[AArch64][SVE] Add integer arithmetic with immediate instructions.

[dancgr]

Summary

Add pattern matching for the following instructions:

• add, sub, subr, sqadd, sqsub, uqadd, uqsub

This patch required complex patterns to match the immediate with optinal left shift.

I re-used the Select function from the other SVE repo to implement the complext pattern.

I plan on doing another patch to also match constant vector of the same immediate.

[efriedma]

It's a little unfortunate you have to use ComplexPattern here, but I guess it's necessary given the way imm8_opt_lsl is represented.

llvm/include/llvm/IR/IntrinsicsAArch64.td

1050

Missing ImmArg marking. Second arg should be llvm_i32_ty.

llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp

2832

Sign-extended where, exactly? I don't see any reason to allow @llvm.aarch64.sve.uqsub.imm.nxv8i16(<vscale x 8 x i16> %a, i32 -256)

[dancgr]

llvm/include/llvm/IR/IntrinsicsAArch64.td

1050

Updated that.

llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp

2832

That makes sense. We don't have negative immediates for the AddSub, so it should be the same as i32 and i64. Will updated in the next patch.

[dancgr]

• [AArch64][SVE] Update intrinsic to include Immediate option and remove sign handling for i16 immediate for add/sub.

[dancgr]

• [AArch64][SVE] Update intrinsic to include Immediate option and remove sign handling for i16 immediate for add/sub.

[efriedma]

llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp

2828

I'd prefer to reject immediates greater than 255 here, instead of masking off the bits.

[c-rhodes]

llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp

2821

can this be getZExtValue now i16 is being handled the same as i32 and i64?

llvm/lib/Target/AArch64/SVEInstrFormats.td

296

SVE_1_Op_Imm_OptLsl_Pat

297

nit: formatting

llvm/test/CodeGen/AArch64/sve-int-imm.ll

9

nit: imm arg isn't directly under first arg, and below.

[dancgr]

Fix some formatting and apply suggested changes.

[dancgr]

llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp

2821

Yes, we don't need sext anymore. Will be updating this.

2828

In the immediate definition we have:

def addsub_imm8_opt_lsl_i8  : imm8_opt_lsl<8,  "uint8_t",  SVEAddSubImmOperand8,  [{
  return AArch64_AM::isSVEAddSubImm<int8_t>(Imm);
}]>;

This checks if the immediate is in the accepted range for AddSub. Would you say we should also add another check here?

I pasted the code of that function below, since it is not in the changed files from this patch:

template <typename T>
static inline bool isSVEAddSubImm(int64_t Imm) {
  bool IsInt8t =
      std::is_same<int8_t, typename std::make_signed<T>::type>::value;
  return uint8_t(Imm) == Imm || (!IsInt8t && uint16_t(Imm & ~0xff) == Imm);
}

llvm/lib/Target/AArch64/SVEInstrFormats.td

193

Check for SVE AddSub Imm

[efriedma]

llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp

2828

Not sure that TableGen code is actually getting called from isel in this situation? If it is, fine, we don't need a redundant check.

[dancgr]

Add extra check to DAGToDAGISel to avoid invalid immediates.

[dancgr]

llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp

2828

I was able to reproduce some instances where invalid immediates would get past that check. I have added that extra check.

[efriedma]

llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp

2828

Please put the "return true" inside the if statement.

2842

Missing "break" here. (It doesn't have any semantic effect, obviously, but you'll get a switch fallthrough on some compilers, I think.)

[dancgr]

Update minor details.

[efriedma]

LGTM

llvm/include/llvm/IR/IntrinsicsAArch64.td

class AdvSIMD_GatherLoad_VecTorBase_Intrinsic
     : Intrinsic<[llvm_anyvector_ty],
                 [
                   LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
                   llvm_anyvector_ty,
                   llvm_i64_ty
                 ],
                 [IntrReadMem, IntrArgMemOnly]>;
 
+class AdvSIMD_1VectorArg_Imm_Intrinsic
+    : Intrinsic<[llvm_anyvector_ty],
+                [LLVMMatchType<0>,
+                 llvm_i32_ty],
+                [IntrNoMem, ImmArg<1>]>;
+
 //
 // 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;
 
+def int_aarch64_sve_add_imm    : AdvSIMD_1VectorArg_Imm_Intrinsic;
+def int_aarch64_sve_sub_imm    : AdvSIMD_1VectorArg_Imm_Intrinsic;
+def int_aarch64_sve_subr_imm   : AdvSIMD_1VectorArg_Imm_Intrinsic;
+def int_aarch64_sve_sqadd_imm  : AdvSIMD_1VectorArg_Imm_Intrinsic;
+def int_aarch64_sve_uqadd_imm  : AdvSIMD_1VectorArg_Imm_Intrinsic;
+def int_aarch64_sve_sqsub_imm  : AdvSIMD_1VectorArg_Imm_Intrinsic;
+def int_aarch64_sve_uqsub_imm  : AdvSIMD_1VectorArg_Imm_Intrinsic;
+
 def int_aarch64_sve_mul        : AdvSIMD_Pred2VectorArg_Intrinsic;
 def int_aarch64_sve_smulh      : AdvSIMD_Pred2VectorArg_Intrinsic;
 def int_aarch64_sve_umulh      : AdvSIMD_Pred2VectorArg_Intrinsic;
 
 def int_aarch64_sve_sdiv       : AdvSIMD_Pred2VectorArg_Intrinsic;
 def int_aarch64_sve_udiv       : AdvSIMD_Pred2VectorArg_Intrinsic;
 def int_aarch64_sve_sdivr      : AdvSIMD_Pred2VectorArg_Intrinsic;
 def int_aarch64_sve_udivr      : AdvSIMD_Pred2VectorArg_Intrinsic;

llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp

bool SelectDupZeroOrUndef(SDValue N) {
     }
     default:
       break;
     }
 
     return false;
   }
 
+  template<MVT::SimpleValueType VT>
+  bool SelectSVEAddSubImm(SDValue N, SDValue &Imm, SDValue &Shift) {
+    return SelectSVEAddSubImm(N, VT, Imm, Shift);
+  }
+
   /// Form sequences of consecutive 64/128-bit registers for use in NEON
   /// instructions making use of a vector-list (e.g. ldN, tbl). Vecs must have
   /// between 1 and 4 elements. If it contains a single element that is returned
   /// unchanged; otherwise a REG_SEQUENCE value is returned.
   SDValue createDTuple(ArrayRef<SDValue> Vecs);
   SDValue createQTuple(ArrayRef<SDValue> Vecs);
 
   /// Generic helper for the createDTuple/createQTuple
private:
   bool SelectCVTFixedPosOperand(SDValue N, SDValue &FixedPos) {
     return SelectCVTFixedPosOperand(N, FixedPos, RegWidth);
   }
 
   bool SelectCVTFixedPosOperand(SDValue N, SDValue &FixedPos, unsigned Width);
 
   bool SelectCMP_SWAP(SDNode *N);
 
+  bool SelectSVEAddSubImm(SDValue N, MVT VT, SDValue &Imm, SDValue &Shift);
 };
 } // end anonymous namespace
 
 /// isIntImmediate - This method tests to see if the node is a constant
 /// operand. If so Imm will receive the 32-bit value.
 static bool isIntImmediate(const SDNode *N, uint64_t &Imm) {
   if (const ConstantSDNode *C = dyn_cast<const ConstantSDNode>(N)) {
     Imm = C->getZExtValue();
bool AArch64DAGToDAGISel::SelectCMP_SWAP(SDNode *N) {
 
   ReplaceUses(SDValue(N, 0), SDValue(CmpSwap, 0));
   ReplaceUses(SDValue(N, 1), SDValue(CmpSwap, 2));
   CurDAG->RemoveDeadNode(N);
 
   return true;
 }
 
+bool AArch64DAGToDAGISel::SelectSVEAddSubImm(SDValue N, MVT VT, SDValue &Imm, SDValue &Shift) {
+  if (auto CNode = dyn_cast<ConstantSDNode>(N)) {
+    const int64_t ImmVal = CNode->getZExtValue();
+    SDLoc DL(N);
+
+    switch (VT.SimpleTy) {
+    case MVT::i8:
+      Shift = CurDAG->getTargetConstant(0, DL, MVT::i32);
+      Imm = CurDAG->getTargetConstant(ImmVal & 0xFF, DL, MVT::i32);
+      return true;
+    case MVT::i16:
+    case MVT::i32:
+    case MVT::i64:
+      if ((ImmVal & 0xFF) == ImmVal) {
+        Shift = CurDAG->getTargetConstant(0, DL, MVT::i32);
+        Imm = CurDAG->getTargetConstant(ImmVal, DL, MVT::i32);
+        return true;
+      } else if ((ImmVal & 0xFF00) == ImmVal) {
+        Shift = CurDAG->getTargetConstant(8, DL, MVT::i32);
+        Imm = CurDAG->getTargetConstant(ImmVal >> 8, DL, MVT::i32);
+        return true;
+      }
+    default:
+      break;
+    }
+  }
+
+  return false;
+}
+
+
 bool AArch64DAGToDAGISel::trySelectStackSlotTagP(SDNode *N) {
   // tagp(FrameIndex, IRGstack, tag_offset):
   // since the offset between FrameIndex and IRGstack is a compile-time
   // constant, this can be lowered to a single ADDG instruction.
   if (!(isa<FrameIndexSDNode>(N->getOperand(1)))) {
     return false;
   }
 

llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td

let Predicates = [HasSVE] in {
   defm SUB_ZPmZ   : sve_int_bin_pred_arit_0<0b001, "sub", int_aarch64_sve_sub>;
   defm SUBR_ZPmZ  : sve_int_bin_pred_arit_0<0b011, "subr", int_aarch64_sve_subr>;
 
   defm ORR_ZPmZ : sve_int_bin_pred_log<0b000, "orr", int_aarch64_sve_or>;
   defm EOR_ZPmZ : sve_int_bin_pred_log<0b001, "eor", int_aarch64_sve_xor>;
   defm AND_ZPmZ : sve_int_bin_pred_log<0b010, "and", int_aarch64_sve_and>;
   defm BIC_ZPmZ : sve_int_bin_pred_log<0b011, "bic", int_aarch64_sve_bic>;
 
-  defm ADD_ZI   : sve_int_arith_imm0<0b000, "add">;
-  defm SUB_ZI   : sve_int_arith_imm0<0b001, "sub">;
-  defm SUBR_ZI  : sve_int_arith_imm0<0b011, "subr">;
-  defm SQADD_ZI : sve_int_arith_imm0<0b100, "sqadd">;
-  defm UQADD_ZI : sve_int_arith_imm0<0b101, "uqadd">;
-  defm SQSUB_ZI : sve_int_arith_imm0<0b110, "sqsub">;
-  defm UQSUB_ZI : sve_int_arith_imm0<0b111, "uqsub">;
+  defm ADD_ZI   : sve_int_arith_imm0<0b000, "add", int_aarch64_sve_add_imm>;
+  defm SUB_ZI   : sve_int_arith_imm0<0b001, "sub", int_aarch64_sve_sub_imm>;
+  defm SUBR_ZI  : sve_int_arith_imm0<0b011, "subr", int_aarch64_sve_subr_imm>;
+  defm SQADD_ZI : sve_int_arith_imm0<0b100, "sqadd", int_aarch64_sve_sqadd_imm>;
+  defm UQADD_ZI : sve_int_arith_imm0<0b101, "uqadd", int_aarch64_sve_uqadd_imm>;
+  defm SQSUB_ZI : sve_int_arith_imm0<0b110, "sqsub", int_aarch64_sve_sqsub_imm>;
+  defm UQSUB_ZI : sve_int_arith_imm0<0b111, "uqsub", int_aarch64_sve_uqsub_imm>;
 
   defm MAD_ZPmZZ : sve_int_mladdsub_vvv_pred<0b0, "mad", int_aarch64_sve_mad>;
   defm MSB_ZPmZZ : sve_int_mladdsub_vvv_pred<0b1, "msb", int_aarch64_sve_msb>;
   defm MLA_ZPmZZ : sve_int_mlas_vvv_pred<0b0, "mla", int_aarch64_sve_mla>;
   defm MLS_ZPmZZ : sve_int_mlas_vvv_pred<0b1, "mls", int_aarch64_sve_mls>;
 
   // SVE predicated integer reductions.
   defm SADDV_VPZ : sve_int_reduce_0_saddv<0b000, "saddv", int_aarch64_sve_saddv>;

llvm/lib/Target/AArch64/SVEInstrFormats.td

 def cpy_imm8_opt_lsl_i32 : imm8_opt_lsl<32, "int32_t", SVECpyImmOperand32, [{
   return AArch64_AM::isSVECpyImm<int32_t>(Imm);
 }]>;
 def cpy_imm8_opt_lsl_i64 : imm8_opt_lsl<64, "int64_t", SVECpyImmOperand64, [{
   return AArch64_AM::isSVECpyImm<int64_t>(Imm);
 }]>;
 
 def addsub_imm8_opt_lsl_i8  : imm8_opt_lsl<8,  "uint8_t",  SVEAddSubImmOperand8,  [{
   return AArch64_AM::isSVEAddSubImm<int8_t>(Imm);
 }]>;
 def addsub_imm8_opt_lsl_i16 : imm8_opt_lsl<16, "uint16_t", SVEAddSubImmOperand16, [{
   return AArch64_AM::isSVEAddSubImm<int16_t>(Imm);
 }]>;
 def addsub_imm8_opt_lsl_i32 : imm8_opt_lsl<32, "uint32_t", SVEAddSubImmOperand32, [{
   return AArch64_AM::isSVEAddSubImm<int32_t>(Imm);
 }]>;
 def addsub_imm8_opt_lsl_i64 : imm8_opt_lsl<64, "uint64_t", SVEAddSubImmOperand64, [{
   return AArch64_AM::isSVEAddSubImm<int64_t>(Imm);
 }]>;
 
+def SVEAddSubImm8Pat  : ComplexPattern<i32, 2, "SelectSVEAddSubImm<MVT::i8>", []>;
+def SVEAddSubImm16Pat : ComplexPattern<i32, 2, "SelectSVEAddSubImm<MVT::i16>", []>;
+def SVEAddSubImm32Pat : ComplexPattern<i32, 2, "SelectSVEAddSubImm<MVT::i32>", []>;
+def SVEAddSubImm64Pat : ComplexPattern<i32, 2, "SelectSVEAddSubImm<MVT::i64>", []>;
+
 class SVEExactFPImm<string Suffix, string ValA, string ValB> : AsmOperandClass {
   let Name = "SVEExactFPImmOperand" # Suffix;
   let DiagnosticType = "Invalid" # Name;
   let ParserMethod = "tryParseFPImm<false>";
   let PredicateMethod = "isExactFPImm<" # ValA # ", " # ValB # ">";
   let RenderMethod = "addExactFPImmOperands<" # ValA # ", " # ValB # ">";
 }
 

 // SVE pattern match helpers.
 //===----------------------------------------------------------------------===//
 
 class SVE_1_Op_Pat<ValueType vtd, SDPatternOperator op, ValueType vt1,
                    Instruction inst>
 : Pat<(vtd (op vt1:$Op1)),
       (inst $Op1)>;
 
+class SVE_1_Op_Imm_OptLsl_Pat<ValueType vt, SDPatternOperator op, ZPRRegOp zprty,
+                              ComplexPattern cpx, Instruction inst>
+  : Pat<(vt (op (vt zprty:$Op1), (i32 (cpx i32:$imm, i32:$shift)))),
+        (inst $Op1, i32:$imm, i32:$shift)>;
+
 class SVE_2_Op_Pat<ValueType vtd, SDPatternOperator op, ValueType vt1,
                    ValueType vt2, Instruction inst>
 : Pat<(vtd (op vt1:$Op1, vt2:$Op2)),
       (inst $Op1, $Op2)>;
 
 class SVE_2_Op_Pat_Reduce_To_Neon<ValueType vtd, SDPatternOperator op, ValueType vt1,
                    ValueType vt2, Instruction inst, SubRegIndex sub>
 : Pat<(vtd (op vt1:$Op1, vt2:$Op2)),
: I<(outs zprty:$Zdn), (ins zprty:$_Zdn, immtype:$imm),
   let Inst{12-5}  = imm{7-0}; // imm8
   let Inst{4-0}   = Zdn;
 
   let Constraints = "$Zdn = $_Zdn";
   let DestructiveInstType = Destructive;
   let ElementSize = ElementSizeNone;
 }
 
-multiclass sve_int_arith_imm0<bits<3> opc, string asm> {
-  def _B : sve_int_arith_imm0<0b00, opc, asm, ZPR8, addsub_imm8_opt_lsl_i8>;
+multiclass sve_int_arith_imm0<bits<3> opc, string asm, SDPatternOperator op> {
+  def _B : sve_int_arith_imm0<0b00, opc, asm, ZPR8,  addsub_imm8_opt_lsl_i8>;
   def _H : sve_int_arith_imm0<0b01, opc, asm, ZPR16, addsub_imm8_opt_lsl_i16>;
   def _S : sve_int_arith_imm0<0b10, opc, asm, ZPR32, addsub_imm8_opt_lsl_i32>;
   def _D : sve_int_arith_imm0<0b11, opc, asm, ZPR64, addsub_imm8_opt_lsl_i64>;
+
+  def : SVE_1_Op_Imm_OptLsl_Pat<nxv16i8, op, ZPR8,  SVEAddSubImm8Pat,  !cast<Instruction>(NAME # _B)>;
+  def : SVE_1_Op_Imm_OptLsl_Pat<nxv8i16, op, ZPR16, SVEAddSubImm16Pat, !cast<Instruction>(NAME # _H)>;
+  def : SVE_1_Op_Imm_OptLsl_Pat<nxv4i32, op, ZPR32, SVEAddSubImm32Pat, !cast<Instruction>(NAME # _S)>;
+  def : SVE_1_Op_Imm_OptLsl_Pat<nxv2i64, op, ZPR64, SVEAddSubImm64Pat, !cast<Instruction>(NAME # _D)>;
 }
 
 class sve_int_arith_imm<bits<2> sz8_64, bits<6> opc, string asm,
                         ZPRRegOp zprty, Operand immtype>
 : I<(outs zprty:$Zdn), (ins zprty:$_Zdn, immtype:$imm),
   asm, "\t$Zdn, $_Zdn, $imm",
   "",
   []>, Sched<[]> {

llvm/test/CodeGen/AArch64/sve-int-imm.ll

+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve < %s | FileCheck %s
+
+define <vscale x 16 x i8> @add_imm_i8_low(<vscale x 16 x i8> %a) {
+; CHECK-LABEL: add_imm_i8_low
+; CHECK: add  z0.b, z0.b, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 16 x i8> @llvm.aarch64.sve.add.imm.nxv16i8(<vscale x 16 x i8> %a,
+                                                                    i32 30)
+  ret <vscale x 16 x i8> %res
+}
+
+define <vscale x 8 x i16> @add_imm_i16_low(<vscale x 8 x i16> %a) {
+; CHECK-LABEL: add_imm_i16_low
+; CHECK: add  z0.h, z0.h, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 8 x i16> @llvm.aarch64.sve.add.imm.nxv8i16(<vscale x 8 x i16> %a,
+                                                                    i32 30)
+  ret <vscale x 8 x i16> %res
+}
+
+define <vscale x 8 x i16> @add_imm_i16_high(<vscale x 8 x i16> %a) {
+; CHECK-LABEL: add_imm_i16_high
+; CHECK: add  z0.h, z0.h, #1024
+; CHECK-NEXT: ret
+  %res =  call <vscale x 8 x i16> @llvm.aarch64.sve.add.imm.nxv8i16(<vscale x 8 x i16> %a,
+                                                                    i32 1024)
+  ret <vscale x 8 x i16> %res
+}
+
+define <vscale x 4 x i32> @add_imm_i32_low(<vscale x 4 x i32> %a) {
+; CHECK-LABEL: add_imm_i32_low
+; CHECK: add  z0.s, z0.s, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 4 x i32> @llvm.aarch64.sve.add.imm.nxv4i32(<vscale x 4 x i32> %a,
+                                                                    i32 30)
+  ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 4 x i32> @add_imm_i32_high(<vscale x 4 x i32> %a) {
+; CHECK-LABEL: add_imm_i32_high
+; CHECK: add  z0.s, z0.s, #1024
+; CHECK-NEXT: ret
+  %res =  call <vscale x 4 x i32> @llvm.aarch64.sve.add.imm.nxv4i32(<vscale x 4 x i32> %a,
+                                                                    i32 1024)
+  ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 2 x i64> @add_imm_i64_low(<vscale x 2 x i64> %a) {
+; CHECK-LABEL: add_imm_i64_low
+; CHECK: add  z0.d, z0.d, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 2 x i64> @llvm.aarch64.sve.add.imm.nxv2i64(<vscale x 2 x i64> %a,
+                                                                    i32 30)
+  ret <vscale x 2 x i64> %res
+}
+
+define <vscale x 2 x i64> @add_imm_i64_high(<vscale x 2 x i64> %a) {
+; CHECK-LABEL: add_imm_i64_high
+; CHECK: add  z0.d, z0.d, #1024
+; CHECK-NEXT: ret
+  %res =  call <vscale x 2 x i64> @llvm.aarch64.sve.add.imm.nxv2i64(<vscale x 2 x i64> %a,
+                                                                    i32 1024)
+  ret <vscale x 2 x i64> %res
+}
+
+define <vscale x 16 x i8> @sub_imm_i8_low(<vscale x 16 x i8> %a) {
+; CHECK-LABEL: sub_imm_i8_low
+; CHECK: sub  z0.b, z0.b, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 16 x i8> @llvm.aarch64.sve.sub.imm.nxv16i8(<vscale x 16 x i8> %a,
+                                                                    i32 30)
+  ret <vscale x 16 x i8> %res
+}
+
+define <vscale x 8 x i16> @sub_imm_i16_low(<vscale x 8 x i16> %a) {
+; CHECK-LABEL: sub_imm_i16_low
+; CHECK: sub  z0.h, z0.h, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 8 x i16> @llvm.aarch64.sve.sub.imm.nxv8i16(<vscale x 8 x i16> %a,
+                                                                    i32 30)
+  ret <vscale x 8 x i16> %res
+}
+
+define <vscale x 8 x i16> @sub_imm_i16_high(<vscale x 8 x i16> %a) {
+; CHECK-LABEL: sub_imm_i16_high
+; CHECK: sub  z0.h, z0.h, #1024
+; CHECK-NEXT: ret
+  %res =  call <vscale x 8 x i16> @llvm.aarch64.sve.sub.imm.nxv8i16(<vscale x 8 x i16> %a,
+                                                                    i32 1024)
+  ret <vscale x 8 x i16> %res
+}
+
+define <vscale x 4 x i32> @sub_imm_i32_low(<vscale x 4 x i32> %a) {
+; CHECK-LABEL: sub_imm_i32_low
+; CHECK: sub  z0.s, z0.s, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 4 x i32> @llvm.aarch64.sve.sub.imm.nxv4i32(<vscale x 4 x i32> %a,
+                                                                    i32 30)
+  ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 4 x i32> @sub_imm_i32_high(<vscale x 4 x i32> %a) {
+; CHECK-LABEL: sub_imm_i32_high
+; CHECK: sub  z0.s, z0.s, #1024
+; CHECK-NEXT: ret
+  %res =  call <vscale x 4 x i32> @llvm.aarch64.sve.sub.imm.nxv4i32(<vscale x 4 x i32> %a,
+                                                                    i32 1024)
+  ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 2 x i64> @sub_imm_i64_low(<vscale x 2 x i64> %a) {
+; CHECK-LABEL: sub_imm_i64_low
+; CHECK: sub  z0.d, z0.d, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 2 x i64> @llvm.aarch64.sve.sub.imm.nxv2i64(<vscale x 2 x i64> %a,
+                                                                    i32 30)
+  ret <vscale x 2 x i64> %res
+}
+
+define <vscale x 2 x i64> @sub_imm_i64_high(<vscale x 2 x i64> %a) {
+; CHECK-LABEL: sub_imm_i64_high
+; CHECK: sub  z0.d, z0.d, #1024
+; CHECK-NEXT: ret
+  %res =  call <vscale x 2 x i64> @llvm.aarch64.sve.sub.imm.nxv2i64(<vscale x 2 x i64> %a,
+                                                                    i32 1024)
+  ret <vscale x 2 x i64> %res
+}
+
+define <vscale x 16 x i8> @subr_imm_i8_low(<vscale x 16 x i8> %a) {
+; CHECK-LABEL: subr_imm_i8_low
+; CHECK: subr  z0.b, z0.b, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 16 x i8> @llvm.aarch64.sve.subr.imm.nxv16i8(<vscale x 16 x i8> %a,
+                                                                     i32 30)
+  ret <vscale x 16 x i8> %res
+}
+
+define <vscale x 8 x i16> @subr_imm_i16_low(<vscale x 8 x i16> %a) {
+; CHECK-LABEL: subr_imm_i16_low
+; CHECK: subr  z0.h, z0.h, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 8 x i16> @llvm.aarch64.sve.subr.imm.nxv8i16(<vscale x 8 x i16> %a,
+                                                                     i32 30)
+  ret <vscale x 8 x i16> %res
+}
+
+define <vscale x 8 x i16> @subr_imm_i16_high(<vscale x 8 x i16> %a) {
+; CHECK-LABEL: subr_imm_i16_high
+; CHECK: subr  z0.h, z0.h, #1024
+; CHECK-NEXT: ret
+  %res =  call <vscale x 8 x i16> @llvm.aarch64.sve.subr.imm.nxv8i16(<vscale x 8 x i16> %a,
+                                                                     i32 1024)
+  ret <vscale x 8 x i16> %res
+}
+
+define <vscale x 4 x i32> @subr_imm_i32_low(<vscale x 4 x i32> %a) {
+; CHECK-LABEL: subr_imm_i32_low
+; CHECK: subr  z0.s, z0.s, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 4 x i32> @llvm.aarch64.sve.subr.imm.nxv4i32(<vscale x 4 x i32> %a,
+                                                                     i32 30)
+  ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 4 x i32> @subr_imm_i32_high(<vscale x 4 x i32> %a) {
+; CHECK-LABEL: subr_imm_i32_high
+; CHECK: subr  z0.s, z0.s, #1024
+; CHECK-NEXT: ret
+  %res =  call <vscale x 4 x i32> @llvm.aarch64.sve.subr.imm.nxv4i32(<vscale x 4 x i32> %a,
+                                                                     i32 1024)
+  ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 2 x i64> @subr_imm_i64_low(<vscale x 2 x i64> %a) {
+; CHECK-LABEL: subr_imm_i64_low
+; CHECK: subr  z0.d, z0.d, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 2 x i64> @llvm.aarch64.sve.subr.imm.nxv2i64(<vscale x 2 x i64> %a,
+                                                                     i32 30)
+  ret <vscale x 2 x i64> %res
+}
+
+define <vscale x 2 x i64> @subr_imm_i64_high(<vscale x 2 x i64> %a) {
+; CHECK-LABEL: subr_imm_i64_high
+; CHECK: subr  z0.d, z0.d, #1024
+; CHECK-NEXT: ret
+  %res =  call <vscale x 2 x i64> @llvm.aarch64.sve.subr.imm.nxv2i64(<vscale x 2 x i64> %a,
+                                                                     i32 1024)
+  ret <vscale x 2 x i64> %res
+}
+
+define <vscale x 16 x i8> @sqadd_imm_i8_low(<vscale x 16 x i8> %a) {
+; CHECK-LABEL: sqadd_imm_i8_low
+; CHECK: sqadd  z0.b, z0.b, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 16 x i8> @llvm.aarch64.sve.sqadd.imm.nxv16i8(<vscale x 16 x i8> %a,
+                                                                      i32 30)
+  ret <vscale x 16 x i8> %res
+}
+
+define <vscale x 8 x i16> @sqadd_imm_i16_low(<vscale x 8 x i16> %a) {
+; CHECK-LABEL: sqadd_imm_i16_low
+; CHECK: sqadd  z0.h, z0.h, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 8 x i16> @llvm.aarch64.sve.sqadd.imm.nxv8i16(<vscale x 8 x i16> %a,
+                                                                      i32 30)
+  ret <vscale x 8 x i16> %res
+}
+
+define <vscale x 8 x i16> @sqadd_imm_i16_high(<vscale x 8 x i16> %a) {
+; CHECK-LABEL: sqadd_imm_i16_high
+; CHECK: sqadd  z0.h, z0.h, #1024
+; CHECK-NEXT: ret
+  %res =  call <vscale x 8 x i16> @llvm.aarch64.sve.sqadd.imm.nxv8i16(<vscale x 8 x i16> %a,
+                                                                      i32 1024)
+  ret <vscale x 8 x i16> %res
+}
+
+define <vscale x 4 x i32> @sqadd_imm_i32_low(<vscale x 4 x i32> %a) {
+; CHECK-LABEL: sqadd_imm_i32_low
+; CHECK: sqadd  z0.s, z0.s, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 4 x i32> @llvm.aarch64.sve.sqadd.imm.nxv4i32(<vscale x 4 x i32> %a,
+                                                                      i32 30)
+  ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 4 x i32> @sqadd_imm_i32_high(<vscale x 4 x i32> %a) {
+; CHECK-LABEL: sqadd_imm_i32_high
+; CHECK: sqadd  z0.s, z0.s, #1024
+; CHECK-NEXT: ret
+  %res =  call <vscale x 4 x i32> @llvm.aarch64.sve.sqadd.imm.nxv4i32(<vscale x 4 x i32> %a,
+                                                                      i32 1024)
+  ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 2 x i64> @sqadd_imm_i64_low(<vscale x 2 x i64> %a) {
+; CHECK-LABEL: sqadd_imm_i64_low
+; CHECK: sqadd  z0.d, z0.d, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 2 x i64> @llvm.aarch64.sve.sqadd.imm.nxv2i64(<vscale x 2 x i64> %a,
+                                                                      i32 30)
+  ret <vscale x 2 x i64> %res
+}
+
+define <vscale x 2 x i64> @sqadd_imm_i64_high(<vscale x 2 x i64> %a) {
+; CHECK-LABEL: sqadd_imm_i64_high
+; CHECK: sqadd  z0.d, z0.d, #1024
+; CHECK-NEXT: ret
+  %res =  call <vscale x 2 x i64> @llvm.aarch64.sve.sqadd.imm.nxv2i64(<vscale x 2 x i64> %a,
+                                                                      i32 1024)
+  ret <vscale x 2 x i64> %res
+}
+
+define <vscale x 16 x i8> @uqadd_imm_i8_low(<vscale x 16 x i8> %a) {
+; CHECK-LABEL: uqadd_imm_i8_low
+; CHECK: uqadd  z0.b, z0.b, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 16 x i8> @llvm.aarch64.sve.uqadd.imm.nxv16i8(<vscale x 16 x i8> %a,
+                                                                      i32 30)
+  ret <vscale x 16 x i8> %res
+}
+
+define <vscale x 8 x i16> @uqadd_imm_i16_low(<vscale x 8 x i16> %a) {
+; CHECK-LABEL: uqadd_imm_i16_low
+; CHECK: uqadd  z0.h, z0.h, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 8 x i16> @llvm.aarch64.sve.uqadd.imm.nxv8i16(<vscale x 8 x i16> %a,
+                                                                      i32 30)
+  ret <vscale x 8 x i16> %res
+}
+
+define <vscale x 8 x i16> @uqadd_imm_i16_high(<vscale x 8 x i16> %a) {
+; CHECK-LABEL: uqadd_imm_i16_high
+; CHECK: uqadd  z0.h, z0.h, #1024
+; CHECK-NEXT: ret
+  %res =  call <vscale x 8 x i16> @llvm.aarch64.sve.uqadd.imm.nxv8i16(<vscale x 8 x i16> %a,
+                                                                      i32 1024)
+  ret <vscale x 8 x i16> %res
+}
+
+define <vscale x 4 x i32> @uqadd_imm_i32_low(<vscale x 4 x i32> %a) {
+; CHECK-LABEL: uqadd_imm_i32_low
+; CHECK: uqadd  z0.s, z0.s, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 4 x i32> @llvm.aarch64.sve.uqadd.imm.nxv4i32(<vscale x 4 x i32> %a,
+                                                                      i32 30)
+  ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 4 x i32> @uqadd_imm_i32_high(<vscale x 4 x i32> %a) {
+; CHECK-LABEL: uqadd_imm_i32_high
+; CHECK: uqadd  z0.s, z0.s, #1024
+; CHECK-NEXT: ret
+  %res =  call <vscale x 4 x i32> @llvm.aarch64.sve.uqadd.imm.nxv4i32(<vscale x 4 x i32> %a,
+                                                                      i32 1024)
+  ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 2 x i64> @uqadd_imm_i64_low(<vscale x 2 x i64> %a) {
+; CHECK-LABEL: uqadd_imm_i64_low
+; CHECK: uqadd  z0.d, z0.d, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 2 x i64> @llvm.aarch64.sve.uqadd.imm.nxv2i64(<vscale x 2 x i64> %a,
+                                                                      i32 30)
+  ret <vscale x 2 x i64> %res
+}
+
+define <vscale x 2 x i64> @uqadd_imm_i64_high(<vscale x 2 x i64> %a) {
+; CHECK-LABEL: uqadd_imm_i64_high
+; CHECK: uqadd  z0.d, z0.d, #1024
+; CHECK-NEXT: ret
+  %res =  call <vscale x 2 x i64> @llvm.aarch64.sve.uqadd.imm.nxv2i64(<vscale x 2 x i64> %a,
+                                                                      i32 1024)
+  ret <vscale x 2 x i64> %res
+}
+
+define <vscale x 16 x i8> @sqsub_imm_i8_low(<vscale x 16 x i8> %a) {
+; CHECK-LABEL: sqsub_imm_i8_low
+; CHECK: sqsub  z0.b, z0.b, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 16 x i8> @llvm.aarch64.sve.sqsub.imm.nxv16i8(<vscale x 16 x i8> %a,
+                                                                      i32 30)
+  ret <vscale x 16 x i8> %res
+}
+
+define <vscale x 8 x i16> @sqsub_imm_i16_low(<vscale x 8 x i16> %a) {
+; CHECK-LABEL: sqsub_imm_i16_low
+; CHECK: sqsub  z0.h, z0.h, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 8 x i16> @llvm.aarch64.sve.sqsub.imm.nxv8i16(<vscale x 8 x i16> %a,
+                                                                      i32 30)
+  ret <vscale x 8 x i16> %res
+}
+
+define <vscale x 8 x i16> @sqsub_imm_i16_high(<vscale x 8 x i16> %a) {
+; CHECK-LABEL: sqsub_imm_i16_high
+; CHECK: sqsub  z0.h, z0.h, #1024
+; CHECK-NEXT: ret
+  %res =  call <vscale x 8 x i16> @llvm.aarch64.sve.sqsub.imm.nxv8i16(<vscale x 8 x i16> %a,
+                                                                      i32 1024)
+  ret <vscale x 8 x i16> %res
+}
+
+define <vscale x 4 x i32> @sqsub_imm_i32_low(<vscale x 4 x i32> %a) {
+; CHECK-LABEL: sqsub_imm_i32_low
+; CHECK: sqsub  z0.s, z0.s, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 4 x i32> @llvm.aarch64.sve.sqsub.imm.nxv4i32(<vscale x 4 x i32> %a,
+                                                                      i32 30)
+  ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 4 x i32> @sqsub_imm_i32_high(<vscale x 4 x i32> %a) {
+; CHECK-LABEL: sqsub_imm_i32_high
+; CHECK: sqsub  z0.s, z0.s, #1024
+; CHECK-NEXT: ret
+  %res =  call <vscale x 4 x i32> @llvm.aarch64.sve.sqsub.imm.nxv4i32(<vscale x 4 x i32> %a,
+                                                                      i32 1024)
+  ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 2 x i64> @sqsub_imm_i64_low(<vscale x 2 x i64> %a) {
+; CHECK-LABEL: sqsub_imm_i64_low
+; CHECK: sqsub  z0.d, z0.d, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 2 x i64> @llvm.aarch64.sve.sqsub.imm.nxv2i64(<vscale x 2 x i64> %a,
+                                                                      i32 30)
+  ret <vscale x 2 x i64> %res
+}
+
+define <vscale x 2 x i64> @sqsub_imm_i64_high(<vscale x 2 x i64> %a) {
+; CHECK-LABEL: sqsub_imm_i64_high
+; CHECK: sqsub  z0.d, z0.d, #1024
+; CHECK-NEXT: ret
+  %res =  call <vscale x 2 x i64> @llvm.aarch64.sve.sqsub.imm.nxv2i64(<vscale x 2 x i64> %a,
+                                                                      i32 1024)
+  ret <vscale x 2 x i64> %res
+}
+
+define <vscale x 16 x i8> @uqsub_imm_i8_low(<vscale x 16 x i8> %a) {
+; CHECK-LABEL: uqsub_imm_i8_low
+; CHECK: uqsub  z0.b, z0.b, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 16 x i8> @llvm.aarch64.sve.uqsub.imm.nxv16i8(<vscale x 16 x i8> %a,
+                                                                      i32 30)
+  ret <vscale x 16 x i8> %res
+}
+
+define <vscale x 8 x i16> @uqsub_imm_i16_low(<vscale x 8 x i16> %a) {
+; CHECK-LABEL: uqsub_imm_i16_low
+; CHECK: uqsub  z0.h, z0.h, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 8 x i16> @llvm.aarch64.sve.uqsub.imm.nxv8i16(<vscale x 8 x i16> %a,
+                                                                      i32 30)
+  ret <vscale x 8 x i16> %res
+}
+
+define <vscale x 8 x i16> @uqsub_imm_i16_high(<vscale x 8 x i16> %a) {
+; CHECK-LABEL: uqsub_imm_i16_high
+; CHECK: uqsub  z0.h, z0.h, #1024
+; CHECK-NEXT: ret
+  %res =  call <vscale x 8 x i16> @llvm.aarch64.sve.uqsub.imm.nxv8i16(<vscale x 8 x i16> %a,
+                                                                      i32 1024)
+  ret <vscale x 8 x i16> %res
+}
+
+define <vscale x 4 x i32> @uqsub_imm_i32_low(<vscale x 4 x i32> %a) {
+; CHECK-LABEL: uqsub_imm_i32_low
+; CHECK: uqsub  z0.s, z0.s, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 4 x i32> @llvm.aarch64.sve.uqsub.imm.nxv4i32(<vscale x 4 x i32> %a,
+                                                                      i32 30)
+  ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 4 x i32> @uqsub_imm_i32_high(<vscale x 4 x i32> %a) {
+; CHECK-LABEL: uqsub_imm_i32_high
+; CHECK: uqsub  z0.s, z0.s, #1024
+; CHECK-NEXT: ret
+  %res =  call <vscale x 4 x i32> @llvm.aarch64.sve.uqsub.imm.nxv4i32(<vscale x 4 x i32> %a,
+                                                                      i32 1024)
+  ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 2 x i64> @uqsub_imm_i64_low(<vscale x 2 x i64> %a) {
+; CHECK-LABEL: uqsub_imm_i64_low
+; CHECK: uqsub  z0.d, z0.d, #30
+; CHECK-NEXT: ret
+  %res =  call <vscale x 2 x i64> @llvm.aarch64.sve.uqsub.imm.nxv2i64(<vscale x 2 x i64> %a,
+                                                                      i32 30)
+  ret <vscale x 2 x i64> %res
+}
+
+define <vscale x 2 x i64> @uqsub_imm_i64_high(<vscale x 2 x i64> %a) {
+; CHECK-LABEL: uqsub_imm_i64_high
+; CHECK: uqsub  z0.d, z0.d, #1024
+; CHECK-NEXT: ret
+  %res =  call <vscale x 2 x i64> @llvm.aarch64.sve.uqsub.imm.nxv2i64(<vscale x 2 x i64> %a,
+                                                                      i32 1024)
+  ret <vscale x 2 x i64> %res
+}
+
+declare <vscale x 16 x i8> @llvm.aarch64.sve.add.imm.nxv16i8(<vscale x 16 x i8>, i32)
+declare <vscale x 8 x i16> @llvm.aarch64.sve.add.imm.nxv8i16(<vscale x 8 x i16>, i32)
+declare <vscale x 4 x i32> @llvm.aarch64.sve.add.imm.nxv4i32(<vscale x 4 x i32>, i32)
+declare <vscale x 2 x i64> @llvm.aarch64.sve.add.imm.nxv2i64(<vscale x 2 x i64>, i32)
+declare <vscale x 16 x i8> @llvm.aarch64.sve.sub.imm.nxv16i8(<vscale x 16 x i8>, i32)
+declare <vscale x 8 x i16> @llvm.aarch64.sve.sub.imm.nxv8i16(<vscale x 8 x i16>, i32)
+declare <vscale x 4 x i32> @llvm.aarch64.sve.sub.imm.nxv4i32(<vscale x 4 x i32>, i32)
+declare <vscale x 2 x i64> @llvm.aarch64.sve.sub.imm.nxv2i64(<vscale x 2 x i64>, i32)
+declare <vscale x 16 x i8> @llvm.aarch64.sve.subr.imm.nxv16i8(<vscale x 16 x i8>, i32)
+declare <vscale x 8 x i16> @llvm.aarch64.sve.subr.imm.nxv8i16(<vscale x 8 x i16>, i32)
+declare <vscale x 4 x i32> @llvm.aarch64.sve.subr.imm.nxv4i32(<vscale x 4 x i32>, i32)
+declare <vscale x 2 x i64> @llvm.aarch64.sve.subr.imm.nxv2i64(<vscale x 2 x i64>, i32)
+declare <vscale x 16 x i8> @llvm.aarch64.sve.sqadd.imm.nxv16i8(<vscale x 16 x i8>, i32)
+declare <vscale x 8 x i16> @llvm.aarch64.sve.sqadd.imm.nxv8i16(<vscale x 8 x i16>, i32)
+declare <vscale x 4 x i32> @llvm.aarch64.sve.sqadd.imm.nxv4i32(<vscale x 4 x i32>, i32)
+declare <vscale x 2 x i64> @llvm.aarch64.sve.sqadd.imm.nxv2i64(<vscale x 2 x i64>, i32)
+declare <vscale x 16 x i8> @llvm.aarch64.sve.uqadd.imm.nxv16i8(<vscale x 16 x i8>, i32)
+declare <vscale x 8 x i16> @llvm.aarch64.sve.uqadd.imm.nxv8i16(<vscale x 8 x i16>, i32)
+declare <vscale x 4 x i32> @llvm.aarch64.sve.uqadd.imm.nxv4i32(<vscale x 4 x i32>, i32)
+declare <vscale x 2 x i64> @llvm.aarch64.sve.uqadd.imm.nxv2i64(<vscale x 2 x i64>, i32)
+declare <vscale x 16 x i8> @llvm.aarch64.sve.sqsub.imm.nxv16i8(<vscale x 16 x i8>, i32)
+declare <vscale x 8 x i16> @llvm.aarch64.sve.sqsub.imm.nxv8i16(<vscale x 8 x i16>, i32)
+declare <vscale x 4 x i32> @llvm.aarch64.sve.sqsub.imm.nxv4i32(<vscale x 4 x i32>, i32)
+declare <vscale x 2 x i64> @llvm.aarch64.sve.sqsub.imm.nxv2i64(<vscale x 2 x i64>, i32)
+declare <vscale x 16 x i8> @llvm.aarch64.sve.uqsub.imm.nxv16i8(<vscale x 16 x i8>, i32)
+declare <vscale x 8 x i16> @llvm.aarch64.sve.uqsub.imm.nxv8i16(<vscale x 8 x i16>, i32)
+declare <vscale x 4 x i32> @llvm.aarch64.sve.uqsub.imm.nxv4i32(<vscale x 4 x i32>, i32)
+declare <vscale x 2 x i64> @llvm.aarch64.sve.uqsub.imm.nxv2i64(<vscale x 2 x i64>, i32)