[SVE] Lower fixed length vector integer ISD::SETCC operations.

[paulwalker-arm]

The floating-point side requires some refactoring so I'll do that separately.

llvm/test/CodeGen/AArch64/sve-fixed-length-int-compares.ll

446

I believe I've covered all the combinations for the EQ tests and there should already be tests for the ISEL patterns required for SETCC_MERGE_ZERO. For these reasons there doesn't look to be a good reason to repeat the vector combinations for each comparison type, so I'm instead just testing a single but different vector length for the following compare tests.

[paulwalker-arm]

Add extra assert to LowerFixedLengthVectorSetccToSVE.

[cameron.mcinally]

LGTM

llvm/test/CodeGen/AArch64/sve-fixed-length-int-compares.ll

446

That sounds fair. @efriedma, are you okay with this? I suppose it can be fixed post-commit if it's a problem...

[efriedma]

llvm/test/CodeGen/AArch64/sve-fixed-length-int-compares.ll

446

Seems fine.

llvm/lib/Target/AArch64/AArch64ISelLowering.h

private:
   SDValue LowerDYNAMIC_STACKALLOC(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerWindowsDYNAMIC_STACKALLOC(SDValue Op, SDValue Chain,
                                          SDValue &Size,
                                          SelectionDAG &DAG) const;
   SDValue LowerSVEStructLoad(unsigned Intrinsic, ArrayRef<SDValue> LoadOps,
                              EVT VT, SelectionDAG &DAG, const SDLoc &DL) const;
 
   SDValue LowerFixedLengthVectorLoadToSVE(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerFixedLengthVectorSetccToSVE(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerFixedLengthVectorStoreToSVE(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerFixedLengthVectorTruncateToSVE(SDValue Op,
                                               SelectionDAG &DAG) const;
 
   SDValue BuildSDIVPow2(SDNode *N, const APInt &Divisor, SelectionDAG &DAG,
                         SmallVectorImpl<SDNode *> &Created) const override;
   SDValue getSqrtEstimate(SDValue Operand, SelectionDAG &DAG, int Enabled,
                           int &ExtraSteps, bool &UseOneConst,

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

void AArch64TargetLowering::addTypeForFixedLengthSVE(MVT VT) {
   setOperationAction(ISD::FADD, VT, Custom);
   setOperationAction(ISD::FDIV, VT, Custom);
   setOperationAction(ISD::FMA, VT, Custom);
   setOperationAction(ISD::FMUL, VT, Custom);
   setOperationAction(ISD::FSUB, VT, Custom);
   setOperationAction(ISD::LOAD, VT, Custom);
   setOperationAction(ISD::MUL, VT, Custom);
   setOperationAction(ISD::OR, VT, Custom);
+  setOperationAction(ISD::SETCC, VT, Custom);
   setOperationAction(ISD::STORE, VT, Custom);
   setOperationAction(ISD::SUB, VT, Custom);
   setOperationAction(ISD::TRUNCATE, VT, Custom);
   setOperationAction(ISD::XOR, VT, Custom);
 }
 
 void AArch64TargetLowering::addDRTypeForNEON(MVT VT) {
   addRegisterClass(VT, &AArch64::FPR64RegClass);

 SDValue AArch64TargetLowering::LowerVSETCC(SDValue Op,
                                            SelectionDAG &DAG) const {
   if (Op.getValueType().isScalableVector()) {
     if (Op.getOperand(0).getValueType().isFloatingPoint())
       return Op;
     return LowerToPredicatedOp(Op, DAG, AArch64ISD::SETCC_MERGE_ZERO);
   }
 
+  if (useSVEForFixedLengthVectorVT(Op.getOperand(0).getValueType()))
+    return LowerFixedLengthVectorSetccToSVE(Op, DAG);
+
   ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(2))->get();
   SDValue LHS = Op.getOperand(0);
   SDValue RHS = Op.getOperand(1);
   EVT CmpVT = LHS.getValueType().changeVectorElementTypeToInteger();
   SDLoc dl(Op);
 
   if (LHS.getValueType().getVectorElementType().isInteger()) {
     assert(LHS.getValueType() == RHS.getValueType());
for (const SDValue &V : Op->op_values()) {
     assert(useSVEForFixedLengthVectorVT(V.getValueType()) &&
            "Only fixed length vectors are supported!");
     Ops.push_back(convertToScalableVector(DAG, ContainerVT, V));
   }
 
   auto ScalableRes = DAG.getNode(Op.getOpcode(), SDLoc(Op), ContainerVT, Ops);
   return convertFromScalableVector(DAG, VT, ScalableRes);
 }
+
+SDValue AArch64TargetLowering::LowerFixedLengthVectorSetccToSVE(
+    SDValue Op, SelectionDAG &DAG) const {
+  SDLoc DL(Op);
+  EVT InVT = Op.getOperand(0).getValueType();
+  EVT ContainerVT = getContainerForFixedLengthVector(DAG, InVT);
+
+  assert(Op.getValueType() == InVT.changeTypeToInteger() &&
+         "Expected integer result of the same bit length as the inputs!");
+
+  // Expand floating point vector comparisons.
+  if (InVT.isFloatingPoint())
+    return SDValue();
+
+  auto Op1 = convertToScalableVector(DAG, ContainerVT, Op.getOperand(0));
+  auto Op2 = convertToScalableVector(DAG, ContainerVT, Op.getOperand(1));
+  auto Pg = getPredicateForFixedLengthVector(DAG, DL, InVT);
+
+  EVT CmpVT = Pg.getValueType();
+  SmallVector<SDValue, 4> CmpOps = {Pg, Op1, Op2, Op.getOperand(2)};
+  auto Cmp = DAG.getNode(AArch64ISD::SETCC_MERGE_ZERO, DL, CmpVT, CmpOps);
+
+  EVT PromoteVT = ContainerVT.changeTypeToInteger();
+  auto Promote = DAG.getBoolExtOrTrunc(Cmp, DL, PromoteVT, InVT);
+  return convertFromScalableVector(DAG, Op.getValueType(), Promote);
+}

llvm/test/CodeGen/AArch64/sve-fixed-length-int-compares.ll

+; RUN: llc -aarch64-sve-vector-bits-min=128  -asm-verbose=0 < %s | FileCheck %s -check-prefix=NO_SVE
+; RUN: llc -aarch64-sve-vector-bits-min=256  -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_EQ_256
+; RUN: llc -aarch64-sve-vector-bits-min=384  -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK
+; RUN: llc -aarch64-sve-vector-bits-min=512  -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=640  -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=768  -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=896  -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=1024 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
+; RUN: llc -aarch64-sve-vector-bits-min=1152 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
+; RUN: llc -aarch64-sve-vector-bits-min=1280 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
+; RUN: llc -aarch64-sve-vector-bits-min=1408 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
+; RUN: llc -aarch64-sve-vector-bits-min=1536 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
+; RUN: llc -aarch64-sve-vector-bits-min=1664 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
+; RUN: llc -aarch64-sve-vector-bits-min=1792 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
+; RUN: llc -aarch64-sve-vector-bits-min=1920 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
+; RUN: llc -aarch64-sve-vector-bits-min=2048 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024,VBITS_GE_2048
+
+target triple = "aarch64-unknown-linux-gnu"
+
+; Don't use SVE when its registers are no bigger than NEON.
+; NO_SVE-NOT: z{0-9}
+
+;
+; ICMP EQ
+;
+
+; Don't use SVE for 64-bit vectors.
+define <8 x i8> @icmp_eq_v8i8(<8 x i8> %op1, <8 x i8> %op2) #0 {
+; CHECK-LABEL: icmp_eq_v8i8:
+; CHECK: cmeq v0.8b, v0.8b, v1.8b
+; CHECK-NEXT: ret
+  %cmp = icmp eq <8 x i8> %op1, %op2
+  %sext = sext <8 x i1> %cmp to <8 x i8>
+  ret <8 x i8> %sext
+}
+
+; Don't use SVE for 128-bit vectors.
+define <16 x i8> @icmp_eq_v16i8(<16 x i8> %op1, <16 x i8> %op2) #0 {
+; CHECK-LABEL: icmp_eq_v16i8:
+; CHECK: cmeq v0.16b, v0.16b, v1.16b
+; CHECK-NEXT: ret
+  %cmp = icmp eq <16 x i8> %op1, %op2
+  %sext = sext <16 x i1> %cmp to <16 x i8>
+  ret <16 x i8> %sext
+}
+
+define void @icmp_eq_v32i8(<32 x i8>* %a, <32 x i8>* %b) #0 {
+; CHECK-LABEL: icmp_eq_v32i8:
+; CHECK: ptrue [[PG:p[0-9]+]].b, vl32
+; CHECK-DAG: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
+; CHECK-DAG: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
+; CHECK-NEXT: cmpeq [[CMP:p[0-9]+]].b, [[PG]]/z, [[OP1]].b, [[OP2]].b
+; CHECK-NEXT: mov [[SEXT:z[0-9]+]].b, [[CMP]]/z, #-1
+; CHECK-NEXT: st1b { [[SEXT]].b }, [[PG]], [x0]
+; CHECK-NEXT: ret
+  %op1 = load <32 x i8>, <32 x i8>* %a
+  %op2 = load <32 x i8>, <32 x i8>* %b
+  %cmp = icmp eq <32 x i8> %op1, %op2
+  %sext = sext <32 x i1> %cmp to <32 x i8>
+  store <32 x i8> %sext, <32 x i8>* %a
+  ret void
+}
+
+define void @icmp_eq_v64i8(<64 x i8>* %a, <64 x i8>* %b) #0 {
+; CHECK-LABEL: icmp_eq_v64i8:
+; VBITS_GE_512: ptrue [[PG:p[0-9]+]].b, vl64
+; VBITS_GE_512-DAG: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
+; VBITS_GE_512-DAG: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
+; VBITS_GE_512-NEXT: cmpeq [[CMP:p[0-9]+]].b, [[PG]]/z, [[OP1]].b, [[OP2]].b
+; VBITS_GE_512-NEXT: mov [[SEXT:z[0-9]+]].b, [[CMP]]/z, #-1
+; VBITS_GE_512-NEXT: st1b { [[SEXT]].b }, [[PG]], [x0]
+; VBITS_GE_512-NEXT: ret
+
+; Ensure sensible type legalisation
+; VBITS_EQ_256-DAG: ptrue [[PG:p[0-9]+]].b, vl32
+; VBITS_EQ_256-DAG: mov w[[OFF_HI:[0-9]+]], #32
+; VBITS_EQ_256-DAG: ld1b { [[OP1_LO:z[0-9]+]].b }, [[PG]]/z, [x0]
+; VBITS_EQ_256-DAG: ld1b { [[OP1_HI:z[0-9]+]].b }, [[PG]]/z, [x0, x[[OFF_HI]]]
+; VBITS_EQ_256-DAG: ld1b { [[OP2_LO:z[0-9]+]].b }, [[PG]]/z, [x1]
+; VBITS_EQ_256-DAG: ld1b { [[OP2_HI:z[0-9]+]].b }, [[PG]]/z, [x1, x[[OFF_HI]]]
+; VBITS_EQ_256-DAG: cmpeq [[CMP_LO:p[0-9]+]].b, [[PG]]/z, [[OP1_LO]].b, [[OP2_LO]].b
+; VBITS_EQ_256-DAG: cmpeq [[CMP_HI:p[0-9]+]].b, [[PG]]/z, [[OP1_HI]].b, [[OP2_HI]].b
+; VBITS_EQ_256-DAG: mov [[SEXT_LO:z[0-9]+]].b, [[CMP_LO]]/z, #-1
+; VBITS_EQ_256-DAG: mov [[SEXT_HI:z[0-9]+]].b, [[CMP_HI]]/z, #-1
+; VBITS_EQ_256-DAG: st1b { [[SEXT_LO]].b }, [[PG]], [x0]
+; VBITS_EQ_256-DAG: st1b { [[SEXT_HI]].b }, [[PG]], [x0, x[[OFF_HI]]]
+; VBITS_EQ_256-NEXT: ret
+  %op1 = load <64 x i8>, <64 x i8>* %a
+  %op2 = load <64 x i8>, <64 x i8>* %b
+  %cmp = icmp eq <64 x i8> %op1, %op2
+  %sext = sext <64 x i1> %cmp to <64 x i8>
+  store <64 x i8> %sext, <64 x i8>* %a
+  ret void
+}
+
+define void @icmp_eq_v128i8(<128 x i8>* %a, <128 x i8>* %b) #0 {
+; CHECK-LABEL: icmp_eq_v128i8:
+; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].b, vl128
+; VBITS_GE_1024-DAG: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
+; VBITS_GE_1024-DAG: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
+; VBITS_GE_1024-NEXT: cmpeq [[CMP:p[0-9]+]].b, [[PG]]/z, [[OP1]].b, [[OP2]].b
+; VBITS_GE_1024-NEXT: mov [[SEXT:z[0-9]+]].b, [[CMP]]/z, #-1
+; VBITS_GE_1024-NEXT: st1b { [[SEXT]].b }, [[PG]], [x0]
+; VBITS_GE_1024-NEXT: ret
+  %op1 = load <128 x i8>, <128 x i8>* %a
+  %op2 = load <128 x i8>, <128 x i8>* %b
+  %cmp = icmp eq <128 x i8> %op1, %op2
+  %sext = sext <128 x i1> %cmp to <128 x i8>
+  store <128 x i8> %sext, <128 x i8>* %a
+  ret void
+}
+
+define void @icmp_eq_v256i8(<256 x i8>* %a, <256 x i8>* %b) #0 {
+; CHECK-LABEL: icmp_eq_v256i8:
+; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].b, vl256
+; VBITS_GE_2048-DAG: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
+; VBITS_GE_2048-DAG: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
+; VBITS_GE_2048-NEXT: cmpeq [[CMP:p[0-9]+]].b, [[PG]]/z, [[OP1]].b, [[OP2]].b
+; VBITS_GE_2048-NEXT: mov [[SEXT:z[0-9]+]].b, [[CMP]]/z, #-1
+; VBITS_GE_2048-NEXT: st1b { [[SEXT]].b }, [[PG]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %op1 = load <256 x i8>, <256 x i8>* %a
+  %op2 = load <256 x i8>, <256 x i8>* %b
+  %cmp = icmp eq <256 x i8> %op1, %op2
+  %sext = sext <256 x i1> %cmp to <256 x i8>
+  store <256 x i8> %sext, <256 x i8>* %a
+  ret void
+}
+
+; Don't use SVE for 64-bit vectors.
+define <4 x i16> @icmp_eq_v4i16(<4 x i16> %op1, <4 x i16> %op2) #0 {
+; CHECK-LABEL: icmp_eq_v4i16:
+; CHECK: cmeq v0.4h, v0.4h, v1.4h
+; CHECK-NEXT: ret
+  %cmp = icmp eq <4 x i16> %op1, %op2
+  %sext = sext <4 x i1> %cmp to <4 x i16>
+  ret <4 x i16> %sext
+}
+
+; Don't use SVE for 128-bit vectors.
+define <8 x i16> @icmp_eq_v8i16(<8 x i16> %op1, <8 x i16> %op2) #0 {
+; CHECK-LABEL: icmp_eq_v8i16:
+; CHECK: cmeq v0.8h, v0.8h, v1.8h
+; CHECK-NEXT: ret
+  %cmp = icmp eq <8 x i16> %op1, %op2
+  %sext = sext <8 x i1> %cmp to <8 x i16>
+  ret <8 x i16> %sext
+}
+
+define void @icmp_eq_v16i16(<16 x i16>* %a, <16 x i16>* %b) #0 {
+; CHECK-LABEL: icmp_eq_v16i16:
+; CHECK: ptrue [[PG:p[0-9]+]].h, vl16
+; CHECK-DAG: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
+; CHECK-DAG: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
+; CHECK-NEXT: cmpeq [[CMP:p[0-9]+]].h, [[PG]]/z, [[OP1]].h, [[OP2]].h
+; CHECK-NEXT: mov [[SEXT:z[0-9]+]].h, [[CMP]]/z, #-1
+; CHECK-NEXT: st1h { [[SEXT]].h }, [[PG]], [x0]
+; CHECK-NEXT: ret
+  %op1 = load <16 x i16>, <16 x i16>* %a
+  %op2 = load <16 x i16>, <16 x i16>* %b
+  %cmp = icmp eq <16 x i16> %op1, %op2
+  %sext = sext <16 x i1> %cmp to <16 x i16>
+  store <16 x i16> %sext, <16 x i16>* %a
+  ret void
+}
+
+define void @icmp_eq_v32i16(<32 x i16>* %a, <32 x i16>* %b) #0 {
+; CHECK-LABEL: icmp_eq_v32i16:
+; VBITS_GE_512: ptrue [[PG:p[0-9]+]].h, vl32
+; VBITS_GE_512-DAG: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
+; VBITS_GE_512-DAG: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
+; VBITS_GE_512-NEXT: cmpeq [[CMP:p[0-9]+]].h, [[PG]]/z, [[OP1]].h, [[OP2]].h
+; VBITS_GE_512-NEXT: mov [[SEXT:z[0-9]+]].h, [[CMP]]/z, #-1
+; VBITS_GE_512-NEXT: st1h { [[SEXT]].h }, [[PG]], [x0]
+; VBITS_GE_512-NEXT: ret
+
+; Ensure sensible type legalisation
+; VBITS_EQ_256-DAG: ptrue [[PG:p[0-9]+]].h, vl16
+; VBITS_EQ_256-DAG: add x[[A_HI:[0-9]+]], x0, #32
+; VBITS_EQ_256-DAG: add x[[B_HI:[0-9]+]], x1, #32
+; VBITS_EQ_256-DAG: ld1h { [[OP1_LO:z[0-9]+]].h }, [[PG]]/z, [x0]
+; VBITS_EQ_256-DAG: ld1h { [[OP1_HI:z[0-9]+]].h }, [[PG]]/z, [x[[A_HI]]]
+; VBITS_EQ_256-DAG: ld1h { [[OP2_LO:z[0-9]+]].h }, [[PG]]/z, [x1]
+; VBITS_EQ_256-DAG: ld1h { [[OP2_HI:z[0-9]+]].h }, [[PG]]/z, [x[[B_HI]]]
+; VBITS_EQ_256-DAG: cmpeq [[CMP_LO:p[0-9]+]].h, [[PG]]/z, [[OP1_LO]].h, [[OP2_LO]].h
+; VBITS_EQ_256-DAG: cmpeq [[CMP_HI:p[0-9]+]].h, [[PG]]/z, [[OP1_HI]].h, [[OP2_HI]].h
+; VBITS_EQ_256-DAG: mov [[SEXT_LO:z[0-9]+]].h, [[CMP_LO]]/z, #-1
+; VBITS_EQ_256-DAG: mov [[SEXT_HI:z[0-9]+]].h, [[CMP_HI]]/z, #-1
+; VBITS_EQ_256-DAG: st1h { [[SEXT_LO]].h }, [[PG]], [x0]
+; VBITS_EQ_256-DAG: st1h { [[SEXT_HI]].h }, [[PG]], [x[[A_HI]]]
+; VBITS_EQ_256-NEXT: ret
+  %op1 = load <32 x i16>, <32 x i16>* %a
+  %op2 = load <32 x i16>, <32 x i16>* %b
+  %cmp = icmp eq <32 x i16> %op1, %op2
+  %sext = sext <32 x i1> %cmp to <32 x i16>
+  store <32 x i16> %sext, <32 x i16>* %a
+  ret void
+}
+
+define void @icmp_eq_v64i16(<64 x i16>* %a, <64 x i16>* %b) #0 {
+; CHECK-LABEL: icmp_eq_v64i16:
+; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].h, vl64
+; VBITS_GE_1024-DAG: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
+; VBITS_GE_1024-DAG: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
+; VBITS_GE_1024-NEXT: cmpeq [[CMP:p[0-9]+]].h, [[PG]]/z, [[OP1]].h, [[OP2]].h
+; VBITS_GE_1024-NEXT: mov [[SEXT:z[0-9]+]].h, [[CMP]]/z, #-1
+; VBITS_GE_1024-NEXT: st1h { [[SEXT]].h }, [[PG]], [x0]
+; VBITS_GE_1024-NEXT: ret
+  %op1 = load <64 x i16>, <64 x i16>* %a
+  %op2 = load <64 x i16>, <64 x i16>* %b
+  %cmp = icmp eq <64 x i16> %op1, %op2
+  %sext = sext <64 x i1> %cmp to <64 x i16>
+  store <64 x i16> %sext, <64 x i16>* %a
+  ret void
+}
+
+define void @icmp_eq_v128i16(<128 x i16>* %a, <128 x i16>* %b) #0 {
+; CHECK-LABEL: icmp_eq_v128i16:
+; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].h, vl128
+; VBITS_GE_2048-DAG: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
+; VBITS_GE_2048-DAG: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
+; VBITS_GE_2048-NEXT: cmpeq [[CMP:p[0-9]+]].h, [[PG]]/z, [[OP1]].h, [[OP2]].h
+; VBITS_GE_2048-NEXT: mov [[SEXT:z[0-9]+]].h, [[CMP]]/z, #-1
+; VBITS_GE_2048-NEXT: st1h { [[SEXT]].h }, [[PG]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %op1 = load <128 x i16>, <128 x i16>* %a
+  %op2 = load <128 x i16>, <128 x i16>* %b
+  %cmp = icmp eq <128 x i16> %op1, %op2
+  %sext = sext <128 x i1> %cmp to <128 x i16>
+  store <128 x i16> %sext, <128 x i16>* %a
+  ret void
+}
+
+; Don't use SVE for 64-bit vectors.
+define <2 x i32> @icmp_eq_v2i32(<2 x i32> %op1, <2 x i32> %op2) #0 {
+; CHECK-LABEL: icmp_eq_v2i32:
+; CHECK: cmeq v0.2s, v0.2s, v1.2s
+; CHECK-NEXT: ret
+  %cmp = icmp eq <2 x i32> %op1, %op2
+  %sext = sext <2 x i1> %cmp to <2 x i32>
+  ret <2 x i32> %sext
+}
+
+; Don't use SVE for 128-bit vectors.
+define <4 x i32> @icmp_eq_v4i32(<4 x i32> %op1, <4 x i32> %op2) #0 {
+; CHECK-LABEL: icmp_eq_v4i32:
+; CHECK: cmeq v0.4s, v0.4s, v1.4s
+; CHECK-NEXT: ret
+  %cmp = icmp eq <4 x i32> %op1, %op2
+  %sext = sext <4 x i1> %cmp to <4 x i32>
+  ret <4 x i32> %sext
+}
+
+define void @icmp_eq_v8i32(<8 x i32>* %a, <8 x i32>* %b) #0 {
+; CHECK-LABEL: icmp_eq_v8i32:
+; CHECK: ptrue [[PG:p[0-9]+]].s, vl8
+; CHECK-DAG: ld1w { [[OP1:z[0-9]+]].s }, [[PG]]/z, [x0]
+; CHECK-DAG: ld1w { [[OP2:z[0-9]+]].s }, [[PG]]/z, [x1]
+; CHECK-NEXT: cmpeq [[CMP:p[0-9]+]].s, [[PG]]/z, [[OP1]].s, [[OP2]].s
+; CHECK-NEXT: mov [[SEXT:z[0-9]+]].s, [[CMP]]/z, #-1
+; CHECK-NEXT: st1w { [[SEXT]].s }, [[PG]], [x0]
+; CHECK-NEXT: ret
+  %op1 = load <8 x i32>, <8 x i32>* %a
+  %op2 = load <8 x i32>, <8 x i32>* %b
+  %cmp = icmp eq <8 x i32> %op1, %op2
+  %sext = sext <8 x i1> %cmp to <8 x i32>
+  store <8 x i32> %sext, <8 x i32>* %a
+  ret void
+}
+
+define void @icmp_eq_v16i32(<16 x i32>* %a, <16 x i32>* %b) #0 {
+; CHECK-LABEL: icmp_eq_v16i32:
+; VBITS_GE_512: ptrue [[PG:p[0-9]+]].s, vl16
+; VBITS_GE_512-DAG: ld1w { [[OP1:z[0-9]+]].s }, [[PG]]/z, [x0]
+; VBITS_GE_512-DAG: ld1w { [[OP2:z[0-9]+]].s }, [[PG]]/z, [x1]
+; VBITS_GE_512-NEXT: cmpeq [[CMP:p[0-9]+]].s, [[PG]]/z, [[OP1]].s, [[OP2]].s
+; VBITS_GE_512-NEXT: mov [[SEXT:z[0-9]+]].s, [[CMP]]/z, #-1
+; VBITS_GE_512-NEXT: st1w { [[SEXT]].s }, [[PG]], [x0]
+; VBITS_GE_512-NEXT: ret
+
+; Ensure sensible type legalisation
+; VBITS_EQ_256-DAG: ptrue [[PG:p[0-9]+]].s, vl8
+; VBITS_EQ_256-DAG: add x[[A_HI:[0-9]+]], x0, #32
+; VBITS_EQ_256-DAG: add x[[B_HI:[0-9]+]], x1, #32
+; VBITS_EQ_256-DAG: ld1w { [[OP1_LO:z[0-9]+]].s }, [[PG]]/z, [x0]
+; VBITS_EQ_256-DAG: ld1w { [[OP1_HI:z[0-9]+]].s }, [[PG]]/z, [x[[A_HI]]]
+; VBITS_EQ_256-DAG: ld1w { [[OP2_LO:z[0-9]+]].s }, [[PG]]/z, [x1]
+; VBITS_EQ_256-DAG: ld1w { [[OP2_HI:z[0-9]+]].s }, [[PG]]/z, [x[[B_HI]]]
+; VBITS_EQ_256-DAG: cmpeq [[CMP_LO:p[0-9]+]].s, [[PG]]/z, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_EQ_256-DAG: cmpeq [[CMP_HI:p[0-9]+]].s, [[PG]]/z, [[OP1_HI]].s, [[OP2_HI]].s
+; VBITS_EQ_256-DAG: mov [[SEXT_LO:z[0-9]+]].s, [[CMP_LO]]/z, #-1
+; VBITS_EQ_256-DAG: mov [[SEXT_HI:z[0-9]+]].s, [[CMP_HI]]/z, #-1
+; VBITS_EQ_256-DAG: st1w { [[SEXT_LO]].s }, [[PG]], [x0]
+; VBITS_EQ_256-DAG: st1w { [[SEXT_HI]].s }, [[PG]], [x[[A_HI]]]
+; VBITS_EQ_256-NEXT: ret
+  %op1 = load <16 x i32>, <16 x i32>* %a
+  %op2 = load <16 x i32>, <16 x i32>* %b
+  %cmp = icmp eq <16 x i32> %op1, %op2
+  %sext = sext <16 x i1> %cmp to <16 x i32>
+  store <16 x i32> %sext, <16 x i32>* %a
+  ret void
+}
+
+define void @icmp_eq_v32i32(<32 x i32>* %a, <32 x i32>* %b) #0 {
+; CHECK-LABEL: icmp_eq_v32i32:
+; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].s, vl32
+; VBITS_GE_1024-DAG: ld1w { [[OP1:z[0-9]+]].s }, [[PG]]/z, [x0]
+; VBITS_GE_1024-DAG: ld1w { [[OP2:z[0-9]+]].s }, [[PG]]/z, [x1]
+; VBITS_GE_1024-NEXT: cmpeq [[CMP:p[0-9]+]].s, [[PG]]/z, [[OP1]].s, [[OP2]].s
+; VBITS_GE_1024-NEXT: mov [[SEXT:z[0-9]+]].s, [[CMP]]/z, #-1
+; VBITS_GE_1024-NEXT: st1w { [[SEXT]].s }, [[PG]], [x0]
+; VBITS_GE_1024-NEXT: ret
+  %op1 = load <32 x i32>, <32 x i32>* %a
+  %op2 = load <32 x i32>, <32 x i32>* %b
+  %cmp = icmp eq <32 x i32> %op1, %op2
+  %sext = sext <32 x i1> %cmp to <32 x i32>
+  store <32 x i32> %sext, <32 x i32>* %a
+  ret void
+}
+
+define void @icmp_eq_v64i32(<64 x i32>* %a, <64 x i32>* %b) #0 {
+; CHECK-LABEL: icmp_eq_v64i32:
+; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].s, vl64
+; VBITS_GE_2048-DAG: ld1w { [[OP1:z[0-9]+]].s }, [[PG]]/z, [x0]
+; VBITS_GE_2048-DAG: ld1w { [[OP2:z[0-9]+]].s }, [[PG]]/z, [x1]
+; VBITS_GE_2048-NEXT: cmpeq [[CMP:p[0-9]+]].s, [[PG]]/z, [[OP1]].s, [[OP2]].s
+; VBITS_GE_2048-NEXT: mov [[SEXT:z[0-9]+]].s, [[CMP]]/z, #-1
+; VBITS_GE_2048-NEXT: st1w { [[SEXT]].s }, [[PG]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %op1 = load <64 x i32>, <64 x i32>* %a
+  %op2 = load <64 x i32>, <64 x i32>* %b
+  %cmp = icmp eq <64 x i32> %op1, %op2
+  %sext = sext <64 x i1> %cmp to <64 x i32>
+  store <64 x i32> %sext, <64 x i32>* %a
+  ret void
+}
+
+; Don't use SVE for 64-bit vectors.
+define <1 x i64> @icmp_eq_v1i64(<1 x i64> %op1, <1 x i64> %op2) #0 {
+; CHECK-LABEL: icmp_eq_v1i64:
+; CHECK: cmeq d0, d0, d1
+; CHECK-NEXT: ret
+  %cmp = icmp eq <1 x i64> %op1, %op2
+  %sext = sext <1 x i1> %cmp to <1 x i64>
+  ret <1 x i64> %sext
+}
+
+; Don't use SVE for 128-bit vectors.
+define <2 x i64> @icmp_eq_v2i64(<2 x i64> %op1, <2 x i64> %op2) #0 {
+; CHECK-LABEL: icmp_eq_v2i64:
+; CHECK: cmeq v0.2d, v0.2d, v1.2d
+; CHECK-NEXT: ret
+  %cmp = icmp eq <2 x i64> %op1, %op2
+  %sext = sext <2 x i1> %cmp to <2 x i64>
+  ret <2 x i64> %sext
+}
+
+define void @icmp_eq_v4i64(<4 x i64>* %a, <4 x i64>* %b) #0 {
+; CHECK-LABEL: icmp_eq_v4i64:
+; CHECK: ptrue [[PG:p[0-9]+]].d, vl4
+; CHECK-DAG: ld1d { [[OP1:z[0-9]+]].d }, [[PG]]/z, [x0]
+; CHECK-DAG: ld1d { [[OP2:z[0-9]+]].d }, [[PG]]/z, [x1]
+; CHECK-NEXT: cmpeq [[CMP:p[0-9]+]].d, [[PG]]/z, [[OP1]].d, [[OP2]].d
+; CHECK-NEXT: mov [[SEXT:z[0-9]+]].d, [[CMP]]/z, #-1
+; CHECK-NEXT: st1d { [[SEXT]].d }, [[PG]], [x0]
+; CHECK-NEXT: ret
+  %op1 = load <4 x i64>, <4 x i64>* %a
+  %op2 = load <4 x i64>, <4 x i64>* %b
+  %cmp = icmp eq <4 x i64> %op1, %op2
+  %sext = sext <4 x i1> %cmp to <4 x i64>
+  store <4 x i64> %sext, <4 x i64>* %a
+  ret void
+}
+
+define void @icmp_eq_v8i64(<8 x i64>* %a, <8 x i64>* %b) #0 {
+; CHECK-LABEL: icmp_eq_v8i64:
+; VBITS_GE_512: ptrue [[PG:p[0-9]+]].d, vl8
+; VBITS_GE_512-DAG: ld1d { [[OP1:z[0-9]+]].d }, [[PG]]/z, [x0]
+; VBITS_GE_512-DAG: ld1d { [[OP2:z[0-9]+]].d }, [[PG]]/z, [x1]
+; VBITS_GE_512-NEXT: cmpeq [[CMP:p[0-9]+]].d, [[PG]]/z, [[OP1]].d, [[OP2]].d
+; VBITS_GE_512-NEXT: mov [[SEXT:z[0-9]+]].d, [[CMP]]/z, #-1
+; VBITS_GE_512-NEXT: st1d { [[SEXT]].d }, [[PG]], [x0]
+; VBITS_GE_512-NEXT: ret
+
+; Ensure sensible type legalisation
+; VBITS_EQ_256-DAG: ptrue [[PG:p[0-9]+]].d, vl4
+; VBITS_EQ_256-DAG: add x[[A_HI:[0-9]+]], x0, #32
+; VBITS_EQ_256-DAG: add x[[B_HI:[0-9]+]], x1, #32
+; VBITS_EQ_256-DAG: ld1d { [[OP1_LO:z[0-9]+]].d }, [[PG]]/z, [x0]
+; VBITS_EQ_256-DAG: ld1d { [[OP1_HI:z[0-9]+]].d }, [[PG]]/z, [x[[A_HI]]]
+; VBITS_EQ_256-DAG: ld1d { [[OP2_LO:z[0-9]+]].d }, [[PG]]/z, [x1]
+; VBITS_EQ_256-DAG: ld1d { [[OP2_HI:z[0-9]+]].d }, [[PG]]/z, [x[[B_HI]]]
+; VBITS_EQ_256-DAG: cmpeq [[CMP_LO:p[0-9]+]].d, [[PG]]/z, [[OP1_LO]].d, [[OP2_LO]].d
+; VBITS_EQ_256-DAG: cmpeq [[CMP_HI:p[0-9]+]].d, [[PG]]/z, [[OP1_HI]].d, [[OP2_HI]].d
+; VBITS_EQ_256-DAG: mov [[SEXT_LO:z[0-9]+]].d, [[CMP_LO]]/z, #-1
+; VBITS_EQ_256-DAG: mov [[SEXT_HI:z[0-9]+]].d, [[CMP_HI]]/z, #-1
+; VBITS_EQ_256-DAG: st1d { [[SEXT_LO]].d }, [[PG]], [x0]
+; VBITS_EQ_256-DAG: st1d { [[SEXT_HI]].d }, [[PG]], [x[[A_HI]]]
+; VBITS_EQ_256-NEXT: ret
+  %op1 = load <8 x i64>, <8 x i64>* %a
+  %op2 = load <8 x i64>, <8 x i64>* %b
+  %cmp = icmp eq <8 x i64> %op1, %op2
+  %sext = sext <8 x i1> %cmp to <8 x i64>
+  store <8 x i64> %sext, <8 x i64>* %a
+  ret void
+}
+
+define void @icmp_eq_v16i64(<16 x i64>* %a, <16 x i64>* %b) #0 {
+; CHECK-LABEL: icmp_eq_v16i64:
+; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].d, vl16
+; VBITS_GE_1024-DAG: ld1d { [[OP1:z[0-9]+]].d }, [[PG]]/z, [x0]
+; VBITS_GE_1024-DAG: ld1d { [[OP2:z[0-9]+]].d }, [[PG]]/z, [x1]
+; VBITS_GE_1024-NEXT: cmpeq [[CMP:p[0-9]+]].d, [[PG]]/z, [[OP1]].d, [[OP2]].d
+; VBITS_GE_1024-NEXT: mov [[SEXT:z[0-9]+]].d, [[CMP]]/z, #-1
+; VBITS_GE_1024-NEXT: st1d { [[SEXT]].d }, [[PG]], [x0]
+; VBITS_GE_1024-NEXT: ret
+  %op1 = load <16 x i64>, <16 x i64>* %a
+  %op2 = load <16 x i64>, <16 x i64>* %b
+  %cmp = icmp eq <16 x i64> %op1, %op2
+  %sext = sext <16 x i1> %cmp to <16 x i64>
+  store <16 x i64> %sext, <16 x i64>* %a
+  ret void
+}
+
+define void @icmp_eq_v32i64(<32 x i64>* %a, <32 x i64>* %b) #0 {
+; CHECK-LABEL: icmp_eq_v32i64:
+; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].d, vl32
+; VBITS_GE_2048-DAG: ld1d { [[OP1:z[0-9]+]].d }, [[PG]]/z, [x0]
+; VBITS_GE_2048-DAG: ld1d { [[OP2:z[0-9]+]].d }, [[PG]]/z, [x1]
+; VBITS_GE_2048-NEXT: cmpeq [[CMP:p[0-9]+]].d, [[PG]]/z, [[OP1]].d, [[OP2]].d
+; VBITS_GE_2048-NEXT: mov [[SEXT:z[0-9]+]].d, [[CMP]]/z, #-1
+; VBITS_GE_2048-NEXT: st1d { [[SEXT]].d }, [[PG]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %op1 = load <32 x i64>, <32 x i64>* %a
+  %op2 = load <32 x i64>, <32 x i64>* %b
+  %cmp = icmp eq <32 x i64> %op1, %op2
+  %sext = sext <32 x i1> %cmp to <32 x i64>
+  store <32 x i64> %sext, <32 x i64>* %a
+  ret void
+}
+
+;
+; ICMP NE
+;
+
+define void @icmp_ne_v32i8(<32 x i8>* %a, <32 x i8>* %b) #0 {
+; CHECK-LABEL: icmp_ne_v32i8:
+; CHECK: ptrue [[PG:p[0-9]+]].b, vl32
+; CHECK-DAG: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
+; CHECK-DAG: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
+; CHECK-NEXT: cmpne [[CMP:p[0-9]+]].b, [[PG]]/z, [[OP1]].b, [[OP2]].b
+; CHECK-NEXT: mov [[SEXT:z[0-9]+]].b, [[CMP]]/z, #-1
+; CHECK-NEXT: st1b { [[SEXT]].b }, [[PG]], [x0]
+; CHECK-NEXT: ret
+  %op1 = load <32 x i8>, <32 x i8>* %a
+  %op2 = load <32 x i8>, <32 x i8>* %b
+  %cmp = icmp ne <32 x i8> %op1, %op2
+  %sext = sext <32 x i1> %cmp to <32 x i8>
+  store <32 x i8> %sext, <32 x i8>* %a
+  ret void
+}
+
+;
+; ICMP SGE
+;
+
+define void @icmp_sge_v32i16(<32 x i16>* %a, <32 x i16>* %b) #0 {
+; CHECK-LABEL: icmp_sge_v32i16:
+; VBITS_GE_512: ptrue [[PG:p[0-9]+]].h, vl32
+; VBITS_GE_512-DAG: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
+; VBITS_GE_512-DAG: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
+; VBITS_GE_512-NEXT: cmpge [[CMP:p[0-9]+]].h, [[PG]]/z, [[OP1]].h, [[OP2]].h
+; VBITS_GE_512-NEXT: mov [[SEXT:z[0-9]+]].h, [[CMP]]/z, #-1
+; VBITS_GE_512-NEXT: st1h { [[SEXT]].h }, [[PG]], [x0]
+; VBITS_GE_512-NEXT: ret
+  %op1 = load <32 x i16>, <32 x i16>* %a
+  %op2 = load <32 x i16>, <32 x i16>* %b
+  %cmp = icmp sge <32 x i16> %op1, %op2
+  %sext = sext <32 x i1> %cmp to <32 x i16>
+  store <32 x i16> %sext, <32 x i16>* %a
+  ret void
+}
+
+;
+; ICMP SGT
+;
+
+define void @icmp_sgt_v16i16(<16 x i16>* %a, <16 x i16>* %b) #0 {
+; CHECK-LABEL: icmp_sgt_v16i16:
+; CHECK: ptrue [[PG:p[0-9]+]].h, vl16
+; CHECK-DAG: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
+; CHECK-DAG: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
+; CHECK-NEXT: cmpgt [[CMP:p[0-9]+]].h, [[PG]]/z, [[OP1]].h, [[OP2]].h
+; CHECK-NEXT: mov [[SEXT:z[0-9]+]].h, [[CMP]]/z, #-1
+; CHECK-NEXT: st1h { [[SEXT]].h }, [[PG]], [x0]
+; CHECK-NEXT: ret
+  %op1 = load <16 x i16>, <16 x i16>* %a
+  %op2 = load <16 x i16>, <16 x i16>* %b
+  %cmp = icmp sgt <16 x i16> %op1, %op2
+  %sext = sext <16 x i1> %cmp to <16 x i16>
+  store <16 x i16> %sext, <16 x i16>* %a
+  ret void
+}
+
+;
+; ICMP SLE
+;
+
+define void @icmp_sle_v16i32(<16 x i32>* %a, <16 x i32>* %b) #0 {
+; CHECK-LABEL: icmp_sle_v16i32:
+; VBITS_GE_512: ptrue [[PG:p[0-9]+]].s, vl16
+; VBITS_GE_512-DAG: ld1w { [[OP1:z[0-9]+]].s }, [[PG]]/z, [x0]
+; VBITS_GE_512-DAG: ld1w { [[OP2:z[0-9]+]].s }, [[PG]]/z, [x1]
+; VBITS_GE_512-NEXT: cmpge [[CMP:p[0-9]+]].s, [[PG]]/z, [[OP2]].s, [[OP1]].s
+; VBITS_GE_512-NEXT: mov [[SEXT:z[0-9]+]].s, [[CMP]]/z, #-1
+; VBITS_GE_512-NEXT: st1w { [[SEXT]].s }, [[PG]], [x0]
+; VBITS_GE_512-NEXT: ret
+  %op1 = load <16 x i32>, <16 x i32>* %a
+  %op2 = load <16 x i32>, <16 x i32>* %b
+  %cmp = icmp sle <16 x i32> %op1, %op2
+  %sext = sext <16 x i1> %cmp to <16 x i32>
+  store <16 x i32> %sext, <16 x i32>* %a
+  ret void
+}
+
+;
+; ICMP SLT
+;
+
+define void @icmp_slt_v8i32(<8 x i32>* %a, <8 x i32>* %b) #0 {
+; CHECK-LABEL: icmp_slt_v8i32:
+; CHECK: ptrue [[PG:p[0-9]+]].s, vl8
+; CHECK-DAG: ld1w { [[OP1:z[0-9]+]].s }, [[PG]]/z, [x0]
+; CHECK-DAG: ld1w { [[OP2:z[0-9]+]].s }, [[PG]]/z, [x1]
+; CHECK-NEXT: cmpgt [[CMP:p[0-9]+]].s, [[PG]]/z, [[OP2]].s, [[OP1]].s
+; CHECK-NEXT: mov [[SEXT:z[0-9]+]].s, [[CMP]]/z, #-1
+; CHECK-NEXT: st1w { [[SEXT]].s }, [[PG]], [x0]
+; CHECK-NEXT: ret
+  %op1 = load <8 x i32>, <8 x i32>* %a
+  %op2 = load <8 x i32>, <8 x i32>* %b
+  %cmp = icmp slt <8 x i32> %op1, %op2
+  %sext = sext <8 x i1> %cmp to <8 x i32>
+  store <8 x i32> %sext, <8 x i32>* %a
+  ret void
+}
+
+;
+; ICMP UGE
+;
+
+define void @icmp_uge_v8i64(<8 x i64>* %a, <8 x i64>* %b) #0 {
+; CHECK-LABEL: icmp_uge_v8i64:
+; VBITS_GE_512: ptrue [[PG:p[0-9]+]].d, vl8
+; VBITS_GE_512-DAG: ld1d { [[OP1:z[0-9]+]].d }, [[PG]]/z, [x0]
+; VBITS_GE_512-DAG: ld1d { [[OP2:z[0-9]+]].d }, [[PG]]/z, [x1]
+; VBITS_GE_512-NEXT: cmphs [[CMP:p[0-9]+]].d, [[PG]]/z, [[OP1]].d, [[OP2]].d
+; VBITS_GE_512-NEXT: mov [[SEXT:z[0-9]+]].d, [[CMP]]/z, #-1
+; VBITS_GE_512-NEXT: st1d { [[SEXT]].d }, [[PG]], [x0]
+; VBITS_GE_512-NEXT: ret
+  %op1 = load <8 x i64>, <8 x i64>* %a
+  %op2 = load <8 x i64>, <8 x i64>* %b
+  %cmp = icmp uge <8 x i64> %op1, %op2
+  %sext = sext <8 x i1> %cmp to <8 x i64>
+  store <8 x i64> %sext, <8 x i64>* %a
+  ret void
+}
+
+;
+; ICMP UGT
+;
+
+define void @icmp_ugt_v4i64(<4 x i64>* %a, <4 x i64>* %b) #0 {
+; CHECK-LABEL: icmp_ugt_v4i64:
+; CHECK: ptrue [[PG:p[0-9]+]].d, vl4
+; CHECK-DAG: ld1d { [[OP1:z[0-9]+]].d }, [[PG]]/z, [x0]
+; CHECK-DAG: ld1d { [[OP2:z[0-9]+]].d }, [[PG]]/z, [x1]
+; CHECK-NEXT: cmphi [[CMP:p[0-9]+]].d, [[PG]]/z, [[OP1]].d, [[OP2]].d
+; CHECK-NEXT: mov [[SEXT:z[0-9]+]].d, [[CMP]]/z, #-1
+; CHECK-NEXT: st1d { [[SEXT]].d }, [[PG]], [x0]
+; CHECK-NEXT: ret
+  %op1 = load <4 x i64>, <4 x i64>* %a
+  %op2 = load <4 x i64>, <4 x i64>* %b
+  %cmp = icmp ugt <4 x i64> %op1, %op2
+  %sext = sext <4 x i1> %cmp to <4 x i64>
+  store <4 x i64> %sext, <4 x i64>* %a
+  ret void
+}
+
+;
+; ICMP ULE
+;
+
+define void @icmp_ule_v16i64(<16 x i64>* %a, <16 x i64>* %b) #0 {
+; CHECK-LABEL: icmp_ule_v16i64:
+; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].d, vl16
+; VBITS_GE_1024-DAG: ld1d { [[OP1:z[0-9]+]].d }, [[PG]]/z, [x0]
+; VBITS_GE_1024-DAG: ld1d { [[OP2:z[0-9]+]].d }, [[PG]]/z, [x1]
+; VBITS_GE_1024-NEXT: cmphs [[CMP:p[0-9]+]].d, [[PG]]/z, [[OP2]].d, [[OP1]].d
+; VBITS_GE_1024-NEXT: mov [[SEXT:z[0-9]+]].d, [[CMP]]/z, #-1
+; VBITS_GE_1024-NEXT: st1d { [[SEXT]].d }, [[PG]], [x0]
+; VBITS_GE_1024-NEXT: ret
+  %op1 = load <16 x i64>, <16 x i64>* %a
+  %op2 = load <16 x i64>, <16 x i64>* %b
+  %cmp = icmp ule <16 x i64> %op1, %op2
+  %sext = sext <16 x i1> %cmp to <16 x i64>
+  store <16 x i64> %sext, <16 x i64>* %a
+  ret void
+}
+
+;
+; ICMP ULT
+;
+
+define void @icmp_ult_v32i64(<32 x i64>* %a, <32 x i64>* %b) #0 {
+; CHECK-LABEL: icmp_ult_v32i64:
+; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].d, vl32
+; VBITS_GE_2048-DAG: ld1d { [[OP1:z[0-9]+]].d }, [[PG]]/z, [x0]
+; VBITS_GE_2048-DAG: ld1d { [[OP2:z[0-9]+]].d }, [[PG]]/z, [x1]
+; VBITS_GE_2048-NEXT: cmphi [[CMP:p[0-9]+]].d, [[PG]]/z, [[OP2]].d, [[OP1]].d
+; VBITS_GE_2048-NEXT: mov [[SEXT:z[0-9]+]].d, [[CMP]]/z, #-1
+; VBITS_GE_2048-NEXT: st1d { [[SEXT]].d }, [[PG]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %op1 = load <32 x i64>, <32 x i64>* %a
+  %op2 = load <32 x i64>, <32 x i64>* %b
+  %cmp = icmp ult <32 x i64> %op1, %op2
+  %sext = sext <32 x i1> %cmp to <32 x i64>
+  store <32 x i64> %sext, <32 x i64>* %a
+  ret void
+}
+
+attributes #0 = { "target-features"="+sve" }

llvm/test/CodeGen/AArch64/sve-fixed-length-subvector.ll

bb1:
   store <64 x i32> %a, <64 x i32>* %out
   ret void
 }
 
 ;
 define <8 x i1> @no_warn_dropped_scalable(<8 x i32>* %in) #0 {
 ; CHECK-LABEL: no_warn_dropped_scalable:
 ; CHECK: ptrue [[PG:p[0-9]+]].s, vl8
-; CHECK: ld1w { z{{[0-9]+}}.s }, [[PG]]/z, [x0]
-; CHECK-COUNT-8: cmp w{{[0-9]+}}, #0
+; CHECK: ld1w { [[A:z[0-9]+]].s }, [[PG]]/z, [x0]
+; CHECK: cmpgt p{{[0-9]}}.s, [[PG]]/z, [[A]].s, z{{[0-9]+}}.s
 ; CHECK: ret
   %a = load <8 x i32>, <8 x i32>* %in
   br label %bb1
 
 bb1:
   %cond = icmp sgt <8 x i32> %a, zeroinitializer
   ret <8 x i1> %cond
 }
 
 attributes #0 = { "target-features"="+sve" }