHECK-MAX-BANDWIDTH-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[N]], 16
+; CHECK-MAX-BANDWIDTH-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; CHECK-MAX-BANDWIDTH: [[VECTOR_PH]]:
+; CHECK-MAX-BANDWIDTH-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[N]], 16
+; CHECK-MAX-BANDWIDTH-NEXT: [[N_VEC:%.*]] = sub i32 [[N]], [[N_MOD_VF]]
+; CHECK-MAX-BANDWIDTH-NEXT: br label %[[VECTOR_BODY:.*]]
+; CHECK-MAX-BANDWIDTH: [[VECTOR_BODY]]:
+; CHECK-MAX-BANDWIDTH-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-MAX-BANDWIDTH-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[PARTIAL_REDUCE2:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-MAX-BANDWIDTH-NEXT: [[TMP2:%.*]] = getelementptr inbounds nuw i8, ptr [[A]], i32 [[INDEX]]
+; CHECK-MAX-BANDWIDTH-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1
+; CHECK-MAX-BANDWIDTH-NEXT: [[TMP5:%.*]] = getelementptr inbounds nuw i8, ptr [[B]], i32 [[INDEX]]
+; CHECK-MAX-BANDWIDTH-NEXT: [[WIDE_LOAD1:%.*]] = load <16 x i8>, ptr [[TMP5]], align 1
+; CHECK-MAX-BANDWIDTH-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32>
+; CHECK-MAX-BANDWIDTH-NEXT: [[PARTIAL_REDUCE:%.*]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP3]])
+; CHECK-MAX-BANDWIDTH-NEXT: [[TMP6:%.*]] = zext <16 x i8> [[WIDE_LOAD1]] to <16 x i32>
+; CHECK-MAX-BANDWIDTH-NEXT: [[PARTIAL_REDUCE2]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[PARTIAL_REDUCE]], <16 x i32> [[TMP6]])
+; CHECK-MAX-BANDWIDTH-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 16
+; CHECK-MAX-BANDWIDTH-NEXT: [[TMP9:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-MAX-BANDWIDTH-NEXT: br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK-MAX-BANDWIDTH: [[MIDDLE_BLOCK]]:
+; CHECK-MAX-BANDWIDTH-NEXT: [[TMP7:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE2]])
+; CHECK-MAX-BANDWIDTH-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[N]], [[N_VEC]]
+; CHECK-MAX-BANDWIDTH-NEXT: br i1 [[CMP_N]], label %[[FOR_COND_CLEANUP:.*]], label %[[SCALAR_PH]]
+; CHECK-MAX-BANDWIDTH: [[SCALAR_PH]]:
+; CHECK-MAX-BANDWIDTH-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
+; CHECK-MAX-BANDWIDTH-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP7]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
+; CHECK-MAX-BANDWIDTH-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK-MAX-BANDWIDTH: [[FOR_COND_CLEANUP]]:
+; CHECK-MAX-BANDWIDTH-NEXT: [[RESULT_0_LCSSA:%.*]] = phi i32 [ [[ADD3:%.*]], %[[FOR_BODY]] ], [ [[TMP7]], %[[MIDDLE_BLOCK]] ]
+; CHECK-MAX-BANDWIDTH-NEXT: ret i32 [[RESULT_0_LCSSA]]
+; CHECK-MAX-BANDWIDTH: [[FOR_BODY]]:
+; CHECK-MAX-BANDWIDTH-NEXT: [[IV:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ]
+; CHECK-MAX-BANDWIDTH-NEXT: [[RED:%.*]] = phi i32 [ [[ADD3]], %[[FOR_BODY]] ], [ [[BC_MERGE_RDX]], %[[SCALAR_PH]] ]
+; CHECK-MAX-BANDWIDTH-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds nuw i8, ptr [[A]], i32 [[IV]]
+; CHECK-MAX-BANDWIDTH-NEXT: [[TMP11:%.*]] = load i8, ptr [[ARRAYIDX]], align 1
+; CHECK-MAX-BANDWIDTH-NEXT: [[CONV:%.*]] = zext i8 [[TMP11]] to i32
+; CHECK-MAX-BANDWIDTH-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds nuw i8, ptr [[B]], i32 [[IV]]
+; CHECK-MAX-BANDWIDTH-NEXT: [[TMP12:%.*]] = load i8, ptr [[ARRAYIDX1]], align 1
+; CHECK-MAX-BANDWIDTH-NEXT: [[CONV2:%.*]] = zext i8 [[TMP12]] to i32
+; CHECK-MAX-BANDWIDTH-NEXT: [[ADD:%.*]] = add i32 [[RED]], [[CONV]]
+; CHECK-MAX-BANDWIDTH-NEXT: [[ADD3]] = add i32 [[ADD]], [[CONV2]]
+; CHECK-MAX-BANDWIDTH-NEXT: [[INC]] = add nuw i32 [[IV]], 1
+; CHECK-MAX-BANDWIDTH-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
+; CHECK-MAX-BANDWIDTH-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
+;
+entry:
+ br label %for.body
+
+for.cond.cleanup: ; preds = %for.body
+ ret i32 %add3
+
+for.body: ; preds = %entry, %for.body
+ %iv = phi i32 [ %inc, %for.body ], [ 0, %entry ]
+ %red = phi i32 [ %add3, %for.body ], [ 0, %entry ]
+ %arrayidx = getelementptr inbounds nuw i8, ptr %a, i32 %iv
+ %0 = load i8, ptr %arrayidx, align 1
+ %conv = zext i8 %0 to i32
+ %arrayidx1 = getelementptr inbounds nuw i8, ptr %b, i32 %iv
+ %1 = load i8, ptr %arrayidx1, align 1
+ %conv2 = zext i8 %1 to i32
+ %add = add i32 %red, %conv
+ %add3 = add i32 %add, %conv2
+ %inc = add nuw i32 %iv, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
More information about the llvm-commits
mailing list