Index: include/llvm/IR/DerivedTypes.h
===================================================================
--- include/llvm/IR/DerivedTypes.h
+++ include/llvm/IR/DerivedTypes.h
@@ -485,6 +485,38 @@
     return VectorType::get(EltTy, VTy->getElementCount());
   }
 
+  /// This static method is like getInteger except that the element types are
+  /// half as wide as the elements in the input type.
+  static VectorType *getNarrowerFpElementVectorType(VectorType *VTy) {
+    Type *EltTy;
+    switch(VTy->getElementType()->getTypeID()) {
+    case DoubleTyID:
+      EltTy = Type::getFloatTy(VTy->getContext());
+      break;
+    case FloatTyID:
+      EltTy = Type::getHalfTy(VTy->getContext());
+      break;
+    default:
+      assert(0 && "Cannot create narrower fp vector element type");
+      break;
+    }
+    return VectorType::get(EltTy, VTy->getElementCount());
+  }
+
+  // This static method returns a VectorType with a smaller number of elements
+  // of a larger type than the input element type. For example, a <16 x i8>
+  // subdivided twice would return <4 x i32>
+  static VectorType *getSubdividedVectorType(VectorType *VTy, int NumSubdivs) {
+    for (int i = 0; i < NumSubdivs; ++i) {
+      VTy = VectorType::getDoubleElementsVectorType(VTy);
+      if (VTy->getElementType()->isFloatingPointTy())
+        VTy = VectorType::getNarrowerFpElementVectorType(VTy);
+      else
+        VTy = VectorType::getTruncatedElementVectorType(VTy);
+    }
+    return VTy;
+  }
+
   /// This static method returns a VectorType with half as many elements as the
   /// input type and the same element type.
   static VectorType *getHalfElementsVectorType(VectorType *VTy) {
Index: include/llvm/IR/Intrinsics.h
===================================================================
--- include/llvm/IR/Intrinsics.h
+++ include/llvm/IR/Intrinsics.h
@@ -100,7 +100,8 @@
       Integer, Vector, Pointer, Struct,
       Argument, ExtendArgument, TruncArgument, HalfVecArgument,
       SameVecWidthArgument, PtrToArgument, PtrToElt, VecOfAnyPtrsToElt,
-      VecElementArgument, ScalableVecArgument
+      VecElementArgument, ScalableVecArgument, Subdivide2Argument,
+      Subdivide4Argument
     } Kind;
 
     union {
@@ -125,14 +126,16 @@
       assert(Kind == Argument || Kind == ExtendArgument ||
              Kind == TruncArgument || Kind == HalfVecArgument ||
              Kind == SameVecWidthArgument || Kind == PtrToArgument ||
-             Kind == PtrToElt || Kind == VecElementArgument);
+             Kind == PtrToElt || Kind == VecElementArgument ||
+             Kind == Subdivide2Argument || Kind == Subdivide4Argument);
       return Argument_Info >> 3;
     }
     ArgKind getArgumentKind() const {
       assert(Kind == Argument || Kind == ExtendArgument ||
              Kind == TruncArgument || Kind == HalfVecArgument ||
              Kind == SameVecWidthArgument || Kind == PtrToArgument ||
-             Kind == VecElementArgument);
+             Kind == VecElementArgument || Kind == Subdivide2Argument ||
+             Kind == Subdivide4Argument);
       return (ArgKind)(Argument_Info & 7);
     }
 
Index: include/llvm/IR/Intrinsics.td
===================================================================
--- include/llvm/IR/Intrinsics.td
+++ include/llvm/IR/Intrinsics.td
@@ -187,6 +187,12 @@
 // vector type, but change the element count to be half as many
 class LLVMHalfElementsVectorType<int num> : LLVMMatchType<num>;
 
+// Match the type of another intrinsic parameter that is expected to be a
+// vector type (i.e. <N x iM>) but with each element subdivided to
+// form a vector with more elements that are smaller than the original.
+class LLVMSubdivide2VectorType<int num> : LLVMMatchType<num>;
+class LLVMSubdivide4VectorType<int num> : LLVMMatchType<num>;
+
 def llvm_void_ty       : LLVMType<isVoid>;
 let isAny = 1 in {
   def llvm_any_ty        : LLVMType<Any>;
Index: lib/IR/Function.cpp
===================================================================
--- lib/IR/Function.cpp
+++ lib/IR/Function.cpp
@@ -703,7 +703,9 @@
   IIT_STRUCT8 = 40,
   IIT_F128 = 41,
   IIT_VEC_ELEMENT = 42,
-  IIT_SCALABLE_VEC = 43
+  IIT_SCALABLE_VEC = 43,
+  IIT_SUBDIVIDE2_ARG = 44,
+  IIT_SUBDIVIDE4_ARG = 45
 };
 
 static void DecodeIITType(unsigned &NextElt, ArrayRef<unsigned char> Infos,
@@ -868,6 +870,18 @@
       DecodeIITType(NextElt, Infos, OutputTable);
     return;
   }
+  case IIT_SUBDIVIDE2_ARG: {
+    unsigned ArgInfo = (NextElt == Infos.size() ? 0 : Infos[NextElt++]);
+    OutputTable.push_back(IITDescriptor::get(IITDescriptor::Subdivide2Argument,
+                                             ArgInfo));
+    return;
+  }
+  case IIT_SUBDIVIDE4_ARG: {
+    unsigned ArgInfo = (NextElt == Infos.size() ? 0 : Infos[NextElt++]);
+    OutputTable.push_back(IITDescriptor::get(IITDescriptor::Subdivide4Argument,
+                                             ArgInfo));
+    return;
+  }
   case IIT_VEC_ELEMENT: {
     unsigned ArgInfo = (NextElt == Infos.size() ? 0 : Infos[NextElt++]);
     OutputTable.push_back(IITDescriptor::get(IITDescriptor::VecElementArgument,
@@ -970,6 +984,18 @@
     assert(ITy->getBitWidth() % 2 == 0);
     return IntegerType::get(Context, ITy->getBitWidth() / 2);
   }
+  case IITDescriptor::Subdivide2Argument: {
+    Type *Ty = Tys[D.getArgumentNumber()];
+    if (VectorType *VTy = dyn_cast<VectorType>(Ty))
+      return VectorType::getSubdividedVectorType(VTy, 1);
+    llvm_unreachable("unhandled");
+  }
+  case IITDescriptor::Subdivide4Argument: {
+    Type *Ty = Tys[D.getArgumentNumber()];
+    if (VectorType *VTy = dyn_cast<VectorType>(Ty))
+      return VectorType::getSubdividedVectorType(VTy, 2);
+    llvm_unreachable("unhandled");
+  }
   case IITDescriptor::HalfVecArgument:
     return VectorType::getHalfElementsVectorType(cast<VectorType>(
                                                   Tys[D.getArgumentNumber()]));
@@ -1269,6 +1295,32 @@
       auto *ReferenceType = dyn_cast<VectorType>(ArgTys[D.getArgumentNumber()]);
       return !ReferenceType || Ty != ReferenceType->getElementType();
     }
+    case IITDescriptor::Subdivide2Argument: {
+      // This may only be used when referring to a previous vector argument.
+      if (D.getArgumentNumber() >= ArgTys.size())
+        return IsDeferredCheck || DeferCheck(Ty);
+
+      Type *NewTy = ArgTys[D.getArgumentNumber()];
+      if (VectorType *VTy = dyn_cast<VectorType>(NewTy))
+        NewTy = VectorType::getSubdividedVectorType(VTy, 1);
+      else
+        return true;
+
+      return Ty != NewTy;
+    }
+    case IITDescriptor::Subdivide4Argument: {
+      // This may only be used when referring to a previous vector argument.
+      if (D.getArgumentNumber() >= ArgTys.size())
+        return IsDeferredCheck || DeferCheck(Ty);
+
+      Type *NewTy = ArgTys[D.getArgumentNumber()];
+      if (VectorType *VTy = dyn_cast<VectorType>(NewTy))
+        NewTy = VectorType::getSubdividedVectorType(VTy, 2);
+      else
+        return true;
+
+      return Ty != NewTy;
+    }
     case IITDescriptor::ScalableVecArgument: {
       VectorType *VTy = dyn_cast<VectorType>(Ty);
       if (!VTy || !VTy->isScalable())
Index: utils/TableGen/IntrinsicEmitter.cpp
===================================================================
--- utils/TableGen/IntrinsicEmitter.cpp
+++ utils/TableGen/IntrinsicEmitter.cpp
@@ -221,7 +221,9 @@
   IIT_STRUCT8 = 40,
   IIT_F128 = 41,
   IIT_VEC_ELEMENT = 42,
-  IIT_SCALABLE_VEC = 43
+  IIT_SCALABLE_VEC = 43,
+  IIT_SUBDIVIDE2_ARG = 44,
+  IIT_SUBDIVIDE4_ARG = 45
 };
 
 static void EncodeFixedValueType(MVT::SimpleValueType VT,
@@ -293,6 +295,10 @@
       Sig.push_back(IIT_PTR_TO_ELT);
     else if (R->isSubClassOf("LLVMVectorElementType"))
       Sig.push_back(IIT_VEC_ELEMENT);
+    else if (R->isSubClassOf("LLVMSubdivide2VectorType"))
+      Sig.push_back(IIT_SUBDIVIDE2_ARG);
+    else if (R->isSubClassOf("LLVMSubdivide4VectorType"))
+      Sig.push_back(IIT_SUBDIVIDE4_ARG);
     else
       Sig.push_back(IIT_ARG);
     return Sig.push_back((Number << 3) | 7 /*IITDescriptor::AK_MatchType*/);