[AArch64][SVE] Add support for fixed length MSCATTER/MGATHER

[bsmith]

Summary

Since gather lowering can now lower to nodes that may need expansion via
the vector legalizer, do MGATHER lowering via vector legalizer.

Additionally, as part of adding passthru support for fixed typed
gathers, fix passthru support for scalable types.

Depends on D104910

[paulwalker-arm]

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

17808–17809

I guess you've moved these lowering functions because of the placement of the fixed length utility functions. I guess you could move the utility functions but I'm wondering if you can just add forward declarations to the top somewhere?

[bsmith]

• Restore file position of LowerMGATHER/LowerMSCATTER and instead add forward declarations of SVE helpers.
• Fix clang format issues

[peterwaller-arm]

Mostly nits -- looks like it needs a rebase.

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

4223

Is it intended to address this in the near future?

Nit: I initially read 'the code should be moved, once <condition>', but I think you mean, 'the code should be moved, so that it executes after <condition>'.

4264

Pondering: Why are these two isFixedLengthVector queries needed? I read: you have a fixed-vector operation, and it has scalable operands? Can that arise? How so?

4273

Nit: } else if and remove a level of indent.

4280

I wonder if the undef-or-zero logic can be cleanly combined into a function with MLOAD. But maybe not. I wouldn't wonder for too long.

Thought (1): is the PassThru value actually needed for the Ops, or could it use a constant there? My thinking is that the underlying operation is invariant of its PassThru operand (hence the need for the select).

If (1) is right, then could you use an empty SDValue, which gets assigned to the thing to be 'selected', and the logic becomes if (PassThru) { insert select }.

I think this is a matter of taste so feel at liberty to ignore the suggestion. If you go for it please update MLOAD.

llvm/lib/Target/AArch64/AArch64TargetTransformInfo.h

246

This was changed in D101834 so it looks like you need a rebase since the MLOAD work.

249

Nit: Potentially confusing comment about the mechanism of the following block, because it's written in actively: "this block avoids scalarization", but actually the only thing the block can do is return false, which causes scalarization. Suggested form: "Fixed-length vectors do not use masked gather/scatter unless ..."

[peterwaller-arm]

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

4223

Is it intended to address this in the near future?

I've thought about/discussed this am happy with having the FIXME here.

[bsmith]

• Simplify passthru for gathers
• Add tests for type legalization
• Clarify some comments

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

4264

selectGatherScatterAddrMode can swap the BasePtr/Index operands, one of which is a scalar, the other a vector. Hence here we don't know which is which, so we check them both.

4280

This makes sense for MGATHER since the passthru value is totally ignored, however for masked_load we don't have target specific isd nodes, hence the passthru value makes it down to selection.

llvm/lib/Target/AArch64/AArch64TargetTransformInfo.h

246

It was isLegalMaskedLoadStore that was changed in D101834, not this function.

[david-arm]

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

1200

Does the comment need updating to reflect gathers/scatters too?

1201

I realise this has nothing to do with your change, but it looks really odd that we have a nested loop here where we override the VT from the outer loop. Essentially, we're registering the exact same custom operation 4 times in a row! Might be worth fixing at some point.

1202

Not sure if it matters, but isLegalMaskedGatherScatter explicitly excludes MVT::v1f64 as it has < 2 elements.

4263

Is it worth asserting here that Subtarget->useSVEForFixedLengthVectors() returns true?

16185

Is there to do this in a separate patch? I think it would be better if possible - I imagine it works as a standalone change where it tidies up the masked load/store tests.

[bsmith]

• Split out SETCC_MERGE_ZERO dag combine into another patch
• Fixed incorrectly nested loops during setOperationAction's
• Add extra asserts in Lowering functions
• Minor comment corrections
• Minor test changes due to other commits

[david-arm]

Could you fix all the formatting issues before submitting? Thanks!

llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp

#include "llvm/IR/ConstrainedOps.def"
   case ISD::SADDSAT:
   case ISD::UADDSAT:
   case ISD::SSUBSAT:
   case ISD::USUBSAT:
   case ISD::SSHLSAT:
   case ISD::USHLSAT:
   case ISD::FP_TO_SINT_SAT:
   case ISD::FP_TO_UINT_SAT:
+  case ISD::MGATHER:
     Action = TLI.getOperationAction(Node->getOpcode(), Node->getValueType(0));
     break;
   case ISD::SMULFIX:
   case ISD::SMULFIXSAT:
   case ISD::UMULFIX:
   case ISD::UMULFIXSAT:
   case ISD::SDIVFIX:
   case ISD::SDIVFIXSAT:

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

for (auto VT : {MVT::nxv16i1, MVT::nxv8i1, MVT::nxv4i1, MVT::nxv2i1}) {
       setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Custom);
 
       // There are no legal MVT::nxv16f## based types.
       if (VT != MVT::nxv16i1) {
         setOperationAction(ISD::SINT_TO_FP, VT, Custom);
         setOperationAction(ISD::UINT_TO_FP, VT, Custom);
       }
 
       // NEON doesn't support masked loads or stores, but SVE does
       for (auto VT :
            {MVT::v4f16, MVT::v8f16, MVT::v2f32, MVT::v4f32, MVT::v1f64,
             MVT::v2f64, MVT::v8i8, MVT::v16i8, MVT::v4i16, MVT::v8i16,
             MVT::v2i32, MVT::v4i32, MVT::v1i64, MVT::v2i64}) {
         setOperationAction(ISD::MLOAD, VT, Custom);
         setOperationAction(ISD::MSTORE, VT, Custom);
+        setOperationAction(ISD::MGATHER, VT, Custom);
+        setOperationAction(ISD::MSCATTER, VT, Custom);
       }
     }
 
     for (MVT VT : MVT::fp_scalable_vector_valuetypes()) {
       for (MVT InnerVT : MVT::fp_scalable_vector_valuetypes()) {
         // Avoid marking truncating FP stores as legal to prevent the
         // DAGCombiner from creating unsupported truncating stores.
         setTruncStoreAction(VT, InnerVT, Expand);
void AArch64TargetLowering::addTypeForFixedLengthSVE(MVT VT) {
   setOperationAction(ISD::FP_TO_UINT, VT, Custom);
   setOperationAction(ISD::FRINT, VT, Custom);
   setOperationAction(ISD::FROUND, VT, Custom);
   setOperationAction(ISD::FROUNDEVEN, VT, Custom);
   setOperationAction(ISD::FSQRT, VT, Custom);
   setOperationAction(ISD::FSUB, VT, Custom);
   setOperationAction(ISD::FTRUNC, VT, Custom);
   setOperationAction(ISD::LOAD, VT, Custom);
+  setOperationAction(ISD::MGATHER, VT, Custom);
   setOperationAction(ISD::MLOAD, VT, Custom);
+  setOperationAction(ISD::MSCATTER, VT, Custom);
   setOperationAction(ISD::MSTORE, VT, Custom);
   setOperationAction(ISD::MUL, VT, Custom);
   setOperationAction(ISD::MULHS, VT, Custom);
   setOperationAction(ISD::MULHU, VT, Custom);
   setOperationAction(ISD::OR, VT, Custom);
   setOperationAction(ISD::SDIV, VT, Custom);
   setOperationAction(ISD::SELECT, VT, Custom);
   setOperationAction(ISD::SETCC, VT, Custom);

 //    getelementptr nullptr, <vscale x N x T> (splat(#x)) + %indices)
 // -> getelementptr #x, <vscale x N x T> %indices
 void selectGatherScatterAddrMode(SDValue &BasePtr, SDValue &Index, EVT MemVT,
                                  unsigned &Opcode, bool IsGather,
                                  SelectionDAG &DAG) {
   if (!isNullConstant(BasePtr))
     return;
 
+  // FIXME: This will not match for fixed vector type codegen as the nodes in
+  // question will have fixed<->scalable conversions around them. This should be
+  // moved to a DAG combine or complex pattern, after all of the fixed vector
+  // insert and extracts have been removed.
   ConstantSDNode *Offset = nullptr;
   if (Index.getOpcode() == ISD::ADD)
     if (auto SplatVal = DAG.getSplatValue(Index.getOperand(1))) {
       if (isa<ConstantSDNode>(SplatVal))
         Offset = cast<ConstantSDNode>(SplatVal);
       else {
         BasePtr = SplatVal;
         Index = Index->getOperand(0);
void selectGatherScatterAddrMode(SDValue &BasePtr, SDValue &Index, EVT MemVT,
   }
 
   // Immediate is in range
   Opcode = NewOp;
   BasePtr = Index->getOperand(0);
   Index = ConstOffset;
 }
 
-SDValue AArch64TargetLowering::LowerMGATHER(SDValue Op,
-                                            SelectionDAG &DAG) const {
-  SDLoc DL(Op);
-  MaskedGatherSDNode *MGT = cast<MaskedGatherSDNode>(Op);
-  assert(MGT && "Can only custom lower gather load nodes");
-
-  SDValue Index = MGT->getIndex();
-  SDValue Chain = MGT->getChain();
-  SDValue PassThru = MGT->getPassThru();
-  SDValue Mask = MGT->getMask();
-  SDValue BasePtr = MGT->getBasePtr();
-  ISD::LoadExtType ExtTy = MGT->getExtensionType();
-
-  ISD::MemIndexType IndexType = MGT->getIndexType();
-  bool IsScaled =
-      IndexType == ISD::SIGNED_SCALED || IndexType == ISD::UNSIGNED_SCALED;
-  bool IsSigned =
-      IndexType == ISD::SIGNED_SCALED || IndexType == ISD::SIGNED_UNSCALED;
-  bool IdxNeedsExtend =
-      getGatherScatterIndexIsExtended(Index) ||
-      Index.getSimpleValueType().getVectorElementType() == MVT::i32;
-  bool ResNeedsSignExtend = ExtTy == ISD::EXTLOAD || ExtTy == ISD::SEXTLOAD;
-
-  EVT VT = PassThru.getSimpleValueType();
-  EVT MemVT = MGT->getMemoryVT();
-  SDValue InputVT = DAG.getValueType(MemVT);
-
-  if (VT.getVectorElementType() == MVT::bf16 &&
-      !static_cast<const AArch64Subtarget &>(DAG.getSubtarget()).hasBF16())
-    return SDValue();
-
-  // Handle FP data by using an integer gather and casting the result.
-  if (VT.isFloatingPoint()) {
-    EVT PassThruVT = getPackedSVEVectorVT(VT.getVectorElementCount());
-    PassThru = getSVESafeBitCast(PassThruVT, PassThru, DAG);
-    InputVT = DAG.getValueType(MemVT.changeVectorElementTypeToInteger());
-  }
-
-  SDVTList VTs = DAG.getVTList(PassThru.getSimpleValueType(), MVT::Other);
-
-  if (getGatherScatterIndexIsExtended(Index))
-    Index = Index.getOperand(0);
-
-  unsigned Opcode = getGatherVecOpcode(IsScaled, IsSigned, IdxNeedsExtend);
-  selectGatherScatterAddrMode(BasePtr, Index, MemVT, Opcode,
-                              /*isGather=*/true, DAG);
-
-  if (ResNeedsSignExtend)
-    Opcode = getSignExtendedGatherOpcode(Opcode);
-
-  SDValue Ops[] = {Chain, Mask, BasePtr, Index, InputVT, PassThru};
-  SDValue Gather = DAG.getNode(Opcode, DL, VTs, Ops);
-
-  if (VT.isFloatingPoint()) {
-    SDValue Cast = getSVESafeBitCast(VT, Gather, DAG);
-    return DAG.getMergeValues({Cast, Gather.getValue(1)}, DL);
-  }
-
-  return Gather;
-}
-
-SDValue AArch64TargetLowering::LowerMSCATTER(SDValue Op,
-                                             SelectionDAG &DAG) const {
-  SDLoc DL(Op);
-  MaskedScatterSDNode *MSC = cast<MaskedScatterSDNode>(Op);
-  assert(MSC && "Can only custom lower scatter store nodes");
-
-  SDValue Index = MSC->getIndex();
-  SDValue Chain = MSC->getChain();
-  SDValue StoreVal = MSC->getValue();
-  SDValue Mask = MSC->getMask();
-  SDValue BasePtr = MSC->getBasePtr();
-
-  ISD::MemIndexType IndexType = MSC->getIndexType();
-  bool IsScaled =
-      IndexType == ISD::SIGNED_SCALED || IndexType == ISD::UNSIGNED_SCALED;
-  bool IsSigned =
-      IndexType == ISD::SIGNED_SCALED || IndexType == ISD::SIGNED_UNSCALED;
-  bool NeedsExtend =
-      getGatherScatterIndexIsExtended(Index) ||
-      Index.getSimpleValueType().getVectorElementType() == MVT::i32;
-
-  EVT VT = StoreVal.getSimpleValueType();
-  SDVTList VTs = DAG.getVTList(MVT::Other);
-  EVT MemVT = MSC->getMemoryVT();
-  SDValue InputVT = DAG.getValueType(MemVT);
-
-  if (VT.getVectorElementType() == MVT::bf16 &&
-      !static_cast<const AArch64Subtarget &>(DAG.getSubtarget()).hasBF16())
-    return SDValue();
-
-  // Handle FP data by casting the data so an integer scatter can be used.
-  if (VT.isFloatingPoint()) {
-    EVT StoreValVT = getPackedSVEVectorVT(VT.getVectorElementCount());
-    StoreVal = getSVESafeBitCast(StoreValVT, StoreVal, DAG);
-    InputVT = DAG.getValueType(MemVT.changeVectorElementTypeToInteger());
-  }
-
-  if (getGatherScatterIndexIsExtended(Index))
-    Index = Index.getOperand(0);
-
-  unsigned Opcode = getScatterVecOpcode(IsScaled, IsSigned, NeedsExtend);
-  selectGatherScatterAddrMode(BasePtr, Index, MemVT, Opcode,
-                              /*isGather=*/false, DAG);
-
-  SDValue Ops[] = {Chain, StoreVal, Mask, BasePtr, Index, InputVT};
-  return DAG.getNode(Opcode, DL, VTs, Ops);
-}
-
 // Custom lower trunc store for v4i8 vectors, since it is promoted to v4i16.
 static SDValue LowerTruncateVectorStore(SDLoc DL, StoreSDNode *ST,
                                         EVT VT, EVT MemVT,
                                         SelectionDAG &DAG) {
   assert(VT.isVector() && "VT should be a vector type");
   assert(MemVT == MVT::v4i8 && VT == MVT::v4i16);
 
   SDValue Value = ST->getValue();
 
   // It first extend the promoted v4i16 to v8i16, truncate to v8i8, and extract
   // the word lane which represent the v4i8 subvector.  It optimizes the store
   // to:
   //
   //   xtn  v0.8b, v0.8h
   //   str  s0, [x0]
 
   SDValue Undef = DAG.getUNDEF(MVT::i16);
   SDValue UndefVec = DAG.getBuildVector(MVT::v4i16, DL,
                                         {Undef, Undef, Undef, Undef});
 
   SDValue TruncExt = DAG.getNode(ISD::CONCAT_VECTORS, DL, MVT::v8i16,
                                  Value, UndefVec);
   SDValue Trunc = DAG.getNode(ISD::TRUNCATE, DL, MVT::v8i8, TruncExt);
 
   Trunc = DAG.getNode(ISD::BITCAST, DL, MVT::v2i32, Trunc);
   SDValue ExtractTrunc = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, MVT::i32,
SDValue CSEL =
                     LHS.getOperand(1), DAG.getConstant(NewCond, DL, MVT::i32),
                     LHS.getOperand(3));
     return DAG.getZExtOrTrunc(CSEL, DL, N->getValueType(0));
   }
 
   return SDValue();
 }
 
+static SDValue performSetccMergeZeroCombine(SDNode *N, SelectionDAG &DAG) {
+  assert(N->getOpcode() == AArch64ISD::SETCC_MERGE_ZERO &&
+         "Unexpected opcode!");
+
+  SDValue Pred = N->getOperand(0);
+  SDValue LHS = N->getOperand(1);
+  SDValue RHS = N->getOperand(2);
+  ISD::CondCode Cond = cast<CondCodeSDNode>(N->getOperand(3))->get();
+
+  // setcc_merge_zero pred (sign_extend (setcc_merge_zero ... pred ...)), 0, ne
+  //    => inner setcc_merge_zero
+  if (Cond == ISD::SETNE && isZerosVector(RHS.getNode()) &&
+      LHS->getOpcode() == ISD::SIGN_EXTEND &&
+      LHS->getOperand(0)->getValueType(0) == N->getValueType(0) &&
+      LHS->getOperand(0)->getOpcode() == AArch64ISD::SETCC_MERGE_ZERO &&
+      LHS->getOperand(0)->getOperand(0) == Pred)
+    return LHS->getOperand(0);
+
+  return SDValue();
+}
+
 // Optimize some simple tbz/tbnz cases.  Returns the new operand and bit to test
 // as well as whether the test should be inverted.  This code is required to
 // catch these cases (as opposed to standard dag combines) because
 // AArch64ISD::TBZ is matched during legalization.
 static SDValue getTestBitOperand(SDValue Op, unsigned &Bit, bool &Invert,
                                  SelectionDAG &DAG) {
 
   if (!Op->hasOneUse())
SDValue AArch64TargetLowering::PerformDAGCombine(SDNode *N,
   case AArch64ISD::DUP:
     return performPostLD1Combine(N, DCI, false);
   case AArch64ISD::NVCAST:
     return performNVCASTCombine(N);
   case AArch64ISD::SPLICE:
     return performSpliceCombine(N, DAG);
   case AArch64ISD::UZP1:
     return performUzpCombine(N, DAG);
+  case AArch64ISD::SETCC_MERGE_ZERO:
+    return performSetccMergeZeroCombine(N, DAG);
   case AArch64ISD::GLD1_MERGE_ZERO:
   case AArch64ISD::GLD1_SCALED_MERGE_ZERO:
   case AArch64ISD::GLD1_UXTW_MERGE_ZERO:
   case AArch64ISD::GLD1_SXTW_MERGE_ZERO:
   case AArch64ISD::GLD1_UXTW_SCALED_MERGE_ZERO:
   case AArch64ISD::GLD1_SXTW_SCALED_MERGE_ZERO:
   case AArch64ISD::GLD1_IMM_MERGE_ZERO:
   case AArch64ISD::GLD1S_MERGE_ZERO:
SDValue AArch64TargetLowering::LowerVECREDUCE_SEQ_FADD(SDValue ScalarOp,
 
   // Perform reduction.
   SDValue Rdx = DAG.getNode(AArch64ISD::FADDA_PRED, DL, ContainerVT,
                             Pg, AccOp, VecOp);
 
   return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, ResVT, Rdx, Zero);
 }
 
+SDValue AArch64TargetLowering::LowerMGATHER(SDValue Op,
+                                            SelectionDAG &DAG) const {
+  SDLoc DL(Op);
+  MaskedGatherSDNode *MGT = cast<MaskedGatherSDNode>(Op);
+  assert(MGT && "Can only custom lower gather load nodes");
+
+  bool IsFixedLength = MGT->getMemoryVT().isFixedLengthVector();
+
+  SDValue Index = MGT->getIndex();
+  SDValue Chain = MGT->getChain();
+  SDValue PassThru = MGT->getPassThru();
+  SDValue Mask = MGT->getMask();
+  SDValue BasePtr = MGT->getBasePtr();
+  ISD::LoadExtType ExtTy = MGT->getExtensionType();
+
+  ISD::MemIndexType IndexType = MGT->getIndexType();
+  bool IsScaled =
+      IndexType == ISD::SIGNED_SCALED || IndexType == ISD::UNSIGNED_SCALED;
+  bool IsSigned =
+      IndexType == ISD::SIGNED_SCALED || IndexType == ISD::SIGNED_UNSCALED;
+  bool IdxNeedsExtend =
+      getGatherScatterIndexIsExtended(Index) ||
+      Index.getSimpleValueType().getVectorElementType() == MVT::i32;
+  bool ResNeedsSignExtend = ExtTy == ISD::EXTLOAD || ExtTy == ISD::SEXTLOAD;
+
+  EVT VT = PassThru.getSimpleValueType();
+  EVT IndexVT = Index.getSimpleValueType();
+  EVT MemVT = MGT->getMemoryVT();
+  SDValue InputVT = DAG.getValueType(MemVT);
+
+  if (VT.getVectorElementType() == MVT::bf16 &&
+      !static_cast<const AArch64Subtarget &>(DAG.getSubtarget()).hasBF16())
+    return SDValue();
+
+  if (IsFixedLength) {
+    IndexVT = getContainerForFixedLengthVector(DAG, IndexVT);
+    MemVT = IndexVT.changeVectorElementType(MemVT.getVectorElementType());
+    InputVT = DAG.getValueType(MemVT.changeTypeToInteger());
+  }
+
+  bool IsPassThruZeroOrUndef = false;
+  if (PassThru->isUndef()) {
+    PassThru = DAG.getUNDEF(IndexVT);
+    IsPassThruZeroOrUndef = true;
+  } else {
+    if (isZerosVector(PassThru.getNode())) {
+     if (IndexVT.isInteger())
+        PassThru = DAG.getConstant(0, DL, IndexVT);
+      else
+        PassThru = DAG.getConstantFP(0, DL, IndexVT);
+      IsPassThruZeroOrUndef = true;
+    }
+  }
+
+  if (VT.isFloatingPoint() && !IsFixedLength) {
+    // Handle FP data by using an integer gather and casting the result.
+    EVT PassThruVT = getPackedSVEVectorVT(VT.getVectorElementCount());
+    PassThru = getSVESafeBitCast(PassThruVT, PassThru, DAG);
+    InputVT = DAG.getValueType(MemVT.changeVectorElementTypeToInteger());
+  }
+
+  SDVTList VTs = DAG.getVTList(IndexVT, MVT::Other);
+
+  if (getGatherScatterIndexIsExtended(Index))
+    Index = Index.getOperand(0);
+
+  unsigned Opcode = getGatherVecOpcode(IsScaled, IsSigned, IdxNeedsExtend);
+  selectGatherScatterAddrMode(BasePtr, Index, MemVT, Opcode,
+                              /*isGather=*/true, DAG);
+
+  if (ResNeedsSignExtend)
+    Opcode = getSignExtendedGatherOpcode(Opcode);
+
+  if (IsFixedLength) {
+    if (Index.getSimpleValueType().isFixedLengthVector())
+      Index = convertToScalableVector(DAG, IndexVT, Index);
+    if (BasePtr.getSimpleValueType().isFixedLengthVector())
+      BasePtr = convertToScalableVector(DAG, IndexVT, BasePtr);
+    Mask = convertFixedMaskToScalableVector(Mask, DAG);
+  }
+
+  SDValue Ops[] = {Chain, Mask, BasePtr, Index, InputVT, PassThru};
+  SDValue Result = DAG.getNode(Opcode, DL, VTs, Ops);
+  Chain = Result.getValue(1);
+
+  if (IsFixedLength) {
+    Result = convertFromScalableVector(
+        DAG, VT.changeVectorElementType(IndexVT.getVectorElementType()),
+        Result);
+    Result = DAG.getNode(ISD::TRUNCATE, DL, VT.changeTypeToInteger(), Result);
+    Result = DAG.getNode(ISD::BITCAST, DL, VT, Result);
+
+    if (!IsPassThruZeroOrUndef)
+      Result = DAG.getSelect(DL, VT, MGT->getMask(), Result, PassThru);
+  } else {
+    if (!IsPassThruZeroOrUndef)
+      Result = DAG.getSelect(DL, IndexVT, Mask, Result, PassThru);
+
+    if (VT.isFloatingPoint())
+      Result = getSVESafeBitCast(VT, Result, DAG);
+  }
+
+  return DAG.getMergeValues({Result, Chain}, DL);
+}
+
+SDValue AArch64TargetLowering::LowerMSCATTER(SDValue Op,
+                                             SelectionDAG &DAG) const {
+  SDLoc DL(Op);
+  MaskedScatterSDNode *MSC = cast<MaskedScatterSDNode>(Op);
+  assert(MSC && "Can only custom lower scatter store nodes");
+
+  bool IsFixedLength = MSC->getMemoryVT().isFixedLengthVector();
+
+  SDValue Index = MSC->getIndex();
+  SDValue Chain = MSC->getChain();
+  SDValue StoreVal = MSC->getValue();
+  SDValue Mask = MSC->getMask();
+  SDValue BasePtr = MSC->getBasePtr();
+
+  ISD::MemIndexType IndexType = MSC->getIndexType();
+  bool IsScaled =
+      IndexType == ISD::SIGNED_SCALED || IndexType == ISD::UNSIGNED_SCALED;
+  bool IsSigned =
+      IndexType == ISD::SIGNED_SCALED || IndexType == ISD::SIGNED_UNSCALED;
+  bool NeedsExtend =
+      getGatherScatterIndexIsExtended(Index) ||
+      Index.getSimpleValueType().getVectorElementType() == MVT::i32;
+
+  EVT VT = StoreVal.getSimpleValueType();
+  EVT IndexVT = Index.getSimpleValueType();
+  SDVTList VTs = DAG.getVTList(MVT::Other);
+  EVT MemVT = MSC->getMemoryVT();
+  SDValue InputVT = DAG.getValueType(MemVT);
+
+  if (VT.getVectorElementType() == MVT::bf16 &&
+      !static_cast<const AArch64Subtarget &>(DAG.getSubtarget()).hasBF16())
+    return SDValue();
+
+  if (IsFixedLength) {
+    IndexVT = getContainerForFixedLengthVector(DAG, IndexVT);
+    MemVT = IndexVT.changeVectorElementType(MemVT.getVectorElementType());
+    InputVT = DAG.getValueType(MemVT.changeTypeToInteger());
+
+    StoreVal =
+        DAG.getNode(ISD::BITCAST, DL, VT.changeTypeToInteger(), StoreVal);
+    StoreVal = DAG.getNode(
+        ISD::ANY_EXTEND, DL,
+        VT.changeVectorElementType(IndexVT.getVectorElementType()), StoreVal);
+    StoreVal = convertToScalableVector(DAG, IndexVT, StoreVal);
+  } else if (VT.isFloatingPoint()) {
+    // Handle FP data by casting the data so an integer scatter can be used.
+    EVT StoreValVT = getPackedSVEVectorVT(VT.getVectorElementCount());
+    StoreVal = getSVESafeBitCast(StoreValVT, StoreVal, DAG);
+    InputVT = DAG.getValueType(MemVT.changeVectorElementTypeToInteger());
+  }
+
+  if (getGatherScatterIndexIsExtended(Index))
+    Index = Index.getOperand(0);
+
+  unsigned Opcode = getScatterVecOpcode(IsScaled, IsSigned, NeedsExtend);
+  selectGatherScatterAddrMode(BasePtr, Index, MemVT, Opcode,
+                              /*isGather=*/false, DAG);
+
+  if (IsFixedLength) {
+    if (Index.getSimpleValueType().isFixedLengthVector())
+      Index = convertToScalableVector(DAG, IndexVT, Index);
+    if (BasePtr.getSimpleValueType().isFixedLengthVector())
+      BasePtr = convertToScalableVector(DAG, IndexVT, BasePtr);
+    Mask = convertFixedMaskToScalableVector(Mask, DAG);
+  }
+
+  SDValue Ops[] = {Chain, StoreVal, Mask, BasePtr, Index, InputVT};
+  return DAG.getNode(Opcode, DL, VTs, Ops);
+}
+
 SDValue AArch64TargetLowering::LowerPredReductionToSVE(SDValue ReduceOp,
                                                        SelectionDAG &DAG) const {
   SDLoc DL(ReduceOp);
   SDValue Op = ReduceOp.getOperand(0);
   EVT OpVT = Op.getValueType();
   EVT VT = ReduceOp.getValueType();
 
   if (!OpVT.isScalableVector() || OpVT.getVectorElementType() != MVT::i1)

llvm/lib/Target/AArch64/AArch64TargetTransformInfo.h

bool isLegalMaskedLoad(Type *DataType, Align Alignment) {
     return isLegalMaskedLoadStore(DataType, Alignment);
   }
 
   bool isLegalMaskedStore(Type *DataType, Align Alignment) {
     return isLegalMaskedLoadStore(DataType, Alignment);
   }
 
   bool isLegalMaskedGatherScatter(Type *DataType) const {
-    if (isa<FixedVectorType>(DataType) || !ST->hasSVE())
+    if (!ST->hasSVE())
       return false;
 
+    // For fixed vectors, avoid scalarization if using SVE for them.
+    auto *DataTypeFVTy = dyn_cast<FixedVectorType>(DataType);
+    if (DataTypeFVTy && (!ST->useSVEForFixedLengthVectors() ||
+                         DataTypeFVTy->getNumElements() < 2))
+      return false; // Fall back to scalarization of masked operations.
+
     return isLegalElementTypeForSVE(DataType->getScalarType());
   }
 
   bool isLegalMaskedGather(Type *DataType, Align Alignment) const {
     return isLegalMaskedGatherScatter(DataType);
   }
   bool isLegalMaskedScatter(Type *DataType, Align Alignment) const {
     return isLegalMaskedGatherScatter(DataType);

llvm/test/CodeGen/AArch64/sve-fixed-length-masked-gather.ll

+; RUN: llc -aarch64-sve-vector-bits-min=128  -asm-verbose=0 < %s | FileCheck %s -check-prefix=NO_SVE
+; RUN: llc -aarch64-sve-vector-bits-min=256  -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK
+; RUN: llc -aarch64-sve-vector-bits-min=384  -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK
+; RUN: llc -aarch64-sve-vector-bits-min=512  -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=640  -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=768  -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=896  -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=1024 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_1024,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=1152 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_1024,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=1280 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_1024,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=1408 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_1024,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=1536 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_1024,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=1664 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_1024,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=1792 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_1024,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=1920 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_1024,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=2048 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_2048,VBITS_GE_1024,VBITS_GE_512
+
+target triple = "aarch64-unknown-linux-gnu"
+
+; Don't use SVE when its registers are no bigger than NEON.
+; NO_SVE-NOT: ptrue
+
+;
+; LD1B
+;
+
+define void @masked_gather_v2i8(<2 x i8>* %a, <2 x i8*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v2i8:
+; CHECK: ldrb [[VALS_LO:w[0-9]+]], [x0]
+; CHECK-NEXT: ldrb [[VALS_HI:w[0-9]+]], [x0, #1]
+; CHECK-NEXT: ldr q[[PTRS:[0-9]+]], [x1]
+; CHECK-NEXT: ptrue [[PG0:p[0-9]+]].s, vl2
+; CHECK-NEXT: fmov s[[VALS:[0-9]+]], [[VALS_LO]]
+; CHECK-NEXT: mov v[[VALS]].s[1], [[VALS_HI]]
+; CHECK-NEXT: cmeq v[[CMP:[0-9]+]].2s, v[[VALS]].2s, #0
+; CHECK-NEXT: cmpne [[MASK:p[0-9]+]].s, [[PG0]]/z, z[[CMP]].s, #0
+; CHECK-NEXT: ld1sb { z[[RES:[0-9]+]].d }, [[MASK]]/z, [z[[PTRS]].d]
+; CHECK-NEXT: xtn v[[XTN:[0-9]+]].2s, v[[RES]].2d
+; CHECK-NEXT: mov [[RES_HI:w[0-9]+]], v[[XTN]].s[1]
+; CHECK-NEXT: fmov [[RES_LO:w[0-9]+]], s[[XTN]]
+; CHECK-NEXT: strb [[RES_LO]], [x0]
+; CHECK-NEXT: strb [[RES_HI]], [x0, #1]
+; CHECK-NEXT: ret
+  %cval = load <2 x i8>, <2 x i8>* %a
+  %ptrs = load <2 x i8*>, <2 x i8*>* %b
+  %mask = icmp eq <2 x i8> %cval, zeroinitializer
+  %vals = call <2 x i8> @llvm.masked.gather.v2i8(<2 x i8*> %ptrs, i32 8, <2 x i1> %mask, <2 x i8> undef)
+  store <2 x i8> %vals, <2 x i8>* %a
+  ret void
+}
+
+define void @masked_gather_v4i8(<4 x i8>* %a, <4 x i8*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v4i8:
+; CHECK: ldrb [[VALS_LO1:w[0-9]+]], [x0]
+; CHECK-NEXT: ldrb [[VALS_LO2:w[0-9]+]], [x0, #1]
+; CHECK-NEXT: ldrb [[VALS_HI1:w[0-9]+]], [x0, #2]
+; CHECK-NEXT: ldrb [[VALS_HI2:w[0-9]+]], [x0, #3]
+; CHECK-NEXT: fmov s[[VALS:[0-9]+]], [[VALS_LO1]]
+; CHECK-NEXT: ptrue [[PG0:p[0-9]+]].d, vl4
+; CHECK-NEXT: mov v[[VALS]].h[1], [[VALS_LO2]]
+; CHECK-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; CHECK-NEXT: mov v[[VALS]].h[2], [[VALS_HI1]]
+; CHECK-NEXT: mov v[[VALS]].h[3], [[VALS_HI2]]
+; CHECK-NEXT: cmeq v[[CMP:[0-9]+]].4h, v[[VALS]].4h, #0
+; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].h, vl4
+; CHECK-NEXT: cmpne [[MASK:p[0-9]+]].h, [[PG1]]/z, z[[CMP]].h, #0
+; CHECK-NEXT: ld1sb { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; CHECK-NEXT: uzp1 [[UZP1:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; CHECK-NEXT: uzp1 z[[UZP2:[0-9]+]].h, [[UZP1]].h, [[UZP1]].h
+; CHECK-NEXT: uzp1 v[[UZP3:[0-9]+]].8b, v[[UZP2]].8b, v[[UZP2]].8b
+; CHECK-NEXT: str s[[UZP3]], [x0]
+; CHECK-NEXT: ret
+  %cval = load <4 x i8>, <4 x i8>* %a
+  %ptrs = load <4 x i8*>, <4 x i8*>* %b
+  %mask = icmp eq <4 x i8> %cval, zeroinitializer
+  %vals = call <4 x i8> @llvm.masked.gather.v4i8(<4 x i8*> %ptrs, i32 8, <4 x i1> %mask, <4 x i8> undef)
+  store <4 x i8> %vals, <4 x i8>* %a
+  ret void
+}
+
+define void @masked_gather_v8i8(<8 x i8>* %a, <8 x i8*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v8i8:
+; VBITS_GE_512: ldr d[[VALS:[0-9]+]], [x0]
+; VBITS_GE_512-NEXT: ptrue [[PG0:p[0-9]+]].d, vl8
+; VBITS_GE_512-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; VBITS_GE_512-NEXT: ptrue [[PG1:p[0-9]+]].b, vl8
+; VBITS_GE_512-NEXT: cmeq v[[CMP:[0-9]+]].8b, v[[VALS]].8b, #0
+; VBITS_GE_512-NEXT: cmpne [[MASK:p[0-9]+]].b, [[PG1]]/z, z[[CMP]].b, #0
+; VBITS_GE_512-NEXT: ld1b { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; VBITS_GE_512-NEXT: uzp1 [[UZP1:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; VBITS_GE_512-NEXT: uzp1 [[UZP2:z[0-9]+]].h, [[UZP1]].h, [[UZP1]].h
+; VBITS_GE_512-NEXT: uzp1 z[[UZP3:[0-9]+]].b, [[UZP2]].b, [[UZP2]].b
+; VBITS_GE_512-NEXT: str d[[UZP3]], [x0]
+; VBITS_GE_512-NEXT: ret
+  %cval = load <8 x i8>, <8 x i8>* %a
+  %ptrs = load <8 x i8*>, <8 x i8*>* %b
+  %mask = icmp eq <8 x i8> %cval, zeroinitializer
+  %vals = call <8 x i8> @llvm.masked.gather.v8i8(<8 x i8*> %ptrs, i32 8, <8 x i1> %mask, <8 x i8> undef)
+  store <8 x i8> %vals, <8 x i8>* %a
+  ret void
+}
+
+define void @masked_gather_v16i8(<16 x i8>* %a, <16 x i8*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v16i8:
+; VBITS_GE_1024: ldr q[[VALS:[0-9]+]], [x0]
+; VBITS_GE_1024-NEXT: ptrue [[PG0:p[0-9]+]].d, vl16
+; VBITS_GE_1024-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; VBITS_GE_1024-NEXT: ptrue [[PG1:p[0-9]+]].b, vl16
+; VBITS_GE_1024-NEXT: cmeq v[[CMP:[0-9]+]].16b, v[[VALS]].16b, #0
+; VBITS_GE_1024-NEXT: cmpne [[MASK:p[0-9]+]].b, [[PG1]]/z, z[[CMP]].b, #0
+; VBITS_GE_1024-NEXT: ld1b { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; VBITS_GE_1024-NEXT: uzp1 [[UZP1:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; VBITS_GE_1024-NEXT: uzp1 [[UZP2:z[0-9]+]].h, [[UZP1]].h, [[UZP1]].h
+; VBITS_GE_1024-NEXT: uzp1 z[[UZP3:[0-9]+]].b, [[UZP2]].b, [[UZP2]].b
+; VBITS_GE_1024-NEXT: str q[[UZP3]], [x0]
+; VBITS_GE_1024-NEXT: ret
+  %cval = load <16 x i8>, <16 x i8>* %a
+  %ptrs = load <16 x i8*>, <16 x i8*>* %b
+  %mask = icmp eq <16 x i8> %cval, zeroinitializer
+  %vals = call <16 x i8> @llvm.masked.gather.v16i8(<16 x i8*> %ptrs, i32 8, <16 x i1> %mask, <16 x i8> undef)
+  store <16 x i8> %vals, <16 x i8>* %a
+  ret void
+}
+
+define void @masked_gather_v32i8(<32 x i8>* %a, <32 x i8*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v32i8:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].b, vl32
+; VBITS_GE_2048-NEXT: ld1b { [[VALS:z[0-9]+]].b }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].d, vl32
+; VBITS_GE_2048-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: cmpeq [[MASK:p[0-9]+]].b, [[PG0]]/z, [[VALS]].b, #0
+; VBITS_GE_2048-NEXT: ld1b { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; VBITS_GE_2048-NEXT: uzp1 [[UZP1:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; VBITS_GE_2048-NEXT: uzp1 [[UZP2:z[0-9]+]].h, [[UZP1]].h, [[UZP1]].h
+; VBITS_GE_2048-NEXT: uzp1 [[UZP3:z[0-9]+]].b, [[UZP2]].b, [[UZP2]].b
+; VBITS_GE_2048-NEXT: st1b { [[UZP3]].b }, [[PG0]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %cval = load <32 x i8>, <32 x i8>* %a
+  %ptrs = load <32 x i8*>, <32 x i8*>* %b
+  %mask = icmp eq <32 x i8> %cval, zeroinitializer
+  %vals = call <32 x i8> @llvm.masked.gather.v32i8(<32 x i8*> %ptrs, i32 8, <32 x i1> %mask, <32 x i8> undef)
+  store <32 x i8> %vals, <32 x i8>* %a
+  ret void
+}
+
+;
+; LD1H
+;
+
+define void @masked_gather_v2i16(<2 x i16>* %a, <2 x i16*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v2i16:
+; CHECK: ldrh [[VALS_LO:w[0-9]+]], [x0]
+; CHECK-NEXT: ldrh [[VALS_HI:w[0-9]+]], [x0, #2]
+; CHECK-NEXT: ldr q[[PTRS:[0-9]+]], [x1]
+; CHECK-NEXT: ptrue [[PG0:p[0-9]+]].s, vl2
+; CHECK-NEXT: fmov s[[VALS:[0-9]+]], [[VALS_LO]]
+; CHECK-NEXT: mov v[[VALS]].s[1], [[VALS_HI]]
+; CHECK-NEXT: cmeq v[[CMP:[0-9]+]].2s, v[[VALS]].2s, #0
+; CHECK-NEXT: cmpne [[MASK:p[0-9]+]].s, [[PG0]]/z, z[[CMP]].s, #0
+; CHECK-NEXT: ld1sh { z[[RES:[0-9]+]].d }, [[MASK]]/z, [z[[PTRS]].d]
+; CHECK-NEXT: xtn v[[XTN:[0-9]+]].2s, v[[RES]].2d
+; CHECK-NEXT: mov [[RES_HI:w[0-9]+]], v[[XTN]].s[1]
+; CHECK-NEXT: fmov [[RES_LO:w[0-9]+]], s[[XTN]]
+; CHECK-NEXT: strh [[RES_LO]], [x0]
+; CHECK-NEXT: strh [[RES_HI]], [x0, #2]
+; CHECK-NEXT: ret
+  %cval = load <2 x i16>, <2 x i16>* %a
+  %ptrs = load <2 x i16*>, <2 x i16*>* %b
+  %mask = icmp eq <2 x i16> %cval, zeroinitializer
+  %vals = call <2 x i16> @llvm.masked.gather.v2i16(<2 x i16*> %ptrs, i32 8, <2 x i1> %mask, <2 x i16> undef)
+  store <2 x i16> %vals, <2 x i16>* %a
+  ret void
+}
+
+define void @masked_gather_v4i16(<4 x i16>* %a, <4 x i16*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v4i16:
+; CHECK: ldr d[[VALS:[0-9]+]], [x0]
+; CHECK-NEXT: ptrue [[PG0:p[0-9]+]].d, vl4
+; CHECK-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].h, vl4
+; CHECK-NEXT: cmeq v[[CMP:[0-9]+]].4h, v[[VALS]].4h, #0
+; CHECK-NEXT: cmpne [[MASK:p[0-9]+]].h, [[PG1]]/z, z[[CMP]].h, #0
+; CHECK-NEXT: ld1h { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; CHECK-NEXT: uzp1 [[UZP1:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; CHECK-NEXT: uzp1 z[[UZP2:[0-9]+]].h, [[UZP1]].h, [[UZP1]].h
+; CHECK-NEXT: str d[[UZP2]], [x0]
+; CHECK-NEXT: ret
+  %cval = load <4 x i16>, <4 x i16>* %a
+  %ptrs = load <4 x i16*>, <4 x i16*>* %b
+  %mask = icmp eq <4 x i16> %cval, zeroinitializer
+  %vals = call <4 x i16> @llvm.masked.gather.v4i16(<4 x i16*> %ptrs, i32 8, <4 x i1> %mask, <4 x i16> undef)
+  store <4 x i16> %vals, <4 x i16>* %a
+  ret void
+}
+
+define void @masked_gather_v8i16(<8 x i16>* %a, <8 x i16*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v8i16:
+; VBITS_GE_512: ldr q[[VALS:[0-9]+]], [x0]
+; VBITS_GE_512-NEXT: ptrue [[PG0:p[0-9]+]].d, vl8
+; VBITS_GE_512-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; VBITS_GE_512-NEXT: ptrue [[PG1:p[0-9]+]].h, vl8
+; VBITS_GE_512-NEXT: cmeq v[[CMP:[0-9]+]].8h, v[[VALS]].8h, #0
+; VBITS_GE_512-NEXT: cmpne [[MASK:p[0-9]+]].h, [[PG1]]/z, z[[CMP]].h, #0
+; VBITS_GE_512-NEXT: ld1h { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; VBITS_GE_512-NEXT: uzp1 [[UZP1:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; VBITS_GE_512-NEXT: uzp1 z[[UZP2:[0-9]+]].h, [[UZP1]].h, [[UZP1]].h
+; VBITS_GE_512-NEXT: str q[[UZP2]], [x0]
+; VBITS_GE_512-NEXT: ret
+  %cval = load <8 x i16>, <8 x i16>* %a
+  %ptrs = load <8 x i16*>, <8 x i16*>* %b
+  %mask = icmp eq <8 x i16> %cval, zeroinitializer
+  %vals = call <8 x i16> @llvm.masked.gather.v8i16(<8 x i16*> %ptrs, i32 8, <8 x i1> %mask, <8 x i16> undef)
+  store <8 x i16> %vals, <8 x i16>* %a
+  ret void
+}
+
+define void @masked_gather_v16i16(<16 x i16>* %a, <16 x i16*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v16i16:
+; VBITS_GE_1024: ptrue [[PG0:p[0-9]+]].h, vl16
+; VBITS_GE_1024-NEXT: ld1h { [[VALS:z[0-9]+]].h }, [[PG0]]/z, [x0]
+; VBITS_GE_1024-NEXT: ptrue [[PG1:p[0-9]+]].d, vl16
+; VBITS_GE_1024-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_1024-NEXT: cmpeq [[MASK:p[0-9]+]].h, [[PG0]]/z, [[VALS]].h, #0
+; VBITS_GE_1024-NEXT: ld1h { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; VBITS_GE_1024-NEXT: uzp1 [[UZP1:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; VBITS_GE_1024-NEXT: uzp1 [[UZP2:z[0-9]+]].h, [[UZP1]].h, [[UZP1]].h
+; VBITS_GE_1024-NEXT: st1h { [[UZP2]].h }, [[PG0]], [x0]
+; VBITS_GE_1024-NEXT: ret
+  %cval = load <16 x i16>, <16 x i16>* %a
+  %ptrs = load <16 x i16*>, <16 x i16*>* %b
+  %mask = icmp eq <16 x i16> %cval, zeroinitializer
+  %vals = call <16 x i16> @llvm.masked.gather.v16i16(<16 x i16*> %ptrs, i32 8, <16 x i1> %mask, <16 x i16> undef)
+  store <16 x i16> %vals, <16 x i16>* %a
+  ret void
+}
+
+define void @masked_gather_v32i16(<32 x i16>* %a, <32 x i16*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v32i16:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].h, vl32
+; VBITS_GE_2048-NEXT: ld1h { [[VALS:z[0-9]+]].h }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].d, vl32
+; VBITS_GE_2048-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: cmpeq [[MASK:p[0-9]+]].h, [[PG0]]/z, [[VALS]].h, #0
+; VBITS_GE_2048-NEXT: ld1h { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; VBITS_GE_2048-NEXT: uzp1 [[UZP1:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; VBITS_GE_2048-NEXT: uzp1 [[UZP2:z[0-9]+]].h, [[UZP1]].h, [[UZP1]].h
+; VBITS_GE_2048-NEXT: st1h { [[UZP2]].h }, [[PG0]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %cval = load <32 x i16>, <32 x i16>* %a
+  %ptrs = load <32 x i16*>, <32 x i16*>* %b
+  %mask = icmp eq <32 x i16> %cval, zeroinitializer
+  %vals = call <32 x i16> @llvm.masked.gather.v32i16(<32 x i16*> %ptrs, i32 8, <32 x i1> %mask, <32 x i16> undef)
+  store <32 x i16> %vals, <32 x i16>* %a
+  ret void
+}
+
+;
+; LD1W
+;
+
+define void @masked_gather_v2i32(<2 x i32>* %a, <2 x i32*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v2i32:
+; CHECK: ldr d[[VALS:[0-9]+]], [x0]
+; CHECK-NEXT: ldr q[[PTRS:[0-9]+]], [x1]
+; CHECK-NEXT: ptrue [[PG0:p[0-9]+]].s, vl2
+; CHECK-NEXT: cmeq v[[CMP:[0-9]+]].2s, v[[VALS]].2s, #0
+; CHECK-NEXT: cmpne [[MASK:p[0-9]+]].s, [[PG0]]/z, z[[CMP]].s, #0
+; CHECK-NEXT: ld1w { z[[RES:[0-9]+]].d }, [[MASK]]/z, [z[[PTRS]].d]
+; CHECK-NEXT: xtn v[[XTN:[0-9]+]].2s, v[[RES]].2d
+; CHECK-NEXT: str d[[XTN]], [x0]
+; CHECK-NEXT: ret
+  %cval = load <2 x i32>, <2 x i32>* %a
+  %ptrs = load <2 x i32*>, <2 x i32*>* %b
+  %mask = icmp eq <2 x i32> %cval, zeroinitializer
+  %vals = call <2 x i32> @llvm.masked.gather.v2i32(<2 x i32*> %ptrs, i32 8, <2 x i1> %mask, <2 x i32> undef)
+  store <2 x i32> %vals, <2 x i32>* %a
+  ret void
+}
+
+define void @masked_gather_v4i32(<4 x i32>* %a, <4 x i32*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v4i32:
+; CHECK: ldr q[[VALS:[0-9]+]], [x0]
+; CHECK-NEXT: ptrue [[PG0:p[0-9]+]].d, vl4
+; CHECK-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].s, vl4
+; CHECK-NEXT: cmeq v[[CMP:[0-9]+]].4s, v[[VALS]].4s, #0
+; CHECK-NEXT: cmpne [[MASK:p[0-9]+]].s, [[PG1]]/z, z[[CMP]].s, #0
+; CHECK-NEXT: ld1w { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; CHECK-NEXT: uzp1 z[[UZP:[0-9]+]].s, [[RES]].s, [[RES]].s
+; CHECK-NEXT: str q[[UZP]], [x0]
+; CHECK-NEXT: ret
+  %cval = load <4 x i32>, <4 x i32>* %a
+  %ptrs = load <4 x i32*>, <4 x i32*>* %b
+  %mask = icmp eq <4 x i32> %cval, zeroinitializer
+  %vals = call <4 x i32> @llvm.masked.gather.v4i32(<4 x i32*> %ptrs, i32 8, <4 x i1> %mask, <4 x i32> undef)
+  store <4 x i32> %vals, <4 x i32>* %a
+  ret void
+}
+
+define void @masked_gather_v8i32(<8 x i32>* %a, <8 x i32*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v8i32:
+; VBITS_GE_512: ptrue [[PG0:p[0-9]+]].s, vl8
+; VBITS_GE_512-NEXT: ld1w { [[VALS:z[0-9]+]].s }, [[PG0]]/z, [x0]
+; VBITS_GE_512-NEXT: ptrue [[PG1:p[0-9]+]].d, vl8
+; VBITS_GE_512-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_512-NEXT: cmpeq [[MASK:p[0-9]+]].s, [[PG0]]/z, [[VALS]].s, #0
+; VBITS_GE_512-NEXT: ld1w { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; VBITS_GE_512-NEXT: uzp1 [[UZP:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; VBITS_GE_512-NEXT: st1w { [[UZP]].s }, [[PG0]], [x0]
+; VBITS_GE_512-NEXT: ret
+  %cval = load <8 x i32>, <8 x i32>* %a
+  %ptrs = load <8 x i32*>, <8 x i32*>* %b
+  %mask = icmp eq <8 x i32> %cval, zeroinitializer
+  %vals = call <8 x i32> @llvm.masked.gather.v8i32(<8 x i32*> %ptrs, i32 8, <8 x i1> %mask, <8 x i32> undef)
+  store <8 x i32> %vals, <8 x i32>* %a
+  ret void
+}
+
+define void @masked_gather_v16i32(<16 x i32>* %a, <16 x i32*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v16i32:
+; VBITS_GE_1024: ptrue [[PG0:p[0-9]+]].s, vl16
+; VBITS_GE_1024-NEXT: ld1w { [[VALS:z[0-9]+]].s }, [[PG0]]/z, [x0]
+; VBITS_GE_1024-NEXT: ptrue [[PG1:p[0-9]+]].d, vl16
+; VBITS_GE_1024-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_1024-NEXT: cmpeq [[MASK:p[0-9]+]].s, [[PG0]]/z, [[VALS]].s, #0
+; VBITS_GE_1024-NEXT: ld1w { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; VBITS_GE_1024-NEXT: uzp1 [[UZP:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; VBITS_GE_1024-NEXT: st1w { [[UZP]].s }, [[PG0]], [x0]
+; VBITS_GE_1024-NEXT: ret
+  %cval = load <16 x i32>, <16 x i32>* %a
+  %ptrs = load <16 x i32*>, <16 x i32*>* %b
+  %mask = icmp eq <16 x i32> %cval, zeroinitializer
+  %vals = call <16 x i32> @llvm.masked.gather.v16i32(<16 x i32*> %ptrs, i32 8, <16 x i1> %mask, <16 x i32> undef)
+  store <16 x i32> %vals, <16 x i32>* %a
+  ret void
+}
+
+define void @masked_gather_v32i32(<32 x i32>* %a, <32 x i32*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v32i32:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].s, vl32
+; VBITS_GE_2048-NEXT: ld1w { [[VALS:z[0-9]+]].s }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].d, vl32
+; VBITS_GE_2048-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: cmpeq [[MASK:p[0-9]+]].s, [[PG0]]/z, [[VALS]].s, #0
+; VBITS_GE_2048-NEXT: ld1w { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; VBITS_GE_2048-NEXT: uzp1 [[UZP:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; VBITS_GE_2048-NEXT: st1w { [[UZP]].s }, [[PG0]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %cval = load <32 x i32>, <32 x i32>* %a
+  %ptrs = load <32 x i32*>, <32 x i32*>* %b
+  %mask = icmp eq <32 x i32> %cval, zeroinitializer
+  %vals = call <32 x i32> @llvm.masked.gather.v32i32(<32 x i32*> %ptrs, i32 8, <32 x i1> %mask, <32 x i32> undef)
+  store <32 x i32> %vals, <32 x i32>* %a
+  ret void
+}
+
+;
+; LD1D
+;
+
+; Scalarize 1 x i64 gathers
+define void @masked_gather_v1i64(<1 x i64>* %a, <1 x i64*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v1i64:
+; CHECK-NOT: ptrue
+  %cval = load <1 x i64>, <1 x i64>* %a
+  %ptrs = load <1 x i64*>, <1 x i64*>* %b
+  %mask = icmp eq <1 x i64> %cval, zeroinitializer
+  %vals = call <1 x i64> @llvm.masked.gather.v1i64(<1 x i64*> %ptrs, i32 8, <1 x i1> %mask, <1 x i64> undef)
+  store <1 x i64> %vals, <1 x i64>* %a
+  ret void
+}
+
+define void @masked_gather_v2i64(<2 x i64>* %a, <2 x i64*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v2i64:
+; CHECK: ldr q[[VALS:[0-9]+]], [x0]
+; CHECK-NEXT: ldr q[[PTRS:[0-9]+]], [x1]
+; CHECK-NEXT: ptrue [[PG0:p[0-9]+]].d, vl2
+; CHECK-NEXT: cmeq v[[CMP:[0-9]+]].2d, v[[VALS]].2d, #0
+; CHECK-NEXT: cmpne [[MASK:p[0-9]+]].d, [[PG0]]/z, z[[CMP]].d, #0
+; CHECK-NEXT: ld1d { z[[RES:[0-9]+]].d }, [[MASK]]/z, [z[[PTRS]].d]
+; CHECK-NEXT: str q[[RES]], [x0]
+; CHECK-NEXT: ret
+  %cval = load <2 x i64>, <2 x i64>* %a
+  %ptrs = load <2 x i64*>, <2 x i64*>* %b
+  %mask = icmp eq <2 x i64> %cval, zeroinitializer
+  %vals = call <2 x i64> @llvm.masked.gather.v2i64(<2 x i64*> %ptrs, i32 8, <2 x i1> %mask, <2 x i64> undef)
+  store <2 x i64> %vals, <2 x i64>* %a
+  ret void
+}
+
+define void @masked_gather_v4i64(<4 x i64>* %a, <4 x i64*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v4i64:
+; CHECK: ptrue [[PG0:p[0-9]+]].d, vl4
+; CHECK-NEXT: ld1d { [[VALS:z[0-9]+]].d }, [[PG0]]/z, [x0]
+; CHECK-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; CHECK-NEXT: cmpeq [[MASK:p[0-9]+]].d, [[PG0]]/z, [[VALS]].d, #0
+; CHECK-NEXT: ld1d { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; CHECK-NEXT: st1d { [[RES]].d }, [[PG0]], [x0]
+; CHECK-NEXT: ret
+  %cval = load <4 x i64>, <4 x i64>* %a
+  %ptrs = load <4 x i64*>, <4 x i64*>* %b
+  %mask = icmp eq <4 x i64> %cval, zeroinitializer
+  %vals = call <4 x i64> @llvm.masked.gather.v4i64(<4 x i64*> %ptrs, i32 8, <4 x i1> %mask, <4 x i64> undef)
+  store <4 x i64> %vals, <4 x i64>* %a
+  ret void
+}
+
+define void @masked_gather_v8i64(<8 x i64>* %a, <8 x i64*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v8i64:
+; VBITS_GE_512: ptrue [[PG0:p[0-9]+]].d, vl8
+; VBITS_GE_512-NEXT: ld1d { [[VALS:z[0-9]+]].d }, [[PG0]]/z, [x0]
+; VBITS_GE_512-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; VBITS_GE_512-NEXT: cmpeq [[MASK:p[0-9]+]].d, [[PG0]]/z, [[VALS]].d, #0
+; VBITS_GE_512-NEXT: ld1d { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; VBITS_GE_512-NEXT: st1d { [[RES]].d }, [[PG0]], [x0]
+; VBITS_GE_512-NEXT: ret
+  %cval = load <8 x i64>, <8 x i64>* %a
+  %ptrs = load <8 x i64*>, <8 x i64*>* %b
+  %mask = icmp eq <8 x i64> %cval, zeroinitializer
+  %vals = call <8 x i64> @llvm.masked.gather.v8i64(<8 x i64*> %ptrs, i32 8, <8 x i1> %mask, <8 x i64> undef)
+  store <8 x i64> %vals, <8 x i64>* %a
+  ret void
+}
+
+define void @masked_gather_v16i64(<16 x i64>* %a, <16 x i64*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v16i64:
+; VBITS_GE_1024: ptrue [[PG0:p[0-9]+]].d, vl16
+; VBITS_GE_1024-NEXT: ld1d { [[VALS:z[0-9]+]].d }, [[PG0]]/z, [x0]
+; VBITS_GE_1024-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; VBITS_GE_1024-NEXT: cmpeq [[MASK:p[0-9]+]].d, [[PG0]]/z, [[VALS]].d, #0
+; VBITS_GE_1024-NEXT: ld1d { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; VBITS_GE_1024-NEXT: st1d { [[RES]].d }, [[PG0]], [x0]
+; VBITS_GE_1024-NEXT: ret
+  %cval = load <16 x i64>, <16 x i64>* %a
+  %ptrs = load <16 x i64*>, <16 x i64*>* %b
+  %mask = icmp eq <16 x i64> %cval, zeroinitializer
+  %vals = call <16 x i64> @llvm.masked.gather.v16i64(<16 x i64*> %ptrs, i32 8, <16 x i1> %mask, <16 x i64> undef)
+  store <16 x i64> %vals, <16 x i64>* %a
+  ret void
+}
+
+define void @masked_gather_v32i64(<32 x i64>* %a, <32 x i64*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v32i64:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].d, vl32
+; VBITS_GE_2048-NEXT: ld1d { [[VALS:z[0-9]+]].d }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; VBITS_GE_2048-NEXT: cmpeq [[MASK:p[0-9]+]].d, [[PG0]]/z, [[VALS]].d, #0
+; VBITS_GE_2048-NEXT: ld1d { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; VBITS_GE_2048-NEXT: st1d { [[RES]].d }, [[PG0]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %cval = load <32 x i64>, <32 x i64>* %a
+  %ptrs = load <32 x i64*>, <32 x i64*>* %b
+  %mask = icmp eq <32 x i64> %cval, zeroinitializer
+  %vals = call <32 x i64> @llvm.masked.gather.v32i64(<32 x i64*> %ptrs, i32 8, <32 x i1> %mask, <32 x i64> undef)
+  store <32 x i64> %vals, <32 x i64>* %a
+  ret void
+}
+
+;
+; LD1H (float)
+;
+
+define void @masked_gather_v2f16(<2 x half>* %a, <2 x half*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v2f16:
+; CHECK: ldr s[[VALS:[0-9]+]], [x0]
+; CHECK-NEXT: movi d[[ZERO:[0-9]+]], #0000000000000000
+; CHECK-NEXT: ldr q[[PTRS:[0-9]+]], [x1]
+; CHECK-NEXT: ptrue [[PG0:p[0-9]+]].h, vl4
+; CHECK-NEXT: fcmeq v[[CMP:[0-9]+]].4h, v[[VALS]].4h, #0.0
+; CHECK-NEXT: umov w8, v[[CMP]].h[0]
+; CHECK-NEXT: umov w9, v[[CMP]].h[1]
+; CHECK-NEXT: fmov s[[CMP]], w8
+; CHECK-NEXT: mov v[[CMP]].s[1], w9
+; CHECK-NEXT: shl v[[CMP]].2s, v[[CMP]].2s, #16
+; CHECK-NEXT: sshr v[[CMP]].2s, v[[CMP]].2s, #16
+; CHECK-NEXT: fmov w9, s[[CMP]]
+; CHECK-NEXT: mov w8, v[[CMP]].s[1]
+; CHECK-NEXT: mov v[[NCMP:[0-9]+]].h[0], w9
+; CHECK-NEXT: mov v[[NCMP]].h[1], w8
+; CHECK-NEXT: shl v[[SHL:[0-9]+]].4h, v[[NCMP]].4h, #15
+; CHECK-NEXT: sshr v[[SHL]].4h, v[[SHL]].4h, #15
+; CHECK-NEXT: cmpne [[MASK:p[0-9]+]].h, [[PG0]]/z, z[[SHL]].h, #0
+; CHECK-NEXT: ld1h { [[RES:z[0-9]+]].d }, [[MASK]]/z, [z[[PTRS]].d]
+; CHECK-NEXT: uzp1 [[UZP1:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; CHECK-NEXT: uzp1 z[[UZP2:[0-9]+]].h, [[UZP1]].h, [[UZP1]].h
+; CHECK-NEXT: str s[[UZP2]], [x0]
+; CHECK-NEXT: ret
+  %cval = load <2 x half>, <2 x half>* %a
+  %ptrs = load <2 x half*>, <2 x half*>* %b
+  %mask = fcmp oeq <2 x half> %cval, zeroinitializer
+  %vals = call <2 x half> @llvm.masked.gather.v2f16(<2 x half*> %ptrs, i32 8, <2 x i1> %mask, <2 x half> undef)
+  store <2 x half> %vals, <2 x half>* %a
+  ret void
+}
+
+define void @masked_gather_v4f16(<4 x half>* %a, <4 x half*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v4f16:
+; CHECK: ldr d[[VALS:[0-9]+]], [x0]
+; CHECK-NEXT: ptrue [[PG0:p[0-9]+]].d, vl4
+; CHECK-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].h, vl4
+; CHECK-NEXT: fcmeq v[[CMP:[0-9]+]].4h, v[[VALS]].4h, #0
+; CHECK-NEXT: cmpne [[MASK:p[0-9]+]].h, [[PG1]]/z, z[[CMP]].h, #0
+; CHECK-NEXT: ld1h { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; CHECK-NEXT: uzp1 [[UZP1:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; CHECK-NEXT: uzp1 z[[UZP2:[0-9]+]].h, [[UZP1]].h, [[UZP1]].h
+; CHECK-NEXT: str d[[UZP2]], [x0]
+; CHECK-NEXT: ret
+  %cval = load <4 x half>, <4 x half>* %a
+  %ptrs = load <4 x half*>, <4 x half*>* %b
+  %mask = fcmp oeq <4 x half> %cval, zeroinitializer
+  %vals = call <4 x half> @llvm.masked.gather.v4f16(<4 x half*> %ptrs, i32 8, <4 x i1> %mask, <4 x half> undef)
+  store <4 x half> %vals, <4 x half>* %a
+  ret void
+}
+
+define void @masked_gather_v8f16(<8 x half>* %a, <8 x half*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v8f16:
+; VBITS_GE_512: ldr q[[VALS:[0-9]+]], [x0]
+; VBITS_GE_512-NEXT: ptrue [[PG0:p[0-9]+]].d, vl8
+; VBITS_GE_512-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; VBITS_GE_512-NEXT: ptrue [[PG1:p[0-9]+]].h, vl8
+; VBITS_GE_512-NEXT: fcmeq v[[CMP:[0-9]+]].8h, v[[VALS]].8h, #0
+; VBITS_GE_512-NEXT: cmpne [[MASK:p[0-9]+]].h, [[PG1]]/z, z[[CMP]].h, #0
+; VBITS_GE_512-NEXT: ld1h { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; VBITS_GE_512-NEXT: uzp1 [[UZP1:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; VBITS_GE_512-NEXT: uzp1 z[[UZP2:[0-9]+]].h, [[UZP1]].h, [[UZP1]].h
+; VBITS_GE_512-NEXT: str q[[UZP2]], [x0]
+; VBITS_GE_512-NEXT: ret
+  %cval = load <8 x half>, <8 x half>* %a
+  %ptrs = load <8 x half*>, <8 x half*>* %b
+  %mask = fcmp oeq <8 x half> %cval, zeroinitializer
+  %vals = call <8 x half> @llvm.masked.gather.v8f16(<8 x half*> %ptrs, i32 8, <8 x i1> %mask, <8 x half> undef)
+  store <8 x half> %vals, <8 x half>* %a
+  ret void
+}
+
+define void @masked_gather_v16f16(<16 x half>* %a, <16 x half*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v16f16:
+; VBITS_GE_1024: ptrue [[PG0:p[0-9]+]].h, vl16
+; VBITS_GE_1024-NEXT: ld1h { [[VALS:z[0-9]+]].h }, [[PG0]]/z, [x0]
+; VBITS_GE_1024-NEXT: ptrue [[PG1:p[0-9]+]].d, vl16
+; VBITS_GE_1024-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_1024-NEXT: mov [[ZERO:z[0-9]+]].h, #0
+; VBITS_GE_1024-NEXT: fcmeq [[MASK:p[0-9]+]].h, [[PG0]]/z, [[VALS]].h, [[ZERO]].h
+; VBITS_GE_1024-NEXT: ld1h { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; VBITS_GE_1024-NEXT: uzp1 [[UZP1:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; VBITS_GE_1024-NEXT: uzp1 [[UZP2:z[0-9]+]].h, [[UZP1]].h, [[UZP1]].h
+; VBITS_GE_1024-NEXT: st1h { [[UZP2]].h }, [[PG0]], [x0]
+; VBITS_GE_1024-NEXT: ret
+  %cval = load <16 x half>, <16 x half>* %a
+  %ptrs = load <16 x half*>, <16 x half*>* %b
+  %mask = fcmp oeq <16 x half> %cval, zeroinitializer
+  %vals = call <16 x half> @llvm.masked.gather.v16f16(<16 x half*> %ptrs, i32 8, <16 x i1> %mask, <16 x half> undef)
+  store <16 x half> %vals, <16 x half>* %a
+  ret void
+}
+
+define void @masked_gather_v32f16(<32 x half>* %a, <32 x half*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v32f16:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].h, vl32
+; VBITS_GE_2048-NEXT: ld1h { [[VALS:z[0-9]+]].h }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].d, vl32
+; VBITS_GE_2048-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: mov [[ZERO:z[0-9]+]].h, #0
+; VBITS_GE_2048-NEXT: fcmeq [[MASK:p[0-9]+]].h, [[PG0]]/z, [[VALS]].h, [[ZERO]].h
+; VBITS_GE_2048-NEXT: ld1h { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; VBITS_GE_2048-NEXT: uzp1 [[UZP1:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; VBITS_GE_2048-NEXT: uzp1 [[UZP2:z[0-9]+]].h, [[UZP1]].h, [[UZP1]].h
+; VBITS_GE_2048-NEXT: st1h { [[UZP2]].h }, [[PG0]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %cval = load <32 x half>, <32 x half>* %a
+  %ptrs = load <32 x half*>, <32 x half*>* %b
+  %mask = fcmp oeq <32 x half> %cval, zeroinitializer
+  %vals = call <32 x half> @llvm.masked.gather.v32f16(<32 x half*> %ptrs, i32 8, <32 x i1> %mask, <32 x half> undef)
+  store <32 x half> %vals, <32 x half>* %a
+  ret void
+}
+
+;
+; LD1W (float)
+;
+
+define void @masked_gather_v2f32(<2 x float>* %a, <2 x float*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v2f32:
+; CHECK: ldr d[[VALS:[0-9]+]], [x0]
+; CHECK-NEXT: ldr q[[PTRS:[0-9]+]], [x1]
+; CHECK-NEXT: ptrue [[PG0:p[0-9]+]].s, vl2
+; CHECK-NEXT: fcmeq v[[CMP:[0-9]+]].2s, v[[VALS]].2s, #0
+; CHECK-NEXT: cmpne [[MASK:p[0-9]+]].s, [[PG0]]/z, z[[CMP]].s, #0
+; CHECK-NEXT: ld1w { z[[RES:[0-9]+]].d }, [[MASK]]/z, [z[[PTRS]].d]
+; CHECK-NEXT: xtn v[[XTN:[0-9]+]].2s, v[[RES]].2d
+; CHECK-NEXT: str d[[XTN]], [x0]
+; CHECK-NEXT: ret
+  %cval = load <2 x float>, <2 x float>* %a
+  %ptrs = load <2 x float*>, <2 x float*>* %b
+  %mask = fcmp oeq <2 x float> %cval, zeroinitializer
+  %vals = call <2 x float> @llvm.masked.gather.v2f32(<2 x float*> %ptrs, i32 8, <2 x i1> %mask, <2 x float> undef)
+  store <2 x float> %vals, <2 x float>* %a
+  ret void
+}
+
+define void @masked_gather_v4f32(<4 x float>* %a, <4 x float*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v4f32:
+; CHECK: ldr q[[VALS:[0-9]+]], [x0]
+; CHECK-NEXT: ptrue [[PG0:p[0-9]+]].d, vl4
+; CHECK-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].s, vl4
+; CHECK-NEXT: fcmeq v[[CMP:[0-9]+]].4s, v[[VALS]].4s, #0
+; CHECK-NEXT: cmpne [[MASK:p[0-9]+]].s, [[PG1]]/z, z[[CMP]].s, #0
+; CHECK-NEXT: ld1w { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; CHECK-NEXT: uzp1 z[[UZP:[0-9]+]].s, [[RES]].s, [[RES]].s
+; CHECK-NEXT: str q[[UZP]], [x0]
+; CHECK-NEXT: ret
+  %cval = load <4 x float>, <4 x float>* %a
+  %ptrs = load <4 x float*>, <4 x float*>* %b
+  %mask = fcmp oeq <4 x float> %cval, zeroinitializer
+  %vals = call <4 x float> @llvm.masked.gather.v4f32(<4 x float*> %ptrs, i32 8, <4 x i1> %mask, <4 x float> undef)
+  store <4 x float> %vals, <4 x float>* %a
+  ret void
+}
+
+define void @masked_gather_v8f32(<8 x float>* %a, <8 x float*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v8f32:
+; VBITS_GE_512: ptrue [[PG0:p[0-9]+]].s, vl8
+; VBITS_GE_512-NEXT: ld1w { [[VALS:z[0-9]+]].s }, [[PG0]]/z, [x0]
+; VBITS_GE_512-NEXT: ptrue [[PG1:p[0-9]+]].d, vl8
+; VBITS_GE_512-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_512-NEXT: mov [[ZERO:z[0-9]+]].s, #0
+; VBITS_GE_512-NEXT: fcmeq [[MASK:p[0-9]+]].s, [[PG0]]/z, [[VALS]].s, [[ZERO]].s
+; VBITS_GE_512-NEXT: ld1w { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; VBITS_GE_512-NEXT: uzp1 [[UZP:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; VBITS_GE_512-NEXT: st1w { [[UZP]].s }, [[PG0]], [x0]
+; VBITS_GE_512-NEXT: ret
+  %cval = load <8 x float>, <8 x float>* %a
+  %ptrs = load <8 x float*>, <8 x float*>* %b
+  %mask = fcmp oeq <8 x float> %cval, zeroinitializer
+  %vals = call <8 x float> @llvm.masked.gather.v8f32(<8 x float*> %ptrs, i32 8, <8 x i1> %mask, <8 x float> undef)
+  store <8 x float> %vals, <8 x float>* %a
+  ret void
+}
+
+define void @masked_gather_v16f32(<16 x float>* %a, <16 x float*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v16f32:
+; VBITS_GE_1024: ptrue [[PG0:p[0-9]+]].s, vl16
+; VBITS_GE_1024-NEXT: ld1w { [[VALS:z[0-9]+]].s }, [[PG0]]/z, [x0]
+; VBITS_GE_1024-NEXT: ptrue [[PG1:p[0-9]+]].d, vl16
+; VBITS_GE_1024-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_1024-NEXT: mov [[ZERO:z[0-9]+]].s, #0
+; VBITS_GE_1024-NEXT: fcmeq [[MASK:p[0-9]+]].s, [[PG0]]/z, [[VALS]].s, [[ZERO]].s
+; VBITS_GE_1024-NEXT: ld1w { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; VBITS_GE_1024-NEXT: uzp1 [[UZP:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; VBITS_GE_1024-NEXT: st1w { [[UZP]].s }, [[PG0]], [x0]
+; VBITS_GE_1024-NEXT: ret
+  %cval = load <16 x float>, <16 x float>* %a
+  %ptrs = load <16 x float*>, <16 x float*>* %b
+  %mask = fcmp oeq <16 x float> %cval, zeroinitializer
+  %vals = call <16 x float> @llvm.masked.gather.v16f32(<16 x float*> %ptrs, i32 8, <16 x i1> %mask, <16 x float> undef)
+  store <16 x float> %vals, <16 x float>* %a
+  ret void
+}
+
+define void @masked_gather_v32f32(<32 x float>* %a, <32 x float*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v32f32:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].s, vl32
+; VBITS_GE_2048-NEXT: ld1w { [[VALS:z[0-9]+]].s }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].d, vl32
+; VBITS_GE_2048-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: mov [[ZERO:z[0-9]+]].s, #0
+; VBITS_GE_2048-NEXT: fcmeq [[MASK:p[0-9]+]].s, [[PG0]]/z, [[VALS]].s, [[ZERO]].s
+; VBITS_GE_2048-NEXT: ld1w { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; VBITS_GE_2048-NEXT: uzp1 [[UZP:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; VBITS_GE_2048-NEXT: st1w { [[UZP]].s }, [[PG0]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %cval = load <32 x float>, <32 x float>* %a
+  %ptrs = load <32 x float*>, <32 x float*>* %b
+  %mask = fcmp oeq <32 x float> %cval, zeroinitializer
+  %vals = call <32 x float> @llvm.masked.gather.v32f32(<32 x float*> %ptrs, i32 8, <32 x i1> %mask, <32 x float> undef)
+  store <32 x float> %vals, <32 x float>* %a
+  ret void
+}
+
+;
+; LD1D (float)
+;
+
+; Scalarize 1 x double gathers
+define void @masked_gather_v1f64(<1 x double>* %a, <1 x double*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v1f64:
+; CHECK-NOT: ptrue
+  %cval = load <1 x double>, <1 x double>* %a
+  %ptrs = load <1 x double*>, <1 x double*>* %b
+  %mask = fcmp oeq <1 x double> %cval, zeroinitializer
+  %vals = call <1 x double> @llvm.masked.gather.v1f64(<1 x double*> %ptrs, i32 8, <1 x i1> %mask, <1 x double> undef)
+  store <1 x double> %vals, <1 x double>* %a
+  ret void
+}
+
+define void @masked_gather_v2f64(<2 x double>* %a, <2 x double*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v2f64:
+; CHECK: ldr q[[VALS:[0-9]+]], [x0]
+; CHECK-NEXT: ldr q[[PTRS:[0-9]+]], [x1]
+; CHECK-NEXT: ptrue [[PG0:p[0-9]+]].d, vl2
+; CHECK-NEXT: fcmeq v[[CMP:[0-9]+]].2d, v[[VALS]].2d, #0
+; CHECK-NEXT: cmpne [[MASK:p[0-9]+]].d, [[PG0]]/z, z[[CMP]].d, #0
+; CHECK-NEXT: ld1d { z[[RES:[0-9]+]].d }, [[MASK]]/z, [z[[PTRS]].d]
+; CHECK-NEXT: str q[[RES]], [x0]
+; CHECK-NEXT: ret
+  %cval = load <2 x double>, <2 x double>* %a
+  %ptrs = load <2 x double*>, <2 x double*>* %b
+  %mask = fcmp oeq <2 x double> %cval, zeroinitializer
+  %vals = call <2 x double> @llvm.masked.gather.v2f64(<2 x double*> %ptrs, i32 8, <2 x i1> %mask, <2 x double> undef)
+  store <2 x double> %vals, <2 x double>* %a
+  ret void
+}
+
+define void @masked_gather_v4f64(<4 x double>* %a, <4 x double*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v4f64:
+; CHECK: ptrue [[PG0:p[0-9]+]].d, vl4
+; CHECK-NEXT: ld1d { [[VALS:z[0-9]+]].d }, [[PG0]]/z, [x0]
+; CHECK-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; CHECK-NEXT: mov [[ZERO:z[0-9]+]].d, #0
+; CHECK-NEXT: fcmeq [[MASK:p[0-9]+]].d, [[PG0]]/z, [[VALS]].d, [[ZERO]].d
+; CHECK-NEXT: ld1d { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; CHECK-NEXT: st1d { [[RES]].d }, [[PG0]], [x0]
+; CHECK-NEXT: ret
+  %cval = load <4 x double>, <4 x double>* %a
+  %ptrs = load <4 x double*>, <4 x double*>* %b
+  %mask = fcmp oeq <4 x double> %cval, zeroinitializer
+  %vals = call <4 x double> @llvm.masked.gather.v4f64(<4 x double*> %ptrs, i32 8, <4 x i1> %mask, <4 x double> undef)
+  store <4 x double> %vals, <4 x double>* %a
+  ret void
+}
+
+define void @masked_gather_v8f64(<8 x double>* %a, <8 x double*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v8f64:
+; VBITS_GE_512: ptrue [[PG0:p[0-9]+]].d, vl8
+; VBITS_GE_512-NEXT: ld1d { [[VALS:z[0-9]+]].d }, [[PG0]]/z, [x0]
+; VBITS_GE_512-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; VBITS_GE_512-NEXT: mov [[ZERO:z[0-9]+]].d, #0
+; VBITS_GE_512-NEXT: fcmeq [[MASK:p[0-9]+]].d, [[PG0]]/z, [[VALS]].d, [[ZERO]].d
+; VBITS_GE_512-NEXT: ld1d { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; VBITS_GE_512-NEXT: st1d { [[RES]].d }, [[PG0]], [x0]
+; VBITS_GE_512-NEXT: ret
+  %cval = load <8 x double>, <8 x double>* %a
+  %ptrs = load <8 x double*>, <8 x double*>* %b
+  %mask = fcmp oeq <8 x double> %cval, zeroinitializer
+  %vals = call <8 x double> @llvm.masked.gather.v8f64(<8 x double*> %ptrs, i32 8, <8 x i1> %mask, <8 x double> undef)
+  store <8 x double> %vals, <8 x double>* %a
+  ret void
+}
+
+define void @masked_gather_v16f64(<16 x double>* %a, <16 x double*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v16f64:
+; VBITS_GE_1024: ptrue [[PG0:p[0-9]+]].d, vl16
+; VBITS_GE_1024-NEXT: ld1d { [[VALS:z[0-9]+]].d }, [[PG0]]/z, [x0]
+; VBITS_GE_1024-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; VBITS_GE_1024-NEXT: mov [[ZERO:z[0-9]+]].d, #0
+; VBITS_GE_1024-NEXT: fcmeq [[MASK:p[0-9]+]].d, [[PG0]]/z, [[VALS]].d, [[ZERO]].d
+; VBITS_GE_1024-NEXT: ld1d { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; VBITS_GE_1024-NEXT: st1d { [[RES]].d }, [[PG0]], [x0]
+; VBITS_GE_1024-NEXT: ret
+  %cval = load <16 x double>, <16 x double>* %a
+  %ptrs = load <16 x double*>, <16 x double*>* %b
+  %mask = fcmp oeq <16 x double> %cval, zeroinitializer
+  %vals = call <16 x double> @llvm.masked.gather.v16f64(<16 x double*> %ptrs, i32 8, <16 x i1> %mask, <16 x double> undef)
+  store <16 x double> %vals, <16 x double>* %a
+  ret void
+}
+
+define void @masked_gather_v32f64(<32 x double>* %a, <32 x double*>* %b) #0 {
+; CHECK-LABEL: masked_gather_v32f64:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].d, vl32
+; VBITS_GE_2048-NEXT: ld1d { [[VALS:z[0-9]+]].d }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; VBITS_GE_2048-NEXT: mov [[ZERO:z[0-9]+]].d, #0
+; VBITS_GE_2048-NEXT: fcmeq [[MASK:p[0-9]+]].d, [[PG0]]/z, [[VALS]].d, [[ZERO]].d
+; VBITS_GE_2048-NEXT: ld1d { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; VBITS_GE_2048-NEXT: st1d { [[RES]].d }, [[PG0]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %cval = load <32 x double>, <32 x double>* %a
+  %ptrs = load <32 x double*>, <32 x double*>* %b
+  %mask = fcmp oeq <32 x double> %cval, zeroinitializer
+  %vals = call <32 x double> @llvm.masked.gather.v32f64(<32 x double*> %ptrs, i32 8, <32 x i1> %mask, <32 x double> undef)
+  store <32 x double> %vals, <32 x double>* %a
+  ret void
+}
+
+; The above tests test the types, the below tests check that the addressing
+; modes still function
+
+define void @masked_gather_32b_scaled_sext(<32 x half>* %a, <32 x i32>* %b, half* %base) #0 {
+; CHECK-LABEL: masked_gather_32b_scaled_sext:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].h, vl32
+; VBITS_GE_2048-NEXT: ld1h { [[VALS:z[0-9]+]].h }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].s, vl32
+; VBITS_GE_2048-NEXT: ld1w { [[PTRS:z[0-9]+]].s }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: mov [[ZERO:z[0-9]+]].h, #0
+; VBITS_GE_2048-NEXT: fcmeq [[MASK:p[0-9]+]].h, [[PG0]]/z, [[VALS]].h, [[ZERO]]
+; VBITS_GE_2048-NEXT: ld1h { [[RES:z[0-9]+]].s }, [[MASK]]/z, [x2, [[PTRS]].s, sxtw #1]
+; VBITS_GE_2048-NEXT: uzp1 [[UZP:z[0-9]+]].h, [[RES]].h, [[RES]].h
+; VBITS_GE_2048-NEXT: st1h { [[UZP]].h }, [[PG0]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %cvals = load <32 x half>, <32 x half>* %a
+  %idxs = load <32 x i32>, <32 x i32>* %b
+  %ext = sext <32 x i32> %idxs to <32 x i64>
+  %ptrs = getelementptr half, half* %base, <32 x i64> %ext
+  %mask = fcmp oeq <32 x half> %cvals, zeroinitializer
+  %vals = call <32 x half> @llvm.masked.gather.v32f16(<32 x half*> %ptrs, i32 8, <32 x i1> %mask, <32 x half> undef)
+  store <32 x half> %vals, <32 x half>* %a
+  ret void
+}
+
+define void @masked_gather_32b_scaled_zext(<32 x half>* %a, <32 x i32>* %b, half* %base) #0 {
+; CHECK-LABEL: masked_gather_32b_scaled_zext:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].h, vl32
+; VBITS_GE_2048-NEXT: ld1h { [[VALS:z[0-9]+]].h }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].s, vl32
+; VBITS_GE_2048-NEXT: ld1w { [[PTRS:z[0-9]+]].s }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: mov [[ZERO:z[0-9]+]].h, #0
+; VBITS_GE_2048-NEXT: fcmeq [[MASK:p[0-9]+]].h, [[PG0]]/z, [[VALS]].h, [[ZERO]]
+; VBITS_GE_2048-NEXT: ld1h { [[RES:z[0-9]+]].s }, [[MASK]]/z, [x2, [[PTRS]].s, uxtw #1]
+; VBITS_GE_2048-NEXT: uzp1 [[UZP:z[0-9]+]].h, [[RES]].h, [[RES]].h
+; VBITS_GE_2048-NEXT: st1h { [[UZP]].h }, [[PG0]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %cvals = load <32 x half>, <32 x half>* %a
+  %idxs = load <32 x i32>, <32 x i32>* %b
+  %ext = zext <32 x i32> %idxs to <32 x i64>
+  %ptrs = getelementptr half, half* %base, <32 x i64> %ext
+  %mask = fcmp oeq <32 x half> %cvals, zeroinitializer
+  %vals = call <32 x half> @llvm.masked.gather.v32f16(<32 x half*> %ptrs, i32 8, <32 x i1> %mask, <32 x half> undef)
+  store <32 x half> %vals, <32 x half>* %a
+  ret void
+}
+
+define void @masked_gather_32b_unscaled_sext(<32 x half>* %a, <32 x i32>* %b, i8* %base) #0 {
+; CHECK-LABEL: masked_gather_32b_unscaled_sext:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].h, vl32
+; VBITS_GE_2048-NEXT: ld1h { [[VALS:z[0-9]+]].h }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].s, vl32
+; VBITS_GE_2048-NEXT: ld1w { [[PTRS:z[0-9]+]].s }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: mov [[ZERO:z[0-9]+]].h, #0
+; VBITS_GE_2048-NEXT: fcmeq [[MASK:p[0-9]+]].h, [[PG0]]/z, [[VALS]].h, [[ZERO]]
+; VBITS_GE_2048-NEXT: ld1h { [[RES:z[0-9]+]].s }, [[MASK]]/z, [x2, [[PTRS]].s, sxtw]
+; VBITS_GE_2048-NEXT: uzp1 [[UZP:z[0-9]+]].h, [[RES]].h, [[RES]].h
+; VBITS_GE_2048-NEXT: st1h { [[UZP]].h }, [[PG0]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %cvals = load <32 x half>, <32 x half>* %a
+  %idxs = load <32 x i32>, <32 x i32>* %b
+  %ext = sext <32 x i32> %idxs to <32 x i64>
+  %byte_ptrs = getelementptr i8, i8* %base, <32 x i64> %ext
+  %ptrs = bitcast <32 x i8*> %byte_ptrs to <32 x half*>
+  %mask = fcmp oeq <32 x half> %cvals, zeroinitializer
+  %vals = call <32 x half> @llvm.masked.gather.v32f16(<32 x half*> %ptrs, i32 8, <32 x i1> %mask, <32 x half> undef)
+  store <32 x half> %vals, <32 x half>* %a
+  ret void
+}
+
+define void @masked_gather_32b_unscaled_zext(<32 x half>* %a, <32 x i32>* %b, i8* %base) #0 {
+; CHECK-LABEL: masked_gather_32b_unscaled_zext:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].h, vl32
+; VBITS_GE_2048-NEXT: ld1h { [[VALS:z[0-9]+]].h }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].s, vl32
+; VBITS_GE_2048-NEXT: ld1w { [[PTRS:z[0-9]+]].s }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: mov [[ZERO:z[0-9]+]].h, #0
+; VBITS_GE_2048-NEXT: fcmeq [[MASK:p[0-9]+]].h, [[PG0]]/z, [[VALS]].h, [[ZERO]]
+; VBITS_GE_2048-NEXT: ld1h { [[RES:z[0-9]+]].s }, [[MASK]]/z, [x2, [[PTRS]].s, uxtw]
+; VBITS_GE_2048-NEXT: uzp1 [[UZP:z[0-9]+]].h, [[RES]].h, [[RES]].h
+; VBITS_GE_2048-NEXT: st1h { [[UZP]].h }, [[PG0]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %cvals = load <32 x half>, <32 x half>* %a
+  %idxs = load <32 x i32>, <32 x i32>* %b
+  %ext = zext <32 x i32> %idxs to <32 x i64>
+  %byte_ptrs = getelementptr i8, i8* %base, <32 x i64> %ext
+  %ptrs = bitcast <32 x i8*> %byte_ptrs to <32 x half*>
+  %mask = fcmp oeq <32 x half> %cvals, zeroinitializer
+  %vals = call <32 x half> @llvm.masked.gather.v32f16(<32 x half*> %ptrs, i32 8, <32 x i1> %mask, <32 x half> undef)
+  store <32 x half> %vals, <32 x half>* %a
+  ret void
+}
+
+define void @masked_gather_64b_scaled(<32 x float>* %a, <32 x i64>* %b, float* %base) #0 {
+; CHECK-LABEL: masked_gather_64b_scaled:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].s, vl32
+; VBITS_GE_2048-NEXT: ld1w { [[VALS:z[0-9]+]].s }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].d, vl32
+; VBITS_GE_2048-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: mov [[ZERO:z[0-9]+]].s, #0
+; VBITS_GE_2048-NEXT: fcmeq [[MASK:p[0-9]+]].s, [[PG0]]/z, [[VALS]].s, [[ZERO]]
+; VBITS_GE_2048-NEXT: ld1w { [[RES:z[0-9]+]].d }, [[MASK]]/z, [x2, [[PTRS]].d, lsl #2]
+; VBITS_GE_2048-NEXT: uzp1 [[UZP:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; VBITS_GE_2048-NEXT: st1w { [[UZP]].s }, [[PG0]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %cvals = load <32 x float>, <32 x float>* %a
+  %idxs = load <32 x i64>, <32 x i64>* %b
+  %ptrs = getelementptr float, float* %base, <32 x i64> %idxs
+  %mask = fcmp oeq <32 x float> %cvals, zeroinitializer
+  %vals = call <32 x float> @llvm.masked.gather.v32f32(<32 x float*> %ptrs, i32 8, <32 x i1> %mask, <32 x float> undef)
+  store <32 x float> %vals, <32 x float>* %a
+  ret void
+}
+
+define void @masked_gather_64b_unscaled(<32 x float>* %a, <32 x i64>* %b, i8* %base) #0 {
+; CHECK-LABEL: masked_gather_64b_unscaled:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].s, vl32
+; VBITS_GE_2048-NEXT: ld1w { [[VALS:z[0-9]+]].s }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].d, vl32
+; VBITS_GE_2048-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: mov [[ZERO:z[0-9]+]].s, #0
+; VBITS_GE_2048-NEXT: fcmeq [[MASK:p[0-9]+]].s, [[PG0]]/z, [[VALS]].s, [[ZERO]]
+; VBITS_GE_2048-NEXT: ld1w { [[RES:z[0-9]+]].d }, [[MASK]]/z, [x2, [[PTRS]].d]
+; VBITS_GE_2048-NEXT: uzp1 [[UZP:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; VBITS_GE_2048-NEXT: st1w { [[UZP]].s }, [[PG0]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %cvals = load <32 x float>, <32 x float>* %a
+  %idxs = load <32 x i64>, <32 x i64>* %b
+  %byte_ptrs = getelementptr i8, i8* %base, <32 x i64> %idxs
+  %ptrs = bitcast <32 x i8*> %byte_ptrs to <32 x float*>
+  %mask = fcmp oeq <32 x float> %cvals, zeroinitializer
+  %vals = call <32 x float> @llvm.masked.gather.v32f32(<32 x float*> %ptrs, i32 8, <32 x i1> %mask, <32 x float> undef)
+  store <32 x float> %vals, <32 x float>* %a
+  ret void
+}
+
+; FIXME: This case does not yet codegen well due to deficiencies in opcode selection
+define void @masked_gather_vec_plus_reg(<32 x float>* %a, <32 x i8*>* %b, i64 %off) #0 {
+; CHECK-LABEL: masked_gather_vec_plus_reg:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].s, vl32
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].d, vl32
+; VBITS_GE_2048-NEXT: ld1w { [[VALS:z[0-9]+]].s }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: mov [[OFF:z[0-9]+]].d, x2
+; VBITS_GE_2048-NEXT: mov [[ZERO:z[0-9]+]].s, #0
+; VBITS_GE_2048-NEXT: add [[PTRS_ADD:z[0-9]+]].d, [[PG1]]/m, [[PTRS]].d, [[OFF]].d
+; VBITS_GE_2048-NEXT: fcmeq [[MASK:p[0-9]+]].s, [[PG0]]/z, [[VALS]].s, [[ZERO]]
+; VBITS_GE_2048-NEXT: ld1w { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS_ADD]].d]
+; VBITS_GE_2048-NEXT: uzp1 [[UZP:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; VBITS_GE_2048-NEXT: st1w { [[UZP]].s }, [[PG0]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %cvals = load <32 x float>, <32 x float>* %a
+  %bases = load <32 x i8*>, <32 x i8*>* %b
+  %byte_ptrs = getelementptr i8, <32 x i8*> %bases, i64 %off
+  %ptrs = bitcast <32 x i8*> %byte_ptrs to <32 x float*>
+  %mask = fcmp oeq <32 x float> %cvals, zeroinitializer
+  %vals = call <32 x float> @llvm.masked.gather.v32f32(<32 x float*> %ptrs, i32 8, <32 x i1> %mask, <32 x float> undef)
+  store <32 x float> %vals, <32 x float>* %a
+  ret void
+}
+
+; FIXME: This case does not yet codegen well due to deficiencies in opcode selection
+define void @masked_gather_vec_plus_imm(<32 x float>* %a, <32 x i8*>* %b) #0 {
+; CHECK-LABEL: masked_gather_vec_plus_imm:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].s, vl32
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].d, vl32
+; VBITS_GE_2048-NEXT: ld1w { [[VALS:z[0-9]+]].s }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: mov [[OFF:z[0-9]+]].d, #4
+; VBITS_GE_2048-NEXT: mov [[ZERO:z[0-9]+]].s, #0
+; VBITS_GE_2048-NEXT: add [[PTRS_ADD:z[0-9]+]].d, [[PG1]]/m, [[PTRS]].d, [[OFF]].d
+; VBITS_GE_2048-NEXT: fcmeq [[MASK:p[0-9]+]].s, [[PG0]]/z, [[VALS]].s, [[ZERO]]
+; VBITS_GE_2048-NEXT: ld1w { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS_ADD]].d]
+; VBITS_GE_2048-NEXT: uzp1 [[UZP:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; VBITS_GE_2048-NEXT: st1w { [[UZP]].s }, [[PG0]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %cvals = load <32 x float>, <32 x float>* %a
+  %bases = load <32 x i8*>, <32 x i8*>* %b
+  %byte_ptrs = getelementptr i8, <32 x i8*> %bases, i64 4
+  %ptrs = bitcast <32 x i8*> %byte_ptrs to <32 x float*>
+  %mask = fcmp oeq <32 x float> %cvals, zeroinitializer
+  %vals = call <32 x float> @llvm.masked.gather.v32f32(<32 x float*> %ptrs, i32 8, <32 x i1> %mask, <32 x float> undef)
+  store <32 x float> %vals, <32 x float>* %a
+  ret void
+}
+
+define void @masked_gather_passthru(<32 x float>* %a, <32 x float*>* %b, <32 x float>* %c) #0 {
+; CHECK-LABEL: masked_gather_passthru:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].s, vl32
+; VBITS_GE_2048-NEXT: ld1w { [[VALS:z[0-9]+]].s }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].d, vl32
+; VBITS_GE_2048-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: mov [[ZERO:z[0-9]+]].s, #0
+; VBITS_GE_2048-NEXT: fcmeq [[MASK:p[0-9]+]].s, [[PG0]]/z, [[VALS]].s, [[ZERO]]
+; VBITS_GE_2048-NEXT: ld1w { [[PT:z[0-9]+]].s }, [[PG0]]/z, [x2]
+; VBITS_GE_2048-NEXT: ld1w { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; VBITS_GE_2048-NEXT: uzp1 [[UZP:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; VBITS_GE_2048-NEXT: sel [[SEL:z[0-9]+]].s, [[PG1]], [[UZP]].s, [[PT]].s
+; VBITS_GE_2048-NEXT: st1w { [[SEL]].s }, [[PG0]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %cvals = load <32 x float>, <32 x float>* %a
+  %ptrs = load <32 x float*>, <32 x float*>* %b
+  %passthru = load <32 x float>, <32 x float>* %c
+  %mask = fcmp oeq <32 x float> %cvals, zeroinitializer
+  %vals = call <32 x float> @llvm.masked.gather.v32f32(<32 x float*> %ptrs, i32 8, <32 x i1> %mask, <32 x float> %passthru)
+  store <32 x float> %vals, <32 x float>* %a
+  ret void
+}
+
+define void @masked_gather_passthru_0(<32 x float>* %a, <32 x float*>* %b) #0 {
+; CHECK-LABEL: masked_gather_passthru_0:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].s, vl32
+; VBITS_GE_2048-NEXT: ld1w { [[VALS:z[0-9]+]].s }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].d, vl32
+; VBITS_GE_2048-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: mov [[ZERO:z[0-9]+]].s, #0
+; VBITS_GE_2048-NEXT: fcmeq [[MASK:p[0-9]+]].s, [[PG0]]/z, [[VALS]].s, [[ZERO]]
+; VBITS_GE_2048-NEXT: ld1w { [[RES:z[0-9]+]].d }, [[MASK]]/z, {{\[}}[[PTRS]].d]
+; VBITS_GE_2048-NEXT: uzp1 [[UZP:z[0-9]+]].s, [[RES]].s, [[RES]].s
+; VBITS_GE_2048-NEXT: st1w { [[UZP]].s }, [[PG0]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %cvals = load <32 x float>, <32 x float>* %a
+  %ptrs = load <32 x float*>, <32 x float*>* %b
+  %mask = fcmp oeq <32 x float> %cvals, zeroinitializer
+  %vals = call <32 x float> @llvm.masked.gather.v32f32(<32 x float*> %ptrs, i32 8, <32 x i1> %mask, <32 x float> zeroinitializer)
+  store <32 x float> %vals, <32 x float>* %a
+  ret void
+}
+
+declare <2 x i8> @llvm.masked.gather.v2i8(<2 x i8*>, i32, <2 x i1>, <2 x i8>)
+declare <4 x i8> @llvm.masked.gather.v4i8(<4 x i8*>, i32, <4 x i1>, <4 x i8>)
+declare <8 x i8> @llvm.masked.gather.v8i8(<8 x i8*>, i32, <8 x i1>, <8 x i8>)
+declare <16 x i8> @llvm.masked.gather.v16i8(<16 x i8*>, i32, <16 x i1>, <16 x i8>)
+declare <32 x i8> @llvm.masked.gather.v32i8(<32 x i8*>, i32, <32 x i1>, <32 x i8>)
+
+declare <2 x i16> @llvm.masked.gather.v2i16(<2 x i16*>, i32, <2 x i1>, <2 x i16>)
+declare <4 x i16> @llvm.masked.gather.v4i16(<4 x i16*>, i32, <4 x i1>, <4 x i16>)
+declare <8 x i16> @llvm.masked.gather.v8i16(<8 x i16*>, i32, <8 x i1>, <8 x i16>)
+declare <16 x i16> @llvm.masked.gather.v16i16(<16 x i16*>, i32, <16 x i1>, <16 x i16>)
+declare <32 x i16> @llvm.masked.gather.v32i16(<32 x i16*>, i32, <32 x i1>, <32 x i16>)
+
+declare <2 x i32> @llvm.masked.gather.v2i32(<2 x i32*>, i32, <2 x i1>, <2 x i32>)
+declare <4 x i32> @llvm.masked.gather.v4i32(<4 x i32*>, i32, <4 x i1>, <4 x i32>)
+declare <8 x i32> @llvm.masked.gather.v8i32(<8 x i32*>, i32, <8 x i1>, <8 x i32>)
+declare <16 x i32> @llvm.masked.gather.v16i32(<16 x i32*>, i32, <16 x i1>, <16 x i32>)
+declare <32 x i32> @llvm.masked.gather.v32i32(<32 x i32*>, i32, <32 x i1>, <32 x i32>)
+
+declare <1 x i64> @llvm.masked.gather.v1i64(<1 x i64*>, i32, <1 x i1>, <1 x i64>)
+declare <2 x i64> @llvm.masked.gather.v2i64(<2 x i64*>, i32, <2 x i1>, <2 x i64>)
+declare <4 x i64> @llvm.masked.gather.v4i64(<4 x i64*>, i32, <4 x i1>, <4 x i64>)
+declare <8 x i64> @llvm.masked.gather.v8i64(<8 x i64*>, i32, <8 x i1>, <8 x i64>)
+declare <16 x i64> @llvm.masked.gather.v16i64(<16 x i64*>, i32, <16 x i1>, <16 x i64>)
+declare <32 x i64> @llvm.masked.gather.v32i64(<32 x i64*>, i32, <32 x i1>, <32 x i64>)
+
+declare <2 x half> @llvm.masked.gather.v2f16(<2 x half*>, i32, <2 x i1>, <2 x half>)
+declare <4 x half> @llvm.masked.gather.v4f16(<4 x half*>, i32, <4 x i1>, <4 x half>)
+declare <8 x half> @llvm.masked.gather.v8f16(<8 x half*>, i32, <8 x i1>, <8 x half>)
+declare <16 x half> @llvm.masked.gather.v16f16(<16 x half*>, i32, <16 x i1>, <16 x half>)
+declare <32 x half> @llvm.masked.gather.v32f16(<32 x half*>, i32, <32 x i1>, <32 x half>)
+
+declare <2 x float> @llvm.masked.gather.v2f32(<2 x float*>, i32, <2 x i1>, <2 x float>)
+declare <4 x float> @llvm.masked.gather.v4f32(<4 x float*>, i32, <4 x i1>, <4 x float>)
+declare <8 x float> @llvm.masked.gather.v8f32(<8 x float*>, i32, <8 x i1>, <8 x float>)
+declare <16 x float> @llvm.masked.gather.v16f32(<16 x float*>, i32, <16 x i1>, <16 x float>)
+declare <32 x float> @llvm.masked.gather.v32f32(<32 x float*>, i32, <32 x i1>, <32 x float>)
+
+declare <1 x double> @llvm.masked.gather.v1f64(<1 x double*>, i32, <1 x i1>, <1 x double>)
+declare <2 x double> @llvm.masked.gather.v2f64(<2 x double*>, i32, <2 x i1>, <2 x double>)
+declare <4 x double> @llvm.masked.gather.v4f64(<4 x double*>, i32, <4 x i1>, <4 x double>)
+declare <8 x double> @llvm.masked.gather.v8f64(<8 x double*>, i32, <8 x i1>, <8 x double>)
+declare <16 x double> @llvm.masked.gather.v16f64(<16 x double*>, i32, <16 x i1>, <16 x double>)
+declare <32 x double> @llvm.masked.gather.v32f64(<32 x double*>, i32, <32 x i1>, <32 x double>)
+
+attributes #0 = { "target-features"="+sve" }

llvm/test/CodeGen/AArch64/sve-fixed-length-masked-loads.ll

 }
 
 define <8 x float> @masked_load_v8f32(<8 x float>* %ap, <8 x float>* %bp) #0 {
 ; CHECK-LABEL: masked_load_v8f32:
 ; CHECK: ptrue [[PG0:p[0-9]+]].s, vl[[#min(div(VBYTES,4),8)]]
 ; CHECK-NEXT: ld1w { [[Z0:z[0-9]+]].s }, p0/z, [x0]
 ; CHECK-NEXT: ld1w { [[Z1:z[0-9]+]].s }, p0/z, [x1]
 ; CHECK-NEXT: fcmeq [[PG1:p[0-9]+]].s, [[PG0]]/z, [[Z0]].s, [[Z1]].s
-; CHECK-NEXT: mov [[Z0]].s, [[PG1]]/z, #-1
-; CHECK-NEXT: cmpne [[PG1]].s, [[PG0]]/z, [[Z0]].s, #0
 ; CHECK-NEXT: ld1w { [[Z0]].s }, [[PG1]]/z, [x0]
 ; CHECK-NEXT: st1w { [[Z0]].s }, [[PG0]], [x8]
 ; CHECK-NEXT: ret
   %a = load <8 x float>, <8 x float>* %ap
   %b = load <8 x float>, <8 x float>* %bp
   %mask = fcmp oeq <8 x float> %a, %b
   %load = call <8 x float> @llvm.masked.load.v8f32(<8 x float>* %ap, i32 8, <8 x i1> %mask, <8 x float> zeroinitializer)
   ret <8 x float> %load
 }
 
 define <16 x float> @masked_load_v16f32(<16 x float>* %ap, <16 x float>* %bp) #0 {
 ; CHECK-LABEL: masked_load_v16f32:
 ; VBITS_GE_512: ptrue [[PG0:p[0-9]+]].s, vl[[#min(div(VBYTES,4),16)]]
 ; VBITS_GE_512-NEXT: ld1w { [[Z0:z[0-9]+]].s }, p0/z, [x0]
 ; VBITS_GE_512-NEXT: ld1w { [[Z1:z[0-9]+]].s }, p0/z, [x1]
 ; VBITS_GE_512-NEXT: fcmeq [[PG1:p[0-9]+]].s, [[PG0]]/z, [[Z0]].s, [[Z1]].s
-; VBITS_GE_512-NEXT: mov [[Z0]].s, [[PG1]]/z, #-1
-; VBITS_GE_512-NEXT: cmpne [[PG1]].s, [[PG0]]/z, [[Z0]].s, #0
 ; VBITS_GE_512-NEXT: ld1w { [[Z0]].s }, [[PG1]]/z, [x{{[0-9]+}}]
 ; VBITS_GE_512-NEXT: st1w { [[Z0]].s }, [[PG0]], [x8]
 ; VBITS_GE_512-NEXT: ret
   %a = load <16 x float>, <16 x float>* %ap
   %b = load <16 x float>, <16 x float>* %bp
   %mask = fcmp oeq <16 x float> %a, %b
   %load = call <16 x float> @llvm.masked.load.v16f32(<16 x float>* %ap, i32 8, <16 x i1> %mask, <16 x float> zeroinitializer)
   ret <16 x float> %load
 }
 
 define <32 x float> @masked_load_v32f32(<32 x float>* %ap, <32 x float>* %bp) #0 {
 ; CHECK-LABEL: masked_load_v32f32:
 ; VBITS_GE_1024: ptrue [[PG0:p[0-9]+]].s, vl[[#min(div(VBYTES,4),32)]]
 ; VBITS_GE_1024-NEXT: ld1w { [[Z0:z[0-9]+]].s }, p0/z, [x0]
 ; VBITS_GE_1024-NEXT: ld1w { [[Z1:z[0-9]+]].s }, p0/z, [x1]
 ; VBITS_GE_1024-NEXT: fcmeq [[PG1:p[0-9]+]].s, [[PG0]]/z, [[Z0]].s, [[Z1]].s
-; VBITS_GE_1024-NEXT: mov [[Z0]].s, [[PG1]]/z, #-1
-; VBITS_GE_1024-NEXT: cmpne [[PG1]].s, [[PG0]]/z, [[Z0]].s, #0
 ; VBITS_GE_1024-NEXT: ld1w { [[Z0]].s }, [[PG1]]/z, [x{{[0-9]+}}]
 ; VBITS_GE_1024-NEXT: st1w { [[Z0]].s }, [[PG0]], [x8]
 ; VBITS_GE_1024-NEXT: ret
   %a = load <32 x float>, <32 x float>* %ap
   %b = load <32 x float>, <32 x float>* %bp
   %mask = fcmp oeq <32 x float> %a, %b
   %load = call <32 x float> @llvm.masked.load.v32f32(<32 x float>* %ap, i32 8, <32 x i1> %mask, <32 x float> zeroinitializer)
   ret <32 x float> %load
 }
 
 define <64 x float> @masked_load_v64f32(<64 x float>* %ap, <64 x float>* %bp) #0 {
 ; CHECK-LABEL: masked_load_v64f32:
 ; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].s, vl[[#min(div(VBYTES,4),64)]]
 ; VBITS_GE_2048-NEXT: ld1w { [[Z0:z[0-9]+]].s }, p0/z, [x0]
 ; VBITS_GE_2048-NEXT: ld1w { [[Z1:z[0-9]+]].s }, p0/z, [x1]
 ; VBITS_GE_2048-NEXT: fcmeq [[PG1:p[0-9]+]].s, [[PG0]]/z, [[Z0]].s, [[Z1]].s
-; VBITS_GE_2048-NEXT: mov [[Z0]].s, [[PG1]]/z, #-1
-; VBITS_GE_2048-NEXT: cmpne [[PG1]].s, [[PG0]]/z, [[Z0]].s, #0
 ; VBITS_GE_2048-NEXT: ld1w { [[Z0]].s }, [[PG1]]/z, [x{{[0-9]+}}]
 ; VBITS_GE_2048-NEXT: st1w { [[Z0]].s }, [[PG0]], [x8]
 ; VBITS_GE_2048-NEXT: ret
 
   %a = load <64 x float>, <64 x float>* %ap
   %b = load <64 x float>, <64 x float>* %bp
   %mask = fcmp oeq <64 x float> %a, %b
   %load = call <64 x float> @llvm.masked.load.v64f32(<64 x float>* %ap, i32 8, <64 x i1> %mask, <64 x float> zeroinitializer)
   ret <64 x float> %load
 }
 
 define <64 x i8> @masked_load_v64i8(<64 x i8>* %ap, <64 x i8>* %bp) #0 {
 ; CHECK-LABEL: masked_load_v64i8:
 ; VBITS_GE_512: ptrue [[PG0:p[0-9]+]].b, vl64
 ; VBITS_GE_512-NEXT: ld1b { [[Z0:z[0-9]+]].b }, p0/z, [x0]
 ; VBITS_GE_512-NEXT: ld1b { [[Z1:z[0-9]+]].b }, p0/z, [x1]
 ; VBITS_GE_512-NEXT: cmpeq [[PG1:p[0-9]+]].b, [[PG0]]/z, [[Z0]].b, [[Z1]].b
-; VBITS_GE_512-NEXT: mov [[Z0]].b, [[PG1]]/z, #-1
-; VBITS_GE_512-NEXT: cmpne [[PG1]].b, [[PG0]]/z, [[Z0]].b, #0
 ; VBITS_GE_512-NEXT: ld1b { [[Z0]].b }, [[PG1]]/z, [x{{[0-9]+}}]
 ; VBITS_GE_512-NEXT: st1b { [[Z0]].b }, [[PG0]], [x8]
 ; VBITS_GE_512-NEXT: ret
   %a = load <64 x i8>, <64 x i8>* %ap
   %b = load <64 x i8>, <64 x i8>* %bp
   %mask = icmp eq <64 x i8> %a, %b
   %load = call <64 x i8> @llvm.masked.load.v64i8(<64 x i8>* %ap, i32 8, <64 x i1> %mask, <64 x i8> undef)
   ret <64 x i8> %load
 }
 
 define <32 x i16> @masked_load_v32i16(<32 x i16>* %ap, <32 x i16>* %bp) #0 {
 ; CHECK-LABEL: masked_load_v32i16:
 ; VBITS_GE_512: ptrue [[PG0:p[0-9]+]].h, vl32
 ; VBITS_GE_512-NEXT: ld1h { [[Z0:z[0-9]+]].h }, p0/z, [x0]
 ; VBITS_GE_512-NEXT: ld1h { [[Z1:z[0-9]+]].h }, p0/z, [x1]
 ; VBITS_GE_512-NEXT: cmpeq [[PG1:p[0-9]+]].h, [[PG0]]/z, [[Z0]].h, [[Z1]].h
-; VBITS_GE_512-NEXT: mov [[Z0]].h, [[PG1]]/z, #-1
-; VBITS_GE_512-NEXT: cmpne [[PG1]].h, [[PG0]]/z, [[Z0]].h, #0
 ; VBITS_GE_512-NEXT: ld1h { [[Z0]].h }, [[PG1]]/z, [x{{[0-9]+}}]
 ; VBITS_GE_512-NEXT: st1h { [[Z0]].h }, [[PG0]], [x8]
 ; VBITS_GE_512: ret
   %a = load <32 x i16>, <32 x i16>* %ap
   %b = load <32 x i16>, <32 x i16>* %bp
   %mask = icmp eq <32 x i16> %a, %b
   %load = call <32 x i16> @llvm.masked.load.v32i16(<32 x i16>* %ap, i32 8, <32 x i1> %mask, <32 x i16> undef)
   ret <32 x i16> %load
 }
 
 define <16 x i32> @masked_load_v16i32(<16 x i32>* %ap, <16 x i32>* %bp) #0 {
 ; CHECK-LABEL: masked_load_v16i32:
 ; VBITS_GE_512: ptrue [[PG0:p[0-9]+]].s, vl16
 ; VBITS_GE_512-NEXT: ld1w { [[Z0:z[0-9]+]].s }, p0/z, [x0]
 ; VBITS_GE_512-NEXT: ld1w { [[Z1:z[0-9]+]].s }, p0/z, [x1]
 ; VBITS_GE_512-NEXT: cmpeq [[PG1:p[0-9]+]].s, [[PG0]]/z, [[Z0]].s, [[Z1]].s
-; VBITS_GE_512-NEXT: mov [[Z0]].s, [[PG1]]/z, #-1
-; VBITS_GE_512-NEXT: cmpne [[PG1]].s, [[PG0]]/z, [[Z0]].s, #0
 ; VBITS_GE_512-NEXT: ld1w { [[Z0]].s }, [[PG1]]/z, [x{{[0-9]+}}]
 ; VBITS_GE_512-NEXT: st1w { [[Z0]].s }, [[PG0]], [x8]
 ; VBITS_GE_512-NEXT: ret
   %a = load <16 x i32>, <16 x i32>* %ap
   %b = load <16 x i32>, <16 x i32>* %bp
   %mask = icmp eq <16 x i32> %a, %b
   %load = call <16 x i32> @llvm.masked.load.v16i32(<16 x i32>* %ap, i32 8, <16 x i1> %mask, <16 x i32> undef)
   ret <16 x i32> %load
 }
 
 define <8 x i64> @masked_load_v8i64(<8 x i64>* %ap, <8 x i64>* %bp) #0 {
 ; CHECK-LABEL: masked_load_v8i64:
 ; VBITS_GE_512: ptrue [[PG0:p[0-9]+]].d, vl8
 ; VBITS_GE_512-NEXT: ld1d { [[Z0:z[0-9]+]].d }, p0/z, [x0]
 ; VBITS_GE_512-NEXT: ld1d { [[Z1:z[0-9]+]].d }, p0/z, [x1]
 ; VBITS_GE_512-NEXT: cmpeq [[PG1:p[0-9]+]].d, [[PG0]]/z, [[Z0]].d, [[Z1]].d
-; VBITS_GE_512-NEXT: mov [[Z0]].d, [[PG1]]/z, #-1
-; VBITS_GE_512-NEXT: cmpne [[PG1]].d, [[PG0]]/z, [[Z0]].d, #0
 ; VBITS_GE_512-NEXT: ld1d { [[Z0]].d }, [[PG1]]/z, [x{{[0-9]+}}]
 ; VBITS_GE_512-NEXT: st1d { [[Z0]].d }, [[PG0]], [x8]
 ; VBITS_GE_512-NEXT: ret
   %a = load <8 x i64>, <8 x i64>* %ap
   %b = load <8 x i64>, <8 x i64>* %bp
   %mask = icmp eq <8 x i64> %a, %b
   %load = call <8 x i64> @llvm.masked.load.v8i64(<8 x i64>* %ap, i32 8, <8 x i1> %mask, <8 x i64> undef)
   ret <8 x i64> %load
 }
 
 define <8 x i64> @masked_load_passthru_v8i64(<8 x i64>* %ap, <8 x i64>* %bp) #0 {
 ; CHECK-LABEL: masked_load_passthru_v8i64:
 ; VBITS_GE_512: ptrue [[PG0:p[0-9]+]].d, vl8
 ; VBITS_GE_512-NEXT: ld1d { [[Z0:z[0-9]+]].d }, p0/z, [x0]
 ; VBITS_GE_512-NEXT: ld1d { [[Z1:z[0-9]+]].d }, p0/z, [x1]
 ; VBITS_GE_512-NEXT: cmpeq [[PG1:p[0-9]+]].d, [[PG0]]/z, [[Z0]].d, [[Z1]].d
-; VBITS_GE_512-NEXT: mov [[Z0]].d, [[PG1]]/z, #-1
-; VBITS_GE_512-NEXT: cmpne [[PG1]].d, [[PG0]]/z, [[Z0]].d, #0
 ; VBITS_GE_512-NEXT: ld1d { [[Z0]].d }, [[PG1]]/z, [x{{[0-9]+}}]
 ; VBITS_GE_512-NEXT: sel [[Z2:z[0-9]+]].d, [[PG1]], [[Z0]].d, [[Z1]].d
 ; VBITS_GE_512-NEXT: st1d { [[Z2]].d }, [[PG0]], [x8]
 ; VBITS_GE_512-NEXT: ret
   %a = load <8 x i64>, <8 x i64>* %ap
   %b = load <8 x i64>, <8 x i64>* %bp
   %mask = icmp eq <8 x i64> %a, %b
   %load = call <8 x i64> @llvm.masked.load.v8i64(<8 x i64>* %ap, i32 8, <8 x i1> %mask, <8 x i64> %b)
   ret <8 x i64> %load
 }
 
 define <8 x double> @masked_load_passthru_v8f64(<8 x double>* %ap, <8 x double>* %bp) #0 {
 ; CHECK-LABEL: masked_load_passthru_v8f64:
 ; VBITS_GE_512: ptrue [[PG0:p[0-9]+]].d, vl8
 ; VBITS_GE_512-NEXT: ld1d { [[Z0:z[0-9]+]].d }, p0/z, [x0]
 ; VBITS_GE_512-NEXT: ld1d { [[Z1:z[0-9]+]].d }, p0/z, [x1]
 ; VBITS_GE_512-NEXT: fcmeq [[PG1:p[0-9]+]].d, [[PG0]]/z, [[Z0]].d, [[Z1]].d
-; VBITS_GE_512-NEXT: mov [[Z0]].d, [[PG1]]/z, #-1
-; VBITS_GE_512-NEXT: cmpne [[PG1]].d, [[PG0]]/z, [[Z0]].d, #0
 ; VBITS_GE_512-NEXT: ld1d { [[Z0]].d }, [[PG1]]/z, [x{{[0-9]+}}]
 ; VBITS_GE_512-NEXT: sel [[Z2:z[0-9]+]].d, [[PG1]], [[Z0]].d, [[Z1]].d
 ; VBITS_GE_512-NEXT: st1d { [[Z2]].d }, [[PG0]], [x8]
 ; VBITS_GE_512-NEXT: ret
   %a = load <8 x double>, <8 x double>* %ap
   %b = load <8 x double>, <8 x double>* %bp
   %mask = fcmp oeq <8 x double> %a, %b
   %load = call <8 x double> @llvm.masked.load.v8f64(<8 x double>* %ap, i32 8, <8 x i1> %mask, <8 x double> %b)
   ret <8 x double> %load
 }
 
 define <32 x i16> @masked_load_sext_v32i8i16(<32 x i8>* %ap, <32 x i8>* %bp) #0 {
 ; CHECK-LABEL: masked_load_sext_v32i8i16:
 ; VBITS_GE_512: ptrue [[PG0:p[0-9]+]].b, vl32
 ; VBITS_GE_512-NEXT: ld1b { [[Z0:z[0-9]+]].b }, p0/z, [x0]
 ; VBITS_GE_512-NEXT: ld1b { [[Z1:z[0-9]+]].b }, p0/z, [x1]
 ; VBITS_GE_512-NEXT: cmpeq [[PG1:p[0-9]+]].b, [[PG0]]/z, [[Z0]].b, [[Z1]].b
-; VBITS_GE_512-NEXT: mov [[Z0]].b, [[PG1]]/z, #-1
-; VBITS_GE_512-NEXT: cmpne [[PG2:p[0-9]+]].b, [[PG0]]/z, [[Z0]].b, #0
-; VBITS_GE_512-NEXT: ld1b { [[Z0]].b }, [[PG2]]/z, [x{{[0-9]+}}]
+; VBITS_GE_512-NEXT: ld1b { [[Z0]].b }, [[PG1]]/z, [x{{[0-9]+}}]
 ; VBITS_GE_512-NEXT: ptrue [[PG2:p[0-9]+]].h, vl32
 ; VBITS_GE_512-NEXT: sunpklo [[Z0]].h, [[Z0]].b
-; VBITS_GE_512-NEXT: st1h { [[Z0]].h }, [[PG2]], [x8]
+; VBITS_GE_512-NEXT: st1h { [[Z0]].h }, [[PG1]], [x8]
 ; VBITS_GE_512-NEXT: ret
   %a = load <32 x i8>, <32 x i8>* %ap
   %b = load <32 x i8>, <32 x i8>* %bp
   %mask = icmp eq <32 x i8> %a, %b
   %load = call <32 x i8> @llvm.masked.load.v32i8(<32 x i8>* %ap, i32 8, <32 x i1> %mask, <32 x i8> undef)
   %ext = sext <32 x i8> %load to <32 x i16>
   ret <32 x i16> %ext
 }

 }
 
 define <16 x i32> @masked_load_sext_v16i16i32(<16 x i16>* %ap, <16 x i16>* %bp) #0 {
 ; CHECK-LABEL: masked_load_sext_v16i16i32:
 ; VBITS_GE_512: ptrue [[PG0:p[0-9]+]].h, vl16
 ; VBITS_GE_512-NEXT: ld1h { [[Z0:z[0-9]+]].h }, p0/z, [x0]
 ; VBITS_GE_512-NEXT: ld1h { [[Z1:z[0-9]+]].h }, p0/z, [x1]
 ; VBITS_GE_512-NEXT: cmpeq [[PG1:p[0-9]+]].h, [[PG0]]/z, [[Z0]].h, [[Z1]].h
-; VBITS_GE_512-NEXT: mov [[Z0]].h, [[PG1]]/z, #-1
-; VBITS_GE_512-NEXT: cmpne [[PG2:p[0-9]+]].h, [[PG0]]/z, [[Z0]].h, #0
-; VBITS_GE_512-NEXT: ld1h { [[Z0]].h }, [[PG2]]/z, [x{{[0-9]+}}]
+; VBITS_GE_512-NEXT: ld1h { [[Z0]].h }, [[PG1]]/z, [x{{[0-9]+}}]
 ; VBITS_GE_512-NEXT: ptrue [[PG2:p[0-9]+]].s, vl16
 ; VBITS_GE_512-NEXT: sunpklo [[Z0]].s, [[Z0]].h
-; VBITS_GE_512-NEXT: st1w { [[Z0]].s }, [[PG2]], [x8]
+; VBITS_GE_512-NEXT: st1w { [[Z0]].s }, [[PG1]], [x8]
 ; VBITS_GE_512-NEXT: ret
   %a = load <16 x i16>, <16 x i16>* %ap
   %b = load <16 x i16>, <16 x i16>* %bp
   %mask = icmp eq <16 x i16> %a, %b
   %load = call <16 x i16> @llvm.masked.load.v16i16(<16 x i16>* %ap, i32 8, <16 x i1> %mask, <16 x i16> undef)
   %ext = sext <16 x i16> %load to <16 x i32>
   ret <16 x i32> %ext
 }

 }
 
 define <8 x i64> @masked_load_sext_v8i32i64(<8 x i32>* %ap, <8 x i32>* %bp) #0 {
 ; CHECK-LABEL: masked_load_sext_v8i32i64:
 ; VBITS_GE_512: ptrue [[PG0:p[0-9]+]].s, vl8
 ; VBITS_GE_512-NEXT: ld1w { [[Z0:z[0-9]+]].s }, p0/z, [x0]
 ; VBITS_GE_512-NEXT: ld1w { [[Z1:z[0-9]+]].s }, p0/z, [x1]
 ; VBITS_GE_512-NEXT: cmpeq [[PG1:p[0-9]+]].s, [[PG0]]/z, [[Z0]].s, [[Z1]].s
-; VBITS_GE_512-NEXT: mov [[Z0]].s, [[PG1]]/z, #-1
-; VBITS_GE_512-NEXT: cmpne [[PG2:p[0-9]+]].s, [[PG0]]/z, [[Z0]].s, #0
-; VBITS_GE_512-NEXT: ld1w { [[Z0]].s }, [[PG2]]/z, [x{{[0-9]+}}]
+; VBITS_GE_512-NEXT: ld1w { [[Z0]].s }, [[PG1]]/z, [x{{[0-9]+}}]
 ; VBITS_GE_512-NEXT: ptrue [[PG2:p[0-9]+]].d, vl8
 ; VBITS_GE_512-NEXT: sunpklo [[Z0]].d, [[Z0]].s
-; VBITS_GE_512-NEXT: st1d { [[Z0]].d }, [[PG2]], [x8]
+; VBITS_GE_512-NEXT: st1d { [[Z0]].d }, [[PG1]], [x8]
 ; VBITS_GE_512-NEXT: ret
   %a = load <8 x i32>, <8 x i32>* %ap
   %b = load <8 x i32>, <8 x i32>* %bp
   %mask = icmp eq <8 x i32> %a, %b
   %load = call <8 x i32> @llvm.masked.load.v8i32(<8 x i32>* %ap, i32 8, <8 x i1> %mask, <8 x i32> undef)
   %ext = sext <8 x i32> %load to <8 x i64>
   ret <8 x i64> %ext
 }
 
 define <32 x i16> @masked_load_zext_v32i8i16(<32 x i8>* %ap, <32 x i8>* %bp) #0 {
 ; CHECK-LABEL: masked_load_zext_v32i8i16:
 ; VBITS_GE_512: ptrue [[PG0:p[0-9]+]].b, vl32
 ; VBITS_GE_512-NEXT: ld1b { [[Z0:z[0-9]+]].b }, p0/z, [x0]
 ; VBITS_GE_512-NEXT: ld1b { [[Z1:z[0-9]+]].b }, p0/z, [x1]
 ; VBITS_GE_512-NEXT: cmpeq [[PG1:p[0-9]+]].b, [[PG0]]/z, [[Z0]].b, [[Z1]].b
-; VBITS_GE_512-NEXT: mov [[Z0]].b, [[PG1]]/z, #-1
-; VBITS_GE_512-NEXT: cmpne [[PG2:p[0-9]+]].b, [[PG0]]/z, [[Z0]].b, #0
-; VBITS_GE_512-NEXT: ld1b { [[Z0]].b }, [[PG2]]/z, [x{{[0-9]+}}]
+; VBITS_GE_512-NEXT: ld1b { [[Z0]].b }, [[PG1]]/z, [x{{[0-9]+}}]
 ; VBITS_GE_512-NEXT: ptrue [[PG2:p[0-9]+]].h, vl32
 ; VBITS_GE_512-NEXT: uunpklo [[Z0]].h, [[Z0]].b
-; VBITS_GE_512-NEXT: st1h { [[Z0]].h }, [[PG2]], [x8]
+; VBITS_GE_512-NEXT: st1h { [[Z0]].h }, [[PG1]], [x8]
 ; VBITS_GE_512-NEXT: ret
   %a = load <32 x i8>, <32 x i8>* %ap
   %b = load <32 x i8>, <32 x i8>* %bp
   %mask = icmp eq <32 x i8> %a, %b
   %load = call <32 x i8> @llvm.masked.load.v32i8(<32 x i8>* %ap, i32 8, <32 x i1> %mask, <32 x i8> undef)
   %ext = zext <32 x i8> %load to <32 x i16>
   ret <32 x i16> %ext
 }

 }
 
 define <16 x i32> @masked_load_zext_v16i16i32(<16 x i16>* %ap, <16 x i16>* %bp) #0 {
 ; CHECK-LABEL: masked_load_zext_v16i16i32:
 ; VBITS_GE_512: ptrue [[PG0:p[0-9]+]].h, vl16
 ; VBITS_GE_512-NEXT: ld1h { [[Z0:z[0-9]+]].h }, p0/z, [x0]
 ; VBITS_GE_512-NEXT: ld1h { [[Z1:z[0-9]+]].h }, p0/z, [x1]
 ; VBITS_GE_512-NEXT: cmpeq [[PG1:p[0-9]+]].h, [[PG0]]/z, [[Z0]].h, [[Z1]].h
-; VBITS_GE_512-NEXT: mov [[Z0]].h, [[PG1]]/z, #-1
-; VBITS_GE_512-NEXT: cmpne [[PG2:p[0-9]+]].h, [[PG0]]/z, [[Z0]].h, #0
-; VBITS_GE_512-NEXT: ld1h { [[Z0]].h }, [[PG2]]/z, [x{{[0-9]+}}]
+; VBITS_GE_512-NEXT: ld1h { [[Z0]].h }, [[PG1]]/z, [x{{[0-9]+}}]
 ; VBITS_GE_512-NEXT: ptrue [[PG2:p[0-9]+]].s, vl16
 ; VBITS_GE_512-NEXT: uunpklo [[Z0]].s, [[Z0]].h
-; VBITS_GE_512-NEXT: st1w { [[Z0]].s }, [[PG2]], [x8]
+; VBITS_GE_512-NEXT: st1w { [[Z0]].s }, [[PG1]], [x8]
 ; VBITS_GE_512-NEXT: ret
   %a = load <16 x i16>, <16 x i16>* %ap
   %b = load <16 x i16>, <16 x i16>* %bp
   %mask = icmp eq <16 x i16> %a, %b
   %load = call <16 x i16> @llvm.masked.load.v16i16(<16 x i16>* %ap, i32 8, <16 x i1> %mask, <16 x i16> undef)
   %ext = zext <16 x i16> %load to <16 x i32>
   ret <16 x i32> %ext
 }

 }
 
 define <8 x i64> @masked_load_zext_v8i32i64(<8 x i32>* %ap, <8 x i32>* %bp) #0 {
 ; CHECK-LABEL: masked_load_zext_v8i32i64:
 ; VBITS_GE_512: ptrue [[PG0:p[0-9]+]].s, vl8
 ; VBITS_GE_512-NEXT: ld1w { [[Z0:z[0-9]+]].s }, p0/z, [x0]
 ; VBITS_GE_512-NEXT: ld1w { [[Z1:z[0-9]+]].s }, p0/z, [x1]
 ; VBITS_GE_512-NEXT: cmpeq [[PG1:p[0-9]+]].s, [[PG0]]/z, [[Z0]].s, [[Z1]].s
-; VBITS_GE_512-NEXT: mov [[Z0]].s, [[PG1]]/z, #-1
-; VBITS_GE_512-NEXT: cmpne [[PG2:p[0-9]+]].s, [[PG0]]/z, [[Z0]].s, #0
-; VBITS_GE_512-NEXT: ld1w { [[Z0]].s }, [[PG2]]/z, [x{{[0-9]+}}]
+; VBITS_GE_512-NEXT: ld1w { [[Z0]].s }, [[PG1]]/z, [x{{[0-9]+}}]
 ; VBITS_GE_512-NEXT: ptrue [[PG2:p[0-9]+]].d, vl8
 ; VBITS_GE_512-NEXT: uunpklo [[Z0]].d, [[Z0]].s
-; VBITS_GE_512-NEXT: st1d { [[Z0]].d }, [[PG2]], [x8]
+; VBITS_GE_512-NEXT: st1d { [[Z0]].d }, [[PG1]], [x8]
 ; VBITS_GE_512-NEXT: ret
   %a = load <8 x i32>, <8 x i32>* %ap
   %b = load <8 x i32>, <8 x i32>* %bp
   %mask = icmp eq <8 x i32> %a, %b
   %load = call <8 x i32> @llvm.masked.load.v8i32(<8 x i32>* %ap, i32 8, <8 x i1> %mask, <8 x i32> undef)
   %ext = zext <8 x i32> %load to <8 x i64>
   ret <8 x i64> %ext
 }

llvm/test/CodeGen/AArch64/sve-fixed-length-masked-scatter.ll

+; RUN: llc -aarch64-sve-vector-bits-min=128  -asm-verbose=0 < %s | FileCheck %s -check-prefix=NO_SVE
+; RUN: llc -aarch64-sve-vector-bits-min=256  -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK
+; RUN: llc -aarch64-sve-vector-bits-min=384  -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK
+; RUN: llc -aarch64-sve-vector-bits-min=512  -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=640  -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=768  -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=896  -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=1024 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_1024,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=1152 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_1024,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=1280 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_1024,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=1408 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_1024,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=1536 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_1024,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=1664 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_1024,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=1792 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_1024,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=1920 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_1024,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=2048 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_2048,VBITS_GE_1024,VBITS_GE_512
+
+target triple = "aarch64-unknown-linux-gnu"
+
+; Don't use SVE when its registers are no bigger than NEON.
+; NO_SVE-NOT: ptrue
+
+;
+; ST1B
+;
+
+define void @masked_scatter_v2i8(<2 x i8>* %a, <2 x i8*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v2i8:
+; CHECK: ldrb [[VALS_LO:w[0-9]+]], [x0]
+; CHECK-NEXT: ldrb [[VALS_HI:w[0-9]+]], [x0, #1]
+; CHECK-NEXT: ldr q[[PTRS:[0-9]+]], [x1]
+; CHECK-NEXT: ptrue [[PG0:p[0-9]+]].s, vl2
+; CHECK-NEXT: fmov s[[VALS:[0-9]+]], [[VALS_LO]]
+; CHECK-NEXT: mov v[[VALS]].s[1], [[VALS_HI]]
+; CHECK-NEXT: cmeq v[[CMP:[0-9]+]].2s, v[[VALS]].2s, #0
+; CHECK-NEXT: cmpne [[MASK:p[0-9]+]].s, [[PG0]]/z, z[[CMP]].s, #0
+; CHECK-NEXT: ushll v[[SHL:[0-9]+]].2d, v[[VALS]].2s, #0
+; CHECK-NEXT: st1b { z[[SHL]].d }, [[MASK]], [z[[PTRS]].d]
+; CHECK-NEXT: ret
+  %vals = load <2 x i8>, <2 x i8>* %a
+  %ptrs = load <2 x i8*>, <2 x i8*>* %b
+  %mask = icmp eq <2 x i8> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v2i8(<2 x i8> %vals, <2 x i8*> %ptrs, i32 8, <2 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v4i8(<4 x i8>* %a, <4 x i8*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v4i8:
+; CHECK: ldrb [[VALS_LO1:w[0-9]+]], [x0]
+; CHECK-NEXT: ldrb [[VALS_LO2:w[0-9]+]], [x0, #1]
+; CHECK-NEXT: ldrb [[VALS_HI1:w[0-9]+]], [x0, #2]
+; CHECK-NEXT: ldrb [[VALS_HI2:w[0-9]+]], [x0, #3]
+; CHECK-NEXT: fmov s[[VALS:[0-9]+]], [[VALS_LO1]]
+; CHECK-NEXT: ptrue [[PG0:p[0-9]+]].d, vl4
+; CHECK-NEXT: mov v[[VALS]].h[1], [[VALS_LO2]]
+; CHECK-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; CHECK-NEXT: mov v[[VALS]].h[2], [[VALS_HI1]]
+; CHECK-NEXT: mov v[[VALS]].h[3], [[VALS_HI2]]
+; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].h, vl4
+; CHECK-NEXT: cmeq v[[CMP:[0-9]+]].4h, v[[VALS]].4h, #0
+; CHECK-NEXT: uunpklo [[UPK1:z[0-9]+]].s, z[[VALS]].h
+; CHECK-NEXT: cmpne [[MASK:p[0-9]+]].h, [[PG1]]/z, z[[CMP]].h, #0
+; CHECK-NEXT: uunpklo [[UPK2:z[0-9]+]].d, [[UPK1]].s
+; CHECK-NEXT: st1b { [[UPK2]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; CHECK-NEXT: ret
+  %vals = load <4 x i8>, <4 x i8>* %a
+  %ptrs = load <4 x i8*>, <4 x i8*>* %b
+  %mask = icmp eq <4 x i8> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v4i8(<4 x i8> %vals, <4 x i8*> %ptrs, i32 8, <4 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v8i8(<8 x i8>* %a, <8 x i8*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v8i8:
+; VBITS_GE_512: ldr d[[VALS:[0-9]+]], [x0]
+; VBITS_GE_512-NEXT: ptrue [[PG0:p[0-9]+]].d, vl8
+; VBITS_GE_512-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; VBITS_GE_512-NEXT: ptrue [[PG1:p[0-9]+]].b, vl8
+; VBITS_GE_512-NEXT: cmeq v[[CMP:[0-9]+]].8b, v[[VALS]].8b, #0
+; VBITS_GE_512-NEXT: uunpklo [[UPK1:z[0-9]+]].h, z[[VALS]].b
+; VBITS_GE_512-NEXT: uunpklo [[UPK2:z[0-9]+]].s, [[UPK1]].h
+; VBITS_GE_512-NEXT: cmpne [[MASK:p[0-9]+]].b, [[PG1]]/z, z[[CMP]].b, #0
+; VBITS_GE_512-NEXT: uunpklo [[UPK3:z[0-9]+]].d, [[UPK2]].s
+; VBITS_GE_512-NEXT: st1b { [[UPK3]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; VBITS_BE_512-NEXT: ret
+  %vals = load <8 x i8>, <8 x i8>* %a
+  %ptrs = load <8 x i8*>, <8 x i8*>* %b
+  %mask = icmp eq <8 x i8> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v8i8(<8 x i8> %vals, <8 x i8*> %ptrs, i32 8, <8 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v16i8(<16 x i8>* %a, <16 x i8*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v16i8:
+; VBITS_GE_1024: ldr q[[VALS:[0-9]+]], [x0]
+; VBITS_GE_1024-NEXT: ptrue [[PG0:p[0-9]+]].d, vl16
+; VBITS_GE_1024-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; VBITS_GE_1024-NEXT: ptrue [[PG1:p[0-9]+]].b, vl16
+; VBITS_GE_1024-NEXT: cmeq v[[CMP:[0-9]+]].16b, v[[VALS]].16b, #0
+; VBITS_GE_1024-NEXT: uunpklo [[UPK1:z[0-9]+]].h, z[[VALS]].b
+; VBITS_GE_1024-NEXT: uunpklo [[UPK2:z[0-9]+]].s, [[UPK1]].h
+; VBITS_GE_1024-NEXT: cmpne [[MASK:p[0-9]+]].b, [[PG1]]/z, z[[CMP]].b, #0
+; VBITS_GE_1024-NEXT: uunpklo [[UPK3:z[0-9]+]].d, [[UPK2]].s
+; VBITS_GE_1024-NEXT: st1b { [[UPK3]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; VBITS_BE_1024-NEXT: ret
+  %vals = load <16 x i8>, <16 x i8>* %a
+  %ptrs = load <16 x i8*>, <16 x i8*>* %b
+  %mask = icmp eq <16 x i8> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v16i8(<16 x i8> %vals, <16 x i8*> %ptrs, i32 8, <16 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v32i8(<32 x i8>* %a, <32 x i8*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v32i8:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].b, vl32
+; VBITS_GE_2048-NEXT: ld1b { [[VALS:z[0-9]+]].b }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].d, vl32
+; VBITS_GE_2048-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: cmpeq [[MASK:p[0-9]+]].b, [[PG0]]/z, [[VALS]].b, #0
+; VBITS_GE_2048-NEXT: uunpklo [[UPK1:z[0-9]+]].h, [[VALS]].b
+; VBITS_GE_2048-NEXT: uunpklo [[UPK2:z[0-9]+]].s, [[UPK1]].h
+; VBITS_GE_2048-NEXT: uunpklo [[UPK3:z[0-9]+]].d, [[UPK2]].s
+; VBITS_GE_2048-NEXT: st1b { [[UPK3]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; VBITS_BE_2048-NEXT: ret
+  %vals = load <32 x i8>, <32 x i8>* %a
+  %ptrs = load <32 x i8*>, <32 x i8*>* %b
+  %mask = icmp eq <32 x i8> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v32i8(<32 x i8> %vals, <32 x i8*> %ptrs, i32 8, <32 x i1> %mask)
+  ret void
+}
+
+;
+; ST1H
+;
+
+define void @masked_scatter_v2i16(<2 x i16>* %a, <2 x i16*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v2i16:
+; CHECK: ldrh [[VALS_LO:w[0-9]+]], [x0]
+; CHECK-NEXT: ldrh [[VALS_HI:w[0-9]+]], [x0, #2]
+; CHECK-NEXT: ldr q[[PTRS:[0-9]+]], [x1]
+; CHECK-NEXT: ptrue [[PG0:p[0-9]+]].s, vl2
+; CHECK-NEXT: fmov s[[VALS:[0-9]+]], [[VALS_LO]]
+; CHECK-NEXT: mov v[[VALS]].s[1], [[VALS_HI]]
+; CHECK-NEXT: cmeq v[[CMP:[0-9]+]].2s, v[[VALS]].2s, #0
+; CHECK-NEXT: cmpne [[MASK:p[0-9]+]].s, [[PG0]]/z, z[[CMP]].s, #0
+; CHECK-NEXT: ushll v[[SHL:[0-9]+]].2d, v[[VALS]].2s, #0
+; CHECK-NEXT: st1h { z[[SHL]].d }, [[MASK]], [z[[PTRS]].d]
+; CHECK-NEXT: ret
+  %vals = load <2 x i16>, <2 x i16>* %a
+  %ptrs = load <2 x i16*>, <2 x i16*>* %b
+  %mask = icmp eq <2 x i16> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v2i16(<2 x i16> %vals, <2 x i16*> %ptrs, i32 8, <2 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v4i16(<4 x i16>* %a, <4 x i16*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v4i16:
+; CHECK: ldr d[[VALS:[0-9]+]], [x0]
+; CHECK-NEXT: ptrue [[PG0:p[0-9]+]].d, vl4
+; CHECK-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].h, vl4
+; CHECK-NEXT: cmeq v[[CMP:[0-9]+]].4h, v[[VALS]].4h, #0
+; CHECK-NEXT: uunpklo [[UPK1:z[0-9]+]].s, z[[VALS]].h
+; CHECK-NEXT: cmpne [[MASK:p[0-9]+]].h, [[PG1]]/z, z[[CMP]].h, #0
+; CHECK-NEXT: uunpklo [[UPK2:z[0-9]+]].d, [[UPK1]].s
+; CHECK-NEXT: st1h { [[UPK2]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; CHECK-NEXT: ret
+  %vals = load <4 x i16>, <4 x i16>* %a
+  %ptrs = load <4 x i16*>, <4 x i16*>* %b
+  %mask = icmp eq <4 x i16> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v4i16(<4 x i16> %vals, <4 x i16*> %ptrs, i32 8, <4 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v8i16(<8 x i16>* %a, <8 x i16*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v8i16:
+; VBITS_GE_512: ldr q[[VALS:[0-9]+]], [x0]
+; VBITS_GE_512-NEXT: ptrue [[PG0:p[0-9]+]].d, vl8
+; VBITS_GE_512-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; VBITS_GE_512-NEXT: ptrue [[PG1:p[0-9]+]].h, vl8
+; VBITS_GE_512-NEXT: cmeq v[[CMP:[0-9]+]].8h, v[[VALS]].8h, #0
+; VBITS_GE_512-NEXT: uunpklo [[UPK1:z[0-9]+]].s, z[[VALS]].h
+; VBITS_GE_512-NEXT: cmpne [[MASK:p[0-9]+]].h, [[PG1]]/z, z[[CMP]].h, #0
+; VBITS_GE_512-NEXT: uunpklo [[UPK2:z[0-9]+]].d, [[UPK1]].s
+; VBITS_GE_512-NEXT: st1h { [[UPK2]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; VBITS_BE_512-NEXT: ret
+  %vals = load <8 x i16>, <8 x i16>* %a
+  %ptrs = load <8 x i16*>, <8 x i16*>* %b
+  %mask = icmp eq <8 x i16> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v8i16(<8 x i16> %vals, <8 x i16*> %ptrs, i32 8, <8 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v16i16(<16 x i16>* %a, <16 x i16*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v16i16:
+; VBITS_GE_1024: ptrue [[PG0:p[0-9]+]].h, vl16
+; VBITS_GE_1024-NEXT: ld1h { [[VALS:z[0-9]+]].h }, [[PG0]]/z, [x0]
+; VBITS_GE_1024-NEXT: ptrue [[PG1:p[0-9]+]].d, vl16
+; VBITS_GE_1024-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_1024-NEXT: cmpeq [[MASK:p[0-9]+]].h, [[PG0]]/z, [[VALS]].h, #0
+; VBITS_GE_1024-NEXT: uunpklo [[UPK1:z[0-9]+]].s, [[VALS]].h
+; VBITS_GE_1024-NEXT: uunpklo [[UPK2:z[0-9]+]].d, [[UPK1]].s
+; VBITS_GE_1024-NEXT: st1h { [[UPK2]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; VBITS_BE_1024-NEXT: ret
+  %vals = load <16 x i16>, <16 x i16>* %a
+  %ptrs = load <16 x i16*>, <16 x i16*>* %b
+  %mask = icmp eq <16 x i16> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v16i16(<16 x i16> %vals, <16 x i16*> %ptrs, i32 8, <16 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v32i16(<32 x i16>* %a, <32 x i16*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v32i16:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].h, vl32
+; VBITS_GE_2048-NEXT: ld1h { [[VALS:z[0-9]+]].h }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].d, vl32
+; VBITS_GE_2048-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: cmpeq [[MASK:p[0-9]+]].h, [[PG0]]/z, [[VALS]].h, #0
+; VBITS_GE_2048-NEXT: uunpklo [[UPK1:z[0-9]+]].s, [[VALS]].h
+; VBITS_GE_2048-NEXT: uunpklo [[UPK2:z[0-9]+]].d, [[UPK1]].s
+; VBITS_GE_2048-NEXT: st1h { [[UPK2]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; VBITS_BE_2048-NEXT: ret
+  %vals = load <32 x i16>, <32 x i16>* %a
+  %ptrs = load <32 x i16*>, <32 x i16*>* %b
+  %mask = icmp eq <32 x i16> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v32i16(<32 x i16> %vals, <32 x i16*> %ptrs, i32 8, <32 x i1> %mask)
+  ret void
+}
+
+;
+; ST1W
+;
+
+define void @masked_scatter_v2i32(<2 x i32>* %a, <2 x i32*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v2i32:
+; CHECK: ldr d[[VALS:[0-9]+]], [x0]
+; CHECK-NEXT: ldr q[[PTRS:[0-9]+]], [x1]
+; CHECK-NEXT: ptrue [[PG0:p[0-9]+]].s, vl2
+; CHECK-NEXT: cmeq v[[CMP:[0-9]+]].2s, v[[VALS]].2s, #0
+; CHECK-NEXT: cmpne [[MASK:p[0-9]+]].s, [[PG0]]/z, z[[CMP]].s, #0
+; CHECK-NEXT: ushll v[[SHL:[0-9]+]].2d, v[[VALS]].2s, #0
+; CHECK-NEXT: st1w { z[[SHL]].d }, [[MASK]], [z[[PTRS]].d]
+; CHECK-NEXT: ret
+  %vals = load <2 x i32>, <2 x i32>* %a
+  %ptrs = load <2 x i32*>, <2 x i32*>* %b
+  %mask = icmp eq <2 x i32> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v2i32(<2 x i32> %vals, <2 x i32*> %ptrs, i32 8, <2 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v4i32(<4 x i32>* %a, <4 x i32*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v4i32:
+; CHECK: ldr q[[VALS:[0-9]+]], [x0]
+; CHECK-NEXT: ptrue [[PG0:p[0-9]+]].d, vl4
+; CHECK-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].s, vl4
+; CHECK-NEXT: cmeq v[[CMP:[0-9]+]].4s, v[[VALS]].4s, #0
+; CHECK-NEXT: cmpne [[MASK:p[0-9]+]].s, [[PG1]]/z, z[[CMP]].s, #0
+; CHECK-NEXT: uunpklo [[UPK:z[0-9]+]].d, z[[VALS]].s
+; CHECK-NEXT: st1w { [[UPK]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; CHECK-NEXT: ret
+  %vals = load <4 x i32>, <4 x i32>* %a
+  %ptrs = load <4 x i32*>, <4 x i32*>* %b
+  %mask = icmp eq <4 x i32> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v4i32(<4 x i32> %vals, <4 x i32*> %ptrs, i32 8, <4 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v8i32(<8 x i32>* %a, <8 x i32*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v8i32:
+; VBITS_GE_512: ptrue [[PG0:p[0-9]+]].s, vl8
+; VBITS_GE_512-NEXT: ld1w { [[VALS:z[0-9]+]].s }, [[PG0]]/z, [x0]
+; VBITS_GE_512-NEXT: ptrue [[PG1:p[0-9]+]].d, vl8
+; VBITS_GE_512-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_512-NEXT: cmpeq [[MASK:p[0-9]+]].s, [[PG0]]/z, [[VALS]].s, #0
+; VBITS_GE_512-NEXT: uunpklo [[UPK:z[0-9]+]].d, [[VALS]].s
+; VBITS_GE_512-NEXT: st1w { [[UPK]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; VBITS_BE_512-NEXT: ret
+  %vals = load <8 x i32>, <8 x i32>* %a
+  %ptrs = load <8 x i32*>, <8 x i32*>* %b
+  %mask = icmp eq <8 x i32> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v8i32(<8 x i32> %vals, <8 x i32*> %ptrs, i32 8, <8 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v16i32(<16 x i32>* %a, <16 x i32*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v16i32:
+; VBITS_GE_1024: ptrue [[PG0:p[0-9]+]].s, vl16
+; VBITS_GE_1024-NEXT: ld1w { [[VALS:z[0-9]+]].s }, [[PG0]]/z, [x0]
+; VBITS_GE_1024-NEXT: ptrue [[PG1:p[0-9]+]].d, vl16
+; VBITS_GE_1024-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_1024-NEXT: cmpeq [[MASK:p[0-9]+]].s, [[PG0]]/z, [[VALS]].s, #0
+; VBITS_GE_1024-NEXT: uunpklo [[UPK:z[0-9]+]].d, [[VALS]].s
+; VBITS_GE_1024-NEXT: st1w { [[UPK]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; VBITS_BE_1024-NEXT: ret
+  %vals = load <16 x i32>, <16 x i32>* %a
+  %ptrs = load <16 x i32*>, <16 x i32*>* %b
+  %mask = icmp eq <16 x i32> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v16i32(<16 x i32> %vals, <16 x i32*> %ptrs, i32 8, <16 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v32i32(<32 x i32>* %a, <32 x i32*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v32i32:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].s, vl32
+; VBITS_GE_2048-NEXT: ld1w { [[VALS:z[0-9]+]].s }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].d, vl32
+; VBITS_GE_2048-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: cmpeq [[MASK:p[0-9]+]].s, [[PG0]]/z, [[VALS]].s, #0
+; VBITS_GE_2048-NEXT: uunpklo [[UPK:z[0-9]+]].d, [[VALS]].s
+; VBITS_GE_2048-NEXT: st1w { [[UPK]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; VBITS_BE_2048-NEXT: ret
+  %vals = load <32 x i32>, <32 x i32>* %a
+  %ptrs = load <32 x i32*>, <32 x i32*>* %b
+  %mask = icmp eq <32 x i32> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v32i32(<32 x i32> %vals, <32 x i32*> %ptrs, i32 8, <32 x i1> %mask)
+  ret void
+}
+
+;
+; ST1D
+;
+
+; Scalarize 1 x i64 scatters
+define void @masked_scatter_v1i64(<1 x i64>* %a, <1 x i64*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v1i64:
+; CHECK-NOT: ptrue
+  %vals = load <1 x i64>, <1 x i64>* %a
+  %ptrs = load <1 x i64*>, <1 x i64*>* %b
+  %mask = icmp eq <1 x i64> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v1i64(<1 x i64> %vals, <1 x i64*> %ptrs, i32 8, <1 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v2i64(<2 x i64>* %a, <2 x i64*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v2i64:
+; CHECK: ldr q[[VALS:[0-9]+]], [x0]
+; CHECK-NEXT: ldr q[[PTRS:[0-9]+]], [x1]
+; CHECK-NEXT: ptrue [[PG0:p[0-9]+]].d, vl2
+; CHECK-NEXT: cmeq v[[CMP:[0-9]+]].2d, v[[VALS]].2d, #0
+; CHECK-NEXT: cmpne [[MASK:p[0-9]+]].d, [[PG0]]/z, z[[CMP]].d, #0
+; CHECK-NEXT: st1d { z[[VALS]].d }, [[MASK]], [z[[PTRS]].d]
+; CHECK-NEXT: ret
+  %vals = load <2 x i64>, <2 x i64>* %a
+  %ptrs = load <2 x i64*>, <2 x i64*>* %b
+  %mask = icmp eq <2 x i64> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v2i64(<2 x i64> %vals, <2 x i64*> %ptrs, i32 8, <2 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v4i64(<4 x i64>* %a, <4 x i64*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v4i64:
+; CHECK: ptrue [[PG0:p[0-9]+]].d, vl4
+; CHECK-NEXT: ld1d { [[VALS:z[0-9]+]].d }, [[PG0]]/z, [x0]
+; CHECK-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; CHECK-NEXT: cmpeq [[MASK:p[0-9]+]].d, [[PG0]]/z, [[VALS]].d, #0
+; CHECK-NEXT: st1d { [[VALS]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; CHECK-NEXT: ret
+  %vals = load <4 x i64>, <4 x i64>* %a
+  %ptrs = load <4 x i64*>, <4 x i64*>* %b
+  %mask = icmp eq <4 x i64> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v4i64(<4 x i64> %vals, <4 x i64*> %ptrs, i32 8, <4 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v8i64(<8 x i64>* %a, <8 x i64*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v8i64:
+; VBITS_GE_512: ptrue [[PG0:p[0-9]+]].d, vl8
+; VBITS_GE_512-NEXT: ld1d { [[VALS:z[0-9]+]].d }, [[PG0]]/z, [x0]
+; VBITS_GE_512-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; VBITS_GE_512-NEXT: cmpeq [[MASK:p[0-9]+]].d, [[PG0]]/z, [[VALS]].d, #0
+; VBITS_GE_512-NEXT: st1d { [[VALS]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; VBITS_BE_512-NEXT: ret
+  %vals = load <8 x i64>, <8 x i64>* %a
+  %ptrs = load <8 x i64*>, <8 x i64*>* %b
+  %mask = icmp eq <8 x i64> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v8i64(<8 x i64> %vals, <8 x i64*> %ptrs, i32 8, <8 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v16i64(<16 x i64>* %a, <16 x i64*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v16i64:
+; VBITS_GE_1024: ptrue [[PG0:p[0-9]+]].d, vl16
+; VBITS_GE_1024-NEXT: ld1d { [[VALS:z[0-9]+]].d }, [[PG0]]/z, [x0]
+; VBITS_GE_1024-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; VBITS_GE_1024-NEXT: cmpeq [[MASK:p[0-9]+]].d, [[PG0]]/z, [[VALS]].d, #0
+; VBITS_GE_1024-NEXT: st1d { [[VALS]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; VBITS_BE_1024-NEXT: ret
+  %vals = load <16 x i64>, <16 x i64>* %a
+  %ptrs = load <16 x i64*>, <16 x i64*>* %b
+  %mask = icmp eq <16 x i64> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v16i64(<16 x i64> %vals, <16 x i64*> %ptrs, i32 8, <16 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v32i64(<32 x i64>* %a, <32 x i64*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v32i64:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].d, vl32
+; VBITS_GE_2048-NEXT: ld1d { [[VALS:z[0-9]+]].d }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; VBITS_GE_2048-NEXT: cmpeq [[MASK:p[0-9]+]].d, [[PG0]]/z, [[VALS]].d, #0
+; VBITS_GE_2048-NEXT: st1d { [[VALS]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; VBITS_BE_2048-NEXT: ret
+  %vals = load <32 x i64>, <32 x i64>* %a
+  %ptrs = load <32 x i64*>, <32 x i64*>* %b
+  %mask = icmp eq <32 x i64> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v32i64(<32 x i64> %vals, <32 x i64*> %ptrs, i32 8, <32 x i1> %mask)
+  ret void
+}
+
+;
+; ST1H (float)
+;
+
+define void @masked_scatter_v2f16(<2 x half>* %a, <2 x half*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v2f16:
+; CHECK: ldr s[[VALS:[0-9]+]], [x0]
+; CHECK-NEXT: movi d2, #0000000000000000
+; CHECK-NEXT: ldr q[[PTRS:[0-9]+]], [x1]
+; CHECK-NEXT: ptrue [[PG0:p[0-9]+]].h, vl4
+; CHECK-NEXT: fcmeq v[[CMP:[0-9]+]].4h, v[[VALS]].4h, #0.0
+; CHECK-NEXT: umov w8, v[[CMP]].h[0]
+; CHECK-NEXT: umov w9, v[[CMP]].h[1]
+; CHECK-NEXT: fmov s[[CMP]], w8
+; CHECK-NEXT: mov v[[CMP]].s[1], w9
+; CHECK-NEXT: shl v[[CMP]].2s, v[[CMP]].2s, #16
+; CHECK-NEXT: sshr v[[CMP]].2s, v[[CMP]].2s, #16
+; CHECK-NEXT: fmov w9, s[[CMP]]
+; CHECK-NEXT: mov w8, v[[CMP]].s[1]
+; CHECK-NEXT: mov v[[NCMP:[0-9]+]].h[0], w9
+; CHECK-NEXT: mov v[[NCMP]].h[1], w8
+; CHECK-NEXT: shl v[[NCMP]].4h, v[[NCMP]].4h, #15
+; CHECK-NEXT: uunpklo [[UPK1]].s, z[[VALS]].h
+; CHECK-NEXT: sshr v[[NCMP]].4h, v[[NCMP]].4h, #15
+; CHECK-NEXT: cmpne [[MASK:p[0-9]+]].h, [[PG0]]/z, z[[NCMP]].h, #0
+; CHECK-NEXT: uunpklo [[UPK2]].d, [[UPK1]].s
+; CHECK-NEXT: st1h { [[UPK2]].d }, [[MASK]], [z[[PTRS]].d]
+; CHECK-NEXT: ret
+  %vals = load <2 x half>, <2 x half>* %a
+  %ptrs = load <2 x half*>, <2 x half*>* %b
+  %mask = fcmp oeq <2 x half> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v2f16(<2 x half> %vals, <2 x half*> %ptrs, i32 8, <2 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v4f16(<4 x half>* %a, <4 x half*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v4f16:
+; CHECK: ldr d[[VALS:[0-9]+]], [x0]
+; CHECK-NEXT: ptrue [[PG0:p[0-9]+]].d, vl4
+; CHECK-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].h, vl4
+; CHECK-NEXT: fcmeq v[[CMP:[0-9]+]].4h, v[[VALS]].4h, #0
+; CHECK-NEXT: uunpklo [[UPK1:z[0-9]+]].s, z[[VALS]].h
+; CHECK-NEXT: cmpne [[MASK:p[0-9]+]].h, [[PG1]]/z, z[[CMP]].h, #0
+; CHECK-NEXT: uunpklo [[UPK2:z[0-9]+]].d, [[UPK1]].s
+; CHECK-NEXT: st1h { [[UPK2]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; CHECK-NEXT: ret
+  %vals = load <4 x half>, <4 x half>* %a
+  %ptrs = load <4 x half*>, <4 x half*>* %b
+  %mask = fcmp oeq <4 x half> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v4f16(<4 x half> %vals, <4 x half*> %ptrs, i32 8, <4 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v8f16(<8 x half>* %a, <8 x half*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v8f16:
+; VBITS_GE_512: ldr q[[VALS:[0-9]+]], [x0]
+; VBITS_GE_512-NEXT: ptrue [[PG0:p[0-9]+]].d, vl8
+; VBITS_GE_512-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; VBITS_GE_512-NEXT: ptrue [[PG1:p[0-9]+]].h, vl8
+; VBITS_GE_512-NEXT: fcmeq v[[CMP:[0-9]+]].8h, v[[VALS]].8h, #0
+; VBITS_GE_512-NEXT: uunpklo [[UPK1:z[0-9]+]].s, z[[VALS]].h
+; VBITS_GE_512-NEXT: cmpne [[MASK:p[0-9]+]].h, [[PG1]]/z, z[[CMP]].h, #0
+; VBITS_GE_512-NEXT: uunpklo [[UPK2:z[0-9]+]].d, [[UPK1]].s
+; VBITS_GE_512-NEXT: st1h { [[UPK2]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; VBITS_BE_512-NEXT: ret
+  %vals = load <8 x half>, <8 x half>* %a
+  %ptrs = load <8 x half*>, <8 x half*>* %b
+  %mask = fcmp oeq <8 x half> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v8f16(<8 x half> %vals, <8 x half*> %ptrs, i32 8, <8 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v16f16(<16 x half>* %a, <16 x half*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v16f16:
+; VBITS_GE_1024: ptrue [[PG0:p[0-9]+]].h, vl16
+; VBITS_GE_1024-NEXT: ld1h { [[VALS:z[0-9]+]].h }, [[PG0]]/z, [x0]
+; VBITS_GE_1024-NEXT: ptrue [[PG1:p[0-9]+]].d, vl16
+; VBITS_GE_1024-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_1024-NEXT: mov [[ZERO:z[0-9]+]].h, #0
+; VBITS_GE_1024-NEXT: fcmeq [[MASK:p[0-9]+]].h, [[PG0]]/z, [[VALS]].h, [[ZERO]]
+; VBITS_GE_1024-NEXT: uunpklo [[UPK1:z[0-9]+]].s, [[VALS]].h
+; VBITS_GE_1024-NEXT: uunpklo [[UPK2:z[0-9]+]].d, [[UPK1]].s
+; VBITS_GE_1024-NEXT: st1h { [[UPK2]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; VBITS_BE_1024-NEXT: ret
+  %vals = load <16 x half>, <16 x half>* %a
+  %ptrs = load <16 x half*>, <16 x half*>* %b
+  %mask = fcmp oeq <16 x half> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v16f16(<16 x half> %vals, <16 x half*> %ptrs, i32 8, <16 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v32f16(<32 x half>* %a, <32 x half*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v32f16:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].h, vl32
+; VBITS_GE_2048-NEXT: ld1h { [[VALS:z[0-9]+]].h }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].d, vl32
+; VBITS_GE_2048-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: mov [[ZERO:z[0-9]+]].h, #0
+; VBITS_GE_2048-NEXT: fcmeq [[MASK:p[0-9]+]].h, [[PG0]]/z, [[VALS]].h, [[ZERO]]
+; VBITS_GE_2048-NEXT: uunpklo [[UPK1:z[0-9]+]].s, [[VALS]].h
+; VBITS_GE_2048-NEXT: uunpklo [[UPK2:z[0-9]+]].d, [[UPK1]].s
+; VBITS_GE_2048-NEXT: st1h { [[UPK2]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; VBITS_BE_2048-NEXT: ret
+  %vals = load <32 x half>, <32 x half>* %a
+  %ptrs = load <32 x half*>, <32 x half*>* %b
+  %mask = fcmp oeq <32 x half> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v32f16(<32 x half> %vals, <32 x half*> %ptrs, i32 8, <32 x i1> %mask)
+  ret void
+}
+
+;
+; ST1W (float)
+;
+
+define void @masked_scatter_v2f32(<2 x float>* %a, <2 x float*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v2f32:
+; CHECK: ldr d[[VALS:[0-9]+]], [x0]
+; CHECK-NEXT: ldr q[[PTRS:[0-9]+]], [x1]
+; CHECK-NEXT: ptrue [[PG0:p[0-9]+]].s, vl2
+; CHECK-NEXT: fcmeq v[[CMP:[0-9]+]].2s, v[[VALS]].2s, #0
+; CHECK-NEXT: cmpne [[MASK:p[0-9]+]].s, [[PG0]]/z, z[[CMP]].s, #0
+; CHECK-NEXT: ushll v[[SHL:[0-9]+]].2d, v[[VALS]].2s, #0
+; CHECK-NEXT: st1w { z[[SHL]].d }, [[MASK]], [z[[PTRS]].d]
+; CHECK-NEXT: ret
+  %vals = load <2 x float>, <2 x float>* %a
+  %ptrs = load <2 x float*>, <2 x float*>* %b
+  %mask = fcmp oeq <2 x float> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v2f32(<2 x float> %vals, <2 x float*> %ptrs, i32 8, <2 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v4f32(<4 x float>* %a, <4 x float*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v4f32:
+; CHECK: ldr q[[VALS:[0-9]+]], [x0]
+; CHECK-NEXT: ptrue [[PG0:p[0-9]+]].d, vl4
+; CHECK-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].s, vl4
+; CHECK-NEXT: fcmeq v[[CMP:[0-9]+]].4s, v[[VALS]].4s, #0
+; CHECK-NEXT: cmpne [[MASK:p[0-9]+]].s, [[PG1]]/z, z[[CMP]].s, #0
+; CHECK-NEXT: uunpklo [[UPK:z[0-9]+]].d, z[[VALS]].s
+; CHECK-NEXT: st1w { [[UPK]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; CHECK-NEXT: ret
+  %vals = load <4 x float>, <4 x float>* %a
+  %ptrs = load <4 x float*>, <4 x float*>* %b
+  %mask = fcmp oeq <4 x float> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v4f32(<4 x float> %vals, <4 x float*> %ptrs, i32 8, <4 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v8f32(<8 x float>* %a, <8 x float*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v8f32:
+; VBITS_GE_512: ptrue [[PG0:p[0-9]+]].s, vl8
+; VBITS_GE_512-NEXT: ld1w { [[VALS:z[0-9]+]].s }, [[PG0]]/z, [x0]
+; VBITS_GE_512-NEXT: ptrue [[PG1:p[0-9]+]].d, vl8
+; VBITS_GE_512-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_512-NEXT: mov [[ZERO:z[0-9]+]].s, #0
+; VBITS_GE_512-NEXT: fcmeq [[MASK:p[0-9]+]].s, [[PG0]]/z, [[VALS]].s, [[ZERO]]
+; VBITS_GE_512-NEXT: uunpklo [[UPK:z[0-9]+]].d, [[VALS]].s
+; VBITS_GE_512-NEXT: st1w { [[UPK]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; VBITS_BE_512-NEXT: ret
+  %vals = load <8 x float>, <8 x float>* %a
+  %ptrs = load <8 x float*>, <8 x float*>* %b
+  %mask = fcmp oeq <8 x float> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v8f32(<8 x float> %vals, <8 x float*> %ptrs, i32 8, <8 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v16f32(<16 x float>* %a, <16 x float*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v16f32:
+; VBITS_GE_1024: ptrue [[PG0:p[0-9]+]].s, vl16
+; VBITS_GE_1024-NEXT: ld1w { [[VALS:z[0-9]+]].s }, [[PG0]]/z, [x0]
+; VBITS_GE_1024-NEXT: ptrue [[PG1:p[0-9]+]].d, vl16
+; VBITS_GE_1024-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_1024-NEXT: mov [[ZERO:z[0-9]+]].s, #0
+; VBITS_GE_1024-NEXT: fcmeq [[MASK:p[0-9]+]].s, [[PG0]]/z, [[VALS]].s, [[ZERO]]
+; VBITS_GE_1024-NEXT: uunpklo [[UPK:z[0-9]+]].d, [[VALS]].s
+; VBITS_GE_1024-NEXT: st1w { [[UPK]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; VBITS_BE_1024-NEXT: ret
+  %vals = load <16 x float>, <16 x float>* %a
+  %ptrs = load <16 x float*>, <16 x float*>* %b
+  %mask = fcmp oeq <16 x float> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v16f32(<16 x float> %vals, <16 x float*> %ptrs, i32 8, <16 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v32f32(<32 x float>* %a, <32 x float*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v32f32:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].s, vl32
+; VBITS_GE_2048-NEXT: ld1w { [[VALS:z[0-9]+]].s }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].d, vl32
+; VBITS_GE_2048-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: mov [[ZERO:z[0-9]+]].s, #0
+; VBITS_GE_2048-NEXT: fcmeq [[MASK:p[0-9]+]].s, [[PG0]]/z, [[VALS]].s, [[ZERO]]
+; VBITS_GE_2048-NEXT: uunpklo [[UPK:z[0-9]+]].d, [[VALS]].s
+; VBITS_GE_2048-NEXT: st1w { [[UPK]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; VBITS_BE_2048-NEXT: ret
+  %vals = load <32 x float>, <32 x float>* %a
+  %ptrs = load <32 x float*>, <32 x float*>* %b
+  %mask = fcmp oeq <32 x float> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v32f32(<32 x float> %vals, <32 x float*> %ptrs, i32 8, <32 x i1> %mask)
+  ret void
+}
+
+;
+; ST1D (float)
+;
+
+; Scalarize 1 x double scatters
+define void @masked_scatter_v1f64(<1 x double>* %a, <1 x double*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v1f64:
+; CHECK-NOT: ptrue
+  %vals = load <1 x double>, <1 x double>* %a
+  %ptrs = load <1 x double*>, <1 x double*>* %b
+  %mask = fcmp oeq <1 x double> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v1f64(<1 x double> %vals, <1 x double*> %ptrs, i32 8, <1 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v2f64(<2 x double>* %a, <2 x double*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v2f64:
+; CHECK: ldr q[[VALS:[0-9]+]], [x0]
+; CHECK-NEXT: ldr q[[PTRS:[0-9]+]], [x1]
+; CHECK-NEXT: ptrue [[PG0:p[0-9]+]].d, vl2
+; CHECK-NEXT: fcmeq v[[CMP:[0-9]+]].2d, v[[VALS]].2d, #0
+; CHECK-NEXT: cmpne [[MASK:p[0-9]+]].d, [[PG0]]/z, z[[CMP]].d, #0
+; CHECK-NEXT: st1d { z[[VALS]].d }, [[MASK]], [z[[PTRS]].d]
+; CHECK-NEXT: ret
+  %vals = load <2 x double>, <2 x double>* %a
+  %ptrs = load <2 x double*>, <2 x double*>* %b
+  %mask = fcmp oeq <2 x double> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v2f64(<2 x double> %vals, <2 x double*> %ptrs, i32 8, <2 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v4f64(<4 x double>* %a, <4 x double*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v4f64:
+; CHECK: ptrue [[PG0:p[0-9]+]].d, vl4
+; CHECK-NEXT: ld1d { [[VALS:z[0-9]+]].d }, [[PG0]]/z, [x0]
+; CHECK-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; CHECK-NEXT: mov [[ZERO:z[0-9]+]].d, #0
+; CHECK-NEXT: fcmeq [[MASK:p[0-9]+]].d, [[PG0]]/z, [[VALS]].d, [[ZERO]].d
+; CHECK-NEXT: st1d { [[VALS]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; CHECK-NEXT: ret
+  %vals = load <4 x double>, <4 x double>* %a
+  %ptrs = load <4 x double*>, <4 x double*>* %b
+  %mask = fcmp oeq <4 x double> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v4f64(<4 x double> %vals, <4 x double*> %ptrs, i32 8, <4 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v8f64(<8 x double>* %a, <8 x double*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v8f64:
+; VBITS_GE_512: ptrue [[PG0:p[0-9]+]].d, vl8
+; VBITS_GE_512-NEXT: ld1d { [[VALS:z[0-9]+]].d }, [[PG0]]/z, [x0]
+; VBITS_GE_512-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; VBITS_GE_512-NEXT: mov [[ZERO:z[0-9]+]].d, #0
+; VBITS_GE_512-NEXT: fcmeq [[MASK:p[0-9]+]].d, [[PG0]]/z, [[VALS]].d, [[ZERO]]
+; VBITS_GE_512-NEXT: st1d { [[VALS]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; VBITS_BE_512-NEXT: ret
+  %vals = load <8 x double>, <8 x double>* %a
+  %ptrs = load <8 x double*>, <8 x double*>* %b
+  %mask = fcmp oeq <8 x double> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v8f64(<8 x double> %vals, <8 x double*> %ptrs, i32 8, <8 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v16f64(<16 x double>* %a, <16 x double*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v16f64:
+; VBITS_GE_1024: ptrue [[PG0:p[0-9]+]].d, vl16
+; VBITS_GE_1024-NEXT: ld1d { [[VALS:z[0-9]+]].d }, [[PG0]]/z, [x0]
+; VBITS_GE_1024-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; VBITS_GE_1024-NEXT: mov [[ZERO:z[0-9]+]].d, #0
+; VBITS_GE_1024-NEXT: fcmeq [[MASK:p[0-9]+]].d, [[PG0]]/z, [[VALS]].d, [[ZERO]]
+; VBITS_GE_1024-NEXT: st1d { [[VALS]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; VBITS_BE_1024-NEXT: ret
+  %vals = load <16 x double>, <16 x double>* %a
+  %ptrs = load <16 x double*>, <16 x double*>* %b
+  %mask = fcmp oeq <16 x double> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v16f64(<16 x double> %vals, <16 x double*> %ptrs, i32 8, <16 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_v32f64(<32 x double>* %a, <32 x double*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_v32f64:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].d, vl32
+; VBITS_GE_2048-NEXT: ld1d { [[VALS:z[0-9]+]].d }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG0]]/z, [x1]
+; VBITS_GE_2048-NEXT: mov [[ZERO:z[0-9]+]].d, #0
+; VBITS_GE_2048-NEXT: fcmeq [[MASK:p[0-9]+]].d, [[PG0]]/z, [[VALS]].d, [[ZERO]]
+; VBITS_GE_2048-NEXT: st1d { [[VALS]].d }, [[MASK]], {{\[}}[[PTRS]].d]
+; VBITS_BE_2048-NEXT: ret
+  %vals = load <32 x double>, <32 x double>* %a
+  %ptrs = load <32 x double*>, <32 x double*>* %b
+  %mask = fcmp oeq <32 x double> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v32f64(<32 x double> %vals, <32 x double*> %ptrs, i32 8, <32 x i1> %mask)
+  ret void
+}
+
+; The above tests test the types, the below tests check that the addressing
+; modes still function
+
+define void @masked_scatter_32b_scaled_sext(<32 x half>* %a, <32 x i32>* %b, half* %base) #0 {
+; CHECK-LABEL: masked_scatter_32b_scaled_sext:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].h, vl32
+; VBITS_GE_2048-NEXT: ld1h { [[VALS:z[0-9]+]].h }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].s, vl32
+; VBITS_GE_2048-NEXT: ld1w { [[PTRS:z[0-9]+]].s }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: mov [[ZERO:z[0-9]+]].h, #0
+; VBITS_GE_2048-NEXT: fcmeq [[MASK:p[0-9]+]].h, [[PG0]]/z, [[VALS]].h, [[ZERO]]
+; VBITS_GE_2048-NEXT: uunpklo [[UPK:z[0-9]+]].s, [[VALS]].h
+; VBITS_GE_2048-NEXT: st1h { [[VALS]].s }, [[MASK]], [x2, [[PTRS]].s, sxtw #1]
+; VBITS_GE_2048-NEXT: ret
+  %vals = load <32 x half>, <32 x half>* %a
+  %idxs = load <32 x i32>, <32 x i32>* %b
+  %ext = sext <32 x i32> %idxs to <32 x i64>
+  %ptrs = getelementptr half, half* %base, <32 x i64> %ext
+  %mask = fcmp oeq <32 x half> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v32f16(<32 x half> %vals, <32 x half*> %ptrs, i32 8, <32 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_32b_scaled_zext(<32 x half>* %a, <32 x i32>* %b, half* %base) #0 {
+; CHECK-LABEL: masked_scatter_32b_scaled_zext:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].h, vl32
+; VBITS_GE_2048-NEXT: ld1h { [[VALS:z[0-9]+]].h }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].s, vl32
+; VBITS_GE_2048-NEXT: ld1w { [[PTRS:z[0-9]+]].s }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: mov [[ZERO:z[0-9]+]].h, #0
+; VBITS_GE_2048-NEXT: fcmeq [[MASK:p[0-9]+]].h, [[PG0]]/z, [[VALS]].h, [[ZERO]]
+; VBITS_GE_2048-NEXT: uunpklo [[UPK:z[0-9]+]].s, [[VALS]].h
+; VBITS_GE_2048-NEXT: st1h { [[VALS]].s }, [[MASK]], [x2, [[PTRS]].s, uxtw #1]
+; VBITS_GE_2048-NEXT: ret
+  %vals = load <32 x half>, <32 x half>* %a
+  %idxs = load <32 x i32>, <32 x i32>* %b
+  %ext = zext <32 x i32> %idxs to <32 x i64>
+  %ptrs = getelementptr half, half* %base, <32 x i64> %ext
+  %mask = fcmp oeq <32 x half> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v32f16(<32 x half> %vals, <32 x half*> %ptrs, i32 8, <32 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_32b_unscaled_sext(<32 x half>* %a, <32 x i32>* %b, i8* %base) #0 {
+; CHECK-LABEL: masked_scatter_32b_unscaled_sext:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].h, vl32
+; VBITS_GE_2048-NEXT: ld1h { [[VALS:z[0-9]+]].h }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].s, vl32
+; VBITS_GE_2048-NEXT: ld1w { [[PTRS:z[0-9]+]].s }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: mov [[ZERO:z[0-9]+]].h, #0
+; VBITS_GE_2048-NEXT: fcmeq [[MASK:p[0-9]+]].h, [[PG0]]/z, [[VALS]].h, [[ZERO]]
+; VBITS_GE_2048-NEXT: uunpklo [[UPK:z[0-9]+]].s, [[VALS]].h
+; VBITS_GE_2048-NEXT: st1h { [[VALS]].s }, [[MASK]], [x2, [[PTRS]].s, sxtw]
+; VBITS_GE_2048-NEXT: ret
+  %vals = load <32 x half>, <32 x half>* %a
+  %idxs = load <32 x i32>, <32 x i32>* %b
+  %ext = sext <32 x i32> %idxs to <32 x i64>
+  %byte_ptrs = getelementptr i8, i8* %base, <32 x i64> %ext
+  %ptrs = bitcast <32 x i8*> %byte_ptrs to <32 x half*>
+  %mask = fcmp oeq <32 x half> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v32f16(<32 x half> %vals, <32 x half*> %ptrs, i32 8, <32 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_32b_unscaled_zext(<32 x half>* %a, <32 x i32>* %b, i8* %base) #0 {
+; CHECK-LABEL: masked_scatter_32b_unscaled_zext:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].h, vl32
+; VBITS_GE_2048-NEXT: ld1h { [[VALS:z[0-9]+]].h }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].s, vl32
+; VBITS_GE_2048-NEXT: ld1w { [[PTRS:z[0-9]+]].s }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: mov [[ZERO:z[0-9]+]].h, #0
+; VBITS_GE_2048-NEXT: fcmeq [[MASK:p[0-9]+]].h, [[PG0]]/z, [[VALS]].h, [[ZERO]]
+; VBITS_GE_2048-NEXT: uunpklo [[UPK:z[0-9]+]].s, [[VALS]].h
+; VBITS_GE_2048-NEXT: st1h { [[VALS]].s }, [[MASK]], [x2, [[PTRS]].s, uxtw]
+; VBITS_GE_2048-NEXT: ret
+  %vals = load <32 x half>, <32 x half>* %a
+  %idxs = load <32 x i32>, <32 x i32>* %b
+  %ext = zext <32 x i32> %idxs to <32 x i64>
+  %byte_ptrs = getelementptr i8, i8* %base, <32 x i64> %ext
+  %ptrs = bitcast <32 x i8*> %byte_ptrs to <32 x half*>
+  %mask = fcmp oeq <32 x half> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v32f16(<32 x half> %vals, <32 x half*> %ptrs, i32 8, <32 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_64b_scaled(<32 x float>* %a, <32 x i64>* %b, float* %base) #0 {
+; CHECK-LABEL: masked_scatter_64b_scaled:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].s, vl32
+; VBITS_GE_2048-NEXT: ld1w { [[VALS:z[0-9]+]].s }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].d, vl32
+; VBITS_GE_2048-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: mov [[ZERO:z[0-9]+]].s, #0
+; VBITS_GE_2048-NEXT: fcmeq [[MASK:p[0-9]+]].s, [[PG0]]/z, [[VALS]].s, [[ZERO]]
+; VBITS_GE_2048-NEXT: uunpklo [[UPK:z[0-9]+]].d, [[VALS]].s
+; VBITS_GE_2048-NEXT: st1w { [[VALS]].d }, [[MASK]], [x2, [[PTRS]].d, lsl #2]
+; VBITS_GE_2048-NEXT: ret
+  %vals = load <32 x float>, <32 x float>* %a
+  %idxs = load <32 x i64>, <32 x i64>* %b
+  %ptrs = getelementptr float, float* %base, <32 x i64> %idxs
+  %mask = fcmp oeq <32 x float> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v32f32(<32 x float> %vals, <32 x float*> %ptrs, i32 8, <32 x i1> %mask)
+  ret void
+}
+
+define void @masked_scatter_64b_unscaled(<32 x float>* %a, <32 x i64>* %b, i8* %base) #0 {
+; CHECK-LABEL: masked_scatter_64b_unscaled:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].s, vl32
+; VBITS_GE_2048-NEXT: ld1w { [[VALS:z[0-9]+]].s }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].d, vl32
+; VBITS_GE_2048-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: mov [[ZERO:z[0-9]+]].s, #0
+; VBITS_GE_2048-NEXT: fcmeq [[MASK:p[0-9]+]].s, [[PG0]]/z, [[VALS]].s, [[ZERO]]
+; VBITS_GE_2048-NEXT: uunpklo [[UPK:z[0-9]+]].d, [[VALS]].s
+; VBITS_GE_2048-NEXT: st1w { [[VALS]].d }, [[MASK]], [x2, [[PTRS]].d]
+; VBITS_GE_2048-NEXT: ret
+  %vals = load <32 x float>, <32 x float>* %a
+  %idxs = load <32 x i64>, <32 x i64>* %b
+  %byte_ptrs = getelementptr i8, i8* %base, <32 x i64> %idxs
+  %ptrs = bitcast <32 x i8*> %byte_ptrs to <32 x float*>
+  %mask = fcmp oeq <32 x float> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v32f32(<32 x float> %vals, <32 x float*> %ptrs, i32 8, <32 x i1> %mask)
+  ret void
+}
+
+; FIXME: This case does not yet codegen well due to deficiencies in opcode selection
+define void @masked_scatter_vec_plus_reg(<32 x float>* %a, <32 x i8*>* %b, i64 %off) #0 {
+; CHECK-LABEL: masked_scatter_vec_plus_reg:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].s, vl32
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].d, vl32
+; VBITS_GE_2048-NEXT: ld1w { [[VALS:z[0-9]+]].s }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: mov [[OFF:z[0-9]+]].d, x2
+; VBITS_GE_2048-NEXT: mov [[ZERO:z[0-9]+]].s, #0
+; VBITS_GE_2048-NEXT: fcmeq [[MASK:p[0-9]+]].s, [[PG0]]/z, [[VALS]].s, [[ZERO]]
+; VBITS_GE_2048-NEXT: add [[PTRS_ADD:z[0-9]+]].d, [[PG1]]/m, [[PTRS]].d, [[OFF]].d
+; VBITS_GE_2048-NEXT: uunpklo [[UPK:z[0-9]+]].d, [[VALS]].s
+; VBITS_GE_2048-NEXT: st1w { [[VALS]].d }, [[MASK]], {{\[}}[[PTRS_ADD]].d]
+; VBITS_GE_2048-NEXT: ret
+  %vals = load <32 x float>, <32 x float>* %a
+  %bases = load <32 x i8*>, <32 x i8*>* %b
+  %byte_ptrs = getelementptr i8, <32 x i8*> %bases, i64 %off
+  %ptrs = bitcast <32 x i8*> %byte_ptrs to <32 x float*>
+  %mask = fcmp oeq <32 x float> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v32f32(<32 x float> %vals, <32 x float*> %ptrs, i32 8, <32 x i1> %mask)
+  ret void
+}
+
+; FIXME: This case does not yet codegen well due to deficiencies in opcode selection
+define void @masked_scatter_vec_plus_imm(<32 x float>* %a, <32 x i8*>* %b) #0 {
+; CHECK-LABEL: masked_scatter_vec_plus_imm:
+; VBITS_GE_2048: ptrue [[PG0:p[0-9]+]].s, vl32
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].d, vl32
+; VBITS_GE_2048-NEXT: ld1w { [[VALS:z[0-9]+]].s }, [[PG0]]/z, [x0]
+; VBITS_GE_2048-NEXT: ld1d { [[PTRS:z[0-9]+]].d }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: mov [[OFF:z[0-9]+]].d, #4
+; VBITS_GE_2048-NEXT: mov [[ZERO:z[0-9]+]].s, #0
+; VBITS_GE_2048-NEXT: fcmeq [[MASK:p[0-9]+]].s, [[PG0]]/z, [[VALS]].s, [[ZERO]]
+; VBITS_GE_2048-NEXT: add [[PTRS_ADD:z[0-9]+]].d, [[PG1]]/m, [[PTRS]].d, [[OFF]].d
+; VBITS_GE_2048-NEXT: uunpklo [[UPK:z[0-9]+]].d, [[VALS]].s
+; VBITS_GE_2048-NEXT: st1w { [[VALS]].d }, [[MASK]], {{\[}}[[PTRS_ADD]].d]
+; VBITS_GE_2048-NEXT: ret
+  %vals = load <32 x float>, <32 x float>* %a
+  %bases = load <32 x i8*>, <32 x i8*>* %b
+  %byte_ptrs = getelementptr i8, <32 x i8*> %bases, i64 4
+  %ptrs = bitcast <32 x i8*> %byte_ptrs to <32 x float*>
+  %mask = fcmp oeq <32 x float> %vals, zeroinitializer
+  call void @llvm.masked.scatter.v32f32(<32 x float> %vals, <32 x float*> %ptrs, i32 8, <32 x i1> %mask)
+  ret void
+}
+
+declare void @llvm.masked.scatter.v2i8(<2 x i8>, <2 x i8*>, i32, <2 x i1>)
+declare void @llvm.masked.scatter.v4i8(<4 x i8>, <4 x i8*>, i32, <4 x i1>)
+declare void @llvm.masked.scatter.v8i8(<8 x i8>, <8 x i8*>, i32, <8 x i1>)
+declare void @llvm.masked.scatter.v16i8(<16 x i8>, <16 x i8*>, i32, <16 x i1>)
+declare void @llvm.masked.scatter.v32i8(<32 x i8>, <32 x i8*>, i32, <32 x i1>)
+
+declare void @llvm.masked.scatter.v2i16(<2 x i16>, <2 x i16*>, i32, <2 x i1>)
+declare void @llvm.masked.scatter.v4i16(<4 x i16>, <4 x i16*>, i32, <4 x i1>)
+declare void @llvm.masked.scatter.v8i16(<8 x i16>, <8 x i16*>, i32, <8 x i1>)
+declare void @llvm.masked.scatter.v16i16(<16 x i16>, <16 x i16*>, i32, <16 x i1>)
+declare void @llvm.masked.scatter.v32i16(<32 x i16>, <32 x i16*>, i32, <32 x i1>)
+
+declare void @llvm.masked.scatter.v2i32(<2 x i32>, <2 x i32*>, i32, <2 x i1>)
+declare void @llvm.masked.scatter.v4i32(<4 x i32>, <4 x i32*>, i32, <4 x i1>)
+declare void @llvm.masked.scatter.v8i32(<8 x i32>, <8 x i32*>, i32, <8 x i1>)
+declare void @llvm.masked.scatter.v16i32(<16 x i32>, <16 x i32*>, i32, <16 x i1>)
+declare void @llvm.masked.scatter.v32i32(<32 x i32>, <32 x i32*>, i32, <32 x i1>)
+
+declare void @llvm.masked.scatter.v1i64(<1 x i64>, <1 x i64*>, i32, <1 x i1>)
+declare void @llvm.masked.scatter.v2i64(<2 x i64>, <2 x i64*>, i32, <2 x i1>)
+declare void @llvm.masked.scatter.v4i64(<4 x i64>, <4 x i64*>, i32, <4 x i1>)
+declare void @llvm.masked.scatter.v8i64(<8 x i64>, <8 x i64*>, i32, <8 x i1>)
+declare void @llvm.masked.scatter.v16i64(<16 x i64>, <16 x i64*>, i32, <16 x i1>)
+declare void @llvm.masked.scatter.v32i64(<32 x i64>, <32 x i64*>, i32, <32 x i1>)
+
+declare void @llvm.masked.scatter.v2f16(<2 x half>, <2 x half*>, i32, <2 x i1>)
+declare void @llvm.masked.scatter.v4f16(<4 x half>, <4 x half*>, i32, <4 x i1>)
+declare void @llvm.masked.scatter.v8f16(<8 x half>, <8 x half*>, i32, <8 x i1>)
+declare void @llvm.masked.scatter.v16f16(<16 x half>, <16 x half*>, i32, <16 x i1>)
+declare void @llvm.masked.scatter.v32f16(<32 x half>, <32 x half*>, i32, <32 x i1>)
+
+declare void @llvm.masked.scatter.v2f32(<2 x float>, <2 x float*>, i32, <2 x i1>)
+declare void @llvm.masked.scatter.v4f32(<4 x float>, <4 x float*>, i32, <4 x i1>)
+declare void @llvm.masked.scatter.v8f32(<8 x float>, <8 x float*>, i32, <8 x i1>)
+declare void @llvm.masked.scatter.v16f32(<16 x float>, <16 x float*>, i32, <16 x i1>)
+declare void @llvm.masked.scatter.v32f32(<32 x float>, <32 x float*>, i32, <32 x i1>)
+
+declare void @llvm.masked.scatter.v1f64(<1 x double>, <1 x double*>, i32, <1 x i1>)
+declare void @llvm.masked.scatter.v2f64(<2 x double>, <2 x double*>, i32, <2 x i1>)
+declare void @llvm.masked.scatter.v4f64(<4 x double>, <4 x double*>, i32, <4 x i1>)
+declare void @llvm.masked.scatter.v8f64(<8 x double>, <8 x double*>, i32, <8 x i1>)
+declare void @llvm.masked.scatter.v16f64(<16 x double>, <16 x double*>, i32, <16 x i1>)
+declare void @llvm.masked.scatter.v32f64(<32 x double>, <32 x double*>, i32, <32 x i1>)
+
+attributes #0 = { "target-features"="+sve" }

llvm/test/CodeGen/AArch64/sve-fixed-length-masked-stores.ll

 }
 
 define void @masked_store_v8f32(<8 x float>* %ap, <8 x float>* %bp) #0 {
 ; CHECK-LABEL: masked_store_v8f32:
 ; CHECK: ptrue [[PG0:p[0-9]+]].s, vl[[#min(div(VBYTES,4),8)]]
 ; CHECK-NEXT: ld1w { [[Z0:z[0-9]+]].s }, [[PG0]]/z, [x0]
 ; CHECK-NEXT: ld1w { [[Z1:z[0-9]+]].s }, [[PG0]]/z, [x1]
 ; CHECK-NEXT: fcmeq [[PG1:p[0-9]+]].s, [[PG0]]/z, [[Z0]].s, [[Z1]].s
-; CHECK-NEXT: mov [[Z2:z[0-9]+]].s, [[PG1]]/z, #-1
-; CHECK-NEXT: cmpne [[PG2:p[0-9]+]].s, [[PG0]]/z, [[Z2]].s, #0
-; CHECK-NEXT: st1w { z0.s }, [[PG2]], [x{{[0-9]+}}]
+; CHECK-NEXT: st1w { z0.s }, [[PG1]], [x{{[0-9]+}}]
 ; CHECK-NEXT: ret
   %a = load <8 x float>, <8 x float>* %ap
   %b = load <8 x float>, <8 x float>* %bp
   %mask = fcmp oeq <8 x float> %a, %b
   call void @llvm.masked.store.v8f32(<8 x float> %a, <8 x float>* %bp, i32 8, <8 x i1> %mask)
   ret void
 }
 
 define void @masked_store_v16f32(<16 x float>* %ap, <16 x float>* %bp) #0 {
 ; CHECK-LABEL: masked_store_v16f32:
 ; VBITS_GE_512: ptrue p[[P0:[0-9]+]].s, vl[[#min(div(VBYTES,4),16)]]
 ; VBITS_GE_512-NEXT: ld1w { [[Z0:z[0-9]+]].s }, [[PG0]]/z, [x0]
 ; VBITS_GE_512-NEXT: ld1w { [[Z1:z[0-9]+]].s }, [[PG0]]/z, [x1]
 ; VBITS_GE_512-NEXT: fcmeq [[PG1:p[0-9]+]].s, [[PG0]]/z, [[Z0]].s, [[Z1]].s
-; VBITS_GE_512-NEXT: mov [[Z2:z[0-9]+]].s, [[PG1]]/z, #-1
-; VBITS_GE_512-NEXT: cmpne [[PG2:p[0-9]+]].s, [[PG0]]/z, [[Z1]].s, #0
-; VBITS_GE_512-NEXT: st1w { z0.s }, [[PG2]], [x{{[0-9]+}}]
+; VBITS_GE_512-NEXT: st1w { z0.s }, [[PG1]], [x{{[0-9]+}}]
 ; VBITS_GE_512-NEXT: ret
   %a = load <16 x float>, <16 x float>* %ap
   %b = load <16 x float>, <16 x float>* %bp
   %mask = fcmp oeq <16 x float> %a, %b
   call void @llvm.masked.store.v16f32(<16 x float> %a, <16 x float>* %ap, i32 8, <16 x i1> %mask)
   ret void
 }
 
 define void @masked_store_v32f32(<32 x float>* %ap, <32 x float>* %bp) #0 {
 ; CHECK-LABEL: masked_store_v32f32:
 ; VBITS_GE_1024: ptrue p[[P0:[0-9]+]].s, vl[[#min(div(VBYTES,4),32)]]
 ; VBITS_GE_1024-NEXT: ld1w { [[Z0:z[0-9]+]].s }, [[PG0]]/z, [x0]
 ; VBITS_GE_1024-NEXT: ld1w { [[Z1:z[0-9]+]].s }, [[PG0]]/z, [x1]
 ; VBITS_GE_1024-NEXT: fcmeq [[PG1:p[0-9]+]].s, [[PG0]]/z, [[Z0]].s, [[Z1]].s
-; VBITS_GE_1024-NEXT: mov [[Z1:z[0-9]+]].s, [[PG1]]/z, #-1
-; VBITS_GE_1024-NEXT: cmpne [[PG2:p[0-9]+]].s, [[PG0]]/z, [[Z1]].s, #0
-; VBITS_GE_1024-NEXT: st1w { z0.s }, [[PG2]], [x{{[0-9]+}}]
+; VBITS_GE_1024-NEXT: st1w { z0.s }, [[PG1]], [x{{[0-9]+}}]
 ; VBITS_GE_1024-NEXT: ret
   %a = load <32 x float>, <32 x float>* %ap
   %b = load <32 x float>, <32 x float>* %bp
   %mask = fcmp oeq <32 x float> %a, %b
   call void @llvm.masked.store.v32f32(<32 x float> %a, <32 x float>* %ap, i32 8, <32 x i1> %mask)
   ret void
 }
 
 define void @masked_store_v64f32(<64 x float>* %ap, <64 x float>* %bp) #0 {
 ; CHECK-LABEL: masked_store_v64f32:
 ; VBITS_GE_2048: ptrue p[[P0:[0-9]+]].s, vl[[#min(div(VBYTES,4),64)]]
 ; VBITS_GE_2048-NEXT: ld1w { [[Z0:z[0-9]+]].s }, [[PG0]]/z, [x0]
 ; VBITS_GE_2048-NEXT: ld1w { [[Z1:z[0-9]+]].s }, [[PG0]]/z, [x1]
 ; VBITS_GE_2048-NEXT: fcmeq [[PG1:p[0-9]+]].s, [[PG0]]/z, [[Z0]].s, [[Z1]].s
-; VBITS_GE_2048-NEXT: mov [[Z1:z[0-9]+]].s, [[PG1]]/z, #-1
-; VBITS_GE_2048-NEXT: cmpne [[PG2:p[0-9]+]].s, [[PG0]]/z, [[Z1]].s, #0
-; VBITS_GE_2048-NEXT: st1w { z0.s }, [[PG2]], [x{{[0-9]+}}]
+; VBITS_GE_2048-NEXT: st1w { z0.s }, [[PG1]], [x{{[0-9]+}}]
 ; VBITS_GE_2048-NEXT: ret
   %a = load <64 x float>, <64 x float>* %ap
   %b = load <64 x float>, <64 x float>* %bp
   %mask = fcmp oeq <64 x float> %a, %b
   call void @llvm.masked.store.v64f32(<64 x float> %a, <64 x float>* %ap, i32 8, <64 x i1> %mask)
   ret void
 }
 

llvm/test/CodeGen/AArch64/sve-masked-gather.ll

 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ld1sw { z0.d }, p0/z, [z0.d]
 ; CHECK-NEXT:    ret
   %vals = call <vscale x 2 x i32> @llvm.masked.gather.nxv2i32(<vscale x 2 x i32*> %ptrs, i32 4, <vscale x 2 x i1> %mask, <vscale x 2 x i32> undef)
   %vals.sext = sext <vscale x 2 x i32> %vals to <vscale x 2 x i64>
   ret <vscale x 2 x i64> %vals.sext
 }
 
+define <vscale x 2 x i64> @masked_gather_passthru(<vscale x 2 x i32*> %ptrs, <vscale x 2 x i1> %mask, <vscale x 2 x i32> %passthru) {
+; CHECK-LABEL: masked_gather_passthru:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ld1sw { z0.d }, p0/z, [z0.d]
+; CHECK-NEXT:    sel z0.d, p0, z0.d, z1.d
+; CHECK-NEXT:    ret
+  %vals = call <vscale x 2 x i32> @llvm.masked.gather.nxv2i32(<vscale x 2 x i32*> %ptrs, i32 4, <vscale x 2 x i1> %mask, <vscale x 2 x i32> %passthru)
+  %vals.sext = sext <vscale x 2 x i32> %vals to <vscale x 2 x i64>
+  ret <vscale x 2 x i64> %vals.sext
+}
+
+define <vscale x 2 x i64> @masked_gather_passthru_0(<vscale x 2 x i32*> %ptrs, <vscale x 2 x i1> %mask) {
+; CHECK-LABEL: masked_gather_passthru_0:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ld1sw { z0.d }, p0/z, [z0.d]
+; CHECK-NEXT:    ret
+  %vals = call <vscale x 2 x i32> @llvm.masked.gather.nxv2i32(<vscale x 2 x i32*> %ptrs, i32 4, <vscale x 2 x i1> %mask, <vscale x 2 x i32> zeroinitializer)
+  %vals.sext = sext <vscale x 2 x i32> %vals to <vscale x 2 x i64>
+  ret <vscale x 2 x i64> %vals.sext
+}
+
 declare <vscale x 2 x i8> @llvm.masked.gather.nxv2i8(<vscale x 2 x i8*>, i32, <vscale x 2 x i1>, <vscale x 2 x i8>)
 declare <vscale x 2 x i16> @llvm.masked.gather.nxv2i16(<vscale x 2 x i16*>, i32, <vscale x 2 x i1>, <vscale x 2 x i16>)
 declare <vscale x 2 x i32> @llvm.masked.gather.nxv2i32(<vscale x 2 x i32*>, i32, <vscale x 2 x i1>, <vscale x 2 x i32>)
 declare <vscale x 2 x i64> @llvm.masked.gather.nxv2i64(<vscale x 2 x i64*>, i32, <vscale x 2 x i1>, <vscale x 2 x i64>)
 declare <vscale x 2 x half> @llvm.masked.gather.nxv2f16(<vscale x 2 x half*>, i32, <vscale x 2 x i1>, <vscale x 2 x half>)
 declare <vscale x 2 x bfloat> @llvm.masked.gather.nxv2bf16(<vscale x 2 x bfloat*>, i32, <vscale x 2 x i1>, <vscale x 2 x bfloat>)
 declare <vscale x 2 x float> @llvm.masked.gather.nxv2f32(<vscale x 2 x float*>, i32, <vscale x 2 x i1>, <vscale x 2 x float>)
 declare <vscale x 2 x double> @llvm.masked.gather.nxv2f64(<vscale x 2 x double*>, i32, <vscale x 2 x i1>, <vscale x 2 x double>)
 attributes #0 = { "target-features"="+sve,+bf16" }