diff --git a/CMakeLists.txt b/CMakeLists.txt
index 8f7db99784..74244c1d58 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -686,6 +686,8 @@ add_subdirectory(include/dxc)
 # really depend on anything else in the build it is safe.
 list(APPEND LLVM_COMMON_DEPENDS HCTGen) 
 
+add_subdirectory(utils/hct)
+
 if(EXISTS "${LLVM_MAIN_SRC_DIR}/external")
   add_subdirectory(external) # SPIRV change
 endif()
diff --git a/include/dxc/DXIL/DxilConstants.h b/include/dxc/DXIL/DxilConstants.h
index f8d5b740f7..db238c99d6 100644
--- a/include/dxc/DXIL/DxilConstants.h
+++ b/include/dxc/DXIL/DxilConstants.h
@@ -147,12 +147,19 @@ const unsigned kMaxMSTotalSigRows = 32;
 const unsigned kMaxMSSMSize = 1024 * 28;
 const unsigned kMinWaveSize = 4;
 const unsigned kMaxWaveSize = 128;
+const unsigned kDefaultMaxVectorLength = 4;
+const unsigned kSM69MaxVectorLength = 1024;
 
 const float kMaxMipLodBias = 15.99f;
 const float kMinMipLodBias = -16.0f;
 
 const unsigned kResRetStatusIndex = 4;
 
+/* <py::lines('OLOAD_DIMS-TEXT')>hctdb_instrhelp.get_max_oload_dims()</py>*/
+// OLOAD_DIMS-TEXT:BEGIN
+const unsigned kDxilMaxOloadDims = 2;
+// OLOAD_DIMS-TEXT:END
+
 enum class ComponentType : uint32_t {
   Invalid = 0,
   I1,
@@ -463,6 +470,11 @@ inline bool IsTBuffer(DXIL::ResourceKind ResourceKind) {
   return ResourceKind == DXIL::ResourceKind::TBuffer;
 }
 
+inline bool IsCTBuffer(DXIL::ResourceKind ResourceKind) {
+  return ResourceKind == DXIL::ResourceKind::CBuffer ||
+         ResourceKind == DXIL::ResourceKind::TBuffer;
+}
+
 /// Whether the resource kind is a FeedbackTexture.
 inline bool IsFeedbackTexture(DXIL::ResourceKind ResourceKind) {
   return ResourceKind == DXIL::ResourceKind::FeedbackTexture2D ||
@@ -475,6 +487,9 @@ inline bool IsFeedbackTexture(DXIL::ResourceKind ResourceKind) {
 // Enumeration for operations specified by DXIL
 enum class OpCode : unsigned {
   //
+  RawBufferVectorLoad = 303, // reads from a raw buffer and structured buffer
+  RawBufferVectorStore =
+      304,           // writes to a RWByteAddressBuffer or RWStructuredBuffer
   Reserved0 = 226,   // Reserved
   Reserved1 = 227,   // Reserved
   Reserved10 = 236,  // Reserved
@@ -1029,8 +1044,9 @@ enum class OpCode : unsigned {
   NumOpCodes_Dxil_1_6 = 222,
   NumOpCodes_Dxil_1_7 = 226,
   NumOpCodes_Dxil_1_8 = 258,
+  NumOpCodes_Dxil_1_9 = 305,
 
-  NumOpCodes = 303 // exclusive last value of enumeration
+  NumOpCodes = 305 // exclusive last value of enumeration
 };
 // OPCODE-ENUM:END
 
@@ -1042,6 +1058,8 @@ enum class OpCode : unsigned {
 // Groups for DXIL operations with equivalent function templates
 enum class OpCodeClass : unsigned {
   //
+  RawBufferVectorLoad,
+  RawBufferVectorStore,
   Reserved,
 
   // Amplification shader instructions
@@ -1337,8 +1355,9 @@ enum class OpCodeClass : unsigned {
   NumOpClasses_Dxil_1_6 = 149,
   NumOpClasses_Dxil_1_7 = 153,
   NumOpClasses_Dxil_1_8 = 174,
+  NumOpClasses_Dxil_1_9 = 177,
 
-  NumOpClasses = 175 // exclusive last value of enumeration
+  NumOpClasses = 177 // exclusive last value of enumeration
 };
 // OPCODECLASS-ENUM:END
 
diff --git a/include/dxc/DXIL/DxilInstructions.h b/include/dxc/DXIL/DxilInstructions.h
index 11ab8e3b8d..c685f64b49 100644
--- a/include/dxc/DXIL/DxilInstructions.h
+++ b/include/dxc/DXIL/DxilInstructions.h
@@ -645,6 +645,42 @@ struct LlvmInst_VAArg {
   bool isAllowed() const { return false; }
 };
 
+/// This instruction extracts from vector
+struct LlvmInst_ExtractElement {
+  llvm::Instruction *Instr;
+  // Construction and identification
+  LlvmInst_ExtractElement(llvm::Instruction *pInstr) : Instr(pInstr) {}
+  operator bool() const {
+    return Instr->getOpcode() == llvm::Instruction::ExtractElement;
+  }
+  // Validation support
+  bool isAllowed() const { return true; }
+};
+
+/// This instruction inserts into vector
+struct LlvmInst_InsertElement {
+  llvm::Instruction *Instr;
+  // Construction and identification
+  LlvmInst_InsertElement(llvm::Instruction *pInstr) : Instr(pInstr) {}
+  operator bool() const {
+    return Instr->getOpcode() == llvm::Instruction::InsertElement;
+  }
+  // Validation support
+  bool isAllowed() const { return true; }
+};
+
+/// This instruction Shuffle two vectors
+struct LlvmInst_ShuffleVector {
+  llvm::Instruction *Instr;
+  // Construction and identification
+  LlvmInst_ShuffleVector(llvm::Instruction *pInstr) : Instr(pInstr) {}
+  operator bool() const {
+    return Instr->getOpcode() == llvm::Instruction::ShuffleVector;
+  }
+  // Validation support
+  bool isAllowed() const { return true; }
+};
+
 /// This instruction extracts from aggregate
 struct LlvmInst_ExtractValue {
   llvm::Instruction *Instr;
@@ -8813,5 +8849,98 @@ struct DxilInst_AllocateRayQuery2 {
                              llvm::APInt(32, (uint64_t)val)));
   }
 };
+
+/// This instruction reads from a raw buffer and structured buffer
+struct DxilInst_RawBufferVectorLoad {
+  llvm::Instruction *Instr;
+  // Construction and identification
+  DxilInst_RawBufferVectorLoad(llvm::Instruction *pInstr) : Instr(pInstr) {}
+  operator bool() const {
+    return hlsl::OP::IsDxilOpFuncCallInst(
+        Instr, hlsl::OP::OpCode::RawBufferVectorLoad);
+  }
+  // Validation support
+  bool isAllowed() const { return true; }
+  bool isArgumentListValid() const {
+    if (5 != llvm::dyn_cast<llvm::CallInst>(Instr)->getNumArgOperands())
+      return false;
+    return true;
+  }
+  // Metadata
+  bool requiresUniformInputs() const { return false; }
+  // Operand indexes
+  enum OperandIdx {
+    arg_srv = 1,
+    arg_index = 2,
+    arg_elementOffset = 3,
+    arg_alignment = 4,
+  };
+  // Accessors
+  llvm::Value *get_srv() const { return Instr->getOperand(1); }
+  void set_srv(llvm::Value *val) { Instr->setOperand(1, val); }
+  llvm::Value *get_index() const { return Instr->getOperand(2); }
+  void set_index(llvm::Value *val) { Instr->setOperand(2, val); }
+  llvm::Value *get_elementOffset() const { return Instr->getOperand(3); }
+  void set_elementOffset(llvm::Value *val) { Instr->setOperand(3, val); }
+  llvm::Value *get_alignment() const { return Instr->getOperand(4); }
+  void set_alignment(llvm::Value *val) { Instr->setOperand(4, val); }
+  int32_t get_alignment_val() const {
+    return (int32_t)(llvm::dyn_cast<llvm::ConstantInt>(Instr->getOperand(4))
+                         ->getZExtValue());
+  }
+  void set_alignment_val(int32_t val) {
+    Instr->setOperand(4, llvm::Constant::getIntegerValue(
+                             llvm::IntegerType::get(Instr->getContext(), 32),
+                             llvm::APInt(32, (uint64_t)val)));
+  }
+};
+
+/// This instruction writes to a RWByteAddressBuffer or RWStructuredBuffer
+struct DxilInst_RawBufferVectorStore {
+  llvm::Instruction *Instr;
+  // Construction and identification
+  DxilInst_RawBufferVectorStore(llvm::Instruction *pInstr) : Instr(pInstr) {}
+  operator bool() const {
+    return hlsl::OP::IsDxilOpFuncCallInst(
+        Instr, hlsl::OP::OpCode::RawBufferVectorStore);
+  }
+  // Validation support
+  bool isAllowed() const { return true; }
+  bool isArgumentListValid() const {
+    if (6 != llvm::dyn_cast<llvm::CallInst>(Instr)->getNumArgOperands())
+      return false;
+    return true;
+  }
+  // Metadata
+  bool requiresUniformInputs() const { return false; }
+  // Operand indexes
+  enum OperandIdx {
+    arg_uav = 1,
+    arg_index = 2,
+    arg_elementOffset = 3,
+    arg_value0 = 4,
+    arg_alignment = 5,
+  };
+  // Accessors
+  llvm::Value *get_uav() const { return Instr->getOperand(1); }
+  void set_uav(llvm::Value *val) { Instr->setOperand(1, val); }
+  llvm::Value *get_index() const { return Instr->getOperand(2); }
+  void set_index(llvm::Value *val) { Instr->setOperand(2, val); }
+  llvm::Value *get_elementOffset() const { return Instr->getOperand(3); }
+  void set_elementOffset(llvm::Value *val) { Instr->setOperand(3, val); }
+  llvm::Value *get_value0() const { return Instr->getOperand(4); }
+  void set_value0(llvm::Value *val) { Instr->setOperand(4, val); }
+  llvm::Value *get_alignment() const { return Instr->getOperand(5); }
+  void set_alignment(llvm::Value *val) { Instr->setOperand(5, val); }
+  int32_t get_alignment_val() const {
+    return (int32_t)(llvm::dyn_cast<llvm::ConstantInt>(Instr->getOperand(5))
+                         ->getZExtValue());
+  }
+  void set_alignment_val(int32_t val) {
+    Instr->setOperand(5, llvm::Constant::getIntegerValue(
+                             llvm::IntegerType::get(Instr->getContext(), 32),
+                             llvm::APInt(32, (uint64_t)val)));
+  }
+};
 // INSTR-HELPER:END
 } // namespace hlsl
diff --git a/include/dxc/DXIL/DxilOperations.h b/include/dxc/DXIL/DxilOperations.h
index 3514701327..955b5449db 100644
--- a/include/dxc/DXIL/DxilOperations.h
+++ b/include/dxc/DXIL/DxilOperations.h
@@ -58,11 +58,14 @@ class OP {
   void RefreshCache();
 
   llvm::Function *GetOpFunc(OpCode OpCode, llvm::Type *pOverloadType);
+  llvm::Function *GetOpFunc(OpCode OpCode,
+                            llvm::ArrayRef<llvm::Type *> ExtendedOverloads);
   const llvm::SmallMapVector<llvm::Type *, llvm::Function *, 8> &
   GetOpFuncList(OpCode OpCode) const;
   bool IsDxilOpUsed(OpCode opcode) const;
   void RemoveFunction(llvm::Function *F);
   llvm::LLVMContext &GetCtx() { return m_Ctx; }
+  llvm::Module *GetModule() { return m_pModule; }
   llvm::Type *GetHandleType() const;
   llvm::Type *GetNodeHandleType() const;
   llvm::Type *GetNodeRecordHandleType() const;
@@ -83,6 +86,10 @@ class OP {
   llvm::Type *GetVectorType(unsigned numElements, llvm::Type *pOverloadType);
   bool IsResRetType(llvm::Type *Ty);
 
+  // Construct an unnamed struct type containing the set of member types.
+  llvm::StructType *
+  GetExtendedOverloadType(llvm::ArrayRef<llvm::Type *> OverloadTypes);
+
   // Try to get the opcode class for a function.
   // Return true and set `opClass` if the given function is a dxil function.
   // Return false if the given function is not a dxil function.
@@ -140,6 +147,8 @@ class OP {
                                        unsigned valMinor, unsigned &major,
                                        unsigned &minor, unsigned &mask);
 
+  static bool IsDxilOpExtendedOverload(OpCode C);
+
 private:
   // Per-module properties.
   llvm::LLVMContext &m_Ctx;
@@ -164,8 +173,10 @@ class OP {
 
   static const unsigned kUserDefineTypeSlot = 9;
   static const unsigned kObjectTypeSlot = 10;
+  static const unsigned kVectorTypeSlot = 11;
+  static const unsigned kExtendedTypeSlot = 12;
   static const unsigned kNumTypeOverloads =
-      11; // void, h,f,d, i1, i8,i16,i32,i64, udt, obj
+      13; // void, h,f,d, i1, i8,i16,i32,i64, udt, obj, vec, extended
 
   llvm::Type *m_pResRetType[kNumTypeOverloads];
   llvm::Type *m_pCBufferRetType[kNumTypeOverloads];
@@ -179,14 +190,39 @@ class OP {
 
 private:
   // Static properties.
+  struct OverloadMask {
+    // mask of type slot bits as (1 << TypeSlot)
+    uint16_t SlotMask;
+    static_assert(kNumTypeOverloads <= (sizeof(SlotMask) * 8));
+    bool operator[](unsigned TypeSlot) const {
+      return (TypeSlot < kNumTypeOverloads) ? (bool)(SlotMask & (1 << TypeSlot))
+                                            : 0;
+    }
+    operator bool() const { return SlotMask != 0; }
+  };
   struct OpCodeProperty {
     OpCode opCode;
     const char *pOpCodeName;
     OpCodeClass opCodeClass;
     const char *pOpCodeClassName;
     bool bAllowOverload[kNumTypeOverloads]; // void, h,f,d, i1, i8,i16,i32,i64,
-                                            // udt
+                                            // udt, obj, vec, extended
     llvm::Attribute::AttrKind FuncAttr;
+
+    // Extended Type Overloads:
+    // This is an encoding for a multi-dimensional overload.
+    // 1. Only bAllowOverload[kExtendedTypeSlot] is set to true
+    // 2. ExtendedOverloads defines allowed types for each overload index
+    // 3. AllowedVectorElements defines allowed vector component types,
+    //    when kVectorTypeSlot bit is set for the corresponding overload index.
+    //    This includes when a single vector overload type is specified with
+    //    bAllowOverload[kVectorTypeSlot].
+
+    // A bit mask of allowed type slots per extended overload
+    OverloadMask ExtendedOverloads[DXIL::kDxilMaxOloadDims];
+    // A bit mask of allowed vector element types for the vector overload
+    // or each corresponding extended vector overload.
+    OverloadMask AllowedVectorElements[DXIL::kDxilMaxOloadDims];
   };
   static const OpCodeProperty m_OpCodeProps[(unsigned)OpCode::NumOpCodes];
 
diff --git a/include/dxc/HlslIntrinsicOp.h b/include/dxc/HlslIntrinsicOp.h
index fcc9bb11b1..41c72d1a51 100644
--- a/include/dxc/HlslIntrinsicOp.h
+++ b/include/dxc/HlslIntrinsicOp.h
@@ -5,378 +5,366 @@
 #pragma once
 namespace hlsl {
 enum class IntrinsicOp {
-  IOP_AcceptHitAndEndSearch,
-  IOP_AddUint64,
-  IOP_AllMemoryBarrier,
-  IOP_AllMemoryBarrierWithGroupSync,
-  IOP_AllocateRayQuery,
-  IOP_Barrier,
-  IOP_CallShader,
-  IOP_CheckAccessFullyMapped,
-  IOP_CreateResourceFromHeap,
-  IOP_D3DCOLORtoUBYTE4,
-  IOP_DeviceMemoryBarrier,
-  IOP_DeviceMemoryBarrierWithGroupSync,
-  IOP_DispatchMesh,
-  IOP_DispatchRaysDimensions,
-  IOP_DispatchRaysIndex,
-  IOP_EvaluateAttributeAtSample,
-  IOP_EvaluateAttributeCentroid,
-  IOP_EvaluateAttributeSnapped,
-  IOP_GeometryIndex,
-  IOP_GetAttributeAtVertex,
-  IOP_GetRemainingRecursionLevels,
-  IOP_GetRenderTargetSampleCount,
-  IOP_GetRenderTargetSamplePosition,
-  IOP_GroupMemoryBarrier,
-  IOP_GroupMemoryBarrierWithGroupSync,
-  IOP_HitKind,
-  IOP_IgnoreHit,
-  IOP_InstanceID,
-  IOP_InstanceIndex,
-  IOP_InterlockedAdd,
-  IOP_InterlockedAnd,
-  IOP_InterlockedCompareExchange,
-  IOP_InterlockedCompareExchangeFloatBitwise,
-  IOP_InterlockedCompareStore,
-  IOP_InterlockedCompareStoreFloatBitwise,
-  IOP_InterlockedExchange,
-  IOP_InterlockedMax,
-  IOP_InterlockedMin,
-  IOP_InterlockedOr,
-  IOP_InterlockedXor,
-  IOP_IsHelperLane,
-  IOP_NonUniformResourceIndex,
-  IOP_ObjectRayDirection,
-  IOP_ObjectRayOrigin,
-  IOP_ObjectToWorld,
-  IOP_ObjectToWorld3x4,
-  IOP_ObjectToWorld4x3,
-  IOP_PrimitiveIndex,
-  IOP_Process2DQuadTessFactorsAvg,
-  IOP_Process2DQuadTessFactorsMax,
-  IOP_Process2DQuadTessFactorsMin,
-  IOP_ProcessIsolineTessFactors,
-  IOP_ProcessQuadTessFactorsAvg,
-  IOP_ProcessQuadTessFactorsMax,
-  IOP_ProcessQuadTessFactorsMin,
-  IOP_ProcessTriTessFactorsAvg,
-  IOP_ProcessTriTessFactorsMax,
-  IOP_ProcessTriTessFactorsMin,
-  IOP_QuadAll,
-  IOP_QuadAny,
-  IOP_QuadReadAcrossDiagonal,
-  IOP_QuadReadAcrossX,
-  IOP_QuadReadAcrossY,
-  IOP_QuadReadLaneAt,
-  IOP_RayFlags,
-  IOP_RayTCurrent,
-  IOP_RayTMin,
-  IOP_ReportHit,
-  IOP_SetMeshOutputCounts,
-  IOP_TraceRay,
-  IOP_WaveActiveAllEqual,
-  IOP_WaveActiveAllTrue,
-  IOP_WaveActiveAnyTrue,
-  IOP_WaveActiveBallot,
-  IOP_WaveActiveBitAnd,
-  IOP_WaveActiveBitOr,
-  IOP_WaveActiveBitXor,
-  IOP_WaveActiveCountBits,
-  IOP_WaveActiveMax,
-  IOP_WaveActiveMin,
-  IOP_WaveActiveProduct,
-  IOP_WaveActiveSum,
-  IOP_WaveGetLaneCount,
-  IOP_WaveGetLaneIndex,
-  IOP_WaveIsFirstLane,
-  IOP_WaveMatch,
-  IOP_WaveMultiPrefixBitAnd,
-  IOP_WaveMultiPrefixBitOr,
-  IOP_WaveMultiPrefixBitXor,
-  IOP_WaveMultiPrefixCountBits,
-  IOP_WaveMultiPrefixProduct,
-  IOP_WaveMultiPrefixSum,
-  IOP_WavePrefixCountBits,
-  IOP_WavePrefixProduct,
-  IOP_WavePrefixSum,
-  IOP_WaveReadLaneAt,
-  IOP_WaveReadLaneFirst,
-  IOP_WorldRayDirection,
-  IOP_WorldRayOrigin,
-  IOP_WorldToObject,
-  IOP_WorldToObject3x4,
-  IOP_WorldToObject4x3,
-  IOP_abort,
-  IOP_abs,
-  IOP_acos,
-  IOP_all,
-  IOP_and,
-  IOP_any,
-  IOP_asdouble,
-  IOP_asfloat,
-  IOP_asfloat16,
-  IOP_asin,
-  IOP_asint,
-  IOP_asint16,
-  IOP_asuint,
-  IOP_asuint16,
-  IOP_atan,
-  IOP_atan2,
-  IOP_ceil,
-  IOP_clamp,
-  IOP_clip,
-  IOP_cos,
-  IOP_cosh,
-  IOP_countbits,
-  IOP_cross,
-  IOP_ddx,
-  IOP_ddx_coarse,
-  IOP_ddx_fine,
-  IOP_ddy,
-  IOP_ddy_coarse,
-  IOP_ddy_fine,
-  IOP_degrees,
-  IOP_determinant,
-  IOP_distance,
-  IOP_dot,
-  IOP_dot2add,
-  IOP_dot4add_i8packed,
-  IOP_dot4add_u8packed,
-  IOP_dst,
-  IOP_exp,
-  IOP_exp2,
-  IOP_f16tof32,
-  IOP_f32tof16,
-  IOP_faceforward,
-  IOP_firstbithigh,
-  IOP_firstbitlow,
-  IOP_floor,
-  IOP_fma,
-  IOP_fmod,
-  IOP_frac,
-  IOP_frexp,
-  IOP_fwidth,
-  IOP_isfinite,
-  IOP_isinf,
-  IOP_isnan,
-  IOP_ldexp,
-  IOP_length,
-  IOP_lerp,
-  IOP_lit,
-  IOP_log,
-  IOP_log10,
-  IOP_log2,
-  IOP_mad,
-  IOP_max,
-  IOP_min,
-  IOP_modf,
-  IOP_msad4,
-  IOP_mul,
-  IOP_normalize,
-  IOP_or,
-  IOP_pack_clamp_s8,
-  IOP_pack_clamp_u8,
-  IOP_pack_s8,
-  IOP_pack_u8,
-  IOP_pow,
-  IOP_printf,
-  IOP_radians,
-  IOP_rcp,
-  IOP_reflect,
-  IOP_refract,
-  IOP_reversebits,
-  IOP_round,
-  IOP_rsqrt,
-  IOP_saturate,
-  IOP_select,
-  IOP_sign,
-  IOP_sin,
-  IOP_sincos,
-  IOP_sinh,
-  IOP_smoothstep,
-  IOP_source_mark,
-  IOP_sqrt,
-  IOP_step,
-  IOP_tan,
-  IOP_tanh,
-  IOP_tex1D,
-  IOP_tex1Dbias,
-  IOP_tex1Dgrad,
-  IOP_tex1Dlod,
-  IOP_tex1Dproj,
-  IOP_tex2D,
-  IOP_tex2Dbias,
-  IOP_tex2Dgrad,
-  IOP_tex2Dlod,
-  IOP_tex2Dproj,
-  IOP_tex3D,
-  IOP_tex3Dbias,
-  IOP_tex3Dgrad,
-  IOP_tex3Dlod,
-  IOP_tex3Dproj,
-  IOP_texCUBE,
-  IOP_texCUBEbias,
-  IOP_texCUBEgrad,
-  IOP_texCUBElod,
-  IOP_texCUBEproj,
-  IOP_transpose,
-  IOP_trunc,
-  IOP_unpack_s8s16,
-  IOP_unpack_s8s32,
-  IOP_unpack_u8u16,
-  IOP_unpack_u8u32,
-#ifdef ENABLE_SPIRV_CODEGEN
-  IOP_VkRawBufferLoad,
-#endif // ENABLE_SPIRV_CODEGEN
-#ifdef ENABLE_SPIRV_CODEGEN
-  IOP_VkRawBufferStore,
-#endif // ENABLE_SPIRV_CODEGEN
-#ifdef ENABLE_SPIRV_CODEGEN
-  IOP_VkReadClock,
-#endif // ENABLE_SPIRV_CODEGEN
-#ifdef ENABLE_SPIRV_CODEGEN
-  IOP_Vkext_execution_mode,
-#endif // ENABLE_SPIRV_CODEGEN
-#ifdef ENABLE_SPIRV_CODEGEN
-  IOP_Vkext_execution_mode_id,
-#endif // ENABLE_SPIRV_CODEGEN
-  MOP_Append,
-  MOP_RestartStrip,
-  MOP_CalculateLevelOfDetail,
-  MOP_CalculateLevelOfDetailUnclamped,
-  MOP_GetDimensions,
-  MOP_Load,
-  MOP_Sample,
-  MOP_SampleBias,
-  MOP_SampleCmp,
-  MOP_SampleCmpBias,
-  MOP_SampleCmpGrad,
-  MOP_SampleCmpLevel,
-  MOP_SampleCmpLevelZero,
-  MOP_SampleGrad,
-  MOP_SampleLevel,
-  MOP_Gather,
-  MOP_GatherAlpha,
-  MOP_GatherBlue,
-  MOP_GatherCmp,
-  MOP_GatherCmpAlpha,
-  MOP_GatherCmpBlue,
-  MOP_GatherCmpGreen,
-  MOP_GatherCmpRed,
-  MOP_GatherGreen,
-  MOP_GatherRaw,
-  MOP_GatherRed,
-  MOP_GetSamplePosition,
-  MOP_Load2,
-  MOP_Load3,
-  MOP_Load4,
-  MOP_InterlockedAdd,
-  MOP_InterlockedAdd64,
-  MOP_InterlockedAnd,
-  MOP_InterlockedAnd64,
-  MOP_InterlockedCompareExchange,
-  MOP_InterlockedCompareExchange64,
-  MOP_InterlockedCompareExchangeFloatBitwise,
-  MOP_InterlockedCompareStore,
-  MOP_InterlockedCompareStore64,
-  MOP_InterlockedCompareStoreFloatBitwise,
-  MOP_InterlockedExchange,
-  MOP_InterlockedExchange64,
-  MOP_InterlockedExchangeFloat,
-  MOP_InterlockedMax,
-  MOP_InterlockedMax64,
-  MOP_InterlockedMin,
-  MOP_InterlockedMin64,
-  MOP_InterlockedOr,
-  MOP_InterlockedOr64,
-  MOP_InterlockedXor,
-  MOP_InterlockedXor64,
-  MOP_Store,
-  MOP_Store2,
-  MOP_Store3,
-  MOP_Store4,
-  MOP_DecrementCounter,
-  MOP_IncrementCounter,
-  MOP_Consume,
-  MOP_WriteSamplerFeedback,
-  MOP_WriteSamplerFeedbackBias,
-  MOP_WriteSamplerFeedbackGrad,
-  MOP_WriteSamplerFeedbackLevel,
-  MOP_Abort,
-  MOP_CandidateGeometryIndex,
-  MOP_CandidateInstanceContributionToHitGroupIndex,
-  MOP_CandidateInstanceID,
-  MOP_CandidateInstanceIndex,
-  MOP_CandidateObjectRayDirection,
-  MOP_CandidateObjectRayOrigin,
-  MOP_CandidateObjectToWorld3x4,
-  MOP_CandidateObjectToWorld4x3,
-  MOP_CandidatePrimitiveIndex,
-  MOP_CandidateProceduralPrimitiveNonOpaque,
-  MOP_CandidateTriangleBarycentrics,
-  MOP_CandidateTriangleFrontFace,
-  MOP_CandidateTriangleRayT,
-  MOP_CandidateType,
-  MOP_CandidateWorldToObject3x4,
-  MOP_CandidateWorldToObject4x3,
-  MOP_CommitNonOpaqueTriangleHit,
-  MOP_CommitProceduralPrimitiveHit,
-  MOP_CommittedGeometryIndex,
-  MOP_CommittedInstanceContributionToHitGroupIndex,
-  MOP_CommittedInstanceID,
-  MOP_CommittedInstanceIndex,
-  MOP_CommittedObjectRayDirection,
-  MOP_CommittedObjectRayOrigin,
-  MOP_CommittedObjectToWorld3x4,
-  MOP_CommittedObjectToWorld4x3,
-  MOP_CommittedPrimitiveIndex,
-  MOP_CommittedRayT,
-  MOP_CommittedStatus,
-  MOP_CommittedTriangleBarycentrics,
-  MOP_CommittedTriangleFrontFace,
-  MOP_CommittedWorldToObject3x4,
-  MOP_CommittedWorldToObject4x3,
-  MOP_Proceed,
-  MOP_RayFlags,
-  MOP_RayTMin,
-  MOP_TraceRayInline,
-  MOP_WorldRayDirection,
-  MOP_WorldRayOrigin,
-  MOP_Count,
-  MOP_FinishedCrossGroupSharing,
-  MOP_GetGroupNodeOutputRecords,
-  MOP_GetThreadNodeOutputRecords,
-  MOP_IsValid,
-  MOP_GroupIncrementOutputCount,
-  MOP_ThreadIncrementOutputCount,
-  MOP_OutputComplete,
-#ifdef ENABLE_SPIRV_CODEGEN
-  MOP_SubpassLoad,
-#endif // ENABLE_SPIRV_CODEGEN
+  IOP_AcceptHitAndEndSearch = 0,
+  IOP_AddUint64 = 1,
+  IOP_AllMemoryBarrier = 2,
+  IOP_AllMemoryBarrierWithGroupSync = 3,
+  IOP_AllocateRayQuery = 4,
+  IOP_Barrier = 5,
+  IOP_CallShader = 6,
+  IOP_CheckAccessFullyMapped = 7,
+  IOP_CreateResourceFromHeap = 8,
+  IOP_D3DCOLORtoUBYTE4 = 9,
+  IOP_DeviceMemoryBarrier = 10,
+  IOP_DeviceMemoryBarrierWithGroupSync = 11,
+  IOP_DispatchMesh = 12,
+  IOP_DispatchRaysDimensions = 13,
+  IOP_DispatchRaysIndex = 14,
+  IOP_EvaluateAttributeAtSample = 15,
+  IOP_EvaluateAttributeCentroid = 16,
+  IOP_EvaluateAttributeSnapped = 17,
+  IOP_GeometryIndex = 18,
+  IOP_GetAttributeAtVertex = 19,
+  IOP_GetRemainingRecursionLevels = 20,
+  IOP_GetRenderTargetSampleCount = 21,
+  IOP_GetRenderTargetSamplePosition = 22,
+  IOP_GroupMemoryBarrier = 23,
+  IOP_GroupMemoryBarrierWithGroupSync = 24,
+  IOP_HitKind = 25,
+  IOP_IgnoreHit = 26,
+  IOP_InstanceID = 27,
+  IOP_InstanceIndex = 28,
+  IOP_InterlockedAdd = 29,
+  IOP_InterlockedAnd = 30,
+  IOP_InterlockedCompareExchange = 31,
+  IOP_InterlockedCompareExchangeFloatBitwise = 32,
+  IOP_InterlockedCompareStore = 33,
+  IOP_InterlockedCompareStoreFloatBitwise = 34,
+  IOP_InterlockedExchange = 35,
+  IOP_InterlockedMax = 36,
+  IOP_InterlockedMin = 37,
+  IOP_InterlockedOr = 38,
+  IOP_InterlockedXor = 39,
+  IOP_IsHelperLane = 40,
+  IOP_NonUniformResourceIndex = 41,
+  IOP_ObjectRayDirection = 42,
+  IOP_ObjectRayOrigin = 43,
+  IOP_ObjectToWorld = 44,
+  IOP_ObjectToWorld3x4 = 45,
+  IOP_ObjectToWorld4x3 = 46,
+  IOP_PrimitiveIndex = 47,
+  IOP_Process2DQuadTessFactorsAvg = 48,
+  IOP_Process2DQuadTessFactorsMax = 49,
+  IOP_Process2DQuadTessFactorsMin = 50,
+  IOP_ProcessIsolineTessFactors = 51,
+  IOP_ProcessQuadTessFactorsAvg = 52,
+  IOP_ProcessQuadTessFactorsMax = 53,
+  IOP_ProcessQuadTessFactorsMin = 54,
+  IOP_ProcessTriTessFactorsAvg = 55,
+  IOP_ProcessTriTessFactorsMax = 56,
+  IOP_ProcessTriTessFactorsMin = 57,
+  IOP_QuadAll = 58,
+  IOP_QuadAny = 59,
+  IOP_QuadReadAcrossDiagonal = 60,
+  IOP_QuadReadAcrossX = 61,
+  IOP_QuadReadAcrossY = 62,
+  IOP_QuadReadLaneAt = 63,
+  IOP_RayFlags = 64,
+  IOP_RayTCurrent = 65,
+  IOP_RayTMin = 66,
+  IOP_ReportHit = 67,
+  IOP_SetMeshOutputCounts = 68,
+  IOP_TraceRay = 69,
+  IOP_WaveActiveAllEqual = 70,
+  IOP_WaveActiveAllTrue = 71,
+  IOP_WaveActiveAnyTrue = 72,
+  IOP_WaveActiveBallot = 73,
+  IOP_WaveActiveBitAnd = 74,
+  IOP_WaveActiveBitOr = 75,
+  IOP_WaveActiveBitXor = 76,
+  IOP_WaveActiveCountBits = 77,
+  IOP_WaveActiveMax = 78,
+  IOP_WaveActiveMin = 79,
+  IOP_WaveActiveProduct = 80,
+  IOP_WaveActiveSum = 81,
+  IOP_WaveGetLaneCount = 82,
+  IOP_WaveGetLaneIndex = 83,
+  IOP_WaveIsFirstLane = 84,
+  IOP_WaveMatch = 85,
+  IOP_WaveMultiPrefixBitAnd = 86,
+  IOP_WaveMultiPrefixBitOr = 87,
+  IOP_WaveMultiPrefixBitXor = 88,
+  IOP_WaveMultiPrefixCountBits = 89,
+  IOP_WaveMultiPrefixProduct = 90,
+  IOP_WaveMultiPrefixSum = 91,
+  IOP_WavePrefixCountBits = 92,
+  IOP_WavePrefixProduct = 93,
+  IOP_WavePrefixSum = 94,
+  IOP_WaveReadLaneAt = 95,
+  IOP_WaveReadLaneFirst = 96,
+  IOP_WorldRayDirection = 97,
+  IOP_WorldRayOrigin = 98,
+  IOP_WorldToObject = 99,
+  IOP_WorldToObject3x4 = 100,
+  IOP_WorldToObject4x3 = 101,
+  IOP_abort = 102,
+  IOP_abs = 103,
+  IOP_acos = 104,
+  IOP_all = 105,
+  IOP_and = 106,
+  IOP_any = 107,
+  IOP_asdouble = 108,
+  IOP_asfloat = 109,
+  IOP_asfloat16 = 110,
+  IOP_asin = 111,
+  IOP_asint = 112,
+  IOP_asint16 = 113,
+  IOP_asuint = 114,
+  IOP_asuint16 = 115,
+  IOP_atan = 116,
+  IOP_atan2 = 117,
+  IOP_ceil = 118,
+  IOP_clamp = 119,
+  IOP_clip = 120,
+  IOP_cos = 121,
+  IOP_cosh = 122,
+  IOP_countbits = 123,
+  IOP_cross = 124,
+  IOP_ddx = 125,
+  IOP_ddx_coarse = 126,
+  IOP_ddx_fine = 127,
+  IOP_ddy = 128,
+  IOP_ddy_coarse = 129,
+  IOP_ddy_fine = 130,
+  IOP_degrees = 131,
+  IOP_determinant = 132,
+  IOP_distance = 133,
+  IOP_dot = 134,
+  IOP_dot2add = 135,
+  IOP_dot4add_i8packed = 136,
+  IOP_dot4add_u8packed = 137,
+  IOP_dst = 138,
+  IOP_exp = 139,
+  IOP_exp2 = 140,
+  IOP_f16tof32 = 141,
+  IOP_f32tof16 = 142,
+  IOP_faceforward = 143,
+  IOP_firstbithigh = 144,
+  IOP_firstbitlow = 145,
+  IOP_floor = 146,
+  IOP_fma = 147,
+  IOP_fmod = 148,
+  IOP_frac = 149,
+  IOP_frexp = 150,
+  IOP_fwidth = 151,
+  IOP_isfinite = 152,
+  IOP_isinf = 153,
+  IOP_isnan = 154,
+  IOP_ldexp = 155,
+  IOP_length = 156,
+  IOP_lerp = 157,
+  IOP_lit = 158,
+  IOP_log = 159,
+  IOP_log10 = 160,
+  IOP_log2 = 161,
+  IOP_mad = 162,
+  IOP_max = 163,
+  IOP_min = 164,
+  IOP_modf = 165,
+  IOP_msad4 = 166,
+  IOP_mul = 167,
+  IOP_normalize = 168,
+  IOP_or = 169,
+  IOP_pack_clamp_s8 = 170,
+  IOP_pack_clamp_u8 = 171,
+  IOP_pack_s8 = 172,
+  IOP_pack_u8 = 173,
+  IOP_pow = 174,
+  IOP_printf = 175,
+  IOP_radians = 176,
+  IOP_rcp = 177,
+  IOP_reflect = 178,
+  IOP_refract = 179,
+  IOP_reversebits = 180,
+  IOP_round = 181,
+  IOP_rsqrt = 182,
+  IOP_saturate = 183,
+  IOP_select = 184,
+  IOP_sign = 185,
+  IOP_sin = 186,
+  IOP_sincos = 187,
+  IOP_sinh = 188,
+  IOP_smoothstep = 189,
+  IOP_source_mark = 190,
+  IOP_sqrt = 191,
+  IOP_step = 192,
+  IOP_tan = 193,
+  IOP_tanh = 194,
+  IOP_tex1D = 195,
+  IOP_tex1Dbias = 196,
+  IOP_tex1Dgrad = 197,
+  IOP_tex1Dlod = 198,
+  IOP_tex1Dproj = 199,
+  IOP_tex2D = 200,
+  IOP_tex2Dbias = 201,
+  IOP_tex2Dgrad = 202,
+  IOP_tex2Dlod = 203,
+  IOP_tex2Dproj = 204,
+  IOP_tex3D = 205,
+  IOP_tex3Dbias = 206,
+  IOP_tex3Dgrad = 207,
+  IOP_tex3Dlod = 208,
+  IOP_tex3Dproj = 209,
+  IOP_texCUBE = 210,
+  IOP_texCUBEbias = 211,
+  IOP_texCUBEgrad = 212,
+  IOP_texCUBElod = 213,
+  IOP_texCUBEproj = 214,
+  IOP_transpose = 215,
+  IOP_trunc = 216,
+  IOP_unpack_s8s16 = 217,
+  IOP_unpack_s8s32 = 218,
+  IOP_unpack_u8u16 = 219,
+  IOP_unpack_u8u32 = 220,
+  IOP_VkRawBufferLoad = 221,
+  IOP_VkRawBufferStore = 222,
+  IOP_VkReadClock = 223,
+  IOP_Vkext_execution_mode = 224,
+  IOP_Vkext_execution_mode_id = 225,
+  MOP_Append = 226,
+  MOP_RestartStrip = 227,
+  MOP_CalculateLevelOfDetail = 228,
+  MOP_CalculateLevelOfDetailUnclamped = 229,
+  MOP_GetDimensions = 230,
+  MOP_Load = 231,
+  MOP_Sample = 232,
+  MOP_SampleBias = 233,
+  MOP_SampleCmp = 234,
+  MOP_SampleCmpBias = 235,
+  MOP_SampleCmpGrad = 236,
+  MOP_SampleCmpLevel = 237,
+  MOP_SampleCmpLevelZero = 238,
+  MOP_SampleGrad = 239,
+  MOP_SampleLevel = 240,
+  MOP_Gather = 241,
+  MOP_GatherAlpha = 242,
+  MOP_GatherBlue = 243,
+  MOP_GatherCmp = 244,
+  MOP_GatherCmpAlpha = 245,
+  MOP_GatherCmpBlue = 246,
+  MOP_GatherCmpGreen = 247,
+  MOP_GatherCmpRed = 248,
+  MOP_GatherGreen = 249,
+  MOP_GatherRaw = 250,
+  MOP_GatherRed = 251,
+  MOP_GetSamplePosition = 252,
+  MOP_Load2 = 253,
+  MOP_Load3 = 254,
+  MOP_Load4 = 255,
+  MOP_InterlockedAdd = 256,
+  MOP_InterlockedAdd64 = 257,
+  MOP_InterlockedAnd = 258,
+  MOP_InterlockedAnd64 = 259,
+  MOP_InterlockedCompareExchange = 260,
+  MOP_InterlockedCompareExchange64 = 261,
+  MOP_InterlockedCompareExchangeFloatBitwise = 262,
+  MOP_InterlockedCompareStore = 263,
+  MOP_InterlockedCompareStore64 = 264,
+  MOP_InterlockedCompareStoreFloatBitwise = 265,
+  MOP_InterlockedExchange = 266,
+  MOP_InterlockedExchange64 = 267,
+  MOP_InterlockedExchangeFloat = 268,
+  MOP_InterlockedMax = 269,
+  MOP_InterlockedMax64 = 270,
+  MOP_InterlockedMin = 271,
+  MOP_InterlockedMin64 = 272,
+  MOP_InterlockedOr = 273,
+  MOP_InterlockedOr64 = 274,
+  MOP_InterlockedXor = 275,
+  MOP_InterlockedXor64 = 276,
+  MOP_Store = 277,
+  MOP_Store2 = 278,
+  MOP_Store3 = 279,
+  MOP_Store4 = 280,
+  MOP_DecrementCounter = 281,
+  MOP_IncrementCounter = 282,
+  MOP_Consume = 283,
+  MOP_WriteSamplerFeedback = 284,
+  MOP_WriteSamplerFeedbackBias = 285,
+  MOP_WriteSamplerFeedbackGrad = 286,
+  MOP_WriteSamplerFeedbackLevel = 287,
+  MOP_Abort = 288,
+  MOP_CandidateGeometryIndex = 289,
+  MOP_CandidateInstanceContributionToHitGroupIndex = 290,
+  MOP_CandidateInstanceID = 291,
+  MOP_CandidateInstanceIndex = 292,
+  MOP_CandidateObjectRayDirection = 293,
+  MOP_CandidateObjectRayOrigin = 294,
+  MOP_CandidateObjectToWorld3x4 = 295,
+  MOP_CandidateObjectToWorld4x3 = 296,
+  MOP_CandidatePrimitiveIndex = 297,
+  MOP_CandidateProceduralPrimitiveNonOpaque = 298,
+  MOP_CandidateTriangleBarycentrics = 299,
+  MOP_CandidateTriangleFrontFace = 300,
+  MOP_CandidateTriangleRayT = 301,
+  MOP_CandidateType = 302,
+  MOP_CandidateWorldToObject3x4 = 303,
+  MOP_CandidateWorldToObject4x3 = 304,
+  MOP_CommitNonOpaqueTriangleHit = 305,
+  MOP_CommitProceduralPrimitiveHit = 306,
+  MOP_CommittedGeometryIndex = 307,
+  MOP_CommittedInstanceContributionToHitGroupIndex = 308,
+  MOP_CommittedInstanceID = 309,
+  MOP_CommittedInstanceIndex = 310,
+  MOP_CommittedObjectRayDirection = 311,
+  MOP_CommittedObjectRayOrigin = 312,
+  MOP_CommittedObjectToWorld3x4 = 313,
+  MOP_CommittedObjectToWorld4x3 = 314,
+  MOP_CommittedPrimitiveIndex = 315,
+  MOP_CommittedRayT = 316,
+  MOP_CommittedStatus = 317,
+  MOP_CommittedTriangleBarycentrics = 318,
+  MOP_CommittedTriangleFrontFace = 319,
+  MOP_CommittedWorldToObject3x4 = 320,
+  MOP_CommittedWorldToObject4x3 = 321,
+  MOP_Proceed = 322,
+  MOP_RayFlags = 323,
+  MOP_RayTMin = 324,
+  MOP_TraceRayInline = 325,
+  MOP_WorldRayDirection = 326,
+  MOP_WorldRayOrigin = 327,
+  MOP_Count = 328,
+  MOP_FinishedCrossGroupSharing = 329,
+  MOP_GetGroupNodeOutputRecords = 330,
+  MOP_GetThreadNodeOutputRecords = 331,
+  MOP_IsValid = 332,
+  MOP_GroupIncrementOutputCount = 333,
+  MOP_ThreadIncrementOutputCount = 334,
+  MOP_OutputComplete = 335,
+  MOP_SubpassLoad = 336,
   // unsigned
-  IOP_InterlockedUMax,
-  IOP_InterlockedUMin,
-  IOP_WaveActiveUMax,
-  IOP_WaveActiveUMin,
-  IOP_WaveActiveUProduct,
-  IOP_WaveActiveUSum,
-  IOP_WaveMultiPrefixUProduct,
-  IOP_WaveMultiPrefixUSum,
-  IOP_WavePrefixUProduct,
-  IOP_WavePrefixUSum,
-  IOP_uabs,
-  IOP_uclamp,
-  IOP_udot,
-  IOP_ufirstbithigh,
-  IOP_umad,
-  IOP_umax,
-  IOP_umin,
-  IOP_umul,
-  IOP_usign,
-  MOP_InterlockedUMax,
-  MOP_InterlockedUMin,
-  Num_Intrinsics,
+  IOP_InterlockedUMax = 337,
+  IOP_InterlockedUMin = 338,
+  IOP_WaveActiveUMax = 339,
+  IOP_WaveActiveUMin = 340,
+  IOP_WaveActiveUProduct = 341,
+  IOP_WaveActiveUSum = 342,
+  IOP_WaveMultiPrefixUProduct = 343,
+  IOP_WaveMultiPrefixUSum = 344,
+  IOP_WavePrefixUProduct = 345,
+  IOP_WavePrefixUSum = 346,
+  IOP_uabs = 347,
+  IOP_uclamp = 348,
+  IOP_udot = 349,
+  IOP_ufirstbithigh = 350,
+  IOP_umad = 351,
+  IOP_umax = 352,
+  IOP_umin = 353,
+  IOP_umul = 354,
+  IOP_usign = 355,
+  MOP_InterlockedUMax = 356,
+  MOP_InterlockedUMin = 357,
+  Num_Intrinsics = 358,
 };
 inline bool HasUnsignedIntrinsicOpcode(IntrinsicOp opcode) {
   switch (opcode) {
diff --git a/lib/DXIL/DxilOperations.cpp b/lib/DXIL/DxilOperations.cpp
index b3e552da18..02e72fb401 100644
--- a/lib/DXIL/DxilOperations.cpp
+++ b/lib/DXIL/DxilOperations.cpp
@@ -43,2981 +43,3040 @@ import hctdb_instrhelp
 const OP::OpCodeProperty OP::m_OpCodeProps[(unsigned)OP::OpCode::NumOpCodes] = {
     //   OpCode                       OpCode name,                OpCodeClass
     //   OpCodeClass name,              void,     h,     f,     d,    i1,    i8,
-    //   i16,   i32,   i64,   udt,   obj,  function attribute
+    //   i16,   i32,   i64,   udt,   obj,   vec,   ext,  function attribute, ext
+    //   oload, vec oload
     // Temporary, indexable, input, output registers void,     h,     f,     d,
-    // i1,    i8,   i16,   i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::TempRegLoad,
-        "TempRegLoad",
-        OCC::TempRegLoad,
-        "tempRegLoad",
-        {false, true, true, false, false, false, true, true, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::TempRegStore,
-        "TempRegStore",
-        OCC::TempRegStore,
-        "tempRegStore",
-        {false, true, true, false, false, false, true, true, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::MinPrecXRegLoad,
-        "MinPrecXRegLoad",
-        OCC::MinPrecXRegLoad,
-        "minPrecXRegLoad",
-        {false, true, false, false, false, false, true, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::MinPrecXRegStore,
-        "MinPrecXRegStore",
-        OCC::MinPrecXRegStore,
-        "minPrecXRegStore",
-        {false, true, false, false, false, false, true, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::LoadInput,
-        "LoadInput",
-        OCC::LoadInput,
-        "loadInput",
-        {false, true, true, false, false, false, true, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::StoreOutput,
-        "StoreOutput",
-        OCC::StoreOutput,
-        "storeOutput",
-        {false, true, true, false, false, false, true, true, false, false,
-         false},
-        Attribute::None,
-    },
+    // i1,    i8,   i16,   i32,   i64,   udt,   obj,   vec,  function attribute,
+    // ext oload, vec oload
+    {OC::TempRegLoad,
+     "TempRegLoad",
+     OCC::TempRegLoad,
+     "tempRegLoad",
+     {false, true, true, false, false, false, true, true, false, false, false,
+      false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::TempRegStore,
+     "TempRegStore",
+     OCC::TempRegStore,
+     "tempRegStore",
+     {false, true, true, false, false, false, true, true, false, false, false,
+      false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::MinPrecXRegLoad,
+     "MinPrecXRegLoad",
+     OCC::MinPrecXRegLoad,
+     "minPrecXRegLoad",
+     {false, true, false, false, false, false, true, false, false, false, false,
+      false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::MinPrecXRegStore,
+     "MinPrecXRegStore",
+     OCC::MinPrecXRegStore,
+     "minPrecXRegStore",
+     {false, true, false, false, false, false, true, false, false, false, false,
+      false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::LoadInput,
+     "LoadInput",
+     OCC::LoadInput,
+     "loadInput",
+     {false, true, true, false, false, false, true, true, false, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::StoreOutput,
+     "StoreOutput",
+     OCC::StoreOutput,
+     "storeOutput",
+     {false, true, true, false, false, false, true, true, false, false, false,
+      false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Unary float void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::FAbs,
-        "FAbs",
-        OCC::Unary,
-        "unary",
-        {false, true, true, true, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Saturate,
-        "Saturate",
-        OCC::Unary,
-        "unary",
-        {false, true, true, true, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::IsNaN,
-        "IsNaN",
-        OCC::IsSpecialFloat,
-        "isSpecialFloat",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::IsInf,
-        "IsInf",
-        OCC::IsSpecialFloat,
-        "isSpecialFloat",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::IsFinite,
-        "IsFinite",
-        OCC::IsSpecialFloat,
-        "isSpecialFloat",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::IsNormal,
-        "IsNormal",
-        OCC::IsSpecialFloat,
-        "isSpecialFloat",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Cos,
-        "Cos",
-        OCC::Unary,
-        "unary",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Sin,
-        "Sin",
-        OCC::Unary,
-        "unary",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Tan,
-        "Tan",
-        OCC::Unary,
-        "unary",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Acos,
-        "Acos",
-        OCC::Unary,
-        "unary",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Asin,
-        "Asin",
-        OCC::Unary,
-        "unary",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Atan,
-        "Atan",
-        OCC::Unary,
-        "unary",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Hcos,
-        "Hcos",
-        OCC::Unary,
-        "unary",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Hsin,
-        "Hsin",
-        OCC::Unary,
-        "unary",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Htan,
-        "Htan",
-        OCC::Unary,
-        "unary",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Exp,
-        "Exp",
-        OCC::Unary,
-        "unary",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Frc,
-        "Frc",
-        OCC::Unary,
-        "unary",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Log,
-        "Log",
-        OCC::Unary,
-        "unary",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Sqrt,
-        "Sqrt",
-        OCC::Unary,
-        "unary",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Rsqrt,
-        "Rsqrt",
-        OCC::Unary,
-        "unary",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::FAbs,
+     "FAbs",
+     OCC::Unary,
+     "unary",
+     {false, true, true, true, false, false, false, false, false, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::Saturate,
+     "Saturate",
+     OCC::Unary,
+     "unary",
+     {false, true, true, true, false, false, false, false, false, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::IsNaN,
+     "IsNaN",
+     OCC::IsSpecialFloat,
+     "isSpecialFloat",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::IsInf,
+     "IsInf",
+     OCC::IsSpecialFloat,
+     "isSpecialFloat",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::IsFinite,
+     "IsFinite",
+     OCC::IsSpecialFloat,
+     "isSpecialFloat",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::IsNormal,
+     "IsNormal",
+     OCC::IsSpecialFloat,
+     "isSpecialFloat",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::Cos,
+     "Cos",
+     OCC::Unary,
+     "unary",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x6}, {0x0}}},
+    {OC::Sin,
+     "Sin",
+     OCC::Unary,
+     "unary",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x6}, {0x0}}},
+    {OC::Tan,
+     "Tan",
+     OCC::Unary,
+     "unary",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x6}, {0x0}}},
+    {OC::Acos,
+     "Acos",
+     OCC::Unary,
+     "unary",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x6}, {0x0}}},
+    {OC::Asin,
+     "Asin",
+     OCC::Unary,
+     "unary",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x6}, {0x0}}},
+    {OC::Atan,
+     "Atan",
+     OCC::Unary,
+     "unary",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x6}, {0x0}}},
+    {OC::Hcos,
+     "Hcos",
+     OCC::Unary,
+     "unary",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x6}, {0x0}}},
+    {OC::Hsin,
+     "Hsin",
+     OCC::Unary,
+     "unary",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x6}, {0x0}}},
+    {OC::Htan,
+     "Htan",
+     OCC::Unary,
+     "unary",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x6}, {0x0}}},
+    {OC::Exp,
+     "Exp",
+     OCC::Unary,
+     "unary",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x6}, {0x0}}},
+    {OC::Frc,
+     "Frc",
+     OCC::Unary,
+     "unary",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x6}, {0x0}}},
+    {OC::Log,
+     "Log",
+     OCC::Unary,
+     "unary",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x6}, {0x0}}},
+    {OC::Sqrt,
+     "Sqrt",
+     OCC::Unary,
+     "unary",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x6}, {0x0}}},
+    {OC::Rsqrt,
+     "Rsqrt",
+     OCC::Unary,
+     "unary",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x6}, {0x0}}},
 
     // Unary float - rounding void,     h,     f,     d,    i1,    i8,   i16,
-    // i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::Round_ne,
-        "Round_ne",
-        OCC::Unary,
-        "unary",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Round_ni,
-        "Round_ni",
-        OCC::Unary,
-        "unary",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Round_pi,
-        "Round_pi",
-        OCC::Unary,
-        "unary",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Round_z,
-        "Round_z",
-        OCC::Unary,
-        "unary",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i32,   i64,   udt,   obj,   vec,  function attribute, ext oload, vec
+    // oload
+    {OC::Round_ne,
+     "Round_ne",
+     OCC::Unary,
+     "unary",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x6}, {0x0}}},
+    {OC::Round_ni,
+     "Round_ni",
+     OCC::Unary,
+     "unary",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x6}, {0x0}}},
+    {OC::Round_pi,
+     "Round_pi",
+     OCC::Unary,
+     "unary",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x6}, {0x0}}},
+    {OC::Round_z,
+     "Round_z",
+     OCC::Unary,
+     "unary",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x6}, {0x0}}},
 
     // Unary int void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::Bfrev,
-        "Bfrev",
-        OCC::Unary,
-        "unary",
-        {false, false, false, false, false, false, true, true, true, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Countbits,
-        "Countbits",
-        OCC::UnaryBits,
-        "unaryBits",
-        {false, false, false, false, false, false, true, true, true, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::FirstbitLo,
-        "FirstbitLo",
-        OCC::UnaryBits,
-        "unaryBits",
-        {false, false, false, false, false, false, true, true, true, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::Bfrev,
+     "Bfrev",
+     OCC::Unary,
+     "unary",
+     {false, false, false, false, false, false, true, true, true, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::Countbits,
+     "Countbits",
+     OCC::UnaryBits,
+     "unaryBits",
+     {false, false, false, false, false, false, true, true, true, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::FirstbitLo,
+     "FirstbitLo",
+     OCC::UnaryBits,
+     "unaryBits",
+     {false, false, false, false, false, false, true, true, true, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Unary uint void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::FirstbitHi,
-        "FirstbitHi",
-        OCC::UnaryBits,
-        "unaryBits",
-        {false, false, false, false, false, false, true, true, true, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::FirstbitHi,
+     "FirstbitHi",
+     OCC::UnaryBits,
+     "unaryBits",
+     {false, false, false, false, false, false, true, true, true, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Unary int void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::FirstbitSHi,
-        "FirstbitSHi",
-        OCC::UnaryBits,
-        "unaryBits",
-        {false, false, false, false, false, false, true, true, true, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::FirstbitSHi,
+     "FirstbitSHi",
+     OCC::UnaryBits,
+     "unaryBits",
+     {false, false, false, false, false, false, true, true, true, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Binary float void,     h,     f,     d,    i1,    i8,   i16,   i32, i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::FMax,
-        "FMax",
-        OCC::Binary,
-        "binary",
-        {false, true, true, true, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::FMin,
-        "FMin",
-        OCC::Binary,
-        "binary",
-        {false, true, true, true, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::FMax,
+     "FMax",
+     OCC::Binary,
+     "binary",
+     {false, true, true, true, false, false, false, false, false, false, false,
+      true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0xe}, {0x0}}},
+    {OC::FMin,
+     "FMin",
+     OCC::Binary,
+     "binary",
+     {false, true, true, true, false, false, false, false, false, false, false,
+      true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0xe}, {0x0}}},
 
     // Binary int void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::IMax,
-        "IMax",
-        OCC::Binary,
-        "binary",
-        {false, false, false, false, false, false, true, true, true, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::IMin,
-        "IMin",
-        OCC::Binary,
-        "binary",
-        {false, false, false, false, false, false, true, true, true, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::IMax,
+     "IMax",
+     OCC::Binary,
+     "binary",
+     {false, false, false, false, false, false, true, true, true, false, false,
+      true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x1c0}, {0x0}}},
+    {OC::IMin,
+     "IMin",
+     OCC::Binary,
+     "binary",
+     {false, false, false, false, false, false, true, true, true, false, false,
+      true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x1c0}, {0x0}}},
 
     // Binary uint void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::UMax,
-        "UMax",
-        OCC::Binary,
-        "binary",
-        {false, false, false, false, false, false, true, true, true, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::UMin,
-        "UMin",
-        OCC::Binary,
-        "binary",
-        {false, false, false, false, false, false, true, true, true, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::UMax,
+     "UMax",
+     OCC::Binary,
+     "binary",
+     {false, false, false, false, false, false, true, true, true, false, false,
+      true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x1c0}, {0x0}}},
+    {OC::UMin,
+     "UMin",
+     OCC::Binary,
+     "binary",
+     {false, false, false, false, false, false, true, true, true, false, false,
+      true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x1c0}, {0x0}}},
 
     // Binary int with two outputs void,     h,     f,     d,    i1,    i8, i16,
-    // i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::IMul,
-        "IMul",
-        OCC::BinaryWithTwoOuts,
-        "binaryWithTwoOuts",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i32,   i64,   udt,   obj,   vec,  function attribute, ext oload, vec
+    // oload
+    {OC::IMul,
+     "IMul",
+     OCC::BinaryWithTwoOuts,
+     "binaryWithTwoOuts",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Binary uint with two outputs void,     h,     f,     d,    i1,    i8,
-    // i16,   i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::UMul,
-        "UMul",
-        OCC::BinaryWithTwoOuts,
-        "binaryWithTwoOuts",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::UDiv,
-        "UDiv",
-        OCC::BinaryWithTwoOuts,
-        "binaryWithTwoOuts",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i16,   i32,   i64,   udt,   obj,   vec,  function attribute, ext oload,
+    // vec oload
+    {OC::UMul,
+     "UMul",
+     OCC::BinaryWithTwoOuts,
+     "binaryWithTwoOuts",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::UDiv,
+     "UDiv",
+     OCC::BinaryWithTwoOuts,
+     "binaryWithTwoOuts",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Binary uint with carry or borrow void,     h,     f,     d,    i1,    i8,
-    // i16,   i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::UAddc,
-        "UAddc",
-        OCC::BinaryWithCarryOrBorrow,
-        "binaryWithCarryOrBorrow",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::USubb,
-        "USubb",
-        OCC::BinaryWithCarryOrBorrow,
-        "binaryWithCarryOrBorrow",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i16,   i32,   i64,   udt,   obj,   vec,  function attribute, ext oload,
+    // vec oload
+    {OC::UAddc,
+     "UAddc",
+     OCC::BinaryWithCarryOrBorrow,
+     "binaryWithCarryOrBorrow",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::USubb,
+     "USubb",
+     OCC::BinaryWithCarryOrBorrow,
+     "binaryWithCarryOrBorrow",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Tertiary float void,     h,     f,     d,    i1,    i8,   i16,   i32,
-    // i64,   udt,   obj ,  function attribute
-    {
-        OC::FMad,
-        "FMad",
-        OCC::Tertiary,
-        "tertiary",
-        {false, true, true, true, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Fma,
-        "Fma",
-        OCC::Tertiary,
-        "tertiary",
-        {false, false, false, true, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::FMad,
+     "FMad",
+     OCC::Tertiary,
+     "tertiary",
+     {false, true, true, true, false, false, false, false, false, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::Fma,
+     "Fma",
+     OCC::Tertiary,
+     "tertiary",
+     {false, false, false, true, false, false, false, false, false, false,
+      false, true, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x8}, {0x0}}},
 
     // Tertiary int void,     h,     f,     d,    i1,    i8,   i16,   i32, i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::IMad,
-        "IMad",
-        OCC::Tertiary,
-        "tertiary",
-        {false, false, false, false, false, false, true, true, true, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::IMad,
+     "IMad",
+     OCC::Tertiary,
+     "tertiary",
+     {false, false, false, false, false, false, true, true, true, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Tertiary uint void,     h,     f,     d,    i1,    i8,   i16,   i32, i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::UMad,
-        "UMad",
-        OCC::Tertiary,
-        "tertiary",
-        {false, false, false, false, false, false, true, true, true, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::UMad,
+     "UMad",
+     OCC::Tertiary,
+     "tertiary",
+     {false, false, false, false, false, false, true, true, true, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Tertiary int void,     h,     f,     d,    i1,    i8,   i16,   i32, i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::Msad,
-        "Msad",
-        OCC::Tertiary,
-        "tertiary",
-        {false, false, false, false, false, false, false, true, true, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Ibfe,
-        "Ibfe",
-        OCC::Tertiary,
-        "tertiary",
-        {false, false, false, false, false, false, false, true, true, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::Msad,
+     "Msad",
+     OCC::Tertiary,
+     "tertiary",
+     {false, false, false, false, false, false, false, true, true, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::Ibfe,
+     "Ibfe",
+     OCC::Tertiary,
+     "tertiary",
+     {false, false, false, false, false, false, false, true, true, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Tertiary uint void,     h,     f,     d,    i1,    i8,   i16,   i32, i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::Ubfe,
-        "Ubfe",
-        OCC::Tertiary,
-        "tertiary",
-        {false, false, false, false, false, false, false, true, true, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::Ubfe,
+     "Ubfe",
+     OCC::Tertiary,
+     "tertiary",
+     {false, false, false, false, false, false, false, true, true, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Quaternary void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::Bfi,
-        "Bfi",
-        OCC::Quaternary,
-        "quaternary",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::Bfi,
+     "Bfi",
+     OCC::Quaternary,
+     "quaternary",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Dot void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,   udt,
-    // obj ,  function attribute
-    {
-        OC::Dot2,
-        "Dot2",
-        OCC::Dot2,
-        "dot2",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Dot3,
-        "Dot3",
-        OCC::Dot3,
-        "dot3",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Dot4,
-        "Dot4",
-        OCC::Dot4,
-        "dot4",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // obj,   vec,  function attribute, ext oload, vec oload
+    {OC::Dot2,
+     "Dot2",
+     OCC::Dot2,
+     "dot2",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::Dot3,
+     "Dot3",
+     OCC::Dot3,
+     "dot3",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::Dot4,
+     "Dot4",
+     OCC::Dot4,
+     "dot4",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Resources void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::CreateHandle,
-        "CreateHandle",
-        OCC::CreateHandle,
-        "createHandle",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::CBufferLoad,
-        "CBufferLoad",
-        OCC::CBufferLoad,
-        "cbufferLoad",
-        {false, true, true, true, false, true, true, true, true, false, false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::CBufferLoadLegacy,
-        "CBufferLoadLegacy",
-        OCC::CBufferLoadLegacy,
-        "cbufferLoadLegacy",
-        {false, true, true, true, false, false, true, true, true, false, false},
-        Attribute::ReadOnly,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::CreateHandle,
+     "CreateHandle",
+     OCC::CreateHandle,
+     "createHandle",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::CBufferLoad,
+     "CBufferLoad",
+     OCC::CBufferLoad,
+     "cbufferLoad",
+     {false, true, true, true, false, true, true, true, true, false, false,
+      false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::CBufferLoadLegacy,
+     "CBufferLoadLegacy",
+     OCC::CBufferLoadLegacy,
+     "cbufferLoadLegacy",
+     {false, true, true, true, false, false, true, true, true, false, false,
+      false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Resources - sample void,     h,     f,     d,    i1,    i8,   i16,   i32,
-    // i64,   udt,   obj ,  function attribute
-    {
-        OC::Sample,
-        "Sample",
-        OCC::Sample,
-        "sample",
-        {false, true, true, false, false, false, true, true, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::SampleBias,
-        "SampleBias",
-        OCC::SampleBias,
-        "sampleBias",
-        {false, true, true, false, false, false, true, true, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::SampleLevel,
-        "SampleLevel",
-        OCC::SampleLevel,
-        "sampleLevel",
-        {false, true, true, false, false, false, true, true, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::SampleGrad,
-        "SampleGrad",
-        OCC::SampleGrad,
-        "sampleGrad",
-        {false, true, true, false, false, false, true, true, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::SampleCmp,
-        "SampleCmp",
-        OCC::SampleCmp,
-        "sampleCmp",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::SampleCmpLevelZero,
-        "SampleCmpLevelZero",
-        OCC::SampleCmpLevelZero,
-        "sampleCmpLevelZero",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
+    // i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::Sample,
+     "Sample",
+     OCC::Sample,
+     "sample",
+     {false, true, true, false, false, false, true, true, false, false, false,
+      false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::SampleBias,
+     "SampleBias",
+     OCC::SampleBias,
+     "sampleBias",
+     {false, true, true, false, false, false, true, true, false, false, false,
+      false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::SampleLevel,
+     "SampleLevel",
+     OCC::SampleLevel,
+     "sampleLevel",
+     {false, true, true, false, false, false, true, true, false, false, false,
+      false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::SampleGrad,
+     "SampleGrad",
+     OCC::SampleGrad,
+     "sampleGrad",
+     {false, true, true, false, false, false, true, true, false, false, false,
+      false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::SampleCmp,
+     "SampleCmp",
+     OCC::SampleCmp,
+     "sampleCmp",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::SampleCmpLevelZero,
+     "SampleCmpLevelZero",
+     OCC::SampleCmpLevelZero,
+     "sampleCmpLevelZero",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Resources void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::TextureLoad,
-        "TextureLoad",
-        OCC::TextureLoad,
-        "textureLoad",
-        {false, true, true, false, false, false, true, true, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::TextureStore,
-        "TextureStore",
-        OCC::TextureStore,
-        "textureStore",
-        {false, true, true, false, false, false, true, true, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::BufferLoad,
-        "BufferLoad",
-        OCC::BufferLoad,
-        "bufferLoad",
-        {false, true, true, false, false, false, true, true, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::BufferStore,
-        "BufferStore",
-        OCC::BufferStore,
-        "bufferStore",
-        {false, true, true, false, false, false, true, true, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::BufferUpdateCounter,
-        "BufferUpdateCounter",
-        OCC::BufferUpdateCounter,
-        "bufferUpdateCounter",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::CheckAccessFullyMapped,
-        "CheckAccessFullyMapped",
-        OCC::CheckAccessFullyMapped,
-        "checkAccessFullyMapped",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::GetDimensions,
-        "GetDimensions",
-        OCC::GetDimensions,
-        "getDimensions",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::TextureLoad,
+     "TextureLoad",
+     OCC::TextureLoad,
+     "textureLoad",
+     {false, true, true, false, false, false, true, true, false, false, false,
+      false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::TextureStore,
+     "TextureStore",
+     OCC::TextureStore,
+     "textureStore",
+     {false, true, true, false, false, false, true, true, false, false, false,
+      false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::BufferLoad,
+     "BufferLoad",
+     OCC::BufferLoad,
+     "bufferLoad",
+     {false, true, true, false, false, false, true, true, false, false, false,
+      false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::BufferStore,
+     "BufferStore",
+     OCC::BufferStore,
+     "bufferStore",
+     {false, true, true, false, false, false, true, true, false, false, false,
+      false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::BufferUpdateCounter,
+     "BufferUpdateCounter",
+     OCC::BufferUpdateCounter,
+     "bufferUpdateCounter",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::CheckAccessFullyMapped,
+     "CheckAccessFullyMapped",
+     OCC::CheckAccessFullyMapped,
+     "checkAccessFullyMapped",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::GetDimensions,
+     "GetDimensions",
+     OCC::GetDimensions,
+     "getDimensions",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Resources - gather void,     h,     f,     d,    i1,    i8,   i16,   i32,
-    // i64,   udt,   obj ,  function attribute
-    {
-        OC::TextureGather,
-        "TextureGather",
-        OCC::TextureGather,
-        "textureGather",
-        {false, true, true, false, false, false, true, true, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::TextureGatherCmp,
-        "TextureGatherCmp",
-        OCC::TextureGatherCmp,
-        "textureGatherCmp",
-        {false, true, true, false, false, false, true, true, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
+    // i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::TextureGather,
+     "TextureGather",
+     OCC::TextureGather,
+     "textureGather",
+     {false, true, true, false, false, false, true, true, false, false, false,
+      false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::TextureGatherCmp,
+     "TextureGatherCmp",
+     OCC::TextureGatherCmp,
+     "textureGatherCmp",
+     {false, true, true, false, false, false, true, true, false, false, false,
+      false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Resources - sample void,     h,     f,     d,    i1,    i8,   i16,   i32,
-    // i64,   udt,   obj ,  function attribute
-    {
-        OC::Texture2DMSGetSamplePosition,
-        "Texture2DMSGetSamplePosition",
-        OCC::Texture2DMSGetSamplePosition,
-        "texture2DMSGetSamplePosition",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RenderTargetGetSamplePosition,
-        "RenderTargetGetSamplePosition",
-        OCC::RenderTargetGetSamplePosition,
-        "renderTargetGetSamplePosition",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RenderTargetGetSampleCount,
-        "RenderTargetGetSampleCount",
-        OCC::RenderTargetGetSampleCount,
-        "renderTargetGetSampleCount",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
+    // i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::Texture2DMSGetSamplePosition,
+     "Texture2DMSGetSamplePosition",
+     OCC::Texture2DMSGetSamplePosition,
+     "texture2DMSGetSamplePosition",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RenderTargetGetSamplePosition,
+     "RenderTargetGetSamplePosition",
+     OCC::RenderTargetGetSamplePosition,
+     "renderTargetGetSamplePosition",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RenderTargetGetSampleCount,
+     "RenderTargetGetSampleCount",
+     OCC::RenderTargetGetSampleCount,
+     "renderTargetGetSampleCount",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Synchronization void,     h,     f,     d,    i1,    i8,   i16,   i32,
-    // i64,   udt,   obj ,  function attribute
-    {
-        OC::AtomicBinOp,
-        "AtomicBinOp",
-        OCC::AtomicBinOp,
-        "atomicBinOp",
-        {false, false, false, false, false, false, false, true, true, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::AtomicCompareExchange,
-        "AtomicCompareExchange",
-        OCC::AtomicCompareExchange,
-        "atomicCompareExchange",
-        {false, false, false, false, false, false, false, true, true, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::Barrier,
-        "Barrier",
-        OCC::Barrier,
-        "barrier",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::NoDuplicate,
-    },
+    // i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::AtomicBinOp,
+     "AtomicBinOp",
+     OCC::AtomicBinOp,
+     "atomicBinOp",
+     {false, false, false, false, false, false, false, true, true, false, false,
+      false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::AtomicCompareExchange,
+     "AtomicCompareExchange",
+     OCC::AtomicCompareExchange,
+     "atomicCompareExchange",
+     {false, false, false, false, false, false, false, true, true, false, false,
+      false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::Barrier,
+     "Barrier",
+     OCC::Barrier,
+     "barrier",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::NoDuplicate,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Derivatives void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::CalculateLOD,
-        "CalculateLOD",
-        OCC::CalculateLOD,
-        "calculateLOD",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::CalculateLOD,
+     "CalculateLOD",
+     OCC::CalculateLOD,
+     "calculateLOD",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Pixel shader void,     h,     f,     d,    i1,    i8,   i16,   i32, i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::Discard,
-        "Discard",
-        OCC::Discard,
-        "discard",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::Discard,
+     "Discard",
+     OCC::Discard,
+     "discard",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Derivatives void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::DerivCoarseX,
-        "DerivCoarseX",
-        OCC::Unary,
-        "unary",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::DerivCoarseY,
-        "DerivCoarseY",
-        OCC::Unary,
-        "unary",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::DerivFineX,
-        "DerivFineX",
-        OCC::Unary,
-        "unary",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::DerivFineY,
-        "DerivFineY",
-        OCC::Unary,
-        "unary",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::DerivCoarseX,
+     "DerivCoarseX",
+     OCC::Unary,
+     "unary",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::DerivCoarseY,
+     "DerivCoarseY",
+     OCC::Unary,
+     "unary",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::DerivFineX,
+     "DerivFineX",
+     OCC::Unary,
+     "unary",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::DerivFineY,
+     "DerivFineY",
+     OCC::Unary,
+     "unary",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Pixel shader void,     h,     f,     d,    i1,    i8,   i16,   i32, i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::EvalSnapped,
-        "EvalSnapped",
-        OCC::EvalSnapped,
-        "evalSnapped",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::EvalSampleIndex,
-        "EvalSampleIndex",
-        OCC::EvalSampleIndex,
-        "evalSampleIndex",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::EvalCentroid,
-        "EvalCentroid",
-        OCC::EvalCentroid,
-        "evalCentroid",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::SampleIndex,
-        "SampleIndex",
-        OCC::SampleIndex,
-        "sampleIndex",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Coverage,
-        "Coverage",
-        OCC::Coverage,
-        "coverage",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::InnerCoverage,
-        "InnerCoverage",
-        OCC::InnerCoverage,
-        "innerCoverage",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::EvalSnapped,
+     "EvalSnapped",
+     OCC::EvalSnapped,
+     "evalSnapped",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::EvalSampleIndex,
+     "EvalSampleIndex",
+     OCC::EvalSampleIndex,
+     "evalSampleIndex",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::EvalCentroid,
+     "EvalCentroid",
+     OCC::EvalCentroid,
+     "evalCentroid",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::SampleIndex,
+     "SampleIndex",
+     OCC::SampleIndex,
+     "sampleIndex",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::Coverage,
+     "Coverage",
+     OCC::Coverage,
+     "coverage",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::InnerCoverage,
+     "InnerCoverage",
+     OCC::InnerCoverage,
+     "innerCoverage",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Compute/Mesh/Amplification/Node shader void,     h,     f,     d,    i1,
-    // i8,   i16,   i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::ThreadId,
-        "ThreadId",
-        OCC::ThreadId,
-        "threadId",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::GroupId,
-        "GroupId",
-        OCC::GroupId,
-        "groupId",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::ThreadIdInGroup,
-        "ThreadIdInGroup",
-        OCC::ThreadIdInGroup,
-        "threadIdInGroup",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::FlattenedThreadIdInGroup,
-        "FlattenedThreadIdInGroup",
-        OCC::FlattenedThreadIdInGroup,
-        "flattenedThreadIdInGroup",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i8,   i16,   i32,   i64,   udt,   obj,   vec,  function attribute, ext
+    // oload, vec oload
+    {OC::ThreadId,
+     "ThreadId",
+     OCC::ThreadId,
+     "threadId",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::GroupId,
+     "GroupId",
+     OCC::GroupId,
+     "groupId",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ThreadIdInGroup,
+     "ThreadIdInGroup",
+     OCC::ThreadIdInGroup,
+     "threadIdInGroup",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::FlattenedThreadIdInGroup,
+     "FlattenedThreadIdInGroup",
+     OCC::FlattenedThreadIdInGroup,
+     "flattenedThreadIdInGroup",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Geometry shader void,     h,     f,     d,    i1,    i8,   i16,   i32,
-    // i64,   udt,   obj ,  function attribute
-    {
-        OC::EmitStream,
-        "EmitStream",
-        OCC::EmitStream,
-        "emitStream",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::CutStream,
-        "CutStream",
-        OCC::CutStream,
-        "cutStream",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::EmitThenCutStream,
-        "EmitThenCutStream",
-        OCC::EmitThenCutStream,
-        "emitThenCutStream",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::GSInstanceID,
-        "GSInstanceID",
-        OCC::GSInstanceID,
-        "gsInstanceID",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::EmitStream,
+     "EmitStream",
+     OCC::EmitStream,
+     "emitStream",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::CutStream,
+     "CutStream",
+     OCC::CutStream,
+     "cutStream",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::EmitThenCutStream,
+     "EmitThenCutStream",
+     OCC::EmitThenCutStream,
+     "emitThenCutStream",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::GSInstanceID,
+     "GSInstanceID",
+     OCC::GSInstanceID,
+     "gsInstanceID",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Double precision void,     h,     f,     d,    i1,    i8,   i16,   i32,
-    // i64,   udt,   obj ,  function attribute
-    {
-        OC::MakeDouble,
-        "MakeDouble",
-        OCC::MakeDouble,
-        "makeDouble",
-        {false, false, false, true, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::SplitDouble,
-        "SplitDouble",
-        OCC::SplitDouble,
-        "splitDouble",
-        {false, false, false, true, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::MakeDouble,
+     "MakeDouble",
+     OCC::MakeDouble,
+     "makeDouble",
+     {false, false, false, true, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::SplitDouble,
+     "SplitDouble",
+     OCC::SplitDouble,
+     "splitDouble",
+     {false, false, false, true, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Domain and hull shader void,     h,     f,     d,    i1,    i8,   i16,
-    // i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::LoadOutputControlPoint,
-        "LoadOutputControlPoint",
-        OCC::LoadOutputControlPoint,
-        "loadOutputControlPoint",
-        {false, true, true, false, false, false, true, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::LoadPatchConstant,
-        "LoadPatchConstant",
-        OCC::LoadPatchConstant,
-        "loadPatchConstant",
-        {false, true, true, false, false, false, true, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i32,   i64,   udt,   obj,   vec,  function attribute, ext oload, vec
+    // oload
+    {OC::LoadOutputControlPoint,
+     "LoadOutputControlPoint",
+     OCC::LoadOutputControlPoint,
+     "loadOutputControlPoint",
+     {false, true, true, false, false, false, true, true, false, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::LoadPatchConstant,
+     "LoadPatchConstant",
+     OCC::LoadPatchConstant,
+     "loadPatchConstant",
+     {false, true, true, false, false, false, true, true, false, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Domain shader void,     h,     f,     d,    i1,    i8,   i16,   i32, i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::DomainLocation,
-        "DomainLocation",
-        OCC::DomainLocation,
-        "domainLocation",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::DomainLocation,
+     "DomainLocation",
+     OCC::DomainLocation,
+     "domainLocation",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Hull shader void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::StorePatchConstant,
-        "StorePatchConstant",
-        OCC::StorePatchConstant,
-        "storePatchConstant",
-        {false, true, true, false, false, false, true, true, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::OutputControlPointID,
-        "OutputControlPointID",
-        OCC::OutputControlPointID,
-        "outputControlPointID",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::StorePatchConstant,
+     "StorePatchConstant",
+     OCC::StorePatchConstant,
+     "storePatchConstant",
+     {false, true, true, false, false, false, true, true, false, false, false,
+      false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::OutputControlPointID,
+     "OutputControlPointID",
+     OCC::OutputControlPointID,
+     "outputControlPointID",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Hull, Domain and Geometry shaders void,     h,     f,     d,    i1, i8,
-    // i16,   i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::PrimitiveID,
-        "PrimitiveID",
-        OCC::PrimitiveID,
-        "primitiveID",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i16,   i32,   i64,   udt,   obj,   vec,  function attribute, ext oload,
+    // vec oload
+    {OC::PrimitiveID,
+     "PrimitiveID",
+     OCC::PrimitiveID,
+     "primitiveID",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Other void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64, udt,
-    // obj ,  function attribute
-    {
-        OC::CycleCounterLegacy,
-        "CycleCounterLegacy",
-        OCC::CycleCounterLegacy,
-        "cycleCounterLegacy",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
+    // obj,   vec,  function attribute, ext oload, vec oload
+    {OC::CycleCounterLegacy,
+     "CycleCounterLegacy",
+     OCC::CycleCounterLegacy,
+     "cycleCounterLegacy",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Wave void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,   udt,
-    // obj ,  function attribute
-    {
-        OC::WaveIsFirstLane,
-        "WaveIsFirstLane",
-        OCC::WaveIsFirstLane,
-        "waveIsFirstLane",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::WaveGetLaneIndex,
-        "WaveGetLaneIndex",
-        OCC::WaveGetLaneIndex,
-        "waveGetLaneIndex",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::WaveGetLaneCount,
-        "WaveGetLaneCount",
-        OCC::WaveGetLaneCount,
-        "waveGetLaneCount",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::WaveAnyTrue,
-        "WaveAnyTrue",
-        OCC::WaveAnyTrue,
-        "waveAnyTrue",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::WaveAllTrue,
-        "WaveAllTrue",
-        OCC::WaveAllTrue,
-        "waveAllTrue",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::WaveActiveAllEqual,
-        "WaveActiveAllEqual",
-        OCC::WaveActiveAllEqual,
-        "waveActiveAllEqual",
-        {false, true, true, true, true, true, true, true, true, false, false},
-        Attribute::None,
-    },
-    {
-        OC::WaveActiveBallot,
-        "WaveActiveBallot",
-        OCC::WaveActiveBallot,
-        "waveActiveBallot",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::WaveReadLaneAt,
-        "WaveReadLaneAt",
-        OCC::WaveReadLaneAt,
-        "waveReadLaneAt",
-        {false, true, true, true, true, true, true, true, true, false, false},
-        Attribute::None,
-    },
-    {
-        OC::WaveReadLaneFirst,
-        "WaveReadLaneFirst",
-        OCC::WaveReadLaneFirst,
-        "waveReadLaneFirst",
-        {false, true, true, true, true, true, true, true, true, false, false},
-        Attribute::None,
-    },
-    {
-        OC::WaveActiveOp,
-        "WaveActiveOp",
-        OCC::WaveActiveOp,
-        "waveActiveOp",
-        {false, true, true, true, true, true, true, true, true, false, false},
-        Attribute::None,
-    },
-    {
-        OC::WaveActiveBit,
-        "WaveActiveBit",
-        OCC::WaveActiveBit,
-        "waveActiveBit",
-        {false, false, false, false, false, true, true, true, true, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::WavePrefixOp,
-        "WavePrefixOp",
-        OCC::WavePrefixOp,
-        "wavePrefixOp",
-        {false, true, true, true, false, true, true, true, true, false, false},
-        Attribute::None,
-    },
+    // obj,   vec,  function attribute, ext oload, vec oload
+    {OC::WaveIsFirstLane,
+     "WaveIsFirstLane",
+     OCC::WaveIsFirstLane,
+     "waveIsFirstLane",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::WaveGetLaneIndex,
+     "WaveGetLaneIndex",
+     OCC::WaveGetLaneIndex,
+     "waveGetLaneIndex",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::WaveGetLaneCount,
+     "WaveGetLaneCount",
+     OCC::WaveGetLaneCount,
+     "waveGetLaneCount",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::WaveAnyTrue,
+     "WaveAnyTrue",
+     OCC::WaveAnyTrue,
+     "waveAnyTrue",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::WaveAllTrue,
+     "WaveAllTrue",
+     OCC::WaveAllTrue,
+     "waveAllTrue",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::WaveActiveAllEqual,
+     "WaveActiveAllEqual",
+     OCC::WaveActiveAllEqual,
+     "waveActiveAllEqual",
+     {false, true, true, true, true, true, true, true, true, false, false,
+      false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::WaveActiveBallot,
+     "WaveActiveBallot",
+     OCC::WaveActiveBallot,
+     "waveActiveBallot",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::WaveReadLaneAt,
+     "WaveReadLaneAt",
+     OCC::WaveReadLaneAt,
+     "waveReadLaneAt",
+     {false, true, true, true, true, true, true, true, true, false, false,
+      false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::WaveReadLaneFirst,
+     "WaveReadLaneFirst",
+     OCC::WaveReadLaneFirst,
+     "waveReadLaneFirst",
+     {false, true, true, true, true, true, true, true, true, false, false,
+      false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::WaveActiveOp,
+     "WaveActiveOp",
+     OCC::WaveActiveOp,
+     "waveActiveOp",
+     {false, true, true, true, true, true, true, true, true, false, false,
+      false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::WaveActiveBit,
+     "WaveActiveBit",
+     OCC::WaveActiveBit,
+     "waveActiveBit",
+     {false, false, false, false, false, true, true, true, true, false, false,
+      false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::WavePrefixOp,
+     "WavePrefixOp",
+     OCC::WavePrefixOp,
+     "wavePrefixOp",
+     {false, true, true, true, false, true, true, true, true, false, false,
+      false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Quad Wave Ops void,     h,     f,     d,    i1,    i8,   i16,   i32, i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::QuadReadLaneAt,
-        "QuadReadLaneAt",
-        OCC::QuadReadLaneAt,
-        "quadReadLaneAt",
-        {false, true, true, true, true, true, true, true, true, false, false},
-        Attribute::None,
-    },
-    {
-        OC::QuadOp,
-        "QuadOp",
-        OCC::QuadOp,
-        "quadOp",
-        {false, true, true, true, false, true, true, true, true, false, false},
-        Attribute::None,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::QuadReadLaneAt,
+     "QuadReadLaneAt",
+     OCC::QuadReadLaneAt,
+     "quadReadLaneAt",
+     {false, true, true, true, true, true, true, true, true, false, false,
+      false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::QuadOp,
+     "QuadOp",
+     OCC::QuadOp,
+     "quadOp",
+     {false, true, true, true, false, true, true, true, true, false, false,
+      false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Bitcasts with different sizes void,     h,     f,     d,    i1,    i8,
-    // i16,   i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::BitcastI16toF16,
-        "BitcastI16toF16",
-        OCC::BitcastI16toF16,
-        "bitcastI16toF16",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::BitcastF16toI16,
-        "BitcastF16toI16",
-        OCC::BitcastF16toI16,
-        "bitcastF16toI16",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::BitcastI32toF32,
-        "BitcastI32toF32",
-        OCC::BitcastI32toF32,
-        "bitcastI32toF32",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::BitcastF32toI32,
-        "BitcastF32toI32",
-        OCC::BitcastF32toI32,
-        "bitcastF32toI32",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::BitcastI64toF64,
-        "BitcastI64toF64",
-        OCC::BitcastI64toF64,
-        "bitcastI64toF64",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::BitcastF64toI64,
-        "BitcastF64toI64",
-        OCC::BitcastF64toI64,
-        "bitcastF64toI64",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i16,   i32,   i64,   udt,   obj,   vec,  function attribute, ext oload,
+    // vec oload
+    {OC::BitcastI16toF16,
+     "BitcastI16toF16",
+     OCC::BitcastI16toF16,
+     "bitcastI16toF16",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::BitcastF16toI16,
+     "BitcastF16toI16",
+     OCC::BitcastF16toI16,
+     "bitcastF16toI16",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::BitcastI32toF32,
+     "BitcastI32toF32",
+     OCC::BitcastI32toF32,
+     "bitcastI32toF32",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::BitcastF32toI32,
+     "BitcastF32toI32",
+     OCC::BitcastF32toI32,
+     "bitcastF32toI32",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::BitcastI64toF64,
+     "BitcastI64toF64",
+     OCC::BitcastI64toF64,
+     "bitcastI64toF64",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::BitcastF64toI64,
+     "BitcastF64toI64",
+     OCC::BitcastF64toI64,
+     "bitcastF64toI64",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Legacy floating-point void,     h,     f,     d,    i1,    i8,   i16,
-    // i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::LegacyF32ToF16,
-        "LegacyF32ToF16",
-        OCC::LegacyF32ToF16,
-        "legacyF32ToF16",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::LegacyF16ToF32,
-        "LegacyF16ToF32",
-        OCC::LegacyF16ToF32,
-        "legacyF16ToF32",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i32,   i64,   udt,   obj,   vec,  function attribute, ext oload, vec
+    // oload
+    {OC::LegacyF32ToF16,
+     "LegacyF32ToF16",
+     OCC::LegacyF32ToF16,
+     "legacyF32ToF16",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::LegacyF16ToF32,
+     "LegacyF16ToF32",
+     OCC::LegacyF16ToF32,
+     "legacyF16ToF32",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Double precision void,     h,     f,     d,    i1,    i8,   i16,   i32,
-    // i64,   udt,   obj ,  function attribute
-    {
-        OC::LegacyDoubleToFloat,
-        "LegacyDoubleToFloat",
-        OCC::LegacyDoubleToFloat,
-        "legacyDoubleToFloat",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::LegacyDoubleToSInt32,
-        "LegacyDoubleToSInt32",
-        OCC::LegacyDoubleToSInt32,
-        "legacyDoubleToSInt32",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::LegacyDoubleToUInt32,
-        "LegacyDoubleToUInt32",
-        OCC::LegacyDoubleToUInt32,
-        "legacyDoubleToUInt32",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::LegacyDoubleToFloat,
+     "LegacyDoubleToFloat",
+     OCC::LegacyDoubleToFloat,
+     "legacyDoubleToFloat",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::LegacyDoubleToSInt32,
+     "LegacyDoubleToSInt32",
+     OCC::LegacyDoubleToSInt32,
+     "legacyDoubleToSInt32",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::LegacyDoubleToUInt32,
+     "LegacyDoubleToUInt32",
+     OCC::LegacyDoubleToUInt32,
+     "legacyDoubleToUInt32",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Wave void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,   udt,
-    // obj ,  function attribute
-    {
-        OC::WaveAllBitCount,
-        "WaveAllBitCount",
-        OCC::WaveAllOp,
-        "waveAllOp",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::WavePrefixBitCount,
-        "WavePrefixBitCount",
-        OCC::WavePrefixOp,
-        "wavePrefixOp",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
+    // obj,   vec,  function attribute, ext oload, vec oload
+    {OC::WaveAllBitCount,
+     "WaveAllBitCount",
+     OCC::WaveAllOp,
+     "waveAllOp",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::WavePrefixBitCount,
+     "WavePrefixBitCount",
+     OCC::WavePrefixOp,
+     "wavePrefixOp",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Pixel shader void,     h,     f,     d,    i1,    i8,   i16,   i32, i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::AttributeAtVertex,
-        "AttributeAtVertex",
-        OCC::AttributeAtVertex,
-        "attributeAtVertex",
-        {false, true, true, false, false, false, true, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::AttributeAtVertex,
+     "AttributeAtVertex",
+     OCC::AttributeAtVertex,
+     "attributeAtVertex",
+     {false, true, true, false, false, false, true, true, false, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Graphics shader void,     h,     f,     d,    i1,    i8,   i16,   i32,
-    // i64,   udt,   obj ,  function attribute
-    {
-        OC::ViewID,
-        "ViewID",
-        OCC::ViewID,
-        "viewID",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::ViewID,
+     "ViewID",
+     OCC::ViewID,
+     "viewID",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Resources void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::RawBufferLoad,
-        "RawBufferLoad",
-        OCC::RawBufferLoad,
-        "rawBufferLoad",
-        {false, true, true, true, false, false, true, true, true, false, false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RawBufferStore,
-        "RawBufferStore",
-        OCC::RawBufferStore,
-        "rawBufferStore",
-        {false, true, true, true, false, false, true, true, true, false, false},
-        Attribute::None,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::RawBufferLoad,
+     "RawBufferLoad",
+     OCC::RawBufferLoad,
+     "rawBufferLoad",
+     {false, true, true, true, false, false, true, true, true, false, false,
+      false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RawBufferStore,
+     "RawBufferStore",
+     OCC::RawBufferStore,
+     "rawBufferStore",
+     {false, true, true, true, false, false, true, true, true, false, false,
+      false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Raytracing object space uint System Values void,     h,     f,     d, i1,
-    // i8,   i16,   i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::InstanceID,
-        "InstanceID",
-        OCC::InstanceID,
-        "instanceID",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::InstanceIndex,
-        "InstanceIndex",
-        OCC::InstanceIndex,
-        "instanceIndex",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i8,   i16,   i32,   i64,   udt,   obj,   vec,  function attribute, ext
+    // oload, vec oload
+    {OC::InstanceID,
+     "InstanceID",
+     OCC::InstanceID,
+     "instanceID",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::InstanceIndex,
+     "InstanceIndex",
+     OCC::InstanceIndex,
+     "instanceIndex",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Raytracing hit uint System Values void,     h,     f,     d,    i1, i8,
-    // i16,   i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::HitKind,
-        "HitKind",
-        OCC::HitKind,
-        "hitKind",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i16,   i32,   i64,   udt,   obj,   vec,  function attribute, ext oload,
+    // vec oload
+    {OC::HitKind,
+     "HitKind",
+     OCC::HitKind,
+     "hitKind",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Raytracing uint System Values void,     h,     f,     d,    i1,    i8,
-    // i16,   i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::RayFlags,
-        "RayFlags",
-        OCC::RayFlags,
-        "rayFlags",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i16,   i32,   i64,   udt,   obj,   vec,  function attribute, ext oload,
+    // vec oload
+    {OC::RayFlags,
+     "RayFlags",
+     OCC::RayFlags,
+     "rayFlags",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Ray Dispatch Arguments void,     h,     f,     d,    i1,    i8,   i16,
-    // i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::DispatchRaysIndex,
-        "DispatchRaysIndex",
-        OCC::DispatchRaysIndex,
-        "dispatchRaysIndex",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::DispatchRaysDimensions,
-        "DispatchRaysDimensions",
-        OCC::DispatchRaysDimensions,
-        "dispatchRaysDimensions",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i32,   i64,   udt,   obj,   vec,  function attribute, ext oload, vec
+    // oload
+    {OC::DispatchRaysIndex,
+     "DispatchRaysIndex",
+     OCC::DispatchRaysIndex,
+     "dispatchRaysIndex",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::DispatchRaysDimensions,
+     "DispatchRaysDimensions",
+     OCC::DispatchRaysDimensions,
+     "dispatchRaysDimensions",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Ray Vectors void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::WorldRayOrigin,
-        "WorldRayOrigin",
-        OCC::WorldRayOrigin,
-        "worldRayOrigin",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::WorldRayDirection,
-        "WorldRayDirection",
-        OCC::WorldRayDirection,
-        "worldRayDirection",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::WorldRayOrigin,
+     "WorldRayOrigin",
+     OCC::WorldRayOrigin,
+     "worldRayOrigin",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::WorldRayDirection,
+     "WorldRayDirection",
+     OCC::WorldRayDirection,
+     "worldRayDirection",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Ray object space Vectors void,     h,     f,     d,    i1,    i8,   i16,
-    // i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::ObjectRayOrigin,
-        "ObjectRayOrigin",
-        OCC::ObjectRayOrigin,
-        "objectRayOrigin",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::ObjectRayDirection,
-        "ObjectRayDirection",
-        OCC::ObjectRayDirection,
-        "objectRayDirection",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i32,   i64,   udt,   obj,   vec,  function attribute, ext oload, vec
+    // oload
+    {OC::ObjectRayOrigin,
+     "ObjectRayOrigin",
+     OCC::ObjectRayOrigin,
+     "objectRayOrigin",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ObjectRayDirection,
+     "ObjectRayDirection",
+     OCC::ObjectRayDirection,
+     "objectRayDirection",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Ray Transforms void,     h,     f,     d,    i1,    i8,   i16,   i32,
-    // i64,   udt,   obj ,  function attribute
-    {
-        OC::ObjectToWorld,
-        "ObjectToWorld",
-        OCC::ObjectToWorld,
-        "objectToWorld",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::WorldToObject,
-        "WorldToObject",
-        OCC::WorldToObject,
-        "worldToObject",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::ObjectToWorld,
+     "ObjectToWorld",
+     OCC::ObjectToWorld,
+     "objectToWorld",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::WorldToObject,
+     "WorldToObject",
+     OCC::WorldToObject,
+     "worldToObject",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // RayT void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,   udt,
-    // obj ,  function attribute
-    {
-        OC::RayTMin,
-        "RayTMin",
-        OCC::RayTMin,
-        "rayTMin",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::RayTCurrent,
-        "RayTCurrent",
-        OCC::RayTCurrent,
-        "rayTCurrent",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
+    // obj,   vec,  function attribute, ext oload, vec oload
+    {OC::RayTMin,
+     "RayTMin",
+     OCC::RayTMin,
+     "rayTMin",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayTCurrent,
+     "RayTCurrent",
+     OCC::RayTCurrent,
+     "rayTCurrent",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // AnyHit Terminals void,     h,     f,     d,    i1,    i8,   i16,   i32,
-    // i64,   udt,   obj ,  function attribute
-    {
-        OC::IgnoreHit,
-        "IgnoreHit",
-        OCC::IgnoreHit,
-        "ignoreHit",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::NoReturn,
-    },
-    {
-        OC::AcceptHitAndEndSearch,
-        "AcceptHitAndEndSearch",
-        OCC::AcceptHitAndEndSearch,
-        "acceptHitAndEndSearch",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::NoReturn,
-    },
+    // i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::IgnoreHit,
+     "IgnoreHit",
+     OCC::IgnoreHit,
+     "ignoreHit",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::NoReturn,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::AcceptHitAndEndSearch,
+     "AcceptHitAndEndSearch",
+     OCC::AcceptHitAndEndSearch,
+     "acceptHitAndEndSearch",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::NoReturn,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Indirect Shader Invocation void,     h,     f,     d,    i1,    i8, i16,
-    // i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::TraceRay,
-        "TraceRay",
-        OCC::TraceRay,
-        "traceRay",
-        {false, false, false, false, false, false, false, false, false, true,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReportHit,
-        "ReportHit",
-        OCC::ReportHit,
-        "reportHit",
-        {false, false, false, false, false, false, false, false, false, true,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::CallShader,
-        "CallShader",
-        OCC::CallShader,
-        "callShader",
-        {false, false, false, false, false, false, false, false, false, true,
-         false},
-        Attribute::None,
-    },
+    // i32,   i64,   udt,   obj,   vec,  function attribute, ext oload, vec
+    // oload
+    {OC::TraceRay,
+     "TraceRay",
+     OCC::TraceRay,
+     "traceRay",
+     {false, false, false, false, false, false, false, false, false, true,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReportHit,
+     "ReportHit",
+     OCC::ReportHit,
+     "reportHit",
+     {false, false, false, false, false, false, false, false, false, true,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::CallShader,
+     "CallShader",
+     OCC::CallShader,
+     "callShader",
+     {false, false, false, false, false, false, false, false, false, true,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Library create handle from resource struct (like HL intrinsic) void, h,
-    // f,     d,    i1,    i8,   i16,   i32,   i64,   udt,   obj ,  function
-    // attribute
-    {
-        OC::CreateHandleForLib,
-        "CreateHandleForLib",
-        OCC::CreateHandleForLib,
-        "createHandleForLib",
-        {false, false, false, false, false, false, false, false, false, false,
-         true},
-        Attribute::ReadOnly,
-    },
+    // f,     d,    i1,    i8,   i16,   i32,   i64,   udt,   obj,   vec,
+    // function attribute, ext oload, vec oload
+    {OC::CreateHandleForLib,
+     "CreateHandleForLib",
+     OCC::CreateHandleForLib,
+     "createHandleForLib",
+     {false, false, false, false, false, false, false, false, false, false,
+      true, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Raytracing object space uint System Values void,     h,     f,     d, i1,
-    // i8,   i16,   i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::PrimitiveIndex,
-        "PrimitiveIndex",
-        OCC::PrimitiveIndex,
-        "primitiveIndex",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i8,   i16,   i32,   i64,   udt,   obj,   vec,  function attribute, ext
+    // oload, vec oload
+    {OC::PrimitiveIndex,
+     "PrimitiveIndex",
+     OCC::PrimitiveIndex,
+     "primitiveIndex",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Dot product with accumulate void,     h,     f,     d,    i1,    i8, i16,
-    // i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::Dot2AddHalf,
-        "Dot2AddHalf",
-        OCC::Dot2AddHalf,
-        "dot2AddHalf",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Dot4AddI8Packed,
-        "Dot4AddI8Packed",
-        OCC::Dot4AddPacked,
-        "dot4AddPacked",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::Dot4AddU8Packed,
-        "Dot4AddU8Packed",
-        OCC::Dot4AddPacked,
-        "dot4AddPacked",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i32,   i64,   udt,   obj,   vec,  function attribute, ext oload, vec
+    // oload
+    {OC::Dot2AddHalf,
+     "Dot2AddHalf",
+     OCC::Dot2AddHalf,
+     "dot2AddHalf",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::Dot4AddI8Packed,
+     "Dot4AddI8Packed",
+     OCC::Dot4AddPacked,
+     "dot4AddPacked",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::Dot4AddU8Packed,
+     "Dot4AddU8Packed",
+     OCC::Dot4AddPacked,
+     "dot4AddPacked",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Wave void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,   udt,
-    // obj ,  function attribute
-    {
-        OC::WaveMatch,
-        "WaveMatch",
-        OCC::WaveMatch,
-        "waveMatch",
-        {false, true, true, true, false, true, true, true, true, false, false},
-        Attribute::None,
-    },
-    {
-        OC::WaveMultiPrefixOp,
-        "WaveMultiPrefixOp",
-        OCC::WaveMultiPrefixOp,
-        "waveMultiPrefixOp",
-        {false, true, true, true, false, true, true, true, true, false, false},
-        Attribute::None,
-    },
-    {
-        OC::WaveMultiPrefixBitCount,
-        "WaveMultiPrefixBitCount",
-        OCC::WaveMultiPrefixBitCount,
-        "waveMultiPrefixBitCount",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
+    // obj,   vec,  function attribute, ext oload, vec oload
+    {OC::WaveMatch,
+     "WaveMatch",
+     OCC::WaveMatch,
+     "waveMatch",
+     {false, true, true, true, false, true, true, true, true, false, false,
+      false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::WaveMultiPrefixOp,
+     "WaveMultiPrefixOp",
+     OCC::WaveMultiPrefixOp,
+     "waveMultiPrefixOp",
+     {false, true, true, true, false, true, true, true, true, false, false,
+      false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::WaveMultiPrefixBitCount,
+     "WaveMultiPrefixBitCount",
+     OCC::WaveMultiPrefixBitCount,
+     "waveMultiPrefixBitCount",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Mesh shader instructions void,     h,     f,     d,    i1,    i8,   i16,
-    // i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::SetMeshOutputCounts,
-        "SetMeshOutputCounts",
-        OCC::SetMeshOutputCounts,
-        "setMeshOutputCounts",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::EmitIndices,
-        "EmitIndices",
-        OCC::EmitIndices,
-        "emitIndices",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::GetMeshPayload,
-        "GetMeshPayload",
-        OCC::GetMeshPayload,
-        "getMeshPayload",
-        {false, false, false, false, false, false, false, false, false, true,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::StoreVertexOutput,
-        "StoreVertexOutput",
-        OCC::StoreVertexOutput,
-        "storeVertexOutput",
-        {false, true, true, false, false, false, true, true, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::StorePrimitiveOutput,
-        "StorePrimitiveOutput",
-        OCC::StorePrimitiveOutput,
-        "storePrimitiveOutput",
-        {false, true, true, false, false, false, true, true, false, false,
-         false},
-        Attribute::None,
-    },
+    // i32,   i64,   udt,   obj,   vec,  function attribute, ext oload, vec
+    // oload
+    {OC::SetMeshOutputCounts,
+     "SetMeshOutputCounts",
+     OCC::SetMeshOutputCounts,
+     "setMeshOutputCounts",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::EmitIndices,
+     "EmitIndices",
+     OCC::EmitIndices,
+     "emitIndices",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::GetMeshPayload,
+     "GetMeshPayload",
+     OCC::GetMeshPayload,
+     "getMeshPayload",
+     {false, false, false, false, false, false, false, false, false, true,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::StoreVertexOutput,
+     "StoreVertexOutput",
+     OCC::StoreVertexOutput,
+     "storeVertexOutput",
+     {false, true, true, false, false, false, true, true, false, false, false,
+      false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::StorePrimitiveOutput,
+     "StorePrimitiveOutput",
+     OCC::StorePrimitiveOutput,
+     "storePrimitiveOutput",
+     {false, true, true, false, false, false, true, true, false, false, false,
+      false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Amplification shader instructions void,     h,     f,     d,    i1, i8,
-    // i16,   i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::DispatchMesh,
-        "DispatchMesh",
-        OCC::DispatchMesh,
-        "dispatchMesh",
-        {false, false, false, false, false, false, false, false, false, true,
-         false},
-        Attribute::None,
-    },
+    // i16,   i32,   i64,   udt,   obj,   vec,  function attribute, ext oload,
+    // vec oload
+    {OC::DispatchMesh,
+     "DispatchMesh",
+     OCC::DispatchMesh,
+     "dispatchMesh",
+     {false, false, false, false, false, false, false, false, false, true,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Sampler Feedback void,     h,     f,     d,    i1,    i8,   i16,   i32,
-    // i64,   udt,   obj ,  function attribute
-    {
-        OC::WriteSamplerFeedback,
-        "WriteSamplerFeedback",
-        OCC::WriteSamplerFeedback,
-        "writeSamplerFeedback",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::WriteSamplerFeedbackBias,
-        "WriteSamplerFeedbackBias",
-        OCC::WriteSamplerFeedbackBias,
-        "writeSamplerFeedbackBias",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::WriteSamplerFeedbackLevel,
-        "WriteSamplerFeedbackLevel",
-        OCC::WriteSamplerFeedbackLevel,
-        "writeSamplerFeedbackLevel",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::WriteSamplerFeedbackGrad,
-        "WriteSamplerFeedbackGrad",
-        OCC::WriteSamplerFeedbackGrad,
-        "writeSamplerFeedbackGrad",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
+    // i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::WriteSamplerFeedback,
+     "WriteSamplerFeedback",
+     OCC::WriteSamplerFeedback,
+     "writeSamplerFeedback",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::WriteSamplerFeedbackBias,
+     "WriteSamplerFeedbackBias",
+     OCC::WriteSamplerFeedbackBias,
+     "writeSamplerFeedbackBias",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::WriteSamplerFeedbackLevel,
+     "WriteSamplerFeedbackLevel",
+     OCC::WriteSamplerFeedbackLevel,
+     "writeSamplerFeedbackLevel",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::WriteSamplerFeedbackGrad,
+     "WriteSamplerFeedbackGrad",
+     OCC::WriteSamplerFeedbackGrad,
+     "writeSamplerFeedbackGrad",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Inline Ray Query void,     h,     f,     d,    i1,    i8,   i16,   i32,
-    // i64,   udt,   obj ,  function attribute
-    {
-        OC::AllocateRayQuery,
-        "AllocateRayQuery",
-        OCC::AllocateRayQuery,
-        "allocateRayQuery",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::RayQuery_TraceRayInline,
-        "RayQuery_TraceRayInline",
-        OCC::RayQuery_TraceRayInline,
-        "rayQuery_TraceRayInline",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::RayQuery_Proceed,
-        "RayQuery_Proceed",
-        OCC::RayQuery_Proceed,
-        "rayQuery_Proceed",
-        {false, false, false, false, true, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::RayQuery_Abort,
-        "RayQuery_Abort",
-        OCC::RayQuery_Abort,
-        "rayQuery_Abort",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::RayQuery_CommitNonOpaqueTriangleHit,
-        "RayQuery_CommitNonOpaqueTriangleHit",
-        OCC::RayQuery_CommitNonOpaqueTriangleHit,
-        "rayQuery_CommitNonOpaqueTriangleHit",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::RayQuery_CommitProceduralPrimitiveHit,
-        "RayQuery_CommitProceduralPrimitiveHit",
-        OCC::RayQuery_CommitProceduralPrimitiveHit,
-        "rayQuery_CommitProceduralPrimitiveHit",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::RayQuery_CommittedStatus,
-        "RayQuery_CommittedStatus",
-        OCC::RayQuery_StateScalar,
-        "rayQuery_StateScalar",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CandidateType,
-        "RayQuery_CandidateType",
-        OCC::RayQuery_StateScalar,
-        "rayQuery_StateScalar",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CandidateObjectToWorld3x4,
-        "RayQuery_CandidateObjectToWorld3x4",
-        OCC::RayQuery_StateMatrix,
-        "rayQuery_StateMatrix",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CandidateWorldToObject3x4,
-        "RayQuery_CandidateWorldToObject3x4",
-        OCC::RayQuery_StateMatrix,
-        "rayQuery_StateMatrix",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CommittedObjectToWorld3x4,
-        "RayQuery_CommittedObjectToWorld3x4",
-        OCC::RayQuery_StateMatrix,
-        "rayQuery_StateMatrix",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CommittedWorldToObject3x4,
-        "RayQuery_CommittedWorldToObject3x4",
-        OCC::RayQuery_StateMatrix,
-        "rayQuery_StateMatrix",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CandidateProceduralPrimitiveNonOpaque,
-        "RayQuery_CandidateProceduralPrimitiveNonOpaque",
-        OCC::RayQuery_StateScalar,
-        "rayQuery_StateScalar",
-        {false, false, false, false, true, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CandidateTriangleFrontFace,
-        "RayQuery_CandidateTriangleFrontFace",
-        OCC::RayQuery_StateScalar,
-        "rayQuery_StateScalar",
-        {false, false, false, false, true, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CommittedTriangleFrontFace,
-        "RayQuery_CommittedTriangleFrontFace",
-        OCC::RayQuery_StateScalar,
-        "rayQuery_StateScalar",
-        {false, false, false, false, true, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CandidateTriangleBarycentrics,
-        "RayQuery_CandidateTriangleBarycentrics",
-        OCC::RayQuery_StateVector,
-        "rayQuery_StateVector",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CommittedTriangleBarycentrics,
-        "RayQuery_CommittedTriangleBarycentrics",
-        OCC::RayQuery_StateVector,
-        "rayQuery_StateVector",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_RayFlags,
-        "RayQuery_RayFlags",
-        OCC::RayQuery_StateScalar,
-        "rayQuery_StateScalar",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_WorldRayOrigin,
-        "RayQuery_WorldRayOrigin",
-        OCC::RayQuery_StateVector,
-        "rayQuery_StateVector",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_WorldRayDirection,
-        "RayQuery_WorldRayDirection",
-        OCC::RayQuery_StateVector,
-        "rayQuery_StateVector",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_RayTMin,
-        "RayQuery_RayTMin",
-        OCC::RayQuery_StateScalar,
-        "rayQuery_StateScalar",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CandidateTriangleRayT,
-        "RayQuery_CandidateTriangleRayT",
-        OCC::RayQuery_StateScalar,
-        "rayQuery_StateScalar",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CommittedRayT,
-        "RayQuery_CommittedRayT",
-        OCC::RayQuery_StateScalar,
-        "rayQuery_StateScalar",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CandidateInstanceIndex,
-        "RayQuery_CandidateInstanceIndex",
-        OCC::RayQuery_StateScalar,
-        "rayQuery_StateScalar",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CandidateInstanceID,
-        "RayQuery_CandidateInstanceID",
-        OCC::RayQuery_StateScalar,
-        "rayQuery_StateScalar",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CandidateGeometryIndex,
-        "RayQuery_CandidateGeometryIndex",
-        OCC::RayQuery_StateScalar,
-        "rayQuery_StateScalar",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CandidatePrimitiveIndex,
-        "RayQuery_CandidatePrimitiveIndex",
-        OCC::RayQuery_StateScalar,
-        "rayQuery_StateScalar",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CandidateObjectRayOrigin,
-        "RayQuery_CandidateObjectRayOrigin",
-        OCC::RayQuery_StateVector,
-        "rayQuery_StateVector",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CandidateObjectRayDirection,
-        "RayQuery_CandidateObjectRayDirection",
-        OCC::RayQuery_StateVector,
-        "rayQuery_StateVector",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CommittedInstanceIndex,
-        "RayQuery_CommittedInstanceIndex",
-        OCC::RayQuery_StateScalar,
-        "rayQuery_StateScalar",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CommittedInstanceID,
-        "RayQuery_CommittedInstanceID",
-        OCC::RayQuery_StateScalar,
-        "rayQuery_StateScalar",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CommittedGeometryIndex,
-        "RayQuery_CommittedGeometryIndex",
-        OCC::RayQuery_StateScalar,
-        "rayQuery_StateScalar",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CommittedPrimitiveIndex,
-        "RayQuery_CommittedPrimitiveIndex",
-        OCC::RayQuery_StateScalar,
-        "rayQuery_StateScalar",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CommittedObjectRayOrigin,
-        "RayQuery_CommittedObjectRayOrigin",
-        OCC::RayQuery_StateVector,
-        "rayQuery_StateVector",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CommittedObjectRayDirection,
-        "RayQuery_CommittedObjectRayDirection",
-        OCC::RayQuery_StateVector,
-        "rayQuery_StateVector",
-        {false, false, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
+    // i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::AllocateRayQuery,
+     "AllocateRayQuery",
+     OCC::AllocateRayQuery,
+     "allocateRayQuery",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_TraceRayInline,
+     "RayQuery_TraceRayInline",
+     OCC::RayQuery_TraceRayInline,
+     "rayQuery_TraceRayInline",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_Proceed,
+     "RayQuery_Proceed",
+     OCC::RayQuery_Proceed,
+     "rayQuery_Proceed",
+     {false, false, false, false, true, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_Abort,
+     "RayQuery_Abort",
+     OCC::RayQuery_Abort,
+     "rayQuery_Abort",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CommitNonOpaqueTriangleHit,
+     "RayQuery_CommitNonOpaqueTriangleHit",
+     OCC::RayQuery_CommitNonOpaqueTriangleHit,
+     "rayQuery_CommitNonOpaqueTriangleHit",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CommitProceduralPrimitiveHit,
+     "RayQuery_CommitProceduralPrimitiveHit",
+     OCC::RayQuery_CommitProceduralPrimitiveHit,
+     "rayQuery_CommitProceduralPrimitiveHit",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CommittedStatus,
+     "RayQuery_CommittedStatus",
+     OCC::RayQuery_StateScalar,
+     "rayQuery_StateScalar",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CandidateType,
+     "RayQuery_CandidateType",
+     OCC::RayQuery_StateScalar,
+     "rayQuery_StateScalar",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CandidateObjectToWorld3x4,
+     "RayQuery_CandidateObjectToWorld3x4",
+     OCC::RayQuery_StateMatrix,
+     "rayQuery_StateMatrix",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CandidateWorldToObject3x4,
+     "RayQuery_CandidateWorldToObject3x4",
+     OCC::RayQuery_StateMatrix,
+     "rayQuery_StateMatrix",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CommittedObjectToWorld3x4,
+     "RayQuery_CommittedObjectToWorld3x4",
+     OCC::RayQuery_StateMatrix,
+     "rayQuery_StateMatrix",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CommittedWorldToObject3x4,
+     "RayQuery_CommittedWorldToObject3x4",
+     OCC::RayQuery_StateMatrix,
+     "rayQuery_StateMatrix",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CandidateProceduralPrimitiveNonOpaque,
+     "RayQuery_CandidateProceduralPrimitiveNonOpaque",
+     OCC::RayQuery_StateScalar,
+     "rayQuery_StateScalar",
+     {false, false, false, false, true, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CandidateTriangleFrontFace,
+     "RayQuery_CandidateTriangleFrontFace",
+     OCC::RayQuery_StateScalar,
+     "rayQuery_StateScalar",
+     {false, false, false, false, true, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CommittedTriangleFrontFace,
+     "RayQuery_CommittedTriangleFrontFace",
+     OCC::RayQuery_StateScalar,
+     "rayQuery_StateScalar",
+     {false, false, false, false, true, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CandidateTriangleBarycentrics,
+     "RayQuery_CandidateTriangleBarycentrics",
+     OCC::RayQuery_StateVector,
+     "rayQuery_StateVector",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CommittedTriangleBarycentrics,
+     "RayQuery_CommittedTriangleBarycentrics",
+     OCC::RayQuery_StateVector,
+     "rayQuery_StateVector",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_RayFlags,
+     "RayQuery_RayFlags",
+     OCC::RayQuery_StateScalar,
+     "rayQuery_StateScalar",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_WorldRayOrigin,
+     "RayQuery_WorldRayOrigin",
+     OCC::RayQuery_StateVector,
+     "rayQuery_StateVector",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_WorldRayDirection,
+     "RayQuery_WorldRayDirection",
+     OCC::RayQuery_StateVector,
+     "rayQuery_StateVector",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_RayTMin,
+     "RayQuery_RayTMin",
+     OCC::RayQuery_StateScalar,
+     "rayQuery_StateScalar",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CandidateTriangleRayT,
+     "RayQuery_CandidateTriangleRayT",
+     OCC::RayQuery_StateScalar,
+     "rayQuery_StateScalar",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CommittedRayT,
+     "RayQuery_CommittedRayT",
+     OCC::RayQuery_StateScalar,
+     "rayQuery_StateScalar",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CandidateInstanceIndex,
+     "RayQuery_CandidateInstanceIndex",
+     OCC::RayQuery_StateScalar,
+     "rayQuery_StateScalar",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CandidateInstanceID,
+     "RayQuery_CandidateInstanceID",
+     OCC::RayQuery_StateScalar,
+     "rayQuery_StateScalar",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CandidateGeometryIndex,
+     "RayQuery_CandidateGeometryIndex",
+     OCC::RayQuery_StateScalar,
+     "rayQuery_StateScalar",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CandidatePrimitiveIndex,
+     "RayQuery_CandidatePrimitiveIndex",
+     OCC::RayQuery_StateScalar,
+     "rayQuery_StateScalar",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CandidateObjectRayOrigin,
+     "RayQuery_CandidateObjectRayOrigin",
+     OCC::RayQuery_StateVector,
+     "rayQuery_StateVector",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CandidateObjectRayDirection,
+     "RayQuery_CandidateObjectRayDirection",
+     OCC::RayQuery_StateVector,
+     "rayQuery_StateVector",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CommittedInstanceIndex,
+     "RayQuery_CommittedInstanceIndex",
+     OCC::RayQuery_StateScalar,
+     "rayQuery_StateScalar",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CommittedInstanceID,
+     "RayQuery_CommittedInstanceID",
+     OCC::RayQuery_StateScalar,
+     "rayQuery_StateScalar",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CommittedGeometryIndex,
+     "RayQuery_CommittedGeometryIndex",
+     OCC::RayQuery_StateScalar,
+     "rayQuery_StateScalar",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CommittedPrimitiveIndex,
+     "RayQuery_CommittedPrimitiveIndex",
+     OCC::RayQuery_StateScalar,
+     "rayQuery_StateScalar",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CommittedObjectRayOrigin,
+     "RayQuery_CommittedObjectRayOrigin",
+     OCC::RayQuery_StateVector,
+     "rayQuery_StateVector",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CommittedObjectRayDirection,
+     "RayQuery_CommittedObjectRayDirection",
+     OCC::RayQuery_StateVector,
+     "rayQuery_StateVector",
+     {false, false, true, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Raytracing object space uint System Values, raytracing tier 1.1 void, h,
-    // f,     d,    i1,    i8,   i16,   i32,   i64,   udt,   obj ,  function
-    // attribute
-    {
-        OC::GeometryIndex,
-        "GeometryIndex",
-        OCC::GeometryIndex,
-        "geometryIndex",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // f,     d,    i1,    i8,   i16,   i32,   i64,   udt,   obj,   vec,
+    // function attribute, ext oload, vec oload
+    {OC::GeometryIndex,
+     "GeometryIndex",
+     OCC::GeometryIndex,
+     "geometryIndex",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Inline Ray Query void,     h,     f,     d,    i1,    i8,   i16,   i32,
-    // i64,   udt,   obj ,  function attribute
-    {
-        OC::RayQuery_CandidateInstanceContributionToHitGroupIndex,
-        "RayQuery_CandidateInstanceContributionToHitGroupIndex",
-        OCC::RayQuery_StateScalar,
-        "rayQuery_StateScalar",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::RayQuery_CommittedInstanceContributionToHitGroupIndex,
-        "RayQuery_CommittedInstanceContributionToHitGroupIndex",
-        OCC::RayQuery_StateScalar,
-        "rayQuery_StateScalar",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
+    // i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::RayQuery_CandidateInstanceContributionToHitGroupIndex,
+     "RayQuery_CandidateInstanceContributionToHitGroupIndex",
+     OCC::RayQuery_StateScalar,
+     "rayQuery_StateScalar",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RayQuery_CommittedInstanceContributionToHitGroupIndex,
+     "RayQuery_CommittedInstanceContributionToHitGroupIndex",
+     OCC::RayQuery_StateScalar,
+     "rayQuery_StateScalar",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Get handle from heap void,     h,     f,     d,    i1,    i8,   i16, i32,
-    // i64,   udt,   obj ,  function attribute
-    {
-        OC::AnnotateHandle,
-        "AnnotateHandle",
-        OCC::AnnotateHandle,
-        "annotateHandle",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::CreateHandleFromBinding,
-        "CreateHandleFromBinding",
-        OCC::CreateHandleFromBinding,
-        "createHandleFromBinding",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::CreateHandleFromHeap,
-        "CreateHandleFromHeap",
-        OCC::CreateHandleFromHeap,
-        "createHandleFromHeap",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::AnnotateHandle,
+     "AnnotateHandle",
+     OCC::AnnotateHandle,
+     "annotateHandle",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::CreateHandleFromBinding,
+     "CreateHandleFromBinding",
+     OCC::CreateHandleFromBinding,
+     "createHandleFromBinding",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::CreateHandleFromHeap,
+     "CreateHandleFromHeap",
+     OCC::CreateHandleFromHeap,
+     "createHandleFromHeap",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Unpacking intrinsics void,     h,     f,     d,    i1,    i8,   i16, i32,
-    // i64,   udt,   obj ,  function attribute
-    {
-        OC::Unpack4x8,
-        "Unpack4x8",
-        OCC::Unpack4x8,
-        "unpack4x8",
-        {false, false, false, false, false, false, true, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::Unpack4x8,
+     "Unpack4x8",
+     OCC::Unpack4x8,
+     "unpack4x8",
+     {false, false, false, false, false, false, true, true, false, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Packing intrinsics void,     h,     f,     d,    i1,    i8,   i16,   i32,
-    // i64,   udt,   obj ,  function attribute
-    {
-        OC::Pack4x8,
-        "Pack4x8",
-        OCC::Pack4x8,
-        "pack4x8",
-        {false, false, false, false, false, false, true, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::Pack4x8,
+     "Pack4x8",
+     OCC::Pack4x8,
+     "pack4x8",
+     {false, false, false, false, false, false, true, true, false, false, false,
+      false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Helper Lanes void,     h,     f,     d,    i1,    i8,   i16,   i32, i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::IsHelperLane,
-        "IsHelperLane",
-        OCC::IsHelperLane,
-        "isHelperLane",
-        {false, false, false, false, true, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::IsHelperLane,
+     "IsHelperLane",
+     OCC::IsHelperLane,
+     "isHelperLane",
+     {false, false, false, false, true, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Quad Wave Ops void,     h,     f,     d,    i1,    i8,   i16,   i32, i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::QuadVote,
-        "QuadVote",
-        OCC::QuadVote,
-        "quadVote",
-        {false, false, false, false, true, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::QuadVote,
+     "QuadVote",
+     OCC::QuadVote,
+     "quadVote",
+     {false, false, false, false, true, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Resources - gather void,     h,     f,     d,    i1,    i8,   i16,   i32,
-    // i64,   udt,   obj ,  function attribute
-    {
-        OC::TextureGatherRaw,
-        "TextureGatherRaw",
-        OCC::TextureGatherRaw,
-        "textureGatherRaw",
-        {false, false, false, false, false, false, true, true, true, false,
-         false},
-        Attribute::ReadOnly,
-    },
+    // i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::TextureGatherRaw,
+     "TextureGatherRaw",
+     OCC::TextureGatherRaw,
+     "textureGatherRaw",
+     {false, false, false, false, false, false, true, true, true, false, false,
+      false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Resources - sample void,     h,     f,     d,    i1,    i8,   i16,   i32,
-    // i64,   udt,   obj ,  function attribute
-    {
-        OC::SampleCmpLevel,
-        "SampleCmpLevel",
-        OCC::SampleCmpLevel,
-        "sampleCmpLevel",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
+    // i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::SampleCmpLevel,
+     "SampleCmpLevel",
+     OCC::SampleCmpLevel,
+     "sampleCmpLevel",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Resources void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,
-    // udt,   obj ,  function attribute
-    {
-        OC::TextureStoreSample,
-        "TextureStoreSample",
-        OCC::TextureStoreSample,
-        "textureStoreSample",
-        {false, true, true, false, false, false, true, true, false, false,
-         false},
-        Attribute::None,
-    },
-
-    //                                                                                                                         void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::Reserved0,
-        "Reserved0",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::Reserved1,
-        "Reserved1",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::Reserved2,
-        "Reserved2",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::Reserved3,
-        "Reserved3",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::Reserved4,
-        "Reserved4",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::Reserved5,
-        "Reserved5",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::Reserved6,
-        "Reserved6",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::Reserved7,
-        "Reserved7",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::Reserved8,
-        "Reserved8",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::Reserved9,
-        "Reserved9",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::Reserved10,
-        "Reserved10",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::Reserved11,
-        "Reserved11",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
+    // udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::TextureStoreSample,
+     "TextureStoreSample",
+     OCC::TextureStoreSample,
+     "textureStoreSample",
+     {false, true, true, false, false, false, true, true, false, false, false,
+      false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+
+    //                                                                                                                         void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::Reserved0,
+     "Reserved0",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::Reserved1,
+     "Reserved1",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::Reserved2,
+     "Reserved2",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::Reserved3,
+     "Reserved3",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::Reserved4,
+     "Reserved4",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::Reserved5,
+     "Reserved5",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::Reserved6,
+     "Reserved6",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::Reserved7,
+     "Reserved7",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::Reserved8,
+     "Reserved8",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::Reserved9,
+     "Reserved9",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::Reserved10,
+     "Reserved10",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::Reserved11,
+     "Reserved11",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Create/Annotate Node Handles void,     h,     f,     d,    i1,    i8,
-    // i16,   i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::AllocateNodeOutputRecords,
-        "AllocateNodeOutputRecords",
-        OCC::AllocateNodeOutputRecords,
-        "allocateNodeOutputRecords",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
+    // i16,   i32,   i64,   udt,   obj,   vec,  function attribute, ext oload,
+    // vec oload
+    {OC::AllocateNodeOutputRecords,
+     "AllocateNodeOutputRecords",
+     OCC::AllocateNodeOutputRecords,
+     "allocateNodeOutputRecords",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Get Pointer to Node Record in Address Space 6 void,     h,     f,     d,
-    // i1,    i8,   i16,   i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::GetNodeRecordPtr,
-        "GetNodeRecordPtr",
-        OCC::GetNodeRecordPtr,
-        "getNodeRecordPtr",
-        {false, false, false, false, false, false, false, false, false, true,
-         false},
-        Attribute::ReadNone,
-    },
+    // i1,    i8,   i16,   i32,   i64,   udt,   obj,   vec,  function attribute,
+    // ext oload, vec oload
+    {OC::GetNodeRecordPtr,
+     "GetNodeRecordPtr",
+     OCC::GetNodeRecordPtr,
+     "getNodeRecordPtr",
+     {false, false, false, false, false, false, false, false, false, true,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Work Graph intrinsics void,     h,     f,     d,    i1,    i8,   i16,
-    // i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::IncrementOutputCount,
-        "IncrementOutputCount",
-        OCC::IncrementOutputCount,
-        "incrementOutputCount",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::OutputComplete,
-        "OutputComplete",
-        OCC::OutputComplete,
-        "outputComplete",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::GetInputRecordCount,
-        "GetInputRecordCount",
-        OCC::GetInputRecordCount,
-        "getInputRecordCount",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::FinishedCrossGroupSharing,
-        "FinishedCrossGroupSharing",
-        OCC::FinishedCrossGroupSharing,
-        "finishedCrossGroupSharing",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
+    // i32,   i64,   udt,   obj,   vec,  function attribute, ext oload, vec
+    // oload
+    {OC::IncrementOutputCount,
+     "IncrementOutputCount",
+     OCC::IncrementOutputCount,
+     "incrementOutputCount",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::OutputComplete,
+     "OutputComplete",
+     OCC::OutputComplete,
+     "outputComplete",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::GetInputRecordCount,
+     "GetInputRecordCount",
+     OCC::GetInputRecordCount,
+     "getInputRecordCount",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::FinishedCrossGroupSharing,
+     "FinishedCrossGroupSharing",
+     OCC::FinishedCrossGroupSharing,
+     "finishedCrossGroupSharing",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Synchronization void,     h,     f,     d,    i1,    i8,   i16,   i32,
-    // i64,   udt,   obj ,  function attribute
-    {
-        OC::BarrierByMemoryType,
-        "BarrierByMemoryType",
-        OCC::BarrierByMemoryType,
-        "barrierByMemoryType",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::NoDuplicate,
-    },
-    {
-        OC::BarrierByMemoryHandle,
-        "BarrierByMemoryHandle",
-        OCC::BarrierByMemoryHandle,
-        "barrierByMemoryHandle",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::NoDuplicate,
-    },
-    {
-        OC::BarrierByNodeRecordHandle,
-        "BarrierByNodeRecordHandle",
-        OCC::BarrierByNodeRecordHandle,
-        "barrierByNodeRecordHandle",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::NoDuplicate,
-    },
+    // i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::BarrierByMemoryType,
+     "BarrierByMemoryType",
+     OCC::BarrierByMemoryType,
+     "barrierByMemoryType",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::NoDuplicate,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::BarrierByMemoryHandle,
+     "BarrierByMemoryHandle",
+     OCC::BarrierByMemoryHandle,
+     "barrierByMemoryHandle",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::NoDuplicate,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::BarrierByNodeRecordHandle,
+     "BarrierByNodeRecordHandle",
+     OCC::BarrierByNodeRecordHandle,
+     "barrierByNodeRecordHandle",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::NoDuplicate,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Create/Annotate Node Handles void,     h,     f,     d,    i1,    i8,
-    // i16,   i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::CreateNodeOutputHandle,
-        "CreateNodeOutputHandle",
-        OCC::createNodeOutputHandle,
-        "createNodeOutputHandle",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::IndexNodeHandle,
-        "IndexNodeHandle",
-        OCC::IndexNodeHandle,
-        "indexNodeHandle",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::AnnotateNodeHandle,
-        "AnnotateNodeHandle",
-        OCC::AnnotateNodeHandle,
-        "annotateNodeHandle",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::CreateNodeInputRecordHandle,
-        "CreateNodeInputRecordHandle",
-        OCC::CreateNodeInputRecordHandle,
-        "createNodeInputRecordHandle",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::AnnotateNodeRecordHandle,
-        "AnnotateNodeRecordHandle",
-        OCC::AnnotateNodeRecordHandle,
-        "annotateNodeRecordHandle",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i16,   i32,   i64,   udt,   obj,   vec,  function attribute, ext oload,
+    // vec oload
+    {OC::CreateNodeOutputHandle,
+     "CreateNodeOutputHandle",
+     OCC::createNodeOutputHandle,
+     "createNodeOutputHandle",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::IndexNodeHandle,
+     "IndexNodeHandle",
+     OCC::IndexNodeHandle,
+     "indexNodeHandle",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::AnnotateNodeHandle,
+     "AnnotateNodeHandle",
+     OCC::AnnotateNodeHandle,
+     "annotateNodeHandle",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::CreateNodeInputRecordHandle,
+     "CreateNodeInputRecordHandle",
+     OCC::CreateNodeInputRecordHandle,
+     "createNodeInputRecordHandle",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::AnnotateNodeRecordHandle,
+     "AnnotateNodeRecordHandle",
+     OCC::AnnotateNodeRecordHandle,
+     "annotateNodeRecordHandle",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Work Graph intrinsics void,     h,     f,     d,    i1,    i8,   i16,
-    // i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::NodeOutputIsValid,
-        "NodeOutputIsValid",
-        OCC::NodeOutputIsValid,
-        "nodeOutputIsValid",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::GetRemainingRecursionLevels,
-        "GetRemainingRecursionLevels",
-        OCC::GetRemainingRecursionLevels,
-        "getRemainingRecursionLevels",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
+    // i32,   i64,   udt,   obj,   vec,  function attribute, ext oload, vec
+    // oload
+    {OC::NodeOutputIsValid,
+     "NodeOutputIsValid",
+     OCC::NodeOutputIsValid,
+     "nodeOutputIsValid",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::GetRemainingRecursionLevels,
+     "GetRemainingRecursionLevels",
+     OCC::GetRemainingRecursionLevels,
+     "getRemainingRecursionLevels",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Comparison Samples void,     h,     f,     d,    i1,    i8,   i16,   i32,
-    // i64,   udt,   obj ,  function attribute
-    {
-        OC::SampleCmpGrad,
-        "SampleCmpGrad",
-        OCC::SampleCmpGrad,
-        "sampleCmpGrad",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
-    {
-        OC::SampleCmpBias,
-        "SampleCmpBias",
-        OCC::SampleCmpBias,
-        "sampleCmpBias",
-        {false, true, true, false, false, false, false, false, false, false,
-         false},
-        Attribute::ReadOnly,
-    },
+    // i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::SampleCmpGrad,
+     "SampleCmpGrad",
+     OCC::SampleCmpGrad,
+     "sampleCmpGrad",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::SampleCmpBias,
+     "SampleCmpBias",
+     OCC::SampleCmpBias,
+     "sampleCmpBias",
+     {false, true, true, false, false, false, false, false, false, false, false,
+      false, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Extended Command Information void,     h,     f,     d,    i1,    i8,
-    // i16,   i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::StartVertexLocation,
-        "StartVertexLocation",
-        OCC::StartVertexLocation,
-        "startVertexLocation",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
-    {
-        OC::StartInstanceLocation,
-        "StartInstanceLocation",
-        OCC::StartInstanceLocation,
-        "startInstanceLocation",
-        {false, false, false, false, false, false, false, true, false, false,
-         false},
-        Attribute::ReadNone,
-    },
+    // i16,   i32,   i64,   udt,   obj,   vec,  function attribute, ext oload,
+    // vec oload
+    {OC::StartVertexLocation,
+     "StartVertexLocation",
+     OCC::StartVertexLocation,
+     "startVertexLocation",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::StartInstanceLocation,
+     "StartInstanceLocation",
+     OCC::StartInstanceLocation,
+     "startInstanceLocation",
+     {false, false, false, false, false, false, false, true, false, false,
+      false, false, false},
+     Attribute::ReadNone,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
 
     // Inline Ray Query void,     h,     f,     d,    i1,    i8,   i16,   i32,
-    // i64,   udt,   obj ,  function attribute
-    {
-        OC::AllocateRayQuery2,
-        "AllocateRayQuery2",
-        OCC::AllocateRayQuery2,
-        "allocateRayQuery2",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-
-    //                                                                                                                         void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,   udt,   obj ,  function attribute
-    {
-        OC::ReservedA0,
-        "ReservedA0",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedA1,
-        "ReservedA1",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedA2,
-        "ReservedA2",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB0,
-        "ReservedB0",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB1,
-        "ReservedB1",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB2,
-        "ReservedB2",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB3,
-        "ReservedB3",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB4,
-        "ReservedB4",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB5,
-        "ReservedB5",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB6,
-        "ReservedB6",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB7,
-        "ReservedB7",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB8,
-        "ReservedB8",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB9,
-        "ReservedB9",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB10,
-        "ReservedB10",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB11,
-        "ReservedB11",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB12,
-        "ReservedB12",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB13,
-        "ReservedB13",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB14,
-        "ReservedB14",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB15,
-        "ReservedB15",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB16,
-        "ReservedB16",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB17,
-        "ReservedB17",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB18,
-        "ReservedB18",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB19,
-        "ReservedB19",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB20,
-        "ReservedB20",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB21,
-        "ReservedB21",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB22,
-        "ReservedB22",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB23,
-        "ReservedB23",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB24,
-        "ReservedB24",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB25,
-        "ReservedB25",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB26,
-        "ReservedB26",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB27,
-        "ReservedB27",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB28,
-        "ReservedB28",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB29,
-        "ReservedB29",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedB30,
-        "ReservedB30",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedC0,
-        "ReservedC0",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedC1,
-        "ReservedC1",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedC2,
-        "ReservedC2",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedC3,
-        "ReservedC3",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedC4,
-        "ReservedC4",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedC5,
-        "ReservedC5",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedC6,
-        "ReservedC6",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedC7,
-        "ReservedC7",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedC8,
-        "ReservedC8",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
-    {
-        OC::ReservedC9,
-        "ReservedC9",
-        OCC::Reserved,
-        "reserved",
-        {true, false, false, false, false, false, false, false, false, false,
-         false},
-        Attribute::None,
-    },
+    // i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::AllocateRayQuery2,
+     "AllocateRayQuery2",
+     OCC::AllocateRayQuery2,
+     "allocateRayQuery2",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+
+    //                                                                                                                         void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload
+    {OC::ReservedA0,
+     "ReservedA0",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedA1,
+     "ReservedA1",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedA2,
+     "ReservedA2",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB0,
+     "ReservedB0",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB1,
+     "ReservedB1",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB2,
+     "ReservedB2",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB3,
+     "ReservedB3",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB4,
+     "ReservedB4",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB5,
+     "ReservedB5",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB6,
+     "ReservedB6",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB7,
+     "ReservedB7",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB8,
+     "ReservedB8",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB9,
+     "ReservedB9",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB10,
+     "ReservedB10",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB11,
+     "ReservedB11",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB12,
+     "ReservedB12",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB13,
+     "ReservedB13",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB14,
+     "ReservedB14",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB15,
+     "ReservedB15",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB16,
+     "ReservedB16",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB17,
+     "ReservedB17",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB18,
+     "ReservedB18",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB19,
+     "ReservedB19",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB20,
+     "ReservedB20",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB21,
+     "ReservedB21",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB22,
+     "ReservedB22",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB23,
+     "ReservedB23",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB24,
+     "ReservedB24",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB25,
+     "ReservedB25",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB26,
+     "ReservedB26",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB27,
+     "ReservedB27",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB28,
+     "ReservedB28",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB29,
+     "ReservedB29",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedB30,
+     "ReservedB30",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedC0,
+     "ReservedC0",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedC1,
+     "ReservedC1",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedC2,
+     "ReservedC2",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedC3,
+     "ReservedC3",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedC4,
+     "ReservedC4",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedC5,
+     "ReservedC5",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedC6,
+     "ReservedC6",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedC7,
+     "ReservedC7",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedC8,
+     "ReservedC8",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::ReservedC9,
+     "ReservedC9",
+     OCC::Reserved,
+     "reserved",
+     {true, false, false, false, false, false, false, false, false, false,
+      false, false, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x0}, {0x0}}},
+    {OC::RawBufferVectorLoad,
+     "RawBufferVectorLoad",
+     OCC::RawBufferVectorLoad,
+     "rawBufferVectorLoad",
+     {false, true, true, true, false, false, true, true, true, false, false,
+      true, false},
+     Attribute::ReadOnly,
+     {{0x0}, {0x0}},
+     {{0x1ce}, {0x0}}},
+    {OC::RawBufferVectorStore,
+     "RawBufferVectorStore",
+     OCC::RawBufferVectorStore,
+     "rawBufferVectorStore",
+     {false, true, true, true, false, false, true, true, true, false, false,
+      true, false},
+     Attribute::None,
+     {{0x0}, {0x0}},
+     {{0x1ce}, {0x0}}},
 };
 // OPCODE-OLOADS:END
 
 const char *OP::m_OverloadTypeName[kNumTypeOverloads] = {
-    "void", "f16", "f32", "f64", "i1",  "i8",
-    "i16",  "i32", "i64", "udt", "obj", // These should not be used
+    "void", "f16", "f32", "f64", "i1",  "i8",  "i16",
+    "i32",  "i64", "udt", "obj", "vec", "ext",
+    // "udt", "obj", "vec", and "ext" should not be used
 };
 
 const char *OP::m_NamePrefix = "dx.op.";
@@ -3067,7 +3126,14 @@ unsigned OP::GetTypeSlot(Type *pType) {
     return GetTypeSlot(pType);
   }
   case Type::StructTyID:
-    return kObjectTypeSlot;
+    // Named struct value (not pointer) indicates a built-in object type.
+    // Anonymous struct value is used to wrap multi-overload dimensions.
+    if (cast<StructType>(pType)->hasName())
+      return kObjectTypeSlot;
+    else
+      return kExtendedTypeSlot;
+  case Type::VectorTyID:
+    return kVectorTypeSlot;
   default:
     break;
   }
@@ -3091,6 +3157,26 @@ llvm::StringRef OP::GetTypeName(Type *Ty, std::string &str) {
   } else if (TypeSlot == kObjectTypeSlot) {
     StructType *ST = cast<StructType>(Ty);
     return ST->getStructName();
+  } else if (TypeSlot == kVectorTypeSlot) {
+    VectorType *VecTy = cast<VectorType>(Ty);
+    str = "v";
+    str += std::to_string(VecTy->getNumElements());
+    str += GetOverloadTypeName(OP::GetTypeSlot(VecTy->getElementType()));
+    return str;
+  } else if (TypeSlot == kExtendedTypeSlot) {
+    DXASSERT(isa<StructType>(Ty),
+             "otherwise, extended overload type not wrapped in struct type.");
+    StructType *ST = cast<StructType>(Ty);
+    DXASSERT(ST->getNumElements() <= DXIL::kDxilMaxOloadDims,
+             "otherwise, extended overload has too many dimensions.");
+    // Iterate extended slots, recurse, separate with '.'
+    for (unsigned I = 0; I < ST->getNumElements(); ++I) {
+      if (I > 0)
+        str += ".";
+      std::string TempStr;
+      str += GetTypeName(ST->getElementType(I), TempStr);
+    }
+    return str;
   } else {
     raw_string_ostream os(str);
     Ty->print(os);
@@ -3138,13 +3224,43 @@ llvm::Attribute::AttrKind OP::GetMemAccessAttr(OpCode opCode) {
 }
 
 bool OP::IsOverloadLegal(OpCode opCode, Type *pType) {
+  auto &OpProps = m_OpCodeProps[static_cast<unsigned>(opCode)];
   if (!pType)
     return false;
   if (opCode == OpCode::NumOpCodes)
     return false;
   unsigned TypeSlot = GetTypeSlot(pType);
-  return TypeSlot != UINT_MAX &&
-         m_OpCodeProps[(unsigned)opCode].bAllowOverload[TypeSlot];
+  if (TypeSlot >= kNumTypeOverloads)
+    return false;
+  if (!OpProps.bAllowOverload[TypeSlot])
+    return false;
+
+  if (TypeSlot == kVectorTypeSlot) {
+    unsigned EltTypeSlot =
+        GetTypeSlot(cast<VectorType>(pType)->getElementType());
+    return OpProps.AllowedVectorElements[0][EltTypeSlot];
+  }
+
+  if (TypeSlot == kExtendedTypeSlot) {
+    StructType *ST = cast<StructType>(pType);
+    if (ST->getNumElements() < 2 ||
+        ST->getNumElements() > DXIL::kDxilMaxOloadDims)
+      return false;
+    for (unsigned I = 0; I < ST->getNumElements(); ++I) {
+      Type *ElTy = ST->getElementType(I);
+      unsigned OloadSlot = GetTypeSlot(ElTy);
+      if (!OpProps.ExtendedOverloads[I][OloadSlot])
+        return false;
+      if (OloadSlot == kVectorTypeSlot) {
+        unsigned EltTypeSlot =
+            GetTypeSlot(cast<VectorType>(ElTy)->getElementType());
+        if (!OpProps.AllowedVectorElements[I][EltTypeSlot])
+          return false;
+      }
+    }
+    return true;
+  }
+  return true;
 }
 
 bool OP::CheckOpCodeTable() {
@@ -3292,6 +3408,13 @@ bool OP::IsDxilOpBarrier(OpCode C) {
   // OPCODE-BARRIER:END
 }
 
+bool OP::IsDxilOpExtendedOverload(OpCode C) {
+  if (C >= OpCode::NumOpCodes)
+    return false;
+  return m_OpCodeProps[static_cast<unsigned>(C)]
+      .bAllowOverload[kExtendedTypeSlot];
+}
+
 static unsigned MaskMemoryTypeFlagsIfAllowed(unsigned memoryTypeFlags,
                                              unsigned allowedMask) {
   // If the memory type is AllMemory, masking inapplicable flags is allowed.
@@ -3930,6 +4053,8 @@ void OP::FixOverloadNames() {
     if (F.isDeclaration() && OP::IsDxilOpFunc(&F) && !F.user_empty()) {
       CallInst *CI = cast<CallInst>(*F.user_begin());
       DXIL::OpCode opCode = OP::GetDxilOpFuncCallInst(CI);
+      if (IsDxilOpExtendedOverload(opCode))
+        continue;
       llvm::Type *Ty = OP::GetOverloadType(opCode, &F);
       if (!OP::IsOverloadLegal(opCode, Ty))
         continue;
@@ -3949,11 +4074,31 @@ void OP::UpdateCache(OpCodeClass opClass, Type *Ty, llvm::Function *F) {
   m_FunctionToOpClass[F] = opClass;
 }
 
+Function *OP::GetOpFunc(OpCode opCode, ArrayRef<Type *> ExtendedOverloads) {
+  DXASSERT(IsDxilOpExtendedOverload(opCode),
+           "otherwise, Dxil Op does not support extended overload");
+  return GetOpFunc(opCode, GetExtendedOverloadType(ExtendedOverloads));
+}
+
 Function *OP::GetOpFunc(OpCode opCode, Type *pOverloadType) {
   if (opCode == OpCode::NumOpCodes)
     return nullptr;
   if (!pOverloadType)
     return nullptr;
+  if (IsDxilOpExtendedOverload(opCode)) {
+    StructType *ST = dyn_cast<StructType>(pOverloadType);
+    DXASSERT(ST != nullptr,
+             "otherwise, extended overload type is not a struct");
+    if (ST == nullptr)
+      return nullptr;
+    bool EltCountValid = ST->getNumElements() > 1 &&
+                         ST->getNumElements() <= DXIL::kDxilMaxOloadDims;
+    DXASSERT(EltCountValid,
+             "otherwise, invalid type count for extended overload.");
+    if (!EltCountValid)
+      return nullptr;
+  }
+
   // Illegal overloads are generated and eliminated by DXIL op constant
   // evaluation for a number of cases where a double overload of an HL intrinsic
   // that otherwise does not support double is used for literal values, when
@@ -4006,6 +4151,9 @@ Function *OP::GetOpFunc(OpCode opCode, Type *pOverloadType) {
 #define CBRT(_y) A(GetCBufferRetType(_y))
 #define VEC4(_y) A(GetVectorType(4, _y))
 
+// Extended Overload types are wrapped in an anonymous struct
+#define EXT(_y) A(cast<StructType>(pOverloadType)->getElementType(_y))
+
   /* <py::lines('OPCODE-OLOAD-FUNCS')>hctdb_instrhelp.get_oloads_funcs()</py>*/
   switch (opCode) { // return     opCode
                     // OPCODE-OLOAD-FUNCS:BEGIN
@@ -6023,6 +6171,23 @@ Function *OP::GetOpFunc(OpCode opCode, Type *pOverloadType) {
     A(pV);
     A(pI32);
     break;
+  case OpCode::RawBufferVectorLoad:
+    RRT(pETy);
+    A(pI32);
+    A(pRes);
+    A(pI32);
+    A(pI32);
+    A(pI32);
+    break;
+  case OpCode::RawBufferVectorStore:
+    A(pV);
+    A(pI32);
+    A(pRes);
+    A(pI32);
+    A(pI32);
+    A(pETy);
+    A(pI32);
+    break;
   // OPCODE-OLOAD-FUNCS:END
   default:
     DXASSERT(false, "otherwise unhandled case");
@@ -6171,6 +6336,7 @@ llvm::Type *OP::GetOverloadType(OpCode opCode, llvm::Function *F) {
   case OpCode::StoreVertexOutput:
   case OpCode::StorePrimitiveOutput:
   case OpCode::DispatchMesh:
+  case OpCode::RawBufferVectorStore:
     if (FT->getNumParams() <= 4)
       return nullptr;
     return FT->getParamType(4);
@@ -6417,7 +6583,8 @@ llvm::Type *OP::GetOverloadType(OpCode opCode, llvm::Function *F) {
   case OpCode::TextureGatherRaw:
   case OpCode::SampleCmpLevel:
   case OpCode::SampleCmpGrad:
-  case OpCode::SampleCmpBias: {
+  case OpCode::SampleCmpBias:
+  case OpCode::RawBufferVectorLoad: {
     StructType *ST = cast<StructType>(Ty);
     return ST->getElementType(0);
   }
@@ -6472,7 +6639,15 @@ bool OP::IsResRetType(llvm::Type *Ty) {
 Type *OP::GetResRetType(Type *pOverloadType) {
   unsigned TypeSlot = GetTypeSlot(pOverloadType);
 
-  if (m_pResRetType[TypeSlot] == nullptr) {
+  if (TypeSlot == kVectorTypeSlot) {
+    string TypeName("dx.types.ResRet.");
+    VectorType *VecTy = cast<VectorType>(pOverloadType);
+    TypeName += "v";
+    TypeName += std::to_string(VecTy->getNumElements());
+    TypeName += GetOverloadTypeName(OP::GetTypeSlot(VecTy->getElementType()));
+    Type *FieldTypes[2] = {pOverloadType, Type::getInt32Ty(m_Ctx)};
+    return GetOrCreateStructType(m_Ctx, FieldTypes, TypeName, m_pModule);
+  } else if (m_pResRetType[TypeSlot] == nullptr) {
     string TypeName("dx.types.ResRet.");
     TypeName += GetOverloadTypeName(TypeSlot);
     Type *FieldTypes[5] = {pOverloadType, pOverloadType, pOverloadType,
@@ -6529,6 +6704,10 @@ Type *OP::GetVectorType(unsigned numElements, Type *pOverloadType) {
   return nullptr;
 }
 
+StructType *OP::GetExtendedOverloadType(ArrayRef<Type *> OverloadTypes) {
+  return StructType::get(m_Ctx, OverloadTypes);
+}
+
 //------------------------------------------------------------------------------
 //
 //  LLVM utility methods.
diff --git a/lib/DXIL/DxilUtil.cpp b/lib/DXIL/DxilUtil.cpp
index 757a0bc3ee..f6ffd7f7e2 100644
--- a/lib/DXIL/DxilUtil.cpp
+++ b/lib/DXIL/DxilUtil.cpp
@@ -426,35 +426,37 @@ GetHLSLResourceProperties(llvm::Type *Ty) {
                                             false, false, false));
 
     if (name == "SamplerComparisonState")
-      return RetType(
-          true, MakeResourceProperties(hlsl::DXIL::ResourceKind::Sampler, false,
-                                       false, /*cmp or counter*/ true));
+      return RetType(true, MakeResourceProperties(
+                               hlsl::DXIL::ResourceKind::Sampler, /*UAV*/ false,
+                               /*ROV*/ false, /*cmp or counter*/ true));
 
     if (name.startswith("AppendStructuredBuffer<"))
-      return RetType(true, MakeResourceProperties(
-                               hlsl::DXIL::ResourceKind::StructuredBuffer,
-                               false, false, /*cmp or counter*/ true));
+      return RetType(true,
+                     MakeResourceProperties(
+                         hlsl::DXIL::ResourceKind::StructuredBuffer,
+                         /*UAV*/ true, /*ROV*/ false, /*cmp or counter*/ true));
 
     if (name.startswith("ConsumeStructuredBuffer<"))
       return RetType(true, MakeResourceProperties(
                                hlsl::DXIL::ResourceKind::StructuredBuffer,
-                               false, false, /*cmp or counter*/ true));
+                               /*UAV*/ false, /*ROV*/ false,
+                               /*cmp or counter*/ true));
 
     if (name == "RaytracingAccelerationStructure")
       return RetType(true,
                      MakeResourceProperties(
                          hlsl::DXIL::ResourceKind::RTAccelerationStructure,
-                         false, false, false));
+                         /*UAV*/ false, /*ROV*/ false, false));
 
     if (name.startswith("ConstantBuffer<"))
-      return RetType(true,
-                     MakeResourceProperties(hlsl::DXIL::ResourceKind::CBuffer,
-                                            false, false, false));
+      return RetType(
+          true, MakeResourceProperties(hlsl::DXIL::ResourceKind::CBuffer,
+                                       /*UAV*/ false, /*ROV*/ false, false));
 
     if (name.startswith("TextureBuffer<"))
-      return RetType(true,
-                     MakeResourceProperties(hlsl::DXIL::ResourceKind::TBuffer,
-                                            false, false, false));
+      return RetType(
+          true, MakeResourceProperties(hlsl::DXIL::ResourceKind::TBuffer,
+                                       /*UAV*/ false, /*ROV*/ false, false));
 
     if (ConsumePrefix(name, "FeedbackTexture2D")) {
       hlsl::DXIL::ResourceKind kind = hlsl::DXIL::ResourceKind::Invalid;
@@ -464,7 +466,9 @@ GetHLSLResourceProperties(llvm::Type *Ty) {
         kind = hlsl::DXIL::ResourceKind::FeedbackTexture2D;
 
       if (name.startswith("<"))
-        return RetType(true, MakeResourceProperties(kind, false, false, false));
+        return RetType(true,
+                       MakeResourceProperties(kind, /*UAV*/ false,
+                                              /*ROV*/ false, /*Cmp*/ false));
 
       return FalseRet;
     }
@@ -475,63 +479,63 @@ GetHLSLResourceProperties(llvm::Type *Ty) {
     if (name == "ByteAddressBuffer")
       return RetType(true,
                      MakeResourceProperties(hlsl::DXIL::ResourceKind::RawBuffer,
-                                            UAV, ROV, false));
+                                            UAV, ROV, /*Cmp*/ false));
 
     if (name.startswith("Buffer<"))
       return RetType(
           true, MakeResourceProperties(hlsl::DXIL::ResourceKind::TypedBuffer,
-                                       UAV, ROV, false));
+                                       UAV, ROV, /*Cmp*/ false));
 
     if (name.startswith("StructuredBuffer<"))
       return RetType(true, MakeResourceProperties(
                                hlsl::DXIL::ResourceKind::StructuredBuffer, UAV,
-                               ROV, false));
+                               ROV, /*Cmp*/ false));
 
     if (ConsumePrefix(name, "Texture")) {
       if (name.startswith("1D<"))
         return RetType(
             true, MakeResourceProperties(hlsl::DXIL::ResourceKind::Texture1D,
-                                         UAV, ROV, false));
+                                         UAV, ROV, /*Cmp*/ false));
 
       if (name.startswith("1DArray<"))
         return RetType(true, MakeResourceProperties(
                                  hlsl::DXIL::ResourceKind::Texture1DArray, UAV,
-                                 ROV, false));
+                                 ROV, /*Cmp*/ false));
 
       if (name.startswith("2D<"))
         return RetType(
             true, MakeResourceProperties(hlsl::DXIL::ResourceKind::Texture2D,
-                                         UAV, ROV, false));
+                                         UAV, ROV, /*Cmp*/ false));
 
       if (name.startswith("2DArray<"))
         return RetType(true, MakeResourceProperties(
                                  hlsl::DXIL::ResourceKind::Texture2DArray, UAV,
-                                 ROV, false));
+                                 ROV, /*Cmp*/ false));
 
       if (name.startswith("3D<"))
         return RetType(
             true, MakeResourceProperties(hlsl::DXIL::ResourceKind::Texture3D,
-                                         UAV, ROV, false));
+                                         UAV, ROV, /*Cmp*/ false));
 
       if (name.startswith("Cube<"))
         return RetType(
             true, MakeResourceProperties(hlsl::DXIL::ResourceKind::TextureCube,
-                                         UAV, ROV, false));
+                                         UAV, ROV, /*Cmp*/ false));
 
       if (name.startswith("CubeArray<"))
         return RetType(true, MakeResourceProperties(
                                  hlsl::DXIL::ResourceKind::TextureCubeArray,
-                                 UAV, ROV, false));
+                                 UAV, ROV, /*Cmp*/ false));
 
       if (name.startswith("2DMS<"))
         return RetType(
             true, MakeResourceProperties(hlsl::DXIL::ResourceKind::Texture2DMS,
-                                         UAV, ROV, false));
+                                         UAV, ROV, /*Cmp*/ false));
 
       if (name.startswith("2DMSArray<"))
         return RetType(true, MakeResourceProperties(
                                  hlsl::DXIL::ResourceKind::Texture2DMSArray,
-                                 UAV, ROV, false));
+                                 UAV, ROV, /*Cmp*/ false));
       return FalseRet;
     }
   }
diff --git a/lib/DxilValidation/DxilValidation.cpp b/lib/DxilValidation/DxilValidation.cpp
index 0a2001a745..9c93b70cce 100644
--- a/lib/DxilValidation/DxilValidation.cpp
+++ b/lib/DxilValidation/DxilValidation.cpp
@@ -2037,7 +2037,7 @@ static void ValidateExternalFunction(Function *F, ValidationContext &ValCtx) {
         ValCtx.EmitInstrError(CI, ValidationRule::InstrOload);
         continue;
       }
-      dxilFunc = hlslOP->GetOpFunc(dxilOpcode, Ty->getScalarType());
+      dxilFunc = hlslOP->GetOpFunc(dxilOpcode, Ty);
     }
 
     if (!dxilFunc) {
@@ -2109,17 +2109,20 @@ static bool IsDxilBuiltinStructType(StructType *ST, hlsl::OP *hlslOP) {
     return true;
 
   unsigned EltNum = ST->getNumElements();
+  Type *EltTy = ST->getElementType(0);
   switch (EltNum) {
   case 2:
+    // Check if it's a native vector resret.
+    if (EltTy->isVectorTy())
+      return ST == hlslOP->GetResRetType(EltTy);
+    LLVM_FALLTHROUGH;
   case 4:
-  case 8: { // 2 for doubles, 8 for halfs.
-    Type *EltTy = ST->getElementType(0);
+  case 8: // 2 for doubles, 8 for halfs.
     return ST == hlslOP->GetCBufferRetType(EltTy);
-  } break;
-  case 5: {
-    Type *EltTy = ST->getElementType(0);
+  break;
+  case 5:
     return ST == hlslOP->GetResRetType(EltTy);
-  } break;
+  break;
   default:
     return false;
   }
@@ -2193,6 +2196,8 @@ static bool ValidateType(Type *Ty, ValidationContext &ValCtx,
     return true;
 
   if (Ty->isVectorTy()) {
+    if (ValCtx.DxilMod.GetShaderModel()->IsSM69Plus())
+      return true;
     ValCtx.EmitTypeError(Ty, ValidationRule::TypesNoVector);
     return false;
   }
diff --git a/lib/HLSL/DxilLinker.cpp b/lib/HLSL/DxilLinker.cpp
index 68c83fc037..ca343662ab 100644
--- a/lib/HLSL/DxilLinker.cpp
+++ b/lib/HLSL/DxilLinker.cpp
@@ -1255,6 +1255,12 @@ void DxilLinkJob::RunPreparePass(Module &M) {
   // For static global handle.
   PM.add(createLowerStaticGlobalIntoAlloca());
 
+  // Change dynamic indexing vector to array where vectors aren't
+  // supported, but might be there from the initial compile.
+  if (!pSM->IsSM69Plus())
+    PM.add(
+        createDynamicIndexingVectorToArrayPass(false /* ReplaceAllVector */));
+
   // Remove MultiDimArray from function call arg.
   PM.add(createMultiDimArrayToOneDimArrayPass());
 
diff --git a/lib/HLSL/HLMatrixBitcastLowerPass.cpp b/lib/HLSL/HLMatrixBitcastLowerPass.cpp
index 93ba3b9816..b708293fca 100644
--- a/lib/HLSL/HLMatrixBitcastLowerPass.cpp
+++ b/lib/HLSL/HLMatrixBitcastLowerPass.cpp
@@ -113,13 +113,13 @@ class MatrixBitcastLowerPass : public FunctionPass {
 
     // Lower matrix first.
     for (BitCastInst *BCI : matCastSet) {
-      lowerMatrix(BCI, BCI->getOperand(0));
+      lowerMatrix(DM, BCI, BCI->getOperand(0));
     }
     return bUpdated;
   }
 
 private:
-  void lowerMatrix(Instruction *M, Value *A);
+  void lowerMatrix(DxilModule &DM, Instruction *M, Value *A);
   bool hasCallUser(Instruction *M);
 };
 
@@ -180,7 +180,8 @@ Value *CreateEltGEP(Value *A, unsigned i, Value *zeroIdx,
 }
 } // namespace
 
-void MatrixBitcastLowerPass::lowerMatrix(Instruction *M, Value *A) {
+void MatrixBitcastLowerPass::lowerMatrix(DxilModule &DM, Instruction *M,
+                                         Value *A) {
   for (auto it = M->user_begin(); it != M->user_end();) {
     User *U = *(it++);
     if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(U)) {
@@ -193,31 +194,42 @@ void MatrixBitcastLowerPass::lowerMatrix(Instruction *M, Value *A) {
         SmallVector<Value *, 2> idxList(GEP->idx_begin(), GEP->idx_end());
         DXASSERT(idxList.size() == 2,
                  "else not one dim matrix array index to matrix");
-
-        HLMatrixType MatTy = HLMatrixType::cast(EltTy);
-        Value *matSize = Builder.getInt32(MatTy.getNumElements());
-        idxList.back() = Builder.CreateMul(idxList.back(), matSize);
+        if (!DM.GetShaderModel()->IsSM69Plus()) {
+          HLMatrixType MatTy = HLMatrixType::cast(EltTy);
+          Value *matSize = Builder.getInt32(MatTy.getNumElements());
+          idxList.back() = Builder.CreateMul(idxList.back(), matSize);
+        }
         Value *NewGEP = Builder.CreateGEP(A, idxList);
-        lowerMatrix(GEP, NewGEP);
+        lowerMatrix(DM, GEP, NewGEP);
         DXASSERT(GEP->user_empty(), "else lower matrix fail");
         GEP->eraseFromParent();
       } else {
         DXASSERT(0, "invalid GEP for matrix");
       }
     } else if (BitCastInst *BCI = dyn_cast<BitCastInst>(U)) {
-      lowerMatrix(BCI, A);
+      lowerMatrix(DM, BCI, A);
       DXASSERT(BCI->user_empty(), "else lower matrix fail");
       BCI->eraseFromParent();
     } else if (LoadInst *LI = dyn_cast<LoadInst>(U)) {
       if (VectorType *Ty = dyn_cast<VectorType>(LI->getType())) {
         IRBuilder<> Builder(LI);
-        Value *zeroIdx = Builder.getInt32(0);
-        unsigned vecSize = Ty->getNumElements();
-        Value *NewVec = UndefValue::get(LI->getType());
-        for (unsigned i = 0; i < vecSize; i++) {
-          Value *GEP = CreateEltGEP(A, i, zeroIdx, Builder);
-          Value *Elt = Builder.CreateLoad(GEP);
-          NewVec = Builder.CreateInsertElement(NewVec, Elt, i);
+        Value *NewVec = nullptr;
+        if (DM.GetShaderModel()->IsSM69Plus()) {
+          // Just create a replacement load using the vector pointer.
+          Instruction *NewLI = LI->clone();
+          unsigned VecIdx = NewLI->getNumOperands() - 1;
+          NewLI->setOperand(VecIdx, A);
+          Builder.Insert(NewLI);
+          NewVec = NewLI;
+        } else {
+          Value *zeroIdx = Builder.getInt32(0);
+          unsigned vecSize = Ty->getNumElements();
+          NewVec = UndefValue::get(LI->getType());
+          for (unsigned i = 0; i < vecSize; i++) {
+            Value *GEP = CreateEltGEP(A, i, zeroIdx, Builder);
+            Value *Elt = Builder.CreateLoad(GEP);
+            NewVec = Builder.CreateInsertElement(NewVec, Elt, i);
+          }
         }
         LI->replaceAllUsesWith(NewVec);
         LI->eraseFromParent();
diff --git a/lib/HLSL/HLOperationLower.cpp b/lib/HLSL/HLOperationLower.cpp
index bc293357d6..d218136bd7 100644
--- a/lib/HLSL/HLOperationLower.cpp
+++ b/lib/HLSL/HLOperationLower.cpp
@@ -481,6 +481,44 @@ Value *TrivialDxilOperation(OP::OpCode opcode, ArrayRef<Value *> refArgs,
   return TrivialDxilOperation(opcode, refArgs, Ty, Inst->getType(), hlslOP, B);
 }
 
+
+Value *TrivialDxilVectorOperation(Function *dxilFunc, OP::OpCode opcode,
+                            ArrayRef<Value *> refArgs, Type *Ty,
+                            OP *hlslOP, IRBuilder<> &Builder) {
+  if (!Ty->isVoidTy()) {
+    Value *retVal =
+      Builder.CreateCall(dxilFunc, refArgs, hlslOP->GetOpCodeName(opcode));
+    return retVal;
+  } else {
+    // Cannot add name to void.
+    return Builder.CreateCall(dxilFunc, refArgs);
+  }
+}
+
+
+Value *TrivialDxilVectorUnaryOperationRet(OP::OpCode opcode, Value *src, Type *Ty,
+					  OP *hlslOP, IRBuilder<> &Builder) {
+
+  Constant *opArg = hlslOP->GetU32Const((unsigned)opcode);
+  Value *args[] = {opArg, src};
+
+  Function *dxilFunc = hlslOP->GetOpFunc(opcode, Ty);
+
+  return TrivialDxilVectorOperation(dxilFunc, opcode, args, Ty, hlslOP, Builder);
+}
+
+Value *TrivialDxilVectorBinaryOperation(OP::OpCode opcode, Value *src0, Value *src1,
+                                  hlsl::OP *hlslOP, IRBuilder<> &Builder) {
+  Type *Ty = src0->getType();
+
+  Constant *opArg = hlslOP->GetU32Const((unsigned)opcode);
+  Value *args[] = {opArg, src0, src1};
+
+  Function *dxilFunc = hlslOP->GetOpFunc(opcode, Ty);
+
+  return TrivialDxilVectorOperation(dxilFunc, opcode, args, Ty, hlslOP, Builder);
+}
+
 Value *TrivialDxilUnaryOperationRet(OP::OpCode opcode, Value *src, Type *RetTy,
                                     hlsl::OP *hlslOP, IRBuilder<> &Builder) {
   Type *Ty = src->getType();
@@ -507,17 +545,26 @@ Value *TrivialDxilBinaryOperation(OP::OpCode opcode, Value *src0, Value *src1,
   return TrivialDxilOperation(opcode, args, Ty, Ty, hlslOP, Builder);
 }
 
-Value *TrivialDxilTrinaryOperation(OP::OpCode opcode, Value *src0, Value *src1,
-                                   Value *src2, hlsl::OP *hlslOP,
-                                   IRBuilder<> &Builder) {
-  Type *Ty = src0->getType();
-
+Value *TrivialDxilTrinaryOperationRet(OP::OpCode opcode, Value *src0, Value *src1,
+				      Value *src2, Type *Ty, hlsl::OP *hlslOP,
+				      IRBuilder<> &Builder) {
   Constant *opArg = hlslOP->GetU32Const((unsigned)opcode);
   Value *args[] = {opArg, src0, src1, src2};
 
   return TrivialDxilOperation(opcode, args, Ty, Ty, hlslOP, Builder);
 }
 
+Value *TrivialDxilVectorTrinaryOperationRet(OP::OpCode opcode, Value *src0, Value *src1,
+					    Value *src2, Type *Ty, hlsl::OP *hlslOP,
+					    IRBuilder<> &Builder) {
+  Constant *opArg = hlslOP->GetU32Const((unsigned)opcode);
+  Value *args[] = {opArg, src0, src1, src2};
+
+  Function *dxilFunc = hlslOP->GetOpFunc(opcode, Ty);
+
+  return TrivialDxilVectorOperation(dxilFunc, opcode, args, Ty, hlslOP, Builder);
+}
+
 Value *TrivialUnaryOperation(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
                              HLOperationLowerHelper &helper,
                              HLObjectOperationLowerHelper *pObjHelper,
@@ -530,6 +577,24 @@ Value *TrivialUnaryOperation(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
   return retVal;
 }
 
+Value *TrivialVectorizableUnaryOperation(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
+					 HLOperationLowerHelper &helper,
+					 HLObjectOperationLowerHelper *pObjHelper,
+					 bool &Translated) {
+  Value *src0 = CI->getArgOperand(HLOperandIndex::kUnaryOpSrc0Idx);
+  Type *Ty = CI->getType();
+  IRBuilder<> Builder(CI);
+  hlsl::OP *hlslOP = &helper.hlslOP;
+
+  if (Ty->isVectorTy() &&
+      helper.M.GetShaderModel()->IsSM69Plus())
+    return TrivialDxilVectorUnaryOperationRet(opcode, src0, Ty,
+					      hlslOP, Builder);
+  else
+    return TrivialDxilUnaryOperationRet(opcode, src0, Ty,
+					hlslOP, Builder);
+}
+
 Value *TrivialBinaryOperation(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
                               HLOperationLowerHelper &helper,
                               HLObjectOperationLowerHelper *pObjHelper,
@@ -544,19 +609,36 @@ Value *TrivialBinaryOperation(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
   return binOp;
 }
 
-Value *TrivialTrinaryOperation(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
-                               HLOperationLowerHelper &helper,
-                               HLObjectOperationLowerHelper *pObjHelper,
-                               bool &Translated) {
+Value *TrivialVectorBinaryOperation(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
+				    HLOperationLowerHelper &helper,
+				    HLObjectOperationLowerHelper *pObjHelper,
+				    bool &Translated) {
   hlsl::OP *hlslOP = &helper.hlslOP;
+  Value *src0 = CI->getArgOperand(HLOperandIndex::kBinaryOpSrc0Idx);
+  Value *src1 = CI->getArgOperand(HLOperandIndex::kBinaryOpSrc1Idx);
+  IRBuilder<> Builder(CI);
+
+  Value *binOp =
+      TrivialDxilVectorBinaryOperation(opcode, src0, src1, hlslOP, Builder);
+  return binOp;
+}
+
+Value *TranslateFMA(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
+		    HLOperationLowerHelper &helper,
+		    HLObjectOperationLowerHelper *pObjHelper,
+		    bool &Translated) {
+  hlsl::OP *hlslOP = &helper.hlslOP;
+  Type *Ty = CI->getType();
   Value *src0 = CI->getArgOperand(HLOperandIndex::kTrinaryOpSrc0Idx);
   Value *src1 = CI->getArgOperand(HLOperandIndex::kTrinaryOpSrc1Idx);
   Value *src2 = CI->getArgOperand(HLOperandIndex::kTrinaryOpSrc2Idx);
   IRBuilder<> Builder(CI);
 
-  Value *triOp =
-      TrivialDxilTrinaryOperation(opcode, src0, src1, src2, hlslOP, Builder);
-  return triOp;
+  if (Ty->isVectorTy() &&
+      helper.M.GetShaderModel()->IsSM69Plus())
+    return TrivialDxilVectorTrinaryOperationRet(opcode, src0, src1, src2, Ty, hlslOP, Builder);
+  else
+    return TrivialDxilTrinaryOperationRet(opcode, src0, src1, src2, Ty, hlslOP, Builder);
 }
 
 Value *TrivialIsSpecialFloat(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
@@ -735,6 +817,12 @@ bool CanUseFxcMulOnlyPatternForPow(IRBuilder<> &Builder, Value *x, Value *pow,
     }
   }
 
+  // Only apply on aggregates of 16 or fewer elements,
+  // representing the max 4x4 matrix size.
+  Type *xTy = x->getType();
+  if (xTy->isVectorTy() && xTy->getVectorNumElements() > 16)
+    return false;
+
   APFloat powAPF = isa<ConstantDataVector>(pow)
                        ? cast<ConstantDataVector>(pow)->getElementAsAPFloat(0)
                        : // should be a splat value
@@ -1896,9 +1984,16 @@ Value *TranslateClamp(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
 
   IRBuilder<> Builder(CI);
   // min(max(x, minVal), maxVal).
-  Value *maxXMinVal =
+  if (Ty->isVectorTy() &&
+      helper.M.GetShaderModel()->IsSM69Plus()) {
+    Value *maxXMinVal =
+      TrivialDxilVectorBinaryOperation(maxOp, x, minVal, hlslOP, Builder);
+    return TrivialDxilVectorBinaryOperation(minOp, maxXMinVal, maxVal, hlslOP, Builder);
+  } else {
+    Value *maxXMinVal =
       TrivialDxilBinaryOperation(maxOp, x, minVal, hlslOP, Builder);
-  return TrivialDxilBinaryOperation(minOp, maxXMinVal, maxVal, hlslOP, Builder);
+    return TrivialDxilBinaryOperation(minOp, maxXMinVal, maxVal, hlslOP, Builder);
+  }
 }
 
 Value *TranslateClip(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
@@ -2211,8 +2306,11 @@ Value *TranslateExp(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
         ConstantVector::getSplat(Ty->getVectorNumElements(), log2eConst);
   }
   val = Builder.CreateFMul(log2eConst, val);
-  Value *exp = TrivialDxilUnaryOperation(OP::OpCode::Exp, val, hlslOP, Builder);
-  return exp;
+  if (Ty->isVectorTy() &&
+      helper.M.GetShaderModel()->IsSM69Plus())
+    return TrivialDxilVectorUnaryOperationRet(OP::OpCode::Exp, val, Ty, hlslOP, Builder);
+  else
+    return TrivialDxilUnaryOperationRet(OP::OpCode::Exp, val, Ty, hlslOP, Builder);
 }
 
 Value *TranslateLog(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
@@ -2227,7 +2325,12 @@ Value *TranslateLog(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
   if (Ty != Ty->getScalarType()) {
     ln2Const = ConstantVector::getSplat(Ty->getVectorNumElements(), ln2Const);
   }
-  Value *log = TrivialDxilUnaryOperation(OP::OpCode::Log, val, hlslOP, Builder);
+  Value *log = nullptr;
+  if (Ty->isVectorTy() &&
+      helper.M.GetShaderModel()->IsSM69Plus())
+    log = TrivialDxilVectorUnaryOperationRet(OP::OpCode::Log, val, Ty, hlslOP, Builder);
+  else
+    log = TrivialDxilUnaryOperationRet(OP::OpCode::Log, val, Ty, hlslOP, Builder);
 
   return Builder.CreateFMul(ln2Const, log);
 }
@@ -2287,8 +2390,13 @@ Value *TranslateFUIBinary(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
       break;
     }
   }
-  return TrivialBinaryOperation(CI, IOP, opcode, helper, pObjHelper,
-                                Translated);
+  if (CI->getType()->isVectorTy() &&
+      helper.M.GetShaderModel()->IsSM69Plus())
+    return TrivialVectorBinaryOperation(CI, IOP, opcode, helper, pObjHelper,
+					Translated);
+  else
+    return TrivialBinaryOperation(CI, IOP, opcode, helper, pObjHelper,
+				  Translated);
 }
 
 Value *TranslateFUITrinary(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
@@ -2305,8 +2413,15 @@ Value *TranslateFUITrinary(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
       break;
     }
   }
-  return TrivialTrinaryOperation(CI, IOP, opcode, helper, pObjHelper,
-                                 Translated);
+
+  hlsl::OP *hlslOP = &helper.hlslOP;
+  Type *Ty = CI->getType();
+  Value *src0 = CI->getArgOperand(HLOperandIndex::kTrinaryOpSrc0Idx);
+  Value *src1 = CI->getArgOperand(HLOperandIndex::kTrinaryOpSrc1Idx);
+  Value *src2 = CI->getArgOperand(HLOperandIndex::kTrinaryOpSrc2Idx);
+  IRBuilder<> Builder(CI);
+
+  return TrivialDxilTrinaryOperationRet(opcode, src0, src1, src2, Ty, hlslOP, Builder);
 }
 
 Value *TranslateFrexp(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
@@ -2428,18 +2543,22 @@ Value *TrivialDotOperation(OP::OpCode opcode, Value *src0, Value *src1,
   return dotOP;
 }
 
-Value *TranslateIDot(Value *arg0, Value *arg1, unsigned vecSize,
-                     hlsl::OP *hlslOP, IRBuilder<> &Builder,
-                     bool Unsigned = false) {
+// Instead of using a DXIL intrinsic, implement a dot product operation using
+// multiply and add operations. Used for integer dots and long vectors.
+Value *ExpandDot(Value *arg0, Value *arg1, unsigned vecSize,
+		 hlsl::OP *hlslOP, IRBuilder<> &Builder,
+		 bool Unsigned = false) {
   auto madOpCode = Unsigned ? DXIL::OpCode::UMad : DXIL::OpCode::IMad;
+  if (arg0->getType()->getScalarType()->isFloatingPointTy())
+    madOpCode = DXIL::OpCode::FMad;
   Value *Elt0 = Builder.CreateExtractElement(arg0, (uint64_t)0);
   Value *Elt1 = Builder.CreateExtractElement(arg1, (uint64_t)0);
   Value *Result = Builder.CreateMul(Elt0, Elt1);
-  for (unsigned iVecElt = 1; iVecElt < vecSize; ++iVecElt) {
-    Elt0 = Builder.CreateExtractElement(arg0, iVecElt);
-    Elt1 = Builder.CreateExtractElement(arg1, iVecElt);
-    Result = TrivialDxilTrinaryOperation(madOpCode, Elt0, Elt1, Result, hlslOP,
-                                         Builder);
+  for (unsigned Elt = 1; Elt < vecSize; ++Elt) {
+    Elt0 = Builder.CreateExtractElement(arg0, Elt);
+    Elt1 = Builder.CreateExtractElement(arg1, Elt);
+    Result = TrivialDxilTrinaryOperationRet(madOpCode, Elt0, Elt1, Result, Elt0->getType(), hlslOP,
+					    Builder);
   }
 
   return Result;
@@ -2477,10 +2596,10 @@ Value *TranslateDot(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
   unsigned vecSize = Ty->getVectorNumElements();
   Value *arg1 = CI->getArgOperand(HLOperandIndex::kBinaryOpSrc1Idx);
   IRBuilder<> Builder(CI);
-  if (Ty->getScalarType()->isFloatingPointTy()) {
+  if (Ty->getScalarType()->isFloatingPointTy() && Ty->getVectorNumElements() <= 4) {
     return TranslateFDot(arg0, arg1, vecSize, hlslOP, Builder);
   } else {
-    return TranslateIDot(arg0, arg1, vecSize, hlslOP, Builder,
+    return ExpandDot(arg0, arg1, vecSize, hlslOP, Builder,
                          IOP == IntrinsicOp::IOP_udot);
   }
 }
@@ -2664,8 +2783,8 @@ Value *TranslateMSad4(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
   byteSrc = Builder.CreateInsertElement(byteSrc, byteSrcElt, 3);
 
   // Msad on vecref and byteSrc.
-  return TrivialDxilTrinaryOperation(DXIL::OpCode::Msad, vecRef, byteSrc, accum,
-                                     hlslOP, Builder);
+  return TrivialDxilTrinaryOperationRet(DXIL::OpCode::Msad, vecRef, byteSrc, accum,
+					vecRef->getType(), hlslOP, Builder);
 }
 
 Value *TranslateRCP(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
@@ -2988,9 +3107,10 @@ static Value *ScalarizeResRet(Type *RetTy, Value *ResRet,
 static Value *ScalarizeElements(Type *RetTy, ArrayRef<Value *> Elts,
                                 IRBuilder<> &Builder) {
   // Extract value part.
-  Value *retVal = llvm::UndefValue::get(RetTy);
+  Value *retVal = nullptr;
   if (RetTy->isVectorTy()) {
     unsigned vecSize = RetTy->getVectorNumElements();
+    retVal = UndefValue::get(VectorType::get(Elts[0]->getType(), vecSize));
     DXASSERT(vecSize <= Elts.size(), "vector size mismatch");
     for (unsigned i = 0; i < vecSize; i++) {
       Value *retComp = Elts[i];
@@ -3046,7 +3166,7 @@ Value *TranslateMul(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
       if (arg0Ty->getScalarType()->isFloatingPointTy()) {
         return TranslateFDot(arg0, arg1, vecSize, hlslOP, Builder);
       } else {
-        return TranslateIDot(arg0, arg1, vecSize, hlslOP, Builder,
+        return ExpandDot(arg0, arg1, vecSize, hlslOP, Builder,
                              IOP == IntrinsicOp::IOP_umul);
       }
     } else {
@@ -3941,14 +4061,40 @@ TranslateWriteSamplerFeedback(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
 }
 
 // Load/Store intrinsics.
+OP::OpCode LoadOpFromResKind(DxilResource::Kind RK) {
+  switch (RK) {
+  case DxilResource::Kind::RawBuffer:
+  case DxilResource::Kind::StructuredBuffer:
+    return OP::OpCode::RawBufferLoad;
+  case DxilResource::Kind::TypedBuffer:
+    return OP::OpCode::BufferLoad;
+  case DxilResource::Kind::Invalid:
+    DXASSERT(0, "invalid resource kind");
+    break;
+  default:
+    return OP::OpCode::TextureLoad;
+  }
+  return OP::OpCode::TextureLoad;
+}
+
 struct ResLoadHelper {
+  // Default constructor uses CI load intrinsic call
+  //  to get the retval and various location indicators.
   ResLoadHelper(CallInst *CI, DxilResource::Kind RK, DxilResourceBase::Class RC,
-                Value *h, IntrinsicOp IOP, bool bForSubscript = false);
-  // For double subscript.
-  ResLoadHelper(Instruction *ldInst, Value *h, Value *idx, Value *mip)
-      : opcode(OP::OpCode::TextureLoad),
-        intrinsicOpCode(IntrinsicOp::Num_Intrinsics), handle(h), retVal(ldInst),
-        addr(idx), offset(nullptr), status(nullptr), mipLevel(mip) {}
+                Value *h, IntrinsicOp IOP, LoadInst *TyBufSubLoad = nullptr);
+  // Alternative constructor explicitly sets the index.
+  // Used for some subscript operators.
+  ResLoadHelper(Instruction *ldInst, DxilResource::Kind RK, Value *h,
+                Value *idx, Value *Offset, Value *mip = nullptr)
+      : intrinsicOpCode(IntrinsicOp::Num_Intrinsics), handle(h), retVal(ldInst),
+        addr(idx), offset(Offset), status(nullptr), mipLevel(mip) {
+    opcode = LoadOpFromResKind(RK);
+    if (opcode == OP::OpCode::RawBufferLoad &&
+        ldInst->getType()->isVectorTy() &&
+        ldInst->getType()->getVectorNumElements() > 1 &&
+        ldInst->getModule()->GetHLModule().GetShaderModel()->IsSM69Plus())
+      opcode = OP::OpCode::RawBufferVectorLoad;
+  }
   OP::OpCode opcode;
   IntrinsicOp intrinsicOpCode;
   unsigned dxilMajor;
@@ -3961,26 +4107,23 @@ struct ResLoadHelper {
   Value *mipLevel;
 };
 
+// Uses CI arguments to determine the index, offset, and mipLevel also depending
+// on the RK/RC resource kind and class, which determine the opcode.
+// Handle and IOP are set explicitly.
+// For typed buffer loads, the call instruction feeds into a load
+// represented by TyBufSubLoad which determines the instruction to replace.
+// Otherwise, CI is replaced.
 ResLoadHelper::ResLoadHelper(CallInst *CI, DxilResource::Kind RK,
                              DxilResourceBase::Class RC, Value *hdl,
-                             IntrinsicOp IOP, bool bForSubscript)
+                             IntrinsicOp IOP, LoadInst *TyBufSubLoad)
     : intrinsicOpCode(IOP), handle(hdl), offset(nullptr), status(nullptr) {
-  switch (RK) {
-  case DxilResource::Kind::RawBuffer:
-  case DxilResource::Kind::StructuredBuffer:
-    opcode = OP::OpCode::RawBufferLoad;
-    break;
-  case DxilResource::Kind::TypedBuffer:
-    opcode = OP::OpCode::BufferLoad;
-    break;
-  case DxilResource::Kind::Invalid:
-    DXASSERT(0, "invalid resource kind");
-    break;
-  default:
-    opcode = OP::OpCode::TextureLoad;
-    break;
-  }
-  retVal = CI;
+  opcode = LoadOpFromResKind(RK);
+  bool bForSubscript = false;
+  if (TyBufSubLoad) {
+    bForSubscript = true;
+    retVal = TyBufSubLoad;
+  } else
+    retVal = CI;
   const unsigned kAddrIdx = HLOperandIndex::kBufLoadAddrOpIdx;
   addr = CI->getArgOperand(kAddrIdx);
   unsigned argc = CI->getNumArgOperands();
@@ -4013,8 +4156,9 @@ ResLoadHelper::ResLoadHelper(CallInst *CI, DxilResource::Kind RK,
           RK == DxilResource::Kind::Texture2DMSArray) {
         offsetIdx = HLOperandIndex::kTex2DMSLoadOffsetOpIdx;
         statusIdx = HLOperandIndex::kTex2DMSLoadStatusOpIdx;
-        mipLevel =
-            CI->getArgOperand(HLOperandIndex::kTex2DMSLoadSampleIdxOpIdx);
+        if (!bForSubscript)
+          mipLevel =
+              CI->getArgOperand(HLOperandIndex::kTex2DMSLoadSampleIdxOpIdx);
       }
 
       if (argc > offsetIdx)
@@ -4026,7 +4170,12 @@ ResLoadHelper::ResLoadHelper(CallInst *CI, DxilResource::Kind RK,
                (RK == DxilResource::Kind::Texture2DMS ||
                 RK == DxilResource::Kind::Texture2DMSArray)) {
       unsigned statusIdx = HLOperandIndex::kTex2DMSLoadStatusOpIdx;
-      mipLevel = CI->getArgOperand(HLOperandIndex::kTex2DMSLoadSampleIdxOpIdx);
+
+      if (!bForSubscript)
+        mipLevel =
+            CI->getArgOperand(HLOperandIndex::kTex2DMSLoadSampleIdxOpIdx);
+      else
+        mipLevel = IRBuilder<>(CI).getInt32(0);
 
       if (argc > statusIdx)
         status = CI->getArgOperand(statusIdx);
@@ -4038,9 +4187,19 @@ ResLoadHelper::ResLoadHelper(CallInst *CI, DxilResource::Kind RK,
         status = CI->getArgOperand(kStatusIdx);
     }
   } else {
+    if (opcode == OP::OpCode::RawBufferLoad &&
+        CI->getType()->isVectorTy() &&
+        CI->getType()->getVectorNumElements() > 1 &&
+        CI->getModule()->GetHLModule().GetShaderModel()->IsSM69Plus())
+      opcode = OP::OpCode::RawBufferVectorLoad;
     const unsigned kStatusIdx = HLOperandIndex::kBufLoadStatusOpIdx;
     if (argc > kStatusIdx)
       status = CI->getArgOperand(kStatusIdx);
+    Type *i32Ty = IRBuilder<>(CI).getInt32Ty();
+    if (DXIL::IsStructuredBuffer(RK))
+      offset = ConstantInt::get(i32Ty, 0U);
+    else
+      offset = UndefValue::get(i32Ty);
   }
 }
 
@@ -4048,35 +4207,6 @@ void TranslateStructBufSubscript(CallInst *CI, Value *handle, Value *status,
                                  hlsl::OP *OP, HLResource::Kind RK,
                                  const DataLayout &DL);
 
-// Create { v0, v1 } from { v0.lo, v0.hi, v1.lo, v1.hi }
-void Make64bitResultForLoad(Type *EltTy, ArrayRef<Value *> resultElts32,
-                            unsigned size, MutableArrayRef<Value *> resultElts,
-                            hlsl::OP *hlslOP, IRBuilder<> &Builder) {
-  Type *i64Ty = Builder.getInt64Ty();
-  Type *doubleTy = Builder.getDoubleTy();
-  if (EltTy == doubleTy) {
-    Function *makeDouble =
-        hlslOP->GetOpFunc(DXIL::OpCode::MakeDouble, doubleTy);
-    Value *makeDoubleOpArg =
-        Builder.getInt32((unsigned)DXIL::OpCode::MakeDouble);
-    for (unsigned i = 0; i < size; i++) {
-      Value *lo = resultElts32[2 * i];
-      Value *hi = resultElts32[2 * i + 1];
-      Value *V = Builder.CreateCall(makeDouble, {makeDoubleOpArg, lo, hi});
-      resultElts[i] = V;
-    }
-  } else {
-    for (unsigned i = 0; i < size; i++) {
-      Value *lo = resultElts32[2 * i];
-      Value *hi = resultElts32[2 * i + 1];
-      lo = Builder.CreateZExt(lo, i64Ty);
-      hi = Builder.CreateZExt(hi, i64Ty);
-      hi = Builder.CreateShl(hi, 32);
-      resultElts[i] = Builder.CreateOr(lo, hi);
-    }
-  }
-}
-
 static Constant *GetRawBufferMaskForETy(Type *Ty, unsigned NumComponents,
                                         hlsl::OP *OP) {
   unsigned mask = 0;
@@ -4108,183 +4238,165 @@ Value *GenerateRawBufLd(Value *handle, Value *bufIdx, Value *offset,
                         IRBuilder<> &Builder, unsigned NumComponents,
                         Constant *alignment);
 
-static Value *TranslateRawBufVecLd(Type *VecEltTy, unsigned VecElemCount,
-                                   IRBuilder<> &Builder, Value *handle,
-                                   hlsl::OP *OP, Value *status, Value *bufIdx,
-                                   Value *baseOffset, const DataLayout &DL,
-                                   std::vector<Value *> &bufLds,
-                                   unsigned baseAlign, bool isScalarTy = false);
-
-void TranslateLoad(ResLoadHelper &helper, HLResource::Kind RK,
-                   IRBuilder<> &Builder, hlsl::OP *OP, const DataLayout &DL) {
-
-  Type *Ty = helper.retVal->getType();
-  if (Ty->isPointerTy()) {
-    DXASSERT(!DxilResource::IsAnyTexture(RK),
-             "Textures should not be treated as structured buffers.");
-    TranslateStructBufSubscript(cast<CallInst>(helper.retVal), helper.handle,
-                                helper.status, OP, RK, DL);
-    return;
-  }
+static Value *GenerateBufLd(hlsl::OP *OP, IRBuilder<> &Builder,
+                            OP::OpCode opcode, Type *RegTy, Type *MemTy,
+                            unsigned NumElements, ArrayRef<Value *> Args,
+                            std::vector<Value *>::iterator EltIt);
 
+// Sets up arguments for buffer load call.
+static SmallVector<Value *, 12> GetBufLoadArgs(ResLoadHelper helper,
+                                               HLResource::Kind RK,
+                                               IRBuilder<> Builder, Type *EltTy,
+                                               unsigned LdSize) {
   OP::OpCode opcode = helper.opcode;
+  llvm::Constant *opArg = Builder.getInt32((uint32_t)opcode);
 
-  Type *i32Ty = Builder.getInt32Ty();
-  Type *i64Ty = Builder.getInt64Ty();
-  Type *doubleTy = Builder.getDoubleTy();
-  Type *EltTy = Ty->getScalarType();
-  unsigned numComponents = 1;
-  if (Ty->isVectorTy()) {
-    numComponents = Ty->getVectorNumElements();
-  }
-
-  if (DXIL::IsStructuredBuffer(RK) || DXIL::IsRawBuffer(RK)) {
-    std::vector<Value *> bufLds;
-    const bool isBool = EltTy->isIntegerTy(1);
+  unsigned alignment = RK == DxilResource::Kind::RawBuffer ? 4U : 8U;
+  alignment = std::min(alignment, LdSize);
+  Constant *alignmentVal = Builder.getInt32(alignment);
 
-    // Bool are represented as i32 in memory
-    Type *MemReprTy = isBool ? Builder.getInt32Ty() : EltTy;
-    bool isScalarTy = !Ty->isVectorTy();
-
-    Value *retValNew = nullptr;
-    if (DXIL::IsStructuredBuffer(RK)) {
-      retValNew = TranslateRawBufVecLd(
-          MemReprTy, numComponents, Builder, helper.handle, OP, helper.status,
-          helper.addr, OP->GetU32Const(0), DL, bufLds,
-          /*baseAlign (in bytes)*/ 8, isScalarTy);
-    } else {
-      retValNew =
-          TranslateRawBufVecLd(MemReprTy, numComponents, Builder, helper.handle,
-                               OP, helper.status, nullptr, helper.addr, DL,
-                               bufLds, /*baseAlign (in bytes)*/ 4, isScalarTy);
-    }
-
-    DXASSERT_NOMSG(!bufLds.empty());
-    dxilutil::MigrateDebugValue(helper.retVal, bufLds.front());
-
-    if (isBool) {
-      // Convert result back to register representation.
-      retValNew = Builder.CreateICmpNE(
-          retValNew, Constant::getNullValue(retValNew->getType()));
-    }
-
-    helper.retVal->replaceAllUsesWith(retValNew);
-    helper.retVal = retValNew;
-    return;
-  }
-
-  bool isTyped = opcode == OP::OpCode::TextureLoad ||
-                 RK == DxilResource::Kind::TypedBuffer;
-  bool is64 = EltTy == i64Ty || EltTy == doubleTy;
-  if (is64 && isTyped) {
-    EltTy = i32Ty;
-  }
-  bool isBool = EltTy->isIntegerTy(1);
-  if (isBool) {
-    // Value will be loaded in its memory representation.
-    EltTy = i32Ty;
-    if (Ty->isVectorTy())
-      Ty = VectorType::get(EltTy, numComponents);
-  }
-
-  Function *F = OP->GetOpFunc(opcode, EltTy);
-  llvm::Constant *opArg = OP->GetU32Const((unsigned)opcode);
-
-  llvm::Value *undefI = llvm::UndefValue::get(i32Ty);
-
-  SmallVector<Value *, 12> loadArgs;
-  loadArgs.emplace_back(opArg);         // opcode
-  loadArgs.emplace_back(helper.handle); // resource handle
+  // Assemble args is specific to the type bab/struct/typed
+  // Typed needs to handle the possibility of vector coords
+  // Raws need to calculate alignment and mask values.
+  SmallVector<Value *, 12> Args;
+  Args.emplace_back(opArg);         // opcode @0
+  Args.emplace_back(helper.handle); // resource handle @1
 
+  // Set offsets appropriate for the load operation.
+  bool isVectorAddr = helper.addr->getType()->isVectorTy();
   if (opcode == OP::OpCode::TextureLoad) {
-    // set mip level
-    loadArgs.emplace_back(helper.mipLevel);
-  }
+    llvm::Value *undefI = llvm::UndefValue::get(Builder.getInt32Ty());
 
-  if (opcode == OP::OpCode::TextureLoad) {
-    // texture coord
+    // Set mip level or sample for MS texutures @3.
+    Args.emplace_back(helper.mipLevel);
+    // Set texture coords according to resource kind @4-6
+    // Coords unused by the resource kind are undefs.
     unsigned coordSize = DxilResource::GetNumCoords(RK);
-    bool isVectorAddr = helper.addr->getType()->isVectorTy();
     for (unsigned i = 0; i < 3; i++) {
       if (i < coordSize) {
-        loadArgs.emplace_back(isVectorAddr
-                                  ? Builder.CreateExtractElement(helper.addr, i)
-                                  : helper.addr);
+        Args.emplace_back(isVectorAddr
+                              ? Builder.CreateExtractElement(helper.addr, i)
+                              : helper.addr);
       } else
-        loadArgs.emplace_back(undefI);
+        Args.emplace_back(undefI);
+    }
+    // Set texture offsets according to resource kind @7-9
+    // Coords unused by the resource kind are undefs.
+    unsigned offsetSize = DxilResource::GetNumOffsets(RK);
+    if (!helper.offset || isa<llvm::UndefValue>(helper.offset))
+      offsetSize = 0;
+    for (unsigned i = 0; i < 3; i++) {
+      if (i < offsetSize)
+        Args.emplace_back(Builder.CreateExtractElement(helper.offset, i));
+      else
+        Args.emplace_back(undefI);
     }
   } else {
-    if (helper.addr->getType()->isVectorTy()) {
-      Value *scalarOffset =
-          Builder.CreateExtractElement(helper.addr, (uint64_t)0);
+    // coord (may be changed later) @2
+    Args.emplace_back(
+        isVectorAddr ? Builder.CreateExtractElement(helper.addr, (uint64_t)0)
+                     : helper.addr);
+    Args.emplace_back(helper.offset); // offset (may be changed later) @3
+
+    if (opcode == OP::OpCode::RawBufferLoad) {
+      // Unlike typed buffer load, raw buffer load has mask and alignment.
+      Args.emplace_back(nullptr);      // mask (to be added later) @4
+      Args.emplace_back(alignmentVal); // alignment @5
+    } else if (opcode == OP::OpCode::RawBufferVectorLoad) {
+      // RawBufferVectorLoad takes no mask argument.
+      Args.emplace_back(alignmentVal); // alignment @4
+    }
 
-      // TODO: calculate the real address based on opcode
+  }
+  return Args;
+}
 
-      loadArgs.emplace_back(scalarOffset); // offset
-    } else {
-      // TODO: calculate the real address based on opcode
+// Emits as many calls as needed to load the full vector
+// Performs any needed extractions and conversions of the results.
+Value *TranslateBufLoad(ResLoadHelper &helper, HLResource::Kind RK,
+                        IRBuilder<> &Builder, hlsl::OP *OP,
+                        const DataLayout &DL) {
+  OP::OpCode opcode = helper.opcode;
+  Type *Ty = helper.retVal->getType();
+  unsigned numComponents = 1;
+  if (Ty->isVectorTy())
+    numComponents = Ty->getVectorNumElements();
 
-      loadArgs.emplace_back(helper.addr); // offset
-    }
-  }
-  // offset 0
-  if (opcode == OP::OpCode::TextureLoad) {
-    if (helper.offset && !isa<llvm::UndefValue>(helper.offset)) {
-      unsigned offsetSize = DxilResource::GetNumOffsets(RK);
-      for (unsigned i = 0; i < 3; i++) {
-        if (i < offsetSize)
-          loadArgs.emplace_back(Builder.CreateExtractElement(helper.offset, i));
-        else
-          loadArgs.emplace_back(undefI);
-      }
-    } else {
-      loadArgs.emplace_back(undefI);
-      loadArgs.emplace_back(undefI);
-      loadArgs.emplace_back(undefI);
-    }
-  }
+  const bool isTyped = DXIL::IsTyped(RK);
+  Type *EltTy = Ty->getScalarType();
+  const bool is64 = (EltTy->isIntegerTy(64) || EltTy->isDoubleTy());
+  const bool isBool = EltTy->isIntegerTy(1);
+  Type *MemTy = EltTy;
+  if (isBool || (is64 && isTyped))
+    // Value will be loaded in its memory representation.
+    MemTy = Builder.getInt32Ty();
+  unsigned LdSize = DL.getTypeAllocSize(MemTy);
 
-  // Offset 1
-  if (RK == DxilResource::Kind::TypedBuffer) {
-    loadArgs.emplace_back(undefI);
-  }
+  std::vector<Value *> elts(numComponents);
+
+  SmallVector<Value *, 12> Args =
+      GetBufLoadArgs(helper, RK, Builder, MemTy, LdSize);
 
-  Value *ResRet = Builder.CreateCall(F, loadArgs, OP->GetOpCodeName(opcode));
-  dxilutil::MigrateDebugValue(helper.retVal, ResRet);
+  const unsigned kCoordIdx = 2;
+  const unsigned kOffsetIdx = 3;
+  const unsigned kMaskIdx = 4;
 
+  // Keep track of the first load for debug info migration.
+  Value *FirstLd = nullptr;
   Value *retValNew = nullptr;
-  if (!is64 || !isTyped) {
-    retValNew = ScalarizeResRet(Ty, ResRet, Builder);
+  if (opcode == OP::OpCode::RawBufferVectorLoad) {
+    FirstLd = GenerateBufLd(OP, Builder, opcode, Ty->getScalarType(),
+                            VectorType::get(MemTy, numComponents), 1, Args,
+                            elts.begin());
+    UpdateStatus(FirstLd, helper.status, Builder, OP);
+    retValNew = elts[0];
   } else {
-    unsigned size = numComponents;
-    DXASSERT(size <= 2, "typed buffer only allow 4 dwords");
-    EltTy = Ty->getScalarType();
-    Value *Elts[2];
-
-    Make64bitResultForLoad(Ty->getScalarType(),
-                           {
-                               Builder.CreateExtractValue(ResRet, 0),
-                               Builder.CreateExtractValue(ResRet, 1),
-                               Builder.CreateExtractValue(ResRet, 2),
-                               Builder.CreateExtractValue(ResRet, 3),
-                           },
-                           size, Elts, OP, Builder);
-
-    retValNew = ScalarizeElements(Ty, Elts, Builder);
+    // Create calls to function object.
+    // Typed buffer loads are limited to one load of up to 4 32-bit values.
+    // Raw buffer loads might need multiple loads in chunks of 4.
+    for (unsigned i = 0; i < numComponents;) {
+      unsigned chunkSize = (numComponents - i) <= 4 ? numComponents - i : 4;
+
+      // Assign mask for raw buffer loads.
+      if (opcode == OP::OpCode::RawBufferLoad)
+        Args[kMaskIdx] = GetRawBufferMaskForETy(MemTy, chunkSize, OP);
+
+      Value *Ld = GenerateBufLd(OP, Builder, opcode, Ty->getScalarType(), MemTy,
+                                chunkSize, Args, elts.begin() + i);
+      i += chunkSize;
+
+      // Update status.
+      UpdateStatus(Ld, helper.status, Builder, OP);
+
+      if (!FirstLd)
+        FirstLd = Ld;
+
+      if (opcode == OP::OpCode::RawBufferLoad && i < numComponents) {
+        if (RK == DxilResource::Kind::RawBuffer)
+          // Raw buffers can't use offset param. Add to coord index.
+          Args[kCoordIdx] =
+            Builder.CreateAdd(Args[kCoordIdx], OP->GetU32Const(4 * LdSize));
+        else
+          // Structured buffers increment the offset parameter.
+          Args[kOffsetIdx] =
+            Builder.CreateAdd(Args[kOffsetIdx], OP->GetU32Const(4 * LdSize));
+      }
+    }
+    retValNew = ScalarizeElements(Ty, elts, Builder);
   }
 
+  DXASSERT(FirstLd, "No loads created by TranslateBufLoad");
+
   if (isBool) {
     // Convert result back to register representation.
     retValNew = Builder.CreateICmpNE(
         retValNew, Constant::getNullValue(retValNew->getType()));
   }
 
-  // replace
   helper.retVal->replaceAllUsesWith(retValNew);
-  // Save new ret val.
   helper.retVal = retValNew;
-  // get status
-  UpdateStatus(ResRet, helper.status, Builder, OP);
+
+  return FirstLd;
 }
 
 Value *TranslateResourceLoad(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
@@ -4292,6 +4404,7 @@ Value *TranslateResourceLoad(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
                              HLObjectOperationLowerHelper *pObjHelper,
                              bool &Translated) {
   hlsl::OP *hlslOP = &helper.hlslOP;
+  DataLayout &DL = helper.dataLayout;
   Value *handle = CI->getArgOperand(HLOperandIndex::kHandleOpIdx);
 
   IRBuilder<> Builder(CI);
@@ -4299,9 +4412,19 @@ Value *TranslateResourceLoad(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
   DXIL::ResourceClass RC = pObjHelper->GetRC(handle);
   DXIL::ResourceKind RK = pObjHelper->GetRK(handle);
 
-  ResLoadHelper loadHelper(CI, RK, RC, handle, IOP);
-  TranslateLoad(loadHelper, RK, Builder, hlslOP, helper.dataLayout);
-  // CI is replaced in TranslateLoad.
+  ResLoadHelper ldHelper(CI, RK, RC, handle, IOP);
+  Type *Ty = CI->getType();
+  Value *Ld = nullptr;
+  if (Ty->isPointerTy()) {
+    DXASSERT(!DxilResource::IsAnyTexture(RK),
+             "Textures should not be treated as structured buffers.");
+    TranslateStructBufSubscript(cast<CallInst>(ldHelper.retVal), handle,
+                                ldHelper.status, hlslOP, RK, DL);
+  } else {
+    Ld = TranslateBufLoad(ldHelper, RK, Builder, hlslOP, DL);
+    dxilutil::MigrateDebugValue(CI, Ld);
+  }
+  // CI is replaced by above translation calls..
   return nullptr;
 }
 
@@ -4345,19 +4468,20 @@ void Split64bitValForStore(Type *EltTy, ArrayRef<Value *> vals, unsigned size,
 }
 
 void TranslateStore(DxilResource::Kind RK, Value *handle, Value *val,
-                    Value *offset, IRBuilder<> &Builder, hlsl::OP *OP,
-                    Value *sampIdx = nullptr) {
+                    Value *Idx, Value *offset, IRBuilder<> &Builder,
+                    hlsl::OP *OP, Value *sampIdx = nullptr) {
   Type *Ty = val->getType();
-
-  // This function is no longer used for lowering stores to a
-  // structured buffer.
-  DXASSERT_NOMSG(RK != DxilResource::Kind::StructuredBuffer);
-
   OP::OpCode opcode = OP::OpCode::NumOpCodes;
+  bool isTyped = true;
   switch (RK) {
   case DxilResource::Kind::RawBuffer:
   case DxilResource::Kind::StructuredBuffer:
+    isTyped = false;
     opcode = OP::OpCode::RawBufferStore;
+    // Where shader model and type allows, use vector store intrinsic.
+    if (OP->GetModule()->GetHLModule().GetShaderModel()->IsSM69Plus() &&
+        Ty->isVectorTy() && Ty->getVectorNumElements() > 1)
+      opcode = OP::OpCode::RawBufferVectorStore;
     break;
   case DxilResource::Kind::TypedBuffer:
     opcode = OP::OpCode::BufferStore;
@@ -4374,10 +4498,6 @@ void TranslateStore(DxilResource::Kind RK, Value *handle, Value *val,
     break;
   }
 
-  bool isTyped = opcode == OP::OpCode::TextureStore ||
-                 opcode == OP::OpCode::TextureStoreSample ||
-                 RK == DxilResource::Kind::TypedBuffer;
-
   Type *i32Ty = Builder.getInt32Ty();
   Type *i64Ty = Builder.getInt64Ty();
   Type *doubleTy = Builder.getDoubleTy();
@@ -4404,7 +4524,6 @@ void TranslateStore(DxilResource::Kind RK, Value *handle, Value *val,
     EltTy = i32Ty;
   }
 
-  Function *F = OP->GetOpFunc(opcode, EltTy);
   llvm::Constant *opArg = OP->GetU32Const((unsigned)opcode);
 
   llvm::Value *undefI =
@@ -4416,44 +4535,58 @@ void TranslateStore(DxilResource::Kind RK, Value *handle, Value *val,
   storeArgs.emplace_back(opArg);  // opcode
   storeArgs.emplace_back(handle); // resource handle
 
-  unsigned offset0Idx = 0;
-  if (RK == DxilResource::Kind::RawBuffer ||
-      RK == DxilResource::Kind::TypedBuffer) {
-    // Offset 0
-    if (offset->getType()->isVectorTy()) {
-      Value *scalarOffset = Builder.CreateExtractElement(offset, (uint64_t)0);
-      storeArgs.emplace_back(scalarOffset); // offset
+  unsigned OffsetIdx = 0;
+  if (opcode == OP::OpCode::RawBufferStore ||
+      opcode == OP::OpCode::RawBufferVectorStore ||
+      opcode == OP::OpCode::BufferStore) {
+    // Append Coord0 (Index) value.
+    if (Idx->getType()->isVectorTy()) {
+      Value *ScalarIdx = Builder.CreateExtractElement(Idx, (uint64_t)0);
+      storeArgs.emplace_back(ScalarIdx); // Coord0 (Index).
     } else {
-      storeArgs.emplace_back(offset); // offset
+      storeArgs.emplace_back(Idx); // Coord0 (Index).
     }
 
-    // Store offset0 for later use
-    offset0Idx = storeArgs.size() - 1;
+    // Store OffsetIdx representing the argument that may need to be incremented
+    // later to load additional chunks of data.
+    // Only structured buffers can use the offset parameter.
+    // Others must increment the index.
+    if (RK == DxilResource::Kind::StructuredBuffer)
+      OffsetIdx = storeArgs.size();
+    else
+      OffsetIdx = storeArgs.size() - 1;
 
-    // Offset 1
-    storeArgs.emplace_back(undefI);
+    // Coord1 (Offset).
+    storeArgs.emplace_back(offset);
   } else {
     // texture store
     unsigned coordSize = DxilResource::GetNumCoords(RK);
 
     // Set x first.
-    if (offset->getType()->isVectorTy())
-      storeArgs.emplace_back(Builder.CreateExtractElement(offset, (uint64_t)0));
+    if (Idx->getType()->isVectorTy())
+      storeArgs.emplace_back(Builder.CreateExtractElement(Idx, (uint64_t)0));
     else
-      storeArgs.emplace_back(offset);
-
-    // Store offset0 for later use
-    offset0Idx = storeArgs.size() - 1;
+      storeArgs.emplace_back(Idx);
 
     for (unsigned i = 1; i < 3; i++) {
       if (i < coordSize)
-        storeArgs.emplace_back(Builder.CreateExtractElement(offset, i));
+        storeArgs.emplace_back(Builder.CreateExtractElement(Idx, i));
       else
         storeArgs.emplace_back(undefI);
     }
     // TODO: support mip for texture ST
   }
 
+  // RawBufferVectorStore only takes a single value and alignment arguments.
+  if (opcode == DXIL::OpCode::RawBufferVectorStore) {
+    storeArgs.emplace_back(val);
+    storeArgs.emplace_back(Alignment);
+    Function *F = OP->GetOpFunc(DXIL::OpCode::RawBufferVectorStore, Ty);
+    Builder.CreateCall(F, storeArgs);
+    return;
+  }
+  Function *F = OP->GetOpFunc(opcode, EltTy);
+
   constexpr unsigned MaxStoreElemCount = 4;
   const unsigned CompCount = Ty->isVectorTy() ? Ty->getVectorNumElements() : 1;
   const unsigned StoreInstCount =
@@ -4474,23 +4607,17 @@ void TranslateStore(DxilResource::Kind RK, Value *handle, Value *val,
   }
 
   for (unsigned j = 0; j < storeArgsList.size(); j++) {
-
-    // For second and subsequent store calls, increment the offset0 (i.e. store
-    // index)
+    // For second and subsequent store calls, increment the resource-appropriate
+    // index or offset parameter.
     if (j > 0) {
-      // Greater than four-components store is not allowed for
-      // TypedBuffer and Textures. So greater than four elements
-      // scenario should only get hit here for RawBuffer.
-      DXASSERT_NOMSG(RK == DxilResource::Kind::RawBuffer);
       unsigned EltSize = OP->GetAllocSizeForType(EltTy);
-      unsigned newOffset = EltSize * MaxStoreElemCount * j;
-      Value *newOffsetVal = ConstantInt::get(Builder.getInt32Ty(), newOffset);
-      newOffsetVal =
-          Builder.CreateAdd(storeArgsList[0][offset0Idx], newOffsetVal);
-      storeArgsList[j][offset0Idx] = newOffsetVal;
+      unsigned NewCoord = EltSize * MaxStoreElemCount * j;
+      Value *NewCoordVal = ConstantInt::get(Builder.getInt32Ty(), NewCoord);
+      NewCoordVal = Builder.CreateAdd(storeArgsList[0][OffsetIdx], NewCoordVal);
+      storeArgsList[j][OffsetIdx] = NewCoordVal;
     }
 
-    // values
+    // Set value parameters.
     uint8_t mask = 0;
     if (Ty->isVectorTy()) {
       unsigned vecSize =
@@ -4586,7 +4713,8 @@ Value *TranslateResourceStore(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
 
   Value *val = CI->getArgOperand(HLOperandIndex::kStoreValOpIdx);
   Value *offset = CI->getArgOperand(HLOperandIndex::kStoreOffsetOpIdx);
-  TranslateStore(RK, handle, val, offset, Builder, hlslOP);
+  Value *UndefI = UndefValue::get(Builder.getInt32Ty());
+  TranslateStore(RK, handle, val, offset, UndefI, Builder, hlslOP);
 
   return nullptr;
 }
@@ -6091,20 +6219,8 @@ Value *TranslateAnd(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
                     bool &Translated) {
   Value *x = CI->getArgOperand(HLOperandIndex::kBinaryOpSrc0Idx);
   Value *y = CI->getArgOperand(HLOperandIndex::kBinaryOpSrc1Idx);
-  Type *Ty = CI->getType();
-  Type *EltTy = Ty->getScalarType();
   IRBuilder<> Builder(CI);
 
-  if (Ty != EltTy) {
-    Value *Result = UndefValue::get(Ty);
-    for (unsigned i = 0; i < Ty->getVectorNumElements(); i++) {
-      Value *EltX = Builder.CreateExtractElement(x, i);
-      Value *EltY = Builder.CreateExtractElement(y, i);
-      Value *tmp = Builder.CreateAnd(EltX, EltY);
-      Result = Builder.CreateInsertElement(Result, tmp, i);
-    }
-    return Result;
-  }
   return Builder.CreateAnd(x, y);
 }
 Value *TranslateOr(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
@@ -6112,20 +6228,8 @@ Value *TranslateOr(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
                    HLObjectOperationLowerHelper *pObjHelper, bool &Translated) {
   Value *x = CI->getArgOperand(HLOperandIndex::kBinaryOpSrc0Idx);
   Value *y = CI->getArgOperand(HLOperandIndex::kBinaryOpSrc1Idx);
-  Type *Ty = CI->getType();
-  Type *EltTy = Ty->getScalarType();
   IRBuilder<> Builder(CI);
 
-  if (Ty != EltTy) {
-    Value *Result = UndefValue::get(Ty);
-    for (unsigned i = 0; i < Ty->getVectorNumElements(); i++) {
-      Value *EltX = Builder.CreateExtractElement(x, i);
-      Value *EltY = Builder.CreateExtractElement(y, i);
-      Value *tmp = Builder.CreateOr(EltX, EltY);
-      Result = Builder.CreateInsertElement(Result, tmp, i);
-    }
-    return Result;
-  }
   return Builder.CreateOr(x, y);
 }
 Value *TranslateSelect(CallInst *CI, IntrinsicOp IOP, OP::OpCode opcode,
@@ -6166,7 +6270,6 @@ Value *EmptyLower(CallInst *CI, IntrinsicOp IOP, DXIL::OpCode opcode,
 }
 
 // SPIRV change starts
-#ifdef ENABLE_SPIRV_CODEGEN
 Value *UnsupportedVulkanIntrinsic(CallInst *CI, IntrinsicOp IOP,
                                   DXIL::OpCode opcode,
                                   HLOperationLowerHelper &helper,
@@ -6176,7 +6279,6 @@ Value *UnsupportedVulkanIntrinsic(CallInst *CI, IntrinsicOp IOP,
   dxilutil::EmitErrorOnInstruction(CI, "Unsupported Vulkan intrinsic.");
   return nullptr;
 }
-#endif // ENABLE_SPIRV_CODEGEN
 // SPIRV change ends
 
 Value *StreamOutputLower(CallInst *CI, IntrinsicOp IOP, DXIL::OpCode opcode,
@@ -6403,7 +6505,7 @@ IntrinsicLower gLowerTable[] = {
     {IntrinsicOp::IOP_asint16, TranslateBitcast, DXIL::OpCode::NumOpCodes},
     {IntrinsicOp::IOP_asuint, TranslateAsUint, DXIL::OpCode::SplitDouble},
     {IntrinsicOp::IOP_asuint16, TranslateAsUint, DXIL::OpCode::NumOpCodes},
-    {IntrinsicOp::IOP_atan, TrivialUnaryOperation, DXIL::OpCode::Atan},
+    {IntrinsicOp::IOP_atan, TrivialVectorizableUnaryOperation, DXIL::OpCode::Atan},
     {IntrinsicOp::IOP_atan2, TranslateAtan2, DXIL::OpCode::NumOpCodes},
     {IntrinsicOp::IOP_ceil, TrivialUnaryOperation, DXIL::OpCode::Round_pi},
     {IntrinsicOp::IOP_clamp, TranslateClamp, DXIL::OpCode::NumOpCodes},
@@ -6446,7 +6548,7 @@ IntrinsicLower gLowerTable[] = {
     {IntrinsicOp::IOP_firstbitlow, TranslateFirstbitLo,
      DXIL::OpCode::FirstbitLo},
     {IntrinsicOp::IOP_floor, TrivialUnaryOperation, DXIL::OpCode::Round_ni},
-    {IntrinsicOp::IOP_fma, TrivialTrinaryOperation, DXIL::OpCode::Fma},
+    {IntrinsicOp::IOP_fma, TranslateFMA, DXIL::OpCode::Fma},
     {IntrinsicOp::IOP_fmod, TranslateFMod, DXIL::OpCode::NumOpCodes},
     {IntrinsicOp::IOP_frac, TrivialUnaryOperation, DXIL::OpCode::Frc},
     {IntrinsicOp::IOP_frexp, TranslateFrexp, DXIL::OpCode::NumOpCodes},
@@ -6494,7 +6596,7 @@ IntrinsicLower gLowerTable[] = {
     {IntrinsicOp::IOP_sqrt, TrivialUnaryOperation, DXIL::OpCode::Sqrt},
     {IntrinsicOp::IOP_step, TranslateStep, DXIL::OpCode::NumOpCodes},
     {IntrinsicOp::IOP_tan, TrivialUnaryOperation, DXIL::OpCode::Tan},
-    {IntrinsicOp::IOP_tanh, TrivialUnaryOperation, DXIL::OpCode::Htan},
+    {IntrinsicOp::IOP_tanh, TrivialVectorizableUnaryOperation, DXIL::OpCode::Htan},
     {IntrinsicOp::IOP_tex1D, EmptyLower, DXIL::OpCode::NumOpCodes},
     {IntrinsicOp::IOP_tex1Dbias, EmptyLower, DXIL::OpCode::NumOpCodes},
     {IntrinsicOp::IOP_tex1Dgrad, EmptyLower, DXIL::OpCode::NumOpCodes},
@@ -6521,7 +6623,6 @@ IntrinsicLower gLowerTable[] = {
     {IntrinsicOp::IOP_unpack_s8s32, TranslateUnpack, DXIL::OpCode::Unpack4x8},
     {IntrinsicOp::IOP_unpack_u8u16, TranslateUnpack, DXIL::OpCode::Unpack4x8},
     {IntrinsicOp::IOP_unpack_u8u32, TranslateUnpack, DXIL::OpCode::Unpack4x8},
-#ifdef ENABLE_SPIRV_CODEGEN
     {IntrinsicOp::IOP_VkRawBufferLoad, UnsupportedVulkanIntrinsic,
      DXIL::OpCode::NumOpCodes},
     {IntrinsicOp::IOP_VkRawBufferStore, UnsupportedVulkanIntrinsic,
@@ -6532,7 +6633,6 @@ IntrinsicLower gLowerTable[] = {
      DXIL::OpCode::NumOpCodes},
     {IntrinsicOp::IOP_Vkext_execution_mode_id, UnsupportedVulkanIntrinsic,
      DXIL::OpCode::NumOpCodes},
-#endif // ENABLE_SPIRV_CODEGEN
     {IntrinsicOp::MOP_Append, StreamOutputLower, DXIL::OpCode::EmitStream},
     {IntrinsicOp::MOP_RestartStrip, StreamOutputLower, DXIL::OpCode::CutStream},
     {IntrinsicOp::MOP_CalculateLevelOfDetail, TranslateCalculateLOD,
@@ -6760,11 +6860,9 @@ IntrinsicLower gLowerTable[] = {
     {IntrinsicOp::MOP_OutputComplete, TranslateNodeOutputComplete,
      DXIL::OpCode::OutputComplete},
 
-// SPIRV change starts
-#ifdef ENABLE_SPIRV_CODEGEN
+    // SPIRV change starts
     {IntrinsicOp::MOP_SubpassLoad, UnsupportedVulkanIntrinsic,
      DXIL::OpCode::NumOpCodes},
-#endif // ENABLE_SPIRV_CODEGEN
     // SPIRV change ends
 
     // Manually added part.
@@ -7830,6 +7928,62 @@ void TranslateCBOperationsLegacy(Value *handle, Value *ptr, OP *hlslOP,
 // Structured buffer.
 namespace {
 
+// Returns load call return value
+//  and the elements extracted from it in `Elts`
+static Value *GenerateBufLd(hlsl::OP *OP, IRBuilder<> &Builder,
+                            OP::OpCode opcode, Type *RegTy, Type *MemTy,
+                            unsigned NumElements, ArrayRef<Value *> Args,
+                            std::vector<Value *>::iterator EltIt) {
+  const bool isTyped =
+      (opcode == OP::OpCode::BufferLoad || opcode == OP::OpCode::TextureLoad);
+  // Value will be loaded in its memory representation.
+  Function *F = OP->GetOpFunc(opcode, MemTy);
+  Value *Ld = Builder.CreateCall(F, Args, OP::GetOpCodeName(opcode));
+
+  // Extract values and convet to register type if needed.
+  // 64-bit types need to have the two i32 merged into their result
+  // bools need to use cmp to convert them to i1s.
+
+  if (isTyped && RegTy->isDoubleTy()) {
+    DXASSERT(NumElements <= 2, "typed buffers only allow 4 dwords");
+    Type *doubleTy = Builder.getDoubleTy();
+    Function *makeDouble = OP->GetOpFunc(DXIL::OpCode::MakeDouble, doubleTy);
+    Value *makeDoubleOpArg =
+        Builder.getInt32((unsigned)DXIL::OpCode::MakeDouble);
+    for (unsigned i = 0; i < NumElements; i++, EltIt++) {
+      Value *lo = Builder.CreateExtractValue(Ld, 2 * i);
+      Value *hi = Builder.CreateExtractValue(Ld, 2 * i + 1);
+      Value *V = Builder.CreateCall(makeDouble, {makeDoubleOpArg, lo, hi});
+      *EltIt = V;
+    }
+    return Ld;
+  }
+
+  if (isTyped && RegTy->isIntegerTy(64)) {
+    DXASSERT(NumElements <= 2, "typed buffers only allow 4 dwords");
+    Type *i64Ty = Builder.getInt64Ty();
+    for (unsigned i = 0; i < NumElements; i++, EltIt++) {
+      Value *lo = Builder.CreateExtractValue(Ld, 2 * i);
+      Value *hi = Builder.CreateExtractValue(Ld, 2 * i + 1);
+      lo = Builder.CreateZExt(lo, i64Ty);
+      hi = Builder.CreateZExt(hi, i64Ty);
+      hi = Builder.CreateShl(hi, 32);
+      *EltIt = Builder.CreateOr(lo, hi);
+    }
+    return Ld;
+  }
+
+  if (opcode == OP::OpCode::RawBufferVectorLoad) {
+    DXASSERT(RegTy != Builder.getInt64Ty() && RegTy != Builder.getDoubleTy(),
+             "64-bit type conversions for longvecs requires op support");
+    *EltIt = Builder.CreateExtractValue(Ld, 0);
+  } else
+    for (unsigned i = 0; i < NumElements; i++, EltIt++)
+      *EltIt = Builder.CreateExtractValue(Ld, i);
+
+  return Ld;
+}
+
 Value *GenerateRawBufLd(Value *handle, Value *bufIdx, Value *offset,
                         Value *status, Type *EltTy,
                         MutableArrayRef<Value *> resultElts, hlsl::OP *OP,
@@ -7887,113 +8041,38 @@ void GenerateStructBufSt(Value *handle, Value *bufIdx, Value *offset,
   Builder.CreateCall(dxilF, Args);
 }
 
-static Value *TranslateRawBufVecLd(Type *VecEltTy, unsigned ElemCount,
-                                   IRBuilder<> &Builder, Value *handle,
-                                   hlsl::OP *OP, Value *status, Value *bufIdx,
-                                   Value *baseOffset, const DataLayout &DL,
-                                   std::vector<Value *> &bufLds,
-                                   unsigned baseAlign, bool isScalarTy) {
-
-  unsigned EltSize = DL.getTypeAllocSize(VecEltTy);
-  unsigned alignment = std::min(baseAlign, EltSize);
-  Constant *alignmentVal = OP->GetI32Const(alignment);
-
-  if (baseOffset == nullptr) {
-    baseOffset = OP->GetU32Const(0);
-  }
-
-  std::vector<Value *> elts(ElemCount);
-  unsigned rest = (ElemCount % 4);
-  for (unsigned i = 0; i < ElemCount - rest; i += 4) {
-    Value *ResultElts[4];
-    Value *bufLd =
-        GenerateRawBufLd(handle, bufIdx, baseOffset, status, VecEltTy,
-                         ResultElts, OP, Builder, 4, alignmentVal);
-    bufLds.emplace_back(bufLd);
-    elts[i] = ResultElts[0];
-    elts[i + 1] = ResultElts[1];
-    elts[i + 2] = ResultElts[2];
-    elts[i + 3] = ResultElts[3];
-
-    baseOffset = Builder.CreateAdd(baseOffset, OP->GetU32Const(4 * EltSize));
-  }
-
-  if (rest) {
-    Value *ResultElts[4];
-    Value *bufLd =
-        GenerateRawBufLd(handle, bufIdx, baseOffset, status, VecEltTy,
-                         ResultElts, OP, Builder, rest, alignmentVal);
-    bufLds.emplace_back(bufLd);
-    for (unsigned i = 0; i < rest; i++)
-      elts[ElemCount - rest + i] = ResultElts[i];
-  }
-
-  // If the expected return type is scalar then skip building a vector
-  if (isScalarTy) {
-    return elts[0];
-  }
-
-  Value *Vec = HLMatrixLower::BuildVector(VecEltTy, elts, Builder);
-  return Vec;
-}
-
-Value *TranslateStructBufMatLd(Type *matType, IRBuilder<> &Builder,
-                               Value *handle, hlsl::OP *OP, Value *status,
-                               Value *bufIdx, Value *baseOffset,
+Value *TranslateStructBufMatLd(CallInst *CI, IRBuilder<> &Builder,
+                               Value *handle, HLResource::Kind RK, hlsl::OP *OP,
+                               Value *status, Value *bufIdx, Value *baseOffset,
                                const DataLayout &DL) {
+
+  ResLoadHelper helper(CI, RK, handle, bufIdx, baseOffset);
+#ifndef NDEBUG
+  Value *ptr = CI->getArgOperand(HLOperandIndex::kMatLoadPtrOpIdx);
+  Type *matType = ptr->getType()->getPointerElementType();
   HLMatrixType MatTy = HLMatrixType::cast(matType);
-  Type *EltTy = MatTy.getElementTypeForMem();
-  unsigned matSize = MatTy.getNumElements();
-  std::vector<Value *> bufLds;
-  Value *Vec =
-      TranslateRawBufVecLd(EltTy, matSize, Builder, handle, OP, status, bufIdx,
-                           baseOffset, DL, bufLds, /*baseAlign (in bytes)*/ 8);
-  Vec = MatTy.emitLoweredMemToReg(Vec, Builder);
-  return Vec;
+  DXASSERT(MatTy.getLoweredVectorType(false /*MemRepr*/) ==
+               helper.retVal->getType(),
+           "helper type should match vectorized matrix");
+#endif
+  return TranslateBufLoad(helper, RK, Builder, OP, DL);
 }
 
 void TranslateStructBufMatSt(Type *matType, IRBuilder<> &Builder, Value *handle,
                              hlsl::OP *OP, Value *bufIdx, Value *baseOffset,
                              Value *val, const DataLayout &DL) {
+#ifndef NDEBUG
   HLMatrixType MatTy = HLMatrixType::cast(matType);
-  Type *EltTy = MatTy.getElementTypeForMem();
-
-  val = MatTy.emitLoweredRegToMem(val, Builder);
-
-  unsigned EltSize = DL.getTypeAllocSize(EltTy);
-  Constant *Alignment = OP->GetI32Const(EltSize);
-  Value *offset = baseOffset;
-  if (baseOffset == nullptr)
-    offset = OP->GetU32Const(0);
-
-  unsigned matSize = MatTy.getNumElements();
-  Value *undefElt = UndefValue::get(EltTy);
-
-  unsigned storeSize = matSize;
-  if (matSize % 4) {
-    storeSize = matSize + 4 - (matSize & 3);
-  }
-  std::vector<Value *> elts(storeSize, undefElt);
-  for (unsigned i = 0; i < matSize; i++)
-    elts[i] = Builder.CreateExtractElement(val, i);
-
-  for (unsigned i = 0; i < matSize; i += 4) {
-    uint8_t mask = 0;
-    for (unsigned j = 0; j < 4 && (i + j) < matSize; j++) {
-      if (elts[i + j] != undefElt)
-        mask |= (1 << j);
-    }
-    GenerateStructBufSt(handle, bufIdx, offset, EltTy, OP, Builder,
-                        {elts[i], elts[i + 1], elts[i + 2], elts[i + 3]}, mask,
-                        Alignment);
-    // Update offset by 4*4bytes.
-    offset = Builder.CreateAdd(offset, OP->GetU32Const(4 * EltSize));
-  }
+  DXASSERT(MatTy.getLoweredVectorType(false /*MemRepr*/) == val->getType(),
+           "helper type should match vectorized matrix");
+#endif
+  TranslateStore(DxilResource::Kind::StructuredBuffer, handle, val, bufIdx,
+                 baseOffset, Builder, OP);
 }
 
-void TranslateStructBufMatLdSt(CallInst *CI, Value *handle, hlsl::OP *OP,
-                               Value *status, Value *bufIdx, Value *baseOffset,
-                               const DataLayout &DL) {
+void TranslateStructBufMatLdSt(CallInst *CI, Value *handle, HLResource::Kind RK,
+                               hlsl::OP *OP, Value *status, Value *bufIdx,
+                               Value *baseOffset, const DataLayout &DL) {
   IRBuilder<> Builder(CI);
   HLOpcodeGroup group = hlsl::GetHLOpcodeGroupByName(CI->getCalledFunction());
   unsigned opcode = GetHLOpcode(CI);
@@ -8006,13 +8085,10 @@ void TranslateStructBufMatLdSt(CallInst *CI, Value *handle, hlsl::OP *OP,
   // orientation.
   switch (matOp) {
   case HLMatLoadStoreOpcode::RowMatLoad:
-  case HLMatLoadStoreOpcode::ColMatLoad: {
-    Value *ptr = CI->getArgOperand(HLOperandIndex::kMatLoadPtrOpIdx);
-    Value *NewLd = TranslateStructBufMatLd(
-        ptr->getType()->getPointerElementType(), Builder, handle, OP, status,
-        bufIdx, baseOffset, DL);
-    CI->replaceAllUsesWith(NewLd);
-  } break;
+  case HLMatLoadStoreOpcode::ColMatLoad:
+    TranslateStructBufMatLd(CI, Builder, handle, RK, OP, status, bufIdx,
+                            baseOffset, DL);
+    break;
   case HLMatLoadStoreOpcode::RowMatStore:
   case HLMatLoadStoreOpcode::ColMatStore: {
     Value *ptr = CI->getArgOperand(HLOperandIndex::kMatStoreDstPtrOpIdx);
@@ -8136,6 +8212,9 @@ void TranslateStructBufMatSubscript(CallInst *CI, Value *handle,
 
       GEP->eraseFromParent();
     } else if (StoreInst *stUser = dyn_cast<StoreInst>(subsUser)) {
+      // Store elements of matrix in a struct. Needs to be done one scalar at a
+      // time even for vectors in the case that matrix orientation spreads the
+      // indexed scalars throughout the matrix vector.
       IRBuilder<> stBuilder(stUser);
       Value *Val = stUser->getValueOperand();
       if (Val->getType()->isVectorTy()) {
@@ -8159,6 +8238,9 @@ void TranslateStructBufMatSubscript(CallInst *CI, Value *handle,
       LoadInst *ldUser = cast<LoadInst>(subsUser);
       IRBuilder<> ldBuilder(ldUser);
       Value *ldData = UndefValue::get(resultType);
+      // Load elements of matrix in a struct. Needs to be done one scalar at a
+      // time even for vectors in the case that matrix orientation spreads the
+      // indexed scalars throughout the matrix vector.
       if (resultType->isVectorTy()) {
         for (unsigned i = 0; i < resultSize; i++) {
           Value *ResultElt;
@@ -8283,57 +8365,26 @@ void TranslateStructBufSubscriptUser(Instruction *user, Value *handle,
       }
       userCall->eraseFromParent();
     } else if (group == HLOpcodeGroup::HLMatLoadStore)
-      TranslateStructBufMatLdSt(userCall, handle, OP, status, bufIdx,
+      // Load/Store matrix within a struct
+      TranslateStructBufMatLdSt(userCall, handle, ResKind, OP, status, bufIdx,
                                 baseOffset, DL);
     else if (group == HLOpcodeGroup::HLSubscript) {
+      // Subscript of matrix within a struct
       TranslateStructBufMatSubscript(userCall, handle, ResKind, bufIdx,
                                      baseOffset, status, OP, DL);
     }
-  } else if (isa<LoadInst>(user) || isa<StoreInst>(user)) {
-    LoadInst *LdInst = dyn_cast<LoadInst>(user);
-    StoreInst *StInst = dyn_cast<StoreInst>(user);
-
-    Type *Ty = isa<LoadInst>(user) ? LdInst->getType()
-                                   : StInst->getValueOperand()->getType();
-    Type *pOverloadTy = Ty->getScalarType();
-    Value *Offset = baseOffset;
-
-    if (LdInst) {
-      unsigned NumComponents = 0;
-      if (VectorType *VTy = dyn_cast<VectorType>(Ty))
-        NumComponents = VTy->getNumElements();
-      else
-        NumComponents = 1;
-      Value *ResultElts[4];
-      Constant *Alignment =
-          OP->GetI32Const(DL.getTypeAllocSize(Ty->getScalarType()));
-      GenerateRawBufLd(handle, bufIdx, Offset, status, pOverloadTy, ResultElts,
-                       OP, Builder, NumComponents, Alignment);
-      Value *NewLd = ScalarizeElements(Ty, ResultElts, Builder);
-      LdInst->replaceAllUsesWith(NewLd);
-    } else {
-      Value *val = StInst->getValueOperand();
-      Value *undefVal = llvm::UndefValue::get(pOverloadTy);
-      Value *vals[] = {undefVal, undefVal, undefVal, undefVal};
-      uint8_t mask = 0;
-      if (Ty->isVectorTy()) {
-        unsigned vectorNumElements = Ty->getVectorNumElements();
-        DXASSERT(vectorNumElements <= 4, "up to 4 elements in vector");
-        assert(vectorNumElements <= 4);
-        for (unsigned i = 0; i < vectorNumElements; i++) {
-          vals[i] = Builder.CreateExtractElement(val, i);
-          mask |= (1 << i);
-        }
-      } else {
-        vals[0] = val;
-        mask = DXIL::kCompMask_X;
-      }
-      Constant *alignment =
-          OP->GetI32Const(DL.getTypeAllocSize(Ty->getScalarType()));
-      GenerateStructBufSt(handle, bufIdx, Offset, pOverloadTy, OP, Builder,
-                          vals, mask, alignment);
-    }
-    user->eraseFromParent();
+  } else if (LoadInst *LdInst = dyn_cast<LoadInst>(user)) {
+    // Load of scalar/vector within a struct or structured raw load.
+    ResLoadHelper helper(LdInst, ResKind, handle, bufIdx, baseOffset);
+    TranslateBufLoad(helper, ResKind, Builder, OP, DL);
+
+    LdInst->eraseFromParent();
+  } else if (StoreInst *StInst = dyn_cast<StoreInst>(user)) {
+    // Store of scalar/vector within a struct or structured raw store.
+    Value *val = StInst->getValueOperand();
+    TranslateStore(DxilResource::Kind::StructuredBuffer, handle, val, bufIdx,
+                   baseOffset, Builder, OP);
+    StInst->eraseFromParent();
   } else if (BitCastInst *BCI = dyn_cast<BitCastInst>(user)) {
     // Recurse users
     for (auto U = BCI->user_begin(); U != BCI->user_end();) {
@@ -8368,13 +8419,18 @@ void TranslateStructBufSubscriptUser(Instruction *user, Value *handle,
     DXASSERT_LOCALVAR(Ty,
                       offset->getType() == Type::getInt32Ty(Ty->getContext()),
                       "else bitness is wrong");
-    offset = Builder.CreateAdd(offset, baseOffset);
+    // Raw buffers can't have defined offsets, apply to index.
+    if (DXIL::IsRawBuffer(ResKind))
+      bufIdx = Builder.CreateAdd(offset, bufIdx);
+    else
+      baseOffset = Builder.CreateAdd(offset, baseOffset);
 
     for (auto U = GEP->user_begin(); U != GEP->user_end();) {
       Value *GEPUser = *(U++);
 
       TranslateStructBufSubscriptUser(cast<Instruction>(GEPUser), handle,
-                                      ResKind, bufIdx, offset, status, OP, DL);
+                                      ResKind, bufIdx, baseOffset, status, OP,
+                                      DL);
     }
     // delete the inst
     GEP->eraseFromParent();
@@ -8388,13 +8444,12 @@ void TranslateStructBufSubscript(CallInst *CI, Value *handle, Value *status,
       CI->getArgOperand(HLOperandIndex::kSubscriptIndexOpIdx);
   Value *bufIdx = nullptr;
   Value *offset = nullptr;
-  if (ResKind == HLResource::Kind::RawBuffer) {
-    offset = subscriptIndex;
-  } else {
+  bufIdx = subscriptIndex;
+  if (ResKind == HLResource::Kind::RawBuffer)
+    offset = UndefValue::get(Type::getInt32Ty(CI->getContext()));
+  else
     // StructuredBuffer, TypedBuffer, etc.
-    bufIdx = subscriptIndex;
     offset = OP->GetU32Const(0);
-  }
 
   for (auto U = CI->user_begin(); U != CI->user_end();) {
     Value *user = *(U++);
@@ -8408,19 +8463,14 @@ void TranslateStructBufSubscript(CallInst *CI, Value *handle, Value *status,
 // HLSubscript.
 namespace {
 
-Value *TranslateTypedBufLoad(CallInst *CI, DXIL::ResourceKind RK,
+Value *TranslateTypedBufSubscript(CallInst *CI, DXIL::ResourceKind RK,
                              DXIL::ResourceClass RC, Value *handle,
                              LoadInst *ldInst, IRBuilder<> &Builder,
                              hlsl::OP *hlslOP, const DataLayout &DL) {
-  ResLoadHelper ldHelper(CI, RK, RC, handle, IntrinsicOp::MOP_Load,
-                         /*bForSubscript*/ true);
-  // Default sampleIdx for 2DMS textures.
-  if (RK == DxilResource::Kind::Texture2DMS ||
-      RK == DxilResource::Kind::Texture2DMSArray)
-    ldHelper.mipLevel = hlslOP->GetU32Const(0);
-  // use ldInst as retVal
-  ldHelper.retVal = ldInst;
-  TranslateLoad(ldHelper, RK, Builder, hlslOP, DL);
+  // The arguments to the call instruction are used to determine the access,
+  // the return value and type come from the load instruction.
+  ResLoadHelper ldHelper(CI, RK, RC, handle, IntrinsicOp::MOP_Load, ldInst);
+  TranslateBufLoad(ldHelper, RK, Builder, hlslOP, DL);
   // delete the ld
   ldInst->eraseFromParent();
   return ldHelper.retVal;
@@ -8463,7 +8513,7 @@ Value *UpdateVectorElt(Value *VecVal, Value *EltVal, Value *EltIdx,
   return VecVal;
 }
 
-void TranslateDefaultSubscript(CallInst *CI, HLOperationLowerHelper &helper,
+void TranslateTypedBufferSubscript(CallInst *CI, HLOperationLowerHelper &helper,
                                HLObjectOperationLowerHelper *pObjHelper,
                                bool &Translated) {
   Value *ptr = CI->getArgOperand(HLOperandIndex::kSubscriptObjectOpIdx);
@@ -8480,14 +8530,15 @@ void TranslateDefaultSubscript(CallInst *CI, HLOperationLowerHelper &helper,
     User *user = *(It++);
     Instruction *I = cast<Instruction>(user);
     IRBuilder<> Builder(I);
+    Value *UndefI = UndefValue::get(Builder.getInt32Ty());
     if (LoadInst *ldInst = dyn_cast<LoadInst>(user)) {
-      TranslateTypedBufLoad(CI, RK, RC, handle, ldInst, Builder, hlslOP,
+      TranslateTypedBufSubscript(CI, RK, RC, handle, ldInst, Builder, hlslOP,
                             helper.dataLayout);
     } else if (StoreInst *stInst = dyn_cast<StoreInst>(user)) {
       Value *val = stInst->getValueOperand();
       TranslateStore(RK, handle, val,
-                     CI->getArgOperand(HLOperandIndex::kStoreOffsetOpIdx),
-                     Builder, hlslOP);
+                     CI->getArgOperand(HLOperandIndex::kSubscriptIndexOpIdx),
+                     UndefI, Builder, hlslOP);
       // delete the st
       stInst->eraseFromParent();
     } else if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(user)) {
@@ -8504,7 +8555,7 @@ void TranslateDefaultSubscript(CallInst *CI, HLOperationLowerHelper &helper,
           // Generate Ld.
           LoadInst *tmpLd = StBuilder.CreateLoad(CI);
 
-          Value *ldVal = TranslateTypedBufLoad(
+          Value *ldVal = TranslateTypedBufSubscript(
               CI, RK, RC, handle, tmpLd, StBuilder, hlslOP, helper.dataLayout);
           // Update vector.
           ldVal = UpdateVectorElt(ldVal, SI->getValueOperand(), EltIdx,
@@ -8512,9 +8563,10 @@ void TranslateDefaultSubscript(CallInst *CI, HLOperationLowerHelper &helper,
           // Generate St.
           // Reset insert point, UpdateVectorElt may move SI to different block.
           StBuilder.SetInsertPoint(SI);
-          TranslateStore(RK, handle, ldVal,
-                         CI->getArgOperand(HLOperandIndex::kStoreOffsetOpIdx),
-                         StBuilder, hlslOP);
+          TranslateStore(
+              RK, handle, ldVal,
+              CI->getArgOperand(HLOperandIndex::kSubscriptIndexOpIdx), UndefI,
+              StBuilder, hlslOP);
           SI->eraseFromParent();
           continue;
         }
@@ -8524,7 +8576,7 @@ void TranslateDefaultSubscript(CallInst *CI, HLOperationLowerHelper &helper,
           // Generate tmp vector load with vector type & translate it
           LoadInst *tmpLd = LdBuilder.CreateLoad(CI);
 
-          Value *ldVal = TranslateTypedBufLoad(
+          Value *ldVal = TranslateTypedBufSubscript(
               CI, RK, RC, handle, tmpLd, LdBuilder, hlslOP, helper.dataLayout);
 
           // get the single element
@@ -8697,15 +8749,17 @@ void TranslateHLSubscript(CallInst *CI, HLSubscriptOpcode opcode,
     DXASSERT(CI->hasOneUse(), "subscript should only have one use");
     IRBuilder<> Builder(CI);
     if (LoadInst *ldInst = dyn_cast<LoadInst>(*U)) {
-      ResLoadHelper ldHelper(ldInst, handle, coord, mipLevel);
-      TranslateLoad(ldHelper, RK, Builder, hlslOP, helper.dataLayout);
+      Value *Offset = UndefValue::get(Builder.getInt32Ty());
+      ResLoadHelper ldHelper(ldInst, RK, handle, coord, Offset, mipLevel);
+      TranslateBufLoad(ldHelper, RK, Builder, hlslOP, helper.dataLayout);
       ldInst->eraseFromParent();
     } else {
       StoreInst *stInst = cast<StoreInst>(*U);
       Value *val = stInst->getValueOperand();
+      Value *UndefI = UndefValue::get(Builder.getInt32Ty());
       TranslateStore(RK, handle, val,
-                     CI->getArgOperand(HLOperandIndex::kStoreOffsetOpIdx),
-                     Builder, hlslOP, mipLevel);
+                     CI->getArgOperand(HLOperandIndex::kSubscriptIndexOpIdx),
+                     UndefI, Builder, hlslOP, mipLevel);
       stInst->eraseFromParent();
     }
     Translated = true;
@@ -8736,7 +8790,7 @@ void TranslateHLSubscript(CallInst *CI, HLSubscriptOpcode opcode,
         TranslateStructBufSubscript(CI, handle, /*status*/ nullptr, hlslOP, RK,
                                     helper.dataLayout);
       else
-        TranslateDefaultSubscript(CI, helper, pObjHelper, Translated);
+        TranslateTypedBufferSubscript(CI, helper, pObjHelper, Translated);
 
       return;
     }
diff --git a/lib/Transforms/Scalar/DxilEliminateVector.cpp b/lib/Transforms/Scalar/DxilEliminateVector.cpp
index 366f011dae..bb9cf43594 100644
--- a/lib/Transforms/Scalar/DxilEliminateVector.cpp
+++ b/lib/Transforms/Scalar/DxilEliminateVector.cpp
@@ -10,6 +10,8 @@
 //                                                                           //
 ///////////////////////////////////////////////////////////////////////////////
 
+#include "dxc/DXIL/DxilModule.h"
+
 #include "llvm/IR/Dominators.h"
 #include "llvm/IR/Instructions.h"
 #include "llvm/Pass.h"
@@ -151,6 +153,10 @@ bool DxilEliminateVector::TryRewriteDebugInfoForVector(InsertElementInst *IE) {
 
 bool DxilEliminateVector::runOnFunction(Function &F) {
 
+  if (F.getParent()->HasDxilModule())
+    if (F.getParent()->GetDxilModule().GetShaderModel()->IsSM69Plus())
+      return false;
+
   auto *DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree();
   DxilValueCache *DVC = &getAnalysis<DxilValueCache>();
 
diff --git a/lib/Transforms/Scalar/LowerTypePasses.cpp b/lib/Transforms/Scalar/LowerTypePasses.cpp
index feeb23a5da..6d6b93f951 100644
--- a/lib/Transforms/Scalar/LowerTypePasses.cpp
+++ b/lib/Transforms/Scalar/LowerTypePasses.cpp
@@ -8,6 +8,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "dxc/DXIL/DxilConstants.h"
+#include "dxc/DXIL/DxilModule.h"
 #include "dxc/DXIL/DxilOperations.h"
 #include "dxc/DXIL/DxilUtil.h"
 #include "dxc/HLSL/HLModule.h"
@@ -180,10 +181,12 @@ bool LowerTypePass::runOnModule(Module &M) {
 namespace {
 class DynamicIndexingVectorToArray : public LowerTypePass {
   bool ReplaceAllVectors;
+  bool SupportsVectors;
 
 public:
   explicit DynamicIndexingVectorToArray(bool ReplaceAll = false)
-      : LowerTypePass(ID), ReplaceAllVectors(ReplaceAll) {}
+      : LowerTypePass(ID), ReplaceAllVectors(ReplaceAll),
+        SupportsVectors(false) {}
   static char ID; // Pass identification, replacement for typeid
   void applyOptions(PassOptions O) override;
   void dumpConfig(raw_ostream &OS) override;
@@ -194,6 +197,7 @@ class DynamicIndexingVectorToArray : public LowerTypePass {
   Type *lowerType(Type *Ty) override;
   Constant *lowerInitVal(Constant *InitVal, Type *NewTy) override;
   StringRef getGlobalPrefix() override { return ".v"; }
+  void initialize(Module &M) override;
 
 private:
   bool HasVectorDynamicIndexing(Value *V);
@@ -207,6 +211,11 @@ class DynamicIndexingVectorToArray : public LowerTypePass {
   void ReplaceAddrSpaceCast(ConstantExpr *CE, Value *A, IRBuilder<> &Builder);
 };
 
+void DynamicIndexingVectorToArray::initialize(Module &M) {
+  if (M.HasHLModule())
+    SupportsVectors = M.GetHLModule().GetShaderModel()->IsSM69Plus();
+}
+
 void DynamicIndexingVectorToArray::applyOptions(PassOptions O) {
   GetPassOptionBool(O, "ReplaceAllVectors", &ReplaceAllVectors,
                     ReplaceAllVectors);
@@ -286,7 +295,7 @@ void DynamicIndexingVectorToArray::ReplaceStaticIndexingOnVector(Value *V) {
             StoreInst *stInst = cast<StoreInst>(GEPUser);
             Value *val = stInst->getValueOperand();
             Value *ldVal = Builder.CreateLoad(V);
-            ldVal = Builder.CreateInsertElement(ldVal, val, constIdx);
+            ldVal = Builder.CreateInsertElement(ldVal, val, constIdx); // UGH
             Builder.CreateStore(ldVal, V);
             stInst->eraseFromParent();
           }
@@ -306,8 +315,11 @@ void DynamicIndexingVectorToArray::ReplaceStaticIndexingOnVector(Value *V) {
 }
 
 bool DynamicIndexingVectorToArray::needToLower(Value *V) {
+  // Only needed where vectors aren't supported.
+  if (SupportsVectors)
+    return false;
   Type *Ty = V->getType()->getPointerElementType();
-  if (dyn_cast<VectorType>(Ty)) {
+  if (isa<VectorType>(Ty)) {
     if (isa<GlobalVariable>(V) || ReplaceAllVectors) {
       return true;
     }
diff --git a/lib/Transforms/Scalar/ScalarReplAggregatesHLSL.cpp b/lib/Transforms/Scalar/ScalarReplAggregatesHLSL.cpp
index 0c3e13f608..6737c9100e 100644
--- a/lib/Transforms/Scalar/ScalarReplAggregatesHLSL.cpp
+++ b/lib/Transforms/Scalar/ScalarReplAggregatesHLSL.cpp
@@ -1869,7 +1869,8 @@ bool SROAGlobalAndAllocas(HLModule &HLM, bool bHasDbgInfo) {
       // if
       // all its users can be transformed, then split up the aggregate into its
       // separate elements.
-      if (ShouldAttemptScalarRepl(AI) && isSafeAllocaToScalarRepl(AI)) {
+      if (!HLM.GetShaderModel()->IsSM69Plus() && ShouldAttemptScalarRepl(AI) &&
+          isSafeAllocaToScalarRepl(AI)) {
         std::vector<Value *> Elts;
         IRBuilder<> Builder(dxilutil::FindAllocaInsertionPt(AI));
         bool hasPrecise = HLModule::HasPreciseAttributeWithMetadata(AI);
@@ -1945,8 +1946,9 @@ bool SROAGlobalAndAllocas(HLModule &HLM, bool bHasDbgInfo) {
         continue;
       }
 
-      // Flat Global vector if no dynamic vector indexing.
-      bool bFlatVector = !hasDynamicVectorIndexing(GV);
+      // Flat Global vector if no dynamic vector indexing and pre-6.9.
+      bool bFlatVector =
+          !hasDynamicVectorIndexing(GV) && !HLM.GetShaderModel()->IsSM69Plus();
 
       if (bFlatVector) {
         GVDbgOffset &dbgOffset = GVDbgOffsetMap[GV];
@@ -1980,10 +1982,12 @@ bool SROAGlobalAndAllocas(HLModule &HLM, bool bHasDbgInfo) {
       } else {
         // SROA_Parameter_HLSL has no access to a domtree, if one is needed,
         // it'll be generated
-        SROAed = SROA_Helper::DoScalarReplacement(
-            GV, Elts, Builder, bFlatVector,
-            // TODO: set precise.
-            /*hasPrecise*/ false, typeSys, DL, DeadInsts, /*DT*/ nullptr);
+        if (!HLM.GetShaderModel()->IsSM69Plus()) {
+          SROAed = SROA_Helper::DoScalarReplacement(
+              GV, Elts, Builder, bFlatVector,
+              // TODO: set precise.
+              /*hasPrecise*/ false, typeSys, DL, DeadInsts, /*DT*/ nullptr);
+        }
       }
 
       if (SROAed) {
diff --git a/lib/Transforms/Scalar/Scalarizer.cpp b/lib/Transforms/Scalar/Scalarizer.cpp
index 729771c7c7..1b07d5f14f 100644
--- a/lib/Transforms/Scalar/Scalarizer.cpp
+++ b/lib/Transforms/Scalar/Scalarizer.cpp
@@ -14,6 +14,8 @@
 //
 //===----------------------------------------------------------------------===//
 
+#include "dxc/DXIL/DxilModule.h"
+
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/IR/IRBuilder.h"
 #include "llvm/IR/InstVisitor.h"
@@ -290,6 +292,10 @@ bool Scalarizer::doInitialization(Module &M) {
 }
 
 bool Scalarizer::runOnFunction(Function &F) {
+  if (F.getParent()->HasDxilModule())
+    if (F.getParent()->GetDxilModule().GetShaderModel()->IsSM69Plus())
+      return false;
+
   for (Function::iterator BBI = F.begin(), BBE = F.end(); BBI != BBE; ++BBI) {
     BasicBlock *BB = BBI;
     for (BasicBlock::iterator II = BB->begin(), IE = BB->end(); II != IE;) {
diff --git a/tools/clang/include/clang/AST/DeclCXX.h b/tools/clang/include/clang/AST/DeclCXX.h
index 3b07576545..36e0f99c82 100644
--- a/tools/clang/include/clang/AST/DeclCXX.h
+++ b/tools/clang/include/clang/AST/DeclCXX.h
@@ -465,6 +465,10 @@ class CXXRecordDecl : public RecordDecl {
     /// \brief Whether we are currently parsing base specifiers.
     bool IsParsingBaseSpecifiers : 1;
 
+    /// \brief Whether this class contains at least one member or base
+    ///  class containing an HLSL vector longer than 4 elements.
+    bool HasHLSLLongVector : 1;
+
     /// \brief The number of base class specifiers in Bases.
     unsigned NumBases;
 
@@ -1018,6 +1022,13 @@ class CXXRecordDecl : public RecordDecl {
     return data().NeedOverloadResolutionForDestructor;
   }
 
+  // HLSL Change add HLSL Long vector bit.
+  /// \brief Determine whether this class contains an HLSL long vector
+  /// of over 4 elements.
+  bool hasHLSLLongVector() { return data().HasHLSLLongVector; }
+  /// \brief Set that this class contains an HLSL long vector of over 4 elements
+  bool setHasHLSLLongVector() { return data().HasHLSLLongVector = true; }
+
   /// \brief Determine whether this class describes a lambda function object.
   bool isLambda() const {
     // An update record can't turn a non-lambda into a lambda.
diff --git a/tools/clang/include/clang/AST/HlslTypes.h b/tools/clang/include/clang/AST/HlslTypes.h
index 2aa9afa5f9..e6a50de8fb 100644
--- a/tools/clang/include/clang/AST/HlslTypes.h
+++ b/tools/clang/include/clang/AST/HlslTypes.h
@@ -348,9 +348,10 @@ void AddHLSLNodeOutputRecordTemplate(
     _Outptr_ clang::ClassTemplateDecl **outputRecordTemplateDecl,
     bool isCompleteType = true);
 
-clang::CXXRecordDecl *DeclareRecordTypeWithHandle(clang::ASTContext &context,
-                                                  llvm::StringRef name,
-                                                  bool isCompleteType = true);
+clang::CXXRecordDecl *
+DeclareRecordTypeWithHandle(clang::ASTContext &context, llvm::StringRef name,
+                            bool isCompleteType = true,
+                            clang::InheritableAttr *Attr = nullptr);
 
 void AddRaytracingConstants(clang::ASTContext &context);
 void AddSamplerFeedbackConstants(clang::ASTContext &context);
@@ -381,14 +382,14 @@ clang::CXXRecordDecl *DeclareTemplateTypeWithHandleInDeclContext(
 
 clang::CXXRecordDecl *DeclareUIntTemplatedTypeWithHandle(
     clang::ASTContext &context, llvm::StringRef typeName,
-    llvm::StringRef templateParamName,
-    clang::TagTypeKind tagKind = clang::TagTypeKind::TTK_Class);
+    llvm::StringRef templateParamName, clang::InheritableAttr *Attr = nullptr);
 clang::CXXRecordDecl *DeclareUIntTemplatedTypeWithHandleInDeclContext(
     clang::ASTContext &context, clang::DeclContext *declContext,
     llvm::StringRef typeName, llvm::StringRef templateParamName,
-    clang::TagTypeKind tagKind = clang::TagTypeKind::TTK_Class);
-clang::CXXRecordDecl *DeclareConstantBufferViewType(clang::ASTContext &context,
-                                                    bool bTBuf);
+    clang::InheritableAttr *Attr = nullptr);
+clang::CXXRecordDecl *
+DeclareConstantBufferViewType(clang::ASTContext &context,
+                              clang::InheritableAttr *Attr);
 clang::CXXRecordDecl *DeclareRayQueryType(clang::ASTContext &context);
 clang::CXXRecordDecl *DeclareResourceType(clang::ASTContext &context,
                                           bool bSampler);
diff --git a/tools/clang/include/clang/Basic/Attr.td b/tools/clang/include/clang/Basic/Attr.td
index 3a6718a339..9e48df51fd 100644
--- a/tools/clang/include/clang/Basic/Attr.td
+++ b/tools/clang/include/clang/Basic/Attr.td
@@ -939,6 +939,52 @@ def HLSLCXXOverload : InheritableAttr {
   let Documentation = [Undocumented];
 }
 
+def HLSLVector : InheritableAttr {
+  let Spellings = []; // No spellings!
+  let Subjects = SubjectList<[CXXRecord]>;
+  let Documentation = [Undocumented];
+}
+
+def HLSLMatrix : InheritableAttr {
+  let Spellings = []; // No spellings!
+  let Subjects = SubjectList<[CXXRecord]>;
+  let Documentation = [Undocumented];
+}
+
+def HLSLTessPatch : InheritableAttr {
+  let Spellings = []; // No spellings!
+  let Args = [BoolArgument<"IsInput">];
+  let Subjects = SubjectList<[CXXRecord]>;
+  let Documentation = [Undocumented];
+}
+
+def HLSLStreamOutput : InheritableAttr {
+  let Spellings = []; // No spellings!
+  // PrimVertices are the number of vertices that make up the streamed
+  // primitive. Points have 1. Lines have 2. Triangles have 3.
+  let Args = [UnsignedArgument<"PrimVertices">];
+  let Subjects = SubjectList<[CXXRecord]>;
+  let Documentation = [Undocumented];
+}
+
+def HLSLResource : InheritableAttr {
+  let Spellings = []; // No spellings!
+  let Args = [UnsignedArgument<"ResKindUint">,
+              UnsignedArgument<"ResClassUint">];
+  let Subjects = SubjectList<[CXXRecord]>;
+  let Documentation = [Undocumented];
+
+  // Add enum typed getters for safety and brevity.
+  let AdditionalMembers = [{
+  hlsl::DXIL::ResourceKind getResKind() const {
+    return (hlsl::DXIL::ResourceKind)getResKindUint();
+  }
+  hlsl::DXIL::ResourceClass getResClass() const {
+    return (hlsl::DXIL::ResourceClass)getResClassUint();
+  }
+  }];
+}
+
 def HLSLNodeLaunch : InheritableAttr {
   let Spellings = [CXX11<"", "nodelaunch", 2017>];
   let Args = [StringArgument<"LaunchType">]; // one of broadcasting, coalescing, thread
@@ -992,13 +1038,6 @@ def HLSLNodeTrackRWInputSharing : InheritableAttr {
   let Documentation = [Undocumented];
 }
 
-def HLSLResource : InheritableAttr {
-  let Spellings = []; // No spellings!
-  let Args = [UnsignedArgument<"ResKind">, UnsignedArgument<"ResClass">];
-  let Subjects = SubjectList<[CXXRecord]>;
-  let Documentation = [Undocumented];
-}
-
 def HLSLNodeObject : InheritableAttr {
   let Spellings = []; // No spellings!
   let Subjects = SubjectList<[CXXRecord]>;
diff --git a/tools/clang/include/clang/Basic/DiagnosticSemaKinds.td b/tools/clang/include/clang/Basic/DiagnosticSemaKinds.td
index 99b6534e1f..d4ebb2ce21 100644
--- a/tools/clang/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/tools/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -7519,8 +7519,8 @@ def err_hlsl_half_load_store: Error<
   "LoadHalf and StoreHalf are not supported for min precision mode">;
 def err_hlsl_interfaces_cannot_inherit: Error<
   "interfaces cannot inherit from other types">;
-def err_hlsl_invalid_range_1_4: Error<
-  "invalid value, valid range is between 1 and 4 inclusive">;
+def err_hlsl_invalid_range_1_to_max
+    : Error<"invalid value, valid range is between 1 and %0 inclusive">;
 def err_hlsl_matrix_member_bad_format: Error<
   "invalid format for matrix subscript '%0'">;
 def err_hlsl_matrix_member_empty: Error<
@@ -7549,6 +7549,8 @@ def err_hlsl_vector_element_index_out_of_bounds: Error<
   "vector element index '%0' is out of bounds">;
 def err_hlsl_vector_member_too_many_positions: Error<
   "more than four positions are referenced in '%0'">;
+def err_hlsl_vector_member_on_long_vector: Error<
+  "Invalid swizzle '%0' on vector of over 4 elements.">;
 def err_hlsl_missing_type_specifier : Error< // Patterened after err_missing_type_specifier
   "HLSL requires a type specifier for all declarations">;
 def err_hlsl_multiple_concrete_bases : Error<
@@ -7701,8 +7703,6 @@ def err_hlsl_control_flow_cond_not_scalar : Error<
   "%0 statement conditional expressions must evaluate to a scalar">;
 def err_hlsl_unsupportedvectortype : Error<
   "%0 is declared with type %1, but only primitive scalar values are supported">;
-def err_hlsl_unsupportedvectorsize : Error<
-  "%0 is declared with size %1, but only values 1 through 4 are supported">;
 def err_hlsl_unsupportedmatrixsize : Error<
   "%0 is declared with size %1x%2, but only values 1 through 4 are supported">;
 def err_hlsl_norm_float_only : Error<
@@ -7853,6 +7853,8 @@ def err_hlsl_load_from_mesh_out_arrays: Error<
    "output arrays of a mesh shader can not be read from">;
 def err_hlsl_out_indices_array_incorrect_access: Error<
    "a vector in out indices array must be accessed as a whole">;
+def err_hlsl_unsupported_long_vector
+    : Error<"vectors of over 4 elements in %0 are not supported">;
 def err_hlsl_logical_binop_scalar : Error<
    "operands for short-circuiting logical binary operator must be scalar, for non-scalar types use '%select{and|or}0'">;
 def err_hlsl_ternary_scalar : Error<
diff --git a/tools/clang/include/clang/Basic/LangOptions.h b/tools/clang/include/clang/Basic/LangOptions.h
index 8dc15da5d8..433b767c8d 100644
--- a/tools/clang/include/clang/Basic/LangOptions.h
+++ b/tools/clang/include/clang/Basic/LangOptions.h
@@ -15,7 +15,7 @@
 #ifndef LLVM_CLANG_BASIC_LANGOPTIONS_H
 #define LLVM_CLANG_BASIC_LANGOPTIONS_H
 
-#include "dxc/DXIL/DxilConstants.h" // For DXIL::DefaultLinkage
+#include "dxc/DXIL/DxilConstants.h" // For DXIL:: default values.
 #include "dxc/Support/HLSLVersion.h"
 #include "clang/Basic/CommentOptions.h"
 #include "clang/Basic/LLVM.h"
@@ -168,6 +168,7 @@ class LangOptions : public LangOptionsBase {
       hlsl::DXIL::DefaultLinkage::Default;
   /// Whether use row major as default matrix major.
   bool HLSLDefaultRowMajor = false;
+  unsigned MaxHLSLVectorLength = hlsl::DXIL::kDefaultMaxVectorLength;
   // HLSL Change Ends
 
   bool SPIRV = false;  // SPIRV Change
diff --git a/tools/clang/include/clang/Sema/SemaHLSL.h b/tools/clang/include/clang/Sema/SemaHLSL.h
index 40b030b430..7e7400d390 100644
--- a/tools/clang/include/clang/Sema/SemaHLSL.h
+++ b/tools/clang/include/clang/Sema/SemaHLSL.h
@@ -128,6 +128,8 @@ unsigned CaculateInitListArraySizeForHLSL(clang::Sema *sema,
                                           const clang::InitListExpr *InitList,
                                           const clang::QualType EltTy);
 
+bool ContainsLongVector(clang::QualType qt);
+
 bool IsConversionToLessOrEqualElements(clang::Sema *self,
                                        const clang::ExprResult &sourceExpr,
                                        const clang::QualType &targetType,
diff --git a/tools/clang/lib/AST/ASTContextHLSL.cpp b/tools/clang/lib/AST/ASTContextHLSL.cpp
index 02125d5a84..870d032d39 100644
--- a/tools/clang/lib/AST/ASTContextHLSL.cpp
+++ b/tools/clang/lib/AST/ASTContextHLSL.cpp
@@ -329,6 +329,9 @@ void hlsl::AddHLSLMatrixTemplate(ASTContext &context,
 
   typeDeclBuilder.addField("h", vectorArrayType);
 
+  typeDeclBuilder.getRecordDecl()->addAttr(
+      HLSLMatrixAttr::CreateImplicit(context));
+
   // Add an operator[]. The operator ranges from zero to rowcount-1, and returns
   // a vector of colcount elements.
   const unsigned int templateDepth = 0;
@@ -385,6 +388,9 @@ void hlsl::AddHLSLVectorTemplate(ASTContext &context,
   // Add an 'h' field to hold the handle.
   typeDeclBuilder.addField("h", vectorType);
 
+  typeDeclBuilder.getRecordDecl()->addAttr(
+      HLSLVectorAttr::CreateImplicit(context));
+
   // Add an operator[]. The operator ranges from zero to colcount-1, and returns
   // a scalar.
 
@@ -525,11 +531,15 @@ hlsl::DeclareRecordTypeWithHandleAndNoMemberFunctions(ASTContext &context,
 /// </summary>
 CXXRecordDecl *
 hlsl::DeclareRecordTypeWithHandle(ASTContext &context, StringRef name,
-                                  bool isCompleteType /*= true */) {
+                                  bool isCompleteType /*= true */,
+                                  InheritableAttr *Attr) {
   BuiltinTypeDeclBuilder typeDeclBuilder(context.getTranslationUnitDecl(), name,
                                          TagDecl::TagKind::TTK_Struct);
   typeDeclBuilder.startDefinition();
   typeDeclBuilder.addField("h", GetHLSLObjectHandleType(context));
+  if (Attr)
+    typeDeclBuilder.getRecordDecl()->addAttr(Attr);
+
   if (isCompleteType)
     return typeDeclBuilder.completeDefinition();
   return typeDeclBuilder.getRecordDecl();
@@ -915,6 +925,7 @@ CXXRecordDecl *hlsl::DeclareTemplateTypeWithHandleInDeclContext(
     ASTContext &context, DeclContext *declContext, StringRef name,
     uint8_t templateArgCount, TypeSourceInfo *defaultTypeArgValue,
     InheritableAttr *Attr) {
+
   DXASSERT(templateArgCount != 0,
            "otherwise caller should be creating a class or struct");
   DXASSERT(templateArgCount <= 2, "otherwise the function needs to be updated "
@@ -938,11 +949,9 @@ CXXRecordDecl *hlsl::DeclareTemplateTypeWithHandleInDeclContext(
   QualType elementType = context.getTemplateTypeParmType(
       /*templateDepth*/ 0, 0, ParameterPackFalse, elementTemplateParamDecl);
 
-  if (templateArgCount > 1 &&
-      // Only need array type for inputpatch and outputpatch.
-      // Avoid Texture2DMS which may use 0 count.
-      // TODO: use hlsl types to do the check.
-      !name.startswith("Texture") && !name.startswith("RWTexture")) {
+  // Only need array type for inputpatch and outputpatch.
+  if (Attr && isa<HLSLTessPatchAttr>(Attr)) {
+    DXASSERT(templateArgCount == 2, "Tess patches need 2 template params");
     Expr *countExpr = DeclRefExpr::Create(
         context, NestedNameSpecifierLoc(), NoLoc, countTemplateParamDecl, false,
         DeclarationNameInfo(countTemplateParamDecl->getDeclName(), NoLoc),
@@ -1098,41 +1107,50 @@ CXXMethodDecl *hlsl::CreateObjectFunctionDeclarationWithParams(
 
 CXXRecordDecl *hlsl::DeclareUIntTemplatedTypeWithHandle(
     ASTContext &context, StringRef typeName, StringRef templateParamName,
-    TagTypeKind tagKind) {
+    InheritableAttr *Attr) {
   return DeclareUIntTemplatedTypeWithHandleInDeclContext(
       context, context.getTranslationUnitDecl(), typeName, templateParamName,
-      tagKind);
+      Attr);
 }
 
 CXXRecordDecl *hlsl::DeclareUIntTemplatedTypeWithHandleInDeclContext(
     ASTContext &context, DeclContext *declContext, StringRef typeName,
-    StringRef templateParamName, TagTypeKind tagKind) {
+    StringRef templateParamName, InheritableAttr *Attr) {
   // template<uint kind> FeedbackTexture2D[Array] { ... }
-  BuiltinTypeDeclBuilder typeDeclBuilder(declContext, typeName, tagKind);
+  BuiltinTypeDeclBuilder typeDeclBuilder(declContext, typeName,
+                                         TagTypeKind::TTK_Class);
   typeDeclBuilder.addIntegerTemplateParam(templateParamName,
                                           context.UnsignedIntTy);
   typeDeclBuilder.startDefinition();
   typeDeclBuilder.addField(
       "h", context.UnsignedIntTy); // Add an 'h' field to hold the handle.
+  if (Attr)
+    typeDeclBuilder.getRecordDecl()->addAttr(Attr);
+
   return typeDeclBuilder.getRecordDecl();
 }
 
 clang::CXXRecordDecl *
-hlsl::DeclareConstantBufferViewType(clang::ASTContext &context, bool bTBuf) {
+hlsl::DeclareConstantBufferViewType(clang::ASTContext &context,
+                                    InheritableAttr *Attr) {
   // Create ConstantBufferView template declaration in translation unit scope
   // like other resource.
   // template<typename T> ConstantBuffer { int h; }
   DeclContext *DC = context.getTranslationUnitDecl();
+  DXASSERT(Attr, "Constbuffer types require an attribute");
 
-  BuiltinTypeDeclBuilder typeDeclBuilder(
-      DC, bTBuf ? "TextureBuffer" : "ConstantBuffer",
-      TagDecl::TagKind::TTK_Struct);
+  const char *TypeName = "ConstantBuffer";
+  if (IsTBuffer(cast<HLSLResourceAttr>(Attr)->getResKind()))
+    TypeName = "TextureBuffer";
+  BuiltinTypeDeclBuilder typeDeclBuilder(DC, TypeName,
+                                         TagDecl::TagKind::TTK_Struct);
   (void)typeDeclBuilder.addTypeTemplateParam("T");
   typeDeclBuilder.startDefinition();
   CXXRecordDecl *templateRecordDecl = typeDeclBuilder.getRecordDecl();
 
   typeDeclBuilder.addField(
       "h", context.UnsignedIntTy); // Add an 'h' field to hold the handle.
+  typeDeclBuilder.getRecordDecl()->addAttr(Attr);
 
   typeDeclBuilder.getRecordDecl();
 
diff --git a/tools/clang/lib/AST/DeclCXX.cpp b/tools/clang/lib/AST/DeclCXX.cpp
index 9ef771b932..5f8c186919 100644
--- a/tools/clang/lib/AST/DeclCXX.cpp
+++ b/tools/clang/lib/AST/DeclCXX.cpp
@@ -48,34 +48,31 @@ void LazyASTUnresolvedSet::getFromExternalSource(ASTContext &C) const {
 }
 
 CXXRecordDecl::DefinitionData::DefinitionData(CXXRecordDecl *D)
-  : UserDeclaredConstructor(false), UserDeclaredSpecialMembers(0),
-    Aggregate(true), PlainOldData(true), Empty(true), Polymorphic(false),
-    Abstract(false), IsStandardLayout(true), HasNoNonEmptyBases(true),
-    HasPrivateFields(false), HasProtectedFields(false), HasPublicFields(false),
-    HasMutableFields(false), HasVariantMembers(false), HasOnlyCMembers(true),
-    HasInClassInitializer(false), HasUninitializedReferenceMember(false),
-    NeedOverloadResolutionForMoveConstructor(false),
-    NeedOverloadResolutionForMoveAssignment(false),
-    NeedOverloadResolutionForDestructor(false),
-    DefaultedMoveConstructorIsDeleted(false),
-    DefaultedMoveAssignmentIsDeleted(false),
-    DefaultedDestructorIsDeleted(false),
-    HasTrivialSpecialMembers(SMF_All),
-    DeclaredNonTrivialSpecialMembers(0),
-    HasIrrelevantDestructor(true),
-    HasConstexprNonCopyMoveConstructor(false),
-    DefaultedDefaultConstructorIsConstexpr(true),
-    HasConstexprDefaultConstructor(false),
-    HasNonLiteralTypeFieldsOrBases(false), ComputedVisibleConversions(false),
-    UserProvidedDefaultConstructor(false), DeclaredSpecialMembers(0),
-    ImplicitCopyConstructorHasConstParam(true),
-    ImplicitCopyAssignmentHasConstParam(true),
-    HasDeclaredCopyConstructorWithConstParam(false),
-    HasDeclaredCopyAssignmentWithConstParam(false),
-    IsLambda(false), IsParsingBaseSpecifiers(false), NumBases(0), NumVBases(0),
-    Bases(), VBases(),
-    Definition(D), FirstFriend() {
-}
+    : UserDeclaredConstructor(false), UserDeclaredSpecialMembers(0),
+      Aggregate(true), PlainOldData(true), Empty(true), Polymorphic(false),
+      Abstract(false), IsStandardLayout(true), HasNoNonEmptyBases(true),
+      HasPrivateFields(false), HasProtectedFields(false),
+      HasPublicFields(false), HasMutableFields(false), HasVariantMembers(false),
+      HasOnlyCMembers(true), HasInClassInitializer(false),
+      HasUninitializedReferenceMember(false),
+      NeedOverloadResolutionForMoveConstructor(false),
+      NeedOverloadResolutionForMoveAssignment(false),
+      NeedOverloadResolutionForDestructor(false),
+      DefaultedMoveConstructorIsDeleted(false),
+      DefaultedMoveAssignmentIsDeleted(false),
+      DefaultedDestructorIsDeleted(false), HasTrivialSpecialMembers(SMF_All),
+      DeclaredNonTrivialSpecialMembers(0), HasIrrelevantDestructor(true),
+      HasConstexprNonCopyMoveConstructor(false),
+      DefaultedDefaultConstructorIsConstexpr(true),
+      HasConstexprDefaultConstructor(false),
+      HasNonLiteralTypeFieldsOrBases(false), ComputedVisibleConversions(false),
+      UserProvidedDefaultConstructor(false), DeclaredSpecialMembers(0),
+      ImplicitCopyConstructorHasConstParam(true),
+      ImplicitCopyAssignmentHasConstParam(true),
+      HasDeclaredCopyConstructorWithConstParam(false),
+      HasDeclaredCopyAssignmentWithConstParam(false), IsLambda(false),
+      IsParsingBaseSpecifiers(false), HasHLSLLongVector(false), NumBases(0),
+      NumVBases(0), Bases(), VBases(), Definition(D), FirstFriend() {}
 
 CXXBaseSpecifier *CXXRecordDecl::DefinitionData::getBasesSlowCase() const {
   return Bases.get(Definition->getASTContext().getExternalSource());
@@ -204,6 +201,10 @@ CXXRecordDecl::setBases(CXXBaseSpecifier const * const *Bases,
     if (!BaseClassDecl->isStandardLayout())
       data().IsStandardLayout = false;
 
+    // Propagate presence of long vector to child classes.
+    if (BaseClassDecl->hasHLSLLongVector())
+      data().HasHLSLLongVector = true;
+
     // Record if this base is the first non-literal field or base.
     if (!hasNonLiteralTypeFieldsOrBases() && !BaseType->isLiteralType(C))
       data().HasNonLiteralTypeFieldsOrBases = true;
@@ -385,6 +386,9 @@ void CXXRecordDecl::addedClassSubobject(CXXRecordDecl *Subobj) {
     data().NeedOverloadResolutionForMoveConstructor = true;
     data().NeedOverloadResolutionForDestructor = true;
   }
+
+  if (Subobj->hasHLSLLongVector())
+    data().HasHLSLLongVector = true;
 }
 
 /// Callback function for CXXRecordDecl::forallBases that acknowledges
diff --git a/tools/clang/lib/AST/HlslTypes.cpp b/tools/clang/lib/AST/HlslTypes.cpp
index d83b307463..41175e3d37 100644
--- a/tools/clang/lib/AST/HlslTypes.cpp
+++ b/tools/clang/lib/AST/HlslTypes.cpp
@@ -53,44 +53,44 @@ ConvertHLSLVecMatTypeToExtVectorType(const clang::ASTContext &context,
   return nullptr;
 }
 
+template <typename AttrType> static AttrType *getAttr(clang::QualType type) {
+  type = type.getCanonicalType();
+  if (const RecordType *RT = type->getAs<RecordType>()) {
+    if (const auto *Spec =
+            dyn_cast<ClassTemplateSpecializationDecl>(RT->getDecl()))
+      if (const auto *Template =
+              dyn_cast<ClassTemplateDecl>(Spec->getSpecializedTemplate()))
+        return Template->getTemplatedDecl()->getAttr<AttrType>();
+    if (const auto *Decl = dyn_cast<CXXRecordDecl>(RT->getDecl()))
+      return Decl->getAttr<AttrType>();
+  }
+  return nullptr;
+}
+
 bool IsHLSLVecMatType(clang::QualType type) {
-  const Type *Ty = type.getCanonicalType().getTypePtr();
-  if (const RecordType *RT = dyn_cast<RecordType>(Ty)) {
-    if (const ClassTemplateSpecializationDecl *templateDecl =
-            dyn_cast<ClassTemplateSpecializationDecl>(RT->getDecl())) {
-      if (templateDecl->getName() == "vector") {
-        return true;
-      } else if (templateDecl->getName() == "matrix") {
-        return true;
-      }
-    }
+  type = type.getCanonicalType();
+  if (const RecordType *RT = type->getAs<RecordType>()) {
+    if (const auto *Spec =
+            dyn_cast<ClassTemplateSpecializationDecl>(RT->getDecl()))
+      if (const auto *Template =
+              dyn_cast<ClassTemplateDecl>(Spec->getSpecializedTemplate()))
+        return Template->getTemplatedDecl()->getAttr<HLSLMatrixAttr>() ||
+               Template->getTemplatedDecl()->getAttr<HLSLVectorAttr>();
+    if (const auto *Decl = dyn_cast<CXXRecordDecl>(RT->getDecl()))
+      return Decl->getAttr<HLSLMatrixAttr>() || Decl->getAttr<HLSLVectorAttr>();
   }
   return false;
 }
 
 bool IsHLSLMatType(clang::QualType type) {
-  const clang::Type *Ty = type.getCanonicalType().getTypePtr();
-  if (const RecordType *RT = dyn_cast<RecordType>(Ty)) {
-    if (const ClassTemplateSpecializationDecl *templateDecl =
-            dyn_cast<ClassTemplateSpecializationDecl>(RT->getDecl())) {
-      if (templateDecl->getName() == "matrix") {
-        return true;
-      }
-    }
-  }
+  if (getAttr<HLSLMatrixAttr>(type))
+    return true;
   return false;
 }
 
 bool IsHLSLVecType(clang::QualType type) {
-  const clang::Type *Ty = type.getCanonicalType().getTypePtr();
-  if (const RecordType *RT = dyn_cast<RecordType>(Ty)) {
-    if (const ClassTemplateSpecializationDecl *templateDecl =
-            dyn_cast<ClassTemplateSpecializationDecl>(RT->getDecl())) {
-      if (templateDecl->getName() == "vector") {
-        return true;
-      }
-    }
-  }
+  if (getAttr<HLSLVectorAttr>(type))
+    return true;
   return false;
 }
 
@@ -474,160 +474,52 @@ clang::QualType GetHLSLMatElementType(clang::QualType type) {
   QualType elemTy = arg0.getAsType();
   return elemTy;
 }
+
 // TODO: Add type cache to ASTContext.
 bool IsHLSLInputPatchType(QualType type) {
-  type = type.getCanonicalType();
-  if (const RecordType *RT = dyn_cast<RecordType>(type)) {
-    if (const ClassTemplateSpecializationDecl *templateDecl =
-            dyn_cast<ClassTemplateSpecializationDecl>(
-                RT->getAsCXXRecordDecl())) {
-      if (templateDecl->getName() == "InputPatch") {
-        return true;
-      }
-    }
-  }
+  if (const HLSLTessPatchAttr *Attr = getAttr<HLSLTessPatchAttr>(type))
+    return Attr->getIsInput();
   return false;
 }
+
 bool IsHLSLOutputPatchType(QualType type) {
-  type = type.getCanonicalType();
-  if (const RecordType *RT = dyn_cast<RecordType>(type)) {
-    if (const ClassTemplateSpecializationDecl *templateDecl =
-            dyn_cast<ClassTemplateSpecializationDecl>(
-                RT->getAsCXXRecordDecl())) {
-      if (templateDecl->getName() == "OutputPatch") {
-        return true;
-      }
-    }
-  }
+  if (const HLSLTessPatchAttr *Attr = getAttr<HLSLTessPatchAttr>(type))
+    return !Attr->getIsInput();
   return false;
 }
+
 bool IsHLSLPointStreamType(QualType type) {
-  type = type.getCanonicalType();
-  if (const RecordType *RT = dyn_cast<RecordType>(type)) {
-    if (const ClassTemplateSpecializationDecl *templateDecl =
-            dyn_cast<ClassTemplateSpecializationDecl>(
-                RT->getAsCXXRecordDecl())) {
-      if (templateDecl->getName() == "PointStream")
-        return true;
-    }
-  }
+  if (const HLSLStreamOutputAttr *Attr = getAttr<HLSLStreamOutputAttr>(type))
+    return Attr->getPrimVertices() == (unsigned)DXIL::InputPrimitive::Point;
   return false;
 }
+
 bool IsHLSLLineStreamType(QualType type) {
-  type = type.getCanonicalType();
-  if (const RecordType *RT = dyn_cast<RecordType>(type)) {
-    if (const ClassTemplateSpecializationDecl *templateDecl =
-            dyn_cast<ClassTemplateSpecializationDecl>(
-                RT->getAsCXXRecordDecl())) {
-      if (templateDecl->getName() == "LineStream")
-        return true;
-    }
-  }
+  if (const HLSLStreamOutputAttr *Attr = getAttr<HLSLStreamOutputAttr>(type))
+    return Attr->getPrimVertices() == (unsigned)DXIL::InputPrimitive::Line;
   return false;
 }
+
 bool IsHLSLTriangleStreamType(QualType type) {
-  type = type.getCanonicalType();
-  if (const RecordType *RT = dyn_cast<RecordType>(type)) {
-    if (const ClassTemplateSpecializationDecl *templateDecl =
-            dyn_cast<ClassTemplateSpecializationDecl>(
-                RT->getAsCXXRecordDecl())) {
-      if (templateDecl->getName() == "TriangleStream")
-        return true;
-    }
-  }
+  if (const HLSLStreamOutputAttr *Attr = getAttr<HLSLStreamOutputAttr>(type))
+    return Attr->getPrimVertices() == (unsigned)DXIL::InputPrimitive::Triangle;
   return false;
 }
+
 bool IsHLSLStreamOutputType(QualType type) {
-  type = type.getCanonicalType();
-  if (const RecordType *RT = dyn_cast<RecordType>(type)) {
-    if (const ClassTemplateSpecializationDecl *templateDecl =
-            dyn_cast<ClassTemplateSpecializationDecl>(
-                RT->getAsCXXRecordDecl())) {
-      if (templateDecl->getName() == "PointStream")
-        return true;
-      if (templateDecl->getName() == "LineStream")
-        return true;
-      if (templateDecl->getName() == "TriangleStream")
-        return true;
-    }
-  }
+  if (getAttr<HLSLStreamOutputAttr>(type))
+    return true;
   return false;
 }
-bool IsHLSLResourceType(clang::QualType type) {
-  if (const RecordType *RT = type->getAs<RecordType>()) {
-    StringRef name = RT->getDecl()->getName();
-    if (name == "Texture1D" || name == "RWTexture1D")
-      return true;
-    if (name == "Texture2D" || name == "RWTexture2D")
-      return true;
-    if (name == "Texture2DMS" || name == "RWTexture2DMS")
-      return true;
-    if (name == "Texture3D" || name == "RWTexture3D")
-      return true;
-    if (name == "TextureCube" || name == "RWTextureCube")
-      return true;
-
-    if (name == "Texture1DArray" || name == "RWTexture1DArray")
-      return true;
-    if (name == "Texture2DArray" || name == "RWTexture2DArray")
-      return true;
-    if (name == "Texture2DMSArray" || name == "RWTexture2DMSArray")
-      return true;
-    if (name == "TextureCubeArray" || name == "RWTextureCubeArray")
-      return true;
-
-    if (name == "FeedbackTexture2D" || name == "FeedbackTexture2DArray")
-      return true;
-
-    if (name == "RasterizerOrderedTexture1D" ||
-        name == "RasterizerOrderedTexture2D" ||
-        name == "RasterizerOrderedTexture3D" ||
-        name == "RasterizerOrderedTexture1DArray" ||
-        name == "RasterizerOrderedTexture2DArray" ||
-        name == "RasterizerOrderedBuffer" ||
-        name == "RasterizerOrderedByteAddressBuffer" ||
-        name == "RasterizerOrderedStructuredBuffer")
-      return true;
-
-    if (name == "ByteAddressBuffer" || name == "RWByteAddressBuffer")
-      return true;
-
-    if (name == "StructuredBuffer" || name == "RWStructuredBuffer")
-      return true;
-
-    if (name == "AppendStructuredBuffer" || name == "ConsumeStructuredBuffer")
-      return true;
-
-    if (name == "Buffer" || name == "RWBuffer")
-      return true;
-
-    if (name == "SamplerState" || name == "SamplerComparisonState")
-      return true;
 
-    if (name == "ConstantBuffer" || name == "TextureBuffer")
-      return true;
-
-    if (name == "RaytracingAccelerationStructure")
-      return true;
-  }
+bool IsHLSLResourceType(clang::QualType type) {
+  if (getAttr<HLSLResourceAttr>(type))
+    return true;
   return false;
 }
 
-static HLSLNodeObjectAttr *getNodeAttr(clang::QualType type) {
-  if (const RecordType *RT = type->getAs<RecordType>()) {
-    if (const auto *Spec =
-            dyn_cast<ClassTemplateSpecializationDecl>(RT->getDecl()))
-      if (const auto *Template =
-              dyn_cast<ClassTemplateDecl>(Spec->getSpecializedTemplate()))
-        return Template->getTemplatedDecl()->getAttr<HLSLNodeObjectAttr>();
-    if (const auto *Decl = dyn_cast<CXXRecordDecl>(RT->getDecl()))
-      return Decl->getAttr<HLSLNodeObjectAttr>();
-  }
-  return nullptr;
-}
-
 DXIL::NodeIOKind GetNodeIOType(clang::QualType type) {
-  if (const HLSLNodeObjectAttr *Attr = getNodeAttr(type))
+  if (const HLSLNodeObjectAttr *Attr = getAttr<HLSLNodeObjectAttr>(type))
     return Attr->getNodeIOType();
   return DXIL::NodeIOKind::Invalid;
 }
@@ -654,27 +546,20 @@ bool IsHLSLDynamicSamplerType(clang::QualType type) {
 }
 
 bool IsHLSLNodeType(clang::QualType type) {
-  if (const HLSLNodeObjectAttr *Attr = getNodeAttr(type))
+  if (const HLSLNodeObjectAttr *Attr = getAttr<HLSLNodeObjectAttr>(type))
     return true;
   return false;
 }
 
 bool IsHLSLObjectWithImplicitMemberAccess(clang::QualType type) {
-  if (const RecordType *RT = type->getAs<RecordType>()) {
-    StringRef name = RT->getDecl()->getName();
-    if (name == "ConstantBuffer" || name == "TextureBuffer")
-      return true;
-  }
+  if (const HLSLResourceAttr *Attr = getAttr<HLSLResourceAttr>(type))
+    return DXIL::IsCTBuffer(Attr->getResKind());
   return false;
 }
 
 bool IsHLSLObjectWithImplicitROMemberAccess(clang::QualType type) {
-  if (const RecordType *RT = type->getAs<RecordType>()) {
-    StringRef name = RT->getDecl()->getName();
-    // Read-only records
-    if (name == "ConstantBuffer" || name == "TextureBuffer")
-      return true;
-  }
+  if (const HLSLResourceAttr *Attr = getAttr<HLSLResourceAttr>(type))
+    return DXIL::IsCTBuffer(Attr->getResKind());
   return false;
 }
 
@@ -701,14 +586,8 @@ bool IsHLSLNodeOutputType(clang::QualType type) {
 }
 
 bool IsHLSLStructuredBufferType(clang::QualType type) {
-  if (const RecordType *RT = type->getAs<RecordType>()) {
-    StringRef name = RT->getDecl()->getName();
-    if (name == "StructuredBuffer" || name == "RWStructuredBuffer")
-      return true;
-
-    if (name == "AppendStructuredBuffer" || name == "ConsumeStructuredBuffer")
-      return true;
-  }
+  if (const HLSLResourceAttr *Attr = getAttr<HLSLResourceAttr>(type))
+    return Attr->getResKind() == DXIL::ResourceKind::StructuredBuffer;
   return false;
 }
 
@@ -914,7 +793,8 @@ QualType GetHLSLResourceResultType(QualType type) {
   if (const ClassTemplateSpecializationDecl *templateDecl =
           dyn_cast<ClassTemplateSpecializationDecl>(RD)) {
 
-    if (RD->getName().startswith("FeedbackTexture")) {
+    const HLSLResourceAttr *Attr = getAttr<HLSLResourceAttr>(type);
+    if (Attr && DXIL::IsFeedbackTexture(Attr->getResKind())) {
       // Feedback textures are write-only and the data is opaque,
       // so there is no result type per se.
       return {};
diff --git a/tools/clang/lib/CodeGen/CGExprScalar.cpp b/tools/clang/lib/CodeGen/CGExprScalar.cpp
index 0cb993e6f4..530c791fcc 100644
--- a/tools/clang/lib/CodeGen/CGExprScalar.cpp
+++ b/tools/clang/lib/CodeGen/CGExprScalar.cpp
@@ -3713,20 +3713,7 @@ VisitAbstractConditionalOperator(const AbstractConditionalOperator *E) {
       llvm::Value *CondV = CGF.EmitScalarExpr(condExpr);
       llvm::Value *LHS = Visit(lhsExpr);
       llvm::Value *RHS = Visit(rhsExpr);
-      if (llvm::VectorType *VT = dyn_cast<llvm::VectorType>(CondV->getType())) {
-        llvm::VectorType *ResultVT = cast<llvm::VectorType>(LHS->getType());
-        llvm::Value *result = llvm::UndefValue::get(ResultVT);
-        for (unsigned i = 0; i < VT->getNumElements(); i++) {
-          llvm::Value *EltCond = Builder.CreateExtractElement(CondV, i);
-          llvm::Value *EltL = Builder.CreateExtractElement(LHS, i);
-          llvm::Value *EltR = Builder.CreateExtractElement(RHS, i);
-          llvm::Value *EltSelect = Builder.CreateSelect(EltCond, EltL, EltR);
-          result = Builder.CreateInsertElement(result, EltSelect, i);
-        }
-        return result;
-      } else {
-        return Builder.CreateSelect(CondV, LHS, RHS);
-      }
+      return Builder.CreateSelect(CondV, LHS, RHS);
     }
     if (hlsl::IsHLSLMatType(E->getType())) {
       llvm::Value *Cond = CGF.EmitScalarExpr(condExpr);
diff --git a/tools/clang/lib/Sema/SemaDXR.cpp b/tools/clang/lib/Sema/SemaDXR.cpp
index 6d838fb203..73ea9dd93c 100644
--- a/tools/clang/lib/Sema/SemaDXR.cpp
+++ b/tools/clang/lib/Sema/SemaDXR.cpp
@@ -810,6 +810,12 @@ void DiagnoseTraceCall(Sema &S, const VarDecl *Payload,
     return;
   }
 
+  if (ContainsLongVector(Payload->getType())) {
+    S.Diag(Payload->getLocation(), diag::err_hlsl_unsupported_long_vector)
+        << "payload parameters";
+    return;
+  }
+
   CollectNonAccessableFields(PayloadType, CallerStage, {}, {},
                              NonWriteableFields, NonReadableFields);
 
diff --git a/tools/clang/lib/Sema/SemaHLSL.cpp b/tools/clang/lib/Sema/SemaHLSL.cpp
index fb3937cfd5..555b0ba44a 100644
--- a/tools/clang/lib/Sema/SemaHLSL.cpp
+++ b/tools/clang/lib/Sema/SemaHLSL.cpp
@@ -363,6 +363,8 @@ enum ArBasicKind {
 
 #define IS_BPROP_STREAM(_Props) (((_Props)&BPROP_STREAM) != 0)
 
+#define IS_BPROP_PATCH(_Props) (((_Props) & BPROP_PATCH) != 0)
+
 #define IS_BPROP_SAMPLER(_Props) (((_Props)&BPROP_SAMPLER) != 0)
 
 #define IS_BPROP_TEXTURE(_Props) (((_Props)&BPROP_TEXTURE) != 0)
@@ -616,6 +618,8 @@ C_ASSERT(ARRAYSIZE(g_uBasicKindProps) == AR_BASIC_MAXIMUM_COUNT);
 
 #define IS_BASIC_STREAM(_Kind) IS_BPROP_STREAM(GetBasicKindProps(_Kind))
 
+#define IS_BASIC_PATCH(_Kind) IS_BPROP_PATCH(GetBasicKindProps(_Kind))
+
 #define IS_BASIC_SAMPLER(_Kind) IS_BPROP_SAMPLER(GetBasicKindProps(_Kind))
 #define IS_BASIC_TEXTURE(_Kind) IS_BPROP_TEXTURE(GetBasicKindProps(_Kind))
 #define IS_BASIC_OBJECT(_Kind) IS_BPROP_OBJECT(GetBasicKindProps(_Kind))
@@ -1014,12 +1018,13 @@ static const ArBasicKind g_UIntCT[] = {AR_BASIC_UINT32, AR_BASIC_LITERAL_INT,
 // should map to int32, not int16 or int64
 static const ArBasicKind g_AnyIntCT[] = {
     AR_BASIC_INT32, AR_BASIC_INT16,  AR_BASIC_UINT32,      AR_BASIC_UINT16,
-    AR_BASIC_INT64, AR_BASIC_UINT64, AR_BASIC_LITERAL_INT, AR_BASIC_UNKNOWN};
+    AR_BASIC_INT64, AR_BASIC_UINT64, AR_BASIC_INT8_4PACKED, AR_BASIC_UINT8_4PACKED,
+    AR_BASIC_LITERAL_INT, AR_BASIC_UNKNOWN};
 
 static const ArBasicKind g_AnyInt32CT[] = {
-    AR_BASIC_INT32, AR_BASIC_UINT32, AR_BASIC_LITERAL_INT, AR_BASIC_UNKNOWN};
+  AR_BASIC_INT32, AR_BASIC_UINT32, AR_BASIC_INT8_4PACKED, AR_BASIC_UINT8_4PACKED, AR_BASIC_LITERAL_INT, AR_BASIC_UNKNOWN};
 
-static const ArBasicKind g_UIntOnlyCT[] = {AR_BASIC_UINT32, AR_BASIC_UINT64,
+static const ArBasicKind g_UIntOnlyCT[] = {AR_BASIC_UINT32, AR_BASIC_UINT64,AR_BASIC_INT8_4PACKED, AR_BASIC_UINT8_4PACKED, 
                                            AR_BASIC_LITERAL_INT,
                                            AR_BASIC_NOCAST, AR_BASIC_UNKNOWN};
 
@@ -1059,18 +1064,20 @@ static const ArBasicKind g_NumericCT[] = {
     AR_BASIC_UINT16,        AR_BASIC_UINT32,
     AR_BASIC_MIN12INT,      AR_BASIC_MIN16INT,
     AR_BASIC_MIN16UINT,     AR_BASIC_INT64,
-    AR_BASIC_UINT64,        AR_BASIC_UNKNOWN};
+    AR_BASIC_UINT64,        AR_BASIC_INT8_4PACKED, AR_BASIC_UINT8_4PACKED, AR_BASIC_UNKNOWN};
 
 static const ArBasicKind g_Numeric32CT[] = {
     AR_BASIC_FLOAT32,       AR_BASIC_FLOAT32_PARTIAL_PRECISION,
     AR_BASIC_LITERAL_FLOAT, AR_BASIC_LITERAL_INT,
     AR_BASIC_INT32,         AR_BASIC_UINT32,
+AR_BASIC_INT8_4PACKED, AR_BASIC_UINT8_4PACKED,     
     AR_BASIC_UNKNOWN};
 
 static const ArBasicKind g_Numeric32OnlyCT[] = {
     AR_BASIC_FLOAT32,       AR_BASIC_FLOAT32_PARTIAL_PRECISION,
     AR_BASIC_LITERAL_FLOAT, AR_BASIC_LITERAL_INT,
     AR_BASIC_INT32,         AR_BASIC_UINT32,
+AR_BASIC_INT8_4PACKED, AR_BASIC_UINT8_4PACKED, 
     AR_BASIC_NOCAST,        AR_BASIC_UNKNOWN};
 
 static const ArBasicKind g_AnyCT[] = {
@@ -1083,6 +1090,7 @@ static const ArBasicKind g_AnyCT[] = {
     AR_BASIC_MIN12INT,      AR_BASIC_MIN16INT,
     AR_BASIC_MIN16UINT,     AR_BASIC_BOOL,
     AR_BASIC_INT64,         AR_BASIC_UINT64,
+AR_BASIC_INT8_4PACKED, AR_BASIC_UINT8_4PACKED, 
     AR_BASIC_UNKNOWN};
 
 static const ArBasicKind g_AnySamplerCT[] = {
@@ -1146,6 +1154,7 @@ static const ArBasicKind g_Numeric16OnlyCT[] = {
     AR_BASIC_UNKNOWN};
 
 static const ArBasicKind g_Int32OnlyCT[] = {AR_BASIC_INT32, AR_BASIC_UINT32,
+AR_BASIC_INT8_4PACKED, AR_BASIC_UINT8_4PACKED, 
                                             AR_BASIC_LITERAL_INT,
                                             AR_BASIC_NOCAST, AR_BASIC_UNKNOWN};
 
@@ -1170,10 +1179,13 @@ static const ArBasicKind g_UInt8_4PackedCT[] = {
 
 static const ArBasicKind g_AnyInt16Or32CT[] = {
     AR_BASIC_INT32,  AR_BASIC_UINT32,      AR_BASIC_INT16,
-    AR_BASIC_UINT16, AR_BASIC_LITERAL_INT, AR_BASIC_UNKNOWN};
+    AR_BASIC_UINT16, 
+AR_BASIC_INT8_4PACKED, AR_BASIC_UINT8_4PACKED, AR_BASIC_LITERAL_INT, AR_BASIC_UNKNOWN};
 
 static const ArBasicKind g_SInt16Or32OnlyCT[] = {
-    AR_BASIC_INT32, AR_BASIC_INT16, AR_BASIC_LITERAL_INT, AR_BASIC_NOCAST,
+    AR_BASIC_INT32, AR_BASIC_INT16, AR_BASIC_LITERAL_INT, 
+AR_BASIC_INT8_4PACKED, AR_BASIC_UINT8_4PACKED, 
+AR_BASIC_NOCAST,
     AR_BASIC_UNKNOWN};
 
 static const ArBasicKind g_ByteAddressBufferCT[] = {
@@ -2856,8 +2868,9 @@ class HLSLExternalSource : public ExternalSemaSource {
   TypedefDecl *m_matrixShorthandTypes[HLSLScalarTypeCount][4][4];
 
   // Vector types already built.
-  QualType m_vectorTypes[HLSLScalarTypeCount][4];
-  TypedefDecl *m_vectorTypedefs[HLSLScalarTypeCount][4];
+  QualType m_vectorTypes[HLSLScalarTypeCount][DXIL::kDefaultMaxVectorLength];
+  TypedefDecl
+      *m_vectorTypedefs[HLSLScalarTypeCount][DXIL::kDefaultMaxVectorLength];
 
   // BuiltinType for each scalar type.
   QualType m_baseTypes[HLSLScalarTypeCount];
@@ -3540,6 +3553,20 @@ class HLSLExternalSource : public ExternalSemaSource {
       if (kind == AR_OBJECT_LEGACY_EFFECT)
         effectKindIndex = i;
 
+      InheritableAttr *Attr = nullptr;
+      if (IS_BASIC_STREAM(kind))
+        Attr = HLSLStreamOutputAttr::CreateImplicit(
+            *m_context, kind - AR_OBJECT_POINTSTREAM + 1);
+      else if (IS_BASIC_PATCH(kind))
+        Attr = HLSLTessPatchAttr::CreateImplicit(*m_context,
+                                                 kind == AR_OBJECT_INPUTPATCH);
+      else {
+        DXIL::ResourceKind ResKind = DXIL::ResourceKind::NumEntries;
+        DXIL::ResourceClass ResClass = DXIL::ResourceClass::Invalid;
+        if (GetBasicKindResourceKindAndClass(kind, ResKind, ResClass))
+          Attr = HLSLResourceAttr::CreateImplicit(*m_context, (unsigned)ResKind,
+                                                  (unsigned)ResClass);
+      }
       DXASSERT(kind < _countof(g_ArBasicTypeNames),
                "g_ArBasicTypeNames has the wrong number of entries");
       assert(kind < _countof(g_ArBasicTypeNames));
@@ -3586,9 +3613,9 @@ class HLSLExternalSource : public ExternalSemaSource {
           break;
         }
       } else if (kind == AR_OBJECT_CONSTANT_BUFFER) {
-        recordDecl = DeclareConstantBufferViewType(*m_context, /*bTBuf*/ false);
+        recordDecl = DeclareConstantBufferViewType(*m_context, Attr);
       } else if (kind == AR_OBJECT_TEXTURE_BUFFER) {
-        recordDecl = DeclareConstantBufferViewType(*m_context, /*bTBuf*/ true);
+        recordDecl = DeclareConstantBufferViewType(*m_context, Attr);
       } else if (kind == AR_OBJECT_RAY_QUERY) {
         recordDecl = DeclareRayQueryType(*m_context);
       } else if (kind == AR_OBJECT_HEAP_RESOURCE) {
@@ -3609,10 +3636,10 @@ class HLSLExternalSource : public ExternalSemaSource {
         }
       } else if (kind == AR_OBJECT_FEEDBACKTEXTURE2D) {
         recordDecl = DeclareUIntTemplatedTypeWithHandle(
-            *m_context, "FeedbackTexture2D", "kind");
+            *m_context, "FeedbackTexture2D", "kind", Attr);
       } else if (kind == AR_OBJECT_FEEDBACKTEXTURE2D_ARRAY) {
         recordDecl = DeclareUIntTemplatedTypeWithHandle(
-            *m_context, "FeedbackTexture2DArray", "kind");
+            *m_context, "FeedbackTexture2DArray", "kind", Attr);
       } else if (kind == AR_OBJECT_EMPTY_NODE_INPUT) {
         recordDecl = DeclareNodeOrRecordType(
             *m_context, DXIL::NodeIOKind::EmptyInput,
@@ -3728,20 +3755,12 @@ class HLSLExternalSource : public ExternalSemaSource {
       }
 #endif
       else if (templateArgCount == 0) {
-        recordDecl = DeclareRecordTypeWithHandle(*m_context, typeName,
-                                                 /*isCompleteType*/ false);
+        recordDecl =
+            DeclareRecordTypeWithHandle(*m_context, typeName,
+                                        /*isCompleteType*/ false, Attr);
       } else {
         DXASSERT(templateArgCount == 1 || templateArgCount == 2,
                  "otherwise a new case has been added");
-
-        InheritableAttr *Attr = nullptr;
-        DXIL::ResourceKind ResKind = DXIL::ResourceKind::NumEntries;
-        DXIL::ResourceClass ResClass = DXIL::ResourceClass::Invalid;
-        if (GetBasicKindResourceKindAndClass(kind, ResKind, ResClass)) {
-          Attr = HLSLResourceAttr::CreateImplicit(*m_context, (unsigned)ResKind,
-                                                  (unsigned)ResClass);
-        }
-
         TypeSourceInfo *typeDefault =
             TemplateHasDefaultType(kind) ? float4TypeSourceInfo : nullptr;
         recordDecl = DeclareTemplateTypeWithHandle(
@@ -3830,7 +3849,7 @@ class HLSLExternalSource : public ExternalSemaSource {
   clang::TypedefDecl *LookupVectorShorthandType(HLSLScalarType scalarType,
                                                 UINT colCount) {
     DXASSERT_NOMSG(scalarType != HLSLScalarType::HLSLScalarType_unknown &&
-                   colCount <= 4);
+                   colCount <= DXIL::kDefaultMaxVectorLength);
     TypedefDecl *qts = m_vectorTypedefs[scalarType][colCount - 1];
     if (qts == nullptr) {
       QualType type = LookupVectorType(scalarType, colCount);
@@ -3937,7 +3956,9 @@ class HLSLExternalSource : public ExternalSemaSource {
   }
 
   QualType LookupVectorType(HLSLScalarType scalarType, unsigned int colCount) {
-    QualType qt = m_vectorTypes[scalarType][colCount - 1];
+    QualType qt;
+    if (colCount < DXIL::kDefaultMaxVectorLength)
+      qt = m_vectorTypes[scalarType][colCount - 1];
     if (qt.isNull()) {
       if (m_scalarTypes[scalarType].isNull()) {
         LookupScalarTypeDef(scalarType);
@@ -3945,7 +3966,8 @@ class HLSLExternalSource : public ExternalSemaSource {
       qt = GetOrCreateVectorSpecialization(*m_context, m_sema,
                                            m_vectorTemplateDecl,
                                            m_scalarTypes[scalarType], colCount);
-      m_vectorTypes[scalarType][colCount - 1] = qt;
+      if (colCount < DXIL::kDefaultMaxVectorLength)
+        m_vectorTypes[scalarType][colCount - 1] = qt;
     }
     return qt;
   }
@@ -4741,7 +4763,7 @@ class HLSLExternalSource : public ExternalSemaSource {
       return true;
     case AR_OBJECT_TEXTURE_BUFFER:
       ResKind = DXIL::ResourceKind::TBuffer;
-      ResClass = DXIL::ResourceClass::CBuffer;
+      ResClass = DXIL::ResourceClass::SRV;
       return true;
     case AR_OBJECT_FEEDBACKTEXTURE2D:
       ResKind = DXIL::ResourceKind::FeedbackTexture2D;
@@ -4751,6 +4773,15 @@ class HLSLExternalSource : public ExternalSemaSource {
       ResKind = DXIL::ResourceKind::FeedbackTexture2DArray;
       ResClass = DXIL::ResourceClass::SRV;
       return true;
+    case AR_OBJECT_SAMPLER:
+    case AR_OBJECT_SAMPLERCOMPARISON:
+      ResKind = DXIL::ResourceKind::Sampler;
+      ResClass = DXIL::ResourceClass::Sampler;
+      return true;
+    case AR_OBJECT_ACCELERATION_STRUCT:
+      ResKind = DXIL::ResourceKind::RTAccelerationStructure;
+      ResClass = DXIL::ResourceClass::SRV;
+      return true;
     default:
       return false;
     }
@@ -5033,10 +5064,6 @@ class HLSLExternalSource : public ExternalSemaSource {
            AR_BASIC_UNKNOWN;
   }
 
-  /// <summary>Checks whether the specified value is a valid vector
-  /// size.</summary>
-  bool IsValidVectorSize(size_t length) { return 1 <= length && length <= 4; }
-
   /// <summary>Checks whether the specified value is a valid matrix row or
   /// column size.</summary>
   bool IsValidMatrixColOrRowSize(size_t length) {
@@ -5072,11 +5099,6 @@ class HLSLExternalSource : public ExternalSemaSource {
                                            false);
       } else if (objectKind == AR_TOBJ_VECTOR) {
         bool valid = true;
-        if (!IsValidVectorSize(GetHLSLVecSize(type))) {
-          valid = false;
-          m_sema->Diag(argLoc, diag::err_hlsl_unsupportedvectorsize)
-              << type << GetHLSLVecSize(type);
-        }
         if (!IsScalarType(GetMatrixOrVectorElementType(type))) {
           valid = false;
           m_sema->Diag(argLoc, diag::err_hlsl_unsupportedvectortype)
@@ -5194,9 +5216,13 @@ class HLSLExternalSource : public ExternalSemaSource {
                                                 SourceLocation Loc);
 
   bool CheckRangedTemplateArgument(SourceLocation diagLoc,
-                                   llvm::APSInt &sintValue) {
-    if (!sintValue.isStrictlyPositive() || sintValue.getLimitedValue() > 4) {
-      m_sema->Diag(diagLoc, diag::err_hlsl_invalid_range_1_4);
+                                   llvm::APSInt &sintValue, bool IsVector) {
+    unsigned MaxLength = DXIL::kDefaultMaxVectorLength;
+    if (IsVector)
+      MaxLength = m_sema->getLangOpts().MaxHLSLVectorLength;
+    if (!sintValue.isStrictlyPositive() ||
+        sintValue.getLimitedValue() > MaxLength) {
+      m_sema->Diag(diagLoc, diag::err_hlsl_invalid_range_1_to_max) << MaxLength;
       return true;
     }
 
@@ -5219,11 +5245,14 @@ class HLSLExternalSource : public ExternalSemaSource {
       return false;
     }
     // Allow object type for Constant/TextureBuffer.
-    if (templateName == "ConstantBuffer" || templateName == "TextureBuffer") {
+    HLSLResourceAttr *ResAttr =
+        Template->getTemplatedDecl()->getAttr<HLSLResourceAttr>();
+    if (ResAttr && DXIL::IsCTBuffer(ResAttr->getResKind())) {
       if (TemplateArgList.size() == 1) {
         const TemplateArgumentLoc &argLoc = TemplateArgList[0];
         const TemplateArgument &arg = argLoc.getArgument();
-        DXASSERT(arg.getKind() == TemplateArgument::ArgKind::Type, "");
+        DXASSERT(arg.getKind() == TemplateArgument::ArgKind::Type,
+                 "cbuffer with non-type template arg");
         QualType argType = arg.getAsType();
         SourceLocation argSrcLoc = argLoc.getLocation();
         if (IsScalarType(argType) || IsVectorType(m_sema, argType) ||
@@ -5233,23 +5262,13 @@ class HLSLExternalSource : public ExternalSemaSource {
               << argType;
           return true;
         }
-        if (auto *TST = dyn_cast<TemplateSpecializationType>(argType)) {
-          // This is a bit of a special case we need to handle. Because the
-          // buffer types don't use their template parameter in a way that would
-          // force instantiation, we need to force specialization here.
-          GetOrCreateTemplateSpecialization(
-              *m_context, *m_sema,
-              cast<ClassTemplateDecl>(
-                  TST->getTemplateName().getAsTemplateDecl()),
-              llvm::ArrayRef<TemplateArgument>(TST->getArgs(),
-                                               TST->getNumArgs()));
-        }
-        if (const RecordType *recordType = argType->getAs<RecordType>()) {
-          if (!recordType->getDecl()->isCompleteDefinition()) {
-            m_sema->Diag(argSrcLoc, diag::err_typecheck_decl_incomplete_type)
-                << argType;
-            return true;
-          }
+        m_sema->RequireCompleteType(argSrcLoc, argType,
+                                    diag::err_typecheck_decl_incomplete_type);
+
+        if (ContainsLongVector(argType)) {
+          m_sema->Diag(argSrcLoc, diag::err_hlsl_unsupported_long_vector)
+              << "ConstantBuffers or TextureBuffers";
+          return true;
         }
       }
       return false;
@@ -5279,22 +5298,13 @@ class HLSLExternalSource : public ExternalSemaSource {
       // template instantiation.
       if (ArgTy->isDependentType())
         return false;
-      if (auto *recordType = ArgTy->getAs<RecordType>()) {
-        if (CXXRecordDecl *cxxRecordDecl =
-                dyn_cast<CXXRecordDecl>(recordType->getDecl())) {
-          if (ClassTemplateSpecializationDecl *templateSpecializationDecl =
-                  dyn_cast<ClassTemplateSpecializationDecl>(cxxRecordDecl)) {
-            if (templateSpecializationDecl->getSpecializationKind() ==
-                TSK_Undeclared) {
-              // Make sure specialization is done before IsTypeNumeric.
-              // If not, ArgTy might be treat as empty struct.
-              m_sema->RequireCompleteType(
-                  ArgLoc.getLocation(), ArgTy,
-                  diag::err_typecheck_decl_incomplete_type);
-            }
-          }
-        }
-      }
+      // Make sure specialization is done before IsTypeNumeric.
+      // If not, ArgTy might be treat as empty struct.
+      m_sema->RequireCompleteType(ArgLoc.getLocation(), ArgTy,
+                                  diag::err_typecheck_decl_incomplete_type);
+      CXXRecordDecl *Decl = ArgTy->getAsCXXRecordDecl();
+      if (Decl && !Decl->isCompleteDefinition())
+        return true;
       // The node record type must be compound - error if it is not.
       if (GetTypeObjectKind(ArgTy) != AR_TOBJ_COMPOUND) {
         m_sema->Diag(ArgLoc.getLocation(), diag::err_hlsl_node_record_type)
@@ -5316,6 +5326,43 @@ class HLSLExternalSource : public ExternalSemaSource {
         return true;
       }
       return false;
+    } else if (Template->getTemplatedDecl()->hasAttr<HLSLTessPatchAttr>()) {
+      DXASSERT(TemplateArgList.size() > 0,
+               "Tessellation patch should have at least one template args");
+      const TemplateArgumentLoc &argLoc = TemplateArgList[0];
+      const TemplateArgument &arg = argLoc.getArgument();
+      DXASSERT(arg.getKind() == TemplateArgument::ArgKind::Type,
+               "Tessellation patch requires type template arg 0");
+
+      m_sema->RequireCompleteType(argLoc.getLocation(), arg.getAsType(),
+                                  diag::err_typecheck_decl_incomplete_type);
+      CXXRecordDecl *Decl = arg.getAsType()->getAsCXXRecordDecl();
+      if (Decl && !Decl->isCompleteDefinition())
+        return true;
+      if (ContainsLongVector(arg.getAsType())) {
+        m_sema->Diag(argLoc.getLocation(),
+                     diag::err_hlsl_unsupported_long_vector)
+            << "tessellation patches";
+        return true;
+      }
+    } else if (Template->getTemplatedDecl()->hasAttr<HLSLStreamOutputAttr>()) {
+      DXASSERT(TemplateArgList.size() > 0,
+               "Geometry streams should have at least one template args");
+      const TemplateArgumentLoc &argLoc = TemplateArgList[0];
+      const TemplateArgument &arg = argLoc.getArgument();
+      DXASSERT(arg.getKind() == TemplateArgument::ArgKind::Type,
+               "Geometry stream requires type template arg 0");
+      m_sema->RequireCompleteType(argLoc.getLocation(), arg.getAsType(),
+                                  diag::err_typecheck_decl_incomplete_type);
+      CXXRecordDecl *Decl = arg.getAsType()->getAsCXXRecordDecl();
+      if (Decl && !Decl->isCompleteDefinition())
+        return true;
+      if (ContainsLongVector(arg.getAsType())) {
+        m_sema->Diag(argLoc.getLocation(),
+                     diag::err_hlsl_unsupported_long_vector)
+            << "geometry streams";
+        return true;
+      }
     }
 
     bool isMatrix = Template->getCanonicalDecl() ==
@@ -5337,9 +5384,7 @@ class HLSLExternalSource : public ExternalSemaSource {
             // NOTE: IsValidTemplateArgumentType emits its own diagnostics
             return true;
           }
-          HLSLResourceAttr *ResAttr =
-              Template->getTemplatedDecl()->getAttr<HLSLResourceAttr>();
-          if (ResAttr && IsTyped((DXIL::ResourceKind)ResAttr->getResKind())) {
+          if (ResAttr && IsTyped(ResAttr->getResKind())) {
             // Check vectors for being too large.
             if (IsVectorType(m_sema, argType)) {
               unsigned NumElt = hlsl::GetElementCount(argType);
@@ -5368,17 +5413,16 @@ class HLSLExternalSource : public ExternalSemaSource {
           llvm::APSInt constantResult;
           if (expr != nullptr &&
               expr->isIntegerConstantExpr(constantResult, *m_context)) {
-            if (CheckRangedTemplateArgument(argSrcLoc, constantResult)) {
+            if (CheckRangedTemplateArgument(argSrcLoc, constantResult,
+                                            isVector))
               return true;
-            }
           }
         }
       } else if (arg.getKind() == TemplateArgument::ArgKind::Integral) {
         if (isMatrix || isVector) {
           llvm::APSInt Val = arg.getAsIntegral();
-          if (CheckRangedTemplateArgument(argSrcLoc, Val)) {
+          if (CheckRangedTemplateArgument(argSrcLoc, Val, isVector))
             return true;
-          }
         }
       }
     }
@@ -6631,6 +6675,9 @@ bool HLSLExternalSource::MatchArguments(
     }
   }
 
+  std::string profile = m_sema->getLangOpts().HLSLProfile;
+  const ShaderModel *SM = hlsl::ShaderModel::GetByName(profile.c_str());
+
   // Populate argTypes.
   for (size_t i = 0; i <= Args.size(); i++) {
     const HLSL_INTRINSIC_ARGUMENT *pArgument = &pIntrinsic->pArgs[i];
@@ -6801,8 +6848,9 @@ bool HLSLExternalSource::MatchArguments(
       }
 
       // Verify that the final results are in bounds.
-      CAB(uCols > 0 && uCols <= MaxVectorSize && uRows > 0 &&
-              uRows <= MaxVectorSize,
+      CAB((uCols > 0 && uRows > 0 &&
+           ((uCols <= MaxVectorSize && uRows <= MaxVectorSize) ||
+            (SM->IsSM69Plus() && uRows == 1))),
           i);
 
       // Const
@@ -8571,6 +8619,10 @@ ExprResult HLSLExternalSource::LookupVectorMemberExprForHLSL(
     llvm_unreachable("Unknown VectorMemberAccessError value");
   }
 
+
+  if (colCount > 4)
+    msg = diag::err_hlsl_vector_member_on_long_vector;
+
   if (msg != 0) {
     m_sema->Diag(MemberLoc, msg) << memberText;
 
@@ -11581,10 +11633,17 @@ bool hlsl::DiagnoseNodeStructArgument(Sema *self, TemplateArgumentLoc ArgLoc,
   HLSLExternalSource *source = HLSLExternalSource::FromSema(self);
   ArTypeObjectKind shapeKind = source->GetTypeObjectKind(ArgTy);
   switch (shapeKind) {
+  case AR_TOBJ_VECTOR:
+    if (GetHLSLVecSize(ArgTy) > DXIL::kDefaultMaxVectorLength) {
+      self->Diag(ArgLoc.getLocation(), diag::err_hlsl_unsupported_long_vector)
+          << "node records";
+      Empty = false;
+      return false;
+    }
+    LLVM_FALLTHROUGH;
   case AR_TOBJ_ARRAY:
   case AR_TOBJ_BASIC:
   case AR_TOBJ_MATRIX:
-  case AR_TOBJ_VECTOR:
     Empty = false;
     return false;
   case AR_TOBJ_OBJECT:
@@ -11603,14 +11662,15 @@ bool hlsl::DiagnoseNodeStructArgument(Sema *self, TemplateArgumentLoc ArgLoc,
     bool ErrorFound = false;
     const RecordDecl *RD = ArgTy->getAs<RecordType>()->getDecl();
     // Check the fields of the RecordDecl
-    RecordDecl::field_iterator begin = RD->field_begin();
-    RecordDecl::field_iterator end = RD->field_end();
-    while (begin != end) {
-      const FieldDecl *FD = *begin;
+    for (auto *FD : RD->fields())
       ErrorFound |=
           DiagnoseNodeStructArgument(self, ArgLoc, FD->getType(), Empty, FD);
-      begin++;
-    }
+    if (RD->isCompleteDefinition())
+      if (auto *Child = dyn_cast<CXXRecordDecl>(RD))
+        // Walk up the inheritance chain and check base class fields
+        for (auto &B : Child->bases())
+          ErrorFound |=
+              DiagnoseNodeStructArgument(self, ArgLoc, B.getType(), Empty);
     return ErrorFound;
   }
   default:
@@ -12046,6 +12106,21 @@ bool hlsl::ShouldSkipNRVO(clang::Sema &sema, clang::QualType returnType,
   return false;
 }
 
+bool hlsl::ContainsLongVector(QualType qt) {
+  if (qt.isNull() || qt->isDependentType())
+    return false;
+
+  while (const ArrayType *Arr = qt->getAsArrayTypeUnsafe())
+    qt = Arr->getElementType();
+
+  if (CXXRecordDecl *Decl = qt->getAsCXXRecordDecl()) {
+    if (!Decl->isCompleteDefinition())
+      return false;
+    return Decl->hasHLSLLongVector();
+  }
+  return false;
+}
+
 bool hlsl::IsConversionToLessOrEqualElements(
     clang::Sema *self, const clang::ExprResult &sourceExpr,
     const clang::QualType &targetType, bool explicitConversion) {
@@ -14369,6 +14444,7 @@ bool Sema::DiagnoseHLSLDecl(Declarator &D, DeclContext *DC, Expr *BitWidth,
                        *pDispatchGrid = nullptr, *pMaxDispatchGrid = nullptr;
   bool usageIn = false;
   bool usageOut = false;
+  bool isGroupShared = false;
 
   for (clang::AttributeList *pAttr = D.getDeclSpec().getAttributes().getList();
        pAttr != NULL; pAttr = pAttr->getNext()) {
@@ -14392,6 +14468,7 @@ bool Sema::DiagnoseHLSLDecl(Declarator &D, DeclContext *DC, Expr *BitWidth,
       }
       break;
     case AttributeList::AT_HLSLGroupShared:
+      isGroupShared = true;
       if (!isGlobal) {
         Diag(pAttr->getLoc(), diag::err_hlsl_varmodifierna)
             << pAttr->getName() << declarationType << pAttr->getRange();
@@ -14672,6 +14749,22 @@ bool Sema::DiagnoseHLSLDecl(Declarator &D, DeclContext *DC, Expr *BitWidth,
     result = false;
   }
 
+  // Disallow long vecs from $Global cbuffers.
+  if (isGlobal && !isStatic && !isGroupShared) {
+    // Suppress actual emitting of errors for incompletable types here
+    // They are redundant to those produced in ActOnUninitializedDecl.
+    struct SilentDiagnoser : public TypeDiagnoser {
+      SilentDiagnoser() : TypeDiagnoser(true) {}
+      virtual void diagnose(Sema &S, SourceLocation Loc, QualType T) {}
+    } SD;
+    RequireCompleteType(D.getLocStart(), qt, SD);
+    if (ContainsLongVector(qt)) {
+      Diag(D.getLocStart(), diag::err_hlsl_unsupported_long_vector)
+          << "cbuffers or tbuffers";
+      result = false;
+    }
+  }
+
   // SPIRV change starts
 #ifdef ENABLE_SPIRV_CODEGEN
   // Validate that Vulkan specific feature is only used when targeting SPIR-V
@@ -15560,6 +15653,16 @@ static bool isRelatedDeclMarkedNointerpolation(Expr *E) {
   return false;
 }
 
+// Verify that user-defined intrinsic struct args contain no long vectors
+static bool CheckUDTIntrinsicArg(Sema *S, Expr *Arg) {
+  if (ContainsLongVector(Arg->getType())) {
+    S->Diag(Arg->getExprLoc(), diag::err_hlsl_unsupported_long_vector)
+        << "user-defined struct parameter";
+    return true;
+  }
+  return false;
+}
+
 static bool CheckIntrinsicGetAttributeAtVertex(Sema *S, FunctionDecl *FDecl,
                                                CallExpr *TheCall) {
   assert(TheCall->getNumArgs() > 0);
@@ -15577,6 +15680,12 @@ static bool CheckIntrinsicGetAttributeAtVertex(Sema *S, FunctionDecl *FDecl,
 bool Sema::CheckHLSLIntrinsicCall(FunctionDecl *FDecl, CallExpr *TheCall) {
   auto attr = FDecl->getAttr<HLSLIntrinsicAttr>();
 
+  if (!attr)
+    return false;
+
+  if (!IsBuiltinTable(attr->getGroup()))
+    return false;
+
   switch (hlsl::IntrinsicOp(attr->getOpcode())) {
   case hlsl::IntrinsicOp::IOP_GetAttributeAtVertex:
     // See #hlsl-specs/issues/181. Feature is broken. For SPIR-V we want
@@ -15588,6 +15697,22 @@ bool Sema::CheckHLSLIntrinsicCall(FunctionDecl *FDecl, CallExpr *TheCall) {
     // existing ones. See the ExtensionTest.EvalAttributeCollision test.
     assert(FDecl->getName() == "GetAttributeAtVertex");
     return CheckIntrinsicGetAttributeAtVertex(this, FDecl, TheCall);
+  case hlsl::IntrinsicOp::IOP_DispatchMesh:
+    assert(TheCall->getNumArgs() > 3);
+    assert(FDecl->getName() == "DispatchMesh");
+    return CheckUDTIntrinsicArg(this, TheCall->getArg(3)->IgnoreCasts());
+  case hlsl::IntrinsicOp::IOP_CallShader:
+    assert(TheCall->getNumArgs() > 1);
+    assert(FDecl->getName() == "CallShader");
+    return CheckUDTIntrinsicArg(this, TheCall->getArg(1)->IgnoreCasts());
+  case hlsl::IntrinsicOp::IOP_TraceRay:
+    assert(TheCall->getNumArgs() > 7);
+    assert(FDecl->getName() == "TraceRay");
+    return CheckUDTIntrinsicArg(this, TheCall->getArg(7)->IgnoreCasts());
+  case hlsl::IntrinsicOp::IOP_ReportHit:
+    assert(TheCall->getNumArgs() > 2);
+    assert(FDecl->getName() == "ReportHit");
+    return CheckUDTIntrinsicArg(this, TheCall->getArg(2)->IgnoreCasts());
   default:
     break;
   }
@@ -16268,6 +16393,19 @@ void DiagnoseEntry(Sema &S, FunctionDecl *FD) {
     return;
   }
 
+  // Check general parameter characteristics
+  // Would be nice to check for resources here as they crash the compiler now.
+  // See issue #7186.
+  for (const auto *param : FD->params()) {
+    if (ContainsLongVector(param->getType()))
+      S.Diag(param->getLocation(), diag::err_hlsl_unsupported_long_vector)
+          << "entry function parameters";
+  }
+
+  if (ContainsLongVector(FD->getReturnType()))
+    S.Diag(FD->getLocation(), diag::err_hlsl_unsupported_long_vector)
+        << "entry function return type";
+
   DXIL::ShaderKind Stage =
       ShaderModel::KindFromFullName(shaderAttr->getStage());
   llvm::StringRef StageName = shaderAttr->getStage();
diff --git a/tools/clang/lib/Sema/SemaHLSLDiagnoseTU.cpp b/tools/clang/lib/Sema/SemaHLSLDiagnoseTU.cpp
index cf5d741541..a11f72b306 100644
--- a/tools/clang/lib/Sema/SemaHLSLDiagnoseTU.cpp
+++ b/tools/clang/lib/Sema/SemaHLSLDiagnoseTU.cpp
@@ -520,6 +520,16 @@ void hlsl::DiagnoseTranslationUnit(clang::Sema *self) {
               << hullPatchCount.value();
         }
       }
+      for (const auto *param : pPatchFnDecl->params())
+        if (ContainsLongVector(param->getType()))
+          self->Diag(param->getLocation(),
+                     diag::err_hlsl_unsupported_long_vector)
+              << "patch constant function parameters";
+
+      if (ContainsLongVector(pPatchFnDecl->getReturnType()))
+        self->Diag(pPatchFnDecl->getLocation(),
+                   diag::err_hlsl_unsupported_long_vector)
+            << "patch constant function return type";
     }
 
     DXIL::ShaderKind EntrySK = shaderModel->GetKind();
diff --git a/tools/clang/lib/Sema/SemaTemplateInstantiate.cpp b/tools/clang/lib/Sema/SemaTemplateInstantiate.cpp
index a6ae05faa5..1eacedbb0b 100644
--- a/tools/clang/lib/Sema/SemaTemplateInstantiate.cpp
+++ b/tools/clang/lib/Sema/SemaTemplateInstantiate.cpp
@@ -2139,6 +2139,18 @@ Sema::InstantiateClass(SourceLocation PointOfInstantiation,
               SourceLocation(), SourceLocation(), nullptr);
   CheckCompletedCXXClass(Instantiation);
 
+  // HLSL Change Begin - set longvec bit for vectors of over 4 elements
+  ClassTemplateSpecializationDecl *Spec =
+      dyn_cast<ClassTemplateSpecializationDecl>(Instantiation);
+  if (Spec && Spec->hasAttr<HLSLVectorAttr>()) {
+    const TemplateArgumentList &argList = Spec->getTemplateArgs();
+    const TemplateArgument &arg1 = argList[1];
+    llvm::APSInt vecSize = arg1.getAsIntegral();
+    if (vecSize.getLimitedValue() > hlsl::DXIL::kDefaultMaxVectorLength)
+      Instantiation->setHasHLSLLongVector();
+  }
+  // HLSL Change End - set longvec bit for vectors of over 4 elements
+
   // Default arguments are parsed, if not instantiated. We can go instantiate
   // default arg exprs for default constructors if necessary now.
   ActOnFinishCXXMemberDefaultArgs(Instantiation);
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/intrinsics/buffer-load-stores-scalars.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/intrinsics/buffer-load-stores-scalars.hlsl
new file mode 100644
index 0000000000..03735cb968
--- /dev/null
+++ b/tools/clang/test/CodeGenDXIL/hlsl/intrinsics/buffer-load-stores-scalars.hlsl
@@ -0,0 +1,162 @@
+// RUN: %dxc -DTYPE=float    -T vs_6_6 %s | FileCheck %s
+// RUN: %dxc -DTYPE=bool     -T vs_6_6 %s | FileCheck %s --check-prefixes=CHECK,I1
+// RUN: %dxc -DTYPE=uint64_t -T vs_6_6 %s | FileCheck %s --check-prefixes=CHECK,I64
+// RUN: %dxc -DTYPE=double   -T vs_6_6 %s | FileCheck %s --check-prefixes=CHECK,F64
+
+// RUN: %dxc -DTYPE=float1    -T vs_6_6 %s | FileCheck %s
+// RUN: %dxc -DTYPE=bool1     -T vs_6_6 %s | FileCheck %s --check-prefixes=CHECK,I1
+// RUN: %dxc -DTYPE=uint64_t1 -T vs_6_6 %s | FileCheck %s --check-prefixes=CHECK,I64
+// RUN: %dxc -DTYPE=double1   -T vs_6_6 %s | FileCheck %s --check-prefixes=CHECK,F64
+
+// Confirm that 6.9 doesn't use vector loads for scalars and vec1s
+// RUN: %dxc -DTYPE=float    -T vs_6_9 %s | FileCheck %s
+// RUN: %dxc -DTYPE=bool     -T vs_6_9 %s | FileCheck %s --check-prefixes=CHECK,I1
+// RUN: %dxc -DTYPE=uint64_t -T vs_6_9 %s | FileCheck %s --check-prefixes=CHECK,I64
+// RUN: %dxc -DTYPE=double   -T vs_6_9 %s | FileCheck %s --check-prefixes=CHECK,F64
+
+// RUN: %dxc -DTYPE=float1    -T vs_6_9 %s | FileCheck %s
+// RUiN: %dxc -DTYPE=bool1     -T vs_6_9 %s | FileCheck %s --check-prefixes=CHECK,I1
+// RUN: %dxc -DTYPE=uint64_t1 -T vs_6_9 %s | FileCheck %s --check-prefixes=CHECK,I64
+// RUN: %dxc -DTYPE=double1   -T vs_6_9 %s | FileCheck %s --check-prefixes=CHECK,F64
+
+///////////////////////////////////////////////////////////////////////
+// Test codegen for various load and store operations and conversions
+//  for different scalar buffer types and confirm that the proper
+//  loads, stores, and conversion operations take place.
+///////////////////////////////////////////////////////////////////////
+
+
+// These -DAGs must match the same line. That is the only reason for the -DAG.
+// The first match will assign [[TY]] to the native type
+// For most runs, the second match will assign [[TY32]] to the same thing.
+// For 64-bit types, the memory representation is i32 and a separate variable is needed.
+// For these cases, there is another line that will always match i32.
+// This line will also force the previous -DAGs to match the same line since the most
+// This shader can produce is two ResRet types.
+// CHECK-DAG: %dx.types.ResRet.[[TY:[a-z][0-9][0-9]]] = type { [[TYPE:[a-z0-9]*]],
+// CHECK-DAG: %dx.types.ResRet.[[TY32:[a-z][0-9][0-9]]] = type { [[TYPE]],
+// I64: %dx.types.ResRet.[[TY32:i32]]
+// F64: %dx.types.ResRet.[[TY32:i32]]
+
+  ByteAddressBuffer RoByBuf : register(t1);
+RWByteAddressBuffer RwByBuf : register(u1);
+
+  StructuredBuffer< TYPE > RoStBuf : register(t2);
+RWStructuredBuffer< TYPE > RwStBuf : register(u2);
+
+  Buffer< TYPE > RoTyBuf : register(t3);
+RWBuffer< TYPE > RwTyBuf : register(u3);
+
+ConsumeStructuredBuffer<TYPE> CnStBuf : register(u4);
+AppendStructuredBuffer<TYPE> ApStBuf  : register(u5);
+
+void main(uint ix[2] : IX) {
+  // ByteAddressBuffer Tests
+
+  // CHECK-DAG: [[HDLROBY:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 1, i32 1, i32 0, i8 0 }, i32 1, i1 false)
+  // CHECK-DAG: [[HDLRWBY:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 1, i32 1, i32 0, i8 1 }, i32 1, i1 false)
+
+  // CHECK-DAG: [[HDLROST:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 2, i32 2, i32 0, i8 0 }, i32 2, i1 false)
+  // CHECK-DAG: [[HDLRWST:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 2, i32 2, i32 0, i8 1 }, i32 2, i1 false)
+
+  // CHECK-DAG: [[HDLROTY:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 3, i32 3, i32 0, i8 0 }, i32 3, i1 false)
+  // CHECK-DAG: [[HDLRWTY:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 3, i32 3, i32 0, i8 1 }, i32 3, i1 false)
+
+  // CHECK-DAG: [[HDLCON:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 4, i32 4, i32 0, i8 1 }, i32 4, i1 false)
+  // CHECK-DAG: [[HDLAPP:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 5, i32 5, i32 0, i8 1 }, i32 5, i1 false)
+
+  // CHECK: [[IX0:%.*]] = call i32 @dx.op.loadInput.i32(i32 4,
+
+  // CHECK: [[ANHDLRWBY:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLRWBY]]
+  // CHECK: call %dx.types.ResRet.[[TY]] @dx.op.rawBufferLoad.[[TY]](i32 139, %dx.types.Handle [[ANHDLRWBY]], i32 [[IX0]]
+  // I1: icmp ne i32 %{{.*}}, 0
+  TYPE babElt1 = RwByBuf.Load< TYPE >(ix[0]);
+
+  // CHECK: [[ANHDLROBY:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLROBY]]
+  // CHECK: call %dx.types.ResRet.[[TY]] @dx.op.rawBufferLoad.[[TY]](i32 139, %dx.types.Handle [[ANHDLROBY]], i32 [[IX0]]
+  // I1: icmp ne i32 %{{.*}}, 0
+  TYPE babElt2 = RoByBuf.Load< TYPE >(ix[0]);
+
+  // I1: zext i1 %{{.*}} to i32
+  // CHECK: all void @dx.op.rawBufferStore.[[TY]](i32 140, %dx.types.Handle [[ANHDLRWBY]], i32 [[IX0]]
+  RwByBuf.Store< TYPE >(ix[0], babElt1 + babElt2);
+
+  // StructuredBuffer Tests
+  // CHECK: [[ANHDLRWST:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLRWST]]
+  // CHECK: call %dx.types.ResRet.[[TY]] @dx.op.rawBufferLoad.[[TY]](i32 139, %dx.types.Handle [[ANHDLRWST]], i32 [[IX0]]
+  // I1: icmp ne i32 %{{.*}}, 0
+  TYPE stbElt1 = RwStBuf.Load(ix[0]);
+  // CHECK: [[IX1:%.*]] = call i32 @dx.op.loadInput.i32(i32 4,
+  // CHECK: call %dx.types.ResRet.[[TY]] @dx.op.rawBufferLoad.[[TY]](i32 139, %dx.types.Handle [[ANHDLRWST]], i32 [[IX1]]
+  // I1: icmp ne i32 %{{.*}}, 0
+  TYPE stbElt2 = RwStBuf[ix[1]];
+
+  // CHECK: [[ANHDLROST:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLROST]]
+  // CHECK: call %dx.types.ResRet.[[TY]] @dx.op.rawBufferLoad.[[TY]](i32 139, %dx.types.Handle [[ANHDLROST]], i32 [[IX0]]
+  // I1: icmp ne i32 %{{.*}}, 0
+  TYPE stbElt3 = RoStBuf.Load(ix[0]);
+  // CHECK: call %dx.types.ResRet.[[TY]] @dx.op.rawBufferLoad.[[TY]](i32 139, %dx.types.Handle [[ANHDLROST]], i32 [[IX1]]
+  // I1: icmp ne i32 %{{.*}}, 0
+  TYPE stbElt4 = RoStBuf[ix[1]];
+
+  // I1: zext i1 %{{.*}} to i32
+  // CHECK: all void @dx.op.rawBufferStore.[[TY]](i32 140, %dx.types.Handle [[ANHDLRWST]], i32 [[IX0]]
+  RwStBuf[ix[0]] = stbElt1 + stbElt2 + stbElt3 + stbElt4;
+
+  // {Append/Consume}StructuredBuffer Tests
+  // CHECK: [[ANHDLCON:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLCON]]
+  // CHECK: [[CONIX:%.*]] = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle [[ANHDLCON]], i8 -1)
+  // CHECK: call %dx.types.ResRet.[[TY]] @dx.op.rawBufferLoad.[[TY]](i32 139, %dx.types.Handle [[ANHDLCON]], i32 [[CONIX]]
+  // I1: icmp ne i32 %{{.*}}, 0
+  TYPE cnElt = CnStBuf.Consume();
+
+  // CHECK: [[ANHDLAPP:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLAPP]]
+  // CHECK: [[APPIX:%.*]] = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle [[ANHDLAPP]], i8 1)
+  // I1: zext i1 %{{.*}} to i32
+  // CHECK: all void @dx.op.rawBufferStore.[[TY]](i32 140, %dx.types.Handle [[ANHDLAPP]], i32 [[APPIX]]
+  ApStBuf.Append(cnElt);
+
+  // TypedBuffer Tests
+  // CHECK: [[ANHDLRWTY:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLRWTY]]
+  // CHECK: call %dx.types.ResRet.[[TY32]] @dx.op.bufferLoad.[[TY32]](i32 68, %dx.types.Handle [[ANHDLRWTY]], i32 [[IX0]]
+  // F64: call double @dx.op.makeDouble.f64(i32 101
+  // I64: zext i32 %{{.*}} to i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: shl nuw i64
+  // I64: or i64
+  // I1: icmp ne i32 %{{.*}}, 0
+  TYPE typElt1 = RwTyBuf.Load(ix[0]);
+  // CHECK: call %dx.types.ResRet.[[TY32]] @dx.op.bufferLoad.[[TY32]](i32 68, %dx.types.Handle [[ANHDLRWTY]], i32 [[IX1]]
+  // F64: call double @dx.op.makeDouble.f64(i32 101
+  // I64: zext i32 %{{.*}} to i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: shl nuw i64
+  // I64: or i64
+  // I1: icmp ne i32 %{{.*}}, 0
+  TYPE typElt2 = RwTyBuf[ix[1]];
+  // CHECK: [[ANHDLROTY:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLROTY]]
+  // CHECK: call %dx.types.ResRet.[[TY32]] @dx.op.bufferLoad.[[TY32]](i32 68, %dx.types.Handle [[ANHDLROTY]], i32 [[IX0]]
+  // F64: call double @dx.op.makeDouble.f64(i32 101
+  // I64: zext i32 %{{.*}} to i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: shl nuw i64
+  // I64: or i64
+  // I1: icmp ne i32 %{{.*}}, 0
+  TYPE typElt3 = RoTyBuf.Load(ix[0]);
+  // CHECK: call %dx.types.ResRet.[[TY32]] @dx.op.bufferLoad.[[TY32]](i32 68, %dx.types.Handle [[ANHDLROTY]], i32 [[IX1]]
+  // F64: call double @dx.op.makeDouble.f64(i32 101
+  // I64: zext i32 %{{.*}} to i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: shl nuw i64
+  // I64: or i64
+  // I1: icmp ne i32 %{{.*}}, 0
+  TYPE typElt4 = RoTyBuf[ix[1]];
+
+  // F64: call %dx.types.splitdouble @dx.op.splitDouble.f64(i32 102
+  // I64: trunc i64 %{{.*}} to i32
+  // I64: lshr i64  %{{.*}}, 32
+  // I64: trunc i64 %{{.*}} to i32
+  // I1: zext i1 %{{.*}} to i32
+  // CHECK: all void @dx.op.bufferStore.[[TY32]](i32 69, %dx.types.Handle [[ANHDLRWTY]], i32 [[IX0]]
+  RwTyBuf[ix[0]] = typElt1 + typElt2 + typElt3 + typElt4;
+}
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/intrinsics/buffer-load-stores-sm69.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/intrinsics/buffer-load-stores-sm69.hlsl
new file mode 100644
index 0000000000..8df8a36cd7
--- /dev/null
+++ b/tools/clang/test/CodeGenDXIL/hlsl/intrinsics/buffer-load-stores-sm69.hlsl
@@ -0,0 +1,139 @@
+// RUN: %dxc -DTYPE=float -DNUM=4    -T vs_6_9 %s | FileCheck %s
+// RUN: %dxc -DTYPE=bool -DNUM=4     -T vs_6_9 %s | FileCheck %s --check-prefixes=CHECK,I1
+// RUN: %dxc -DTYPE=uint8_t4_packed -DNUM=4     -T vs_6_9 %s | FileCheck %s --check-prefixes=CHECK
+
+// 64-bit types require operation/intrinsic support to convert the values to/from the i32 memory representations.
+// RUN: %dxc -DTYPE=uint64_t -DNUM=2 -T vs_6_9 %s | FileCheck %s --check-prefixes=CHECK,I64
+// RUN: %dxc -DTYPE=double -DNUM=2   -T vs_6_9 %s | FileCheck %s --check-prefixes=CHECK,F64
+
+///////////////////////////////////////////////////////////////////////
+// Test codegen for various load and store operations and conversions
+//  for different scalar/vector buffer types and indices.
+///////////////////////////////////////////////////////////////////////
+
+// CHECK-DAG: %dx.types.ResRet.[[VTY:v[0-9]*[a-z][0-9][0-9]]] = type { [[VTYPE:<[a-z 0-9]*>]],
+// CHECK-DAG: %dx.types.ResRet.[[TY:[if][0-9][0-9]]] = type
+// CHECK: %"class.StructuredBuffer<vector<{{.*}}, [[NUM:[0-9]*]]>
+
+  ByteAddressBuffer RoByBuf : register(t1);
+RWByteAddressBuffer RwByBuf : register(u1);
+
+StructuredBuffer< vector<TYPE, NUM> > RoStBuf : register(t2);
+RWStructuredBuffer< vector<TYPE, NUM>  > RwStBuf : register(u2);
+
+  Buffer< vector<TYPE, NUM>  > RoTyBuf : register(t3);
+RWBuffer< vector<TYPE, NUM>  > RwTyBuf : register(u3);
+
+ConsumeStructuredBuffer<vector<TYPE, NUM> > CnStBuf : register(u4);
+AppendStructuredBuffer<vector<TYPE, NUM> > ApStBuf  : register(u5);
+
+void main(uint ix[2] : IX) {
+  // ByteAddressBuffer Tests
+
+  // CHECK-DAG: [[HDLROBY:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 1, i32 1, i32 0, i8 0 }, i32 1, i1 false)
+  // CHECK-DAG: [[HDLRWBY:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 1, i32 1, i32 0, i8 1 }, i32 1, i1 false)
+
+  // CHECK-DAG: [[HDLROST:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 2, i32 2, i32 0, i8 0 }, i32 2, i1 false)
+  // CHECK-DAG: [[HDLRWST:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 2, i32 2, i32 0, i8 1 }, i32 2, i1 false)
+
+  // CHECK-DAG: [[HDLROTY:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 3, i32 3, i32 0, i8 0 }, i32 3, i1 false)
+  // CHECK-DAG: [[HDLRWTY:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 3, i32 3, i32 0, i8 1 }, i32 3, i1 false)
+
+  // CHECK-DAG: [[HDLCON:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 4, i32 4, i32 0, i8 1 }, i32 4, i1 false)
+  // CHECK-DAG: [[HDLAPP:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 5, i32 5, i32 0, i8 1 }, i32 5, i1 false)
+
+  // CHECK: [[IX0:%.*]] = call i32 @dx.op.loadInput.i32(i32 4,
+
+  // CHECK: [[ANHDLRWBY:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLRWBY]]
+  // CHECK: call %dx.types.ResRet.[[VTY]] @dx.op.rawBufferVectorLoad.[[VTY]](i32 303, %dx.types.Handle [[ANHDLRWBY]], i32 [[IX0]]
+  // I1: icmp ne [[VTYPE]] %{{.*}}, zeroinitializer
+  vector<TYPE, NUM>  babElt1 = RwByBuf.Load< vector<TYPE, NUM>  >(ix[0]);
+
+  // CHECK: [[ANHDLROBY:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLROBY]]
+  // CHECK: call %dx.types.ResRet.[[VTY]] @dx.op.rawBufferVectorLoad.[[VTY]](i32 303, %dx.types.Handle [[ANHDLROBY]], i32 [[IX0]]
+  // I1: icmp ne [[VTYPE]] %{{.*}}, zeroinitializer
+  vector<TYPE, NUM>  babElt2 = RoByBuf.Load< vector<TYPE, NUM>  >(ix[0]);
+
+  // I1: zext <[[NUM]] x i1> %{{.*}} to [[VTYPE]]
+  // CHECK: all void @dx.op.rawBufferVectorStore.[[VTY]](i32 304, %dx.types.Handle [[ANHDLRWBY]], i32 [[IX0]]
+  RwByBuf.Store< vector<TYPE, NUM>  >(ix[0], babElt1 + babElt2);
+
+  // StructuredBuffer Tests
+  // CHECK: [[ANHDLRWST:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLRWST]]
+  // CHECK: call %dx.types.ResRet.[[VTY]] @dx.op.rawBufferVectorLoad.[[VTY]](i32 303, %dx.types.Handle [[ANHDLRWST]], i32 [[IX0]]
+  // I1: icmp ne [[VTYPE]] %{{.*}}, zeroinitializer
+  vector<TYPE, NUM>  stbElt1 = RwStBuf.Load(ix[0]);
+  // CHECK: [[IX1:%.*]] = call i32 @dx.op.loadInput.i32(i32 4,
+  // CHECK: call %dx.types.ResRet.[[VTY]] @dx.op.rawBufferVectorLoad.[[VTY]](i32 303, %dx.types.Handle [[ANHDLRWST]], i32 [[IX1]]
+  // I1: icmp ne [[VTYPE]] %{{.*}}, zeroinitializer
+  vector<TYPE, NUM>  stbElt2 = RwStBuf[ix[1]];
+
+  // CHECK: [[ANHDLROST:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLROST]]
+  // CHECK: call %dx.types.ResRet.[[VTY]] @dx.op.rawBufferVectorLoad.[[VTY]](i32 303, %dx.types.Handle [[ANHDLROST]], i32 [[IX0]]
+  // I1: icmp ne [[VTYPE]] %{{.*}}, zeroinitializer
+  vector<TYPE, NUM>  stbElt3 = RoStBuf.Load(ix[0]);
+  // CHECK: call %dx.types.ResRet.[[VTY]] @dx.op.rawBufferVectorLoad.[[VTY]](i32 303, %dx.types.Handle [[ANHDLROST]], i32 [[IX1]]
+  // I1: icmp ne [[VTYPE]] %{{.*}}, zeroinitializer
+  vector<TYPE, NUM>  stbElt4 = RoStBuf[ix[1]];
+
+  // I1: zext <[[NUM]] x i1> %{{.*}} to [[VTYPE]]
+  // CHECK: all void @dx.op.rawBufferVectorStore.[[VTY]](i32 304, %dx.types.Handle [[ANHDLRWST]], i32 [[IX0]]
+  RwStBuf[ix[0]] = stbElt1 + stbElt2 + stbElt3 + stbElt4;
+
+  // {Append/Consume}StructuredBuffer Tests
+  // CHECK: [[ANHDLCON:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLCON]]
+  // CHECK: [[CONIX:%.*]] = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle [[ANHDLCON]], i8 -1) 
+  // CHECK: call %dx.types.ResRet.[[VTY]] @dx.op.rawBufferVectorLoad.[[VTY]](i32 303, %dx.types.Handle [[ANHDLCON]], i32 [[CONIX]]
+  // I1: icmp ne [[VTYPE]] %{{.*}}, zeroinitializer
+  vector<TYPE, NUM>  cnElt = CnStBuf.Consume();
+
+  // CHECK: [[ANHDLAPP:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLAPP]]
+  // CHECK: [[APPIX:%.*]] = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle [[ANHDLAPP]], i8 1) 
+  // I1: zext <[[NUM]] x i1> %{{.*}} to [[VTYPE]]
+  // CHECK: all void @dx.op.rawBufferVectorStore.[[VTY]](i32 304, %dx.types.Handle [[ANHDLAPP]], i32 [[APPIX]]
+  ApStBuf.Append(cnElt);
+
+  // TypedBuffer Tests
+  // CHECK: [[ANHDLRWTY:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLRWTY]]
+  // CHECK: call %dx.types.ResRet.[[TY]] @dx.op.bufferLoad.[[TY]](i32 68, %dx.types.Handle [[ANHDLRWTY]], i32 [[IX0]]
+  // F64: call double @dx.op.makeDouble.f64(i32 101
+  // I64: zext i32 %{{.*}} to i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: shl nuw i64
+  // I64: or i64
+  // I1: icmp ne [[VTYPE]] %{{.*}}, zeroinitializer
+  vector<TYPE, NUM>  typElt1 = RwTyBuf.Load(ix[0]);
+  // CHECK: call %dx.types.ResRet.[[TY]] @dx.op.bufferLoad.[[TY]](i32 68, %dx.types.Handle [[ANHDLRWTY]], i32 [[IX1]]
+  // F64: call double @dx.op.makeDouble.f64(i32 101
+  // I64: zext i32 %{{.*}} to i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: shl nuw i64
+  // I64: or i64
+  // I1: icmp ne [[VTYPE]] %{{.*}}, zeroinitializer
+  vector<TYPE, NUM>  typElt2 = RwTyBuf[ix[1]];
+  // CHECK: [[ANHDLROTY:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLROTY]]
+  // CHECK: call %dx.types.ResRet.[[TY]] @dx.op.bufferLoad.[[TY]](i32 68, %dx.types.Handle [[ANHDLROTY]], i32 [[IX0]]
+  // F64: call double @dx.op.makeDouble.f64(i32 101
+  // I64: zext i32 %{{.*}} to i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: shl nuw i64
+  // I64: or i64
+  // I1: icmp ne [[VTYPE]] %{{.*}}, zeroinitializer
+  vector<TYPE, NUM>  typElt3 = RoTyBuf.Load(ix[0]);
+  // CHECK: call %dx.types.ResRet.[[TY]] @dx.op.bufferLoad.[[TY]](i32 68, %dx.types.Handle [[ANHDLROTY]], i32 [[IX1]]
+  // F64: call double @dx.op.makeDouble.f64(i32 101
+  // I64: zext i32 %{{.*}} to i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: shl nuw i64
+  // I64: or i64
+  // I1: icmp ne [[VTYPE]] %{{.*}}, zeroinitializer
+  vector<TYPE, NUM>  typElt4 = RoTyBuf[ix[1]];
+
+  // F64: call %dx.types.splitdouble @dx.op.splitDouble.f64(i32 102
+  // I64: trunc i64 %{{.*}} to i32
+  // lshr i64  %{{.*}}, 32
+  // I64: trunc i64 %{{.*}} to i32
+  // I1: zext <[[NUM]] x i1> %{{.*}} to [[VTYPE]]
+  // CHECK: all void @dx.op.bufferStore.[[TY]](i32 69, %dx.types.Handle [[ANHDLRWTY]], i32 [[IX0]]
+  RwTyBuf[ix[0]] = typElt1 + typElt2 + typElt3 + typElt4;
+}
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/intrinsics/buffer-load-stores.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/intrinsics/buffer-load-stores.hlsl
index ea44fef604..8dcf5ead1c 100644
--- a/tools/clang/test/CodeGenDXIL/hlsl/intrinsics/buffer-load-stores.hlsl
+++ b/tools/clang/test/CodeGenDXIL/hlsl/intrinsics/buffer-load-stores.hlsl
@@ -27,13 +27,20 @@ RWByteAddressBuffer RwByBuf : register(u1);
   StructuredBuffer< TYPE > RoStBuf : register(t2);
 RWStructuredBuffer< TYPE > RwStBuf : register(u2);
 
-  Buffer< TYPE > RoTyBuf : register(t3);
-RWBuffer< TYPE > RwTyBuf : register(u3);
+ConsumeStructuredBuffer<TYPE> CnStBuf : register(u3);
+AppendStructuredBuffer<TYPE> ApStBuf  : register(u4);
 
-ConsumeStructuredBuffer<TYPE> CnStBuf : register(u4);
-AppendStructuredBuffer<TYPE> ApStBuf  : register(u5);
+  Buffer< TYPE > RoTyBuf : register(t5);
+RWBuffer< TYPE > RwTyBuf : register(u5);
 
-void main(uint ix[2] : IX) {
+  Texture1D< TYPE > RoTex1d : register(t6);
+RWTexture1D< TYPE > RwTex1d : register(u6);
+  Texture2D< TYPE > RoTex2d : register(t7);
+RWTexture2D< TYPE > RwTex2d : register(u7);
+  Texture3D< TYPE > RoTex3d : register(t8);
+RWTexture3D< TYPE > RwTex3d : register(u8);
+
+void main(uint ix0 : IX0, uint ix1 : IX1, uint2 ix2 : IX2, uint3 ix3 : IX3) {
   // ByteAddressBuffer Tests
 
   // CHECK-DAG: [[HDLROBY:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 1, i32 1, i32 0, i8 0 }, i32 1, i1 false)
@@ -42,13 +49,27 @@ void main(uint ix[2] : IX) {
   // CHECK-DAG: [[HDLROST:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 2, i32 2, i32 0, i8 0 }, i32 2, i1 false)
   // CHECK-DAG: [[HDLRWST:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 2, i32 2, i32 0, i8 1 }, i32 2, i1 false)
 
-  // CHECK-DAG: [[HDLROTY:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 3, i32 3, i32 0, i8 0 }, i32 3, i1 false)
-  // CHECK-DAG: [[HDLRWTY:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 3, i32 3, i32 0, i8 1 }, i32 3, i1 false)
+  // CHECK-DAG: [[HDLCON:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 3, i32 3, i32 0, i8 1 }, i32 3, i1 false)
+  // CHECK-DAG: [[HDLAPP:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 4, i32 4, i32 0, i8 1 }, i32 4, i1 false)
+
+  // CHECK-DAG: [[HDLROTY:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 5, i32 5, i32 0, i8 0 }, i32 5, i1 false)
+  // CHECK-DAG: [[HDLRWTY:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 5, i32 5, i32 0, i8 1 }, i32 5, i1 false)
 
-  // CHECK-DAG: [[HDLCON:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 4, i32 4, i32 0, i8 1 }, i32 4, i1 false)
-  // CHECK-DAG: [[HDLAPP:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 5, i32 5, i32 0, i8 1 }, i32 5, i1 false)
+  // CHECK-DAG: [[HDLROTX1:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 6, i32 6, i32 0, i8 0 }, i32 6, i1 false)
+  // CHECK-DAG: [[HDLRWTX1:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 6, i32 6, i32 0, i8 1 }, i32 6, i1 false)
+  // CHECK-DAG: [[HDLROTX2:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 7, i32 7, i32 0, i8 0 }, i32 7, i1 false)
+  // CHECK-DAG: [[HDLRWTX2:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 7, i32 7, i32 0, i8 1 }, i32 7, i1 false)
+  // CHECK-DAG: [[HDLROTX3:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 8, i32 8, i32 0, i8 0 }, i32 8, i1 false)
+  // CHECK-DAG: [[HDLRWTX3:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 8, i32 8, i32 0, i8 1 }, i32 8, i1 false)
 
-  // CHECK: [[IX0:%.*]] = call i32 @dx.op.loadInput.i32(i32 4,
+
+  // CHECK-DAG: [[IX0:%.*]] = call i32 @dx.op.loadInput.i32(i32 4, i32 0, i32 0, i8 0
+  // CHECK-DAG: [[IX1:%.*]] = call i32 @dx.op.loadInput.i32(i32 4, i32 1, i32 0, i8 0
+  // CHECK-DAG: [[IX20:%.*]] = call i32 @dx.op.loadInput.i32(i32 4, i32 2, i32 0, i8 0
+  // CHECK-DAG: [[IX21:%.*]] = call i32 @dx.op.loadInput.i32(i32 4, i32 2, i32 0, i8 1
+  // CHECK-DAG: [[IX30:%.*]] = call i32 @dx.op.loadInput.i32(i32 4, i32 3, i32 0, i8 0
+  // CHECK-DAG: [[IX31:%.*]] = call i32 @dx.op.loadInput.i32(i32 4, i32 3, i32 0, i8 1
+  // CHECK-DAG: [[IX32:%.*]] = call i32 @dx.op.loadInput.i32(i32 4, i32 3, i32 0, i8 2
 
   // CHECK: [[ANHDLRWBY:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLRWBY]]
   // CHECK: call %dx.types.ResRet.[[TY]] @dx.op.rawBufferLoad.[[TY]](i32 139, %dx.types.Handle [[ANHDLRWBY]], i32 [[IX0]]
@@ -56,7 +77,7 @@ void main(uint ix[2] : IX) {
   // I1: icmp ne i32 %{{.*}}, 0
   // I1: icmp ne i32 %{{.*}}, 0
   // I1: icmp ne i32 %{{.*}}, 0
-  TYPE babElt1 = RwByBuf.Load< TYPE >(ix[0]);
+  TYPE babElt1 = RwByBuf.Load< TYPE >(ix0);
 
   // CHECK: [[ANHDLROBY:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLROBY]]
   // CHECK: call %dx.types.ResRet.[[TY]] @dx.op.rawBufferLoad.[[TY]](i32 139, %dx.types.Handle [[ANHDLROBY]], i32 [[IX0]]
@@ -64,14 +85,14 @@ void main(uint ix[2] : IX) {
   // I1: icmp ne i32 %{{.*}}, 0
   // I1: icmp ne i32 %{{.*}}, 0
   // I1: icmp ne i32 %{{.*}}, 0
-  TYPE babElt2 = RoByBuf.Load< TYPE >(ix[0]);
+  TYPE babElt2 = RoByBuf.Load< TYPE >(ix0);
 
   // I1: zext i1 %{{.*}} to i32
   // I1: zext i1 %{{.*}} to i32
   // I1: zext i1 %{{.*}} to i32
   // I1: zext i1 %{{.*}} to i32
   // CHECK: all void @dx.op.rawBufferStore.[[TY]](i32 140, %dx.types.Handle [[ANHDLRWBY]], i32 [[IX0]]
-  RwByBuf.Store< TYPE >(ix[0], babElt1 + babElt2);
+  RwByBuf.Store< TYPE >(ix0, babElt1 + babElt2);
 
   // StructuredBuffer Tests
   // CHECK: [[ANHDLRWST:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLRWST]]
@@ -80,14 +101,13 @@ void main(uint ix[2] : IX) {
   // I1: icmp ne i32 %{{.*}}, 0
   // I1: icmp ne i32 %{{.*}}, 0
   // I1: icmp ne i32 %{{.*}}, 0
-  TYPE stbElt1 = RwStBuf.Load(ix[0]);
-  // CHECK: [[IX1:%.*]] = call i32 @dx.op.loadInput.i32(i32 4,
+  TYPE stbElt1 = RwStBuf.Load(ix0);
   // CHECK: call %dx.types.ResRet.[[TY]] @dx.op.rawBufferLoad.[[TY]](i32 139, %dx.types.Handle [[ANHDLRWST]], i32 [[IX1]]
   // I1: icmp ne i32 %{{.*}}, 0
   // I1: icmp ne i32 %{{.*}}, 0
   // I1: icmp ne i32 %{{.*}}, 0
   // I1: icmp ne i32 %{{.*}}, 0
-  TYPE stbElt2 = RwStBuf[ix[1]];
+  TYPE stbElt2 = RwStBuf[ix1];
 
   // CHECK: [[ANHDLROST:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLROST]]
   // CHECK: call %dx.types.ResRet.[[TY]] @dx.op.rawBufferLoad.[[TY]](i32 139, %dx.types.Handle [[ANHDLROST]], i32 [[IX0]]
@@ -95,20 +115,20 @@ void main(uint ix[2] : IX) {
   // I1: icmp ne i32 %{{.*}}, 0
   // I1: icmp ne i32 %{{.*}}, 0
   // I1: icmp ne i32 %{{.*}}, 0
-  TYPE stbElt3 = RoStBuf.Load(ix[0]);
+  TYPE stbElt3 = RoStBuf.Load(ix0);
   // CHECK: call %dx.types.ResRet.[[TY]] @dx.op.rawBufferLoad.[[TY]](i32 139, %dx.types.Handle [[ANHDLROST]], i32 [[IX1]]
   // I1: icmp ne i32 %{{.*}}, 0
   // I1: icmp ne i32 %{{.*}}, 0
   // I1: icmp ne i32 %{{.*}}, 0
   // I1: icmp ne i32 %{{.*}}, 0
-  TYPE stbElt4 = RoStBuf[ix[1]];
+  TYPE stbElt4 = RoStBuf[ix1];
 
   // I1: zext i1 %{{.*}} to i32
   // I1: zext i1 %{{.*}} to i32
   // I1: zext i1 %{{.*}} to i32
   // I1: zext i1 %{{.*}} to i32
   // CHECK: all void @dx.op.rawBufferStore.[[TY]](i32 140, %dx.types.Handle [[ANHDLRWST]], i32 [[IX0]]
-  RwStBuf[ix[0]] = stbElt1 + stbElt2 + stbElt3 + stbElt4;
+  RwStBuf[ix0] = stbElt1 + stbElt2 + stbElt3 + stbElt4;
 
   // {Append/Consume}StructuredBuffer Tests
   // CHECK: [[ANHDLCON:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLCON]]
@@ -146,7 +166,7 @@ void main(uint ix[2] : IX) {
   // I1: icmp ne i32 %{{.*}}, 0
   // I1: icmp ne i32 %{{.*}}, 0
   // I1: icmp ne i32 %{{.*}}, 0
-  TYPE typElt1 = RwTyBuf.Load(ix[0]);
+  TYPE typElt1 = RwTyBuf.Load(ix0);
   // CHECK: call %dx.types.ResRet.[[TY32]] @dx.op.bufferLoad.[[TY32]](i32 68, %dx.types.Handle [[ANHDLRWTY]], i32 [[IX1]]
   // F64: call double @dx.op.makeDouble.f64(i32 101
   // F64: call double @dx.op.makeDouble.f64(i32 101
@@ -162,7 +182,7 @@ void main(uint ix[2] : IX) {
   // I1: icmp ne i32 %{{.*}}, 0
   // I1: icmp ne i32 %{{.*}}, 0
   // I1: icmp ne i32 %{{.*}}, 0
-  TYPE typElt2 = RwTyBuf[ix[1]];
+  TYPE typElt2 = RwTyBuf[ix1];
   // CHECK: [[ANHDLROTY:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLROTY]]
   // CHECK: call %dx.types.ResRet.[[TY32]] @dx.op.bufferLoad.[[TY32]](i32 68, %dx.types.Handle [[ANHDLROTY]], i32 [[IX0]]
   // F64: call double @dx.op.makeDouble.f64(i32 101
@@ -179,7 +199,7 @@ void main(uint ix[2] : IX) {
   // I1: icmp ne i32 %{{.*}}, 0
   // I1: icmp ne i32 %{{.*}}, 0
   // I1: icmp ne i32 %{{.*}}, 0
-  TYPE typElt3 = RoTyBuf.Load(ix[0]);
+  TYPE typElt3 = RoTyBuf.Load(ix0);
   // CHECK: call %dx.types.ResRet.[[TY32]] @dx.op.bufferLoad.[[TY32]](i32 68, %dx.types.Handle [[ANHDLROTY]], i32 [[IX1]]
   // F64: call double @dx.op.makeDouble.f64(i32 101
   // F64: call double @dx.op.makeDouble.f64(i32 101
@@ -195,7 +215,7 @@ void main(uint ix[2] : IX) {
   // I1: icmp ne i32 %{{.*}}, 0
   // I1: icmp ne i32 %{{.*}}, 0
   // I1: icmp ne i32 %{{.*}}, 0
-  TYPE typElt4 = RoTyBuf[ix[1]];
+  TYPE typElt4 = RoTyBuf[ix1];
 
   // F64: call %dx.types.splitdouble @dx.op.splitDouble.f64(i32 102
   // F64: call %dx.types.splitdouble @dx.op.splitDouble.f64(i32 102
@@ -210,5 +230,126 @@ void main(uint ix[2] : IX) {
   // I1: zext i1 %{{.*}} to i32
   // I1: zext i1 %{{.*}} to i32
   // CHECK: all void @dx.op.bufferStore.[[TY32]](i32 69, %dx.types.Handle [[ANHDLRWTY]], i32 [[IX0]]
-  RwTyBuf[ix[0]] = typElt1 + typElt2 + typElt3 + typElt4;
+  RwTyBuf[ix0] = typElt1 + typElt2 + typElt3 + typElt4;
+
+  // Texture Tests
+  // CHECK: [[ANHDLROTX1:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLROTX1]]
+  // CHECK: call %dx.types.ResRet.[[TY32]] @dx.op.textureLoad.[[TY32]](i32 66, %dx.types.Handle [[ANHDLROTX1]], i32 0, i32 [[IX0]], i32 undef, i32 undef
+  // F64: call double @dx.op.makeDouble.f64(i32 101
+  // F64: call double @dx.op.makeDouble.f64(i32 101
+  // I64: zext i32 %{{.*}} to i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: shl nuw i64
+  // I64: or i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: shl nuw i64
+  // I64: or i64
+  // I1: icmp ne i32 %{{.*}}, 0
+  // I1: icmp ne i32 %{{.*}}, 0
+  // I1: icmp ne i32 %{{.*}}, 0
+  // I1: icmp ne i32 %{{.*}}, 0
+  TYPE texElt1 = RoTex1d[ix0];
+  // CHECK: [[ANHDLRWTX1:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLRWTX1]]
+  // CHECK: call %dx.types.ResRet.[[TY32]] @dx.op.textureLoad.[[TY32]](i32 66, %dx.types.Handle [[ANHDLRWTX1]], i32 undef, i32 [[IX0]], i32 undef, i32 undef
+  // F64: call double @dx.op.makeDouble.f64(i32 101
+  // F64: call double @dx.op.makeDouble.f64(i32 101
+  // I64: zext i32 %{{.*}} to i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: shl nuw i64
+  // I64: or i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: shl nuw i64
+  // I64: or i64
+  // I1: icmp ne i32 %{{.*}}, 0
+  // I1: icmp ne i32 %{{.*}}, 0
+  // I1: icmp ne i32 %{{.*}}, 0
+  // I1: icmp ne i32 %{{.*}}, 0
+  TYPE texElt2 = RwTex1d[ix0];
+
+  // CHECK: [[ANHDLROTX2:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLROTX2]]
+  // CHECK: call %dx.types.ResRet.[[TY32]] @dx.op.textureLoad.[[TY32]](i32 66, %dx.types.Handle [[ANHDLROTX2]], i32 0, i32 [[IX20]], i32 [[IX21]], i32 undef
+  // F64: call double @dx.op.makeDouble.f64(i32 101
+  // F64: call double @dx.op.makeDouble.f64(i32 101
+  // I64: zext i32 %{{.*}} to i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: shl nuw i64
+  // I64: or i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: shl nuw i64
+  // I64: or i64
+  // I1: icmp ne i32 %{{.*}}, 0
+  // I1: icmp ne i32 %{{.*}}, 0
+  // I1: icmp ne i32 %{{.*}}, 0
+  // I1: icmp ne i32 %{{.*}}, 0
+  TYPE texElt3 = RoTex2d[ix2];
+  // CHECK: [[ANHDLRWTX2:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLRWTX2]]
+  // CHECK: call %dx.types.ResRet.[[TY32]] @dx.op.textureLoad.[[TY32]](i32 66, %dx.types.Handle [[ANHDLRWTX2]], i32 undef, i32 [[IX20]], i32 [[IX21]], i32 undef
+  // F64: call double @dx.op.makeDouble.f64(i32 101
+  // F64: call double @dx.op.makeDouble.f64(i32 101
+  // I64: zext i32 %{{.*}} to i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: shl nuw i64
+  // I64: or i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: shl nuw i64
+  // I64: or i64
+  // I1: icmp ne i32 %{{.*}}, 0
+  // I1: icmp ne i32 %{{.*}}, 0
+  // I1: icmp ne i32 %{{.*}}, 0
+  // I1: icmp ne i32 %{{.*}}, 0
+  TYPE texElt4 = RwTex2d[ix2];
+
+  // CHECK: [[ANHDLROTX3:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLROTX3]]
+  // CHECK: call %dx.types.ResRet.[[TY32]] @dx.op.textureLoad.[[TY32]](i32 66, %dx.types.Handle [[ANHDLROTX3]], i32 0, i32 [[IX30]], i32 [[IX31]], i32 [[IX32]]
+  // F64: call double @dx.op.makeDouble.f64(i32 101
+  // F64: call double @dx.op.makeDouble.f64(i32 101
+  // I64: zext i32 %{{.*}} to i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: shl nuw i64
+  // I64: or i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: shl nuw i64
+  // I64: or i64
+  // I1: icmp ne i32 %{{.*}}, 0
+  // I1: icmp ne i32 %{{.*}}, 0
+  // I1: icmp ne i32 %{{.*}}, 0
+  // I1: icmp ne i32 %{{.*}}, 0
+  TYPE texElt5 = RoTex3d[ix3];
+  // CHECK: [[ANHDLRWTX3:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[HDLRWTX3]]
+  // CHECK: call %dx.types.ResRet.[[TY32]] @dx.op.textureLoad.[[TY32]](i32 66, %dx.types.Handle [[ANHDLRWTX3]], i32 undef, i32 [[IX30]], i32 [[IX31]], i32 [[IX32]]
+  // F64: call double @dx.op.makeDouble.f64(i32 101
+  // F64: call double @dx.op.makeDouble.f64(i32 101
+  // I64: zext i32 %{{.*}} to i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: shl nuw i64
+  // I64: or i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: zext i32 %{{.*}} to i64
+  // I64: shl nuw i64
+  // I64: or i64
+  // I1: icmp ne i32 %{{.*}}, 0
+  // I1: icmp ne i32 %{{.*}}, 0
+  // I1: icmp ne i32 %{{.*}}, 0
+  // I1: icmp ne i32 %{{.*}}, 0
+  TYPE texElt6 = RwTex3d[ix3];
+
+  // F64: call %dx.types.splitdouble @dx.op.splitDouble.f64(i32 102
+  // F64: call %dx.types.splitdouble @dx.op.splitDouble.f64(i32 102
+  // I64: trunc i64 %{{.*}} to i32
+  // lshr i64  %{{.*}}, 32
+  // I64: trunc i64 %{{.*}} to i32
+  // I64: trunc i64 %{{.*}} to i32
+  // lshr i64  %{{.*}}, 32
+  // I64: trunc i64 %{{.*}} to i32
+  // I1: zext i1 %{{.*}} to i32
+  // I1: zext i1 %{{.*}} to i32
+  // I1: zext i1 %{{.*}} to i32
+  // I1: zext i1 %{{.*}} to i32
+  // CHECK: call void @dx.op.textureStore.[[TY32]](i32 67, %dx.types.Handle [[ANHDLRWTX3]], i32 [[IX30]], i32 [[IX31]], i32 [[IX32]]
+  RwTex3d[ix3] = texElt1 + texElt2 + texElt3 + texElt4 + texElt5 + texElt6;
 }
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/types/longvec-decls.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/types/longvec-decls.hlsl
new file mode 100644
index 0000000000..8bc7b9e73d
--- /dev/null
+++ b/tools/clang/test/CodeGenDXIL/hlsl/types/longvec-decls.hlsl
@@ -0,0 +1,322 @@
+// RUN: %dxc -fcgl -T lib_6_9 -DTYPE=float     -DNUM=5 %s | FileCheck %s -check-prefixes=CHECK,F5
+// RUN: %dxc -fcgl -T lib_6_9 -DTYPE=bool      -DNUM=7 %s | FileCheck %s -check-prefixes=CHECK,B7
+// RUN: %dxc -fcgl -T lib_6_9 -DTYPE=uint64_t  -DNUM=9 %s | FileCheck %s -check-prefixes=CHECK,L9
+// RUN: %dxc -fcgl -T lib_6_9 -DTYPE=double    -DNUM=17 %s | FileCheck %s -check-prefixes=CHECK,D17
+// RUN: %dxc -fcgl -T lib_6_9 -DTYPE=float16_t -DNUM=256 -enable-16bit-types %s | FileCheck %s -check-prefixes=CHECK,H256
+// RUN: %dxc -fcgl -T lib_6_9 -DTYPE=int16_t   -DNUM=1024 -enable-16bit-types %s | FileCheck %s -check-prefixes=CHECK,S1024
+
+// A test to verify that declarations of longvecs are permitted in all the accepted places.
+// Only tests for acceptance, most codegen is ignored for now.
+
+// CHECK: %struct.LongVec = type { <4 x float>, <[[NUM:[0-9]*]] x [[STY:[a-z0-9]*]]> }
+struct LongVec {
+  float4 f;
+  vector<TYPE,NUM> vec;
+};
+
+struct LongVecSub : LongVec {
+  int3 is;
+};
+
+template<int N>
+struct LongVecTpl {
+  float4 f;
+  vector<float,N> vec;
+};
+
+// Just some dummies to capture the types and mangles.
+// CHECK: @"\01?dummy@@3[[MNG:F|M|N|_N|_K|\$f16@]]A" = external addrspace(3) global [[STY]]
+groupshared TYPE dummy;
+
+// Use the first groupshared to establish mangles and sizes
+// F5-DAG: @"\01?gs_vec@@3V?$vector@[[MNG:M]]$[[VS:04]]@@A" = external addrspace(3) global <[[NUM]] x [[STY]]>
+// B7-DAG: @"\01?gs_vec@@3V?$vector@[[MNG:_N]]$[[VS:06]]@@A" = external addrspace(3) global <[[NUM]] x [[STY]]>
+// L9-DAG: @"\01?gs_vec@@3V?$vector@[[MNG:_K]]$[[VS:08]]@@A" = external addrspace(3) global <[[NUM]] x [[STY]]>
+// D17-DAG: @"\01?gs_vec@@3V?$vector@[[MNG:N]]$[[VS:0BB@]]@@A" = external addrspace(3) global <[[NUM]] x [[STY]]>
+// H256-DAG: @"\01?gs_vec@@3V?$vector@[[MNG:\$f16@]]$[[VS:0BAA@]]@@A" = external addrspace(3) global <[[NUM]] x [[STY]]>
+// S1024-DAG: @"\01?gs_vec@@3V?$vector@[[MNG:F]]$[[VS:0EAA@]]@@A" = external addrspace(3) global <[[NUM]] x [[STY]]>
+groupshared vector<TYPE, NUM> gs_vec;
+
+// CHECK-DAG: @"\01?gs_vec_arr@@3PAV?$vector@[[MNG]]$[[VS]]@@A" = external addrspace(3) global [10 x <[[NUM]] x [[STY]]>]
+groupshared vector<TYPE, NUM> gs_vec_arr[10];
+// CHECK-DAG: @"\01?gs_vec_rec@@3ULongVec@@A" = external addrspace(3) global %struct.LongVec
+groupshared LongVec gs_vec_rec;
+// CHECK-DAG: @"\01?gs_vec_sub@@3ULongVecSub@@A" = external addrspace(3) global %struct.LongVecSub
+groupshared LongVecSub gs_vec_sub;
+// CHECK-DAG: @"\01?gs_vec_tpl@@3U?$LongVecTpl@$[[VS]]@@A" = external addrspace(3) global %"struct.LongVecTpl<[[NUM]]>"
+groupshared LongVecTpl<NUM> gs_vec_tpl;
+
+// CHECK-DAG: @static_vec = internal global <[[NUM]] x [[STY]]>
+static vector<TYPE, NUM> static_vec;
+// CHECK-DAG: @static_vec_arr = internal global [10 x <[[NUM]] x [[STY]]>] zeroinitializer
+static vector<TYPE, NUM> static_vec_arr[10];
+// CHECK-DAG: @static_vec_rec = internal global %struct.LongVec
+static LongVec static_vec_rec;
+// CHECK-DAG: @static_vec_sub = internal global %struct.LongVecSub
+static LongVecSub static_vec_sub;
+// CHECK-DAG: @static_vec_tpl = internal global %"struct.LongVecTpl<[[NUM]]>"
+static LongVecTpl<NUM> static_vec_tpl;
+
+// CHECK: define [[RTY:[a-z0-9]*]] @"\01?getVal@@YA[[MNG]][[MNG]]@Z"([[RTY]] {{.*}}%t)
+export TYPE getVal(TYPE t) {TYPE ret = dummy; dummy = t; return ret;}
+
+// CHECK: define <[[NUM]] x [[RTY]]>
+// CHECK-LABEL: @"\01?lv_param_passthru
+// CHECK-SAME: @@YA?AV?$vector@[[MNG]]$[[VS]]@@V1@@Z"(<[[NUM]] x [[RTY]]> %vec1)
+// CHECK:   ret <[[NUM]] x [[RTY]]>
+export vector<TYPE, NUM> lv_param_passthru(vector<TYPE, NUM> vec1) {
+  return vec1;
+}
+
+// CHECK-LABEL: define void @"\01?lv_param_arr_passthru
+// CHECK-SAME: @@YA$$BY09V?$vector@[[MNG]]$[[VS]]@@Y09V1@@Z"([10 x <[[NUM]] x [[STY]]>]* noalias sret %agg.result, [10 x <[[NUM]] x [[STY]]>]* %vec)
+// CHECK: ret void
+export vector<TYPE, NUM> lv_param_arr_passthru(vector<TYPE, NUM> vec[10])[10] {
+  return vec;
+}
+
+// CHECK-LABEL: define void @"\01?lv_param_rec_passthru@@YA?AULongVec@@U1@@Z"(%struct.LongVec* noalias sret %agg.result, %struct.LongVec* %vec)
+// CHECK: memcpy
+// CHECK:   ret void
+export LongVec lv_param_rec_passthru(LongVec vec) {
+  return vec;
+}
+
+// CHECK-LABEL: define void @"\01?lv_param_sub_passthru@@YA?AULongVec@@U1@@Z"(%struct.LongVec* noalias sret %agg.result, %struct.LongVec* %vec)
+// CHECK: memcpy
+// CHECK:   ret void
+export LongVec lv_param_sub_passthru(LongVec vec) {
+  return vec;
+}
+
+// CHECK-LABEL: define void @"\01?lv_param_tpl_passthru@@YA?AULongVec@@U1@@Z"(%struct.LongVec* noalias sret %agg.result, %struct.LongVec* %vec)
+// CHECK: memcpy
+// CHECK:   ret void
+export LongVec lv_param_tpl_passthru(LongVec vec) {
+  return vec;
+}
+
+// CHECK-LABEL: define void @"\01?lv_param_in_out
+// CHECK-SAME: @@YAXV?$vector@[[MNG]]$[[VS]]@@AIAV1@@Z"(<[[NUM]] x [[RTY]]> %vec1, <[[NUM]] x [[STY]]>* noalias dereferenceable({{[0-9]*}}) %vec2)
+// CHECK:   store <[[NUM]] x [[STY]]> {{%.*}}, <[[NUM]] x [[STY]]>* %vec2, align 4
+// CHECK:   ret void
+export void lv_param_in_out(in vector<TYPE, NUM> vec1, out vector<TYPE, NUM> vec2) {
+  vec2 = vec1;
+}
+
+// CHECK-LABEL: define void @"\01?lv_param_in_out_rec@@YAXULongVec@@U1@@Z"(%struct.LongVec* %vec1, %struct.LongVec* noalias %vec2)
+// CHECK: memcpy
+// CHECK:   ret void
+export void lv_param_in_out_rec(in LongVec vec1, out LongVec vec2) {
+  vec2 = vec1;
+}
+
+// CHECK-LABEL: define void @"\01?lv_param_in_out_sub@@YAXULongVec@@U1@@Z"(%struct.LongVec* %vec1, %struct.LongVec* noalias %vec2)
+// CHECK: memcpy
+// CHECK:   ret void
+export void lv_param_in_out_sub(in LongVec vec1, out LongVec vec2) {
+  vec2 = vec1;
+}
+
+// CHECK-LABEL: define void @"\01?lv_param_in_out_tpl@@YAXULongVec@@U1@@Z"(%struct.LongVec* %vec1, %struct.LongVec* noalias %vec2)
+// CHECK: memcpy
+// CHECK:   ret void
+export void lv_param_in_out_tpl(in LongVec vec1, out LongVec vec2) {
+  vec2 = vec1;
+}
+
+
+// CHECK-LABEL: define void @"\01?lv_param_inout
+// CHECK-SAME: @@YAXAIAV?$vector@[[MNG]]$[[VS]]@@0@Z"(<[[NUM]] x [[STY]]>* noalias dereferenceable({{[0-9]*}}) %vec1, <[[NUM]] x [[STY]]>* noalias dereferenceable({{[0-9]*}}) %vec2)
+// CHECK:   load <[[NUM]] x [[STY]]>, <[[NUM]] x [[STY]]>* %vec1, align 4
+// CHECK:   load <[[NUM]] x [[STY]]>, <[[NUM]] x [[STY]]>* %vec2, align 4
+// CHECK:   store <[[NUM]] x [[STY]]> {{%.*}}, <[[NUM]] x [[STY]]>* %vec1, align 4
+// CHECK:   store <[[NUM]] x [[STY]]> {{%.*}}, <[[NUM]] x [[STY]]>* %vec2, align 4
+// CHECK:   ret void
+export void lv_param_inout(inout vector<TYPE, NUM> vec1, inout vector<TYPE, NUM> vec2) {
+  vector<TYPE, NUM> tmp = vec1;
+  vec1 = vec2;
+  vec2 = tmp;
+}
+
+// CHECK-LABEL: define void @"\01?lv_param_inout_rec@@YAXULongVec@@0@Z"(%struct.LongVec* noalias %vec1, %struct.LongVec* noalias %vec2)
+// CHECK: memcpy
+// CHECK:   ret void
+export void lv_param_inout_rec(inout LongVec vec1, inout LongVec vec2) {
+  LongVec tmp = vec1;
+  vec1 = vec2;
+  vec2 = tmp;
+}
+
+// CHECK-LABEL: define void @"\01?lv_param_inout_sub@@YAXULongVec@@0@Z"(%struct.LongVec* noalias %vec1, %struct.LongVec* noalias %vec2)
+// CHECK: memcpy
+// CHECK:   ret void
+export void lv_param_inout_sub(inout LongVec vec1, inout LongVec vec2) {
+  LongVec tmp = vec1;
+  vec1 = vec2;
+  vec2 = tmp;
+}
+
+// CHECK-LABEL: define void @"\01?lv_param_inout_tpl@@YAXULongVec@@0@Z"(%struct.LongVec* noalias %vec1, %struct.LongVec* noalias %vec2)
+// CHECK: memcpy
+// CHECK:   ret void
+export void lv_param_inout_tpl(inout LongVec vec1, inout LongVec vec2) {
+  LongVec tmp = vec1;
+  vec1 = vec2;
+  vec2 = tmp;
+}
+
+// CHECK-LABEL: define void @"\01?lv_global_assign
+// CHECK-SAME: @@YAXV?$vector@[[MNG]]$[[VS]]@@Y09V1@ULongVec@@ULongVecSub@@U?$LongVecTpl@$[[VS]]@@@Z"(<[[NUM]] x [[RTY]]> %vec, [10 x <[[NUM]] x [[STY]]>]* %arr, %struct.LongVec* %rec, %struct.LongVecSub* %sub, %"struct.LongVecTpl<[[NUM]]>"* %tpl)
+// CHECK:   store <[[NUM]] x [[STY]]> {{%.*}}, <[[NUM]] x [[STY]]>* @static_vec
+// CHECK:   ret void
+export void lv_global_assign(vector<TYPE, NUM> vec, vector<TYPE, NUM> arr[10],
+                             LongVec rec, LongVecSub sub, LongVecTpl<NUM> tpl) {
+  static_vec = vec;
+  static_vec_arr = arr;
+  static_vec_rec = rec;
+  static_vec_sub = sub;
+  static_vec_tpl = tpl;
+}
+
+// CHECK-LABEL: define void @"\01?lv_gs_assign
+// CHECK-SAME: @@YAXV?$vector@[[MNG]]$[[VS]]@@Y09V1@ULongVec@@ULongVecSub@@U?$LongVecTpl@$[[VS]]@@@Z"(<[[NUM]] x [[RTY]]> %vec, [10 x <[[NUM]] x [[STY]]>]* %arr, %struct.LongVec* %rec, %struct.LongVecSub* %sub, %"struct.LongVecTpl<[[NUM]]>"* %tpl)
+// CHECK:   store <[[NUM]] x [[STY]]> {{%.*}}, <[[NUM]] x [[STY]]> addrspace(3)* @"\01?gs_vec@@3V?$vector@[[MNG]]$[[VS]]@@A"
+// CHECK:   ret void
+export void lv_gs_assign(vector<TYPE, NUM> vec, vector<TYPE, NUM> arr[10],
+                         LongVec rec, LongVecSub sub, LongVecTpl<NUM> tpl) {
+  gs_vec = vec;
+  gs_vec_arr = arr;
+  gs_vec_rec = sub;
+  gs_vec_tpl = tpl;
+}
+
+// CHECK: define <[[NUM]] x [[RTY]]>
+// CHECK-LABEL: @"\01?lv_global_ret
+// CHECK-SAME: @@YA?AV?$vector@[[MNG]]$[[VS]]@@XZ"()
+// CHECK:   load <[[NUM]] x [[STY]]>, <[[NUM]] x [[STY]]>* @static_vec
+// CHECK:   ret <[[NUM]] x [[RTY]]>
+export vector<TYPE, NUM> lv_global_ret() {
+  return static_vec;
+}
+
+// CHECK-LABEL: define void @"\01?lv_global_arr_ret
+// CHECK-SAME: @@YA$$BY09V?$vector@[[MNG]]$[[VS]]@@XZ"([10 x <[[NUM]] x [[STY]]>]* noalias sret %agg.result)
+// CHECK: ret void
+export vector<TYPE, NUM> lv_global_arr_ret()[10] {
+  return static_vec_arr;
+}
+
+// CHECK-LABEL: define void @"\01?lv_global_rec_ret@@YA?AULongVec@@XZ"(%struct.LongVec* noalias sret %agg.result)
+// CHECK: memcpy
+// CHECK:   ret void
+export LongVec lv_global_rec_ret() {
+  return static_vec_rec;
+}
+
+// CHECK-LABEL: define void @"\01?lv_global_sub_ret@@YA?AULongVecSub@@XZ"(%struct.LongVecSub* noalias sret %agg.result)
+// CHECK: memcpy
+// CHECK:   ret void
+export LongVecSub lv_global_sub_ret() {
+  return static_vec_sub;
+}
+
+// CHECK-LABEL: define void @"\01?lv_global_tpl_ret
+// CHECK-SAME: @@YA?AU?$LongVecTpl@$[[VS]]@@XZ"(%"struct.LongVecTpl<[[NUM]]>"* noalias sret %agg.result)
+// CHECK: memcpy
+// CHECK:   ret void
+export LongVecTpl<NUM> lv_global_tpl_ret() {
+  return static_vec_tpl;
+}
+
+// CHECK: define <[[NUM]] x [[RTY]]>
+// CHECK-LABEL: @"\01?lv_gs_ret
+// CHECK-SAME: @@YA?AV?$vector@[[MNG]]$[[VS]]@@XZ"()
+// CHECK:   load <[[NUM]] x [[STY]]>, <[[NUM]] x [[STY]]> addrspace(3)* @"\01?gs_vec@@3V?$vector@[[MNG]]$[[VS]]@@A"
+// CHECK:   ret <[[NUM]] x [[RTY]]>
+export vector<TYPE, NUM> lv_gs_ret() {
+  return gs_vec;
+}
+
+// CHECK-LABEL: define void @"\01?lv_gs_arr_ret
+// CHECK-SAME: @@YA$$BY09V?$vector@[[MNG]]$[[VS]]@@XZ"([10 x <[[NUM]] x [[STY]]>]* noalias sret %agg.result)
+// CHECK: ret void
+export vector<TYPE, NUM> lv_gs_arr_ret()[10] {
+  return gs_vec_arr;
+}
+
+// CHECK-LABEL: define void @"\01?lv_gs_rec_ret@@YA?AULongVec@@XZ"(%struct.LongVec* noalias sret %agg.result)
+// CHECK: memcpy
+// CHECK:   ret void
+export LongVec lv_gs_rec_ret() {
+  return gs_vec_rec;
+}
+
+// CHECK-LABEL: define void @"\01?lv_gs_sub_ret@@YA?AULongVecSub@@XZ"(%struct.LongVecSub* noalias sret %agg.result)
+// CHECK: memcpy
+// CHECK:   ret void
+export LongVecSub lv_gs_sub_ret() {
+  return gs_vec_sub;
+}
+
+// CHECK-LABEL: define void @"\01?lv_gs_tpl_ret
+// CHECK-SAME: @@YA?AU?$LongVecTpl@$[[VS]]@@XZ"(%"struct.LongVecTpl<[[NUM]]>"* noalias sret %agg.result)
+// CHECK: memcpy
+// CHECK:   ret void
+export LongVecTpl<NUM> lv_gs_tpl_ret() {
+  return gs_vec_tpl;
+}
+
+// CHECK: define <[[NUM]] x [[RTY]]>
+// CHECK-LABEL: @"\01?lv_splat
+// CHECK-SAME: @@YA?AV?$vector@[[MNG]]$[[VS]]@@[[MNG]]@Z"([[RTY]] {{.*}}%scalar)
+// CHECK:   ret <[[NUM]] x [[RTY]]>
+export vector<TYPE,NUM> lv_splat(TYPE scalar) {
+  vector<TYPE,NUM> ret = scalar;
+  return ret;
+}
+
+// CHECK: define <6 x [[RTY]]>
+// CHECK-LABEL: @"\01?lv_initlist
+// CHECK-SAME: @@YA?AV?$vector@[[MNG]]$05@@XZ"()
+// CHECK:   ret <6 x [[RTY]]>
+export vector<TYPE, 6> lv_initlist() {
+  vector<TYPE, 6> ret = {1, 2, 3, 4, 5, 6};
+  return ret;
+}
+
+// CHECK: define <6 x [[RTY]]>
+// CHECK-LABEL: @"\01?lv_initlist_vec
+// CHECK-SAME: @@YA?AV?$vector@[[MNG]]$05@@V?$vector@[[MNG]]$02@@@Z"(<3 x [[RTY]]> %vec)
+// CHECK:   ret <6 x [[RTY]]>
+export vector<TYPE, 6> lv_initlist_vec(vector<TYPE, 3> vec) {
+  vector<TYPE, 6> ret = {vec, 4.0, 5.0, 6.0};
+  return ret;
+}
+
+// CHECK: define <6 x [[RTY]]>
+// CHECK-LABEL: @"\01?lv_vec_vec
+// CHECK-SAME: @@YA?AV?$vector@[[MNG]]$05@@V?$vector@[[MNG]]$02@@0@Z"(<3 x [[RTY]]> %vec1, <3 x [[RTY]]> %vec2)
+// CHECK:   ret <6 x [[RTY]]>
+export vector<TYPE, 6> lv_vec_vec(vector<TYPE, 3> vec1, vector<TYPE, 3> vec2) {
+  vector<TYPE, 6> ret = {vec1, vec2};
+  return ret;
+}
+
+// CHECK: define <[[NUM]] x [[RTY]]>
+// CHECK-LABEL: @"\01?lv_array_cast
+// CHECK-SAME: @@YA?AV?$vector@[[MNG]]$[[VS]]@@Y[[VS]][[MNG]]@Z"({{\[}}[[NUM]] x [[STY]]]* %arr)
+// CHECK:   ret <[[NUM]] x [[RTY]]>
+export vector<TYPE, NUM> lv_array_cast(TYPE arr[NUM]) {
+  vector<TYPE, NUM> ret = (vector<TYPE,NUM>)arr;
+  return ret;
+}
+
+// CHECK: define <6 x [[RTY]]>
+// CHECK-LABEL: @"\01?lv_ctor
+// CHECK-SAME: @@YA?AV?$vector@[[MNG]]$05@@[[MNG]]@Z"([[RTY]] {{.*}}%s)
+// CHECK:  ret <6 x [[RTY]]>
+export vector<TYPE, 6> lv_ctor(TYPE s) {
+  vector<TYPE, 6> ret = vector<TYPE,6>(1.0, 2.0, 3.0, 4.0, 5.0, s);
+  return ret;
+}
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/types/longvec-operators-bool.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/types/longvec-operators-bool.hlsl
new file mode 100644
index 0000000000..bb2cae6756
--- /dev/null
+++ b/tools/clang/test/CodeGenDXIL/hlsl/types/longvec-operators-bool.hlsl
@@ -0,0 +1,463 @@
+// RUN: %dxc -HV 2018 -T lib_6_9 -DNUM=2 %s | FileCheck %s
+// RUN: %dxc -HV 2018 -T lib_6_9 -DNUM=5 %s | FileCheck %s
+// RUN: %dxc -HV 2018 -T lib_6_9 -DNUM=3 %s | FileCheck %s
+// RUN: %dxc -HV 2018 -T lib_6_9 -DNUM=9 %s | FileCheck %s
+
+// Test relevant operators on an assortment bool vector sizes with 6.9 native vectors.
+
+// Just a trick to capture the needed type spellings since the DXC version of FileCheck can't do that explicitly.
+// Uses non vector buffer to avoid interacting with that implementation.
+// CHECK: %dx.types.ResRet.[[TY:[a-z0-9]*]] = type { [[TYPE:[a-z_0-9]*]]
+RWStructuredBuffer< bool > buf;
+
+groupshared vector<bool, NUM> gs_vec1, gs_vec2;
+groupshared vector<bool, NUM+1> gs_vec3;
+
+
+// A mixed-type overload to test overload resolution and mingle different vector element types in ops
+// Test assignment operators.
+// CHECK-LABEL: define void @"\01?assignments
+export void assignments(inout vector<bool, NUM> things[10], bool scales[10]) {
+
+  // Another trick to capture the size.
+  // CHECK: [[res:%[0-9]*]] = call %dx.types.ResRet.i32 @dx.op.rawBufferLoad.i32(i32 139, %dx.types.Handle %{{[^,]*}}, i32 [[NUM:[0-9]*]]
+  // CHECK: [[scl:%[0-9]*]] = extractvalue %dx.types.ResRet.i32 [[res]], 0
+  // CHECK: [[bscl:%[0-9]*]] = icmp ne i32 [[scl]], 0
+  bool scalar = buf.Load(NUM);
+
+  // CHECK: [[add9:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %things, i32 0, i32 9
+  // CHECK: [[vec9:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add9]]
+  // CHECK: [[bvec9:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec9]], zeroinitializer
+  // CHECK: [[add0:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %things, i32 0, i32 0
+  // CHECK: [[res0:%[0-9]*]] = zext <[[NUM]] x i1> [[bvec9]] to <[[NUM]] x i32>
+  // CHECK: store <[[NUM]] x i32> [[res0]], <[[NUM]] x i32>* [[add0]]
+  things[0] = things[9];
+
+  // CHECK: [[spt:%[0-9]*]] = insertelement <[[NUM]] x i1> undef, i1 [[bscl]], i32 0
+  // CHECK: [[res:%[0-9]*]] = shufflevector <[[NUM]] x i1> [[spt]], <[[NUM]] x i1> undef, <[[NUM]] x i32> zeroinitializer
+  // CHECK: [[add5:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %things, i32 0, i32 5
+  // CHECK: [[res5:%[0-9]*]] = zext <[[NUM]] x i1> [[res]] to <[[NUM]] x i32>
+  // CHECK: store <[[NUM]] x i32> [[res5]], <[[NUM]] x i32>* [[add5]]
+  things[5] = scalar;
+
+}
+
+// Test arithmetic operators.
+// CHECK-LABEL: define void @"\01?arithmetic
+export vector<bool, NUM> arithmetic(inout vector<bool, NUM> things[10])[10] {
+  vector<bool, NUM> res[10];
+  // CHECK: [[add0:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %things, i32 0, i32 0
+  // CHECK: [[vec0:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add0]]
+  // CHECK: [[add1:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %things, i32 0, i32 1
+  // CHECK: [[vec1:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add1]]
+  // CHECK: [[add2:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %things, i32 0, i32 2
+  // CHECK: [[vec2:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add2]]
+  // CHECK: [[add3:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %things, i32 0, i32 3
+  // CHECK: [[vec3:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add3]]
+  // CHECK: [[add4:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %things, i32 0, i32 4
+  // CHECK: [[vec4:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add4]]
+  // CHECK: [[add5:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %things, i32 0, i32 5
+  // CHECK: [[vec5:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add5]]
+  // CHECK: [[add6:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %things, i32 0, i32 6
+  // CHECK: [[vec6:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add6]]
+
+  // CHECK: [[bvec0:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec0]], zeroinitializer
+  // CHECK: [[svec0:%[0-9]*]] = sext <[[NUM]] x i1> [[bvec0]] to <[[NUM]] x i32>
+  // CHECK: [[bsvec0:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[svec0]], zeroinitializer
+  // CHECK: [[res0:%[0-9]*]] = zext <[[NUM]] x i1> [[bsvec0]] to <[[NUM]] x i32>
+  res[0] = -things[0];
+
+  // CHECK: [[vec0:%[0-9]*]] = zext <[[NUM]] x i1> [[bvec0]] to <[[NUM]] x i32>
+  // CHECK: [[bvec0:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec0]], zeroinitializer
+  // CHECK: [[res1:%[0-9]*]] = zext <[[NUM]] x i1> [[bvec0]] to <[[NUM]] x i32>
+  res[1] = +things[0];
+
+  // CHECK: [[bvec1:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec1]], zeroinitializer
+  // CHECK: [[vec1:%[0-9]*]] = zext <[[NUM]] x i1> [[bvec1]] to <[[NUM]] x i32>
+  // CHECK: [[bvec2:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec2]], zeroinitializer
+  // CHECK: [[vec2:%[0-9]*]] = zext <[[NUM]] x i1> [[bvec2]] to <[[NUM]] x i32>
+  // CHECK: [[res2:%[0-9]*]] = add nuw nsw <[[NUM]] x i32> [[vec2]], [[vec1]]
+  // CHECK: [[bres2:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[res2]], zeroinitializer
+  // CHECK: [[res2:%[0-9][0-9]*]] = zext <[[NUM]] x i1> [[bres2]] to <[[NUM]] x i32>
+  res[2] = things[1] + things[2];
+
+  // CHECK: [[bvec3:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec3]], zeroinitializer
+  // CHECK: [[vec3:%[0-9]*]] = zext <[[NUM]] x i1> [[bvec3]] to <[[NUM]] x i32>
+  // CHECK: [[res3:%[0-9]*]] = sub nsw <[[NUM]] x i32> [[vec2]], [[vec3]]
+  // CHECK: [[bres3:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[res3]], zeroinitializer
+  // CHECK: [[res3:%[0-9][0-9]*]] = zext <[[NUM]] x i1> [[bres3]] to <[[NUM]] x i32>
+  res[3] = things[2] - things[3];
+
+  // CHECK: [[bvec4:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec4]], zeroinitializer
+  // CHECK: [[vec4:%[0-9]*]] = zext <[[NUM]] x i1> [[bvec4]] to <[[NUM]] x i32>
+  // CHECK: [[res4:%[0-9]*]] = mul nuw nsw <[[NUM]] x i32> [[vec4]], [[vec3]]
+  // CHECK: [[bres4:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[res4]], zeroinitializer
+  // CHECK: [[res4:%[0-9][0-9]*]] = zext <[[NUM]] x i1> [[bres4]] to <[[NUM]] x i32>
+  res[4] = things[3] * things[4];
+
+  // CHECK: [[bvec5:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec5]], zeroinitializer
+  // CHECK: [[vec5:%[0-9]*]] = zext <[[NUM]] x i1> [[bvec5]] to <[[NUM]] x i32>
+  // CHECK: [[res5:%[0-9]*]] = sdiv <[[NUM]] x i32> [[vec4]], [[vec5]]
+  // CHECK: [[bres5:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[res5]], zeroinitializer
+  // CHECK: [[res5:%[0-9][0-9]*]] = zext <[[NUM]] x i1> [[bres5]] to <[[NUM]] x i32>
+  res[5] = things[4] / things[5];
+
+  // CHECK: [[bvec6:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec6]], zeroinitializer
+  // CHECK: [[vec6:%[0-9]*]] = zext <[[NUM]] x i1> [[bvec6]] to <[[NUM]] x i32>
+  // CHECK: [[res6:%[0-9]*]] = {{[ufs]?rem( fast)?}} <[[NUM]] x i32> [[vec5]], [[vec6]]
+  res[6] = things[5] % things[6];
+
+  // Stores into res[]. Previous were for things[] inout.
+  // CHECK: [[add0:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 0
+  // CHECK: store <[[NUM]] x i32> [[res0]], <[[NUM]] x i32>* [[add0]]
+  // CHECK: [[add1:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 1
+  // CHECK: store <[[NUM]] x i32> [[res1]], <[[NUM]] x i32>* [[add1]]
+  // CHECK: [[add2:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 2
+  // CHECK: store <[[NUM]] x i32> [[res2]], <[[NUM]] x i32>* [[add2]]
+  // CHECK: [[add3:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 3
+  // CHECK: store <[[NUM]] x i32> [[res3]], <[[NUM]] x i32>* [[add3]]
+  // CHECK: [[add4:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 4
+  // CHECK: store <[[NUM]] x i32> [[res4]], <[[NUM]] x i32>* [[add4]]
+  // CHECK: [[add5:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 5
+  // CHECK: store <[[NUM]] x i32> [[res5]], <[[NUM]] x i32>* [[add5]]
+  // CHECK: ret void
+
+
+  return res;
+}
+
+// Test arithmetic operators with scalars.
+// CHECK-LABEL: define void @"\01?scarithmetic
+export vector<bool, NUM> scarithmetic(inout vector<bool, NUM> things[10], bool scales[10])[10] {
+  vector<bool, NUM> res[10];
+  // CHECK: [[add0:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %things, i32 0, i32 0
+  // CHECK: [[vec0:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add0]]
+  // CHECK: [[add1:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %things, i32 0, i32 1
+  // CHECK: [[vec1:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add1]]
+  // CHECK: [[add2:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %things, i32 0, i32 2
+  // CHECK: [[vec2:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add2]]
+  // CHECK: [[add3:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %things, i32 0, i32 3
+  // CHECK: [[vec3:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add3]]
+  // CHECK: [[add4:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %things, i32 0, i32 4
+  // CHECK: [[vec4:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add4]]
+  // CHECK: [[add5:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %things, i32 0, i32 5
+  // CHECK: [[vec5:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add5]]
+  // CHECK: [[add6:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %things, i32 0, i32 6
+  // CHECK: [[vec6:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add6]]
+
+  // CHECK: [[bvec0:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec0]], zeroinitializer
+  // CHECK: [[vec0:%[0-9]*]] = zext <[[NUM]] x i1> [[bvec0]] to <[[NUM]] x i32>
+  // CHECK: [[add0:%[0-9]*]] = getelementptr inbounds [10 x i32], [10 x i32]* %scales, i32 0, i32 0
+  // CHECK: [[scl0:%[0-9]*]] = load i32, i32* [[add0]]
+  // CHECK: [[spt:%[0-9]*]] = insertelement <[[NUM]] x i32> undef, i32 [[scl0]], i32 0
+  // CHECK: [[spt0:%[0-9]*]] = shufflevector <[[NUM]] x i32> [[spt]], <[[NUM]] x i32> undef, <[[NUM]] x i32> zeroinitializer
+  // CHECK: [[res0:%[0-9]*]] = add <[[NUM]] x i32> [[spt0]], [[vec0]]
+  // CHECK: [[bres0:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[res0]], zeroinitializer
+  // CHECK: [[res0:%[0-9]*]] = zext <[[NUM]] x i1> [[bres0]] to <[[NUM]] x i32>
+  res[0] = things[0] + scales[0];
+
+  // CHECK: [[bvec1:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec1]], zeroinitializer
+  // CHECK: [[vec1:%[0-9]*]] = zext <[[NUM]] x i1> [[bvec1]] to <[[NUM]] x i32>
+  // CHECK: [[add1:%[0-9]*]] = getelementptr inbounds [10 x i32], [10 x i32]* %scales, i32 0, i32 1
+  // CHECK: [[scl1:%[0-9]*]] = load i32, i32* [[add1]]
+  // CHECK: [[spt:%[0-9]*]] = insertelement <[[NUM]] x i32> undef, i32 [[scl1]], i32 0
+  // CHECK: [[spt1:%[0-9]*]] = shufflevector <[[NUM]] x i32> [[spt]], <[[NUM]] x i32> undef, <[[NUM]] x i32> zeroinitializer
+  // CHECK: [[res1:%[0-9]*]] = sub <[[NUM]] x i32> [[vec1]], [[spt1]]
+  // CHECK: [[bres1:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[res1]], zeroinitializer
+  // CHECK: [[res1:%[0-9]*]] = zext <[[NUM]] x i1> [[bres1]] to <[[NUM]] x i32>
+  res[1] = things[1] - scales[1];
+
+
+  // CHECK: [[bvec2:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec2]], zeroinitializer
+  // CHECK: [[vec2:%[0-9]*]] = zext <[[NUM]] x i1> [[bvec2]] to <[[NUM]] x i32>
+  // CHECK: [[add2:%[0-9]*]] = getelementptr inbounds [10 x i32], [10 x i32]* %scales, i32 0, i32 2
+  // CHECK: [[scl2:%[0-9]*]] = load i32, i32* [[add2]]
+  // CHECK: [[spt:%[0-9]*]] = insertelement <[[NUM]] x i32> undef, i32 [[scl2]], i32 0
+  // CHECK: [[spt2:%[0-9]*]] = shufflevector <[[NUM]] x i32> [[spt]], <[[NUM]] x i32> undef, <[[NUM]] x i32> zeroinitializer
+  // CHECK: [[res2:%[0-9]*]] = mul nuw <[[NUM]] x i32> [[spt2]], [[vec2]]
+  // CHECK: [[bres2:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[res2]], zeroinitializer
+  // CHECK: [[res2:%[0-9]*]] = zext <[[NUM]] x i1> [[bres2]] to <[[NUM]] x i32>
+  res[2] = things[2] * scales[2];
+
+  // CHECK: [[bvec3:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec3]], zeroinitializer
+  // CHECK: [[vec3:%[0-9]*]] = zext <[[NUM]] x i1> [[bvec3]] to <[[NUM]] x i32>
+  // CHECK: [[add3:%[0-9]*]] = getelementptr inbounds [10 x i32], [10 x i32]* %scales, i32 0, i32 3
+  // CHECK: [[scl3:%[0-9]*]] = load i32, i32* [[add3]]
+  // CHECK: [[spt:%[0-9]*]] = insertelement <[[NUM]] x i32> undef, i32 [[scl3]], i32 0
+  // CHECK: [[spt3:%[0-9]*]] = shufflevector <[[NUM]] x i32> [[spt]], <[[NUM]] x i32> undef, <[[NUM]] x i32> zeroinitializer
+  // CHECK: [[res3:%[0-9]*]] = sdiv <[[NUM]] x i32> [[vec3]], [[spt3]]
+  // CHECK: [[bres3:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[res3]], zeroinitializer
+  // CHECK: [[res3:%[0-9]*]] = zext <[[NUM]] x i1> [[bres3]] to <[[NUM]] x i32>
+  res[3] = things[3] / scales[3];
+
+  // CHECK: [[add4:%[0-9]*]] = getelementptr inbounds [10 x i32], [10 x i32]* %scales, i32 0, i32 4
+  // CHECK: [[scl4:%[0-9]*]] = load i32, i32* [[add4]]
+  // CHECK: [[spt:%[0-9]*]] = insertelement <[[NUM]] x i32> undef, i32 [[scl4]], i32 0
+  // CHECK: [[spt4:%[0-9]*]] = shufflevector <[[NUM]] x i32> [[spt]], <[[NUM]] x i32> undef, <[[NUM]] x i32> zeroinitializer
+  // CHECK: [[bvec4:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec4]], zeroinitializer
+  // CHECK: [[vec4:%[0-9]*]] = zext <[[NUM]] x i1> [[bvec4]] to <[[NUM]] x i32>
+  // CHECK: [[res4:%[0-9]*]] = add <[[NUM]] x i32> [[spt4]], [[vec4]]
+  // CHECK: [[bres4:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[res4]], zeroinitializer
+  // CHECK: [[res4:%[0-9]*]] = zext <[[NUM]] x i1> [[bres4]] to <[[NUM]] x i32>
+  res[4] = scales[4] + things[4];
+
+  // CHECK: [[add5:%[0-9]*]] = getelementptr inbounds [10 x i32], [10 x i32]* %scales, i32 0, i32 5
+  // CHECK: [[scl5:%[0-9]*]] = load i32, i32* [[add5]]
+  // CHECK: [[spt:%[0-9]*]] = insertelement <[[NUM]] x i32> undef, i32 [[scl5]], i32 0
+  // CHECK: [[spt5:%[0-9]*]] = shufflevector <[[NUM]] x i32> [[spt]], <[[NUM]] x i32> undef, <[[NUM]] x i32> zeroinitializer
+  // CHECK: [[bvec5:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec5]], zeroinitializer
+  // CHECK: [[vec5:%[0-9]*]] = zext <[[NUM]] x i1> [[bvec5]] to <[[NUM]] x i32>
+  // CHECK: [[res5:%[0-9]*]] = sub <[[NUM]] x i32> [[spt5]], [[vec5]]
+  // CHECK: [[bres5:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[res5]], zeroinitializer
+  // CHECK: [[res5:%[0-9]*]] = zext <[[NUM]] x i1> [[bres5]] to <[[NUM]] x i32>
+  res[5] = scales[5] - things[5];
+
+  // CHECK: [[add6:%[0-9]*]] = getelementptr inbounds [10 x i32], [10 x i32]* %scales, i32 0, i32 6
+  // CHECK: [[scl6:%[0-9]*]] = load i32, i32* [[add6]]
+  // CHECK: [[spt:%[0-9]*]] = insertelement <[[NUM]] x i32> undef, i32 [[scl6]], i32 0
+  // CHECK: [[spt6:%[0-9]*]] = shufflevector <[[NUM]] x i32> [[spt]], <[[NUM]] x i32> undef, <[[NUM]] x i32> zeroinitializer
+  // CHECK: [[bvec6:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec6]], zeroinitializer
+  // CHECK: [[vec6:%[0-9]*]] = zext <[[NUM]] x i1> [[bvec6]] to <[[NUM]] x i32>
+  // CHECK: [[res6:%[0-9]*]] = mul nuw <[[NUM]] x i32> [[spt6]], [[vec6]]
+  // CHECK: [[bres6:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[res6]], zeroinitializer
+  // CHECK: [[res6:%[0-9]*]] = zext <[[NUM]] x i1> [[bres6]] to <[[NUM]] x i32>
+  res[6] = scales[6] * things[6];
+
+  // CHECK: [[add0:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 0
+  // CHECK: store <[[NUM]] x i32> [[res0]], <[[NUM]] x i32>* [[add0]]
+  // CHECK: [[add1:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 1
+  // CHECK: store <[[NUM]] x i32> [[res1]], <[[NUM]] x i32>* [[add1]]
+  // CHECK: [[add2:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 2
+  // CHECK: store <[[NUM]] x i32> [[res2]], <[[NUM]] x i32>* [[add2]]
+  // CHECK: [[add3:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 3
+  // CHECK: store <[[NUM]] x i32> [[res3]], <[[NUM]] x i32>* [[add3]]
+  // CHECK: [[add4:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 4
+  // CHECK: store <[[NUM]] x i32> [[res4]], <[[NUM]] x i32>* [[add4]]
+  // CHECK: [[add5:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 5
+  // CHECK: store <[[NUM]] x i32> [[res5]], <[[NUM]] x i32>* [[add5]]
+  // CHECK: [[add6:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 6
+  // CHECK: store <[[NUM]] x i32> [[res6]], <[[NUM]] x i32>* [[add6]]
+  // CHECK: ret void
+
+
+  return res;
+}
+
+// Test logic operators.
+// Only permissable in pre-HLSL2021
+// CHECK-LABEL: define void @"\01?logic
+export vector<bool, NUM> logic(vector<bool, NUM> truth[10], vector<bool, NUM> consequences[10])[10] {
+  vector<bool, NUM> res[10];
+  // CHECK: [[add0:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %truth, i32 0, i32 0
+  // CHECK: [[vec0:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add0]]
+  // CHECK: [[cmp:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec0]], zeroinitializer
+  // CHECK: [[cmp0:%[0-9]*]] = icmp eq <[[NUM]] x i1> [[cmp]], zeroinitializer
+  // CHECK: [[res0:%[0-9]*]] = zext <[[NUM]] x i1> [[cmp0]] to <[[NUM]] x i32>
+  res[0] = !truth[0];
+
+  // CHECK: [[add1:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %truth, i32 0, i32 1
+  // CHECK: [[vec1:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add1]]
+  // CHECK: [[bvec1:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec1]], zeroinitializer
+  // CHECK: [[add2:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %truth, i32 0, i32 2
+  // CHECK: [[vec2:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add2]]
+  // CHECK: [[bvec2:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec2]], zeroinitializer
+  // CHECK: [[bres1:%[0-9]*]] = or <[[NUM]] x i1> [[bvec2]], [[bvec1]]
+  // CHECK: [[res1:%[0-9]*]] = zext <[[NUM]] x i1> [[bres1]] to <[[NUM]] x i32>
+  res[1] = truth[1] || truth[2];
+
+  // CHECK: [[add3:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %truth, i32 0, i32 3
+  // CHECK: [[vec3:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add3]]
+  // CHECK: [[bvec3:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec3]], zeroinitializer
+  // CHECK: [[bres2:%[0-9]*]] = and <[[NUM]] x i1> [[bvec3]], [[bvec2]]
+  // CHECK: [[res2:%[0-9]*]] = zext <[[NUM]] x i1> [[bres2]] to <[[NUM]] x i32>
+  res[2] = truth[2] && truth[3];
+
+  // CHECK: [[add4:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %truth, i32 0, i32 4
+  // CHECK: [[vec4:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add4]]
+  // CHECK: [[bvec4:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec4]], zeroinitializer
+  // CHECK: [[add5:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %truth, i32 0, i32 5
+  // CHECK: [[vec5:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add5]]
+  // CHECK: [[bvec5:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec5]], zeroinitializer
+  // MORE STUFF
+
+  res[3] = truth[3] ? truth[4] : truth[5];
+
+  // CHECK: [[add0:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %consequences, i32 0, i32 0
+  // CHECK: [[vec0:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add0]]
+  // CHECK: [[bvec0:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec0]], zeroinitializer
+
+  // CHECK: [[add1:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %consequences, i32 0, i32 1
+  // CHECK: [[vec1:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add1]]
+  // CHECK: [[bvec1:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec1]], zeroinitializer
+  // CHECK: [[bres4:%[0-9]*]] = icmp eq <[[NUM]] x i1> [[bvec0]], [[bvec1]]
+  // CHECK: [[res4:%[0-9]*]] = zext <[[NUM]] x i1> [[bres4]] to <[[NUM]] x i32>
+  res[4] = consequences[0] == consequences[1];
+
+  // CHECK: [[add2:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %consequences, i32 0, i32 2
+  // CHECK: [[vec2:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add2]]
+  // CHECK: [[bvec2:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec2]], zeroinitializer
+  // CHECK: [[bres5:%[0-9]*]] = icmp {{u?}}ne <[[NUM]] x i1> [[bvec1]], [[bvec2]]
+  // CHECK: [[res5:%[0-9]*]] = zext <[[NUM]] x i1> [[bres5]] to <[[NUM]] x i32>
+  res[5] = consequences[1] != consequences[2];
+
+  // CHECK: [[add3:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %consequences, i32 0, i32 3
+  // CHECK: [[vec3:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add3]]
+  // CHECK: [[bvec3:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec3]], zeroinitializer
+  // CHECK: [[bres6:%[0-9]*]] = icmp {{[osu]?}}lt <[[NUM]] x i1> [[bvec2]], [[bvec3]]
+  // CHECK: [[res6:%[0-9]*]] = zext <[[NUM]] x i1> [[bres6]] to <[[NUM]] x i32>
+  res[6] = consequences[2] <  consequences[3];
+
+  // CHECK: [[add4:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %consequences, i32 0, i32 4
+  // CHECK: [[vec4:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add4]]
+  // CHECK: [[bvec4:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec4]], zeroinitializer
+  // CHECK: [[bres7:%[0-9]*]] = icmp {{[osu]]?}}gt <[[NUM]] x i1> [[bvec3]], [[bvec4]]
+  // CHECK: [[res7:%[0-9]*]] = zext <[[NUM]] x i1> [[bres7]] to <[[NUM]] x i32>
+  res[7] = consequences[3] >  consequences[4];
+
+  // CHECK: [[add5:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %consequences, i32 0, i32 5
+  // CHECK: [[vec5:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add5]]
+  // CHECK: [[bvec5:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec5]], zeroinitializer
+  // CHECK: [[bres8:%[0-9]*]] = icmp {{[osu]]?}}le <[[NUM]] x i1> [[bvec4]], [[bvec5]]
+  // CHECK: [[res8:%[0-9]*]] = zext <[[NUM]] x i1> [[bres8]] to <[[NUM]] x i32>
+  res[8] = consequences[4] <= consequences[5];
+
+  // CHECK: [[add6:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %consequences, i32 0, i32 6
+  // CHECK: [[vec6:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add6]]
+  // CHECK: [[bvec6:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec6]], zeroinitializer
+  // CHECK: [[bres9:%[0-9]*]] = icmp {{[osu]?}}ge <[[NUM]] x i1> [[bvec5]], [[bvec6]]
+  // CHECK: [[res9:%[0-9]*]] = zext <[[NUM]] x i1> [[bres9]] to <[[NUM]] x i32>
+  res[9] = consequences[5] >= consequences[6];
+
+  // CHECK: [[add0:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 0
+  // CHECK: store <[[NUM]] x i32> [[res0]], <[[NUM]] x i32>* [[add0]]
+  // CHECK: [[add4:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 4
+  // CHECK: store <[[NUM]] x i32> [[res4]], <[[NUM]] x i32>* [[add4]]
+  // CHECK: [[add5:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 5
+  // CHECK: store <[[NUM]] x i32> [[res5]], <[[NUM]] x i32>* [[add5]]
+  // CHECK: [[add6:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 6
+  // CHECK: store <[[NUM]] x i32> [[res6]], <[[NUM]] x i32>* [[add6]]
+  // CHECK: [[add7:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 7
+  // CHECK: store <[[NUM]] x i32> [[res7]], <[[NUM]] x i32>* [[add7]]
+  // CHECK: [[add8:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 8
+  // CHECK: store <[[NUM]] x i32> [[res8]], <[[NUM]] x i32>* [[add8]]
+  // CHECK: [[add9:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 9
+  // CHECK: store <[[NUM]] x i32> [[res9]], <[[NUM]] x i32>* [[add9]]
+  // CHECK: ret void
+
+  return res;
+}
+
+static const int Ix = 2;
+
+// Test indexing operators
+// CHECK-LABEL: define void @"\01?index
+export vector<bool, NUM> index(vector<bool, NUM> things[10], int i, bool val)[10] {
+  vector<bool, NUM> res[10];
+
+  // CHECK: [[res:%[0-9]*]] = alloca [10 x <[[NUM]] x i32>]
+  // CHECK: [[res0:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* [[res]], i32 0, i32 0
+  // CHECK: store <[[NUM]] x i32> zeroinitializer, <[[NUM]] x i32>* [[res0]]
+  res[0] = 0;
+
+  // CHECK: [[resi:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* [[res]], i32 0, i32 %i
+  // CHECK: store <[[NUM]] x i32> <i32 1{{.*}}>, <[[NUM]] x i32>* [[resi]]
+  res[i] = 1;
+
+  // CHECK: [[res2:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* [[res]], i32 0, i32 2
+  // CHECK: store <[[NUM]] x i32> <i32 1{{.*}}>, <[[NUM]] x i32>* [[res2]]
+  res[Ix] = true;
+
+  // CHECK: [[add0:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %things, i32 0, i32 0
+  // CHECK: [[thg0:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add0]]
+  // CHECK: [[bthg0:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[thg0]], zeroinitializer
+  // CHECK: [[res3:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* [[res]], i32 0, i32 3
+  // CHECK: [[thg0:%[0-9]*]] = zext <[[NUM]] x i1> [[bthg0]] to <[[NUM]] x i32>
+  // CHECK: store <[[NUM]] x i32> [[thg0]], <[[NUM]] x i32>* [[res3]]
+  res[3] = things[0];
+
+  // CHECK: [[addi:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %things, i32 0, i32 %i
+  // CHECK: [[thgi:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[addi]]
+  // CHECK: [[bthgi:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[thgi]], zeroinitializer
+  // CHECK: [[res4:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* [[res]], i32 0, i32 4
+  // CHECK: [[thgi:%[0-9]*]] = zext <[[NUM]] x i1> [[bthgi]] to <[[NUM]] x i32>
+  // CHECK: store <[[NUM]] x i32> [[thgi]], <[[NUM]] x i32>* [[res4]]
+  res[4] = things[i];
+
+  // CHECK: [[add2:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %things, i32 0, i32 2
+  // CHECK: [[thg2:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add2]]
+  // CHECK: [[bthg2:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[thg2]], zeroinitializer
+  // CHECK: [[res5:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* [[res]], i32 0, i32 5
+  // CHECK: [[thg2:%[0-9]*]] = zext <[[NUM]] x i1> [[bthg2]] to <[[NUM]] x i32>
+  // CHECK: store <[[NUM]] x i32> [[thg2]], <[[NUM]] x i32>* [[res5]]
+  res[5] = things[Ix];
+  // CHECK: ret void
+  return res;
+
+}
+
+// Test bit twiddling operators.
+// CHECK-LABEL: define void @"\01?bittwiddlers
+export void bittwiddlers(inout vector<bool, NUM> things[10]) {
+  // CHECK: [[add2:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 2
+  // CHECK: [[vec2:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add2]]
+  // CHECK: [[bvec2:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec2]], zeroinitializer
+  // CHECK: [[vec2:%[0-9]*]] = zext <[[NUM]] x i1> [[bvec2]] to <[[NUM]] x i32>
+
+  // CHECK: [[add3:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 3
+  // CHECK: [[vec3:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add3]]
+  // CHECK: [[bvec3:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec3]], zeroinitializer
+  // CHECK: [[vec3:%[0-9]*]] = zext <[[NUM]] x i1> [[bvec3]] to <[[NUM]] x i32>
+  // CHECK: [[res1:%[0-9]*]] = or <[[NUM]] x [[TYPE]]> [[vec3]], [[vec2]]
+  // CHECK: [[bres1:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[res1]], zeroinitializer
+  // CHECK: [[add1:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %things, i32 0, i32 1
+  // CHECK: [[res1:%[0-9]*]] = zext <[[NUM]] x i1> [[bres1]] to <[[NUM]] x i32>
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res1]], <[[NUM]] x [[TYPE]]>* [[add1]]
+  things[1] = things[2] | things[3];
+
+  // CHECK: [[add4:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 4
+  // CHECK: [[vec4:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add4]]
+  // CHECK: [[bvec4:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec4]], zeroinitializer
+  // CHECK: [[bres2:%[0-9]*]] = and <[[NUM]] x i1> [[bvec4]], [[bvec3]]
+  // CHECK: [[res2:%[0-9]*]] = zext <[[NUM]] x i1> [[bres2]] to <[[NUM]] x i32>
+  // CHECK: [[bres2:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[res2]], zeroinitializer
+  // CHECK: [[res2:%[0-9]*]] = zext <[[NUM]] x i1> [[bres2]] to <[[NUM]] x i32>
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res2]], <[[NUM]] x [[TYPE]]>* [[add2]]
+  things[2] = things[3] & things[4];
+
+  // CHECK: [[vec4:%[0-9]*]] = zext <[[NUM]] x i1> [[bvec4]] to <[[NUM]] x i32>
+  // CHECK: [[add5:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 5
+  // CHECK: [[vec5:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add5]]
+  // CHECK: [[bvec5:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec5]], zeroinitializer
+  // CHECK: [[vec5:%[0-9]*]] = zext <[[NUM]] x i1> [[bvec5]] to <[[NUM]] x i32>
+  // CHECK: [[res3:%[0-9]*]] = xor <[[NUM]] x [[TYPE]]> [[vec4]], [[vec5]]
+  // CHECK: [[bres3:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[res3]], zeroinitializer
+  // CHECK: [[res3:%[0-9]*]] = zext <[[NUM]] x i1> [[bres3]] to <[[NUM]] x i32>
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res3]], <[[NUM]] x [[TYPE]]>* [[add3]]
+  things[3] = things[4] ^ things[5];
+
+  // CHECK: [[add6:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 6
+  // CHECK: [[vec6:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add6]]
+  // CHECK: [[bvec6:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec6]], zeroinitializer
+  // CHECK: [[bres4:%[0-9]*]] = or <[[NUM]] x i1> [[bvec6]], [[bvec4]]
+  // CHECK: [[res4:%[0-9]*]] = zext <[[NUM]] x i1> [[bres4]] to <[[NUM]] x i32>
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res4]], <[[NUM]] x [[TYPE]]>* [[add4]]
+  things[4] |= things[6];
+
+  // CHECK: [[add7:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 7
+  // CHECK: [[vec7:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add7]]
+  // CHECK: [[bvec7:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec7]], zeroinitializer
+  // CHECK: [[bres5:%[0-9]*]] = and <[[NUM]] x i1> [[bvec7]], [[bvec5]]
+  // CHECK: [[res5:%[0-9]*]] = zext <[[NUM]] x i1> [[bres5]] to <[[NUM]] x i32>
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res5]], <[[NUM]] x [[TYPE]]>* [[add5]]
+  things[5] &= things[7];
+
+  // CHECK: [[add8:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 8
+  // CHECK: [[vec8:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add8]]
+  // CHECK: [[bvec8:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec8]], zeroinitializer
+  // CHECK: [[bres6:%[0-9]*]] = xor <[[NUM]] x i1> [[bvec6]], [[bvec8]]
+  // CHECK: [[res6:%[0-9]*]] = zext <[[NUM]] x i1> [[bres6]] to <[[NUM]] x i32>
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res6]], <[[NUM]] x [[TYPE]]>* [[add6]]
+  things[6] ^= things[8];
+
+  // CHECK: ret void
+}
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/types/longvec-operators-int.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/types/longvec-operators-int.hlsl
new file mode 100644
index 0000000000..8c07f40af7
--- /dev/null
+++ b/tools/clang/test/CodeGenDXIL/hlsl/types/longvec-operators-int.hlsl
@@ -0,0 +1,58 @@
+// RUN: %dxc -HV 2018 -T lib_6_9   -DTYPE=uint     -DNUM=5 %s | FileCheck %s
+// RUN: %dxc -HV 2018 -T lib_6_9   -DTYPE=int64_t  -DNUM=3 %s | FileCheck %s
+// RUN: %dxc -HV 2018 -T lib_6_9   -DTYPE=uint16_t -DNUM=9 -enable-16bit-types %s | FileCheck %s
+
+// Test bitwise operators on an assortment vector sizes and integer types with 6.9 native vectors.
+
+// Test bit twiddling operators.
+// CHECK-LABEL: define void @"\01?bittwiddlers
+// CHECK-SAME: ([10 x <[[NUM:[0-9][0-9]*]] x [[TYPE:[a-z0-9]*]]>]*
+export void bittwiddlers(inout vector<TYPE, NUM> things[10]) {
+  // CHECK: [[add1:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 1
+  // CHECK: [[vec1:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add1]]
+  // CHECK: [[res1:%[0-9]*]] = xor <[[NUM]] x [[TYPE]]> [[vec1]], <[[TYPE]] -1,
+  // CHECK: [[add0:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 0
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res1]], <[[NUM]] x [[TYPE]]>* [[add0]]
+  things[0] = ~things[1];
+
+  // CHECK: [[add2:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 2
+  // CHECK: [[vec2:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add2]]
+
+  // CHECK: [[add3:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 3
+  // CHECK: [[vec3:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add3]]
+  // CHECK: [[res1:%[0-9]*]] = or <[[NUM]] x [[TYPE]]> [[vec3]], [[vec2]]
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res1]], <[[NUM]] x [[TYPE]]>* [[add1]]
+  things[1] = things[2] | things[3];
+
+  // CHECK: [[add4:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 4
+  // CHECK: [[vec4:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add4]]
+  // CHECK: [[res2:%[0-9]*]] = and <[[NUM]] x [[TYPE]]> [[vec4]], [[vec3]]
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res2]], <[[NUM]] x [[TYPE]]>* [[add2]]
+  things[2] = things[3] & things[4];
+
+  // CHECK: [[add5:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 5
+  // CHECK: [[vec5:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add5]]
+  // CHECK: [[res3:%[0-9]*]] = xor <[[NUM]] x [[TYPE]]> [[vec4]], [[vec5]]
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res3]], <[[NUM]] x [[TYPE]]>* [[add3]]
+  things[3] = things[4] ^ things[5];
+
+  // CHECK: [[add6:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 6
+  // CHECK: [[vec6:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add6]]
+  // CHECK: [[res4:%[0-9]*]] = or <[[NUM]] x [[TYPE]]> [[vec6]], [[vec4]]
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res4]], <[[NUM]] x [[TYPE]]>* [[add4]]
+  things[4] |= things[6];
+
+  // CHECK: [[add7:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 7
+  // CHECK: [[vec7:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add7]]
+  // CHECK: [[res5:%[0-9]*]] = and <[[NUM]] x [[TYPE]]> [[vec7]], [[vec5]]
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res5]], <[[NUM]] x [[TYPE]]>* [[add5]]
+  things[5] &= things[7];
+
+  // CHECK: [[add8:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 8
+  // CHECK: [[vec8:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add8]]
+  // CHECK: [[res6:%[0-9]*]] = xor <[[NUM]] x [[TYPE]]> [[vec6]], [[vec8]]
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res6]], <[[NUM]] x [[TYPE]]>* [[add6]]
+  things[6] ^= things[8];
+
+  // CHECK: ret void
+}
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/types/longvec-operators.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/types/longvec-operators.hlsl
new file mode 100644
index 0000000000..b3285267c9
--- /dev/null
+++ b/tools/clang/test/CodeGenDXIL/hlsl/types/longvec-operators.hlsl
@@ -0,0 +1,492 @@
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=float -DNUM=2 %s | FileCheck %s --check-prefixes=CHECK,NODBL
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=float -DNUM=3 %s | FileCheck %s --check-prefixes=CHECK,NODBL
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=float -DNUM=4 %s | FileCheck %s --check-prefixes=CHECK,NODBL
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=float -DNUM=5 %s | FileCheck %s --check-prefixes=CHECK,NODBL
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=float -DNUM=6 %s | FileCheck %s --check-prefixes=CHECK,NODBL
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=float -DNUM=7 %s | FileCheck %s --check-prefixes=CHECK,NODBL
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=float -DNUM=8 %s | FileCheck %s --check-prefixes=CHECK,NODBL
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=float -DNUM=9 %s | FileCheck %s --check-prefixes=CHECK,NODBL
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=float -DNUM=10 %s | FileCheck %s --check-prefixes=CHECK,NODBL
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=float -DNUM=11 %s | FileCheck %s --check-prefixes=CHECK,NODBL
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=float -DNUM=12 %s | FileCheck %s --check-prefixes=CHECK,NODBL
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=float -DNUM=13 %s | FileCheck %s --check-prefixes=CHECK,NODBL
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=float -DNUM=14 %s | FileCheck %s --check-prefixes=CHECK,NODBL
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=float -DNUM=15 %s | FileCheck %s --check-prefixes=CHECK,NODBL
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=float -DNUM=16 %s | FileCheck %s --check-prefixes=CHECK,NODBL
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=float -DNUM=17 %s | FileCheck %s --check-prefixes=CHECK,NODBL
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=float -DNUM=18 %s | FileCheck %s --check-prefixes=CHECK,NODBL
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=float -DNUM=128 %s | FileCheck %s --check-prefixes=CHECK,NODBL
+
+// Less exhaustive testing for some other types.
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=int      -DNUM=2 %s | FileCheck %s --check-prefixes=CHECK,NODBL
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=uint8_t4_packed      -DNUM=4 %s | FileCheck %s --check-prefixes=CHECK,NODBL
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=uint     -DNUM=5 %s | FileCheck %s --check-prefixes=CHECK,NODBL
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=double   -DNUM=3 -DDBL %s | FileCheck %s --check-prefixes=CHECK,DBL
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=uint64_t -DNUM=9 %s | FileCheck %s --check-prefixes=CHECK,NODBL
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=float16_t -DNUM=17 -enable-16bit-types %s | FileCheck %s --check-prefixes=CHECK,NODBL
+// RUN: %dxc -HV 2018 -T lib_6_9 -DTYPE=int16_t   -DNUM=177 -enable-16bit-types %s | FileCheck %s --check-prefixes=CHECK,NODBL
+
+// Test relevant operators on an assortment bool vector sizes and types with 6.9 native vectors.
+
+// Just a trick to capture the needed type spellings since the DXC version of FileCheck can't do that explicitly.
+// Uses non vector buffer to avoid interacting with that implementation.
+// CHECK: %dx.types.ResRet.[[TY:[a-z0-9]*]] = type { [[TYPE:[a-z_0-9]*]]
+
+RWStructuredBuffer< TYPE > buf;
+
+export void assignments(inout vector<TYPE, NUM> things[10], TYPE scales[10]);
+export vector<TYPE, NUM> arithmetic(inout vector<TYPE, NUM> things[11])[11];
+export vector<TYPE, NUM> scarithmetic(inout vector<TYPE, NUM> things[10], TYPE scales[10])[10];
+export vector<bool, NUM> logic(vector<bool, NUM> truth[10], vector<TYPE, NUM> consequences[10])[10];
+export vector<TYPE, NUM> index(vector<TYPE, NUM> things[10], int i, TYPE val)[10];
+
+struct Interface {
+  vector<TYPE, NUM> assigned[10];
+  vector<TYPE, NUM> arithmeticked[11];
+  vector<TYPE, NUM> scarithmeticked[10];
+  vector<bool, NUM> logicked[10];
+  vector<TYPE, NUM> indexed[10];
+  TYPE scales[10];
+};
+
+#if 0
+// Requires vector loading support. Enable when available.
+RWStructuredBuffer<Interface> Input;
+RWStructuredBuffer<Interface> Output;
+
+TYPE g_val;
+
+[shader("compute")]
+[numthreads(8,1,1)]
+void main(uint GI : SV_GroupIndex) {
+  assignments(Output[GI].assigned, Input[GI].scales);
+  Output[GI].arithmeticked = arithmetic(Input[GI].arithmeticked);
+  Output[GI].scarithmeticked = scarithmetic(Input[GI].scarithmeticked, Input[GI].scales);
+  Output[GI].logicked = logic(Input[GI].logicked, Input[GI].assigned);
+  Output[GI].indexed = index(Input[GI].indexed, GI, g_val);
+}
+#endif
+
+// A mixed-type overload to test overload resolution and mingle different vector element types in ops
+// Test assignment operators.
+// CHECK-LABEL: define void @"\01?assignments
+export void assignments(inout vector<TYPE, NUM> things[10], TYPE scales[10]) {
+
+  // Another trick to capture the size.
+  // CHECK: [[res:%[0-9]*]] = call %dx.types.ResRet.[[TY]] @dx.op.rawBufferLoad.[[TY]](i32 139, %dx.types.Handle %{{[^,]*}}, i32 [[NUM:[0-9]*]]
+  // CHECK: [[scl:%[0-9]*]] = extractvalue %dx.types.ResRet.[[TY]] [[res]], 0
+  TYPE scalar = buf.Load(NUM);
+
+  // CHECK: [[spt:%[0-9]*]] = insertelement <[[NUM]] x [[TYPE]]> undef, [[TYPE]] [[scl]], i32 0
+  // CHECK: [[res0:%[0-9]*]] = shufflevector <[[NUM]] x [[TYPE]]> [[spt]], <[[NUM]] x [[TYPE]]> undef, <[[NUM]] x i32> zeroinitializer
+  // CHECK: [[add0:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 0
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res0]], <[[NUM]] x [[TYPE]]>* [[add0]]
+  things[0] = scalar;
+
+  // CHECK: [[add5:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 5
+  // CHECK: [[vec5:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add5]]
+  // CHECK: [[add1:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 1
+  // CHECK: [[vec1:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add1]]
+  // CHECK: [[res1:%[0-9]*]] = [[ADD:f?add( fast)?]] <[[NUM]] x [[TYPE]]> [[vec1]], [[vec5]]
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res1]], <[[NUM]] x [[TYPE]]>* [[add1]]
+  things[1] += things[5];
+
+   // CHECK: [[add6:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 6
+  // CHECK: [[vec6:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add6]]
+  // CHECK: [[add2:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 2
+  // CHECK: [[vec2:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add2]]
+  // CHECK: [[res2:%[0-9]*]] = [[SUB:f?sub( fast)?]] <[[NUM]] x [[TYPE]]> [[vec2]], [[vec6]]
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res2]], <[[NUM]] x [[TYPE]]>* [[add2]]
+  things[2] -= things[6];
+
+  // CHECK: [[add7:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 7
+  // CHECK: [[vec7:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add7]]
+  // CHECK: [[add3:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 3
+  // CHECK: [[vec3:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add3]]
+  // CHECK: [[res3:%[0-9]*]] = [[MUL:f?mul( fast)?]] <[[NUM]] x [[TYPE]]> [[vec3]], [[vec7]]
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res3]], <[[NUM]] x [[TYPE]]>* [[add3]]
+  things[3] *= things[7];
+
+  // CHECK: [[add8:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 8
+  // CHECK: [[vec8:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add8]]
+  // CHECK: [[add4:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 4
+  // CHECK: [[vec4:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add4]]
+  // CHECK: [[res4:%[0-9]*]] = [[DIV:[ufs]?div( fast)?]] <[[NUM]] x [[TYPE]]> [[vec4]], [[vec8]]
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res4]], <[[NUM]] x [[TYPE]]>* [[add4]]
+  things[4] /= things[8];
+
+  // CHECK: [[add9:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 9
+  // CHECK: [[vec9:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add9]]
+#ifdef DBL
+  // DBL: [[fvec9:%[0-9]*]] = fptrunc <[[NUM]] x double> [[vec9]] to <[[NUM]] x float>
+  // DBL: [[fvec5:%[0-9]*]] = fptrunc <[[NUM]] x double> [[vec5]] to <[[NUM]] x float>
+  // DBL: [[fres5:%[0-9]*]] = [[REM:[ufs]?rem( fast)?]] <[[NUM]] x float> [[fvec5]], [[fvec9]]
+  // DBL: [[res5:%[0-9]*]] = fpext <[[NUM]] x float> [[fres5]] to <[[NUM]] x double>
+  vector<float,NUM> f9 = things[9];
+  vector<float,NUM> f5 = things[5];
+  f5 %= f9;
+  things[5] = f5;
+#else
+  // NODBL: [[res5:%[0-9]*]] = [[REM:[ufs]?rem( fast)?]] <[[NUM]] x [[TYPE]]> [[vec5]], [[vec9]]
+  things[5] %= things[9];
+#endif
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res5]], <[[NUM]] x [[TYPE]]>* [[add5]]
+
+  // CHECK: [[add1:%[0-9]*]] = getelementptr inbounds [10 x [[TYPE]]], [10 x [[TYPE]]]* %scales, i32 0, i32 1
+  // CHECK: [[scl1:%[0-9]*]] = load [[TYPE]], [[TYPE]]* [[add1]]
+  // CHECK: [[spt:%[0-9]*]] = insertelement <[[NUM]] x [[TYPE]]> undef, [[TYPE]] [[scl1]], i32 0
+  // CHECK: [[spt1:%[0-9]*]] = shufflevector <[[NUM]] x [[TYPE]]> [[spt]], <[[NUM]] x [[TYPE]]> undef, <[[NUM]] x i32> zeroinitializer
+  // CHECK: [[res6:%[0-9]*]] = [[ADD]] <[[NUM]] x [[TYPE]]> [[spt1]], [[vec6]]
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res6]], <[[NUM]] x [[TYPE]]>* [[add6]]
+  things[6] += scales[1];
+
+  // CHECK: [[add2:%[0-9]*]] = getelementptr inbounds [10 x [[TYPE]]], [10 x [[TYPE]]]* %scales, i32 0, i32 2
+  // CHECK: [[scl2:%[0-9]*]] = load [[TYPE]], [[TYPE]]* [[add2]]
+  // CHECK: [[spt:%[0-9]*]] = insertelement <[[NUM]] x [[TYPE]]> undef, [[TYPE]] [[scl2]], i32 0
+  // CHECK: [[spt2:%[0-9]*]] = shufflevector <[[NUM]] x [[TYPE]]> [[spt]], <[[NUM]] x [[TYPE]]> undef, <[[NUM]] x i32> zeroinitializer
+  // CHECK: [[res7:%[0-9]*]] = [[SUB]] <[[NUM]] x [[TYPE]]> [[vec7]], [[spt2]]
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res7]], <[[NUM]] x [[TYPE]]>* [[add7]]
+  things[7] -= scales[2];
+
+  // CHECK: [[add3:%[0-9]*]] = getelementptr inbounds [10 x [[TYPE]]], [10 x [[TYPE]]]* %scales, i32 0, i32 3
+  // CHECK: [[scl3:%[0-9]*]] = load [[TYPE]], [[TYPE]]* [[add3]]
+  // CHECK: [[spt:%[0-9]*]] = insertelement <[[NUM]] x [[TYPE]]> undef, [[TYPE]] [[scl3]], i32 0
+  // CHECK: [[spt3:%[0-9]*]] = shufflevector <[[NUM]] x [[TYPE]]> [[spt]], <[[NUM]] x [[TYPE]]> undef, <[[NUM]] x i32> zeroinitializer
+  // CHECK: [[res8:%[0-9]*]] = [[MUL]] <[[NUM]] x [[TYPE]]> [[spt3]], [[vec8]]
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res8]], <[[NUM]] x [[TYPE]]>* [[add8]]
+  things[8] *= scales[3];
+
+  // CHECK: [[add4:%[0-9]*]] = getelementptr inbounds [10 x [[TYPE]]], [10 x [[TYPE]]]* %scales, i32 0, i32 4
+  // CHECK: [[scl4:%[0-9]*]] = load [[TYPE]], [[TYPE]]* [[add4]]
+  // CHECK: [[spt:%[0-9]*]] = insertelement <[[NUM]] x [[TYPE]]> undef, [[TYPE]] [[scl4]], i32 0
+  // CHECK: [[spt4:%[0-9]*]] = shufflevector <[[NUM]] x [[TYPE]]> [[spt]], <[[NUM]] x [[TYPE]]> undef, <[[NUM]] x i32> zeroinitializer
+  // CHECK: [[res9:%[0-9]*]] = [[DIV]] <[[NUM]] x [[TYPE]]> [[vec9]], [[spt4]]
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res9]], <[[NUM]] x [[TYPE]]>* [[add9]]
+  things[9] /= scales[4];
+
+}
+
+// Test arithmetic operators.
+// CHECK-LABEL: define void @"\01?arithmetic
+export vector<TYPE, NUM> arithmetic(inout vector<TYPE, NUM> things[11])[11] {
+  vector<TYPE, NUM> res[11];
+  // CHECK: [[add0:%[0-9]*]] = getelementptr inbounds [11 x <[[NUM]] x [[TYPE]]>], [11 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 0
+  // CHECK: [[res1:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add0]]
+  // CHECK: [[res0:%[0-9]*]] = [[SUB]] <[[NUM]] x [[TYPE]]>
+  // CHECK: [[add1:%[0-9]*]] = getelementptr inbounds [11 x <[[NUM]] x [[TYPE]]>], [11 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 1
+  // CHECK: [[vec1:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add1]]
+  res[0] = -things[0];
+  res[1] = +things[0];
+
+  // CHECK: [[add2:%[0-9]*]] = getelementptr inbounds [11 x <[[NUM]] x [[TYPE]]>], [11 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 2
+  // CHECK: [[vec2:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add2]]
+  // CHECK: [[res2:%[0-9]*]] = [[ADD]] <[[NUM]] x [[TYPE]]> [[vec2]], [[vec1]]
+  res[2] = things[1] + things[2];
+
+  // CHECK: [[add3:%[0-9]*]] = getelementptr inbounds [11 x <[[NUM]] x [[TYPE]]>], [11 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 3
+  // CHECK: [[vec3:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add3]]
+  // CHECK: [[res3:%[0-9]*]] = [[SUB]] <[[NUM]] x [[TYPE]]> [[vec2]], [[vec3]]
+  res[3] = things[2] - things[3];
+
+  // CHECK: [[add4:%[0-9]*]] = getelementptr inbounds [11 x <[[NUM]] x [[TYPE]]>], [11 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 4
+  // CHECK: [[vec4:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add4]]
+  // CHECK: [[res4:%[0-9]*]] = [[MUL]] <[[NUM]] x [[TYPE]]> [[vec4]], [[vec3]]
+  res[4] = things[3] * things[4];
+
+  // CHECK: [[add5:%[0-9]*]] = getelementptr inbounds [11 x <[[NUM]] x [[TYPE]]>], [11 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 5
+  // CHECK: [[vec5:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add5]]
+  // CHECK: [[res5:%[0-9]*]] = [[DIV]] <[[NUM]] x [[TYPE]]> [[vec4]], [[vec5]]
+  res[5] = things[4] / things[5];
+
+  // DBL: [[fvec5:%[0-9]*]] = fptrunc <[[NUM]] x double> [[vec5]] to <[[NUM]] x float>
+  // CHECK: [[add6:%[0-9]*]] = getelementptr inbounds [11 x <[[NUM]] x [[TYPE]]>], [11 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 6
+  // CHECK: [[vec6:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add6]]
+#ifdef DBL
+  // DBL: [[fvec6:%[0-9]*]] = fptrunc <[[NUM]] x double> [[vec6]] to <[[NUM]] x float>
+  // DBL: [[fres6:%[0-9]*]] = [[REM]] <[[NUM]] x float> [[fvec5]], [[fvec6]]
+  // DBL: [[res6:%[0-9]*]] = fpext <[[NUM]] x float> [[fres6]] to <[[NUM]] x double>
+  res[6] = (vector<float,NUM>)things[5] % (vector<float,NUM>)things[6];
+#else
+  // NODBL: [[res6:%[0-9]*]] = [[REM]] <[[NUM]] x [[TYPE]]> [[vec5]], [[vec6]]
+  res[6] = things[5] % things[6];
+#endif
+
+  // CHECK: [[add7:%[0-9]*]] = getelementptr inbounds [11 x <[[NUM]] x [[TYPE]]>], [11 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 7
+  // CHECK: [[vec7:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add7]]
+  // CHECK: [[res7:%[0-9]*]] = [[ADD]] <[[NUM]] x [[TYPE]]> [[vec7]]
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res7]], <[[NUM]] x [[TYPE]]>* [[add7]]
+  res[7] = things[7]++;
+
+  // CHECK: [[add8:%[0-9]*]] = getelementptr inbounds [11 x <[[NUM]] x [[TYPE]]>], [11 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 8
+  // CHECK: [[vec8:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add8]]
+  // CHECK: [[res8:%[0-9]*]] = [[ADD]] <[[NUM]] x [[TYPE]]> [[vec8]]
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res8]], <[[NUM]] x [[TYPE]]>* [[add8]]
+  res[8] = things[8]--;
+
+  // CHECK: [[add9:%[0-9]*]] = getelementptr inbounds [11 x <[[NUM]] x [[TYPE]]>], [11 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 9
+  // CHECK: [[vec9:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add9]]
+  // CHECK: [[res9:%[0-9]*]] = [[ADD]] <[[NUM]] x [[TYPE]]> [[vec9]]
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res9]], <[[NUM]] x [[TYPE]]>* [[add9]]
+  res[9] = ++things[9];
+
+  // CHECK: [[add10:%[0-9]*]] = getelementptr inbounds [11 x <[[NUM]] x [[TYPE]]>], [11 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 10
+  // CHECK: [[vec10:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add10]]
+  // CHECK: [[res10:%[0-9]*]] = [[ADD]] <[[NUM]] x [[TYPE]]> [[vec10]]
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res10]], <[[NUM]] x [[TYPE]]>* [[add10]]
+  res[10] = --things[10];
+
+  // Stores into res[]. Previous were for things[] inout.
+  // CHECK: [[add0:%[0-9]*]] = getelementptr inbounds [11 x <[[NUM]] x [[TYPE]]>], [11 x <[[NUM]] x [[TYPE]]>]* %agg.result, i32 0, i32 0
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res0]], <[[NUM]] x [[TYPE]]>* [[add0]]
+  // CHECK: [[add1:%[0-9]*]] = getelementptr inbounds [11 x <[[NUM]] x [[TYPE]]>], [11 x <[[NUM]] x [[TYPE]]>]* %agg.result, i32 0, i32 1
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res1]], <[[NUM]] x [[TYPE]]>* [[add1]]
+  // CHECK: [[add2:%[0-9]*]] = getelementptr inbounds [11 x <[[NUM]] x [[TYPE]]>], [11 x <[[NUM]] x [[TYPE]]>]* %agg.result, i32 0, i32 2
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res2]], <[[NUM]] x [[TYPE]]>* [[add2]]
+  // CHECK: [[add3:%[0-9]*]] = getelementptr inbounds [11 x <[[NUM]] x [[TYPE]]>], [11 x <[[NUM]] x [[TYPE]]>]* %agg.result, i32 0, i32 3
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res3]], <[[NUM]] x [[TYPE]]>* [[add3]]
+  // CHECK: [[add4:%[0-9]*]] = getelementptr inbounds [11 x <[[NUM]] x [[TYPE]]>], [11 x <[[NUM]] x [[TYPE]]>]* %agg.result, i32 0, i32 4
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res4]], <[[NUM]] x [[TYPE]]>* [[add4]]
+  // CHECK: [[add5:%[0-9]*]] = getelementptr inbounds [11 x <[[NUM]] x [[TYPE]]>], [11 x <[[NUM]] x [[TYPE]]>]* %agg.result, i32 0, i32 5
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res5]], <[[NUM]] x [[TYPE]]>* [[add5]]
+  // CHECK: [[add6:%[0-9]*]] = getelementptr inbounds [11 x <[[NUM]] x [[TYPE]]>], [11 x <[[NUM]] x [[TYPE]]>]* %agg.result, i32 0, i32 6
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res6]], <[[NUM]] x [[TYPE]]>* [[add6]]
+  // These two were post ops, so the original value goes into res[].
+  // CHECK: [[add7:%[0-9]*]] = getelementptr inbounds [11 x <[[NUM]] x [[TYPE]]>], [11 x <[[NUM]] x [[TYPE]]>]* %agg.result, i32 0, i32 7
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[vec7]], <[[NUM]] x [[TYPE]]>* [[add7]]
+  // CHECK: [[add8:%[0-9]*]] = getelementptr inbounds [11 x <[[NUM]] x [[TYPE]]>], [11 x <[[NUM]] x [[TYPE]]>]* %agg.result, i32 0, i32 8
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[vec8]], <[[NUM]] x [[TYPE]]>* [[add8]]
+  // CHECK: [[add9:%[0-9]*]] = getelementptr inbounds [11 x <[[NUM]] x [[TYPE]]>], [11 x <[[NUM]] x [[TYPE]]>]* %agg.result, i32 0, i32 9
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res9]], <[[NUM]] x [[TYPE]]>* [[add9]]
+  // CHECK: [[add10:%[0-9]*]] = getelementptr inbounds [11 x <[[NUM]] x [[TYPE]]>], [11 x <[[NUM]] x [[TYPE]]>]* %agg.result, i32 0, i32 10
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res10]], <[[NUM]] x [[TYPE]]>* [[add10]]
+  // CHECK: ret void
+
+
+  return res;
+}
+
+// Test arithmetic operators with scalars.
+// CHECK-LABEL: define void @"\01?scarithmetic
+export vector<TYPE, NUM> scarithmetic(inout vector<TYPE, NUM> things[10], TYPE scales[10])[10] {
+  vector<TYPE, NUM> res[10];
+  // CHECK: [[add0:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 0
+  // CHECK: [[vec0:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add0]]
+  // CHECK: [[add1:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 1
+  // CHECK: [[vec1:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add1]]
+  // CHECK: [[add2:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 2
+  // CHECK: [[vec2:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add2]]
+  // CHECK: [[add3:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 3
+  // CHECK: [[vec3:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add3]]
+  // CHECK: [[add4:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 4
+  // CHECK: [[vec4:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add4]]
+  // CHECK: [[add5:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 5
+  // CHECK: [[vec5:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add5]]
+  // CHECK: [[add6:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 6
+  // CHECK: [[vec6:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add6]]
+  // CHECK: [[add0:%[0-9]*]] = getelementptr inbounds [10 x [[TYPE]]], [10 x [[TYPE]]]* %scales, i32 0, i32 0
+  // CHECK: [[scl0:%[0-9]*]] = load [[TYPE]], [[TYPE]]* [[add0]]
+  // CHECK: [[spt:%[0-9]*]] = insertelement <[[NUM]] x [[TYPE]]> undef, [[TYPE]] [[scl0]], i32 0
+  // CHECK: [[spt0:%[0-9]*]] = shufflevector <[[NUM]] x [[TYPE]]> [[spt]], <[[NUM]] x [[TYPE]]> undef, <[[NUM]] x i32> zeroinitializer
+  // CHECK: [[res0:%[0-9]*]] = [[ADD]] <[[NUM]] x [[TYPE]]> [[spt0]], [[vec0]]
+  res[0] = things[0] + scales[0];
+
+  // CHECK: [[add1:%[0-9]*]] = getelementptr inbounds [10 x [[TYPE]]], [10 x [[TYPE]]]* %scales, i32 0, i32 1
+  // CHECK: [[scl1:%[0-9]*]] = load [[TYPE]], [[TYPE]]* [[add1]]
+  // CHECK: [[spt:%[0-9]*]] = insertelement <[[NUM]] x [[TYPE]]> undef, [[TYPE]] [[scl1]], i32 0
+  // CHECK: [[spt1:%[0-9]*]] = shufflevector <[[NUM]] x [[TYPE]]> [[spt]], <[[NUM]] x [[TYPE]]> undef, <[[NUM]] x i32> zeroinitializer
+  // CHECK: [[res1:%[0-9]*]] = [[SUB]] <[[NUM]] x [[TYPE]]> [[vec1]], [[spt1]]
+  res[1] = things[1] - scales[1];
+
+
+  // CHECK: [[add2:%[0-9]*]] = getelementptr inbounds [10 x [[TYPE]]], [10 x [[TYPE]]]* %scales, i32 0, i32 2
+  // CHECK: [[scl2:%[0-9]*]] = load [[TYPE]], [[TYPE]]* [[add2]]
+  // CHECK: [[spt:%[0-9]*]] = insertelement <[[NUM]] x [[TYPE]]> undef, [[TYPE]] [[scl2]], i32 0
+  // CHECK: [[spt2:%[0-9]*]] = shufflevector <[[NUM]] x [[TYPE]]> [[spt]], <[[NUM]] x [[TYPE]]> undef, <[[NUM]] x i32> zeroinitializer
+  // CHECK: [[res2:%[0-9]*]] = [[MUL]] <[[NUM]] x [[TYPE]]> [[spt2]], [[vec2]]
+  res[2] = things[2] * scales[2];
+
+  // CHECK: [[add3:%[0-9]*]] = getelementptr inbounds [10 x [[TYPE]]], [10 x [[TYPE]]]* %scales, i32 0, i32 3
+  // CHECK: [[scl3:%[0-9]*]] = load [[TYPE]], [[TYPE]]* [[add3]]
+  // CHECK: [[spt:%[0-9]*]] = insertelement <[[NUM]] x [[TYPE]]> undef, [[TYPE]] [[scl3]], i32 0
+  // CHECK: [[spt3:%[0-9]*]] = shufflevector <[[NUM]] x [[TYPE]]> [[spt]], <[[NUM]] x [[TYPE]]> undef, <[[NUM]] x i32> zeroinitializer
+  // CHECK: [[res3:%[0-9]*]] = [[DIV]] <[[NUM]] x [[TYPE]]> [[vec3]], [[spt3]]
+  res[3] = things[3] / scales[3];
+
+  // CHECK: [[add4:%[0-9]*]] = getelementptr inbounds [10 x [[TYPE]]], [10 x [[TYPE]]]* %scales, i32 0, i32 4
+  // CHECK: [[scl4:%[0-9]*]] = load [[TYPE]], [[TYPE]]* [[add4]]
+  // CHECK: [[spt:%[0-9]*]] = insertelement <[[NUM]] x [[TYPE]]> undef, [[TYPE]] [[scl4]], i32 0
+  // CHECK: [[spt4:%[0-9]*]] = shufflevector <[[NUM]] x [[TYPE]]> [[spt]], <[[NUM]] x [[TYPE]]> undef, <[[NUM]] x i32> zeroinitializer
+  // CHECK: [[res4:%[0-9]*]] = [[ADD]] <[[NUM]] x [[TYPE]]> [[spt4]], [[vec4]]
+  res[4] = scales[4] + things[4];
+
+  // CHECK: [[add5:%[0-9]*]] = getelementptr inbounds [10 x [[TYPE]]], [10 x [[TYPE]]]* %scales, i32 0, i32 5
+  // CHECK: [[scl5:%[0-9]*]] = load [[TYPE]], [[TYPE]]* [[add5]]
+  // CHECK: [[spt:%[0-9]*]] = insertelement <[[NUM]] x [[TYPE]]> undef, [[TYPE]] [[scl5]], i32 0
+  // CHECK: [[spt5:%[0-9]*]] = shufflevector <[[NUM]] x [[TYPE]]> [[spt]], <[[NUM]] x [[TYPE]]> undef, <[[NUM]] x i32> zeroinitializer
+  // CHECK: [[res5:%[0-9]*]] = [[SUB]] <[[NUM]] x [[TYPE]]> [[spt5]], [[vec5]]
+  res[5] = scales[5] - things[5];
+
+  // CHECK: [[add6:%[0-9]*]] = getelementptr inbounds [10 x [[TYPE]]], [10 x [[TYPE]]]* %scales, i32 0, i32 6
+  // CHECK: [[scl6:%[0-9]*]] = load [[TYPE]], [[TYPE]]* [[add6]]
+  // CHECK: [[spt:%[0-9]*]] = insertelement <[[NUM]] x [[TYPE]]> undef, [[TYPE]] [[scl6]], i32 0
+  // CHECK: [[spt6:%[0-9]*]] = shufflevector <[[NUM]] x [[TYPE]]> [[spt]], <[[NUM]] x [[TYPE]]> undef, <[[NUM]] x i32> zeroinitializer
+  // CHECK: [[res6:%[0-9]*]] = [[MUL]] <[[NUM]] x [[TYPE]]> [[spt6]], [[vec6]]
+  res[6] = scales[6] * things[6];
+
+  // CHECK: [[add0:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %agg.result, i32 0, i32 0
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res0]], <[[NUM]] x [[TYPE]]>* [[add0]]
+  // CHECK: [[add1:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %agg.result, i32 0, i32 1
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res1]], <[[NUM]] x [[TYPE]]>* [[add1]]
+  // CHECK: [[add2:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %agg.result, i32 0, i32 2
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res2]], <[[NUM]] x [[TYPE]]>* [[add2]]
+  // CHECK: [[add3:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %agg.result, i32 0, i32 3
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res3]], <[[NUM]] x [[TYPE]]>* [[add3]]
+  // CHECK: [[add4:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %agg.result, i32 0, i32 4
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res4]], <[[NUM]] x [[TYPE]]>* [[add4]]
+  // CHECK: [[add5:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %agg.result, i32 0, i32 5
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res5]], <[[NUM]] x [[TYPE]]>* [[add5]]
+  // CHECK: [[add6:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %agg.result, i32 0, i32 6
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[res6]], <[[NUM]] x [[TYPE]]>* [[add6]]
+  // CHECK: ret void
+
+
+  return res;
+}
+
+// Test logic operators.
+// Only permissable in pre-HLSL2021
+// CHECK-LABEL: define void @"\01?logic
+export vector<bool, NUM> logic(vector<bool, NUM> truth[10], vector<TYPE, NUM> consequences[10])[10] {
+  vector<bool, NUM> res[10];
+  // CHECK: [[add0:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %truth, i32 0, i32 0
+  // CHECK: [[vec0:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add0]]
+  // CHECK: [[cmp:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec0]], zeroinitializer
+  // CHECK: [[cmp0:%[0-9]*]] = icmp eq <[[NUM]] x i1> [[cmp]], zeroinitializer
+  // CHECK: [[res0:%[0-9]*]] = zext <[[NUM]] x i1> [[cmp0]] to <[[NUM]] x i32>
+  res[0] = !truth[0];
+
+  // CHECK: [[add1:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %truth, i32 0, i32 1
+  // CHECK: [[vec1:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add1]]
+  // CHECK: [[bvec1:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec1]], zeroinitializer
+  // CHECK: [[add2:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %truth, i32 0, i32 2
+  // CHECK: [[vec2:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add2]]
+  // CHECK: [[bvec2:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec2]], zeroinitializer
+  // CHECK: [[bres1:%[0-9]*]] = or <[[NUM]] x i1> [[bvec2]], [[bvec1]]
+  // CHECK: [[res1:%[0-9]*]] = zext <[[NUM]] x i1> [[bres1]] to <[[NUM]] x i32>
+  res[1] = truth[1] || truth[2];
+
+  // CHECK: [[add3:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %truth, i32 0, i32 3
+  // CHECK: [[vec3:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add3]]
+  // CHECK: [[bvec3:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec3]], zeroinitializer
+  // CHECK: [[bres2:%[0-9]*]] = and <[[NUM]] x i1> [[bvec3]], [[bvec2]]
+  // CHECK: [[res2:%[0-9]*]] = zext <[[NUM]] x i1> [[bres2]] to <[[NUM]] x i32>
+  res[2] = truth[2] && truth[3];
+
+  // CHECK: [[add4:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %truth, i32 0, i32 4
+  // CHECK: [[vec4:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add4]]
+  // CHECK: [[bvec4:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec4]], zeroinitializer
+  // CHECK: [[add5:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %truth, i32 0, i32 5
+  // CHECK: [[vec5:%[0-9]*]] = load <[[NUM]] x i32>, <[[NUM]] x i32>* [[add5]]
+  // CHECK: [[bvec5:%[0-9]*]] = icmp ne <[[NUM]] x i32> [[vec5]], zeroinitializer
+
+  // NOT RIGHT STUFF.. Select is still extracting everything, slows WAY down with over 100 elements
+
+  res[3] = truth[3] ? truth[4] : truth[5];
+
+  // CHECK: [[add0:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %consequences, i32 0, i32 0
+  // CHECK: [[vec0:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add0]]
+  // CHECK: [[add1:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %consequences, i32 0, i32 1
+  // CHECK: [[vec1:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add1]]
+  // CHECK: [[cmp4:%[0-9]*]] = [[CMP:[fi]?cmp( fast)?]] {{o?}}eq <[[NUM]] x [[TYPE]]> [[vec0]], [[vec1]]
+  // CHECK: [[res4:%[0-9]*]] = zext <[[NUM]] x i1> [[cmp4]] to <[[NUM]] x i32>
+  res[4] = consequences[0] == consequences[1];
+
+  // CHECK: [[add2:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %consequences, i32 0, i32 2
+  // CHECK: [[vec2:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add2]]
+  // CHECK: [[cmp5:%[0-9]*]] = [[CMP]] {{u?}}ne <[[NUM]] x [[TYPE]]> [[vec1]], [[vec2]]
+  // CHECK: [[res5:%[0-9]*]] = zext <[[NUM]] x i1> [[cmp5]] to <[[NUM]] x i32>
+  res[5] = consequences[1] != consequences[2];
+
+  // CHECK: [[add3:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %consequences, i32 0, i32 3
+  // CHECK: [[vec3:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add3]]
+  // CHECK: [[cmp6:%[0-9]*]] = [[CMP]] {{[osu]?}}lt <[[NUM]] x [[TYPE]]> [[vec2]], [[vec3]]
+  // CHECK: [[res6:%[0-9]*]] = zext <[[NUM]] x i1> [[cmp6]] to <[[NUM]] x i32>
+  res[6] = consequences[2] <  consequences[3];
+
+  // CHECK: [[add4:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %consequences, i32 0, i32 4
+  // CHECK: [[vec4:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add4]]
+  // CHECK: [[cmp7:%[0-9]*]] = [[CMP]] {{[osu]]?}}gt <[[NUM]] x [[TYPE]]> [[vec3]], [[vec4]]
+  // CHECK: [[res7:%[0-9]*]] = zext <[[NUM]] x i1> [[cmp7]] to <[[NUM]] x i32>
+  res[7] = consequences[3] >  consequences[4];
+
+  // CHECK: [[add5:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %consequences, i32 0, i32 5
+  // CHECK: [[vec5:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add5]]
+  // CHECK: [[cmp8:%[0-9]*]] = [[CMP]] {{[osu]]?}}le <[[NUM]] x [[TYPE]]> [[vec4]], [[vec5]]
+  // CHECK: [[res8:%[0-9]*]] = zext <[[NUM]] x i1> [[cmp8]] to <[[NUM]] x i32>
+  res[8] = consequences[4] <= consequences[5];
+
+  // CHECK: [[add6:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %consequences, i32 0, i32 6
+  // CHECK: [[vec6:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add6]]
+  // CHECK: [[cmp9:%[0-9]*]] = [[CMP]] {{[osu]?}}ge <[[NUM]] x [[TYPE]]> [[vec5]], [[vec6]]
+  // CHECK: [[res9:%[0-9]*]] = zext <[[NUM]] x i1> [[cmp9]] to <[[NUM]] x i32>
+  res[9] = consequences[5] >= consequences[6];
+
+  // CHECK: [[add0:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 0
+  // CHECK: store <[[NUM]] x i32> [[res0]], <[[NUM]] x i32>* [[add0]]
+  // CHECK: [[add4:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 4
+  // CHECK: store <[[NUM]] x i32> [[res4]], <[[NUM]] x i32>* [[add4]]
+  // CHECK: [[add5:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 5
+  // CHECK: store <[[NUM]] x i32> [[res5]], <[[NUM]] x i32>* [[add5]]
+  // CHECK: [[add6:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 6
+  // CHECK: store <[[NUM]] x i32> [[res6]], <[[NUM]] x i32>* [[add6]]
+  // CHECK: [[add7:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 7
+  // CHECK: store <[[NUM]] x i32> [[res7]], <[[NUM]] x i32>* [[add7]]
+  // CHECK: [[add8:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 8
+  // CHECK: store <[[NUM]] x i32> [[res8]], <[[NUM]] x i32>* [[add8]]
+  // CHECK: [[add9:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x i32>], [10 x <[[NUM]] x i32>]* %agg.result, i32 0, i32 9
+  // CHECK: store <[[NUM]] x i32> [[res9]], <[[NUM]] x i32>* [[add9]]
+  // CHECK: ret void
+
+  return res;
+}
+
+static const int Ix = 2;
+
+// Test indexing operators
+// CHECK-LABEL: define void @"\01?index
+export vector<TYPE, NUM> index(vector<TYPE, NUM> things[10], int i, TYPE val)[10] {
+  vector<TYPE, NUM> res[10];
+
+  // CHECK: [[res:%[0-9]*]] = alloca [10 x <[[NUM]] x [[TYPE]]>]
+  // CHECK: [[res0:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* [[res]], i32 0, i32 0
+  // CHECK: store <[[NUM]] x [[TYPE]]> zeroinitializer, <[[NUM]] x [[TYPE]]>* [[res0]]
+  res[0] = 0;
+
+  // CHECK: [[resi:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* [[res]], i32 0, i32 %i
+  // CHECK: store <[[NUM]] x [[TYPE]]> <[[TYPE]] {{(1|0xH3C00).*}}>, <[[NUM]] x [[TYPE]]>* [[resi]]
+  res[i] = 1;
+
+  // CHECK: [[res2:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* [[res]], i32 0, i32 2
+  // CHECK: store <[[NUM]] x [[TYPE]]> <[[TYPE]] {{(2|0xH4000).*}}>, <[[NUM]] x [[TYPE]]>* [[res2]]
+  res[Ix] = 2;
+
+  // CHECK: [[add0:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 0
+  // CHECK: [[thg0:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add0]]
+  // CHECK: [[res3:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* [[res]], i32 0, i32 3
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[thg0]], <[[NUM]] x [[TYPE]]>* [[res3]]
+  res[3] = things[0];
+
+  // CHECK: [[addi:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 %i
+  // CHECK: [[thgi:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[addi]]
+  // CHECK: [[res4:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* [[res]], i32 0, i32 4
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[thgi]], <[[NUM]] x [[TYPE]]>* [[res4]]
+  res[4] = things[i];
+
+  // CHECK: [[add2:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* %things, i32 0, i32 2
+  // CHECK: [[thg2:%[0-9]*]] = load <[[NUM]] x [[TYPE]]>, <[[NUM]] x [[TYPE]]>* [[add2]]
+  // CHECK: [[res5:%[0-9]*]] = getelementptr inbounds [10 x <[[NUM]] x [[TYPE]]>], [10 x <[[NUM]] x [[TYPE]]>]* [[res]], i32 0, i32 5
+  // CHECK: store <[[NUM]] x [[TYPE]]> [[thg2]], <[[NUM]] x [[TYPE]]>* [[res5]]
+  res[5] = things[Ix];
+  // CHECK: ret void
+  return res;
+}
diff --git a/tools/clang/test/CodeGenDXIL/hlsl/types/longvecs.hlsl b/tools/clang/test/CodeGenDXIL/hlsl/types/longvecs.hlsl
new file mode 100644
index 0000000000..1910e08a25
--- /dev/null
+++ b/tools/clang/test/CodeGenDXIL/hlsl/types/longvecs.hlsl
@@ -0,0 +1,154 @@
+// RUN: %dxc -Wno-conversion -T cs_6_9       %s | FileCheck %s --check-prefixes=CHECK,F32
+// RUN: %dxc -Wno-conversion -T cs_6_9 -DF64 %s | FileCheck %s --check-prefixes=CHECK,F64
+
+RWByteAddressBuffer buf;
+
+// "TYPE" is the mainly focused test type.
+// "UNTYPE" is the other type used for mixed precision testing.
+#ifdef F64
+typedef double TYPE;
+typedef float UNTYPE;
+#else
+typedef float TYPE;
+typedef double UNTYPE;
+#endif
+
+// Two main test function overloads. One expects matching element types.
+// The other uses different types to test ops and overload resolution.
+template <typename T, int N> vector<T, N> dostuff(vector<T, N> thing1, vector<T, N> thing2, vector<T, N> thing3);
+vector<TYPE, 8> dostuff(vector<TYPE, 8> thing1, vector<UNTYPE, 8> thing2, vector<TYPE, 8> thing3);
+
+// Just a trick to capture the needed type spellings since the DXC version of FileCheck can't do that explicitly.
+// F32-DAG: %dx.types.ResRet.[[TY:v8f32]] = type { [[TYPE:<8 x float>]]
+// F32-DAG: %dx.types.ResRet.[[UNTY:v8f64]] = type { [[UNTYPE:<8 x double>]]
+// F64-DAG: %dx.types.ResRet.[[TY:v8f64]] = type { [[TYPE:<8 x double>]]
+// F64-DAG: %dx.types.ResRet.[[UNTY:v8f32]] = type { [[UNTYPE:<8 x float>]]
+
+// Verify that groupshared vectors are kept as aggregates
+// CHECK: @"\01?gs_vec1@@3V?$vector@{{M|N}}$07@@A" = external addrspace(3) global [[TYPE]]
+// CHECK: @"\01?gs_vec2@@3V?$vector@{{M|N}}$07@@A" = external addrspace(3) global [[TYPE]]
+// CHECK: @"\01?gs_vec3@@3V?$vector@{{M|N}}$07@@A" = external addrspace(3) global [[TYPE]]
+groupshared vector<TYPE, 8> gs_vec1, gs_vec2, gs_vec3;
+
+[numthreads(8,1,1)]
+void main() {
+  // CHECK: [[buf:%.*]] = call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle %1, %dx.types.ResourceProperties { i32 4107, i32 0 })  ; AnnotateHandle(res,props)  resource: RWByteAddressBuffer
+
+  // CHECK: [[vec1_res:%.*]] = call %dx.types.ResRet.[[TY]] @dx.op.rawBufferVectorLoad.[[TY]](i32 303, %dx.types.Handle [[buf]], i32 0
+  // CHECK-DAG: [[vec1:%.*]] = extractvalue %dx.types.ResRet.[[TY]] [[vec1_res]], 0
+  // F32-DAG: [[vec1_32:%.*]] = extractvalue %dx.types.ResRet.[[TY]] [[vec1_res]], 0
+  // F64-DAG: [[vec1_64:%.*]] = extractvalue %dx.types.ResRet.[[TY]] [[vec1_res]], 0
+  vector<TYPE, 8> vec1 = buf.Load<vector<TYPE, 8> >(0);
+
+  // CHECK: [[vec2_res:%.*]] = call %dx.types.ResRet.[[TY]] @dx.op.rawBufferVectorLoad.[[TY]](i32 303, %dx.types.Handle [[buf]], i32 60
+  // CHECK-DAG: [[vec2:%.*]] = extractvalue %dx.types.ResRet.[[TY]] [[vec2_res]], 0
+  // F32-DAG: [[vec2_32:%.*]] = extractvalue %dx.types.ResRet.[[TY]] [[vec2_res]], 0
+  // F64-DAG: [[vec2_64:%.*]] = extractvalue %dx.types.ResRet.[[TY]] [[vec2_res]], 0
+  vector<TYPE, 8> vec2 = buf.Load<vector<TYPE, 8> >(60);
+
+  // CHECK: [[vec3_res:%.*]] = call %dx.types.ResRet.[[TY]] @dx.op.rawBufferVectorLoad.[[TY]](i32 303, %dx.types.Handle [[buf]], i32 120
+  // CHECK-DAG: [[vec3:%.*]] = extractvalue %dx.types.ResRet.[[TY]] [[vec3_res]], 0
+  // F64-DAG: [[vec3_64:%.*]] = extractvalue %dx.types.ResRet.[[TY]] [[vec3_res]], 0
+  vector<TYPE, 8> vec3 = buf.Load<vector<TYPE, 8> >(120);
+
+  // CHECK: [[unvec_res:%.*]] = call %dx.types.ResRet.[[UNTY]] @dx.op.rawBufferVectorLoad.[[UNTY]](i32 303, %dx.types.Handle [[buf]], i32 180
+  // CHECK-DAG: [[unvec:%.*]] = extractvalue %dx.types.ResRet.[[UNTY]] [[unvec_res]], 0
+  // F32-DAG: [[unvec_64:%.*]] = extractvalue %dx.types.ResRet.[[UNTY]] [[unvec_res]], 0
+  // F64-DAG: [[unvec_32:%.*]] = extractvalue %dx.types.ResRet.[[UNTY]] [[unvec_res]], 0
+  vector<UNTYPE, 8> unvec = buf.Load<vector<UNTYPE, 8> >(180);
+
+  vec1 = dostuff(vec1, vec2, vec3);
+
+  // Test mixed type operations
+  vec2 = dostuff(vec2, unvec, vec3);
+
+  gs_vec2 = dostuff(gs_vec1, gs_vec2, gs_vec3);
+
+  // mix groupshared and non
+  //vec1 = dostuff(vec1, gs_vec2, vec3);
+
+  buf.Store<vector<TYPE, 8> >(240, vec1 * vec2 - vec3 * gs_vec1 + gs_vec2 / gs_vec3);
+}
+
+//  Test the required ops on long vectors and confirm correct lowering.
+template <typename T, int N>
+vector<T, N> dostuff(vector<T, N> thing1, vector<T, N> thing2, vector<T, N> thing3) {
+  vector<T, N> res = 0;
+
+  // CHECK: call [[TYPE]] @dx.op.binary.[[TY]](i32 36, [[TYPE]] [[vec1]], [[TYPE]] [[vec2]])  ; FMin(a,b)
+  res += min(thing1, thing2);
+  // CHECK: call [[TYPE]] @dx.op.binary.[[TY]](i32 35, [[TYPE]] [[vec1]], [[TYPE]] [[vec3]])  ; FMax(a,b)
+  res += max(thing1, thing3);
+
+  // CHECK: [[tmp:%.*]] = call [[TYPE]] @dx.op.binary.[[TY]](i32 35, [[TYPE]] [[vec1]], [[TYPE]] [[vec2]])  ; FMax(a,b)
+  // CHECK: call [[TYPE]] @dx.op.binary.[[TY]](i32 36, [[TYPE]] [[tmp]], [[TYPE]] [[vec3]])  ; FMin(a,b)
+  res += clamp(thing1, thing2, thing3);
+
+  // F32: [[vec3_64:%.*]] = fpext <8 x float> [[vec3]] to <8 x double>
+  // F32: [[vec2_64:%.*]] = fpext <8 x float> [[vec2]] to <8 x double>
+  // F32: [[vec1_64:%.*]] = fpext <8 x float> [[vec1]] to <8 x double>
+  // CHECK: call <8 x double> @dx.op.tertiary.v8f64(i32 47, <8 x double> [[vec1_64]], <8 x double> [[vec2_64]], <8 x double> [[vec3_64]]) ; Fma(a,b,c)
+  res += (vector<T, N>)fma((vector<double, N>)thing1, (vector<double, N>)(thing2), (vector<double, N>)thing3);
+
+  // Even in the double test, these will be downconverted because these builtins only take floats.
+  // F64: [[vec2_32:%.*]] = fptrunc <8 x double> [[vec2]] to <8 x float>
+  // F64: [[vec1_32:%.*]] = fptrunc <8 x double> [[vec1]] to <8 x float>
+
+  // CHECK: [[tmp:%.*]] = fcmp fast olt <8 x float> [[vec2_32]], [[vec1_32]]
+  // CHECK: select <8 x i1> [[tmp]], [[TYPE]] zeroinitializer, [[TYPE]]
+  res += step(thing1, thing2);
+
+  // CHECK: [[tmp:%.*]] = fmul fast <8 x float> [[vec1_32]], <float 0x
+  // CHECK: call <8 x float> @dx.op.unary.v8f32(i32 21, <8 x float> [[tmp]])  ; Exp(value)
+  res += exp(thing1);
+
+  // CHECK: [[tmp:%.*]] = call <8 x float> @dx.op.unary.v8f32(i32 23, <8 x float> [[vec1_32]])  ; Log(value)
+  // CHECK: fmul fast <8 x float> [[tmp]], <float 0x
+  res += log(thing1);
+
+  // CHECK: call <8 x float> @dx.op.unary.v8f32(i32 20, <8 x float> [[vec1_32]])  ; Htan(value)
+  res += tanh(thing1);
+  // CHECK: call <8 x float> @dx.op.unary.v8f32(i32 17, <8 x float> [[vec1_32]])  ; Atan(value)
+  res += atan(thing1);
+
+  return res;
+}
+
+// A mixed-type overload to test overload resolution and mingle different vector element types in ops
+vector<TYPE, 8> dostuff(vector<TYPE, 8> thing1, vector<UNTYPE, 8> thing2, vector<TYPE, 8> thing3) {
+  vector<TYPE, 8> res = 0;
+
+  // F64: [[unvec_64:%.*]] = fpext <8 x float> [[unvec]] to <8 x double>
+  // CHECK: call <8 x double> @dx.op.binary.v8f64(i32 36, <8 x double> [[vec2_64]], <8 x double> [[unvec_64]])  ; FMin(a,b)
+  res += min(thing1, thing2);
+
+  // CHECK: call [[TYPE]] @dx.op.binary.[[TY]](i32 35, [[TYPE]] [[vec2]], [[TYPE]] [[vec3]]) ; FMax(a,b)
+  res += max(thing1, thing3);
+
+  // CHECK: [[tmp:%.*]] = call <8 x double> @dx.op.binary.v8f64(i32 35, <8 x double> [[vec2_64]], <8 x double> [[unvec_64]])  ; FMax(a,b)
+  // CHECK: call <8 x double> @dx.op.binary.v8f64(i32 36, <8 x double> [[tmp]], <8 x double> [[vec3_64]])  ; FMin(a,b)
+  res += clamp(thing1, thing2, thing3);
+
+  // CHECK: call <8 x double> @dx.op.tertiary.v8f64(i32 47, <8 x double> [[vec2_64]], <8 x double> [[unvec_64]], <8 x double> [[vec3_64]]) ; Fma(a,b,c)
+  res += (vector<TYPE, 8>)fma((vector<double,8>)thing1, (vector<double,8>)(thing2), (vector<double,8>)thing3);
+
+  // F32: [[unvec_32:%.*]] = fptrunc <8 x double> [[unvec]] to <8 x float>
+  // CHECK: [[tmp:%.*]] = fcmp fast olt <8 x float> [[unvec_32]], [[vec2_32]]
+  // CHECK: select <8 x i1> [[tmp]], [[TYPE]] zeroinitializer, [[TYPE]]
+  res += step(thing1, thing2);
+
+  // CHECK: [[tmp:%.*]] = fmul fast <8 x float> [[vec2_32]], <float 0x
+  // CHECK: call <8 x float> @dx.op.unary.v8f32(i32 21, <8 x float> [[tmp]])  ; Exp(value)
+  res += exp(thing1);
+
+  // CHECK: [[tmp:%.*]] = call <8 x float> @dx.op.unary.v8f32(i32 23, <8 x float> [[vec2_32]])  ; Log(value)
+  // CHECK: fmul fast <8 x float> [[tmp]], <float 0x
+  res += log(thing1);
+
+  // CHECK: call <8 x float> @dx.op.unary.v8f32(i32 20, <8 x float> [[vec2_32]])  ; Htan(value)
+  res += tanh(thing1);
+  // CHECK: call <8 x float> @dx.op.unary.v8f32(i32 17, <8 x float> [[vec2_32]])  ; Atan(value)
+  res += atan(thing1);
+
+  return res;
+}
diff --git a/tools/clang/test/HLSLFileCheck/hlsl/types/matrix/matrix-ast.hlsl b/tools/clang/test/HLSLFileCheck/hlsl/types/matrix/matrix-ast.hlsl
index 33086852ab..5443ada0c9 100644
--- a/tools/clang/test/HLSLFileCheck/hlsl/types/matrix/matrix-ast.hlsl
+++ b/tools/clang/test/HLSLFileCheck/hlsl/types/matrix/matrix-ast.hlsl
@@ -15,6 +15,7 @@
 // ext_vector array.
 // CHECK-NEXT: CXXRecordDecl {{0x[0-9a-fA-F]+}} <<invalid sloc>> <invalid sloc> implicit class matrix definition
 // CHECK-NEXT: FinalAttr {{0x[0-9a-fA-F]+}} <<invalid sloc>> Implicit final
+// CHECK-NEXT: HLSLMatrixAttr {{0x[0-9a-fA-F]+}} <<invalid sloc>> Implicit
 // CHECK-NEXT: FieldDecl {{0x[0-9a-fA-F]+}} <<invalid sloc>> <invalid sloc> implicit h 'element [row_count] __attribute__((ext_vector_type(col_count)))'
 
 
diff --git a/tools/clang/test/HLSLFileCheck/hlsl/types/vector/vector-ast.hlsl b/tools/clang/test/HLSLFileCheck/hlsl/types/vector/vector-ast.hlsl
index 0ad236a4b2..12859b7eda 100644
--- a/tools/clang/test/HLSLFileCheck/hlsl/types/vector/vector-ast.hlsl
+++ b/tools/clang/test/HLSLFileCheck/hlsl/types/vector/vector-ast.hlsl
@@ -12,6 +12,7 @@
 // Verify the class, final attribute and ext_vector field decl.
 // CHECK-NEXT: CXXRecordDecl {{0x[0-9a-fA-F]+}} <<invalid sloc>> <invalid sloc> implicit class vector definition
 // CHECK-NEXT: FinalAttr {{0x[0-9a-fA-F]+}} <<invalid sloc>> Implicit final
+// CHECK-NEXT: HLSLVectorAttr {{0x[0-9a-fA-F]+}} <<invalid sloc>> Implicit
 // CHECK-NEXT: FieldDecl {{0x[0-9a-fA-F]+}} <<invalid sloc>> <invalid sloc> implicit h 'element __attribute__((ext_vector_type(element_count)))'
 
 // Verify operator overloads for const vector subscript operators.
diff --git a/tools/clang/test/SemaHLSL/const-default.hlsl b/tools/clang/test/SemaHLSL/const-default.hlsl
index 2ebb6fe52e..6b5e43e0e9 100644
--- a/tools/clang/test/SemaHLSL/const-default.hlsl
+++ b/tools/clang/test/SemaHLSL/const-default.hlsl
@@ -33,7 +33,11 @@ class MyClass {
 ConstantBuffer<MyClass> g_const_buffer2;
 TextureBuffer<MyClass> g_texture_buffer2;
 
+// expected-note@+2 {{forward declaration of 'FWDDeclStruct'}}
+// expected-note@+1 {{forward declaration of 'FWDDeclStruct'}}
 struct FWDDeclStruct;
+// expected-note@+2 {{forward declaration of 'FWDDeclClass'}}
+// expected-note@+1 {{forward declaration of 'FWDDeclClass'}}
 class FWDDeclClass;
 
 // Ensure forward declared struct/class fails as expected
diff --git a/tools/clang/test/SemaHLSL/hlsl/types/invalid-longvec-decls-hs.hlsl b/tools/clang/test/SemaHLSL/hlsl/types/invalid-longvec-decls-hs.hlsl
new file mode 100644
index 0000000000..1625454360
--- /dev/null
+++ b/tools/clang/test/SemaHLSL/hlsl/types/invalid-longvec-decls-hs.hlsl
@@ -0,0 +1,24 @@
+// RUN: %dxc -DTYPE=float -DNUM=7 -T hs_6_9 -verify %s
+
+struct HsConstantData {
+  float Edges[3] : SV_TessFactor;
+  vector <float, 7> vec;
+};
+
+struct LongVec {
+  float4 f;
+  vector<TYPE,NUM> vec;
+};
+
+HsConstantData PatchConstantFunction( // expected-error{{vectors of over 4 elements in patch constant function return type are not supported}}
+				      vector<TYPE,NUM> vec : V, // expected-error{{vectors of over 4 elements in patch constant function parameters are not supported}}
+				      LongVec lv : L) { // expected-error{{vectors of over 4 elements in patch constant function parameters are not supported}}
+  return (HsConstantData)0;
+}
+
+[domain("tri")]
+[outputtopology("triangle_cw")]
+[outputcontrolpoints(32)]
+[patchconstantfunc("PatchConstantFunction")]
+void main() {
+}
diff --git a/tools/clang/test/SemaHLSL/hlsl/types/invalid-longvec-decls.hlsl b/tools/clang/test/SemaHLSL/hlsl/types/invalid-longvec-decls.hlsl
new file mode 100644
index 0000000000..0604feeaec
--- /dev/null
+++ b/tools/clang/test/SemaHLSL/hlsl/types/invalid-longvec-decls.hlsl
@@ -0,0 +1,200 @@
+// RUN: %dxc -T ps_6_9 -DTYPE=LongVec    -DNUM=5   -verify %s
+// RUN: %dxc -T ps_6_9 -DTYPE=LongVecSub -DNUM=128 -verify %s
+// RUN: %dxc -T ps_6_9                   -DNUM=1024 -verify %s
+
+// Add tests for base types and instantiated template classes with longvecs
+// Size of the vector shouldn't matter, but using a few different ones just in case.
+
+#define PASTE_(x,y) x##y
+#define PASTE(x,y) PASTE_(x,y)
+
+#ifndef TYPE
+#define TYPE LongVecTpl<NUM>
+#endif
+
+struct LongVec {
+  float4 f;
+  vector<float,NUM> vec;
+};
+
+struct LongVecSub : LongVec {
+  int3 is;
+};
+
+template <int N>
+struct LongVecTpl {
+  float4 f;
+  vector<float,N> vec;
+};
+
+vector<float, NUM> global_vec; // expected-error{{vectors of over 4 elements in cbuffers or tbuffers are not supported}}
+vector<float, NUM> global_vec_arr[10]; // expected-error{{vectors of over 4 elements in cbuffers or tbuffers are not supported}}
+TYPE global_vec_rec; // expected-error{{vectors of over 4 elements in cbuffers or tbuffers are not supported}}
+TYPE global_vec_rec_arr[10]; // expected-error{{vectors of over 4 elements in cbuffers or tbuffers are not supported}}
+
+cbuffer BadBuffy {
+  vector<float, NUM> cb_vec; // expected-error{{vectors of over 4 elements in cbuffers or tbuffers are not supported}}
+  vector<float, NUM> cb_vec_arr[10]; // expected-error{{vectors of over 4 elements in cbuffers or tbuffers are not supported}}
+  TYPE cb_vec_rec; // expected-error{{vectors of over 4 elements in cbuffers or tbuffers are not supported}}
+  TYPE cb_vec_rec_arr[10]; // expected-error{{vectors of over 4 elements in cbuffers or tbuffers are not supported}}
+};
+
+tbuffer BadTuffy {
+  vector<float, NUM> tb_vec; // expected-error{{vectors of over 4 elements in cbuffers or tbuffers are not supported}}
+  vector<float, NUM> tb_vec_arr[10]; // expected-error{{vectors of over 4 elements in cbuffers or tbuffers are not supported}}
+  TYPE tb_vec_rec; // expected-error{{vectors of over 4 elements in cbuffers or tbuffers are not supported}}
+  TYPE tb_vec_rec_arr[10]; // expected-error{{vectors of over 4 elements in cbuffers or tbuffers are not supported}}
+};
+
+ConstantBuffer< TYPE > const_buf; // expected-error{{vectors of over 4 elements in ConstantBuffers or TextureBuffers are not supported}}
+TextureBuffer< TYPE > tex_buf; // expected-error{{vectors of over 4 elements in ConstantBuffers or TextureBuffers are not supported}}
+
+[shader("pixel")]
+vector<float, NUM> main( // expected-error{{vectors of over 4 elements in entry function return type are not supported}}
+                     vector<float, NUM> vec : V) : SV_Target { // expected-error{{vectors of over 4 elements in entry function parameters are not supported}}
+  return vec;
+}
+
+[shader("vertex")]
+TYPE vs_main( // expected-error{{vectors of over 4 elements in entry function return type are not supported}}
+                     TYPE parm : P) : SV_Target { // expected-error{{vectors of over 4 elements in entry function parameters are not supported}}
+  parm.f = 0;
+  return parm;
+}
+
+
+[shader("geometry")]
+[maxvertexcount(3)]
+void gs_point(line TYPE e, // expected-error{{vectors of over 4 elements in entry function parameters are not supported}}
+              inout PointStream<TYPE> OutputStream0) {} // expected-error{{vectors of over 4 elements in geometry streams are not supported}}
+
+[shader("geometry")]
+[maxvertexcount(12)]
+void gs_line(line TYPE a, // expected-error{{vectors of over 4 elements in entry function parameters are not supported}}
+             inout LineStream<TYPE> OutputStream0) {} // expected-error{{vectors of over 4 elements in geometry streams are not supported}}
+
+
+[shader("geometry")]
+[maxvertexcount(12)]
+void gs_line(line TYPE a, // expected-error{{vectors of over 4 elements in entry function parameters are not supported}}
+             inout TriangleStream<TYPE> OutputStream0) {} // expected-error{{vectors of over 4 elements in geometry streams are not supported}}
+
+[shader("domain")]
+[domain("tri")]
+void ds_main(OutputPatch<TYPE, 3> TrianglePatch) {} // expected-error{{vectors of over 4 elements in tessellation patches are not supported}}
+
+void patch_const(InputPatch<TYPE, 3> inpatch, // expected-error{{vectors of over 4 elements in tessellation patches are not supported}}
+			   OutputPatch<TYPE, 3> outpatch) {} // expected-error{{vectors of over 4 elements in tessellation patches are not supported}}
+
+[shader("hull")]
+[domain("tri")]
+[outputtopology("triangle_cw")]
+[outputcontrolpoints(32)]
+[patchconstantfunc("patch_const")]
+void hs_main(InputPatch<TYPE, 3> TrianglePatch) {} // expected-error{{vectors of over 4 elements in tessellation patches are not supported}}
+
+RaytracingAccelerationStructure RTAS;
+
+struct [raypayload] DXRLongVec {
+  float4 f : write(closesthit) : read(caller);
+  vector<float,NUM> vec : write(closesthit) : read(caller);
+};
+
+struct [raypayload] DXRLongVecSub : DXRLongVec {
+  int3 is : write(closesthit) : read(caller);
+};
+
+template<int N>
+struct [raypayload] DXRLongVecTpl {
+  float4 f : write(closesthit) : read(caller);
+  vector<float,N> vec : write(closesthit) : read(caller);
+};
+
+#define RTTYPE PASTE(DXR,TYPE)
+
+[shader("raygeneration")]
+void raygen() {
+  RTTYPE p = (RTTYPE)0;
+  RayDesc ray = (RayDesc)0;
+  TraceRay(RTAS, RAY_FLAG_NONE, 0, 0, 1, 0, ray, p); // expected-error{{vectors of over 4 elements in user-defined struct parameter are not supported}}
+  CallShader(0, p); // expected-error{{vectors of over 4 elements in user-defined struct parameter are not supported}}
+}
+
+
+[shader("closesthit")]
+void closesthit(inout RTTYPE payload, // expected-error{{vectors of over 4 elements in entry function parameters are not supported}}
+		in RTTYPE attribs ) { // expected-error{{vectors of over 4 elements in entry function parameters are not supported}}
+  RayDesc ray;
+  TraceRay( RTAS, RAY_FLAG_NONE, 0xff, 0, 1, 0, ray, payload ); // expected-error{{vectors of over 4 elements in user-defined struct parameter are not supported}}
+  CallShader(0, payload); // expected-error{{vectors of over 4 elements in user-defined struct parameter are not supported}}
+}
+
+[shader("anyhit")]
+void AnyHit( inout RTTYPE payload, // expected-error{{vectors of over 4 elements in entry function parameters are not supported}}
+	      in RTTYPE attribs  ) // expected-error{{vectors of over 4 elements in entry function parameters are not supported}}
+{
+}
+
+[shader("miss")]
+void Miss(inout RTTYPE payload){ // expected-error{{vectors of over 4 elements in entry function parameters are not supported}}
+  RayDesc ray;
+  TraceRay( RTAS, RAY_FLAG_NONE, 0xff, 0, 1, 0, ray, payload ); // expected-error{{vectors of over 4 elements in user-defined struct parameter are not supported}}
+  CallShader(0, payload); // expected-error{{vectors of over 4 elements in user-defined struct parameter are not supported}}
+}
+
+[shader("intersection")]
+void Intersection() {
+  float hitT = RayTCurrent();
+  RTTYPE attr = (RTTYPE)0;
+  bool bReported = ReportHit(hitT, 0, attr); // expected-error{{vectors of over 4 elements in user-defined struct parameter are not supported}}
+}
+
+[shader("callable")]
+void callable1(inout RTTYPE p) { // expected-error{{vectors of over 4 elements in entry function parameters are not supported}}
+  CallShader(0, p); // expected-error{{vectors of over 4 elements in user-defined struct parameter are not supported}}
+}
+
+groupshared LongVec as_pld;
+
+[shader("amplification")]
+[numthreads(1,1,1)]
+void Amp() {
+  DispatchMesh(1,1,1,as_pld); // expected-error{{vectors of over 4 elements in user-defined struct parameter are not supported}}
+}
+
+struct NodeLongVec {
+  uint3 grid : SV_DispatchGrid;
+  vector<float,NUM> vec;
+};
+
+struct NodeLongVecSub : NodeLongVec {
+  int3 is;
+};
+
+template<int N>
+struct NodeLongVecTpl {
+  uint3 grid : SV_DispatchGrid;
+  vector<float,N> vec;
+};
+
+#define NTYPE PASTE(Node,TYPE)
+
+[Shader("node")]
+[NodeLaunch("broadcasting")]
+[NumThreads(8,1,1)]
+[NodeMaxDispatchGrid(8,1,1)]
+void broadcast(DispatchNodeInputRecord<NTYPE> input,  // expected-error{{vectors of over 4 elements in node records are not supported}}
+                NodeOutput<TYPE> output) // expected-error{{vectors of over 4 elements in node records are not supported}}
+{
+  ThreadNodeOutputRecords<TYPE> touts; // expected-error{{vectors of over 4 elements in node records are not supported}}
+  GroupNodeOutputRecords<TYPE> gouts; // expected-error{{vectors of over 4 elements in node records are not supported}}
+}
+
+[Shader("node")]
+[NodeLaunch("coalescing")]
+[NumThreads(8,1,1)]
+void coalesce(GroupNodeInputRecords<TYPE> input) {} // expected-error{{vectors of over 4 elements in node records are not supported}}
+
+[Shader("node")]
+[NodeLaunch("thread")]
+void threader(ThreadNodeInputRecord<TYPE> input) {} // expected-error{{vectors of over 4 elements in node records are not supported}}
diff --git a/tools/clang/test/SemaHLSL/hlsl/types/invalid-longvec-swizzle.hlsl b/tools/clang/test/SemaHLSL/hlsl/types/invalid-longvec-swizzle.hlsl
new file mode 100644
index 0000000000..472a192677
--- /dev/null
+++ b/tools/clang/test/SemaHLSL/hlsl/types/invalid-longvec-swizzle.hlsl
@@ -0,0 +1,7 @@
+// RUN: %dxc -Tlib_6_9 -verify %s
+
+export
+vector<double, 3> doit(vector<double, 5> vec5) {
+  vec5.x = 1; // expected-error {{Invalid swizzle 'x' on vector of over 4 elements.}}
+  return vec5.xyw; // expected-error {{Invalid swizzle 'xyw' on vector of over 4 elements.}}
+}
diff --git a/tools/clang/test/SemaHLSL/hlsl/types/invalid-longvecs-sm68.hlsl b/tools/clang/test/SemaHLSL/hlsl/types/invalid-longvecs-sm68.hlsl
new file mode 100644
index 0000000000..42eb6b077c
--- /dev/null
+++ b/tools/clang/test/SemaHLSL/hlsl/types/invalid-longvecs-sm68.hlsl
@@ -0,0 +1,34 @@
+// RUN: %dxc -T ps_6_8 -verify %s
+
+#define TYPE float
+#define NUM 5
+
+struct LongVec {
+  float4 f;
+  vector<TYPE,NUM> vec; // expected-error{{invalid value, valid range is between 1 and 4 inclusive}}
+};
+groupshared vector<TYPE, NUM> gs_vec; // expected-error{{invalid value, valid range is between 1 and 4 inclusive}}
+groupshared vector<TYPE, NUM> gs_vec_arr[10]; // expected-error{{invalid value, valid range is between 1 and 4 inclusive}}
+
+static vector<TYPE, NUM> static_vec; // expected-error{{invalid value, valid range is between 1 and 4 inclusive}}
+static vector<TYPE, NUM> static_vec_arr[10]; // expected-error{{invalid value, valid range is between 1 and 4 inclusive}}
+
+export vector<TYPE, NUM> lv_param_passthru( // expected-error{{invalid value, valid range is between 1 and 4 inclusive}}
+                                           vector<TYPE, NUM> vec1) { // expected-error{{invalid value, valid range is between 1 and 4 inclusive}}
+  vector<TYPE, NUM> ret = vec1; // expected-error{{invalid value, valid range is between 1 and 4 inclusive}}
+  vector<TYPE, NUM> arr[10]; // expected-error{{invalid value, valid range is between 1 and 4 inclusive}}
+  arr[1]= vec1;
+  return ret;
+}
+
+export void lv_param_in_out(in vector<TYPE, NUM> vec1, // expected-error{{invalid value, valid range is between 1 and 4 inclusive}}
+                            out vector<TYPE, NUM> vec2) { // expected-error{{invalid value, valid range is between 1 and 4 inclusive}}
+  vec2 = vec1;
+}
+
+export void lv_param_inout(inout vector<TYPE, NUM> vec1, // expected-error{{invalid value, valid range is between 1 and 4 inclusive}}
+                           inout vector<TYPE, NUM> vec2) { // expected-error{{invalid value, valid range is between 1 and 4 inclusive}}
+  vector<TYPE, NUM> tmp = vec1; // expected-error{{invalid value, valid range is between 1 and 4 inclusive}}
+  vec1 = vec2;
+  vec2 = tmp;
+}
diff --git a/tools/clang/test/SemaHLSL/hlsl/types/toolong-vectors.hlsl b/tools/clang/test/SemaHLSL/hlsl/types/toolong-vectors.hlsl
new file mode 100644
index 0000000000..c1da348695
--- /dev/null
+++ b/tools/clang/test/SemaHLSL/hlsl/types/toolong-vectors.hlsl
@@ -0,0 +1,116 @@
+// RUN: %dxc -T lib_6_9 -DTYPE=float -DNUM=1025 -verify %s
+// RUN: %dxc -T ps_6_9  -DTYPE=float -DNUM=1025 -verify %s
+
+// A test to verify that declarations of longvecs are permitted in all the accepted places.
+// Only tests for acceptance, most codegen is ignored for now.
+
+struct LongVec {
+  float4 f;
+  vector<TYPE,NUM> vec; // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+};
+
+template <int N>
+struct LongVecTpl {
+  float4 f;
+  vector<TYPE,N> vec; // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+};
+
+template <int N>
+struct LongVecTpl2 {
+  float4 f;
+  vector<TYPE,N> vec; // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+};
+
+groupshared vector<TYPE, NUM> gs_vec; // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+groupshared vector<TYPE, NUM> gs_vec_arr[10]; // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+groupshared LongVecTpl<NUM> gs_vec_tpl; // expected-note{{in instantiation of template class 'LongVecTpl<1025>' requested here}}
+
+static vector<TYPE, NUM> static_vec; // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+static vector<TYPE, NUM> static_vec_arr[10]; // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+static LongVecTpl2<NUM> static_vec_tpl; // expected-note{{in instantiation of template class 'LongVecTpl2<1025>' requested here}}
+
+export vector<TYPE, NUM> // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+lv_param_passthru(vector<TYPE, NUM> vec1) { // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+  vector<TYPE, NUM> ret = vec1; // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+  return ret;
+}
+
+export void lv_param_in_out(in vector<TYPE, NUM> vec1, // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+                            out vector<TYPE, NUM> vec2) { // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+  vec2 = vec1;
+}
+
+export void lv_param_inout(inout vector<TYPE, NUM> vec1, // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+                           inout vector<TYPE, NUM> vec2) { // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+  vector<TYPE, NUM> tmp = vec1; // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+  vec1 = vec2;
+  vec2 = tmp;
+}
+
+export void lv_global_assign(vector<TYPE, NUM> vec) { // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+  static_vec = vec;
+}
+
+export vector<TYPE, NUM> lv_global_ret() { // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+  vector<TYPE, NUM> ret = static_vec; // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+  return ret;
+}
+
+export void lv_gs_assign(vector<TYPE, NUM> vec) { // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+  gs_vec = vec;
+}
+
+export vector<TYPE, NUM> lv_gs_ret() { // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+  vector<TYPE, NUM> ret = gs_vec; // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+  return ret;
+}
+
+#define DIMS 10
+
+export vector<TYPE, NUM> // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+lv_param_arr_passthru(vector<TYPE, NUM> vec)[10] { // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+  vector<TYPE, NUM> ret[10]; // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+  for (int i = 0; i < DIMS; i++)
+    ret[i] = vec;
+  return ret;
+}
+
+export void lv_global_arr_assign(vector<TYPE, NUM> vec[10]) { // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+  for (int i = 0; i < DIMS; i++)
+    static_vec_arr[i] = vec[i];
+}
+
+export vector<TYPE, NUM> lv_global_arr_ret()[10] { // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+  vector<TYPE, NUM> ret[10]; // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+  for (int i = 0; i < DIMS; i++)
+    ret[i] = static_vec_arr[i];
+  return ret;
+}
+
+export void lv_gs_arr_assign(vector<TYPE, NUM> vec[10]) { // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+  for (int i = 0; i < DIMS; i++)
+    gs_vec_arr[i] = vec[i];
+}
+
+export vector<TYPE, NUM> lv_gs_arr_ret()[10] { // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+  vector<TYPE, NUM> ret[10]; // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+  for (int i = 0; i < DIMS; i++)
+    ret[i] = gs_vec_arr[i];
+  return ret;
+}
+
+export LongVec lv_param_rec_passthru(LongVec vec) {
+  LongVec ret = vec;
+  return ret;
+}
+
+export vector<TYPE,NUM> lv_splat(TYPE scalar) { // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+  vector<TYPE,NUM> ret = scalar; // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+  return ret;
+}
+
+export vector<TYPE, NUM> lv_array_cast(TYPE arr[NUM]) { // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+  vector<TYPE, NUM> ret = (vector<TYPE,NUM>)arr; // expected-error{{invalid value, valid range is between 1 and 1024 inclusive}}
+  return ret;
+}
+
diff --git a/tools/clang/test/SemaHLSL/hlsl/workgraph/dependent_type_for_node_object_template_arg.hlsl b/tools/clang/test/SemaHLSL/hlsl/workgraph/dependent_type_for_node_object_template_arg.hlsl
index 40e0452719..05ec268a0c 100644
--- a/tools/clang/test/SemaHLSL/hlsl/workgraph/dependent_type_for_node_object_template_arg.hlsl
+++ b/tools/clang/test/SemaHLSL/hlsl/workgraph/dependent_type_for_node_object_template_arg.hlsl
@@ -60,12 +60,9 @@ void woo() {
 }
 
 template<typename T>
-// expected-note@+1{{zero sized record defined here}}
 struct ForwardDecl; // expected-note{{template is declared here}}
 
 void woot() {
-  // Forward decl fails because forcing completion to check empty size for node object.
-  // expected-error@+1{{record used in GroupNodeInputRecords may not have zero size}}
   GroupNodeInputRecords<ForwardDecl<int> > data; // expected-error{{implicit instantiation of undefined template 'ForwardDecl<int>'}}
   foo(data);
 }
diff --git a/tools/clang/test/SemaHLSL/incomplete-type.hlsl b/tools/clang/test/SemaHLSL/incomplete-type.hlsl
index 8869b80400..b0d4f1da7f 100644
--- a/tools/clang/test/SemaHLSL/incomplete-type.hlsl
+++ b/tools/clang/test/SemaHLSL/incomplete-type.hlsl
@@ -1,17 +1,83 @@
-// RUN: %dxc -Tlib_6_3 -Wno-unused-value -verify %s
+// RUN: %dxc -Tlib_6_8 -Wno-unused-value -verify %s
 
 // Tests that the compiler is well-behaved with regard to uses of incomplete types.
 // Regression test for GitHub #2058, which crashed in this case.
 
-// expected-note@+4 {{forward declaration of 'S'}}
-// expected-note@+3 {{forward declaration of 'S'}}
-// expected-note@+2 {{forward declaration of 'S'}}
-// expected-note@+1 {{forward declaration of 'S'}}
-struct S;
+
+struct S; // expected-note 24 {{forward declaration of 'S'}}
+template <int N> struct T; // expected-note 4 {{template is declared here}}
+
 ConstantBuffer<S> CB; // expected-error {{variable has incomplete type 'S'}}
+ConstantBuffer<T<1> > TB; // expected-error {{implicit instantiation of undefined template 'T<1>'}}
+
+S s; // expected-error {{variable has incomplete type 'S'}}
+T<1> t; // expected-error {{implicit instantiation of undefined template 'T<1>'}}
+
+cbuffer BadBuffy {
+  S cb_s; // expected-error {{variable has incomplete type 'S'}}
+  T<1> cb_t; // expected-error {{implicit instantiation of undefined template 'T<1>'}}
+};
+
+tbuffer BadTuffy {
+  S tb_s; // expected-error {{variable has incomplete type 'S'}}
+  T<1> tb_t; // expected-error {{implicit instantiation of undefined template 'T<1>'}}
+};
+
 S func( // expected-error {{incomplete result type 'S' in function definition}}
   S param) // expected-error {{variable has incomplete type 'S'}}
 {
   S local; // expected-error {{variable has incomplete type 'S'}}
   return (S)0; // expected-error {{'S' is an incomplete type}}
 }
+
+[shader("geometry")]
+[maxvertexcount(3)]
+void gs_point(line S e, // expected-error {{variable has incomplete type 'S'}}
+              inout PointStream<S> OutputStream0) {} // expected-error {{variable has incomplete type 'S'}}
+
+[shader("geometry")]
+[maxvertexcount(12)]
+void gs_line(line S a, // expected-error {{variable has incomplete type 'S'}}
+             inout LineStream<S> OutputStream0) {} // expected-error {{variable has incomplete type 'S'}}
+
+
+[shader("geometry")]
+[maxvertexcount(12)]
+void gs_line(line S a, // expected-error {{variable has incomplete type 'S'}}
+             inout TriangleStream<S> OutputStream0) {} // expected-error {{variable has incomplete type 'S'}}
+
+
+[shader("domain")]
+[domain("tri")]
+void ds_main(OutputPatch<S, 3> TrianglePatch) {} // expected-error{{variable has incomplete type 'S'}}
+
+void patch_const(InputPatch<S, 3> inpatch, // expected-error{{variable has incomplete type 'S'}}
+                 OutputPatch<S, 3> outpatch) {} // expected-error{{variable has incomplete type 'S'}}
+
+[shader("hull")]
+[domain("tri")]
+[outputtopology("triangle_cw")]
+[outputcontrolpoints(32)]
+[patchconstantfunc("patch_const")]
+void hs_main(InputPatch<S, 3> TrianglePatch) {} // expected-error{{variable has incomplete type 'S'}}
+
+[Shader("node")]
+[NodeLaunch("broadcasting")]
+[NumThreads(8,1,1)]
+[NodeMaxDispatchGrid(8,1,1)]
+// expected-error@+1{{Broadcasting node shader 'broadcast' with NodeMaxDispatchGrid attribute must declare an input record containing a field with SV_DispatchGrid semantic}}
+void broadcast(DispatchNodeInputRecord<S> input,  // expected-error{{variable has incomplete type 'S'}}
+                NodeOutput<S> output) // expected-error{{variable has incomplete type 'S'}}
+{
+  ThreadNodeOutputRecords<S> touts; // expected-error{{variable has incomplete type 'S'}}
+  GroupNodeOutputRecords<S> gouts; // expected-error{{variable has incomplete type 'S'}}
+}
+
+[Shader("node")]
+[NodeLaunch("coalescing")]
+[NumThreads(8,1,1)]
+void coalesce(GroupNodeInputRecords<S> input) {} // expected-error{{variable has incomplete type 'S'}}
+
+[Shader("node")]
+[NodeLaunch("thread")]
+void threader(ThreadNodeInputRecord<S> input) {} // expected-error{{variable has incomplete type 'S'}}
diff --git a/tools/clang/tools/dxcompiler/dxcompilerobj.cpp b/tools/clang/tools/dxcompiler/dxcompilerobj.cpp
index c1c844d4be..11effb645b 100644
--- a/tools/clang/tools/dxcompiler/dxcompilerobj.cpp
+++ b/tools/clang/tools/dxcompiler/dxcompilerobj.cpp
@@ -1440,6 +1440,13 @@ class DxcCompiler : public IDxcCompiler3,
         Opts.EnablePayloadQualifiers;
     compiler.getLangOpts().HLSLProfile = compiler.getCodeGenOpts().HLSLProfile =
         Opts.TargetProfile;
+    const ShaderModel *SM = hlsl::ShaderModel::GetByName(
+        compiler.getLangOpts().HLSLProfile.c_str());
+    if (SM->IsSM69Plus())
+      compiler.getLangOpts().MaxHLSLVectorLength = DXIL::kSM69MaxVectorLength;
+    else
+      compiler.getLangOpts().MaxHLSLVectorLength =
+          DXIL::kDefaultMaxVectorLength;
 
     // Enable dumping implicit top level decls either if it was specifically
     // requested or if we are not dumping the ast from the command line. That
diff --git a/tools/clang/unittests/HLSL/LinkerTest.cpp b/tools/clang/unittests/HLSL/LinkerTest.cpp
index 7cafa0db06..df8bb644e1 100644
--- a/tools/clang/unittests/HLSL/LinkerTest.cpp
+++ b/tools/clang/unittests/HLSL/LinkerTest.cpp
@@ -526,6 +526,11 @@ TEST_F(LinkerTest, RunLinkMatArrayParam) {
   Link(L"main", L"ps_6_0", pLinker, {libName, libName2},
        {"alloca [24 x float]", "getelementptr [12 x float], [12 x float]*"},
        {});
+
+  Link(L"main", L"ps_6_9", pLinker, {libName, libName2},
+       {"alloca [2 x <12 x float>]",
+        "getelementptr [12 x float], [12 x float]*"},
+       {});
 }
 
 TEST_F(LinkerTest, RunLinkMatParam) {
diff --git a/utils/hct/CMakeLists.txt b/utils/hct/CMakeLists.txt
new file mode 100644
index 0000000000..41e6b494e6
--- /dev/null
+++ b/utils/hct/CMakeLists.txt
@@ -0,0 +1,3 @@
+# generate hlsl_intrinsic_opcodes.json to preserve high level intrinsic opcodes
+# This uses CODE_TAG because the file exists in the source tree.
+add_hlsl_hctgen(HlslIntrinsicOpcodes OUTPUT hlsl_intrinsic_opcodes.json CODE_TAG)
diff --git a/utils/hct/hctdb.py b/utils/hct/hctdb.py
index 66376c3b9b..66b897945e 100644
--- a/utils/hct/hctdb.py
+++ b/utils/hct/hctdb.py
@@ -37,6 +37,32 @@
     "array_local_ldst",
 ]
 
+# These are the valid overload type characters for DXIL instructions.
+# - "v" is for void, and is mutually exclusive with the other types.
+# - "u" is for user defined type (UDT), and is mutually exclusive with the other
+#   types.
+# - "o" is for an HLSL object type (e.g. Texture, Sampler, etc.), and is
+#   mutually exclusive with the other types.
+# - "<" is for vector overloads, and may be followed by a set of supported
+#   component types.
+#   - If "<" is not followed by any component types, any preceding scalar types
+#     are used.
+#   - Vector component types are captured into a separate list during
+#     processing.
+# - "x" is for extended overloads, in order to support multiple overload
+#   dimensions, and is mutually exclusive with the other types.
+# - "x" is not supplied manually, but automatically used when processing
+#   overloads that use "," to separate multiple overload dimensions, which are
+#   captured into a separate list.
+# - "," is used to separate multiple overload dimensions, will be converted
+#   to a use of a single "x" overload string during processing.
+dxil_all_user_oload_chars = "vhfd18wiluo<,"
+dxil_scalar_oload_chars = "hfd18wil"
+
+# Maximum number of overload dimensions supported through the extended overload
+# in DXIL instructions.
+dxil_max_overload_dims = 2
+
 
 class db_dxil_enum_value(object):
     "A representation for a value in an enumeration type"
@@ -81,6 +107,9 @@ def __init__(self, name, **kwargs):
         self.ops = []  # the operands that this instruction takes
         self.is_allowed = True  # whether this instruction is allowed in a DXIL program
         self.oload_types = ""  # overload types if applicable
+        # Always call process_oload_types() after setting oload_types.
+        self.extended_oload_types = None  # extended overload types if applicable
+        self.vector_oload_types = None  # vector overload types if applicable
         self.fn_attr = ""  # attribute shorthands: rn=does not access memory,ro=only reads from memory,
         self.is_deriv = False  # whether this is some kind of derivative
         self.is_gradient = False  # whether this requires a gradient calculation
@@ -98,6 +127,8 @@ def __init__(self, name, **kwargs):
         self.is_reserved = self.dxil_class == "Reserved"
         self.shader_model_translated = ()  # minimum shader model required with translation by linker
         self.props = {}  # extra properties
+        if self.is_dxil_op:
+            self.process_oload_types()
 
     def __str__(self):
         return self.name
@@ -105,6 +136,125 @@ def __str__(self):
     def fully_qualified_name(self):
         return "{}::{}".format(self.fully_qualified_name_prefix, self.name)
 
+    def process_oload_types(self):
+        if type(self.oload_types) is not str:
+            raise ValueError(
+                f"overload for '{self.name}' should be a string - use empty if n/a"
+            )
+        # Early out for LLVM instructions
+        if not self.is_dxil_op:
+            return
+
+        self.extended_oload_types = [""] * dxil_max_overload_dims
+        self.vector_oload_types = [""] * dxil_max_overload_dims
+
+        # Early out for void overloads.
+        if self.oload_types == "v":
+            return
+
+        if self.oload_types == "":
+            raise ValueError(
+                f"overload for '{self.name}' should not be empty - use void if n/a"
+            )
+        if "v" in self.oload_types:
+            raise ValueError(
+                f"void overload should be exclusive to other types for '({self.name})'"
+            )
+
+        # Process oload_types for extended and vector overloads.
+        # Contrived example: "hf<,<fd,i<1"
+        #   - "," splits multiple overload dimensions, setting the main overload
+        #     types to "x" for extended.
+        #   - In the first overload dimension "hf<":
+        #     - "hf" means overloads for scalar half and float
+        #     - ending with "<" means vector overload supporting the same
+        #       components as defined for the scalar overload types.
+        #   - In the second overload dimension "<f":
+        #     - starting with "<" means only vector overloads are supported.
+        #     - "fd" means the vector supports float or double components.
+        #   - In the third overload dimension "i<1":
+        #     - "i" means it supports a scalar i32 overload
+        #     - "<1" means it also supports a vector overload with an i1
+        #       component type.
+        used = set()
+        for c in self.oload_types:
+            if c in used:
+                raise ValueError(
+                    f"Duplicate overload type character '{c}' used for DXIL op "
+                    f"{self.name}: '{self.oload_types}'"
+                )
+            if c not in dxil_all_user_oload_chars:
+                raise ValueError(
+                    "Invalid overload type character used for DXIL op "
+                    f"{self.name}: {c} in '{self.oload_types}'"
+                )
+        oload_types = self.oload_types.split(",")
+        if len(oload_types) > dxil_max_overload_dims:
+            raise ValueError(
+                "Too many overload dimensions for DXIL op "
+                f"{self.name}: '{self.oload_types}'"
+            )
+        for n, oloads in enumerate(oload_types):
+            if len(oloads) == 0:
+                raise ValueError(
+                    f"Invalid extended overload type syntax for DXIL op "
+                    f"{self.name}: '{self.oload_types}'"
+                )
+            # split at vector for component overloads, if vector specified
+            # without following components, use the scalar overloads that
+            # precede the vector character.
+            split = oloads.split("<")
+            if len(split) == 1:
+                # No vector overload.
+                continue
+            elif len(split) != 2:
+                raise ValueError(
+                    f"Invalid overload types for DXIL op {self.name}: "
+                    f"{self.oload_types}"
+                )
+
+            # Split into scalar and vector component overloads.
+            scalars, vector_oloads = split
+            if not vector_oloads:
+                vector_oloads = scalars
+            if not vector_oloads:
+                raise ValueError(
+                    "No scalar overload types provided with vector overload "
+                    f"for DXIL op {self.name}: '{self.oload_types}'"
+                )
+            for c in scalars:
+                if c not in dxil_scalar_oload_chars:
+                    raise ValueError(
+                        "Invalid overload type character used with vector for "
+                        f"DXIL op {self.name}: {c} in '{self.oload_types}'"
+                    )
+            oload_types[n] = scalars + "<"
+            self.vector_oload_types[n] = vector_oloads
+        if len(oload_types) > 1:
+            self.oload_types = "x"
+            self.extended_oload_types[: len(oload_types)] = oload_types
+            self.check_extended_oload_ops()
+        else:
+            self.oload_types = oload_types[0]
+
+    def check_extended_oload_ops(self):
+        "Ensure ops has sequential extended overload references with $x0, $x1, etc."
+        next_oload_idx = 0
+        for i in self.ops:
+            if i.llvm_type.startswith("$x"):
+                if i.llvm_type != "$x" + str(next_oload_idx):
+                    raise ValueError(
+                        "Extended overloads are not sequentially referenced in "
+                        f"DXIL op {self.name}: {i.llvm_type} != $x{next_oload_idx}"
+                    )
+                next_oload_idx += 1
+        if next_oload_idx != len(self.extended_oload_types):
+            raise ValueError(
+                "Extended overloads are not referenced for all overload "
+                f"dimensions in DXIL op {self.name}: {next_oload_idx} != "
+                f"{len(self.extended_oload_types)}"
+            )
+
 
 class db_dxil_metadata(object):
     "A representation for a metadata record"
@@ -477,9 +627,7 @@ def populate_categories_and_models(self):
                 "closesthit",
             )
         for i in "GeometryIndex".split(","):
-            self.name_idx[
-                i
-            ].category = (
+            self.name_idx[i].category = (
                 "Raytracing object space uint System Values, raytracing tier 1.1"
             )
             self.name_idx[i].shader_model = 6, 5
@@ -574,9 +722,7 @@ def populate_categories_and_models(self):
             self.name_idx[i].shader_model = 6, 3
             self.name_idx[i].shader_stages = ("library", "intersection")
         for i in "CreateHandleForLib".split(","):
-            self.name_idx[
-                i
-            ].category = (
+            self.name_idx[i].category = (
                 "Library create handle from resource struct (like HL intrinsic)"
             )
             self.name_idx[i].shader_model = 6, 3
@@ -1175,6 +1321,37 @@ def populate_llvm_instructions(self):
         self.add_llvm_instr(
             "OTHER", 53, "VAArg", "VAArgInst", "vaarg instruction", "", []
         )
+
+        self.add_llvm_instr(
+            "OTHER",
+            54,
+            "ExtractElement",
+            "ExtractElementInst",
+            "extracts from vector",
+            "",
+            [],
+        )
+
+        self.add_llvm_instr(
+            "OTHER",
+            55,
+            "InsertElement",
+            "InsertElementInst",
+            "inserts into vector",
+            "",
+            [],
+        )
+
+        self.add_llvm_instr(
+            "OTHER",
+            56,
+            "ShuffleVector",
+            "ShuffleVectorInst",
+            "Shuffle two vectors",
+            "",
+            [],
+        )
+
         self.add_llvm_instr(
             "OTHER",
             57,
@@ -1348,7 +1525,7 @@ def UFI(name, **mappings):
                 next_op_idx,
                 "Unary",
                 "returns the " + i,
-                "hf",
+                "hf<",
                 "rn",
                 [
                     db_dxil_param(0, "$o", "", "operation result"),
@@ -1412,7 +1589,7 @@ def UFI(name, **mappings):
                 next_op_idx,
                 "Binary",
                 "returns the " + i + " of the input values",
-                "hfd",
+                "hfd<",
                 "rn",
                 [
                     db_dxil_param(0, "$o", "", "operation result"),
@@ -1430,7 +1607,7 @@ def UFI(name, **mappings):
                 next_op_idx,
                 "Binary",
                 "returns the " + i + " of the input values",
-                "wil",
+                "wil<",
                 "rn",
                 [
                     db_dxil_param(0, "$o", "", "operation result"),
@@ -1502,7 +1679,7 @@ def UFI(name, **mappings):
             next_op_idx,
             "Tertiary",
             "performs a fused multiply add (FMA) of the form a * b + c",
-            "d",
+            "d<",
             "rn",
             [
                 db_dxil_param(
@@ -5555,6 +5732,83 @@ def UFI(name, **mappings):
         # Reserved block C
         next_op_idx = self.reserve_dxil_op_range("ReservedC", next_op_idx, 10)
 
+        # Long Vectors
+        self.add_dxil_op(
+            "RawBufferVectorLoad",
+            next_op_idx,
+            "RawBufferVectorLoad",
+            "reads from a raw buffer and structured buffer",
+            "hfwidl<",
+            "ro",
+            [
+                db_dxil_param(0, "$r", "", "the loaded value"),
+                db_dxil_param(2, "res", "srv", "handle of TypedBuffer SRV to sample"),
+                db_dxil_param(
+                    3,
+                    "i32",
+                    "index",
+                    "element index for StructuredBuffer, or byte offset for ByteAddressBuffer",
+                ),
+                db_dxil_param(
+                    4,
+                    "i32",
+                    "elementOffset",
+                    "offset into element for StructuredBuffer, or undef for ByteAddressBuffer",
+                ),
+                db_dxil_param(
+                    5,
+                    "i32",
+                    "alignment",
+                    "relative load access alignment",
+                    is_const=True,
+                ),
+            ],
+            counters=("tex_load",),
+        )
+        next_op_idx += 1
+
+        self.add_dxil_op(
+            "RawBufferVectorStore",
+            next_op_idx,
+            "RawBufferVectorStore",
+            "writes to a RWByteAddressBuffer or RWStructuredBuffer",
+            "hfwidl<",
+            "",
+            [
+                db_dxil_param(0, "v", "", ""),
+                db_dxil_param(2, "res", "uav", "handle of UAV to store to"),
+                db_dxil_param(
+                    3,
+                    "i32",
+                    "index",
+                    "element index for StructuredBuffer, or byte offset for ByteAddressBuffer",
+                ),
+                db_dxil_param(
+                    4,
+                    "i32",
+                    "elementOffset",
+                    "offset into element for StructuredBuffer, or undef for ByteAddressBuffer",
+                ),
+                db_dxil_param(5, "$o", "value0", "value"),
+                db_dxil_param(
+                    6,
+                    "i32",
+                    "alignment",
+                    "relative store access alignment",
+                    is_const=True,
+                ),
+            ],
+            counters=("tex_store",),
+        )
+        next_op_idx += 1
+
+        # End of DXIL 1.9 opcodes.
+        self.set_op_count_for_version(1, 9, next_op_idx)
+        assert next_op_idx == 305, (
+            "260 is expected next operation index but encountered %d and thus opcodes are broken"
+            % next_op_idx
+        )
+
         # Set interesting properties.
         self.build_indices()
         for (
@@ -5576,18 +5830,6 @@ def UFI(name, **mappings):
         )
         for i in self.instr:
             self.verify_dense(i.ops, lambda x: x.pos, lambda x: i.name)
-        for i in self.instr:
-            if i.is_dxil_op:
-                assert i.oload_types != "", (
-                    "overload for DXIL operation %s should not be empty - use void if n/a"
-                    % (i.name)
-                )
-                assert i.oload_types == "v" or i.oload_types.find("v") < 0, (
-                    "void overload should be exclusive to other types (%s)" % i.name
-                )
-            assert (
-                type(i.oload_types) is str
-            ), "overload for %s should be a string - use empty if n/a" % (i.name)
 
         # Verify that all operations in each class have the same signature.
         import itertools
@@ -8247,6 +8489,8 @@ def __init__(
         self.vulkanSpecific = ns.startswith(
             "Vk"
         )  # Vulkan specific intrinsic - SPIRV change
+        self.opcode = None  # high-level opcode assigned later
+        self.unsigned_opcode = None  # unsigned high-level opcode if appicable
 
 
 class db_hlsl_namespace(object):
@@ -8292,7 +8536,7 @@ def __init__(
 class db_hlsl(object):
     "A database of HLSL language data"
 
-    def __init__(self, intrinsic_defs):
+    def __init__(self, intrinsic_defs, opcode_data):
         self.base_types = {
             "bool": "LICOMPTYPE_BOOL",
             "int": "LICOMPTYPE_INT",
@@ -8365,6 +8609,13 @@ def __init__(self, intrinsic_defs):
         self.populate_attributes()
         self.opcode_namespace = "hlsl::IntrinsicOp"
 
+        # Populate opcode data for HLSL intrinsics.
+        self.opcode_data = opcode_data
+        # If opcode data is empty, create the default structure.
+        if not self.opcode_data:
+            self.opcode_data["IntrinsicOpCodes"] = {"Num_Intrinsics": 0}
+        self.assign_opcodes()
+
     def create_namespaces(self):
         last_ns = None
         self.namespaces = {}
@@ -8869,6 +9120,29 @@ def add_attr_arg(title_name, scope, args, doc):
         )
         self.attributes = attributes
 
+    # Iterate through all intrinsics, assigning opcodes to each one.
+    # This uses the opcode_data to preserve already-assigned opcodes.
+    def assign_opcodes(self):
+        "Assign opcodes to the intrinsics."
+        IntrinsicOpDict = self.opcode_data["IntrinsicOpCodes"]
+        Num_Intrinsics = self.opcode_data["IntrinsicOpCodes"]["Num_Intrinsics"]
+
+        def add_intrinsic(name):
+            nonlocal Num_Intrinsics
+            opcode = IntrinsicOpDict.setdefault(name, Num_Intrinsics)
+            if opcode == Num_Intrinsics:
+                Num_Intrinsics += 1
+            return opcode
+
+        sorted_intrinsics = sorted(self.intrinsics, key=lambda x: x.key)
+        for i in sorted_intrinsics:
+            i.opcode = add_intrinsic(i.enum_name)
+        for i in sorted_intrinsics:
+            if i.unsigned_op == "":
+                continue
+            i.unsigned_opcode = add_intrinsic(i.unsigned_op)
+        self.opcode_data["IntrinsicOpCodes"]["Num_Intrinsics"] = Num_Intrinsics
+
 
 if __name__ == "__main__":
     db = db_dxil()
diff --git a/utils/hct/hctdb_instrhelp.py b/utils/hct/hctdb_instrhelp.py
index 17eefd4918..cf81dd04fc 100644
--- a/utils/hct/hctdb_instrhelp.py
+++ b/utils/hct/hctdb_instrhelp.py
@@ -18,6 +18,29 @@ def get_db_dxil():
     return g_db_dxil
 
 
+# opcode data contains fixed opcode assignments for HLSL intrinsics.
+g_hlsl_opcode_data = None
+
+
+def get_hlsl_opcode_data():
+    global g_hlsl_opcode_data
+    if g_hlsl_opcode_data is None:
+        # Load the intrinsic opcodes from the JSON file.
+        json_filepath = os.path.join(
+            os.path.dirname(__file__), "hlsl_intrinsic_opcodes.json"
+        )
+        try:
+            with open(json_filepath, "r") as file:
+                g_hlsl_opcode_data = json.load(file)
+        except FileNotFoundError:
+            print(f"File not found: {json_filepath}")
+        except json.JSONDecodeError as e:
+            print(f"Error decoding JSON from {json_filepath}: {e}")
+        if not g_hlsl_opcode_data:
+            g_hlsl_opcode_data = {}
+    return g_hlsl_opcode_data
+
+
 g_db_hlsl = None
 
 
@@ -26,10 +49,15 @@ def get_db_hlsl():
     if g_db_hlsl is None:
         thisdir = os.path.dirname(os.path.realpath(__file__))
         with open(os.path.join(thisdir, "gen_intrin_main.txt"), "r") as f:
-            g_db_hlsl = db_hlsl(f)
+            g_db_hlsl = db_hlsl(f, get_hlsl_opcode_data())
     return g_db_hlsl
 
 
+def get_max_oload_dims():
+    db = get_db_dxil()
+    return f"const unsigned kDxilMaxOloadDims = {dxil_max_overload_dims};"
+
+
 def format_comment(prefix, val):
     "Formats a value with a line-comment prefix."
     result = ""
@@ -487,25 +515,23 @@ def print_opfunc_props(self):
             )
         )
         print(
-            "//   OpCode                       OpCode name,                OpCodeClass                    OpCodeClass name,              void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,   udt,   obj,  function attribute"
+            "//   OpCode                       OpCode name,                OpCodeClass                    OpCodeClass name,              void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,   udt,   obj,   vec,   ext,  function attribute, ext oload, vec oload"
         )
         # Example formatted string:
-        #   {  OC::TempRegLoad,             "TempRegLoad",              OCC::TempRegLoad,              "tempRegLoad",                false,  true,  true, false,  true, false,  true,  true, false, Attribute::ReadOnly, },
-        # 012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789
-        # 0         1         2         3         4         5         6         7         8         9         0         1         2         3         4         5         6         7         8         9         0
+        #   {  OC::TempRegLoad,             "TempRegLoad",              OCC::TempRegLoad,              "tempRegLoad",                false,  true,  true, false,  true, false,  true,  true, false, false, false, false, false, Attribute::ReadOnly, {}, {0x00} },
+        # 01234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890
+        # 0         1         2         3         4         5         6         7         8         9         0         1         2         3         4         5         6         7         8         9         0         1         2         3         4         5
 
         last_category = None
-        # overload types are a string of (v)oid, (h)alf, (f)loat, (d)ouble, (1)-bit, (8)-bit, (w)ord, (i)nt, (l)ong, u(dt)
+        # overload types are a string of (v)oid, (h)alf, (f)loat, (d)ouble, (1)-bit, (8)-bit, (w)ord, (i)nt, (l)ong, u(dt), o(bj), vec(t)or, e(x)tended
         f = lambda i, c: "true" if i.oload_types.find(c) >= 0 else "false"
         lower_exceptions = {
             "CBufferLoad": "cbufferLoad",
             "CBufferLoadLegacy": "cbufferLoadLegacy",
             "GSInstanceID": "gsInstanceID",
         }
-        lower_fn = (
-            lambda t: lower_exceptions[t]
-            if t in lower_exceptions
-            else t[:1].lower() + t[1:]
+        lower_fn = lambda t: (
+            lower_exceptions[t] if t in lower_exceptions else t[:1].lower() + t[1:]
         )
         attr_dict = {
             "": "None",
@@ -516,19 +542,37 @@ def print_opfunc_props(self):
             "nr": "NoReturn",
             "wv": "None",
         }
-        attr_fn = lambda i: "Attribute::" + attr_dict[i.fn_attr] + ","
+        attr_fn = lambda i: "Attribute::" + attr_dict[i.fn_attr]
+        oload_to_mask = lambda oload: sum(
+            [1 << dxil_all_user_oload_chars.find(c) for c in oload]
+        )
+        ext_oload_fn = (
+            lambda i: "{"
+            + ",".join(["{0x%x}" % oload_to_mask(o) for o in i.extended_oload_types])
+            + "}"
+        )
+        vec_oload_fn = (
+            lambda i: "{"
+            + ",".join(["{0x%x}" % oload_to_mask(o) for o in i.vector_oload_types])
+            + "}"
+        )
         for i in self.instrs:
             if last_category != i.category:
                 if last_category != None:
                     print("")
                 print(
-                    "  // {category:118}  void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,   udt,   obj ,  function attribute".format(
+                    "  // {category:118}  void,     h,     f,     d,    i1,    i8,   i16,   i32,   i64,   udt,   obj,   vec,  function attribute, ext oload, vec oload".format(
                         category=i.category
                     )
                 )
                 last_category = i.category
             print(
-                "  {{  {OC}::{name:24} {quotName:27} {OCC}::{className:25} {classNameQuot:28} {{{v:>6},{h:>6},{f:>6},{d:>6},{b:>6},{e:>6},{w:>6},{i:>6},{l:>6},{u:>6},{o:>6}}}, {attr:20} }},".format(
+                (
+                    "  {{  {OC}::{name:24} {quotName:27} {OCC}::{className:25} "
+                    + "{classNameQuot:28} {{{v:>6},{h:>6},{f:>6},{d:>6},{b:>6},"
+                    + "{e:>6},{w:>6},{i:>6},{l:>6},{u:>6},{o:>6},{t:>6},"
+                    + "{x:>6}}}, {attr:20}, {ext_oload:2}, {vec_oload:6} }},"
+                ).format(
                     name=i.name + ",",
                     quotName='"' + i.name + '",',
                     className=i.dxil_class + ",",
@@ -544,9 +588,13 @@ def print_opfunc_props(self):
                     l=f(i, "l"),
                     u=f(i, "u"),
                     o=f(i, "o"),
+                    t=f(i, "<"),
+                    x=f(i, "x"),
                     attr=attr_fn(i),
                     OC=self.OC,
                     OCC=self.OCC,
+                    ext_oload=ext_oload_fn(i),
+                    vec_oload=vec_oload_fn(i),
                 )
             )
         print("};")
@@ -599,6 +647,9 @@ def print_opfunc_table(self):
             "noderecordhandle": "A(pNodeRecordHandle);",
             "nodeproperty": "A(nodeProperty);",
             "noderecordproperty": "A(nodeRecordProperty);",
+            # Extended overload slots, extend as needed:
+            "$x0": "EXT(0);",
+            "$x1": "EXT(1);",
         }
         last_category = None
         for i in self.instrs:
@@ -629,12 +680,15 @@ def print_opfunc_oload_type(self):
         obj_ty = "obj"
         vec_ty = "$vec"
         gsptr_ty = "$gsptr"
+        extended_ty = "$x"
         last_category = None
 
         index_dict = collections.OrderedDict()
         ptr_index_dict = collections.OrderedDict()
         single_dict = collections.OrderedDict()
+        extended_dict = collections.OrderedDict()
         struct_list = []
+        extended_list = []
 
         for instr in self.instrs:
             ret_ty = instr.ops[0].llvm_type
@@ -654,6 +708,10 @@ def print_opfunc_oload_type(self):
                 struct_list.append(instr.name)
                 continue
 
+            if instr.oload_types == "x":
+                extended_list.append(instr)
+                continue
+
             in_param_ty = False
             # Try to find elt_ty in parameter types.
             for index, op in enumerate(instr.ops):
@@ -708,8 +766,7 @@ def print_opfunc_oload_type(self):
                 "i": "IntegerType::get(Ctx, 32)",
                 "l": "IntegerType::get(Ctx, 64)",
                 "v": "Type::getVoidTy(Ctx)",
-                "u": "Type::getInt32PtrTy(Ctx)",
-                "o": "Type::getInt32PtrTy(Ctx)",
+                # No other types should be referenced here.
             }
             assert ty in type_code_texts, "llvm type %s is unknown" % (ty)
             ty_code = type_code_texts[ty]
@@ -769,6 +826,61 @@ def print_opfunc_oload_type(self):
         line = line + "}"
         print(line)
 
+        for instr in extended_list:
+            # Collect indices for overloaded return and types, make a tuple of
+            # indices the key, and add the opcode to a list of opcodes for that
+            # key.  Indices start with 0 for return type, and 1 for the first
+            # function parameter, which is the DXIL OpCode.
+            indices = []
+            for index, op in enumerate(instr.ops):
+                # Skip dxil opcode.
+                if op.pos == 1:
+                    continue
+
+                op_type = op.llvm_type
+                if op_type.startswith(extended_ty):
+                    try:
+                        extended_index = int(op_type[2:])
+                    except:
+                        raise ValueError(
+                            "Error parsing extended operand type "
+                            + f"'{op_type}' for DXIL op '{instr.name}'"
+                        )
+                    if extended_index != len(indices):
+                        raise ValueError(
+                            f"'$x{extended_index}' is not in sequential "
+                            + f"order for DXIL op '{instr.name}'"
+                        )
+                    indices.append(op.pos)
+
+            if len(indices) != len(instr.extended_oload_types):
+                raise ValueError(
+                    f"DXIL op {instr.name}: extended overload count "
+                    + "mismatches the number of overload types"
+                )
+            extended_dict.setdefault(tuple(indices), []).append(instr.name)
+
+        def GetTypeAtIndex(index):
+            if index == 0:
+                return "FT->getReturnType()"
+            return f"FT->getParamType({index - 1})"
+
+        for index_tuple, opcodes in extended_dict.items():
+            line = ""
+            for opcode in opcodes:
+                line = line + f"case OpCode::{opcode}:\n"
+            if index_tuple[-1] > 0:
+                line += (
+                    f"  if (FT->getNumParams() < {index_tuple[-1]})\n"
+                    + "    return nullptr;\n"
+                )
+            line += (
+                "  return llvm::StructType::get(Ctx, {"
+                + ", ".join([GetTypeAtIndex(index) for index in index_tuple])
+                + "});\n"
+            )
+            print(line)
+
 
 class db_valfns_gen:
     "A generator of validation functions."
@@ -1045,22 +1157,22 @@ def wrap_with_ifdef_if_vulkan_specific(intrinsic, text):
 def enum_hlsl_intrinsics():
     db = get_db_hlsl()
     result = ""
-    enumed = []
+    enumed = set()
     for i in sorted(db.intrinsics, key=lambda x: x.key):
         if i.enum_name not in enumed:
-            enumerant = "  %s,\n" % (i.enum_name)
-            result += wrap_with_ifdef_if_vulkan_specific(i, enumerant)  # SPIRV Change
-            enumed.append(i.enum_name)
+            result += "  %s = %d,\n" % (i.enum_name, i.opcode)
+            enumed.add(i.enum_name)
     # unsigned
     result += "  // unsigned\n"
 
     for i in sorted(db.intrinsics, key=lambda x: x.key):
         if i.unsigned_op != "":
             if i.unsigned_op not in enumed:
-                result += "  %s,\n" % (i.unsigned_op)
-                enumed.append(i.unsigned_op)
+                result += "  %s = %d,\n" % (i.unsigned_op, i.unsigned_opcode)
+                enumed.add(i.unsigned_op)
 
-    result += "  Num_Intrinsics,\n"
+    Num_Intrinsics = get_hlsl_opcode_data()["IntrinsicOpCodes"]["Num_Intrinsics"]
+    result += "  Num_Intrinsics = %d,\n" % (Num_Intrinsics)
     return result
 
 
@@ -1570,6 +1682,7 @@ def get_highest_released_shader_model():
     )
     return result
 
+
 def get_highest_shader_model():
     result = """static const unsigned kHighestMajor = %d;
 static const unsigned kHighestMinor = %d;""" % (
@@ -1578,6 +1691,7 @@ def get_highest_shader_model():
     )
     return result
 
+
 def get_dxil_version_minor():
     return "const unsigned kDxilMinor = %d;" % highest_minor
 
diff --git a/utils/hct/hctgen.py b/utils/hct/hctgen.py
index dbb7e3a745..1421fbfad5 100755
--- a/utils/hct/hctgen.py
+++ b/utils/hct/hctgen.py
@@ -2,6 +2,7 @@
 import argparse
 from hctdb_instrhelp import *
 from hctdb import *
+import json
 import sys
 import os
 import CodeTags
@@ -28,6 +29,7 @@
         "DxilCounters",
         "DxilMetadata",
         "RDAT_LibraryTypes",
+        "HlslIntrinsicOpcodes",
     ],
 )
 parser.add_argument("--output", required=True)
@@ -232,6 +234,14 @@ def writeDxilPIXPasses(args):
     return 0
 
 
+def writeHlslIntrinsicOpcodes(args):
+    out = openOutput(args)
+    # get_db_hlsl() initializes the hlsl intrinsic database and opcode_data.
+    get_db_hlsl()
+    json.dump(get_hlsl_opcode_data(), out, indent=2)
+    out.write("\n")
+    return 0
+
 args = parser.parse_args()
 if args.force_lf and args.force_crlf:
     eprint("--force-lf and --force-crlf are mutually exclusive, only pass one")
diff --git a/utils/hct/hlsl_intrinsic_opcodes.json b/utils/hct/hlsl_intrinsic_opcodes.json
new file mode 100644
index 0000000000..48a0b74c17
--- /dev/null
+++ b/utils/hct/hlsl_intrinsic_opcodes.json
@@ -0,0 +1,363 @@
+{
+  "IntrinsicOpCodes": {
+    "Num_Intrinsics": 358,
+    "IOP_AcceptHitAndEndSearch": 0,
+    "IOP_AddUint64": 1,
+    "IOP_AllMemoryBarrier": 2,
+    "IOP_AllMemoryBarrierWithGroupSync": 3,
+    "IOP_AllocateRayQuery": 4,
+    "IOP_Barrier": 5,
+    "IOP_CallShader": 6,
+    "IOP_CheckAccessFullyMapped": 7,
+    "IOP_CreateResourceFromHeap": 8,
+    "IOP_D3DCOLORtoUBYTE4": 9,
+    "IOP_DeviceMemoryBarrier": 10,
+    "IOP_DeviceMemoryBarrierWithGroupSync": 11,
+    "IOP_DispatchMesh": 12,
+    "IOP_DispatchRaysDimensions": 13,
+    "IOP_DispatchRaysIndex": 14,
+    "IOP_EvaluateAttributeAtSample": 15,
+    "IOP_EvaluateAttributeCentroid": 16,
+    "IOP_EvaluateAttributeSnapped": 17,
+    "IOP_GeometryIndex": 18,
+    "IOP_GetAttributeAtVertex": 19,
+    "IOP_GetRemainingRecursionLevels": 20,
+    "IOP_GetRenderTargetSampleCount": 21,
+    "IOP_GetRenderTargetSamplePosition": 22,
+    "IOP_GroupMemoryBarrier": 23,
+    "IOP_GroupMemoryBarrierWithGroupSync": 24,
+    "IOP_HitKind": 25,
+    "IOP_IgnoreHit": 26,
+    "IOP_InstanceID": 27,
+    "IOP_InstanceIndex": 28,
+    "IOP_InterlockedAdd": 29,
+    "IOP_InterlockedAnd": 30,
+    "IOP_InterlockedCompareExchange": 31,
+    "IOP_InterlockedCompareExchangeFloatBitwise": 32,
+    "IOP_InterlockedCompareStore": 33,
+    "IOP_InterlockedCompareStoreFloatBitwise": 34,
+    "IOP_InterlockedExchange": 35,
+    "IOP_InterlockedMax": 36,
+    "IOP_InterlockedMin": 37,
+    "IOP_InterlockedOr": 38,
+    "IOP_InterlockedXor": 39,
+    "IOP_IsHelperLane": 40,
+    "IOP_NonUniformResourceIndex": 41,
+    "IOP_ObjectRayDirection": 42,
+    "IOP_ObjectRayOrigin": 43,
+    "IOP_ObjectToWorld": 44,
+    "IOP_ObjectToWorld3x4": 45,
+    "IOP_ObjectToWorld4x3": 46,
+    "IOP_PrimitiveIndex": 47,
+    "IOP_Process2DQuadTessFactorsAvg": 48,
+    "IOP_Process2DQuadTessFactorsMax": 49,
+    "IOP_Process2DQuadTessFactorsMin": 50,
+    "IOP_ProcessIsolineTessFactors": 51,
+    "IOP_ProcessQuadTessFactorsAvg": 52,
+    "IOP_ProcessQuadTessFactorsMax": 53,
+    "IOP_ProcessQuadTessFactorsMin": 54,
+    "IOP_ProcessTriTessFactorsAvg": 55,
+    "IOP_ProcessTriTessFactorsMax": 56,
+    "IOP_ProcessTriTessFactorsMin": 57,
+    "IOP_QuadAll": 58,
+    "IOP_QuadAny": 59,
+    "IOP_QuadReadAcrossDiagonal": 60,
+    "IOP_QuadReadAcrossX": 61,
+    "IOP_QuadReadAcrossY": 62,
+    "IOP_QuadReadLaneAt": 63,
+    "IOP_RayFlags": 64,
+    "IOP_RayTCurrent": 65,
+    "IOP_RayTMin": 66,
+    "IOP_ReportHit": 67,
+    "IOP_SetMeshOutputCounts": 68,
+    "IOP_TraceRay": 69,
+    "IOP_WaveActiveAllEqual": 70,
+    "IOP_WaveActiveAllTrue": 71,
+    "IOP_WaveActiveAnyTrue": 72,
+    "IOP_WaveActiveBallot": 73,
+    "IOP_WaveActiveBitAnd": 74,
+    "IOP_WaveActiveBitOr": 75,
+    "IOP_WaveActiveBitXor": 76,
+    "IOP_WaveActiveCountBits": 77,
+    "IOP_WaveActiveMax": 78,
+    "IOP_WaveActiveMin": 79,
+    "IOP_WaveActiveProduct": 80,
+    "IOP_WaveActiveSum": 81,
+    "IOP_WaveGetLaneCount": 82,
+    "IOP_WaveGetLaneIndex": 83,
+    "IOP_WaveIsFirstLane": 84,
+    "IOP_WaveMatch": 85,
+    "IOP_WaveMultiPrefixBitAnd": 86,
+    "IOP_WaveMultiPrefixBitOr": 87,
+    "IOP_WaveMultiPrefixBitXor": 88,
+    "IOP_WaveMultiPrefixCountBits": 89,
+    "IOP_WaveMultiPrefixProduct": 90,
+    "IOP_WaveMultiPrefixSum": 91,
+    "IOP_WavePrefixCountBits": 92,
+    "IOP_WavePrefixProduct": 93,
+    "IOP_WavePrefixSum": 94,
+    "IOP_WaveReadLaneAt": 95,
+    "IOP_WaveReadLaneFirst": 96,
+    "IOP_WorldRayDirection": 97,
+    "IOP_WorldRayOrigin": 98,
+    "IOP_WorldToObject": 99,
+    "IOP_WorldToObject3x4": 100,
+    "IOP_WorldToObject4x3": 101,
+    "IOP_abort": 102,
+    "IOP_abs": 103,
+    "IOP_acos": 104,
+    "IOP_all": 105,
+    "IOP_and": 106,
+    "IOP_any": 107,
+    "IOP_asdouble": 108,
+    "IOP_asfloat": 109,
+    "IOP_asfloat16": 110,
+    "IOP_asin": 111,
+    "IOP_asint": 112,
+    "IOP_asint16": 113,
+    "IOP_asuint": 114,
+    "IOP_asuint16": 115,
+    "IOP_atan": 116,
+    "IOP_atan2": 117,
+    "IOP_ceil": 118,
+    "IOP_clamp": 119,
+    "IOP_clip": 120,
+    "IOP_cos": 121,
+    "IOP_cosh": 122,
+    "IOP_countbits": 123,
+    "IOP_cross": 124,
+    "IOP_ddx": 125,
+    "IOP_ddx_coarse": 126,
+    "IOP_ddx_fine": 127,
+    "IOP_ddy": 128,
+    "IOP_ddy_coarse": 129,
+    "IOP_ddy_fine": 130,
+    "IOP_degrees": 131,
+    "IOP_determinant": 132,
+    "IOP_distance": 133,
+    "IOP_dot": 134,
+    "IOP_dot2add": 135,
+    "IOP_dot4add_i8packed": 136,
+    "IOP_dot4add_u8packed": 137,
+    "IOP_dst": 138,
+    "IOP_exp": 139,
+    "IOP_exp2": 140,
+    "IOP_f16tof32": 141,
+    "IOP_f32tof16": 142,
+    "IOP_faceforward": 143,
+    "IOP_firstbithigh": 144,
+    "IOP_firstbitlow": 145,
+    "IOP_floor": 146,
+    "IOP_fma": 147,
+    "IOP_fmod": 148,
+    "IOP_frac": 149,
+    "IOP_frexp": 150,
+    "IOP_fwidth": 151,
+    "IOP_isfinite": 152,
+    "IOP_isinf": 153,
+    "IOP_isnan": 154,
+    "IOP_ldexp": 155,
+    "IOP_length": 156,
+    "IOP_lerp": 157,
+    "IOP_lit": 158,
+    "IOP_log": 159,
+    "IOP_log10": 160,
+    "IOP_log2": 161,
+    "IOP_mad": 162,
+    "IOP_max": 163,
+    "IOP_min": 164,
+    "IOP_modf": 165,
+    "IOP_msad4": 166,
+    "IOP_mul": 167,
+    "IOP_normalize": 168,
+    "IOP_or": 169,
+    "IOP_pack_clamp_s8": 170,
+    "IOP_pack_clamp_u8": 171,
+    "IOP_pack_s8": 172,
+    "IOP_pack_u8": 173,
+    "IOP_pow": 174,
+    "IOP_printf": 175,
+    "IOP_radians": 176,
+    "IOP_rcp": 177,
+    "IOP_reflect": 178,
+    "IOP_refract": 179,
+    "IOP_reversebits": 180,
+    "IOP_round": 181,
+    "IOP_rsqrt": 182,
+    "IOP_saturate": 183,
+    "IOP_select": 184,
+    "IOP_sign": 185,
+    "IOP_sin": 186,
+    "IOP_sincos": 187,
+    "IOP_sinh": 188,
+    "IOP_smoothstep": 189,
+    "IOP_source_mark": 190,
+    "IOP_sqrt": 191,
+    "IOP_step": 192,
+    "IOP_tan": 193,
+    "IOP_tanh": 194,
+    "IOP_tex1D": 195,
+    "IOP_tex1Dbias": 196,
+    "IOP_tex1Dgrad": 197,
+    "IOP_tex1Dlod": 198,
+    "IOP_tex1Dproj": 199,
+    "IOP_tex2D": 200,
+    "IOP_tex2Dbias": 201,
+    "IOP_tex2Dgrad": 202,
+    "IOP_tex2Dlod": 203,
+    "IOP_tex2Dproj": 204,
+    "IOP_tex3D": 205,
+    "IOP_tex3Dbias": 206,
+    "IOP_tex3Dgrad": 207,
+    "IOP_tex3Dlod": 208,
+    "IOP_tex3Dproj": 209,
+    "IOP_texCUBE": 210,
+    "IOP_texCUBEbias": 211,
+    "IOP_texCUBEgrad": 212,
+    "IOP_texCUBElod": 213,
+    "IOP_texCUBEproj": 214,
+    "IOP_transpose": 215,
+    "IOP_trunc": 216,
+    "IOP_unpack_s8s16": 217,
+    "IOP_unpack_s8s32": 218,
+    "IOP_unpack_u8u16": 219,
+    "IOP_unpack_u8u32": 220,
+    "IOP_VkRawBufferLoad": 221,
+    "IOP_VkRawBufferStore": 222,
+    "IOP_VkReadClock": 223,
+    "IOP_Vkext_execution_mode": 224,
+    "IOP_Vkext_execution_mode_id": 225,
+    "MOP_Append": 226,
+    "MOP_RestartStrip": 227,
+    "MOP_CalculateLevelOfDetail": 228,
+    "MOP_CalculateLevelOfDetailUnclamped": 229,
+    "MOP_GetDimensions": 230,
+    "MOP_Load": 231,
+    "MOP_Sample": 232,
+    "MOP_SampleBias": 233,
+    "MOP_SampleCmp": 234,
+    "MOP_SampleCmpBias": 235,
+    "MOP_SampleCmpGrad": 236,
+    "MOP_SampleCmpLevel": 237,
+    "MOP_SampleCmpLevelZero": 238,
+    "MOP_SampleGrad": 239,
+    "MOP_SampleLevel": 240,
+    "MOP_Gather": 241,
+    "MOP_GatherAlpha": 242,
+    "MOP_GatherBlue": 243,
+    "MOP_GatherCmp": 244,
+    "MOP_GatherCmpAlpha": 245,
+    "MOP_GatherCmpBlue": 246,
+    "MOP_GatherCmpGreen": 247,
+    "MOP_GatherCmpRed": 248,
+    "MOP_GatherGreen": 249,
+    "MOP_GatherRaw": 250,
+    "MOP_GatherRed": 251,
+    "MOP_GetSamplePosition": 252,
+    "MOP_Load2": 253,
+    "MOP_Load3": 254,
+    "MOP_Load4": 255,
+    "MOP_InterlockedAdd": 256,
+    "MOP_InterlockedAdd64": 257,
+    "MOP_InterlockedAnd": 258,
+    "MOP_InterlockedAnd64": 259,
+    "MOP_InterlockedCompareExchange": 260,
+    "MOP_InterlockedCompareExchange64": 261,
+    "MOP_InterlockedCompareExchangeFloatBitwise": 262,
+    "MOP_InterlockedCompareStore": 263,
+    "MOP_InterlockedCompareStore64": 264,
+    "MOP_InterlockedCompareStoreFloatBitwise": 265,
+    "MOP_InterlockedExchange": 266,
+    "MOP_InterlockedExchange64": 267,
+    "MOP_InterlockedExchangeFloat": 268,
+    "MOP_InterlockedMax": 269,
+    "MOP_InterlockedMax64": 270,
+    "MOP_InterlockedMin": 271,
+    "MOP_InterlockedMin64": 272,
+    "MOP_InterlockedOr": 273,
+    "MOP_InterlockedOr64": 274,
+    "MOP_InterlockedXor": 275,
+    "MOP_InterlockedXor64": 276,
+    "MOP_Store": 277,
+    "MOP_Store2": 278,
+    "MOP_Store3": 279,
+    "MOP_Store4": 280,
+    "MOP_DecrementCounter": 281,
+    "MOP_IncrementCounter": 282,
+    "MOP_Consume": 283,
+    "MOP_WriteSamplerFeedback": 284,
+    "MOP_WriteSamplerFeedbackBias": 285,
+    "MOP_WriteSamplerFeedbackGrad": 286,
+    "MOP_WriteSamplerFeedbackLevel": 287,
+    "MOP_Abort": 288,
+    "MOP_CandidateGeometryIndex": 289,
+    "MOP_CandidateInstanceContributionToHitGroupIndex": 290,
+    "MOP_CandidateInstanceID": 291,
+    "MOP_CandidateInstanceIndex": 292,
+    "MOP_CandidateObjectRayDirection": 293,
+    "MOP_CandidateObjectRayOrigin": 294,
+    "MOP_CandidateObjectToWorld3x4": 295,
+    "MOP_CandidateObjectToWorld4x3": 296,
+    "MOP_CandidatePrimitiveIndex": 297,
+    "MOP_CandidateProceduralPrimitiveNonOpaque": 298,
+    "MOP_CandidateTriangleBarycentrics": 299,
+    "MOP_CandidateTriangleFrontFace": 300,
+    "MOP_CandidateTriangleRayT": 301,
+    "MOP_CandidateType": 302,
+    "MOP_CandidateWorldToObject3x4": 303,
+    "MOP_CandidateWorldToObject4x3": 304,
+    "MOP_CommitNonOpaqueTriangleHit": 305,
+    "MOP_CommitProceduralPrimitiveHit": 306,
+    "MOP_CommittedGeometryIndex": 307,
+    "MOP_CommittedInstanceContributionToHitGroupIndex": 308,
+    "MOP_CommittedInstanceID": 309,
+    "MOP_CommittedInstanceIndex": 310,
+    "MOP_CommittedObjectRayDirection": 311,
+    "MOP_CommittedObjectRayOrigin": 312,
+    "MOP_CommittedObjectToWorld3x4": 313,
+    "MOP_CommittedObjectToWorld4x3": 314,
+    "MOP_CommittedPrimitiveIndex": 315,
+    "MOP_CommittedRayT": 316,
+    "MOP_CommittedStatus": 317,
+    "MOP_CommittedTriangleBarycentrics": 318,
+    "MOP_CommittedTriangleFrontFace": 319,
+    "MOP_CommittedWorldToObject3x4": 320,
+    "MOP_CommittedWorldToObject4x3": 321,
+    "MOP_Proceed": 322,
+    "MOP_RayFlags": 323,
+    "MOP_RayTMin": 324,
+    "MOP_TraceRayInline": 325,
+    "MOP_WorldRayDirection": 326,
+    "MOP_WorldRayOrigin": 327,
+    "MOP_Count": 328,
+    "MOP_FinishedCrossGroupSharing": 329,
+    "MOP_GetGroupNodeOutputRecords": 330,
+    "MOP_GetThreadNodeOutputRecords": 331,
+    "MOP_IsValid": 332,
+    "MOP_GroupIncrementOutputCount": 333,
+    "MOP_ThreadIncrementOutputCount": 334,
+    "MOP_OutputComplete": 335,
+    "MOP_SubpassLoad": 336,
+    "IOP_InterlockedUMax": 337,
+    "IOP_InterlockedUMin": 338,
+    "IOP_WaveActiveUMax": 339,
+    "IOP_WaveActiveUMin": 340,
+    "IOP_WaveActiveUProduct": 341,
+    "IOP_WaveActiveUSum": 342,
+    "IOP_WaveMultiPrefixUProduct": 343,
+    "IOP_WaveMultiPrefixUSum": 344,
+    "IOP_WavePrefixUProduct": 345,
+    "IOP_WavePrefixUSum": 346,
+    "IOP_uabs": 347,
+    "IOP_uclamp": 348,
+    "IOP_udot": 349,
+    "IOP_ufirstbithigh": 350,
+    "IOP_umad": 351,
+    "IOP_umax": 352,
+    "IOP_umin": 353,
+    "IOP_umul": 354,
+    "IOP_usign": 355,
+    "MOP_InterlockedUMax": 356,
+    "MOP_InterlockedUMin": 357
+  }
+}