diff --git a/lib/SPIRV/OCLUtil.cpp b/lib/SPIRV/OCLUtil.cpp
index aadfb0ceae..a403975eb8 100644
--- a/lib/SPIRV/OCLUtil.cpp
+++ b/lib/SPIRV/OCLUtil.cpp
@@ -898,6 +898,7 @@ SPIRAddressSpace getOCLOpaqueTypeAddrSpace(Op OpCode) {
   case OpTypeSampler:
     return SPIRV_SAMPLER_T_ADDR_SPACE;
   case internal::OpTypeJointMatrixINTEL:
+  case internal::OpTypeJointMatrixINTELv2:
   case OpTypeCooperativeMatrixKHR:
     return SPIRAS_Global;
   default:
diff --git a/lib/SPIRV/SPIRVInternal.h b/lib/SPIRV/SPIRVInternal.h
index ced8c147a6..b00938255e 100644
--- a/lib/SPIRV/SPIRVInternal.h
+++ b/lib/SPIRV/SPIRVInternal.h
@@ -318,6 +318,7 @@ const static char PipeStorage[] = "PipeStorage";
 const static char ConstantPipeStorage[] = "ConstantPipeStorage";
 const static char VmeImageINTEL[] = "VmeImageINTEL";
 const static char JointMatrixINTEL[] = "JointMatrixINTEL";
+const static char BufferSurfaceINTEL[] = "BufferSurfaceINTEL";
 const static char CooperativeMatrixKHR[] = "CooperativeMatrixKHR";
 } // namespace kSPIRVTypeName
 
@@ -976,6 +977,7 @@ template <> inline void SPIRVMap<std::string, Op, SPIRVOpaqueType>::init() {
   _SPIRV_OP(AvcRefResultINTEL)
   _SPIRV_OP(AvcSicResultINTEL)
   _SPIRV_OP(VmeImageINTEL)
+  _SPIRV_OP(BufferSurfaceINTEL)
   _SPIRV_OP(CooperativeMatrixKHR)
 #undef _SPIRV_OP
   add("JointMatrixINTEL", internal::OpTypeJointMatrixINTEL);
diff --git a/lib/SPIRV/SPIRVReader.cpp b/lib/SPIRV/SPIRVReader.cpp
index 581f5ee192..0a64d9647c 100644
--- a/lib/SPIRV/SPIRVReader.cpp
+++ b/lib/SPIRV/SPIRVReader.cpp
@@ -472,15 +472,26 @@ Type *SPIRVToLLVM::transType(SPIRVType *T, bool UseTPT) {
     auto *MT = static_cast<SPIRVTypeJointMatrixINTEL *>(T);
     auto R = static_cast<SPIRVConstant *>(MT->getRows())->getZExtIntValue();
     auto C = static_cast<SPIRVConstant *>(MT->getColumns())->getZExtIntValue();
-    auto L = static_cast<SPIRVConstant *>(MT->getLayout())->getZExtIntValue();
-    auto S = static_cast<SPIRVConstant *>(MT->getScope())->getZExtIntValue();
-    SmallVector<unsigned, 5> Params = {(unsigned)R, (unsigned)C, (unsigned)L,
-                                       (unsigned)S};
+    std::vector<unsigned> Params = {(unsigned)R, (unsigned)C};
+    if (auto *Layout = MT->getLayout())
+      Params.push_back(static_cast<SPIRVConstant *>(Layout)->getZExtIntValue());
+    Params.push_back(
+        static_cast<SPIRVConstant *>(MT->getScope())->getZExtIntValue());
     if (auto *Use = MT->getUse())
       Params.push_back(static_cast<SPIRVConstant *>(Use)->getZExtIntValue());
-    return mapType(T, getSPIRVType(internal::OpTypeJointMatrixINTEL,
-                                   transTypeToOCLTypeName(MT->getCompType()),
-                                   Params, !UseTPT));
+    auto *CTI = MT->getComponentTypeInterpretation();
+    if (!CTI)
+      return mapType(
+          T, llvm::TargetExtType::get(*Context, "spirv.JointMatrixINTEL",
+                                      transType(MT->getCompType()), Params));
+    const unsigned CTIValue =
+        static_cast<SPIRVConstant *>(CTI)->getZExtIntValue();
+    assert(CTIValue <= internal::InternalJointMatrixCTI::PackedInt4 &&
+           "Unknown matrix component type interpretation");
+    Params.push_back(CTIValue);
+    return mapType(
+        T, llvm::TargetExtType::get(*Context, "spirv.JointMatrixINTEL",
+                                    transType(MT->getCompType()), Params));
   }
   case OpTypeCooperativeMatrixKHR: {
     auto *MT = static_cast<SPIRVTypeCooperativeMatrixKHR *>(T);
diff --git a/lib/SPIRV/SPIRVWriter.cpp b/lib/SPIRV/SPIRVWriter.cpp
index ac3d805d4d..268d356bc4 100644
--- a/lib/SPIRV/SPIRVWriter.cpp
+++ b/lib/SPIRV/SPIRVWriter.cpp
@@ -649,21 +649,6 @@ SPIRVType *LLVMToSPIRVBase::transPointerType(Type *ET, unsigned AddrSpc) {
           transType(ET)));
     }
   } else {
-    // JointMatrixINTEL type is not necessarily an opaque type, it can be
-    // represented as a structure with pointer to a multidimensional array
-    // member.
-    if (ST && ST->hasName()) {
-      StringRef STName = ST->getName();
-      if (STName.startswith(kSPIRVTypeName::PrefixAndDelim)) {
-        SmallVector<std::string, 8> Postfixes;
-        auto TN = decodeSPIRVTypeName(STName, Postfixes);
-        if (TN == kSPIRVTypeName::JointMatrixINTEL) {
-          SPIRVType *TranslatedTy = transSPIRVJointMatrixINTELType(Postfixes);
-          PointeeTypeMap[TypeKey] = TranslatedTy;
-          return TranslatedTy;
-        }
-      }
-    }
     SPIRVType *ElementType = transType(ET);
     // ET, as a recursive type, may contain exactly the same pointer T, so it
     // may happen that after translation of ET we already have translated T,
@@ -698,56 +683,6 @@ SPIRVType *LLVMToSPIRVBase::transPointerType(SPIRVType *ET, unsigned AddrSpc) {
   return TranslatedTy;
 }
 
-// Representation in LLVM IR before the translator is a pointer to an opaque
-// structure:
-// %spirv.JointMatrixINTEL._%element_type%_%rows%_%cols%_%scope%_%use%
-// Here we check the structure name yet again. Another option would be to
-// check SPIR-V friendly function calls (by their name) and obtain return
-// or their parameter types, assuming, that the appropriate types are Matrix
-// structure type. But in the near future, we will reuse Composite
-// instructions to do, for example, matrix initialization directly on AMX
-// register by OpCompositeConstruct. And we can't claim, that the Result type
-// of OpCompositeConstruct instruction is always the joint matrix type, it's
-// simply not true.
-SPIRVType *LLVMToSPIRVBase::transSPIRVJointMatrixINTELType(
-    SmallVector<std::string, 8> Postfixes) {
-  Type *ElemTy = nullptr;
-  StringRef Ty{Postfixes[0]};
-  auto NumBits = llvm::StringSwitch<unsigned>(Ty)
-                     .Case("char", 8)
-                     .Case("short", 16)
-                     .Case("int", 32)
-                     .Case("long", 64)
-                     .Default(0);
-  if (NumBits)
-    ElemTy = IntegerType::get(M->getContext(), NumBits);
-  else if (Ty == "half")
-    ElemTy = Type::getHalfTy(M->getContext());
-  else if (Ty == "float")
-    ElemTy = Type::getFloatTy(M->getContext());
-  else if (Ty == "double")
-    ElemTy = Type::getDoubleTy(M->getContext());
-  else if (Ty == "bfloat16")
-    ElemTy = Type::getInt16Ty(M->getContext());
-  else
-    llvm_unreachable("Unexpected type for matrix!");
-
-  auto ParseInteger = [this](StringRef Postfix) -> ConstantInt * {
-    unsigned long long N = 0;
-    if (consumeUnsignedInteger(Postfix, 10, N)) {
-      BM->getErrorLog().checkError(
-          false, SPIRVEC_InvalidLlvmModule,
-          "TypeJointMatrixINTEL expects integer parameters");
-      return 0;
-    }
-    return getUInt32(M, N);
-  };
-  std::vector<SPIRVValue *> Args;
-  for (size_t I = 1; I != Postfixes.size(); ++I)
-    Args.emplace_back(transConstant(ParseInteger(Postfixes[I])));
-  return BM->addJointMatrixINTELType(transType(ElemTy), Args);
-}
-
 SPIRVType *LLVMToSPIRVBase::transSPIRVOpaqueType(StringRef STName,
                                                  unsigned AddrSpace) {
   std::pair<StringRef, unsigned> Key = {STName, AddrSpace};
@@ -804,9 +739,7 @@ SPIRVType *LLVMToSPIRVBase::transSPIRVOpaqueType(StringRef STName,
     return SaveType(BM->addQueueType());
   else if (TN == kSPIRVTypeName::PipeStorage)
     return SaveType(BM->addPipeStorageType());
-  else if (TN == kSPIRVTypeName::JointMatrixINTEL) {
-    return SaveType(transSPIRVJointMatrixINTELType(Postfixes));
-  } else
+  else
     return SaveType(
         BM->addOpaqueGenericType(SPIRVOpaqueTypeOpCodeMap::map(TN)));
 }
diff --git a/lib/SPIRV/libSPIRV/SPIRVEntry.cpp b/lib/SPIRV/libSPIRV/SPIRVEntry.cpp
index 16a918b18e..b860eb9d6e 100644
--- a/lib/SPIRV/libSPIRV/SPIRVEntry.cpp
+++ b/lib/SPIRV/libSPIRV/SPIRVEntry.cpp
@@ -84,6 +84,10 @@ SPIRVEntry *SPIRVEntry::create(Op OpCode) {
   static const OpToFactoryMapTy OpToFactoryMap(std::begin(Table),
                                                std::end(Table));
 
+  // TODO: To remove this when we make a switch to new version
+  if (OpCode == internal::OpTypeJointMatrixINTELv2)
+    OpCode = internal::OpTypeJointMatrixINTEL;
+
   OpToFactoryMapTy::const_iterator Loc = OpToFactoryMap.find(OpCode);
   if (Loc != OpToFactoryMap.end())
     return Loc->second();
diff --git a/lib/SPIRV/libSPIRV/SPIRVEnum.h b/lib/SPIRV/libSPIRV/SPIRVEnum.h
index c691c4e4f4..8bbb3b7bb0 100644
--- a/lib/SPIRV/libSPIRV/SPIRVEnum.h
+++ b/lib/SPIRV/libSPIRV/SPIRVEnum.h
@@ -207,10 +207,20 @@ template <> inline void SPIRVMap<SPIRVCapabilityKind, SPIRVCapVec>::init() {
                {CapabilitySubgroupAvcMotionEstimationIntraINTEL});
   ADD_VEC_INIT(internal::CapabilityJointMatrixWIInstructionsINTEL,
                {internal::CapabilityJointMatrixINTEL});
+  ADD_VEC_INIT(internal::CapabilityJointMatrixTF32ComponentTypeINTEL,
+               {internal::CapabilityJointMatrixINTEL});
+  ADD_VEC_INIT(internal::CapabilityJointMatrixBF16ComponentTypeINTEL,
+               {internal::CapabilityJointMatrixINTEL});
+  ADD_VEC_INIT(internal::CapabilityJointMatrixPackedInt2ComponentTypeINTEL,
+               {internal::CapabilityJointMatrixINTEL});
+  ADD_VEC_INIT(internal::CapabilityJointMatrixPackedInt4ComponentTypeINTEL,
+               {internal::CapabilityJointMatrixINTEL});
   ADD_VEC_INIT(internal::CapabilityCooperativeMatrixCheckedInstructionsINTEL,
                {CapabilityCooperativeMatrixKHR});
   ADD_VEC_INIT(internal::CapabilityCooperativeMatrixPrefetchINTEL,
                {CapabilityCooperativeMatrixKHR});
+  ADD_VEC_INIT(internal::CapabilityCooperativeMatrixInvocationInstructionsINTEL,
+               {CapabilityCooperativeMatrixKHR});
 }
 
 template <> inline void SPIRVMap<SPIRVExecutionModelKind, SPIRVCapVec>::init() {
diff --git a/lib/SPIRV/libSPIRV/SPIRVInstruction.h b/lib/SPIRV/libSPIRV/SPIRVInstruction.h
index c6fc1bd678..82c7f33f19 100644
--- a/lib/SPIRV/libSPIRV/SPIRVInstruction.h
+++ b/lib/SPIRV/libSPIRV/SPIRVInstruction.h
@@ -1991,6 +1991,7 @@ class SPIRVCompositeConstruct : public SPIRVInstruction {
     case OpTypeArray:
     case OpTypeStruct:
     case internal::OpTypeJointMatrixINTEL:
+    case internal::OpTypeJointMatrixINTELv2:
     case OpTypeCooperativeMatrixKHR:
       break;
     default:
@@ -3406,10 +3407,17 @@ template <Op OC>
 class SPIRVBfloat16ConversionINTELInstBase : public SPIRVUnaryInst<OC> {
 protected:
   SPIRVCapVec getRequiredCapability() const override {
+    SPIRVType *ResCompTy = this->getType();
+    if (ResCompTy->isTypeCooperativeMatrixKHR())
+      return getVec(internal::CapabilityBfloat16ConversionINTEL,
+                    internal::CapabilityJointMatrixBF16ComponentTypeINTEL);
     return getVec(internal::CapabilityBfloat16ConversionINTEL);
   }
 
   std::optional<ExtensionID> getRequiredExtension() const override {
+    SPIRVType *ResCompTy = this->getType();
+    if (ResCompTy->isTypeCooperativeMatrixKHR())
+      this->getModule()->addExtension(ExtensionID::SPV_INTEL_joint_matrix);
     return ExtensionID::SPV_INTEL_bfloat16_conversion;
   }
 
@@ -3438,8 +3446,25 @@ class SPIRVBfloat16ConversionINTELInstBase : public SPIRVUnaryInst<OC> {
     }
 
     auto InstName = OpCodeNameMap::map(OC);
-    SPIRVErrorLog &SPVErrLog = this->getModule()->getErrorLog();
+    auto *Module = this->getModule();
+    SPIRVErrorLog &SPVErrLog = Module->getErrorLog();
 
+    // Cooperative matrix type is allowed as input/output of the instruction
+    // if SPV_INTEL_joint_matrix is enabled
+    if (ResCompTy->isTypeCooperativeMatrixKHR()) {
+      SPVErrLog.checkError(
+          Module->isAllowedToUseExtension(ExtensionID::SPV_INTEL_joint_matrix),
+          SPIRVEC_InvalidInstruction,
+          InstName + "\nCan be used with "
+                     "cooperative matrices only when SPV_INTEL_joint_matrix is "
+                     "enabled\n");
+      assert(InCompTy->isTypeCooperativeMatrixKHR() &&
+             "Input must also be a cooperative matrix");
+      ResCompTy = static_cast<SPIRVTypeCooperativeMatrixKHR *>(ResCompTy)
+                      ->getCompType();
+      InCompTy =
+          static_cast<SPIRVTypeCooperativeMatrixKHR *>(InCompTy)->getCompType();
+    }
     if (OC == internal::OpConvertFToBF16INTEL) {
       SPVErrLog.checkError(
           ResCompTy->isTypeInt(16), SPIRVEC_InvalidInstruction,
@@ -3492,10 +3517,10 @@ class SPIRVJointMatrixINTELInst : public SPIRVJointMatrixINTELInstBase {
       SPIRV##x##INTEL;
 _SPIRV_OP(JointMatrixLoad, true, 6, true)
 _SPIRV_OP(JointMatrixStore, false, 5, true)
-_SPIRV_OP(JointMatrixMad, true, 7)
-_SPIRV_OP(JointMatrixSUMad, true, 7)
-_SPIRV_OP(JointMatrixUSMad, true, 7)
-_SPIRV_OP(JointMatrixUUMad, true, 7)
+_SPIRV_OP(JointMatrixMad, true, 6, true)
+_SPIRV_OP(JointMatrixSUMad, true, 6, true)
+_SPIRV_OP(JointMatrixUSMad, true, 6, true)
+_SPIRV_OP(JointMatrixUUMad, true, 6, true)
 // TODO: move to SPIRVJointMatrixINTELWorkItemInst
 _SPIRV_OP(JointMatrixWorkItemLength, true, 4)
 #undef _SPIRV_OP
@@ -3529,7 +3554,27 @@ class SPIRVCooperativeMatrixPrefetchINTELInstBase
   typedef SPIRVInstTemplate<SPIRVCooperativeMatrixPrefetchINTELInstBase,       \
                             internal::Op##x##INTEL, __VA_ARGS__>               \
       SPIRV##x##INTEL;
-_SPIRV_OP(CooperativeMatrixPrefetch, false, 8, true, 5)
+_SPIRV_OP(CooperativeMatrixPrefetch, false, 6, true, 3)
+#undef _SPIRV_OP
+
+class SPIRVCooperativeMatrixInvocationInstructionsINTELInstBase
+    : public SPIRVInstTemplateBase {
+protected:
+  std::optional<ExtensionID> getRequiredExtension() const override {
+    return ExtensionID::SPV_INTEL_joint_matrix;
+  }
+  SPIRVCapVec getRequiredCapability() const override {
+    return getVec(
+        internal::CapabilityCooperativeMatrixInvocationInstructionsINTEL);
+  }
+};
+
+#define _SPIRV_OP(x, ...)                                                      \
+  typedef SPIRVInstTemplate<                                                   \
+      SPIRVCooperativeMatrixInvocationInstructionsINTELInstBase,               \
+      internal::Op##x##INTEL, __VA_ARGS__>                                     \
+      SPIRV##x##INTEL;
+_SPIRV_OP(CooperativeMatrixApplyFunction, true, 5)
 #undef _SPIRV_OP
 
 class SPIRVCooperativeMatrixKHRInstBase : public SPIRVInstTemplateBase {
@@ -3813,10 +3858,17 @@ template <Op OC>
 class SPIRVTensorFloat32RoundingINTELInstBase : public SPIRVUnaryInst<OC> {
 protected:
   SPIRVCapVec getRequiredCapability() const override {
+    SPIRVType *ResCompTy = this->getType();
+    if (ResCompTy->isTypeCooperativeMatrixKHR())
+      return getVec(internal::CapabilityTensorFloat32RoundingINTEL,
+                    internal::CapabilityJointMatrixTF32ComponentTypeINTEL);
     return getVec(internal::CapabilityTensorFloat32RoundingINTEL);
   }
 
   std::optional<ExtensionID> getRequiredExtension() const override {
+    SPIRVType *ResCompTy = this->getType();
+    if (ResCompTy->isTypeCooperativeMatrixKHR())
+      this->getModule()->addExtension(ExtensionID::SPV_INTEL_joint_matrix);
     return ExtensionID::SPV_INTEL_tensor_float32_conversion;
   }
 
@@ -3845,7 +3897,25 @@ class SPIRVTensorFloat32RoundingINTELInstBase : public SPIRVUnaryInst<OC> {
     }
 
     auto InstName = OpCodeNameMap::map(OC);
-    SPIRVErrorLog &SPVErrLog = this->getModule()->getErrorLog();
+    auto *Module = this->getModule();
+    SPIRVErrorLog &SPVErrLog = Module->getErrorLog();
+
+    // Cooperative matrix type is allowed as input/output of the instruction
+    // if SPV_INTEL_joint_matrix is enabled
+    if (ResCompTy->isTypeCooperativeMatrixKHR()) {
+      SPVErrLog.checkError(
+          Module->isAllowedToUseExtension(ExtensionID::SPV_INTEL_joint_matrix),
+          SPIRVEC_InvalidInstruction,
+          InstName + "\nCan be used with "
+                     "cooperative matrices only when SPV_INTEL_joint_matrix is "
+                     "enabled\n");
+      assert(InCompTy->isTypeCooperativeMatrixKHR() &&
+             "Input must also be a cooperative matrix");
+      ResCompTy = static_cast<SPIRVTypeCooperativeMatrixKHR *>(ResCompTy)
+                      ->getCompType();
+      InCompTy =
+          static_cast<SPIRVTypeCooperativeMatrixKHR *>(InCompTy)->getCompType();
+    }
 
     SPVErrLog.checkError(
         ResCompTy->isTypeFloat(32), SPIRVEC_InvalidInstruction,
diff --git a/lib/SPIRV/libSPIRV/SPIRVNameMapEnum.h b/lib/SPIRV/libSPIRV/SPIRVNameMapEnum.h
index a0f1e4e1e2..b37434796f 100644
--- a/lib/SPIRV/libSPIRV/SPIRVNameMapEnum.h
+++ b/lib/SPIRV/libSPIRV/SPIRVNameMapEnum.h
@@ -652,8 +652,18 @@ template <> inline void SPIRVMap<Capability, std::string>::init() {
   add(internal::CapabilityCacheControlsINTEL, "CacheControlsINTEL");
   add(internal::CapabilityJointMatrixWIInstructionsINTEL,
       "JointMatrixWIInstructionsINTEL");
+  add(internal::CapabilityJointMatrixTF32ComponentTypeINTEL,
+      "JointMatrixTF32ComponentTypeINTEL");
+  add(internal::CapabilityJointMatrixBF16ComponentTypeINTEL,
+      "JointMatrixBF16ComponentTypeINTEL");
+  add(internal::CapabilityJointMatrixPackedInt2ComponentTypeINTEL,
+      "JointMatrixPackedInt2ComponentTypeINTEL");
+  add(internal::CapabilityJointMatrixPackedInt4ComponentTypeINTEL,
+      "JointMatrixPackedInt4ComponentTypeINTEL");
   add(internal::CapabilityCooperativeMatrixPrefetchINTEL,
       "CooperativeMatrixPrefetchINTEL");
+  add(internal::CapabilityCooperativeMatrixInvocationInstructionsINTEL,
+      "CooperativeMatrixInvocationInstructionsINTEL");
   add(internal::CapabilityCooperativeMatrixCheckedInstructionsINTEL,
       "CooperativeMatrixCheckedInstructionsINTEL");
   add(internal::CapabilityBindlessImagesINTEL, "BindlessImagesINTEL");
diff --git a/lib/SPIRV/libSPIRV/SPIRVOpCode.h b/lib/SPIRV/libSPIRV/SPIRVOpCode.h
index 7af20dd43c..cfa480faf0 100644
--- a/lib/SPIRV/libSPIRV/SPIRVOpCode.h
+++ b/lib/SPIRV/libSPIRV/SPIRVOpCode.h
@@ -230,6 +230,7 @@ inline bool isTypeOpCode(Op OpCode) {
          isSubgroupAvcINTELTypeOpCode(OpCode) || OC == OpTypeVmeImageINTEL ||
          isVCOpCode(OpCode) || OC == internal::OpTypeTokenINTEL ||
          OC == internal::OpTypeJointMatrixINTEL ||
+         OC == internal::OpTypeJointMatrixINTELv2 ||
          OC == OpTypeCooperativeMatrixKHR;
 }
 
diff --git a/lib/SPIRV/libSPIRV/SPIRVOpCodeEnumInternal.h b/lib/SPIRV/libSPIRV/SPIRVOpCodeEnumInternal.h
index 9fb2825b1b..a84ee56507 100644
--- a/lib/SPIRV/libSPIRV/SPIRVOpCodeEnumInternal.h
+++ b/lib/SPIRV/libSPIRV/SPIRVOpCodeEnumInternal.h
@@ -6,6 +6,7 @@ _SPIRV_OP_INTERNAL(ArithmeticFenceINTEL, internal::OpArithmeticFenceINTEL)
 _SPIRV_OP_INTERNAL(ConvertFToBF16INTEL, internal::OpConvertFToBF16INTEL)
 _SPIRV_OP_INTERNAL(ConvertBF16ToFINTEL, internal::OpConvertBF16ToFINTEL)
 _SPIRV_OP_INTERNAL(TypeJointMatrixINTEL, internal::OpTypeJointMatrixINTEL)
+_SPIRV_OP_INTERNAL(TypeJointMatrixINTEL, internal::OpTypeJointMatrixINTEL)
 _SPIRV_OP_INTERNAL(JointMatrixLoadINTEL, internal::OpJointMatrixLoadINTEL)
 _SPIRV_OP_INTERNAL(JointMatrixStoreINTEL, internal::OpJointMatrixStoreINTEL)
 _SPIRV_OP_INTERNAL(JointMatrixMadINTEL, internal::OpJointMatrixMadINTEL)
@@ -24,6 +25,8 @@ _SPIRV_OP_INTERNAL(CooperativeMatrixConstructCheckedINTEL,
                    internal::OpCooperativeMatrixConstructCheckedINTEL)
 _SPIRV_OP_INTERNAL(CooperativeMatrixPrefetchINTEL,
                    internal::OpCooperativeMatrixPrefetchINTEL)
+_SPIRV_OP_INTERNAL(CooperativeMatrixApplyFunctionINTEL,
+                   internal::OpCooperativeMatrixApplyFunctionINTEL)
 _SPIRV_OP_INTERNAL(ComplexFMulINTEL, internal::ComplexFMulINTEL)
 _SPIRV_OP_INTERNAL(ComplexFDivINTEL, internal::ComplexFDivINTEL)
 _SPIRV_OP_INTERNAL(MaskedGatherINTEL, internal::OpMaskedGatherINTEL)
diff --git a/lib/SPIRV/libSPIRV/SPIRVType.cpp b/lib/SPIRV/libSPIRV/SPIRVType.cpp
index 9f7aac0b2c..51779146c9 100644
--- a/lib/SPIRV/libSPIRV/SPIRVType.cpp
+++ b/lib/SPIRV/libSPIRV/SPIRVType.cpp
@@ -206,7 +206,8 @@ bool SPIRVType::isTypeStruct() const { return OpCode == OpTypeStruct; }
 bool SPIRVType::isTypeVector() const { return OpCode == OpTypeVector; }
 
 bool SPIRVType::isTypeJointMatrixINTEL() const {
-  return OpCode == internal::OpTypeJointMatrixINTEL;
+  return OpCode == internal::OpTypeJointMatrixINTEL ||
+         OpCode == internal::OpTypeJointMatrixINTELv2;
 }
 
 bool SPIRVType::isTypeCooperativeMatrixKHR() const {
@@ -290,13 +291,20 @@ void SPIRVTypeForwardPointer::decode(std::istream &I) {
 }
 
 SPIRVTypeJointMatrixINTEL::SPIRVTypeJointMatrixINTEL(
-    SPIRVModule *M, SPIRVId TheId, SPIRVType *CompType,
+    SPIRVModule *M, SPIRVId TheId, Op OC, SPIRVType *CompType,
     std::vector<SPIRVValue *> Args)
     : SPIRVType(M, FixedWC + Args.size(), OC, TheId), CompType(CompType),
-      Args(Args) {}
+      Args(std::move(Args)) {}
+
+SPIRVTypeJointMatrixINTEL::SPIRVTypeJointMatrixINTEL(
+    SPIRVModule *M, SPIRVId TheId, SPIRVType *CompType,
+    std::vector<SPIRVValue *> Args)
+    : SPIRVType(M, FixedWC + Args.size(), internal::OpTypeJointMatrixINTEL,
+                TheId),
+      CompType(CompType), Args(std::move(Args)) {}
 
 SPIRVTypeJointMatrixINTEL::SPIRVTypeJointMatrixINTEL()
-    : SPIRVType(OC), CompType(nullptr),
+    : SPIRVType(internal::OpTypeJointMatrixINTEL), CompType(nullptr),
       Args({nullptr, nullptr, nullptr, nullptr}) {}
 
 void SPIRVTypeJointMatrixINTEL::encode(spv_ostream &O) const {
@@ -329,4 +337,22 @@ void SPIRVTypeCooperativeMatrixKHR::decode(std::istream &I) {
   Decoder >> Id >> CompType >> Args;
 }
 
+void SPIRVTypeCooperativeMatrixKHR::validate() const {
+  SPIRVEntry::validate();
+  SPIRVErrorLog &SPVErrLog = this->getModule()->getErrorLog();
+  SPIRVConstant *UseConst = static_cast<SPIRVConstant *>(this->getUse());
+  auto InstName = OpCodeNameMap::map(OC);
+  uint64_t UseValue = UseConst->getZExtIntValue();
+  SPVErrLog.checkError(
+      (UseValue <= CooperativeMatrixUseMatrixAccumulatorKHR),
+      SPIRVEC_InvalidInstruction,
+      InstName + "\nIncorrect Use parameter, should be MatrixA, MatrixB or "
+                 "Accumulator\n");
+  SPIRVConstant *ScopeConst = static_cast<SPIRVConstant *>(this->getScope());
+  uint64_t ScopeValue = ScopeConst->getZExtIntValue();
+  SPVErrLog.checkError((ScopeValue <= ScopeInvocation),
+                       SPIRVEC_InvalidInstruction,
+                       InstName + "\nUnsupported Scope parameter\n");
+}
+
 } // namespace SPIRV
diff --git a/lib/SPIRV/libSPIRV/SPIRVType.h b/lib/SPIRV/libSPIRV/SPIRVType.h
index 9c9325d88b..1e4789d799 100644
--- a/lib/SPIRV/libSPIRV/SPIRVType.h
+++ b/lib/SPIRV/libSPIRV/SPIRVType.h
@@ -1062,13 +1062,18 @@ class SPIRVTypeTokenINTEL : public SPIRVType {
 };
 
 class SPIRVTypeJointMatrixINTEL : public SPIRVType {
+  Op OC;
   SPIRVType *CompType;
   std::vector<SPIRVValue *> Args;
 
 public:
-  const static Op OC = internal::OpTypeJointMatrixINTEL;
   const static SPIRVWord FixedWC = 3;
-  // Complete constructor
+  // Complete constructor with non-default OC
+  SPIRVTypeJointMatrixINTEL(SPIRVModule *M, SPIRVId TheId, Op OC,
+                            SPIRVType *CompType,
+                            std::vector<SPIRVValue *> Args);
+
+  // Incomplete constructor for default OC
   SPIRVTypeJointMatrixINTEL(SPIRVModule *M, SPIRVId TheId, SPIRVType *CompType,
                             std::vector<SPIRVValue *> Args);
   // Incomplete constructor
@@ -1087,15 +1092,43 @@ class SPIRVTypeJointMatrixINTEL : public SPIRVType {
   SPIRVType *getCompType() const { return CompType; }
   SPIRVValue *getRows() const { return Args[0]; }
   SPIRVValue *getColumns() const { return Args[1]; }
-  SPIRVValue *getLayout() const { return Args[2]; }
-  SPIRVValue *getScope() const { return Args[3]; }
-  SPIRVValue *getUse() const { return Args.size() > 4 ? Args[4] : nullptr; }
+
+  SPIRVValue *getLayout() const {
+    if (this->getOpCode() == internal::OpTypeJointMatrixINTEL)
+      return Args[2];
+    return nullptr;
+  }
+
+  SPIRVValue *getScope() const {
+    if (this->getOpCode() == internal::OpTypeJointMatrixINTEL)
+      return Args[3];
+    return Args[2];
+  }
+
+  SPIRVValue *getUse() const {
+    if (this->getOpCode() == internal::OpTypeJointMatrixINTEL)
+      return Args.size() > 4 ? Args[4] : nullptr;
+    return Args[3];
+  }
+
+  SPIRVValue *getComponentTypeInterpretation() const {
+    if (this->getOpCode() == internal::OpTypeJointMatrixINTEL)
+      return Args.size() > 5 ? Args[5] : nullptr;
+    return Args.size() > 4 ? Args[4] : nullptr;
+  }
+
+  std::vector<SPIRVEntry *> getNonLiteralOperands() const override {
+    return std::vector<SPIRVEntry *>(1, CompType);
+  }
 };
 
 class SPIRVTypeCooperativeMatrixKHR : public SPIRVType {
   SPIRVType *CompType;
   std::vector<SPIRVValue *> Args;
 
+protected:
+  void validate() const override;
+
 public:
   const static Op OC = OpTypeCooperativeMatrixKHR;
   const static SPIRVWord FixedWC = 7;
@@ -1118,6 +1151,10 @@ class SPIRVTypeCooperativeMatrixKHR : public SPIRVType {
   SPIRVValue *getRows() const { return Args[1]; }
   SPIRVValue *getColumns() const { return Args[2]; }
   SPIRVValue *getUse() const { return Args[3]; }
+
+  std::vector<SPIRVEntry *> getNonLiteralOperands() const override {
+    return std::vector<SPIRVEntry *>(1, CompType);
+  }
 };
 
 } // namespace SPIRV
diff --git a/lib/SPIRV/libSPIRV/spirv_internal.hpp b/lib/SPIRV/libSPIRV/spirv_internal.hpp
index b1f91405fd..07d75a2db1 100644
--- a/lib/SPIRV/libSPIRV/spirv_internal.hpp
+++ b/lib/SPIRV/libSPIRV/spirv_internal.hpp
@@ -68,6 +68,7 @@ enum InternalOp {
   IOpJointMatrixSUMadINTEL = 6128,
   IOpJointMatrixUSMadINTEL = 6129,
   IOpJointMatrixUUMadINTEL = 6130,
+  IOpTypeJointMatrixINTELv2 = 6184,
   IOpArithmeticFenceINTEL = 6145,
   IOpCooperativeMatrixLoadCheckedINTEL = 6193,
   IOpCooperativeMatrixStoreCheckedINTEL = 6194,
@@ -79,6 +80,7 @@ enum InternalOp {
   IOpMaskedGatherINTEL = 6428,
   IOpMaskedScatterINTEL = 6429,
   IOpJointMatrixGetElementCoordINTEL = 6440,
+  IOpCooperativeMatrixApplyFunctionINTEL = 6448,
   IOpCooperativeMatrixPrefetchINTEL = 6449,
   IOpConvertHandleToImageINTEL = 6529,
   IOpConvertHandleToSamplerINTEL = 6530,
@@ -113,6 +115,11 @@ enum InternalCapability {
   ICapabilityTensorFloat32RoundingINTEL = 6425,
   ICapabilityMaskedGatherScatterINTEL = 6427,
   ICapabilityJointMatrixWIInstructionsINTEL = 6435,
+  ICapabilityCooperativeMatrixInvocationInstructionsINTEL = 6435,
+  ICapabilityJointMatrixTF32ComponentTypeINTEL = 6436,
+  ICapabilityJointMatrixBF16ComponentTypeINTEL = 6437,
+  ICapabilityJointMatrixPackedInt2ComponentTypeINTEL = 6438,
+  ICapabilityJointMatrixPackedInt4ComponentTypeINTEL = 6439,
   ICapabilityCacheControlsINTEL = 6441,
   ICapRegisterLimitsINTEL = 6460,
   ICapabilityBindlessImagesINTEL = 6528
@@ -139,6 +146,14 @@ enum InternalJointMatrixLayout {
 
 enum InternalJointMatrixUse { MatrixA = 0, MatrixB = 1, Accumulator = 2 };
 
+enum InternalJointMatrixCTI {
+  None = 0,
+  TF32 = 1,
+  Bfloat16 = 2,
+  PackedInt2 = 3,
+  PackedInt4 = 4
+};
+
 enum InternalBuiltIn {
   IBuiltInSubDeviceIDINTEL = 6135,
   IBuiltInGlobalHWThreadIDINTEL = 6136,
@@ -162,7 +177,12 @@ enum class StoreCacheControlINTEL {
 #define _SPIRV_OP(x, y) constexpr x x##y = static_cast<x>(I##x##y);
 _SPIRV_OP(Capability, JointMatrixINTEL)
 _SPIRV_OP(Capability, JointMatrixWIInstructionsINTEL)
+_SPIRV_OP(Capability, JointMatrixTF32ComponentTypeINTEL)
+_SPIRV_OP(Capability, JointMatrixBF16ComponentTypeINTEL)
+_SPIRV_OP(Capability, JointMatrixPackedInt2ComponentTypeINTEL)
+_SPIRV_OP(Capability, JointMatrixPackedInt4ComponentTypeINTEL)
 _SPIRV_OP(Op, TypeJointMatrixINTEL)
+_SPIRV_OP(Op, TypeJointMatrixINTELv2)
 _SPIRV_OP(Op, JointMatrixLoadINTEL)
 _SPIRV_OP(Op, JointMatrixStoreINTEL)
 _SPIRV_OP(Op, JointMatrixMadINTEL)
@@ -179,6 +199,9 @@ _SPIRV_OP(Op, CooperativeMatrixConstructCheckedINTEL)
 _SPIRV_OP(Capability, CooperativeMatrixPrefetchINTEL)
 _SPIRV_OP(Op, CooperativeMatrixPrefetchINTEL)
 
+_SPIRV_OP(Capability, CooperativeMatrixInvocationInstructionsINTEL)
+_SPIRV_OP(Op, CooperativeMatrixApplyFunctionINTEL)
+
 _SPIRV_OP(Capability, HWThreadQueryINTEL)
 _SPIRV_OP(BuiltIn, SubDeviceIDINTEL)
 _SPIRV_OP(BuiltIn, GlobalHWThreadIDINTEL)
diff --git a/test/extensions/INTEL/SPV_INTEL_joint_matrix/array_of_matrices.ll b/test/extensions/INTEL/SPV_INTEL_joint_matrix/array_of_matrices.ll
new file mode 100644
index 0000000000..0571af5dd9
--- /dev/null
+++ b/test/extensions/INTEL/SPV_INTEL_joint_matrix/array_of_matrices.ll
@@ -0,0 +1,436 @@
+;; Compiled from joint_matrix_bf16_fill_k_cache.cpp from https://github.com/intel/llvm
+;; command: clang++ -fsycl -DSYCL_EXT_ONEAPI_MATRIX_VERSION=4 llvm/sycl/test-e2e/Matrix/joint_matrix_bf16_fill_k_cache.cpp -fsycl-device-only -o test.bc
+
+; RUN: llvm-as < %s -o %t.bc
+; RUN: llvm-spirv %t.bc --spirv-ext=+SPV_INTEL_joint_matrix -o %t.spv
+; RUN: llvm-spirv %t.spv -to-text -o %t.spt
+; RUN: FileCheck < %t.spt %s --check-prefix=CHECK-SPIRV
+
+; RUN: llvm-spirv -r %t.spv -o %t.rev.bc
+; RUN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM
+
+; CHECK-SPIRV-DAG: Capability JointMatrixINTEL
+; CHECK-SPIRV-DAG: Extension "SPV_INTEL_joint_matrix"
+; CHECK-SPIRV: TypeInt [[#Int16Ty:]] 16 0
+; CHECK-SPIRV: TypeFloat [[#FloatTy:]] 32
+; CHECK-SPIRV: TypeJointMatrixINTEL [[#MatTy1:]] [[#FloatTy]]
+; CHECK-SPIRV: TypeStruct [[#StructTy1:]] [[#MatTy1]]
+; CHECK-SPIRV: TypeArray [[#ArrayTy1:]] [[#StructTy1]] [[#]]
+; CHECK-SPIRV: TypeArray [[#]] [[#ArrayTy1]] [[#]]
+; CHECK-SPIRV: TypeJointMatrixINTEL [[#MatTy2:]] [[#Int16Ty]]
+; CHECK-SPIRV: TypeStruct [[#StructTy2:]] [[#MatTy2]]
+; CHECK-SPIRV: TypeArray [[#ArrayTy2:]] [[#StructTy2]] [[#]]
+; CHECK-SPIRV: TypeArray [[#]] [[#ArrayTy2]] [[#]]
+; CHECK-SPIRV: TypeJointMatrixINTEL [[#MatTy3:]] [[#Int16Ty]]
+; CHECK-SPIRV: TypeStruct [[#StructTy3:]] [[#MatTy3]]
+; CHECK-SPIRV: TypeArray [[#ArrayTy3:]] [[#StructTy3]] [[#]]
+; CHECK-SPIRV: TypeArray [[#]] [[#ArrayTy3]] [[#]]
+
+; CHECK-LLVM: %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix" = type { target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) }
+; CHECK-LLVM: %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix.5" = type { target("spirv.JointMatrixINTEL", i16, 8, 16, 0, 3, 0, 1) }
+; CHECK-LLVM: %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix.6" = type { target("spirv.JointMatrixINTEL", i16, 16, 16, 2, 3, 1, 1) }
+; CHECK-LLVM: alloca [4 x [4 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix"]]
+; CHECK-LLVM: alloca [4 x [2 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix.5"]]
+; CHECK-LLVM: alloca [4 x [2 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix.6"]]
+
+; ModuleID = 'test.bc'
+source_filename = "llvm/sycl/test-e2e/Matrix/joint_matrix_bf16_fill_k_cache.cpp"
+target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-n8:16:32:64"
+target triple = "spir64-unknown-unknown"
+
+%"class.sycl::_V1::__generated_multi_ptr" = type { ptr addrspace(1) }
+%"class.sycl::_V1::__generated_multi_ptr.0" = type { ptr addrspace(1) }
+%"class.sycl::_V1::__generated_multi_ptr.1" = type { ptr addrspace(1) }
+%"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix" = type { target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) }
+%"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix.5" = type { target("spirv.JointMatrixINTEL", i16, 8, 16, 0, 3, 0, 1) }
+%"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix.6" = type { target("spirv.JointMatrixINTEL", i16, 16, 16, 2, 3, 1, 1) }
+%"class.sycl::_V1::ext::oneapi::bfloat16" = type { i16 }
+
+$_ZTSZZ12joint_matmulILj256ELj256ELj256ELj256ELj2EN4sycl3_V13ext6oneapi8bfloat16EfLj16EEdPT4_S6_PT5_RNS1_5queueEiENKUlRNS1_7handlerEE_clESC_EUlNS1_7nd_itemILi2EEEE_ = comdat any
+
+@__spirv_BuiltInWorkgroupId = external dso_local local_unnamed_addr addrspace(1) constant <3 x i64>, align 32
+@__spirv_BuiltInLocalInvocationId = external dso_local local_unnamed_addr addrspace(1) constant <3 x i64>, align 32
+
+; Function Attrs: convergent norecurse nounwind
+define weak_odr dso_local spir_kernel void @_ZTSZZ12joint_matmulILj256ELj256ELj256ELj256ELj2EN4sycl3_V13ext6oneapi8bfloat16EfLj16EEdPT4_S6_PT5_RNS1_5queueEiENKUlRNS1_7handlerEE_clESC_EUlNS1_7nd_itemILi2EEEE_(ptr noundef byval(%"class.sycl::_V1::__generated_multi_ptr") align 8 %_arg_pA, ptr noundef byval(%"class.sycl::_V1::__generated_multi_ptr.0") align 8 %_arg_pB, ptr noundef byval(%"class.sycl::_V1::__generated_multi_ptr.1") align 8 %_arg_pC) local_unnamed_addr #0 comdat !srcloc !59 !kernel_arg_buffer_location !60 !intel_reqd_sub_group_size !61 !sycl_fixed_targets !62 !sycl_kernel_omit_args !63 {
+entry:
+  call void @__itt_offload_wi_start_wrapper()
+  %tC.i = alloca [4 x [4 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix"]], align 8
+  %tA.i = alloca [4 x [2 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix.5"]], align 8
+  %tB.i = alloca [4 x [2 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix.6"]], align 8
+  %0 = load i64, ptr %_arg_pA, align 8, !tbaa !64
+  %1 = inttoptr i64 %0 to ptr addrspace(1)
+  %2 = load i64, ptr %_arg_pB, align 8, !tbaa !64
+  %3 = inttoptr i64 %2 to ptr addrspace(1)
+  %4 = load i64, ptr %_arg_pC, align 8, !tbaa !64
+  %5 = inttoptr i64 %4 to ptr addrspace(1)
+  %6 = load i64, ptr addrspace(1) getelementptr inbounds (i8, ptr addrspace(1) @__spirv_BuiltInWorkgroupId, i64 8), align 8, !noalias !68
+  %7 = load i64, ptr addrspace(1) @__spirv_BuiltInWorkgroupId, align 32, !noalias !68
+  %8 = load i64, ptr addrspace(1) getelementptr inbounds (i8, ptr addrspace(1) @__spirv_BuiltInLocalInvocationId, i64 8), align 8, !noalias !75
+  %9 = load i64, ptr addrspace(1) @__spirv_BuiltInLocalInvocationId, align 32, !noalias !75
+  %cmp.i.i = icmp ult i64 %6, 2147483648
+  tail call void @llvm.assume(i1 %cmp.i.i)
+  %cmp.i208.i = icmp ult i64 %7, 2147483648
+  tail call void @llvm.assume(i1 %cmp.i208.i)
+  %cmp.i209.i = icmp ult i64 %8, 2147483648
+  tail call void @llvm.assume(i1 %cmp.i209.i)
+  %cmp.i212.i = icmp ult i64 %9, 2147483648
+  tail call void @llvm.assume(i1 %cmp.i212.i)
+  %div205.i = lshr i64 %9, 4
+  call void @llvm.lifetime.start.p0(i64 128, ptr nonnull %tC.i) #4
+  br label %arrayctor.loop.i
+
+arrayctor.loop.i:                                 ; preds = %arrayctor.loop.i, %entry
+  %arrayctor.cur.idx.i = phi i64 [ 0, %entry ], [ %arrayctor.cur.add.i, %arrayctor.loop.i ]
+  %arrayctor.cur.add.i = add nuw nsw i64 %arrayctor.cur.idx.i, 1
+  %arrayctor.done.i = icmp eq i64 %arrayctor.cur.add.i, 16
+  br i1 %arrayctor.done.i, label %for.cond.i, label %arrayctor.loop.i
+
+for.cond.i:                                       ; preds = %arrayctor.loop.i, %for.cond.cleanup7.i
+  %m.0.i = phi i32 [ %inc12.i, %for.cond.cleanup7.i ], [ 0, %arrayctor.loop.i ]
+  %cmp.i = icmp ult i32 %m.0.i, 4
+  br i1 %cmp.i, label %for.cond5.preheader.i, label %for.cond14.preheader.i
+
+for.cond5.preheader.i:                            ; preds = %for.cond.i
+  %idxprom.i = zext i32 %m.0.i to i64
+  br label %for.cond5.i
+
+for.cond14.preheader.i:                           ; preds = %for.cond.i
+  %mul50.i = shl nuw nsw i64 %6, 8
+  %mul51.i = shl nuw nsw i64 %8, 5
+  %add52.i = add nuw nsw i64 %mul50.i, %mul51.i
+  %mul80.i = shl nuw nsw i64 %div205.i, 7
+  %10 = shl nuw nsw i64 %7, 9
+  %11 = add nuw nsw i64 %10, %mul80.i
+  br label %for.cond14.i
+
+for.cond5.i:                                      ; preds = %for.body8.i, %for.cond5.preheader.i
+  %n.0.i = phi i32 [ %inc.i, %for.body8.i ], [ 0, %for.cond5.preheader.i ]
+  %cmp6.i = icmp ult i32 %n.0.i, 4
+  br i1 %cmp6.i, label %for.body8.i, label %for.cond.cleanup7.i
+
+for.cond.cleanup7.i:                              ; preds = %for.cond5.i
+  %inc12.i = add nuw nsw i32 %m.0.i, 1
+  br label %for.cond.i, !llvm.loop !80
+
+for.body8.i:                                      ; preds = %for.cond5.i
+  %conv.i = zext i32 %n.0.i to i64
+  %arrayidx10.i = getelementptr inbounds [4 x [4 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix"]], ptr %tC.i, i64 0, i64 %idxprom.i, i64 %conv.i
+  %call.i.i = tail call spir_func noundef target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) @_Z26__spirv_CompositeConstructIffLm8ELm16ELN5__spv9MatrixUseE2ELNS0_12MatrixLayoutE3ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEET_(float noundef 0.000000e+00) #5
+  store target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) %call.i.i, ptr %arrayidx10.i, align 8, !tbaa !82
+  %inc.i = add nuw nsw i32 %n.0.i, 1
+  br label %for.cond5.i, !llvm.loop !84
+
+for.cond14.i:                                     ; preds = %for.cond.cleanup34.i, %for.cond14.preheader.i
+  %k2.0.i = phi i32 [ %inc129.i, %for.cond.cleanup34.i ], [ 0, %for.cond14.preheader.i ]
+  %cmp15.i = icmp ult i32 %k2.0.i, 8
+  br i1 %cmp15.i, label %for.body17.i, label %for.cond132.preheader.i
+
+for.cond132.preheader.i:                          ; preds = %for.cond14.i
+  %mul156.i = shl nuw nsw i64 %7, 8
+  %mul157.i = shl nuw nsw i64 %div205.i, 6
+  %add158.i = add nuw nsw i64 %mul156.i, %mul157.i
+  br label %for.cond132.i
+
+for.body17.i:                                     ; preds = %for.cond14.i
+  call void @llvm.lifetime.start.p0(i64 64, ptr nonnull %tA.i) #4
+  br label %arrayctor.loop20.i
+
+arrayctor.loop20.i:                               ; preds = %arrayctor.loop20.i, %for.body17.i
+  %arrayctor.cur21.idx.i = phi i64 [ 0, %for.body17.i ], [ %arrayctor.cur21.add.i, %arrayctor.loop20.i ]
+  %arrayctor.cur21.add.i = add nuw nsw i64 %arrayctor.cur21.idx.i, 1
+  %arrayctor.done23.i = icmp eq i64 %arrayctor.cur21.add.i, 8
+  br i1 %arrayctor.done23.i, label %arrayctor.cont24.i, label %arrayctor.loop20.i
+
+arrayctor.cont24.i:                               ; preds = %arrayctor.loop20.i
+  call void @llvm.lifetime.start.p0(i64 64, ptr nonnull %tB.i) #4
+  br label %arrayctor.loop27.i
+
+arrayctor.loop27.i:                               ; preds = %arrayctor.loop27.i, %arrayctor.cont24.i
+  %arrayctor.cur28.idx.i = phi i64 [ 0, %arrayctor.cont24.i ], [ %arrayctor.cur28.add.i, %arrayctor.loop27.i ]
+  %arrayctor.cur28.add.i = add nuw nsw i64 %arrayctor.cur28.idx.i, 1
+  %arrayctor.done30.i = icmp eq i64 %arrayctor.cur28.add.i, 8
+  br i1 %arrayctor.done30.i, label %for.cond32.preheader.i, label %arrayctor.loop27.i
+
+for.cond32.preheader.i:                           ; preds = %arrayctor.loop27.i
+  %12 = shl nuw i32 %k2.0.i, 1
+  br label %for.cond32.i
+
+for.cond32.i:                                     ; preds = %for.cond.cleanup92.i, %for.cond32.preheader.i
+  %k1.0.i = phi i32 [ %inc126.i, %for.cond.cleanup92.i ], [ 0, %for.cond32.preheader.i ]
+  %cmp33.i = icmp ult i32 %k1.0.i, 2
+  br i1 %cmp33.i, label %for.body35.i, label %for.cond.cleanup34.i
+
+for.cond.cleanup34.i:                             ; preds = %for.cond32.i
+  call void @llvm.lifetime.end.p0(i64 64, ptr nonnull %tB.i) #4
+  call void @llvm.lifetime.end.p0(i64 64, ptr nonnull %tA.i) #4
+  %inc129.i = add nuw nsw i32 %k2.0.i, 1
+  br label %for.cond14.i, !llvm.loop !85
+
+for.body35.i:                                     ; preds = %for.cond32.i
+  %13 = add nuw i32 %12, %k1.0.i
+  %div37206.i = and i32 %13, 268435455
+  %idxprom46.i = zext i32 %k1.0.i to i64
+  %mul57.i = shl nuw nsw i32 %div37206.i, 4
+  %conv58.i = zext i32 %mul57.i to i64
+  %invariant.gep = getelementptr inbounds %"class.sycl::_V1::ext::oneapi::bfloat16", ptr addrspace(1) %1, i64 %conv58.i
+  br label %for.cond39.i
+
+for.cond39.i:                                     ; preds = %for.body42.i, %for.body35.i
+  %m38.0.i = phi i32 [ 0, %for.body35.i ], [ %inc60.i, %for.body42.i ]
+  %cmp40.i = icmp ult i32 %m38.0.i, 4
+  br i1 %cmp40.i, label %for.body42.i, label %for.cond63.preheader.i
+
+for.cond63.preheader.i:                           ; preds = %for.cond39.i
+  %mul77.i = shl nuw nsw i32 %div37206.i, 12
+  %conv78.i = zext i32 %mul77.i to i64
+  %add.ptr.i225.i = getelementptr inbounds %"class.sycl::_V1::ext::oneapi::bfloat16", ptr addrspace(1) %3, i64 %conv78.i
+  br label %for.cond63.i
+
+for.body42.i:                                     ; preds = %for.cond39.i
+  %idxprom44.i = zext i32 %m38.0.i to i64
+  %arrayidx47.i = getelementptr inbounds [4 x [2 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix.5"]], ptr %tA.i, i64 0, i64 %idxprom44.i, i64 %idxprom46.i
+  %mul53.i = shl nuw nsw i32 %m38.0.i, 3
+  %conv54.i = zext i32 %mul53.i to i64
+  %add55.i = add nuw nsw i64 %add52.i, %conv54.i
+  %mul56.i = shl nuw nsw i64 %add55.i, 8
+  %gep = getelementptr inbounds %"class.sycl::_V1::ext::oneapi::bfloat16", ptr addrspace(1) %invariant.gep, i64 %mul56.i
+  %call1.i.i = tail call spir_func noundef target("spirv.JointMatrixINTEL", i16, 8, 16, 0, 3, 0, 1) @_Z28__spirv_JointMatrixLoadINTELIU3AS1N4sycl3_V13ext6oneapi8bfloat16ES4_Lm8ELm16ELN5__spv9MatrixUseE0ELNS6_12MatrixLayoutE0ELNS6_5Scope4FlagE3EEPNS6_24__spirv_JointMatrixINTELIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEPT_mS8_SA_i(ptr addrspace(1) noundef %gep, i64 noundef 256, i32 noundef 0, i32 noundef 3, i32 noundef 0) #5
+  store target("spirv.JointMatrixINTEL", i16, 8, 16, 0, 3, 0, 1) %call1.i.i, ptr %arrayidx47.i, align 8, !tbaa !86
+  %inc60.i = add nuw nsw i32 %m38.0.i, 1
+  br label %for.cond39.i, !llvm.loop !88
+
+for.cond63.i:                                     ; preds = %for.body67.i, %for.cond63.preheader.i
+  %n62.0.i = phi i32 [ %inc87.i, %for.body67.i ], [ 0, %for.cond63.preheader.i ]
+  %cmp65.i = icmp ult i32 %n62.0.i, 4
+  br i1 %cmp65.i, label %for.body67.i, label %for.cond90.i
+
+for.body67.i:                                     ; preds = %for.cond63.i
+  %conv64.i = zext i32 %n62.0.i to i64
+  %arrayidx72.i = getelementptr inbounds [4 x [2 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix.6"]], ptr %tB.i, i64 0, i64 %conv64.i, i64 %idxprom46.i
+  %14 = shl nuw nsw i64 %conv64.i, 5
+  %mul85.i = add nuw nsw i64 %14, %11
+  %add.ptr.i226.i = getelementptr inbounds %"class.sycl::_V1::ext::oneapi::bfloat16", ptr addrspace(1) %add.ptr.i225.i, i64 %mul85.i
+  %call1.i219.i = tail call spir_func noundef target("spirv.JointMatrixINTEL", i16, 16, 16, 2, 3, 1, 1) @_Z28__spirv_JointMatrixLoadINTELIU3AS1N4sycl3_V13ext6oneapi8bfloat16ES4_Lm16ELm16ELN5__spv9MatrixUseE1ELNS6_12MatrixLayoutE2ELNS6_5Scope4FlagE3EEPNS6_24__spirv_JointMatrixINTELIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEPT_mS8_SA_i(ptr addrspace(1) noundef %add.ptr.i226.i, i64 noundef 512, i32 noundef 2, i32 noundef 3, i32 noundef 0) #5
+  store target("spirv.JointMatrixINTEL", i16, 16, 16, 2, 3, 1, 1) %call1.i219.i, ptr %arrayidx72.i, align 8, !tbaa !89
+  %inc87.i = add nuw nsw i32 %n62.0.i, 1
+  br label %for.cond63.i, !llvm.loop !91
+
+for.cond90.i:                                     ; preds = %for.cond63.i, %for.cond.cleanup98.i
+  %m89.0.i = phi i32 [ %inc123.i, %for.cond.cleanup98.i ], [ 0, %for.cond63.i ]
+  %cmp91.i = icmp ult i32 %m89.0.i, 4
+  br i1 %cmp91.i, label %for.cond95.preheader.i, label %for.cond.cleanup92.i
+
+for.cond95.preheader.i:                           ; preds = %for.cond90.i
+  %idxprom102.i = zext i32 %m89.0.i to i64
+  %arrayidx105.i = getelementptr inbounds [4 x [2 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix.5"]], ptr %tA.i, i64 0, i64 %idxprom102.i, i64 %idxprom46.i
+  %15 = load target("spirv.JointMatrixINTEL", i16, 8, 16, 0, 3, 0, 1), ptr %arrayidx105.i, align 8, !tbaa !86, !noalias !92
+  br label %for.cond95.i
+
+for.cond.cleanup92.i:                             ; preds = %for.cond90.i
+  %inc126.i = add nuw nsw i32 %k1.0.i, 1
+  br label %for.cond32.i, !llvm.loop !95
+
+for.cond95.i:                                     ; preds = %for.body99.i, %for.cond95.preheader.i
+  %n94.0.i = phi i32 [ %inc120.i, %for.body99.i ], [ 0, %for.cond95.preheader.i ]
+  %cmp97.i = icmp ult i32 %n94.0.i, 4
+  br i1 %cmp97.i, label %for.body99.i, label %for.cond.cleanup98.i
+
+for.cond.cleanup98.i:                             ; preds = %for.cond95.i
+  %inc123.i = add nuw nsw i32 %m89.0.i, 1
+  br label %for.cond90.i, !llvm.loop !96
+
+for.body99.i:                                     ; preds = %for.cond95.i
+  %conv96.i = zext i32 %n94.0.i to i64
+  %arrayidx109.i = getelementptr inbounds [4 x [2 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix.6"]], ptr %tB.i, i64 0, i64 %conv96.i, i64 %idxprom46.i
+  %arrayidx113.i = getelementptr inbounds [4 x [4 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix"]], ptr %tC.i, i64 0, i64 %idxprom102.i, i64 %conv96.i
+  %16 = load target("spirv.JointMatrixINTEL", i16, 16, 16, 2, 3, 1, 1), ptr %arrayidx109.i, align 8, !tbaa !89, !noalias !92
+  %17 = load target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2), ptr %arrayidx113.i, align 8, !tbaa !82, !noalias !92
+  %call.i221.i = tail call spir_func noundef target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) @_Z27__spirv_JointMatrixMadINTELIN4sycl3_V13ext6oneapi8bfloat16EfLm8ELm16ELm16ELN5__spv9MatrixUseE0ELS6_1ELS6_2ELNS5_12MatrixLayoutE0ELS7_2ELS7_3ELNS5_5Scope4FlagE3EEPNS5_24__spirv_JointMatrixINTELIT0_XT1_EXT3_EXT9_EXT10_EXT6_EEEPNSA_IT_XT1_EXT2_EXT7_EXT10_EXT4_EEEPNSA_ISE_XT2_EXT3_EXT8_EXT10_EXT5_EEESD_S9_(target("spirv.JointMatrixINTEL", i16, 8, 16, 0, 3, 0, 1) noundef %15, target("spirv.JointMatrixINTEL", i16, 16, 16, 2, 3, 1, 1) noundef %16, target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) noundef %17, i32 noundef 3) #5, !noalias !92
+  store target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) %call.i221.i, ptr %arrayidx113.i, align 8, !tbaa !82
+  %inc120.i = add nuw nsw i32 %n94.0.i, 1
+  br label %for.cond95.i, !llvm.loop !97
+
+for.cond132.i:                                    ; preds = %for.cond.cleanup140.i, %for.cond132.preheader.i
+  %m131.0.i = phi i32 [ %inc166.i, %for.cond.cleanup140.i ], [ 0, %for.cond132.preheader.i ]
+  %cmp133.i = icmp ult i32 %m131.0.i, 4
+  br i1 %cmp133.i, label %for.cond137.preheader.i, label %_ZZZ12joint_matmulILj256ELj256ELj256ELj256ELj2EN4sycl3_V13ext6oneapi8bfloat16EfLj16EEdPT4_S6_PT5_RNS1_5queueEiENKUlRNS1_7handlerEE_clESC_ENKUlNS1_7nd_itemILi2EEEE_clESF_.exit
+
+for.cond137.preheader.i:                          ; preds = %for.cond132.i
+  %idxprom143.i = zext i32 %m131.0.i to i64
+  %mul152.i = shl nuw nsw i32 %m131.0.i, 3
+  %conv153.i = zext i32 %mul152.i to i64
+  %add154.i = add nuw nsw i64 %add52.i, %conv153.i
+  %mul155.i = shl nuw nsw i64 %add154.i, 8
+  %add.ptr.i227.i = getelementptr inbounds float, ptr addrspace(1) %5, i64 %mul155.i
+  br label %for.cond137.i
+
+for.cond137.i:                                    ; preds = %for.body141.i, %for.cond137.preheader.i
+  %n136.0.i = phi i32 [ %inc163.i, %for.body141.i ], [ 0, %for.cond137.preheader.i ]
+  %cmp139.i = icmp ult i32 %n136.0.i, 4
+  br i1 %cmp139.i, label %for.body141.i, label %for.cond.cleanup140.i
+
+for.cond.cleanup140.i:                            ; preds = %for.cond137.i
+  %inc166.i = add nuw nsw i32 %m131.0.i, 1
+  br label %for.cond132.i, !llvm.loop !98
+
+for.body141.i:                                    ; preds = %for.cond137.i
+  %conv138.i = zext i32 %n136.0.i to i64
+  %arrayidx146.i = getelementptr inbounds [4 x [4 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix"]], ptr %tC.i, i64 0, i64 %idxprom143.i, i64 %conv138.i
+  %mul160.i = shl nuw nsw i64 %conv138.i, 4
+  %add161.i = add nuw nsw i64 %add158.i, %mul160.i
+  %add.ptr.i228.i = getelementptr inbounds float, ptr addrspace(1) %add.ptr.i227.i, i64 %add161.i
+  %18 = load target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2), ptr %arrayidx146.i, align 8, !tbaa !82
+  tail call spir_func void @_Z29__spirv_JointMatrixStoreINTELIU3AS1ffLm8ELm16ELN5__spv9MatrixUseE2ELNS1_12MatrixLayoutE3ELNS1_5Scope4FlagE3EEvPT_PNS1_24__spirv_JointMatrixINTELIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEmS3_S5_i(ptr addrspace(1) noundef %add.ptr.i228.i, target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) noundef %18, i64 noundef 256, i32 noundef 0, i32 noundef 3, i32 noundef 0) #5
+  %inc163.i = add nuw nsw i32 %n136.0.i, 1
+  br label %for.cond137.i, !llvm.loop !99
+
+_ZZZ12joint_matmulILj256ELj256ELj256ELj256ELj2EN4sycl3_V13ext6oneapi8bfloat16EfLj16EEdPT4_S6_PT5_RNS1_5queueEiENKUlRNS1_7handlerEE_clESC_ENKUlNS1_7nd_itemILi2EEEE_clESF_.exit: ; preds = %for.cond132.i
+  call void @llvm.lifetime.end.p0(i64 128, ptr nonnull %tC.i) #4
+  call void @__itt_offload_wi_finish_wrapper()
+  ret void
+}
+
+; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
+declare void @llvm.lifetime.start.p0(i64 immarg, ptr nocapture) #1
+
+; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
+declare void @llvm.lifetime.end.p0(i64 immarg, ptr nocapture) #1
+
+; Function Attrs: nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: write)
+declare void @llvm.assume(i1 noundef) #2
+
+; Function Attrs: convergent nounwind
+declare dso_local spir_func noundef target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) @_Z26__spirv_CompositeConstructIffLm8ELm16ELN5__spv9MatrixUseE2ELNS0_12MatrixLayoutE3ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEET_(float noundef) local_unnamed_addr #3
+
+; Function Attrs: convergent nounwind
+declare dso_local spir_func noundef target("spirv.JointMatrixINTEL", i16, 8, 16, 0, 3, 0, 1) @_Z28__spirv_JointMatrixLoadINTELIU3AS1N4sycl3_V13ext6oneapi8bfloat16ES4_Lm8ELm16ELN5__spv9MatrixUseE0ELNS6_12MatrixLayoutE0ELNS6_5Scope4FlagE3EEPNS6_24__spirv_JointMatrixINTELIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEPT_mS8_SA_i(ptr addrspace(1) noundef, i64 noundef, i32 noundef, i32 noundef, i32 noundef) local_unnamed_addr #3
+
+; Function Attrs: convergent nounwind
+declare dso_local spir_func noundef target("spirv.JointMatrixINTEL", i16, 16, 16, 2, 3, 1, 1) @_Z28__spirv_JointMatrixLoadINTELIU3AS1N4sycl3_V13ext6oneapi8bfloat16ES4_Lm16ELm16ELN5__spv9MatrixUseE1ELNS6_12MatrixLayoutE2ELNS6_5Scope4FlagE3EEPNS6_24__spirv_JointMatrixINTELIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEPT_mS8_SA_i(ptr addrspace(1) noundef, i64 noundef, i32 noundef, i32 noundef, i32 noundef) local_unnamed_addr #3
+
+; Function Attrs: convergent nounwind
+declare dso_local spir_func noundef target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) @_Z27__spirv_JointMatrixMadINTELIN4sycl3_V13ext6oneapi8bfloat16EfLm8ELm16ELm16ELN5__spv9MatrixUseE0ELS6_1ELS6_2ELNS5_12MatrixLayoutE0ELS7_2ELS7_3ELNS5_5Scope4FlagE3EEPNS5_24__spirv_JointMatrixINTELIT0_XT1_EXT3_EXT9_EXT10_EXT6_EEEPNSA_IT_XT1_EXT2_EXT7_EXT10_EXT4_EEEPNSA_ISE_XT2_EXT3_EXT8_EXT10_EXT5_EEESD_S9_(target("spirv.JointMatrixINTEL", i16, 8, 16, 0, 3, 0, 1) noundef, target("spirv.JointMatrixINTEL", i16, 16, 16, 2, 3, 1, 1) noundef, target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) noundef, i32 noundef) local_unnamed_addr #3
+
+; Function Attrs: convergent nounwind
+declare dso_local spir_func void @_Z29__spirv_JointMatrixStoreINTELIU3AS1ffLm8ELm16ELN5__spv9MatrixUseE2ELNS1_12MatrixLayoutE3ELNS1_5Scope4FlagE3EEvPT_PNS1_24__spirv_JointMatrixINTELIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEmS3_S5_i(ptr addrspace(1) noundef, target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) noundef, i64 noundef, i32 noundef, i32 noundef, i32 noundef) local_unnamed_addr #3
+
+declare dso_local spir_func i32 @_Z18__spirv_ocl_printfPU3AS2Kcz(ptr addrspace(2), ...)
+
+declare void @__itt_offload_wi_start_wrapper()
+
+declare void @__itt_offload_wi_finish_wrapper()
+
+attributes #0 = { convergent norecurse nounwind "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "sycl-module-id"="llvm/sycl/test-e2e/Matrix/joint_matrix_bf16_fill_k_cache.cpp" "sycl-optlevel"="2" "uniform-work-group-size"="true" }
+attributes #1 = { nocallback nofree nosync nounwind willreturn memory(argmem: readwrite) }
+attributes #2 = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: write) }
+attributes #3 = { convergent nounwind "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #4 = { nounwind }
+attributes #5 = { convergent nounwind }
+
+!llvm.module.flags = !{!0, !1}
+!opencl.spir.version = !{!2}
+!spirv.Source = !{!3}
+!sycl_aspects = !{!4, !5, !6, !7, !8, !9, !10, !11, !12, !13, !14, !15, !16, !17, !18, !19, !20, !21, !22, !23, !24, !25, !26, !27, !28, !29, !30, !31, !32, !33, !34, !35, !36, !37, !38, !39, !40, !41, !42, !43, !44, !45, !46, !47, !48, !49, !50, !51, !52, !53, !54, !55, !56, !57}
+!llvm.ident = !{!58}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{i32 7, !"frame-pointer", i32 2}
+!2 = !{i32 1, i32 2}
+!3 = !{i32 4, i32 100000}
+!4 = !{!"cpu", i32 1}
+!5 = !{!"gpu", i32 2}
+!6 = !{!"accelerator", i32 3}
+!7 = !{!"custom", i32 4}
+!8 = !{!"fp16", i32 5}
+!9 = !{!"fp64", i32 6}
+!10 = !{!"image", i32 9}
+!11 = !{!"online_compiler", i32 10}
+!12 = !{!"online_linker", i32 11}
+!13 = !{!"queue_profiling", i32 12}
+!14 = !{!"usm_device_allocations", i32 13}
+!15 = !{!"usm_host_allocations", i32 14}
+!16 = !{!"usm_shared_allocations", i32 15}
+!17 = !{!"usm_system_allocations", i32 17}
+!18 = !{!"ext_intel_pci_address", i32 18}
+!19 = !{!"ext_intel_gpu_eu_count", i32 19}
+!20 = !{!"ext_intel_gpu_eu_simd_width", i32 20}
+!21 = !{!"ext_intel_gpu_slices", i32 21}
+!22 = !{!"ext_intel_gpu_subslices_per_slice", i32 22}
+!23 = !{!"ext_intel_gpu_eu_count_per_subslice", i32 23}
+!24 = !{!"ext_intel_max_mem_bandwidth", i32 24}
+!25 = !{!"ext_intel_mem_channel", i32 25}
+!26 = !{!"usm_atomic_host_allocations", i32 26}
+!27 = !{!"usm_atomic_shared_allocations", i32 27}
+!28 = !{!"atomic64", i32 28}
+!29 = !{!"ext_intel_device_info_uuid", i32 29}
+!30 = !{!"ext_oneapi_srgb", i32 30}
+!31 = !{!"ext_oneapi_native_assert", i32 31}
+!32 = !{!"host_debuggable", i32 32}
+!33 = !{!"ext_intel_gpu_hw_threads_per_eu", i32 33}
+!34 = !{!"ext_oneapi_cuda_async_barrier", i32 34}
+!35 = !{!"ext_oneapi_bfloat16_math_functions", i32 35}
+!36 = !{!"ext_intel_free_memory", i32 36}
+!37 = !{!"ext_intel_device_id", i32 37}
+!38 = !{!"ext_intel_memory_clock_rate", i32 38}
+!39 = !{!"ext_intel_memory_bus_width", i32 39}
+!40 = !{!"emulated", i32 40}
+!41 = !{!"ext_intel_legacy_image", i32 41}
+!42 = !{!"ext_oneapi_bindless_images", i32 42}
+!43 = !{!"ext_oneapi_bindless_images_shared_usm", i32 43}
+!44 = !{!"ext_oneapi_bindless_images_1d_usm", i32 44}
+!45 = !{!"ext_oneapi_bindless_images_2d_usm", i32 45}
+!46 = !{!"ext_oneapi_interop_memory_import", i32 46}
+!47 = !{!"ext_oneapi_interop_memory_export", i32 47}
+!48 = !{!"ext_oneapi_interop_semaphore_import", i32 48}
+!49 = !{!"ext_oneapi_interop_semaphore_export", i32 49}
+!50 = !{!"ext_oneapi_mipmap", i32 50}
+!51 = !{!"ext_oneapi_mipmap_anisotropy", i32 51}
+!52 = !{!"ext_oneapi_mipmap_level_reference", i32 52}
+!53 = !{!"int64_base_atomics", i32 7}
+!54 = !{!"int64_extended_atomics", i32 8}
+!55 = !{!"usm_system_allocator", i32 17}
+!56 = !{!"usm_restricted_shared_allocations", i32 16}
+!57 = !{!"host", i32 0}
+!58 = !{!"clang version 18.0.0 (https://github.com/intel/llvm.git cc440821c30daabef517c7c8ff75546719f8094c)"}
+!59 = !{i32 242145}
+!60 = !{i32 -1, i32 -1, i32 -1}
+!61 = !{i32 16}
+!62 = !{}
+!63 = !{i1 false, i1 false, i1 false}
+!64 = !{!65, !65, i64 0}
+!65 = !{!"any pointer", !66, i64 0}
+!66 = !{!"omnipotent char", !67, i64 0}
+!67 = !{!"Simple C++ TBAA"}
+!68 = !{!69, !71, !73}
+!69 = distinct !{!69, !70, !"_ZN7__spirv22InitSizesSTWorkgroupIdILi2EN4sycl3_V12idILi2EEEE8initSizeEv: %agg.result"}
+!70 = distinct !{!70, !"_ZN7__spirv22InitSizesSTWorkgroupIdILi2EN4sycl3_V12idILi2EEEE8initSizeEv"}
+!71 = distinct !{!71, !72, !"_ZN7__spirv15initWorkgroupIdILi2EN4sycl3_V12idILi2EEEEET0_v: %agg.result"}
+!72 = distinct !{!72, !"_ZN7__spirv15initWorkgroupIdILi2EN4sycl3_V12idILi2EEEEET0_v"}
+!73 = distinct !{!73, !74, !"_ZN4sycl3_V16detail7Builder10getElementILi2EEEKNS0_7nd_itemIXT_EEEPS5_: %agg.result"}
+!74 = distinct !{!74, !"_ZN4sycl3_V16detail7Builder10getElementILi2EEEKNS0_7nd_itemIXT_EEEPS5_"}
+!75 = !{!76, !78, !73}
+!76 = distinct !{!76, !77, !"_ZN7__spirv28InitSizesSTLocalInvocationIdILi2EN4sycl3_V12idILi2EEEE8initSizeEv: %agg.result"}
+!77 = distinct !{!77, !"_ZN7__spirv28InitSizesSTLocalInvocationIdILi2EN4sycl3_V12idILi2EEEE8initSizeEv"}
+!78 = distinct !{!78, !79, !"_ZN7__spirv21initLocalInvocationIdILi2EN4sycl3_V12idILi2EEEEET0_v: %agg.result"}
+!79 = distinct !{!79, !"_ZN7__spirv21initLocalInvocationIdILi2EN4sycl3_V12idILi2EEEEET0_v"}
+!80 = distinct !{!80, !81}
+!81 = !{!"llvm.loop.mustprogress"}
+!82 = !{!83, !65, i64 0}
+!83 = !{!"_ZTSN4sycl3_V13ext6oneapi12experimental6matrix12joint_matrixINS0_9sub_groupEfLNS4_3useE2ELm8ELm16ELNS4_6layoutE3EEE", !65, i64 0}
+!84 = distinct !{!84, !81}
+!85 = distinct !{!85, !81}
+!86 = !{!87, !65, i64 0}
+!87 = !{!"_ZTSN4sycl3_V13ext6oneapi12experimental6matrix12joint_matrixINS0_9sub_groupENS2_8bfloat16ELNS4_3useE0ELm8ELm16ELNS4_6layoutE0EEE", !65, i64 0}
+!88 = distinct !{!88, !81}
+!89 = !{!90, !65, i64 0}
+!90 = !{!"_ZTSN4sycl3_V13ext6oneapi12experimental6matrix12joint_matrixINS0_9sub_groupENS2_8bfloat16ELNS4_3useE1ELm16ELm16ELNS4_6layoutE2EEE", !65, i64 0}
+!91 = distinct !{!91, !81}
+!92 = !{!93}
+!93 = distinct !{!93, !94, !"_ZN4sycl3_V13ext6oneapi12experimental6matrix16joint_matrix_madINS0_9sub_groupENS2_8bfloat16ES7_fLm8ELm16ELm16ELNS4_6layoutE0ELS8_2EEENS4_12joint_matrixIT_T2_LNS4_3useE2EXT3_EXT5_ELS8_3EEESA_RNS9_ISA_T0_LSC_0EXT3_EXT4_EXT6_EEERNS9_ISA_T1_LSC_1EXT4_EXT5_EXT7_EEERSD_: %agg.result"}
+!94 = distinct !{!94, !"_ZN4sycl3_V13ext6oneapi12experimental6matrix16joint_matrix_madINS0_9sub_groupENS2_8bfloat16ES7_fLm8ELm16ELm16ELNS4_6layoutE0ELS8_2EEENS4_12joint_matrixIT_T2_LNS4_3useE2EXT3_EXT5_ELS8_3EEESA_RNS9_ISA_T0_LSC_0EXT3_EXT4_EXT6_EEERNS9_ISA_T1_LSC_1EXT4_EXT5_EXT7_EEERSD_"}
+!95 = distinct !{!95, !81}
+!96 = distinct !{!96, !81}
+!97 = distinct !{!97, !81}
+!98 = distinct !{!98, !81}
+!99 = distinct !{!99, !81}
diff --git a/test/extensions/INTEL/SPV_INTEL_joint_matrix/bf16_conversion_instructions.ll b/test/extensions/INTEL/SPV_INTEL_joint_matrix/bf16_conversion_instructions.ll
new file mode 100644
index 0000000000..237c05688b
--- /dev/null
+++ b/test/extensions/INTEL/SPV_INTEL_joint_matrix/bf16_conversion_instructions.ll
@@ -0,0 +1,79 @@
+; RUN: llvm-as < %s -o %t.bc
+; RUN: llvm-spirv %t.bc --spirv-ext=+SPV_KHR_cooperative_matrix,+SPV_INTEL_joint_matrix,+SPV_INTEL_bfloat16_conversion -o %t.spv
+; RUN: llvm-spirv %t.spv -to-text -o %t.spt
+; RUN: FileCheck < %t.spt %s --check-prefix=CHECK-SPIRV
+
+; RUN: llvm-spirv -r %t.spv -o %t.rev.bc
+; RUN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-OCL-IR
+
+; RUN: llvm-spirv -r %t.spv -o %t.rev.bc --spirv-target-env=SPV-IR
+; RUN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-SPV-IR
+
+; RUN: not llvm-spirv %t.bc --spirv-ext=+SPV_KHR_cooperative_matrix,+SPV_INTEL_bfloat16_conversion 2>&1 \
+; RUN: | FileCheck %s --check-prefix=CHECK-ERROR
+
+; CHECK-ERROR: InvalidInstruction: Can't translate llvm instruction:
+; CHECK-ERROR-NEXT: ConvertFToBF16INTEL
+; CHECK-ERROR-NEXT: Can be used with cooperative matrices only when SPV_INTEL_joint_matrix is enabled
+
+; CHECK-SPIRV-DAG: Capability CooperativeMatrixKHR
+; CHECK-SPIRV-DAG: Capability Bfloat16ConversionINTEL
+; CHECK-SPIRV-DAG: Capability JointMatrixBF16ComponentTypeINTEL
+; CHECK-SPIRV-DAG: Extension "SPV_INTEL_bfloat16_conversion"
+; CHECK-SPIRV-DAG: Extension "SPV_KHR_cooperative_matrix"
+; CHECK-SPIRV-DAG: Extension "SPV_INTEL_joint_matrix"
+; CHECK-SPIRV-DAG: TypeInt [[#ShortTy:]] 16 0
+; CHECK-SPIRV-DAG: TypeFloat [[#FP32Ty:]] 32
+; CHECK-SPIRV-DAG: TypeCooperativeMatrixKHR [[#FP32MatTy:]] [[#FP32Ty]]
+; CHECK-SPIRV-DAG: TypeCooperativeMatrixKHR [[#ShortMatTy:]] [[#ShortTy]]
+; CHECK-SPIRV: CompositeConstruct [[#FP32MatTy]] [[#FP32Mat:]]
+; CHECK-SPIRV: ConvertFToBF16INTEL [[#ShortMatTy]] [[#]] [[#FP32Mat]]
+; CHECK-SPIRV: CompositeConstruct [[#ShortMatTy]] [[#ShortMat:]]
+; CHECK-SPIRV: ConvertBF16ToFINTEL [[#FP32MatTy]] [[#]] [[#ShortMat]]
+
+; CHECK-OCL-IR: %[[#FP32Matrix:]] = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructf(float 0.000000e+00)
+; CHECK-OCL-IR: call spir_func target("spirv.CooperativeMatrixKHR", i16, 3, 12, 12, 2) @_Z32intel_convert_bfloat16_as_ushortPU3AS145__spirv_CooperativeMatrixKHR__float_3_12_12_2(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %[[#FP32Matrix]])
+; CHECK-OCL-IR: %[[#ShortMatrix:]] = call spir_func target("spirv.CooperativeMatrixKHR", i16, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructs(i16 0)
+; CHECK-OCL-IR: call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z31intel_convert_as_bfloat16_floatPU3AS145__spirv_CooperativeMatrixKHR__short_3_12_12_2(target("spirv.CooperativeMatrixKHR", i16, 3, 12, 12, 2) %[[#ShortMatrix]])
+
+
+; CHECK-SPV-IR: %[[#FP32Matrix:]] = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructf(float 0.000000e+00)
+; CHECK-SPV-IR: call spir_func target("spirv.CooperativeMatrixKHR", i16, 3, 12, 12, 2) @_Z27__spirv_ConvertFToBF16INTELPU3AS145__spirv_CooperativeMatrixKHR__float_3_12_12_2(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %[[#FP32Matrix]])
+; CHECK-SPV-IR: %[[#ShortMatrix:]] = call spir_func target("spirv.CooperativeMatrixKHR", i16, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructs(i16 0)
+; CHECK-SPV-IR: call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z27__spirv_ConvertBF16ToFINTELPU3AS145__spirv_CooperativeMatrixKHR__short_3_12_12_2(target("spirv.CooperativeMatrixKHR", i16, 3, 12, 12, 2) %[[#ShortMatrix]])
+
+
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "spir64-unknown-unknown"
+
+define void @convert_f_to_bf() {
+entry:
+  %0 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructFloat(float 0.000000e+00)
+  %call = call spir_func target("spirv.CooperativeMatrixKHR", i16, 3, 12, 12, 2) @_Z27__spirv_ConvertFToBF16INTEL(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %0)
+  ret void
+}
+
+define void @convert_bf_to_f() {
+entry:
+  %0 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i16, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructInt16(i16 0)
+  %call = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z27__spirv_ConvertBF16ToFINTEL(target("spirv.CooperativeMatrixKHR", i16, 3, 12, 12, 2) %0)
+  ret void
+}
+
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructFloat(float noundef)
+
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", i16, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructInt16(i16 noundef)
+
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", i16, 3, 12, 12, 2) @_Z27__spirv_ConvertFToBF16INTEL(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) noundef)
+
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z27__spirv_ConvertBF16ToFINTEL(target("spirv.CooperativeMatrixKHR", i16, 3, 12, 12, 2) noundef)
+
+!llvm.module.flags = !{!0, !1, !2, !3, !4}
+!llvm.ident = !{!5}
+
+!0 = !{i32 7, !"Dwarf Version", i32 4}
+!1 = !{i32 1, !"wchar_size", i32 4}
+!2 = !{i32 8, !"PIC Level", i32 2}
+!3 = !{i32 7, !"PIE Level", i32 2}
+!4 = !{i32 7, !"uwtable", i32 2}
+!5 = !{!"clang version 17.0.0"}
diff --git a/test/extensions/INTEL/SPV_INTEL_joint_matrix/cooperative_matrix_apply.ll b/test/extensions/INTEL/SPV_INTEL_joint_matrix/cooperative_matrix_apply.ll
new file mode 100644
index 0000000000..b0f97b74d7
--- /dev/null
+++ b/test/extensions/INTEL/SPV_INTEL_joint_matrix/cooperative_matrix_apply.ll
@@ -0,0 +1,149 @@
+;; compiled from joint_matrix_apply_bf16.cpp from intel/llvm with some modifications
+
+; RUN: llvm-as < %s -o %t.bc
+; RUN: llvm-spirv %t.bc --spirv-ext=+SPV_KHR_cooperative_matrix,+SPV_INTEL_joint_matrix -o %t.spv
+; RUN: llvm-spirv %t.spv -to-text -o %t.spt
+; RUN: FileCheck < %t.spt %s --check-prefix=CHECK-SPIRV
+
+; RUN: llvm-spirv -r %t.spv -o %t.rev.bc
+; RUN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM
+
+; CHECK-SPIRV-DAG: Capability CooperativeMatrixKHR
+; CHECK-SPIRV-DAG: Capability CooperativeMatrixInvocationInstructionsINTEL
+; CHECK-SPIRV-DAG: Extension "SPV_INTEL_joint_matrix"
+; CHECK-SPIRV-DAG: Extension "SPV_KHR_cooperative_matrix"
+; CHECK-SPIRV-DAG: TypeCooperativeMatrixKHR [[#MatTy:]]
+; CHECK-SPIRV: CompositeConstruct [[#MatTy]] [[#Mat:]]
+; CHECK-SPIRV: PtrCastToGeneric [[#]] [[#Ptr:]] [[#]]
+; CHECK-SPIRV: CooperativeMatrixApplyFunctionINTEL [[#MatTy]] [[#Apply:]] [[#Ptr]] [[#Mat]]
+; CHECK-SPIRV: CooperativeMatrixStoreKHR [[#]] [[#Apply]]
+
+; CHECK-LLVM: %[[Mat:[%0-9a-z.]+]] = call spir_func target("spirv.CooperativeMatrixKHR", i16, 3, 8, 16, 0) @"_Z26__spirv_CompositeConstructP38class.sycl::_V1::ext::oneapi::bfloat16"
+; CHECK-LLVM: %[[Apply:[%0-9a-z.]+]] = call spir_func target("spirv.CooperativeMatrixKHR", i16, 3, 8, 16, 0) @"_Z43__spirv_CooperativeMatrixApplyFunctionINTELPU3AS477class.sycl::_V1::ext::oneapi::experimental::matrix::helper::reference_wrapperPU3AS144__spirv_CooperativeMatrixKHR__short_3_8_16_0"(ptr addrspace(4) %ref.tmp.ascast.i21, target("spirv.CooperativeMatrixKHR", i16, 3, 8, 16, 0) %[[Mat]])
+; CHECK-LLVM: call spir_func void @"_Z33__spirv_CooperativeMatrixStoreKHRPU3AS138class.sycl::_V1::ext::oneapi::bfloat16PU3AS144__spirv_CooperativeMatrixKHR__short_3_8_16_0il"(ptr addrspace(1) %{{.*}}, target("spirv.CooperativeMatrixKHR", i16, 3, 8, 16, 0) %[[Apply]], i32 0, i64 0)
+
+; ModuleID = 'matrix_apply.bc'
+source_filename = "../llvm/sycl/test-e2e/Matrix/joint_matrix_apply_bf16.cpp"
+target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-n8:16:32:64"
+target triple = "spir64-unknown-unknown"
+
+%"class.sycl::_V1::range" = type { %"class.sycl::_V1::detail::array" }
+%"class.sycl::_V1::detail::array" = type { [2 x i64] }
+%"class.sycl::_V1::id" = type { %"class.sycl::_V1::detail::array" }
+%"class.sycl::_V1::ext::oneapi::experimental::matrix::helper::reference_wrapper" = type { ptr addrspace(4) }
+%"class.sycl::_V1::ext::oneapi::bfloat16" = type { i16 }
+%class.anon.0 = type <{ %"class.sycl::_V1::accessor", %class.anon, [7 x i8] }>
+%"class.sycl::_V1::accessor" = type { %"class.sycl::_V1::detail::AccessorImplDevice", %union.anon }
+%"class.sycl::_V1::detail::AccessorImplDevice" = type { %"class.sycl::_V1::id", %"class.sycl::_V1::range", %"class.sycl::_V1::range" }
+%union.anon = type { ptr addrspace(1) }
+%class.anon = type { i8 }
+
+$_ZTSZZ17matrix_verify_addIN4sycl3_V13ext6oneapi8bfloat16ELm16ELm32EZ4mainEUlRS4_E_EvNS1_5queueER10big_matrixIT_XT0_EXT1_EERNS1_8nd_rangeILi2EEEfOT2_ENKUlRNS1_7handlerEE_clESI_EUlNS1_7nd_itemILi2EEEE_ = comdat any
+
+@__spirv_BuiltInGlobalInvocationId = external dso_local local_unnamed_addr addrspace(1) constant <3 x i64>, align 32
+@__spirv_BuiltInLocalInvocationId = external dso_local local_unnamed_addr addrspace(1) constant <3 x i64>, align 32
+
+; Function Attrs: convergent norecurse nounwind
+define weak_odr dso_local spir_kernel void @_ZTSZZ17matrix_verify_addIN4sycl3_V13ext6oneapi8bfloat16ELm16ELm32EZ4mainEUlRS4_E_EvNS1_5queueER10big_matrixIT_XT0_EXT1_EERNS1_8nd_rangeILi2EEEfOT2_ENKUlRNS1_7handlerEE_clESI_EUlNS1_7nd_itemILi2EEEE_(ptr addrspace(1) noundef align 2 %_arg_accA, ptr noundef byval(%"class.sycl::_V1::range") align 8 %_arg_accA1, ptr noundef byval(%"class.sycl::_V1::range") align 8 %_arg_accA2, ptr noundef byval(%"class.sycl::_V1::id") align 8 %_arg_accA3) local_unnamed_addr {
+entry:
+  %ref.tmp.i20 = alloca %"class.sycl::_V1::ext::oneapi::experimental::matrix::helper::reference_wrapper", align 8
+  %agg.tmp.i17 = alloca %"class.sycl::_V1::ext::oneapi::bfloat16", align 2
+  %ref.tmp6.i = alloca float, align 4
+  %__SYCLKernel = alloca %class.anon.0, align 8
+  %__SYCLKernel.ascast = addrspacecast ptr %__SYCLKernel to ptr addrspace(4)
+  call void @llvm.lifetime.start.p0(i64 64, ptr nonnull %__SYCLKernel)
+  %agg.tmp.sroa.0.sroa.0.0.copyload = load i64, ptr %_arg_accA1, align 8
+  %agg.tmp.sroa.0.sroa.2.0._arg_accA1.ascast.sroa_idx = getelementptr inbounds i8, ptr %_arg_accA1, i64 8
+  %agg.tmp.sroa.0.sroa.2.0.copyload = load i64, ptr %agg.tmp.sroa.0.sroa.2.0._arg_accA1.ascast.sroa_idx, align 8
+  %agg.tmp5.sroa.0.sroa.0.0.copyload = load i64, ptr %_arg_accA2, align 8
+  %agg.tmp5.sroa.0.sroa.2.0._arg_accA2.ascast.sroa_idx = getelementptr inbounds i8, ptr %_arg_accA2, i64 8
+  %agg.tmp5.sroa.0.sroa.2.0.copyload = load i64, ptr %agg.tmp5.sroa.0.sroa.2.0._arg_accA2.ascast.sroa_idx, align 8
+  %agg.tmp6.sroa.0.sroa.0.0.copyload = load i64, ptr %_arg_accA3, align 8
+  %agg.tmp6.sroa.0.sroa.2.0._arg_accA3.ascast.sroa_idx = getelementptr inbounds i8, ptr %_arg_accA3, i64 8
+  %agg.tmp6.sroa.0.sroa.2.0.copyload = load i64, ptr %agg.tmp6.sroa.0.sroa.2.0._arg_accA3.ascast.sroa_idx, align 8
+  %0 = getelementptr inbounds %"class.sycl::_V1::accessor", ptr %__SYCLKernel, i64 0, i32 1
+  store i64 %agg.tmp6.sroa.0.sroa.0.0.copyload, ptr %__SYCLKernel, align 8
+  %AccessRange.i.i.i.i.i = getelementptr inbounds %"class.sycl::_V1::detail::AccessorImplDevice", ptr %__SYCLKernel, i64 0, i32 1
+  store i64 %agg.tmp.sroa.0.sroa.0.0.copyload, ptr %AccessRange.i.i.i.i.i, align 8
+  %MemRange.i.i.i.i.i = getelementptr inbounds %"class.sycl::_V1::detail::AccessorImplDevice", ptr %__SYCLKernel, i64 0, i32 2
+  store i64 %agg.tmp5.sroa.0.sroa.0.0.copyload, ptr %MemRange.i.i.i.i.i, align 8
+  %arrayidx.i21.i.i.i.i = getelementptr inbounds [2 x i64], ptr %__SYCLKernel, i64 0, i64 1
+  store i64 %agg.tmp6.sroa.0.sroa.2.0.copyload, ptr %arrayidx.i21.i.i.i.i, align 8
+  %arrayidx.i25.i.i.i.i = getelementptr inbounds %"class.sycl::_V1::detail::AccessorImplDevice", ptr %__SYCLKernel, i64 0, i32 1, i32 0, i32 0, i64 1
+  store i64 %agg.tmp.sroa.0.sroa.2.0.copyload, ptr %arrayidx.i25.i.i.i.i, align 8
+  %arrayidx.i29.i.i.i.i = getelementptr inbounds %"class.sycl::_V1::detail::AccessorImplDevice", ptr %__SYCLKernel, i64 0, i32 2, i32 0, i32 0, i64 1
+  store i64 %agg.tmp5.sroa.0.sroa.2.0.copyload, ptr %arrayidx.i29.i.i.i.i, align 8
+  %mul.i6.i.i.i.i = mul i64 %agg.tmp6.sroa.0.sroa.0.0.copyload, %agg.tmp5.sroa.0.sroa.2.0.copyload
+  %1 = getelementptr %"class.sycl::_V1::ext::oneapi::bfloat16", ptr addrspace(1) %_arg_accA, i64 %mul.i6.i.i.i.i
+  %add.ptr.i = getelementptr %"class.sycl::_V1::ext::oneapi::bfloat16", ptr addrspace(1) %1, i64 %agg.tmp6.sroa.0.sroa.2.0.copyload
+  store ptr addrspace(1) %add.ptr.i, ptr %0, align 8
+  %2 = load i64, ptr addrspace(1) getelementptr inbounds (i8, ptr addrspace(1) @__spirv_BuiltInGlobalInvocationId, i64 8), align 8
+  %3 = load i64, ptr addrspace(1) @__spirv_BuiltInGlobalInvocationId, align 32
+  %4 = load i64, ptr addrspace(1) getelementptr inbounds (i8, ptr addrspace(1) @__spirv_BuiltInLocalInvocationId, i64 8), align 8
+  %5 = load i64, ptr addrspace(1) @__spirv_BuiltInLocalInvocationId, align 32
+  %ref.tmp6.ascast.i = addrspacecast ptr %ref.tmp6.i to ptr addrspace(4)
+  %cmp.i11 = icmp ult i64 %2, 2147483648
+  %cmp.i = icmp ult i64 %3, 2147483648
+  %cmp.i15 = icmp ult i64 %4, 2147483648
+  %sub.i = sub nsw i64 %2, %4
+  %cmp.i12 = icmp ult i64 %5, 2147483648
+  %sub5.i = sub nsw i64 %3, %5
+  call void @llvm.lifetime.start.p0(i64 4, ptr nonnull %ref.tmp6.i)
+  store float 5.000000e+00, ptr %ref.tmp6.i, align 4
+  %call.i.i = call spir_func noundef zeroext i16 @__devicelib_ConvertFToBF16INTEL(ptr addrspace(4) noundef align 4 dereferenceable(4) %ref.tmp6.ascast.i)
+  call void @llvm.lifetime.start.p0(i64 2, ptr nonnull %agg.tmp.i17)
+  store i16 %call.i.i, ptr %agg.tmp.i17, align 2
+  %call.i18 = call spir_func noundef target("spirv.CooperativeMatrixKHR", i16, 3, 8, 16, 0) @_Z26__spirv_CompositeConstruct(ptr noundef nonnull byval(%"class.sycl::_V1::ext::oneapi::bfloat16") align 2 %agg.tmp.i17)
+  call void @llvm.lifetime.end.p0(i64 2, ptr nonnull %agg.tmp.i17)
+  call void @llvm.lifetime.end.p0(i64 4, ptr nonnull %ref.tmp6.i)
+  %lambda.i = getelementptr inbounds %class.anon.0, ptr addrspace(4) %__SYCLKernel.ascast, i64 0, i32 1
+  %ref.tmp.ascast.i21 = addrspacecast ptr %ref.tmp.i20 to ptr addrspace(4)
+  call void @llvm.lifetime.start.p0(i64 8, ptr nonnull %ref.tmp.i20)
+  store ptr addrspace(4) %lambda.i, ptr %ref.tmp.i20, align 8
+  %call.i22 = call spir_func noundef target("spirv.CooperativeMatrixKHR", i16, 3, 8, 16, 0) @_Z43__spirv_CooperativeMatrixApplyFunctionINTEL(ptr addrspace(4) noundef align 8 dereferenceable(8) %ref.tmp.ascast.i21, target("spirv.CooperativeMatrixKHR", i16, 3, 8, 16, 0) noundef %call.i18)
+  call void @llvm.lifetime.end.p0(i64 8, ptr nonnull %ref.tmp.i20)
+  %6 = load ptr addrspace(1), ptr %0, align 8
+  %7 = load i64, ptr %__SYCLKernel, align 8
+  %8 = load i64, ptr %arrayidx.i29.i.i.i.i, align 8
+  %mul.i6.i.i.i.i.i = mul i64 %7, %8
+  %9 = load i64, ptr %arrayidx.i21.i.i.i.i, align 8
+  %add.i7.i.i.i.i.i = add i64 %mul.i6.i.i.i.i.i, %9
+  %idx.neg.i.i = sub i64 0, %add.i7.i.i.i.i.i
+  %add.ptr.i.i = getelementptr inbounds %"class.sycl::_V1::ext::oneapi::bfloat16", ptr addrspace(1) %6, i64 %idx.neg.i.i
+  %mul12.i = shl nsw i64 %sub.i, 8
+  %add.ptr.i43 = getelementptr inbounds %"class.sycl::_V1::ext::oneapi::bfloat16", ptr addrspace(1) %add.ptr.i.i, i64 %mul12.i
+  %div14.i = and i64 %sub5.i, -16
+  %add.ptr.i44 = getelementptr inbounds %"class.sycl::_V1::ext::oneapi::bfloat16", ptr addrspace(1) %add.ptr.i43, i64 %div14.i
+  call spir_func void @_Z33__spirv_CooperativeMatrixStoreKHRPU3AS4iPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_3ili(ptr addrspace(1) noundef %add.ptr.i44, target("spirv.CooperativeMatrixKHR", i16, 3, 8, 16, 0) noundef %call.i22, i32 noundef 0, i64 noundef 0)
+  call void @llvm.lifetime.end.p0(i64 64, ptr nonnull %__SYCLKernel)
+  ret void
+}
+
+; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
+declare void @llvm.lifetime.start.p0(i64 immarg, ptr nocapture)
+
+; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
+declare void @llvm.lifetime.end.p0(i64 immarg, ptr nocapture)
+
+; Function Attrs: convergent nounwind
+declare dso_local spir_func noundef target("spirv.CooperativeMatrixKHR", i16, 3, 8, 16, 0) @_Z26__spirv_CompositeConstruct(ptr noundef byval(%"class.sycl::_V1::ext::oneapi::bfloat16") align 2) local_unnamed_addr
+
+; Function Attrs: convergent nounwind
+declare dso_local spir_func zeroext i16 @__devicelib_ConvertFToBF16INTEL(ptr addrspace(4) noundef align 4 dereferenceable(4)) local_unnamed_addr
+
+; Function Attrs: convergent nounwind
+declare dso_local spir_func noundef target("spirv.CooperativeMatrixKHR", i16, 3, 8, 16, 0) @_Z43__spirv_CooperativeMatrixApplyFunctionINTEL(ptr addrspace(4) noundef align 8 dereferenceable(8), target("spirv.CooperativeMatrixKHR", i16, 3, 8, 16, 0) noundef) local_unnamed_addr
+
+; Function Attrs: convergent nounwind
+declare dso_local spir_func void @_Z33__spirv_CooperativeMatrixStoreKHRPU3AS4iPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_3ili(ptr addrspace(1) noundef, target("spirv.CooperativeMatrixKHR", i16, 3, 8, 16, 0) noundef, i32 noundef, i64 noundef) local_unnamed_addr
+
+!llvm.module.flags = !{!0, !1}
+!opencl.spir.version = !{!2}
+!spirv.Source = !{!3}
+!llvm.ident = !{!4}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{i32 7, !"frame-pointer", i32 2}
+!2 = !{i32 1, i32 2}
+!3 = !{i32 4, i32 100000}
+!4 = !{!"clang version 18.0.0 (https://github.com/intel/llvm.git)"}
diff --git a/test/extensions/INTEL/SPV_INTEL_joint_matrix/cooperative_matrix_prefetch.ll b/test/extensions/INTEL/SPV_INTEL_joint_matrix/cooperative_matrix_prefetch.ll
index a6f24bc596..03dfbdfeb8 100644
--- a/test/extensions/INTEL/SPV_INTEL_joint_matrix/cooperative_matrix_prefetch.ll
+++ b/test/extensions/INTEL/SPV_INTEL_joint_matrix/cooperative_matrix_prefetch.ll
@@ -20,9 +20,9 @@
 ; CHECK-SPIRV-DAG: Constant [[#Int32Ty]] [[#Const3:]] 3
 ; CHECK-SPIRV-DAG: Constant [[#Int32Ty]] [[#Const2:]] 2
 ; CHECK-SPIRV-DAG: Constant [[#Int32Ty]] [[#Const1:]] 1
-; CHECK-SPIRV-DAG: TypeCooperativeMatrixKHR [[#MatTy1:]] [[#Int32Ty]] [[#Const3]] [[#Const12]] [[#Const12]] [[#Const3]]
-; CHECK-SPIRV-DAG: TypeCooperativeMatrixKHR [[#MatTy2:]] [[#Int8Ty]] [[#Const0]] [[#Const12]] [[#Const48]] [[#Const3]]
-; CHECK-SPIRV-DAG: TypeCooperativeMatrixKHR [[#MatTy3:]] [[#Int8Ty]] [[#Const2]] [[#Const48]] [[#Const12]] [[#Const3]]
+; CHECK-SPIRV-DAG: TypeCooperativeMatrixKHR [[#MatTy1:]] [[#Int32Ty]] [[#Const3]] [[#Const12]] [[#Const12]] [[#Const2]]
+; CHECK-SPIRV-DAG: TypeCooperativeMatrixKHR [[#MatTy2:]] [[#Int8Ty]] [[#Const3]] [[#Const12]] [[#Const48]] [[#Const0]]
+; CHECK-SPIRV-DAG: TypeCooperativeMatrixKHR [[#MatTy3:]] [[#Int8Ty]] [[#Const2]] [[#Const48]] [[#Const12]] [[#Const1]]
 ; CHECK-SPIRV: CompositeConstruct [[#MatTy1]]
 ; CHECK-SPIRV: CooperativeMatrixLoadKHR [[#MatTy2]] [[#Load1:]]
 ; CHECK-SPIRV: CooperativeMatrixLengthKHR [[#Int32Ty]] [[#]] [[#MatTy2]]
@@ -31,14 +31,14 @@
 ; CHECK-SPIRV: CooperativeMatrixMulAddKHR [[#MatTy1]]
 ; CHECK-SPIRV: CooperativeMatrixStoreKHR
 
-
-; CHECK-LLVM: call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructi(i32 0)
-; CHECK-LLVM: call spir_func void @_Z38__spirv_CooperativeMatrixPrefetchINTELPU3AS4ciiiiil(ptr addrspace(4) %[[MatrixPtr:[%0-9a-z.]+]], i32 0, i32 0, i32 1, i32 1, i32 0, i64 %_arg_K)
-; CHECK-LLVM: call spir_func target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 3) @_Z86__spirv_CooperativeMatrixLoadKHR_RPU3AS144__spirv_CooperativeMatrixKHR__char_0_12_48_3PU3AS4clii(ptr addrspace(4) %[[MatrixPtr:[%0-9a-z.]+]], i64 %_arg_K, i32 0, i32 1)
-; CHECK-LLVM: call spir_func i32 @_Z34__spirv_CooperativeMatrixLengthKHRPU3AS144__spirv_CooperativeMatrixKHR__char_0_12_48_3(target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 3)
-; CHECK-LLVM: call spir_func target("spirv.CooperativeMatrixKHR", i8, 2, 48, 12, 3) @_Z86__spirv_CooperativeMatrixLoadKHR_RPU3AS144__spirv_CooperativeMatrixKHR__char_2_48_12_3PU3AS4cl
-; CHECK-LLVM: call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z34__spirv_CooperativeMatrixMulAddKHRPU3AS144__spirv_CooperativeMatrixKHR__char_0_12_48_3PU3AS144__spirv_CooperativeMatrixKHR__char_2_48_12_3PU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_3i(target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 3) %{{.*}}, target("spirv.CooperativeMatrixKHR", i8, 2, 48, 12, 3) %{{.*}}, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3)
-; CHECK-LLVM: call spir_func void @_Z33__spirv_CooperativeMatrixStoreKHRPU3AS4iPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_3ili(ptr addrspace(4) %{{.*}}, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3)
+; CHECK-LLVM: call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructi(i32 0)
+; CHECK-LLVM: call spir_func void @_Z38__spirv_CooperativeMatrixPrefetchINTELPU3AS4ciiiil(ptr addrspace(4) %[[MatrixPtr:[%0-9a-z.]+]], i32 12, i32 48, i32 0, i32 0, i64 %_arg_K)
+; CHECK-LLVM: call spir_func target("spirv.CooperativeMatrixKHR", i8, 3, 12, 48, 0) @_Z86__spirv_CooperativeMatrixLoadKHR_RPU3AS144__spirv_CooperativeMatrixKHR__char_3_12_48_0PU3AS4cili(ptr addrspace(4) %[[MatrixPtr:[%0-9a-z.]+]], i32 0, i64 %_arg_K, i32 1)
+; CHECK-LLVM: call spir_func i32 @_Z34__spirv_CooperativeMatrixLengthKHRPU3AS144__spirv_CooperativeMatrixKHR__char_3_12_48_0(target("spirv.CooperativeMatrixKHR", i8, 3, 12, 48, 0)
+; CHECK-LLVM: call spir_func void @_Z38__spirv_CooperativeMatrixPrefetchINTELPU3AS4ciiiil(ptr addrspace(4) %[[MatrixPtr:[%0-9a-z.]+]], i32 12, i32 48, i32 0, i32 0, i64 %mul22.i)
+; CHECK-LLVM: call spir_func target("spirv.CooperativeMatrixKHR", i8, 2, 48, 12, 1) @_Z86__spirv_CooperativeMatrixLoadKHR_RPU3AS144__spirv_CooperativeMatrixKHR__char_2_48_12_1PU3AS4cil
+; CHECK-LLVM: call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z34__spirv_CooperativeMatrixMulAddKHRPU3AS144__spirv_CooperativeMatrixKHR__char_3_12_48_0PU3AS144__spirv_CooperativeMatrixKHR__char_2_48_12_1PU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_2i(target("spirv.CooperativeMatrixKHR", i8, 3, 12, 48, 0) %{{.*}}, target("spirv.CooperativeMatrixKHR", i8, 2, 48, 12, 1) %{{.*}}, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2)
+; CHECK-LLVM: call spir_func void @_Z33__spirv_CooperativeMatrixStoreKHRPU3AS4iPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_2ili(ptr addrspace(4) %{{.*}}, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2)
 
 ; ModuleID = 'test-matrix-opaque.bc'
 source_filename = "matrix-int8-test.cpp"
@@ -57,8 +57,8 @@ $_ZTSZZ15matrix_multiply = comdat any
 ; Function Attrs: convergent norecurse
 define weak_odr dso_local spir_kernel void @_ZTSZZ15matrix_multiply(ptr addrspace(1) noundef align 1 %_arg_accA, ptr addrspace(1) noundef align 1 %_arg_accB, ptr noundef byval(%"class.sycl::_V1::range") align 8 %_arg_accB5, ptr noundef byval(%"class.sycl::_V1::id") align 8 %_arg_accB6, ptr addrspace(1) noundef align 4 %_arg_accC, i64 noundef %_arg_N, i64 noundef %_arg_K) local_unnamed_addr #0 comdat {
 entry:
-  %sub_c.sroa.0.i = alloca target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3), align 8
-  %ref.tmp29.sroa.0.i = alloca target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3), align 8
+  %sub_c.sroa.0.i = alloca target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2), align 8
+  %ref.tmp29.sroa.0.i = alloca target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2), align 8
   %agg.tmp15.sroa.0.sroa.2.0..sroa_idx = getelementptr inbounds %"class.sycl::_V1::range", ptr %_arg_accB5, i64 0, i32 0, i32 0, i64 1
   %agg.tmp15.sroa.0.sroa.2.0.copyload = load i64, ptr %agg.tmp15.sroa.0.sroa.2.0..sroa_idx, align 8
   %agg.tmp16.sroa.0.sroa.0.0.copyload = load i64, ptr %_arg_accB6, align 8
@@ -80,8 +80,8 @@ entry:
   %cmp.i58.i = icmp ult i64 %5, 2147483648
   %sub5.i = sub nsw i64 %2, %5
   call void @llvm.lifetime.start.p0(i64 8, ptr nonnull %sub_c.sroa.0.i)
-  %call.i.i = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z26__spirv_CompositeConstruct(i32 noundef 0) #4
-  store target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %call.i.i, ptr %sub_c.sroa.0.i, align 8
+  %call.i.i = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z26__spirv_CompositeConstruct(i32 noundef 0) #4
+  store target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %call.i.i, ptr %sub_c.sroa.0.i, align 8
   %mul.i = mul nsw i64 %sub.i, 12
   %div2452.i = lshr i64 %sub5.i, 4
   %mul26.i = mul i64 %div2452.i, 48
@@ -105,20 +105,20 @@ for.body.i:                                       ; preds = %for.cond.i
   %conv13.i = zext i32 %mul12.i to i64
   %add.ptr.i96.i = getelementptr inbounds i8, ptr addrspace(1) %add.ptr.i93.i, i64 %conv13.i
   %call.ascast.i66.i = addrspacecast ptr addrspace(1) %add.ptr.i96.i to ptr addrspace(4)
-  tail call spir_func void @_Z38__spirv_CooperativeMatrixPrefetchINTEL(ptr addrspace(4) noundef %call.ascast.i66.i, i32 noundef 0, i32 noundef 0, i32 noundef 1, i32 noundef 1, i32 noundef 0, i64 noundef %_arg_K) #4
-  %call1.i.i = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 3) @_Z32__spirv_CooperativeMatrixLoadKHR_1(ptr addrspace(4) noundef %call.ascast.i66.i, i64 noundef %_arg_K, i32 noundef 0, i32 noundef 1) #4
-  %len = tail call spir_func noundef i32 @_Z34__spirv_CooperativeMatrixLengthKHR(target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 3) %call1.i.i)
+  tail call spir_func void @_Z38__spirv_CooperativeMatrixPrefetchINTEL(ptr addrspace(4) noundef %call.ascast.i66.i, i32 noundef 12, i32 noundef 48, i32 noundef 0, i32 noundef 0, i64 noundef %_arg_K)
+  %call1.i.i = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i8, 3, 12, 48, 0) @_Z32__spirv_CooperativeMatrixLoadKHR_1(ptr addrspace(4) noundef %call.ascast.i66.i, i32 noundef 0, i64 noundef %_arg_K, i32 noundef 1) #4
+  %len = tail call spir_func noundef i32 @_Z34__spirv_CooperativeMatrixLengthKHR(target("spirv.CooperativeMatrixKHR", i8, 3, 12, 48, 0) %call1.i.i)
   %div20.i = mul nsw i32 %k.0.i, 12
   %conv21.i = zext i32 %div20.i to i64
   %mul23.i = mul i64 %mul22.i, %conv21.i
   %add.ptr.i111.i = getelementptr i8, ptr addrspace(1) %add.ptr.i108140.i, i64 %mul23.i
   %call.ascast.i72.i = addrspacecast ptr addrspace(1) %add.ptr.i111.i to ptr addrspace(4)
-  tail call spir_func void @_Z38__spirv_CooperativeMatrixPrefetchINTEL(ptr addrspace(4) noundef %call.ascast.i72.i, i32 noundef 0, i32 noundef 0, i32 noundef 1, i32 noundef 1, i32 noundef 0, i64 noundef %mul22.i) #4
-  %call1.i73.i = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i8, 2, 48, 12, 3) @_Z32__spirv_CooperativeMatrixLoadKHR_2(ptr addrspace(4) noundef %call.ascast.i72.i, i64 noundef %mul22.i) #4
+  tail call spir_func void @_Z38__spirv_CooperativeMatrixPrefetchINTEL(ptr addrspace(4) noundef %call.ascast.i72.i, i32 noundef 12, i32 noundef 48, i32 noundef 0, i32 noundef 0, i64 noundef %mul22.i)
+  %call1.i73.i = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i8, 2, 48, 12, 1) @_Z32__spirv_CooperativeMatrixLoadKHR_2(ptr addrspace(4) noundef %call.ascast.i72.i, i32 noundef 0, i64 noundef %mul22.i) #4
   call void @llvm.lifetime.start.p0(i64 8, ptr nonnull %ref.tmp29.sroa.0.i)
-  %sub_c.sroa.0.i.0.sub_c.sroa.0.i.0.sub_c.sroa.0.0.sub_c.sroa.0.0.sub_c.sroa.0.0.125.i = load target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3), ptr %sub_c.sroa.0.i, align 8
-  %call.i77.i = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z34__spirv_CooperativeMatrixMulAddKHR(target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 3) noundef %call1.i.i, target("spirv.CooperativeMatrixKHR", i8, 2, 48, 12, 3) noundef %call1.i73.i, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) noundef %sub_c.sroa.0.i.0.sub_c.sroa.0.i.0.sub_c.sroa.0.0.sub_c.sroa.0.0.sub_c.sroa.0.0.125.i, i32 noundef 12) #4
-  store target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %call.i77.i, ptr %ref.tmp29.sroa.0.i, align 8
+  %sub_c.sroa.0.i.0.sub_c.sroa.0.i.0.sub_c.sroa.0.0.sub_c.sroa.0.0.sub_c.sroa.0.0.125.i = load target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2), ptr %sub_c.sroa.0.i, align 8
+  %call.i77.i = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z34__spirv_CooperativeMatrixMulAddKHR(target("spirv.CooperativeMatrixKHR", i8, 3, 12, 48, 0) noundef %call1.i.i, target("spirv.CooperativeMatrixKHR", i8, 2, 48, 12, 1) noundef %call1.i73.i, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) noundef %sub_c.sroa.0.i.0.sub_c.sroa.0.i.0.sub_c.sroa.0.0.sub_c.sroa.0.0.sub_c.sroa.0.0.125.i, i32 noundef 12) #4
+  store target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %call.i77.i, ptr %ref.tmp29.sroa.0.i, align 8
   %ref.tmp29.sroa.0.i.0.ref.tmp29.sroa.0.i.0.ref.tmp29.sroa.0.0.ref.tmp29.sroa.0.0.ref.tmp29.sroa.0.0..i = load i64, ptr %ref.tmp29.sroa.0.i, align 8
   store i64 %ref.tmp29.sroa.0.i.0.ref.tmp29.sroa.0.i.0.ref.tmp29.sroa.0.0.ref.tmp29.sroa.0.0.ref.tmp29.sroa.0.0..i, ptr %sub_c.sroa.0.i, align 8
   call void @llvm.lifetime.end.p0(i64 8, ptr nonnull %ref.tmp29.sroa.0.i)
@@ -131,31 +131,31 @@ _ZZZ15matrix_multiplyIiaLm24ELm96ELm24ELm96ELm24ELm24EEvR10big_matrixIT_XT5_EXT6
   %mul39.i = mul nuw i64 %div2452.i, 12
   %add.ptr.i81.i = getelementptr inbounds i32, ptr addrspace(1) %add.ptr.i.i, i64 %mul39.i
   %call.ascast.i.i = addrspacecast ptr addrspace(1) %add.ptr.i81.i to ptr addrspace(4)
-  %sub_c.sroa.0.i.0.sub_c.sroa.0.i.0.sub_c.sroa.0.0.sub_c.sroa.0.0.sub_c.sroa.0.0..i = load target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3), ptr %sub_c.sroa.0.i, align 8
-  tail call spir_func void @_Z33__spirv_CooperativeMatrixStoreKHR(ptr addrspace(4) noundef %call.ascast.i.i, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) noundef %sub_c.sroa.0.i.0.sub_c.sroa.0.i.0.sub_c.sroa.0.0.sub_c.sroa.0.0.sub_c.sroa.0.0..i, i32 noundef 0, i64 noundef %_arg_N, i32 noundef 1) #4
+  %sub_c.sroa.0.i.0.sub_c.sroa.0.i.0.sub_c.sroa.0.0.sub_c.sroa.0.0.sub_c.sroa.0.0..i = load target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2), ptr %sub_c.sroa.0.i, align 8
+  tail call spir_func void @_Z33__spirv_CooperativeMatrixStoreKHR(ptr addrspace(4) noundef %call.ascast.i.i, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) noundef %sub_c.sroa.0.i.0.sub_c.sroa.0.i.0.sub_c.sroa.0.0.sub_c.sroa.0.0.sub_c.sroa.0.0..i, i32 noundef 0, i64 noundef %_arg_N, i32 noundef 1) #4
   call void @llvm.lifetime.end.p0(i64 8, ptr nonnull %sub_c.sroa.0.i)
   ret void
 }
 
 ; Function Attrs: convergent
-declare dso_local spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z26__spirv_CompositeConstruct(i32 noundef) local_unnamed_addr #2
+declare dso_local spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z26__spirv_CompositeConstruct(i32 noundef) local_unnamed_addr #2
 
-declare dso_local spir_func noundef i32 @_Z34__spirv_CooperativeMatrixLengthKHR(target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 3) noundef)
+declare dso_local spir_func noundef i32 @_Z34__spirv_CooperativeMatrixLengthKHR(target("spirv.CooperativeMatrixKHR", i8, 3, 12, 48, 0) noundef)
 
-; Function Attrs: convergent
-declare dso_local spir_func void @_Z38__spirv_CooperativeMatrixPrefetchINTEL(ptr addrspace(4) noundef, i32 noundef, i32 noundef, i32 noundef, i32 noundef, i32 noundef, i64 noundef) local_unnamed_addr #2
+; Function Attrs: convergent nounwind
+declare dso_local spir_func void @_Z38__spirv_CooperativeMatrixPrefetchINTEL(ptr addrspace(4) noundef, i32 noundef, i32 noundef, i32 noundef, i32 noundef, i64 noundef) local_unnamed_addr #2
 
 ; Function Attrs: convergent
-declare dso_local spir_func noundef target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 3) @_Z32__spirv_CooperativeMatrixLoadKHR_1(ptr addrspace(4) noundef, i64 noundef, i32 noundef, i32 noundef) local_unnamed_addr #2
+declare dso_local spir_func noundef target("spirv.CooperativeMatrixKHR", i8, 3, 12, 48, 0) @_Z32__spirv_CooperativeMatrixLoadKHR_1(ptr addrspace(4) noundef, i32 noundef, i64 noundef, i32 noundef) local_unnamed_addr #2
 
 ; Function Attrs: convergent
-declare dso_local spir_func noundef target("spirv.CooperativeMatrixKHR", i8, 2, 48, 12, 3) @_Z32__spirv_CooperativeMatrixLoadKHR_2(ptr addrspace(4) noundef, i64 noundef) local_unnamed_addr #2
+declare dso_local spir_func noundef target("spirv.CooperativeMatrixKHR", i8, 2, 48, 12, 1) @_Z32__spirv_CooperativeMatrixLoadKHR_2(ptr addrspace(4) noundef, i32 noundef, i64 noundef) local_unnamed_addr #2
 
 ; Function Attrs: convergent
-declare dso_local spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z34__spirv_CooperativeMatrixMulAddKHR(target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 3) noundef, target("spirv.CooperativeMatrixKHR", i8, 2, 48, 12, 3) noundef, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) noundef, i32 noundef) local_unnamed_addr #2
+declare dso_local spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z34__spirv_CooperativeMatrixMulAddKHR(target("spirv.CooperativeMatrixKHR", i8, 3, 12, 48, 0) noundef, target("spirv.CooperativeMatrixKHR", i8, 2, 48, 12, 1) noundef, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) noundef, i32 noundef) local_unnamed_addr #2
 
 ; Function Attrs: convergent
-declare dso_local spir_func void @_Z33__spirv_CooperativeMatrixStoreKHR(ptr addrspace(4) noundef, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) noundef, i32 noundef, i64 noundef, i32 noundef) local_unnamed_addr #2
+declare dso_local spir_func void @_Z33__spirv_CooperativeMatrixStoreKHR(ptr addrspace(4) noundef, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) noundef, i32 noundef, i64 noundef, i32 noundef) local_unnamed_addr #2
 
 ; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
 declare void @llvm.lifetime.start.p0(i64 immarg, ptr nocapture) #3
diff --git a/test/extensions/INTEL/SPV_INTEL_joint_matrix/joint_matrix.ll b/test/extensions/INTEL/SPV_INTEL_joint_matrix/joint_matrix.ll
index 72010ed93e..71ea0a8afe 100644
--- a/test/extensions/INTEL/SPV_INTEL_joint_matrix/joint_matrix.ll
+++ b/test/extensions/INTEL/SPV_INTEL_joint_matrix/joint_matrix.ll
@@ -1,207 +1,158 @@
-; RUN: llvm-as -opaque-pointers=0 < %s -o %t.bc
-; RUN: llvm-spirv %t.bc -opaque-pointers=0 -spirv-ext=+SPV_INTEL_joint_matrix -o %t.spv
-; RUN: llvm-spirv %t.spv -to-text -o - | FileCheck %s --check-prefix=CHECK-SPIRV
+; RUN: llvm-as < %s -o %t.bc
+; RUN: llvm-spirv %t.bc --spirv-ext=+SPV_INTEL_joint_matrix -o %t.spv
+; RUN: llvm-spirv %t.spv -to-text -o %t.spt
+; RUN: FileCheck < %t.spt %s --check-prefix=CHECK-SPIRV
 
 ; RUN: llvm-spirv -r %t.spv -o %t.rev.bc
-; RUN: llvm-dis -opaque-pointers=0 %t.rev.bc -o - | FileCheck %s --check-prefix=CHECK-LLVM
-
-; CHECK-SPIRV: Capability JointMatrixINTEL
-; CHECK-SPIRV: Extension "SPV_INTEL_joint_matrix"
-; CHECK-SPIRV: Name [[#Kernel:]] "_ZTSZ4mainE11matrix_test"
-
-; CHECK-SPIRV-DAG: TypeInt [[#ShortTy:]] 16 0
-; CHECK-SPIRV-DAG: TypeInt [[#CharTy:]] 8 0
-; CHECK-SPIRV-DAG: TypeInt [[#IntTy:]] 32 0
-; CHECK-SPIRV-DAG: Constant [[#IntTy]] [[#Zero:]] 0
-; CHECK-SPIRV-DAG: Constant [[#IntTy]] [[#Two:]] 2
-; CHECK-SPIRV-DAG: Constant [[#IntTy]] [[#Three:]] 3
-; CHECK-SPIRV-DAG: Constant [[#IntTy]] [[#Sixteen:]] 16
-; CHECK-SPIRV-DAG: Constant [[#IntTy]] [[#FortyTwo:]] 42
-; CHECK-SPIRV: TypeJointMatrixINTEL [[#CTy:]] [[#ShortTy]] [[#Two]] [[#Two]] [[#Zero]] [[#Three]]
-; CHECK-SPIRV: TypeJointMatrixINTEL [[#ATy:]] [[#CharTy]] [[#Two]] [[#Sixteen]] [[#Zero]] [[#Three]] [[#Zero]]
-; CHECK-SPIRV: TypeJointMatrixINTEL [[#BTy:]] [[#CharTy]] [[#Sixteen]] [[#Two]] [[#Three]] [[#Three]]
-
-; CHECK-SPIRV: Function [[#]] [[#Kernel]]
-; CHECK-SPIRV: FunctionParameter
-; CHECK-SPIRV: FunctionParameter [[#]] [[#Stride:]]
-
-; CHECK-SPIRV: Label [[#Entry:]]
-; CHECK-SPIRV: JointMatrixLoadINTEL [[#CTy]] [[#CLoaded:]] [[#Cptr:]] [[#Stride]] [[#Zero]] [[#Three]] [[#Zero]]
-
-; CHECK-SPIRV: Phi [[#CTy]] [[#C:]] [[#CLoaded]] [[#Entry]] [[#CMad:]] [[#ForBody:]]
-
-; CHECK-SPIRV: Label [[#ForBody]]
-; CHECK-SPIRV: JointMatrixLoadINTEL [[#ATy]] [[#A:]] [[#Aptr:]] [[#Stride]] [[#Zero]] [[#Three]] [[#Zero]]
-; CHECK-SPIRV: JointMatrixLoadINTEL [[#BTy]] [[#B:]] [[#Bptr:]] [[#Stride]] [[#Zero]] [[#Three]] [[#Zero]]
-; CHECK-SPIRV: JointMatrixMadINTEL [[#CTy]] [[#CMad]] [[#A]] [[#B]] [[#C]] [[#Three]]
-; CHECK-SPIRV: JointMatrixSUMadINTEL [[#CTy]] [[#UnusedMad1:]] [[#A]] [[#B]] [[#C]] [[#Three]]
-; CHECK-SPIRV: JointMatrixUSMadINTEL [[#CTy]] [[#UnusedMad2:]] [[#A]] [[#B]] [[#C]] [[#Three]]
-; CHECK-SPIRV: JointMatrixUUMadINTEL [[#CTy]] [[#UnusedMad3:]] [[#A]] [[#B]] [[#C]] [[#Three]]
-
-; CHECK-SPIRV: JointMatrixStoreINTEL [[#Cptr:]] [[#C]] [[#Stride]] [[#Zero]] [[#Three]] [[#Zero]]
-; CHECK-SPIRV: CompositeConstruct [[#CTy]] [[#Cnew:]] [[#FortyTwo]]
-; CHECK-SPIRV: Store [[#PtrToZero:]] [[#Zero]]
-; CHECK-SPIRV: Load [[#]] [[#ZeroLoad:]] [[#PtrToZero]]
-; CHECK-SPIRV: CompositeConstruct [[#CTy]] [[#CnewLoad:]] [[#ZeroLoad]]
-
-
-; CHECK-LLVM: %spirv.JointMatrixINTEL._short_2_2_0_3
-; CHECK-LLVM: %spirv.JointMatrixINTEL._char_2_16_0_3_0
-; CHECK-LLVM: %spirv.JointMatrixINTEL._char_16_2_3_3
-
-; CHECK-LLVM: [[CLoaded:%.*]] = call spir_func %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(1)* @_Z77__spirv_JointMatrixLoadINTEL_RPU3AS139__spirv_JointMatrixINTEL__short_2_2_0_3PU3AS4sliii(i16 addrspace(4)* [[CPtr:%.*]], i64 [[Stride:%.*]], i32 0, i32 3, i32 0)
-; CHECK-LLVM: [[C:%.*]] = phi %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(1)* [ [[CLoaded]], %entry ], [ [[CMad:%.*]], %for.body.i ]
-; CHECK-LLVM: [[A:%.*]] = call spir_func %spirv.JointMatrixINTEL._char_2_16_0_3_0 addrspace(1)* @_Z79__spirv_JointMatrixLoadINTEL_RPU3AS141__spirv_JointMatrixINTEL__char_2_16_0_3_0PU3AS4cliii(i8 addrspace(4)* [[APtr:%.*]], i64 [[Stride]], i32 0, i32 3, i32 0)
-; CHECK-LLVM: [[B:%.*]] = call spir_func %spirv.JointMatrixINTEL._char_16_2_3_3 addrspace(1)* @_Z77__spirv_JointMatrixLoadINTEL_RPU3AS139__spirv_JointMatrixINTEL__char_16_2_3_3PU3AS4cliii(i8 addrspace(4)* [[BPtr:%.*]], i64 [[Stride]], i32 0, i32 3, i32 0)
-; CHECK-LLVM: [[CMad1:%.*]] = call spir_func %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(1)* @_Z27__spirv_JointMatrixMadINTELPU3AS141__spirv_JointMatrixINTEL__char_2_16_0_3_0PU3AS139__spirv_JointMatrixINTEL__char_16_2_3_3PU3AS139__spirv_JointMatrixINTEL__short_2_2_0_3i(%spirv.JointMatrixINTEL._char_2_16_0_3_0 addrspace(1)* [[A]], %spirv.JointMatrixINTEL._char_16_2_3_3 addrspace(1)* [[B]], %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(1)* [[C]], i32 3)
-; CHECK-LLVM: [[CMad2:%.*]] = call spir_func %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(1)* @_Z29__spirv_JointMatrixSUMadINTELPU3AS141__spirv_JointMatrixINTEL__char_2_16_0_3_0PU3AS139__spirv_JointMatrixINTEL__char_16_2_3_3PU3AS139__spirv_JointMatrixINTEL__short_2_2_0_3i(%spirv.JointMatrixINTEL._char_2_16_0_3_0 addrspace(1)* [[A]], %spirv.JointMatrixINTEL._char_16_2_3_3 addrspace(1)* [[B]], %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(1)* [[C]], i32 3)
-; CHECK-LLVM: [[CMad3:%.*]] = call spir_func %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(1)* @_Z29__spirv_JointMatrixUSMadINTELPU3AS141__spirv_JointMatrixINTEL__char_2_16_0_3_0PU3AS139__spirv_JointMatrixINTEL__char_16_2_3_3PU3AS139__spirv_JointMatrixINTEL__short_2_2_0_3i(%spirv.JointMatrixINTEL._char_2_16_0_3_0 addrspace(1)* [[A]], %spirv.JointMatrixINTEL._char_16_2_3_3 addrspace(1)* [[B]], %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(1)* [[C]], i32 3)
-; CHECK-LLVM: [[CMad4:%.*]] = call spir_func %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(1)* @_Z29__spirv_JointMatrixUUMadINTELPU3AS141__spirv_JointMatrixINTEL__char_2_16_0_3_0PU3AS139__spirv_JointMatrixINTEL__char_16_2_3_3PU3AS139__spirv_JointMatrixINTEL__short_2_2_0_3i(%spirv.JointMatrixINTEL._char_2_16_0_3_0 addrspace(1)* [[A]], %spirv.JointMatrixINTEL._char_16_2_3_3 addrspace(1)* [[B]], %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(1)* [[C]], i32 3)
-
-; CHECK-LLVM: call spir_func void @_Z29__spirv_JointMatrixStoreINTELPU3AS4sPU3AS139__spirv_JointMatrixINTEL__short_2_2_0_3liii(i16 addrspace(4)* [[CPtr]], %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(1)* [[C]], i64 [[Stride]], i32 0, i32 3, i32 0)
-; CHECK-LLVM: call spir_func %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(1)* @_Z26__spirv_CompositeConstructi(i32 42)
-; CHECK-LLVM: store i32 0, i32 addrspace(4)* [[StoredZero:%.*]], align 4
-; CHECK-LLVM: [[LoadedZero:%.*]] = load i32, i32 addrspace(4)* [[StoredZero]], align 8
-; CHECK-LLVM: call spir_func %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(1)* @_Z26__spirv_CompositeConstructi(i32 [[LoadedZero]])
-
-; ModuleID = 'joint_matrix_test-sycl-spir64-unknown-unknown.bc'
-source_filename = "./joint_matrix_test.cpp"
+; RUN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM
+
+; CHECK-SPIRV-DAG: Capability JointMatrixINTEL
+; CHECK-SPIRV-DAG: Extension "SPV_INTEL_joint_matrix"
+; CHECK-SPIRV-DAG: TypeInt [[#Int8Ty:]] 8 0
+; CHECK-SPIRV-DAG: TypeInt [[#Int32Ty:]] 32 0
+; CHECK-SPIRV-DAG: Constant [[#Int32Ty]] [[#Const12:]] 12
+; CHECK-SPIRV-DAG: Constant [[#Int32Ty]] [[#Const3:]] 3
+; CHECK-SPIRV-DAG: Constant [[#Int32Ty]] [[#Const2:]] 2
+; CHECK-SPIRV-DAG: Constant [[#Int32Ty]] [[#Const0:]] 0
+; CHECK-SPIRV-DAG: Constant [[#Int32Ty]] [[#Const48:]] 48
+; CHECK-SPIRV-DAG: Constant [[#Int32Ty]] [[#Const1:]] 1
+; CHECK-SPIRV-DAG: TypeJointMatrixINTEL [[#MatTy1:]] [[#Int32Ty]] [[#Const12]] [[#Const12]] [[#Const3]] [[#Const3]] [[#Const2]]
+; CHECK-SPIRV-DAG: TypeJointMatrixINTEL [[#MatTy2:]] [[#Int8Ty]] [[#Const12]] [[#Const48]] [[#Const0]] [[#Const3]] [[#Const0]]
+; CHECK-SPIRV-DAG: TypeJointMatrixINTEL [[#MatTy3:]] [[#Int8Ty]] [[#Const48]] [[#Const12]] [[#Const2]] [[#Const3]] [[#Const1]]
+; CHECK-SPIRV: CompositeConstruct [[#MatTy1]]
+; CHECK-SPIRV: JointMatrixLoadINTEL [[#MatTy2]]
+; CHECK-SPIRV: JointMatrixLoadINTEL [[#MatTy3]]
+; CHECK-SPIRV: JointMatrixMadINTEL [[#MatTy1]]
+; CHECK-SPIRV: JointMatrixStoreINTEL
+
+; CHECK-LLVM: call spir_func target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2) @_Z26__spirv_CompositeConstructi(i32 0)
+; CHECK-LLVM: call spir_func target("spirv.JointMatrixINTEL", i8, 12, 48, 0, 3, 0) @_Z80__spirv_JointMatrixLoadINTEL_RPU3AS142__spirv_JointMatrixINTEL__char_12_48_0_3_0PU3AS4cliii
+; CHECK-LLVM: call spir_func target("spirv.JointMatrixINTEL", i8, 48, 12, 2, 3, 1) @_Z80__spirv_JointMatrixLoadINTEL_RPU3AS142__spirv_JointMatrixINTEL__char_48_12_2_3_1PU3AS4cliii
+; CHECK-LLVM: call spir_func target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2) @_Z27__spirv_JointMatrixMadINTELPU3AS142__spirv_JointMatrixINTEL__char_12_48_0_3_0PU3AS142__spirv_JointMatrixINTEL__char_48_12_2_3_1PU3AS142__spirv_JointMatrixINTEL__uint_12_12_3_3_2i(target("spirv.JointMatrixINTEL", i8, 12, 48, 0, 3, 0) %{{.*}}, target("spirv.JointMatrixINTEL", i8, 48, 12, 2, 3, 1) %{{.*}}, target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2)
+; CHECK-LLVM: call spir_func void @_Z29__spirv_JointMatrixStoreINTELPU3AS4iPU3AS142__spirv_JointMatrixINTEL__uint_12_12_3_3_2liii(ptr addrspace(4) %call.ascast.i.i, target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2)
+
+; ModuleID = 'test-matrix-opaque.bc'
+source_filename = "matrix-int8-test.cpp"
 target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-n8:16:32:64"
 target triple = "spir64-unknown-unknown"
 
-%spirv.JointMatrixINTEL._short_2_2_0_3 = type { [2 x [2 x [1 x [4 x [4 x i16]]]]]* }
-%spirv.JointMatrixINTEL._char_2_16_0_3_0 = type { [2 x [16 x [1 x [4 x [1 x i8]]]]]* }
-%spirv.JointMatrixINTEL._char_16_2_3_3 = type { [16 x [2 x [4 x [4 x i8]]]]* }
+%"class.sycl::_V1::range" = type { %"class.sycl::_V1::detail::array" }
+%"class.sycl::_V1::detail::array" = type { [2 x i64] }
+%"class.sycl::_V1::id" = type { %"class.sycl::_V1::detail::array" }
 
-$_ZTSZ4mainE11matrix_test = comdat any
+$_ZTSZZ15matrix_multiplyIiaLm24ELm96ELm24ELm96ELm24ELm24EEvR10big_matrixIT_XT5_EXT6_EERS0_IT0_XT1_EXT2_EERS0_IS4_XT3_EXT4_EEENKUlRN4sycl3_V17handlerEE_clESC_E7imatrix = comdat any
 
 @__spirv_BuiltInGlobalInvocationId = external dso_local local_unnamed_addr addrspace(1) constant <3 x i64>, align 32
 @__spirv_BuiltInLocalInvocationId = external dso_local local_unnamed_addr addrspace(1) constant <3 x i64>, align 32
 
 ; Function Attrs: convergent norecurse
-define weak_odr dso_local spir_kernel void @_ZTSZ4mainE11matrix_test(i16 addrspace(1)* %_arg_, i64 %_arg_1, i8 addrspace(1)* %_arg_3, i8 addrspace(1)* %_arg_5) local_unnamed_addr #0 comdat !kernel_arg_buffer_location !5 !intel_reqd_sub_group_size !6 {
+define weak_odr dso_local spir_kernel void @_ZTSZZ15matrix_multiplyIiaLm24ELm96ELm24ELm96ELm24ELm24EEvR10big_matrixIT_XT5_EXT6_EERS0_IT0_XT1_EXT2_EERS0_IS4_XT3_EXT4_EEENKUlRN4sycl3_V17handlerEE_clESC_E7imatrix(ptr addrspace(1) noundef align 1 %_arg_accA, ptr addrspace(1) noundef align 1 %_arg_accB, ptr noundef byval(%"class.sycl::_V1::range") align 8 %_arg_accB5, ptr noundef byval(%"class.sycl::_V1::id") align 8 %_arg_accB6, ptr addrspace(1) noundef align 4 %_arg_accC, i64 noundef %_arg_N, i64 noundef %_arg_K) local_unnamed_addr #0 comdat {
 entry:
-  %0 = load <3 x i64>, <3 x i64> addrspace(4)* addrspacecast (<3 x i64> addrspace(1)* @__spirv_BuiltInGlobalInvocationId to <3 x i64> addrspace(4)*), align 32, !noalias !7
-  %1 = extractelement <3 x i64> %0, i64 1
-  %2 = extractelement <3 x i64> %0, i64 0
-  %3 = load <3 x i64>, <3 x i64> addrspace(4)* addrspacecast (<3 x i64> addrspace(1)* @__spirv_BuiltInLocalInvocationId to <3 x i64> addrspace(4)*), align 32, !noalias !14
-  %4 = extractelement <3 x i64> %3, i64 1
-  %5 = extractelement <3 x i64> %3, i64 0
-  %cmp.i.i = icmp ult i64 %1, 2147483648
-  tail call void @llvm.assume(i1 %cmp.i.i)
-  %cmp.i45.i = icmp ult i64 %2, 2147483648
-  tail call void @llvm.assume(i1 %cmp.i45.i)
-  %cmp.i43.i = icmp ult i64 %4, 2147483648
-  tail call void @llvm.assume(i1 %cmp.i43.i)
-  %sub.i = sub nsw i64 %1, %4
-  %cmp.i41.i = icmp ult i64 %5, 2147483648
-  tail call void @llvm.assume(i1 %cmp.i41.i)
-  %sub5.i = sub nsw i64 %2, %5
-  %mul6.i = shl nsw i64 %sub.i, 6
-  %add.ptr.i51 = getelementptr inbounds i16, i16 addrspace(1)* %_arg_, i64 %mul6.i
-  %add.ptr7.i52 = getelementptr inbounds i16, i16 addrspace(1)* %add.ptr.i51, i64 %sub5.i
-  %add.ptr7.i = addrspacecast i16 addrspace(1)* %add.ptr7.i52 to i16 addrspace(4)*
-  %call8.i = tail call spir_func %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(4)* @_Z28__spirv_JointMatrixLoadINTELIsLm2ELm2ELN5__spv12MatrixLayoutE0ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEEPS5_mS1_S3_i(i16 addrspace(4)* %add.ptr7.i, i64 %_arg_1, i32 0, i32 3, i32 0) #3
-  %add.ptr11.i53 = getelementptr inbounds i8, i8 addrspace(1)* %_arg_3, i64 %mul6.i
-  %add.ptr16.i55 = getelementptr inbounds i8, i8 addrspace(1)* %_arg_5, i64 %sub5.i
+  %sub_c.sroa.0.i = alloca target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2), align 8
+  %ref.tmp29.sroa.0.i = alloca target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2), align 8
+  %agg.tmp15.sroa.0.sroa.2.0..sroa_idx = getelementptr inbounds %"class.sycl::_V1::range", ptr %_arg_accB5, i64 0, i32 0, i32 0, i64 1
+  %agg.tmp15.sroa.0.sroa.2.0.copyload = load i64, ptr %agg.tmp15.sroa.0.sroa.2.0..sroa_idx, align 8
+  %0 = getelementptr inbounds %"class.sycl::_V1::id", ptr %_arg_accB6, i64 0, i32 0, i32 0, i64 0
+  %agg.tmp16.sroa.0.sroa.0.0.copyload = load i64, ptr %0, align 8
+  %agg.tmp16.sroa.0.sroa.2.0..sroa_idx = getelementptr inbounds %"class.sycl::_V1::id", ptr %_arg_accB6, i64 0, i32 0, i32 0, i64 1
+  %agg.tmp16.sroa.0.sroa.2.0.copyload = load i64, ptr %agg.tmp16.sroa.0.sroa.2.0..sroa_idx, align 8
+  %mul.i4.i.i.i.i45 = mul i64 %agg.tmp16.sroa.0.sroa.0.0.copyload, %agg.tmp15.sroa.0.sroa.2.0.copyload
+  %add.i6.i.i.i.i46 = add i64 %mul.i4.i.i.i.i45, %agg.tmp16.sroa.0.sroa.2.0.copyload
+  %add.ptr.i47 = getelementptr inbounds i8, ptr addrspace(1) %_arg_accB, i64 %add.i6.i.i.i.i46
+  %1 = load <3 x i64>, ptr addrspace(1) @__spirv_BuiltInGlobalInvocationId, align 32
+  %2 = extractelement <3 x i64> %1, i64 1
+  %3 = extractelement <3 x i64> %1, i64 0
+  %4 = load <3 x i64>, ptr addrspace(1) @__spirv_BuiltInLocalInvocationId, align 32
+  %5 = extractelement <3 x i64> %4, i64 1
+  %6 = extractelement <3 x i64> %4, i64 0
+  %cmp.i.i = icmp ult i64 %2, 2147483648
+  %cmp.i54.i = icmp ult i64 %3, 2147483648
+  %cmp.i56.i = icmp ult i64 %5, 2147483648
+  %sub.i = sub nsw i64 %2, %5
+  %cmp.i58.i = icmp ult i64 %6, 2147483648
+  %sub5.i = sub nsw i64 %3, %6
+  %sub_c.sroa.0.i.0.i.0..sroa_cast = bitcast ptr %sub_c.sroa.0.i to ptr
+  call void @llvm.lifetime.start.p0(i64 8, ptr nonnull %sub_c.sroa.0.i.0.i.0..sroa_cast)
+  %call.i.i = tail call spir_func noundef target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2) @_Z26__spirv_CompositeConstructIiLm12ELm12ELN5__spv9MatrixUseE2ELNS0_12MatrixLayoutE3ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT3_EXT4_EXT2_EEES6_(i32 noundef 0) #4
+  store target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2) %call.i.i, ptr %sub_c.sroa.0.i, align 8
+  %mul.i = mul nsw i64 %sub.i, 12
+  %div2452.i = lshr i64 %sub5.i, 4
+  %mul26.i = mul i64 %div2452.i, 48
+  %div.i = udiv i64 %_arg_K, 48
+  %mul11.i = mul i64 %mul.i, %_arg_K
+  %add.ptr.i93.i = getelementptr inbounds i8, ptr addrspace(1) %_arg_accA, i64 %mul11.i
+  %idx.neg.i.i104.i = sub i64 0, %add.i6.i.i.i.i46
+  %add.ptr.i.i105141.i = getelementptr i8, ptr addrspace(1) %add.ptr.i47, i64 %mul26.i
+  %mul22.i = shl i64 %_arg_N, 2
+  %add.ptr.i108140.i = getelementptr i8, ptr addrspace(1) %add.ptr.i.i105141.i, i64 %idx.neg.i.i104.i
+  %ref.tmp29.sroa.0.i.0.i.0..sroa_cast = bitcast ptr %ref.tmp29.sroa.0.i to ptr
+  %7 = bitcast ptr %ref.tmp29.sroa.0.i to ptr
+  %8 = bitcast ptr %sub_c.sroa.0.i to ptr
   br label %for.cond.i
 
 for.cond.i:                                       ; preds = %for.body.i, %entry
   %k.0.i = phi i32 [ 0, %entry ], [ %add.i, %for.body.i ]
-  %C.0.i = phi %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(4)* [ %call8.i, %entry ], [ %call19.i, %for.body.i ]
-  %cmp.i = icmp ult i32 %k.0.i, 32
-  br i1 %cmp.i, label %for.body.i, label %_ZZ4mainENKUlN2cl4sycl7nd_itemILi2EEEE_clES2_.exit
+  %conv.i = zext i32 %k.0.i to i64
+  %cmp.i = icmp ugt i64 %div.i, %conv.i
+  br i1 %cmp.i, label %for.body.i, label %_ZZZ15matrix_multiplyIiaLm24ELm96ELm24ELm96ELm24ELm24EEvR10big_matrixIT_XT5_EXT6_EERS0_IT0_XT1_EXT2_EERS0_IS4_XT3_EXT4_EEENKUlRN4sycl3_V17handlerEE_clESC_ENKUlNSA_7nd_itemILi2EEEE_clESF_.exit
 
 for.body.i:                                       ; preds = %for.cond.i
-  %idx.ext.i = zext i32 %k.0.i to i64
-  %add.ptr12.i54 = getelementptr inbounds i8, i8 addrspace(1)* %add.ptr11.i53, i64 %idx.ext.i
-  %add.ptr12.i = addrspacecast i8 addrspace(1)* %add.ptr12.i54 to i8 addrspace(4)*
-  %call13.i = tail call spir_func %spirv.JointMatrixINTEL._char_2_16_0_3_0 addrspace(4)* @_Z28__spirv_JointMatrixLoadINTELIaLm2ELm16ELN5__spv12MatrixLayoutE0ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEEPS5_mS1_S3_i(i8 addrspace(4)* %add.ptr12.i, i64 %_arg_1, i32 0, i32 3, i32 0) #3
-  %mul14.i = shl nuw nsw i32 %k.0.i, 5
-  %idx.ext15.i = zext i32 %mul14.i to i64
-  %add.ptr17.i56 = getelementptr inbounds i8, i8 addrspace(1)* %add.ptr16.i55, i64 %idx.ext15.i
-  %add.ptr17.i = addrspacecast i8 addrspace(1)* %add.ptr17.i56 to i8 addrspace(4)*
-  %call18.i = tail call spir_func %spirv.JointMatrixINTEL._char_16_2_3_3 addrspace(4)* @_Z28__spirv_JointMatrixLoadINTELIaLm16ELm2ELN5__spv12MatrixLayoutE3ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEEPS5_mS1_S3_i(i8 addrspace(4)* %add.ptr17.i, i64 %_arg_1, i32 0, i32 3, i32 0) #3
-  %call19.i = tail call spir_func %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(4)* @_Z27__spirv_JointMatrixMadINTELIasLm2ELm16ELm2ELN5__spv12MatrixLayoutE0ELS1_3ELS1_0ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT0_XT1_EXT3_EXT6_EXT7_EEEPNS4_IT_XT1_EXT2_EXT4_EXT7_EEEPNS4_IS8_XT2_EXT3_EXT5_EXT7_EEES7_S3_(%spirv.JointMatrixINTEL._char_2_16_0_3_0 addrspace(4)* %call13.i, %spirv.JointMatrixINTEL._char_16_2_3_3 addrspace(4)* %call18.i, %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(4)* %C.0.i, i32 3) #3
-  %call20.i = tail call spir_func %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(4)* @_Z29__spirv_JointMatrixSUMadINTELIasLm2ELm16ELm2ELN5__spv12MatrixLayoutE0ELS1_3ELS1_0ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT0_XT1_EXT3_EXT6_EXT7_EEEPNS4_IT_XT1_EXT2_EXT4_EXT7_EEEPNS4_IS8_XT2_EXT3_EXT5_EXT7_EEES7_S3_(%spirv.JointMatrixINTEL._char_2_16_0_3_0 addrspace(4)* %call13.i, %spirv.JointMatrixINTEL._char_16_2_3_3 addrspace(4)* %call18.i, %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(4)* %C.0.i, i32 3) #3
-  %call21.i = tail call spir_func %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(4)* @_Z29__spirv_JointMatrixUSMadINTELIasLm2ELm16ELm2ELN5__spv12MatrixLayoutE0ELS1_3ELS1_0ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT0_XT1_EXT3_EXT6_EXT7_EEEPNS4_IT_XT1_EXT2_EXT4_EXT7_EEEPNS4_IS8_XT2_EXT3_EXT5_EXT7_EEES7_S3_(%spirv.JointMatrixINTEL._char_2_16_0_3_0 addrspace(4)* %call13.i, %spirv.JointMatrixINTEL._char_16_2_3_3 addrspace(4)* %call18.i, %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(4)* %C.0.i, i32 3) #3
-  %call22.i = tail call spir_func %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(4)* @_Z29__spirv_JointMatrixUUMadINTELIasLm2ELm16ELm2ELN5__spv12MatrixLayoutE0ELS1_3ELS1_0ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT0_XT1_EXT3_EXT6_EXT7_EEEPNS4_IT_XT1_EXT2_EXT4_EXT7_EEEPNS4_IS8_XT2_EXT3_EXT5_EXT7_EEES7_S3_(%spirv.JointMatrixINTEL._char_2_16_0_3_0 addrspace(4)* %call13.i, %spirv.JointMatrixINTEL._char_16_2_3_3 addrspace(4)* %call18.i, %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(4)* %C.0.i, i32 3) #3
-  %add.i = add nuw nsw i32 %k.0.i, 16
-  br label %for.cond.i, !llvm.loop !19
-
-_ZZ4mainENKUlN2cl4sycl7nd_itemILi2EEEE_clES2_.exit: ; preds = %for.cond.i
-  tail call spir_func void @_Z29__spirv_JointMatrixStoreINTELIsLm2ELm2ELN5__spv12MatrixLayoutE0ELNS0_5Scope4FlagE3EEvPT_PNS0_24__spirv_JointMatrixINTELIS4_XT0_EXT1_EXT2_EXT3_EEEmS1_S3_i(i16 addrspace(4)* %add.ptr7.i, %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(4)* %C.0.i, i64 %_arg_1, i32 0, i32 3, i32 0) #3
-  %C.0.i.new = call spir_func %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(4)* @_Z26__spirv_CompositeConstructi(i32 42) #1
-  %ref.tmp = alloca i32, align 4
-  %ref.tmp.ascast = addrspacecast i32* %ref.tmp to i32 addrspace(4)*
-  store i32 0, i32 addrspace(4)* %ref.tmp.ascast, align 4
-  %zero = load i32, i32 addrspace(4)* %ref.tmp.ascast, align 8
-  %C.0.i.new.load = call spir_func %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(4)* @_Z26__spirv_CompositeConstructi(i32 %zero) #1
+  %mul12.i = mul nsw i32 %k.0.i, 48
+  %conv13.i = zext i32 %mul12.i to i64
+  %add.ptr.i96.i = getelementptr inbounds i8, ptr addrspace(1) %add.ptr.i93.i, i64 %conv13.i
+  %call.ascast.i66.i = addrspacecast ptr addrspace(1) %add.ptr.i96.i to ptr addrspace(4)
+  %call1.i.i = tail call spir_func noundef target("spirv.JointMatrixINTEL", i8, 12, 48, 0, 3, 0) @_Z28__spirv_JointMatrixLoadINTELIaLm12ELm48ELN5__spv9MatrixUseE0ELNS0_12MatrixLayoutE0ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT3_EXT4_EXT2_EEEPS6_mS2_S4_i(ptr addrspace(4) noundef %call.ascast.i66.i, i64 noundef %_arg_K, i32 noundef 0, i32 noundef 3, i32 noundef 0) #4
+  %div20.i = mul nsw i32 %k.0.i, 12
+  %conv21.i = zext i32 %div20.i to i64
+  %mul23.i = mul i64 %mul22.i, %conv21.i
+  %add.ptr.i111.i = getelementptr i8, ptr addrspace(1) %add.ptr.i108140.i, i64 %mul23.i
+  %call.ascast.i72.i = addrspacecast ptr addrspace(1) %add.ptr.i111.i to ptr addrspace(4)
+  %call1.i73.i = tail call spir_func noundef target("spirv.JointMatrixINTEL", i8, 48, 12, 2, 3, 1) @_Z28__spirv_JointMatrixLoadINTELIaLm48ELm12ELN5__spv9MatrixUseE1ELNS0_12MatrixLayoutE2ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT3_EXT4_EXT2_EEEPS6_mS2_S4_i(ptr addrspace(4) noundef %call.ascast.i72.i, i64 noundef %mul22.i, i32 noundef 2, i32 noundef 3, i32 noundef 0) #4
+  call void @llvm.lifetime.start.p0(i64 8, ptr nonnull %ref.tmp29.sroa.0.i.0.i.0..sroa_cast)
+  %sub_c.sroa.0.i.0.sub_c.sroa.0.i.0.sub_c.sroa.0.0.sub_c.sroa.0.0.sub_c.sroa.0.0.125.i = load target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2), ptr %sub_c.sroa.0.i, align 8
+  %call.i77.i = tail call spir_func noundef target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2) @_Z27__spirv_JointMatrixMadINTELIaiLm12ELm48ELm12ELN5__spv9MatrixUseE0ELS1_1ELS1_2ELNS0_12MatrixLayoutE0ELS2_2ELS2_3ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT0_XT1_EXT3_EXT9_EXT10_EXT6_EEEPNS5_IT_XT1_EXT2_EXT7_EXT10_EXT4_EEEPNS5_IS9_XT2_EXT3_EXT8_EXT10_EXT5_EEES8_S4_(target("spirv.JointMatrixINTEL", i8, 12, 48, 0, 3, 0) noundef %call1.i.i, target("spirv.JointMatrixINTEL", i8, 48, 12, 2, 3, 1) noundef %call1.i73.i, target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2) noundef %sub_c.sroa.0.i.0.sub_c.sroa.0.i.0.sub_c.sroa.0.0.sub_c.sroa.0.0.sub_c.sroa.0.0.125.i, i32 noundef 3) #4
+  store target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2) %call.i77.i, ptr %ref.tmp29.sroa.0.i, align 8
+  %ref.tmp29.sroa.0.i.0.ref.tmp29.sroa.0.i.0.ref.tmp29.sroa.0.0.ref.tmp29.sroa.0.0.ref.tmp29.sroa.0.0..i = load i64, ptr %7, align 8
+  store i64 %ref.tmp29.sroa.0.i.0.ref.tmp29.sroa.0.i.0.ref.tmp29.sroa.0.0.ref.tmp29.sroa.0.0.ref.tmp29.sroa.0.0..i, ptr %8, align 8
+  call void @llvm.lifetime.end.p0(i64 8, ptr nonnull %ref.tmp29.sroa.0.i.0.i.0..sroa_cast)
+  %add.i = add nuw nsw i32 %k.0.i, 1
+  br label %for.cond.i
 
+_ZZZ15matrix_multiplyIiaLm24ELm96ELm24ELm96ELm24ELm24EEvR10big_matrixIT_XT5_EXT6_EERS0_IT0_XT1_EXT2_EERS0_IS4_XT3_EXT4_EEENKUlRN4sycl3_V17handlerEE_clESC_ENKUlNSA_7nd_itemILi2EEEE_clESF_.exit: ; preds = %for.cond.i
+  %mul37.i = mul i64 %mul.i, %_arg_N
+  %add.ptr.i.i = getelementptr inbounds i32, ptr addrspace(1) %_arg_accC, i64 %mul37.i
+  %mul39.i = mul nuw i64 %div2452.i, 12
+  %add.ptr.i81.i = getelementptr inbounds i32, ptr addrspace(1) %add.ptr.i.i, i64 %mul39.i
+  %call.ascast.i.i = addrspacecast ptr addrspace(1) %add.ptr.i81.i to ptr addrspace(4)
+  %sub_c.sroa.0.i.0.sub_c.sroa.0.i.0.sub_c.sroa.0.0.sub_c.sroa.0.0.sub_c.sroa.0.0..i = load target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2), ptr %sub_c.sroa.0.i, align 8
+  tail call spir_func void @_Z29__spirv_JointMatrixStoreINTELIiLm12ELm12ELN5__spv9MatrixUseE2ELNS0_12MatrixLayoutE3ELNS0_5Scope4FlagE3EEvPT_PNS0_24__spirv_JointMatrixINTELIS5_XT0_EXT1_EXT3_EXT4_EXT2_EEEmS2_S4_i(ptr addrspace(4) noundef %call.ascast.i.i, target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2) noundef %sub_c.sroa.0.i.0.sub_c.sroa.0.i.0.sub_c.sroa.0.0.sub_c.sroa.0.0.sub_c.sroa.0.0..i, i64 noundef %_arg_N, i32 noundef 0, i32 noundef 3, i32 noundef 0) #4
+  call void @llvm.lifetime.end.p0(i64 8, ptr nonnull %sub_c.sroa.0.i.0.i.0..sroa_cast)
   ret void
 }
 
 ; Function Attrs: convergent
-declare dso_local spir_func %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(4)* @_Z28__spirv_JointMatrixLoadINTELIsLm2ELm2ELN5__spv12MatrixLayoutE0ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEEPS5_mS1_S3_i(i16 addrspace(4)*, i64, i32, i32, i32) local_unnamed_addr #1
-
-; Function Attrs: convergent
-declare dso_local spir_func %spirv.JointMatrixINTEL._char_2_16_0_3_0 addrspace(4)* @_Z28__spirv_JointMatrixLoadINTELIaLm2ELm16ELN5__spv12MatrixLayoutE0ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEEPS5_mS1_S3_i(i8 addrspace(4)*, i64, i32, i32, i32) local_unnamed_addr #1
-
-; Function Attrs: convergent
-declare dso_local spir_func %spirv.JointMatrixINTEL._char_16_2_3_3 addrspace(4)* @_Z28__spirv_JointMatrixLoadINTELIaLm16ELm2ELN5__spv12MatrixLayoutE3ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEEPS5_mS1_S3_i(i8 addrspace(4)*, i64, i32, i32, i32) local_unnamed_addr #1
-
-; Function Attrs: convergent
-declare dso_local spir_func %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(4)* @_Z27__spirv_JointMatrixMadINTELIasLm2ELm16ELm2ELN5__spv12MatrixLayoutE0ELS1_3ELS1_0ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT0_XT1_EXT3_EXT6_EXT7_EEEPNS4_IT_XT1_EXT2_EXT4_EXT7_EEEPNS4_IS8_XT2_EXT3_EXT5_EXT7_EEES7_S3_(%spirv.JointMatrixINTEL._char_2_16_0_3_0 addrspace(4)*, %spirv.JointMatrixINTEL._char_16_2_3_3 addrspace(4)*, %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(4)*, i32) local_unnamed_addr #1
+declare dso_local spir_func noundef target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2) @_Z26__spirv_CompositeConstructIiLm12ELm12ELN5__spv9MatrixUseE2ELNS0_12MatrixLayoutE3ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT3_EXT4_EXT2_EEES6_(i32 noundef) local_unnamed_addr #2
 
 ; Function Attrs: convergent
-declare dso_local spir_func %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(4)* @_Z29__spirv_JointMatrixSUMadINTELIasLm2ELm16ELm2ELN5__spv12MatrixLayoutE0ELS1_3ELS1_0ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT0_XT1_EXT3_EXT6_EXT7_EEEPNS4_IT_XT1_EXT2_EXT4_EXT7_EEEPNS4_IS8_XT2_EXT3_EXT5_EXT7_EEES7_S3_(%spirv.JointMatrixINTEL._char_2_16_0_3_0 addrspace(4)*, %spirv.JointMatrixINTEL._char_16_2_3_3 addrspace(4)*, %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(4)*, i32) local_unnamed_addr #1
+declare dso_local spir_func noundef target("spirv.JointMatrixINTEL", i8, 12, 48, 0, 3, 0) @_Z28__spirv_JointMatrixLoadINTELIaLm12ELm48ELN5__spv9MatrixUseE0ELNS0_12MatrixLayoutE0ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT3_EXT4_EXT2_EEEPS6_mS2_S4_i(ptr addrspace(4) noundef, i64 noundef, i32 noundef, i32 noundef, i32 noundef) local_unnamed_addr #2
 
 ; Function Attrs: convergent
-declare dso_local spir_func %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(4)* @_Z29__spirv_JointMatrixUSMadINTELIasLm2ELm16ELm2ELN5__spv12MatrixLayoutE0ELS1_3ELS1_0ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT0_XT1_EXT3_EXT6_EXT7_EEEPNS4_IT_XT1_EXT2_EXT4_EXT7_EEEPNS4_IS8_XT2_EXT3_EXT5_EXT7_EEES7_S3_(%spirv.JointMatrixINTEL._char_2_16_0_3_0 addrspace(4)*, %spirv.JointMatrixINTEL._char_16_2_3_3 addrspace(4)*, %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(4)*, i32) local_unnamed_addr #1
+declare dso_local spir_func noundef target("spirv.JointMatrixINTEL", i8, 48, 12, 2, 3, 1) @_Z28__spirv_JointMatrixLoadINTELIaLm48ELm12ELN5__spv9MatrixUseE1ELNS0_12MatrixLayoutE2ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT3_EXT4_EXT2_EEEPS6_mS2_S4_i(ptr addrspace(4) noundef, i64 noundef, i32 noundef, i32 noundef, i32 noundef) local_unnamed_addr #2
 
 ; Function Attrs: convergent
-declare dso_local spir_func %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(4)* @_Z29__spirv_JointMatrixUUMadINTELIasLm2ELm16ELm2ELN5__spv12MatrixLayoutE0ELS1_3ELS1_0ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT0_XT1_EXT3_EXT6_EXT7_EEEPNS4_IT_XT1_EXT2_EXT4_EXT7_EEEPNS4_IS8_XT2_EXT3_EXT5_EXT7_EEES7_S3_(%spirv.JointMatrixINTEL._char_2_16_0_3_0 addrspace(4)*, %spirv.JointMatrixINTEL._char_16_2_3_3 addrspace(4)*, %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(4)*, i32) local_unnamed_addr #1
+declare dso_local spir_func noundef target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2) @_Z27__spirv_JointMatrixMadINTELIaiLm12ELm48ELm12ELN5__spv9MatrixUseE0ELS1_1ELS1_2ELNS0_12MatrixLayoutE0ELS2_2ELS2_3ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT0_XT1_EXT3_EXT9_EXT10_EXT6_EEEPNS5_IT_XT1_EXT2_EXT7_EXT10_EXT4_EEEPNS5_IS9_XT2_EXT3_EXT8_EXT10_EXT5_EEES8_S4_(target("spirv.JointMatrixINTEL", i8, 12, 48, 0, 3, 0) noundef, target("spirv.JointMatrixINTEL", i8, 48, 12, 2, 3, 1) noundef, target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2) noundef, i32 noundef) local_unnamed_addr #2
 
 ; Function Attrs: convergent
-declare dso_local spir_func void @_Z29__spirv_JointMatrixStoreINTELIsLm2ELm2ELN5__spv12MatrixLayoutE0ELNS0_5Scope4FlagE3EEvPT_PNS0_24__spirv_JointMatrixINTELIS4_XT0_EXT1_EXT2_EXT3_EEEmS1_S3_i(i16 addrspace(4)*, %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(4)*, i64, i32, i32, i32) local_unnamed_addr #1
+declare dso_local spir_func void @_Z29__spirv_JointMatrixStoreINTELIiLm12ELm12ELN5__spv9MatrixUseE2ELNS0_12MatrixLayoutE3ELNS0_5Scope4FlagE3EEvPT_PNS0_24__spirv_JointMatrixINTELIS5_XT0_EXT1_EXT3_EXT4_EXT2_EEEmS2_S4_i(ptr addrspace(4) noundef, target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2) noundef, i64 noundef, i32 noundef, i32 noundef, i32 noundef) local_unnamed_addr #2
 
-; Function Attrs: convergent
-declare dso_local spir_func %spirv.JointMatrixINTEL._short_2_2_0_3 addrspace(4)* @_Z26__spirv_CompositeConstructi(i32) #1
-
-; Function Attrs: inaccessiblememonly nofree nosync nounwind willreturn
-declare void @llvm.assume(i1 noundef) #2
-
-attributes #0 = { convergent norecurse "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "sycl-module-id"="./joint_matrix_test.cpp" "uniform-work-group-size"="true" }
-attributes #1 = { convergent "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
-attributes #2 = { inaccessiblememonly nofree nosync nounwind willreturn }
-attributes #3 = { convergent }
-
-!llvm.module.flags = !{!0, !1}
-!opencl.spir.version = !{!2}
-!spirv.Source = !{!3}
-!llvm.ident = !{!4}
-
-!0 = !{i32 1, !"wchar_size", i32 4}
-!1 = !{i32 7, !"frame-pointer", i32 2}
-!2 = !{i32 1, i32 2}
-!3 = !{i32 4, i32 100000}
-!4 = !{!"clang version 13.0.0 (https://github.com/intel/llvm.git b3243d9f711a1cd80681530d6017324796668d51)"}
-!5 = !{i32 -1, i32 -1, i32 -1, i32 -1}
-!6 = !{i32 16}
-!7 = !{!8, !10, !12}
-!8 = distinct !{!8, !9, !"_ZN7__spirv29InitSizesSTGlobalInvocationIdILi2EN2cl4sycl2idILi2EEEE8initSizeEv: %agg.result"}
-!9 = distinct !{!9, !"_ZN7__spirv29InitSizesSTGlobalInvocationIdILi2EN2cl4sycl2idILi2EEEE8initSizeEv"}
-!10 = distinct !{!10, !11, !"_ZN7__spirvL22initGlobalInvocationIdILi2EN2cl4sycl2idILi2EEEEET0_v: %agg.result"}
-!11 = distinct !{!11, !"_ZN7__spirvL22initGlobalInvocationIdILi2EN2cl4sycl2idILi2EEEEET0_v"}
-!12 = distinct !{!12, !13, !"_ZN2cl4sycl6detail7Builder10getElementILi2EEEKNS0_7nd_itemIXT_EEEPS5_: %agg.result"}
-!13 = distinct !{!13, !"_ZN2cl4sycl6detail7Builder10getElementILi2EEEKNS0_7nd_itemIXT_EEEPS5_"}
-!14 = !{!15, !17, !12}
-!15 = distinct !{!15, !16, !"_ZN7__spirv28InitSizesSTLocalInvocationIdILi2EN2cl4sycl2idILi2EEEE8initSizeEv: %agg.result"}
-!16 = distinct !{!16, !"_ZN7__spirv28InitSizesSTLocalInvocationIdILi2EN2cl4sycl2idILi2EEEE8initSizeEv"}
-!17 = distinct !{!17, !18, !"_ZN7__spirvL21initLocalInvocationIdILi2EN2cl4sycl2idILi2EEEEET0_v: %agg.result"}
-!18 = distinct !{!18, !"_ZN7__spirvL21initLocalInvocationIdILi2EN2cl4sycl2idILi2EEEEET0_v"}
-!19 = distinct !{!19, !20, !21}
-!20 = !{!"llvm.loop.mustprogress"}
-!21 = !{!"llvm.loop.unroll.disable"}
+; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
+declare void @llvm.lifetime.start.p0(i64 immarg, ptr nocapture) #3
 
+; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
+declare void @llvm.lifetime.end.p0(i64 immarg, ptr nocapture) #3
 
+attributes #0 = { convergent norecurse "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "sycl-module-id"="matrix-int8-test.cpp" "uniform-work-group-size"="true" }
+attributes #1 = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: readwrite) }
+attributes #2 = { convergent "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #3 = { nocallback nofree nosync nounwind willreturn memory(argmem: readwrite) }
+attributes #4 = { convergent }
diff --git a/test/extensions/INTEL/SPV_INTEL_joint_matrix/joint_matrix_bfloat16.ll b/test/extensions/INTEL/SPV_INTEL_joint_matrix/joint_matrix_bfloat16.ll
index 8109cb0ed1..80c014b689 100644
--- a/test/extensions/INTEL/SPV_INTEL_joint_matrix/joint_matrix_bfloat16.ll
+++ b/test/extensions/INTEL/SPV_INTEL_joint_matrix/joint_matrix_bfloat16.ll
@@ -1,218 +1,202 @@
-; RUN: llvm-as -opaque-pointers=0 < %s -o %t.bc
+; compiled from joint_matrix_bfloat16.cpp test from intel/llvm
 
-; RUN: llvm-spirv -s %t.bc -opaque-pointers=0 -o %t.regularized.bc
-; RUN: llvm-dis -opaque-pointers=0 %t.regularized.bc -o %t.regularized.ll
-; RUN: FileCheck < %t.regularized.ll %s --check-prefix=CHECK-REGULARIZED
+; RUN: llvm-as < %s -o %t.bc
 
-; RUN: llvm-spirv %t.bc -opaque-pointers=0 --spirv-ext=+SPV_INTEL_bfloat16_conversion,+SPV_INTEL_joint_matrix -o %t.spv
+; RUN: llvm-spirv %t.bc --spirv-ext=+SPV_INTEL_bfloat16_conversion,+SPV_INTEL_joint_matrix -o %t.spv
 ; RUN: llvm-spirv %t.spv -to-text -o %t.spt
 ; RUN: FileCheck < %t.spt %s --check-prefix=CHECK-SPIRV
 
 ; RUN: llvm-spirv -r %t.spv -o %t.rev.bc
-; RUN: llvm-dis -opaque-pointers=0 < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM
-
-; CHECK-REGULARIZED: %[[Alloca:.*]] = alloca %"class.cl::sycl::ext::intel::experimental::bfloat16", align 2
-; CHECK-REGULARIZED: %[[ASCast:.*]] = addrspacecast %"class.cl::sycl::ext::intel::experimental::bfloat16"* %[[Alloca]] to %"class.cl::sycl::ext::intel::experimental::bfloat16" addrspace(4)*
-; CHECK-REGULARIZED: %[[GEP1:.*]] = getelementptr inbounds %"class.cl::sycl::ext::intel::experimental::bfloat16", %"class.cl::sycl::ext::intel::experimental::bfloat16" addrspace(4)* %[[ASCast]], i64 0, i32 0
-; CHECK-REGULARIZED: %[[#Extract:]] = call spir_func i16 @_Z28__spirv_VectorExtractDynamicIN2cl4sycl3ext5intel12experimental8bfloat16ELm8ELm16ELN5__spv12MatrixLayoutE0ELNS6_5Scope4FlagE3EET_PNS6_24__spirv_JointMatrixINTELISA_XT0_EXT1_EXT2_EXT3_EEEm(%spirv.JointMatrixINTEL._bfloat16_8_16_0_3 addrspace(4)* align 2 %{{.*}}, i64 noundef %{{.*}})
-; CHECK-REGULARIZED: %[[#GEP2:]] = getelementptr inbounds %"class.cl::sycl::ext::intel::experimental::bfloat16", %"class.cl::sycl::ext::intel::experimental::bfloat16" addrspace(4)* %[[ASCast]], i32 0, i32 0
-; CHECK-REGULARIZED: store i16 %[[#Extract]], i16 addrspace(4)* %[[#GEP2]], align 2
-; CHECK-REGULARIZED: %[[#Load:]] = load i16, i16 addrspace(4)* %[[GEP1]], align 2
-; CHECK-REGULARIZED: %[[ConvertVal:.*]] = call spir_func noundef float @_Z27__spirv_ConvertBF16ToFINTELt(i16 noundef zeroext %[[#Load]])
-; CHECK-REGULARIZED: %{{.*}} = fadd float %[[ConvertVal]], %{{.*}}
-
-; CHECK-SPIRV: TypeInt [[#TypeI16ID:]] 16 0
-; CHECK-SPIRV: TypeFloat [[#TypeFID:]] 32
-; CHECK-SPIRV: TypeJointMatrixINTEL [[#TypeJointMID:]] [[#TypeI16ID]] [[#]] [[#]] [[#]] [[#]]
-; CHECK-SPIRV: Phi [[#TypeJointMID]] [[#PhiID:]] [[#]] [[#]] [[#]] [[#]]
-; CHECK-SPIRV: VectorExtractDynamic [[#TypeI16ID]] [[#ExtractID:]] [[#PhiID]] [[#]]
-; CHECK-SPIRV: Store [[#PtrID:]] [[#ExtractID]] [[#]] [[#]]
-; CHECK-SPIRV: ConvertBF16ToFINTEL [[#TypeFID]] [[#Conv1ID:]] [[#]]
-; CHECK-SPIRV: ConvertBF16ToFINTEL [[#TypeFID]] [[#Conv2ID:]] [[#]]
-; CHECK-SPIRV: FAdd [[#TypeFID]] [[#ResId:]] [[#Conv1ID]] [[#Conv2ID]]
-; CHECK-SPIRV: ConvertFToBF16INTEL [[#TypeI16ID]] [[#]] [[#ResId]]
-; CHECK-SPIRV: Load [[#TypeI16ID]] [[#LoadID:]] [[#]] [[#]] [[#]]
-; CHECK-SPIRV: VectorInsertDynamic [[#TypeJointMID]] [[#]] [[#PhiID]] [[#LoadID]] [[#]]
-
-; CHECK-LLVM: %spirv.JointMatrixINTEL._short_8_16_0_3
-; CHECK-LLVM: %[[GEP1:.*]] = getelementptr inbounds %"class.cl::sycl::ext::intel::experimental::bfloat16", %"class.cl::sycl::ext::intel::experimental::bfloat16" addrspace(4)* %{{.*}}, i64 0, i32 0
-; CHECK-LLVM: %[[GEP2:.*]] = getelementptr inbounds %"class.cl::sycl::ext::intel::experimental::bfloat16", %"class.cl::sycl::ext::intel::experimental::bfloat16" addrspace(4)* %{{.*}}, i64 0, i32 0
-; CHECK-LLVM: %[[ConvertConst:.*]] = call spir_func i16 @_Z32intel_convert_bfloat16_as_ushortf(float 2.000000e+00)
-; CHECK-LLVM: %[[#LoadGEP:]] = load i16, i16 addrspace(4)* %[[GEP2]], align 2
-; CHECK-LLVM: %[[ConvertVal:.*]] = call spir_func float @_Z31intel_convert_as_bfloat16_floats(i16 %[[#LoadGEP]])
-; CHECK-LLVM: %[[ConvertConstToF:.*]] = call spir_func float @_Z31intel_convert_as_bfloat16_floats(i16 %[[ConvertConst]])
-; CHECK-LLVM: %[[FAddRes:.*]] = fadd float %[[ConvertVal]], %[[ConvertConstToF]]
-; CHECK-LLVM: %[[ConvertResToBF:.*]] = call spir_func i16 @_Z32intel_convert_bfloat16_as_ushortf(float %[[FAddRes]])
-; CHECK-LLVM: store i16 %[[ConvertResToBF]], i16 addrspace(4)* %[[#]], align 2
-
-; ModuleID = 'joint_matrix_bfloat16_test.bc'
-source_filename = "joint_matrix_bfloat16_test.cpp"
+; RUN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM
+
+; CHECK-SPIRV-DAG: TypeInt [[#SHORT:]] 16
+; CHECK-SPIRV-DAG: TypeInt [[#INT:]] 32
+; CHECK-SPIRV-DAG: TypeFloat [[#Float:]] 32
+; CHECK-SPIRV-DAG: Constant [[#INT]] [[#CONST8:]] 8
+; CHECK-SPIRV-DAG: Constant [[#INT]] [[#CONST16:]] 16
+; CHECK-SPIRV-DAG: Constant [[#INT]] [[#CONST3:]] 3
+; CHECK-SPIRV-DAG: Constant [[#INT]] [[#CONST2:]] 2
+; CHECK-SPIRV-DAG: Constant [[#INT]] [[#CONST1:]] 1
+; CHECK-SPIRV-DAG: Constant [[#INT]] [[#CONST0:]] 0
+; CHECK-SPIRV-DAG: TypeJointMatrixINTEL [[#MatTy1:]] [[#Float]] [[#CONST8]] [[#CONST16]] [[#CONST3]] [[#CONST3]] [[#CONST2]]
+; CHECK-SPIRV-DAG: TypeJointMatrixINTEL [[#MatTy2:]] [[#SHORT]] [[#CONST8]] [[#CONST16]] [[#CONST0]] [[#CONST3]] [[#CONST0]] [[#CONST1]]
+; CHECK-SPIRV-DAG: TypeJointMatrixINTEL [[#MatTy3:]] [[#SHORT]] [[#CONST16]] [[#CONST16]] [[#CONST2]] [[#CONST3]] [[#CONST1]] [[#CONST1]]
+
+; CHECK-LLVM: call spir_func target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) @_Z80__spirv_JointMatrixLoadINTEL_RPU3AS142__spirv_JointMatrixINTEL__float_8_16_3_3_2PU3AS1fliii(ptr addrspace(1) %{{.*}}, i64 32, i32 0, i32 3, i32 0)
+; CHECK-LLVM: call spir_func target("spirv.JointMatrixINTEL", i16, 8, 16, 0, 3, 0, 1) @"_Z82__spirv_JointMatrixLoadINTEL_RPU3AS144__spirv_JointMatrixINTEL__short_8_16_0_3_0_1PU3AS138class.sycl::_V1::ext::oneapi::bfloat16liii"(ptr addrspace(1) %{{.*}}, i64 32, i32 0, i32 3, i32 0)
+; CHECK-LLVM: call spir_func target("spirv.JointMatrixINTEL", i16, 16, 16, 2, 3, 1, 1) @"_Z83__spirv_JointMatrixLoadINTEL_RPU3AS145__spirv_JointMatrixINTEL__short_16_16_2_3_1_1PU3AS138class.sycl::_V1::ext::oneapi::bfloat16liii"(ptr addrspace(1) %{{.*}}, i64 64, i32 2, i32 3, i32 0)
+; CHECK-LLVM: call spir_func target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) @_Z27__spirv_JointMatrixMadINTELPU3AS144__spirv_JointMatrixINTEL__short_8_16_0_3_0_1PU3AS145__spirv_JointMatrixINTEL__short_16_16_2_3_1_1PU3AS142__spirv_JointMatrixINTEL__float_8_16_3_3_2i(target("spirv.JointMatrixINTEL", i16, 8, 16, 0, 3, 0, 1) %{{.*}}, target("spirv.JointMatrixINTEL", i16, 16, 16, 2, 3, 1, 1) %{{.*}}, target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) %{{.*}}, i32 3)
+; CHECK-LLVM: call spir_func void @_Z29__spirv_JointMatrixStoreINTELPU3AS1fPU3AS142__spirv_JointMatrixINTEL__float_8_16_3_3_2liii(ptr addrspace(1) %{{.*}}, target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) %{{.*}}, i64 32, i32 0, i32 3, i32 0)
+
+; ModuleID = 'joint_matrix_bfloat16-sycl-spir64-unknown-unknown.bc'
+source_filename = "../joint_matrix_bfloat16.cpp"
 target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-n8:16:32:64"
 target triple = "spir64-unknown-unknown"
 
-%class.anon = type { %"class.cl::sycl::accessor" }
-%"class.cl::sycl::accessor" = type { %"class.cl::sycl::detail::AccessorImplDevice", %union.anon }
-%"class.cl::sycl::detail::AccessorImplDevice" = type { %"class.cl::sycl::id", %"class.cl::sycl::range", %"class.cl::sycl::range" }
-%"class.cl::sycl::id" = type { %"class.cl::sycl::detail::array" }
-%"class.cl::sycl::detail::array" = type { [2 x i64] }
-%"class.cl::sycl::range" = type { %"class.cl::sycl::detail::array" }
-%union.anon = type { %"class.cl::sycl::ext::intel::experimental::bfloat16" addrspace(1)* }
-%"class.cl::sycl::ext::intel::experimental::bfloat16" = type { i16 }
-%"class.cl::sycl::nd_item" = type { %"class.cl::sycl::item", %"class.cl::sycl::item.0", %"class.cl::sycl::group" }
-%"class.cl::sycl::item" = type { %"struct.cl::sycl::detail::ItemBase" }
-%"struct.cl::sycl::detail::ItemBase" = type { %"class.cl::sycl::range", %"class.cl::sycl::id", %"class.cl::sycl::id" }
-%"class.cl::sycl::item.0" = type { %"struct.cl::sycl::detail::ItemBase.1" }
-%"struct.cl::sycl::detail::ItemBase.1" = type { %"class.cl::sycl::range", %"class.cl::sycl::id" }
-%"class.cl::sycl::group" = type { %"class.cl::sycl::range", %"class.cl::sycl::range", %"class.cl::sycl::range", %"class.cl::sycl::id" }
-%spirv.JointMatrixINTEL._bfloat16_8_16_0_3 = type opaque
-
-$_ZZZ17matrix_verify_addIN2cl4sycl3ext5intel12experimental8bfloat16ELm16ELm16EEvNS1_5queueER10big_matrixIT_XT0_EXT1_EERNS1_8nd_rangeILi2EEEfENKUlRNS1_7handlerEE_clESF_ENKUlNS1_7nd_itemILi2EEEE_clESI_ = comdat any
-
-; Function Attrs: argmemonly nofree nosync nounwind willreturn
-declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #0
-
-; Function Attrs: convergent inlinehint norecurse
-define linkonce_odr dso_local spir_func void @_ZZZ17matrix_verify_addIN2cl4sycl3ext5intel12experimental8bfloat16ELm16ELm16EEvNS1_5queueER10big_matrixIT_XT0_EXT1_EERNS1_8nd_rangeILi2EEEfENKUlRNS1_7handlerEE_clESF_ENKUlNS1_7nd_itemILi2EEEE_clESI_(%class.anon addrspace(4)* noundef align 8 dereferenceable_or_null(56) %this, %"class.cl::sycl::nd_item"* noundef byval(%"class.cl::sycl::nd_item") align 8 %spmd_item) local_unnamed_addr #1 comdat align 2 {
+%"class.sycl::_V1::ext::oneapi::bfloat16" = type { i16 }
+
+$_ZTSZZ15matrix_multiplyIfN4sycl3_V13ext6oneapi8bfloat16ELm16ELm32ELm32EEvR10big_matrixIT_XT1_EXT2_EERS5_IT0_XT1_EXT3_EERS5_IS9_XdvT3_Li2EEXmlT2_Li2EEEENKUlRNS1_7handlerEE_clESF_E7imatrix = comdat any
+
+@__spirv_BuiltInGlobalInvocationId = external dso_local local_unnamed_addr addrspace(1) constant <3 x i64>, align 32
+@__spirv_BuiltInLocalInvocationId = external dso_local local_unnamed_addr addrspace(1) constant <3 x i64>, align 32
+
+; Function Attrs: convergent norecurse nounwind
+define weak_odr dso_local spir_kernel void @_ZTSZZ15matrix_multiplyIfN4sycl3_V13ext6oneapi8bfloat16ELm16ELm32ELm32EEvR10big_matrixIT_XT1_EXT2_EERS5_IT0_XT1_EXT3_EERS5_IS9_XdvT3_Li2EEXmlT2_Li2EEEENKUlRNS1_7handlerEE_clESF_E7imatrix(ptr addrspace(1) noundef align 4 %_arg_accC, ptr addrspace(1) noundef align 2 %_arg_accA, ptr addrspace(1) noundef align 2 %_arg_accB) local_unnamed_addr #0 comdat !srcloc !48 !kernel_arg_buffer_location !49 !kernel_arg_runtime_aligned !50 !kernel_arg_exclusive_ptr !50 !intel_reqd_sub_group_size !51 !sycl_fixed_targets !52 !sycl_kernel_omit_args !53 {
 entry:
-  %ref.tmp.i = alloca %"class.cl::sycl::ext::intel::experimental::bfloat16", align 2
-  %agg.tmp.i54 = alloca %"class.cl::sycl::ext::intel::experimental::bfloat16", align 2
-  %agg.tmp.i = alloca %"class.cl::sycl::ext::intel::experimental::bfloat16", align 2
-  %spmd_item.ascast = addrspacecast %"class.cl::sycl::nd_item"* %spmd_item to %"class.cl::sycl::nd_item" addrspace(4)*
-  %arrayidx.i.i.i = getelementptr inbounds %"class.cl::sycl::nd_item", %"class.cl::sycl::nd_item" addrspace(4)* %spmd_item.ascast, i64 0, i32 0, i32 0, i32 1, i32 0, i32 0, i64 0
-  %0 = load i64, i64 addrspace(4)* %arrayidx.i.i.i, align 8, !tbaa !5
-  %cmp.i = icmp ult i64 %0, 2147483648
-  tail call void @llvm.assume(i1 %cmp.i)
-  %arrayidx.i.i.i29 = getelementptr inbounds %"class.cl::sycl::nd_item", %"class.cl::sycl::nd_item" addrspace(4)* %spmd_item.ascast, i64 0, i32 0, i32 0, i32 1, i32 0, i32 0, i64 1
-  %1 = load i64, i64 addrspace(4)* %arrayidx.i.i.i29, align 8, !tbaa !5
-  %cmp.i30 = icmp ult i64 %1, 2147483648
-  tail call void @llvm.assume(i1 %cmp.i30)
-  %arrayidx.i.i.i31 = getelementptr inbounds %"class.cl::sycl::nd_item", %"class.cl::sycl::nd_item" addrspace(4)* %spmd_item.ascast, i64 0, i32 1, i32 0, i32 1, i32 0, i32 0, i64 0
-  %2 = load i64, i64 addrspace(4)* %arrayidx.i.i.i31, align 8, !tbaa !5
-  %cmp.i32 = icmp ult i64 %2, 2147483648
-  tail call void @llvm.assume(i1 %cmp.i32)
-  %arrayidx.i.i.i33 = getelementptr inbounds %"class.cl::sycl::nd_item", %"class.cl::sycl::nd_item" addrspace(4)* %spmd_item.ascast, i64 0, i32 1, i32 0, i32 1, i32 0, i32 0, i64 1
-  %3 = load i64, i64 addrspace(4)* %arrayidx.i.i.i33, align 8, !tbaa !5
-  %cmp.i34 = icmp ult i64 %3, 2147483648
-  tail call void @llvm.assume(i1 %cmp.i34)
-  %4 = bitcast %"class.cl::sycl::ext::intel::experimental::bfloat16"* %agg.tmp.i to i8*
-  call void @llvm.lifetime.start.p0i8(i64 2, i8* nonnull %4)
-  %agg.tmp.ascast.i = addrspacecast %"class.cl::sycl::ext::intel::experimental::bfloat16"* %agg.tmp.i to %"class.cl::sycl::ext::intel::experimental::bfloat16" addrspace(4)*
-  %call.i.i.i = tail call spir_func noundef zeroext i16 @_Z27__spirv_ConvertFToBF16INTELf(float noundef 5.000000e+00) #6
-  %value.i.i = getelementptr inbounds %"class.cl::sycl::ext::intel::experimental::bfloat16", %"class.cl::sycl::ext::intel::experimental::bfloat16" addrspace(4)* %agg.tmp.ascast.i, i64 0, i32 0
-  store i16 %call.i.i.i, i16 addrspace(4)* %value.i.i, align 2, !tbaa !9
-  %call.i = tail call spir_func noundef %spirv.JointMatrixINTEL._bfloat16_8_16_0_3 addrspace(4)* @_Z26__spirv_CompositeConstructIN2cl4sycl3ext5intel12experimental8bfloat16ELm8ELm16ELN5__spv12MatrixLayoutE0ELNS6_5Scope4FlagE3EEPNS6_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEESB_(%"class.cl::sycl::ext::intel::experimental::bfloat16"* noundef nonnull byval(%"class.cl::sycl::ext::intel::experimental::bfloat16") align 2 %agg.tmp.i) #7
-  call void @llvm.lifetime.end.p0i8(i64 2, i8* nonnull %4)
-  %ref.tmp.ascast.i = addrspacecast %"class.cl::sycl::ext::intel::experimental::bfloat16"* %ref.tmp.i to %"class.cl::sycl::ext::intel::experimental::bfloat16" addrspace(4)*
-  %5 = bitcast %"class.cl::sycl::ext::intel::experimental::bfloat16"* %ref.tmp.i to i8*
-  %value.i.i.i = getelementptr inbounds %"class.cl::sycl::ext::intel::experimental::bfloat16", %"class.cl::sycl::ext::intel::experimental::bfloat16" addrspace(4)* %ref.tmp.ascast.i, i64 0, i32 0
-  %6 = bitcast %"class.cl::sycl::ext::intel::experimental::bfloat16"* %agg.tmp.i54 to i8*
-  %7 = getelementptr inbounds %"class.cl::sycl::ext::intel::experimental::bfloat16", %"class.cl::sycl::ext::intel::experimental::bfloat16"* %agg.tmp.i54, i64 0, i32 0
-  %8 = addrspacecast i16* %7 to i16 addrspace(4)*
-  br label %for.cond
-
-for.cond:                                         ; preds = %for.body, %entry
-  %sub_a.sroa.0.0 = phi %spirv.JointMatrixINTEL._bfloat16_8_16_0_3 addrspace(4)* [ %call.i, %entry ], [ %call.i58, %for.body ]
-  %i.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
-  %conv = zext i32 %i.0 to i64
-  %call.i41 = call spir_func noundef i64 @_Z38__spirv_JointMatrixWorkItemLengthINTELIN2cl4sycl3ext5intel12experimental8bfloat16ELm8ELm16ELN5__spv12MatrixLayoutE0ELNS6_5Scope4FlagE3EEmPNS6_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEE(%spirv.JointMatrixINTEL._bfloat16_8_16_0_3 addrspace(4)* noundef %sub_a.sroa.0.0) #7
-  %cmp = icmp ugt i64 %call.i41, %conv
-  br i1 %cmp, label %for.body, label %for.cond.cleanup
-
-for.cond.cleanup:                                 ; preds = %for.cond
-  %sub5 = sub nsw i64 %1, %3
-  %sub = sub nsw i64 %0, %2
-  %MData.i.i.i = getelementptr inbounds %class.anon, %class.anon addrspace(4)* %this, i64 0, i32 0, i32 1, i32 0
-  %9 = load %"class.cl::sycl::ext::intel::experimental::bfloat16" addrspace(1)*, %"class.cl::sycl::ext::intel::experimental::bfloat16" addrspace(1)* addrspace(4)* %MData.i.i.i, align 8, !tbaa !12, !noalias !13
-  %mul19 = shl nsw i64 %sub, 7
-  %add.ptr.i = getelementptr inbounds %"class.cl::sycl::ext::intel::experimental::bfloat16", %"class.cl::sycl::ext::intel::experimental::bfloat16" addrspace(1)* %9, i64 %mul19
-  %div = and i64 %sub5, -8
-  %add.ptr.i45 = getelementptr inbounds %"class.cl::sycl::ext::intel::experimental::bfloat16", %"class.cl::sycl::ext::intel::experimental::bfloat16" addrspace(1)* %add.ptr.i, i64 %div
-  %call.ascast.i = addrspacecast %"class.cl::sycl::ext::intel::experimental::bfloat16" addrspace(1)* %add.ptr.i45 to %"class.cl::sycl::ext::intel::experimental::bfloat16" addrspace(4)*
-  call spir_func void @_Z29__spirv_JointMatrixStoreINTELIN2cl4sycl3ext5intel12experimental8bfloat16ELm8ELm16ELN5__spv12MatrixLayoutE0ELNS6_5Scope4FlagE3EEvPT_PNS6_24__spirv_JointMatrixINTELISA_XT0_EXT1_EXT2_EXT3_EEEmS7_S9_i(%"class.cl::sycl::ext::intel::experimental::bfloat16" addrspace(4)* noundef %call.ascast.i, %spirv.JointMatrixINTEL._bfloat16_8_16_0_3 addrspace(4)* noundef %sub_a.sroa.0.0, i64 noundef 16, i32 noundef 0, i32 noundef 3, i32 noundef 0) #7
+  call void @__itt_offload_wi_start_wrapper()
+  %0 = load i64, ptr addrspace(1) getelementptr inbounds (i8, ptr addrspace(1) @__spirv_BuiltInGlobalInvocationId, i64 8), align 8, !noalias !54
+  %1 = load i64, ptr addrspace(1) @__spirv_BuiltInGlobalInvocationId, align 32, !noalias !54
+  %2 = load i64, ptr addrspace(1) getelementptr inbounds (i8, ptr addrspace(1) @__spirv_BuiltInLocalInvocationId, i64 8), align 8, !noalias !61
+  %3 = load i64, ptr addrspace(1) @__spirv_BuiltInLocalInvocationId, align 32, !noalias !61
+  %cmp.i.i = icmp ult i64 %0, 2147483648
+  tail call void @llvm.assume(i1 %cmp.i.i)
+  %cmp.i50.i = icmp ult i64 %1, 2147483648
+  tail call void @llvm.assume(i1 %cmp.i50.i)
+  %cmp.i52.i = icmp ult i64 %2, 2147483648
+  tail call void @llvm.assume(i1 %cmp.i52.i)
+  %sub.i = sub nsw i64 %0, %2
+  %cmp.i55.i = icmp ult i64 %3, 2147483648
+  tail call void @llvm.assume(i1 %cmp.i55.i)
+  %sub5.i = sub nsw i64 %1, %3
+  %mul8.i = shl nsw i64 %sub.i, 8
+  %add.ptr.i.i = getelementptr inbounds float, ptr addrspace(1) %_arg_accC, i64 %mul8.i
+  %div48.i = and i64 %sub5.i, -16
+  %add.ptr.i69.i = getelementptr inbounds float, ptr addrspace(1) %add.ptr.i.i, i64 %div48.i
+  %call1.i.i = tail call spir_func noundef target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) @_Z28__spirv_JointMatrixLoadINTELIU3AS1ffLm8ELm16ELN5__spv9MatrixUseE2ELNS1_12MatrixLayoutE3ELNS1_5Scope4FlagE3EEPNS1_24__spirv_JointMatrixINTELIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEPT_mS3_S5_i(ptr addrspace(1) noundef %add.ptr.i69.i, i64 noundef 32, i32 noundef 0, i32 noundef 3, i32 noundef 0) #3
+  %mul28.i = shl nsw i64 %div48.i, 1
+  %add.ptr.i84.i = getelementptr inbounds %"class.sycl::_V1::ext::oneapi::bfloat16", ptr addrspace(1) %_arg_accA, i64 %mul8.i
+  %invariant.gep = getelementptr %"class.sycl::_V1::ext::oneapi::bfloat16", ptr addrspace(1) %_arg_accB, i64 %mul28.i
+  br label %for.cond.i
+
+for.cond.i:                                       ; preds = %for.body.i, %entry
+  %sub_c.sroa.0.0.i = phi target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) [ %call1.i.i, %entry ], [ %call.i.i, %for.body.i ]
+  %k.0.i = phi i32 [ 0, %entry ], [ %add.i, %for.body.i ]
+  %cmp.i = icmp ult i32 %k.0.i, 2
+  br i1 %cmp.i, label %for.body.i, label %_ZZZ15matrix_multiplyIfN4sycl3_V13ext6oneapi8bfloat16ELm16ELm32ELm32EEvR10big_matrixIT_XT1_EXT2_EERS5_IT0_XT1_EXT3_EERS5_IS9_XdvT3_Li2EEXmlT2_Li2EEEENKUlRNS1_7handlerEE_clESF_ENKUlNS1_7nd_itemILi2EEEE_clESI_.exit
+
+for.body.i:                                       ; preds = %for.cond.i
+  %mul16.i = shl nuw nsw i32 %k.0.i, 4
+  %conv17.i = zext i32 %mul16.i to i64
+  %add.ptr.i85.i = getelementptr inbounds %"class.sycl::_V1::ext::oneapi::bfloat16", ptr addrspace(1) %add.ptr.i84.i, i64 %conv17.i
+  %call1.i63.i = tail call spir_func noundef target("spirv.JointMatrixINTEL", i16, 8, 16, 0, 3, 0, 1) @_Z28__spirv_JointMatrixLoadINTELIU3AS1N4sycl3_V13ext6oneapi8bfloat16ES4_Lm8ELm16ELN5__spv9MatrixUseE0ELNS6_12MatrixLayoutE0ELNS6_5Scope4FlagE3EEPNS6_24__spirv_JointMatrixINTELIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEPT_mS8_SA_i(ptr addrspace(1) noundef %add.ptr.i85.i, i64 noundef 32, i32 noundef 0, i32 noundef 3, i32 noundef 0) #3
+  %div23.i = shl nuw nsw i32 %k.0.i, 3
+  %conv24.i = zext i32 %div23.i to i64
+  %mul25.i = shl nuw nsw i64 %conv24.i, 6
+  %gep = getelementptr %"class.sycl::_V1::ext::oneapi::bfloat16", ptr addrspace(1) %invariant.gep, i64 %mul25.i
+  %call1.i67.i = tail call spir_func noundef target("spirv.JointMatrixINTEL", i16, 16, 16, 2, 3, 1, 1) @_Z28__spirv_JointMatrixLoadINTELIU3AS1N4sycl3_V13ext6oneapi8bfloat16ES4_Lm16ELm16ELN5__spv9MatrixUseE1ELNS6_12MatrixLayoutE2ELNS6_5Scope4FlagE3EEPNS6_24__spirv_JointMatrixINTELIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEPT_mS8_SA_i(ptr addrspace(1) noundef %gep, i64 noundef 64, i32 noundef 2, i32 noundef 3, i32 noundef 0) #3
+  %call.i.i = tail call spir_func noundef target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) @_Z27__spirv_JointMatrixMadINTELIN4sycl3_V13ext6oneapi8bfloat16EfLm8ELm16ELm16ELN5__spv9MatrixUseE0ELS6_1ELS6_2ELNS5_12MatrixLayoutE0ELS7_2ELS7_3ELNS5_5Scope4FlagE3EEPNS5_24__spirv_JointMatrixINTELIT0_XT1_EXT3_EXT9_EXT10_EXT6_EEEPNSA_IT_XT1_EXT2_EXT7_EXT10_EXT4_EEEPNSA_ISE_XT2_EXT3_EXT8_EXT10_EXT5_EEESD_S9_(target("spirv.JointMatrixINTEL", i16, 8, 16, 0, 3, 0, 1) noundef %call1.i63.i, target("spirv.JointMatrixINTEL", i16, 16, 16, 2, 3, 1, 1) noundef %call1.i67.i, target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) noundef %sub_c.sroa.0.0.i, i32 noundef 3) #3, !noalias !66
+  %add.i = add nuw nsw i32 %k.0.i, 1
+  br label %for.cond.i, !llvm.loop !69
+
+_ZZZ15matrix_multiplyIfN4sycl3_V13ext6oneapi8bfloat16ELm16ELm32ELm32EEvR10big_matrixIT_XT1_EXT2_EERS5_IT0_XT1_EXT3_EERS5_IS9_XdvT3_Li2EEXmlT2_Li2EEEENKUlRNS1_7handlerEE_clESF_ENKUlNS1_7nd_itemILi2EEEE_clESI_.exit: ; preds = %for.cond.i
+  tail call spir_func void @_Z29__spirv_JointMatrixStoreINTELIU3AS1ffLm8ELm16ELN5__spv9MatrixUseE2ELNS1_12MatrixLayoutE3ELNS1_5Scope4FlagE3EEvPT_PNS1_24__spirv_JointMatrixINTELIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEmS3_S5_i(ptr addrspace(1) noundef %add.ptr.i69.i, target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) noundef %sub_c.sroa.0.0.i, i64 noundef 32, i32 noundef 0, i32 noundef 3, i32 noundef 0) #3
+  call void @__itt_offload_wi_finish_wrapper()
   ret void
-
-for.body:                                         ; preds = %for.cond
-  %call.i.i = call spir_func noundef zeroext i16 @_Z27__spirv_ConvertFToBF16INTELf(float noundef 2.000000e+00) #6
-  call void @llvm.lifetime.start.p0i8(i64 2, i8* nonnull %5) #8, !noalias !16
-  call spir_func void @_Z28__spirv_VectorExtractDynamicIN2cl4sycl3ext5intel12experimental8bfloat16ELm8ELm16ELN5__spv12MatrixLayoutE0ELNS6_5Scope4FlagE3EET_PNS6_24__spirv_JointMatrixINTELISA_XT0_EXT1_EXT2_EXT3_EEEm(%"class.cl::sycl::ext::intel::experimental::bfloat16" addrspace(4)* sret(%"class.cl::sycl::ext::intel::experimental::bfloat16") align 2 %ref.tmp.ascast.i, %spirv.JointMatrixINTEL._bfloat16_8_16_0_3 addrspace(4)* noundef %sub_a.sroa.0.0, i64 noundef %conv) #7, !noalias !16
-  %10 = load i16, i16 addrspace(4)* %value.i.i.i, align 2, !tbaa !19, !noalias !20
-  %call.i.i.i.i = call spir_func noundef float @_Z27__spirv_ConvertBF16ToFINTELt(i16 noundef zeroext %10) #6, !noalias !20
-  %call.i.i3.i.i = call spir_func noundef float @_Z27__spirv_ConvertBF16ToFINTELt(i16 noundef zeroext %call.i.i) #6, !noalias !20
-  %add.i.i = fadd float %call.i.i.i.i, %call.i.i3.i.i
-  %call.i.i4.i.i = call spir_func noundef zeroext i16 @_Z27__spirv_ConvertFToBF16INTELf(float noundef %add.i.i) #6, !noalias !20
-  call void @llvm.lifetime.end.p0i8(i64 2, i8* nonnull %5) #8, !noalias !16
-  call void @llvm.lifetime.start.p0i8(i64 2, i8* nonnull %6)
-  store i16 %call.i.i4.i.i, i16 addrspace(4)* %8, align 2, !tbaa !19
-  %call.i58 = call spir_func noundef %spirv.JointMatrixINTEL._bfloat16_8_16_0_3 addrspace(4)* @_Z27__spirv_VectorInsertDynamicIN2cl4sycl3ext5intel12experimental8bfloat16ELm8ELm16ELN5__spv12MatrixLayoutE0ELNS6_5Scope4FlagE3EEPNS6_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEESD_SB_m(%spirv.JointMatrixINTEL._bfloat16_8_16_0_3 addrspace(4)* noundef %sub_a.sroa.0.0, %"class.cl::sycl::ext::intel::experimental::bfloat16"* noundef nonnull byval(%"class.cl::sycl::ext::intel::experimental::bfloat16") align 2 %agg.tmp.i54, i64 noundef %conv) #7
-  call void @llvm.lifetime.end.p0i8(i64 2, i8* nonnull %6)
-  %inc = add nuw nsw i32 %i.0, 1
-  br label %for.cond, !llvm.loop !23
 }
 
-; Function Attrs: argmemonly nofree nosync nounwind willreturn
-declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #0
-
-; Function Attrs: inaccessiblememonly nofree nosync nounwind willreturn
-declare void @llvm.assume(i1 noundef) #3
+; Function Attrs: nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: readwrite)
+declare void @llvm.assume(i1 noundef) #1
 
-; Function Attrs: convergent
-declare dso_local spir_func noundef %spirv.JointMatrixINTEL._bfloat16_8_16_0_3 addrspace(4)* @_Z26__spirv_CompositeConstructIN2cl4sycl3ext5intel12experimental8bfloat16ELm8ELm16ELN5__spv12MatrixLayoutE0ELNS6_5Scope4FlagE3EEPNS6_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEESB_(%"class.cl::sycl::ext::intel::experimental::bfloat16"* noundef byval(%"class.cl::sycl::ext::intel::experimental::bfloat16") align 2) local_unnamed_addr #4
+; Function Attrs: convergent nounwind
+declare dso_local spir_func noundef target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) @_Z28__spirv_JointMatrixLoadINTELIU3AS1ffLm8ELm16ELN5__spv9MatrixUseE2ELNS1_12MatrixLayoutE3ELNS1_5Scope4FlagE3EEPNS1_24__spirv_JointMatrixINTELIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEPT_mS3_S5_i(ptr addrspace(1) noundef, i64 noundef, i32 noundef, i32 noundef, i32 noundef) local_unnamed_addr #2
 
-; Function Attrs: convergent
-declare dso_local spir_func noundef i64 @_Z38__spirv_JointMatrixWorkItemLengthINTELIN2cl4sycl3ext5intel12experimental8bfloat16ELm8ELm16ELN5__spv12MatrixLayoutE0ELNS6_5Scope4FlagE3EEmPNS6_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEE(%spirv.JointMatrixINTEL._bfloat16_8_16_0_3 addrspace(4)* noundef) local_unnamed_addr #4
+; Function Attrs: convergent nounwind
+declare dso_local spir_func noundef target("spirv.JointMatrixINTEL", i16, 8, 16, 0, 3, 0, 1) @_Z28__spirv_JointMatrixLoadINTELIU3AS1N4sycl3_V13ext6oneapi8bfloat16ES4_Lm8ELm16ELN5__spv9MatrixUseE0ELNS6_12MatrixLayoutE0ELNS6_5Scope4FlagE3EEPNS6_24__spirv_JointMatrixINTELIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEPT_mS8_SA_i(ptr addrspace(1) noundef, i64 noundef, i32 noundef, i32 noundef, i32 noundef) local_unnamed_addr #2
 
-; Function Attrs: convergent
-declare dso_local spir_func void @_Z28__spirv_VectorExtractDynamicIN2cl4sycl3ext5intel12experimental8bfloat16ELm8ELm16ELN5__spv12MatrixLayoutE0ELNS6_5Scope4FlagE3EET_PNS6_24__spirv_JointMatrixINTELISA_XT0_EXT1_EXT2_EXT3_EEEm(%"class.cl::sycl::ext::intel::experimental::bfloat16" addrspace(4)* sret(%"class.cl::sycl::ext::intel::experimental::bfloat16") align 2, %spirv.JointMatrixINTEL._bfloat16_8_16_0_3 addrspace(4)* noundef, i64 noundef) local_unnamed_addr #4
+; Function Attrs: convergent nounwind
+declare dso_local spir_func noundef target("spirv.JointMatrixINTEL", i16, 16, 16, 2, 3, 1, 1) @_Z28__spirv_JointMatrixLoadINTELIU3AS1N4sycl3_V13ext6oneapi8bfloat16ES4_Lm16ELm16ELN5__spv9MatrixUseE1ELNS6_12MatrixLayoutE2ELNS6_5Scope4FlagE3EEPNS6_24__spirv_JointMatrixINTELIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEPT_mS8_SA_i(ptr addrspace(1) noundef, i64 noundef, i32 noundef, i32 noundef, i32 noundef) local_unnamed_addr #2
 
 ; Function Attrs: convergent nounwind
-declare dso_local spir_func noundef float @_Z27__spirv_ConvertBF16ToFINTELt(i16 noundef zeroext) local_unnamed_addr #5
+declare dso_local spir_func noundef target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) @_Z27__spirv_JointMatrixMadINTELIN4sycl3_V13ext6oneapi8bfloat16EfLm8ELm16ELm16ELN5__spv9MatrixUseE0ELS6_1ELS6_2ELNS5_12MatrixLayoutE0ELS7_2ELS7_3ELNS5_5Scope4FlagE3EEPNS5_24__spirv_JointMatrixINTELIT0_XT1_EXT3_EXT9_EXT10_EXT6_EEEPNSA_IT_XT1_EXT2_EXT7_EXT10_EXT4_EEEPNSA_ISE_XT2_EXT3_EXT8_EXT10_EXT5_EEESD_S9_(target("spirv.JointMatrixINTEL", i16, 8, 16, 0, 3, 0, 1) noundef, target("spirv.JointMatrixINTEL", i16, 16, 16, 2, 3, 1, 1) noundef, target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) noundef, i32 noundef) local_unnamed_addr #2
 
 ; Function Attrs: convergent nounwind
-declare dso_local spir_func noundef zeroext i16 @_Z27__spirv_ConvertFToBF16INTELf(float noundef) local_unnamed_addr #5
+declare dso_local spir_func void @_Z29__spirv_JointMatrixStoreINTELIU3AS1ffLm8ELm16ELN5__spv9MatrixUseE2ELNS1_12MatrixLayoutE3ELNS1_5Scope4FlagE3EEvPT_PNS1_24__spirv_JointMatrixINTELIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEmS3_S5_i(ptr addrspace(1) noundef, target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) noundef, i64 noundef, i32 noundef, i32 noundef, i32 noundef) local_unnamed_addr #2
+
+declare dso_local spir_func i32 @_Z18__spirv_ocl_printfPU3AS2Kcz(ptr addrspace(2), ...)
 
-; Function Attrs: convergent
-declare dso_local spir_func noundef %spirv.JointMatrixINTEL._bfloat16_8_16_0_3 addrspace(4)* @_Z27__spirv_VectorInsertDynamicIN2cl4sycl3ext5intel12experimental8bfloat16ELm8ELm16ELN5__spv12MatrixLayoutE0ELNS6_5Scope4FlagE3EEPNS6_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEESD_SB_m(%spirv.JointMatrixINTEL._bfloat16_8_16_0_3 addrspace(4)* noundef, %"class.cl::sycl::ext::intel::experimental::bfloat16"* noundef byval(%"class.cl::sycl::ext::intel::experimental::bfloat16") align 2, i64 noundef) local_unnamed_addr #4
+declare void @__itt_offload_wi_start_wrapper()
 
-; Function Attrs: convergent
-declare dso_local spir_func void @_Z29__spirv_JointMatrixStoreINTELIN2cl4sycl3ext5intel12experimental8bfloat16ELm8ELm16ELN5__spv12MatrixLayoutE0ELNS6_5Scope4FlagE3EEvPT_PNS6_24__spirv_JointMatrixINTELISA_XT0_EXT1_EXT2_EXT3_EEEmS7_S9_i(%"class.cl::sycl::ext::intel::experimental::bfloat16" addrspace(4)* noundef, %spirv.JointMatrixINTEL._bfloat16_8_16_0_3 addrspace(4)* noundef, i64 noundef, i32 noundef, i32 noundef, i32 noundef) local_unnamed_addr #4
+declare void @__itt_offload_wi_finish_wrapper()
 
-attributes #0 = { argmemonly nofree nosync nounwind willreturn }
-attributes #1 = { convergent inlinehint norecurse "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
-attributes #2 = { argmemonly nofree nounwind willreturn writeonly }
-attributes #3 = { inaccessiblememonly nofree nosync nounwind willreturn }
-attributes #4 = { convergent "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
-attributes #5 = { convergent nounwind "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
-attributes #6 = { convergent nounwind }
-attributes #7 = { convergent }
-attributes #8 = { nounwind }
+attributes #0 = { convergent norecurse nounwind "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "sycl-module-id"="../joint_matrix_bfloat16.cpp" "sycl-optlevel"="2" "uniform-work-group-size"="true" }
+attributes #1 = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: readwrite) }
+attributes #2 = { convergent nounwind "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #3 = { convergent nounwind }
 
 !llvm.module.flags = !{!0, !1}
 !opencl.spir.version = !{!2}
 !spirv.Source = !{!3}
-!llvm.ident = !{!4}
+!sycl_aspects = !{!4, !5, !6, !7, !8, !9, !10, !11, !12, !13, !14, !15, !16, !17, !18, !19, !20, !21, !22, !23, !24, !25, !26, !27, !28, !29, !30, !31, !32, !33, !34, !35, !36, !37, !38, !39, !40, !41, !42, !43, !44, !45, !46}
+!llvm.ident = !{!47}
 
 !0 = !{i32 1, !"wchar_size", i32 4}
 !1 = !{i32 7, !"frame-pointer", i32 2}
 !2 = !{i32 1, i32 2}
 !3 = !{i32 4, i32 100000}
-!4 = !{!"clang version 15.0.0 (https://github.com/pauzinl/llvm.git fb27c655023f19ff91f09413a0c51f0a37071cff)"}
-!5 = !{!6, !6, i64 0}
-!6 = !{!"long", !7, i64 0}
-!7 = !{!"omnipotent char", !8, i64 0}
-!8 = !{!"Simple C++ TBAA"}
-!9 = !{!10, !11, i64 0}
-!10 = !{!"_ZTSN2cl4sycl3ext5intel12experimental8bfloat16E", !11, i64 0}
-!11 = !{!"short", !7, i64 0}
-!12 = !{!7, !7, i64 0}
-!13 = !{!14}
-!14 = distinct !{!14, !15, !"_ZNK2cl4sycl8accessorINS0_3ext5intel12experimental8bfloat16ELi2ELNS0_6access4modeE1026ELNS6_6targetE2014ELNS6_11placeholderE0ENS2_6oneapi22accessor_property_listIJEEEE11get_pointerILS8_2014EvEENS0_9multi_ptrIS5_LNS6_13address_spaceE1EEEv: %agg.result"}
-!15 = distinct !{!15, !"_ZNK2cl4sycl8accessorINS0_3ext5intel12experimental8bfloat16ELi2ELNS0_6access4modeE1026ELNS6_6targetE2014ELNS6_11placeholderE0ENS2_6oneapi22accessor_property_listIJEEEE11get_pointerILS8_2014EvEENS0_9multi_ptrIS5_LNS6_13address_spaceE1EEEv"}
-!16 = !{!17}
-!17 = distinct !{!17, !18, !"_ZN2cl4sycl3ext6oneapi12experimental6matrixplERKNS4_10wi_elementINS1_5intel12experimental8bfloat16ELm8ELm16ELNS4_13matrix_layoutE0ENS2_9sub_groupEEERKS8_: %agg.result"}
-!18 = distinct !{!18, !"_ZN2cl4sycl3ext6oneapi12experimental6matrixplERKNS4_10wi_elementINS1_5intel12experimental8bfloat16ELm8ELm16ELNS4_13matrix_layoutE0ENS2_9sub_groupEEERKS8_"}
-!19 = !{!11, !11, i64 0}
-!20 = !{!21, !17}
-!21 = distinct !{!21, !22, !"_ZN2cl4sycl3ext5intel12experimentalplERKNS3_8bfloat16ES6_: %agg.result"}
-!22 = distinct !{!22, !"_ZN2cl4sycl3ext5intel12experimentalplERKNS3_8bfloat16ES6_"}
-!23 = distinct !{!23, !24}
-!24 = !{!"llvm.loop.mustprogress"}
+!4 = !{!"cpu", i32 1}
+!5 = !{!"gpu", i32 2}
+!6 = !{!"accelerator", i32 3}
+!7 = !{!"custom", i32 4}
+!8 = !{!"fp16", i32 5}
+!9 = !{!"fp64", i32 6}
+!10 = !{!"image", i32 9}
+!11 = !{!"online_compiler", i32 10}
+!12 = !{!"online_linker", i32 11}
+!13 = !{!"queue_profiling", i32 12}
+!14 = !{!"usm_device_allocations", i32 13}
+!15 = !{!"usm_host_allocations", i32 14}
+!16 = !{!"usm_shared_allocations", i32 15}
+!17 = !{!"usm_system_allocations", i32 17}
+!18 = !{!"ext_intel_pci_address", i32 18}
+!19 = !{!"ext_intel_gpu_eu_count", i32 19}
+!20 = !{!"ext_intel_gpu_eu_simd_width", i32 20}
+!21 = !{!"ext_intel_gpu_slices", i32 21}
+!22 = !{!"ext_intel_gpu_subslices_per_slice", i32 22}
+!23 = !{!"ext_intel_gpu_eu_count_per_subslice", i32 23}
+!24 = !{!"ext_intel_max_mem_bandwidth", i32 24}
+!25 = !{!"ext_intel_mem_channel", i32 25}
+!26 = !{!"usm_atomic_host_allocations", i32 26}
+!27 = !{!"usm_atomic_shared_allocations", i32 27}
+!28 = !{!"atomic64", i32 28}
+!29 = !{!"ext_intel_device_info_uuid", i32 29}
+!30 = !{!"ext_oneapi_srgb", i32 30}
+!31 = !{!"ext_oneapi_native_assert", i32 31}
+!32 = !{!"host_debuggable", i32 32}
+!33 = !{!"ext_intel_gpu_hw_threads_per_eu", i32 33}
+!34 = !{!"ext_oneapi_cuda_async_barrier", i32 34}
+!35 = !{!"ext_oneapi_bfloat16_math_functions", i32 35}
+!36 = !{!"ext_intel_free_memory", i32 36}
+!37 = !{!"ext_intel_device_id", i32 37}
+!38 = !{!"ext_intel_memory_clock_rate", i32 38}
+!39 = !{!"ext_intel_memory_bus_width", i32 39}
+!40 = !{!"emulated", i32 40}
+!41 = !{!"ext_intel_legacy_image", i32 41}
+!42 = !{!"int64_base_atomics", i32 7}
+!43 = !{!"int64_extended_atomics", i32 8}
+!44 = !{!"usm_system_allocator", i32 17}
+!45 = !{!"usm_restricted_shared_allocations", i32 16}
+!46 = !{!"host", i32 0}
+!47 = !{!"clang version 17.0.0 (https://github.com/intel/llvm.git 93f477358d74ae90024f758e7eeb97d4b13cea42)"}
+!48 = !{i32 10642943}
+!49 = !{i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1}
+!50 = !{i1 true, i1 false, i1 false, i1 false, i1 true, i1 false, i1 false, i1 false, i1 true, i1 false, i1 false, i1 false}
+!51 = !{i32 16}
+!52 = !{}
+!53 = !{i1 false, i1 true, i1 true, i1 true, i1 false, i1 true, i1 true, i1 true, i1 false, i1 true, i1 true, i1 true}
+!54 = !{!55, !57, !59}
+!55 = distinct !{!55, !56, !"_ZN7__spirv29InitSizesSTGlobalInvocationIdILi2EN4sycl3_V12idILi2EEEE8initSizeEv: %agg.result"}
+!56 = distinct !{!56, !"_ZN7__spirv29InitSizesSTGlobalInvocationIdILi2EN4sycl3_V12idILi2EEEE8initSizeEv"}
+!57 = distinct !{!57, !58, !"_ZN7__spirvL22initGlobalInvocationIdILi2EN4sycl3_V12idILi2EEEEET0_v: %agg.result"}
+!58 = distinct !{!58, !"_ZN7__spirvL22initGlobalInvocationIdILi2EN4sycl3_V12idILi2EEEEET0_v"}
+!59 = distinct !{!59, !60, !"_ZN4sycl3_V16detail7Builder10getElementILi2EEEKNS0_7nd_itemIXT_EEEPS5_: %agg.result"}
+!60 = distinct !{!60, !"_ZN4sycl3_V16detail7Builder10getElementILi2EEEKNS0_7nd_itemIXT_EEEPS5_"}
+!61 = !{!62, !64, !59}
+!62 = distinct !{!62, !63, !"_ZN7__spirv28InitSizesSTLocalInvocationIdILi2EN4sycl3_V12idILi2EEEE8initSizeEv: %agg.result"}
+!63 = distinct !{!63, !"_ZN7__spirv28InitSizesSTLocalInvocationIdILi2EN4sycl3_V12idILi2EEEE8initSizeEv"}
+!64 = distinct !{!64, !65, !"_ZN7__spirvL21initLocalInvocationIdILi2EN4sycl3_V12idILi2EEEEET0_v: %agg.result"}
+!65 = distinct !{!65, !"_ZN7__spirvL21initLocalInvocationIdILi2EN4sycl3_V12idILi2EEEEET0_v"}
+!66 = !{!67}
+!67 = distinct !{!67, !68, !"_ZN4sycl3_V13ext6oneapi12experimental6matrix16joint_matrix_madINS0_9sub_groupENS2_8bfloat16ES7_fLm8ELm16ELm16ELNS4_6layoutE0ELS8_2EEENS4_12joint_matrixIT_T2_LNS4_3useE2EXT3_EXT5_ELS8_3EEESA_RNS9_ISA_T0_LSC_0EXT3_EXT4_EXT6_EEERNS9_ISA_T1_LSC_1EXT4_EXT5_EXT7_EEERSD_: %agg.result"}
+!68 = distinct !{!68, !"_ZN4sycl3_V13ext6oneapi12experimental6matrix16joint_matrix_madINS0_9sub_groupENS2_8bfloat16ES7_fLm8ELm16ELm16ELNS4_6layoutE0ELS8_2EEENS4_12joint_matrixIT_T2_LNS4_3useE2EXT3_EXT5_ELS8_3EEESA_RNS9_ISA_T0_LSC_0EXT3_EXT4_EXT6_EEERNS9_ISA_T1_LSC_1EXT4_EXT5_EXT7_EEERSD_"}
+!69 = distinct !{!69, !70}
+!70 = !{!"llvm.loop.mustprogress"}
diff --git a/test/extensions/INTEL/SPV_INTEL_joint_matrix/joint_matrix_element.ll b/test/extensions/INTEL/SPV_INTEL_joint_matrix/joint_matrix_element.ll
deleted file mode 100644
index 9ddff0fc2e..0000000000
--- a/test/extensions/INTEL/SPV_INTEL_joint_matrix/joint_matrix_element.ll
+++ /dev/null
@@ -1,121 +0,0 @@
-; RUN: llvm-as -opaque-pointers=0 < %s -o %t.bc
-; RUN: llvm-spirv %t.bc -opaque-pointers=0 -spirv-ext=+all -o %t.spv
-; RUN: llvm-spirv %t.spv -to-text -o - | FileCheck %s --check-prefix=CHECK-SPIRV
-
-; RUN: llvm-spirv -r -emit-opaque-pointers %t.spv -o %t.rev.bc
-; RUN: llvm-dis %t.rev.bc -o - | FileCheck %s --check-prefix=CHECK-LLVM
-
-; CHECK-SPIRV-DAG: Capability JointMatrixINTEL
-; CHECK-SPIRV-DAG: Capability JointMatrixWIInstructionsINTEL
-; CHECK-SPIRV-DAG: Extension "SPV_INTEL_joint_matrix"
-; CHECK-SPIRV-DAG: TypeInt [[#TypeInt32:]] 32
-; CHECK-SPIRV-DAG: TypeInt [[#TypeInt64:]] 64
-; CHECK-SPIRV-DAG: TypeFloat [[#TypeFloat:]] 32
-; CHECK-SPIRV-DAG: TypeJointMatrixINTEL [[#TypeMatrix:]] [[#TypeFloat]] [[#]] [[#]] [[#]] [[#]]
-; CHECK-SPIRV-DAG: TypeVector [[#TypeVec:]] [[#TypeInt32]] 2
-; CHECK-SPIRV: Phi [[#TypeMatrix]] [[#Matrix:]]
-; CHECK-SPIRV: JointMatrixWorkItemLengthINTEL [[#TypeInt64]] [[#]] [[#Matrix]]
-; CHECK-SPIRV: VectorExtractDynamic [[#TypeFloat]] [[#]] [[#Matrix]] [[#Index:]]
-; CHECK-SPIRV: FMul [[#TypeFloat]] [[#NewVal:]] [[#]] [[#]]
-; CHECK-SPIRV: VectorInsertDynamic [[#TypeMatrix]] [[#]] [[#Matrix]] [[#NewVal]] [[#Index]]
-; CHECK-SPIRV: JointMatrixGetElementCoordINTEL [[#TypeVec]] [[#]] [[#Matrix]] [[#Index]]
-
-; CHECK-LLVM: [[Length:%.*]] = call spir_func i64 @_Z38__spirv_JointMatrixWorkItemLengthINTELPU3AS141__spirv_JointMatrixINTEL__float_16_16_0_3(ptr addrspace(1) [[Matrix:%.*]])
-; CHECK-LLVM: [[Elem:%.*]] = call spir_func float @_Z28__spirv_VectorExtractDynamicPU3AS141__spirv_JointMatrixINTEL__float_16_16_0_3l(ptr addrspace(1) [[Matrix]], i64 [[Index:%.*]])
-; CHECK-LLVM: [[NewVal:%.*]] = fmul float [[Elem]], 5.000000e+00
-; CHECK-LLVM: {{%.*}} = call spir_func ptr addrspace(1) @_Z27__spirv_VectorInsertDynamicPU3AS141__spirv_JointMatrixINTEL__float_16_16_0_3fl(ptr addrspace(1) [[Matrix]], float [[NewVal]], i64 [[Index]])
-; CHECK-LLVM: {{%.*}} = call spir_func <2 x i32> @_Z39__spirv_JointMatrixGetElementCoordINTELPU3AS141__spirv_JointMatrixINTEL__float_16_16_0_3l(ptr addrspace(1) [[Matrix]], i64 [[Index]])
-
-source_filename = "/work/tmp/matrix-slice.cpp"
-target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-n8:16:32:64"
-target triple = "spir64-unknown-unknown"
-
-%"struct.cl::sycl::detail::AssertHappened" = type { i32, [257 x i8], [257 x i8], [129 x i8], i32, i64, i64, i64, i64, i64, i64 }
-%"class.cl::sycl::range" = type { %"class.cl::sycl::detail::array" }
-%"class.cl::sycl::detail::array" = type { [1 x i64] }
-%"class.cl::sycl::id" = type { %"class.cl::sycl::detail::array" }
-%spirv.JointMatrixINTEL._float_16_16_0_3 = type opaque
-
-$_ZTSN2cl4sycl6detail23__sycl_service_kernel__16AssertInfoCopierE = comdat any
-
-$_ZTSZZ4mainENKUlRN2cl4sycl7handlerEE_clES2_E6matrix = comdat any
-
-; Function Attrs: convergent norecurse
-define weak_odr dso_local spir_kernel void @_ZTSN2cl4sycl6detail23__sycl_service_kernel__16AssertInfoCopierE(%"struct.cl::sycl::detail::AssertHappened" addrspace(1)* %_arg_, %"class.cl::sycl::range"* byval(%"class.cl::sycl::range") align 8 %_arg_1, %"class.cl::sycl::range"* byval(%"class.cl::sycl::range") align 8 %_arg_2, %"class.cl::sycl::id"* byval(%"class.cl::sycl::id") align 8 %_arg_3) local_unnamed_addr #0 comdat !kernel_arg_buffer_location !5 {
-entry:
-  %0 = getelementptr inbounds %"class.cl::sycl::id", %"class.cl::sycl::id"* %_arg_3, i64 0, i32 0, i32 0, i64 0
-  %1 = addrspacecast i64* %0 to i64 addrspace(4)*
-  %2 = load i64, i64 addrspace(4)* %1, align 8
-  %add.ptr.i = getelementptr inbounds %"struct.cl::sycl::detail::AssertHappened", %"struct.cl::sycl::detail::AssertHappened" addrspace(1)* %_arg_, i64 %2
-  %3 = bitcast %"struct.cl::sycl::detail::AssertHappened" addrspace(1)* %add.ptr.i to i8 addrspace(1)*
-  %4 = addrspacecast i8 addrspace(1)* %3 to i8 addrspace(4)*
-  tail call spir_func void @__devicelib_assert_read(i8 addrspace(4)* %4) #2
-  ret void
-}
-
-; Function Attrs: convergent
-declare extern_weak dso_local spir_func void @__devicelib_assert_read(i8 addrspace(4)*) local_unnamed_addr #1
-
-; Function Attrs: convergent norecurse
-define weak_odr dso_local spir_kernel void @_ZTSZZ4mainENKUlRN2cl4sycl7handlerEE_clES2_E6matrix() local_unnamed_addr #0 comdat !kernel_arg_buffer_location !6 {
-entry:
-  %call9.i.i = tail call spir_func %spirv.JointMatrixINTEL._float_16_16_0_3 addrspace(4)* @_Z28__spirv_JointMatrixLoadINTELIfLm16ELm16ELN5__spv12MatrixLayoutE0ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEEPS5_mS1_S3_i(float addrspace(4)* addrspacecast (float addrspace(1)* null to float addrspace(4)*), i64 1, i32 0, i32 3, i32 0) #2
-  br label %for.cond.i
-
-for.cond.i:                                       ; preds = %for.body.i, %entry
-  %A.sroa.0.0.i = phi %spirv.JointMatrixINTEL._float_16_16_0_3 addrspace(4)* [ %call9.i.i, %entry ], [ %call5.i.i, %for.body.i ]
-  %i.0.i = phi i32 [ 0, %entry ], [ %inc.i, %for.body.i ]
-  %conv.i = zext i32 %i.0.i to i64
-  %call.i12.i = tail call spir_func i64 @_Z38__spirv_JointMatrixWorkItemLengthINTELIfLm16ELm16ELN5__spv12MatrixLayoutE0ELNS0_5Scope4FlagE3EEmPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEE(%spirv.JointMatrixINTEL._float_16_16_0_3 addrspace(4)* %A.sroa.0.0.i) #2
-  %cmp.i = icmp ugt i64 %call.i12.i, %conv.i
-  br i1 %cmp.i, label %for.body.i, label %_ZZZ4mainENKUlRN2cl4sycl7handlerEE_clES2_ENKUlNS0_7nd_itemILi2EEEE_clES5_.exit
-
-for.body.i:                                       ; preds = %for.cond.i
-  %call.i.i = tail call spir_func float @_Z28__spirv_VectorExtractDynamicIfLm16ELm16ELN5__spv12MatrixLayoutE0ELNS0_5Scope4FlagE3EmET_PNS0_24__spirv_JointMatrixINTELIS4_XT0_EXT1_EXT2_EXT3_EEET4_(%spirv.JointMatrixINTEL._float_16_16_0_3 addrspace(4)* %A.sroa.0.0.i, i64 %conv.i) #2
-  %mul.i.i = fmul float %call.i.i, 5.000000e+00
-  %call5.i.i = tail call spir_func %spirv.JointMatrixINTEL._float_16_16_0_3 addrspace(4)* @_Z27__spirv_VectorInsertDynamicIfLm16ELm16ELN5__spv12MatrixLayoutE0ELNS0_5Scope4FlagE3EmEPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEES7_T4_S5_(%spirv.JointMatrixINTEL._float_16_16_0_3 addrspace(4)* %A.sroa.0.0.i, float %mul.i.i, i64 %conv.i) #2
-  %call6 = tail call spir_func <2 x i32> @_Z39__spirv_JointMatrixGetElementCoordINTELIaLm8ELm32ELN5__spv9MatrixUseE0ELNS0_12MatrixLayoutE0ELNS0_5Scope4FlagE3EEDv2_jPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT3_EXT4_EXT2_EEEm(%spirv.JointMatrixINTEL._float_16_16_0_3 addrspace(4)* %A.sroa.0.0.i, i64 %conv.i) #2
-  %inc.i = add nuw nsw i32 %i.0.i, 1
-  br label %for.cond.i, !llvm.loop !7
-
-_ZZZ4mainENKUlRN2cl4sycl7handlerEE_clES2_ENKUlNS0_7nd_itemILi2EEEE_clES5_.exit: ; preds = %for.cond.i
-  tail call spir_func void @_Z29__spirv_JointMatrixStoreINTELIfLm16ELm16ELN5__spv12MatrixLayoutE0ELNS0_5Scope4FlagE3EEvPT_PNS0_24__spirv_JointMatrixINTELIS4_XT0_EXT1_EXT2_EXT3_EEEmS1_S3_i(float addrspace(4)* addrspacecast (float addrspace(1)* null to float addrspace(4)*), %spirv.JointMatrixINTEL._float_16_16_0_3 addrspace(4)* %A.sroa.0.0.i, i64 1, i32 0, i32 3, i32 0) #2
-  ret void
-}
-
-; Function Attrs: convergent
-declare dso_local spir_func %spirv.JointMatrixINTEL._float_16_16_0_3 addrspace(4)* @_Z28__spirv_JointMatrixLoadINTELIfLm16ELm16ELN5__spv12MatrixLayoutE0ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEEPS5_mS1_S3_i(float addrspace(4)*, i64, i32, i32, i32) local_unnamed_addr #1
-
-; Function Attrs: convergent
-declare dso_local spir_func i64 @_Z38__spirv_JointMatrixWorkItemLengthINTELIfLm16ELm16ELN5__spv12MatrixLayoutE0ELNS0_5Scope4FlagE3EEmPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEE(%spirv.JointMatrixINTEL._float_16_16_0_3 addrspace(4)*) local_unnamed_addr #1
-
-; Function Attrs: convergent
-declare dso_local spir_func float @_Z28__spirv_VectorExtractDynamicIfLm16ELm16ELN5__spv12MatrixLayoutE0ELNS0_5Scope4FlagE3EmET_PNS0_24__spirv_JointMatrixINTELIS4_XT0_EXT1_EXT2_EXT3_EEET4_(%spirv.JointMatrixINTEL._float_16_16_0_3 addrspace(4)*, i64) local_unnamed_addr #1
-
-; Function Attrs: convergent
-declare dso_local spir_func %spirv.JointMatrixINTEL._float_16_16_0_3 addrspace(4)* @_Z27__spirv_VectorInsertDynamicIfLm16ELm16ELN5__spv12MatrixLayoutE0ELNS0_5Scope4FlagE3EmEPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEES7_T4_S5_(%spirv.JointMatrixINTEL._float_16_16_0_3 addrspace(4)*, float, i64) local_unnamed_addr #1
-
-; Function Attrs: convergent
-declare dso_local spir_func void @_Z29__spirv_JointMatrixStoreINTELIfLm16ELm16ELN5__spv12MatrixLayoutE0ELNS0_5Scope4FlagE3EEvPT_PNS0_24__spirv_JointMatrixINTELIS4_XT0_EXT1_EXT2_EXT3_EEEmS1_S3_i(float addrspace(4)*, %spirv.JointMatrixINTEL._float_16_16_0_3 addrspace(4)*, i64, i32, i32, i32) local_unnamed_addr #1
-
-; Function Attrs: convergent
-declare dso_local spir_func <2 x i32> @_Z39__spirv_JointMatrixGetElementCoordINTELIaLm8ELm32ELN5__spv9MatrixUseE0ELNS0_12MatrixLayoutE0ELNS0_5Scope4FlagE3EEDv2_jPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT3_EXT4_EXT2_EEEm(%spirv.JointMatrixINTEL._float_16_16_0_3 addrspace(4)*, i64) #2
-
-attributes #0 = { convergent norecurse "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "sycl-module-id"="/work/tmp/matrix-slice.cpp" "uniform-work-group-size"="true" }
-attributes #1 = { convergent "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
-attributes #2 = { convergent }
-
-!llvm.module.flags = !{!0, !1}
-!opencl.spir.version = !{!2}
-!spirv.Source = !{!3}
-!llvm.ident = !{!4}
-
-!0 = !{i32 1, !"wchar_size", i32 4}
-!1 = !{i32 7, !"frame-pointer", i32 2}
-!2 = !{i32 1, i32 2}
-!3 = !{i32 4, i32 100000}
-!4 = !{!"clang version 14.0.0 (https://github.com/intel/llvm.git 3648adf79e4fdb619fdbe41d63bc39f456b5be8c)"}
-!5 = !{i32 -1, i32 -1, i32 -1, i32 -1}
-!6 = !{}
-!7 = distinct !{!7, !8}
-!8 = !{!"llvm.loop.mustprogress"}
diff --git a/test/extensions/INTEL/SPV_INTEL_joint_matrix/joint_matrix_extract_insert_element_of_sycl_half.ll b/test/extensions/INTEL/SPV_INTEL_joint_matrix/joint_matrix_extract_insert_element_of_sycl_half.ll
deleted file mode 100644
index a0f259934f..0000000000
--- a/test/extensions/INTEL/SPV_INTEL_joint_matrix/joint_matrix_extract_insert_element_of_sycl_half.ll
+++ /dev/null
@@ -1,130 +0,0 @@
-; RUN: llvm-as -opaque-pointers=0 %s -o %t.bc
-
-; RUN: llvm-spirv -s %t.bc -opaque-pointers=0 -o %t.regularized.bc
-; RUN: llvm-dis -opaque-pointers=0 %t.regularized.bc -o %t.regularized.ll
-; RUN: FileCheck < %t.regularized.ll %s --check-prefix=CHECK-REGULARIZED
-
-; RUN: llvm-spirv %t.bc -opaque-pointers=0 --spirv-ext=+SPV_INTEL_joint_matrix -o %t.spv
-; RUN: llvm-spirv -to-text %t.spv -o %t.spt
-; RUN: FileCheck < %t.spt %s --check-prefix=CHECK-SPIRV
-
-; RUN: llvm-spirv -r -emit-opaque-pointers %t.spv -o %t.rev.bc
-; RUN: llvm-dis %t.rev.bc -o %t.rev.ll
-; RUN: FileCheck < %t.rev.ll %s --check-prefix=CHECK-LLVM
-
-; CHECK-REGULARIZED: %[[#ExtractElementCall:]] = call spir_func half @_Z28__spirv_VectorExtractDynamicIN2cl4sycl6detail9half_impl4halfELm8ELm16ELN5__spv12MatrixLayoutE0ELNS5_5Scope4FlagE3EET_PNS5_24__spirv_JointMatrixINTELIS9_XT0_EXT1_EXT2_EXT3_EEEm(%spirv.JointMatrixINTEL._half_8_16_0_3 addrspace(4)* align 2{{.*}}, i64{{.*}})
-; CHECK-REGULARIZED: %[[#GEP:]] = getelementptr inbounds %"class.cl::sycl::detail::half_impl::half", %"class.cl::sycl::detail::half_impl::half" addrspace(4)*{{.*}}, i32 0, i32 0
-; CHECK-REGULARIZED: store half %[[#ExtractElementCall]], half addrspace(4)* %[[#GEP]]
-; CHECK-REGULARIZED: %[[#GEP:]] = getelementptr inbounds %"class.cl::sycl::detail::half_impl::half", %"class.cl::sycl::detail::half_impl::half"*{{.*}}, i32 0, i32 0
-; CHECK-REGULARIZED: %[[#Component:]] = load half, half*{{.*}}, align 2
-; CHECK-REGULARIZED: call spir_func %spirv.JointMatrixINTEL._half_8_16_0_3 addrspace(4)* @_Z27__spirv_VectorInsertDynamicIN2cl4sycl6detail9half_impl4halfELm8ELm16ELN5__spv12MatrixLayoutE0ELNS5_5Scope4FlagE3EEPNS5_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEESC_SA_m(%spirv.JointMatrixINTEL._half_8_16_0_3 addrspace(4)*{{.*}}, half %[[#Component]], i64{{.*}})
-; CHECK-REGULARIZED: declare dso_local spir_func half @_Z28__spirv_VectorExtractDynamicIN2cl4sycl6detail9half_impl4halfELm8ELm16ELN5__spv12MatrixLayoutE0ELNS5_5Scope4FlagE3EET_PNS5_24__spirv_JointMatrixINTELIS9_XT0_EXT1_EXT2_EXT3_EEEm(%spirv.JointMatrixINTEL._half_8_16_0_3 addrspace(4)* align 2, i64)
-; CHECK-REGULARIZED: declare dso_local spir_func %spirv.JointMatrixINTEL._half_8_16_0_3 addrspace(4)* @_Z27__spirv_VectorInsertDynamicIN2cl4sycl6detail9half_impl4halfELm8ELm16ELN5__spv12MatrixLayoutE0ELNS5_5Scope4FlagE3EEPNS5_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEESC_SA_m(%spirv.JointMatrixINTEL._half_8_16_0_3 addrspace(4)*, half, i64)
-
-; CHECK-SPIRV: Name [[#VIDValueId:]] "agg.tmp.ascast.ascast"
-; CHECK-SPIRV: TypeFloat [[#Float16Id:]] 16
-; CHECK-SPIRV: TypeJointMatrixINTEL [[#JointMatrixTyId:]] [[#]] [[#]] [[#]] [[#]] [[#]]
-; CHECK-SPIRV: VectorExtractDynamic [[#Float16Id]] [[#VEDId:]] [[#]] [[#]]
-; CHECK-SPIRV: Store [[#]] [[#VEDId]]
-; CHECK-SPIRV: PtrAccessChain [[#]] [[#GEPId:]] [[#VIDValueId]] [[#]] [[#]]
-; CHECK-SPIRV: Load [[#Float16Id]] [[#ComponentId:]] [[#GEPId]]
-; CHECK-SPIRV: VectorInsertDynamic [[#JointMatrixTyId]] [[#]] [[#]] [[#ComponentId]] [[#]]
-
-; CHECK-LLVM: %[[#ExtractElementCall:]] = call spir_func half @_Z28__spirv_VectorExtractDynamicPU3AS139__spirv_JointMatrixINTEL__half_8_16_0_3l(ptr addrspace(1){{.*}}, i64{{.*}})
-; CHECK-LLVM: %[[#GEP:]] = getelementptr inbounds %"class.cl::sycl::detail::half_impl::half", ptr addrspace(4) {{.*}}, i32 0, i32 0
-; CHECK-LLVM: store half %[[#ExtractElementCall]], ptr addrspace(4) %[[#GEP]]
-
-; CHECK-LLVM: %[[#GEP:]] = getelementptr inbounds %"class.cl::sycl::detail::half_impl::half", ptr{{.*}}, i32 0, i32 0
-; CHECK-LLVM: %[[#Component:]] = load half, ptr %[[#GEP]]
-; CHECK-LLVM: spir_func ptr addrspace(1) @_Z27__spirv_VectorInsertDynamicPU3AS139__spirv_JointMatrixINTEL__half_8_16_0_3Dhl(ptr addrspace(1){{.*}}, half %[[#Component]], i64{{.*}})
-
-; ModuleID = 'element_wise_all_ops_half.bc'
-source_filename = "llvm-link"
-target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-n8:16:32:64"
-target triple = "spir64-unknown-unknown"
-
-%"class.cl::sycl::detail::half_impl::half" = type { half }
-%"struct.cl::sycl::ext::oneapi::experimental::matrix::joint_matrix" = type { %spirv.JointMatrixINTEL._half_8_16_0_3 addrspace(4)* }
-%spirv.JointMatrixINTEL._half_8_16_0_3 = type opaque
-%"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element" = type { %"struct.cl::sycl::ext::oneapi::experimental::matrix::joint_matrix" addrspace(4)*, i64 }
-
-$_ZN2cl4sycl3ext6oneapi12experimental6matrixplERKNS4_10wi_elementINS0_6detail9half_impl4halfELm8ELm16ELNS4_13matrix_layoutE0ENS2_9sub_groupEEERKS8_ = comdat any
-
-$_ZN2cl4sycl3ext6oneapi12experimental6matrix10wi_elementINS0_6detail9half_impl4halfELm8ELm16ELNS4_13matrix_layoutE0ENS2_9sub_groupEEaSERKS8_ = comdat any
-
-; Function Attrs: convergent mustprogress norecurse
-define linkonce_odr dso_local spir_func void @_ZN2cl4sycl3ext6oneapi12experimental6matrixplERKNS4_10wi_elementINS0_6detail9half_impl4halfELm8ELm16ELNS4_13matrix_layoutE0ENS2_9sub_groupEEERKS8_(%"class.cl::sycl::detail::half_impl::half" addrspace(4)* noalias sret(%"class.cl::sycl::detail::half_impl::half") align 2 %agg.result, %"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element" addrspace(4)* align 8 dereferenceable(16) %lhs, %"class.cl::sycl::detail::half_impl::half" addrspace(4)* align 2 dereferenceable(2) %rhs) #0 comdat {
-entry:
-  %lhs.addr = alloca %"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element" addrspace(4)*, align 8
-  %ref.tmp1 = alloca %"class.cl::sycl::detail::half_impl::half", align 2
-  %lhs.addr.ascast = addrspacecast %"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element" addrspace(4)** %lhs.addr to %"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element" addrspace(4)* addrspace(4)*
-  %ref.tmp1.ascast = addrspacecast %"class.cl::sycl::detail::half_impl::half"* %ref.tmp1 to %"class.cl::sycl::detail::half_impl::half" addrspace(4)*
-  %0 = load %"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element" addrspace(4)*, %"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element" addrspace(4)* addrspace(4)* %lhs.addr.ascast, align 8, !tbaa !8
-  %M = getelementptr inbounds %"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element", %"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element" addrspace(4)* %0, i32 0, i32 0
-  %1 = load %"struct.cl::sycl::ext::oneapi::experimental::matrix::joint_matrix" addrspace(4)*, %"struct.cl::sycl::ext::oneapi::experimental::matrix::joint_matrix" addrspace(4)* addrspace(4)* %M, align 8, !tbaa !15
-  %spvm = getelementptr inbounds %"struct.cl::sycl::ext::oneapi::experimental::matrix::joint_matrix", %"struct.cl::sycl::ext::oneapi::experimental::matrix::joint_matrix" addrspace(4)* %1, i32 0, i32 0
-  %2 = load %spirv.JointMatrixINTEL._half_8_16_0_3 addrspace(4)*, %spirv.JointMatrixINTEL._half_8_16_0_3 addrspace(4)* addrspace(4)* %spvm, align 8, !tbaa !13
-  %3 = load %"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element" addrspace(4)*, %"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element" addrspace(4)* addrspace(4)* %lhs.addr.ascast, align 8, !tbaa !8
-  %idx = getelementptr inbounds %"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element", %"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element" addrspace(4)* %3, i32 0, i32 1
-  %4 = load i64, i64 addrspace(4)* %idx, align 8, !tbaa !17
-  call spir_func void @_Z28__spirv_VectorExtractDynamicIN2cl4sycl6detail9half_impl4halfELm8ELm16ELN5__spv12MatrixLayoutE0ELNS5_5Scope4FlagE3EET_PNS5_24__spirv_JointMatrixINTELIS9_XT0_EXT1_EXT2_EXT3_EEEm(%"class.cl::sycl::detail::half_impl::half" addrspace(4)* sret(%"class.cl::sycl::detail::half_impl::half") align 2 %ref.tmp1.ascast, %spirv.JointMatrixINTEL._half_8_16_0_3 addrspace(4)* %2, i64 %4) #2
-  ret void
-}
-
-; Function Attrs: convergent mustprogress norecurse
-define linkonce_odr dso_local spir_func align 8 dereferenceable(16) %"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element" addrspace(4)* @_ZN2cl4sycl3ext6oneapi12experimental6matrix10wi_elementINS0_6detail9half_impl4halfELm8ELm16ELNS4_13matrix_layoutE0ENS2_9sub_groupEEaSERKS8_(%"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element" addrspace(4)* align 8 dereferenceable_or_null(16) %this, %"class.cl::sycl::detail::half_impl::half" addrspace(4)* align 2 dereferenceable(2) %rhs) #0 comdat align 2 {
-entry:
-  %this.addr = alloca %"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element" addrspace(4)*, align 8
-  %agg.tmp = alloca %"class.cl::sycl::detail::half_impl::half", align 2
-  %this.addr.ascast = addrspacecast %"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element" addrspace(4)** %this.addr to %"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element" addrspace(4)* addrspace(4)*
-  %agg.tmp.ascast = addrspacecast %"class.cl::sycl::detail::half_impl::half"* %agg.tmp to %"class.cl::sycl::detail::half_impl::half" addrspace(4)*
-  %this1 = load %"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element" addrspace(4)*, %"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element" addrspace(4)* addrspace(4)* %this.addr.ascast, align 8
-  %M = getelementptr inbounds %"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element", %"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element" addrspace(4)* %this1, i32 0, i32 0
-  %0 = load %"struct.cl::sycl::ext::oneapi::experimental::matrix::joint_matrix" addrspace(4)*, %"struct.cl::sycl::ext::oneapi::experimental::matrix::joint_matrix" addrspace(4)* addrspace(4)* %M, align 8, !tbaa !15
-  %spvm = getelementptr inbounds %"struct.cl::sycl::ext::oneapi::experimental::matrix::joint_matrix", %"struct.cl::sycl::ext::oneapi::experimental::matrix::joint_matrix" addrspace(4)* %0, i32 0, i32 0
-  %1 = load %spirv.JointMatrixINTEL._half_8_16_0_3 addrspace(4)*, %spirv.JointMatrixINTEL._half_8_16_0_3 addrspace(4)* addrspace(4)* %spvm, align 8, !tbaa !13
-  %idx = getelementptr inbounds %"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element", %"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element" addrspace(4)* %this1, i32 0, i32 1
-  %2 = load i64, i64 addrspace(4)* %idx, align 8, !tbaa !17
-  %agg.tmp.ascast.ascast = addrspacecast %"class.cl::sycl::detail::half_impl::half" addrspace(4)* %agg.tmp.ascast to %"class.cl::sycl::detail::half_impl::half"*
-  %call = call spir_func %spirv.JointMatrixINTEL._half_8_16_0_3 addrspace(4)* @_Z27__spirv_VectorInsertDynamicIN2cl4sycl6detail9half_impl4halfELm8ELm16ELN5__spv12MatrixLayoutE0ELNS5_5Scope4FlagE3EEPNS5_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEESC_SA_m(%spirv.JointMatrixINTEL._half_8_16_0_3 addrspace(4)* %1, %"class.cl::sycl::detail::half_impl::half"* byval(%"class.cl::sycl::detail::half_impl::half") align 2 %agg.tmp.ascast.ascast, i64 %2) #2
-  ret %"class.cl::sycl::ext::oneapi::experimental::matrix::wi_element" addrspace(4)* %this1
-}
-
-; Function Attrs: convergent
-declare dso_local spir_func %spirv.JointMatrixINTEL._half_8_16_0_3 addrspace(4)* @_Z27__spirv_VectorInsertDynamicIN2cl4sycl6detail9half_impl4halfELm8ELm16ELN5__spv12MatrixLayoutE0ELNS5_5Scope4FlagE3EEPNS5_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEESC_SA_m(%spirv.JointMatrixINTEL._half_8_16_0_3 addrspace(4)*, %"class.cl::sycl::detail::half_impl::half"* byval(%"class.cl::sycl::detail::half_impl::half") align 2, i64) #1
-
-; Function Attrs: convergent
-declare dso_local spir_func void @_Z28__spirv_VectorExtractDynamicIN2cl4sycl6detail9half_impl4halfELm8ELm16ELN5__spv12MatrixLayoutE0ELNS5_5Scope4FlagE3EET_PNS5_24__spirv_JointMatrixINTELIS9_XT0_EXT1_EXT2_EXT3_EEEm(%"class.cl::sycl::detail::half_impl::half" addrspace(4)* sret(%"class.cl::sycl::detail::half_impl::half") align 2, %spirv.JointMatrixINTEL._half_8_16_0_3 addrspace(4)*, i64) #1
-
-attributes #0 = { convergent mustprogress norecurse "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
-attributes #1 = { convergent "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
-attributes #2 = { convergent }
-
-!opencl.spir.version = !{!0, !0, !0, !0, !0, !0, !0, !0, !0, !0, !0, !0}
-!spirv.Source = !{!1, !1, !1, !1, !1, !1, !1, !1, !1, !1, !1, !1}
-!opencl.used.extensions = !{!2, !3, !3, !3, !3, !3, !3, !3, !3, !3, !3, !3}
-!opencl.used.optional.core.features = !{!4, !3, !3, !4, !3, !4, !3, !3, !3, !4, !3, !4}
-!opencl.compiler.options = !{!3, !3, !3, !3, !3, !3, !3, !3, !3, !3, !3, !3}
-!llvm.ident = !{!5, !5, !5, !5, !5, !5, !5, !5, !5, !5, !5, !5}
-!llvm.module.flags = !{!6, !7}
-!sycl.specialization-constants = !{}
-!sycl.specialization-constants-default-values = !{}
-
-!0 = !{i32 1, i32 2}
-!1 = !{i32 4, i32 100000}
-!2 = !{!"cl_khr_fp16"}
-!3 = !{}
-!4 = !{!"cl_doubles"}
-!5 = !{!"Compiler"}
-!6 = !{i32 1, !"wchar_size", i32 4}
-!7 = !{i32 7, !"frame-pointer", i32 2}
-!8 = !{!9, !9, i64 0}
-!9 = !{!"any pointer", !10, i64 0}
-!10 = !{!"omnipotent char", !11, i64 0}
-!11 = !{!"Simple C++ TBAA"}
-!12 = !{!"long", !10, i64 0}
-!13 = !{!14, !9, i64 0}
-!14 = !{!"_ZTSN2cl4sycl3ext6oneapi12experimental6matrix12joint_matrixINS0_6detail9half_impl4halfELm8ELm16ELNS4_13matrix_layoutE0ENS2_9sub_groupEEE", !9, i64 0}
-!15 = !{!16, !9, i64 0}
-!16 = !{!"_ZTSN2cl4sycl3ext6oneapi12experimental6matrix10wi_elementINS0_6detail9half_impl4halfELm8ELm16ELNS4_13matrix_layoutE0ENS2_9sub_groupEEE", !9, i64 0, !12, i64 8}
-!17 = !{!16, !12, i64 8}
diff --git a/test/extensions/INTEL/SPV_INTEL_joint_matrix/joint_matrix_half.ll b/test/extensions/INTEL/SPV_INTEL_joint_matrix/joint_matrix_half.ll
index 8ea9b3041d..6b5c380de5 100644
--- a/test/extensions/INTEL/SPV_INTEL_joint_matrix/joint_matrix_half.ll
+++ b/test/extensions/INTEL/SPV_INTEL_joint_matrix/joint_matrix_half.ll
@@ -1,166 +1,210 @@
-; RUN: llvm-as -opaque-pointers=0 < %s -o %t.bc
-; RUN: llvm-spirv %t.bc -opaque-pointers=0 -spirv-ext=+SPV_INTEL_joint_matrix -o %t.spv
-; RUN: llvm-spirv %t.spv -to-text -o - | FileCheck %s --check-prefix=CHECK-SPIRV
+; compiled from joint_matrix_half.cpp test from intel/llvm
+
+; RUN: llvm-as < %s -o %t.bc
+
+; RUN: llvm-spirv %t.bc --spirv-ext=+SPV_INTEL_bfloat16_conversion,+SPV_INTEL_joint_matrix -o %t.spv
+; RUN: llvm-spirv %t.spv -to-text -o %t.spt
+; RUN: FileCheck < %t.spt %s --check-prefix=CHECK-SPIRV
 
 ; RUN: llvm-spirv -r %t.spv -o %t.rev.bc
-; RUN: llvm-dis -opaque-pointers=0 %t.rev.bc -o - | FileCheck %s --check-prefix=CHECK-LLVM
-
-; CHECK-SPIRV-DAG: TypeFloat [[#FloatTy:]] 32
-; CHECK-SPIRV-DAG: TypeFloat [[#HalfTy:]] 16
-; CHECK-SPIRV-DAG: TypeInt [[#IntTy:]] 32 0
-; CHECK-SPIRV-DAG: Constant [[#IntTy]] [[#Zero:]] 0
-; CHECK-SPIRV-DAG: Constant [[#IntTy]] [[#Two:]] 2
-; CHECK-SPIRV-DAG: Constant [[#IntTy]] [[#Three:]] 3
-; CHECK-SPIRV-DAG: Constant [[#IntTy]] [[#Sixteen:]] 16
-; CHECK-SPIRV: TypeJointMatrixINTEL [[#CTy:]] [[#FloatTy]] [[#Two]] [[#Two]] [[#Zero]] [[#Three]]
-; CHECK-SPIRV: TypeJointMatrixINTEL [[#ATy:]] [[#HalfTy]] [[#Two]] [[#Sixteen]] [[#Zero]] [[#Three]]
-; CHECK-SPIRV: TypeJointMatrixINTEL [[#BTy:]] [[#HalfTy]] [[#Sixteen]] [[#Two]] [[#Three]] [[#Three]]
-
-; CHECK-LLVM: %spirv.JointMatrixINTEL._float_2_2_0_3
-; CHECK-LLVM: %spirv.JointMatrixINTEL._half_2_16_0_3
-; CHECK-LLVM: %spirv.JointMatrixINTEL._half_16_2_3_3
-
-; ModuleID = 'joint_matrix_test-sycl-spir64-unknown-unknown.bc'
-source_filename = "joint_matrix_test.cpp"
+; RUN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM
+
+; CHECK-SPIRV-DAG: TypeInt [[#INT:]] 32
+; CHECK-SPIRV-DAG: TypeFloat [[#Half:]] 16
+; CHECK-SPIRV-DAG: TypeFloat [[#Float:]] 32
+; CHECK-SPIRV-DAG: Constant [[#INT]] [[#CONST8:]] 8
+; CHECK-SPIRV-DAG: Constant [[#INT]] [[#CONST16:]] 16
+; CHECK-SPIRV-DAG: Constant [[#INT]] [[#CONST3:]] 3
+; CHECK-SPIRV-DAG: Constant [[#INT]] [[#CONST2:]] 2
+; CHECK-SPIRV-DAG: Constant [[#INT]] [[#CONST1:]] 1
+; CHECK-SPIRV-DAG: Constant [[#INT]] [[#CONST0:]] 0
+; CHECK-SPIRV-DAG: TypeJointMatrixINTEL [[#MatTy1:]] [[#Float]] [[#CONST8]] [[#CONST16]] [[#CONST3]] [[#CONST3]] [[#CONST2]]
+; CHECK-SPIRV-DAG: TypeJointMatrixINTEL [[#MatTy2:]] [[#Half]] [[#CONST8]] [[#CONST16]] [[#CONST0]] [[#CONST3]] [[#CONST0]]
+; CHECK-SPIRV-DAG: TypeJointMatrixINTEL [[#MatTy3:]] [[#Half]] [[#CONST16]] [[#CONST16]] [[#CONST2]] [[#CONST3]] [[#CONST1]]
+
+; CHECK-LLVM: call spir_func target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) @_Z80__spirv_JointMatrixLoadINTEL_RPU3AS142__spirv_JointMatrixINTEL__float_8_16_3_3_2PU3AS1fliii(ptr addrspace(1) %{{.*}}, i64 %{{.*}}, i32 0, i32 3, i32 0)
+; CHECK-LLVM: call spir_func target("spirv.JointMatrixINTEL", half, 8, 16, 0, 3, 0) @"_Z79__spirv_JointMatrixLoadINTEL_RPU3AS141__spirv_JointMatrixINTEL__half_8_16_0_3_0PU3AS140class.sycl::_V1::detail::half_impl::halfliii"(ptr addrspace(1) %{{.*}}, i64 %{{.*}}, i32 0, i32 3, i32 0)
+; CHECK-LLVM: call spir_func target("spirv.JointMatrixINTEL", half, 16, 16, 2, 3, 1) @"_Z80__spirv_JointMatrixLoadINTEL_RPU3AS142__spirv_JointMatrixINTEL__half_16_16_2_3_1PU3AS140class.sycl::_V1::detail::half_impl::halfliii"(ptr addrspace(1) %{{.*}}, i64 %{{.*}}, i32 2, i32 3, i32 0)
+; CHECK-LLVM: call spir_func target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) @_Z27__spirv_JointMatrixMadINTELPU3AS141__spirv_JointMatrixINTEL__half_8_16_0_3_0PU3AS142__spirv_JointMatrixINTEL__half_16_16_2_3_1PU3AS142__spirv_JointMatrixINTEL__float_8_16_3_3_2i(target("spirv.JointMatrixINTEL", half, 8, 16, 0, 3, 0) %{{.*}}, target("spirv.JointMatrixINTEL", half, 16, 16, 2, 3, 1) %{{.*}}, target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) %{{.*}}, i32 3)
+; CHECK-LLVM: call spir_func void @_Z29__spirv_JointMatrixStoreINTELPU3AS1fPU3AS142__spirv_JointMatrixINTEL__float_8_16_3_3_2liii(ptr addrspace(1) %{{.*}}, target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) %{{.*}}, i64 %{{.*}}, i32 0, i32 3, i32 0)
+
+; ModuleID = 'half.bc'
+source_filename = "../joint_matrix_half.cpp"
 target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-n8:16:32:64"
 target triple = "spir64-unknown-unknown"
 
-%"struct._ZTSN2cl4sycl6detail14AssertHappenedE.cl::sycl::detail::AssertHappened" = type { i32, [257 x i8], [257 x i8], [129 x i8], i32, i64, i64, i64, i64, i64, i64 }
-%"class._ZTSN2cl4sycl5rangeILi1EEE.cl::sycl::range" = type { %"class._ZTSN2cl4sycl6detail5arrayILi1EEE.cl::sycl::detail::array" }
-%"class._ZTSN2cl4sycl6detail5arrayILi1EEE.cl::sycl::detail::array" = type { [1 x i64] }
-%"class._ZTSN2cl4sycl2idILi1EEE.cl::sycl::id" = type { %"class._ZTSN2cl4sycl6detail5arrayILi1EEE.cl::sycl::detail::array" }
-%"class.cl::sycl::detail::half_impl::half" = type { half }
-%spirv.JointMatrixINTEL._float_2_2_0_3 = type opaque
-%spirv.JointMatrixINTEL._half_2_16_0_3 = type opaque
-%spirv.JointMatrixINTEL._half_16_2_3_3 = type opaque
-
-$_ZTSN2cl4sycl6detail16AssertInfoCopierE = comdat any
+%"class.sycl::_V1::detail::half_impl::half" = type { half }
 
-$_ZTSZ4mainE11matrix_test = comdat any
+$_ZTSZZ15matrix_multiplyIfN4sycl3_V16detail9half_impl4halfELm16ELm32ELm16ELm64ELm16ELm32EEvR10big_matrixIT_XT5_EXT6_EERS5_IT0_XT1_EXT2_EERS5_IS9_XT3_EXT4_EEENKUlRNS1_7handlerEE_clESF_E7imatrix = comdat any
 
 @__spirv_BuiltInGlobalInvocationId = external dso_local local_unnamed_addr addrspace(1) constant <3 x i64>, align 32
 @__spirv_BuiltInLocalInvocationId = external dso_local local_unnamed_addr addrspace(1) constant <3 x i64>, align 32
 
-; Function Attrs: convergent norecurse
-define weak_odr dso_local spir_kernel void @_ZTSN2cl4sycl6detail16AssertInfoCopierE(%"struct._ZTSN2cl4sycl6detail14AssertHappenedE.cl::sycl::detail::AssertHappened" addrspace(1)* %_arg_, %"class._ZTSN2cl4sycl5rangeILi1EEE.cl::sycl::range"* byval(%"class._ZTSN2cl4sycl5rangeILi1EEE.cl::sycl::range") align 8 %_arg_1, %"class._ZTSN2cl4sycl5rangeILi1EEE.cl::sycl::range"* byval(%"class._ZTSN2cl4sycl5rangeILi1EEE.cl::sycl::range") align 8 %_arg_2, %"class._ZTSN2cl4sycl2idILi1EEE.cl::sycl::id"* byval(%"class._ZTSN2cl4sycl2idILi1EEE.cl::sycl::id") align 8 %_arg_3) local_unnamed_addr #0 comdat !kernel_arg_buffer_location !6 {
+; Function Attrs: convergent norecurse nounwind
+define weak_odr dso_local spir_kernel void @_ZTSZZ15matrix_multiplyIfN4sycl3_V16detail9half_impl4halfELm16ELm32ELm16ELm64ELm16ELm32EEvR10big_matrixIT_XT5_EXT6_EERS5_IT0_XT1_EXT2_EERS5_IS9_XT3_EXT4_EEENKUlRNS1_7handlerEE_clESF_E7imatrix(ptr addrspace(1) noundef align 2 %_arg_accA, ptr addrspace(1) noundef align 2 %_arg_accB, ptr addrspace(1) noundef align 4 %_arg_accC, i64 noundef %_arg_N, i64 noundef %_arg_K) local_unnamed_addr #0 comdat !srcloc !49 !kernel_arg_buffer_location !50 !kernel_arg_runtime_aligned !51 !kernel_arg_exclusive_ptr !51 !intel_reqd_sub_group_size !52 !sycl_used_aspects !53 !sycl_fixed_targets !54 !sycl_kernel_omit_args !55 {
 entry:
-  %0 = getelementptr inbounds %"class._ZTSN2cl4sycl2idILi1EEE.cl::sycl::id", %"class._ZTSN2cl4sycl2idILi1EEE.cl::sycl::id"* %_arg_3, i64 0, i32 0, i32 0, i64 0
-  %1 = addrspacecast i64* %0 to i64 addrspace(4)*
-  %2 = load i64, i64 addrspace(4)* %1, align 8
-  %add.ptr.i = getelementptr inbounds %"struct._ZTSN2cl4sycl6detail14AssertHappenedE.cl::sycl::detail::AssertHappened", %"struct._ZTSN2cl4sycl6detail14AssertHappenedE.cl::sycl::detail::AssertHappened" addrspace(1)* %_arg_, i64 %2
-  %3 = bitcast %"struct._ZTSN2cl4sycl6detail14AssertHappenedE.cl::sycl::detail::AssertHappened" addrspace(1)* %add.ptr.i to i8 addrspace(1)*
-  %4 = addrspacecast i8 addrspace(1)* %3 to i8 addrspace(4)*
-  tail call spir_func void @__devicelib_assert_read(i8 addrspace(4)* %4) #3
-  ret void
-}
-
-; Function Attrs: convergent
-declare extern_weak dso_local spir_func void @__devicelib_assert_read(i8 addrspace(4)*) local_unnamed_addr #1
-
-; Function Attrs: convergent norecurse
-define weak_odr dso_local spir_kernel void @_ZTSZ4mainE11matrix_test(float addrspace(1)* %_arg_, i64 %_arg_1, %"class.cl::sycl::detail::half_impl::half" addrspace(1)* %_arg_3, %"class.cl::sycl::detail::half_impl::half" addrspace(1)* %_arg_5) local_unnamed_addr #0 comdat !kernel_arg_buffer_location !6 !intel_reqd_sub_group_size !7 {
-entry:
-  %0 = load <3 x i64>, <3 x i64> addrspace(4)* addrspacecast (<3 x i64> addrspace(1)* @__spirv_BuiltInGlobalInvocationId to <3 x i64> addrspace(4)*), align 32, !noalias !8
-  %1 = extractelement <3 x i64> %0, i64 1
-  %2 = extractelement <3 x i64> %0, i64 0
-  %3 = load <3 x i64>, <3 x i64> addrspace(4)* addrspacecast (<3 x i64> addrspace(1)* @__spirv_BuiltInLocalInvocationId to <3 x i64> addrspace(4)*), align 32, !noalias !15
-  %4 = extractelement <3 x i64> %3, i64 1
-  %5 = extractelement <3 x i64> %3, i64 0
-  %cmp.i.i = icmp ult i64 %1, 2147483648
+  call void @__itt_offload_wi_start_wrapper()
+  %0 = load i64, ptr addrspace(1) getelementptr inbounds (i8, ptr addrspace(1) @__spirv_BuiltInGlobalInvocationId, i64 8), align 8, !noalias !56
+  %1 = load i64, ptr addrspace(1) @__spirv_BuiltInGlobalInvocationId, align 32, !noalias !56
+  %2 = load i64, ptr addrspace(1) getelementptr inbounds (i8, ptr addrspace(1) @__spirv_BuiltInLocalInvocationId, i64 8), align 8, !noalias !63
+  %3 = load i64, ptr addrspace(1) @__spirv_BuiltInLocalInvocationId, align 32, !noalias !63
+  %cmp.i.i = icmp ult i64 %0, 2147483648
   tail call void @llvm.assume(i1 %cmp.i.i)
-  %cmp.i45.i = icmp ult i64 %2, 2147483648
-  tail call void @llvm.assume(i1 %cmp.i45.i)
-  %cmp.i43.i = icmp ult i64 %4, 2147483648
-  tail call void @llvm.assume(i1 %cmp.i43.i)
-  %sub.i = sub nsw i64 %1, %4
-  %cmp.i41.i = icmp ult i64 %5, 2147483648
-  tail call void @llvm.assume(i1 %cmp.i41.i)
-  %sub5.i = sub nsw i64 %2, %5
-  %mul6.i = shl nsw i64 %sub.i, 6
-  %add.ptr.i51 = getelementptr inbounds float, float addrspace(1)* %_arg_, i64 %mul6.i
-  %add.ptr7.i52 = getelementptr inbounds float, float addrspace(1)* %add.ptr.i51, i64 %sub5.i
-  %add.ptr7.i = addrspacecast float addrspace(1)* %add.ptr7.i52 to float addrspace(4)*
-  %call8.i = tail call spir_func %spirv.JointMatrixINTEL._float_2_2_0_3 addrspace(4)* @_Z28__spirv_JointMatrixLoadINTELIfLm2ELm2ELN5__spv12MatrixLayoutE0ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEEPS5_mS1_S3_i(float addrspace(4)* %add.ptr7.i, i64 %_arg_1, i32 0, i32 3, i32 0) #3
-  %add.ptr11.i53 = getelementptr inbounds %"class.cl::sycl::detail::half_impl::half", %"class.cl::sycl::detail::half_impl::half" addrspace(1)* %_arg_3, i64 %mul6.i
-  %add.ptr16.i55 = getelementptr inbounds %"class.cl::sycl::detail::half_impl::half", %"class.cl::sycl::detail::half_impl::half" addrspace(1)* %_arg_5, i64 %sub5.i
+  %cmp.i61.i = icmp ult i64 %1, 2147483648
+  tail call void @llvm.assume(i1 %cmp.i61.i)
+  %cmp.i63.i = icmp ult i64 %2, 2147483648
+  tail call void @llvm.assume(i1 %cmp.i63.i)
+  %sub.i = sub nsw i64 %0, %2
+  %cmp.i66.i = icmp ult i64 %3, 2147483648
+  tail call void @llvm.assume(i1 %cmp.i66.i)
+  %sub5.i = sub nsw i64 %1, %3
+  %mul.i = shl nsw i64 %sub.i, 3
+  %mul8.i = mul i64 %mul.i, %_arg_N
+  %add.ptr.i.i = getelementptr inbounds float, ptr addrspace(1) %_arg_accC, i64 %mul8.i
+  %div58.i = and i64 %sub5.i, -16
+  %add.ptr.i80.i = getelementptr inbounds float, ptr addrspace(1) %add.ptr.i.i, i64 %div58.i
+  %call1.i.i = tail call spir_func noundef target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) @_Z28__spirv_JointMatrixLoadINTELIU3AS1ffLm8ELm16ELN5__spv9MatrixUseE2ELNS1_12MatrixLayoutE3ELNS1_5Scope4FlagE3EEPNS1_24__spirv_JointMatrixINTELIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEPT_mS3_S5_i(ptr addrspace(1) noundef %add.ptr.i80.i, i64 noundef %_arg_N, i32 noundef 0, i32 noundef 3, i32 noundef 0) #3
+  %mul34.i = shl nsw i64 %div58.i, 1
+  %div1159.i = lshr i64 %_arg_K, 4
+  %mul18.i = mul i64 %mul.i, %_arg_K
+  %add.ptr.i95.i = getelementptr inbounds %"class.sycl::_V1::detail::half_impl::half", ptr addrspace(1) %_arg_accA, i64 %mul18.i
+  %mul30.i = shl i64 %_arg_N, 1
+  %invariant.gep = getelementptr %"class.sycl::_V1::detail::half_impl::half", ptr addrspace(1) %_arg_accB, i64 %mul34.i
   br label %for.cond.i
 
 for.cond.i:                                       ; preds = %for.body.i, %entry
+  %sub_c.sroa.0.0.i = phi target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) [ %call1.i.i, %entry ], [ %call.i.i, %for.body.i ]
   %k.0.i = phi i32 [ 0, %entry ], [ %add.i, %for.body.i ]
-  %C.0.i = phi %spirv.JointMatrixINTEL._float_2_2_0_3 addrspace(4)* [ %call8.i, %entry ], [ %call19.i, %for.body.i ]
-  %cmp.i = icmp ult i32 %k.0.i, 32
-  br i1 %cmp.i, label %for.body.i, label %_ZZ4mainENKUlN2cl4sycl7nd_itemILi2EEEE_clES2_.exit
+  %conv.i = zext i32 %k.0.i to i64
+  %cmp.i = icmp ugt i64 %div1159.i, %conv.i
+  br i1 %cmp.i, label %for.body.i, label %_ZZZ15matrix_multiplyIfN4sycl3_V16detail9half_impl4halfELm16ELm32ELm16ELm64ELm16ELm32EEvR10big_matrixIT_XT5_EXT6_EERS5_IT0_XT1_EXT2_EERS5_IS9_XT3_EXT4_EEENKUlRNS1_7handlerEE_clESF_ENKUlNS1_7nd_itemILi2EEEE_clESI_.exit
 
 for.body.i:                                       ; preds = %for.cond.i
-  %idx.ext46.i = zext i32 %k.0.i to i64
-  %add.ptr12.i54 = getelementptr inbounds %"class.cl::sycl::detail::half_impl::half", %"class.cl::sycl::detail::half_impl::half" addrspace(1)* %add.ptr11.i53, i64 %idx.ext46.i
-  %add.ptr12.i = addrspacecast %"class.cl::sycl::detail::half_impl::half" addrspace(1)* %add.ptr12.i54 to %"class.cl::sycl::detail::half_impl::half" addrspace(4)*
-  %call13.i = tail call spir_func %spirv.JointMatrixINTEL._half_2_16_0_3 addrspace(4)* @_Z28__spirv_JointMatrixLoadINTELIN2cl4sycl6detail9half_impl4halfELm2ELm16ELN5__spv12MatrixLayoutE0ELNS5_5Scope4FlagE3EEPNS5_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEEPSA_mS6_S8_i(%"class.cl::sycl::detail::half_impl::half" addrspace(4)* %add.ptr12.i, i64 %_arg_1, i32 0, i32 3, i32 0) #3
-  %mul14.i = shl nuw nsw i32 %k.0.i, 5
-  %idx.ext1547.i = zext i32 %mul14.i to i64
-  %add.ptr17.i56 = getelementptr inbounds %"class.cl::sycl::detail::half_impl::half", %"class.cl::sycl::detail::half_impl::half" addrspace(1)* %add.ptr16.i55, i64 %idx.ext1547.i
-  %add.ptr17.i = addrspacecast %"class.cl::sycl::detail::half_impl::half" addrspace(1)* %add.ptr17.i56 to %"class.cl::sycl::detail::half_impl::half" addrspace(4)*
-  %call18.i = tail call spir_func %spirv.JointMatrixINTEL._half_16_2_3_3 addrspace(4)* @_Z28__spirv_JointMatrixLoadINTELIN2cl4sycl6detail9half_impl4halfELm16ELm2ELN5__spv12MatrixLayoutE3ELNS5_5Scope4FlagE3EEPNS5_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEEPSA_mS6_S8_i(%"class.cl::sycl::detail::half_impl::half" addrspace(4)* %add.ptr17.i, i64 %_arg_1, i32 0, i32 3, i32 0) #3
-  %call19.i = tail call spir_func %spirv.JointMatrixINTEL._float_2_2_0_3 addrspace(4)* @_Z27__spirv_JointMatrixMadINTELIN2cl4sycl6detail9half_impl4halfEfLm2ELm16ELm2ELN5__spv12MatrixLayoutE0ELS6_3ELS6_0ELNS5_5Scope4FlagE3EEPNS5_24__spirv_JointMatrixINTELIT0_XT1_EXT3_EXT6_EXT7_EEEPNS9_IT_XT1_EXT2_EXT4_EXT7_EEEPNS9_ISD_XT2_EXT3_EXT5_EXT7_EEESC_S8_(%spirv.JointMatrixINTEL._half_2_16_0_3 addrspace(4)* %call13.i, %spirv.JointMatrixINTEL._half_16_2_3_3 addrspace(4)* %call18.i, %spirv.JointMatrixINTEL._float_2_2_0_3 addrspace(4)* %C.0.i, i32 3) #3
-  %add.i = add nuw nsw i32 %k.0.i, 16
-  br label %for.cond.i, !llvm.loop !20
-
-_ZZ4mainENKUlN2cl4sycl7nd_itemILi2EEEE_clES2_.exit: ; preds = %for.cond.i
-  tail call spir_func void @_Z29__spirv_JointMatrixStoreINTELIfLm2ELm2ELN5__spv12MatrixLayoutE0ELNS0_5Scope4FlagE3EEvPT_PNS0_24__spirv_JointMatrixINTELIS4_XT0_EXT1_EXT2_EXT3_EEEmS1_S3_i(float addrspace(4)* %add.ptr7.i, %spirv.JointMatrixINTEL._float_2_2_0_3 addrspace(4)* %C.0.i, i64 %_arg_1, i32 0, i32 3, i32 0) #3
+  %mul19.i = shl nsw i32 %k.0.i, 4
+  %conv20.i = zext i32 %mul19.i to i64
+  %add.ptr.i96.i = getelementptr inbounds %"class.sycl::_V1::detail::half_impl::half", ptr addrspace(1) %add.ptr.i95.i, i64 %conv20.i
+  %call1.i74.i = tail call spir_func noundef target("spirv.JointMatrixINTEL", half, 8, 16, 0, 3, 0) @_Z28__spirv_JointMatrixLoadINTELIU3AS1N4sycl3_V16detail9half_impl4halfES4_Lm8ELm16ELN5__spv9MatrixUseE0ELNS6_12MatrixLayoutE0ELNS6_5Scope4FlagE3EEPNS6_24__spirv_JointMatrixINTELIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEPT_mS8_SA_i(ptr addrspace(1) noundef %add.ptr.i96.i, i64 noundef %_arg_K, i32 noundef 0, i32 noundef 3, i32 noundef 0) #3
+  %div27.i = shl nsw i32 %k.0.i, 3
+  %conv28.i = zext i32 %div27.i to i64
+  %mul31.i = mul i64 %mul30.i, %conv28.i
+  %gep = getelementptr %"class.sycl::_V1::detail::half_impl::half", ptr addrspace(1) %invariant.gep, i64 %mul31.i
+  %call1.i78.i = tail call spir_func noundef target("spirv.JointMatrixINTEL", half, 16, 16, 2, 3, 1) @_Z28__spirv_JointMatrixLoadINTELIU3AS1N4sycl3_V16detail9half_impl4halfES4_Lm16ELm16ELN5__spv9MatrixUseE1ELNS6_12MatrixLayoutE2ELNS6_5Scope4FlagE3EEPNS6_24__spirv_JointMatrixINTELIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEPT_mS8_SA_i(ptr addrspace(1) noundef %gep, i64 noundef %mul30.i, i32 noundef 2, i32 noundef 3, i32 noundef 0) #3
+  %call.i.i = tail call spir_func noundef target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) @_Z27__spirv_JointMatrixMadINTELIN4sycl3_V16detail9half_impl4halfEfLm8ELm16ELm16ELN5__spv9MatrixUseE0ELS6_1ELS6_2ELNS5_12MatrixLayoutE0ELS7_2ELS7_3ELNS5_5Scope4FlagE3EEPNS5_24__spirv_JointMatrixINTELIT0_XT1_EXT3_EXT9_EXT10_EXT6_EEEPNSA_IT_XT1_EXT2_EXT7_EXT10_EXT4_EEEPNSA_ISE_XT2_EXT3_EXT8_EXT10_EXT5_EEESD_S9_(target("spirv.JointMatrixINTEL", half, 8, 16, 0, 3, 0) noundef %call1.i74.i, target("spirv.JointMatrixINTEL", half, 16, 16, 2, 3, 1) noundef %call1.i78.i, target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) noundef %sub_c.sroa.0.0.i, i32 noundef 3) #3, !noalias !68
+  %add.i = add nuw nsw i32 %k.0.i, 1
+  br label %for.cond.i, !llvm.loop !71
+
+_ZZZ15matrix_multiplyIfN4sycl3_V16detail9half_impl4halfELm16ELm32ELm16ELm64ELm16ELm32EEvR10big_matrixIT_XT5_EXT6_EERS5_IT0_XT1_EXT2_EERS5_IS9_XT3_EXT4_EEENKUlRNS1_7handlerEE_clESF_ENKUlNS1_7nd_itemILi2EEEE_clESI_.exit: ; preds = %for.cond.i
+  tail call spir_func void @_Z29__spirv_JointMatrixStoreINTELIU3AS1ffLm8ELm16ELN5__spv9MatrixUseE2ELNS1_12MatrixLayoutE3ELNS1_5Scope4FlagE3EEvPT_PNS1_24__spirv_JointMatrixINTELIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEmS3_S5_i(ptr addrspace(1) noundef %add.ptr.i80.i, target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) noundef %sub_c.sroa.0.0.i, i64 noundef %_arg_N, i32 noundef 0, i32 noundef 3, i32 noundef 0) #3
+  call void @__itt_offload_wi_finish_wrapper()
   ret void
 }
 
-; Function Attrs: convergent
-declare dso_local spir_func %spirv.JointMatrixINTEL._float_2_2_0_3 addrspace(4)* @_Z28__spirv_JointMatrixLoadINTELIfLm2ELm2ELN5__spv12MatrixLayoutE0ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEEPS5_mS1_S3_i(float addrspace(4)*, i64, i32, i32, i32) local_unnamed_addr #1
+; Function Attrs: nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: readwrite)
+declare void @llvm.assume(i1 noundef) #1
+
+; Function Attrs: convergent nounwind
+declare dso_local spir_func noundef target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) @_Z28__spirv_JointMatrixLoadINTELIU3AS1ffLm8ELm16ELN5__spv9MatrixUseE2ELNS1_12MatrixLayoutE3ELNS1_5Scope4FlagE3EEPNS1_24__spirv_JointMatrixINTELIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEPT_mS3_S5_i(ptr addrspace(1) noundef, i64 noundef, i32 noundef, i32 noundef, i32 noundef) local_unnamed_addr #2
+
+; Function Attrs: convergent nounwind
+declare dso_local spir_func noundef target("spirv.JointMatrixINTEL", half, 8, 16, 0, 3, 0) @_Z28__spirv_JointMatrixLoadINTELIU3AS1N4sycl3_V16detail9half_impl4halfES4_Lm8ELm16ELN5__spv9MatrixUseE0ELNS6_12MatrixLayoutE0ELNS6_5Scope4FlagE3EEPNS6_24__spirv_JointMatrixINTELIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEPT_mS8_SA_i(ptr addrspace(1) noundef, i64 noundef, i32 noundef, i32 noundef, i32 noundef) local_unnamed_addr #2
+
+; Function Attrs: convergent nounwind
+declare dso_local spir_func noundef target("spirv.JointMatrixINTEL", half, 16, 16, 2, 3, 1) @_Z28__spirv_JointMatrixLoadINTELIU3AS1N4sycl3_V16detail9half_impl4halfES4_Lm16ELm16ELN5__spv9MatrixUseE1ELNS6_12MatrixLayoutE2ELNS6_5Scope4FlagE3EEPNS6_24__spirv_JointMatrixINTELIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEPT_mS8_SA_i(ptr addrspace(1) noundef, i64 noundef, i32 noundef, i32 noundef, i32 noundef) local_unnamed_addr #2
 
-; Function Attrs: convergent
-declare dso_local spir_func %spirv.JointMatrixINTEL._half_2_16_0_3 addrspace(4)* @_Z28__spirv_JointMatrixLoadINTELIN2cl4sycl6detail9half_impl4halfELm2ELm16ELN5__spv12MatrixLayoutE0ELNS5_5Scope4FlagE3EEPNS5_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEEPSA_mS6_S8_i(%"class.cl::sycl::detail::half_impl::half" addrspace(4)*, i64, i32, i32, i32) local_unnamed_addr #1
+; Function Attrs: convergent nounwind
+declare dso_local spir_func noundef target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) @_Z27__spirv_JointMatrixMadINTELIN4sycl3_V16detail9half_impl4halfEfLm8ELm16ELm16ELN5__spv9MatrixUseE0ELS6_1ELS6_2ELNS5_12MatrixLayoutE0ELS7_2ELS7_3ELNS5_5Scope4FlagE3EEPNS5_24__spirv_JointMatrixINTELIT0_XT1_EXT3_EXT9_EXT10_EXT6_EEEPNSA_IT_XT1_EXT2_EXT7_EXT10_EXT4_EEEPNSA_ISE_XT2_EXT3_EXT8_EXT10_EXT5_EEESD_S9_(target("spirv.JointMatrixINTEL", half, 8, 16, 0, 3, 0) noundef, target("spirv.JointMatrixINTEL", half, 16, 16, 2, 3, 1) noundef, target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) noundef, i32 noundef) local_unnamed_addr #2
 
-; Function Attrs: convergent
-declare dso_local spir_func %spirv.JointMatrixINTEL._half_16_2_3_3 addrspace(4)* @_Z28__spirv_JointMatrixLoadINTELIN2cl4sycl6detail9half_impl4halfELm16ELm2ELN5__spv12MatrixLayoutE3ELNS5_5Scope4FlagE3EEPNS5_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT2_EXT3_EEEPSA_mS6_S8_i(%"class.cl::sycl::detail::half_impl::half" addrspace(4)*, i64, i32, i32, i32) local_unnamed_addr #1
+; Function Attrs: convergent nounwind
+declare dso_local spir_func void @_Z29__spirv_JointMatrixStoreINTELIU3AS1ffLm8ELm16ELN5__spv9MatrixUseE2ELNS1_12MatrixLayoutE3ELNS1_5Scope4FlagE3EEvPT_PNS1_24__spirv_JointMatrixINTELIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEmS3_S5_i(ptr addrspace(1) noundef, target("spirv.JointMatrixINTEL", float, 8, 16, 3, 3, 2) noundef, i64 noundef, i32 noundef, i32 noundef, i32 noundef) local_unnamed_addr #2
 
-; Function Attrs: convergent
-declare dso_local spir_func %spirv.JointMatrixINTEL._float_2_2_0_3 addrspace(4)* @_Z27__spirv_JointMatrixMadINTELIN2cl4sycl6detail9half_impl4halfEfLm2ELm16ELm2ELN5__spv12MatrixLayoutE0ELS6_3ELS6_0ELNS5_5Scope4FlagE3EEPNS5_24__spirv_JointMatrixINTELIT0_XT1_EXT3_EXT6_EXT7_EEEPNS9_IT_XT1_EXT2_EXT4_EXT7_EEEPNS9_ISD_XT2_EXT3_EXT5_EXT7_EEESC_S8_(%spirv.JointMatrixINTEL._half_2_16_0_3 addrspace(4)*, %spirv.JointMatrixINTEL._half_16_2_3_3 addrspace(4)*, %spirv.JointMatrixINTEL._float_2_2_0_3 addrspace(4)*, i32) local_unnamed_addr #1
+declare dso_local spir_func i32 @_Z18__spirv_ocl_printfPU3AS2Kcz(ptr addrspace(2), ...)
 
-; Function Attrs: convergent
-declare dso_local spir_func void @_Z29__spirv_JointMatrixStoreINTELIfLm2ELm2ELN5__spv12MatrixLayoutE0ELNS0_5Scope4FlagE3EEvPT_PNS0_24__spirv_JointMatrixINTELIS4_XT0_EXT1_EXT2_EXT3_EEEmS1_S3_i(float addrspace(4)*, %spirv.JointMatrixINTEL._float_2_2_0_3 addrspace(4)*, i64, i32, i32, i32) local_unnamed_addr #1
+declare void @__itt_offload_wi_start_wrapper()
 
-; Function Attrs: inaccessiblememonly nofree nosync nounwind willreturn
-declare void @llvm.assume(i1 noundef) #2
+declare void @__itt_offload_wi_finish_wrapper()
 
-attributes #0 = { convergent norecurse "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "sycl-module-id"="/work/intel/build/joint_matrix_test.cpp" "uniform-work-group-size"="true" }
-attributes #1 = { convergent "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
-attributes #2 = { inaccessiblememonly nofree nosync nounwind willreturn }
-attributes #3 = { convergent }
+attributes #0 = { convergent norecurse nounwind "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "sycl-module-id"="../joint_matrix_half.cpp" "sycl-optlevel"="2" "uniform-work-group-size"="true" }
+attributes #1 = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: readwrite) }
+attributes #2 = { convergent nounwind "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #3 = { convergent nounwind }
 
 !llvm.module.flags = !{!0, !1}
 !opencl.spir.version = !{!2}
 !spirv.Source = !{!3}
-!opencl.used.extensions = !{!4}
-!opencl.used.optional.core.features = !{!4}
-!opencl.compiler.options = !{!4}
-!llvm.ident = !{!5}
+!sycl_types_that_use_aspects = !{!4}
+!sycl_aspects = !{!5, !6, !7, !8, !9, !10, !11, !12, !13, !14, !15, !16, !17, !18, !19, !20, !21, !22, !23, !24, !25, !26, !27, !28, !29, !30, !31, !32, !33, !34, !35, !36, !37, !38, !39, !40, !41, !42, !43, !44, !45, !46, !47}
+!llvm.ident = !{!48}
 
 !0 = !{i32 1, !"wchar_size", i32 4}
 !1 = !{i32 7, !"frame-pointer", i32 2}
 !2 = !{i32 1, i32 2}
 !3 = !{i32 4, i32 100000}
-!4 = !{}
-!5 = !{!"Clang"}
-!6 = !{i32 -1, i32 -1, i32 -1, i32 -1}
-!7 = !{i32 16}
-!8 = !{!9, !11, !13}
-!9 = distinct !{!9, !10, !"_ZN7__spirv29InitSizesSTGlobalInvocationIdILi2EN2cl4sycl2idILi2EEEE8initSizeEv: %agg.result"}
-!10 = distinct !{!10, !"_ZN7__spirv29InitSizesSTGlobalInvocationIdILi2EN2cl4sycl2idILi2EEEE8initSizeEv"}
-!11 = distinct !{!11, !12, !"_ZN7__spirvL22initGlobalInvocationIdILi2EN2cl4sycl2idILi2EEEEET0_v: %agg.result"}
-!12 = distinct !{!12, !"_ZN7__spirvL22initGlobalInvocationIdILi2EN2cl4sycl2idILi2EEEEET0_v"}
-!13 = distinct !{!13, !14, !"_ZN2cl4sycl6detail7Builder10getElementILi2EEEKNS0_7nd_itemIXT_EEEPS5_: %agg.result"}
-!14 = distinct !{!14, !"_ZN2cl4sycl6detail7Builder10getElementILi2EEEKNS0_7nd_itemIXT_EEEPS5_"}
-!15 = !{!16, !18, !13}
-!16 = distinct !{!16, !17, !"_ZN7__spirv28InitSizesSTLocalInvocationIdILi2EN2cl4sycl2idILi2EEEE8initSizeEv: %agg.result"}
-!17 = distinct !{!17, !"_ZN7__spirv28InitSizesSTLocalInvocationIdILi2EN2cl4sycl2idILi2EEEE8initSizeEv"}
-!18 = distinct !{!18, !19, !"_ZN7__spirvL21initLocalInvocationIdILi2EN2cl4sycl2idILi2EEEEET0_v: %agg.result"}
-!19 = distinct !{!19, !"_ZN7__spirvL21initLocalInvocationIdILi2EN2cl4sycl2idILi2EEEEET0_v"}
-!20 = distinct !{!20, !21, !22}
-!21 = !{!"llvm.loop.mustprogress"}
-!22 = !{!"llvm.loop.unroll.disable"}
+!4 = !{!"class.sycl::_V1::detail::half_impl::half", i32 5}
+!5 = !{!"cpu", i32 1}
+!6 = !{!"gpu", i32 2}
+!7 = !{!"accelerator", i32 3}
+!8 = !{!"custom", i32 4}
+!9 = !{!"fp16", i32 5}
+!10 = !{!"fp64", i32 6}
+!11 = !{!"image", i32 9}
+!12 = !{!"online_compiler", i32 10}
+!13 = !{!"online_linker", i32 11}
+!14 = !{!"queue_profiling", i32 12}
+!15 = !{!"usm_device_allocations", i32 13}
+!16 = !{!"usm_host_allocations", i32 14}
+!17 = !{!"usm_shared_allocations", i32 15}
+!18 = !{!"usm_system_allocations", i32 17}
+!19 = !{!"ext_intel_pci_address", i32 18}
+!20 = !{!"ext_intel_gpu_eu_count", i32 19}
+!21 = !{!"ext_intel_gpu_eu_simd_width", i32 20}
+!22 = !{!"ext_intel_gpu_slices", i32 21}
+!23 = !{!"ext_intel_gpu_subslices_per_slice", i32 22}
+!24 = !{!"ext_intel_gpu_eu_count_per_subslice", i32 23}
+!25 = !{!"ext_intel_max_mem_bandwidth", i32 24}
+!26 = !{!"ext_intel_mem_channel", i32 25}
+!27 = !{!"usm_atomic_host_allocations", i32 26}
+!28 = !{!"usm_atomic_shared_allocations", i32 27}
+!29 = !{!"atomic64", i32 28}
+!30 = !{!"ext_intel_device_info_uuid", i32 29}
+!31 = !{!"ext_oneapi_srgb", i32 30}
+!32 = !{!"ext_oneapi_native_assert", i32 31}
+!33 = !{!"host_debuggable", i32 32}
+!34 = !{!"ext_intel_gpu_hw_threads_per_eu", i32 33}
+!35 = !{!"ext_oneapi_cuda_async_barrier", i32 34}
+!36 = !{!"ext_oneapi_bfloat16_math_functions", i32 35}
+!37 = !{!"ext_intel_free_memory", i32 36}
+!38 = !{!"ext_intel_device_id", i32 37}
+!39 = !{!"ext_intel_memory_clock_rate", i32 38}
+!40 = !{!"ext_intel_memory_bus_width", i32 39}
+!41 = !{!"emulated", i32 40}
+!42 = !{!"ext_intel_legacy_image", i32 41}
+!43 = !{!"int64_base_atomics", i32 7}
+!44 = !{!"int64_extended_atomics", i32 8}
+!45 = !{!"usm_system_allocator", i32 17}
+!46 = !{!"usm_restricted_shared_allocations", i32 16}
+!47 = !{!"host", i32 0}
+!48 = !{!"clang version 17.0.0 (https://github.com/intel/llvm.git 93f477358d74ae90024f758e7eeb97d4b13cea42)"}
+!49 = !{i32 10643216}
+!50 = !{i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1}
+!51 = !{i1 true, i1 false, i1 false, i1 false, i1 true, i1 false, i1 false, i1 false, i1 true, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false}
+!52 = !{i32 16}
+!53 = !{i32 5}
+!54 = !{}
+!55 = !{i1 false, i1 true, i1 true, i1 true, i1 false, i1 true, i1 true, i1 true, i1 false, i1 true, i1 true, i1 true, i1 true, i1 false, i1 false}
+!56 = !{!57, !59, !61}
+!57 = distinct !{!57, !58, !"_ZN7__spirv29InitSizesSTGlobalInvocationIdILi2EN4sycl3_V12idILi2EEEE8initSizeEv: %agg.result"}
+!58 = distinct !{!58, !"_ZN7__spirv29InitSizesSTGlobalInvocationIdILi2EN4sycl3_V12idILi2EEEE8initSizeEv"}
+!59 = distinct !{!59, !60, !"_ZN7__spirvL22initGlobalInvocationIdILi2EN4sycl3_V12idILi2EEEEET0_v: %agg.result"}
+!60 = distinct !{!60, !"_ZN7__spirvL22initGlobalInvocationIdILi2EN4sycl3_V12idILi2EEEEET0_v"}
+!61 = distinct !{!61, !62, !"_ZN4sycl3_V16detail7Builder10getElementILi2EEEKNS0_7nd_itemIXT_EEEPS5_: %agg.result"}
+!62 = distinct !{!62, !"_ZN4sycl3_V16detail7Builder10getElementILi2EEEKNS0_7nd_itemIXT_EEEPS5_"}
+!63 = !{!64, !66, !61}
+!64 = distinct !{!64, !65, !"_ZN7__spirv28InitSizesSTLocalInvocationIdILi2EN4sycl3_V12idILi2EEEE8initSizeEv: %agg.result"}
+!65 = distinct !{!65, !"_ZN7__spirv28InitSizesSTLocalInvocationIdILi2EN4sycl3_V12idILi2EEEE8initSizeEv"}
+!66 = distinct !{!66, !67, !"_ZN7__spirvL21initLocalInvocationIdILi2EN4sycl3_V12idILi2EEEEET0_v: %agg.result"}
+!67 = distinct !{!67, !"_ZN7__spirvL21initLocalInvocationIdILi2EN4sycl3_V12idILi2EEEEET0_v"}
+!68 = !{!69}
+!69 = distinct !{!69, !70, !"_ZN4sycl3_V13ext6oneapi12experimental6matrix16joint_matrix_madINS0_9sub_groupENS0_6detail9half_impl4halfES9_fLm8ELm16ELm16ELNS4_6layoutE0ELSA_2EEENS4_12joint_matrixIT_T2_LNS4_3useE2EXT3_EXT5_ELSA_3EEESC_RNSB_ISC_T0_LSE_0EXT3_EXT4_EXT6_EEERNSB_ISC_T1_LSE_1EXT4_EXT5_EXT7_EEERSF_: %agg.result"}
+!70 = distinct !{!70, !"_ZN4sycl3_V13ext6oneapi12experimental6matrix16joint_matrix_madINS0_9sub_groupENS0_6detail9half_impl4halfES9_fLm8ELm16ELm16ELNS4_6layoutE0ELSA_2EEENS4_12joint_matrixIT_T2_LNS4_3useE2EXT3_EXT5_ELSA_3EEESC_RNSB_ISC_T0_LSE_0EXT3_EXT4_EXT6_EEERNSB_ISC_T1_LSE_1EXT4_EXT5_EXT7_EEERSF_"}
+!71 = distinct !{!71, !72}
+!72 = !{!"llvm.loop.mustprogress"}
diff --git a/test/extensions/INTEL/SPV_INTEL_joint_matrix/opaque_joint_matrix.ll b/test/extensions/INTEL/SPV_INTEL_joint_matrix/opaque_joint_matrix.ll
deleted file mode 100644
index 5b59c2d8f1..0000000000
--- a/test/extensions/INTEL/SPV_INTEL_joint_matrix/opaque_joint_matrix.ll
+++ /dev/null
@@ -1,151 +0,0 @@
-; RUN: llvm-as  < %s -o %t.bc
-; RUN: llvm-spirv %t.bc --spirv-ext=+SPV_INTEL_joint_matrix -o %t.spv
-; RUN: llvm-spirv %t.spv -to-text -o %t.spt
-; RUN: FileCheck < %t.spt %s --check-prefix=CHECK-SPIRV
-
-; RUN: llvm-spirv -r %t.spv -o %t.rev.bc -opaque-pointers=0
-; RUN: llvm-dis -opaque-pointers=0 < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM
-
-; CHECK-SPIRV-DAG: Capability JointMatrixINTEL
-; CHECK-SPIRV-DAG: Extension "SPV_INTEL_joint_matrix"
-; CHECK-SPIRV-DAG: TypeInt [[#Int8Ty:]] 8 0
-; CHECK-SPIRV-DAG: TypeInt [[#Int32Ty:]] 32 0
-; CHECK-SPIRV-DAG: Constant [[#Int32Ty]] [[#Const12:]] 12
-; CHECK-SPIRV-DAG: Constant [[#Int32Ty]] [[#Const3:]] 3
-; CHECK-SPIRV-DAG: Constant [[#Int32Ty]] [[#Const2:]] 2
-; CHECK-SPIRV-DAG: Constant [[#Int32Ty]] [[#Const0:]] 0
-; CHECK-SPIRV-DAG: Constant [[#Int32Ty]] [[#Const48:]] 48
-; CHECK-SPIRV-DAG: Constant [[#Int32Ty]] [[#Const1:]] 1
-; CHECK-SPIRV-DAG: TypeJointMatrixINTEL [[#MatTy1:]] [[#Int32Ty]] [[#Const12]] [[#Const12]] [[#Const3]] [[#Const3]] [[#Const2]]
-; CHECK-SPIRV-DAG: TypeJointMatrixINTEL [[#MatTy2:]] [[#Int8Ty]] [[#Const12]] [[#Const48]] [[#Const0]] [[#Const3]] [[#Const0]]
-; CHECK-SPIRV-DAG: TypeJointMatrixINTEL [[#MatTy3:]] [[#Int8Ty]] [[#Const48]] [[#Const12]] [[#Const2]] [[#Const3]] [[#Const1]]
-
-; CHECK-LLVM-DAG: %spirv.JointMatrixINTEL._int_12_12_3_3_2 = type opaque
-; CHECK-LLVM-DAG: %spirv.JointMatrixINTEL._char_12_48_0_3_0 = type opaque
-; CHECK-LLVM-DAG: %spirv.JointMatrixINTEL._char_48_12_2_3_1 = type opaque
-
-; ModuleID = 'test-matrix-opaque.bc'
-source_filename = "matrix-int8-test.cpp"
-target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-n8:16:32:64"
-target triple = "spir64-unknown-unknown"
-
-%"class.sycl::_V1::range" = type { %"class.sycl::_V1::detail::array" }
-%"class.sycl::_V1::detail::array" = type { [2 x i64] }
-%"class.sycl::_V1::id" = type { %"class.sycl::_V1::detail::array" }
-
-$_ZTSZZ15matrix_multiplyIiaLm24ELm96ELm24ELm96ELm24ELm24EEvR10big_matrixIT_XT5_EXT6_EERS0_IT0_XT1_EXT2_EERS0_IS4_XT3_EXT4_EEENKUlRN4sycl3_V17handlerEE_clESC_E7imatrix = comdat any
-
-@__spirv_BuiltInGlobalInvocationId = external dso_local local_unnamed_addr addrspace(1) constant <3 x i64>, align 32
-@__spirv_BuiltInLocalInvocationId = external dso_local local_unnamed_addr addrspace(1) constant <3 x i64>, align 32
-
-; Function Attrs: convergent norecurse
-define weak_odr dso_local spir_kernel void @_ZTSZZ15matrix_multiplyIiaLm24ELm96ELm24ELm96ELm24ELm24EEvR10big_matrixIT_XT5_EXT6_EERS0_IT0_XT1_EXT2_EERS0_IS4_XT3_EXT4_EEENKUlRN4sycl3_V17handlerEE_clESC_E7imatrix(ptr addrspace(1) noundef align 1 %_arg_accA, ptr addrspace(1) noundef align 1 %_arg_accB, ptr noundef byval(%"class.sycl::_V1::range") align 8 %_arg_accB5, ptr noundef byval(%"class.sycl::_V1::id") align 8 %_arg_accB6, ptr addrspace(1) noundef align 4 %_arg_accC, i64 noundef %_arg_N, i64 noundef %_arg_K) local_unnamed_addr #0 comdat {
-entry:
-  %sub_c.sroa.0.i = alloca target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2), align 8
-  %ref.tmp29.sroa.0.i = alloca target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2), align 8
-  %agg.tmp15.sroa.0.sroa.2.0..sroa_idx = getelementptr inbounds %"class.sycl::_V1::range", ptr %_arg_accB5, i64 0, i32 0, i32 0, i64 1
-  %agg.tmp15.sroa.0.sroa.2.0.copyload = load i64, ptr %agg.tmp15.sroa.0.sroa.2.0..sroa_idx, align 8
-  %0 = getelementptr inbounds %"class.sycl::_V1::id", ptr %_arg_accB6, i64 0, i32 0, i32 0, i64 0
-  %agg.tmp16.sroa.0.sroa.0.0.copyload = load i64, ptr %0, align 8
-  %agg.tmp16.sroa.0.sroa.2.0..sroa_idx = getelementptr inbounds %"class.sycl::_V1::id", ptr %_arg_accB6, i64 0, i32 0, i32 0, i64 1
-  %agg.tmp16.sroa.0.sroa.2.0.copyload = load i64, ptr %agg.tmp16.sroa.0.sroa.2.0..sroa_idx, align 8
-  %mul.i4.i.i.i.i45 = mul i64 %agg.tmp16.sroa.0.sroa.0.0.copyload, %agg.tmp15.sroa.0.sroa.2.0.copyload
-  %add.i6.i.i.i.i46 = add i64 %mul.i4.i.i.i.i45, %agg.tmp16.sroa.0.sroa.2.0.copyload
-  %add.ptr.i47 = getelementptr inbounds i8, ptr addrspace(1) %_arg_accB, i64 %add.i6.i.i.i.i46
-  %1 = load <3 x i64>, ptr addrspace(1) @__spirv_BuiltInGlobalInvocationId, align 32
-  %2 = extractelement <3 x i64> %1, i64 1
-  %3 = extractelement <3 x i64> %1, i64 0
-  %4 = load <3 x i64>, ptr addrspace(1) @__spirv_BuiltInLocalInvocationId, align 32
-  %5 = extractelement <3 x i64> %4, i64 1
-  %6 = extractelement <3 x i64> %4, i64 0
-  %cmp.i.i = icmp ult i64 %2, 2147483648
-  %cmp.i54.i = icmp ult i64 %3, 2147483648
-  %cmp.i56.i = icmp ult i64 %5, 2147483648
-  %sub.i = sub nsw i64 %2, %5
-  %cmp.i58.i = icmp ult i64 %6, 2147483648
-  %sub5.i = sub nsw i64 %3, %6
-  %sub_c.sroa.0.i.0.i.0..sroa_cast = bitcast ptr %sub_c.sroa.0.i to ptr
-  call void @llvm.lifetime.start.p0(i64 8, ptr nonnull %sub_c.sroa.0.i.0.i.0..sroa_cast)
-  %call.i.i = tail call spir_func noundef target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2) @_Z26__spirv_CompositeConstructIiLm12ELm12ELN5__spv9MatrixUseE2ELNS0_12MatrixLayoutE3ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT3_EXT4_EXT2_EEES6_(i32 noundef 0) #4
-  store target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2) %call.i.i, ptr %sub_c.sroa.0.i, align 8
-  %mul.i = mul nsw i64 %sub.i, 12
-  %div2452.i = lshr i64 %sub5.i, 4
-  %mul26.i = mul i64 %div2452.i, 48
-  %div.i = udiv i64 %_arg_K, 48
-  %mul11.i = mul i64 %mul.i, %_arg_K
-  %add.ptr.i93.i = getelementptr inbounds i8, ptr addrspace(1) %_arg_accA, i64 %mul11.i
-  %idx.neg.i.i104.i = sub i64 0, %add.i6.i.i.i.i46
-  %add.ptr.i.i105141.i = getelementptr i8, ptr addrspace(1) %add.ptr.i47, i64 %mul26.i
-  %mul22.i = shl i64 %_arg_N, 2
-  %add.ptr.i108140.i = getelementptr i8, ptr addrspace(1) %add.ptr.i.i105141.i, i64 %idx.neg.i.i104.i
-  %ref.tmp29.sroa.0.i.0.i.0..sroa_cast = bitcast ptr %ref.tmp29.sroa.0.i to ptr
-  %7 = bitcast ptr %ref.tmp29.sroa.0.i to ptr
-  %8 = bitcast ptr %sub_c.sroa.0.i to ptr
-  br label %for.cond.i
-
-for.cond.i:                                       ; preds = %for.body.i, %entry
-  %k.0.i = phi i32 [ 0, %entry ], [ %add.i, %for.body.i ]
-  %conv.i = zext i32 %k.0.i to i64
-  %cmp.i = icmp ugt i64 %div.i, %conv.i
-  br i1 %cmp.i, label %for.body.i, label %_ZZZ15matrix_multiplyIiaLm24ELm96ELm24ELm96ELm24ELm24EEvR10big_matrixIT_XT5_EXT6_EERS0_IT0_XT1_EXT2_EERS0_IS4_XT3_EXT4_EEENKUlRN4sycl3_V17handlerEE_clESC_ENKUlNSA_7nd_itemILi2EEEE_clESF_.exit
-
-for.body.i:                                       ; preds = %for.cond.i
-  %mul12.i = mul nsw i32 %k.0.i, 48
-  %conv13.i = zext i32 %mul12.i to i64
-  %add.ptr.i96.i = getelementptr inbounds i8, ptr addrspace(1) %add.ptr.i93.i, i64 %conv13.i
-  %call.ascast.i66.i = addrspacecast ptr addrspace(1) %add.ptr.i96.i to ptr addrspace(4)
-  %call1.i.i = tail call spir_func noundef target("spirv.JointMatrixINTEL", i8, 12, 48, 0, 3, 0) @_Z28__spirv_JointMatrixLoadINTELIaLm12ELm48ELN5__spv9MatrixUseE0ELNS0_12MatrixLayoutE0ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT3_EXT4_EXT2_EEEPS6_mS2_S4_i(ptr addrspace(4) noundef %call.ascast.i66.i, i64 noundef %_arg_K, i32 noundef 0, i32 noundef 3, i32 noundef 0) #4
-  %div20.i = mul nsw i32 %k.0.i, 12
-  %conv21.i = zext i32 %div20.i to i64
-  %mul23.i = mul i64 %mul22.i, %conv21.i
-  %add.ptr.i111.i = getelementptr i8, ptr addrspace(1) %add.ptr.i108140.i, i64 %mul23.i
-  %call.ascast.i72.i = addrspacecast ptr addrspace(1) %add.ptr.i111.i to ptr addrspace(4)
-  %call1.i73.i = tail call spir_func noundef target("spirv.JointMatrixINTEL", i8, 48, 12, 2, 3, 1) @_Z28__spirv_JointMatrixLoadINTELIaLm48ELm12ELN5__spv9MatrixUseE1ELNS0_12MatrixLayoutE2ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT3_EXT4_EXT2_EEEPS6_mS2_S4_i(ptr addrspace(4) noundef %call.ascast.i72.i, i64 noundef %mul22.i, i32 noundef 2, i32 noundef 3, i32 noundef 0) #4
-  call void @llvm.lifetime.start.p0(i64 8, ptr nonnull %ref.tmp29.sroa.0.i.0.i.0..sroa_cast)
-  %sub_c.sroa.0.i.0.sub_c.sroa.0.i.0.sub_c.sroa.0.0.sub_c.sroa.0.0.sub_c.sroa.0.0.125.i = load target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2), ptr %sub_c.sroa.0.i, align 8
-  %call.i77.i = tail call spir_func noundef target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2) @_Z27__spirv_JointMatrixMadINTELIaiLm12ELm48ELm12ELN5__spv9MatrixUseE0ELS1_1ELS1_2ELNS0_12MatrixLayoutE0ELS2_2ELS2_3ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT0_XT1_EXT3_EXT9_EXT10_EXT6_EEEPNS5_IT_XT1_EXT2_EXT7_EXT10_EXT4_EEEPNS5_IS9_XT2_EXT3_EXT8_EXT10_EXT5_EEES8_S4_(target("spirv.JointMatrixINTEL", i8, 12, 48, 0, 3, 0) noundef %call1.i.i, target("spirv.JointMatrixINTEL", i8, 48, 12, 2, 3, 1) noundef %call1.i73.i, target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2) noundef %sub_c.sroa.0.i.0.sub_c.sroa.0.i.0.sub_c.sroa.0.0.sub_c.sroa.0.0.sub_c.sroa.0.0.125.i, i32 noundef 3) #4
-  store target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2) %call.i77.i, ptr %ref.tmp29.sroa.0.i, align 8
-  %ref.tmp29.sroa.0.i.0.ref.tmp29.sroa.0.i.0.ref.tmp29.sroa.0.0.ref.tmp29.sroa.0.0.ref.tmp29.sroa.0.0..i = load i64, ptr %7, align 8
-  store i64 %ref.tmp29.sroa.0.i.0.ref.tmp29.sroa.0.i.0.ref.tmp29.sroa.0.0.ref.tmp29.sroa.0.0.ref.tmp29.sroa.0.0..i, ptr %8, align 8
-  call void @llvm.lifetime.end.p0(i64 8, ptr nonnull %ref.tmp29.sroa.0.i.0.i.0..sroa_cast)
-  %add.i = add nuw nsw i32 %k.0.i, 1
-  br label %for.cond.i
-
-_ZZZ15matrix_multiplyIiaLm24ELm96ELm24ELm96ELm24ELm24EEvR10big_matrixIT_XT5_EXT6_EERS0_IT0_XT1_EXT2_EERS0_IS4_XT3_EXT4_EEENKUlRN4sycl3_V17handlerEE_clESC_ENKUlNSA_7nd_itemILi2EEEE_clESF_.exit: ; preds = %for.cond.i
-  %mul37.i = mul i64 %mul.i, %_arg_N
-  %add.ptr.i.i = getelementptr inbounds i32, ptr addrspace(1) %_arg_accC, i64 %mul37.i
-  %mul39.i = mul nuw i64 %div2452.i, 12
-  %add.ptr.i81.i = getelementptr inbounds i32, ptr addrspace(1) %add.ptr.i.i, i64 %mul39.i
-  %call.ascast.i.i = addrspacecast ptr addrspace(1) %add.ptr.i81.i to ptr addrspace(4)
-  %sub_c.sroa.0.i.0.sub_c.sroa.0.i.0.sub_c.sroa.0.0.sub_c.sroa.0.0.sub_c.sroa.0.0..i = load target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2), ptr %sub_c.sroa.0.i, align 8
-  tail call spir_func void @_Z29__spirv_JointMatrixStoreINTELIiLm12ELm12ELN5__spv9MatrixUseE2ELNS0_12MatrixLayoutE3ELNS0_5Scope4FlagE3EEvPT_PNS0_24__spirv_JointMatrixINTELIS5_XT0_EXT1_EXT3_EXT4_EXT2_EEEmS2_S4_i(ptr addrspace(4) noundef %call.ascast.i.i, target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2) noundef %sub_c.sroa.0.i.0.sub_c.sroa.0.i.0.sub_c.sroa.0.0.sub_c.sroa.0.0.sub_c.sroa.0.0..i, i64 noundef %_arg_N, i32 noundef 0, i32 noundef 3, i32 noundef 0) #4
-  call void @llvm.lifetime.end.p0(i64 8, ptr nonnull %sub_c.sroa.0.i.0.i.0..sroa_cast)
-  ret void
-}
-
-; Function Attrs: convergent
-declare dso_local spir_func noundef target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2) @_Z26__spirv_CompositeConstructIiLm12ELm12ELN5__spv9MatrixUseE2ELNS0_12MatrixLayoutE3ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT3_EXT4_EXT2_EEES6_(i32 noundef) local_unnamed_addr #2
-
-; Function Attrs: convergent
-declare dso_local spir_func noundef target("spirv.JointMatrixINTEL", i8, 12, 48, 0, 3, 0) @_Z28__spirv_JointMatrixLoadINTELIaLm12ELm48ELN5__spv9MatrixUseE0ELNS0_12MatrixLayoutE0ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT3_EXT4_EXT2_EEEPS6_mS2_S4_i(ptr addrspace(4) noundef, i64 noundef, i32 noundef, i32 noundef, i32 noundef) local_unnamed_addr #2
-
-; Function Attrs: convergent
-declare dso_local spir_func noundef target("spirv.JointMatrixINTEL", i8, 48, 12, 2, 3, 1) @_Z28__spirv_JointMatrixLoadINTELIaLm48ELm12ELN5__spv9MatrixUseE1ELNS0_12MatrixLayoutE2ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT_XT0_EXT1_EXT3_EXT4_EXT2_EEEPS6_mS2_S4_i(ptr addrspace(4) noundef, i64 noundef, i32 noundef, i32 noundef, i32 noundef) local_unnamed_addr #2
-
-; Function Attrs: convergent
-declare dso_local spir_func noundef target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2) @_Z27__spirv_JointMatrixMadINTELIaiLm12ELm48ELm12ELN5__spv9MatrixUseE0ELS1_1ELS1_2ELNS0_12MatrixLayoutE0ELS2_2ELS2_3ELNS0_5Scope4FlagE3EEPNS0_24__spirv_JointMatrixINTELIT0_XT1_EXT3_EXT9_EXT10_EXT6_EEEPNS5_IT_XT1_EXT2_EXT7_EXT10_EXT4_EEEPNS5_IS9_XT2_EXT3_EXT8_EXT10_EXT5_EEES8_S4_(target("spirv.JointMatrixINTEL", i8, 12, 48, 0, 3, 0) noundef, target("spirv.JointMatrixINTEL", i8, 48, 12, 2, 3, 1) noundef, target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2) noundef, i32 noundef) local_unnamed_addr #2
-
-; Function Attrs: convergent
-declare dso_local spir_func void @_Z29__spirv_JointMatrixStoreINTELIiLm12ELm12ELN5__spv9MatrixUseE2ELNS0_12MatrixLayoutE3ELNS0_5Scope4FlagE3EEvPT_PNS0_24__spirv_JointMatrixINTELIS5_XT0_EXT1_EXT3_EXT4_EXT2_EEEmS2_S4_i(ptr addrspace(4) noundef, target("spirv.JointMatrixINTEL", i32, 12, 12, 3, 3, 2) noundef, i64 noundef, i32 noundef, i32 noundef, i32 noundef) local_unnamed_addr #2
-
-; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
-declare void @llvm.lifetime.start.p0(i64 immarg, ptr nocapture) #3
-
-; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
-declare void @llvm.lifetime.end.p0(i64 immarg, ptr nocapture) #3
-
-attributes #0 = { convergent norecurse "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "sycl-module-id"="matrix-int8-test.cpp" "uniform-work-group-size"="true" }
-attributes #1 = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: readwrite) }
-attributes #2 = { convergent "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
-attributes #3 = { nocallback nofree nosync nounwind willreturn memory(argmem: readwrite) }
-attributes #4 = { convergent }
diff --git a/test/extensions/INTEL/SPV_INTEL_joint_matrix/sycl_2020_namespace.ll b/test/extensions/INTEL/SPV_INTEL_joint_matrix/sycl_2020_namespace.ll
deleted file mode 100644
index 2a813950fd..0000000000
--- a/test/extensions/INTEL/SPV_INTEL_joint_matrix/sycl_2020_namespace.ll
+++ /dev/null
@@ -1,24 +0,0 @@
-; RUN: llvm-as -opaque-pointers=0 %s -o %t.bc
-; RUN: llvm-spirv %t.bc -opaque-pointers=0 -spirv-ext=+SPV_INTEL_joint_matrix -o %t.spv
-; RUN: llvm-spirv %t.spv -to-text -o - | FileCheck %s
-target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-n8:16:32:64"
-target triple = "spir64-unknown-unknown"
-
-; Ensure that ::sycl::_V1.*{half|bfloat16} are parsed as SYCL types.
-
-; CHECK-DAG: TypeFloat [[#HalfTy:]] 16
-; CHECK-DAG: TypeInt [[#BFloat16Ty:]] 16
-
-%"class.sycl::_V1::anything::half" = type { half }
-%"class.sycl::_V1::anything::bfloat16" = type { i16 }
-
-%"struct.__spv::__spirv_JointMatrixINTEL.half" = type { [2 x [2 x [1 x [4 x %"class.sycl::_V1::anything::half"]]]]* }
-%"struct.__spv::__spirv_JointMatrixINTEL.bfloat16" = type { [2 x [2 x [1 x [4 x %"class.sycl::_V1::anything::bfloat16"]]]]* }
-
-define spir_func void @foo(%"struct.__spv::__spirv_JointMatrixINTEL.half" *) {
-  ret void
-}
-
-define spir_func void @bar(%"struct.__spv::__spirv_JointMatrixINTEL.bfloat16" *) {
-  ret void
-}
diff --git a/test/extensions/INTEL/SPV_INTEL_joint_matrix/tf32_conversion_instructions.ll b/test/extensions/INTEL/SPV_INTEL_joint_matrix/tf32_conversion_instructions.ll
new file mode 100644
index 0000000000..8d0dcbfde4
--- /dev/null
+++ b/test/extensions/INTEL/SPV_INTEL_joint_matrix/tf32_conversion_instructions.ll
@@ -0,0 +1,53 @@
+; RUN: llvm-as < %s -o %t.bc
+; RUN: llvm-spirv %t.bc --spirv-ext=+SPV_KHR_cooperative_matrix,+SPV_INTEL_joint_matrix,+SPV_INTEL_tensor_float32_conversion -o %t.spv
+; RUN: llvm-spirv %t.spv -to-text -o %t.spt
+; RUN: FileCheck < %t.spt %s --check-prefix=CHECK-SPIRV
+
+; RUN: llvm-spirv -r %t.spv -o %t.rev.bc
+; RUN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM
+
+; RUN: not llvm-spirv %t.bc --spirv-ext=+SPV_KHR_cooperative_matrix,+SPV_INTEL_tensor_float32_conversion 2>&1 \
+; RUN: | FileCheck %s --check-prefix=CHECK-ERROR
+
+; CHECK-ERROR: InvalidInstruction: Can't translate llvm instruction:
+; CHECK-ERROR-NEXT: RoundFToTF32INTEL
+; CHECK-ERROR-NEXT: Can be used with cooperative matrices only when SPV_INTEL_joint_matrix is enabled
+
+; CHECK-SPIRV-DAG: Capability CooperativeMatrixKHR
+; CHECK-SPIRV-DAG: Capability TensorFloat32RoundingINTEL
+; CHECK-SPIRV-DAG: Capability JointMatrixTF32ComponentTypeINTEL
+; CHECK-SPIRV-DAG: Extension "SPV_INTEL_tensor_float32_conversion"
+; CHECK-SPIRV-DAG: Extension "SPV_KHR_cooperative_matrix"
+; CHECK-SPIRV-DAG: Extension "SPV_INTEL_joint_matrix"
+; CHECK-SPIRV-DAG: TypeFloat [[#FP32Ty:]] 32
+; CHECK-SPIRV-DAG: TypeCooperativeMatrixKHR [[#FP32MatTy:]] [[#FP32Ty]]
+; CHECK-SPIRV: CompositeConstruct [[#FP32MatTy]] [[#FP32Mat:]]
+; CHECK-SPIRV: RoundFToTF32INTEL [[#FP32MatTy]] [[#]] [[#FP32Mat]]
+
+; CHECK-LLVM: %[[#Mat:]] = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructf(float 0.000000e+00)
+; CHECK-LLVM: call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z25__spirv_RoundFToTF32INTELPU3AS145__spirv_CooperativeMatrixKHR__float_3_12_12_2(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %[[#Mat]])
+
+
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "spir64-unknown-unknown"
+
+define void @convert_f_to_tf() {
+entry:
+  %0 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructFloat(float 0.000000e+00)
+  %call = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z25__spirv_RoundFToTF32INTEL(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %0)
+  ret void
+}
+
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructFloat(float noundef)
+
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z25__spirv_RoundFToTF32INTEL(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) noundef)
+
+!llvm.module.flags = !{!0, !1, !2, !3, !4}
+!llvm.ident = !{!5}
+
+!0 = !{i32 7, !"Dwarf Version", i32 4}
+!1 = !{i32 1, !"wchar_size", i32 4}
+!2 = !{i32 8, !"PIC Level", i32 2}
+!3 = !{i32 7, !"PIE Level", i32 2}
+!4 = !{i32 7, !"uwtable", i32 2}
+!5 = !{!"clang version 17.0.0"}
diff --git a/test/extensions/KHR/SPV_KHR_cooperative_matrix/arithmetic_instructions.ll b/test/extensions/KHR/SPV_KHR_cooperative_matrix/arithmetic_instructions.ll
index 87e4fc17dd..40b8749e11 100644
--- a/test/extensions/KHR/SPV_KHR_cooperative_matrix/arithmetic_instructions.ll
+++ b/test/extensions/KHR/SPV_KHR_cooperative_matrix/arithmetic_instructions.ll
@@ -20,79 +20,79 @@ target triple = "spir-unknown-unknown"
 
 ; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeInt]] [[#MatrixIn:]] [[#]] {{$}}
 ; CHECK-SPIRV: SNegate [[#MatrixTypeInt]] [[#]] [[#MatrixIn]]
-; CHECK-LLVM: %1 = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructi(i32 0)
-; CHECK-LLVM: %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z15__spirv_SNegatePU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_3(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %1)
+; CHECK-LLVM: %1 = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructi(i32 0)
+; CHECK-LLVM: %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z15__spirv_SNegatePU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_2(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %1)
 define spir_kernel void @testSNegate(i32 %a) #0 !kernel_arg_addr_space !10 !kernel_arg_access_qual !11 !kernel_arg_type !12 !kernel_arg_type_qual !9 !kernel_arg_base_type !12 {
-  %1 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructInt32(i32 0)
-  %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z15__spirv_SNegate(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %1)
+  %1 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructInt32(i32 0)
+  %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z15__spirv_SNegate(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %1)
   ret void
 }
 
 ; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeFloat]] [[#MatrixIn:]] [[#]] {{$}}
 ; CHECK-SPIRV: FNegate [[#MatrixTypeFloat]] [[#]] [[#MatrixIn]]
-; CHECK-LLVM: %0 = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructf(float 0.000000e+00)
-; CHECK-LLVM: %call = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z15__spirv_FNegatePU3AS145__spirv_CooperativeMatrixKHR__float_3_12_12_3(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) %0)
+; CHECK-LLVM: %0 = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructf(float 0.000000e+00)
+; CHECK-LLVM: %call = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z15__spirv_FNegatePU3AS145__spirv_CooperativeMatrixKHR__float_3_12_12_2(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %0)
 define spir_kernel void @testFNeg(float %a) local_unnamed_addr #0 !kernel_arg_addr_space !2 !kernel_arg_access_qual !3 !kernel_arg_type !4 !kernel_arg_base_type !4 !kernel_arg_type_qual !9 {
 entry:
-  %0 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructFloat(float 0.000000e+00)
-  %call = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z15__spirv_FNegate(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) %0)
+  %0 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructFloat(float 0.000000e+00)
+  %call = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z15__spirv_FNegate(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %0)
   ret void
 }
 
 ; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeInt]] [[#MatrixA:]] [[#]] {{$}}
 ; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeInt]] [[#MatrixB:]] [[#]] {{$}}
 ; CHECK-SPIRV: IAdd [[#MatrixTypeInt]] [[#]] [[#MatrixA]] [[#MatrixB]]
-; CHECK-LLVM: %1 = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructi(i32 0)
-; CHECK-LLVM: %2 = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructi(i32 0)
-; CHECK-LLVM: %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z12__spirv_IAddPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_3S1_(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %1, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %2)
+; CHECK-LLVM: %1 = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructi(i32 0)
+; CHECK-LLVM: %2 = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructi(i32 0)
+; CHECK-LLVM: %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z12__spirv_IAddPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_2S1_(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %1, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %2)
 define spir_kernel void @testIAdd(i32 %a, i32 %b) #0 !kernel_arg_addr_space !4 !kernel_arg_access_qual !5 !kernel_arg_type !6 !kernel_arg_type_qual !7 !kernel_arg_base_type !6 {
-  %1 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructInt32(i32 0)
-  %2 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructInt32(i32 0)
-  %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z12__spirv_IAdd(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %1, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %2)
+  %1 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructInt32(i32 0)
+  %2 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructInt32(i32 0)
+  %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z12__spirv_IAdd(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %1, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %2)
   ret void
 }
 
 ; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeInt]] [[#MatrixA:]] [[#]] {{$}}
 ; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeInt]] [[#MatrixB:]] [[#]] {{$}}
 ; CHECK-SPIRV: ISub [[#MatrixTypeInt]] [[#]] [[#MatrixA]] [[#MatrixB]]
-; CHECK-LLVM: %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z12__spirv_ISubPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_3S1_(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %1, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %2)
+; CHECK-LLVM: %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z12__spirv_ISubPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_2S1_(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %1, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %2)
 define spir_kernel void @testISub(i32 %a, i32 %b) #0 !kernel_arg_addr_space !4 !kernel_arg_access_qual !5 !kernel_arg_type !6 !kernel_arg_type_qual !7 !kernel_arg_base_type !6 {
-  %1 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructInt32(i32 0)
-  %2 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructInt32(i32 0)
-  %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z12__spirv_ISub(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %1, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %2)
+  %1 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructInt32(i32 0)
+  %2 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructInt32(i32 0)
+  %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z12__spirv_ISub(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %1, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %2)
   ret void
 }
 
 ; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeInt]] [[#MatrixA:]] [[#]] {{$}}
 ; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeInt]] [[#MatrixB:]] [[#]] {{$}}
 ; CHECK-SPIRV: IMul [[#MatrixTypeInt]] [[#]] [[#MatrixA]] [[#MatrixB]]
-; CHECK-LLVM: %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z12__spirv_IMulPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_3S1_(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %1, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %2)
+; CHECK-LLVM: %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z12__spirv_IMulPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_2S1_(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %1, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %2)
 define spir_kernel void @testIMul(i32 %a, i32 %b) #0 !kernel_arg_addr_space !4 !kernel_arg_access_qual !5 !kernel_arg_type !6 !kernel_arg_type_qual !7 !kernel_arg_base_type !6 {
-  %1 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructInt32(i32 0)
-  %2 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructInt32(i32 0)
-  %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z12__spirv_IMul(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %1, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %2)
+  %1 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructInt32(i32 0)
+  %2 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructInt32(i32 0)
+  %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z12__spirv_IMul(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %1, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %2)
   ret void
 }
 
 ; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeInt]] [[#MatrixA:]] [[#]] {{$}}
 ; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeInt]] [[#MatrixB:]] [[#]] {{$}}
 ; CHECK-SPIRV: SDiv [[#MatrixTypeInt]] [[#]] [[#MatrixA]] [[#MatrixB]]
-; CHECK-LLVM: %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z12__spirv_SDivPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_3S1_(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %1, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %2)
+; CHECK-LLVM: %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z12__spirv_SDivPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_2S1_(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %1, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %2)
 define void @testSDiv(i32 %a, i32 %b) {
-  %1 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructInt32(i32 0)
-  %2 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructInt32(i32 0)
-  %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z12__spirv_SDiv(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %1, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %2)
+  %1 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructInt32(i32 0)
+  %2 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructInt32(i32 0)
+  %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z12__spirv_SDiv(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %1, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %2)
   ret void
 }
 
 ; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeInt]] [[#MatrixA:]] [[#]] {{$}}
 ; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeInt]] [[#MatrixB:]] [[#]] {{$}}
 ; CHECK-SPIRV: UDiv [[#MatrixTypeInt]] [[#]] [[#MatrixA]] [[#MatrixB]]
-; CHECK-LLVM: %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z12__spirv_UDivPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_3S1_(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %1, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %2)
+; CHECK-LLVM: %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z12__spirv_UDivPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_2S1_(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %1, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %2)
 define void @testUDiv(i32 %a, i32 %b) {
-  %1 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructInt32(i32 0)
-  %2 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructInt32(i32 0)
-  %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z12__spirv_UDiv(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %1, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %2)
+  %1 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructInt32(i32 0)
+  %2 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructInt32(i32 0)
+  %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z12__spirv_UDiv(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %1, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %2)
   ret void
 }
 
@@ -100,69 +100,69 @@ define void @testUDiv(i32 %a, i32 %b) {
 ; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeFloat]] [[#MatrixA:]] [[#]] {{$}}
 ; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeFloat]] [[#MatrixB:]] [[#]] {{$}}
 ; CHECK-SPIRV: FAdd [[#MatrixTypeFloat]] [[#]] [[#MatrixA]] [[#MatrixB]]
-; CHECK-LLVM: %0 = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructf(float 0.000000e+00)
-; CHECK-LLVM: %1 = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructf(float 0.000000e+00)
-; CHECK-LLVM: %call = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z12__spirv_FAddPU3AS145__spirv_CooperativeMatrixKHR__float_3_12_12_3S1_(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) %0, target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) %1)
+; CHECK-LLVM: %0 = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructf(float 0.000000e+00)
+; CHECK-LLVM: %1 = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructf(float 0.000000e+00)
+; CHECK-LLVM: %call = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z12__spirv_FAddPU3AS145__spirv_CooperativeMatrixKHR__float_3_12_12_2S1_(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %0, target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %1)
 define spir_kernel void @testFAdd(float %a, float %b) local_unnamed_addr #0 !kernel_arg_addr_space !2 !kernel_arg_access_qual !3 !kernel_arg_type !4 !kernel_arg_base_type !4 !kernel_arg_type_qual !5 {
 entry:
-  %0 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructFloat(float 0.000000e+00)
-  %1 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructFloat(float 0.000000e+00)
-  %call = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z12__spirv_FAdd(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) %0, target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) %1)
+  %0 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructFloat(float 0.000000e+00)
+  %1 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructFloat(float 0.000000e+00)
+  %call = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z12__spirv_FAdd(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %0, target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %1)
   ret void
 }
 
 ; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeFloat]] [[#MatrixA:]] [[#]] {{$}}
 ; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeFloat]] [[#MatrixB:]] [[#]] {{$}}
 ; CHECK-SPIRV: FSub [[#MatrixTypeFloat]] [[#]] [[#MatrixA]] [[#MatrixB]]
-; CHECK-LLVM: %call = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z12__spirv_FSubPU3AS145__spirv_CooperativeMatrixKHR__float_3_12_12_3S1_(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) %0, target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) %1)
+; CHECK-LLVM: %call = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z12__spirv_FSubPU3AS145__spirv_CooperativeMatrixKHR__float_3_12_12_2S1_(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %0, target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %1)
 define spir_kernel void @testFSub(float %a, float %b) local_unnamed_addr #0 !kernel_arg_addr_space !2 !kernel_arg_access_qual !3 !kernel_arg_type !4 !kernel_arg_base_type !4 !kernel_arg_type_qual !5 {
 entry:
-  %0 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructFloat(float 0.000000e+00)
-  %1 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructFloat(float 0.000000e+00)
-  %call = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z12__spirv_FSub(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) %0, target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) %1)
+  %0 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructFloat(float 0.000000e+00)
+  %1 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructFloat(float 0.000000e+00)
+  %call = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z12__spirv_FSub(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %0, target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %1)
   ret void
 }
 
 ; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeFloat]] [[#MatrixA:]] [[#]] {{$}}
 ; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeFloat]] [[#MatrixB:]] [[#]] {{$}}
 ; CHECK-SPIRV: FMul [[#MatrixTypeFloat]] [[#]] [[#MatrixA]] [[#MatrixB]]
-; CHECK-LLVM: %call = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z12__spirv_FMulPU3AS145__spirv_CooperativeMatrixKHR__float_3_12_12_3S1_(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) %0, target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) %1)
+; CHECK-LLVM: %call = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z12__spirv_FMulPU3AS145__spirv_CooperativeMatrixKHR__float_3_12_12_2S1_(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %0, target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %1)
 define spir_kernel void @testFMul(float %a, float %b) local_unnamed_addr #0 !kernel_arg_addr_space !2 !kernel_arg_access_qual !3 !kernel_arg_type !4 !kernel_arg_base_type !4 !kernel_arg_type_qual !5 {
 entry:
-  %0 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructFloat(float 0.000000e+00)
-  %1 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructFloat(float 0.000000e+00)
-  %call = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z12__spirv_FMul(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) %0, target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) %1)
+  %0 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructFloat(float 0.000000e+00)
+  %1 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructFloat(float 0.000000e+00)
+  %call = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z12__spirv_FMul(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %0, target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %1)
   ret void
 }
 
 ; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeFloat]] [[#MatrixA:]] [[#]] {{$}}
 ; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeFloat]] [[#MatrixB:]] [[#]] {{$}}
 ; CHECK-SPIRV: FDiv [[#MatrixTypeFloat]] [[#]] [[#MatrixA]] [[#MatrixB]]
-; CHECK-LLVM: %call = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z12__spirv_FDivPU3AS145__spirv_CooperativeMatrixKHR__float_3_12_12_3S1_(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) %0, target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) %1)
+; CHECK-LLVM: %call = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z12__spirv_FDivPU3AS145__spirv_CooperativeMatrixKHR__float_3_12_12_2S1_(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %0, target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %1)
 define spir_kernel void @testFDiv(float %a, float %b) local_unnamed_addr #0 !kernel_arg_addr_space !2 !kernel_arg_access_qual !3 !kernel_arg_type !4 !kernel_arg_base_type !4 !kernel_arg_type_qual !5 {
 entry:
-  %0 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructFloat(float 0.000000e+00)
-  %1 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructFloat(float 0.000000e+00)
-  %call = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z12__spirv_FDiv(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) %0, target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) %1)
+  %0 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructFloat(float 0.000000e+00)
+  %1 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructFloat(float 0.000000e+00)
+  %call = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z12__spirv_FDiv(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %0, target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %1)
   ret void
 }
 
-declare spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructFloat(float noundef)
-declare spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructInt32(i32 noundef)
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructFloat(float noundef)
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructInt32(i32 noundef)
 
-declare spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z15__spirv_FNegate(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) noundef)
-declare spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z15__spirv_SNegate(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) noundef)
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z15__spirv_FNegate(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) noundef)
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z15__spirv_SNegate(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) noundef)
 
-declare spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z12__spirv_IAdd(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) noundef, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) noundef)
-declare spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z12__spirv_ISub(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) noundef, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) noundef)
-declare spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z12__spirv_IMul(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) noundef, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) noundef)
-declare spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z12__spirv_SDiv(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) noundef, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) noundef)
-declare spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z12__spirv_UDiv(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) noundef, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) noundef)
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z12__spirv_IAdd(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) noundef, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) noundef)
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z12__spirv_ISub(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) noundef, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) noundef)
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z12__spirv_IMul(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) noundef, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) noundef)
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z12__spirv_SDiv(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) noundef, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) noundef)
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z12__spirv_UDiv(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) noundef, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) noundef)
 
-declare spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z12__spirv_FAdd(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) noundef, target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) noundef)
-declare spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z12__spirv_FSub(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) noundef, target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) noundef)
-declare spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z12__spirv_FMul(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) noundef, target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) noundef)
-declare spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z12__spirv_FDiv(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) noundef, target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) noundef)
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z12__spirv_FAdd(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) noundef, target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) noundef)
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z12__spirv_FSub(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) noundef, target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) noundef)
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z12__spirv_FMul(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) noundef, target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) noundef)
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z12__spirv_FDiv(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) noundef, target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) noundef)
 
 
 attributes #0 = { nounwind }
diff --git a/test/extensions/KHR/SPV_KHR_cooperative_matrix/array_of_matrices.ll b/test/extensions/KHR/SPV_KHR_cooperative_matrix/array_of_matrices.ll
new file mode 100644
index 0000000000..28979b4eb1
--- /dev/null
+++ b/test/extensions/KHR/SPV_KHR_cooperative_matrix/array_of_matrices.ll
@@ -0,0 +1,437 @@
+;; Compiled from joint_matrix_bf16_fill_k_cache.cpp from https://github.com/intel/llvm
+;; command: clang++ -fsycl -DSYCL_EXT_ONEAPI_MATRIX_VERSION=4 llvm/sycl/test-e2e/Matrix/joint_matrix_bf16_fill_k_cache.cpp -fsycl-device-only -o test.bc
+;; and then JointMatrixINTEL target ext type was replaced with CooperativeMatrixKHR
+
+; RUN: llvm-as < %s -o %t.bc
+; RUN: llvm-spirv %t.bc --spirv-ext=+SPV_KHR_cooperative_matrix -o %t.spv
+; RUN: llvm-spirv %t.spv -to-text -o %t.spt
+; RUN: FileCheck < %t.spt %s --check-prefix=CHECK-SPIRV
+
+; RUN: llvm-spirv -r %t.spv -o %t.rev.bc
+; RUN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM
+
+; CHECK-SPIRV-DAG: Capability CooperativeMatrixKHR
+; CHECK-SPIRV-DAG: Extension "SPV_KHR_cooperative_matrix"
+; CHECK-SPIRV: TypeInt [[#Int16Ty:]] 16 0
+; CHECK-SPIRV: TypeFloat [[#FloatTy:]] 32
+; CHECK-SPIRV: TypeCooperativeMatrixKHR [[#MatTy1:]] [[#FloatTy]]
+; CHECK-SPIRV: TypeCooperativeMatrixKHR [[#MatTy2:]] [[#Int16Ty]]
+; CHECK-SPIRV: TypeCooperativeMatrixKHR [[#MatTy3:]] [[#Int16Ty]]
+; CHECK-SPIRV: TypeStruct [[#StructTy1:]] [[#MatTy1]]
+; CHECK-SPIRV: TypeArray [[#ArrayTy1:]] [[#StructTy1]] [[#]]
+; CHECK-SPIRV: TypeArray [[#]] [[#ArrayTy1]] [[#]]
+; CHECK-SPIRV: TypeStruct [[#StructTy2:]] [[#MatTy2]]
+; CHECK-SPIRV: TypeArray [[#ArrayTy2:]] [[#StructTy2]] [[#]]
+; CHECK-SPIRV: TypeArray [[#]] [[#ArrayTy2]] [[#]]
+; CHECK-SPIRV: TypeStruct [[#StructTy3:]] [[#MatTy3]]
+; CHECK-SPIRV: TypeArray [[#ArrayTy3:]] [[#StructTy3]] [[#]]
+; CHECK-SPIRV: TypeArray [[#]] [[#ArrayTy3]] [[#]]
+
+; CHECK-LLVM: %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix" = type { target("spirv.CooperativeMatrixKHR", float, 3, 8, 16, 2) }
+; CHECK-LLVM: %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix.5" = type { target("spirv.CooperativeMatrixKHR", i16, 3, 8, 16, 0) }
+; CHECK-LLVM: %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix.6" = type { target("spirv.CooperativeMatrixKHR", i16, 3, 16, 16, 1) }
+; CHECK-LLVM: alloca [4 x [4 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix"]]
+; CHECK-LLVM: alloca [4 x [2 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix.5"]]
+; CHECK-LLVM: alloca [4 x [2 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix.6"]]
+
+; ModuleID = 'test.bc'
+source_filename = "llvm/sycl/test-e2e/Matrix/joint_matrix_bf16_fill_k_cache.cpp"
+target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-n8:16:32:64"
+target triple = "spir64-unknown-unknown"
+
+%"class.sycl::_V1::__generated_multi_ptr" = type { ptr addrspace(1) }
+%"class.sycl::_V1::__generated_multi_ptr.0" = type { ptr addrspace(1) }
+%"class.sycl::_V1::__generated_multi_ptr.1" = type { ptr addrspace(1) }
+%"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix" = type { target("spirv.CooperativeMatrixKHR", float, 3, 8, 16, 2) }
+%"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix.5" = type { target("spirv.CooperativeMatrixKHR", i16, 3, 8, 16, 0) }
+%"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix.6" = type { target("spirv.CooperativeMatrixKHR", i16, 3, 16, 16, 1) }
+%"class.sycl::_V1::ext::oneapi::bfloat16" = type { i16 }
+
+$_ZTSZZ12joint_matmulILj256ELj256ELj256ELj256ELj2EN4sycl3_V13ext6oneapi8bfloat16EfLj16EEdPT4_S6_PT5_RNS1_5queueEiENKUlRNS1_7handlerEE_clESC_EUlNS1_7nd_itemILi2EEEE_ = comdat any
+
+@__spirv_BuiltInWorkgroupId = external dso_local local_unnamed_addr addrspace(1) constant <3 x i64>, align 32
+@__spirv_BuiltInLocalInvocationId = external dso_local local_unnamed_addr addrspace(1) constant <3 x i64>, align 32
+
+; Function Attrs: convergent norecurse nounwind
+define weak_odr dso_local spir_kernel void @_ZTSZZ12joint_matmulILj256ELj256ELj256ELj256ELj2EN4sycl3_V13ext6oneapi8bfloat16EfLj16EEdPT4_S6_PT5_RNS1_5queueEiENKUlRNS1_7handlerEE_clESC_EUlNS1_7nd_itemILi2EEEE_(ptr noundef byval(%"class.sycl::_V1::__generated_multi_ptr") align 8 %_arg_pA, ptr noundef byval(%"class.sycl::_V1::__generated_multi_ptr.0") align 8 %_arg_pB, ptr noundef byval(%"class.sycl::_V1::__generated_multi_ptr.1") align 8 %_arg_pC) local_unnamed_addr #0 comdat !srcloc !59 !kernel_arg_buffer_location !60 !intel_reqd_sub_group_size !61 !sycl_fixed_targets !62 !sycl_kernel_omit_args !63 {
+entry:
+  call void @__itt_offload_wi_start_wrapper()
+  %tC.i = alloca [4 x [4 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix"]], align 8
+  %tA.i = alloca [4 x [2 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix.5"]], align 8
+  %tB.i = alloca [4 x [2 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix.6"]], align 8
+  %0 = load i64, ptr %_arg_pA, align 8, !tbaa !64
+  %1 = inttoptr i64 %0 to ptr addrspace(1)
+  %2 = load i64, ptr %_arg_pB, align 8, !tbaa !64
+  %3 = inttoptr i64 %2 to ptr addrspace(1)
+  %4 = load i64, ptr %_arg_pC, align 8, !tbaa !64
+  %5 = inttoptr i64 %4 to ptr addrspace(1)
+  %6 = load i64, ptr addrspace(1) getelementptr inbounds (i8, ptr addrspace(1) @__spirv_BuiltInWorkgroupId, i64 8), align 8, !noalias !68
+  %7 = load i64, ptr addrspace(1) @__spirv_BuiltInWorkgroupId, align 32, !noalias !68
+  %8 = load i64, ptr addrspace(1) getelementptr inbounds (i8, ptr addrspace(1) @__spirv_BuiltInLocalInvocationId, i64 8), align 8, !noalias !75
+  %9 = load i64, ptr addrspace(1) @__spirv_BuiltInLocalInvocationId, align 32, !noalias !75
+  %cmp.i.i = icmp ult i64 %6, 2147483648
+  tail call void @llvm.assume(i1 %cmp.i.i)
+  %cmp.i208.i = icmp ult i64 %7, 2147483648
+  tail call void @llvm.assume(i1 %cmp.i208.i)
+  %cmp.i209.i = icmp ult i64 %8, 2147483648
+  tail call void @llvm.assume(i1 %cmp.i209.i)
+  %cmp.i212.i = icmp ult i64 %9, 2147483648
+  tail call void @llvm.assume(i1 %cmp.i212.i)
+  %div205.i = lshr i64 %9, 4
+  call void @llvm.lifetime.start.p0(i64 128, ptr nonnull %tC.i) #4
+  br label %arrayctor.loop.i
+
+arrayctor.loop.i:                                 ; preds = %arrayctor.loop.i, %entry
+  %arrayctor.cur.idx.i = phi i64 [ 0, %entry ], [ %arrayctor.cur.add.i, %arrayctor.loop.i ]
+  %arrayctor.cur.add.i = add nuw nsw i64 %arrayctor.cur.idx.i, 1
+  %arrayctor.done.i = icmp eq i64 %arrayctor.cur.add.i, 16
+  br i1 %arrayctor.done.i, label %for.cond.i, label %arrayctor.loop.i
+
+for.cond.i:                                       ; preds = %arrayctor.loop.i, %for.cond.cleanup7.i
+  %m.0.i = phi i32 [ %inc12.i, %for.cond.cleanup7.i ], [ 0, %arrayctor.loop.i ]
+  %cmp.i = icmp ult i32 %m.0.i, 4
+  br i1 %cmp.i, label %for.cond5.preheader.i, label %for.cond14.preheader.i
+
+for.cond5.preheader.i:                            ; preds = %for.cond.i
+  %idxprom.i = zext i32 %m.0.i to i64
+  br label %for.cond5.i
+
+for.cond14.preheader.i:                           ; preds = %for.cond.i
+  %mul50.i = shl nuw nsw i64 %6, 8
+  %mul51.i = shl nuw nsw i64 %8, 5
+  %add52.i = add nuw nsw i64 %mul50.i, %mul51.i
+  %mul80.i = shl nuw nsw i64 %div205.i, 7
+  %10 = shl nuw nsw i64 %7, 9
+  %11 = add nuw nsw i64 %10, %mul80.i
+  br label %for.cond14.i
+
+for.cond5.i:                                      ; preds = %for.body8.i, %for.cond5.preheader.i
+  %n.0.i = phi i32 [ %inc.i, %for.body8.i ], [ 0, %for.cond5.preheader.i ]
+  %cmp6.i = icmp ult i32 %n.0.i, 4
+  br i1 %cmp6.i, label %for.body8.i, label %for.cond.cleanup7.i
+
+for.cond.cleanup7.i:                              ; preds = %for.cond5.i
+  %inc12.i = add nuw nsw i32 %m.0.i, 1
+  br label %for.cond.i, !llvm.loop !80
+
+for.body8.i:                                      ; preds = %for.cond5.i
+  %conv.i = zext i32 %n.0.i to i64
+  %arrayidx10.i = getelementptr inbounds [4 x [4 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix"]], ptr %tC.i, i64 0, i64 %idxprom.i, i64 %conv.i
+  %call.i.i = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 8, 16, 2) @_Z26__spirv_CompositeConstructIffLm8ELm16ELN5__spv9MatrixUseE2ELNS0_12MatrixLayoutE3ELNS0_5Scope4FlagE3EEPNS0_24__spirv_CooperativeMatrixKHRIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEET_(float noundef 0.000000e+00) #5
+  store target("spirv.CooperativeMatrixKHR", float, 3, 8, 16, 2) %call.i.i, ptr %arrayidx10.i, align 8, !tbaa !82
+  %inc.i = add nuw nsw i32 %n.0.i, 1
+  br label %for.cond5.i, !llvm.loop !84
+
+for.cond14.i:                                     ; preds = %for.cond.cleanup34.i, %for.cond14.preheader.i
+  %k2.0.i = phi i32 [ %inc129.i, %for.cond.cleanup34.i ], [ 0, %for.cond14.preheader.i ]
+  %cmp15.i = icmp ult i32 %k2.0.i, 8
+  br i1 %cmp15.i, label %for.body17.i, label %for.cond132.preheader.i
+
+for.cond132.preheader.i:                          ; preds = %for.cond14.i
+  %mul156.i = shl nuw nsw i64 %7, 8
+  %mul157.i = shl nuw nsw i64 %div205.i, 6
+  %add158.i = add nuw nsw i64 %mul156.i, %mul157.i
+  br label %for.cond132.i
+
+for.body17.i:                                     ; preds = %for.cond14.i
+  call void @llvm.lifetime.start.p0(i64 64, ptr nonnull %tA.i) #4
+  br label %arrayctor.loop20.i
+
+arrayctor.loop20.i:                               ; preds = %arrayctor.loop20.i, %for.body17.i
+  %arrayctor.cur21.idx.i = phi i64 [ 0, %for.body17.i ], [ %arrayctor.cur21.add.i, %arrayctor.loop20.i ]
+  %arrayctor.cur21.add.i = add nuw nsw i64 %arrayctor.cur21.idx.i, 1
+  %arrayctor.done23.i = icmp eq i64 %arrayctor.cur21.add.i, 8
+  br i1 %arrayctor.done23.i, label %arrayctor.cont24.i, label %arrayctor.loop20.i
+
+arrayctor.cont24.i:                               ; preds = %arrayctor.loop20.i
+  call void @llvm.lifetime.start.p0(i64 64, ptr nonnull %tB.i) #4
+  br label %arrayctor.loop27.i
+
+arrayctor.loop27.i:                               ; preds = %arrayctor.loop27.i, %arrayctor.cont24.i
+  %arrayctor.cur28.idx.i = phi i64 [ 0, %arrayctor.cont24.i ], [ %arrayctor.cur28.add.i, %arrayctor.loop27.i ]
+  %arrayctor.cur28.add.i = add nuw nsw i64 %arrayctor.cur28.idx.i, 1
+  %arrayctor.done30.i = icmp eq i64 %arrayctor.cur28.add.i, 8
+  br i1 %arrayctor.done30.i, label %for.cond32.preheader.i, label %arrayctor.loop27.i
+
+for.cond32.preheader.i:                           ; preds = %arrayctor.loop27.i
+  %12 = shl nuw i32 %k2.0.i, 1
+  br label %for.cond32.i
+
+for.cond32.i:                                     ; preds = %for.cond.cleanup92.i, %for.cond32.preheader.i
+  %k1.0.i = phi i32 [ %inc126.i, %for.cond.cleanup92.i ], [ 0, %for.cond32.preheader.i ]
+  %cmp33.i = icmp ult i32 %k1.0.i, 2
+  br i1 %cmp33.i, label %for.body35.i, label %for.cond.cleanup34.i
+
+for.cond.cleanup34.i:                             ; preds = %for.cond32.i
+  call void @llvm.lifetime.end.p0(i64 64, ptr nonnull %tB.i) #4
+  call void @llvm.lifetime.end.p0(i64 64, ptr nonnull %tA.i) #4
+  %inc129.i = add nuw nsw i32 %k2.0.i, 1
+  br label %for.cond14.i, !llvm.loop !85
+
+for.body35.i:                                     ; preds = %for.cond32.i
+  %13 = add nuw i32 %12, %k1.0.i
+  %div37206.i = and i32 %13, 268435455
+  %idxprom46.i = zext i32 %k1.0.i to i64
+  %mul57.i = shl nuw nsw i32 %div37206.i, 4
+  %conv58.i = zext i32 %mul57.i to i64
+  %invariant.gep = getelementptr inbounds %"class.sycl::_V1::ext::oneapi::bfloat16", ptr addrspace(1) %1, i64 %conv58.i
+  br label %for.cond39.i
+
+for.cond39.i:                                     ; preds = %for.body42.i, %for.body35.i
+  %m38.0.i = phi i32 [ 0, %for.body35.i ], [ %inc60.i, %for.body42.i ]
+  %cmp40.i = icmp ult i32 %m38.0.i, 4
+  br i1 %cmp40.i, label %for.body42.i, label %for.cond63.preheader.i
+
+for.cond63.preheader.i:                           ; preds = %for.cond39.i
+  %mul77.i = shl nuw nsw i32 %div37206.i, 12
+  %conv78.i = zext i32 %mul77.i to i64
+  %add.ptr.i225.i = getelementptr inbounds %"class.sycl::_V1::ext::oneapi::bfloat16", ptr addrspace(1) %3, i64 %conv78.i
+  br label %for.cond63.i
+
+for.body42.i:                                     ; preds = %for.cond39.i
+  %idxprom44.i = zext i32 %m38.0.i to i64
+  %arrayidx47.i = getelementptr inbounds [4 x [2 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix.5"]], ptr %tA.i, i64 0, i64 %idxprom44.i, i64 %idxprom46.i
+  %mul53.i = shl nuw nsw i32 %m38.0.i, 3
+  %conv54.i = zext i32 %mul53.i to i64
+  %add55.i = add nuw nsw i64 %add52.i, %conv54.i
+  %mul56.i = shl nuw nsw i64 %add55.i, 8
+  %gep = getelementptr inbounds %"class.sycl::_V1::ext::oneapi::bfloat16", ptr addrspace(1) %invariant.gep, i64 %mul56.i
+  %call1.i.i = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i16, 3, 8, 16, 0) @_Z32__spirv_CooperativeMatrixLoadKHRIU3AS1N4sycl3_V13ext6oneapi8bfloat16ES4_Lm8ELm16ELN5__spv9MatrixUseE0ELNS6_12MatrixLayoutE0ELNS6_5Scope4FlagE3EEPNS6_24__spirv_CooperativeMatrixKHRIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEPT_mS8_SA_i(ptr addrspace(1) noundef %gep, i64 noundef 256, i32 noundef 0, i32 noundef 3, i32 noundef 0) #5
+  store target("spirv.CooperativeMatrixKHR", i16, 3, 8, 16, 0) %call1.i.i, ptr %arrayidx47.i, align 8, !tbaa !86
+  %inc60.i = add nuw nsw i32 %m38.0.i, 1
+  br label %for.cond39.i, !llvm.loop !88
+
+for.cond63.i:                                     ; preds = %for.body67.i, %for.cond63.preheader.i
+  %n62.0.i = phi i32 [ %inc87.i, %for.body67.i ], [ 0, %for.cond63.preheader.i ]
+  %cmp65.i = icmp ult i32 %n62.0.i, 4
+  br i1 %cmp65.i, label %for.body67.i, label %for.cond90.i
+
+for.body67.i:                                     ; preds = %for.cond63.i
+  %conv64.i = zext i32 %n62.0.i to i64
+  %arrayidx72.i = getelementptr inbounds [4 x [2 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix.6"]], ptr %tB.i, i64 0, i64 %conv64.i, i64 %idxprom46.i
+  %14 = shl nuw nsw i64 %conv64.i, 5
+  %mul85.i = add nuw nsw i64 %14, %11
+  %add.ptr.i226.i = getelementptr inbounds %"class.sycl::_V1::ext::oneapi::bfloat16", ptr addrspace(1) %add.ptr.i225.i, i64 %mul85.i
+  %call1.i219.i = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i16, 3, 16, 16, 1) @_Z32__spirv_CooperativeMatrixLoadKHRIU3AS1N4sycl3_V13ext6oneapi8bfloat16ES4_Lm16ELm16ELN5__spv9MatrixUseE1ELNS6_12MatrixLayoutE2ELNS6_5Scope4FlagE3EEPNS6_24__spirv_CooperativeMatrixKHRIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEPT_mS8_SA_i(ptr addrspace(1) noundef %add.ptr.i226.i, i64 noundef 512, i32 noundef 2, i32 noundef 3, i32 noundef 0) #5
+  store target("spirv.CooperativeMatrixKHR", i16, 3, 16, 16, 1) %call1.i219.i, ptr %arrayidx72.i, align 8, !tbaa !89
+  %inc87.i = add nuw nsw i32 %n62.0.i, 1
+  br label %for.cond63.i, !llvm.loop !91
+
+for.cond90.i:                                     ; preds = %for.cond63.i, %for.cond.cleanup98.i
+  %m89.0.i = phi i32 [ %inc123.i, %for.cond.cleanup98.i ], [ 0, %for.cond63.i ]
+  %cmp91.i = icmp ult i32 %m89.0.i, 4
+  br i1 %cmp91.i, label %for.cond95.preheader.i, label %for.cond.cleanup92.i
+
+for.cond95.preheader.i:                           ; preds = %for.cond90.i
+  %idxprom102.i = zext i32 %m89.0.i to i64
+  %arrayidx105.i = getelementptr inbounds [4 x [2 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix.5"]], ptr %tA.i, i64 0, i64 %idxprom102.i, i64 %idxprom46.i
+  %15 = load target("spirv.CooperativeMatrixKHR", i16, 3, 8, 16, 0), ptr %arrayidx105.i, align 8, !tbaa !86, !noalias !92
+  br label %for.cond95.i
+
+for.cond.cleanup92.i:                             ; preds = %for.cond90.i
+  %inc126.i = add nuw nsw i32 %k1.0.i, 1
+  br label %for.cond32.i, !llvm.loop !95
+
+for.cond95.i:                                     ; preds = %for.body99.i, %for.cond95.preheader.i
+  %n94.0.i = phi i32 [ %inc120.i, %for.body99.i ], [ 0, %for.cond95.preheader.i ]
+  %cmp97.i = icmp ult i32 %n94.0.i, 4
+  br i1 %cmp97.i, label %for.body99.i, label %for.cond.cleanup98.i
+
+for.cond.cleanup98.i:                             ; preds = %for.cond95.i
+  %inc123.i = add nuw nsw i32 %m89.0.i, 1
+  br label %for.cond90.i, !llvm.loop !96
+
+for.body99.i:                                     ; preds = %for.cond95.i
+  %conv96.i = zext i32 %n94.0.i to i64
+  %arrayidx109.i = getelementptr inbounds [4 x [2 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix.6"]], ptr %tB.i, i64 0, i64 %conv96.i, i64 %idxprom46.i
+  %arrayidx113.i = getelementptr inbounds [4 x [4 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix"]], ptr %tC.i, i64 0, i64 %idxprom102.i, i64 %conv96.i
+  %16 = load target("spirv.CooperativeMatrixKHR", i16, 3, 16, 16, 1), ptr %arrayidx109.i, align 8, !tbaa !89, !noalias !92
+  %17 = load target("spirv.CooperativeMatrixKHR", float, 3, 8, 16, 2), ptr %arrayidx113.i, align 8, !tbaa !82, !noalias !92
+  %call.i221.i = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 8, 16, 2) @_Z31__spirv_CooperativeMatrixMadKHRIN4sycl3_V13ext6oneapi8bfloat16EfLm8ELm16ELm16ELN5__spv9MatrixUseE0ELS6_1ELS6_2ELNS5_12MatrixLayoutE0ELS7_2ELS7_3ELNS5_5Scope4FlagE3EEPNS5_24__spirv_CooperativeMatrixKHRIT0_XT1_EXT3_EXT9_EXT10_EXT6_EEEPNSA_IT_XT1_EXT2_EXT7_EXT10_EXT4_EEEPNSA_ISE_XT2_EXT3_EXT8_EXT10_EXT5_EEESD_S9_(target("spirv.CooperativeMatrixKHR", i16, 3, 8, 16, 0) noundef %15, target("spirv.CooperativeMatrixKHR", i16, 3, 16, 16, 1) noundef %16, target("spirv.CooperativeMatrixKHR", float, 3, 8, 16, 2) noundef %17, i32 noundef 3) #5, !noalias !92
+  store target("spirv.CooperativeMatrixKHR", float, 3, 8, 16, 2) %call.i221.i, ptr %arrayidx113.i, align 8, !tbaa !82
+  %inc120.i = add nuw nsw i32 %n94.0.i, 1
+  br label %for.cond95.i, !llvm.loop !97
+
+for.cond132.i:                                    ; preds = %for.cond.cleanup140.i, %for.cond132.preheader.i
+  %m131.0.i = phi i32 [ %inc166.i, %for.cond.cleanup140.i ], [ 0, %for.cond132.preheader.i ]
+  %cmp133.i = icmp ult i32 %m131.0.i, 4
+  br i1 %cmp133.i, label %for.cond137.preheader.i, label %_ZZZ12joint_matmulILj256ELj256ELj256ELj256ELj2EN4sycl3_V13ext6oneapi8bfloat16EfLj16EEdPT4_S6_PT5_RNS1_5queueEiENKUlRNS1_7handlerEE_clESC_ENKUlNS1_7nd_itemILi2EEEE_clESF_.exit
+
+for.cond137.preheader.i:                          ; preds = %for.cond132.i
+  %idxprom143.i = zext i32 %m131.0.i to i64
+  %mul152.i = shl nuw nsw i32 %m131.0.i, 3
+  %conv153.i = zext i32 %mul152.i to i64
+  %add154.i = add nuw nsw i64 %add52.i, %conv153.i
+  %mul155.i = shl nuw nsw i64 %add154.i, 8
+  %add.ptr.i227.i = getelementptr inbounds float, ptr addrspace(1) %5, i64 %mul155.i
+  br label %for.cond137.i
+
+for.cond137.i:                                    ; preds = %for.body141.i, %for.cond137.preheader.i
+  %n136.0.i = phi i32 [ %inc163.i, %for.body141.i ], [ 0, %for.cond137.preheader.i ]
+  %cmp139.i = icmp ult i32 %n136.0.i, 4
+  br i1 %cmp139.i, label %for.body141.i, label %for.cond.cleanup140.i
+
+for.cond.cleanup140.i:                            ; preds = %for.cond137.i
+  %inc166.i = add nuw nsw i32 %m131.0.i, 1
+  br label %for.cond132.i, !llvm.loop !98
+
+for.body141.i:                                    ; preds = %for.cond137.i
+  %conv138.i = zext i32 %n136.0.i to i64
+  %arrayidx146.i = getelementptr inbounds [4 x [4 x %"struct.sycl::_V1::ext::oneapi::experimental::matrix::joint_matrix"]], ptr %tC.i, i64 0, i64 %idxprom143.i, i64 %conv138.i
+  %mul160.i = shl nuw nsw i64 %conv138.i, 4
+  %add161.i = add nuw nsw i64 %add158.i, %mul160.i
+  %add.ptr.i228.i = getelementptr inbounds float, ptr addrspace(1) %add.ptr.i227.i, i64 %add161.i
+  %18 = load target("spirv.CooperativeMatrixKHR", float, 3, 8, 16, 2), ptr %arrayidx146.i, align 8, !tbaa !82
+  tail call spir_func void @_Z33__spirv_CooperativeMatrixStoreKHRIU3AS1ffLm8ELm16ELN5__spv9MatrixUseE2ELNS1_12MatrixLayoutE3ELNS1_5Scope4FlagE3EEvPT_PNS1_24__spirv_CooperativeMatrixKHRIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEmS3_S5_i(ptr addrspace(1) noundef %add.ptr.i228.i, target("spirv.CooperativeMatrixKHR", float, 3, 8, 16, 2) noundef %18, i64 noundef 256, i32 noundef 0, i32 noundef 3, i32 noundef 0) #5
+  %inc163.i = add nuw nsw i32 %n136.0.i, 1
+  br label %for.cond137.i, !llvm.loop !99
+
+_ZZZ12joint_matmulILj256ELj256ELj256ELj256ELj2EN4sycl3_V13ext6oneapi8bfloat16EfLj16EEdPT4_S6_PT5_RNS1_5queueEiENKUlRNS1_7handlerEE_clESC_ENKUlNS1_7nd_itemILi2EEEE_clESF_.exit: ; preds = %for.cond132.i
+  call void @llvm.lifetime.end.p0(i64 128, ptr nonnull %tC.i) #4
+  call void @__itt_offload_wi_finish_wrapper()
+  ret void
+}
+
+; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
+declare void @llvm.lifetime.start.p0(i64 immarg, ptr nocapture) #1
+
+; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
+declare void @llvm.lifetime.end.p0(i64 immarg, ptr nocapture) #1
+
+; Function Attrs: nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: write)
+declare void @llvm.assume(i1 noundef) #2
+
+; Function Attrs: convergent nounwind
+declare dso_local spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 8, 16, 2) @_Z26__spirv_CompositeConstructIffLm8ELm16ELN5__spv9MatrixUseE2ELNS0_12MatrixLayoutE3ELNS0_5Scope4FlagE3EEPNS0_24__spirv_CooperativeMatrixKHRIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEET_(float noundef) local_unnamed_addr #3
+
+; Function Attrs: convergent nounwind
+declare dso_local spir_func noundef target("spirv.CooperativeMatrixKHR", i16, 3, 8, 16, 0) @_Z32__spirv_CooperativeMatrixLoadKHRIU3AS1N4sycl3_V13ext6oneapi8bfloat16ES4_Lm8ELm16ELN5__spv9MatrixUseE0ELNS6_12MatrixLayoutE0ELNS6_5Scope4FlagE3EEPNS6_24__spirv_CooperativeMatrixKHRIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEPT_mS8_SA_i(ptr addrspace(1) noundef, i64 noundef, i32 noundef, i32 noundef, i32 noundef) local_unnamed_addr #3
+
+; Function Attrs: convergent nounwind
+declare dso_local spir_func noundef target("spirv.CooperativeMatrixKHR", i16, 3, 16, 16, 1) @_Z32__spirv_CooperativeMatrixLoadKHRIU3AS1N4sycl3_V13ext6oneapi8bfloat16ES4_Lm16ELm16ELN5__spv9MatrixUseE1ELNS6_12MatrixLayoutE2ELNS6_5Scope4FlagE3EEPNS6_24__spirv_CooperativeMatrixKHRIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEPT_mS8_SA_i(ptr addrspace(1) noundef, i64 noundef, i32 noundef, i32 noundef, i32 noundef) local_unnamed_addr #3
+
+; Function Attrs: convergent nounwind
+declare dso_local spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 8, 16, 2) @_Z31__spirv_CooperativeMatrixMadKHRIN4sycl3_V13ext6oneapi8bfloat16EfLm8ELm16ELm16ELN5__spv9MatrixUseE0ELS6_1ELS6_2ELNS5_12MatrixLayoutE0ELS7_2ELS7_3ELNS5_5Scope4FlagE3EEPNS5_24__spirv_CooperativeMatrixKHRIT0_XT1_EXT3_EXT9_EXT10_EXT6_EEEPNSA_IT_XT1_EXT2_EXT7_EXT10_EXT4_EEEPNSA_ISE_XT2_EXT3_EXT8_EXT10_EXT5_EEESD_S9_(target("spirv.CooperativeMatrixKHR", i16, 3, 8, 16, 0) noundef, target("spirv.CooperativeMatrixKHR", i16, 3, 16, 16, 1) noundef, target("spirv.CooperativeMatrixKHR", float, 3, 8, 16, 2) noundef, i32 noundef) local_unnamed_addr #3
+
+; Function Attrs: convergent nounwind
+declare dso_local spir_func void @_Z33__spirv_CooperativeMatrixStoreKHRIU3AS1ffLm8ELm16ELN5__spv9MatrixUseE2ELNS1_12MatrixLayoutE3ELNS1_5Scope4FlagE3EEvPT_PNS1_24__spirv_CooperativeMatrixKHRIT0_XT1_EXT2_EXT4_EXT5_EXT3_EEEmS3_S5_i(ptr addrspace(1) noundef, target("spirv.CooperativeMatrixKHR", float, 3, 8, 16, 2) noundef, i64 noundef, i32 noundef, i32 noundef, i32 noundef) local_unnamed_addr #3
+
+declare dso_local spir_func i32 @_Z18__spirv_ocl_printfPU3AS2Kcz(ptr addrspace(2), ...)
+
+declare void @__itt_offload_wi_start_wrapper()
+
+declare void @__itt_offload_wi_finish_wrapper()
+
+attributes #0 = { convergent norecurse nounwind "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "sycl-module-id"="llvm/sycl/test-e2e/Matrix/joint_matrix_bf16_fill_k_cache.cpp" "sycl-optlevel"="2" "uniform-work-group-size"="true" }
+attributes #1 = { nocallback nofree nosync nounwind willreturn memory(argmem: readwrite) }
+attributes #2 = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: write) }
+attributes #3 = { convergent nounwind "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #4 = { nounwind }
+attributes #5 = { convergent nounwind }
+
+!llvm.module.flags = !{!0, !1}
+!opencl.spir.version = !{!2}
+!spirv.Source = !{!3}
+!sycl_aspects = !{!4, !5, !6, !7, !8, !9, !10, !11, !12, !13, !14, !15, !16, !17, !18, !19, !20, !21, !22, !23, !24, !25, !26, !27, !28, !29, !30, !31, !32, !33, !34, !35, !36, !37, !38, !39, !40, !41, !42, !43, !44, !45, !46, !47, !48, !49, !50, !51, !52, !53, !54, !55, !56, !57}
+!llvm.ident = !{!58}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{i32 7, !"frame-pointer", i32 2}
+!2 = !{i32 1, i32 2}
+!3 = !{i32 4, i32 100000}
+!4 = !{!"cpu", i32 1}
+!5 = !{!"gpu", i32 2}
+!6 = !{!"accelerator", i32 3}
+!7 = !{!"custom", i32 4}
+!8 = !{!"fp16", i32 5}
+!9 = !{!"fp64", i32 6}
+!10 = !{!"image", i32 9}
+!11 = !{!"online_compiler", i32 10}
+!12 = !{!"online_linker", i32 11}
+!13 = !{!"queue_profiling", i32 12}
+!14 = !{!"usm_device_allocations", i32 13}
+!15 = !{!"usm_host_allocations", i32 14}
+!16 = !{!"usm_shared_allocations", i32 15}
+!17 = !{!"usm_system_allocations", i32 17}
+!18 = !{!"ext_intel_pci_address", i32 18}
+!19 = !{!"ext_intel_gpu_eu_count", i32 19}
+!20 = !{!"ext_intel_gpu_eu_simd_width", i32 20}
+!21 = !{!"ext_intel_gpu_slices", i32 21}
+!22 = !{!"ext_intel_gpu_subslices_per_slice", i32 22}
+!23 = !{!"ext_intel_gpu_eu_count_per_subslice", i32 23}
+!24 = !{!"ext_intel_max_mem_bandwidth", i32 24}
+!25 = !{!"ext_intel_mem_channel", i32 25}
+!26 = !{!"usm_atomic_host_allocations", i32 26}
+!27 = !{!"usm_atomic_shared_allocations", i32 27}
+!28 = !{!"atomic64", i32 28}
+!29 = !{!"ext_intel_device_info_uuid", i32 29}
+!30 = !{!"ext_oneapi_srgb", i32 30}
+!31 = !{!"ext_oneapi_native_assert", i32 31}
+!32 = !{!"host_debuggable", i32 32}
+!33 = !{!"ext_intel_gpu_hw_threads_per_eu", i32 33}
+!34 = !{!"ext_oneapi_cuda_async_barrier", i32 34}
+!35 = !{!"ext_oneapi_bfloat16_math_functions", i32 35}
+!36 = !{!"ext_intel_free_memory", i32 36}
+!37 = !{!"ext_intel_device_id", i32 37}
+!38 = !{!"ext_intel_memory_clock_rate", i32 38}
+!39 = !{!"ext_intel_memory_bus_width", i32 39}
+!40 = !{!"emulated", i32 40}
+!41 = !{!"ext_intel_legacy_image", i32 41}
+!42 = !{!"ext_oneapi_bindless_images", i32 42}
+!43 = !{!"ext_oneapi_bindless_images_shared_usm", i32 43}
+!44 = !{!"ext_oneapi_bindless_images_1d_usm", i32 44}
+!45 = !{!"ext_oneapi_bindless_images_2d_usm", i32 45}
+!46 = !{!"ext_oneapi_interop_memory_import", i32 46}
+!47 = !{!"ext_oneapi_interop_memory_export", i32 47}
+!48 = !{!"ext_oneapi_interop_semaphore_import", i32 48}
+!49 = !{!"ext_oneapi_interop_semaphore_export", i32 49}
+!50 = !{!"ext_oneapi_mipmap", i32 50}
+!51 = !{!"ext_oneapi_mipmap_anisotropy", i32 51}
+!52 = !{!"ext_oneapi_mipmap_level_reference", i32 52}
+!53 = !{!"int64_base_atomics", i32 7}
+!54 = !{!"int64_extended_atomics", i32 8}
+!55 = !{!"usm_system_allocator", i32 17}
+!56 = !{!"usm_restricted_shared_allocations", i32 16}
+!57 = !{!"host", i32 0}
+!58 = !{!"clang version 18.0.0 (https://github.com/intel/llvm.git cc440821c30daabef517c7c8ff75546719f8094c)"}
+!59 = !{i32 242145}
+!60 = !{i32 -1, i32 -1, i32 -1}
+!61 = !{i32 16}
+!62 = !{}
+!63 = !{i1 false, i1 false, i1 false}
+!64 = !{!65, !65, i64 0}
+!65 = !{!"any pointer", !66, i64 0}
+!66 = !{!"omnipotent char", !67, i64 0}
+!67 = !{!"Simple C++ TBAA"}
+!68 = !{!69, !71, !73}
+!69 = distinct !{!69, !70, !"_ZN7__spirv22InitSizesSTWorkgroupIdILi2EN4sycl3_V12idILi2EEEE8initSizeEv: %agg.result"}
+!70 = distinct !{!70, !"_ZN7__spirv22InitSizesSTWorkgroupIdILi2EN4sycl3_V12idILi2EEEE8initSizeEv"}
+!71 = distinct !{!71, !72, !"_ZN7__spirv15initWorkgroupIdILi2EN4sycl3_V12idILi2EEEEET0_v: %agg.result"}
+!72 = distinct !{!72, !"_ZN7__spirv15initWorkgroupIdILi2EN4sycl3_V12idILi2EEEEET0_v"}
+!73 = distinct !{!73, !74, !"_ZN4sycl3_V16detail7Builder10getElementILi2EEEKNS0_7nd_itemIXT_EEEPS5_: %agg.result"}
+!74 = distinct !{!74, !"_ZN4sycl3_V16detail7Builder10getElementILi2EEEKNS0_7nd_itemIXT_EEEPS5_"}
+!75 = !{!76, !78, !73}
+!76 = distinct !{!76, !77, !"_ZN7__spirv28InitSizesSTLocalInvocationIdILi2EN4sycl3_V12idILi2EEEE8initSizeEv: %agg.result"}
+!77 = distinct !{!77, !"_ZN7__spirv28InitSizesSTLocalInvocationIdILi2EN4sycl3_V12idILi2EEEE8initSizeEv"}
+!78 = distinct !{!78, !79, !"_ZN7__spirv21initLocalInvocationIdILi2EN4sycl3_V12idILi2EEEEET0_v: %agg.result"}
+!79 = distinct !{!79, !"_ZN7__spirv21initLocalInvocationIdILi2EN4sycl3_V12idILi2EEEEET0_v"}
+!80 = distinct !{!80, !81}
+!81 = !{!"llvm.loop.mustprogress"}
+!82 = !{!83, !65, i64 0}
+!83 = !{!"_ZTSN4sycl3_V13ext6oneapi12experimental6matrix12joint_matrixINS0_9sub_groupEfLNS4_3useE2ELm8ELm16ELNS4_6layoutE3EEE", !65, i64 0}
+!84 = distinct !{!84, !81}
+!85 = distinct !{!85, !81}
+!86 = !{!87, !65, i64 0}
+!87 = !{!"_ZTSN4sycl3_V13ext6oneapi12experimental6matrix12joint_matrixINS0_9sub_groupENS2_8bfloat16ELNS4_3useE0ELm8ELm16ELNS4_6layoutE0EEE", !65, i64 0}
+!88 = distinct !{!88, !81}
+!89 = !{!90, !65, i64 0}
+!90 = !{!"_ZTSN4sycl3_V13ext6oneapi12experimental6matrix12joint_matrixINS0_9sub_groupENS2_8bfloat16ELNS4_3useE1ELm16ELm16ELNS4_6layoutE2EEE", !65, i64 0}
+!91 = distinct !{!91, !81}
+!92 = !{!93}
+!93 = distinct !{!93, !94, !"_ZN4sycl3_V13ext6oneapi12experimental6matrix16joint_matrix_madINS0_9sub_groupENS2_8bfloat16ES7_fLm8ELm16ELm16ELNS4_6layoutE0ELS8_2EEENS4_12joint_matrixIT_T2_LNS4_3useE2EXT3_EXT5_ELS8_3EEESA_RNS9_ISA_T0_LSC_0EXT3_EXT4_EXT6_EEERNS9_ISA_T1_LSC_1EXT4_EXT5_EXT7_EEERSD_: %agg.result"}
+!94 = distinct !{!94, !"_ZN4sycl3_V13ext6oneapi12experimental6matrix16joint_matrix_madINS0_9sub_groupENS2_8bfloat16ES7_fLm8ELm16ELm16ELNS4_6layoutE0ELS8_2EEENS4_12joint_matrixIT_T2_LNS4_3useE2EXT3_EXT5_ELS8_3EEESA_RNS9_ISA_T0_LSC_0EXT3_EXT4_EXT6_EEERNS9_ISA_T1_LSC_1EXT4_EXT5_EXT7_EEERSD_"}
+!95 = distinct !{!95, !81}
+!96 = distinct !{!96, !81}
+!97 = distinct !{!97, !81}
+!98 = distinct !{!98, !81}
+!99 = distinct !{!99, !81}
diff --git a/test/extensions/KHR/SPV_KHR_cooperative_matrix/conversion_instructions.ll b/test/extensions/KHR/SPV_KHR_cooperative_matrix/conversion_instructions.ll
new file mode 100644
index 0000000000..39bc1763e9
--- /dev/null
+++ b/test/extensions/KHR/SPV_KHR_cooperative_matrix/conversion_instructions.ll
@@ -0,0 +1,149 @@
+; RUN: llvm-as < %s -o %t.bc
+; RUN: llvm-spirv %t.bc --spirv-ext=+SPV_KHR_cooperative_matrix -o %t.spv
+; TODO: Validation is disabled till the moment the tools in CI are updated (passes locally)
+; R/UN: spirv-val %t.spv
+; RUN: llvm-spirv %t.spv -to-text -o %t.spt
+; RUN: FileCheck < %t.spt %s --check-prefix=CHECK-SPIRV
+
+; RUN: llvm-spirv -r --spirv-target-env=SPV-IR %t.spv -o %t.rev.bc
+; RUN: llvm-dis %t.rev.bc 
+; RUN: FileCheck < %t.rev.ll %s --check-prefix=CHECK-LLVM
+
+; CHECK-SPIRV: TypeInt [[#TypeInt32:]] 32 0
+; CHECK-SPIRV: TypeInt [[#TypeInt16:]] 16 0
+; CHECK-SPIRV: TypeInt [[#TypeInt8:]] 8 0
+; CHECK-SPIRV: TypeFloat [[#TypeFloat:]] 32
+; CHECK-SPIRV: TypeCooperativeMatrixKHR [[#MatrixTypeFloat:]] [[#TypeFloat]]
+; CHECK-SPIRV: TypeCooperativeMatrixKHR [[#MatrixTypeInt32:]] [[#TypeInt32]]
+; CHECK-SPIRV: TypeCooperativeMatrixKHR [[#MatrixTypeInt16:]] [[#TypeInt16]]
+; CHECK-SPIRV: TypeFloat [[#TypeFloat16:]] 16
+; CHECK-SPIRV: TypeCooperativeMatrixKHR [[#MatrixTypeFloat16:]] [[#TypeFloat16]]
+; CHECK-SPIRV: TypeCooperativeMatrixKHR [[#MatrixTypeInt8:]] [[#TypeInt8]]
+
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "spir64-unknown-unknown"
+
+; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeFloat]] [[#MatrixIn:]] [[#]] {{$}}
+; CHECK-SPIRV: ConvertFToU [[#MatrixTypeInt32]] [[#]] [[#MatrixIn]]
+
+; CHECK-LLVM: %[[#Matrix:]] = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructf(float 0.000000e+00)
+; CHECK-LLVM: call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z77__spirv_ConvertFToU_RPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_2_rtpPU3AS145__spirv_CooperativeMatrixKHR__float_3_12_12_2(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %[[#Matrix]])
+
+define void @convert_f_to_u() {
+entry:
+  %0 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructFloat(float 0.000000e+00)
+  %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z77__spirv_ConvertFToU_RPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_2_rtpPU3AS145__spirv_CooperativeMatrixKHR__float_3_12_12_2(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %0)
+  ret void
+}
+
+; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeFloat]] [[#MatrixIn:]] [[#]] {{$}}
+; CHECK-SPIRV: ConvertFToS [[#MatrixTypeInt32]] [[#]] [[#MatrixIn]]
+
+; CHECK-LLVM: %[[#Matrix:]] = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructf(float 0.000000e+00)
+; CHECK-LLVM: call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z77__spirv_ConvertFToS_RPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_2_rtpPU3AS145__spirv_CooperativeMatrixKHR__float_3_12_12_2(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %[[#Matrix]])
+
+define void @convert_f_to_s() {
+entry:
+  %0 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructFloat(float 0.000000e+00)
+  %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z77__spirv_ConvertFToS_RPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_2_rtpPU3AS145__spirv_CooperativeMatrixKHR__float_3_12_12_2(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) %0)
+  ret void
+}
+
+; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeInt16]] [[#MatrixIn:]] [[#]] {{$}}
+; CHECK-SPIRV: ConvertSToF [[#MatrixTypeFloat16]] [[#]] [[#MatrixIn]]
+
+; CHECK-LLVM: %[[#Matrix:]] = call spir_func target("spirv.CooperativeMatrixKHR", i16, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructs(i16 0)
+; CHECK-LLVM: call spir_func target("spirv.CooperativeMatrixKHR", half, 3, 12, 12, 2) @_Z77__spirv_ConvertSToF_RPU3AS144__spirv_CooperativeMatrixKHR__half_3_12_12_2_rtpPU3AS145__spirv_CooperativeMatrixKHR__short_3_12_12_2(target("spirv.CooperativeMatrixKHR", i16, 3, 12, 12, 2) %[[#Matrix]])
+
+define void @convert_s_to_f() {
+entry:
+  %0 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i16, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructInt16(i16 0)
+  %call = call spir_func target("spirv.CooperativeMatrixKHR", half, 3, 12, 12, 2) @_Z77__spirv_ConvertSToF_RPU3AS144__spirv_CooperativeMatrixKHR__half_3_12_12_2_rtpPU3AS145__spirv_CooperativeMatrixKHR__short_3_12_12_2(target("spirv.CooperativeMatrixKHR", i16, 3, 12, 12, 2) %0)
+  ret void
+}
+
+; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeInt16]] [[#MatrixIn:]] [[#]] {{$}}
+; CHECK-SPIRV: ConvertUToF [[#MatrixTypeFloat16]] [[#]] [[#MatrixIn]]
+
+; CHECK-LLVM: %[[#Matrix:]] = call spir_func target("spirv.CooperativeMatrixKHR", i16, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructs(i16 0)
+; CHECK-LLVM: call spir_func target("spirv.CooperativeMatrixKHR", half, 3, 12, 12, 2) @_Z77__spirv_ConvertUToF_RPU3AS144__spirv_CooperativeMatrixKHR__half_3_12_12_2_rtpPU3AS145__spirv_CooperativeMatrixKHR__short_3_12_12_2(target("spirv.CooperativeMatrixKHR", i16, 3, 12, 12, 2) %[[#Matrix]])
+
+define void @convert_u_to_f() {
+entry:
+  %0 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i16, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructInt16(i16 0)
+  %call = call spir_func target("spirv.CooperativeMatrixKHR", half, 3, 12, 12, 2) @_Z77__spirv_ConvertUToF_RPU3AS144__spirv_CooperativeMatrixKHR__half_3_12_12_2_rtpPU3AS145__spirv_CooperativeMatrixKHR__short_3_12_12_2(target("spirv.CooperativeMatrixKHR", i16, 3, 12, 12, 2) %0)
+  ret void
+}
+
+; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeInt32]] [[#MatrixIn:]] [[#]] {{$}}
+; CHECK-SPIRV: UConvert [[#MatrixTypeInt8]] [[#]] [[#MatrixIn]]
+
+; CHECK-LLVM: %[[#Matrix:]] = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructi(i32 0)
+; CHECK-LLVM: call spir_func target("spirv.CooperativeMatrixKHR", i8, 3, 12, 12, 2) @_Z74__spirv_UConvert_RPU3AS144__spirv_CooperativeMatrixKHR__char_3_12_12_2_satPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_2(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %[[#Matrix]])
+
+define void @u_convert() {
+entry:
+  %0 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructInt32(i32 0)
+  %call = call spir_func target("spirv.CooperativeMatrixKHR", i8, 3, 12, 12, 2) @_Z74__spirv_UConvert_RPU3AS144__spirv_CooperativeMatrixKHR__char_3_12_12_2_satPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_2(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %0)
+  ret void
+}
+
+; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeInt8]] [[#MatrixIn:]] [[#]] {{$}}
+; CHECK-SPIRV: SConvert [[#MatrixTypeInt32]] [[#]] [[#MatrixIn]]
+
+; CHECK-LLVM: %[[#Matrix:]] = call spir_func target("spirv.CooperativeMatrixKHR", i8, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructc(i8 0)
+; CHECK-LLVM: call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z74__spirv_SConvert_RPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_2_satPU3AS144__spirv_CooperativeMatrixKHR__char_3_12_12_2(target("spirv.CooperativeMatrixKHR", i8, 3, 12, 12, 2) %[[#Matrix]])
+
+define void @s_convert() {
+entry:
+  %0 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i8, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructInt8(i8 0)
+  %call = call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z74__spirv_SConvert_RPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_2_satPU3AS144__spirv_CooperativeMatrixKHR__char_3_12_12_2(target("spirv.CooperativeMatrixKHR", i8, 3, 12, 12, 2) %0)
+  ret void
+}
+
+; CHECK-SPIRV: CompositeConstruct [[#MatrixTypeFloat16]] [[#MatrixIn:]] [[#]] {{$}}
+; CHECK-SPIRV: FConvert [[#MatrixTypeFloat]] [[#]] [[#MatrixIn]]
+
+; CHECK-LLVM: %[[#Matrix:]] = call spir_func target("spirv.CooperativeMatrixKHR", half, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructDh(half 0xH0000)
+; CHECK-LLVM: call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z75__spirv_FConvert_RPU3AS145__spirv_CooperativeMatrixKHR__float_3_12_12_2_satPU3AS144__spirv_CooperativeMatrixKHR__half_3_12_12_2(target("spirv.CooperativeMatrixKHR", half, 3, 12, 12, 2) %[[#Matrix]])
+
+define void @f_convert() {
+entry:
+  %0 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", half, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructHalf(half 0xH0000)
+  %call = call spir_func target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z75__spirv_FConvert_RPU3AS145__spirv_CooperativeMatrixKHR__float_3_12_12_2_satPU3AS144__spirv_CooperativeMatrixKHR__half_3_12_12_2(target("spirv.CooperativeMatrixKHR", half, 3, 12, 12, 2) %0)
+  ret void
+}
+
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructFloat(float noundef)
+
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", half, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructHalf(half noundef)
+
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructInt32(i32 noundef)
+
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", i16, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructInt16(i16 noundef)
+
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", i8, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructInt8(i8 noundef)
+
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z77__spirv_ConvertFToU_RPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_2_rtpPU3AS145__spirv_CooperativeMatrixKHR__float_3_12_12_2(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) noundef)
+
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z77__spirv_ConvertFToS_RPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_2_rtpPU3AS145__spirv_CooperativeMatrixKHR__float_3_12_12_2(target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) noundef)
+
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", half, 3, 12, 12, 2) @_Z77__spirv_ConvertSToF_RPU3AS144__spirv_CooperativeMatrixKHR__half_3_12_12_2_rtpPU3AS145__spirv_CooperativeMatrixKHR__short_3_12_12_2(target("spirv.CooperativeMatrixKHR", i16, 3, 12, 12, 2) noundef)
+
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", half, 3, 12, 12, 2) @_Z77__spirv_ConvertUToF_RPU3AS144__spirv_CooperativeMatrixKHR__half_3_12_12_2_rtpPU3AS145__spirv_CooperativeMatrixKHR__short_3_12_12_2(target("spirv.CooperativeMatrixKHR", i16, 3, 12, 12, 2) noundef)
+
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", i8, 3, 12, 12, 2) @_Z74__spirv_UConvert_RPU3AS144__spirv_CooperativeMatrixKHR__char_3_12_12_2_satPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_2(target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) noundef)
+
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z74__spirv_SConvert_RPU3AS144__spirv_CooperativeMatrixKHR__uint_3_12_12_2_satPU3AS144__spirv_CooperativeMatrixKHR__char_3_12_12_2(target("spirv.CooperativeMatrixKHR", i8, 3, 12, 12, 2) noundef)
+
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 2) @_Z75__spirv_FConvert_RPU3AS145__spirv_CooperativeMatrixKHR__float_3_12_12_2_satPU3AS144__spirv_CooperativeMatrixKHR__half_3_12_12_2(target("spirv.CooperativeMatrixKHR", half, 3, 12, 12, 2) noundef)
+
+!llvm.module.flags = !{!0, !1, !2, !3, !4}
+!llvm.ident = !{!5}
+
+!0 = !{i32 7, !"Dwarf Version", i32 4}
+!1 = !{i32 1, !"wchar_size", i32 4}
+!2 = !{i32 8, !"PIC Level", i32 2}
+!3 = !{i32 7, !"PIE Level", i32 2}
+!4 = !{i32 7, !"uwtable", i32 2}
+!5 = !{!"clang version 16.0.0 (https://github.com/llvm/llvm-project.git 08d094a0e457360ad8b94b017d2dc277e697ca76)"}
diff --git a/test/extensions/KHR/SPV_KHR_cooperative_matrix/cooperative_matrix.ll b/test/extensions/KHR/SPV_KHR_cooperative_matrix/cooperative_matrix.ll
index 9c7e9d9f91..18897f6203 100644
--- a/test/extensions/KHR/SPV_KHR_cooperative_matrix/cooperative_matrix.ll
+++ b/test/extensions/KHR/SPV_KHR_cooperative_matrix/cooperative_matrix.ll
@@ -18,9 +18,9 @@
 ; CHECK-SPIRV-DAG: Constant [[#Int32Ty]] [[#Const3:]] 3
 ; CHECK-SPIRV-DAG: Constant [[#Int32Ty]] [[#Const2:]] 2
 ; CHECK-SPIRV-DAG: Constant [[#Int32Ty]] [[#Const1:]] 1
-; CHECK-SPIRV-DAG: TypeCooperativeMatrixKHR [[#MatTy1:]] [[#Int32Ty]] [[#Const3]] [[#Const12]] [[#Const12]] [[#Const3]]
-; CHECK-SPIRV-DAG: TypeCooperativeMatrixKHR [[#MatTy2:]] [[#Int8Ty]] [[#Const0]] [[#Const12]] [[#Const48]] [[#Const3]]
-; CHECK-SPIRV-DAG: TypeCooperativeMatrixKHR [[#MatTy3:]] [[#Int8Ty]] [[#Const2]] [[#Const48]] [[#Const12]] [[#Const3]]
+; CHECK-SPIRV-DAG: TypeCooperativeMatrixKHR [[#MatTy1:]] [[#Int32Ty]] [[#Const3]] [[#Const12]] [[#Const12]] [[#Const2]]
+; CHECK-SPIRV-DAG: TypeCooperativeMatrixKHR [[#MatTy2:]] [[#Int8Ty]] [[#Const0]] [[#Const12]] [[#Const48]] [[#Const0]]
+; CHECK-SPIRV-DAG: TypeCooperativeMatrixKHR [[#MatTy3:]] [[#Int8Ty]] [[#Const2]] [[#Const48]] [[#Const12]] [[#Const1]]
 ; CHECK-SPIRV: CompositeConstruct [[#MatTy1]]
 ; CHECK-SPIRV: CooperativeMatrixLoadKHR [[#MatTy2]]
 ; CHECK-SPIRV: CooperativeMatrixLengthKHR [[#Int32Ty]] [[#]] [[#MatTy2]]
@@ -29,12 +29,13 @@
 ; CHECK-SPIRV: CooperativeMatrixStoreKHR
 
 
-; CHECK-LLVM: call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructi(i32 0)
-; CHECK-LLVM: call spir_func target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 3) @_Z86__spirv_CooperativeMatrixLoadKHR_RPU3AS144__spirv_CooperativeMatrixKHR__char_0_12_48_3PU3AS4clii
-; CHECK-LLVM: call spir_func i32 @_Z34__spirv_CooperativeMatrixLengthKHRPU3AS144__spirv_CooperativeMatrixKHR__char_0_12_48_3(target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 3)
-; CHECK-LLVM: call spir_func target("spirv.CooperativeMatrixKHR", i8, 2, 48, 12, 3) @_Z86__spirv_CooperativeMatrixLoadKHR_RPU3AS144__spirv_CooperativeMatrixKHR__char_2_48_12_3PU3AS4cl
-; CHECK-LLVM: call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z34__spirv_CooperativeMatrixMulAddKHR{{.*}}(target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 3) %{{.*}}, target("spirv.CooperativeMatrixKHR", i8, 2, 48, 12, 3) %{{.*}}, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3)
-; CHECK-LLVM: call spir_func void @_Z33__spirv_CooperativeMatrixStoreKHR{{.*}}(ptr addrspace(4) %call.ascast.i.i, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3)
+; CHECK-LLVM: call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z26__spirv_CompositeConstructi(i32 0)
+; CHECK-LLVM: call spir_func target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 0) @_Z86__spirv_CooperativeMatrixLoadKHR_RPU3AS144__spirv_CooperativeMatrixKHR__char_0_12_48_0PU3AS4clii
+; CHECK-LLVM: call spir_func i32 @_Z34__spirv_CooperativeMatrixLengthKHRPU3AS144__spirv_CooperativeMatrixKHR__char_0_12_48_0(target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 0)
+; CHECK-LLVM: call spir_func target("spirv.CooperativeMatrixKHR", i8, 2, 48, 12, 1) @_Z86__spirv_CooperativeMatrixLoadKHR_RPU3AS144__spirv_CooperativeMatrixKHR__char_2_48_12_1PU3AS4cl
+; CHECK-LLVM: call spir_func target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z34__spirv_CooperativeMatrixMulAddKHR{{.*}}(target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 0) %{{.*}}, target("spirv.CooperativeMatrixKHR", i8, 2, 48, 12, 1) %{{.*}}, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2)
+; CHECK-LLVM: call spir_func void @_Z33__spirv_CooperativeMatrixStoreKHR{{.*}}(ptr addrspace(4) %call.ascast.i.i, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2)
+
 
 ; ModuleID = 'test-matrix-opaque.bc'
 source_filename = "matrix-int8-test.cpp"
@@ -53,8 +54,8 @@ $_ZTSZZ15matrix_multiply = comdat any
 ; Function Attrs: convergent norecurse
 define weak_odr dso_local spir_kernel void @_ZTSZZ15matrix_multiply(ptr addrspace(1) noundef align 1 %_arg_accA, ptr addrspace(1) noundef align 1 %_arg_accB, ptr noundef byval(%"class.sycl::_V1::range") align 8 %_arg_accB5, ptr noundef byval(%"class.sycl::_V1::id") align 8 %_arg_accB6, ptr addrspace(1) noundef align 4 %_arg_accC, i64 noundef %_arg_N, i64 noundef %_arg_K) local_unnamed_addr #0 comdat {
 entry:
-  %sub_c.sroa.0.i = alloca target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3), align 8
-  %ref.tmp29.sroa.0.i = alloca target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3), align 8
+  %sub_c.sroa.0.i = alloca target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2), align 8
+  %ref.tmp29.sroa.0.i = alloca target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2), align 8
   %agg.tmp15.sroa.0.sroa.2.0..sroa_idx = getelementptr inbounds %"class.sycl::_V1::range", ptr %_arg_accB5, i64 0, i32 0, i32 0, i64 1
   %agg.tmp15.sroa.0.sroa.2.0.copyload = load i64, ptr %agg.tmp15.sroa.0.sroa.2.0..sroa_idx, align 8
   %0 = getelementptr inbounds %"class.sycl::_V1::id", ptr %_arg_accB6, i64 0, i32 0, i32 0, i64 0
@@ -78,8 +79,8 @@ entry:
   %sub5.i = sub nsw i64 %3, %6
   %sub_c.sroa.0.i.0.i.0..sroa_cast = bitcast ptr %sub_c.sroa.0.i to ptr
   call void @llvm.lifetime.start.p0(i64 8, ptr nonnull %sub_c.sroa.0.i.0.i.0..sroa_cast)
-  %call.i.i = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z26__spirv_CompositeConstruct(i32 noundef 0) #4
-  store target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %call.i.i, ptr %sub_c.sroa.0.i, align 8
+  %call.i.i = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z26__spirv_CompositeConstruct(i32 noundef 0) #4
+  store target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %call.i.i, ptr %sub_c.sroa.0.i, align 8
   %mul.i = mul nsw i64 %sub.i, 12
   %div2452.i = lshr i64 %sub5.i, 4
   %mul26.i = mul i64 %div2452.i, 48
@@ -106,18 +107,18 @@ for.body.i:                                       ; preds = %for.cond.i
   %conv13.i = zext i32 %mul12.i to i64
   %add.ptr.i96.i = getelementptr inbounds i8, ptr addrspace(1) %add.ptr.i93.i, i64 %conv13.i
   %call.ascast.i66.i = addrspacecast ptr addrspace(1) %add.ptr.i96.i to ptr addrspace(4)
-  %call1.i.i = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 3) @_Z32__spirv_CooperativeMatrixLoadKHR_1(ptr addrspace(4) noundef %call.ascast.i66.i, i64 noundef %_arg_K, i32 noundef 0, i32 noundef 1) #4
-  %len = tail call spir_func noundef i32 @_Z34__spirv_CooperativeMatrixLengthKHR(target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 3) %call1.i.i)
+  %call1.i.i = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 0) @_Z32__spirv_CooperativeMatrixLoadKHR_1(ptr addrspace(4) noundef %call.ascast.i66.i, i64 noundef %_arg_K, i32 noundef 0, i32 noundef 1) #4
+  %len = tail call spir_func noundef i32 @_Z34__spirv_CooperativeMatrixLengthKHR(target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 0) %call1.i.i)
   %div20.i = mul nsw i32 %k.0.i, 12
   %conv21.i = zext i32 %div20.i to i64
   %mul23.i = mul i64 %mul22.i, %conv21.i
   %add.ptr.i111.i = getelementptr i8, ptr addrspace(1) %add.ptr.i108140.i, i64 %mul23.i
   %call.ascast.i72.i = addrspacecast ptr addrspace(1) %add.ptr.i111.i to ptr addrspace(4)
-  %call1.i73.i = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i8, 2, 48, 12, 3) @_Z32__spirv_CooperativeMatrixLoadKHR_2(ptr addrspace(4) noundef %call.ascast.i72.i, i64 noundef %mul22.i) #4
+  %call1.i73.i = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i8, 2, 48, 12, 1) @_Z32__spirv_CooperativeMatrixLoadKHR_2(ptr addrspace(4) noundef %call.ascast.i72.i, i64 noundef %mul22.i) #4
   call void @llvm.lifetime.start.p0(i64 8, ptr nonnull %ref.tmp29.sroa.0.i.0.i.0..sroa_cast)
-  %sub_c.sroa.0.i.0.sub_c.sroa.0.i.0.sub_c.sroa.0.0.sub_c.sroa.0.0.sub_c.sroa.0.0.125.i = load target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3), ptr %sub_c.sroa.0.i, align 8
-  %call.i77.i = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z34__spirv_CooperativeMatrixMulAddKHR(target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 3) noundef %call1.i.i, target("spirv.CooperativeMatrixKHR", i8, 2, 48, 12, 3) noundef %call1.i73.i, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) noundef %sub_c.sroa.0.i.0.sub_c.sroa.0.i.0.sub_c.sroa.0.0.sub_c.sroa.0.0.sub_c.sroa.0.0.125.i, i32 noundef 12) #4
-  store target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) %call.i77.i, ptr %ref.tmp29.sroa.0.i, align 8
+  %sub_c.sroa.0.i.0.sub_c.sroa.0.i.0.sub_c.sroa.0.0.sub_c.sroa.0.0.sub_c.sroa.0.0.125.i = load target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2), ptr %sub_c.sroa.0.i, align 8
+  %call.i77.i = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z34__spirv_CooperativeMatrixMulAddKHR(target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 0) noundef %call1.i.i, target("spirv.CooperativeMatrixKHR", i8, 2, 48, 12, 1) noundef %call1.i73.i, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) noundef %sub_c.sroa.0.i.0.sub_c.sroa.0.i.0.sub_c.sroa.0.0.sub_c.sroa.0.0.sub_c.sroa.0.0.125.i, i32 noundef 12) #4
+  store target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) %call.i77.i, ptr %ref.tmp29.sroa.0.i, align 8
   %ref.tmp29.sroa.0.i.0.ref.tmp29.sroa.0.i.0.ref.tmp29.sroa.0.0.ref.tmp29.sroa.0.0.ref.tmp29.sroa.0.0..i = load i64, ptr %7, align 8
   store i64 %ref.tmp29.sroa.0.i.0.ref.tmp29.sroa.0.i.0.ref.tmp29.sroa.0.0.ref.tmp29.sroa.0.0.ref.tmp29.sroa.0.0..i, ptr %8, align 8
   call void @llvm.lifetime.end.p0(i64 8, ptr nonnull %ref.tmp29.sroa.0.i.0.i.0..sroa_cast)
@@ -130,28 +131,28 @@ _ZZZ15matrix_multiplyIiaLm24ELm96ELm24ELm96ELm24ELm24EEvR10big_matrixIT_XT5_EXT6
   %mul39.i = mul nuw i64 %div2452.i, 12
   %add.ptr.i81.i = getelementptr inbounds i32, ptr addrspace(1) %add.ptr.i.i, i64 %mul39.i
   %call.ascast.i.i = addrspacecast ptr addrspace(1) %add.ptr.i81.i to ptr addrspace(4)
-  %sub_c.sroa.0.i.0.sub_c.sroa.0.i.0.sub_c.sroa.0.0.sub_c.sroa.0.0.sub_c.sroa.0.0..i = load target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3), ptr %sub_c.sroa.0.i, align 8
-  tail call spir_func void @_Z33__spirv_CooperativeMatrixStoreKHR(ptr addrspace(4) noundef %call.ascast.i.i, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) noundef %sub_c.sroa.0.i.0.sub_c.sroa.0.i.0.sub_c.sroa.0.0.sub_c.sroa.0.0.sub_c.sroa.0.0..i, i32 noundef 0, i64 noundef %_arg_N, i32 noundef 1) #4
+  %sub_c.sroa.0.i.0.sub_c.sroa.0.i.0.sub_c.sroa.0.0.sub_c.sroa.0.0.sub_c.sroa.0.0..i = load target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2), ptr %sub_c.sroa.0.i, align 8
+  tail call spir_func void @_Z33__spirv_CooperativeMatrixStoreKHR(ptr addrspace(4) noundef %call.ascast.i.i, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) noundef %sub_c.sroa.0.i.0.sub_c.sroa.0.i.0.sub_c.sroa.0.0.sub_c.sroa.0.0.sub_c.sroa.0.0..i, i32 noundef 0, i64 noundef %_arg_N, i32 noundef 1) #4
   call void @llvm.lifetime.end.p0(i64 8, ptr nonnull %sub_c.sroa.0.i.0.i.0..sroa_cast)
   ret void
 }
 
 ; Function Attrs: convergent
-declare dso_local spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z26__spirv_CompositeConstruct(i32 noundef) local_unnamed_addr #2
+declare dso_local spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z26__spirv_CompositeConstruct(i32 noundef) local_unnamed_addr #2
 
-declare dso_local spir_func noundef i32 @_Z34__spirv_CooperativeMatrixLengthKHR(target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 3) noundef)
+declare dso_local spir_func noundef i32 @_Z34__spirv_CooperativeMatrixLengthKHR(target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 0) noundef)
 
 ; Function Attrs: convergent
-declare dso_local spir_func noundef target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 3) @_Z32__spirv_CooperativeMatrixLoadKHR_1(ptr addrspace(4) noundef, i64 noundef, i32 noundef, i32 noundef) local_unnamed_addr #2
+declare dso_local spir_func noundef target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 0) @_Z32__spirv_CooperativeMatrixLoadKHR_1(ptr addrspace(4) noundef, i64 noundef, i32 noundef, i32 noundef) local_unnamed_addr #2
 
 ; Function Attrs: convergent
-declare dso_local spir_func noundef target("spirv.CooperativeMatrixKHR", i8, 2, 48, 12, 3) @_Z32__spirv_CooperativeMatrixLoadKHR_2(ptr addrspace(4) noundef, i64 noundef) local_unnamed_addr #2
+declare dso_local spir_func noundef target("spirv.CooperativeMatrixKHR", i8, 2, 48, 12, 1) @_Z32__spirv_CooperativeMatrixLoadKHR_2(ptr addrspace(4) noundef, i64 noundef) local_unnamed_addr #2
 
 ; Function Attrs: convergent
-declare dso_local spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) @_Z34__spirv_CooperativeMatrixMulAddKHR(target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 3) noundef, target("spirv.CooperativeMatrixKHR", i8, 2, 48, 12, 3) noundef, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) noundef, i32 noundef) local_unnamed_addr #2
+declare dso_local spir_func noundef target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) @_Z34__spirv_CooperativeMatrixMulAddKHR(target("spirv.CooperativeMatrixKHR", i8, 0, 12, 48, 0) noundef, target("spirv.CooperativeMatrixKHR", i8, 2, 48, 12, 1) noundef, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) noundef, i32 noundef) local_unnamed_addr #2
 
 ; Function Attrs: convergent
-declare dso_local spir_func void @_Z33__spirv_CooperativeMatrixStoreKHR(ptr addrspace(4) noundef, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 3) noundef, i32 noundef, i64 noundef, i32 noundef) local_unnamed_addr #2
+declare dso_local spir_func void @_Z33__spirv_CooperativeMatrixStoreKHR(ptr addrspace(4) noundef, target("spirv.CooperativeMatrixKHR", i32, 3, 12, 12, 2) noundef, i32 noundef, i64 noundef, i32 noundef) local_unnamed_addr #2
 
 ; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
 declare void @llvm.lifetime.start.p0(i64 immarg, ptr nocapture) #3
diff --git a/test/extensions/KHR/SPV_KHR_cooperative_matrix/cooperative_matrix_wrong_scope.ll b/test/extensions/KHR/SPV_KHR_cooperative_matrix/cooperative_matrix_wrong_scope.ll
new file mode 100644
index 0000000000..ef25389724
--- /dev/null
+++ b/test/extensions/KHR/SPV_KHR_cooperative_matrix/cooperative_matrix_wrong_scope.ll
@@ -0,0 +1,17 @@
+; RUN: llvm-as < %s -o %t.bc
+; RUN: not llvm-spirv %t.bc --spirv-ext=+SPV_KHR_cooperative_matrix -o %t.spv 2>&1 | FileCheck %s
+
+; CHECK: InvalidInstruction: Can't translate llvm instruction:
+; CHECK: TypeCooperativeMatrixKHR
+; CHECK: Unsupported Scope parameter
+
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "spir64-unknown-unknown"
+
+define void @convert_f_to_u() {
+entry:
+  %0 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 8, 12, 12, 2) @_Z26__spirv_CompositeConstructFloat(float 0.000000e+00)
+  ret void
+}
+
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", float, 8, 12, 12, 2) @_Z26__spirv_CompositeConstructFloat(float noundef)
diff --git a/test/extensions/KHR/SPV_KHR_cooperative_matrix/cooperative_matrix_wrong_use.ll b/test/extensions/KHR/SPV_KHR_cooperative_matrix/cooperative_matrix_wrong_use.ll
new file mode 100644
index 0000000000..0016f888cb
--- /dev/null
+++ b/test/extensions/KHR/SPV_KHR_cooperative_matrix/cooperative_matrix_wrong_use.ll
@@ -0,0 +1,17 @@
+; RUN: llvm-as < %s -o %t.bc
+; RUN: not llvm-spirv %t.bc --spirv-ext=+SPV_KHR_cooperative_matrix -o %t.spv 2>&1 | FileCheck %s
+
+; CHECK: InvalidInstruction: Can't translate llvm instruction:
+; CHECK: TypeCooperativeMatrixKHR
+; CHECK: Incorrect Use parameter, should be MatrixA, MatrixB or Accumulator
+
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "spir64-unknown-unknown"
+
+define void @convert_f_to_u() {
+entry:
+  %0 = tail call spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructFloat(float 0.000000e+00)
+  ret void
+}
+
+declare spir_func noundef target("spirv.CooperativeMatrixKHR", float, 3, 12, 12, 3) @_Z26__spirv_CompositeConstructFloat(float noundef)
diff --git a/test/transcoding/spirv-target-types-buffer.ll b/test/transcoding/spirv-target-types-buffer.ll
new file mode 100644
index 0000000000..4de5ad32fb
--- /dev/null
+++ b/test/transcoding/spirv-target-types-buffer.ll
@@ -0,0 +1,23 @@
+; Check translation of the buffer surface target extension type
+;
+; RUN: llvm-as %s -o %t.bc
+; RUN: llvm-spirv -spirv-ext=+SPV_INTEL_vector_compute %t.bc -spirv-text -o %t.spv.txt
+; RUN: FileCheck < %t.spv.txt %s --check-prefix=CHECK-SPIRV
+target datalayout = "e-p:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024"
+target triple = "spir-unknown-unknown"
+
+; CHECK-SPIRV: Capability VectorComputeINTEL
+; CHECK-SPIRV: Extension "SPV_INTEL_vector_compute"
+; CHECK-SPIRV: Name [[#FuncName:]] "foo"
+; CHECK-SPIRV: Name [[#ParamName:]] "a"
+; CHECK-SPIRV: TypeVoid  [[#VoidT:]]
+; CHECK-SPIRV: TypeBufferSurfaceINTEL [[#BufferID:]]
+; CHECK-SPIRV: Function [[#VoidT]] [[#FuncID:]]
+; CHECK-SPIRV-NEXT: FunctionParameter [[#BufferID]] [[#ParamName]]
+
+define spir_kernel void @foo(target("spirv.BufferSurfaceINTEL", 0) %a) #0 {
+  entry:
+  ret void
+ }
+
+attributes #0 = { noinline norecurse nounwind readnone "VCFunction"}
\ No newline at end of file