[RISCV] Support masked load/store for fixed vectors.

[frasercrmck]

Generally, I'd go with the suggestions that clang-tidy has come up with.

llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h

64

Personal preference, but since this supports both load and store, for clarity we could make this a separate isLegalMaskedLoadStore and have isLegalMaskedLoad call that too?

Additionally, is "legality" here just that it shouldn't scalarize? Will the legalizer split too-wide vectors?

llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-load-fp.ll

2

It would be good to check that illegal vectors are correctly split in twain and we don't crash/scalarize.

[craig.topper]

clang-format. Add tests for splitting.

[craig.topper]

llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h

64

I believe the loop vectorizer calls this with a scalar type. There's a separate cost function too. The ScalarizeMaskedMemIntrinsic pass will call it with a fixed vector type to know if it needs to scalarize and insert conditional branches. The type legalizer in SelectionDAG can handle splitting, but not scalarizing.

[craig.topper]

Add isLegalMaskedLoadStore call it from isLegalMaskedLoad/isLegalMaskedStore

[frasercrmck]

LGTM. Is there a reason the scalable-vector versions weren't lowered to intrinsics?

[craig.topper]

In D98561#2631121, @frasercrmck wrote:

LGTM. Is there a reason the scalable-vector versions weren't lowered to intrinsics?

It didn't require custom handling just pattern matching. I can go back and make them the same.

llvm/lib/Target/RISCV/RISCVISelLowering.h

private:
   SDValue lowerINSERT_SUBVECTOR(SDValue Op, SelectionDAG &DAG) const;
   SDValue lowerEXTRACT_SUBVECTOR(SDValue Op, SelectionDAG &DAG) const;
   SDValue lowerVECTOR_REVERSE(SDValue Op, SelectionDAG &DAG) const;
   SDValue lowerABS(SDValue Op, SelectionDAG &DAG) const;
   SDValue lowerFixedLengthVectorFCOPYSIGNToRVV(SDValue Op,
                                                SelectionDAG &DAG) const;
   SDValue lowerFixedLengthVectorLoadToRVV(SDValue Op, SelectionDAG &DAG) const;
   SDValue lowerFixedLengthVectorStoreToRVV(SDValue Op, SelectionDAG &DAG) const;
+  SDValue lowerFixedLengthVectorMaskedLoadToRVV(SDValue Op,
+                                                SelectionDAG &DAG) const;
+  SDValue lowerFixedLengthVectorMaskedStoreToRVV(SDValue Op,
+                                                 SelectionDAG &DAG) const;
   SDValue lowerFixedLengthVectorSetccToRVV(SDValue Op, SelectionDAG &DAG) const;
   SDValue lowerFixedLengthVectorLogicOpToRVV(SDValue Op, SelectionDAG &DAG,
                                              unsigned MaskOpc,
                                              unsigned VecOpc) const;
   SDValue lowerFixedLengthVectorSelectToRVV(SDValue Op,
                                             SelectionDAG &DAG) const;
   SDValue lowerToScalableOp(SDValue Op, SelectionDAG &DAG, unsigned NewOpc,
                             bool HasMask = true) const;

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

if (Subtarget.useRVVForFixedLengthVectors()) {
           setOperationAction(ISD::OR, VT, Custom);
           setOperationAction(ISD::XOR, VT, Custom);
           continue;
         }
 
         setOperationAction(ISD::VECTOR_SHUFFLE, VT, Custom);
         setOperationAction(ISD::INSERT_VECTOR_ELT, VT, Custom);
 
+        setOperationAction(ISD::MLOAD, VT, Custom);
+        setOperationAction(ISD::MSTORE, VT, Custom);
         setOperationAction(ISD::ADD, VT, Custom);
         setOperationAction(ISD::MUL, VT, Custom);
         setOperationAction(ISD::SUB, VT, Custom);
         setOperationAction(ISD::AND, VT, Custom);
         setOperationAction(ISD::OR, VT, Custom);
         setOperationAction(ISD::XOR, VT, Custom);
         setOperationAction(ISD::SDIV, VT, Custom);
         setOperationAction(ISD::SREM, VT, Custom);
if (Subtarget.useRVVForFixedLengthVectors()) {
 
         setOperationAction(ISD::BUILD_VECTOR, VT, Custom);
         setOperationAction(ISD::VECTOR_SHUFFLE, VT, Custom);
         setOperationAction(ISD::INSERT_VECTOR_ELT, VT, Custom);
         setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Custom);
 
         setOperationAction(ISD::LOAD, VT, Custom);
         setOperationAction(ISD::STORE, VT, Custom);
+        setOperationAction(ISD::MLOAD, VT, Custom);
+        setOperationAction(ISD::MSTORE, VT, Custom);
         setOperationAction(ISD::FADD, VT, Custom);
         setOperationAction(ISD::FSUB, VT, Custom);
         setOperationAction(ISD::FMUL, VT, Custom);
         setOperationAction(ISD::FDIV, VT, Custom);
         setOperationAction(ISD::FNEG, VT, Custom);
         setOperationAction(ISD::FABS, VT, Custom);
         setOperationAction(ISD::FCOPYSIGN, VT, Custom);
         setOperationAction(ISD::FSQRT, VT, Custom);
for (const auto &OpIdx : enumerate(Op->ops()))
       Vec = DAG.getNode(ISD::INSERT_SUBVECTOR, DL, VT, Vec, OpIdx.value(),
                         DAG.getIntPtrConstant(OpIdx.index() * NumOpElts, DL));
     return Vec;
   }
   case ISD::LOAD:
     return lowerFixedLengthVectorLoadToRVV(Op, DAG);
   case ISD::STORE:
     return lowerFixedLengthVectorStoreToRVV(Op, DAG);
+  case ISD::MLOAD:
+    return lowerFixedLengthVectorMaskedLoadToRVV(Op, DAG);
+  case ISD::MSTORE:
+    return lowerFixedLengthVectorMaskedStoreToRVV(Op, DAG);
   case ISD::SETCC:
     return lowerFixedLengthVectorSetccToRVV(Op, DAG);
   case ISD::ADD:
     return lowerToScalableOp(Op, DAG, RISCVISD::ADD_VL);
   case ISD::SUB:
     return lowerToScalableOp(Op, DAG, RISCVISD::SUB_VL);
   case ISD::MUL:
     return lowerToScalableOp(Op, DAG, RISCVISD::MUL_VL);
RISCVTargetLowering::lowerFixedLengthVectorStoreToRVV(SDValue Op,
   SDValue NewValue =
       convertToScalableVector(ContainerVT, Store->getValue(), DAG, Subtarget);
   return DAG.getMemIntrinsicNode(
       RISCVISD::VSE_VL, DL, DAG.getVTList(MVT::Other),
       {Store->getChain(), NewValue, Store->getBasePtr(), VL},
       Store->getMemoryVT(), Store->getMemOperand());
 }
 
+SDValue RISCVTargetLowering::lowerFixedLengthVectorMaskedLoadToRVV(
+    SDValue Op, SelectionDAG &DAG) const {
+  auto *Load = cast<MaskedLoadSDNode>(Op);
+
+  SDLoc DL(Op);
+  MVT VT = Op.getSimpleValueType();
+  MVT ContainerVT = getContainerForFixedLengthVector(VT);
+  MVT MaskVT = MVT::getVectorVT(MVT::i1, ContainerVT.getVectorElementCount());
+  MVT XLenVT = Subtarget.getXLenVT();
+
+  SDValue Mask =
+      convertToScalableVector(MaskVT, Load->getMask(), DAG, Subtarget);
+  SDValue PassThru =
+      convertToScalableVector(ContainerVT, Load->getPassThru(), DAG, Subtarget);
+  SDValue VL = DAG.getConstant(VT.getVectorNumElements(), DL, XLenVT);
+
+  SDVTList VTs = DAG.getVTList({ContainerVT, MVT::Other});
+  SDValue IntID = DAG.getTargetConstant(Intrinsic::riscv_vle_mask, DL, XLenVT);
+  SDValue Ops[] = {Load->getChain(),   IntID, PassThru,
+                   Load->getBasePtr(), Mask,  VL};
+  SDValue NewLoad =
+      DAG.getMemIntrinsicNode(ISD::INTRINSIC_W_CHAIN, DL, VTs, Ops,
+                              Load->getMemoryVT(), Load->getMemOperand());
+
+  SDValue Result = convertFromScalableVector(VT, NewLoad, DAG, Subtarget);
+  return DAG.getMergeValues({Result, Load->getChain()}, DL);
+}
+
+SDValue RISCVTargetLowering::lowerFixedLengthVectorMaskedStoreToRVV(
+    SDValue Op, SelectionDAG &DAG) const {
+  auto *Store = cast<MaskedStoreSDNode>(Op);
+
+  SDLoc DL(Op);
+  SDValue Val = Store->getValue();
+  MVT VT = Val.getSimpleValueType();
+  MVT ContainerVT = getContainerForFixedLengthVector(VT);
+  MVT MaskVT = MVT::getVectorVT(MVT::i1, ContainerVT.getVectorElementCount());
+  MVT XLenVT = Subtarget.getXLenVT();
+
+  Val = convertToScalableVector(ContainerVT, Val, DAG, Subtarget);
+  SDValue Mask =
+      convertToScalableVector(MaskVT, Store->getMask(), DAG, Subtarget);
+
+  SDValue VL = DAG.getConstant(VT.getVectorNumElements(), DL, XLenVT);
+
+  SDValue IntID = DAG.getTargetConstant(Intrinsic::riscv_vse_mask, DL, XLenVT);
+  return DAG.getMemIntrinsicNode(
+      ISD::INTRINSIC_VOID, DL, DAG.getVTList(MVT::Other),
+      {Store->getChain(), IntID, Val, Store->getBasePtr(), Mask, VL},
+      Store->getMemoryVT(), Store->getMemOperand());
+}
+
 SDValue
 RISCVTargetLowering::lowerFixedLengthVectorSetccToRVV(SDValue Op,
                                                       SelectionDAG &DAG) const {
   MVT InVT = Op.getOperand(0).getSimpleValueType();
   MVT ContainerVT = getContainerForFixedLengthVector(InVT);
 
   MVT VT = Op.getSimpleValueType();
 

llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h

public:
   unsigned getRegisterBitWidth(bool Vector) const {
     if (Vector) {
       if (ST->hasStdExtV())
         return ST->getMinRVVVectorSizeInBits();
       return 0;
     }
     return ST->getXLen();
   }
+
+  bool isLegalMaskedLoadStore(Type *DataType, Align Alignment) {
+    if (!ST->hasStdExtV())
+      return false;
+
+    // Only support fixed vectors if we know the minimum vector size.
+    if (isa<FixedVectorType>(DataType) && ST->getMinRVVVectorSizeInBits() == 0)
+      return false;
+
+    Type *ScalarTy = DataType->getScalarType();
+    if (ScalarTy->isPointerTy())
+      return true;
+
+    if (ScalarTy->isIntegerTy(8) || ScalarTy->isIntegerTy(16) ||
+        ScalarTy->isIntegerTy(32) || ScalarTy->isIntegerTy(64))
+      return true;
+
+    if (ScalarTy->isHalfTy())
+      return ST->hasStdExtZfh();
+    if (ScalarTy->isFloatTy())
+      return ST->hasStdExtF();
+    if (ScalarTy->isDoubleTy())
+      return ST->hasStdExtD();
+
+    return false;
+  }
+
+  bool isLegalMaskedLoad(Type *DataType, Align Alignment) {
+    return isLegalMaskedLoadStore(DataType, Alignment);
+  }
+  bool isLegalMaskedStore(Type *DataType, Align Alignment) {
+    return isLegalMaskedLoadStore(DataType, Alignment);
+  }
 };
 
 } // end namespace llvm
 
 #endif // LLVM_LIB_TARGET_RISCV_RISCVTARGETTRANSFORMINFO_H

llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-load-fp.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -mattr=+m,+experimental-v,+f,+d,+experimental-zfh -verify-machineinstrs -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV32
+; RUN: llc -mtriple=riscv64 -mattr=+m,+experimental-v,+f,+d,+experimental-zfh -verify-machineinstrs -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV64
+
+define void @masked_load_v1f16(<1 x half>* %a, <1 x half>* %m_ptr, <1 x half>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v1f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 1, e16,m1,ta,mu
+; CHECK-NEXT:    vle16.v v25, (a1)
+; CHECK-NEXT:    fmv.h.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v25, ft0
+; CHECK-NEXT:    vle16.v v25, (a0), v0.t
+; CHECK-NEXT:    vse16.v v25, (a2)
+; CHECK-NEXT:    ret
+  %m = load <1 x half>, <1 x half>* %m_ptr
+  %mask = fcmp oeq <1 x half> %m, zeroinitializer
+  %load = call <1 x half> @llvm.masked.load.v1f16(<1 x half>* %a, i32 8, <1 x i1> %mask, <1 x half> undef)
+  store <1 x half> %load, <1 x half>* %res_ptr
+  ret void
+}
+declare <1 x half> @llvm.masked.load.v1f16(<1 x half>*, i32, <1 x i1>, <1 x half>)
+
+define void @masked_load_v1f32(<1 x float>* %a, <1 x float>* %m_ptr, <1 x float>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v1f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 1, e32,m1,ta,mu
+; CHECK-NEXT:    vle32.v v25, (a1)
+; CHECK-NEXT:    fmv.w.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v25, ft0
+; CHECK-NEXT:    vle32.v v25, (a0), v0.t
+; CHECK-NEXT:    vse32.v v25, (a2)
+; CHECK-NEXT:    ret
+  %m = load <1 x float>, <1 x float>* %m_ptr
+  %mask = fcmp oeq <1 x float> %m, zeroinitializer
+  %load = call <1 x float> @llvm.masked.load.v1f32(<1 x float>* %a, i32 8, <1 x i1> %mask, <1 x float> undef)
+  store <1 x float> %load, <1 x float>* %res_ptr
+  ret void
+}
+declare <1 x float> @llvm.masked.load.v1f32(<1 x float>*, i32, <1 x i1>, <1 x float>)
+
+define void @masked_load_v1f64(<1 x double>* %a, <1 x double>* %m_ptr, <1 x double>* %res_ptr) nounwind {
+; RV32-LABEL: masked_load_v1f64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    vsetivli a3, 1, e64,m1,ta,mu
+; RV32-NEXT:    vle64.v v25, (a1)
+; RV32-NEXT:    fcvt.d.w ft0, zero
+; RV32-NEXT:    vmfeq.vf v0, v25, ft0
+; RV32-NEXT:    vle64.v v25, (a0), v0.t
+; RV32-NEXT:    vse64.v v25, (a2)
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: masked_load_v1f64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli a3, 1, e64,m1,ta,mu
+; RV64-NEXT:    vle64.v v25, (a1)
+; RV64-NEXT:    fmv.d.x ft0, zero
+; RV64-NEXT:    vmfeq.vf v0, v25, ft0
+; RV64-NEXT:    vle64.v v25, (a0), v0.t
+; RV64-NEXT:    vse64.v v25, (a2)
+; RV64-NEXT:    ret
+  %m = load <1 x double>, <1 x double>* %m_ptr
+  %mask = fcmp oeq <1 x double> %m, zeroinitializer
+  %load = call <1 x double> @llvm.masked.load.v1f64(<1 x double>* %a, i32 8, <1 x i1> %mask, <1 x double> undef)
+  store <1 x double> %load, <1 x double>* %res_ptr
+  ret void
+}
+declare <1 x double> @llvm.masked.load.v1f64(<1 x double>*, i32, <1 x i1>, <1 x double>)
+
+define void @masked_load_v2f16(<2 x half>* %a, <2 x half>* %m_ptr, <2 x half>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v2f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 2, e16,m1,ta,mu
+; CHECK-NEXT:    vle16.v v25, (a1)
+; CHECK-NEXT:    fmv.h.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v25, ft0
+; CHECK-NEXT:    vle16.v v25, (a0), v0.t
+; CHECK-NEXT:    vse16.v v25, (a2)
+; CHECK-NEXT:    ret
+  %m = load <2 x half>, <2 x half>* %m_ptr
+  %mask = fcmp oeq <2 x half> %m, zeroinitializer
+  %load = call <2 x half> @llvm.masked.load.v2f16(<2 x half>* %a, i32 8, <2 x i1> %mask, <2 x half> undef)
+  store <2 x half> %load, <2 x half>* %res_ptr
+  ret void
+}
+declare <2 x half> @llvm.masked.load.v2f16(<2 x half>*, i32, <2 x i1>, <2 x half>)
+
+define void @masked_load_v2f32(<2 x float>* %a, <2 x float>* %m_ptr, <2 x float>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v2f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 2, e32,m1,ta,mu
+; CHECK-NEXT:    vle32.v v25, (a1)
+; CHECK-NEXT:    fmv.w.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v25, ft0
+; CHECK-NEXT:    vle32.v v25, (a0), v0.t
+; CHECK-NEXT:    vse32.v v25, (a2)
+; CHECK-NEXT:    ret
+  %m = load <2 x float>, <2 x float>* %m_ptr
+  %mask = fcmp oeq <2 x float> %m, zeroinitializer
+  %load = call <2 x float> @llvm.masked.load.v2f32(<2 x float>* %a, i32 8, <2 x i1> %mask, <2 x float> undef)
+  store <2 x float> %load, <2 x float>* %res_ptr
+  ret void
+}
+declare <2 x float> @llvm.masked.load.v2f32(<2 x float>*, i32, <2 x i1>, <2 x float>)
+
+define void @masked_load_v2f64(<2 x double>* %a, <2 x double>* %m_ptr, <2 x double>* %res_ptr) nounwind {
+; RV32-LABEL: masked_load_v2f64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    vsetivli a3, 2, e64,m1,ta,mu
+; RV32-NEXT:    vle64.v v25, (a1)
+; RV32-NEXT:    fcvt.d.w ft0, zero
+; RV32-NEXT:    vmfeq.vf v0, v25, ft0
+; RV32-NEXT:    vle64.v v25, (a0), v0.t
+; RV32-NEXT:    vse64.v v25, (a2)
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: masked_load_v2f64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli a3, 2, e64,m1,ta,mu
+; RV64-NEXT:    vle64.v v25, (a1)
+; RV64-NEXT:    fmv.d.x ft0, zero
+; RV64-NEXT:    vmfeq.vf v0, v25, ft0
+; RV64-NEXT:    vle64.v v25, (a0), v0.t
+; RV64-NEXT:    vse64.v v25, (a2)
+; RV64-NEXT:    ret
+  %m = load <2 x double>, <2 x double>* %m_ptr
+  %mask = fcmp oeq <2 x double> %m, zeroinitializer
+  %load = call <2 x double> @llvm.masked.load.v2f64(<2 x double>* %a, i32 8, <2 x i1> %mask, <2 x double> undef)
+  store <2 x double> %load, <2 x double>* %res_ptr
+  ret void
+}
+declare <2 x double> @llvm.masked.load.v2f64(<2 x double>*, i32, <2 x i1>, <2 x double>)
+
+define void @masked_load_v4f16(<4 x half>* %a, <4 x half>* %m_ptr, <4 x half>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v4f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 4, e16,m1,ta,mu
+; CHECK-NEXT:    vle16.v v25, (a1)
+; CHECK-NEXT:    fmv.h.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v25, ft0
+; CHECK-NEXT:    vle16.v v25, (a0), v0.t
+; CHECK-NEXT:    vse16.v v25, (a2)
+; CHECK-NEXT:    ret
+  %m = load <4 x half>, <4 x half>* %m_ptr
+  %mask = fcmp oeq <4 x half> %m, zeroinitializer
+  %load = call <4 x half> @llvm.masked.load.v4f16(<4 x half>* %a, i32 8, <4 x i1> %mask, <4 x half> undef)
+  store <4 x half> %load, <4 x half>* %res_ptr
+  ret void
+}
+declare <4 x half> @llvm.masked.load.v4f16(<4 x half>*, i32, <4 x i1>, <4 x half>)
+
+define void @masked_load_v4f32(<4 x float>* %a, <4 x float>* %m_ptr, <4 x float>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v4f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 4, e32,m1,ta,mu
+; CHECK-NEXT:    vle32.v v25, (a1)
+; CHECK-NEXT:    fmv.w.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v25, ft0
+; CHECK-NEXT:    vle32.v v25, (a0), v0.t
+; CHECK-NEXT:    vse32.v v25, (a2)
+; CHECK-NEXT:    ret
+  %m = load <4 x float>, <4 x float>* %m_ptr
+  %mask = fcmp oeq <4 x float> %m, zeroinitializer
+  %load = call <4 x float> @llvm.masked.load.v4f32(<4 x float>* %a, i32 8, <4 x i1> %mask, <4 x float> undef)
+  store <4 x float> %load, <4 x float>* %res_ptr
+  ret void
+}
+declare <4 x float> @llvm.masked.load.v4f32(<4 x float>*, i32, <4 x i1>, <4 x float>)
+
+define void @masked_load_v4f64(<4 x double>* %a, <4 x double>* %m_ptr, <4 x double>* %res_ptr) nounwind {
+; RV32-LABEL: masked_load_v4f64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    vsetivli a3, 4, e64,m2,ta,mu
+; RV32-NEXT:    vle64.v v26, (a1)
+; RV32-NEXT:    fcvt.d.w ft0, zero
+; RV32-NEXT:    vmfeq.vf v0, v26, ft0
+; RV32-NEXT:    vle64.v v26, (a0), v0.t
+; RV32-NEXT:    vse64.v v26, (a2)
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: masked_load_v4f64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli a3, 4, e64,m2,ta,mu
+; RV64-NEXT:    vle64.v v26, (a1)
+; RV64-NEXT:    fmv.d.x ft0, zero
+; RV64-NEXT:    vmfeq.vf v0, v26, ft0
+; RV64-NEXT:    vle64.v v26, (a0), v0.t
+; RV64-NEXT:    vse64.v v26, (a2)
+; RV64-NEXT:    ret
+  %m = load <4 x double>, <4 x double>* %m_ptr
+  %mask = fcmp oeq <4 x double> %m, zeroinitializer
+  %load = call <4 x double> @llvm.masked.load.v4f64(<4 x double>* %a, i32 8, <4 x i1> %mask, <4 x double> undef)
+  store <4 x double> %load, <4 x double>* %res_ptr
+  ret void
+}
+declare <4 x double> @llvm.masked.load.v4f64(<4 x double>*, i32, <4 x i1>, <4 x double>)
+
+define void @masked_load_v8f16(<8 x half>* %a, <8 x half>* %m_ptr, <8 x half>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v8f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 8, e16,m1,ta,mu
+; CHECK-NEXT:    vle16.v v25, (a1)
+; CHECK-NEXT:    fmv.h.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v25, ft0
+; CHECK-NEXT:    vle16.v v25, (a0), v0.t
+; CHECK-NEXT:    vse16.v v25, (a2)
+; CHECK-NEXT:    ret
+  %m = load <8 x half>, <8 x half>* %m_ptr
+  %mask = fcmp oeq <8 x half> %m, zeroinitializer
+  %load = call <8 x half> @llvm.masked.load.v8f16(<8 x half>* %a, i32 8, <8 x i1> %mask, <8 x half> undef)
+  store <8 x half> %load, <8 x half>* %res_ptr
+  ret void
+}
+declare <8 x half> @llvm.masked.load.v8f16(<8 x half>*, i32, <8 x i1>, <8 x half>)
+
+define void @masked_load_v8f32(<8 x float>* %a, <8 x float>* %m_ptr, <8 x float>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v8f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 8, e32,m2,ta,mu
+; CHECK-NEXT:    vle32.v v26, (a1)
+; CHECK-NEXT:    fmv.w.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v26, ft0
+; CHECK-NEXT:    vle32.v v26, (a0), v0.t
+; CHECK-NEXT:    vse32.v v26, (a2)
+; CHECK-NEXT:    ret
+  %m = load <8 x float>, <8 x float>* %m_ptr
+  %mask = fcmp oeq <8 x float> %m, zeroinitializer
+  %load = call <8 x float> @llvm.masked.load.v8f32(<8 x float>* %a, i32 8, <8 x i1> %mask, <8 x float> undef)
+  store <8 x float> %load, <8 x float>* %res_ptr
+  ret void
+}
+declare <8 x float> @llvm.masked.load.v8f32(<8 x float>*, i32, <8 x i1>, <8 x float>)
+
+define void @masked_load_v8f64(<8 x double>* %a, <8 x double>* %m_ptr, <8 x double>* %res_ptr) nounwind {
+; RV32-LABEL: masked_load_v8f64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    vsetivli a3, 8, e64,m4,ta,mu
+; RV32-NEXT:    vle64.v v28, (a1)
+; RV32-NEXT:    fcvt.d.w ft0, zero
+; RV32-NEXT:    vmfeq.vf v0, v28, ft0
+; RV32-NEXT:    vle64.v v28, (a0), v0.t
+; RV32-NEXT:    vse64.v v28, (a2)
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: masked_load_v8f64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli a3, 8, e64,m4,ta,mu
+; RV64-NEXT:    vle64.v v28, (a1)
+; RV64-NEXT:    fmv.d.x ft0, zero
+; RV64-NEXT:    vmfeq.vf v0, v28, ft0
+; RV64-NEXT:    vle64.v v28, (a0), v0.t
+; RV64-NEXT:    vse64.v v28, (a2)
+; RV64-NEXT:    ret
+  %m = load <8 x double>, <8 x double>* %m_ptr
+  %mask = fcmp oeq <8 x double> %m, zeroinitializer
+  %load = call <8 x double> @llvm.masked.load.v8f64(<8 x double>* %a, i32 8, <8 x i1> %mask, <8 x double> undef)
+  store <8 x double> %load, <8 x double>* %res_ptr
+  ret void
+}
+declare <8 x double> @llvm.masked.load.v8f64(<8 x double>*, i32, <8 x i1>, <8 x double>)
+
+define void @masked_load_v16f16(<16 x half>* %a, <16 x half>* %m_ptr, <16 x half>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v16f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 16, e16,m2,ta,mu
+; CHECK-NEXT:    vle16.v v26, (a1)
+; CHECK-NEXT:    fmv.h.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v26, ft0
+; CHECK-NEXT:    vle16.v v26, (a0), v0.t
+; CHECK-NEXT:    vse16.v v26, (a2)
+; CHECK-NEXT:    ret
+  %m = load <16 x half>, <16 x half>* %m_ptr
+  %mask = fcmp oeq <16 x half> %m, zeroinitializer
+  %load = call <16 x half> @llvm.masked.load.v16f16(<16 x half>* %a, i32 8, <16 x i1> %mask, <16 x half> undef)
+  store <16 x half> %load, <16 x half>* %res_ptr
+  ret void
+}
+declare <16 x half> @llvm.masked.load.v16f16(<16 x half>*, i32, <16 x i1>, <16 x half>)
+
+define void @masked_load_v16f32(<16 x float>* %a, <16 x float>* %m_ptr, <16 x float>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v16f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 16, e32,m4,ta,mu
+; CHECK-NEXT:    vle32.v v28, (a1)
+; CHECK-NEXT:    fmv.w.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v28, ft0
+; CHECK-NEXT:    vle32.v v28, (a0), v0.t
+; CHECK-NEXT:    vse32.v v28, (a2)
+; CHECK-NEXT:    ret
+  %m = load <16 x float>, <16 x float>* %m_ptr
+  %mask = fcmp oeq <16 x float> %m, zeroinitializer
+  %load = call <16 x float> @llvm.masked.load.v16f32(<16 x float>* %a, i32 8, <16 x i1> %mask, <16 x float> undef)
+  store <16 x float> %load, <16 x float>* %res_ptr
+  ret void
+}
+declare <16 x float> @llvm.masked.load.v16f32(<16 x float>*, i32, <16 x i1>, <16 x float>)
+
+define void @masked_load_v16f64(<16 x double>* %a, <16 x double>* %m_ptr, <16 x double>* %res_ptr) nounwind {
+; RV32-LABEL: masked_load_v16f64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    vsetivli a3, 16, e64,m8,ta,mu
+; RV32-NEXT:    vle64.v v8, (a1)
+; RV32-NEXT:    fcvt.d.w ft0, zero
+; RV32-NEXT:    vmfeq.vf v0, v8, ft0
+; RV32-NEXT:    vle64.v v8, (a0), v0.t
+; RV32-NEXT:    vse64.v v8, (a2)
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: masked_load_v16f64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli a3, 16, e64,m8,ta,mu
+; RV64-NEXT:    vle64.v v8, (a1)
+; RV64-NEXT:    fmv.d.x ft0, zero
+; RV64-NEXT:    vmfeq.vf v0, v8, ft0
+; RV64-NEXT:    vle64.v v8, (a0), v0.t
+; RV64-NEXT:    vse64.v v8, (a2)
+; RV64-NEXT:    ret
+  %m = load <16 x double>, <16 x double>* %m_ptr
+  %mask = fcmp oeq <16 x double> %m, zeroinitializer
+  %load = call <16 x double> @llvm.masked.load.v16f64(<16 x double>* %a, i32 8, <16 x i1> %mask, <16 x double> undef)
+  store <16 x double> %load, <16 x double>* %res_ptr
+  ret void
+}
+declare <16 x double> @llvm.masked.load.v16f64(<16 x double>*, i32, <16 x i1>, <16 x double>)
+
+define void @masked_load_v32f16(<32 x half>* %a, <32 x half>* %m_ptr, <32 x half>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v32f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, zero, 32
+; CHECK-NEXT:    vsetvli a3, a3, e16,m4,ta,mu
+; CHECK-NEXT:    vle16.v v28, (a1)
+; CHECK-NEXT:    fmv.h.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v28, ft0
+; CHECK-NEXT:    vle16.v v28, (a0), v0.t
+; CHECK-NEXT:    vse16.v v28, (a2)
+; CHECK-NEXT:    ret
+  %m = load <32 x half>, <32 x half>* %m_ptr
+  %mask = fcmp oeq <32 x half> %m, zeroinitializer
+  %load = call <32 x half> @llvm.masked.load.v32f16(<32 x half>* %a, i32 8, <32 x i1> %mask, <32 x half> undef)
+  store <32 x half> %load, <32 x half>* %res_ptr
+  ret void
+}
+declare <32 x half> @llvm.masked.load.v32f16(<32 x half>*, i32, <32 x i1>, <32 x half>)
+
+define void @masked_load_v32f32(<32 x float>* %a, <32 x float>* %m_ptr, <32 x float>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v32f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, zero, 32
+; CHECK-NEXT:    vsetvli a3, a3, e32,m8,ta,mu
+; CHECK-NEXT:    vle32.v v8, (a1)
+; CHECK-NEXT:    fmv.w.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v8, ft0
+; CHECK-NEXT:    vle32.v v8, (a0), v0.t
+; CHECK-NEXT:    vse32.v v8, (a2)
+; CHECK-NEXT:    ret
+  %m = load <32 x float>, <32 x float>* %m_ptr
+  %mask = fcmp oeq <32 x float> %m, zeroinitializer
+  %load = call <32 x float> @llvm.masked.load.v32f32(<32 x float>* %a, i32 8, <32 x i1> %mask, <32 x float> undef)
+  store <32 x float> %load, <32 x float>* %res_ptr
+  ret void
+}
+declare <32 x float> @llvm.masked.load.v32f32(<32 x float>*, i32, <32 x i1>, <32 x float>)
+
+define void @masked_load_v32f64(<32 x double>* %a, <32 x double>* %m_ptr, <32 x double>* %res_ptr) nounwind {
+; RV32-LABEL: masked_load_v32f64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    addi a3, a1, 128
+; RV32-NEXT:    vsetivli a4, 16, e64,m8,ta,mu
+; RV32-NEXT:    vle64.v v8, (a1)
+; RV32-NEXT:    vle64.v v16, (a3)
+; RV32-NEXT:    fcvt.d.w ft0, zero
+; RV32-NEXT:    vmfeq.vf v0, v8, ft0
+; RV32-NEXT:    vmfeq.vf v26, v16, ft0
+; RV32-NEXT:    vle64.v v8, (a0), v0.t
+; RV32-NEXT:    addi a0, a0, 128
+; RV32-NEXT:    vmv1r.v v0, v26
+; RV32-NEXT:    vle64.v v16, (a0), v0.t
+; RV32-NEXT:    vse64.v v8, (a2)
+; RV32-NEXT:    addi a0, a2, 128
+; RV32-NEXT:    vse64.v v16, (a0)
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: masked_load_v32f64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    addi a3, a1, 128
+; RV64-NEXT:    vsetivli a4, 16, e64,m8,ta,mu
+; RV64-NEXT:    vle64.v v8, (a1)
+; RV64-NEXT:    vle64.v v16, (a3)
+; RV64-NEXT:    fmv.d.x ft0, zero
+; RV64-NEXT:    vmfeq.vf v0, v8, ft0
+; RV64-NEXT:    vmfeq.vf v26, v16, ft0
+; RV64-NEXT:    vle64.v v8, (a0), v0.t
+; RV64-NEXT:    addi a0, a0, 128
+; RV64-NEXT:    vmv1r.v v0, v26
+; RV64-NEXT:    vle64.v v16, (a0), v0.t
+; RV64-NEXT:    vse64.v v8, (a2)
+; RV64-NEXT:    addi a0, a2, 128
+; RV64-NEXT:    vse64.v v16, (a0)
+; RV64-NEXT:    ret
+  %m = load <32 x double>, <32 x double>* %m_ptr
+  %mask = fcmp oeq <32 x double> %m, zeroinitializer
+  %load = call <32 x double> @llvm.masked.load.v32f64(<32 x double>* %a, i32 8, <32 x i1> %mask, <32 x double> undef)
+  store <32 x double> %load, <32 x double>* %res_ptr
+  ret void
+}
+declare <32 x double> @llvm.masked.load.v32f64(<32 x double>*, i32, <32 x i1>, <32 x double>)
+
+define void @masked_load_v64f16(<64 x half>* %a, <64 x half>* %m_ptr, <64 x half>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v64f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, zero, 64
+; CHECK-NEXT:    vsetvli a3, a3, e16,m8,ta,mu
+; CHECK-NEXT:    vle16.v v8, (a1)
+; CHECK-NEXT:    fmv.h.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v8, ft0
+; CHECK-NEXT:    vle16.v v8, (a0), v0.t
+; CHECK-NEXT:    vse16.v v8, (a2)
+; CHECK-NEXT:    ret
+  %m = load <64 x half>, <64 x half>* %m_ptr
+  %mask = fcmp oeq <64 x half> %m, zeroinitializer
+  %load = call <64 x half> @llvm.masked.load.v64f16(<64 x half>* %a, i32 8, <64 x i1> %mask, <64 x half> undef)
+  store <64 x half> %load, <64 x half>* %res_ptr
+  ret void
+}
+declare <64 x half> @llvm.masked.load.v64f16(<64 x half>*, i32, <64 x i1>, <64 x half>)
+
+define void @masked_load_v64f32(<64 x float>* %a, <64 x float>* %m_ptr, <64 x float>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v64f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, a1, 128
+; CHECK-NEXT:    addi a4, zero, 32
+; CHECK-NEXT:    vsetvli a4, a4, e32,m8,ta,mu
+; CHECK-NEXT:    vle32.v v8, (a1)
+; CHECK-NEXT:    vle32.v v16, (a3)
+; CHECK-NEXT:    fmv.w.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v8, ft0
+; CHECK-NEXT:    vmfeq.vf v26, v16, ft0
+; CHECK-NEXT:    vle32.v v8, (a0), v0.t
+; CHECK-NEXT:    addi a0, a0, 128
+; CHECK-NEXT:    vmv1r.v v0, v26
+; CHECK-NEXT:    vle32.v v16, (a0), v0.t
+; CHECK-NEXT:    vse32.v v8, (a2)
+; CHECK-NEXT:    addi a0, a2, 128
+; CHECK-NEXT:    vse32.v v16, (a0)
+; CHECK-NEXT:    ret
+  %m = load <64 x float>, <64 x float>* %m_ptr
+  %mask = fcmp oeq <64 x float> %m, zeroinitializer
+  %load = call <64 x float> @llvm.masked.load.v64f32(<64 x float>* %a, i32 8, <64 x i1> %mask, <64 x float> undef)
+  store <64 x float> %load, <64 x float>* %res_ptr
+  ret void
+}
+declare <64 x float> @llvm.masked.load.v64f32(<64 x float>*, i32, <64 x i1>, <64 x float>)
+
+define void @masked_load_v128f16(<128 x half>* %a, <128 x half>* %m_ptr, <128 x half>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v128f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, a1, 128
+; CHECK-NEXT:    addi a4, zero, 64
+; CHECK-NEXT:    vsetvli a4, a4, e16,m8,ta,mu
+; CHECK-NEXT:    vle16.v v8, (a1)
+; CHECK-NEXT:    vle16.v v16, (a3)
+; CHECK-NEXT:    fmv.h.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v8, ft0
+; CHECK-NEXT:    vmfeq.vf v26, v16, ft0
+; CHECK-NEXT:    vle16.v v8, (a0), v0.t
+; CHECK-NEXT:    addi a0, a0, 128
+; CHECK-NEXT:    vmv1r.v v0, v26
+; CHECK-NEXT:    vle16.v v16, (a0), v0.t
+; CHECK-NEXT:    vse16.v v8, (a2)
+; CHECK-NEXT:    addi a0, a2, 128
+; CHECK-NEXT:    vse16.v v16, (a0)
+; CHECK-NEXT:    ret
+  %m = load <128 x half>, <128 x half>* %m_ptr
+  %mask = fcmp oeq <128 x half> %m, zeroinitializer
+  %load = call <128 x half> @llvm.masked.load.v128f16(<128 x half>* %a, i32 8, <128 x i1> %mask, <128 x half> undef)
+  store <128 x half> %load, <128 x half>* %res_ptr
+  ret void
+}
+declare <128 x half> @llvm.masked.load.v128f16(<128 x half>*, i32, <128 x i1>, <128 x half>)

llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-load-int.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -mattr=+m,+experimental-v -verify-machineinstrs -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV32
+; RUN: llc -mtriple=riscv64 -mattr=+m,+experimental-v -verify-machineinstrs -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV64
+
+define void @masked_load_v1i8(<1 x i8>* %a, <1 x i8>* %m_ptr, <1 x i8>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v1i8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 1, e8,m1,ta,mu
+; CHECK-NEXT:    vle8.v v25, (a1)
+; CHECK-NEXT:    vmseq.vi v0, v25, 0
+; CHECK-NEXT:    vle8.v v25, (a0), v0.t
+; CHECK-NEXT:    vse8.v v25, (a2)
+; CHECK-NEXT:    ret
+  %m = load <1 x i8>, <1 x i8>* %m_ptr
+  %mask = icmp eq <1 x i8> %m, zeroinitializer
+  %load = call <1 x i8> @llvm.masked.load.v1i8(<1 x i8>* %a, i32 8, <1 x i1> %mask, <1 x i8> undef)
+  store <1 x i8> %load, <1 x i8>* %res_ptr
+  ret void
+}
+declare <1 x i8> @llvm.masked.load.v1i8(<1 x i8>*, i32, <1 x i1>, <1 x i8>)
+
+define void @masked_load_v1i16(<1 x i16>* %a, <1 x i16>* %m_ptr, <1 x i16>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v1i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 1, e16,m1,ta,mu
+; CHECK-NEXT:    vle16.v v25, (a1)
+; CHECK-NEXT:    vmseq.vi v0, v25, 0
+; CHECK-NEXT:    vle16.v v25, (a0), v0.t
+; CHECK-NEXT:    vse16.v v25, (a2)
+; CHECK-NEXT:    ret
+  %m = load <1 x i16>, <1 x i16>* %m_ptr
+  %mask = icmp eq <1 x i16> %m, zeroinitializer
+  %load = call <1 x i16> @llvm.masked.load.v1i16(<1 x i16>* %a, i32 8, <1 x i1> %mask, <1 x i16> undef)
+  store <1 x i16> %load, <1 x i16>* %res_ptr
+  ret void
+}
+declare <1 x i16> @llvm.masked.load.v1i16(<1 x i16>*, i32, <1 x i1>, <1 x i16>)
+
+define void @masked_load_v1i32(<1 x i32>* %a, <1 x i32>* %m_ptr, <1 x i32>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v1i32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 1, e32,m1,ta,mu
+; CHECK-NEXT:    vle32.v v25, (a1)
+; CHECK-NEXT:    vmseq.vi v0, v25, 0
+; CHECK-NEXT:    vle32.v v25, (a0), v0.t
+; CHECK-NEXT:    vse32.v v25, (a2)
+; CHECK-NEXT:    ret
+  %m = load <1 x i32>, <1 x i32>* %m_ptr
+  %mask = icmp eq <1 x i32> %m, zeroinitializer
+  %load = call <1 x i32> @llvm.masked.load.v1i32(<1 x i32>* %a, i32 8, <1 x i1> %mask, <1 x i32> undef)
+  store <1 x i32> %load, <1 x i32>* %res_ptr
+  ret void
+}
+declare <1 x i32> @llvm.masked.load.v1i32(<1 x i32>*, i32, <1 x i1>, <1 x i32>)
+
+define void @masked_load_v1i64(<1 x i64>* %a, <1 x i64>* %m_ptr, <1 x i64>* %res_ptr) nounwind {
+; RV32-LABEL: masked_load_v1i64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    vsetivli a3, 1, e64,m1,ta,mu
+; RV32-NEXT:    vle64.v v25, (a1)
+; RV32-NEXT:    vsetivli a1, 2, e32,m1,ta,mu
+; RV32-NEXT:    vmv.v.i v26, 0
+; RV32-NEXT:    vsetivli a1, 1, e64,m1,ta,mu
+; RV32-NEXT:    vmseq.vv v0, v25, v26
+; RV32-NEXT:    vle64.v v25, (a0), v0.t
+; RV32-NEXT:    vse64.v v25, (a2)
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: masked_load_v1i64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli a3, 1, e64,m1,ta,mu
+; RV64-NEXT:    vle64.v v25, (a1)
+; RV64-NEXT:    vmseq.vi v0, v25, 0
+; RV64-NEXT:    vle64.v v25, (a0), v0.t
+; RV64-NEXT:    vse64.v v25, (a2)
+; RV64-NEXT:    ret
+  %m = load <1 x i64>, <1 x i64>* %m_ptr
+  %mask = icmp eq <1 x i64> %m, zeroinitializer
+  %load = call <1 x i64> @llvm.masked.load.v1i64(<1 x i64>* %a, i32 8, <1 x i1> %mask, <1 x i64> undef)
+  store <1 x i64> %load, <1 x i64>* %res_ptr
+  ret void
+}
+declare <1 x i64> @llvm.masked.load.v1i64(<1 x i64>*, i32, <1 x i1>, <1 x i64>)
+
+define void @masked_load_v2i8(<2 x i8>* %a, <2 x i8>* %m_ptr, <2 x i8>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v2i8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 2, e8,m1,ta,mu
+; CHECK-NEXT:    vle8.v v25, (a1)
+; CHECK-NEXT:    vmseq.vi v0, v25, 0
+; CHECK-NEXT:    vle8.v v25, (a0), v0.t
+; CHECK-NEXT:    vse8.v v25, (a2)
+; CHECK-NEXT:    ret
+  %m = load <2 x i8>, <2 x i8>* %m_ptr
+  %mask = icmp eq <2 x i8> %m, zeroinitializer
+  %load = call <2 x i8> @llvm.masked.load.v2i8(<2 x i8>* %a, i32 8, <2 x i1> %mask, <2 x i8> undef)
+  store <2 x i8> %load, <2 x i8>* %res_ptr
+  ret void
+}
+declare <2 x i8> @llvm.masked.load.v2i8(<2 x i8>*, i32, <2 x i1>, <2 x i8>)
+
+define void @masked_load_v2i16(<2 x i16>* %a, <2 x i16>* %m_ptr, <2 x i16>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v2i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 2, e16,m1,ta,mu
+; CHECK-NEXT:    vle16.v v25, (a1)
+; CHECK-NEXT:    vmseq.vi v0, v25, 0
+; CHECK-NEXT:    vle16.v v25, (a0), v0.t
+; CHECK-NEXT:    vse16.v v25, (a2)
+; CHECK-NEXT:    ret
+  %m = load <2 x i16>, <2 x i16>* %m_ptr
+  %mask = icmp eq <2 x i16> %m, zeroinitializer
+  %load = call <2 x i16> @llvm.masked.load.v2i16(<2 x i16>* %a, i32 8, <2 x i1> %mask, <2 x i16> undef)
+  store <2 x i16> %load, <2 x i16>* %res_ptr
+  ret void
+}
+declare <2 x i16> @llvm.masked.load.v2i16(<2 x i16>*, i32, <2 x i1>, <2 x i16>)
+
+define void @masked_load_v2i32(<2 x i32>* %a, <2 x i32>* %m_ptr, <2 x i32>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v2i32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 2, e32,m1,ta,mu
+; CHECK-NEXT:    vle32.v v25, (a1)
+; CHECK-NEXT:    vmseq.vi v0, v25, 0
+; CHECK-NEXT:    vle32.v v25, (a0), v0.t
+; CHECK-NEXT:    vse32.v v25, (a2)
+; CHECK-NEXT:    ret
+  %m = load <2 x i32>, <2 x i32>* %m_ptr
+  %mask = icmp eq <2 x i32> %m, zeroinitializer
+  %load = call <2 x i32> @llvm.masked.load.v2i32(<2 x i32>* %a, i32 8, <2 x i1> %mask, <2 x i32> undef)
+  store <2 x i32> %load, <2 x i32>* %res_ptr
+  ret void
+}
+declare <2 x i32> @llvm.masked.load.v2i32(<2 x i32>*, i32, <2 x i1>, <2 x i32>)
+
+define void @masked_load_v2i64(<2 x i64>* %a, <2 x i64>* %m_ptr, <2 x i64>* %res_ptr) nounwind {
+; RV32-LABEL: masked_load_v2i64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    vsetivli a3, 2, e64,m1,ta,mu
+; RV32-NEXT:    vle64.v v25, (a1)
+; RV32-NEXT:    vsetivli a1, 4, e32,m1,ta,mu
+; RV32-NEXT:    vmv.v.i v26, 0
+; RV32-NEXT:    vsetivli a1, 2, e64,m1,ta,mu
+; RV32-NEXT:    vmseq.vv v0, v25, v26
+; RV32-NEXT:    vle64.v v25, (a0), v0.t
+; RV32-NEXT:    vse64.v v25, (a2)
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: masked_load_v2i64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli a3, 2, e64,m1,ta,mu
+; RV64-NEXT:    vle64.v v25, (a1)
+; RV64-NEXT:    vmseq.vi v0, v25, 0
+; RV64-NEXT:    vle64.v v25, (a0), v0.t
+; RV64-NEXT:    vse64.v v25, (a2)
+; RV64-NEXT:    ret
+  %m = load <2 x i64>, <2 x i64>* %m_ptr
+  %mask = icmp eq <2 x i64> %m, zeroinitializer
+  %load = call <2 x i64> @llvm.masked.load.v2i64(<2 x i64>* %a, i32 8, <2 x i1> %mask, <2 x i64> undef)
+  store <2 x i64> %load, <2 x i64>* %res_ptr
+  ret void
+}
+declare <2 x i64> @llvm.masked.load.v2i64(<2 x i64>*, i32, <2 x i1>, <2 x i64>)
+
+define void @masked_load_v4i8(<4 x i8>* %a, <4 x i8>* %m_ptr, <4 x i8>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v4i8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 4, e8,m1,ta,mu
+; CHECK-NEXT:    vle8.v v25, (a1)
+; CHECK-NEXT:    vmseq.vi v0, v25, 0
+; CHECK-NEXT:    vle8.v v25, (a0), v0.t
+; CHECK-NEXT:    vse8.v v25, (a2)
+; CHECK-NEXT:    ret
+  %m = load <4 x i8>, <4 x i8>* %m_ptr
+  %mask = icmp eq <4 x i8> %m, zeroinitializer
+  %load = call <4 x i8> @llvm.masked.load.v4i8(<4 x i8>* %a, i32 8, <4 x i1> %mask, <4 x i8> undef)
+  store <4 x i8> %load, <4 x i8>* %res_ptr
+  ret void
+}
+declare <4 x i8> @llvm.masked.load.v4i8(<4 x i8>*, i32, <4 x i1>, <4 x i8>)
+
+define void @masked_load_v4i16(<4 x i16>* %a, <4 x i16>* %m_ptr, <4 x i16>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v4i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 4, e16,m1,ta,mu
+; CHECK-NEXT:    vle16.v v25, (a1)
+; CHECK-NEXT:    vmseq.vi v0, v25, 0
+; CHECK-NEXT:    vle16.v v25, (a0), v0.t
+; CHECK-NEXT:    vse16.v v25, (a2)
+; CHECK-NEXT:    ret
+  %m = load <4 x i16>, <4 x i16>* %m_ptr
+  %mask = icmp eq <4 x i16> %m, zeroinitializer
+  %load = call <4 x i16> @llvm.masked.load.v4i16(<4 x i16>* %a, i32 8, <4 x i1> %mask, <4 x i16> undef)
+  store <4 x i16> %load, <4 x i16>* %res_ptr
+  ret void
+}
+declare <4 x i16> @llvm.masked.load.v4i16(<4 x i16>*, i32, <4 x i1>, <4 x i16>)
+
+define void @masked_load_v4i32(<4 x i32>* %a, <4 x i32>* %m_ptr, <4 x i32>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v4i32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 4, e32,m1,ta,mu
+; CHECK-NEXT:    vle32.v v25, (a1)
+; CHECK-NEXT:    vmseq.vi v0, v25, 0
+; CHECK-NEXT:    vle32.v v25, (a0), v0.t
+; CHECK-NEXT:    vse32.v v25, (a2)
+; CHECK-NEXT:    ret
+  %m = load <4 x i32>, <4 x i32>* %m_ptr
+  %mask = icmp eq <4 x i32> %m, zeroinitializer
+  %load = call <4 x i32> @llvm.masked.load.v4i32(<4 x i32>* %a, i32 8, <4 x i1> %mask, <4 x i32> undef)
+  store <4 x i32> %load, <4 x i32>* %res_ptr
+  ret void
+}
+declare <4 x i32> @llvm.masked.load.v4i32(<4 x i32>*, i32, <4 x i1>, <4 x i32>)
+
+define void @masked_load_v4i64(<4 x i64>* %a, <4 x i64>* %m_ptr, <4 x i64>* %res_ptr) nounwind {
+; RV32-LABEL: masked_load_v4i64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    vsetivli a3, 4, e64,m2,ta,mu
+; RV32-NEXT:    vle64.v v26, (a1)
+; RV32-NEXT:    vsetivli a1, 8, e32,m2,ta,mu
+; RV32-NEXT:    vmv.v.i v28, 0
+; RV32-NEXT:    vsetivli a1, 4, e64,m2,ta,mu
+; RV32-NEXT:    vmseq.vv v0, v26, v28
+; RV32-NEXT:    vle64.v v26, (a0), v0.t
+; RV32-NEXT:    vse64.v v26, (a2)
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: masked_load_v4i64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli a3, 4, e64,m2,ta,mu
+; RV64-NEXT:    vle64.v v26, (a1)
+; RV64-NEXT:    vmseq.vi v0, v26, 0
+; RV64-NEXT:    vle64.v v26, (a0), v0.t
+; RV64-NEXT:    vse64.v v26, (a2)
+; RV64-NEXT:    ret
+  %m = load <4 x i64>, <4 x i64>* %m_ptr
+  %mask = icmp eq <4 x i64> %m, zeroinitializer
+  %load = call <4 x i64> @llvm.masked.load.v4i64(<4 x i64>* %a, i32 8, <4 x i1> %mask, <4 x i64> undef)
+  store <4 x i64> %load, <4 x i64>* %res_ptr
+  ret void
+}
+declare <4 x i64> @llvm.masked.load.v4i64(<4 x i64>*, i32, <4 x i1>, <4 x i64>)
+
+define void @masked_load_v8i8(<8 x i8>* %a, <8 x i8>* %m_ptr, <8 x i8>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v8i8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 8, e8,m1,ta,mu
+; CHECK-NEXT:    vle8.v v25, (a1)
+; CHECK-NEXT:    vmseq.vi v0, v25, 0
+; CHECK-NEXT:    vle8.v v25, (a0), v0.t
+; CHECK-NEXT:    vse8.v v25, (a2)
+; CHECK-NEXT:    ret
+  %m = load <8 x i8>, <8 x i8>* %m_ptr
+  %mask = icmp eq <8 x i8> %m, zeroinitializer
+  %load = call <8 x i8> @llvm.masked.load.v8i8(<8 x i8>* %a, i32 8, <8 x i1> %mask, <8 x i8> undef)
+  store <8 x i8> %load, <8 x i8>* %res_ptr
+  ret void
+}
+declare <8 x i8> @llvm.masked.load.v8i8(<8 x i8>*, i32, <8 x i1>, <8 x i8>)
+
+define void @masked_load_v8i16(<8 x i16>* %a, <8 x i16>* %m_ptr, <8 x i16>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v8i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 8, e16,m1,ta,mu
+; CHECK-NEXT:    vle16.v v25, (a1)
+; CHECK-NEXT:    vmseq.vi v0, v25, 0
+; CHECK-NEXT:    vle16.v v25, (a0), v0.t
+; CHECK-NEXT:    vse16.v v25, (a2)
+; CHECK-NEXT:    ret
+  %m = load <8 x i16>, <8 x i16>* %m_ptr
+  %mask = icmp eq <8 x i16> %m, zeroinitializer
+  %load = call <8 x i16> @llvm.masked.load.v8i16(<8 x i16>* %a, i32 8, <8 x i1> %mask, <8 x i16> undef)
+  store <8 x i16> %load, <8 x i16>* %res_ptr
+  ret void
+}
+declare <8 x i16> @llvm.masked.load.v8i16(<8 x i16>*, i32, <8 x i1>, <8 x i16>)
+
+define void @masked_load_v8i32(<8 x i32>* %a, <8 x i32>* %m_ptr, <8 x i32>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v8i32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 8, e32,m2,ta,mu
+; CHECK-NEXT:    vle32.v v26, (a1)
+; CHECK-NEXT:    vmseq.vi v0, v26, 0
+; CHECK-NEXT:    vle32.v v26, (a0), v0.t
+; CHECK-NEXT:    vse32.v v26, (a2)
+; CHECK-NEXT:    ret
+  %m = load <8 x i32>, <8 x i32>* %m_ptr
+  %mask = icmp eq <8 x i32> %m, zeroinitializer
+  %load = call <8 x i32> @llvm.masked.load.v8i32(<8 x i32>* %a, i32 8, <8 x i1> %mask, <8 x i32> undef)
+  store <8 x i32> %load, <8 x i32>* %res_ptr
+  ret void
+}
+declare <8 x i32> @llvm.masked.load.v8i32(<8 x i32>*, i32, <8 x i1>, <8 x i32>)
+
+define void @masked_load_v8i64(<8 x i64>* %a, <8 x i64>* %m_ptr, <8 x i64>* %res_ptr) nounwind {
+; RV32-LABEL: masked_load_v8i64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    vsetivli a3, 8, e64,m4,ta,mu
+; RV32-NEXT:    vle64.v v28, (a1)
+; RV32-NEXT:    vsetivli a1, 16, e32,m4,ta,mu
+; RV32-NEXT:    vmv.v.i v8, 0
+; RV32-NEXT:    vsetivli a1, 8, e64,m4,ta,mu
+; RV32-NEXT:    vmseq.vv v0, v28, v8
+; RV32-NEXT:    vle64.v v28, (a0), v0.t
+; RV32-NEXT:    vse64.v v28, (a2)
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: masked_load_v8i64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli a3, 8, e64,m4,ta,mu
+; RV64-NEXT:    vle64.v v28, (a1)
+; RV64-NEXT:    vmseq.vi v0, v28, 0
+; RV64-NEXT:    vle64.v v28, (a0), v0.t
+; RV64-NEXT:    vse64.v v28, (a2)
+; RV64-NEXT:    ret
+  %m = load <8 x i64>, <8 x i64>* %m_ptr
+  %mask = icmp eq <8 x i64> %m, zeroinitializer
+  %load = call <8 x i64> @llvm.masked.load.v8i64(<8 x i64>* %a, i32 8, <8 x i1> %mask, <8 x i64> undef)
+  store <8 x i64> %load, <8 x i64>* %res_ptr
+  ret void
+}
+declare <8 x i64> @llvm.masked.load.v8i64(<8 x i64>*, i32, <8 x i1>, <8 x i64>)
+
+define void @masked_load_v16i8(<16 x i8>* %a, <16 x i8>* %m_ptr, <16 x i8>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v16i8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 16, e8,m1,ta,mu
+; CHECK-NEXT:    vle8.v v25, (a1)
+; CHECK-NEXT:    vmseq.vi v0, v25, 0
+; CHECK-NEXT:    vle8.v v25, (a0), v0.t
+; CHECK-NEXT:    vse8.v v25, (a2)
+; CHECK-NEXT:    ret
+  %m = load <16 x i8>, <16 x i8>* %m_ptr
+  %mask = icmp eq <16 x i8> %m, zeroinitializer
+  %load = call <16 x i8> @llvm.masked.load.v16i8(<16 x i8>* %a, i32 8, <16 x i1> %mask, <16 x i8> undef)
+  store <16 x i8> %load, <16 x i8>* %res_ptr
+  ret void
+}
+declare <16 x i8> @llvm.masked.load.v16i8(<16 x i8>*, i32, <16 x i1>, <16 x i8>)
+
+define void @masked_load_v16i16(<16 x i16>* %a, <16 x i16>* %m_ptr, <16 x i16>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v16i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 16, e16,m2,ta,mu
+; CHECK-NEXT:    vle16.v v26, (a1)
+; CHECK-NEXT:    vmseq.vi v0, v26, 0
+; CHECK-NEXT:    vle16.v v26, (a0), v0.t
+; CHECK-NEXT:    vse16.v v26, (a2)
+; CHECK-NEXT:    ret
+  %m = load <16 x i16>, <16 x i16>* %m_ptr
+  %mask = icmp eq <16 x i16> %m, zeroinitializer
+  %load = call <16 x i16> @llvm.masked.load.v16i16(<16 x i16>* %a, i32 8, <16 x i1> %mask, <16 x i16> undef)
+  store <16 x i16> %load, <16 x i16>* %res_ptr
+  ret void
+}
+declare <16 x i16> @llvm.masked.load.v16i16(<16 x i16>*, i32, <16 x i1>, <16 x i16>)
+
+define void @masked_load_v16i32(<16 x i32>* %a, <16 x i32>* %m_ptr, <16 x i32>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v16i32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 16, e32,m4,ta,mu
+; CHECK-NEXT:    vle32.v v28, (a1)
+; CHECK-NEXT:    vmseq.vi v0, v28, 0
+; CHECK-NEXT:    vle32.v v28, (a0), v0.t
+; CHECK-NEXT:    vse32.v v28, (a2)
+; CHECK-NEXT:    ret
+  %m = load <16 x i32>, <16 x i32>* %m_ptr
+  %mask = icmp eq <16 x i32> %m, zeroinitializer
+  %load = call <16 x i32> @llvm.masked.load.v16i32(<16 x i32>* %a, i32 8, <16 x i1> %mask, <16 x i32> undef)
+  store <16 x i32> %load, <16 x i32>* %res_ptr
+  ret void
+}
+declare <16 x i32> @llvm.masked.load.v16i32(<16 x i32>*, i32, <16 x i1>, <16 x i32>)
+
+define void @masked_load_v16i64(<16 x i64>* %a, <16 x i64>* %m_ptr, <16 x i64>* %res_ptr) nounwind {
+; RV32-LABEL: masked_load_v16i64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    vsetivli a3, 16, e64,m8,ta,mu
+; RV32-NEXT:    vle64.v v8, (a1)
+; RV32-NEXT:    addi a1, zero, 32
+; RV32-NEXT:    vsetvli a1, a1, e32,m8,ta,mu
+; RV32-NEXT:    vmv.v.i v16, 0
+; RV32-NEXT:    vsetivli a1, 16, e64,m8,ta,mu
+; RV32-NEXT:    vmseq.vv v0, v8, v16
+; RV32-NEXT:    vle64.v v8, (a0), v0.t
+; RV32-NEXT:    vse64.v v8, (a2)
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: masked_load_v16i64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli a3, 16, e64,m8,ta,mu
+; RV64-NEXT:    vle64.v v8, (a1)
+; RV64-NEXT:    vmseq.vi v0, v8, 0
+; RV64-NEXT:    vle64.v v8, (a0), v0.t
+; RV64-NEXT:    vse64.v v8, (a2)
+; RV64-NEXT:    ret
+  %m = load <16 x i64>, <16 x i64>* %m_ptr
+  %mask = icmp eq <16 x i64> %m, zeroinitializer
+  %load = call <16 x i64> @llvm.masked.load.v16i64(<16 x i64>* %a, i32 8, <16 x i1> %mask, <16 x i64> undef)
+  store <16 x i64> %load, <16 x i64>* %res_ptr
+  ret void
+}
+declare <16 x i64> @llvm.masked.load.v16i64(<16 x i64>*, i32, <16 x i1>, <16 x i64>)
+
+define void @masked_load_v32i8(<32 x i8>* %a, <32 x i8>* %m_ptr, <32 x i8>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v32i8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, zero, 32
+; CHECK-NEXT:    vsetvli a3, a3, e8,m2,ta,mu
+; CHECK-NEXT:    vle8.v v26, (a1)
+; CHECK-NEXT:    vmseq.vi v0, v26, 0
+; CHECK-NEXT:    vle8.v v26, (a0), v0.t
+; CHECK-NEXT:    vse8.v v26, (a2)
+; CHECK-NEXT:    ret
+  %m = load <32 x i8>, <32 x i8>* %m_ptr
+  %mask = icmp eq <32 x i8> %m, zeroinitializer
+  %load = call <32 x i8> @llvm.masked.load.v32i8(<32 x i8>* %a, i32 8, <32 x i1> %mask, <32 x i8> undef)
+  store <32 x i8> %load, <32 x i8>* %res_ptr
+  ret void
+}
+declare <32 x i8> @llvm.masked.load.v32i8(<32 x i8>*, i32, <32 x i1>, <32 x i8>)
+
+define void @masked_load_v32i16(<32 x i16>* %a, <32 x i16>* %m_ptr, <32 x i16>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v32i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, zero, 32
+; CHECK-NEXT:    vsetvli a3, a3, e16,m4,ta,mu
+; CHECK-NEXT:    vle16.v v28, (a1)
+; CHECK-NEXT:    vmseq.vi v0, v28, 0
+; CHECK-NEXT:    vle16.v v28, (a0), v0.t
+; CHECK-NEXT:    vse16.v v28, (a2)
+; CHECK-NEXT:    ret
+  %m = load <32 x i16>, <32 x i16>* %m_ptr
+  %mask = icmp eq <32 x i16> %m, zeroinitializer
+  %load = call <32 x i16> @llvm.masked.load.v32i16(<32 x i16>* %a, i32 8, <32 x i1> %mask, <32 x i16> undef)
+  store <32 x i16> %load, <32 x i16>* %res_ptr
+  ret void
+}
+declare <32 x i16> @llvm.masked.load.v32i16(<32 x i16>*, i32, <32 x i1>, <32 x i16>)
+
+define void @masked_load_v32i32(<32 x i32>* %a, <32 x i32>* %m_ptr, <32 x i32>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v32i32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, zero, 32
+; CHECK-NEXT:    vsetvli a3, a3, e32,m8,ta,mu
+; CHECK-NEXT:    vle32.v v8, (a1)
+; CHECK-NEXT:    vmseq.vi v0, v8, 0
+; CHECK-NEXT:    vle32.v v8, (a0), v0.t
+; CHECK-NEXT:    vse32.v v8, (a2)
+; CHECK-NEXT:    ret
+  %m = load <32 x i32>, <32 x i32>* %m_ptr
+  %mask = icmp eq <32 x i32> %m, zeroinitializer
+  %load = call <32 x i32> @llvm.masked.load.v32i32(<32 x i32>* %a, i32 8, <32 x i1> %mask, <32 x i32> undef)
+  store <32 x i32> %load, <32 x i32>* %res_ptr
+  ret void
+}
+declare <32 x i32> @llvm.masked.load.v32i32(<32 x i32>*, i32, <32 x i1>, <32 x i32>)
+
+define void @masked_load_v32i64(<32 x i64>* %a, <32 x i64>* %m_ptr, <32 x i64>* %res_ptr) nounwind {
+; RV32-LABEL: masked_load_v32i64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    addi a3, a1, 128
+; RV32-NEXT:    vsetivli a4, 16, e64,m8,ta,mu
+; RV32-NEXT:    vle64.v v8, (a3)
+; RV32-NEXT:    vle64.v v16, (a1)
+; RV32-NEXT:    addi a1, zero, 32
+; RV32-NEXT:    vsetvli a1, a1, e32,m8,ta,mu
+; RV32-NEXT:    vmv.v.i v24, 0
+; RV32-NEXT:    vsetivli a1, 16, e64,m8,ta,mu
+; RV32-NEXT:    vmseq.vv v0, v16, v24
+; RV32-NEXT:    vmseq.vv v16, v8, v24
+; RV32-NEXT:    vle64.v v8, (a0), v0.t
+; RV32-NEXT:    addi a0, a0, 128
+; RV32-NEXT:    vmv1r.v v0, v16
+; RV32-NEXT:    vle64.v v16, (a0), v0.t
+; RV32-NEXT:    vse64.v v8, (a2)
+; RV32-NEXT:    addi a0, a2, 128
+; RV32-NEXT:    vse64.v v16, (a0)
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: masked_load_v32i64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    addi a3, a1, 128
+; RV64-NEXT:    vsetivli a4, 16, e64,m8,ta,mu
+; RV64-NEXT:    vle64.v v8, (a1)
+; RV64-NEXT:    vle64.v v16, (a3)
+; RV64-NEXT:    vmseq.vi v0, v8, 0
+; RV64-NEXT:    vmseq.vi v26, v16, 0
+; RV64-NEXT:    vle64.v v8, (a0), v0.t
+; RV64-NEXT:    addi a0, a0, 128
+; RV64-NEXT:    vmv1r.v v0, v26
+; RV64-NEXT:    vle64.v v16, (a0), v0.t
+; RV64-NEXT:    vse64.v v8, (a2)
+; RV64-NEXT:    addi a0, a2, 128
+; RV64-NEXT:    vse64.v v16, (a0)
+; RV64-NEXT:    ret
+  %m = load <32 x i64>, <32 x i64>* %m_ptr
+  %mask = icmp eq <32 x i64> %m, zeroinitializer
+  %load = call <32 x i64> @llvm.masked.load.v32i64(<32 x i64>* %a, i32 8, <32 x i1> %mask, <32 x i64> undef)
+  store <32 x i64> %load, <32 x i64>* %res_ptr
+  ret void
+}
+declare <32 x i64> @llvm.masked.load.v32i64(<32 x i64>*, i32, <32 x i1>, <32 x i64>)
+
+define void @masked_load_v64i8(<64 x i8>* %a, <64 x i8>* %m_ptr, <64 x i8>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v64i8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, zero, 64
+; CHECK-NEXT:    vsetvli a3, a3, e8,m4,ta,mu
+; CHECK-NEXT:    vle8.v v28, (a1)
+; CHECK-NEXT:    vmseq.vi v0, v28, 0
+; CHECK-NEXT:    vle8.v v28, (a0), v0.t
+; CHECK-NEXT:    vse8.v v28, (a2)
+; CHECK-NEXT:    ret
+  %m = load <64 x i8>, <64 x i8>* %m_ptr
+  %mask = icmp eq <64 x i8> %m, zeroinitializer
+  %load = call <64 x i8> @llvm.masked.load.v64i8(<64 x i8>* %a, i32 8, <64 x i1> %mask, <64 x i8> undef)
+  store <64 x i8> %load, <64 x i8>* %res_ptr
+  ret void
+}
+declare <64 x i8> @llvm.masked.load.v64i8(<64 x i8>*, i32, <64 x i1>, <64 x i8>)
+
+define void @masked_load_v64i16(<64 x i16>* %a, <64 x i16>* %m_ptr, <64 x i16>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v64i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, zero, 64
+; CHECK-NEXT:    vsetvli a3, a3, e16,m8,ta,mu
+; CHECK-NEXT:    vle16.v v8, (a1)
+; CHECK-NEXT:    vmseq.vi v0, v8, 0
+; CHECK-NEXT:    vle16.v v8, (a0), v0.t
+; CHECK-NEXT:    vse16.v v8, (a2)
+; CHECK-NEXT:    ret
+  %m = load <64 x i16>, <64 x i16>* %m_ptr
+  %mask = icmp eq <64 x i16> %m, zeroinitializer
+  %load = call <64 x i16> @llvm.masked.load.v64i16(<64 x i16>* %a, i32 8, <64 x i1> %mask, <64 x i16> undef)
+  store <64 x i16> %load, <64 x i16>* %res_ptr
+  ret void
+}
+declare <64 x i16> @llvm.masked.load.v64i16(<64 x i16>*, i32, <64 x i1>, <64 x i16>)
+
+define void @masked_load_v64i32(<64 x i32>* %a, <64 x i32>* %m_ptr, <64 x i32>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v64i32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, a1, 128
+; CHECK-NEXT:    addi a4, zero, 32
+; CHECK-NEXT:    vsetvli a4, a4, e32,m8,ta,mu
+; CHECK-NEXT:    vle32.v v8, (a1)
+; CHECK-NEXT:    vle32.v v16, (a3)
+; CHECK-NEXT:    vmseq.vi v0, v8, 0
+; CHECK-NEXT:    vmseq.vi v26, v16, 0
+; CHECK-NEXT:    vle32.v v8, (a0), v0.t
+; CHECK-NEXT:    addi a0, a0, 128
+; CHECK-NEXT:    vmv1r.v v0, v26
+; CHECK-NEXT:    vle32.v v16, (a0), v0.t
+; CHECK-NEXT:    vse32.v v8, (a2)
+; CHECK-NEXT:    addi a0, a2, 128
+; CHECK-NEXT:    vse32.v v16, (a0)
+; CHECK-NEXT:    ret
+  %m = load <64 x i32>, <64 x i32>* %m_ptr
+  %mask = icmp eq <64 x i32> %m, zeroinitializer
+  %load = call <64 x i32> @llvm.masked.load.v64i32(<64 x i32>* %a, i32 8, <64 x i1> %mask, <64 x i32> undef)
+  store <64 x i32> %load, <64 x i32>* %res_ptr
+  ret void
+}
+declare <64 x i32> @llvm.masked.load.v64i32(<64 x i32>*, i32, <64 x i1>, <64 x i32>)
+
+define void @masked_load_v128i8(<128 x i8>* %a, <128 x i8>* %m_ptr, <128 x i8>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v128i8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, zero, 128
+; CHECK-NEXT:    vsetvli a3, a3, e8,m8,ta,mu
+; CHECK-NEXT:    vle8.v v8, (a1)
+; CHECK-NEXT:    vmseq.vi v0, v8, 0
+; CHECK-NEXT:    vle8.v v8, (a0), v0.t
+; CHECK-NEXT:    vse8.v v8, (a2)
+; CHECK-NEXT:    ret
+  %m = load <128 x i8>, <128 x i8>* %m_ptr
+  %mask = icmp eq <128 x i8> %m, zeroinitializer
+  %load = call <128 x i8> @llvm.masked.load.v128i8(<128 x i8>* %a, i32 8, <128 x i1> %mask, <128 x i8> undef)
+  store <128 x i8> %load, <128 x i8>* %res_ptr
+  ret void
+}
+declare <128 x i8> @llvm.masked.load.v128i8(<128 x i8>*, i32, <128 x i1>, <128 x i8>)
+
+define void @masked_load_v256i8(<256 x i8>* %a, <256 x i8>* %m_ptr, <256 x i8>* %res_ptr) nounwind {
+; CHECK-LABEL: masked_load_v256i8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, a1, 128
+; CHECK-NEXT:    addi a4, zero, 128
+; CHECK-NEXT:    vsetvli a4, a4, e8,m8,ta,mu
+; CHECK-NEXT:    vle8.v v8, (a1)
+; CHECK-NEXT:    vle8.v v16, (a3)
+; CHECK-NEXT:    vmseq.vi v0, v8, 0
+; CHECK-NEXT:    vmseq.vi v26, v16, 0
+; CHECK-NEXT:    vle8.v v8, (a0), v0.t
+; CHECK-NEXT:    addi a0, a0, 128
+; CHECK-NEXT:    vmv1r.v v0, v26
+; CHECK-NEXT:    vle8.v v16, (a0), v0.t
+; CHECK-NEXT:    vse8.v v8, (a2)
+; CHECK-NEXT:    addi a0, a2, 128
+; CHECK-NEXT:    vse8.v v16, (a0)
+; CHECK-NEXT:    ret
+  %m = load <256 x i8>, <256 x i8>* %m_ptr
+  %mask = icmp eq <256 x i8> %m, zeroinitializer
+  %load = call <256 x i8> @llvm.masked.load.v256i8(<256 x i8>* %a, i32 8, <256 x i1> %mask, <256 x i8> undef)
+  store <256 x i8> %load, <256 x i8>* %res_ptr
+  ret void
+}
+declare <256 x i8> @llvm.masked.load.v256i8(<256 x i8>*, i32, <256 x i1>, <256 x i8>)

llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-store-fp.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -mattr=+m,+experimental-v,+f,+d,+experimental-zfh -verify-machineinstrs -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV32
+; RUN: llc -mtriple=riscv64 -mattr=+m,+experimental-v,+f,+d,+experimental-zfh -verify-machineinstrs -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV64
+
+define void @masked_store_v1f16(<1 x half>* %val_ptr, <1 x half>* %a, <1 x half>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v1f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 1, e16,m1,ta,mu
+; CHECK-NEXT:    vle16.v v25, (a2)
+; CHECK-NEXT:    vle16.v v26, (a0)
+; CHECK-NEXT:    fmv.h.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v25, ft0
+; CHECK-NEXT:    vse16.v v26, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <1 x half>, <1 x half>* %m_ptr
+  %mask = fcmp oeq <1 x half> %m, zeroinitializer
+  %val = load <1 x half>, <1 x half>* %val_ptr
+  call void @llvm.masked.store.v1f16.p0v1f16(<1 x half> %val, <1 x half>* %a, i32 8, <1 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v1f16.p0v1f16(<1 x half>, <1 x half>*, i32, <1 x i1>)
+
+define void @masked_store_v1f32(<1 x float>* %val_ptr, <1 x float>* %a, <1 x float>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v1f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 1, e32,m1,ta,mu
+; CHECK-NEXT:    vle32.v v25, (a2)
+; CHECK-NEXT:    vle32.v v26, (a0)
+; CHECK-NEXT:    fmv.w.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v25, ft0
+; CHECK-NEXT:    vse32.v v26, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <1 x float>, <1 x float>* %m_ptr
+  %mask = fcmp oeq <1 x float> %m, zeroinitializer
+  %val = load <1 x float>, <1 x float>* %val_ptr
+  call void @llvm.masked.store.v1f32.p0v1f32(<1 x float> %val, <1 x float>* %a, i32 8, <1 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v1f32.p0v1f32(<1 x float>, <1 x float>*, i32, <1 x i1>)
+
+define void @masked_store_v1f64(<1 x double>* %val_ptr, <1 x double>* %a, <1 x double>* %m_ptr) nounwind {
+; RV32-LABEL: masked_store_v1f64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    vsetivli a3, 1, e64,m1,ta,mu
+; RV32-NEXT:    vle64.v v25, (a2)
+; RV32-NEXT:    vle64.v v26, (a0)
+; RV32-NEXT:    fcvt.d.w ft0, zero
+; RV32-NEXT:    vmfeq.vf v0, v25, ft0
+; RV32-NEXT:    vse64.v v26, (a1), v0.t
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: masked_store_v1f64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli a3, 1, e64,m1,ta,mu
+; RV64-NEXT:    vle64.v v25, (a2)
+; RV64-NEXT:    vle64.v v26, (a0)
+; RV64-NEXT:    fmv.d.x ft0, zero
+; RV64-NEXT:    vmfeq.vf v0, v25, ft0
+; RV64-NEXT:    vse64.v v26, (a1), v0.t
+; RV64-NEXT:    ret
+  %m = load <1 x double>, <1 x double>* %m_ptr
+  %mask = fcmp oeq <1 x double> %m, zeroinitializer
+  %val = load <1 x double>, <1 x double>* %val_ptr
+  call void @llvm.masked.store.v1f64.p0v1f64(<1 x double> %val, <1 x double>* %a, i32 8, <1 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v1f64.p0v1f64(<1 x double>, <1 x double>*, i32, <1 x i1>)
+
+define void @masked_store_v2f16(<2 x half>* %val_ptr, <2 x half>* %a, <2 x half>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v2f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 2, e16,m1,ta,mu
+; CHECK-NEXT:    vle16.v v25, (a2)
+; CHECK-NEXT:    vle16.v v26, (a0)
+; CHECK-NEXT:    fmv.h.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v25, ft0
+; CHECK-NEXT:    vse16.v v26, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <2 x half>, <2 x half>* %m_ptr
+  %mask = fcmp oeq <2 x half> %m, zeroinitializer
+  %val = load <2 x half>, <2 x half>* %val_ptr
+  call void @llvm.masked.store.v2f16.p0v2f16(<2 x half> %val, <2 x half>* %a, i32 8, <2 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v2f16.p0v2f16(<2 x half>, <2 x half>*, i32, <2 x i1>)
+
+define void @masked_store_v2f32(<2 x float>* %val_ptr, <2 x float>* %a, <2 x float>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v2f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 2, e32,m1,ta,mu
+; CHECK-NEXT:    vle32.v v25, (a2)
+; CHECK-NEXT:    vle32.v v26, (a0)
+; CHECK-NEXT:    fmv.w.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v25, ft0
+; CHECK-NEXT:    vse32.v v26, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <2 x float>, <2 x float>* %m_ptr
+  %mask = fcmp oeq <2 x float> %m, zeroinitializer
+  %val = load <2 x float>, <2 x float>* %val_ptr
+  call void @llvm.masked.store.v2f32.p0v2f32(<2 x float> %val, <2 x float>* %a, i32 8, <2 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v2f32.p0v2f32(<2 x float>, <2 x float>*, i32, <2 x i1>)
+
+define void @masked_store_v2f64(<2 x double>* %val_ptr, <2 x double>* %a, <2 x double>* %m_ptr) nounwind {
+; RV32-LABEL: masked_store_v2f64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    vsetivli a3, 2, e64,m1,ta,mu
+; RV32-NEXT:    vle64.v v25, (a2)
+; RV32-NEXT:    vle64.v v26, (a0)
+; RV32-NEXT:    fcvt.d.w ft0, zero
+; RV32-NEXT:    vmfeq.vf v0, v25, ft0
+; RV32-NEXT:    vse64.v v26, (a1), v0.t
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: masked_store_v2f64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli a3, 2, e64,m1,ta,mu
+; RV64-NEXT:    vle64.v v25, (a2)
+; RV64-NEXT:    vle64.v v26, (a0)
+; RV64-NEXT:    fmv.d.x ft0, zero
+; RV64-NEXT:    vmfeq.vf v0, v25, ft0
+; RV64-NEXT:    vse64.v v26, (a1), v0.t
+; RV64-NEXT:    ret
+  %m = load <2 x double>, <2 x double>* %m_ptr
+  %mask = fcmp oeq <2 x double> %m, zeroinitializer
+  %val = load <2 x double>, <2 x double>* %val_ptr
+  call void @llvm.masked.store.v2f64.p0v2f64(<2 x double> %val, <2 x double>* %a, i32 8, <2 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v2f64.p0v2f64(<2 x double>, <2 x double>*, i32, <2 x i1>)
+
+define void @masked_store_v4f16(<4 x half>* %val_ptr, <4 x half>* %a, <4 x half>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v4f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 4, e16,m1,ta,mu
+; CHECK-NEXT:    vle16.v v25, (a2)
+; CHECK-NEXT:    vle16.v v26, (a0)
+; CHECK-NEXT:    fmv.h.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v25, ft0
+; CHECK-NEXT:    vse16.v v26, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <4 x half>, <4 x half>* %m_ptr
+  %mask = fcmp oeq <4 x half> %m, zeroinitializer
+  %val = load <4 x half>, <4 x half>* %val_ptr
+  call void @llvm.masked.store.v4f16.p0v4f16(<4 x half> %val, <4 x half>* %a, i32 8, <4 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v4f16.p0v4f16(<4 x half>, <4 x half>*, i32, <4 x i1>)
+
+define void @masked_store_v4f32(<4 x float>* %val_ptr, <4 x float>* %a, <4 x float>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v4f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 4, e32,m1,ta,mu
+; CHECK-NEXT:    vle32.v v25, (a2)
+; CHECK-NEXT:    vle32.v v26, (a0)
+; CHECK-NEXT:    fmv.w.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v25, ft0
+; CHECK-NEXT:    vse32.v v26, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <4 x float>, <4 x float>* %m_ptr
+  %mask = fcmp oeq <4 x float> %m, zeroinitializer
+  %val = load <4 x float>, <4 x float>* %val_ptr
+  call void @llvm.masked.store.v4f32.p0v4f32(<4 x float> %val, <4 x float>* %a, i32 8, <4 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v4f32.p0v4f32(<4 x float>, <4 x float>*, i32, <4 x i1>)
+
+define void @masked_store_v4f64(<4 x double>* %val_ptr, <4 x double>* %a, <4 x double>* %m_ptr) nounwind {
+; RV32-LABEL: masked_store_v4f64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    vsetivli a3, 4, e64,m2,ta,mu
+; RV32-NEXT:    vle64.v v26, (a2)
+; RV32-NEXT:    vle64.v v28, (a0)
+; RV32-NEXT:    fcvt.d.w ft0, zero
+; RV32-NEXT:    vmfeq.vf v0, v26, ft0
+; RV32-NEXT:    vse64.v v28, (a1), v0.t
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: masked_store_v4f64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli a3, 4, e64,m2,ta,mu
+; RV64-NEXT:    vle64.v v26, (a2)
+; RV64-NEXT:    vle64.v v28, (a0)
+; RV64-NEXT:    fmv.d.x ft0, zero
+; RV64-NEXT:    vmfeq.vf v0, v26, ft0
+; RV64-NEXT:    vse64.v v28, (a1), v0.t
+; RV64-NEXT:    ret
+  %m = load <4 x double>, <4 x double>* %m_ptr
+  %mask = fcmp oeq <4 x double> %m, zeroinitializer
+  %val = load <4 x double>, <4 x double>* %val_ptr
+  call void @llvm.masked.store.v4f64.p0v4f64(<4 x double> %val, <4 x double>* %a, i32 8, <4 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v4f64.p0v4f64(<4 x double>, <4 x double>*, i32, <4 x i1>)
+
+define void @masked_store_v8f16(<8 x half>* %val_ptr, <8 x half>* %a, <8 x half>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v8f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 8, e16,m1,ta,mu
+; CHECK-NEXT:    vle16.v v25, (a2)
+; CHECK-NEXT:    vle16.v v26, (a0)
+; CHECK-NEXT:    fmv.h.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v25, ft0
+; CHECK-NEXT:    vse16.v v26, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <8 x half>, <8 x half>* %m_ptr
+  %mask = fcmp oeq <8 x half> %m, zeroinitializer
+  %val = load <8 x half>, <8 x half>* %val_ptr
+  call void @llvm.masked.store.v8f16.p0v8f16(<8 x half> %val, <8 x half>* %a, i32 8, <8 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v8f16.p0v8f16(<8 x half>, <8 x half>*, i32, <8 x i1>)
+
+define void @masked_store_v8f32(<8 x float>* %val_ptr, <8 x float>* %a, <8 x float>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v8f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 8, e32,m2,ta,mu
+; CHECK-NEXT:    vle32.v v26, (a2)
+; CHECK-NEXT:    vle32.v v28, (a0)
+; CHECK-NEXT:    fmv.w.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v26, ft0
+; CHECK-NEXT:    vse32.v v28, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <8 x float>, <8 x float>* %m_ptr
+  %mask = fcmp oeq <8 x float> %m, zeroinitializer
+  %val = load <8 x float>, <8 x float>* %val_ptr
+  call void @llvm.masked.store.v8f32.p0v8f32(<8 x float> %val, <8 x float>* %a, i32 8, <8 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v8f32.p0v8f32(<8 x float>, <8 x float>*, i32, <8 x i1>)
+
+define void @masked_store_v8f64(<8 x double>* %val_ptr, <8 x double>* %a, <8 x double>* %m_ptr) nounwind {
+; RV32-LABEL: masked_store_v8f64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    vsetivli a3, 8, e64,m4,ta,mu
+; RV32-NEXT:    vle64.v v28, (a2)
+; RV32-NEXT:    vle64.v v8, (a0)
+; RV32-NEXT:    fcvt.d.w ft0, zero
+; RV32-NEXT:    vmfeq.vf v0, v28, ft0
+; RV32-NEXT:    vse64.v v8, (a1), v0.t
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: masked_store_v8f64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli a3, 8, e64,m4,ta,mu
+; RV64-NEXT:    vle64.v v28, (a2)
+; RV64-NEXT:    vle64.v v8, (a0)
+; RV64-NEXT:    fmv.d.x ft0, zero
+; RV64-NEXT:    vmfeq.vf v0, v28, ft0
+; RV64-NEXT:    vse64.v v8, (a1), v0.t
+; RV64-NEXT:    ret
+  %m = load <8 x double>, <8 x double>* %m_ptr
+  %mask = fcmp oeq <8 x double> %m, zeroinitializer
+  %val = load <8 x double>, <8 x double>* %val_ptr
+  call void @llvm.masked.store.v8f64.p0v8f64(<8 x double> %val, <8 x double>* %a, i32 8, <8 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v8f64.p0v8f64(<8 x double>, <8 x double>*, i32, <8 x i1>)
+
+define void @masked_store_v16f16(<16 x half>* %val_ptr, <16 x half>* %a, <16 x half>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v16f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 16, e16,m2,ta,mu
+; CHECK-NEXT:    vle16.v v26, (a2)
+; CHECK-NEXT:    vle16.v v28, (a0)
+; CHECK-NEXT:    fmv.h.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v26, ft0
+; CHECK-NEXT:    vse16.v v28, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <16 x half>, <16 x half>* %m_ptr
+  %mask = fcmp oeq <16 x half> %m, zeroinitializer
+  %val = load <16 x half>, <16 x half>* %val_ptr
+  call void @llvm.masked.store.v16f16.p0v16f16(<16 x half> %val, <16 x half>* %a, i32 8, <16 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v16f16.p0v16f16(<16 x half>, <16 x half>*, i32, <16 x i1>)
+
+define void @masked_store_v16f32(<16 x float>* %val_ptr, <16 x float>* %a, <16 x float>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v16f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 16, e32,m4,ta,mu
+; CHECK-NEXT:    vle32.v v28, (a2)
+; CHECK-NEXT:    vle32.v v8, (a0)
+; CHECK-NEXT:    fmv.w.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v28, ft0
+; CHECK-NEXT:    vse32.v v8, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <16 x float>, <16 x float>* %m_ptr
+  %mask = fcmp oeq <16 x float> %m, zeroinitializer
+  %val = load <16 x float>, <16 x float>* %val_ptr
+  call void @llvm.masked.store.v16f32.p0v16f32(<16 x float> %val, <16 x float>* %a, i32 8, <16 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v16f32.p0v16f32(<16 x float>, <16 x float>*, i32, <16 x i1>)
+
+define void @masked_store_v16f64(<16 x double>* %val_ptr, <16 x double>* %a, <16 x double>* %m_ptr) nounwind {
+; RV32-LABEL: masked_store_v16f64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    vsetivli a3, 16, e64,m8,ta,mu
+; RV32-NEXT:    vle64.v v8, (a2)
+; RV32-NEXT:    vle64.v v16, (a0)
+; RV32-NEXT:    fcvt.d.w ft0, zero
+; RV32-NEXT:    vmfeq.vf v0, v8, ft0
+; RV32-NEXT:    vse64.v v16, (a1), v0.t
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: masked_store_v16f64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli a3, 16, e64,m8,ta,mu
+; RV64-NEXT:    vle64.v v8, (a2)
+; RV64-NEXT:    vle64.v v16, (a0)
+; RV64-NEXT:    fmv.d.x ft0, zero
+; RV64-NEXT:    vmfeq.vf v0, v8, ft0
+; RV64-NEXT:    vse64.v v16, (a1), v0.t
+; RV64-NEXT:    ret
+  %m = load <16 x double>, <16 x double>* %m_ptr
+  %mask = fcmp oeq <16 x double> %m, zeroinitializer
+  %val = load <16 x double>, <16 x double>* %val_ptr
+  call void @llvm.masked.store.v16f64.p0v16f64(<16 x double> %val, <16 x double>* %a, i32 8, <16 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v16f64.p0v16f64(<16 x double>, <16 x double>*, i32, <16 x i1>)
+
+define void @masked_store_v32f16(<32 x half>* %val_ptr, <32 x half>* %a, <32 x half>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v32f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, zero, 32
+; CHECK-NEXT:    vsetvli a3, a3, e16,m4,ta,mu
+; CHECK-NEXT:    vle16.v v28, (a2)
+; CHECK-NEXT:    vle16.v v8, (a0)
+; CHECK-NEXT:    fmv.h.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v28, ft0
+; CHECK-NEXT:    vse16.v v8, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <32 x half>, <32 x half>* %m_ptr
+  %mask = fcmp oeq <32 x half> %m, zeroinitializer
+  %val = load <32 x half>, <32 x half>* %val_ptr
+  call void @llvm.masked.store.v32f16.p0v32f16(<32 x half> %val, <32 x half>* %a, i32 8, <32 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v32f16.p0v32f16(<32 x half>, <32 x half>*, i32, <32 x i1>)
+
+define void @masked_store_v32f32(<32 x float>* %val_ptr, <32 x float>* %a, <32 x float>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v32f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, zero, 32
+; CHECK-NEXT:    vsetvli a3, a3, e32,m8,ta,mu
+; CHECK-NEXT:    vle32.v v8, (a2)
+; CHECK-NEXT:    vle32.v v16, (a0)
+; CHECK-NEXT:    fmv.w.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v8, ft0
+; CHECK-NEXT:    vse32.v v16, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <32 x float>, <32 x float>* %m_ptr
+  %mask = fcmp oeq <32 x float> %m, zeroinitializer
+  %val = load <32 x float>, <32 x float>* %val_ptr
+  call void @llvm.masked.store.v32f32.p0v32f32(<32 x float> %val, <32 x float>* %a, i32 8, <32 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v32f32.p0v32f32(<32 x float>, <32 x float>*, i32, <32 x i1>)
+
+define void @masked_store_v32f64(<32 x double>* %val_ptr, <32 x double>* %a, <32 x double>* %m_ptr) nounwind {
+; RV32-LABEL: masked_store_v32f64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    vsetivli a3, 16, e64,m8,ta,mu
+; RV32-NEXT:    addi a3, a2, 128
+; RV32-NEXT:    vle64.v v8, (a3)
+; RV32-NEXT:    vle64.v v16, (a2)
+; RV32-NEXT:    fcvt.d.w ft0, zero
+; RV32-NEXT:    vmfeq.vf v1, v8, ft0
+; RV32-NEXT:    vle64.v v8, (a0)
+; RV32-NEXT:    addi a0, a0, 128
+; RV32-NEXT:    vle64.v v24, (a0)
+; RV32-NEXT:    vmfeq.vf v0, v16, ft0
+; RV32-NEXT:    vse64.v v8, (a1), v0.t
+; RV32-NEXT:    addi a0, a1, 128
+; RV32-NEXT:    vmv1r.v v0, v1
+; RV32-NEXT:    vse64.v v24, (a0), v0.t
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: masked_store_v32f64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli a3, 16, e64,m8,ta,mu
+; RV64-NEXT:    addi a3, a2, 128
+; RV64-NEXT:    vle64.v v8, (a3)
+; RV64-NEXT:    vle64.v v16, (a2)
+; RV64-NEXT:    fmv.d.x ft0, zero
+; RV64-NEXT:    vmfeq.vf v1, v8, ft0
+; RV64-NEXT:    vle64.v v8, (a0)
+; RV64-NEXT:    addi a0, a0, 128
+; RV64-NEXT:    vle64.v v24, (a0)
+; RV64-NEXT:    vmfeq.vf v0, v16, ft0
+; RV64-NEXT:    vse64.v v8, (a1), v0.t
+; RV64-NEXT:    addi a0, a1, 128
+; RV64-NEXT:    vmv1r.v v0, v1
+; RV64-NEXT:    vse64.v v24, (a0), v0.t
+; RV64-NEXT:    ret
+  %m = load <32 x double>, <32 x double>* %m_ptr
+  %mask = fcmp oeq <32 x double> %m, zeroinitializer
+  %val = load <32 x double>, <32 x double>* %val_ptr
+  call void @llvm.masked.store.v32f32.p0v32f64(<32 x double> %val, <32 x double>* %a, i32 8, <32 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v32f32.p0v32f64(<32 x double>, <32 x double>*, i32, <32 x i1>)
+
+define void @masked_store_v64f16(<64 x half>* %val_ptr, <64 x half>* %a, <64 x half>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v64f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, zero, 64
+; CHECK-NEXT:    vsetvli a3, a3, e16,m8,ta,mu
+; CHECK-NEXT:    vle16.v v8, (a2)
+; CHECK-NEXT:    vle16.v v16, (a0)
+; CHECK-NEXT:    fmv.h.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v0, v8, ft0
+; CHECK-NEXT:    vse16.v v16, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <64 x half>, <64 x half>* %m_ptr
+  %mask = fcmp oeq <64 x half> %m, zeroinitializer
+  %val = load <64 x half>, <64 x half>* %val_ptr
+  call void @llvm.masked.store.v64f16.p0v64f16(<64 x half> %val, <64 x half>* %a, i32 8, <64 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v64f16.p0v64f16(<64 x half>, <64 x half>*, i32, <64 x i1>)
+
+define void @masked_store_v64f32(<64 x float>* %val_ptr, <64 x float>* %a, <64 x float>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v64f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, zero, 32
+; CHECK-NEXT:    vsetvli a3, a3, e32,m8,ta,mu
+; CHECK-NEXT:    addi a3, a2, 128
+; CHECK-NEXT:    vle32.v v8, (a3)
+; CHECK-NEXT:    vle32.v v16, (a2)
+; CHECK-NEXT:    fmv.w.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v1, v8, ft0
+; CHECK-NEXT:    vle32.v v8, (a0)
+; CHECK-NEXT:    addi a0, a0, 128
+; CHECK-NEXT:    vle32.v v24, (a0)
+; CHECK-NEXT:    vmfeq.vf v0, v16, ft0
+; CHECK-NEXT:    vse32.v v8, (a1), v0.t
+; CHECK-NEXT:    addi a0, a1, 128
+; CHECK-NEXT:    vmv1r.v v0, v1
+; CHECK-NEXT:    vse32.v v24, (a0), v0.t
+; CHECK-NEXT:    ret
+  %m = load <64 x float>, <64 x float>* %m_ptr
+  %mask = fcmp oeq <64 x float> %m, zeroinitializer
+  %val = load <64 x float>, <64 x float>* %val_ptr
+  call void @llvm.masked.store.v64f16.p0v64f32(<64 x float> %val, <64 x float>* %a, i32 8, <64 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v64f16.p0v64f32(<64 x float>, <64 x float>*, i32, <64 x i1>)
+
+define void @masked_store_v128f16(<128 x half>* %val_ptr, <128 x half>* %a, <128 x half>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v128f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, zero, 64
+; CHECK-NEXT:    vsetvli a3, a3, e16,m8,ta,mu
+; CHECK-NEXT:    addi a3, a2, 128
+; CHECK-NEXT:    vle16.v v8, (a3)
+; CHECK-NEXT:    vle16.v v16, (a2)
+; CHECK-NEXT:    fmv.h.x ft0, zero
+; CHECK-NEXT:    vmfeq.vf v1, v8, ft0
+; CHECK-NEXT:    vle16.v v8, (a0)
+; CHECK-NEXT:    addi a0, a0, 128
+; CHECK-NEXT:    vle16.v v24, (a0)
+; CHECK-NEXT:    vmfeq.vf v0, v16, ft0
+; CHECK-NEXT:    vse16.v v8, (a1), v0.t
+; CHECK-NEXT:    addi a0, a1, 128
+; CHECK-NEXT:    vmv1r.v v0, v1
+; CHECK-NEXT:    vse16.v v24, (a0), v0.t
+; CHECK-NEXT:    ret
+  %m = load <128 x half>, <128 x half>* %m_ptr
+  %mask = fcmp oeq <128 x half> %m, zeroinitializer
+  %val = load <128 x half>, <128 x half>* %val_ptr
+  call void @llvm.masked.store.v128f16.p0v128f16(<128 x half> %val, <128 x half>* %a, i32 8, <128 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v128f16.p0v128f16(<128 x half>, <128 x half>*, i32, <128 x i1>)

llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-store-int.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -mattr=+m,+experimental-v -verify-machineinstrs -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV32
+; RUN: llc -mtriple=riscv64 -mattr=+m,+experimental-v -verify-machineinstrs -riscv-v-vector-bits-min=128 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV64
+
+define void @masked_store_v1i8(<1 x i8>* %val_ptr, <1 x i8>* %a, <1 x i8>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v1i8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 1, e8,m1,ta,mu
+; CHECK-NEXT:    vle8.v v25, (a2)
+; CHECK-NEXT:    vle8.v v26, (a0)
+; CHECK-NEXT:    vmseq.vi v0, v25, 0
+; CHECK-NEXT:    vse8.v v26, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <1 x i8>, <1 x i8>* %m_ptr
+  %mask = icmp eq <1 x i8> %m, zeroinitializer
+  %val = load <1 x i8>, <1 x i8>* %val_ptr
+  call void @llvm.masked.store.v1i8.p0v1i8(<1 x i8> %val, <1 x i8>* %a, i32 8, <1 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v1i8.p0v1i8(<1 x i8>, <1 x i8>*, i32, <1 x i1>)
+
+define void @masked_store_v1i16(<1 x i16>* %val_ptr, <1 x i16>* %a, <1 x i16>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v1i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 1, e16,m1,ta,mu
+; CHECK-NEXT:    vle16.v v25, (a2)
+; CHECK-NEXT:    vle16.v v26, (a0)
+; CHECK-NEXT:    vmseq.vi v0, v25, 0
+; CHECK-NEXT:    vse16.v v26, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <1 x i16>, <1 x i16>* %m_ptr
+  %mask = icmp eq <1 x i16> %m, zeroinitializer
+  %val = load <1 x i16>, <1 x i16>* %val_ptr
+  call void @llvm.masked.store.v1i16.p0v1i16(<1 x i16> %val, <1 x i16>* %a, i32 8, <1 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v1i16.p0v1i16(<1 x i16>, <1 x i16>*, i32, <1 x i1>)
+
+define void @masked_store_v1i32(<1 x i32>* %val_ptr, <1 x i32>* %a, <1 x i32>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v1i32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 1, e32,m1,ta,mu
+; CHECK-NEXT:    vle32.v v25, (a2)
+; CHECK-NEXT:    vle32.v v26, (a0)
+; CHECK-NEXT:    vmseq.vi v0, v25, 0
+; CHECK-NEXT:    vse32.v v26, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <1 x i32>, <1 x i32>* %m_ptr
+  %mask = icmp eq <1 x i32> %m, zeroinitializer
+  %val = load <1 x i32>, <1 x i32>* %val_ptr
+  call void @llvm.masked.store.v1i32.p0v1i32(<1 x i32> %val, <1 x i32>* %a, i32 8, <1 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v1i32.p0v1i32(<1 x i32>, <1 x i32>*, i32, <1 x i1>)
+
+define void @masked_store_v1i64(<1 x i64>* %val_ptr, <1 x i64>* %a, <1 x i64>* %m_ptr) nounwind {
+; RV32-LABEL: masked_store_v1i64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    vsetivli a3, 1, e64,m1,ta,mu
+; RV32-NEXT:    vle64.v v25, (a2)
+; RV32-NEXT:    vsetivli a2, 2, e32,m1,ta,mu
+; RV32-NEXT:    vmv.v.i v26, 0
+; RV32-NEXT:    vsetivli a2, 1, e64,m1,ta,mu
+; RV32-NEXT:    vle64.v v27, (a0)
+; RV32-NEXT:    vmseq.vv v0, v25, v26
+; RV32-NEXT:    vse64.v v27, (a1), v0.t
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: masked_store_v1i64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli a3, 1, e64,m1,ta,mu
+; RV64-NEXT:    vle64.v v25, (a2)
+; RV64-NEXT:    vle64.v v26, (a0)
+; RV64-NEXT:    vmseq.vi v0, v25, 0
+; RV64-NEXT:    vse64.v v26, (a1), v0.t
+; RV64-NEXT:    ret
+  %m = load <1 x i64>, <1 x i64>* %m_ptr
+  %mask = icmp eq <1 x i64> %m, zeroinitializer
+  %val = load <1 x i64>, <1 x i64>* %val_ptr
+  call void @llvm.masked.store.v1i64.p0v1i64(<1 x i64> %val, <1 x i64>* %a, i32 8, <1 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v1i64.p0v1i64(<1 x i64>, <1 x i64>*, i32, <1 x i1>)
+
+define void @masked_store_v2i8(<2 x i8>* %val_ptr, <2 x i8>* %a, <2 x i8>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v2i8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 2, e8,m1,ta,mu
+; CHECK-NEXT:    vle8.v v25, (a2)
+; CHECK-NEXT:    vle8.v v26, (a0)
+; CHECK-NEXT:    vmseq.vi v0, v25, 0
+; CHECK-NEXT:    vse8.v v26, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <2 x i8>, <2 x i8>* %m_ptr
+  %mask = icmp eq <2 x i8> %m, zeroinitializer
+  %val = load <2 x i8>, <2 x i8>* %val_ptr
+  call void @llvm.masked.store.v2i8.p0v2i8(<2 x i8> %val, <2 x i8>* %a, i32 8, <2 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v2i8.p0v2i8(<2 x i8>, <2 x i8>*, i32, <2 x i1>)
+
+define void @masked_store_v2i16(<2 x i16>* %val_ptr, <2 x i16>* %a, <2 x i16>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v2i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 2, e16,m1,ta,mu
+; CHECK-NEXT:    vle16.v v25, (a2)
+; CHECK-NEXT:    vle16.v v26, (a0)
+; CHECK-NEXT:    vmseq.vi v0, v25, 0
+; CHECK-NEXT:    vse16.v v26, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <2 x i16>, <2 x i16>* %m_ptr
+  %mask = icmp eq <2 x i16> %m, zeroinitializer
+  %val = load <2 x i16>, <2 x i16>* %val_ptr
+  call void @llvm.masked.store.v2i16.p0v2i16(<2 x i16> %val, <2 x i16>* %a, i32 8, <2 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v2i16.p0v2i16(<2 x i16>, <2 x i16>*, i32, <2 x i1>)
+
+define void @masked_store_v2i32(<2 x i32>* %val_ptr, <2 x i32>* %a, <2 x i32>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v2i32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 2, e32,m1,ta,mu
+; CHECK-NEXT:    vle32.v v25, (a2)
+; CHECK-NEXT:    vle32.v v26, (a0)
+; CHECK-NEXT:    vmseq.vi v0, v25, 0
+; CHECK-NEXT:    vse32.v v26, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <2 x i32>, <2 x i32>* %m_ptr
+  %mask = icmp eq <2 x i32> %m, zeroinitializer
+  %val = load <2 x i32>, <2 x i32>* %val_ptr
+  call void @llvm.masked.store.v2i32.p0v2i32(<2 x i32> %val, <2 x i32>* %a, i32 8, <2 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v2i32.p0v2i32(<2 x i32>, <2 x i32>*, i32, <2 x i1>)
+
+define void @masked_store_v2i64(<2 x i64>* %val_ptr, <2 x i64>* %a, <2 x i64>* %m_ptr) nounwind {
+; RV32-LABEL: masked_store_v2i64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    vsetivli a3, 2, e64,m1,ta,mu
+; RV32-NEXT:    vle64.v v25, (a2)
+; RV32-NEXT:    vsetivli a2, 4, e32,m1,ta,mu
+; RV32-NEXT:    vmv.v.i v26, 0
+; RV32-NEXT:    vsetivli a2, 2, e64,m1,ta,mu
+; RV32-NEXT:    vle64.v v27, (a0)
+; RV32-NEXT:    vmseq.vv v0, v25, v26
+; RV32-NEXT:    vse64.v v27, (a1), v0.t
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: masked_store_v2i64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli a3, 2, e64,m1,ta,mu
+; RV64-NEXT:    vle64.v v25, (a2)
+; RV64-NEXT:    vle64.v v26, (a0)
+; RV64-NEXT:    vmseq.vi v0, v25, 0
+; RV64-NEXT:    vse64.v v26, (a1), v0.t
+; RV64-NEXT:    ret
+  %m = load <2 x i64>, <2 x i64>* %m_ptr
+  %mask = icmp eq <2 x i64> %m, zeroinitializer
+  %val = load <2 x i64>, <2 x i64>* %val_ptr
+  call void @llvm.masked.store.v2i64.p0v2i64(<2 x i64> %val, <2 x i64>* %a, i32 8, <2 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v2i64.p0v2i64(<2 x i64>, <2 x i64>*, i32, <2 x i1>)
+
+define void @masked_store_v4i8(<4 x i8>* %val_ptr, <4 x i8>* %a, <4 x i8>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v4i8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 4, e8,m1,ta,mu
+; CHECK-NEXT:    vle8.v v25, (a2)
+; CHECK-NEXT:    vle8.v v26, (a0)
+; CHECK-NEXT:    vmseq.vi v0, v25, 0
+; CHECK-NEXT:    vse8.v v26, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <4 x i8>, <4 x i8>* %m_ptr
+  %mask = icmp eq <4 x i8> %m, zeroinitializer
+  %val = load <4 x i8>, <4 x i8>* %val_ptr
+  call void @llvm.masked.store.v4i8.p0v4i8(<4 x i8> %val, <4 x i8>* %a, i32 8, <4 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v4i8.p0v4i8(<4 x i8>, <4 x i8>*, i32, <4 x i1>)
+
+define void @masked_store_v4i16(<4 x i16>* %val_ptr, <4 x i16>* %a, <4 x i16>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v4i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 4, e16,m1,ta,mu
+; CHECK-NEXT:    vle16.v v25, (a2)
+; CHECK-NEXT:    vle16.v v26, (a0)
+; CHECK-NEXT:    vmseq.vi v0, v25, 0
+; CHECK-NEXT:    vse16.v v26, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <4 x i16>, <4 x i16>* %m_ptr
+  %mask = icmp eq <4 x i16> %m, zeroinitializer
+  %val = load <4 x i16>, <4 x i16>* %val_ptr
+  call void @llvm.masked.store.v4i16.p0v4i16(<4 x i16> %val, <4 x i16>* %a, i32 8, <4 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v4i16.p0v4i16(<4 x i16>, <4 x i16>*, i32, <4 x i1>)
+
+define void @masked_store_v4i32(<4 x i32>* %val_ptr, <4 x i32>* %a, <4 x i32>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v4i32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 4, e32,m1,ta,mu
+; CHECK-NEXT:    vle32.v v25, (a2)
+; CHECK-NEXT:    vle32.v v26, (a0)
+; CHECK-NEXT:    vmseq.vi v0, v25, 0
+; CHECK-NEXT:    vse32.v v26, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <4 x i32>, <4 x i32>* %m_ptr
+  %mask = icmp eq <4 x i32> %m, zeroinitializer
+  %val = load <4 x i32>, <4 x i32>* %val_ptr
+  call void @llvm.masked.store.v4i32.p0v4i32(<4 x i32> %val, <4 x i32>* %a, i32 8, <4 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v4i32.p0v4i32(<4 x i32>, <4 x i32>*, i32, <4 x i1>)
+
+define void @masked_store_v4i64(<4 x i64>* %val_ptr, <4 x i64>* %a, <4 x i64>* %m_ptr) nounwind {
+; RV32-LABEL: masked_store_v4i64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    vsetivli a3, 4, e64,m2,ta,mu
+; RV32-NEXT:    vle64.v v26, (a2)
+; RV32-NEXT:    vsetivli a2, 8, e32,m2,ta,mu
+; RV32-NEXT:    vmv.v.i v28, 0
+; RV32-NEXT:    vsetivli a2, 4, e64,m2,ta,mu
+; RV32-NEXT:    vle64.v v30, (a0)
+; RV32-NEXT:    vmseq.vv v0, v26, v28
+; RV32-NEXT:    vse64.v v30, (a1), v0.t
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: masked_store_v4i64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli a3, 4, e64,m2,ta,mu
+; RV64-NEXT:    vle64.v v26, (a2)
+; RV64-NEXT:    vle64.v v28, (a0)
+; RV64-NEXT:    vmseq.vi v0, v26, 0
+; RV64-NEXT:    vse64.v v28, (a1), v0.t
+; RV64-NEXT:    ret
+  %m = load <4 x i64>, <4 x i64>* %m_ptr
+  %mask = icmp eq <4 x i64> %m, zeroinitializer
+  %val = load <4 x i64>, <4 x i64>* %val_ptr
+  call void @llvm.masked.store.v4i64.p0v4i64(<4 x i64> %val, <4 x i64>* %a, i32 8, <4 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v4i64.p0v4i64(<4 x i64>, <4 x i64>*, i32, <4 x i1>)
+
+define void @masked_store_v8i8(<8 x i8>* %val_ptr, <8 x i8>* %a, <8 x i8>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v8i8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 8, e8,m1,ta,mu
+; CHECK-NEXT:    vle8.v v25, (a2)
+; CHECK-NEXT:    vle8.v v26, (a0)
+; CHECK-NEXT:    vmseq.vi v0, v25, 0
+; CHECK-NEXT:    vse8.v v26, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <8 x i8>, <8 x i8>* %m_ptr
+  %mask = icmp eq <8 x i8> %m, zeroinitializer
+  %val = load <8 x i8>, <8 x i8>* %val_ptr
+  call void @llvm.masked.store.v8i8.p0v8i8(<8 x i8> %val, <8 x i8>* %a, i32 8, <8 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v8i8.p0v8i8(<8 x i8>, <8 x i8>*, i32, <8 x i1>)
+
+define void @masked_store_v8i16(<8 x i16>* %val_ptr, <8 x i16>* %a, <8 x i16>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v8i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 8, e16,m1,ta,mu
+; CHECK-NEXT:    vle16.v v25, (a2)
+; CHECK-NEXT:    vle16.v v26, (a0)
+; CHECK-NEXT:    vmseq.vi v0, v25, 0
+; CHECK-NEXT:    vse16.v v26, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <8 x i16>, <8 x i16>* %m_ptr
+  %mask = icmp eq <8 x i16> %m, zeroinitializer
+  %val = load <8 x i16>, <8 x i16>* %val_ptr
+  call void @llvm.masked.store.v8i16.p0v8i16(<8 x i16> %val, <8 x i16>* %a, i32 8, <8 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v8i16.p0v8i16(<8 x i16>, <8 x i16>*, i32, <8 x i1>)
+
+define void @masked_store_v8i32(<8 x i32>* %val_ptr, <8 x i32>* %a, <8 x i32>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v8i32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 8, e32,m2,ta,mu
+; CHECK-NEXT:    vle32.v v26, (a2)
+; CHECK-NEXT:    vle32.v v28, (a0)
+; CHECK-NEXT:    vmseq.vi v0, v26, 0
+; CHECK-NEXT:    vse32.v v28, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <8 x i32>, <8 x i32>* %m_ptr
+  %mask = icmp eq <8 x i32> %m, zeroinitializer
+  %val = load <8 x i32>, <8 x i32>* %val_ptr
+  call void @llvm.masked.store.v8i32.p0v8i32(<8 x i32> %val, <8 x i32>* %a, i32 8, <8 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v8i32.p0v8i32(<8 x i32>, <8 x i32>*, i32, <8 x i1>)
+
+define void @masked_store_v8i64(<8 x i64>* %val_ptr, <8 x i64>* %a, <8 x i64>* %m_ptr) nounwind {
+; RV32-LABEL: masked_store_v8i64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    vsetivli a3, 8, e64,m4,ta,mu
+; RV32-NEXT:    vle64.v v28, (a2)
+; RV32-NEXT:    vsetivli a2, 16, e32,m4,ta,mu
+; RV32-NEXT:    vmv.v.i v8, 0
+; RV32-NEXT:    vsetivli a2, 8, e64,m4,ta,mu
+; RV32-NEXT:    vle64.v v12, (a0)
+; RV32-NEXT:    vmseq.vv v0, v28, v8
+; RV32-NEXT:    vse64.v v12, (a1), v0.t
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: masked_store_v8i64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli a3, 8, e64,m4,ta,mu
+; RV64-NEXT:    vle64.v v28, (a2)
+; RV64-NEXT:    vle64.v v8, (a0)
+; RV64-NEXT:    vmseq.vi v0, v28, 0
+; RV64-NEXT:    vse64.v v8, (a1), v0.t
+; RV64-NEXT:    ret
+  %m = load <8 x i64>, <8 x i64>* %m_ptr
+  %mask = icmp eq <8 x i64> %m, zeroinitializer
+  %val = load <8 x i64>, <8 x i64>* %val_ptr
+  call void @llvm.masked.store.v8i64.p0v8i64(<8 x i64> %val, <8 x i64>* %a, i32 8, <8 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v8i64.p0v8i64(<8 x i64>, <8 x i64>*, i32, <8 x i1>)
+
+define void @masked_store_v16i8(<16 x i8>* %val_ptr, <16 x i8>* %a, <16 x i8>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v16i8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 16, e8,m1,ta,mu
+; CHECK-NEXT:    vle8.v v25, (a2)
+; CHECK-NEXT:    vle8.v v26, (a0)
+; CHECK-NEXT:    vmseq.vi v0, v25, 0
+; CHECK-NEXT:    vse8.v v26, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <16 x i8>, <16 x i8>* %m_ptr
+  %mask = icmp eq <16 x i8> %m, zeroinitializer
+  %val = load <16 x i8>, <16 x i8>* %val_ptr
+  call void @llvm.masked.store.v16i8.p0v16i8(<16 x i8> %val, <16 x i8>* %a, i32 8, <16 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v16i8.p0v16i8(<16 x i8>, <16 x i8>*, i32, <16 x i1>)
+
+define void @masked_store_v16i16(<16 x i16>* %val_ptr, <16 x i16>* %a, <16 x i16>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v16i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 16, e16,m2,ta,mu
+; CHECK-NEXT:    vle16.v v26, (a2)
+; CHECK-NEXT:    vle16.v v28, (a0)
+; CHECK-NEXT:    vmseq.vi v0, v26, 0
+; CHECK-NEXT:    vse16.v v28, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <16 x i16>, <16 x i16>* %m_ptr
+  %mask = icmp eq <16 x i16> %m, zeroinitializer
+  %val = load <16 x i16>, <16 x i16>* %val_ptr
+  call void @llvm.masked.store.v16i16.p0v16i16(<16 x i16> %val, <16 x i16>* %a, i32 8, <16 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v16i16.p0v16i16(<16 x i16>, <16 x i16>*, i32, <16 x i1>)
+
+define void @masked_store_v16i32(<16 x i32>* %val_ptr, <16 x i32>* %a, <16 x i32>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v16i32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetivli a3, 16, e32,m4,ta,mu
+; CHECK-NEXT:    vle32.v v28, (a2)
+; CHECK-NEXT:    vle32.v v8, (a0)
+; CHECK-NEXT:    vmseq.vi v0, v28, 0
+; CHECK-NEXT:    vse32.v v8, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <16 x i32>, <16 x i32>* %m_ptr
+  %mask = icmp eq <16 x i32> %m, zeroinitializer
+  %val = load <16 x i32>, <16 x i32>* %val_ptr
+  call void @llvm.masked.store.v16i32.p0v16i32(<16 x i32> %val, <16 x i32>* %a, i32 8, <16 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v16i32.p0v16i32(<16 x i32>, <16 x i32>*, i32, <16 x i1>)
+
+define void @masked_store_v16i64(<16 x i64>* %val_ptr, <16 x i64>* %a, <16 x i64>* %m_ptr) nounwind {
+; RV32-LABEL: masked_store_v16i64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    vsetivli a3, 16, e64,m8,ta,mu
+; RV32-NEXT:    vle64.v v8, (a2)
+; RV32-NEXT:    addi a2, zero, 32
+; RV32-NEXT:    vsetvli a2, a2, e32,m8,ta,mu
+; RV32-NEXT:    vmv.v.i v16, 0
+; RV32-NEXT:    vsetivli a2, 16, e64,m8,ta,mu
+; RV32-NEXT:    vle64.v v24, (a0)
+; RV32-NEXT:    vmseq.vv v0, v8, v16
+; RV32-NEXT:    vse64.v v24, (a1), v0.t
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: masked_store_v16i64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli a3, 16, e64,m8,ta,mu
+; RV64-NEXT:    vle64.v v8, (a2)
+; RV64-NEXT:    vle64.v v16, (a0)
+; RV64-NEXT:    vmseq.vi v0, v8, 0
+; RV64-NEXT:    vse64.v v16, (a1), v0.t
+; RV64-NEXT:    ret
+  %m = load <16 x i64>, <16 x i64>* %m_ptr
+  %mask = icmp eq <16 x i64> %m, zeroinitializer
+  %val = load <16 x i64>, <16 x i64>* %val_ptr
+  call void @llvm.masked.store.v16i64.p0v16i64(<16 x i64> %val, <16 x i64>* %a, i32 8, <16 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v16i64.p0v16i64(<16 x i64>, <16 x i64>*, i32, <16 x i1>)
+
+define void @masked_store_v32i8(<32 x i8>* %val_ptr, <32 x i8>* %a, <32 x i8>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v32i8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, zero, 32
+; CHECK-NEXT:    vsetvli a3, a3, e8,m2,ta,mu
+; CHECK-NEXT:    vle8.v v26, (a2)
+; CHECK-NEXT:    vle8.v v28, (a0)
+; CHECK-NEXT:    vmseq.vi v0, v26, 0
+; CHECK-NEXT:    vse8.v v28, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <32 x i8>, <32 x i8>* %m_ptr
+  %mask = icmp eq <32 x i8> %m, zeroinitializer
+  %val = load <32 x i8>, <32 x i8>* %val_ptr
+  call void @llvm.masked.store.v32i8.p0v32i8(<32 x i8> %val, <32 x i8>* %a, i32 8, <32 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v32i8.p0v32i8(<32 x i8>, <32 x i8>*, i32, <32 x i1>)
+
+define void @masked_store_v32i16(<32 x i16>* %val_ptr, <32 x i16>* %a, <32 x i16>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v32i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, zero, 32
+; CHECK-NEXT:    vsetvli a3, a3, e16,m4,ta,mu
+; CHECK-NEXT:    vle16.v v28, (a2)
+; CHECK-NEXT:    vle16.v v8, (a0)
+; CHECK-NEXT:    vmseq.vi v0, v28, 0
+; CHECK-NEXT:    vse16.v v8, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <32 x i16>, <32 x i16>* %m_ptr
+  %mask = icmp eq <32 x i16> %m, zeroinitializer
+  %val = load <32 x i16>, <32 x i16>* %val_ptr
+  call void @llvm.masked.store.v32i16.p0v32i16(<32 x i16> %val, <32 x i16>* %a, i32 8, <32 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v32i16.p0v32i16(<32 x i16>, <32 x i16>*, i32, <32 x i1>)
+
+define void @masked_store_v32i32(<32 x i32>* %val_ptr, <32 x i32>* %a, <32 x i32>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v32i32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, zero, 32
+; CHECK-NEXT:    vsetvli a3, a3, e32,m8,ta,mu
+; CHECK-NEXT:    vle32.v v8, (a2)
+; CHECK-NEXT:    vle32.v v16, (a0)
+; CHECK-NEXT:    vmseq.vi v0, v8, 0
+; CHECK-NEXT:    vse32.v v16, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <32 x i32>, <32 x i32>* %m_ptr
+  %mask = icmp eq <32 x i32> %m, zeroinitializer
+  %val = load <32 x i32>, <32 x i32>* %val_ptr
+  call void @llvm.masked.store.v32i32.p0v32i32(<32 x i32> %val, <32 x i32>* %a, i32 8, <32 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v32i32.p0v32i32(<32 x i32>, <32 x i32>*, i32, <32 x i1>)
+
+define void @masked_store_v32i64(<32 x i64>* %val_ptr, <32 x i64>* %a, <32 x i64>* %m_ptr) nounwind {
+; RV32-LABEL: masked_store_v32i64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    addi sp, sp, -16
+; RV32-NEXT:    csrr a3, vlenb
+; RV32-NEXT:    slli a3, a3, 4
+; RV32-NEXT:    sub sp, sp, a3
+; RV32-NEXT:    addi a3, a2, 128
+; RV32-NEXT:    vsetivli a4, 16, e64,m8,ta,mu
+; RV32-NEXT:    vle64.v v8, (a3)
+; RV32-NEXT:    csrr a3, vlenb
+; RV32-NEXT:    slli a3, a3, 3
+; RV32-NEXT:    add a3, sp, a3
+; RV32-NEXT:    addi a3, a3, 16
+; RV32-NEXT:    vs8r.v v8, (a3) # Unknown-size Folded Spill
+; RV32-NEXT:    vle64.v v16, (a2)
+; RV32-NEXT:    addi a2, zero, 32
+; RV32-NEXT:    vsetvli a2, a2, e32,m8,ta,mu
+; RV32-NEXT:    vmv.v.i v24, 0
+; RV32-NEXT:    vsetivli a2, 16, e64,m8,ta,mu
+; RV32-NEXT:    vmseq.vv v1, v16, v24
+; RV32-NEXT:    addi a2, a0, 128
+; RV32-NEXT:    vle64.v v16, (a2)
+; RV32-NEXT:    vle64.v v8, (a0)
+; RV32-NEXT:    addi a0, sp, 16
+; RV32-NEXT:    vs8r.v v8, (a0) # Unknown-size Folded Spill
+; RV32-NEXT:    csrr a0, vlenb
+; RV32-NEXT:    slli a0, a0, 3
+; RV32-NEXT:    add a0, sp, a0
+; RV32-NEXT:    addi a0, a0, 16
+; RV32-NEXT:    vl8re8.v v8, (a0) # Unknown-size Folded Reload
+; RV32-NEXT:    vmseq.vv v0, v8, v24
+; RV32-NEXT:    addi a0, a1, 128
+; RV32-NEXT:    vse64.v v16, (a0), v0.t
+; RV32-NEXT:    vmv1r.v v0, v1
+; RV32-NEXT:    addi a0, sp, 16
+; RV32-NEXT:    vl8re8.v v8, (a0) # Unknown-size Folded Reload
+; RV32-NEXT:    vse64.v v8, (a1), v0.t
+; RV32-NEXT:    csrr a0, vlenb
+; RV32-NEXT:    slli a0, a0, 4
+; RV32-NEXT:    add sp, sp, a0
+; RV32-NEXT:    addi sp, sp, 16
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: masked_store_v32i64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    vsetivli a3, 16, e64,m8,ta,mu
+; RV64-NEXT:    addi a3, a2, 128
+; RV64-NEXT:    vle64.v v8, (a3)
+; RV64-NEXT:    vle64.v v16, (a2)
+; RV64-NEXT:    vmseq.vi v1, v8, 0
+; RV64-NEXT:    vle64.v v8, (a0)
+; RV64-NEXT:    addi a0, a0, 128
+; RV64-NEXT:    vle64.v v24, (a0)
+; RV64-NEXT:    vmseq.vi v0, v16, 0
+; RV64-NEXT:    vse64.v v8, (a1), v0.t
+; RV64-NEXT:    addi a0, a1, 128
+; RV64-NEXT:    vmv1r.v v0, v1
+; RV64-NEXT:    vse64.v v24, (a0), v0.t
+; RV64-NEXT:    ret
+  %m = load <32 x i64>, <32 x i64>* %m_ptr
+  %mask = icmp eq <32 x i64> %m, zeroinitializer
+  %val = load <32 x i64>, <32 x i64>* %val_ptr
+  call void @llvm.masked.store.v32i64.p0v32i64(<32 x i64> %val, <32 x i64>* %a, i32 8, <32 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v32i64.p0v32i64(<32 x i64>, <32 x i64>*, i32, <32 x i1>)
+
+define void @masked_store_v64i8(<64 x i8>* %val_ptr, <64 x i8>* %a, <64 x i8>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v64i8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, zero, 64
+; CHECK-NEXT:    vsetvli a3, a3, e8,m4,ta,mu
+; CHECK-NEXT:    vle8.v v28, (a2)
+; CHECK-NEXT:    vle8.v v8, (a0)
+; CHECK-NEXT:    vmseq.vi v0, v28, 0
+; CHECK-NEXT:    vse8.v v8, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <64 x i8>, <64 x i8>* %m_ptr
+  %mask = icmp eq <64 x i8> %m, zeroinitializer
+  %val = load <64 x i8>, <64 x i8>* %val_ptr
+  call void @llvm.masked.store.v64i8.p0v64i8(<64 x i8> %val, <64 x i8>* %a, i32 8, <64 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v64i8.p0v64i8(<64 x i8>, <64 x i8>*, i32, <64 x i1>)
+
+define void @masked_store_v64i16(<64 x i16>* %val_ptr, <64 x i16>* %a, <64 x i16>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v64i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, zero, 64
+; CHECK-NEXT:    vsetvli a3, a3, e16,m8,ta,mu
+; CHECK-NEXT:    vle16.v v8, (a2)
+; CHECK-NEXT:    vle16.v v16, (a0)
+; CHECK-NEXT:    vmseq.vi v0, v8, 0
+; CHECK-NEXT:    vse16.v v16, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <64 x i16>, <64 x i16>* %m_ptr
+  %mask = icmp eq <64 x i16> %m, zeroinitializer
+  %val = load <64 x i16>, <64 x i16>* %val_ptr
+  call void @llvm.masked.store.v64i16.p0v64i16(<64 x i16> %val, <64 x i16>* %a, i32 8, <64 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v64i16.p0v64i16(<64 x i16>, <64 x i16>*, i32, <64 x i1>)
+
+define void @masked_store_v64i32(<64 x i32>* %val_ptr, <64 x i32>* %a, <64 x i32>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v64i32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, zero, 32
+; CHECK-NEXT:    vsetvli a3, a3, e32,m8,ta,mu
+; CHECK-NEXT:    addi a3, a2, 128
+; CHECK-NEXT:    vle32.v v8, (a3)
+; CHECK-NEXT:    vle32.v v16, (a2)
+; CHECK-NEXT:    vmseq.vi v1, v8, 0
+; CHECK-NEXT:    vle32.v v8, (a0)
+; CHECK-NEXT:    addi a0, a0, 128
+; CHECK-NEXT:    vle32.v v24, (a0)
+; CHECK-NEXT:    vmseq.vi v0, v16, 0
+; CHECK-NEXT:    vse32.v v8, (a1), v0.t
+; CHECK-NEXT:    addi a0, a1, 128
+; CHECK-NEXT:    vmv1r.v v0, v1
+; CHECK-NEXT:    vse32.v v24, (a0), v0.t
+; CHECK-NEXT:    ret
+  %m = load <64 x i32>, <64 x i32>* %m_ptr
+  %mask = icmp eq <64 x i32> %m, zeroinitializer
+  %val = load <64 x i32>, <64 x i32>* %val_ptr
+  call void @llvm.masked.store.v64i32.p0v64i32(<64 x i32> %val, <64 x i32>* %a, i32 8, <64 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v64i32.p0v64i32(<64 x i32>, <64 x i32>*, i32, <64 x i1>)
+
+define void @masked_store_v128i8(<128 x i8>* %val_ptr, <128 x i8>* %a, <128 x i8>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v128i8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, zero, 128
+; CHECK-NEXT:    vsetvli a3, a3, e8,m8,ta,mu
+; CHECK-NEXT:    vle8.v v8, (a2)
+; CHECK-NEXT:    vle8.v v16, (a0)
+; CHECK-NEXT:    vmseq.vi v0, v8, 0
+; CHECK-NEXT:    vse8.v v16, (a1), v0.t
+; CHECK-NEXT:    ret
+  %m = load <128 x i8>, <128 x i8>* %m_ptr
+  %mask = icmp eq <128 x i8> %m, zeroinitializer
+  %val = load <128 x i8>, <128 x i8>* %val_ptr
+  call void @llvm.masked.store.v128i8.p0v128i8(<128 x i8> %val, <128 x i8>* %a, i32 8, <128 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v128i8.p0v128i8(<128 x i8>, <128 x i8>*, i32, <128 x i1>)
+
+define void @masked_store_v128i16(<128 x i16>* %val_ptr, <128 x i16>* %a, <128 x i16>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v128i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, zero, 64
+; CHECK-NEXT:    vsetvli a3, a3, e16,m8,ta,mu
+; CHECK-NEXT:    addi a3, a2, 128
+; CHECK-NEXT:    vle16.v v8, (a3)
+; CHECK-NEXT:    vle16.v v16, (a2)
+; CHECK-NEXT:    vmseq.vi v1, v8, 0
+; CHECK-NEXT:    vle16.v v8, (a0)
+; CHECK-NEXT:    addi a0, a0, 128
+; CHECK-NEXT:    vle16.v v24, (a0)
+; CHECK-NEXT:    vmseq.vi v0, v16, 0
+; CHECK-NEXT:    vse16.v v8, (a1), v0.t
+; CHECK-NEXT:    addi a0, a1, 128
+; CHECK-NEXT:    vmv1r.v v0, v1
+; CHECK-NEXT:    vse16.v v24, (a0), v0.t
+; CHECK-NEXT:    ret
+  %m = load <128 x i16>, <128 x i16>* %m_ptr
+  %mask = icmp eq <128 x i16> %m, zeroinitializer
+  %val = load <128 x i16>, <128 x i16>* %val_ptr
+  call void @llvm.masked.store.v128i16.p0v128i16(<128 x i16> %val, <128 x i16>* %a, i32 8, <128 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v128i16.p0v128i16(<128 x i16>, <128 x i16>*, i32, <128 x i1>)
+
+define void @masked_store_v256i8(<256 x i8>* %val_ptr, <256 x i8>* %a, <256 x i8>* %m_ptr) nounwind {
+; CHECK-LABEL: masked_store_v256i8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a3, zero, 128
+; CHECK-NEXT:    vsetvli a3, a3, e8,m8,ta,mu
+; CHECK-NEXT:    addi a3, a2, 128
+; CHECK-NEXT:    vle8.v v8, (a3)
+; CHECK-NEXT:    vle8.v v16, (a2)
+; CHECK-NEXT:    vmseq.vi v1, v8, 0
+; CHECK-NEXT:    vle8.v v8, (a0)
+; CHECK-NEXT:    addi a0, a0, 128
+; CHECK-NEXT:    vle8.v v24, (a0)
+; CHECK-NEXT:    vmseq.vi v0, v16, 0
+; CHECK-NEXT:    vse8.v v8, (a1), v0.t
+; CHECK-NEXT:    addi a0, a1, 128
+; CHECK-NEXT:    vmv1r.v v0, v1
+; CHECK-NEXT:    vse8.v v24, (a0), v0.t
+; CHECK-NEXT:    ret
+  %m = load <256 x i8>, <256 x i8>* %m_ptr
+  %mask = icmp eq <256 x i8> %m, zeroinitializer
+  %val = load <256 x i8>, <256 x i8>* %val_ptr
+  call void @llvm.masked.store.v256i8.p0v256i8(<256 x i8> %val, <256 x i8>* %a, i32 8, <256 x i1> %mask)
+  ret void
+}
+declare void @llvm.masked.store.v256i8.p0v256i8(<256 x i8>, <256 x i8>*, i32, <256 x i1>)