[Aarch64][SVE] Add intrinsics for scatter stores

[andwar]

Summary

This patch adds the following SVE intrinsics for scatter stores:

• 64-bit offsets:
  • @llvm.aarch64.sve.st1.scatter (unscaled)
  • @llvm.aarch64.sve.st1.scatter.index (scaled)
• 32-bit unscaled offsets:
  • @llvm.aarch64.sve.st1.scatter.uxtw (zero-extended offset)
  • @llvm.aarch64.sve.st1.scatter.sxtw (sign-extended-offset)
• 32-bit scaled offsets:
  • @llvm.aarch64.sve.st1.scatter.uxtw.index (zero-extended offset)
  • @llvm.aarch64.sve.st1.scatter.sxtw.index (sign-extended offset)
• vector base + immediate:
  • @llvm.aarch64.sve.st1.scatter.imm

NFCs: Move the TableGen classes for gather load intrnisics so that they
are all in one places. Fixed typo in names.

[efriedma]

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

12039

Is the getKnownMinSize comparison trying to reject illegal types? Or something else?

12067

Does this work correctly for nxv2f32? It looks like we're missing test coverage, also.

12079

INTRINSIC_VOID has one result: the chain. You don't need to MergeValues here.

llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td

605

"unpacked"

llvm/test/CodeGen/AArch64/sve-intrinsics-scatter-stores-32bit-scaled-offsets.ll

158

Why are the offsets here using <vscale x 2 x i64>? <vscale x 2 x i32> seems more natural.

[andwar]

Thank you for taking a look @eli.friedman!

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

12039

It's meant to reject illegal types that won't fit into an SVE register (i.e. nxv8i32). I've added a comment to clarify this.

12067

That's a good catch, thanks! For scatter stores ACLE only supports _packed_ vectors of single or double precision floats. I will add a check for this.

I also missed this when implementing performLD1GatherLoad (which is very similar to this method, but it felt worthwhile to keep them seperate). When this patch is approved I'll prepare an update for performLD1GatherLoad.

llvm/test/CodeGen/AArch64/sve-intrinsics-scatter-stores-32bit-scaled-offsets.ll

158

• Although these _are_ 32-bit wide offsets, they are stored in 64-bit registers (and are implicitly sign or zero-extended)
• By using <vscale x 2 x i64> we can rely on LLVMScalarOrSameVectorWidth to capture error (i.e. incorrect argument type)
• This is consistent with the implementation for gather loads

Do you see any disadvantages of using <vscale x 2 x i64> instead of <vscale x 2 x i32>?

[andwar]

• make sure performST1ScatterCombine bails out for unpacked floats
• remove MergeValues
• rebase on top of master

[efriedma]

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

12067

We could support this in the backend even if clang can't generate it... but I guess it's fine to leave it out.

llvm/test/CodeGen/AArch64/sve-intrinsics-scatter-stores-32bit-scaled-offsets.ll

158

The biggest advantage of using <vscale x 2 x i32> is that it's obvious the high bits are unused, so you don't end up with extra mask/64-bit multiply/etc. That said, we can recover this in other ways.

I'm mostly concerned that we should try to be consistent here. We use <vscale x 2 x i32> for the stored value in llvm.aarch64.sve.st1.scatter.index.nxv2i32.

[sdesmalen]

llvm/include/llvm/IR/IntrinsicsAArch64.td

981

nit: The formatting for these intrinsics is odd.

1456

nit: unrelated whitespace change.

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

12028

It is unclear what N->getOperand(2) is:

EVT SrcVT = N->getOperand(2)->getValueType(0);
const SDValue Dst = N->getOperand(2);

12038

Should we simply compare:

SrcVT.getSizeInBits().getScalableSize() > AArch64::SVEBitsPerBlock

?

12038

Sorry, I meant:

SrcVT.getSizeInBits().getKnownMinSize() > AArch64::SVEBitsPerBlock

llvm/test/CodeGen/AArch64/sve-intrinsics-scatter-stores-32bit-scaled-offsets.ll

158

We envisioned the type for the indices generated by Clang to be <vscale x 2 x i64>, since the ACLE does not have any knowledge of unpacked types (like <vscale x 2 x i32>). The uxtw in the intrinsic would make it clear that the indices will need to be zero-extended from from word to double-word.

However, I agree this isn't very clear when reading/writing the intrinsics in IR and it should be simple to change the patch to take <vscale x 2 x i32> indices instead and generating a truncate in Clang. If the scatterSt1Combine then ANY_EXTENDs the data, the truncate will likely fall away.

If we change this, we should be consistent and update the loads as well.

[andwar]

llvm/include/llvm/IR/IntrinsicsAArch64.td

981

Fixed - I try to use clang-format (which IMHO does a good job here), but sadly the result is inconsistent with the rest of the file.

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

12028

That's a copy & paste leftover, sorry! I've renamed Dst as well as few other variables for consistency.

llvm/test/CodeGen/AArch64/sve-intrinsics-scatter-stores-32bit-scaled-offsets.ll

158

Thank you both, I'll update accordingly. I suggest that gather loads are updated in a separate patch.

[andwar]

• Make sure that 32 bit unpacked offsets are passed as nxv2i32 (instead of nxv2i64)
• Removed NFCs (landed in a seperate patch)
• Refactored performST1ScatterCombine (better variable names)

[efriedma]

LGTM with one minor suggestion

llvm/include/llvm/IR/IntrinsicsAArch64.td

1072

ImmArg?

[andwar]

• add ImmArg in the defintion of AdvSIMD_ScatterStore_VectorBase_Intrinsic
• duplicated imm0_31, uimm5s2, uimm5s4, uimm5s8 with TImmLeaf-based equivalents (required after adding ImmArg above)
• rebased on top of trun
• removed a whitespace

[andwar]

llvm/include/llvm/IR/IntrinsicsAArch64.td

1072

Sorry, you mentioned that earlier and I missed that. I've updated this patch accordingly.

I also had to make sure that the affected instruction multiclasses in _AArch64SVEInstrInfo.td_ use TImmLeaf instead of ImmLeaf (this came up in other patches similar to this one). To this end, I duplicated the following definitions that use ImmLeaf:

• imm0_31
• uimm5s2
• uimm5s4
• uimm5s8

and defined equivalents using TImmLeaf:

• timm0_31
• tuimm5s2
• tuimm5s4
• tuimm5s8

We may implement this later with a ComplexPattern, instead of duplicating these. I think that @kmclaughlin might already be looking into it.

[efriedma]

LGTM

[sdesmalen]

LGTM!

llvm/include/llvm/IR/IntrinsicsAArch64.td

: Intrinsic<[llvm_anyvector_ty],
                  LLVMSubdivide2VectorType<0>,
                  llvm_i32_ty],
                 [IntrNoMem]>;
 
   class SVE2_1VectorArg_Narrowing_Intrinsic
     : Intrinsic<[LLVMSubdivide2VectorType<0>],
                 [llvm_anyvector_ty],
                 [IntrNoMem]>;
 
   class SVE2_Merged1VectorArg_Narrowing_Intrinsic
     : Intrinsic<[LLVMSubdivide2VectorType<0>],
                 [LLVMSubdivide2VectorType<0>,
                  llvm_anyvector_ty],
                 [IntrNoMem]>;
 
   // NOTE: There is no relationship between these intrinsics beyond an attempt
   // to reuse currently identical class definitions.
class AdvSIMD_GatherLoad_VecTorBase_Intrinsic
     : Intrinsic<[llvm_anyvector_ty],
                 [
                   LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
                   llvm_anyvector_ty,
                   llvm_i64_ty
                 ],
                 [IntrReadMem, IntrArgMemOnly]>;
 
+class AdvSIMD_ScatterStore_64bitOffset_Intrinsic
+    : Intrinsic<[],
+               [
+                 llvm_anyvector_ty,
+                 LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
+                 LLVMPointerToElt<0>,
+                 LLVMScalarOrSameVectorWidth<0, llvm_i64_ty>
+               ],
+               [IntrWriteMem, IntrArgMemOnly]>;
+
+class AdvSIMD_ScatterStore_32bitOffset_Intrinsic
+    : Intrinsic<[],
+               [
+                 llvm_anyvector_ty,
+                 LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
+                 LLVMPointerToElt<0>,
+                 LLVMScalarOrSameVectorWidth<0, llvm_i32_ty>
+               ],
+               [IntrWriteMem, IntrArgMemOnly]>;
+
+class AdvSIMD_ScatterStore_VectorBase_Intrinsic
+    : Intrinsic<[],
+               [
+                 llvm_anyvector_ty,
+                 LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
+                 llvm_anyvector_ty, llvm_i64_ty
+               ],
+               [IntrWriteMem, IntrArgMemOnly]>;
+
 //
 // Integer arithmetic
 //
 
 def int_aarch64_sve_add   : AdvSIMD_Pred2VectorArg_Intrinsic;
 def int_aarch64_sve_sub   : AdvSIMD_Pred2VectorArg_Intrinsic;
 def int_aarch64_sve_subr  : AdvSIMD_Pred2VectorArg_Intrinsic;
 

 //  to 64 bits
 def int_aarch64_sve_ld1_gather_sxtw_index : AdvSIMD_GatherLoad_32bitOffset_Intrinsic;
 def int_aarch64_sve_ld1_gather_uxtw_index : AdvSIMD_GatherLoad_32bitOffset_Intrinsic;
 
 // vector base + immediate index
 def int_aarch64_sve_ld1_gather_imm : AdvSIMD_GatherLoad_VecTorBase_Intrinsic;
 
 //
+// Scatter stores:
+//
+
+// scalar + vector, 64 bit unscaled offsets
+def int_aarch64_sve_st1_scatter : AdvSIMD_ScatterStore_64bitOffset_Intrinsic;
+
+// scalar + vector, 64 bit scaled offsets
+def int_aarch64_sve_st1_scatter_index
+    : AdvSIMD_ScatterStore_64bitOffset_Intrinsic;
+
+//  scalar + vector, 32 bit unscaled offsets, sign (sxtw) or zero (zxtw)
+//  extended to 64 bits
+def int_aarch64_sve_st1_scatter_sxtw
+    : AdvSIMD_ScatterStore_32bitOffset_Intrinsic;
+
+def int_aarch64_sve_st1_scatter_uxtw
+    : AdvSIMD_ScatterStore_32bitOffset_Intrinsic;
+
+//  scalar + vector, 32 bit scaled offsets, sign (sxtw) or zero (zxtw) extended
+//  to 64 bits
+def int_aarch64_sve_st1_scatter_sxtw_index
+    : AdvSIMD_ScatterStore_32bitOffset_Intrinsic;
+
+def int_aarch64_sve_st1_scatter_uxtw_index
+    : AdvSIMD_ScatterStore_32bitOffset_Intrinsic;
+
+// vector base + immediate index
+def int_aarch64_sve_st1_scatter_imm : AdvSIMD_ScatterStore_VectorBase_Intrinsic;
+
+//
 // SVE2 - Non-widening pairwise arithmetic
 //
 
 def int_aarch64_sve_faddp   : AdvSIMD_Pred2VectorArg_Intrinsic;
 def int_aarch64_sve_fmaxp   : AdvSIMD_Pred2VectorArg_Intrinsic;
 def int_aarch64_sve_fmaxnmp : AdvSIMD_Pred2VectorArg_Intrinsic;
 def int_aarch64_sve_fminp   : AdvSIMD_Pred2VectorArg_Intrinsic;
 def int_aarch64_sve_fminnmp : AdvSIMD_Pred2VectorArg_Intrinsic;

 def int_aarch64_sve_fmlslt        : SVE2_3VectorArg_Long_Intrinsic;
 def int_aarch64_sve_fmlslt_lane   : SVE2_3VectorArgIndexed_Long_Intrinsic;
 
 //
 // SVE2 - Floating-point integer binary logarithm
 //
 
 def int_aarch64_sve_flogb : AdvSIMD_SVE_LOGB_Intrinsic;
 
 //
 // SVE2 - Unary narrowing operations
 //
 
 def int_aarch64_sve_sqxtnb  : SVE2_1VectorArg_Narrowing_Intrinsic;
 def int_aarch64_sve_sqxtnt  : SVE2_Merged1VectorArg_Narrowing_Intrinsic;
 def int_aarch64_sve_sqxtunb : SVE2_1VectorArg_Narrowing_Intrinsic;
 def int_aarch64_sve_sqxtunt : SVE2_Merged1VectorArg_Narrowing_Intrinsic;
 def int_aarch64_sve_uqxtnb  : SVE2_1VectorArg_Narrowing_Intrinsic;
 def int_aarch64_sve_uqxtnt  : SVE2_Merged1VectorArg_Narrowing_Intrinsic;
 }

llvm/lib/Target/AArch64/AArch64ISelLowering.h

enum NodeType : unsigned {
   // Signed gather loads
   GLD1S,
   GLD1S_SCALED,
   GLD1S_UXTW,
   GLD1S_SXTW,
   GLD1S_UXTW_SCALED,
   GLD1S_SXTW_SCALED,
   GLD1S_IMM,
+  // Scatter store
+  SST1,
+  SST1_SCALED,
+  SST1_UXTW,
+  SST1_SXTW,
+  SST1_UXTW_SCALED,
+  SST1_SXTW_SCALED,
+  SST1_IMM,
 
   // NEON Load/Store with post-increment base updates
   LD2post = ISD::FIRST_TARGET_MEMORY_OPCODE,
   LD3post,
   LD4post,
   ST2post,
   ST3post,
   ST4post,

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

const char *AArch64TargetLowering::getTargetNodeName(unsigned Opcode) const {
   case AArch64ISD::GLD1_IMM:          return "AArch64ISD::GLD1_IMM";
   case AArch64ISD::GLD1S:             return "AArch64ISD::GLD1S";
   case AArch64ISD::GLD1S_SCALED:      return "AArch64ISD::GLD1S_SCALED";
   case AArch64ISD::GLD1S_SXTW:        return "AArch64ISD::GLD1S_SXTW";
   case AArch64ISD::GLD1S_UXTW:        return "AArch64ISD::GLD1S_UXTW";
   case AArch64ISD::GLD1S_SXTW_SCALED: return "AArch64ISD::GLD1S_SXTW_SCALED";
   case AArch64ISD::GLD1S_UXTW_SCALED: return "AArch64ISD::GLD1S_UXTW_SCALED";
   case AArch64ISD::GLD1S_IMM:         return "AArch64ISD::GLD1S_IMM";
+  case AArch64ISD::SST1:              return "AArch64ISD::SST1";
+  case AArch64ISD::SST1_SCALED:       return "AArch64ISD::SST1_SCALED";
+  case AArch64ISD::SST1_SXTW:         return "AArch64ISD::SST1_SXTW";
+  case AArch64ISD::SST1_UXTW:         return "AArch64ISD::SST1_UXTW";
+  case AArch64ISD::SST1_SXTW_SCALED:  return "AArch64ISD::SST1_SXTW_SCALED";
+  case AArch64ISD::SST1_UXTW_SCALED:  return "AArch64ISD::SST1_UXTW_SCALED";
+  case AArch64ISD::SST1_IMM:          return "AArch64ISD::SST1_IMM";
   }
   return nullptr;
 }
 
 MachineBasicBlock *
 AArch64TargetLowering::EmitF128CSEL(MachineInstr &MI,
                                     MachineBasicBlock *MBB) const {
   // We materialise the F128CSEL pseudo-instruction as some control flow and a
static MVT getSVEContainerType(EVT ContentTy) {
   case MVT::nxv4i8:
   case MVT::nxv4i16:
   case MVT::nxv4i32:
   case MVT::nxv4f32:
     return MVT::nxv4i32;
   }
 }
 
+static SDValue performST1ScatterCombine(SDNode *N, SelectionDAG &DAG,
+                                        unsigned Opcode,
+                                        bool OnlyPackedOffsets = true) {
+  const SDValue Src = N->getOperand(2);
+  const EVT SrcVT = Src->getValueType(0);
+  assert(SrcVT.isScalableVector() &&
+         "Scatter stores are only possible for SVE vectors");
+
+  SDLoc DL(N);
+  MVT SrcElVT = SrcVT.getVectorElementType().getSimpleVT();
+
+  // Make sure that source data will fit into an SVE register
+  if (SrcVT.getSizeInBits().getKnownMinSize() > AArch64::SVEBitsPerBlock)
+    return SDValue();
+
+  // For FPs, ACLE only supports _packed_ single and double precision types.
+  if (SrcElVT.isFloatingPoint())
+    if ((SrcVT != MVT::nxv4f32) && (SrcVT != MVT::nxv2f64))
+      return SDValue();
+
+  // Depending on the addressing mode, this is either a pointer or a vector of
+  // pointers (that fits into one register)
+  const SDValue Base = N->getOperand(4);
+  // Depending on the addressing mode, this is either a single offset or a
+  // vector of offsets  (that fits into one register)
+  SDValue Offset = N->getOperand(5);
+
+  auto &TLI = DAG.getTargetLoweringInfo();
+  if (!TLI.isTypeLegal(Base.getValueType()))
+    return SDValue();
+
+  // Some scatter store variants allow unpacked offsets, but only as nxv2i32
+  // vectors. These are implicitly sign (sxtw) or zero (zxtw) extend to
+  // nxv2i64. Legalize accordingly.
+  if (!OnlyPackedOffsets &&
+      Offset.getValueType().getSimpleVT().SimpleTy == MVT::nxv2i32)
+    Offset = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::nxv2i64, Offset).getValue(0);
+
+  if (!TLI.isTypeLegal(Offset.getValueType()))
+    return SDValue();
+
+  // Source value type that is representable in hardware
+  EVT HwSrcVt = getSVEContainerType(SrcVT);
+
+  // Keep the original type of the input data to store - this is needed to
+  // differentiate between ST1B, ST1H, ST1W and ST1D. For FP values we want the
+  // integer equivalent, so just use HwSrcVt.
+  SDValue InputVT = DAG.getValueType(SrcVT);
+  if (SrcVT.isFloatingPoint())
+    InputVT = DAG.getValueType(HwSrcVt);
+
+  SDVTList VTs = DAG.getVTList(MVT::Other);
+  SDValue SrcNew;
+
+  if (Src.getValueType().isFloatingPoint())
+    SrcNew = DAG.getNode(ISD::BITCAST, DL, HwSrcVt, Src);
+  else
+    SrcNew = DAG.getNode(ISD::ANY_EXTEND, DL, HwSrcVt, Src);
+
+  SDValue Ops[] = {N->getOperand(0), // Chain
+                   SrcNew,
+                   N->getOperand(3), // Pg
+                   Base,
+                   Offset,
+                   InputVT};
+
+  return DAG.getNode(Opcode, DL, VTs, Ops);
+}
+
 static SDValue performLD1GatherCombine(SDNode *N, SelectionDAG &DAG,
                                        unsigned Opcode) {
   EVT RetVT = N->getValueType(0);
   assert(RetVT.isScalableVector() &&
          "Gather loads are only possible for SVE vectors");
 
   SDLoc DL(N);
   MVT RetElVT = RetVT.getVectorElementType().getSimpleVT();
case ISD::INTRINSIC_W_CHAIN:
     case Intrinsic::aarch64_sve_ld1_gather_uxtw:
       return performLD1GatherCombine(N, DAG, AArch64ISD::GLD1_UXTW);
     case Intrinsic::aarch64_sve_ld1_gather_sxtw_index:
       return performLD1GatherCombine(N, DAG, AArch64ISD::GLD1_SXTW_SCALED);
     case Intrinsic::aarch64_sve_ld1_gather_uxtw_index:
       return performLD1GatherCombine(N, DAG, AArch64ISD::GLD1_UXTW_SCALED);
     case Intrinsic::aarch64_sve_ld1_gather_imm:
       return performLD1GatherCombine(N, DAG, AArch64ISD::GLD1_IMM);
+    case Intrinsic::aarch64_sve_st1_scatter:
+      return performST1ScatterCombine(N, DAG, AArch64ISD::SST1);
+    case Intrinsic::aarch64_sve_st1_scatter_index:
+      return performST1ScatterCombine(N, DAG, AArch64ISD::SST1_SCALED);
+    case Intrinsic::aarch64_sve_st1_scatter_sxtw:
+      return performST1ScatterCombine(N, DAG, AArch64ISD::SST1_SXTW,
+                                      /*OnlyPackedOffsets=*/false);
+    case Intrinsic::aarch64_sve_st1_scatter_uxtw:
+      return performST1ScatterCombine(N, DAG, AArch64ISD::SST1_UXTW,
+                                      /*OnlyPackedOffsets=*/false);
+    case Intrinsic::aarch64_sve_st1_scatter_sxtw_index:
+      return performST1ScatterCombine(N, DAG, AArch64ISD::SST1_SXTW_SCALED,
+                                      /*OnlyPackedOffsets=*/false);
+    case Intrinsic::aarch64_sve_st1_scatter_uxtw_index:
+      return performST1ScatterCombine(N, DAG, AArch64ISD::SST1_UXTW_SCALED,
+                                      /*OnlyPackedOffsets=*/false);
+    case Intrinsic::aarch64_sve_st1_scatter_imm:
+      return performST1ScatterCombine(N, DAG, AArch64ISD::SST1_IMM);
     default:
       break;
     }
     break;
   case ISD::GlobalAddress:
     return performGlobalAddressCombine(N, DAG, Subtarget, getTargetMachine());
   }
   return SDValue();

llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td

def SDT_AArch64_GLD1 : SDTypeProfile<1, 4, [
   SDTCVecEltisVT<1,i1>, SDTCisSameNumEltsAs<0,1>
 ]>;
 
 def SDT_AArch64_GLD1_IMM : SDTypeProfile<1, 4, [
   SDTCisVec<0>, SDTCisVec<1>, SDTCisVec<2>, SDTCisInt<3>, SDTCisVT<4, OtherVT>,
   SDTCVecEltisVT<1,i1>, SDTCisSameNumEltsAs<0,1>
 ]>;
 
+def SDT_AArch64_SST1 : SDTypeProfile<0, 5, [
+  SDTCisVec<0>, SDTCisVec<1>, SDTCisPtrTy<2>, SDTCisVec<3>, SDTCisVT<4, OtherVT>,
+  SDTCVecEltisVT<1,i1>, SDTCisSameNumEltsAs<0,1>
+]>;
+
+def SDT_AArch64_SST1_IMM : SDTypeProfile<0, 5, [
+  SDTCisVec<0>, SDTCisVec<1>, SDTCisVec<2>, SDTCisInt<3>, SDTCisVT<4, OtherVT>,
+  SDTCVecEltisVT<1,i1>, SDTCisSameNumEltsAs<0,1>
+]>;
+
+def AArch64st1_scatter               : SDNode<"AArch64ISD::SST1",               SDT_AArch64_SST1,     [SDNPHasChain, SDNPMayStore, SDNPOptInGlue]>;
+def AArch64st1_scatter_scaled        : SDNode<"AArch64ISD::SST1_SCALED",        SDT_AArch64_SST1,     [SDNPHasChain, SDNPMayStore, SDNPOptInGlue]>;
+def AArch64st1_scatter_uxtw          : SDNode<"AArch64ISD::SST1_UXTW",          SDT_AArch64_SST1,     [SDNPHasChain, SDNPMayStore, SDNPOptInGlue]>;
+def AArch64st1_scatter_sxtw          : SDNode<"AArch64ISD::SST1_SXTW",          SDT_AArch64_SST1,     [SDNPHasChain, SDNPMayStore, SDNPOptInGlue]>;
+def AArch64st1_scatter_uxtw_scaled   : SDNode<"AArch64ISD::SST1_UXTW_SCALED",   SDT_AArch64_SST1,     [SDNPHasChain, SDNPMayStore, SDNPOptInGlue]>;
+def AArch64st1_scatter_sxtw_scaled   : SDNode<"AArch64ISD::SST1_SXTW_SCALED",   SDT_AArch64_SST1,     [SDNPHasChain, SDNPMayStore, SDNPOptInGlue]>;
+def AArch64st1_scatter_imm           : SDNode<"AArch64ISD::SST1_IMM",           SDT_AArch64_SST1_IMM, [SDNPHasChain, SDNPMayStore, SDNPOptInGlue]>;
+
 def AArch64ld1_gather                : SDNode<"AArch64ISD::GLD1",               SDT_AArch64_GLD1,     [SDNPHasChain, SDNPMayLoad, SDNPOptInGlue]>;
 def AArch64ld1_gather_scaled         : SDNode<"AArch64ISD::GLD1_SCALED",        SDT_AArch64_GLD1,     [SDNPHasChain, SDNPMayLoad, SDNPOptInGlue]>;
 def AArch64ld1_gather_uxtw           : SDNode<"AArch64ISD::GLD1_UXTW",          SDT_AArch64_GLD1,     [SDNPHasChain, SDNPMayLoad, SDNPOptInGlue]>;
 def AArch64ld1_gather_sxtw           : SDNode<"AArch64ISD::GLD1_SXTW",          SDT_AArch64_GLD1,     [SDNPHasChain, SDNPMayLoad, SDNPOptInGlue]>;
 def AArch64ld1_gather_uxtw_scaled    : SDNode<"AArch64ISD::GLD1_UXTW_SCALED",   SDT_AArch64_GLD1,     [SDNPHasChain, SDNPMayLoad, SDNPOptInGlue]>;
 def AArch64ld1_gather_sxtw_scaled    : SDNode<"AArch64ISD::GLD1_SXTW_SCALED",   SDT_AArch64_GLD1,     [SDNPHasChain, SDNPMayLoad, SDNPOptInGlue]>;
 def AArch64ld1_gather_imm            : SDNode<"AArch64ISD::GLD1_IMM",           SDT_AArch64_GLD1_IMM, [SDNPHasChain, SDNPMayLoad, SDNPOptInGlue]>;
 
let Predicates = [HasSVE] in {
   defm ST1B_D : sve_mem_cst_ss<0b0011, "st1b", Z_d, ZPR64, GPR64NoXZRshifted8>;
   defm ST1H   : sve_mem_cst_ss<0b0101, "st1h", Z_h, ZPR16, GPR64NoXZRshifted16>;
   defm ST1H_S : sve_mem_cst_ss<0b0110, "st1h", Z_s, ZPR32, GPR64NoXZRshifted16>;
   defm ST1H_D : sve_mem_cst_ss<0b0111, "st1h", Z_d, ZPR64, GPR64NoXZRshifted16>;
   defm ST1W   : sve_mem_cst_ss<0b1010, "st1w", Z_s, ZPR32, GPR64NoXZRshifted32>;
   defm ST1W_D : sve_mem_cst_ss<0b1011, "st1w", Z_d, ZPR64, GPR64NoXZRshifted32>;
   defm ST1D   : sve_mem_cst_ss<0b1111, "st1d", Z_d, ZPR64, GPR64NoXZRshifted64>;
 
-  // Scatters using unscaled 32-bit offsets, e.g.
-  //    st1h z0.s, p0, [x0, z0.s, uxtw]
-  // and unpacked:
+  // Scatters using unpacked, unscaled 32-bit offsets, e.g.
   //    st1h z0.d, p0, [x0, z0.d, uxtw]
-  defm SST1B_D : sve_mem_sst_sv_32_unscaled<0b000, "st1b", Z_d, ZPR64, ZPR64ExtSXTW8Only, ZPR64ExtUXTW8Only>;
-  defm SST1B_S : sve_mem_sst_sv_32_unscaled<0b001, "st1b", Z_s, ZPR32, ZPR32ExtSXTW8Only, ZPR32ExtUXTW8Only>;
-  defm SST1H_D : sve_mem_sst_sv_32_unscaled<0b010, "st1h", Z_d, ZPR64, ZPR64ExtSXTW8, ZPR64ExtUXTW8>;
-  defm SST1H_S : sve_mem_sst_sv_32_unscaled<0b011, "st1h", Z_s, ZPR32, ZPR32ExtSXTW8, ZPR32ExtUXTW8>;
-  defm SST1W_D : sve_mem_sst_sv_32_unscaled<0b100, "st1w", Z_d, ZPR64, ZPR64ExtSXTW8, ZPR64ExtUXTW8>;
-  defm SST1W   : sve_mem_sst_sv_32_unscaled<0b101, "st1w", Z_s, ZPR32, ZPR32ExtSXTW8, ZPR32ExtUXTW8>;
-  defm SST1D   : sve_mem_sst_sv_32_unscaled<0b110, "st1d", Z_d, ZPR64, ZPR64ExtSXTW8, ZPR64ExtUXTW8>;
+  defm SST1B_D : sve_mem_64b_sst_sv_32_unscaled<0b000, "st1b", AArch64st1_scatter_sxtw, AArch64st1_scatter_uxtw, ZPR64ExtSXTW8Only, ZPR64ExtUXTW8Only, nxv2i8>;
+  defm SST1H_D : sve_mem_64b_sst_sv_32_unscaled<0b010, "st1h", AArch64st1_scatter_sxtw, AArch64st1_scatter_uxtw, ZPR64ExtSXTW8, ZPR64ExtUXTW8, nxv2i16>;
+  defm SST1W_D : sve_mem_64b_sst_sv_32_unscaled<0b100, "st1w", AArch64st1_scatter_sxtw, AArch64st1_scatter_uxtw, ZPR64ExtSXTW8, ZPR64ExtUXTW8,nxv2i32>;
+  defm SST1D   : sve_mem_64b_sst_sv_32_unscaled<0b110, "st1d", AArch64st1_scatter_sxtw, AArch64st1_scatter_uxtw, ZPR64ExtSXTW8, ZPR64ExtUXTW8, nxv2i64>;
+
+  // Scatters using packed, unscaled 32-bit offsets, e.g.
+  //    st1h z0.s, p0, [x0, z0.s, uxtw]
+  defm SST1B_S : sve_mem_32b_sst_sv_32_unscaled<0b001, "st1b", AArch64st1_scatter_sxtw, AArch64st1_scatter_uxtw, ZPR32ExtSXTW8Only, ZPR32ExtUXTW8Only, nxv4i8>;
+  defm SST1H_S : sve_mem_32b_sst_sv_32_unscaled<0b011, "st1h", AArch64st1_scatter_sxtw, AArch64st1_scatter_uxtw, ZPR32ExtSXTW8, ZPR32ExtUXTW8, nxv4i16>;
+  defm SST1W   : sve_mem_32b_sst_sv_32_unscaled<0b101, "st1w", AArch64st1_scatter_sxtw, AArch64st1_scatter_uxtw, ZPR32ExtSXTW8, ZPR32ExtUXTW8, nxv4i32>;
 
-  // Scatters using scaled 32-bit offsets, e.g.
+  // Scatters using packed, scaled 32-bit offsets, e.g.
   //    st1h z0.s, p0, [x0, z0.s, uxtw #1]
-  // and unpacked:
+  defm SST1H_S : sve_mem_32b_sst_sv_32_scaled<0b011, "st1h", AArch64st1_scatter_sxtw_scaled, AArch64st1_scatter_uxtw_scaled, ZPR32ExtSXTW16, ZPR32ExtUXTW16, nxv4i16>;
+  defm SST1W   : sve_mem_32b_sst_sv_32_scaled<0b101, "st1w", AArch64st1_scatter_sxtw_scaled, AArch64st1_scatter_uxtw_scaled, ZPR32ExtSXTW32, ZPR32ExtUXTW32, nxv4i32>;
+
+  // Scatters using unpacked, scaled 32-bit offsets, e.g.
   //    st1h z0.d, p0, [x0, z0.d, uxtw #1]
-  defm SST1H_D : sve_mem_sst_sv_32_scaled<0b010, "st1h", Z_d, ZPR64, ZPR64ExtSXTW16, ZPR64ExtUXTW16>;
-  defm SST1H_S : sve_mem_sst_sv_32_scaled<0b011, "st1h", Z_s, ZPR32, ZPR32ExtSXTW16, ZPR32ExtUXTW16>;
-  defm SST1W_D : sve_mem_sst_sv_32_scaled<0b100, "st1w", Z_d, ZPR64, ZPR64ExtSXTW32, ZPR64ExtUXTW32>;
-  defm SST1W   : sve_mem_sst_sv_32_scaled<0b101, "st1w", Z_s, ZPR32, ZPR32ExtSXTW32, ZPR32ExtUXTW32>;
-  defm SST1D   : sve_mem_sst_sv_32_scaled<0b110, "st1d", Z_d, ZPR64, ZPR64ExtSXTW64, ZPR64ExtUXTW64>;
+  defm SST1H_D : sve_mem_64b_sst_sv_32_scaled<0b010, "st1h", AArch64st1_scatter_sxtw_scaled, AArch64st1_scatter_uxtw_scaled, ZPR64ExtSXTW16, ZPR64ExtUXTW16, nxv2i16>;
+  defm SST1W_D : sve_mem_64b_sst_sv_32_scaled<0b100, "st1w", AArch64st1_scatter_sxtw_scaled, AArch64st1_scatter_uxtw_scaled, ZPR64ExtSXTW32, ZPR64ExtUXTW32, nxv2i32>;
+  defm SST1D   : sve_mem_64b_sst_sv_32_scaled<0b110, "st1d", AArch64st1_scatter_sxtw_scaled, AArch64st1_scatter_uxtw_scaled, ZPR64ExtSXTW64, ZPR64ExtUXTW64, nxv2i64>;
 
   // Scatters using 32/64-bit pointers with offset, e.g.
   //    st1h z0.s, p0, [z0.s, #16]
+  defm SST1B_S : sve_mem_32b_sst_vi_ptrs<0b001, "st1b", imm0_31, AArch64st1_scatter_imm, nxv4i8>;
+  defm SST1H_S : sve_mem_32b_sst_vi_ptrs<0b011, "st1h", uimm5s2, AArch64st1_scatter_imm, nxv4i16>;
+  defm SST1W   : sve_mem_32b_sst_vi_ptrs<0b101, "st1w", uimm5s4, AArch64st1_scatter_imm, nxv4i32>;
+
+  // Scatters using 32/64-bit pointers with offset, e.g.
   //    st1h z0.d, p0, [z0.d, #16]
-  defm SST1B_D : sve_mem_sst_vi_ptrs<0b000, "st1b", Z_d, ZPR64, imm0_31>;
-  defm SST1B_S : sve_mem_sst_vi_ptrs<0b001, "st1b", Z_s, ZPR32, imm0_31>;
-  defm SST1H_D : sve_mem_sst_vi_ptrs<0b010, "st1h", Z_d, ZPR64, uimm5s2>;
-  defm SST1H_S : sve_mem_sst_vi_ptrs<0b011, "st1h", Z_s, ZPR32, uimm5s2>;
-  defm SST1W_D : sve_mem_sst_vi_ptrs<0b100, "st1w", Z_d, ZPR64, uimm5s4>;
-  defm SST1W   : sve_mem_sst_vi_ptrs<0b101, "st1w", Z_s, ZPR32, uimm5s4>;
-  defm SST1D   : sve_mem_sst_vi_ptrs<0b110, "st1d", Z_d, ZPR64, uimm5s8>;
+  defm SST1B_D : sve_mem_64b_sst_vi_ptrs<0b000, "st1b", imm0_31, AArch64st1_scatter_imm, nxv2i8>;
+  defm SST1H_D : sve_mem_64b_sst_vi_ptrs<0b010, "st1h", uimm5s2, AArch64st1_scatter_imm, nxv2i16>;
+  defm SST1W_D : sve_mem_64b_sst_vi_ptrs<0b100, "st1w", uimm5s4, AArch64st1_scatter_imm, nxv2i32>;
+  defm SST1D   : sve_mem_64b_sst_vi_ptrs<0b110, "st1d", uimm5s8, AArch64st1_scatter_imm, nxv2i64>;
 
   // Scatters using unscaled 64-bit offsets, e.g.
   //    st1h z0.d, p0, [x0, z0.d]
-  defm SST1B_D : sve_mem_sst_sv_64_unscaled<0b00, "st1b">;
-  defm SST1H_D : sve_mem_sst_sv_64_unscaled<0b01, "st1h">;
-  defm SST1W_D : sve_mem_sst_sv_64_unscaled<0b10, "st1w">;
-  defm SST1D   : sve_mem_sst_sv_64_unscaled<0b11, "st1d">;
+  defm SST1B_D : sve_mem_sst_sv_64_unscaled<0b00, "st1b", AArch64st1_scatter, nxv2i8>;
+  defm SST1H_D : sve_mem_sst_sv_64_unscaled<0b01, "st1h", AArch64st1_scatter, nxv2i16>;
+  defm SST1W_D : sve_mem_sst_sv_64_unscaled<0b10, "st1w", AArch64st1_scatter, nxv2i32>;
+  defm SST1D   : sve_mem_sst_sv_64_unscaled<0b11, "st1d", AArch64st1_scatter, nxv2i64>;
 
   // Scatters using scaled 64-bit offsets, e.g.
   //    st1h z0.d, p0, [x0, z0.d, lsl #1]
-  defm SST1H_D_SCALED : sve_mem_sst_sv_64_scaled<0b01, "st1h", ZPR64ExtLSL16>;
-  defm SST1W_D_SCALED : sve_mem_sst_sv_64_scaled<0b10, "st1w", ZPR64ExtLSL32>;
-  defm SST1D_SCALED   : sve_mem_sst_sv_64_scaled<0b11, "st1d", ZPR64ExtLSL64>;
+  defm SST1H_D_SCALED : sve_mem_sst_sv_64_scaled<0b01, "st1h", AArch64st1_scatter_scaled, ZPR64ExtLSL16, nxv2i16>;
+  defm SST1W_D_SCALED : sve_mem_sst_sv_64_scaled<0b10, "st1w", AArch64st1_scatter_scaled, ZPR64ExtLSL32, nxv2i32>;
+  defm SST1D_SCALED   : sve_mem_sst_sv_64_scaled<0b11, "st1d", AArch64st1_scatter_scaled, ZPR64ExtLSL64, nxv2i64>;
 
   // ST(2|3|4) structured stores (register + immediate)
   defm ST2B_IMM : sve_mem_est_si<0b00, 0b01, ZZ_b,   "st2b", simm4s2>;
   defm ST3B_IMM : sve_mem_est_si<0b00, 0b10, ZZZ_b,  "st3b", simm4s3>;
   defm ST4B_IMM : sve_mem_est_si<0b00, 0b11, ZZZZ_b, "st4b", simm4s4>;
   defm ST2H_IMM : sve_mem_est_si<0b01, 0b01, ZZ_h,   "st2h", simm4s2>;
   defm ST3H_IMM : sve_mem_est_si<0b01, 0b10, ZZZ_h,  "st3h", simm4s3>;
   defm ST4H_IMM : sve_mem_est_si<0b01, 0b11, ZZZZ_h, "st4h", simm4s4>;

llvm/lib/Target/AArch64/SVEInstrFormats.td

: I<(outs), (ins VecList:$Zt, PPR3bAny:$Pg, GPR64sp:$Rn, zprext:$Zm),
   let Inst{13}    = 0;
   let Inst{12-10} = Pg;
   let Inst{9-5}   = Rn;
   let Inst{4-0}   = Zt;
 
   let mayStore = 1;
 }
 
-multiclass sve_mem_sst_sv_32_scaled<bits<3> opc, string asm,
-                                    RegisterOperand listty,
-                                    ZPRRegOp zprty,
+multiclass sve_mem_32b_sst_sv_32_scaled<bits<3> opc, string asm,
+                                    SDPatternOperator sxtw_op,
+                                    SDPatternOperator uxtw_op,
                                     RegisterOperand sxtw_opnd,
-                                    RegisterOperand uxtw_opnd > {
-  def _UXTW_SCALED : sve_mem_sst_sv<opc, 0, 1, asm, listty, uxtw_opnd>;
-  def _SXTW_SCALED : sve_mem_sst_sv<opc, 1, 1, asm, listty, sxtw_opnd>;
+                                    RegisterOperand uxtw_opnd,
+                                    ValueType vt > {
+  def _UXTW_SCALED : sve_mem_sst_sv<opc, 0, 1, asm, Z_s, uxtw_opnd>;
+  def _SXTW_SCALED : sve_mem_sst_sv<opc, 1, 1, asm, Z_s, sxtw_opnd>;
 
   def : InstAlias<asm # "\t$Zt, $Pg, [$Rn, $Zm]",
-                 (!cast<Instruction>(NAME # _UXTW_SCALED) zprty:$Zt, PPR3bAny:$Pg, GPR64sp:$Rn, uxtw_opnd:$Zm), 0>;
+                 (!cast<Instruction>(NAME # _UXTW_SCALED) ZPR32:$Zt, PPR3bAny:$Pg, GPR64sp:$Rn, uxtw_opnd:$Zm), 0>;
   def : InstAlias<asm # "\t$Zt, $Pg, [$Rn, $Zm]",
-                 (!cast<Instruction>(NAME # _SXTW_SCALED) zprty:$Zt, PPR3bAny:$Pg, GPR64sp:$Rn, sxtw_opnd:$Zm), 0>;
+                 (!cast<Instruction>(NAME # _SXTW_SCALED) ZPR32:$Zt, PPR3bAny:$Pg, GPR64sp:$Rn, sxtw_opnd:$Zm), 0>;
+
+  def : Pat<(uxtw_op (nxv4i32 ZPR:$data), (nxv4i1 PPR:$gp), GPR64sp:$base, (nxv4i32 ZPR:$offsets), vt),
+            (!cast<Instruction>(NAME # _UXTW_SCALED) ZPR:$data, PPR:$gp, GPR64sp:$base, ZPR:$offsets)>;
+  def : Pat<(sxtw_op (nxv4i32 ZPR:$data), (nxv4i1 PPR:$gp), GPR64sp:$base, (nxv4i32 ZPR:$offsets), vt),
+            (!cast<Instruction>(NAME # _SXTW_SCALED) ZPR:$data, PPR:$gp, GPR64sp:$base, ZPR:$offsets)>;
 }
 
-multiclass sve_mem_sst_sv_32_unscaled<bits<3> opc, string asm,
-                                      RegisterOperand listty,
-                                      ZPRRegOp zprty,
-                                      RegisterOperand sxtw_opnd,
-                                      RegisterOperand uxtw_opnd> {
-  def _UXTW : sve_mem_sst_sv<opc, 0, 0, asm, listty, uxtw_opnd>;
-  def _SXTW : sve_mem_sst_sv<opc, 1, 0, asm, listty, sxtw_opnd>;
+multiclass sve_mem_64b_sst_sv_32_scaled<bits<3> opc, string asm,
+                                    SDPatternOperator sxtw_op,
+                                    SDPatternOperator uxtw_op,
+                                    RegisterOperand uxtw_opnd,
+                                    ValueType vt > {
+  def _UXTW_SCALED : sve_mem_sst_sv<opc, 0, 1, asm, Z_d, uxtw_opnd>;
+  def _SXTW_SCALED : sve_mem_sst_sv<opc, 1, 1, asm, Z_d, sxtw_opnd>;
+
+  def : InstAlias<asm # "\t$Zt, $Pg, [$Rn, $Zm]",
+                 (!cast<Instruction>(NAME # _UXTW_SCALED) ZPR64:$Zt, PPR3bAny:$Pg, GPR64sp:$Rn, uxtw_opnd:$Zm), 0>;
+  def : InstAlias<asm # "\t$Zt, $Pg, [$Rn, $Zm]",
+                 (!cast<Instruction>(NAME # _SXTW_SCALED) ZPR64:$Zt, PPR3bAny:$Pg, GPR64sp:$Rn, sxtw_opnd:$Zm), 0>;
+
+  def : Pat<(uxtw_op (nxv2i64 ZPR:$data), (nxv2i1 PPR:$gp), GPR64sp:$base, (nxv2i64 ZPR:$offsets), vt),
+            (!cast<Instruction>(NAME # _UXTW_SCALED) ZPR:$data, PPR:$gp, GPR64sp:$base, ZPR:$offsets)>;
+  def : Pat<(sxtw_op (nxv2i64 ZPR:$data), (nxv2i1 PPR:$gp), GPR64sp:$base, (nxv2i64 ZPR:$offsets), vt),
+            (!cast<Instruction>(NAME # _SXTW_SCALED) ZPR:$data, PPR:$gp, GPR64sp:$base, ZPR:$offsets)>;
+}
+
+multiclass sve_mem_64b_sst_sv_32_unscaled<bits<3> opc, string asm,
+                                         SDPatternOperator sxtw_op,
+                                         SDPatternOperator uxtw_op,
+                                         RegisterOperand sxtw_opnd,
+                                         RegisterOperand uxtw_opnd,
+                                         ValueType vt> {
+  def _UXTW : sve_mem_sst_sv<opc, 0, 0, asm, Z_d, uxtw_opnd>;
+  def _SXTW : sve_mem_sst_sv<opc, 1, 0, asm, Z_d, sxtw_opnd>;
+
+  def : InstAlias<asm # "\t$Zt, $Pg, [$Rn, $Zm]",
+                 (!cast<Instruction>(NAME # _UXTW) ZPR64:$Zt, PPR3bAny:$Pg, GPR64sp:$Rn, uxtw_opnd:$Zm), 0>;
+  def : InstAlias<asm # "\t$Zt, $Pg, [$Rn, $Zm]",
+                 (!cast<Instruction>(NAME # _SXTW) ZPR64:$Zt, PPR3bAny:$Pg, GPR64sp:$Rn, sxtw_opnd:$Zm), 0>;
+
+  def : Pat<(uxtw_op (nxv2i64 ZPR:$data), (nxv2i1 PPR:$gp), GPR64sp:$base, (nxv2i64 ZPR:$offsets), vt),
+            (!cast<Instruction>(NAME # _UXTW) ZPR:$data, PPR:$gp, GPR64sp:$base, ZPR:$offsets)>;
+  def : Pat<(sxtw_op (nxv2i64 ZPR:$data), (nxv2i1 PPR:$gp), GPR64sp:$base, (nxv2i64 ZPR:$offsets), vt),
+            (!cast<Instruction>(NAME # _SXTW) ZPR:$data, PPR:$gp, GPR64sp:$base, ZPR:$offsets)>;
+}
+
+multiclass sve_mem_32b_sst_sv_32_unscaled<bits<3> opc, string asm,
+                                          SDPatternOperator sxtw_op,
+                                          SDPatternOperator uxtw_op,
+                                          RegisterOperand sxtw_opnd,
+                                          RegisterOperand uxtw_opnd,
+                                          ValueType vt> {
+  def _UXTW : sve_mem_sst_sv<opc, 0, 0, asm, Z_s, uxtw_opnd>;
+  def _SXTW : sve_mem_sst_sv<opc, 1, 0, asm, Z_s, sxtw_opnd>;
 
   def : InstAlias<asm # "\t$Zt, $Pg, [$Rn, $Zm]",
-                 (!cast<Instruction>(NAME # _UXTW) zprty:$Zt, PPR3bAny:$Pg, GPR64sp:$Rn, uxtw_opnd:$Zm), 0>;
+                 (!cast<Instruction>(NAME # _UXTW) ZPR32:$Zt, PPR3bAny:$Pg, GPR64sp:$Rn, uxtw_opnd:$Zm), 0>;
   def : InstAlias<asm # "\t$Zt, $Pg, [$Rn, $Zm]",
-                 (!cast<Instruction>(NAME # _SXTW) zprty:$Zt, PPR3bAny:$Pg, GPR64sp:$Rn, sxtw_opnd:$Zm), 0>;
+                 (!cast<Instruction>(NAME # _SXTW) ZPR32:$Zt, PPR3bAny:$Pg, GPR64sp:$Rn, sxtw_opnd:$Zm), 0>;
+
+  def : Pat<(uxtw_op (nxv4i32 ZPR:$data), (nxv4i1 PPR:$gp), GPR64sp:$base, (nxv4i32 ZPR:$offsets), vt),
+            (!cast<Instruction>(NAME # _UXTW) ZPR:$data, PPR:$gp, GPR64sp:$base, ZPR:$offsets)>;
+  def : Pat<(sxtw_op (nxv4i32 ZPR:$data), (nxv4i1 PPR:$gp), GPR64sp:$base, (nxv4i32 ZPR:$offsets), vt),
+            (!cast<Instruction>(NAME # _SXTW) ZPR:$data, PPR:$gp, GPR64sp:$base, ZPR:$offsets)>;
 }
 
 class sve_mem_sst_sv2<bits<2> msz, bit scaled, string asm,
                       RegisterOperand zprext>
 : I<(outs), (ins Z_d:$Zt, PPR3bAny:$Pg, GPR64sp:$Rn, zprext:$Zm),
   asm, "\t$Zt, $Pg, [$Rn, $Zm]",
   "",
   []>, Sched<[]> {
: I<(outs), (ins Z_d:$Zt, PPR3bAny:$Pg, GPR64sp:$Rn, zprext:$Zm),
   let Inst{12-10} = Pg;
   let Inst{9-5}   = Rn;
   let Inst{4-0}   = Zt;
 
   let mayStore = 1;
 }
 
 multiclass sve_mem_sst_sv_64_scaled<bits<2> msz, string asm,
-                                    RegisterOperand zprext> {
-  def "" : sve_mem_sst_sv2<msz, 1, asm, zprext>;
+                                    SDPatternOperator op,
+                                    RegisterOperand zprext,
+                                    ValueType vt> {
+  def _SCALED_REAL : sve_mem_sst_sv2<msz, 1, asm, zprext>;
 
   def : InstAlias<asm # "\t$Zt, $Pg, [$Rn, $Zm]",
-                 (!cast<Instruction>(NAME) ZPR64:$Zt, PPR3bAny:$Pg, GPR64sp:$Rn, zprext:$Zm), 0>;
+                 (!cast<Instruction>(NAME # _SCALED_REAL) ZPR64:$Zt, PPR3bAny:$Pg, GPR64sp:$Rn, zprext:$Zm), 0>;
 
+  def : Pat<(op (nxv2i64 ZPR:$data), (nxv2i1 PPR:$gp), GPR64sp:$base, (nxv2i64 ZPR:$indices), vt),
+            (!cast<Instruction>(NAME # _SCALED_REAL) ZPR:$data, PPR:$gp, GPR64sp:$base, ZPR:$indices)>;
 }
 
-multiclass sve_mem_sst_sv_64_unscaled<bits<2> msz, string asm> {
-  def "" : sve_mem_sst_sv2<msz, 0, asm, ZPR64ExtLSL8>;
+multiclass sve_mem_sst_sv_64_unscaled<bits<2> msz, string asm,
+                                      SDPatternOperator op,
+                                      ValueType vt> {
+  def _REAL : sve_mem_sst_sv2<msz, 0, asm, ZPR64ExtLSL8>;
 
   def : InstAlias<asm # "\t$Zt, $Pg, [$Rn, $Zm]",
-                 (!cast<Instruction>(NAME) ZPR64:$Zt, PPR3bAny:$Pg, GPR64sp:$Rn, ZPR64ExtLSL8:$Zm), 0>;
+                 (!cast<Instruction>(NAME # _REAL) ZPR64:$Zt, PPR3bAny:$Pg, GPR64sp:$Rn, ZPR64ExtLSL8:$Zm), 0>;
+
+  def : Pat<(op (nxv2i64 ZPR:$data), (nxv2i1 PPR:$gp), GPR64sp:$base, (nxv2i64 ZPR:$offsets), vt),
+            (!cast<Instruction>(NAME # _REAL) ZPR:$data, PPR:$gp, GPR64sp:$base, ZPR:$offsets)>;
 }
 
 class sve_mem_sst_vi<bits<3> opc, string asm, ZPRRegOp zprty,
                      RegisterOperand VecList, Operand imm_ty>
 : I<(outs), (ins VecList:$Zt, PPR3bAny:$Pg, zprty:$Zn, imm_ty:$imm5),
   asm, "\t$Zt, $Pg, [$Zn, $imm5]",
   "",
   []>, Sched<[]> {
: I<(outs), (ins VecList:$Zt, PPR3bAny:$Pg, zprty:$Zn, imm_ty:$imm5),
   let Inst{15-13} = 0b101;
   let Inst{12-10} = Pg;
   let Inst{9-5}   = Zn;
   let Inst{4-0}   = Zt;
 
   let mayStore = 1;
 }
 
-multiclass sve_mem_sst_vi_ptrs<bits<3> opc, string asm, RegisterOperand listty,
-                               ZPRRegOp zprty, Operand imm_ty> {
-  def _IMM : sve_mem_sst_vi<opc, asm, zprty, listty, imm_ty>;
+multiclass sve_mem_32b_sst_vi_ptrs<bits<3> opc, string asm,
+                                   Operand imm_ty,
+                                   SDPatternOperator op,
+                                   ValueType vt> {
+  def _IMM : sve_mem_sst_vi<opc, asm, ZPR32, Z_s, imm_ty>;
+
+  def : InstAlias<asm # "\t$Zt, $Pg, [$Zn]",
+                  (!cast<Instruction>(NAME # _IMM) ZPR32:$Zt, PPR3bAny:$Pg, ZPR32:$Zn, 0), 0>;
+  def : InstAlias<asm # "\t$Zt, $Pg, [$Zn, $imm5]",
+                  (!cast<Instruction>(NAME # _IMM) ZPR32:$Zt, PPR3bAny:$Pg, ZPR32:$Zn, imm_ty:$imm5), 0>;
+  def : InstAlias<asm # "\t$Zt, $Pg, [$Zn]",
+                  (!cast<Instruction>(NAME # _IMM) Z_s:$Zt, PPR3bAny:$Pg, ZPR32:$Zn, 0), 1>;
+
+  def : Pat<(op (nxv4i32 ZPR:$data), (nxv4i1 PPR:$gp), (nxv4i32 ZPR:$ptrs), imm_ty:$index, vt),
+            (!cast<Instruction>(NAME # _IMM) ZPR:$data, PPR:$gp, ZPR:$ptrs, imm_ty:$index)>;
+}
+
+multiclass sve_mem_64b_sst_vi_ptrs<bits<3> opc, string asm,
+                                   Operand imm_ty,
+                                   SDPatternOperator op,
+                                   ValueType vt> {
+  def _IMM : sve_mem_sst_vi<opc, asm, ZPR64, Z_d, imm_ty>;
 
   def : InstAlias<asm # "\t$Zt, $Pg, [$Zn]",
-                  (!cast<Instruction>(NAME # _IMM) zprty:$Zt, PPR3bAny:$Pg, zprty:$Zn, 0), 0>;
+                  (!cast<Instruction>(NAME # _IMM) ZPR64:$Zt, PPR3bAny:$Pg, ZPR64:$Zn, 0), 0>;
   def : InstAlias<asm # "\t$Zt, $Pg, [$Zn, $imm5]",
-                  (!cast<Instruction>(NAME # _IMM) zprty:$Zt, PPR3bAny:$Pg, zprty:$Zn, imm_ty:$imm5), 0>;
+                  (!cast<Instruction>(NAME # _IMM) ZPR64:$Zt, PPR3bAny:$Pg, ZPR64:$Zn, imm_ty:$imm5), 0>;
   def : InstAlias<asm # "\t$Zt, $Pg, [$Zn]",
-                  (!cast<Instruction>(NAME # _IMM) listty:$Zt, PPR3bAny:$Pg, zprty:$Zn, 0), 1>;
+                  (!cast<Instruction>(NAME # _IMM) Z_s:$Zt, PPR3bAny:$Pg, ZPR64:$Zn, 0), 1>;
+
+  def : Pat<(op (nxv2i64 ZPR:$data), (nxv2i1 PPR:$gp), (nxv2i64 ZPR:$ptrs), imm_ty:$index, vt),
+            (!cast<Instruction>(NAME # _IMM) ZPR:$data, PPR:$gp, ZPR:$ptrs, imm_ty:$index)>;
 }
 
 class sve_mem_z_spill<string asm>
 : I<(outs), (ins ZPRAny:$Zt, GPR64sp:$Rn, simm9:$imm9),
   asm, "\t$Zt, [$Rn, $imm9, mul vl]",
   "",
   []>, Sched<[]> {
   bits<5> Rn;

llvm/test/CodeGen/AArch64/sve-intrinsics-scatter-stores-32bit-scaled-offsets.ll

+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve < %s | FileCheck %s
+
+;
+; ST1H, ST1W, ST1D: base + 32-bit scaled offset, sign (sxtw) or zero
+; (uxtw) extended to 64 bits.
+;   e.g. st1h { z0.d }, p0, [x0, z1.d, uxtw #1]
+;
+
+; ST1H
+define void @sst1h_s_uxtw(<vscale x 4 x i32> %data, <vscale x 4 x i1> %pg, i16* %base, <vscale x 4 x i32> %indices) {
+; CHECK-LABEL: sst1h_s_uxtw:
+; CHECK: st1h { z0.s }, p0, [x0, z1.s, uxtw #1]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 4 x i32> %data to <vscale x 4 x i16>
+  call void @llvm.aarch64.sve.st1.scatter.uxtw.index.nxv4i16(<vscale x 4 x i16> %data_trunc,
+                                                             <vscale x 4 x i1> %pg,
+                                                             i16* %base,
+                                                             <vscale x 4 x i32> %indices)
+  ret void
+}
+
+define void @sst1h_s_sxtw(<vscale x 4 x i32> %data, <vscale x 4 x i1> %pg, i16* %base, <vscale x 4 x i32> %indices) {
+; CHECK-LABEL: sst1h_s_sxtw:
+; CHECK: st1h { z0.s }, p0, [x0, z1.s, sxtw #1]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 4 x i32> %data to <vscale x 4 x i16>
+  call void @llvm.aarch64.sve.st1.scatter.sxtw.index.nxv4i16(<vscale x 4 x i16> %data_trunc,
+                                                             <vscale x 4 x i1> %pg,
+                                                             i16* %base,
+                                                             <vscale x 4 x i32> %indices)
+  ret void
+}
+
+define void @sst1h_d_uxtw(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i16* %base, <vscale x 2 x i32> %indices) {
+; CHECK-LABEL: sst1h_d_uxtw:
+; CHECK: st1h { z0.d }, p0, [x0, z1.d, uxtw #1]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 2 x i64> %data to <vscale x 2 x i16>
+  call void @llvm.aarch64.sve.st1.scatter.uxtw.index.nxv2i16(<vscale x 2 x i16> %data_trunc,
+                                                             <vscale x 2 x i1> %pg,
+                                                             i16* %base,
+                                                             <vscale x 2 x i32> %indices)
+  ret void
+}
+
+define void @sst1h_d_sxtw(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i16* %base, <vscale x 2 x i32> %indices) {
+; CHECK-LABEL: sst1h_d_sxtw:
+; CHECK: st1h { z0.d }, p0, [x0, z1.d, sxtw #1]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 2 x i64> %data to <vscale x 2 x i16>
+  call void @llvm.aarch64.sve.st1.scatter.sxtw.index.nxv2i16(<vscale x 2 x i16> %data_trunc,
+                                                             <vscale x 2 x i1> %pg,
+                                                             i16* %base,
+                                                             <vscale x 2 x i32> %indices)
+  ret void
+}
+
+; ST1W
+define void @sst1w_s_uxtw(<vscale x 4 x i32> %data, <vscale x 4 x i1> %pg, i32* %base, <vscale x 4 x i32> %indices) {
+; CHECK-LABEL: sst1w_s_uxtw:
+; CHECK: st1w { z0.s }, p0, [x0, z1.s, uxtw #2]
+; CHECK-NEXT: ret
+  call void @llvm.aarch64.sve.st1.scatter.uxtw.index.nxv4i32(<vscale x 4 x i32> %data,
+                                                             <vscale x 4 x i1> %pg,
+                                                             i32* %base,
+                                                             <vscale x 4 x i32> %indices)
+  ret void
+}
+
+define void @sst1w_s_sxtw(<vscale x 4 x i32> %data, <vscale x 4 x i1> %pg, i32* %base, <vscale x 4 x i32> %indices) {
+; CHECK-LABEL: sst1w_s_sxtw:
+; CHECK: st1w { z0.s }, p0, [x0, z1.s, sxtw #2]
+; CHECK-NEXT: ret
+  call void @llvm.aarch64.sve.st1.scatter.sxtw.index.nxv4i32(<vscale x 4 x i32> %data,
+                                                             <vscale x 4 x i1> %pg,
+                                                             i32* %base,
+                                                             <vscale x 4 x i32> %indices)
+  ret void
+}
+
+define void @sst1w_d_uxtw(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i32* %base, <vscale x 2 x i32> %indices) {
+; CHECK-LABEL: sst1w_d_uxtw:
+; CHECK: st1w { z0.d }, p0, [x0, z1.d, uxtw #2]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 2 x i64> %data to <vscale x 2 x i32>
+  call void @llvm.aarch64.sve.st1.scatter.uxtw.index.nxv2i32(<vscale x 2 x i32> %data_trunc,
+                                                             <vscale x 2 x i1> %pg,
+                                                             i32* %base,
+                                                             <vscale x 2 x i32> %indices)
+  ret void
+}
+
+define void @sst1w_d_sxtw(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i32* %base, <vscale x 2 x i32> %indices) {
+; CHECK-LABEL: sst1w_d_sxtw:
+; CHECK: st1w { z0.d }, p0, [x0, z1.d, sxtw #2]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 2 x i64> %data to <vscale x 2 x i32>
+  call void @llvm.aarch64.sve.st1.scatter.sxtw.index.nxv2i32(<vscale x 2 x i32> %data_trunc,
+                                                             <vscale x 2 x i1> %pg,
+                                                             i32* %base,
+                                                             <vscale x 2 x i32> %indices)
+  ret void
+}
+
+define void @sst1w_s_uxtw_float(<vscale x 4 x float> %data, <vscale x 4 x i1> %pg, float* %base, <vscale x 4 x i32> %indices) {
+; CHECK-LABEL: sst1w_s_uxtw_float:
+; CHECK: st1w { z0.s }, p0, [x0, z1.s, uxtw #2]
+; CHECK-NEXT: ret
+  call void @llvm.aarch64.sve.st1.scatter.uxtw.index.nxv4f32(<vscale x 4 x float> %data,
+                                                             <vscale x 4 x i1> %pg,
+                                                             float* %base,
+                                                             <vscale x 4 x i32> %indices)
+  ret void
+}
+
+define void @sst1w_s_sxtw_float(<vscale x 4 x float> %data, <vscale x 4 x i1> %pg, float* %base, <vscale x 4 x i32> %indices) {
+; CHECK-LABEL: sst1w_s_sxtw_float:
+; CHECK: st1w { z0.s }, p0, [x0, z1.s, sxtw #2]
+; CHECK-NEXT: ret
+  call void @llvm.aarch64.sve.st1.scatter.sxtw.index.nxv4f32(<vscale x 4 x float> %data,
+                                                             <vscale x 4 x i1> %pg,
+                                                             float* %base,
+                                                             <vscale x 4 x i32> %indices)
+  ret void
+}
+
+; ST1D
+define void @sst1d_d_uxtw(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i64* %base, <vscale x 2 x i32> %indices) {
+; CHECK-LABEL: sst1d_d_uxtw:
+; CHECK: st1d { z0.d }, p0, [x0, z1.d, uxtw #3]
+; CHECK-NEXT: ret
+  call void @llvm.aarch64.sve.st1.scatter.uxtw.index.nxv2i64(<vscale x 2 x i64> %data,
+                                                             <vscale x 2 x i1> %pg,
+                                                             i64* %base,
+                                                             <vscale x 2 x i32> %indices)
+  ret void
+}
+
+define void @sst1d_d_sxtw(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i64* %base, <vscale x 2 x i32> %indices) {
+; CHECK-LABEL: sst1d_d_sxtw:
+; CHECK: st1d { z0.d }, p0, [x0, z1.d, sxtw #3]
+; CHECK-NEXT: ret
+  call void @llvm.aarch64.sve.st1.scatter.sxtw.index.nxv2i64(<vscale x 2 x i64> %data,
+                                                             <vscale x 2 x i1> %pg,
+                                                             i64* %base,
+                                                             <vscale x 2 x i32> %indices)
+  ret void
+}
+
+define void @sst1d_d_uxtw_double(<vscale x 2 x double> %data, <vscale x 2 x i1> %pg, double* %base, <vscale x 2 x i32> %indices) {
+; CHECK-LABEL: sst1d_d_uxtw_double:
+; CHECK: st1d { z0.d }, p0, [x0, z1.d, uxtw #3]
+; CHECK-NEXT: ret
+  call void @llvm.aarch64.sve.st1.scatter.uxtw.index.nxv2f64(<vscale x 2 x double> %data,
+                                                             <vscale x 2 x i1> %pg,
+                                                             double* %base,
+                                                             <vscale x 2 x i32> %indices)
+  ret void
+}
+
+define void @sst1d_d_sxtw_double(<vscale x 2 x double> %data, <vscale x 2 x i1> %pg, double* %base, <vscale x 2 x i32> %indices) {
+; CHECK-LABEL: sst1d_d_sxtw_double:
+; CHECK: st1d { z0.d }, p0, [x0, z1.d, sxtw #3]
+; CHECK-NEXT: ret
+  call void @llvm.aarch64.sve.st1.scatter.sxtw.index.nxv2f64(<vscale x 2 x double> %data,
+                                                             <vscale x 2 x i1> %pg,
+                                                             double* %base,
+                                                             <vscale x 2 x i32> %indices)
+  ret void
+}
+
+
+; ST1H
+declare void @llvm.aarch64.sve.st1.scatter.sxtw.index.nxv4i16(<vscale x 4 x i16>, <vscale x 4 x i1>, i16*, <vscale x 4 x i32>)
+declare void @llvm.aarch64.sve.st1.scatter.sxtw.index.nxv2i16(<vscale x 2 x i16>, <vscale x 2 x i1>, i16*, <vscale x 2 x i32>)
+declare void @llvm.aarch64.sve.st1.scatter.uxtw.index.nxv4i16(<vscale x 4 x i16>, <vscale x 4 x i1>, i16*, <vscale x 4 x i32>)
+declare void @llvm.aarch64.sve.st1.scatter.uxtw.index.nxv2i16(<vscale x 2 x i16>, <vscale x 2 x i1>, i16*, <vscale x 2 x i32>)
+
+; ST1W
+declare void @llvm.aarch64.sve.st1.scatter.sxtw.index.nxv4i32(<vscale x 4 x i32>, <vscale x 4 x i1>, i32*, <vscale x 4 x i32>)
+declare void @llvm.aarch64.sve.st1.scatter.sxtw.index.nxv2i32(<vscale x 2 x i32>, <vscale x 2 x i1>, i32*, <vscale x 2 x i32>)
+declare void @llvm.aarch64.sve.st1.scatter.uxtw.index.nxv4i32(<vscale x 4 x i32>, <vscale x 4 x i1>, i32*, <vscale x 4 x i32>)
+declare void @llvm.aarch64.sve.st1.scatter.uxtw.index.nxv2i32(<vscale x 2 x i32>, <vscale x 2 x i1>, i32*, <vscale x 2 x i32>)
+
+declare void @llvm.aarch64.sve.st1.scatter.sxtw.index.nxv4f32(<vscale x 4 x float>, <vscale x 4 x i1>, float*, <vscale x 4 x i32>)
+declare void @llvm.aarch64.sve.st1.scatter.uxtw.index.nxv4f32(<vscale x 4 x float>, <vscale x 4 x i1>, float*, <vscale x 4 x i32>)
+
+; ST1D
+declare void @llvm.aarch64.sve.st1.scatter.sxtw.index.nxv2i64(<vscale x 2 x i64>, <vscale x 2 x i1>, i64*, <vscale x 2 x i32>)
+declare void @llvm.aarch64.sve.st1.scatter.uxtw.index.nxv2i64(<vscale x 2 x i64>, <vscale x 2 x i1>, i64*, <vscale x 2 x i32>)
+
+declare void @llvm.aarch64.sve.st1.scatter.sxtw.index.nxv2f64(<vscale x 2 x double>, <vscale x 2 x i1>, double*, <vscale x 2 x i32>)
+declare void @llvm.aarch64.sve.st1.scatter.uxtw.index.nxv2f64(<vscale x 2 x double>, <vscale x 2 x i1>, double*, <vscale x 2 x i32>)

llvm/test/CodeGen/AArch64/sve-intrinsics-scatter-stores-32bit-unscaled-offsets.ll

+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve < %s | FileCheck %s
+
+;
+; ST1B, ST1W, ST1H, ST1D: base + 32-bit unscaled offset, sign (sxtw) or zero
+; (uxtw) extended to 64 bits.
+;   e.g. st1h { z0.d }, p0, [x0, z1.d, uxtw]
+;
+
+; ST1B
+define void @sst1b_s_uxtw(<vscale x 4 x i32> %data, <vscale x 4 x i1> %pg, i8* %base, <vscale x 4 x i32> %offsets) {
+; CHECK-LABEL: sst1b_s_uxtw:
+; CHECK: st1b { z0.s }, p0, [x0, z1.s, uxtw]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 4 x i32> %data to <vscale x 4 x i8>
+  call void  @llvm.aarch64.sve.st1.scatter.uxtw.nxv4i8(<vscale x 4 x i8> %data_trunc,
+                                                       <vscale x 4 x i1> %pg,
+                                                       i8* %base,
+                                                       <vscale x 4 x i32> %offsets)
+  ret void
+}
+
+define void @sst1b_s_sxtw(<vscale x 4 x i32> %data, <vscale x 4 x i1> %pg, i8* %base, <vscale x 4 x i32> %offsets) {
+; CHECK-LABEL: sst1b_s_sxtw:
+; CHECK: st1b { z0.s }, p0, [x0, z1.s, sxtw]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 4 x i32> %data to <vscale x 4 x i8>
+  call void @llvm.aarch64.sve.st1.scatter.sxtw.nxv4i8(<vscale x 4 x i8> %data_trunc,
+                                                      <vscale x 4 x i1> %pg,
+                                                      i8* %base,
+                                                      <vscale x 4 x i32> %offsets)
+  ret void
+}
+
+define void @sst1b_d_uxtw(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i8* %base, <vscale x 2 x i32> %offsets) {
+; CHECK-LABEL: sst1b_d_uxtw:
+; CHECK: st1b { z0.d }, p0, [x0, z1.d, uxtw]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 2 x i64> %data to <vscale x 2 x i8>
+  call void @llvm.aarch64.sve.st1.scatter.uxtw.nxv2i8(<vscale x 2 x i8> %data_trunc,
+                                                      <vscale x 2 x i1> %pg,
+                                                      i8* %base,
+                                                      <vscale x 2 x i32> %offsets)
+  ret void
+}
+
+define void @sst1b_d_sxtw(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i8* %base, <vscale x 2 x i32> %offsets) {
+; CHECK-LABEL: sst1b_d_sxtw:
+; CHECK: st1b { z0.d }, p0, [x0, z1.d, sxtw]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 2 x i64> %data to <vscale x 2 x i8>
+  call void  @llvm.aarch64.sve.st1.scatter.sxtw.nxv2i8(<vscale x 2 x i8> %data_trunc,
+                                                       <vscale x 2 x i1> %pg,
+                                                       i8* %base,
+                                                       <vscale x 2 x i32> %offsets)
+  ret void
+}
+
+; ST1H
+define void @sst1h_s_uxtw(<vscale x 4 x i32> %data, <vscale x 4 x i1> %pg, i16* %base, <vscale x 4 x i32> %offsets) {
+; CHECK-LABEL: sst1h_s_uxtw:
+; CHECK: st1h { z0.s }, p0, [x0, z1.s, uxtw]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 4 x i32> %data to <vscale x 4 x i16>
+  call void @llvm.aarch64.sve.st1.scatter.uxtw.nxv4i16(<vscale x 4 x i16> %data_trunc,
+                                                       <vscale x 4 x i1> %pg,
+                                                       i16* %base,
+                                                       <vscale x 4 x i32> %offsets)
+  ret void
+}
+
+define void @sst1h_s_sxtw(<vscale x 4 x i32> %data, <vscale x 4 x i1> %pg, i16* %base, <vscale x 4 x i32> %offsets) {
+; CHECK-LABEL: sst1h_s_sxtw:
+; CHECK: st1h { z0.s }, p0, [x0, z1.s, sxtw]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 4 x i32> %data to <vscale x 4 x i16>
+  call void @llvm.aarch64.sve.st1.scatter.sxtw.nxv4i16(<vscale x 4 x i16> %data_trunc,
+                                                       <vscale x 4 x i1> %pg,
+                                                       i16* %base,
+                                                       <vscale x 4 x i32> %offsets)
+  ret void
+}
+
+define void @sst1h_d_uxtw(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i16* %base, <vscale x 2 x i32> %offsets) {
+; CHECK-LABEL: sst1h_d_uxtw:
+; CHECK: st1h { z0.d }, p0, [x0, z1.d, uxtw]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 2 x i64> %data to <vscale x 2 x i16>
+  call void @llvm.aarch64.sve.st1.scatter.uxtw.nxv2i16(<vscale x 2 x i16> %data_trunc,
+                                                       <vscale x 2 x i1> %pg,
+                                                       i16* %base,
+                                                       <vscale x 2 x i32> %offsets)
+  ret void
+}
+
+define void @sst1h_d_sxtw(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i16* %base, <vscale x 2 x i32> %offsets) {
+; CHECK-LABEL: sst1h_d_sxtw:
+; CHECK: st1h { z0.d }, p0, [x0, z1.d, sxtw]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 2 x i64> %data to <vscale x 2 x i16>
+  call void @llvm.aarch64.sve.st1.scatter.sxtw.nxv2i16(<vscale x 2 x i16> %data_trunc,
+                                                       <vscale x 2 x i1> %pg,
+                                                       i16* %base,
+                                                       <vscale x 2 x i32> %offsets)
+  ret void
+}
+
+; ST1W
+define void @sst1w_s_uxtw(<vscale x 4 x i32> %data, <vscale x 4 x i1> %pg, i32* %base, <vscale x 4 x i32> %offsets) {
+; CHECK-LABEL: sst1w_s_uxtw:
+; CHECK: st1w { z0.s }, p0, [x0, z1.s, uxtw]
+; CHECK-NEXT: ret
+  call void @llvm.aarch64.sve.st1.scatter.uxtw.nxv4i32(<vscale x 4 x i32> %data,
+                                                       <vscale x 4 x i1> %pg,
+                                                       i32* %base,
+                                                       <vscale x 4 x i32> %offsets)
+  ret void
+}
+
+define void @sst1w_s_sxtw(<vscale x 4 x i32> %data, <vscale x 4 x i1> %pg, i32* %base, <vscale x 4 x i32> %offsets) {
+; CHECK-LABEL: sst1w_s_sxtw:
+; CHECK: st1w { z0.s }, p0, [x0, z1.s, sxtw]
+; CHECK-NEXT: ret
+  call void @llvm.aarch64.sve.st1.scatter.sxtw.nxv4i32(<vscale x 4 x i32> %data,
+                                                       <vscale x 4 x i1> %pg,
+                                                       i32* %base,
+                                                       <vscale x 4 x i32> %offsets)
+  ret void
+}
+
+define void @sst1w_d_uxtw(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i32* %base, <vscale x 2 x i32> %offsets) {
+; CHECK-LABEL: sst1w_d_uxtw:
+; CHECK: st1w { z0.d }, p0, [x0, z1.d, uxtw]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 2 x i64> %data to <vscale x 2 x i32>
+  call void @llvm.aarch64.sve.st1.scatter.uxtw.nxv2i32(<vscale x 2 x i32> %data_trunc,
+                                                       <vscale x 2 x i1> %pg,
+                                                       i32* %base,
+                                                       <vscale x 2 x i32> %offsets)
+  ret void
+}
+
+define void @sst1w_d_sxtw(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i32* %base, <vscale x 2 x i32> %offsets) {
+; CHECK-LABEL: sst1w_d_sxtw:
+; CHECK: st1w { z0.d }, p0, [x0, z1.d, sxtw]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 2 x i64> %data to <vscale x 2 x i32>
+  call void @llvm.aarch64.sve.st1.scatter.sxtw.nxv2i32(<vscale x 2 x i32> %data_trunc,
+                                                       <vscale x 2 x i1> %pg,
+                                                       i32* %base,
+                                                       <vscale x 2 x i32> %offsets)
+  ret void
+}
+
+define void @sst1w_s_uxtw_float(<vscale x 4 x float> %data, <vscale x 4 x i1> %pg, float* %base, <vscale x 4 x i32> %offsets) {
+; CHECK-LABEL: sst1w_s_uxtw_float:
+; CHECK: st1w { z0.s }, p0, [x0, z1.s, uxtw]
+; CHECK-NEXT: ret
+  call void @llvm.aarch64.sve.st1.scatter.uxtw.nxv4f32(<vscale x 4 x float> %data,
+                                                       <vscale x 4 x i1> %pg,
+                                                       float* %base,
+                                                       <vscale x 4 x i32> %offsets)
+  ret void
+}
+
+define void @sst1w_s_sxtw_float(<vscale x 4 x float> %data, <vscale x 4 x i1> %pg, float* %base, <vscale x 4 x i32> %offsets) {
+; CHECK-LABEL: sst1w_s_sxtw_float:
+; CHECK: st1w { z0.s }, p0, [x0, z1.s, sxtw]
+; CHECK-NEXT: ret
+  call void @llvm.aarch64.sve.st1.scatter.sxtw.nxv4f32(<vscale x 4 x float> %data,
+                                                       <vscale x 4 x i1> %pg,
+                                                       float* %base,
+                                                       <vscale x 4 x i32> %offsets)
+  ret void
+}
+
+; ST1D
+define void @sst1d_d_uxtw(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i64* %base, <vscale x 2 x i32> %offsets) {
+; CHECK-LABEL: sst1d_d_uxtw:
+; CHECK: st1d { z0.d }, p0, [x0, z1.d, uxtw]
+; CHECK-NEXT: ret
+  call void @llvm.aarch64.sve.st1.scatter.uxtw.nxv2i64(<vscale x 2 x i64> %data,
+                                                       <vscale x 2 x i1> %pg,
+                                                       i64* %base,
+                                                       <vscale x 2 x i32> %offsets)
+  ret void
+}
+
+define void @sst1d_d_sxtw(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i64* %base, <vscale x 2 x i32> %offsets) {
+; CHECK-LABEL: sst1d_d_sxtw:
+; CHECK: st1d { z0.d }, p0, [x0, z1.d, sxtw]
+; CHECK-NEXT: ret
+  call void @llvm.aarch64.sve.st1.scatter.sxtw.nxv2i64(<vscale x 2 x i64> %data,
+                                                       <vscale x 2 x i1> %pg,
+                                                       i64* %base,
+                                                       <vscale x 2 x i32> %offsets)
+  ret void
+}
+
+define void @sst1d_d_uxtw_double(<vscale x 2 x double> %data, <vscale x 2 x i1> %pg, double* %base, <vscale x 2 x i32> %offsets) {
+; CHECK-LABEL: sst1d_d_uxtw_double:
+; CHECK: st1d { z0.d }, p0, [x0, z1.d, uxtw]
+; CHECK-NEXT: ret
+  call void @llvm.aarch64.sve.st1.scatter.uxtw.nxv2f64(<vscale x 2 x double> %data,
+                                                       <vscale x 2 x i1> %pg,
+                                                       double* %base,
+                                                       <vscale x 2 x i32> %offsets)
+  ret void
+}
+
+define void @sst1d_d_sxtw_double(<vscale x 2 x double> %data, <vscale x 2 x i1> %pg, double* %base, <vscale x 2 x i32> %offsets) {
+; CHECK-LABEL: sst1d_d_sxtw_double:
+; CHECK: st1d { z0.d }, p0, [x0, z1.d, sxtw]
+; CHECK-NEXT: ret
+  call void @llvm.aarch64.sve.st1.scatter.sxtw.nxv2f64(<vscale x 2 x double> %data,
+                                                       <vscale x 2 x i1> %pg,
+                                                       double* %base,
+                                                       <vscale x 2 x i32> %offsets)
+  ret void
+}
+
+
+; ST1B
+declare void @llvm.aarch64.sve.st1.scatter.uxtw.nxv4i8(<vscale x 4 x i8>, <vscale x 4 x i1>, i8*, <vscale x 4 x i32>)
+declare void @llvm.aarch64.sve.st1.scatter.uxtw.nxv2i8(<vscale x 2 x i8>, <vscale x 2 x i1>, i8*, <vscale x 2 x i32>)
+declare void @llvm.aarch64.sve.st1.scatter.sxtw.nxv4i8(<vscale x 4 x i8>, <vscale x 4 x i1>, i8*, <vscale x 4 x i32>)
+declare void @llvm.aarch64.sve.st1.scatter.sxtw.nxv2i8(<vscale x 2 x i8>, <vscale x 2 x i1>, i8*, <vscale x 2 x i32>)
+
+; ST1H
+declare void @llvm.aarch64.sve.st1.scatter.sxtw.nxv4i16(<vscale x 4 x i16>, <vscale x 4 x i1>, i16*, <vscale x 4 x i32>)
+declare void @llvm.aarch64.sve.st1.scatter.sxtw.nxv2i16(<vscale x 2 x i16>, <vscale x 2 x i1>, i16*, <vscale x 2 x i32>)
+declare void @llvm.aarch64.sve.st1.scatter.uxtw.nxv4i16(<vscale x 4 x i16>, <vscale x 4 x i1>, i16*, <vscale x 4 x i32>)
+declare void @llvm.aarch64.sve.st1.scatter.uxtw.nxv2i16(<vscale x 2 x i16>, <vscale x 2 x i1>, i16*, <vscale x 2 x i32>)
+
+; ST1W
+declare void @llvm.aarch64.sve.st1.scatter.sxtw.nxv4i32(<vscale x 4 x i32>, <vscale x 4 x i1>, i32*, <vscale x 4 x i32>)
+declare void @llvm.aarch64.sve.st1.scatter.sxtw.nxv2i32(<vscale x 2 x i32>, <vscale x 2 x i1>, i32*, <vscale x 2 x i32>)
+declare void @llvm.aarch64.sve.st1.scatter.uxtw.nxv4i32(<vscale x 4 x i32>, <vscale x 4 x i1>, i32*, <vscale x 4 x i32>)
+declare void @llvm.aarch64.sve.st1.scatter.uxtw.nxv2i32(<vscale x 2 x i32>, <vscale x 2 x i1>, i32*, <vscale x 2 x i32>)
+
+declare void @llvm.aarch64.sve.st1.scatter.sxtw.nxv4f32(<vscale x 4 x float>, <vscale x 4 x i1>, float*, <vscale x 4 x i32>)
+declare void @llvm.aarch64.sve.st1.scatter.uxtw.nxv4f32(<vscale x 4 x float>, <vscale x 4 x i1>, float*, <vscale x 4 x i32>)
+
+; ST1D
+declare void @llvm.aarch64.sve.st1.scatter.sxtw.nxv2i64(<vscale x 2 x i64>, <vscale x 2 x i1>, i64*, <vscale x 2 x i32>)
+declare void @llvm.aarch64.sve.st1.scatter.uxtw.nxv2i64(<vscale x 2 x i64>, <vscale x 2 x i1>, i64*, <vscale x 2 x i32>)
+
+declare void @llvm.aarch64.sve.st1.scatter.sxtw.nxv2f64(<vscale x 2 x double>, <vscale x 2 x i1>, double*, <vscale x 2 x i32>)
+declare void @llvm.aarch64.sve.st1.scatter.uxtw.nxv2f64(<vscale x 2 x double>, <vscale x 2 x i1>, double*, <vscale x 2 x i32>)

llvm/test/CodeGen/AArch64/sve-intrinsics-scatter-stores-64bit-scaled-offset.ll

+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve < %s | FileCheck %s
+
+;
+; ST1H, ST1W, ST1D: base + 64-bit scaled offset
+;   e.g. st1h { z0.d }, p0, [x0, z0.d, lsl #1]
+;
+
+define void @sst1h_index(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i16* %base, <vscale x 2 x i64> %offsets) {
+; CHECK-LABEL: sst1h_index
+; CHECK:      st1h  { z0.d }, p0, [x0, z1.d, lsl #1]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 2 x i64> %data to <vscale x 2 x i16>
+  call void @llvm.aarch64.sve.st1.scatter.index.nxv2i16(<vscale x 2 x i16> %data_trunc, 
+                                                        <vscale x 2 x i1> %pg,
+                                                        i16* %base,
+                                                        <vscale x 2 x i64> %offsets)
+  ret void
+}
+
+define void @sst1w_index(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i32* %base, <vscale x 2 x i64> %offsets) {
+; CHECK-LABEL: sst1w_index
+; CHECK:      st1w  { z0.d }, p0, [x0, z1.d, lsl #2]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 2 x i64> %data to <vscale x 2 x i32>
+  call void @llvm.aarch64.sve.st1.scatter.index.nxv2i32(<vscale x 2 x i32> %data_trunc,
+                                                        <vscale x 2 x i1> %pg,
+                                                        i32* %base,
+                                                        <vscale x 2 x i64> %offsets)
+  ret void
+}
+
+define void  @sst1d_index(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i64* %base, <vscale x 2 x i64> %offsets) {
+; CHECK-LABEL: sst1d_index
+; CHECK:      st1d  { z0.d }, p0, [x0, z1.d, lsl #3]
+; CHECK-NEXT: ret
+  call void @llvm.aarch64.sve.st1.scatter.index.nxv2i64(<vscale x 2 x i64> %data,
+                                                        <vscale x 2 x i1> %pg,
+                                                        i64* %base,
+                                                        <vscale x 2 x i64> %offsets)
+  ret void
+}
+
+define void  @sst1d_index_double(<vscale x 2 x double> %data, <vscale x 2 x i1> %pg, double* %base, <vscale x 2 x i64> %offsets) {
+; CHECK-LABEL: sst1d_index_double
+; CHECK:      st1d  { z0.d }, p0, [x0, z1.d, lsl #3]
+; CHECK-NEXT: ret
+  call void @llvm.aarch64.sve.st1.scatter.index.nxv2f64(<vscale x 2 x double> %data,
+                                                        <vscale x 2 x i1> %pg,
+                                                        double* %base,
+                                                        <vscale x 2 x i64> %offsets)
+  ret void
+}
+
+
+declare void @llvm.aarch64.sve.st1.scatter.index.nxv2i16(<vscale x 2 x i16>, <vscale x 2 x i1>, i16*, <vscale x 2 x i64>)
+declare void @llvm.aarch64.sve.st1.scatter.index.nxv2i32(<vscale x 2 x i32>, <vscale x 2 x i1>, i32*, <vscale x 2 x i64>)
+declare void @llvm.aarch64.sve.st1.scatter.index.nxv2i64(<vscale x 2 x i64>, <vscale x 2 x i1>, i64*, <vscale x 2 x i64>)
+declare void @llvm.aarch64.sve.st1.scatter.index.nxv2f64(<vscale x 2 x double>, <vscale x 2 x i1>, double*, <vscale x 2 x i64>)

llvm/test/CodeGen/AArch64/sve-intrinsics-scatter-stores-64bit-unscaled-offset.ll

+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve < %s | FileCheck %s
+
+;
+; ST1B, ST1W, ST1H, ST1D: base + 64-bit unscaled offset
+;   e.g. st1h { z0.d }, p0, [x0, z1.d]
+;
+
+define void @sst1b_d(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i8* %base, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: sst1b_d:
+; CHECK: st1b { z0.d }, p0, [x0, z1.d]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 2 x i64> %data to <vscale x 2 x i8>
+  call void @llvm.aarch64.sve.st1.scatter.nxv2i8(<vscale x 2 x i8> %data_trunc,
+                                                 <vscale x 2 x i1> %pg,
+                                                 i8* %base,
+                                                 <vscale x 2 x i64> %b)
+  ret void
+}
+
+define void @sst1h_d(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i16* %base, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: sst1h_d:
+; CHECK: st1h { z0.d }, p0, [x0, z1.d]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 2 x i64> %data to <vscale x 2 x i16>
+  call void @llvm.aarch64.sve.st1.scatter.nxv2i16(<vscale x 2 x i16> %data_trunc,
+                                                 <vscale x 2 x i1> %pg,
+                                                 i16* %base,
+                                                 <vscale x 2 x i64> %b)
+  ret void
+}
+
+define void @sst1w_d(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i32* %base, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: sst1w_d:
+; CHECK: st1w { z0.d }, p0, [x0, z1.d]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 2 x i64> %data to <vscale x 2 x i32>
+  call void @llvm.aarch64.sve.st1.scatter.nxv2i32(<vscale x 2 x i32> %data_trunc,
+                                                 <vscale x 2 x i1> %pg,
+                                                 i32* %base,
+                                                 <vscale x 2 x i64> %b)
+  ret void
+}
+
+define void @sst1d_d(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i64* %base, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: sst1d_d:
+; CHECK: st1d { z0.d }, p0, [x0, z1.d]
+; CHECK-NEXT: ret
+  call void @llvm.aarch64.sve.st1.scatter.nxv2i64(<vscale x 2 x i64> %data,
+                                                 <vscale x 2 x i1> %pg,
+                                                 i64* %base,
+                                                 <vscale x 2 x i64> %b)
+  ret void
+}
+
+define void @sst1d_d_double(<vscale x 2 x double> %data, <vscale x 2 x i1> %pg, double* %base, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: sst1d_d_double:
+; CHECK: st1d { z0.d }, p0, [x0, z1.d]
+; CHECK-NEXT: ret
+  call void @llvm.aarch64.sve.st1.scatter.nxv2f64(<vscale x 2 x double> %data,
+                                                 <vscale x 2 x i1> %pg,
+                                                 double* %base,
+                                                 <vscale x 2 x i64> %b)
+  ret void
+}
+
+declare void @llvm.aarch64.sve.st1.scatter.nxv2i8(<vscale x 2 x i8>, <vscale x 2 x i1>, i8*, <vscale x 2 x i64>)
+declare void @llvm.aarch64.sve.st1.scatter.nxv2i16(<vscale x 2 x i16>, <vscale x 2 x i1>, i16*, <vscale x 2 x i64>)
+declare void @llvm.aarch64.sve.st1.scatter.nxv2i32(<vscale x 2 x i32>, <vscale x 2 x i1>, i32*, <vscale x 2 x i64>)
+declare void @llvm.aarch64.sve.st1.scatter.nxv2i64(<vscale x 2 x i64>, <vscale x 2 x i1>, i64*, <vscale x 2 x i64>)
+declare void @llvm.aarch64.sve.st1.scatter.nxv2f64(<vscale x 2 x double>, <vscale x 2 x i1>, double*, <vscale x 2 x i64>)

llvm/test/CodeGen/AArch64/sve-intrinsics-scatter-stores-vector-base.ll

+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve < %s | FileCheck %s
+
+;
+; ST1B, ST1W, ST1H, ST1D: vector + immediate (index)
+;   e.g. st1h { z0.s }, p0, [z1.s, #16]
+;
+
+; ST1B
+define void @sst1b_s_imm(<vscale x 4 x i32> %data, <vscale x 4 x i1> %pg, <vscale x 4 x i32> %base) {
+; CHECK-LABEL: sst1b_s_imm:
+; CHECK: st1b { z0.s }, p0, [z1.s, #16]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 4 x i32> %data to <vscale x 4 x i8>
+  call void @llvm.aarch64.sve.st1.scatter.imm.nxv4i8.nxv4i32(<vscale x 4 x i8> %data_trunc,
+                                                             <vscale x 4 x i1> %pg,
+                                                             <vscale x 4 x i32> %base,
+                                                             i64 16)
+  ret void
+}
+
+define void @sst1b_d_imm(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, <vscale x 2 x i64> %base) {
+; CHECK-LABEL: sst1b_d_imm:
+; CHECK: st1b { z0.d }, p0, [z1.d, #16]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 2 x i64> %data to <vscale x 2 x i8>
+  call void @llvm.aarch64.sve.st1.scatter.imm.nxv2i8.nxv2i64(<vscale x 2 x i8> %data_trunc,
+                                                             <vscale x 2 x i1> %pg,
+                                                             <vscale x 2 x i64> %base,
+                                                             i64 16)
+  ret void
+}
+
+; ST1H
+define void @sst1h_s_imm(<vscale x 4 x i32> %data, <vscale x 4 x i1> %pg, <vscale x 4 x i32> %base) {
+; CHECK-LABEL: sst1h_s_imm:
+; CHECK: st1h { z0.s }, p0, [z1.s, #16]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 4 x i32> %data to <vscale x 4 x i16>
+  call void @llvm.aarch64.sve.st1.scatter.imm.nxv4i16.nxv4i32(<vscale x 4 x i16> %data_trunc,
+                                                              <vscale x 4 x i1> %pg,
+                                                              <vscale x 4 x i32> %base,
+                                                              i64 16)
+  ret void
+}
+
+define void @sst1h_d_imm(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, <vscale x 2 x i64> %base) {
+; CHECK-LABEL: sst1h_d_imm:
+; CHECK: st1h { z0.d }, p0, [z1.d, #16]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 2 x i64> %data to <vscale x 2 x i16>
+  call void @llvm.aarch64.sve.st1.scatter.imm.nxv2i16.nxv2i64(<vscale x 2 x i16> %data_trunc,
+                                                              <vscale x 2 x i1> %pg,
+                                                              <vscale x 2 x i64> %base,
+                                                              i64 16)
+  ret void
+}
+
+; ST1W
+define void @sst1w_s_imm(<vscale x 4 x i32> %data, <vscale x 4 x i1> %pg, <vscale x 4 x i32> %base) {
+; CHECK-LABEL: sst1w_s_imm:
+; CHECK: st1w { z0.s }, p0, [z1.s, #16]
+; CHECK-NEXT: ret
+  call void @llvm.aarch64.sve.st1.scatter.imm.nxv4i32.nxv4i32(<vscale x 4 x i32> %data,
+                                                              <vscale x 4 x i1> %pg,
+                                                              <vscale x 4 x i32> %base,
+                                                              i64 16)
+  ret void
+}
+
+define void @sst1w_d_imm(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, <vscale x 2 x i64> %base) {
+; CHECK-LABEL: sst1w_d_imm:
+; CHECK: st1w { z0.d }, p0, [z1.d, #16]
+; CHECK-NEXT: ret
+  %data_trunc = trunc <vscale x 2 x i64> %data to <vscale x 2 x i32>
+  call void @llvm.aarch64.sve.st1.scatter.imm.nxv2i32.nxv2i64(<vscale x 2 x i32> %data_trunc,
+                                                              <vscale x 2 x i1> %pg,
+                                                              <vscale x 2 x i64> %base,
+                                                              i64 16)
+  ret void
+}
+
+define void @sst1w_s_imm_float(<vscale x 4 x float> %data, <vscale x 4 x i1> %pg, <vscale x 4 x i32> %base) {
+; CHECK-LABEL: sst1w_s_imm_float:
+; CHECK: st1w { z0.s }, p0, [z1.s, #16]
+; CHECK-NEXT: ret
+  call void @llvm.aarch64.sve.st1.scatter.imm.nxv4f32.nxv4i32(<vscale x 4 x float> %data,
+                                                              <vscale x 4 x i1> %pg,
+                                                              <vscale x 4 x i32> %base,
+                                                              i64 16)
+  ret void
+}
+
+; ST1D
+define void @sst1d_d_imm(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, <vscale x 2 x i64> %base) {
+; CHECK-LABEL: sst1d_d_imm:
+; CHECK: st1d { z0.d }, p0, [z1.d, #16]
+; CHECK-NEXT: ret
+  call void @llvm.aarch64.sve.st1.scatter.imm.nxv2i64.nxv2i64(<vscale x 2 x i64> %data,
+                                                              <vscale x 2 x i1> %pg,
+                                                              <vscale x 2 x i64> %base,
+                                                              i64 16)
+  ret void
+}
+
+define void @sst1d_d_imm_double(<vscale x 2 x double> %data, <vscale x 2 x i1> %pg, <vscale x 2 x i64> %base) {
+; CHECK-LABEL: sst1d_d_imm_double:
+; CHECK: st1d { z0.d }, p0, [z1.d, #16]
+; CHECK-NEXT: ret
+  call void @llvm.aarch64.sve.st1.scatter.imm.nxv2f64.nxv2i64(<vscale x 2 x double> %data,
+                                                              <vscale x 2 x i1> %pg,
+                                                              <vscale x 2 x i64> %base,
+                                                              i64 16)
+  ret void
+}
+
+; ST1B
+declare void @llvm.aarch64.sve.st1.scatter.imm.nxv4i8.nxv4i32(<vscale x 4 x i8>, <vscale x 4 x i1>, <vscale x 4 x i32>, i64)
+declare void @llvm.aarch64.sve.st1.scatter.imm.nxv2i8.nxv2i64(<vscale x 2 x i8>, <vscale x 2 x i1>, <vscale x 2 x i64>, i64)
+
+; ST1H
+declare void @llvm.aarch64.sve.st1.scatter.imm.nxv4i16.nxv4i32(<vscale x 4 x i16>, <vscale x 4 x i1>, <vscale x 4 x i32>, i64)
+declare void @llvm.aarch64.sve.st1.scatter.imm.nxv2i16.nxv2i64(<vscale x 2 x i16>, <vscale x 2 x i1>, <vscale x 2 x i64>, i64)
+
+; ST1W
+declare void @llvm.aarch64.sve.st1.scatter.imm.nxv4i32.nxv4i32(<vscale x 4 x i32>, <vscale x 4 x i1>, <vscale x 4 x i32>, i64)
+declare void @llvm.aarch64.sve.st1.scatter.imm.nxv2i32.nxv2i64(<vscale x 2 x i32>, <vscale x 2 x i1>, <vscale x 2 x i64>, i64)
+
+declare void @llvm.aarch64.sve.st1.scatter.imm.nxv4f32.nxv4i32(<vscale x 4 x float>, <vscale x 4 x i1>, <vscale x 4 x i32>, i64)
+
+; ST1D
+declare void @llvm.aarch64.sve.st1.scatter.imm.nxv2i64.nxv2i64(<vscale x 2 x i64>, <vscale x 2 x i1>, <vscale x 2 x i64>, i64)
+
+declare void @llvm.aarch64.sve.st1.scatter.imm.nxv2f64.nxv2i64(<vscale x 2 x double>, <vscale x 2 x i1>, <vscale x 2 x i64>, i64)