[ARM][MVE] Enable masked scatter

[anwel]

Summary

This patch extends the gather/scatter pass in MVEGatherScatterLowering.cpp to enable the transformation of masked scatters into calls to arm's masked scatter intrinsic.

[dmgreen]

llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp

542 ↗

(On Diff #238523)

This can just call isLegalMaskedGather, like we do for isLegalMaskedStore

llvm/lib/Target/ARM/MVEGatherScatterLowering.cpp

292

This looks like it's gone back to an older version?

459–467

This might be better as

bool Changed = false;
for (IntrinsicInst *I : Gathers)
  Changed |= lowerGather(I);
for (IntrinsicInst *I : Scatters)
  Changed |= lowerScatter(I);

As we might not be altering the gathers/scatters that we see, leaving them for the legalise pass to scalarise.

[anwel]

llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp

542 ↗

(On Diff #238523)

Fair point.

llvm/lib/Target/ARM/MVEGatherScatterLowering.cpp

292

I think you are right. Thanks for catching that.

459–467

Good point. We do get the booleans from the functions' return value, so we should be using it.

[dmgreen]

LGTM. Thanks.

llvm/lib/Target/ARM/MVEGatherScatterLowering.cpp

459–460

You probably don't need these two lines any more.

[anwel]

llvm/lib/Target/ARM/MVEGatherScatterLowering.cpp

459–460

No, I don't. Thanks for spotting.

llvm/lib/Target/ARM/ARMTargetTransformInfo.h

public:
   bool isLegalMaskedLoad(Type *DataTy, MaybeAlign Alignment);
 
   bool isLegalMaskedStore(Type *DataTy, MaybeAlign Alignment) {
     return isLegalMaskedLoad(DataTy, Alignment);
   }
 
   bool isLegalMaskedGather(Type *Ty, MaybeAlign Alignment);
 
-  bool isLegalMaskedScatter(Type *Ty, MaybeAlign Alignment) { return false; }
+  bool isLegalMaskedScatter(Type *Ty, MaybeAlign Alignment) {
+    return isLegalMaskedGather(Ty, Alignment);
+  }
 
   int getMemcpyCost(const Instruction *I);
 
   int getShuffleCost(TTI::ShuffleKind Kind, Type *Tp, int Index, Type *SubTp);
 
   bool useReductionIntrinsic(unsigned Opcode, Type *Ty,
                              TTI::ReductionFlags Flags) const;
 

llvm/lib/Target/ARM/MVEGatherScatterLowering.cpp

private:
   bool isLegalTypeAndAlignment(unsigned NumElements, unsigned ElemSize,
                                unsigned Alignment);
   // Check whether Ptr is hidden behind a bitcast and look through it
   void lookThroughBitcast(Value *&Ptr);
   // Check for a getelementptr and deduce base and offsets from it, on success
   // returning the base directly and the offsets indirectly using the Offsets
   // argument
   Value *checkGEP(Value *&Offsets, Type *Ty, Value *Ptr, IRBuilder<> Builder);
+  // Compute the scale of this gather/scatter instruction
+  int computeScale(unsigned GEPElemSize, unsigned MemoryElemSize);
 
   bool lowerGather(IntrinsicInst *I);
   // Create a gather from a base + vector of offsets
   Value *tryCreateMaskedGatherOffset(IntrinsicInst *I, Value *Ptr,
                                      Instruction *&Root, IRBuilder<> Builder);
   // Create a gather from a vector of pointers
   Value *tryCreateMaskedGatherBase(IntrinsicInst *I, Value *Ptr,
                                    IRBuilder<> Builder);
+
+  bool lowerScatter(IntrinsicInst *I);
+  // Create a scatter to a base + vector of offsets
+  Value *tryCreateMaskedScatterOffset(IntrinsicInst *I, Value *Ptr,
+                                      IRBuilder<> Builder);
+  // Create a scatter to a vector of pointers
+  Value *tryCreateMaskedScatterBase(IntrinsicInst *I, Value *Ptr,
+                                    IRBuilder<> Builder);
 };
 
 } // end anonymous namespace
 
 char MVEGatherScatterLowering::ID = 0;
 
 INITIALIZE_PASS(MVEGatherScatterLowering, DEBUG_TYPE,
                 "MVE gather/scattering lowering pass", false, false)
 
 Pass *llvm::createMVEGatherScatterLoweringPass() {
   return new MVEGatherScatterLowering();
 }
 
 bool MVEGatherScatterLowering::isLegalTypeAndAlignment(unsigned NumElements,
                                                        unsigned ElemSize,
                                                        unsigned Alignment) {
   if (((NumElements == 4 &&
         (ElemSize == 32 || ElemSize == 16 || ElemSize == 8)) ||
        (NumElements == 8 && (ElemSize == 16 || ElemSize == 8)) ||
        (NumElements == 16 && ElemSize == 8)) &&
       ElemSize / 8 <= Alignment)
     return true;
-  LLVM_DEBUG(dbgs() << "masked gathers: instruction does not have valid "
-                    << "alignment or vector type \n");
+  LLVM_DEBUG(dbgs() << "masked gathers/scatters: instruction does not have "
+                    << "valid alignment or vector type \n");
   return false;
 }
 
 Value *MVEGatherScatterLowering::checkGEP(Value *&Offsets, Type *Ty, Value *Ptr,
                                           IRBuilder<> Builder) {
   GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(Ptr);
   if (!GEP) {
-    LLVM_DEBUG(dbgs() << "masked gathers: no getelementpointer found\n");
+    LLVM_DEBUG(
+        dbgs() << "masked gathers/scatters: no getelementpointer found\n");
     return nullptr;
   }
-  LLVM_DEBUG(dbgs() << "masked gathers: getelementpointer found. Loading"
-                    << " from base + vector of offsets\n");
+  LLVM_DEBUG(dbgs() << "masked gathers/scatters: getelementpointer found."
+                    << " Looking at intrinsic for base + vector of offsets\n");
   Value *GEPPtr = GEP->getPointerOperand();
   if (GEPPtr->getType()->isVectorTy()) {
     return nullptr;
   }
   if (GEP->getNumOperands() != 2) {
-    LLVM_DEBUG(dbgs() << "masked gathers: getelementptr with too many"
+    LLVM_DEBUG(dbgs() << "masked gathers/scatters: getelementptr with too many"
                       << " operands. Expanding.\n");
     return nullptr;
   }
   Offsets = GEP->getOperand(1);
   // SExt offsets inside masked gathers are not permitted by the architecture;
   // we therefore can't fold them
   if (ZExtInst *ZextOffs = dyn_cast<ZExtInst>(Offsets))
     Offsets = ZextOffs->getOperand(0);
   Type *OffsType = VectorType::getInteger(cast<VectorType>(Ty));
   // If the offset we found does not have the type the intrinsic expects,
-  // i.e., the same type as the gather itself, we need to convert it (only i
-  // types) or fall back to expanding the gather
+  // i.e., the same type as the gather (or scatter input) itself, we need to
+  // convert it (only i types) or fall back to expanding the gather
   if (OffsType != Offsets->getType()) {
     if (OffsType->getScalarSizeInBits() >
         Offsets->getType()->getScalarSizeInBits()) {
-      LLVM_DEBUG(dbgs() << "masked gathers: extending offsets\n");
+      LLVM_DEBUG(dbgs() << "masked gathers/scatters: extending offsets\n");
       Offsets = Builder.CreateZExt(Offsets, OffsType, "");
     } else {
-      LLVM_DEBUG(dbgs() << "masked gathers: no correct offset type. Can't"
-                        << " create masked gather\n");
+      LLVM_DEBUG(dbgs() << "masked gathers/scatters: no correct offset type."
+                        << " Can't create intrinsic.\n");
       return nullptr;
     }
   }
   // If none of the checks failed, return the gep's base pointer
   return GEPPtr;
 }
 
 void MVEGatherScatterLowering::lookThroughBitcast(Value *&Ptr) {
   // Look through bitcast instruction if #elements is the same
   if (auto *BitCast = dyn_cast<BitCastInst>(Ptr)) {
     Type *BCTy = BitCast->getType();
     Type *BCSrcTy = BitCast->getOperand(0)->getType();
     if (BCTy->getVectorNumElements() == BCSrcTy->getVectorNumElements()) {
-      LLVM_DEBUG(dbgs() << "masked gathers: looking through bitcast\n");
+      LLVM_DEBUG(
+          dbgs() << "masked gathers/scatters: looking through bitcast\n");
       Ptr = BitCast->getOperand(0);
     }
   }
 }
 
+int MVEGatherScatterLowering::computeScale(unsigned GEPElemSize,
+                                           unsigned MemoryElemSize) {
+  // This can be a 32bit load/store scaled by 4, a 16bit load/store scaled by 2,
+  // or a 8bit, 16bit or 32bit load/store scaled by 1
+  if (GEPElemSize == 32 && MemoryElemSize == 32)
+    return 2;
+  else if (GEPElemSize == 16 && MemoryElemSize == 16)
+    return 1;
+  else if (GEPElemSize == 8)
+    return 0;
+  LLVM_DEBUG(dbgs() << "masked gathers/scatters: incorrect scale. Can't "
+                    << "create intrinsic\n");
+  return -1;
+}
+
 bool MVEGatherScatterLowering::lowerGather(IntrinsicInst *I) {
   using namespace PatternMatch;
   LLVM_DEBUG(dbgs() << "masked gathers: checking transform preconditions\n");
 
   // @llvm.masked.gather.*(Ptrs, alignment, Mask, Src0)
   // Attempt to turn the masked gather in I into a MVE intrinsic
   // Potentially optimising the addressing modes as we do so.
   Type *Ty = I->getType();
   Value *Ptr = I->getArgOperand(0);
   unsigned Alignment = cast<ConstantInt>(I->getArgOperand(1))->getZExtValue();
   Value *Mask = I->getArgOperand(2);
   Value *PassThru = I->getArgOperand(3);
 
   if (!isLegalTypeAndAlignment(Ty->getVectorNumElements(),
                                Ty->getScalarSizeInBits(), Alignment))
     return false;
   lookThroughBitcast(Ptr);
   assert(Ptr->getType()->isVectorTy() && "Unexpected pointer type");
 
   IRBuilder<> Builder(I->getContext());
   Builder.SetInsertPoint(I);
   Builder.SetCurrentDebugLocation(I->getDebugLoc());
 
   Instruction *Root = I;
-
   Value *Load = tryCreateMaskedGatherOffset(I, Ptr, Root, Builder);
   if (!Load)
     Load = tryCreateMaskedGatherBase(I, Ptr, Builder);
   if (!Load)
     return false;
 
   if (!isa<UndefValue>(PassThru) && !match(PassThru, m_Zero())) {
     LLVM_DEBUG(dbgs() << "masked gathers: found non-trivial passthru - "
if (Root != I)
     I->eraseFromParent();
   LLVM_DEBUG(dbgs() << "masked gathers: successfully built masked gather\n");
   return true;
 }
 
 Value *MVEGatherScatterLowering::tryCreateMaskedGatherBase(
     IntrinsicInst *I, Value *Ptr, IRBuilder<> Builder) {
   using namespace PatternMatch;
-
   Type *Ty = I->getType();
-
   LLVM_DEBUG(dbgs() << "masked gathers: loading from vector of pointers\n");
   if (Ty->getVectorNumElements() != 4 || Ty->getScalarSizeInBits() != 32)
     // Can't build an intrinsic for this
     return nullptr;
   Value *Mask = I->getArgOperand(2);
   if (match(Mask, m_One()))
     return Builder.CreateIntrinsic(Intrinsic::arm_mve_vldr_gather_base,
                                    {Ty, Ptr->getType()},
if (OriginalTy->getPrimitiveSizeInBits() < 128) {
     if (isa<SExtInst>(Extend)) {
       Unsigned = 0;
     } else if (!isa<ZExtInst>(Extend)) {
       LLVM_DEBUG(dbgs() << "masked gathers: extend needed but not provided. "
                         << "Expanding\n");
       return nullptr;
     }
     LLVM_DEBUG(dbgs() << "masked gathers: found an extending gather\n");
     ResultTy = Extend->getType();
     // The final size of the gather must be a full vector width
     if (ResultTy->getPrimitiveSizeInBits() != 128) {
       LLVM_DEBUG(dbgs() << "masked gathers: extending from the wrong type. "
                         << "Expanding\n");
       return nullptr;
     }
   }
 
   Value *Offsets;
   Value *BasePtr = checkGEP(Offsets, ResultTy, Ptr, Builder);
   if (!BasePtr)
     return nullptr;
 
-  unsigned Scale;
-  int GEPElemSize =
-      BasePtr->getType()->getPointerElementType()->getPrimitiveSizeInBits();
-  int MemoryElemSize = OriginalTy->getScalarSizeInBits();
-  // This can be a 32bit load scaled by 4, a 16bit load scaled by 2, or a
-  // 8bit, 16bit or 32bit load scaled by 1
-  if (GEPElemSize == 32 && MemoryElemSize == 32) {
-    Scale = 2;
-  } else if (GEPElemSize == 16 && MemoryElemSize == 16) {
-    Scale = 1;
-  } else if (GEPElemSize == 8) {
-    Scale = 0;
-  } else {
-    LLVM_DEBUG(dbgs() << "masked gathers: incorrect scale for load. Can't"
-                      << " create masked gather\n");
+  int Scale = computeScale(
+      BasePtr->getType()->getPointerElementType()->getPrimitiveSizeInBits(),
+      OriginalTy->getScalarSizeInBits());
+  if (Scale == -1)
     return nullptr;
-  }
   Root = Extend;
 
   Value *Mask = I->getArgOperand(2);
   if (!match(Mask, m_One()))
     return Builder.CreateIntrinsic(
         Intrinsic::arm_mve_vldr_gather_offset_predicated,
         {ResultTy, BasePtr->getType(), Offsets->getType(), Mask->getType()},
         {BasePtr, Offsets, Builder.getInt32(OriginalTy->getScalarSizeInBits()),
          Builder.getInt32(Scale), Builder.getInt32(Unsigned), Mask});
   else
     return Builder.CreateIntrinsic(
         Intrinsic::arm_mve_vldr_gather_offset,
         {ResultTy, BasePtr->getType(), Offsets->getType()},
         {BasePtr, Offsets, Builder.getInt32(OriginalTy->getScalarSizeInBits()),
          Builder.getInt32(Scale), Builder.getInt32(Unsigned)});
 }
 
+bool MVEGatherScatterLowering::lowerScatter(IntrinsicInst *I) {
+  using namespace PatternMatch;
+  LLVM_DEBUG(dbgs() << "masked scatters: checking transform preconditions\n");
+
+  // @llvm.masked.scatter.*(data, ptrs, alignment, mask)
+  // Attempt to turn the masked scatter in I into a MVE intrinsic
+  // Potentially optimising the addressing modes as we do so.
+  Value *Input = I->getArgOperand(0);
+  Value *Ptr = I->getArgOperand(1);
+  unsigned Alignment = cast<ConstantInt>(I->getArgOperand(2))->getZExtValue();
+  Type *Ty = Input->getType();
+
+  if (!isLegalTypeAndAlignment(Ty->getVectorNumElements(),
+                               Ty->getScalarSizeInBits(), Alignment))
+    return false;
+  lookThroughBitcast(Ptr);
+  assert(Ptr->getType()->isVectorTy() && "Unexpected pointer type");
+
+  IRBuilder<> Builder(I->getContext());
+  Builder.SetInsertPoint(I);
+  Builder.SetCurrentDebugLocation(I->getDebugLoc());
+
+  Value *Store = tryCreateMaskedScatterOffset(I, Ptr, Builder);
+  if (!Store)
+    Store = tryCreateMaskedScatterBase(I, Ptr, Builder);
+  if (!Store)
+    return false;
+
+  LLVM_DEBUG(dbgs() << "masked scatters: successfully built masked scatter\n");
+  I->replaceAllUsesWith(Store);
+  I->eraseFromParent();
+  return true;
+}
+
+Value *MVEGatherScatterLowering::tryCreateMaskedScatterBase(
+    IntrinsicInst *I, Value *Ptr, IRBuilder<> Builder) {
+  using namespace PatternMatch;
+  Value *Input = I->getArgOperand(0);
+  Value *Mask = I->getArgOperand(3);
+  Type *Ty = Input->getType();
+  // Only QR variants allow truncating
+  if (!(Ty->getVectorNumElements() == 4 && Ty->getScalarSizeInBits() == 32)) {
+    // Can't build an intrinsic for this
+    return nullptr;
+  }
+  //  int_arm_mve_vstr_scatter_base(_predicated) addr, offset, data(, mask)
+  LLVM_DEBUG(dbgs() << "masked scatters: storing to a vector of pointers\n");
+  if (match(Mask, m_One()))
+    return Builder.CreateIntrinsic(Intrinsic::arm_mve_vstr_scatter_base,
+                                   {Ptr->getType(), Input->getType()},
+                                   {Ptr, Builder.getInt32(0), Input});
+  else
+    return Builder.CreateIntrinsic(
+        Intrinsic::arm_mve_vstr_scatter_base_predicated,
+        {Ptr->getType(), Input->getType(), Mask->getType()},
+        {Ptr, Builder.getInt32(0), Input, Mask});
+}
+
+Value *MVEGatherScatterLowering::tryCreateMaskedScatterOffset(
+    IntrinsicInst *I, Value *Ptr, IRBuilder<> Builder) {
+  using namespace PatternMatch;
+  Value *Input = I->getArgOperand(0);
+  Value *Mask = I->getArgOperand(3);
+  Type *InputTy = Input->getType();
+  Type *MemoryTy = InputTy;
+  LLVM_DEBUG(dbgs() << "masked scatters: getelementpointer found. Storing"
+                    << " to base + vector of offsets\n");
+  // If the input has been truncated, try to integrate that trunc into the
+  // scatter instruction (we don't care about alignment here)
+  if (TruncInst *Trunc = dyn_cast<TruncInst>(Input)) {
+    Value *PreTrunc = Trunc->getOperand(0);
+    Type *PreTruncTy = PreTrunc->getType();
+    if (PreTruncTy->getPrimitiveSizeInBits() == 128) {
+      Input = PreTrunc;
+      InputTy = PreTruncTy;
+    }
+  }
+  if (InputTy->getPrimitiveSizeInBits() != 128) {
+    LLVM_DEBUG(
+        dbgs() << "masked scatters: cannot create scatters for non-standard"
+               << " input types. Expanding.\n");
+    return nullptr;
+  }
+
+  Value *Offsets;
+  Value *BasePtr = checkGEP(Offsets, InputTy, Ptr, Builder);
+  if (!BasePtr)
+    return nullptr;
+  int Scale = computeScale(
+      BasePtr->getType()->getPointerElementType()->getPrimitiveSizeInBits(),
+      MemoryTy->getScalarSizeInBits());
+  if (Scale == -1)
+    return nullptr;
+
+  if (!match(Mask, m_One()))
+    return Builder.CreateIntrinsic(
+        Intrinsic::arm_mve_vstr_scatter_offset_predicated,
+        {BasePtr->getType(), Offsets->getType(), Input->getType(),
+         Mask->getType()},
+        {BasePtr, Offsets, Input,
+         Builder.getInt32(MemoryTy->getScalarSizeInBits()),
+         Builder.getInt32(Scale), Mask});
+  else
+    return Builder.CreateIntrinsic(
+        Intrinsic::arm_mve_vstr_scatter_offset,
+        {BasePtr->getType(), Offsets->getType(), Input->getType()},
+        {BasePtr, Offsets, Input,
+         Builder.getInt32(MemoryTy->getScalarSizeInBits()),
+         Builder.getInt32(Scale)});
+}
+
 bool MVEGatherScatterLowering::runOnFunction(Function &F) {
   if (!EnableMaskedGatherScatters)
     return false;
   auto &TPC = getAnalysis<TargetPassConfig>();
   auto &TM = TPC.getTM<TargetMachine>();
   auto *ST = &TM.getSubtarget<ARMSubtarget>(F);
   if (!ST->hasMVEIntegerOps())
     return false;
   SmallVector<IntrinsicInst *, 4> Gathers;
+  SmallVector<IntrinsicInst *, 4> Scatters;
   for (BasicBlock &BB : F) {
     for (Instruction &I : BB) {
       IntrinsicInst *II = dyn_cast<IntrinsicInst>(&I);
       if (II && II->getIntrinsicID() == Intrinsic::masked_gather)
         Gathers.push_back(II);
+      else if (II && II->getIntrinsicID() == Intrinsic::masked_scatter)
+        Scatters.push_back(II);
     }
   }
 
-  if (Gathers.empty())
+  if (Gathers.empty() && Scatters.empty())
     return false;
 
+  bool Changed = false;
   for (IntrinsicInst *I : Gathers)
-    lowerGather(I);
+    Changed |= lowerGather(I);
+  for (IntrinsicInst *I : Scatters)
+    Changed |= lowerScatter(I);
 
-  return true;
+  return Changed;
 }

llvm/test/CodeGen/Thumb2/mve-scatter-ind16-scaled.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=thumbv8.1m.main-arm-none-eabi -mattr=+mve.fp -enable-arm-maskedgatscat %s -o - | FileCheck %s
 
 ; VLDRH.16 Qd, [base, offs, uxtw #1]
 define arm_aapcs_vfpcc void @scaled_v8i16_i16(i16* %base, <8 x i16>* %offptr, <8 x i16> %input) {
 ; CHECK-LABEL: scaled_v8i16_i16:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrh.u32 q2, [r1]
-; CHECK-NEXT:    vldrh.u32 q1, [r1, #8]
-; CHECK-NEXT:    vmov.u16 r1, q0[0]
-; CHECK-NEXT:    vshl.i32 q2, q2, #1
-; CHECK-NEXT:    vshl.i32 q1, q1, #1
-; CHECK-NEXT:    vadd.i32 q2, q2, r0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s9
-; CHECK-NEXT:    vmov.u16 r1, q0[1]
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s10
-; CHECK-NEXT:    vmov.u16 r1, q0[2]
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s11
-; CHECK-NEXT:    vmov.u16 r1, q0[3]
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov.u16 r1, q0[4]
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov.u16 r1, q0[5]
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov.u16 r1, q0[6]
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov.u16 r1, q0[7]
-; CHECK-NEXT:    strh r1, [r0]
+; CHECK-NEXT:    vldrh.u16 q1, [r1]
+; CHECK-NEXT:    vstrh.16 q0, [r0, q1, uxtw #1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i16>, <8 x i16>* %offptr, align 2
   %offs.zext = zext <8 x i16> %offs to <8 x i32>
   %ptrs = getelementptr inbounds i16, i16* %base, <8 x i32> %offs.zext
   call void @llvm.masked.scatter.v8i16.v8p0i16(<8 x i16> %input, <8 x i16*> %ptrs, i32 2, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VLDRH.16 Qd, [base, offs, uxtw #1]
 define arm_aapcs_vfpcc void @scaled_v8f16_i16(i16* %base, <8 x i16>* %offptr, <8 x half> %input) {
 ; CHECK-LABEL: scaled_v8f16_i16:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrh.u32 q1, [r1]
-; CHECK-NEXT:    vmovx.f16 s12, s0
-; CHECK-NEXT:    vshl.i32 q2, q1, #1
-; CHECK-NEXT:    vldrh.u32 q1, [r1, #8]
-; CHECK-NEXT:    vadd.i32 q2, q2, r0
-; CHECK-NEXT:    vmov r1, s8
-; CHECK-NEXT:    vshl.i32 q1, q1, #1
-; CHECK-NEXT:    vstr.16 s0, [r1]
-; CHECK-NEXT:    vmov r1, s9
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vstr.16 s12, [r1]
-; CHECK-NEXT:    vmov r1, s10
-; CHECK-NEXT:    vmovx.f16 s0, s3
-; CHECK-NEXT:    vstr.16 s1, [r1]
-; CHECK-NEXT:    vmov r1, s11
-; CHECK-NEXT:    vmovx.f16 s8, s1
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vstr.16 s8, [r1]
-; CHECK-NEXT:    vstr.16 s2, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmovx.f16 s8, s2
-; CHECK-NEXT:    vstr.16 s8, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vstr.16 s3, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vstr.16 s0, [r0]
+; CHECK-NEXT:    vldrh.u16 q1, [r1]
+; CHECK-NEXT:    vstrh.16 q0, [r0, q1, uxtw #1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i16>, <8 x i16>* %offptr, align 2
   %offs.zext = zext <8 x i16> %offs to <8 x i32>
   %i16_ptrs = getelementptr inbounds i16, i16* %base, <8 x i32> %offs.zext
   %ptrs = bitcast <8 x i16*> %i16_ptrs to <8 x half*>
   call void @llvm.masked.scatter.v8f16.v8p0f16(<8 x half> %input, <8 x half*> %ptrs, i32 2, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VLDRH.16 Qd, [base, offs, uxtw #1]
 define arm_aapcs_vfpcc void @scaled_v8f16_half(half* %base, <8 x i16>* %offptr, <8 x half> %input) {
 ; CHECK-LABEL: scaled_v8f16_half:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrh.u32 q1, [r1]
-; CHECK-NEXT:    vmovx.f16 s12, s0
-; CHECK-NEXT:    vshl.i32 q2, q1, #1
-; CHECK-NEXT:    vldrh.u32 q1, [r1, #8]
-; CHECK-NEXT:    vadd.i32 q2, q2, r0
-; CHECK-NEXT:    vmov r1, s8
-; CHECK-NEXT:    vshl.i32 q1, q1, #1
-; CHECK-NEXT:    vstr.16 s0, [r1]
-; CHECK-NEXT:    vmov r1, s9
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vstr.16 s12, [r1]
-; CHECK-NEXT:    vmov r1, s10
-; CHECK-NEXT:    vmovx.f16 s0, s3
-; CHECK-NEXT:    vstr.16 s1, [r1]
-; CHECK-NEXT:    vmov r1, s11
-; CHECK-NEXT:    vmovx.f16 s8, s1
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vstr.16 s8, [r1]
-; CHECK-NEXT:    vstr.16 s2, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmovx.f16 s8, s2
-; CHECK-NEXT:    vstr.16 s8, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vstr.16 s3, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vstr.16 s0, [r0]
+; CHECK-NEXT:    vldrh.u16 q1, [r1]
+; CHECK-NEXT:    vstrh.16 q0, [r0, q1, uxtw #1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i16>, <8 x i16>* %offptr, align 2
   %offs.zext = zext <8 x i16> %offs to <8 x i32>
   %ptrs = getelementptr inbounds half, half* %base, <8 x i32> %offs.zext
   call void @llvm.masked.scatter.v8f16.v8p0f16(<8 x half> %input, <8 x half*> %ptrs, i32 2, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
entry:
   call void @llvm.masked.scatter.v8f16.v8p0f16(<8 x half> %input, <8 x half*> %ptrs, i32 2, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VLDRH.16 Qd, [base, zext(offs), uxtw #1]
 define arm_aapcs_vfpcc void @unsigned_scaled_v8i16_i8(i16* %base, <8 x i8>* %offptr, <8 x i16> %input) {
 ; CHECK-LABEL: unsigned_scaled_v8i16_i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrb.u32 q2, [r1]
-; CHECK-NEXT:    vldrb.u32 q1, [r1, #4]
-; CHECK-NEXT:    vmov.u16 r1, q0[0]
-; CHECK-NEXT:    vshl.i32 q2, q2, #1
-; CHECK-NEXT:    vshl.i32 q1, q1, #1
-; CHECK-NEXT:    vadd.i32 q2, q2, r0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s9
-; CHECK-NEXT:    vmov.u16 r1, q0[1]
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s10
-; CHECK-NEXT:    vmov.u16 r1, q0[2]
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s11
-; CHECK-NEXT:    vmov.u16 r1, q0[3]
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov.u16 r1, q0[4]
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov.u16 r1, q0[5]
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov.u16 r1, q0[6]
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov.u16 r1, q0[7]
-; CHECK-NEXT:    strh r1, [r0]
+; CHECK-NEXT:    vldrb.u16 q1, [r1]
+; CHECK-NEXT:    vstrh.16 q0, [r0, q1, uxtw #1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i8>, <8 x i8>* %offptr, align 1
   %offs.zext = zext <8 x i8> %offs to <8 x i32>
   %ptrs = getelementptr inbounds i16, i16* %base, <8 x i32> %offs.zext
   call void @llvm.masked.scatter.v8i16.v8p0i16(<8 x i16> %input, <8 x i16*> %ptrs, i32 2, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VLDRH.16 Qd, [base, zext(offs), uxtw #1]
 define arm_aapcs_vfpcc void @unsigned_scaled_v8f16_i8(i16* %base, <8 x i8>* %offptr, <8 x half> %input) {
 ; CHECK-LABEL: unsigned_scaled_v8f16_i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrb.u32 q1, [r1]
-; CHECK-NEXT:    vmovx.f16 s12, s0
-; CHECK-NEXT:    vshl.i32 q2, q1, #1
-; CHECK-NEXT:    vldrb.u32 q1, [r1, #4]
-; CHECK-NEXT:    vadd.i32 q2, q2, r0
-; CHECK-NEXT:    vmov r1, s8
-; CHECK-NEXT:    vshl.i32 q1, q1, #1
-; CHECK-NEXT:    vstr.16 s0, [r1]
-; CHECK-NEXT:    vmov r1, s9
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vstr.16 s12, [r1]
-; CHECK-NEXT:    vmov r1, s10
-; CHECK-NEXT:    vmovx.f16 s0, s3
-; CHECK-NEXT:    vstr.16 s1, [r1]
-; CHECK-NEXT:    vmov r1, s11
-; CHECK-NEXT:    vmovx.f16 s8, s1
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vstr.16 s8, [r1]
-; CHECK-NEXT:    vstr.16 s2, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmovx.f16 s8, s2
-; CHECK-NEXT:    vstr.16 s8, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vstr.16 s3, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vstr.16 s0, [r0]
+; CHECK-NEXT:    vldrb.u16 q1, [r1]
+; CHECK-NEXT:    vstrh.16 q0, [r0, q1, uxtw #1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i8>, <8 x i8>* %offptr, align 1
   %offs.zext = zext <8 x i8> %offs to <8 x i32>
   %i16_ptrs = getelementptr inbounds i16, i16* %base, <8 x i32> %offs.zext
   %ptrs = bitcast <8 x i16*> %i16_ptrs to <8 x half*>
   call void @llvm.masked.scatter.v8f16.v8p0f16(<8 x half> %input, <8 x half*> %ptrs, i32 2, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 define arm_aapcs_vfpcc void @scaled_v8i16_i16_passthru_icmp0(i16* %base, <8 x i16>* %offptr, <8 x i16> %input) {
 ; CHECK-LABEL: scaled_v8i16_i16_passthru_icmp0:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .save {r7, lr}
-; CHECK-NEXT:    push {r7, lr}
-; CHECK-NEXT:    .pad #8
-; CHECK-NEXT:    sub sp, #8
-; CHECK-NEXT:    vldrh.u16 q2, [r1]
-; CHECK-NEXT:    movs r2, #0
-; CHECK-NEXT:    vcmp.s16 gt, q2, zr
-; CHECK-NEXT:    vmov.u16 r3, q2[4]
-; CHECK-NEXT:    vmrs r1, p0
-; CHECK-NEXT:    vmov.32 q1[0], r3
-; CHECK-NEXT:    vmov.u16 r3, q2[5]
-; CHECK-NEXT:    vmov.u16 r12, q2[7]
-; CHECK-NEXT:    vmov.32 q1[1], r3
-; CHECK-NEXT:    vmov.u16 r3, q2[6]
-; CHECK-NEXT:    vmov.32 q1[2], r3
-; CHECK-NEXT:    vmov.u16 r3, q2[0]
-; CHECK-NEXT:    vmov.32 q3[0], r3
-; CHECK-NEXT:    vmov.u16 r3, q2[1]
-; CHECK-NEXT:    vmov.32 q3[1], r3
-; CHECK-NEXT:    vmov.u16 r3, q2[2]
-; CHECK-NEXT:    vmov.32 q3[2], r3
-; CHECK-NEXT:    vmov.u16 r3, q2[3]
-; CHECK-NEXT:    vmov.32 q3[3], r3
-; CHECK-NEXT:    vmov.32 q1[3], r12
-; CHECK-NEXT:    vmovlb.u16 q2, q3
-; CHECK-NEXT:    vmovlb.u16 q1, q1
-; CHECK-NEXT:    vshl.i32 q2, q2, #1
-; CHECK-NEXT:    vshl.i32 q1, q1, #1
-; CHECK-NEXT:    vadd.i32 q2, q2, r0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    and lr, r1, #1
-; CHECK-NEXT:    rsb.w r3, lr, #0
-; CHECK-NEXT:    bfi r2, r3, #0, #1
-; CHECK-NEXT:    ubfx r3, r1, #2, #1
-; CHECK-NEXT:    rsbs r3, r3, #0
-; CHECK-NEXT:    bfi r2, r3, #1, #1
-; CHECK-NEXT:    ubfx r3, r1, #4, #1
-; CHECK-NEXT:    rsbs r3, r3, #0
-; CHECK-NEXT:    bfi r2, r3, #2, #1
-; CHECK-NEXT:    ubfx r3, r1, #6, #1
-; CHECK-NEXT:    rsbs r3, r3, #0
-; CHECK-NEXT:    bfi r2, r3, #3, #1
-; CHECK-NEXT:    ubfx r3, r1, #8, #1
-; CHECK-NEXT:    rsbs r3, r3, #0
-; CHECK-NEXT:    bfi r2, r3, #4, #1
-; CHECK-NEXT:    ubfx r3, r1, #10, #1
-; CHECK-NEXT:    rsbs r3, r3, #0
-; CHECK-NEXT:    bfi r2, r3, #5, #1
-; CHECK-NEXT:    ubfx r3, r1, #12, #1
-; CHECK-NEXT:    ubfx r1, r1, #14, #1
-; CHECK-NEXT:    rsbs r3, r3, #0
-; CHECK-NEXT:    bfi r2, r3, #6, #1
-; CHECK-NEXT:    rsbs r1, r1, #0
-; CHECK-NEXT:    bfi r2, r1, #7, #1
-; CHECK-NEXT:    uxtb r1, r2
-; CHECK-NEXT:    lsls r2, r2, #31
-; CHECK-NEXT:    ittt ne
-; CHECK-NEXT:    vmovne r2, s8
-; CHECK-NEXT:    vmovne.u16 r3, q0[0]
-; CHECK-NEXT:    strhne r3, [r2]
-; CHECK-NEXT:    lsls r2, r1, #30
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi r2, s9
-; CHECK-NEXT:    vmovmi.u16 r3, q0[1]
-; CHECK-NEXT:    strhmi r3, [r2]
-; CHECK-NEXT:    lsls r2, r1, #29
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi r2, s10
-; CHECK-NEXT:    vmovmi.u16 r3, q0[2]
-; CHECK-NEXT:    strhmi r3, [r2]
-; CHECK-NEXT:    lsls r2, r1, #28
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi r2, s11
-; CHECK-NEXT:    vmovmi.u16 r3, q0[3]
-; CHECK-NEXT:    strhmi r3, [r2]
-; CHECK-NEXT:    lsls r0, r1, #27
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi.u16 r0, q0[4]
-; CHECK-NEXT:    vmovmi r2, s4
-; CHECK-NEXT:    strhmi r0, [r2]
-; CHECK-NEXT:    lsls r0, r1, #26
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi.u16 r0, q0[5]
-; CHECK-NEXT:    vmovmi r2, s5
-; CHECK-NEXT:    strhmi r0, [r2]
-; CHECK-NEXT:    lsls r0, r1, #25
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi.u16 r0, q0[6]
-; CHECK-NEXT:    vmovmi r2, s6
-; CHECK-NEXT:    strhmi r0, [r2]
-; CHECK-NEXT:    lsls r0, r1, #24
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi.u16 r0, q0[7]
-; CHECK-NEXT:    vmovmi r1, s7
-; CHECK-NEXT:    strhmi r0, [r1]
-; CHECK-NEXT:    add sp, #8
-; CHECK-NEXT:    pop {r7, pc}
+; CHECK-NEXT:    vldrh.u16 q1, [r1]
+; CHECK-NEXT:    vpt.s16 gt, q1, zr
+; CHECK-NEXT:    vstrht.16 q0, [r0, q1, uxtw #1]
+; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i16>, <8 x i16>* %offptr, align 2
   %offs.zext = zext <8 x i16> %offs to <8 x i32>
   %ptrs = getelementptr inbounds i16, i16* %base, <8 x i32> %offs.zext
   %mask = icmp sgt <8 x i16> %offs, zeroinitializer
   call void @llvm.masked.scatter.v8i16.v8p0i16(<8 x i16> %input, <8 x i16*> %ptrs, i32 2, <8 x i1> %mask)
   ret void
 }
 
 declare void @llvm.masked.scatter.v8i16.v8p0i16(<8 x i16>, <8 x i16*>, i32, <8 x i1>)
 declare void @llvm.masked.scatter.v8f16.v8p0f16(<8 x half>, <8 x half*>, i32, <8 x i1>)

llvm/test/CodeGen/Thumb2/mve-scatter-ind16-unscaled.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=thumbv8.1m.main-arm-none-eabi -mattr=+mve.fp -enable-arm-maskedgatscat %s -o - | FileCheck %s
 
 ; VLDRB.u16 Qd, [base, offs]
 define arm_aapcs_vfpcc void @ext_unscaled_i8_i16(i8* %base, <8 x i16>* %offptr, <8 x i16> %input) {
 ; CHECK-LABEL: ext_unscaled_i8_i16:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrh.u32 q2, [r1]
-; CHECK-NEXT:    vldrh.u32 q1, [r1, #8]
-; CHECK-NEXT:    vmov.u16 r1, q0[0]
-; CHECK-NEXT:    vadd.i32 q2, q2, r0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s9
-; CHECK-NEXT:    vmov.u16 r1, q0[1]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s10
-; CHECK-NEXT:    vmov.u16 r1, q0[2]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s11
-; CHECK-NEXT:    vmov.u16 r1, q0[3]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov.u16 r1, q0[4]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov.u16 r1, q0[5]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov.u16 r1, q0[6]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov.u16 r1, q0[7]
-; CHECK-NEXT:    strb r1, [r0]
+; CHECK-NEXT:    vldrh.u16 q1, [r1]
+; CHECK-NEXT:    vstrb.16 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i16>, <8 x i16>* %offptr, align 2
   %offs.zext = zext <8 x i16> %offs to <8 x i32>
   %ptrs = getelementptr inbounds i8, i8* %base, <8 x i32> %offs.zext
   %t = trunc <8 x i16> %input to <8 x i8>
   call void @llvm.masked.scatter.v8i8(<8 x i8> %t, <8 x i8*> %ptrs, i32 2, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VLDRB.u16 Qd, [base, offs]
 define arm_aapcs_vfpcc void @trunc_unsigned_unscaled_i16_i8(i8* %base, <8 x i8>* %offptr, <8 x i16> %input) {
 ; CHECK-LABEL: trunc_unsigned_unscaled_i16_i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrb.u32 q2, [r1]
-; CHECK-NEXT:    vldrb.u32 q1, [r1, #4]
-; CHECK-NEXT:    vmov.u16 r1, q0[0]
-; CHECK-NEXT:    vadd.i32 q2, q2, r0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s9
-; CHECK-NEXT:    vmov.u16 r1, q0[1]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s10
-; CHECK-NEXT:    vmov.u16 r1, q0[2]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s11
-; CHECK-NEXT:    vmov.u16 r1, q0[3]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov.u16 r1, q0[4]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov.u16 r1, q0[5]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov.u16 r1, q0[6]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov.u16 r1, q0[7]
-; CHECK-NEXT:    strb r1, [r0]
+; CHECK-NEXT:    vldrb.u16 q1, [r1]
+; CHECK-NEXT:    vstrb.16 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i8>, <8 x i8>* %offptr, align 1
   %offs.zext = zext <8 x i8> %offs to <8 x i32>
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <8 x i32> %offs.zext
   %input.trunc = trunc <8 x i16> %input to <8 x i8>
   call void @llvm.masked.scatter.v8i8(<8 x i8> %input.trunc, <8 x i8*> %byte_ptrs, i32 2, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VLDRH.16 Qd, [base, offs]
 define arm_aapcs_vfpcc void @unscaled_i16_i16(i8* %base, <8 x i16>* %offptr, <8 x i16> %input) {
 ; CHECK-LABEL: unscaled_i16_i16:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrh.u32 q2, [r1]
-; CHECK-NEXT:    vldrh.u32 q1, [r1, #8]
-; CHECK-NEXT:    vmov.u16 r1, q0[0]
-; CHECK-NEXT:    vadd.i32 q2, q2, r0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s9
-; CHECK-NEXT:    vmov.u16 r1, q0[1]
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s10
-; CHECK-NEXT:    vmov.u16 r1, q0[2]
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s11
-; CHECK-NEXT:    vmov.u16 r1, q0[3]
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov.u16 r1, q0[4]
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov.u16 r1, q0[5]
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov.u16 r1, q0[6]
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov.u16 r1, q0[7]
-; CHECK-NEXT:    strh r1, [r0]
+; CHECK-NEXT:    vldrh.u16 q1, [r1]
+; CHECK-NEXT:    vstrh.16 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i16>, <8 x i16>* %offptr, align 2
   %offs.zext = zext <8 x i16> %offs to <8 x i32>
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <8 x i32> %offs.zext
   %ptrs = bitcast <8 x i8*> %byte_ptrs to <8 x i16*>
   call void @llvm.masked.scatter.v8i16(<8 x i16> %input, <8 x i16*> %ptrs, i32 2, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VLDRH.s16 Qd, [base, offs]
 define arm_aapcs_vfpcc void @unscaled_v8f16_i16(i8* %base, <8 x i16>* %offptr, <8 x half> %input) {
 ; CHECK-LABEL: unscaled_v8f16_i16:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrh.u32 q2, [r1]
-; CHECK-NEXT:    vldrh.u32 q1, [r1, #8]
-; CHECK-NEXT:    vmovx.f16 s12, s0
-; CHECK-NEXT:    vadd.i32 q2, q2, r0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r1, s8
-; CHECK-NEXT:    vstr.16 s0, [r1]
-; CHECK-NEXT:    vmov r1, s9
-; CHECK-NEXT:    vstr.16 s12, [r1]
-; CHECK-NEXT:    vmov r1, s10
-; CHECK-NEXT:    vstr.16 s1, [r1]
-; CHECK-NEXT:    vmov r1, s11
-; CHECK-NEXT:    vmovx.f16 s8, s1
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vstr.16 s8, [r1]
-; CHECK-NEXT:    vstr.16 s2, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmovx.f16 s8, s2
-; CHECK-NEXT:    vstr.16 s8, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vstr.16 s3, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmovx.f16 s0, s3
-; CHECK-NEXT:    vstr.16 s0, [r0]
+; CHECK-NEXT:    vldrh.u16 q1, [r1]
+; CHECK-NEXT:    vstrh.16 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i16>, <8 x i16>* %offptr, align 2
   %offs.zext = zext <8 x i16> %offs to <8 x i32>
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <8 x i32> %offs.zext
   %ptrs = bitcast <8 x i8*> %byte_ptrs to <8 x half*>
   call void @llvm.masked.scatter.v8f16(<8 x half> %input, <8 x half*> %ptrs, i32 2, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
   ret void
entry:
   call void @llvm.masked.scatter.v8f16(<8 x half> %input, <8 x half*> %ptrs, i32 2, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VLDRH.16 Qd, [base, zext(offs)]
 define arm_aapcs_vfpcc void @unsigned_unscaled_i16_i8(i8* %base, <8 x i8>* %offptr, <8 x i16> %input) {
 ; CHECK-LABEL: unsigned_unscaled_i16_i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrb.u32 q2, [r1]
-; CHECK-NEXT:    vldrb.u32 q1, [r1, #4]
-; CHECK-NEXT:    vmov.u16 r1, q0[0]
-; CHECK-NEXT:    vadd.i32 q2, q2, r0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s9
-; CHECK-NEXT:    vmov.u16 r1, q0[1]
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s10
-; CHECK-NEXT:    vmov.u16 r1, q0[2]
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s11
-; CHECK-NEXT:    vmov.u16 r1, q0[3]
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov.u16 r1, q0[4]
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov.u16 r1, q0[5]
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov.u16 r1, q0[6]
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov.u16 r1, q0[7]
-; CHECK-NEXT:    strh r1, [r0]
+; CHECK-NEXT:    vldrb.u16 q1, [r1]
+; CHECK-NEXT:    vstrh.16 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i8>, <8 x i8>* %offptr, align 1
   %offs.zext = zext <8 x i8> %offs to <8 x i32>
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <8 x i32> %offs.zext
   %ptrs = bitcast <8 x i8*> %byte_ptrs to <8 x i16*>
   call void @llvm.masked.scatter.v8i16(<8 x i16> %input, <8 x i16*> %ptrs, i32 2, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VLDRH.16 Qd, [base, zext(offs)]
 define arm_aapcs_vfpcc void @unsigned_unscaled_f16_i8(i8* %base, <8 x i8>* %offptr, <8 x half> %input) {
 ; CHECK-LABEL: unsigned_unscaled_f16_i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrb.u32 q2, [r1]
-; CHECK-NEXT:    vldrb.u32 q1, [r1, #4]
-; CHECK-NEXT:    vmovx.f16 s12, s0
-; CHECK-NEXT:    vadd.i32 q2, q2, r0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r1, s8
-; CHECK-NEXT:    vstr.16 s0, [r1]
-; CHECK-NEXT:    vmov r1, s9
-; CHECK-NEXT:    vstr.16 s12, [r1]
-; CHECK-NEXT:    vmov r1, s10
-; CHECK-NEXT:    vstr.16 s1, [r1]
-; CHECK-NEXT:    vmov r1, s11
-; CHECK-NEXT:    vmovx.f16 s8, s1
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vstr.16 s8, [r1]
-; CHECK-NEXT:    vstr.16 s2, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmovx.f16 s8, s2
-; CHECK-NEXT:    vstr.16 s8, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vstr.16 s3, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmovx.f16 s0, s3
-; CHECK-NEXT:    vstr.16 s0, [r0]
+; CHECK-NEXT:    vldrb.u16 q1, [r1]
+; CHECK-NEXT:    vstrh.16 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i8>, <8 x i8>* %offptr, align 1
   %offs.zext = zext <8 x i8> %offs to <8 x i32>
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <8 x i32> %offs.zext
   %ptrs = bitcast <8 x i8*> %byte_ptrs to <8 x half*>
   call void @llvm.masked.scatter.v8f16(<8 x half> %input, <8 x half*> %ptrs, i32 2, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
   ret void
entry:
   call void @llvm.masked.scatter.v8i16(<8 x i16> %input.trunc, <8 x i16*> %ptrs, i32 2, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; Expand ?
 define arm_aapcs_vfpcc void @trunc_unsigned_unscaled_i64_i8(i8* %base, <8 x i8>* %offptr, <8 x i64> %input) {
 ; CHECK-LABEL: trunc_unsigned_unscaled_i64_i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .vsave {d8, d9, d10, d11}
-; CHECK-NEXT:    vpush {d8, d9, d10, d11}
-; CHECK-NEXT:    vldrb.u32 q5, [r1]
-; CHECK-NEXT:    vldrb.u32 q4, [r1, #4]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vadd.i32 q5, q5, r0
-; CHECK-NEXT:    vadd.i32 q4, q4, r0
-; CHECK-NEXT:    vmov r0, s20
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s21
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s22
-; CHECK-NEXT:    vmov r1, s4
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s23
-; CHECK-NEXT:    vmov r1, s6
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s16
-; CHECK-NEXT:    vmov r1, s8
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s17
-; CHECK-NEXT:    vmov r1, s10
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s18
-; CHECK-NEXT:    vmov r1, s12
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s19
-; CHECK-NEXT:    vmov r1, s14
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vpop {d8, d9, d10, d11}
+; CHECK-NEXT:    .vsave {d8, d9}
+; CHECK-NEXT:    vpush {d8, d9}
+; CHECK-NEXT:    vmov r3, s0
+; CHECK-NEXT:    vmov.16 q4[0], r3
+; CHECK-NEXT:    vmov r3, s2
+; CHECK-NEXT:    vmov.16 q4[1], r3
+; CHECK-NEXT:    vmov r3, s4
+; CHECK-NEXT:    vmov.16 q4[2], r3
+; CHECK-NEXT:    vmov r3, s6
+; CHECK-NEXT:    vmov.16 q4[3], r3
+; CHECK-NEXT:    vmov r3, s8
+; CHECK-NEXT:    vmov.16 q4[4], r3
+; CHECK-NEXT:    vmov r3, s10
+; CHECK-NEXT:    vmov.16 q4[5], r3
+; CHECK-NEXT:    vmov r3, s12
+; CHECK-NEXT:    vmov r2, s14
+; CHECK-NEXT:    vmov.16 q4[6], r3
+; CHECK-NEXT:    vldrb.u16 q0, [r1]
+; CHECK-NEXT:    vmov.16 q4[7], r2
+; CHECK-NEXT:    vstrh.16 q4, [r0, q0]
+; CHECK-NEXT:    vpop {d8, d9}
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i8>, <8 x i8>* %offptr, align 1
   %offs.zext = zext <8 x i8> %offs to <8 x i32>
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <8 x i32> %offs.zext
   %ptrs = bitcast <8 x i8*> %byte_ptrs to <8 x i16*>
   %input.trunc = trunc <8 x i64> %input to <8 x i16>
   call void @llvm.masked.scatter.v8i16(<8 x i16> %input.trunc, <8 x i16*> %ptrs, i32 2, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
entry:
   call void @llvm.masked.scatter.v8i16(<8 x i16> %input.trunc, <8 x i16*> %ptrs, i32 2, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; Expand ?
 define arm_aapcs_vfpcc void @trunc_unsigned_unscaled_i32_i8(i8* %base, <8 x i8>* %offptr, <8 x i32> %input) {
 ; CHECK-LABEL: trunc_unsigned_unscaled_i32_i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrb.u32 q3, [r1]
-; CHECK-NEXT:    vldrb.u32 q2, [r1, #4]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vadd.i32 q3, q3, r0
-; CHECK-NEXT:    vadd.i32 q2, q2, r0
-; CHECK-NEXT:    vmov r0, s12
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s13
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s14
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s15
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vmov r1, s4
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s9
-; CHECK-NEXT:    vmov r1, s5
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s10
-; CHECK-NEXT:    vmov r1, s6
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s11
-; CHECK-NEXT:    vmov r1, s7
-; CHECK-NEXT:    strh r1, [r0]
+; CHECK-NEXT:    vmov r3, s0
+; CHECK-NEXT:    vmov.16 q2[0], r3
+; CHECK-NEXT:    vmov r3, s1
+; CHECK-NEXT:    vmov.16 q2[1], r3
+; CHECK-NEXT:    vmov r3, s2
+; CHECK-NEXT:    vmov.16 q2[2], r3
+; CHECK-NEXT:    vmov r3, s3
+; CHECK-NEXT:    vmov.16 q2[3], r3
+; CHECK-NEXT:    vmov r3, s4
+; CHECK-NEXT:    vmov.16 q2[4], r3
+; CHECK-NEXT:    vmov r3, s5
+; CHECK-NEXT:    vmov.16 q2[5], r3
+; CHECK-NEXT:    vmov r3, s6
+; CHECK-NEXT:    vmov r2, s7
+; CHECK-NEXT:    vmov.16 q2[6], r3
+; CHECK-NEXT:    vldrb.u16 q0, [r1]
+; CHECK-NEXT:    vmov.16 q2[7], r2
+; CHECK-NEXT:    vstrh.16 q2, [r0, q0]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i8>, <8 x i8>* %offptr, align 1
   %offs.zext = zext <8 x i8> %offs to <8 x i32>
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <8 x i32> %offs.zext
   %ptrs = bitcast <8 x i8*> %byte_ptrs to <8 x i16*>
   %input.trunc = trunc <8 x i32> %input to <8 x i16>
   call void @llvm.masked.scatter.v8i16(<8 x i16> %input.trunc, <8 x i16*> %ptrs, i32 2, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)

llvm/test/CodeGen/Thumb2/mve-scatter-ind32-scaled.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=thumbv8.1m.main-arm-none-eabi -mattr=+mve.fp -enable-arm-maskedgatscat %s -o 2>/dev/null - | FileCheck %s
 
 ; VLDRH.u32 Qd, [base, offs, #uxtw #1]
 define arm_aapcs_vfpcc void @ext_scaled_i16_i32(i16* %base, <4 x i32>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: ext_scaled_i16_i32:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrw.u32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vshl.i32 q1, q1, #1
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    strh r1, [r0]
+; CHECK-NEXT:    vstrh.32 q0, [r0, q1, uxtw #1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i32>, <4 x i32>* %offptr, align 4
   %ptrs = getelementptr inbounds i16, i16* %base, <4 x i32> %offs
   %t = trunc <4 x i32> %input to <4 x i16>
   call void @llvm.masked.scatter.v4i16.v4p0i16(<4 x i16> %t, <4 x i16*> %ptrs, i32 2, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VSTRW.32 Qd, [base, offs, uxtw #2]
 define arm_aapcs_vfpcc void @scaled_i32_i32(i32* %base, <4 x i32>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: scaled_i32_i32:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrw.u32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vshl.i32 q1, q1, #2
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    str r1, [r0]
+; CHECK-NEXT:    vstrw.32 q0, [r0, q1, uxtw #2]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i32>, <4 x i32>* %offptr, align 4
   %ptrs = getelementptr inbounds i32, i32* %base, <4 x i32> %offs
   call void @llvm.masked.scatter.v4i32.v4p0i32(<4 x i32> %input, <4 x i32*> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VSTRW.32 Qd, [base, offs, uxtw #2]
 define arm_aapcs_vfpcc void @scaled_f32_i32(i32* %base, <4 x i32>* %offptr, <4 x float> %input) {
 ; CHECK-LABEL: scaled_f32_i32:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrw.u32 q1, [r1]
-; CHECK-NEXT:    vshl.i32 q1, q1, #2
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r3, s4
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s6
-; CHECK-NEXT:    vmov r2, s5
-; CHECK-NEXT:    vstr s0, [r3]
-; CHECK-NEXT:    vstr s1, [r2]
-; CHECK-NEXT:    vstr s2, [r1]
-; CHECK-NEXT:    vstr s3, [r0]
+; CHECK-NEXT:    vstrw.32 q0, [r0, q1, uxtw #2]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i32>, <4 x i32>* %offptr, align 4
   %i32_ptrs = getelementptr inbounds i32, i32* %base, <4 x i32> %offs
   %ptrs = bitcast <4 x i32*> %i32_ptrs to <4 x float*>
   call void @llvm.masked.scatter.v4f32.v4p0f32(<4 x float> %input, <4 x float*> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VSTRW.32 Qd, [base, offs.zext, uxtw #2]
 define arm_aapcs_vfpcc void @unsigned_scaled_b_i32_i16(i32* %base, <4 x i16>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: unsigned_scaled_b_i32_i16:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrh.u32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vshl.i32 q1, q1, #2
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    str r1, [r0]
+; CHECK-NEXT:    vstrw.32 q0, [r0, q1, uxtw #2]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i16>, <4 x i16>* %offptr, align 2
   %offs.zext = zext <4 x i16> %offs to <4 x i32>
   %ptrs = getelementptr inbounds i32, i32* %base, <4 x i32> %offs.zext
   call void @llvm.masked.scatter.v4i32.v4p0i32(<4 x i32> %input, <4 x i32*> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VSTRW.32 Qd, [base, offs.sext, uxtw #2]
 define arm_aapcs_vfpcc void @signed_scaled_i32_i16(i32* %base, <4 x i16>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: signed_scaled_i32_i16:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrh.s32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vshl.i32 q1, q1, #2
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    str r1, [r0]
+; CHECK-NEXT:    vstrw.32 q0, [r0, q1, uxtw #2]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i16>, <4 x i16>* %offptr, align 2
   %offs.sext = sext <4 x i16> %offs to <4 x i32>
   %ptrs = getelementptr inbounds i32, i32* %base, <4 x i32> %offs.sext
   call void @llvm.masked.scatter.v4i32.v4p0i32(<4 x i32> %input, <4 x i32*> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VSTRW.32 Qd, [base, offs.zext, uxtw #2]
 define arm_aapcs_vfpcc void @a_unsigned_scaled_f32_i16(i32* %base, <4 x i16>* %offptr, <4 x float> %input) {
 ; CHECK-LABEL: a_unsigned_scaled_f32_i16:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrh.u32 q1, [r1]
-; CHECK-NEXT:    vshl.i32 q1, q1, #2
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r3, s4
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s6
-; CHECK-NEXT:    vmov r2, s5
-; CHECK-NEXT:    vstr s0, [r3]
-; CHECK-NEXT:    vstr s1, [r2]
-; CHECK-NEXT:    vstr s2, [r1]
-; CHECK-NEXT:    vstr s3, [r0]
+; CHECK-NEXT:    vstrw.32 q0, [r0, q1, uxtw #2]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i16>, <4 x i16>* %offptr, align 2
   %offs.zext = zext <4 x i16> %offs to <4 x i32>
   %i32_ptrs = getelementptr inbounds i32, i32* %base, <4 x i32> %offs.zext
   %ptrs = bitcast <4 x i32*> %i32_ptrs to <4 x float*>
   call void @llvm.masked.scatter.v4f32.v4p0f32(<4 x float> %input, <4 x float*> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VSTRW.32 Qd, [base, offs.sext, uxtw #2]
 define arm_aapcs_vfpcc void @b_signed_scaled_f32_i16(i32* %base, <4 x i16>* %offptr, <4 x float> %input) {
 ; CHECK-LABEL: b_signed_scaled_f32_i16:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrh.s32 q1, [r1]
-; CHECK-NEXT:    vshl.i32 q1, q1, #2
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r3, s4
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s6
-; CHECK-NEXT:    vmov r2, s5
-; CHECK-NEXT:    vstr s0, [r3]
-; CHECK-NEXT:    vstr s1, [r2]
-; CHECK-NEXT:    vstr s2, [r1]
-; CHECK-NEXT:    vstr s3, [r0]
+; CHECK-NEXT:    vstrw.32 q0, [r0, q1, uxtw #2]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i16>, <4 x i16>* %offptr, align 2
   %offs.sext = sext <4 x i16> %offs to <4 x i32>
   %i32_ptrs = getelementptr inbounds i32, i32* %base, <4 x i32> %offs.sext
   %ptrs = bitcast <4 x i32*> %i32_ptrs to <4 x float*>
   call void @llvm.masked.scatter.v4f32.v4p0f32(<4 x float> %input, <4 x float*> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VLDRH.u32 Qd, [base, offs.sext, uxtw #1]
 define arm_aapcs_vfpcc void @ext_signed_scaled_i16_i16(i16* %base, <4 x i16>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: ext_signed_scaled_i16_i16:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrh.s32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vshl.i32 q1, q1, #1
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    strh r1, [r0]
+; CHECK-NEXT:    vstrh.32 q0, [r0, q1, uxtw #1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i16>, <4 x i16>* %offptr, align 2
   %offs.sext = sext <4 x i16> %offs to <4 x i32>
   %ptrs = getelementptr inbounds i16, i16* %base, <4 x i32> %offs.sext
   %t = trunc <4 x i32> %input to <4 x i16>
   call void @llvm.masked.scatter.v4i16.v4p0i16(<4 x i16> %t, <4 x i16*> %ptrs, i32 2, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VSTRH.32 Qd, [base, offs.sext, uxtw #1]
 define arm_aapcs_vfpcc void @ext_unsigned_scaled_i16_i16(i16* %base, <4 x i16>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: ext_unsigned_scaled_i16_i16:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrh.u32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vshl.i32 q1, q1, #1
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    strh r1, [r0]
+; CHECK-NEXT:    vstrh.32 q0, [r0, q1, uxtw #1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i16>, <4 x i16>* %offptr, align 2
   %offs.zext = zext <4 x i16> %offs to <4 x i32>
   %ptrs = getelementptr inbounds i16, i16* %base, <4 x i32> %offs.zext
   %t = trunc <4 x i32> %input to <4 x i16>
   call void @llvm.masked.scatter.v4i16.v4p0i16(<4 x i16> %t, <4 x i16*> %ptrs, i32 2, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VSTRW.32 Qd, [base, offs.zext, uxtw #2]
 define arm_aapcs_vfpcc void @unsigned_scaled_b_i32_i8(i32* %base, <4 x i8>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: unsigned_scaled_b_i32_i8:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrb.u32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vshl.i32 q1, q1, #2
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    str r1, [r0]
+; CHECK-NEXT:    vstrw.32 q0, [r0, q1, uxtw #2]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i8>, <4 x i8>* %offptr, align 1
   %offs.zext = zext <4 x i8> %offs to <4 x i32>
   %ptrs = getelementptr inbounds i32, i32* %base, <4 x i32> %offs.zext
   call void @llvm.masked.scatter.v4i32.v4p0i32(<4 x i32> %input, <4 x i32*> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VSTRW.32 Qd, [base, offs.sext, uxtw #2]
 define arm_aapcs_vfpcc void @signed_scaled_i32_i8(i32* %base, <4 x i8>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: signed_scaled_i32_i8:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrb.s32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vshl.i32 q1, q1, #2
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    str r1, [r0]
+; CHECK-NEXT:    vstrw.32 q0, [r0, q1, uxtw #2]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i8>, <4 x i8>* %offptr, align 1
   %offs.sext = sext <4 x i8> %offs to <4 x i32>
   %ptrs = getelementptr inbounds i32, i32* %base, <4 x i32> %offs.sext
   call void @llvm.masked.scatter.v4i32.v4p0i32(<4 x i32> %input, <4 x i32*> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VSTRW.32 Qd, [base, offs.zext, uxtw #2]
 define arm_aapcs_vfpcc void @a_unsigned_scaled_f32_i8(i32* %base, <4 x i8>* %offptr, <4 x float> %input) {
 ; CHECK-LABEL: a_unsigned_scaled_f32_i8:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrb.u32 q1, [r1]
-; CHECK-NEXT:    vshl.i32 q1, q1, #2
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r3, s4
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s6
-; CHECK-NEXT:    vmov r2, s5
-; CHECK-NEXT:    vstr s0, [r3]
-; CHECK-NEXT:    vstr s1, [r2]
-; CHECK-NEXT:    vstr s2, [r1]
-; CHECK-NEXT:    vstr s3, [r0]
+; CHECK-NEXT:    vstrw.32 q0, [r0, q1, uxtw #2]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i8>, <4 x i8>* %offptr, align 1
   %offs.zext = zext <4 x i8> %offs to <4 x i32>
   %i32_ptrs = getelementptr inbounds i32, i32* %base, <4 x i32> %offs.zext
   %ptrs = bitcast <4 x i32*> %i32_ptrs to <4 x float*>
   call void @llvm.masked.scatter.v4f32.v4p0f32(<4 x float> %input, <4 x float*> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VSTRW.32 Qd, [base, offs.sext, uxtw #2]
 define arm_aapcs_vfpcc void @b_signed_scaled_f32_i8(i32* %base, <4 x i8>* %offptr, <4 x float> %input) {
 ; CHECK-LABEL: b_signed_scaled_f32_i8:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrb.s32 q1, [r1]
-; CHECK-NEXT:    vshl.i32 q1, q1, #2
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r3, s4
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s6
-; CHECK-NEXT:    vmov r2, s5
-; CHECK-NEXT:    vstr s0, [r3]
-; CHECK-NEXT:    vstr s1, [r2]
-; CHECK-NEXT:    vstr s2, [r1]
-; CHECK-NEXT:    vstr s3, [r0]
+; CHECK-NEXT:    vstrw.32 q0, [r0, q1, uxtw #2]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i8>, <4 x i8>* %offptr, align 1
   %offs.sext = sext <4 x i8> %offs to <4 x i32>
   %i32_ptrs = getelementptr inbounds i32, i32* %base, <4 x i32> %offs.sext
   %ptrs = bitcast <4 x i32*> %i32_ptrs to <4 x float*>
   call void @llvm.masked.scatter.v4f32.v4p0f32(<4 x float> %input, <4 x float*> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VLDRH.z32 Qd, [base, offs.sext, uxtw #1]
 define arm_aapcs_vfpcc void @ext_signed_scaled_i16_i8(i16* %base, <4 x i8>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: ext_signed_scaled_i16_i8:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrb.s32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vshl.i32 q1, q1, #1
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    strh r1, [r0]
+; CHECK-NEXT:    vstrh.32 q0, [r0, q1, uxtw #1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i8>, <4 x i8>* %offptr, align 1
   %offs.sext = sext <4 x i8> %offs to <4 x i32>
   %ptrs = getelementptr inbounds i16, i16* %base, <4 x i32> %offs.sext
   %t = trunc <4 x i32> %input to <4 x i16>
   call void @llvm.masked.scatter.v4i16.v4p0i16(<4 x i16> %t, <4 x i16*> %ptrs, i32 2, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VLDRH.z32 Qd, [base, offs.zext, uxtw #1]
 define arm_aapcs_vfpcc void @ext_unsigned_scaled_i16_i8(i16* %base, <4 x i8>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: ext_unsigned_scaled_i16_i8:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrb.u32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vshl.i32 q1, q1, #1
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    strh r1, [r0]
+; CHECK-NEXT:    vstrh.32 q0, [r0, q1, uxtw #1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i8>, <4 x i8>* %offptr, align 1
   %offs.zext = zext <4 x i8> %offs to <4 x i32>
   %ptrs = getelementptr inbounds i16, i16* %base, <4 x i32> %offs.zext
   %t = trunc <4 x i32> %input to <4 x i16>
   call void @llvm.masked.scatter.v4i16.v4p0i16(<4 x i16> %t, <4 x i16*> %ptrs, i32 2, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 declare void @llvm.masked.scatter.v4i8.v4p0i8(<4 x i8>, <4 x i8*>, i32, <4 x i1>)
 declare void @llvm.masked.scatter.v4i16.v4p0i16(<4 x i16>, <4 x i16*>, i32, <4 x i1>)
 declare void @llvm.masked.scatter.v4f16.v4p0f16(<4 x half>, <4 x half*>, i32, <4 x i1>)
 declare void @llvm.masked.scatter.v4i32.v4p0i32(<4 x i32>, <4 x i32*>, i32, <4 x i1>)
 declare void @llvm.masked.scatter.v4f32.v4p0f32(<4 x float>, <4 x float*>, i32, <4 x i1>)

llvm/test/CodeGen/Thumb2/mve-scatter-ind32-unscaled.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=thumbv8.1m.main-arm-none-eabi -mattr=+mve.fp -enable-arm-maskedgatscat %s -o 2>/dev/null - | FileCheck %s
 
 ; VLDRB.u32 Qd, [base, offs]
 define arm_aapcs_vfpcc void @ext_unscaled_i8_i32(i8* %base, <4 x i32>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: ext_unscaled_i8_i32:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrw.u32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    strb r1, [r0]
+; CHECK-NEXT:    vstrb.32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i32>, <4 x i32>* %offptr, align 4
   %ptrs = getelementptr inbounds i8, i8* %base, <4 x i32> %offs
   %t = trunc <4 x i32> %input to <4 x i8>
   call void @llvm.masked.scatter.v4i8.v4p0i8(<4 x i8> %t, <4 x i8*> %ptrs, i32 2, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VLDRH.u32 Qd, [base, offs]
 define arm_aapcs_vfpcc void @ext_unscaled_i16_i32(i8* %base, <4 x i32>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: ext_unscaled_i16_i32:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrw.u32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    strh r1, [r0]
+; CHECK-NEXT:    vstrh.32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i32>, <4 x i32>* %offptr, align 4
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <4 x i32> %offs
   %ptrs = bitcast <4 x i8*> %byte_ptrs to <4 x i16*>
   %t = trunc <4 x i32> %input to <4 x i16>
   call void @llvm.masked.scatter.v4i16.v4p0i16(<4 x i16> %t, <4 x i16*> %ptrs, i32 2, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VSTRW.32 Qd, [base, offs]
 define arm_aapcs_vfpcc void @unscaled_i32_i32(i8* %base, <4 x i32>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: unscaled_i32_i32:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrw.u32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    str r1, [r0]
+; CHECK-NEXT:    vstrw.32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i32>, <4 x i32>* %offptr, align 4
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <4 x i32> %offs
   %ptrs = bitcast <4 x i8*> %byte_ptrs to <4 x i32*>
   call void @llvm.masked.scatter.v4i32.v4p0i32(<4 x i32> %input, <4 x i32*> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VSTRW.32 Qd, [base, offs]
 define arm_aapcs_vfpcc void @unscaled_f32_i32(i8* %base, <4 x i32>* %offptr, <4 x float> %input) {
 ; CHECK-LABEL: unscaled_f32_i32:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrw.u32 q1, [r1]
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r3, s4
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s6
-; CHECK-NEXT:    vmov r2, s5
-; CHECK-NEXT:    vstr s0, [r3]
-; CHECK-NEXT:    vstr s1, [r2]
-; CHECK-NEXT:    vstr s2, [r1]
-; CHECK-NEXT:    vstr s3, [r0]
+; CHECK-NEXT:    vstrw.32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i32>, <4 x i32>* %offptr, align 4
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <4 x i32> %offs
   %ptrs = bitcast <4 x i8*> %byte_ptrs to <4 x float*>
   call void @llvm.masked.scatter.v4f32.v4p0f32(<4 x float> %input, <4 x float*> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VSTRW.32 Qd, [base, offs.zext]
 define arm_aapcs_vfpcc void @unsigned_unscaled_b_i32_i16(i8* %base, <4 x i16>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: unsigned_unscaled_b_i32_i16:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrh.u32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    str r1, [r0]
+; CHECK-NEXT:    vstrw.32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i16>, <4 x i16>* %offptr, align 2
   %offs.zext = zext <4 x i16> %offs to <4 x i32>
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <4 x i32> %offs.zext
   %ptrs = bitcast <4 x i8*> %byte_ptrs to <4 x i32*>
   call void @llvm.masked.scatter.v4i32.v4p0i32(<4 x i32> %input, <4 x i32*> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VSTRW.32 Qd, [base, offs.sext]
 define arm_aapcs_vfpcc void @signed_unscaled_i32_i16(i8* %base, <4 x i16>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: signed_unscaled_i32_i16:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrh.s32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    str r1, [r0]
+; CHECK-NEXT:    vstrw.32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i16>, <4 x i16>* %offptr, align 2
   %offs.sext = sext <4 x i16> %offs to <4 x i32>
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <4 x i32> %offs.sext
   %ptrs = bitcast <4 x i8*> %byte_ptrs to <4 x i32*>
   call void @llvm.masked.scatter.v4i32.v4p0i32(<4 x i32> %input, <4 x i32*> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VSTRW.32 Qd, [base, offs.zext]
 define arm_aapcs_vfpcc void @a_unsigned_unscaled_f32_i16(i8* %base, <4 x i16>* %offptr, <4 x float> %input) {
 ; CHECK-LABEL: a_unsigned_unscaled_f32_i16:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrh.u32 q1, [r1]
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r3, s4
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s6
-; CHECK-NEXT:    vmov r2, s5
-; CHECK-NEXT:    vstr s0, [r3]
-; CHECK-NEXT:    vstr s1, [r2]
-; CHECK-NEXT:    vstr s2, [r1]
-; CHECK-NEXT:    vstr s3, [r0]
+; CHECK-NEXT:    vstrw.32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i16>, <4 x i16>* %offptr, align 2
   %offs.zext = zext <4 x i16> %offs to <4 x i32>
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <4 x i32> %offs.zext
   %ptrs = bitcast <4 x i8*> %byte_ptrs to <4 x float*>
   call void @llvm.masked.scatter.v4f32.v4p0f32(<4 x float> %input, <4 x float*> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VSTRW.32 Qd, [base, offs.sext]
 define arm_aapcs_vfpcc void @b_signed_unscaled_f32_i16(i8* %base, <4 x i16>* %offptr, <4 x float> %input) {
 ; CHECK-LABEL: b_signed_unscaled_f32_i16:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrh.s32 q1, [r1]
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r3, s4
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s6
-; CHECK-NEXT:    vmov r2, s5
-; CHECK-NEXT:    vstr s0, [r3]
-; CHECK-NEXT:    vstr s1, [r2]
-; CHECK-NEXT:    vstr s2, [r1]
-; CHECK-NEXT:    vstr s3, [r0]
+; CHECK-NEXT:    vstrw.32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i16>, <4 x i16>* %offptr, align 2
   %offs.sext = sext <4 x i16> %offs to <4 x i32>
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <4 x i32> %offs.sext
   %ptrs = bitcast <4 x i8*> %byte_ptrs to <4 x float*>
   call void @llvm.masked.scatter.v4f32.v4p0f32(<4 x float> %input, <4 x float*> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VLDRH.u32 Qd, [base, offs.sext]
 define arm_aapcs_vfpcc void @ext_signed_unscaled_i16_i16(i8* %base, <4 x i16>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: ext_signed_unscaled_i16_i16:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrh.s32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    strh r1, [r0]
+; CHECK-NEXT:    vstrh.32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i16>, <4 x i16>* %offptr, align 2
   %offs.sext = sext <4 x i16> %offs to <4 x i32>
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <4 x i32> %offs.sext
   %ptrs = bitcast <4 x i8*> %byte_ptrs to <4 x i16*>
   %t = trunc <4 x i32> %input to <4 x i16>
   call void @llvm.masked.scatter.v4i16.v4p0i16(<4 x i16> %t, <4 x i16*> %ptrs, i32 2, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VLDRH.u32 Qd, [base, offs.zext]
 define arm_aapcs_vfpcc void @ext_unsigned_unscaled_i16_i16(i8* %base, <4 x i16>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: ext_unsigned_unscaled_i16_i16:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrh.u32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    strh r1, [r0]
+; CHECK-NEXT:    vstrh.32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i16>, <4 x i16>* %offptr, align 2
   %offs.zext = zext <4 x i16> %offs to <4 x i32>
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <4 x i32> %offs.zext
   %ptrs = bitcast <4 x i8*> %byte_ptrs to <4 x i16*>
   %t = trunc <4 x i32> %input to <4 x i16>
   call void @llvm.masked.scatter.v4i16.v4p0i16(<4 x i16> %t, <4 x i16*> %ptrs, i32 2, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VLDRB.u32 Qd, [base, offs.sext]
 define arm_aapcs_vfpcc void @ext_signed_unscaled_i8_i16(i8* %base, <4 x i16>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: ext_signed_unscaled_i8_i16:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrh.s32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    strb r1, [r0]
+; CHECK-NEXT:    vstrb.32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i16>, <4 x i16>* %offptr, align 2
   %offs.sext = sext <4 x i16> %offs to <4 x i32>
   %ptrs = getelementptr inbounds i8, i8* %base, <4 x i32> %offs.sext
   %t = trunc <4 x i32> %input to <4 x i8>
   call void @llvm.masked.scatter.v4i8.v4p0i8(<4 x i8> %t, <4 x i8*> %ptrs, i32 2, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VLDRB.s32 Qd, [base, offs.zext]
 define arm_aapcs_vfpcc void @ext_unsigned_unscaled_i8_i16(i8* %base, <4 x i16>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: ext_unsigned_unscaled_i8_i16:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrh.u32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    strb r1, [r0]
+; CHECK-NEXT:    vstrb.32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i16>, <4 x i16>* %offptr, align 2
   %offs.zext = zext <4 x i16> %offs to <4 x i32>
   %ptrs = getelementptr inbounds i8, i8* %base, <4 x i32> %offs.zext
   %t = trunc <4 x i32> %input to <4 x i8>
   call void @llvm.masked.scatter.v4i8.v4p0i8(<4 x i8> %t, <4 x i8*> %ptrs, i32 2, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VSTRW.32 Qd, [base, offs.zext]
 define arm_aapcs_vfpcc void @unsigned_unscaled_b_i32_i8(i8* %base, <4 x i8>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: unsigned_unscaled_b_i32_i8:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrb.u32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    str r1, [r0]
+; CHECK-NEXT:    vstrw.32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i8>, <4 x i8>* %offptr, align 1
   %offs.zext = zext <4 x i8> %offs to <4 x i32>
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <4 x i32> %offs.zext
   %ptrs = bitcast <4 x i8*> %byte_ptrs to <4 x i32*>
   call void @llvm.masked.scatter.v4i32.v4p0i32(<4 x i32> %input, <4 x i32*> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VSTRW.32 Qd, [base, offs.sext]
 define arm_aapcs_vfpcc void @signed_unscaled_i32_i8(i8* %base, <4 x i8>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: signed_unscaled_i32_i8:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrb.s32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    str r1, [r0]
+; CHECK-NEXT:    vstrw.32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i8>, <4 x i8>* %offptr, align 1
   %offs.sext = sext <4 x i8> %offs to <4 x i32>
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <4 x i32> %offs.sext
   %ptrs = bitcast <4 x i8*> %byte_ptrs to <4 x i32*>
   call void @llvm.masked.scatter.v4i32.v4p0i32(<4 x i32> %input, <4 x i32*> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VSTRW.32 Qd, [base, offs.zext]
 define arm_aapcs_vfpcc void @a_unsigned_unscaled_f32_i8(i8* %base, <4 x i8>* %offptr, <4 x float> %input) {
 ; CHECK-LABEL: a_unsigned_unscaled_f32_i8:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrb.u32 q1, [r1]
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r3, s4
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s6
-; CHECK-NEXT:    vmov r2, s5
-; CHECK-NEXT:    vstr s0, [r3]
-; CHECK-NEXT:    vstr s1, [r2]
-; CHECK-NEXT:    vstr s2, [r1]
-; CHECK-NEXT:    vstr s3, [r0]
+; CHECK-NEXT:    vstrw.32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i8>, <4 x i8>* %offptr, align 1
   %offs.zext = zext <4 x i8> %offs to <4 x i32>
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <4 x i32> %offs.zext
   %ptrs = bitcast <4 x i8*> %byte_ptrs to <4 x float*>
   call void @llvm.masked.scatter.v4f32.v4p0f32(<4 x float> %input, <4 x float*> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VSTRW.32 Qd, [base, offs.sext]
 define arm_aapcs_vfpcc void @b_signed_unscaled_f32_i8(i8* %base, <4 x i8>* %offptr, <4 x float> %input) {
 ; CHECK-LABEL: b_signed_unscaled_f32_i8:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrb.s32 q1, [r1]
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r3, s4
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s6
-; CHECK-NEXT:    vmov r2, s5
-; CHECK-NEXT:    vstr s0, [r3]
-; CHECK-NEXT:    vstr s1, [r2]
-; CHECK-NEXT:    vstr s2, [r1]
-; CHECK-NEXT:    vstr s3, [r0]
+; CHECK-NEXT:    vstrw.32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i8>, <4 x i8>* %offptr, align 1
   %offs.sext = sext <4 x i8> %offs to <4 x i32>
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <4 x i32> %offs.sext
   %ptrs = bitcast <4 x i8*> %byte_ptrs to <4 x float*>
   call void @llvm.masked.scatter.v4f32.v4p0f32(<4 x float> %input, <4 x float*> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VLDRH.u32 Qd, [base, offs.sext]
 define arm_aapcs_vfpcc void @ext_signed_unscaled_i8_i8(i8* %base, <4 x i8>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: ext_signed_unscaled_i8_i8:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrb.s32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    strb r1, [r0]
+; CHECK-NEXT:    vstrb.32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i8>, <4 x i8>* %offptr, align 1
   %offs.sext = sext <4 x i8> %offs to <4 x i32>
   %ptrs = getelementptr inbounds i8, i8* %base, <4 x i32> %offs.sext
   %t = trunc <4 x i32> %input to <4 x i8>
   call void @llvm.masked.scatter.v4i8.v4p0i8(<4 x i8> %t, <4 x i8*> %ptrs, i32 2, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; VLDRH.u32 Qd, [base, offs.zext]
 define arm_aapcs_vfpcc void @ext_unsigned_unscaled_i8_i8(i8* %base, <4 x i8>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: ext_unsigned_unscaled_i8_i8:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrb.u32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    strb r1, [r0]
+; CHECK-NEXT:    vstrb.32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i8>, <4 x i8>* %offptr, align 1
   %offs.zext = zext <4 x i8> %offs to <4 x i32>
   %ptrs = getelementptr inbounds i8, i8* %base, <4 x i32> %offs.zext
   %t = trunc <4 x i32> %input to <4 x i8>
   call void @llvm.masked.scatter.v4i8.v4p0i8(<4 x i8> %t, <4 x i8*> %ptrs, i32 2, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 define arm_aapcs_vfpcc void @trunc_signed_unscaled_i64_i8(i8* %base, <4 x i8>* %offptr, <4 x i64> %input) {
 ; CHECK-LABEL: trunc_signed_unscaled_i64_i8:
 ; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vmov.f32 s1, s2
 ; CHECK-NEXT:    vldrb.s32 q2, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vadd.i32 q2, q2, r0
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s9
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s10
-; CHECK-NEXT:    vmov r1, s4
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s11
-; CHECK-NEXT:    vmov r1, s6
-; CHECK-NEXT:    str r1, [r0]
+; CHECK-NEXT:    vmov.f32 s2, s4
+; CHECK-NEXT:    vmov.f32 s3, s6
+; CHECK-NEXT:    vstrw.32 q0, [r0, q2]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i8>, <4 x i8>* %offptr, align 1
   %offs.sext = sext <4 x i8> %offs to <4 x i32>
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <4 x i32> %offs.sext
   %ptrs = bitcast <4 x i8*> %byte_ptrs to <4 x i32*>
   %input.trunc = trunc <4 x i64> %input to <4 x i32>
   call void @llvm.masked.scatter.v4i32.v4p0i32(<4 x i32> %input.trunc, <4 x i32*> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 define arm_aapcs_vfpcc void @trunc_unsigned_unscaled_i64_i8(i8* %base, <4 x i8>* %offptr, <4 x i64> %input) {
 ; CHECK-LABEL: trunc_unsigned_unscaled_i64_i8:
 ; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vmov.f32 s1, s2
 ; CHECK-NEXT:    vldrb.u32 q2, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vadd.i32 q2, q2, r0
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s9
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s10
-; CHECK-NEXT:    vmov r1, s4
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s11
-; CHECK-NEXT:    vmov r1, s6
-; CHECK-NEXT:    str r1, [r0]
+; CHECK-NEXT:    vmov.f32 s2, s4
+; CHECK-NEXT:    vmov.f32 s3, s6
+; CHECK-NEXT:    vstrw.32 q0, [r0, q2]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i8>, <4 x i8>* %offptr, align 1
   %offs.zext = zext <4 x i8> %offs to <4 x i32>
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <4 x i32> %offs.zext
   %ptrs = bitcast <4 x i8*> %byte_ptrs to <4 x i32*>
   %input.trunc = trunc <4 x i64> %input to <4 x i32>
   call void @llvm.masked.scatter.v4i32.v4p0i32(<4 x i32> %input.trunc, <4 x i32*> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 define arm_aapcs_vfpcc void @trunc_signed_unscaled_i32_i8(i8* %base, <4 x i8>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: trunc_signed_unscaled_i32_i8:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrb.s32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    strh r1, [r0]
+; CHECK-NEXT:    vstrh.32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i8>, <4 x i8>* %offptr, align 1
   %offs.sext = sext <4 x i8> %offs to <4 x i32>
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <4 x i32> %offs.sext
   %ptrs = bitcast <4 x i8*> %byte_ptrs to <4 x i16*>
   %input.trunc = trunc <4 x i32> %input to <4 x i16>
   call void @llvm.masked.scatter.v4i16.v4p0i16(<4 x i16> %input.trunc, <4 x i16*> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 define arm_aapcs_vfpcc void @trunc_unsigned_unscaled_i32_i8(i8* %base, <4 x i8>* %offptr, <4 x i32> %input) {
 ; CHECK-LABEL: trunc_unsigned_unscaled_i32_i8:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrb.u32 q1, [r1]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    strh r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    strh r1, [r0]
+; CHECK-NEXT:    vstrh.32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i8>, <4 x i8>* %offptr, align 1
   %offs.zext = zext <4 x i8> %offs to <4 x i32>
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <4 x i32> %offs.zext
   %ptrs = bitcast <4 x i8*> %byte_ptrs to <4 x i16*>
   %input.trunc = trunc <4 x i32> %input to <4 x i16>
   call void @llvm.masked.scatter.v4i16.v4p0i16(<4 x i16> %input.trunc, <4 x i16*> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)

llvm/test/CodeGen/Thumb2/mve-scatter-ind8-unscaled.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=thumbv8.1m.main-arm-none-eabi -mattr=+mve.fp -enable-arm-maskedgatscat %s -o - | FileCheck %s
 
 ; VLDRB.8
 define arm_aapcs_vfpcc void @unscaled_v16i8_i8(i8* %base, <16 x i8>* %offptr, <16 x i8> %input) {
 ; CHECK-LABEL: unscaled_v16i8_i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .vsave {d8, d9}
-; CHECK-NEXT:    vpush {d8, d9}
-; CHECK-NEXT:    vldrb.u32 q4, [r1]
-; CHECK-NEXT:    vldrb.u32 q1, [r1, #12]
-; CHECK-NEXT:    vldrb.u32 q2, [r1, #8]
-; CHECK-NEXT:    vldrb.u32 q3, [r1, #4]
-; CHECK-NEXT:    vadd.i32 q4, q4, r0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vadd.i32 q2, q2, r0
-; CHECK-NEXT:    vadd.i32 q3, q3, r0
-; CHECK-NEXT:    vmov r0, s16
-; CHECK-NEXT:    vmov.u8 r1, q0[0]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s17
-; CHECK-NEXT:    vmov.u8 r1, q0[1]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s18
-; CHECK-NEXT:    vmov.u8 r1, q0[2]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s19
-; CHECK-NEXT:    vmov.u8 r1, q0[3]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s12
-; CHECK-NEXT:    vmov.u8 r1, q0[4]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s13
-; CHECK-NEXT:    vmov.u8 r1, q0[5]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s14
-; CHECK-NEXT:    vmov.u8 r1, q0[6]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s15
-; CHECK-NEXT:    vmov.u8 r1, q0[7]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vmov.u8 r1, q0[8]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s9
-; CHECK-NEXT:    vmov.u8 r1, q0[9]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s10
-; CHECK-NEXT:    vmov.u8 r1, q0[10]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s11
-; CHECK-NEXT:    vmov.u8 r1, q0[11]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov.u8 r1, q0[12]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov.u8 r1, q0[13]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov.u8 r1, q0[14]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov.u8 r1, q0[15]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vpop {d8, d9}
+; CHECK-NEXT:    vldrb.u8 q1, [r1]
+; CHECK-NEXT:    vstrb.8 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <16 x i8>, <16 x i8>* %offptr, align 1
   %offs.zext = zext <16 x i8> %offs to <16 x i32>
   %ptrs = getelementptr inbounds i8, i8* %base, <16 x i32> %offs.zext
   call void @llvm.masked.scatter.v16i8.v16p0i8(<16 x i8> %input, <16 x i8*> %ptrs, i32 1, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
entry:
   call void @llvm.masked.scatter.v16i8.v16p0i8(<16 x i8> %input, <16 x i8*> %ptrs, i32 1, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; Expand
 define arm_aapcs_vfpcc void @trunc_unsigned_unscaled_i64_i8(i8* %base, <16 x i8>* %offptr, <16 x i64> %input) {
 ; CHECK-LABEL: trunc_unsigned_unscaled_i64_i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .vsave {d8, d9, d10, d11, d12, d13, d14, d15}
-; CHECK-NEXT:    vpush {d8, d9, d10, d11, d12, d13, d14, d15}
-; CHECK-NEXT:    vldrb.u32 q7, [r1]
-; CHECK-NEXT:    vldrb.u32 q4, [r1, #12]
-; CHECK-NEXT:    vldrb.u32 q5, [r1, #8]
-; CHECK-NEXT:    vldrb.u32 q6, [r1, #4]
-; CHECK-NEXT:    vadd.i32 q7, q7, r0
-; CHECK-NEXT:    vadd.i32 q4, q4, r0
-; CHECK-NEXT:    vadd.i32 q5, q5, r0
-; CHECK-NEXT:    vadd.i32 q6, q6, r0
-; CHECK-NEXT:    vmov r0, s28
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s29
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s30
-; CHECK-NEXT:    vmov r1, s4
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s31
-; CHECK-NEXT:    vmov r1, s6
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s24
-; CHECK-NEXT:    vmov r1, s8
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s25
-; CHECK-NEXT:    vmov r1, s10
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s26
-; CHECK-NEXT:    vmov r1, s12
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s27
-; CHECK-NEXT:    vmov r1, s14
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    add r1, sp, #64
-; CHECK-NEXT:    vldrw.u32 q0, [r1]
-; CHECK-NEXT:    vmov r0, s20
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s21
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    add r1, sp, #80
-; CHECK-NEXT:    vldrw.u32 q0, [r1]
-; CHECK-NEXT:    vmov r0, s22
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s23
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    add r1, sp, #96
-; CHECK-NEXT:    vldrw.u32 q0, [r1]
-; CHECK-NEXT:    vmov r0, s16
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s17
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    add r1, sp, #112
-; CHECK-NEXT:    vldrw.u32 q0, [r1]
-; CHECK-NEXT:    vmov r0, s18
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s19
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vpop {d8, d9, d10, d11, d12, d13, d14, d15}
-; CHECK-NEXT:    bx lr
+; CHECK-NEXT:    .save {r4, lr}
+; CHECK-NEXT:    push {r4, lr}
+; CHECK-NEXT:    .vsave {d8, d9, d10, d11}
+; CHECK-NEXT:    vpush {d8, d9, d10, d11}
+; CHECK-NEXT:    vmov r4, s0
+; CHECK-NEXT:    add r3, sp, #40
+; CHECK-NEXT:    vmov.8 q5[0], r4
+; CHECK-NEXT:    vmov r4, s2
+; CHECK-NEXT:    vmov.8 q5[1], r4
+; CHECK-NEXT:    vmov r4, s4
+; CHECK-NEXT:    vmov.8 q5[2], r4
+; CHECK-NEXT:    vmov r4, s6
+; CHECK-NEXT:    vmov.8 q5[3], r4
+; CHECK-NEXT:    vmov r4, s8
+; CHECK-NEXT:    vmov.8 q5[4], r4
+; CHECK-NEXT:    vmov r4, s10
+; CHECK-NEXT:    vldrw.u32 q0, [r3]
+; CHECK-NEXT:    vmov.8 q5[5], r4
+; CHECK-NEXT:    vmov r4, s12
+; CHECK-NEXT:    add.w lr, sp, #56
+; CHECK-NEXT:    vmov.8 q5[6], r4
+; CHECK-NEXT:    vmov r4, s14
+; CHECK-NEXT:    vmov.8 q5[7], r4
+; CHECK-NEXT:    vmov r3, s0
+; CHECK-NEXT:    vmov.8 q5[8], r3
+; CHECK-NEXT:    vmov r3, s2
+; CHECK-NEXT:    vldrw.u32 q0, [lr]
+; CHECK-NEXT:    vmov.8 q5[9], r3
+; CHECK-NEXT:    add.w r12, sp, #72
+; CHECK-NEXT:    add r2, sp, #88
+; CHECK-NEXT:    vmov r3, s0
+; CHECK-NEXT:    vldrw.u32 q4, [r2]
+; CHECK-NEXT:    vmov.8 q5[10], r3
+; CHECK-NEXT:    vmov r3, s2
+; CHECK-NEXT:    vldrw.u32 q0, [r12]
+; CHECK-NEXT:    vmov.8 q5[11], r3
+; CHECK-NEXT:    vmov r2, s18
+; CHECK-NEXT:    vmov r3, s0
+; CHECK-NEXT:    vmov.8 q5[12], r3
+; CHECK-NEXT:    vmov r3, s2
+; CHECK-NEXT:    vmov.8 q5[13], r3
+; CHECK-NEXT:    vmov r3, s16
+; CHECK-NEXT:    vmov.8 q5[14], r3
+; CHECK-NEXT:    vldrb.u8 q0, [r1]
+; CHECK-NEXT:    vmov.8 q5[15], r2
+; CHECK-NEXT:    vstrb.8 q5, [r0, q0]
+; CHECK-NEXT:    vpop {d8, d9, d10, d11}
+; CHECK-NEXT:    pop {r4, pc}
 entry:
   %offs = load <16 x i8>, <16 x i8>* %offptr, align 1
   %offs.zext = zext <16 x i8> %offs to <16 x i32>
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <16 x i32> %offs.zext
   %input.trunc = trunc <16 x i64> %input to <16 x i8>
   call void @llvm.masked.scatter.v16i8.v16p0i8(<16 x i8> %input.trunc, <16 x i8*> %byte_ptrs, i32 1, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; Expand
 define arm_aapcs_vfpcc void @trunc_unsigned_unscaled_i32_i8(i8* %base, <16 x i8>* %offptr, <16 x i32> %input) {
 ; CHECK-LABEL: trunc_unsigned_unscaled_i32_i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .vsave {d8, d9, d10, d11, d12, d13, d14, d15}
-; CHECK-NEXT:    vpush {d8, d9, d10, d11, d12, d13, d14, d15}
-; CHECK-NEXT:    vldrb.u32 q7, [r1]
-; CHECK-NEXT:    vldrb.u32 q4, [r1, #12]
-; CHECK-NEXT:    vldrb.u32 q5, [r1, #8]
-; CHECK-NEXT:    vldrb.u32 q6, [r1, #4]
-; CHECK-NEXT:    vadd.i32 q7, q7, r0
-; CHECK-NEXT:    vadd.i32 q4, q4, r0
-; CHECK-NEXT:    vadd.i32 q5, q5, r0
-; CHECK-NEXT:    vadd.i32 q6, q6, r0
-; CHECK-NEXT:    vmov r0, s28
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s29
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s30
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s31
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s24
-; CHECK-NEXT:    vmov r1, s4
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s25
-; CHECK-NEXT:    vmov r1, s5
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s26
-; CHECK-NEXT:    vmov r1, s6
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s27
-; CHECK-NEXT:    vmov r1, s7
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s20
-; CHECK-NEXT:    vmov r1, s8
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s21
-; CHECK-NEXT:    vmov r1, s9
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s22
-; CHECK-NEXT:    vmov r1, s10
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s23
-; CHECK-NEXT:    vmov r1, s11
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s16
-; CHECK-NEXT:    vmov r1, s12
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s17
-; CHECK-NEXT:    vmov r1, s13
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s18
-; CHECK-NEXT:    vmov r1, s14
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s19
-; CHECK-NEXT:    vmov r1, s15
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vpop {d8, d9, d10, d11, d12, d13, d14, d15}
+; CHECK-NEXT:    .vsave {d8, d9}
+; CHECK-NEXT:    vpush {d8, d9}
+; CHECK-NEXT:    vmov r3, s0
+; CHECK-NEXT:    vmov.8 q4[0], r3
+; CHECK-NEXT:    vmov r3, s1
+; CHECK-NEXT:    vmov.8 q4[1], r3
+; CHECK-NEXT:    vmov r3, s2
+; CHECK-NEXT:    vmov.8 q4[2], r3
+; CHECK-NEXT:    vmov r3, s3
+; CHECK-NEXT:    vmov.8 q4[3], r3
+; CHECK-NEXT:    vmov r3, s4
+; CHECK-NEXT:    vmov.8 q4[4], r3
+; CHECK-NEXT:    vmov r3, s5
+; CHECK-NEXT:    vmov.8 q4[5], r3
+; CHECK-NEXT:    vmov r3, s6
+; CHECK-NEXT:    vmov.8 q4[6], r3
+; CHECK-NEXT:    vmov r3, s7
+; CHECK-NEXT:    vmov.8 q4[7], r3
+; CHECK-NEXT:    vmov r3, s8
+; CHECK-NEXT:    vmov.8 q4[8], r3
+; CHECK-NEXT:    vmov r3, s9
+; CHECK-NEXT:    vmov.8 q4[9], r3
+; CHECK-NEXT:    vmov r3, s10
+; CHECK-NEXT:    vmov.8 q4[10], r3
+; CHECK-NEXT:    vmov r3, s11
+; CHECK-NEXT:    vmov.8 q4[11], r3
+; CHECK-NEXT:    vmov r3, s12
+; CHECK-NEXT:    vmov.8 q4[12], r3
+; CHECK-NEXT:    vmov r3, s13
+; CHECK-NEXT:    vmov.8 q4[13], r3
+; CHECK-NEXT:    vmov r3, s14
+; CHECK-NEXT:    vmov r2, s15
+; CHECK-NEXT:    vmov.8 q4[14], r3
+; CHECK-NEXT:    vldrb.u8 q0, [r1]
+; CHECK-NEXT:    vmov.8 q4[15], r2
+; CHECK-NEXT:    vstrb.8 q4, [r0, q0]
+; CHECK-NEXT:    vpop {d8, d9}
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <16 x i8>, <16 x i8>* %offptr, align 1
   %offs.zext = zext <16 x i8> %offs to <16 x i32>
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <16 x i32> %offs.zext
   %input.trunc = trunc <16 x i32> %input to <16 x i8>
   call void @llvm.masked.scatter.v16i8.v16p0i8(<16 x i8> %input.trunc, <16 x i8*> %byte_ptrs, i32 1, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; Expand
 define arm_aapcs_vfpcc void @trunc_unsigned_unscaled_i16_i8(i8* %base, <16 x i8>* %offptr, <16 x i16> %input) {
 ; CHECK-LABEL: trunc_unsigned_unscaled_i16_i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .vsave {d8, d9, d10, d11}
-; CHECK-NEXT:    vpush {d8, d9, d10, d11}
-; CHECK-NEXT:    vldrb.u32 q5, [r1]
-; CHECK-NEXT:    vldrb.u32 q2, [r1, #12]
-; CHECK-NEXT:    vldrb.u32 q3, [r1, #8]
-; CHECK-NEXT:    vldrb.u32 q4, [r1, #4]
-; CHECK-NEXT:    vadd.i32 q5, q5, r0
-; CHECK-NEXT:    vadd.i32 q2, q2, r0
-; CHECK-NEXT:    vadd.i32 q3, q3, r0
-; CHECK-NEXT:    vadd.i32 q4, q4, r0
-; CHECK-NEXT:    vmov r0, s20
-; CHECK-NEXT:    vmov.u16 r1, q0[0]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s21
-; CHECK-NEXT:    vmov.u16 r1, q0[1]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s22
-; CHECK-NEXT:    vmov.u16 r1, q0[2]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s23
-; CHECK-NEXT:    vmov.u16 r1, q0[3]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s16
-; CHECK-NEXT:    vmov.u16 r1, q0[4]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s17
-; CHECK-NEXT:    vmov.u16 r1, q0[5]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s18
-; CHECK-NEXT:    vmov.u16 r1, q0[6]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s19
-; CHECK-NEXT:    vmov.u16 r1, q0[7]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s12
-; CHECK-NEXT:    vmov.u16 r1, q1[0]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s13
-; CHECK-NEXT:    vmov.u16 r1, q1[1]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s14
-; CHECK-NEXT:    vmov.u16 r1, q1[2]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s15
-; CHECK-NEXT:    vmov.u16 r1, q1[3]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vmov.u16 r1, q1[4]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s9
-; CHECK-NEXT:    vmov.u16 r1, q1[5]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s10
-; CHECK-NEXT:    vmov.u16 r1, q1[6]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vmov r0, s11
-; CHECK-NEXT:    vmov.u16 r1, q1[7]
-; CHECK-NEXT:    strb r1, [r0]
-; CHECK-NEXT:    vpop {d8, d9, d10, d11}
+; CHECK-NEXT:    vmov.u16 r3, q0[0]
+; CHECK-NEXT:    vmov.u16 r2, q1[7]
+; CHECK-NEXT:    vmov.8 q2[0], r3
+; CHECK-NEXT:    vmov.u16 r3, q0[1]
+; CHECK-NEXT:    vmov.8 q2[1], r3
+; CHECK-NEXT:    vmov.u16 r3, q0[2]
+; CHECK-NEXT:    vmov.8 q2[2], r3
+; CHECK-NEXT:    vmov.u16 r3, q0[3]
+; CHECK-NEXT:    vmov.8 q2[3], r3
+; CHECK-NEXT:    vmov.u16 r3, q0[4]
+; CHECK-NEXT:    vmov.8 q2[4], r3
+; CHECK-NEXT:    vmov.u16 r3, q0[5]
+; CHECK-NEXT:    vmov.8 q2[5], r3
+; CHECK-NEXT:    vmov.u16 r3, q0[6]
+; CHECK-NEXT:    vmov.8 q2[6], r3
+; CHECK-NEXT:    vmov.u16 r3, q0[7]
+; CHECK-NEXT:    vmov.8 q2[7], r3
+; CHECK-NEXT:    vmov.u16 r3, q1[0]
+; CHECK-NEXT:    vmov.8 q2[8], r3
+; CHECK-NEXT:    vmov.u16 r3, q1[1]
+; CHECK-NEXT:    vmov.8 q2[9], r3
+; CHECK-NEXT:    vmov.u16 r3, q1[2]
+; CHECK-NEXT:    vmov.8 q2[10], r3
+; CHECK-NEXT:    vmov.u16 r3, q1[3]
+; CHECK-NEXT:    vmov.8 q2[11], r3
+; CHECK-NEXT:    vmov.u16 r3, q1[4]
+; CHECK-NEXT:    vmov.8 q2[12], r3
+; CHECK-NEXT:    vmov.u16 r3, q1[5]
+; CHECK-NEXT:    vmov.8 q2[13], r3
+; CHECK-NEXT:    vmov.u16 r3, q1[6]
+; CHECK-NEXT:    vmov.8 q2[14], r3
+; CHECK-NEXT:    vldrb.u8 q0, [r1]
+; CHECK-NEXT:    vmov.8 q2[15], r2
+; CHECK-NEXT:    vstrb.8 q2, [r0, q0]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <16 x i8>, <16 x i8>* %offptr, align 1
   %offs.zext = zext <16 x i8> %offs to <16 x i32>
   %byte_ptrs = getelementptr inbounds i8, i8* %base, <16 x i32> %offs.zext
   %input.trunc = trunc <16 x i16> %input to <16 x i8>
   call void @llvm.masked.scatter.v16i8.v16p0i8(<16 x i8> %input.trunc, <16 x i8*> %byte_ptrs, i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 
 declare void @llvm.masked.scatter.v2i8.v2p0i8(<2 x i8>, <2 x i8*>, i32, <2 x i1>)
 declare void @llvm.masked.scatter.v8i8.v8p0i8(<8 x i8>, <8 x i8*>, i32, <8 x i1>)
 declare void @llvm.masked.scatter.v16i8.v16p0i8(<16 x i8>, <16 x i8*>, i32, <16 x i1>)

llvm/test/CodeGen/Thumb2/mve-scatter-ptrs.ll

entry:
   ret void
 }
 
 ; VSTRW.32 Qd, [offs, 0]
 define arm_aapcs_vfpcc void @ptr_v4i32(<4 x i32> %v, <4 x i32*>* %offptr) {
 ; CHECK-LABEL: ptr_v4i32:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrw.u32 q1, [r0]
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    str r1, [r0]
+; CHECK-NEXT:    vstrw.32 q0, [q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i32*>, <4 x i32*>* %offptr, align 4
   call void @llvm.masked.scatter.v4i32.v4p0i32(<4 x i32> %v, <4 x i32*> %offs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; Expand
entry:
   ret void
 }
 
 ; VSTRW.32 Qd, [offs, 0]
 define arm_aapcs_vfpcc void @ptr_v4f32(<4 x float> %v, <4 x float*>* %offptr) {
 ; CHECK-LABEL: ptr_v4f32:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vldrw.u32 q1, [r0]
-; CHECK-NEXT:    vmov r3, s4
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s6
-; CHECK-NEXT:    vmov r2, s5
-; CHECK-NEXT:    vstr s0, [r3]
-; CHECK-NEXT:    vstr s1, [r2]
-; CHECK-NEXT:    vstr s2, [r1]
-; CHECK-NEXT:    vstr s3, [r0]
+; CHECK-NEXT:    vstrw.32 q0, [q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x float*>, <4 x float*>* %offptr, align 4
   call void @llvm.masked.scatter.v4f32.v4p0f32(<4 x float> %v, <4 x float*> %offs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 ; Expand
entry:
   ret void
 }
 
 ; loops
 
 define void @foo_ptr_p_int32_t(i32* %dest, i32** %src, i32 %n) {
 ; CHECK-LABEL: foo_ptr_p_int32_t:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .save {r4, lr}
-; CHECK-NEXT:    push {r4, lr}
-; CHECK-NEXT:    .pad #4
-; CHECK-NEXT:    sub sp, #4
 ; CHECK-NEXT:    bic r3, r2, #15
 ; CHECK-NEXT:    cmp r3, #1
-; CHECK-NEXT:    blt .LBB16_3
-; CHECK-NEXT:  @ %bb.1: @ %vector.body.preheader
-; CHECK-NEXT:    sub.w r12, r1, #16
-; CHECK-NEXT:  .LBB16_2: @ %vector.body
+; CHECK-NEXT:    it lt
+; CHECK-NEXT:    bxlt lr
+; CHECK-NEXT:  .LBB16_1: @ %vector.body
 ; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
-; CHECK-NEXT:    vldrw.u32 q0, [r12, #16]
-; CHECK-NEXT:    vpt.i32 ne, q0, zr
-; CHECK-NEXT:    vldrwt.u32 q1, [r0]
-; CHECK-NEXT:    vmrs r4, p0
-; CHECK-NEXT:    add.w r12, r12, #16
-; CHECK-NEXT:    adds r0, #16
-; CHECK-NEXT:    and r3, r4, #1
-; CHECK-NEXT:    ubfx r1, r4, #4, #1
-; CHECK-NEXT:    rsb.w lr, r3, #0
-; CHECK-NEXT:    movs r3, #0
-; CHECK-NEXT:    rsbs r1, r1, #0
-; CHECK-NEXT:    bfi r3, lr, #0, #1
-; CHECK-NEXT:    bfi r3, r1, #1, #1
-; CHECK-NEXT:    ubfx r1, r4, #8, #1
-; CHECK-NEXT:    rsbs r1, r1, #0
-; CHECK-NEXT:    bfi r3, r1, #2, #1
-; CHECK-NEXT:    ubfx r1, r4, #12, #1
-; CHECK-NEXT:    rsbs r1, r1, #0
-; CHECK-NEXT:    bfi r3, r1, #3, #1
-; CHECK-NEXT:    lsls r1, r3, #31
-; CHECK-NEXT:    ittt ne
-; CHECK-NEXT:    vmovne r1, s0
-; CHECK-NEXT:    vmovne r4, s4
-; CHECK-NEXT:    strne r4, [r1]
-; CHECK-NEXT:    lsls r1, r3, #30
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi r1, s1
-; CHECK-NEXT:    vmovmi r4, s5
-; CHECK-NEXT:    strmi r4, [r1]
-; CHECK-NEXT:    lsls r1, r3, #29
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi r1, s2
-; CHECK-NEXT:    vmovmi r4, s6
-; CHECK-NEXT:    strmi r4, [r1]
-; CHECK-NEXT:    lsls r1, r3, #28
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi r1, s3
-; CHECK-NEXT:    vmovmi r3, s7
-; CHECK-NEXT:    strmi r3, [r1]
+; CHECK-NEXT:    vldrw.u32 q0, [r1], #16
 ; CHECK-NEXT:    subs r2, #4
-; CHECK-NEXT:    bne .LBB16_2
-; CHECK-NEXT:  .LBB16_3: @ %for.end
-; CHECK-NEXT:    add sp, #4
-; CHECK-NEXT:    pop {r4, pc}
+; CHECK-NEXT:    vptt.i32 ne, q0, zr
+; CHECK-NEXT:    vldrwt.u32 q1, [r0], #16
+; CHECK-NEXT:    vstrwt.32 q1, [q0]
+; CHECK-NEXT:    bne .LBB16_1
+; CHECK-NEXT:  @ %bb.2: @ %for.end
+; CHECK-NEXT:    bx lr
 entry:
   %and = and i32 %n, -16
   %cmp11 = icmp sgt i32 %and, 0
   br i1 %cmp11, label %vector.body, label %for.end
 
 vector.body:                                      ; preds = %entry, %vector.body
   %index = phi i32 [ 0, %entry ], [ %index.next, %vector.body ]
   %0 = getelementptr inbounds i32*, i32** %src, i32 %index

 
 for.end:                                          ; preds = %vector.body, %entry
   ret void
 }
 
 define void @foo_ptr_p_float(float* %dest, float** %src, i32 %n) {
 ; CHECK-LABEL: foo_ptr_p_float:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .save {r4, lr}
-; CHECK-NEXT:    push {r4, lr}
-; CHECK-NEXT:    .pad #4
-; CHECK-NEXT:    sub sp, #4
 ; CHECK-NEXT:    bic r3, r2, #15
 ; CHECK-NEXT:    cmp r3, #1
-; CHECK-NEXT:    blt .LBB17_3
-; CHECK-NEXT:  @ %bb.1: @ %vector.body.preheader
-; CHECK-NEXT:    sub.w r12, r1, #16
-; CHECK-NEXT:  .LBB17_2: @ %vector.body
+; CHECK-NEXT:    it lt
+; CHECK-NEXT:    bxlt lr
+; CHECK-NEXT:  .LBB17_1: @ %vector.body
 ; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
-; CHECK-NEXT:    vldrw.u32 q0, [r12, #16]
-; CHECK-NEXT:    vpt.i32 ne, q0, zr
-; CHECK-NEXT:    vldrwt.u32 q1, [r0]
-; CHECK-NEXT:    vmrs r4, p0
-; CHECK-NEXT:    add.w r12, r12, #16
-; CHECK-NEXT:    adds r0, #16
-; CHECK-NEXT:    and r3, r4, #1
-; CHECK-NEXT:    ubfx r1, r4, #4, #1
-; CHECK-NEXT:    rsb.w lr, r3, #0
-; CHECK-NEXT:    movs r3, #0
-; CHECK-NEXT:    rsbs r1, r1, #0
-; CHECK-NEXT:    bfi r3, lr, #0, #1
-; CHECK-NEXT:    bfi r3, r1, #1, #1
-; CHECK-NEXT:    ubfx r1, r4, #8, #1
-; CHECK-NEXT:    rsbs r1, r1, #0
-; CHECK-NEXT:    bfi r3, r1, #2, #1
-; CHECK-NEXT:    ubfx r1, r4, #12, #1
-; CHECK-NEXT:    rsbs r1, r1, #0
-; CHECK-NEXT:    bfi r3, r1, #3, #1
-; CHECK-NEXT:    lsls r1, r3, #31
-; CHECK-NEXT:    ittt ne
-; CHECK-NEXT:    vmovne r1, s0
-; CHECK-NEXT:    vmovne r4, s4
-; CHECK-NEXT:    strne r4, [r1]
-; CHECK-NEXT:    lsls r1, r3, #30
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi r1, s1
-; CHECK-NEXT:    vmovmi r4, s5
-; CHECK-NEXT:    strmi r4, [r1]
-; CHECK-NEXT:    lsls r1, r3, #29
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi r1, s2
-; CHECK-NEXT:    vmovmi r4, s6
-; CHECK-NEXT:    strmi r4, [r1]
-; CHECK-NEXT:    lsls r1, r3, #28
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi r1, s3
-; CHECK-NEXT:    vmovmi r3, s7
-; CHECK-NEXT:    strmi r3, [r1]
+; CHECK-NEXT:    vldrw.u32 q0, [r1], #16
 ; CHECK-NEXT:    subs r2, #4
-; CHECK-NEXT:    bne .LBB17_2
-; CHECK-NEXT:  .LBB17_3: @ %for.end
-; CHECK-NEXT:    add sp, #4
-; CHECK-NEXT:    pop {r4, pc}
+; CHECK-NEXT:    vptt.i32 ne, q0, zr
+; CHECK-NEXT:    vldrwt.u32 q1, [r0], #16
+; CHECK-NEXT:    vstrwt.32 q1, [q0]
+; CHECK-NEXT:    bne .LBB17_1
+; CHECK-NEXT:  @ %bb.2: @ %for.end
+; CHECK-NEXT:    bx lr
 entry:
   %and = and i32 %n, -16
   %cmp11 = icmp sgt i32 %and, 0
   br i1 %cmp11, label %vector.body, label %for.end
 
 vector.body:                                      ; preds = %entry, %vector.body
   %index = phi i32 [ 0, %entry ], [ %index.next, %vector.body ]
   %0 = getelementptr inbounds float*, float** %src, i32 %index
for.end:                                          ; preds = %vector.body, %entry
   ret void
 }
 
 ; VLSTW.u32 Qd, [P, 4]
 define arm_aapcs_vfpcc void @qi4(<4 x i32> %v, <4 x i32*> %p) {
 ; CHECK-LABEL: qi4:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    vmov.i32 q2, #0x10
-; CHECK-NEXT:    vmov r1, s0
 ; CHECK-NEXT:    vadd.i32 q1, q1, q2
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov r1, s1
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    str r1, [r0]
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vmov r1, s3
-; CHECK-NEXT:    str r1, [r0]
+; CHECK-NEXT:    vstrw.32 q0, [q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %g = getelementptr inbounds i32, <4 x i32*> %p, i32 4
   call void @llvm.masked.scatter.v4i32.v4p0i32(<4 x i32> %v, <4 x i32*> %g, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)
   ret void
 }
 
 declare void @llvm.masked.scatter.v2i16.v2p0i16(<2 x i16>, <2 x i16*>, i32, <2 x i1>)