Implementation of expandload and compressstore intrinsics.

[delena]

Summary

Includes syntax, DAG builder, operation lowering and tests.
Does not include:

• handling of illegal data types
• codegen prepare pass
• cost model

[ABataev]

../lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

3698

sdl->Sdl

3859

sdl->Sdl

../lib/IR/Function.cpp

1084

auto *ReferenceType, and this line seems not formatted properly

1086

auto *ThisArgType

../test/CodeGen/X86/compress_expand.ll

6–9

Do you really need these lines?

[delena]

../lib/IR/Function.cpp

1086

I'll align the code.

../test/CodeGen/X86/compress_expand.ll

6–9

Yes, on Windows %rdi will be replaced with %rcx, and some other patterns may fail on mac. We always define triple.

[delena]

updated according to Alexey's comments

[igorb]

Hello Elena,
Would you consider to merge visitMaskedLoad(I) / visitMaskedExpandLoad and visitMaskedStore(I) / visitMaskedCompressStore ?

LGTM.

[delena]

In D26129#584495, @igorb wrote:

Hello Elena,
Would you consider to merge visitMaskedLoad(I) / visitMaskedExpandLoad and visitMaskedStore(I) / visitMaskedCompressStore ?

Maked load has one more argument - alignment. And it is positioned in the middle. I'll check..

[delena]

I included compressing-store visitor in the general masked-store, the same for expanding load as Igor suggested.
Igor, please look and the new code. Does it look better than the previous?

[igorb]

In D26129#586127, @delena wrote:

I included compressing-store visitor in the general masked-store, the same for expanding load as Igor suggested.
Igor, please look and the new code. Does it look better than the previous?

Hi Elena , I believe so .
Thanks!

[RKSimon]

Closed by rL285876

../include/llvm/IR/Intrinsics.h

#undef GET_INTRINSIC_ENUM_VALUES
 
   /// This is a type descriptor which explains the type requirements of an
   /// intrinsic. This is returned by getIntrinsicInfoTableEntries.
   struct IITDescriptor {
     enum IITDescriptorKind {
       Void, VarArg, MMX, Token, Metadata, Half, Float, Double,
       Integer, Vector, Pointer, Struct,
       Argument, ExtendArgument, TruncArgument, HalfVecArgument,
-      SameVecWidthArgument, PtrToArgument, VecOfPtrsToElt
+      SameVecWidthArgument, PtrToArgument, PtrToElt, VecOfPtrsToElt
     } Kind;
 
     union {
       unsigned Integer_Width;
       unsigned Float_Width;
       unsigned Vector_Width;
       unsigned Pointer_AddressSpace;
       unsigned Struct_NumElements;
       unsigned Argument_Info;
     };
 
     enum ArgKind {
       AK_Any,
       AK_AnyInteger,
       AK_AnyFloat,
       AK_AnyVector,
       AK_AnyPointer
     };
     unsigned getArgumentNumber() const {
       assert(Kind == Argument || Kind == ExtendArgument ||
              Kind == TruncArgument || Kind == HalfVecArgument ||
              Kind == SameVecWidthArgument || Kind == PtrToArgument ||
-             Kind == VecOfPtrsToElt);
+             Kind == PtrToElt || Kind == VecOfPtrsToElt);
       return Argument_Info >> 3;
     }
     ArgKind getArgumentKind() const {
       assert(Kind == Argument || Kind == ExtendArgument ||
              Kind == TruncArgument || Kind == HalfVecArgument ||
              Kind == SameVecWidthArgument || Kind == PtrToArgument ||
              Kind == VecOfPtrsToElt);
       return (ArgKind)(Argument_Info & 7);

../include/llvm/IR/Intrinsics.td

 // the intrinsic is overloaded, so the matched type should be declared as iAny.
 class LLVMExtendedType<int num> : LLVMMatchType<num>;
 class LLVMTruncatedType<int num> : LLVMMatchType<num>;
 class LLVMVectorSameWidth<int num, LLVMType elty>
   : LLVMMatchType<num> {
   ValueType ElTy = elty.VT;
 }
 class LLVMPointerTo<int num> : LLVMMatchType<num>;
+class LLVMPointerToElt<int num> : LLVMMatchType<num>;
 class LLVMVectorOfPointersToElt<int num> : LLVMMatchType<num>;
 
 // Match the type of another intrinsic parameter that is expected to be a
 // vector type, but change the element count to be half as many
 class LLVMHalfElementsVectorType<int num> : LLVMMatchType<num>;
 
 def llvm_void_ty       : LLVMType<isVoid>;
 def llvm_any_ty        : LLVMType<Any>;
def int_masked_load  : Intrinsic<[llvm_anyvector_ty],
                                  [LLVMAnyPointerType<LLVMMatchType<0>>, llvm_i32_ty,
                                   LLVMVectorSameWidth<0, llvm_i1_ty>, LLVMMatchType<0>],
                                  [IntrReadMem, IntrArgMemOnly]>;
 
 def int_masked_gather: Intrinsic<[llvm_anyvector_ty],
                                  [LLVMVectorOfPointersToElt<0>, llvm_i32_ty,
                                   LLVMVectorSameWidth<0, llvm_i1_ty>,
                                   LLVMMatchType<0>],
-          [IntrReadMem]>;
+                                 [IntrReadMem]>;
 
 def int_masked_scatter: Intrinsic<[],
                                   [llvm_anyvector_ty,
                                    LLVMVectorOfPointersToElt<0>, llvm_i32_ty,
                                    LLVMVectorSameWidth<0, llvm_i1_ty>]>;
 
+def int_masked_expandload: Intrinsic<[llvm_anyvector_ty],
+                                     [LLVMPointerToElt<0>,
+                                      LLVMVectorSameWidth<0, llvm_i1_ty>,
+                                      LLVMMatchType<0>],
+                                     [IntrReadMem]>;
+
+def int_masked_compressstore: Intrinsic<[],
+                                     [llvm_anyvector_ty,
+                                      LLVMPointerToElt<0>,
+                                      LLVMVectorSameWidth<0, llvm_i1_ty>],
+                                     [IntrArgMemOnly]>;
+
 // Test whether a pointer is associated with a type metadata identifier.
 def int_type_test : Intrinsic<[llvm_i1_ty], [llvm_ptr_ty, llvm_metadata_ty],
                               [IntrNoMem]>;
 
 // Safely loads a function pointer from a virtual table pointer using type metadata.
 def int_type_checked_load : Intrinsic<[llvm_ptr_ty, llvm_i1_ty],
                                       [llvm_ptr_ty, llvm_i32_ty, llvm_metadata_ty],
                                       [IntrNoMem]>;

../lib/CodeGen/SelectionDAG/DAGCombiner.cpp

if (Mask.getOpcode() == ISD::SETCC) {
     std::tie(DataLo, DataHi) = DAG.SplitVector(Data, DL);
 
     MachineMemOperand *MMO = DAG.getMachineFunction().
       getMachineMemOperand(MST->getPointerInfo(),
                            MachineMemOperand::MOStore,  LoMemVT.getStoreSize(),
                            Alignment, MST->getAAInfo(), MST->getRanges());
 
     Lo = DAG.getMaskedStore(Chain, DL, DataLo, Ptr, MaskLo, LoMemVT, MMO,
-                            MST->isTruncatingStore());
+                            MST->isTruncatingStore(), MST->isCompressingStore());
 
     unsigned IncrementSize = LoMemVT.getSizeInBits()/8;
     Ptr = DAG.getNode(ISD::ADD, DL, Ptr.getValueType(), Ptr,
                       DAG.getConstant(IncrementSize, DL, Ptr.getValueType()));
 
     MMO = DAG.getMachineFunction().
       getMachineMemOperand(MST->getPointerInfo(),
                            MachineMemOperand::MOStore,  HiMemVT.getStoreSize(),
                            SecondHalfAlignment, MST->getAAInfo(),
                            MST->getRanges());
 
     Hi = DAG.getMaskedStore(Chain, DL, DataHi, Ptr, MaskHi, HiMemVT, MMO,
-                            MST->isTruncatingStore());
+                            MST->isTruncatingStore(), MST->isCompressingStore());
 
     AddToWorklist(Lo.getNode());
     AddToWorklist(Hi.getNode());
 
     return DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Lo, Hi);
   }
   return SDValue();
 }

../lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp

if (OpNo == 2) {
     assert(OpNo == 3 && "Unexpected operand for promotion");
     DataOp = GetPromotedInteger(DataOp);
     Mask = PromoteTargetBoolean(Mask, DataOp.getValueType());
     TruncateStore = true;
   }
 
   return DAG.getMaskedStore(N->getChain(), dl, DataOp, N->getBasePtr(), Mask,
                             N->getMemoryVT(), N->getMemOperand(),
-                            TruncateStore);
+                            TruncateStore, N->isCompressingStore());
 }
 
 SDValue DAGTypeLegalizer::PromoteIntOp_MLOAD(MaskedLoadSDNode *N,
                                              unsigned OpNo) {
   assert(OpNo == 2 && "Only know how to promote the mask!");
   EVT DataVT = N->getValueType(0);
   SDValue Mask = PromoteTargetBoolean(N->getOperand(OpNo), DataVT);
   SmallVector<SDValue, 4> NewOps(N->op_begin(), N->op_end());

../lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h

private:
 
   void visitAlloca(const AllocaInst &I);
   void visitLoad(const LoadInst &I);
   void visitStore(const StoreInst &I);
   void visitMaskedLoad(const CallInst &I);
   void visitMaskedStore(const CallInst &I);
   void visitMaskedGather(const CallInst &I);
   void visitMaskedScatter(const CallInst &I);
+  void visitMaskedExpandLoad(const CallInst &I);
+  void visitMaskedCompressStore(const CallInst &I);
   void visitAtomicCmpXchg(const AtomicCmpXchgInst &I);
   void visitAtomicRMW(const AtomicRMWInst &I);
   void visitFence(const FenceInst &I);
   void visitPHI(const PHINode &I);
   void visitCall(const CallInst &I);
   bool visitMemCmpCall(const CallInst &I);
   bool visitMemPCpyCall(const CallInst &I);
   bool visitMemChrCall(const CallInst &I);

../lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

void SelectionDAGBuilder::visitMaskedStore(const CallInst &I) {
   I.getAAMetadata(AAInfo);
 
   MachineMemOperand *MMO =
     DAG.getMachineFunction().
     getMachineMemOperand(MachinePointerInfo(PtrOperand),
                           MachineMemOperand::MOStore,  VT.getStoreSize(),
                           Alignment, AAInfo);
   SDValue StoreNode = DAG.getMaskedStore(getRoot(), sdl, Src0, Ptr, Mask, VT,
-                                         MMO, false);
+                                         MMO);
+  DAG.setRoot(StoreNode);
+  setValue(&I, StoreNode);
+}
+
+void SelectionDAGBuilder::visitMaskedCompressStore(const CallInst &I) {
+  SDLoc Sdl = getCurSDLoc();
+
+  // void llvm.masked.compressstore.*(Src0, Ptr, Mask)
+  Value  *PtrOperand = I.getArgOperand(1);
+  SDValue Ptr = getValue(PtrOperand);
+  SDValue Src0 = getValue(I.getArgOperand(0));
+  SDValue Mask = getValue(I.getArgOperand(2));
+  EVT VT = Src0.getValueType();
+  unsigned Alignment = DAG.getEVTAlignment(VT.getVectorElementType());
+
+  AAMDNodes AAInfo;
+  I.getAAMetadata(AAInfo);
+
+  MachineMemOperand *MMO =
+    DAG.getMachineFunction().
+    getMachineMemOperand(MachinePointerInfo(PtrOperand),
+                          MachineMemOperand::MOStore,  VT.getStoreSize(),
+                          Alignment, AAInfo);
+  SDValue StoreNode = DAG.getMaskedStore(getRoot(), Sdl, Src0, Ptr, Mask, VT,
+                                         MMO, false, /* truncating */
+                                         true /* compressing */);
   DAG.setRoot(StoreNode);
   setValue(&I, StoreNode);
 }
 
 // Get a uniform base for the Gather/Scatter intrinsic.
 // The first argument of the Gather/Scatter intrinsic is a vector of pointers.
 // We try to represent it as a base pointer + vector of indices.
 // Usually, the vector of pointers comes from a 'getelementptr' instruction.
void SelectionDAGBuilder::visitMaskedLoad(const CallInst &I) {
 
   MachineMemOperand *MMO =
     DAG.getMachineFunction().
     getMachineMemOperand(MachinePointerInfo(PtrOperand),
                           MachineMemOperand::MOLoad,  VT.getStoreSize(),
                           Alignment, AAInfo, Ranges);
 
   SDValue Load = DAG.getMaskedLoad(VT, sdl, InChain, Ptr, Mask, Src0, VT, MMO,
-                                   ISD::NON_EXTLOAD, false);
+                                   ISD::NON_EXTLOAD, false /* IsExpanding */);
+  if (AddToChain) {
+    SDValue OutChain = Load.getValue(1);
+    DAG.setRoot(OutChain);
+  }
+  setValue(&I, Load);
+}
+
+void SelectionDAGBuilder::visitMaskedExpandLoad(const CallInst &I) {
+  SDLoc Sdl = getCurSDLoc();
+
+  // @llvm.masked.expandload.*(Ptr, Mask, Src0)
+  Value  *PtrOperand = I.getArgOperand(0);
+  SDValue Ptr = getValue(PtrOperand);
+  SDValue Src0 = getValue(I.getArgOperand(2));
+  SDValue Mask = getValue(I.getArgOperand(1));
+
+  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+  EVT VT = TLI.getValueType(DAG.getDataLayout(), I.getType());
+  unsigned Alignment = DAG.getEVTAlignment(VT.getVectorElementType());
+
+  AAMDNodes AAInfo;
+  I.getAAMetadata(AAInfo);
+  const MDNode *Ranges = I.getMetadata(LLVMContext::MD_range);
+
+  // Do not serialize masked loads of constant memory with anything.
+  bool AddToChain = !AA->pointsToConstantMemory(MemoryLocation(
+      PtrOperand, DAG.getDataLayout().getTypeStoreSize(I.getType()), AAInfo));
+  SDValue InChain = AddToChain ? DAG.getRoot() : DAG.getEntryNode();
+
+  MachineMemOperand *MMO =
+    DAG.getMachineFunction().
+    getMachineMemOperand(MachinePointerInfo(PtrOperand),
+                          MachineMemOperand::MOLoad,  VT.getStoreSize(),
+                          Alignment, AAInfo, Ranges);
+
+  SDValue Load = DAG.getMaskedLoad(VT, Sdl, InChain, Ptr, Mask, Src0, VT, MMO,
+                                   ISD::NON_EXTLOAD, true /* IsExpanding */);
   if (AddToChain) {
     SDValue OutChain = Load.getValue(1);
     DAG.setRoot(OutChain);
   }
   setValue(&I, Load);
 }
 
 void SelectionDAGBuilder::visitMaskedGather(const CallInst &I) {
case Intrinsic::masked_load:
     visitMaskedLoad(I);
     return nullptr;
   case Intrinsic::masked_scatter:
     visitMaskedScatter(I);
     return nullptr;
   case Intrinsic::masked_store:
     visitMaskedStore(I);
     return nullptr;
+  case Intrinsic::masked_expandload:
+    visitMaskedExpandLoad(I);
+    return nullptr;
+  case Intrinsic::masked_compressstore:
+    visitMaskedCompressStore(I);
+    return nullptr;
   case Intrinsic::x86_mmx_pslli_w:
   case Intrinsic::x86_mmx_pslli_d:
   case Intrinsic::x86_mmx_pslli_q:
   case Intrinsic::x86_mmx_psrli_w:
   case Intrinsic::x86_mmx_psrli_d:
   case Intrinsic::x86_mmx_psrli_q:
   case Intrinsic::x86_mmx_psrai_w:
   case Intrinsic::x86_mmx_psrai_d: {

../lib/IR/Function.cpp

enum IIT_Info {
   IIT_EXTEND_ARG = 25,
   IIT_TRUNC_ARG = 26,
   IIT_ANYPTR = 27,
   IIT_V1   = 28,
   IIT_VARARG = 29,
   IIT_HALF_VEC_ARG = 30,
   IIT_SAME_VEC_WIDTH_ARG = 31,
   IIT_PTR_TO_ARG = 32,
-  IIT_VEC_OF_PTRS_TO_ELT = 33,
-  IIT_I128 = 34,
-  IIT_V512 = 35,
-  IIT_V1024 = 36
+  IIT_PTR_TO_ELT = 33,
+  IIT_VEC_OF_PTRS_TO_ELT = 34,
+  IIT_I128 = 35,
+  IIT_V512 = 36,
+  IIT_V1024 = 37
 };
 
 
 static void DecodeIITType(unsigned &NextElt, ArrayRef<unsigned char> Infos,
                       SmallVectorImpl<Intrinsic::IITDescriptor> &OutputTable) {
   IIT_Info Info = IIT_Info(Infos[NextElt++]);
   unsigned StructElts = 2;
   using namespace Intrinsic;
case IIT_SAME_VEC_WIDTH_ARG: {
     return;
   }
   case IIT_PTR_TO_ARG: {
     unsigned ArgInfo = (NextElt == Infos.size() ? 0 : Infos[NextElt++]);
     OutputTable.push_back(IITDescriptor::get(IITDescriptor::PtrToArgument,
                                              ArgInfo));
     return;
   }
+  case IIT_PTR_TO_ELT: {
+    unsigned ArgInfo = (NextElt == Infos.size() ? 0 : Infos[NextElt++]);
+    OutputTable.push_back(IITDescriptor::get(IITDescriptor::PtrToElt, ArgInfo));
+    return;
+  }
   case IIT_VEC_OF_PTRS_TO_ELT: {
     unsigned ArgInfo = (NextElt == Infos.size() ? 0 : Infos[NextElt++]);
     OutputTable.push_back(IITDescriptor::get(IITDescriptor::VecOfPtrsToElt,
                                              ArgInfo));
     return;
   }
   case IIT_EMPTYSTRUCT:
     OutputTable.push_back(IITDescriptor::get(IITDescriptor::Struct, 0));
if (VectorType *VTy = dyn_cast<VectorType>(Ty)) {
       return VectorType::get(EltTy, VTy->getNumElements());
     }
     llvm_unreachable("unhandled");
   }
   case IITDescriptor::PtrToArgument: {
     Type *Ty = Tys[D.getArgumentNumber()];
     return PointerType::getUnqual(Ty);
   }
+  case IITDescriptor::PtrToElt: {
+    Type *Ty = Tys[D.getArgumentNumber()];
+    VectorType *VTy = dyn_cast<VectorType>(Ty);
+    if (!VTy)
+      llvm_unreachable("Expected an argument of Vector Type");
+    Type *EltTy = VTy->getVectorElementType();
+    return PointerType::getUnqual(EltTy);
+  }
   case IITDescriptor::VecOfPtrsToElt: {
     Type *Ty = Tys[D.getArgumentNumber()];
     VectorType *VTy = dyn_cast<VectorType>(Ty);
     if (!VTy)
       llvm_unreachable("Expected an argument of Vector Type");
     Type *EltTy = VTy->getVectorElementType();
     return VectorType::get(PointerType::getUnqual(EltTy),
                            VTy->getNumElements());
case IITDescriptor::HalfVecArgument:
       return D.getArgumentNumber() >= ArgTys.size() ||
              !isa<VectorType>(ArgTys[D.getArgumentNumber()]) ||
              VectorType::getHalfElementsVectorType(
                      cast<VectorType>(ArgTys[D.getArgumentNumber()])) != Ty;
     case IITDescriptor::SameVecWidthArgument: {
       if (D.getArgumentNumber() >= ArgTys.size())
         return true;
       VectorType * ReferenceType =
-              dyn_cast<VectorType>(ArgTys[D.getArgumentNumber()]);
       VectorType *ThisArgType = dyn_cast<VectorType>(Ty);
       if (!ThisArgType || !ReferenceType ||
           (ReferenceType->getVectorNumElements() !=
            ThisArgType->getVectorNumElements()))
         return true;
       return matchIntrinsicType(ThisArgType->getVectorElementType(),
                                 Infos, ArgTys);
     }
     case IITDescriptor::PtrToArgument: {
       if (D.getArgumentNumber() >= ArgTys.size())
         return true;
       Type * ReferenceType = ArgTys[D.getArgumentNumber()];
       PointerType *ThisArgType = dyn_cast<PointerType>(Ty);
       return (!ThisArgType || ThisArgType->getElementType() != ReferenceType);
     }
+    case IITDescriptor::PtrToElt: {
+      if (D.getArgumentNumber() >= ArgTys.size())
+        return true;
+      VectorType * ReferenceType =
+        dyn_cast<VectorType> (ArgTys[D.getArgumentNumber()]);
+      PointerType *ThisArgType = dyn_cast<PointerType>(Ty);
+
+      return (!ThisArgType || !ReferenceType ||
+              ThisArgType->getElementType() != ReferenceType->getElementType());
+    }
     case IITDescriptor::VecOfPtrsToElt: {
       if (D.getArgumentNumber() >= ArgTys.size())
         return true;
       VectorType * ReferenceType =
               dyn_cast<VectorType> (ArgTys[D.getArgumentNumber()]);
       VectorType *ThisArgVecTy = dyn_cast<VectorType>(Ty);
       if (!ThisArgVecTy || !ReferenceType ||
           (ReferenceType->getVectorNumElements() !=

../lib/Target/X86/X86ISelLowering.cpp

if (Subtarget.hasVLX()){
       setTruncStoreAction(MVT::v8i32, MVT::v8i16, Legal);
 
       setTruncStoreAction(MVT::v2i64, MVT::v2i8,  Legal);
       setTruncStoreAction(MVT::v2i64, MVT::v2i16, Legal);
       setTruncStoreAction(MVT::v2i64, MVT::v2i32, Legal);
       setTruncStoreAction(MVT::v4i32, MVT::v4i8,  Legal);
       setTruncStoreAction(MVT::v4i32, MVT::v4i16, Legal);
     } else {
-      setOperationAction(ISD::MLOAD,    MVT::v8i32, Custom);
-      setOperationAction(ISD::MLOAD,    MVT::v8f32, Custom);
-      setOperationAction(ISD::MSTORE,   MVT::v8i32, Custom);
-      setOperationAction(ISD::MSTORE,   MVT::v8f32, Custom);
+      for (auto VT : {MVT::v4i32, MVT::v8i32, MVT::v2i64, MVT::v4i64,
+           MVT::v4f32, MVT::v8f32, MVT::v2f64, MVT::v4f64}) {
+        setOperationAction(ISD::MLOAD,  VT, Custom);
+        setOperationAction(ISD::MSTORE, VT, Custom);
+      }
     }
     setOperationAction(ISD::TRUNCATE,           MVT::i1, Custom);
     setOperationAction(ISD::TRUNCATE,           MVT::v16i8, Custom);
     setOperationAction(ISD::TRUNCATE,           MVT::v8i32, Custom);
     setOperationAction(ISD::VECTOR_SHUFFLE,     MVT::v8i1,  Custom);
     setOperationAction(ISD::VECTOR_SHUFFLE,     MVT::v16i1, Custom);
     setOperationAction(ISD::VSELECT,            MVT::v8i1,  Expand);
     setOperationAction(ISD::VSELECT,            MVT::v16i1, Expand);
static SDValue LowerMLOAD(SDValue Op, const X86Subtarget &Subtarget,
                           SelectionDAG &DAG) {
 
   MaskedLoadSDNode *N = cast<MaskedLoadSDNode>(Op.getNode());
   MVT VT = Op.getSimpleValueType();
   MVT ScalarVT = VT.getScalarType();
   SDValue Mask = N->getMask();
   SDLoc dl(Op);
 
+  assert((!N->isExpandingLoad() || Subtarget.hasAVX512()) &&
+         "Expanding masked load is supported on AVX-512 target only!");
+
+  assert((!N->isExpandingLoad() || ScalarVT.getSizeInBits() >= 32) &&
+         "Expanding masked load is supported for 32 and 64-bit types only!");
+
+  // 4x32, 4x64 and 2x64 vectors of non-expanding loads are legal regardless of
+  // VLX. These types for exp-loads are handled here.
+  if (!N->isExpandingLoad() && VT.getVectorNumElements() <= 4)
+    return Op;
+
   assert(Subtarget.hasAVX512() && !Subtarget.hasVLX() && !VT.is512BitVector() &&
          "Cannot lower masked load op.");
 
-  assert(((ScalarVT == MVT::i32 || ScalarVT == MVT::f32) ||
+  assert((ScalarVT.getSizeInBits() >= 32 ||
           (Subtarget.hasBWI() &&
               (ScalarVT == MVT::i8 || ScalarVT == MVT::i16))) &&
          "Unsupported masked load op.");
 
   // This operation is legal for targets with VLX, but without
   // VLX the vector should be widened to 512 bit
-  unsigned NumEltsInWideVec = 512/VT.getScalarSizeInBits();
+  unsigned NumEltsInWideVec = 512 / VT.getScalarSizeInBits();
   MVT WideDataVT = MVT::getVectorVT(ScalarVT, NumEltsInWideVec);
-  MVT WideMaskVT = MVT::getVectorVT(MVT::i1, NumEltsInWideVec);
   SDValue Src0 = N->getSrc0();
   Src0 = ExtendToType(Src0, WideDataVT, DAG);
+
+  // Mask element has to be i1 
+  MVT MaskEltTy = Mask.getSimpleValueType().getScalarType();
+  assert((MaskEltTy == MVT::i1 || VT.getVectorNumElements() <= 4) &&
+         "We handle 4x32, 4x64 and 2x64 vectors only in this casse");
+
+  MVT WideMaskVT = MVT::getVectorVT(MaskEltTy, NumEltsInWideVec);
+
   Mask = ExtendToType(Mask, WideMaskVT, DAG, true);
+  if (MaskEltTy != MVT::i1)
+    Mask = DAG.getNode(ISD::TRUNCATE, dl,
+                       MVT::getVectorVT(MVT::i1, NumEltsInWideVec), Mask);
   SDValue NewLoad = DAG.getMaskedLoad(WideDataVT, dl, N->getChain(),
                                       N->getBasePtr(), Mask, Src0,
                                       N->getMemoryVT(), N->getMemOperand(),
-                                      N->getExtensionType());
+                                      N->getExtensionType(),
+                                      N->isExpandingLoad());
 
   SDValue Exract = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT,
                                NewLoad.getValue(0),
                                DAG.getIntPtrConstant(0, dl));
   SDValue RetOps[] = {Exract, NewLoad.getValue(1)};
   return DAG.getMergeValues(RetOps, dl);
 }
 
 static SDValue LowerMSTORE(SDValue Op, const X86Subtarget &Subtarget,
                            SelectionDAG &DAG) {
   MaskedStoreSDNode *N = cast<MaskedStoreSDNode>(Op.getNode());
   SDValue DataToStore = N->getValue();
   MVT VT = DataToStore.getSimpleValueType();
   MVT ScalarVT = VT.getScalarType();
   SDValue Mask = N->getMask();
   SDLoc dl(Op);
 
+  assert((!N->isCompressingStore() || Subtarget.hasAVX512()) &&
+         "Expanding masked load is supported on AVX-512 target only!");
+
+  assert((!N->isCompressingStore() || ScalarVT.getSizeInBits() >= 32) &&
+         "Expanding masked load is supported for 32 and 64-bit types only!");
+
+  // 4x32 and 2x64 vectors of non-compressing stores are legal regardless to VLX.
+  if (!N->isCompressingStore() && VT.getVectorNumElements() <= 4)
+    return Op;
+
   assert(Subtarget.hasAVX512() && !Subtarget.hasVLX() && !VT.is512BitVector() &&
          "Cannot lower masked store op.");
 
-  assert(((ScalarVT == MVT::i32 || ScalarVT == MVT::f32) ||
+  assert((ScalarVT.getSizeInBits() >= 32 ||
           (Subtarget.hasBWI() &&
               (ScalarVT == MVT::i8 || ScalarVT == MVT::i16))) &&
           "Unsupported masked store op.");
 
   // This operation is legal for targets with VLX, but without
   // VLX the vector should be widened to 512 bit
   unsigned NumEltsInWideVec = 512/VT.getScalarSizeInBits();
   MVT WideDataVT = MVT::getVectorVT(ScalarVT, NumEltsInWideVec);
-  MVT WideMaskVT = MVT::getVectorVT(MVT::i1, NumEltsInWideVec);
+
+  // Mask element has to be i1 
+  MVT MaskEltTy = Mask.getSimpleValueType().getScalarType();
+  assert((MaskEltTy == MVT::i1 || VT.getVectorNumElements() <= 4) &&
+         "We handle 4x32, 4x64 and 2x64 vectors only in this casse");
+
+  MVT WideMaskVT = MVT::getVectorVT(MaskEltTy, NumEltsInWideVec);
+
   DataToStore = ExtendToType(DataToStore, WideDataVT, DAG);
   Mask = ExtendToType(Mask, WideMaskVT, DAG, true);
+  if (MaskEltTy != MVT::i1)
+    Mask = DAG.getNode(ISD::TRUNCATE, dl,
+                       MVT::getVectorVT(MVT::i1, NumEltsInWideVec), Mask);
   return DAG.getMaskedStore(N->getChain(), dl, DataToStore, N->getBasePtr(),
                             Mask, N->getMemoryVT(), N->getMemOperand(),
-                            N->isTruncatingStore());
+                            N->isTruncatingStore(), N->isCompressingStore());
 }
 
 static SDValue LowerMGATHER(SDValue Op, const X86Subtarget &Subtarget,
                             SelectionDAG &DAG) {
   assert(Subtarget.hasAVX512() &&
          "MGATHER/MSCATTER are supported on AVX-512 arch only");
 
   MaskedGatherSDNode *N = cast<MaskedGatherSDNode>(Op.getNode());
combineMaskedLoadConstantMask(MaskedLoadSDNode *ML, SelectionDAG &DAG,
 
   return DCI.CombineTo(ML, Blend, NewML.getValue(1), true);
 }
 
 static SDValue combineMaskedLoad(SDNode *N, SelectionDAG &DAG,
                                  TargetLowering::DAGCombinerInfo &DCI,
                                  const X86Subtarget &Subtarget) {
   MaskedLoadSDNode *Mld = cast<MaskedLoadSDNode>(N);
+
+  // TODO: Expanding load with constant mask may be optimized as well.
+  if (Mld->isExpandingLoad())
+    return SDValue();
+
   if (Mld->getExtensionType() == ISD::NON_EXTLOAD) {
     if (SDValue ScalarLoad = reduceMaskedLoadToScalarLoad(Mld, DAG, DCI))
       return ScalarLoad;
     // TODO: Do some AVX512 subsets benefit from this transform?
     if (!Subtarget.hasAVX512())
       if (SDValue Blend = combineMaskedLoadConstantMask(Mld, DAG, DCI))
         return Blend;
   }
static SDValue reduceMaskedStoreToScalarStore(MaskedStoreSDNode *MS,
   // Store that element at the appropriate offset from the base pointer.
   return DAG.getStore(MS->getChain(), DL, Extract, Addr, MS->getPointerInfo(),
                       Alignment, MS->getMemOperand()->getFlags());
 }
 
 static SDValue combineMaskedStore(SDNode *N, SelectionDAG &DAG,
                                   const X86Subtarget &Subtarget) {
   MaskedStoreSDNode *Mst = cast<MaskedStoreSDNode>(N);
+
+  if (Mst->isCompressingStore())
+    return SDValue();
+    
   if (!Mst->isTruncatingStore())
     return reduceMaskedStoreToScalarStore(Mst, DAG);
 
   // Resolve truncating stores.
   EVT VT = Mst->getValue().getValueType();
   unsigned NumElems = VT.getVectorNumElements();
   EVT StVT = Mst->getMemoryVT();
   SDLoc dl(Mst);

../lib/Target/X86/X86InstrFragmentsSIMD.td

 def X86mstore : PatFrag<(ops node:$src1, node:$src2, node:$src3),
                         (masked_store node:$src1, node:$src2, node:$src3), [{
   return (!cast<MaskedStoreSDNode>(N)->isTruncatingStore()) &&
          (!cast<MaskedStoreSDNode>(N)->isCompressingStore());
 }]>;
 
 def masked_store_aligned128 : PatFrag<(ops node:$src1, node:$src2, node:$src3),
                          (X86mstore node:$src1, node:$src2, node:$src3), [{
-  if (auto *Store = dyn_cast<MaskedStoreSDNode>(N))
-    return Store->getAlignment() >= 16;
-  return false;
+  return cast<MaskedStoreSDNode>(N)->getAlignment() >= 16;
 }]>;
 
 def masked_store_aligned256 : PatFrag<(ops node:$src1, node:$src2, node:$src3),
                          (X86mstore node:$src1, node:$src2, node:$src3), [{
-  if (auto *Store = dyn_cast<MaskedStoreSDNode>(N))
-    return Store->getAlignment() >= 32;
-  return false;
+  return cast<MaskedStoreSDNode>(N)->getAlignment() >= 32;
 }]>;
 
 def masked_store_aligned512 : PatFrag<(ops node:$src1, node:$src2, node:$src3),
                          (X86mstore node:$src1, node:$src2, node:$src3), [{
-  if (auto *Store = dyn_cast<MaskedStoreSDNode>(N))
-    return Store->getAlignment() >= 64;
-  return false;
+  return cast<MaskedStoreSDNode>(N)->getAlignment() >= 64;
 }]>;
 
 def masked_store_unaligned : PatFrag<(ops node:$src1, node:$src2, node:$src3),
-                         (X86mstore node:$src1, node:$src2, node:$src3), [{
-  return isa<MaskedStoreSDNode>(N);
+                         (masked_store node:$src1, node:$src2, node:$src3), [{
+  return (!cast<MaskedStoreSDNode>(N)->isTruncatingStore()) &&
+         (!cast<MaskedStoreSDNode>(N)->isCompressingStore());
 }]>;
 
 def X86mCompressingStore : PatFrag<(ops node:$src1, node:$src2, node:$src3),
                              (masked_store node:$src1, node:$src2, node:$src3), [{
     return cast<MaskedStoreSDNode>(N)->isCompressingStore();
 }]>;
 
 // masked truncstore fragments

../test/CodeGen/X86/compress_expand.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mattr=+avx512vl,+avx512dq,+avx512bw < %s | FileCheck %s --check-prefix=ALL --check-prefix=SKX
+; RUN: llc -mattr=+avx512f < %s | FileCheck %s --check-prefix=ALL --check-prefix=KNL
+
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+
+
+define <16 x float> @test1(float* %base) {
+; ALL-LABEL: test1:
+; ALL:       # BB#0:
+; ALL-NEXT:    movw $-2049, %ax # imm = 0xF7FF
+; ALL-NEXT:    kmovw %eax, %k1
+; ALL-NEXT:    vexpandps (%rdi), %zmm0 {%k1} {z}
+; ALL-NEXT:    retq
+  %res = call <16 x float> @llvm.masked.expandload.v16f32(float* %base, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 false, i1 true, i1 true, i1 true, i1 true>, <16 x float> undef)
+  ret <16 x float>%res
+}
+
+define <16 x float> @test2(float* %base, <16 x float> %src0) {
+; ALL-LABEL: test2:
+; ALL:       # BB#0:
+; ALL-NEXT:    movw $30719, %ax # imm = 0x77FF
+; ALL-NEXT:    kmovw %eax, %k1
+; ALL-NEXT:    vexpandps (%rdi), %zmm0 {%k1}
+; ALL-NEXT:    retq
+  %res = call <16 x float> @llvm.masked.expandload.v16f32(float* %base, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 false, i1 true, i1 true, i1 true, i1 false>, <16 x float> %src0)
+  ret <16 x float>%res
+}
+
+define <8 x double> @test3(double* %base, <8 x double> %src0, <8 x i1> %mask) {
+; SKX-LABEL: test3:
+; SKX:       # BB#0:
+; SKX-NEXT:    vpsllw $15, %xmm1, %xmm1
+; SKX-NEXT:    vpmovw2m %xmm1, %k1
+; SKX-NEXT:    vexpandpd (%rdi), %zmm0 {%k1}
+; SKX-NEXT:    retq
+;
+; KNL-LABEL: test3:
+; KNL:       # BB#0:
+; KNL-NEXT:    vpmovsxwq %xmm1, %zmm1
+; KNL-NEXT:    vpsllq $63, %zmm1, %zmm1
+; KNL-NEXT:    vptestmq %zmm1, %zmm1, %k1
+; KNL-NEXT:    vexpandpd (%rdi), %zmm0 {%k1}
+; KNL-NEXT:    retq
+  %res = call <8 x double> @llvm.masked.expandload.v8f64(double* %base, <8 x i1> %mask, <8 x double> %src0)
+  ret <8 x double>%res
+}
+
+define <4 x float> @test4(float* %base, <4 x float> %src0) {
+; SKX-LABEL: test4:
+; SKX:       # BB#0:
+; SKX-NEXT:    movb $7, %al
+; SKX-NEXT:    kmovb %eax, %k1
+; SKX-NEXT:    vexpandps (%rdi), %xmm0 {%k1}
+; SKX-NEXT:    retq
+;
+; KNL-LABEL: test4:
+; KNL:       # BB#0:
+; KNL-NEXT:    # kill: %XMM0<def> %XMM0<kill> %ZMM0<def>
+; KNL-NEXT:    movw $7, %ax
+; KNL-NEXT:    kmovw %eax, %k1
+; KNL-NEXT:    vexpandps (%rdi), %zmm0 {%k1}
+; KNL-NEXT:    # kill: %XMM0<def> %XMM0<kill> %ZMM0<kill>
+; KNL-NEXT:    retq
+  %res = call <4 x float> @llvm.masked.expandload.v4f32(float* %base, <4 x i1> <i1 true, i1 true, i1 true, i1 false>, <4 x float> %src0)
+  ret <4 x float>%res
+}
+
+define <2 x i64> @test5(i64* %base, <2 x i64> %src0) {
+; SKX-LABEL: test5:
+; SKX:       # BB#0:
+; SKX-NEXT:    movb $2, %al
+; SKX-NEXT:    kmovb %eax, %k1
+; SKX-NEXT:    vpexpandq (%rdi), %xmm0 {%k1}
+; SKX-NEXT:    retq
+;
+; KNL-LABEL: test5:
+; KNL:       # BB#0:
+; KNL-NEXT:    # kill: %XMM0<def> %XMM0<kill> %ZMM0<def>
+; KNL-NEXT:    movb $2, %al
+; KNL-NEXT:    kmovw %eax, %k1
+; KNL-NEXT:    vpexpandq (%rdi), %zmm0 {%k1}
+; KNL-NEXT:    # kill: %XMM0<def> %XMM0<kill> %ZMM0<kill>
+; KNL-NEXT:    retq
+  %res = call <2 x i64> @llvm.masked.expandload.v2i64(i64* %base, <2 x i1> <i1 false, i1 true>, <2 x i64> %src0)
+  ret <2 x i64>%res
+}
+
+declare <16 x float> @llvm.masked.expandload.v16f32(float*, <16 x i1>, <16 x float>)
+declare <8 x double> @llvm.masked.expandload.v8f64(double*, <8 x i1>, <8 x double>)
+declare <4 x float>  @llvm.masked.expandload.v4f32(float*, <4 x i1>, <4 x float>)
+declare <2 x i64>    @llvm.masked.expandload.v2i64(i64*, <2 x i1>, <2 x i64>)
+
+define void @test6(float* %base, <16 x float> %V) {
+; ALL-LABEL: test6:
+; ALL:       # BB#0:
+; ALL-NEXT:    movw $-2049, %ax # imm = 0xF7FF
+; ALL-NEXT:    kmovw %eax, %k1
+; ALL-NEXT:    vcompressps %zmm0, (%rdi) {%k1}
+; ALL-NEXT:    retq
+  call void @llvm.masked.compressstore.v16f32(<16 x float> %V, float* %base, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 false, i1 true, i1 true, i1 true, i1 true>)
+  ret void
+}
+
+define void @test7(float* %base, <8 x float> %V, <8 x i1> %mask) {
+; SKX-LABEL: test7:
+; SKX:       # BB#0:
+; SKX-NEXT:    vpsllw $15, %xmm1, %xmm1
+; SKX-NEXT:    vpmovw2m %xmm1, %k1
+; SKX-NEXT:    vcompressps %ymm0, (%rdi) {%k1}
+; SKX-NEXT:    retq
+;
+; KNL-LABEL: test7:
+; KNL:       # BB#0:
+; KNL-NEXT:    # kill: %YMM0<def> %YMM0<kill> %ZMM0<def>
+; KNL-NEXT:    vpmovsxwq %xmm1, %zmm1
+; KNL-NEXT:    vpsllq $63, %zmm1, %zmm1
+; KNL-NEXT:    vptestmq %zmm1, %zmm1, %k0
+; KNL-NEXT:    kshiftlw $8, %k0, %k0
+; KNL-NEXT:    kshiftrw $8, %k0, %k1
+; KNL-NEXT:    vcompressps %zmm0, (%rdi) {%k1}
+; KNL-NEXT:    retq
+  call void @llvm.masked.compressstore.v8f32(<8 x float> %V, float* %base, <8 x i1> %mask)
+  ret void
+}
+
+define void @test8(double* %base, <8 x double> %V, <8 x i1> %mask) {
+; SKX-LABEL: test8:
+; SKX:       # BB#0:
+; SKX-NEXT:    vpsllw $15, %xmm1, %xmm1
+; SKX-NEXT:    vpmovw2m %xmm1, %k1
+; SKX-NEXT:    vcompresspd %zmm0, (%rdi) {%k1}
+; SKX-NEXT:    retq
+;
+; KNL-LABEL: test8:
+; KNL:       # BB#0:
+; KNL-NEXT:    vpmovsxwq %xmm1, %zmm1
+; KNL-NEXT:    vpsllq $63, %zmm1, %zmm1
+; KNL-NEXT:    vptestmq %zmm1, %zmm1, %k1
+; KNL-NEXT:    vcompresspd %zmm0, (%rdi) {%k1}
+; KNL-NEXT:    retq
+  call void @llvm.masked.compressstore.v8f64(<8 x double> %V, double* %base, <8 x i1> %mask)
+  ret void
+}
+
+define void @test9(i64* %base, <8 x i64> %V, <8 x i1> %mask) {
+; SKX-LABEL: test9:
+; SKX:       # BB#0:
+; SKX-NEXT:    vpsllw $15, %xmm1, %xmm1
+; SKX-NEXT:    vpmovw2m %xmm1, %k1
+; SKX-NEXT:    vpcompressq %zmm0, (%rdi) {%k1}
+; SKX-NEXT:    retq
+;
+; KNL-LABEL: test9:
+; KNL:       # BB#0:
+; KNL-NEXT:    vpmovsxwq %xmm1, %zmm1
+; KNL-NEXT:    vpsllq $63, %zmm1, %zmm1
+; KNL-NEXT:    vptestmq %zmm1, %zmm1, %k1
+; KNL-NEXT:    vpcompressq %zmm0, (%rdi) {%k1}
+; KNL-NEXT:    retq
+    call void @llvm.masked.compressstore.v8i64(<8 x i64> %V, i64* %base, <8 x i1> %mask)
+  ret void
+}
+
+define void @test10(i64* %base, <4 x i64> %V, <4 x i1> %mask) {
+; SKX-LABEL: test10:
+; SKX:       # BB#0:
+; SKX-NEXT:    vpslld $31, %xmm1, %xmm1
+; SKX-NEXT:    vptestmd %xmm1, %xmm1, %k1
+; SKX-NEXT:    vpcompressq %ymm0, (%rdi) {%k1}
+; SKX-NEXT:    retq
+;
+; KNL-LABEL: test10:
+; KNL:       # BB#0:
+; KNL-NEXT:    # kill: %YMM0<def> %YMM0<kill> %ZMM0<def>
+; KNL-NEXT:    vpslld $31, %xmm1, %xmm1
+; KNL-NEXT:    vpsrad $31, %xmm1, %xmm1
+; KNL-NEXT:    vpmovsxdq %xmm1, %ymm1
+; KNL-NEXT:    vpxord %zmm2, %zmm2, %zmm2
+; KNL-NEXT:    vinserti64x4 $0, %ymm1, %zmm2, %zmm1
+; KNL-NEXT:    vpsllq $63, %zmm1, %zmm1
+; KNL-NEXT:    vptestmq %zmm1, %zmm1, %k1
+; KNL-NEXT:    vpcompressq %zmm0, (%rdi) {%k1}
+; KNL-NEXT:    retq
+    call void @llvm.masked.compressstore.v4i64(<4 x i64> %V, i64* %base, <4 x i1> %mask)
+  ret void
+}
+
+define void @test11(i64* %base, <2 x i64> %V, <2 x i1> %mask) {
+; SKX-LABEL: test11:
+; SKX:       # BB#0:
+; SKX-NEXT:    vpsllq $63, %xmm1, %xmm1
+; SKX-NEXT:    vptestmq %xmm1, %xmm1, %k1
+; SKX-NEXT:    vpcompressq %xmm0, (%rdi) {%k1}
+; SKX-NEXT:    retq
+;
+; KNL-LABEL: test11:
+; KNL:       # BB#0:
+; KNL-NEXT:    # kill: %XMM0<def> %XMM0<kill> %ZMM0<def>
+; KNL-NEXT:    vpsllq $63, %xmm1, %xmm1
+; KNL-NEXT:    vpsrad $31, %xmm1, %xmm1
+; KNL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; KNL-NEXT:    vpxord %zmm2, %zmm2, %zmm2
+; KNL-NEXT:    vinserti32x4 $0, %xmm1, %zmm2, %zmm1
+; KNL-NEXT:    vpsllq $63, %zmm1, %zmm1
+; KNL-NEXT:    vptestmq %zmm1, %zmm1, %k1
+; KNL-NEXT:    vpcompressq %zmm0, (%rdi) {%k1}
+; KNL-NEXT:    retq
+    call void @llvm.masked.compressstore.v2i64(<2 x i64> %V, i64* %base, <2 x i1> %mask)
+  ret void
+}
+
+define void @test12(float* %base, <4 x float> %V, <4 x i1> %mask) {
+; SKX-LABEL: test12:
+; SKX:       # BB#0:
+; SKX-NEXT:    vpslld $31, %xmm1, %xmm1
+; SKX-NEXT:    vptestmd %xmm1, %xmm1, %k1
+; SKX-NEXT:    vcompressps %xmm0, (%rdi) {%k1}
+; SKX-NEXT:    retq
+;
+; KNL-LABEL: test12:
+; KNL:       # BB#0:
+; KNL-NEXT:    # kill: %XMM0<def> %XMM0<kill> %ZMM0<def>
+; KNL-NEXT:    vpslld $31, %xmm1, %xmm1
+; KNL-NEXT:    vpsrad $31, %xmm1, %xmm1
+; KNL-NEXT:    vpxord %zmm2, %zmm2, %zmm2
+; KNL-NEXT:    vinserti32x4 $0, %xmm1, %zmm2, %zmm1
+; KNL-NEXT:    vpslld $31, %zmm1, %zmm1
+; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k1
+; KNL-NEXT:    vcompressps %zmm0, (%rdi) {%k1}
+; KNL-NEXT:    retq
+    call void @llvm.masked.compressstore.v4f32(<4 x float> %V, float* %base, <4 x i1> %mask)
+  ret void
+}
+
+declare void @llvm.masked.compressstore.v16f32(<16 x float>, float* , <16 x i1>)
+declare void @llvm.masked.compressstore.v8f32(<8 x float>, float* , <8 x i1>)
+declare void @llvm.masked.compressstore.v8f64(<8 x double>, double* , <8 x i1>)
+declare void @llvm.masked.compressstore.v16i32(<16 x i32>, i32* , <16 x i1>)
+declare void @llvm.masked.compressstore.v8i32(<8 x i32>, i32* , <8 x i1>)
+declare void @llvm.masked.compressstore.v8i64(<8 x i64>, i64* , <8 x i1>)
+declare void @llvm.masked.compressstore.v4i32(<4 x i32>, i32* , <4 x i1>)
+declare void @llvm.masked.compressstore.v4f32(<4 x float>, float* , <4 x i1>)
+declare void @llvm.masked.compressstore.v4i64(<4 x i64>, i64* , <4 x i1>)
+declare void @llvm.masked.compressstore.v2i64(<2 x i64>, i64* , <2 x i1>)

../utils/TableGen/CodeGenTarget.cpp

if (TyEl->isSubClassOf("LLVMMatchType")) {
       assert(MatchTy < OverloadedVTs.size() &&
              "Invalid matching number!");
       VT = OverloadedVTs[MatchTy];
       // It only makes sense to use the extended and truncated vector element
       // variants with iAny types; otherwise, if the intrinsic is not
       // overloaded, all the types can be specified directly.
       assert(((!TyEl->isSubClassOf("LLVMExtendedType") &&
                !TyEl->isSubClassOf("LLVMTruncatedType") &&
-               !TyEl->isSubClassOf("LLVMVectorSameWidth") &&
-               !TyEl->isSubClassOf("LLVMPointerToElt")) ||
+               !TyEl->isSubClassOf("LLVMVectorSameWidth")) ||
               VT == MVT::iAny || VT == MVT::vAny) &&
              "Expected iAny or vAny type");
     } else
       VT = getValueType(TyEl->getValueAsDef("VT"));
 
     if (MVT(VT).isOverloaded()) {
       OverloadedVTs.push_back(VT);
       isOverloaded = true;

../utils/TableGen/IntrinsicEmitter.cpp

enum IIT_Info {
   IIT_EXTEND_ARG = 25,
   IIT_TRUNC_ARG = 26,
   IIT_ANYPTR = 27,
   IIT_V1   = 28,
   IIT_VARARG = 29,
   IIT_HALF_VEC_ARG = 30,
   IIT_SAME_VEC_WIDTH_ARG = 31,
   IIT_PTR_TO_ARG = 32,
-  IIT_VEC_OF_PTRS_TO_ELT = 33,
-  IIT_I128 = 34,
-  IIT_V512 = 35,
-  IIT_V1024 = 36
+  IIT_PTR_TO_ELT = 33,
+  IIT_VEC_OF_PTRS_TO_ELT = 34,
+  IIT_I128 = 35,
+  IIT_V512 = 36,
+  IIT_V1024 = 37
 };
 
 
 static void EncodeFixedValueType(MVT::SimpleValueType VT,
                                  std::vector<unsigned char> &Sig) {
   if (MVT(VT).isInteger()) {
     unsigned BitWidth = MVT(VT).getSizeInBits();
     switch (BitWidth) {
else if (R->isSubClassOf("LLVMVectorSameWidth")) {
       MVT::SimpleValueType VT = getValueType(R->getValueAsDef("ElTy"));
       EncodeFixedValueType(VT, Sig);
       return;
     }
     else if (R->isSubClassOf("LLVMPointerTo"))
       Sig.push_back(IIT_PTR_TO_ARG);
     else if (R->isSubClassOf("LLVMVectorOfPointersToElt"))
       Sig.push_back(IIT_VEC_OF_PTRS_TO_ELT);
+    else if (R->isSubClassOf("LLVMPointerToElt"))
+      Sig.push_back(IIT_PTR_TO_ELT);
     else
       Sig.push_back(IIT_ARG);
     return Sig.push_back((Number << 3) | ArgCodes[Number]);
   }
 
   MVT::SimpleValueType VT = getValueType(R->getValueAsDef("VT"));
 
   unsigned Tmp = 0;