[OPENMP][NVPTX]Allow to use shared memory for the

target|teams|distribute variables. If the total size of the variables, declared in target|teams|distribute regions, is less than the maximal size of shared memory available, the buffer is allocated in the shared memory. llvm-svn: 346507
2018-11-09 16:18:04 +00:00 · 2018-11-09 16:18:04 +00:00 · 09c9eea78f
parent b2091c930b
commit 09c9eea78f
9 changed files with 143 additions and 98 deletions
--- a/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
@ -176,6 +176,9 @@ enum MachineConfiguration : unsigned {

  /// Global memory alignment for performance.
  GlobalMemoryAlignment = 128,
+
+  /// Maximal size of the shared memory buffer.
+  SharedMemorySize = 128,
 };

 enum NamedBarrier : unsigned {
@ -1143,13 +1146,6 @@ void CGOpenMPRuntimeNVPTX::emitNonSPMDKernel(const OMPExecutableDirective &D,
  IsInTTDRegion = true;
  // Reserve place for the globalized memory.
  GlobalizedRecords.emplace_back();
-  if (!StaticGlobalized) {
-    StaticGlobalized = new llvm::GlobalVariable(
-        CGM.getModule(), CGM.VoidPtrTy, /*isConstant=*/true,
-        llvm::GlobalValue::WeakAnyLinkage, nullptr,
-        "_openmp_static_glob_rd$ptr");
-    StaticGlobalized->setUnnamedAddr(llvm::GlobalValue::UnnamedAddr::Global);
-  }
  if (!KernelStaticGlobalized) {
    KernelStaticGlobalized = new llvm::GlobalVariable(
        CGM.getModule(), CGM.VoidPtrTy, /*isConstant=*/false,
@ -1277,13 +1273,6 @@ void CGOpenMPRuntimeNVPTX::emitSPMDKernel(const OMPExecutableDirective &D,
  IsInTTDRegion = true;
  // Reserve place for the globalized memory.
  GlobalizedRecords.emplace_back();
-  if (!StaticGlobalized) {
-    StaticGlobalized = new llvm::GlobalVariable(
-        CGM.getModule(), CGM.VoidPtrTy, /*isConstant=*/true,
-        llvm::GlobalValue::WeakAnyLinkage, nullptr,
-        "_openmp_static_glob_rd$ptr");
-    StaticGlobalized->setUnnamedAddr(llvm::GlobalValue::UnnamedAddr::Global);
-  }
  if (!KernelStaticGlobalized) {
    KernelStaticGlobalized = new llvm::GlobalVariable(
        CGM.getModule(), CGM.VoidPtrTy, /*isConstant=*/false,
@ -2138,30 +2127,41 @@ void CGOpenMPRuntimeNVPTX::emitGenericVarsProlog(CodeGenFunction &CGF,
      GlobalizedRecords.back().Records.push_back(GlobalizedVarsRecord);
      ++GlobalizedRecords.back().RegionCounter;
      if (GlobalizedRecords.back().Records.size() == 1) {
-        assert(StaticGlobalized &&
-               "Static pointer must be initialized already.");
-        Address Buffer = CGF.EmitLoadOfPointer(
-            Address(StaticGlobalized, CGM.getPointerAlign()),
-            CGM.getContext()
-                .getPointerType(CGM.getContext().VoidPtrTy)
-                .castAs<PointerType>());
+        assert(KernelStaticGlobalized &&
+               "Kernel static pointer must be initialized already.");
+        auto *UseSharedMemory = new llvm::GlobalVariable(
+            CGM.getModule(), CGM.Int16Ty, /*isConstant=*/true,
+            llvm::GlobalValue::InternalLinkage, nullptr,
+            "_openmp_static_kernel$is_shared");
+        UseSharedMemory->setUnnamedAddr(llvm::GlobalValue::UnnamedAddr::Global);
+        QualType Int16Ty = CGM.getContext().getIntTypeForBitwidth(
+            /*DestWidth=*/16, /*Signed=*/0);
+        llvm::Value *IsInSharedMemory = CGF.EmitLoadOfScalar(
+            Address(UseSharedMemory,
+                    CGM.getContext().getTypeAlignInChars(Int16Ty)),
+            /*Volatile=*/false, Int16Ty, Loc);
+        auto *StaticGlobalized = new llvm::GlobalVariable(
+            CGM.getModule(), CGM.Int8Ty, /*isConstant=*/false,
+            llvm::GlobalValue::WeakAnyLinkage, nullptr);
        auto *RecSize = new llvm::GlobalVariable(
            CGM.getModule(), CGM.SizeTy, /*isConstant=*/true,
            llvm::GlobalValue::InternalLinkage, nullptr,
            "_openmp_static_kernel$size");
        RecSize->setUnnamedAddr(llvm::GlobalValue::UnnamedAddr::Global);
        llvm::Value *Ld = CGF.EmitLoadOfScalar(
-            Address(RecSize, CGM.getPointerAlign()), /*Volatile=*/false,
+            Address(RecSize, CGM.getSizeAlign()), /*Volatile=*/false,
            CGM.getContext().getSizeType(), Loc);
        llvm::Value *ResAddr = Bld.CreatePointerBitCastOrAddrSpaceCast(
            KernelStaticGlobalized, CGM.VoidPtrPtrTy);
-        llvm::Value *GlobalRecordSizeArg[] = {
-            Buffer.getPointer(), Ld,
-            llvm::ConstantInt::getNullValue(CGM.Int16Ty), ResAddr};
+        llvm::Value *GlobalRecordSizeArg[] = {StaticGlobalized, Ld,
+                                              IsInSharedMemory, ResAddr};
        CGF.EmitRuntimeCall(createNVPTXRuntimeFunction(
                                OMPRTL_NVPTX__kmpc_get_team_static_memory),
                            GlobalRecordSizeArg);
+        GlobalizedRecords.back().Buffer = StaticGlobalized;
        GlobalizedRecords.back().RecSize = RecSize;
+        GlobalizedRecords.back().UseSharedMemory = UseSharedMemory;
+        GlobalizedRecords.back().Loc = Loc;
      }
      assert(KernelStaticGlobalized && "Global address must be set already.");
      Address FrameAddr = CGF.EmitLoadOfPointer(
@ -2336,10 +2336,16 @@ void CGOpenMPRuntimeNVPTX::emitGenericVarsEpilog(CodeGenFunction &CGF,
        --GlobalizedRecords.back().RegionCounter;
        // Emit the restore function only in the target region.
        if (GlobalizedRecords.back().RegionCounter == 0) {
+          QualType Int16Ty = CGM.getContext().getIntTypeForBitwidth(
+              /*DestWidth=*/16, /*Signed=*/0);
+          llvm::Value *IsInSharedMemory = CGF.EmitLoadOfScalar(
+              Address(GlobalizedRecords.back().UseSharedMemory,
+                      CGM.getContext().getTypeAlignInChars(Int16Ty)),
+              /*Volatile=*/false, Int16Ty, GlobalizedRecords.back().Loc);
          CGF.EmitRuntimeCall(
              createNVPTXRuntimeFunction(
                  OMPRTL_NVPTX__kmpc_restore_team_static_memory),
-              llvm::ConstantInt::getNullValue(CGM.Int16Ty));
+              IsInSharedMemory);
        }
      } else {
        CGF.EmitRuntimeCall(createNVPTXRuntimeFunction(
@ -4507,21 +4513,24 @@ static std::pair<unsigned, unsigned> getSMsBlocksPerSM(CodeGenModule &CGM) {
 void CGOpenMPRuntimeNVPTX::clear() {
  if (!GlobalizedRecords.empty()) {
    ASTContext &C = CGM.getContext();
+    llvm::SmallVector<const GlobalPtrSizeRecsTy *, 4> GlobalRecs;
+    llvm::SmallVector<const GlobalPtrSizeRecsTy *, 4> SharedRecs;
    RecordDecl *StaticRD = C.buildImplicitRecord(
        "_openmp_static_memory_type_$_", RecordDecl::TagKind::TTK_Union);
    StaticRD->startDefinition();
+    RecordDecl *SharedStaticRD = C.buildImplicitRecord(
+        "_shared_openmp_static_memory_type_$_", RecordDecl::TagKind::TTK_Union);
+    SharedStaticRD->startDefinition();
    for (const GlobalPtrSizeRecsTy &Records : GlobalizedRecords) {
      if (Records.Records.empty())
        continue;
      unsigned Size = 0;
      unsigned RecAlignment = 0;
      for (const RecordDecl *RD : Records.Records) {
-        QualType RDTy = CGM.getContext().getRecordType(RD);
-        unsigned Alignment =
-            CGM.getContext().getTypeAlignInChars(RDTy).getQuantity();
+        QualType RDTy = C.getRecordType(RD);
+        unsigned Alignment = C.getTypeAlignInChars(RDTy).getQuantity();
        RecAlignment = std::max(RecAlignment, Alignment);
-        unsigned RecSize =
-            CGM.getContext().getTypeSizeInChars(RDTy).getQuantity();
+        unsigned RecSize = C.getTypeSizeInChars(RDTy).getQuantity();
        Size =
            llvm::alignTo(llvm::alignTo(Size, Alignment) + RecSize, Alignment);
      }
@ -4529,32 +4538,67 @@ void CGOpenMPRuntimeNVPTX::clear() {
      llvm::APInt ArySize(/*numBits=*/64, Size);
      QualType SubTy = C.getConstantArrayType(
          C.CharTy, ArySize, ArrayType::Normal, /*IndexTypeQuals=*/0);
-      auto *Field = FieldDecl::Create(
-          C, StaticRD, SourceLocation(), SourceLocation(), nullptr, SubTy,
-          C.getTrivialTypeSourceInfo(SubTy, SourceLocation()),
-          /*BW=*/nullptr, /*Mutable=*/false,
-          /*InitStyle=*/ICIS_NoInit);
+      const bool UseSharedMemory = Size <= SharedMemorySize;
+      auto *Field =
+          FieldDecl::Create(C, UseSharedMemory ? SharedStaticRD : StaticRD,
+                            SourceLocation(), SourceLocation(), nullptr, SubTy,
+                            C.getTrivialTypeSourceInfo(SubTy, SourceLocation()),
+                            /*BW=*/nullptr, /*Mutable=*/false,
+                            /*InitStyle=*/ICIS_NoInit);
      Field->setAccess(AS_public);
-      StaticRD->addDecl(Field);
+      if (UseSharedMemory) {
+        SharedStaticRD->addDecl(Field);
+        SharedRecs.push_back(&Records);
+      } else {
+        StaticRD->addDecl(Field);
+        GlobalRecs.push_back(&Records);
+      }
      Records.RecSize->setInitializer(llvm::ConstantInt::get(CGM.SizeTy, Size));
+      Records.UseSharedMemory->setInitializer(
+          llvm::ConstantInt::get(CGM.Int16Ty, UseSharedMemory ? 1 : 0));
+    }
+    SharedStaticRD->completeDefinition();
+    if (!SharedStaticRD->field_empty()) {
+      QualType StaticTy = C.getRecordType(SharedStaticRD);
+      llvm::Type *LLVMStaticTy = CGM.getTypes().ConvertTypeForMem(StaticTy);
+      auto *GV = new llvm::GlobalVariable(
+          CGM.getModule(), LLVMStaticTy,
+          /*isConstant=*/false, llvm::GlobalValue::WeakAnyLinkage,
+          llvm::Constant::getNullValue(LLVMStaticTy),
+          "_openmp_shared_static_glob_rd_$_", /*InsertBefore=*/nullptr,
+          llvm::GlobalValue::NotThreadLocal,
+          C.getTargetAddressSpace(LangAS::cuda_shared));
+      auto *Replacement = llvm::ConstantExpr::getPointerBitCastOrAddrSpaceCast(
+          GV, CGM.VoidPtrTy);
+      for (const GlobalPtrSizeRecsTy *Rec : SharedRecs) {
+        Rec->Buffer->replaceAllUsesWith(Replacement);
+        Rec->Buffer->eraseFromParent();
+      }
    }
    StaticRD->completeDefinition();
-    QualType StaticTy = C.getRecordType(StaticRD);
-    std::pair<unsigned, unsigned> SMsBlockPerSM = getSMsBlocksPerSM(CGM);
-    llvm::APInt Size1(32, SMsBlockPerSM.second);
-    QualType Arr1Ty = C.getConstantArrayType(StaticTy, Size1, ArrayType::Normal,
-                                             /*IndexTypeQuals=*/0);
-    llvm::APInt Size2(32, SMsBlockPerSM.first);
-    QualType Arr2Ty = C.getConstantArrayType(Arr1Ty, Size2, ArrayType::Normal,
-                                             /*IndexTypeQuals=*/0);
-    llvm::Type *LLVMArr2Ty = CGM.getTypes().ConvertTypeForMem(Arr2Ty);
-    auto *GV = new llvm::GlobalVariable(
-        CGM.getModule(), LLVMArr2Ty,
-        /*isConstant=*/false, llvm::GlobalValue::WeakAnyLinkage,
-        llvm::Constant::getNullValue(LLVMArr2Ty), "_openmp_static_glob_rd_$_");
-    StaticGlobalized->setInitializer(
-        llvm::ConstantExpr::getPointerBitCastOrAddrSpaceCast(GV,
-                                                             CGM.VoidPtrTy));
+    if (!StaticRD->field_empty()) {
+      QualType StaticTy = C.getRecordType(StaticRD);
+      std::pair<unsigned, unsigned> SMsBlockPerSM = getSMsBlocksPerSM(CGM);
+      llvm::APInt Size1(32, SMsBlockPerSM.second);
+      QualType Arr1Ty =
+          C.getConstantArrayType(StaticTy, Size1, ArrayType::Normal,
+                                 /*IndexTypeQuals=*/0);
+      llvm::APInt Size2(32, SMsBlockPerSM.first);
+      QualType Arr2Ty = C.getConstantArrayType(Arr1Ty, Size2, ArrayType::Normal,
+                                               /*IndexTypeQuals=*/0);
+      llvm::Type *LLVMArr2Ty = CGM.getTypes().ConvertTypeForMem(Arr2Ty);
+      auto *GV = new llvm::GlobalVariable(
+          CGM.getModule(), LLVMArr2Ty,
+          /*isConstant=*/false, llvm::GlobalValue::WeakAnyLinkage,
+          llvm::Constant::getNullValue(LLVMArr2Ty),
+          "_openmp_static_glob_rd_$_");
+      auto *Replacement = llvm::ConstantExpr::getPointerBitCastOrAddrSpaceCast(
+          GV, CGM.VoidPtrTy);
+      for (const GlobalPtrSizeRecsTy *Rec : GlobalRecs) {
+        Rec->Buffer->replaceAllUsesWith(Replacement);
+        Rec->Buffer->eraseFromParent();
+      }
+    }
  }
  CGOpenMPRuntime::clear();
 }
--- a/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.h
+++ b/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.h
@ -420,14 +420,14 @@ private:
  /// union. This resulting union (one per CU) is the entry point for the static
  /// memory management runtime functions.
  struct GlobalPtrSizeRecsTy {
+    llvm::GlobalVariable *UseSharedMemory = nullptr;
    llvm::GlobalVariable *RecSize = nullptr;
+    llvm::GlobalVariable *Buffer = nullptr;
+    SourceLocation Loc;
    llvm::SmallVector<const RecordDecl *, 2> Records;
    unsigned RegionCounter = 0;
  };
  llvm::SmallVector<GlobalPtrSizeRecsTy, 8> GlobalizedRecords;
-  /// Global variable used for staticlly allocated global memoryused for
-  /// globalization in target/teams/distribute regions.
-  llvm::GlobalVariable *StaticGlobalized = nullptr;
  /// Shared pointer for the global memory in the global memory buffer used for
  /// the given kernel.
  llvm::GlobalVariable *KernelStaticGlobalized = nullptr;
--- a/clang/test/OpenMP/nvptx_data_sharing.cpp
+++ b/clang/test/OpenMP/nvptx_data_sharing.cpp
@ -27,10 +27,10 @@ void test_ds(){
  }
 }
 // CK1: [[MEM_TY:%.+]] = type { [8 x i8] }
-// CK1-DAG: [[GLOBAL_RD:@.+]] = weak global [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]] zeroinitializer
-// CK1-DAG: [[GLOBAL_RD_PTR:@.+]] = weak unnamed_addr constant i8* getelementptr inbounds ([{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]], [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]]* [[GLOBAL_RD]], i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0)
+// CK1-DAG: [[SHARED_GLOBAL_RD:@.+]] = weak addrspace(3) global [[MEM_TY]] zeroinitializer
 // CK1-DAG: [[KERNEL_PTR:@.+]] = internal addrspace(3) global i8* null
 // CK1-DAG: [[KERNEL_SIZE:@.+]] = internal unnamed_addr constant i64 8
+// CK1-DAG: [[KERNEL_SHARED:@.+]] = internal unnamed_addr constant i16 1

 /// ========= In the worker function ========= ///
 // CK1: {{.*}}define internal void @__omp_offloading{{.*}}test_ds{{.*}}_worker()
@ -44,9 +44,9 @@ void test_ds(){
 // CK1: [[SHAREDARGS2:%.+]] = alloca i8**
 // CK1: call void @__kmpc_kernel_init
 // CK1: call void @__kmpc_data_sharing_init_stack
-// CK1: [[GLOBAL_RD:%.+]] = load i8*, i8** [[GLOBAL_RD_PTR]],
+// CK1: [[SHARED_MEM_FLAG:%.+]] = load i16, i16* [[KERNEL_SHARED]],
 // CK1: [[SIZE:%.+]] = load i64, i64* [[KERNEL_SIZE]],
-// CK1: call void @__kmpc_get_team_static_memory(i8* [[GLOBAL_RD]], i64 [[SIZE]], i16 0, i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
+// CK1: call void @__kmpc_get_team_static_memory(i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([[MEM_TY]], [[MEM_TY]] addrspace(3)* [[SHARED_GLOBAL_RD]], i32 0, i32 0, i32 0) to i8*), i64 [[SIZE]], i16 [[SHARED_MEM_FLAG]], i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
 // CK1: [[KERNEL_RD:%.+]] = load i8*, i8* addrspace(3)* [[KERNEL_PTR]],
 // CK1: [[GLOBALSTACK:%.+]] = getelementptr inbounds i8, i8* [[KERNEL_RD]], i64 0
 // CK1: [[GLOBALSTACK2:%.+]] = bitcast i8* [[GLOBALSTACK]] to %struct._globalized_locals_ty*
@ -75,7 +75,8 @@ void test_ds(){
 // CK1: call void @llvm.nvvm.barrier0()
 // CK1: call void @llvm.nvvm.barrier0()
 // CK1: call void @__kmpc_end_sharing_variables()
-// CK1: call void @__kmpc_restore_team_static_memory(i16 0)
+// CK1: [[SHARED_MEM_FLAG:%.+]] = load i16, i16* [[KERNEL_SHARED]],
+// CK1: call void @__kmpc_restore_team_static_memory(i16 [[SHARED_MEM_FLAG]])
 // CK1: call void @__kmpc_kernel_deinit(i16 1)

 /// ========= In the data sharing wrapper function ========= ///
--- a/clang/test/OpenMP/nvptx_distribute_parallel_generic_mode_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_distribute_parallel_generic_mode_codegen.cpp
@ -22,8 +22,7 @@ int main(int argc, char **argv) {
 }

 // CHECK: [[MEM_TY:%.+]] = type { [84 x i8] }
-// CHECK-DAG: [[GLOBAL_RD:@.+]] = weak global [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]] zeroinitializer
-// CHECK-DAG: [[GLOBAL_RD_PTR:@.+]] = weak unnamed_addr constant i8* getelementptr inbounds ([{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]], [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]]* [[GLOBAL_RD]], i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0)
+// CHECK-DAG: [[SHARED_GLOBAL_RD:@.+]] = weak addrspace(3) global [[MEM_TY]] zeroinitializer
 // CHECK-DAG: [[KERNEL_PTR:@.+]] = internal addrspace(3) global i8* null
 // CHECK-DAG: [[KERNEL_SIZE:@.+]] = internal unnamed_addr constant i{{64|32}} 84
 // CHECK-DAG: @__omp_offloading_{{.*}}_main_l17_exec_mode = weak constant i8 1
@ -31,8 +30,7 @@ int main(int argc, char **argv) {
 // CHECK-LABEL: define internal void @__omp_offloading_{{.*}}_main_l17_worker(

 // CHECK: define weak void @__omp_offloading_{{.*}}_main_l17([10 x i32]* dereferenceable(40) %{{.+}}, [10 x i32]* dereferenceable(40) %{{.+}}, i32* dereferenceable(4) %{{.+}}, i{{64|32}} %{{.+}}, [10 x i32]* dereferenceable(40) %{{.+}})
-// CHECK: [[GLOBAL_RD:%.+]] = load i8*, i8** [[GLOBAL_RD_PTR]],
-// CHECK: call void @__kmpc_get_team_static_memory(i8* [[GLOBAL_RD]], i{{64|32}} 84, i16 0, i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
+// CHECK: call void @__kmpc_get_team_static_memory(i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([[MEM_TY]], [[MEM_TY]] addrspace(3)* [[SHARED_GLOBAL_RD]], i32 0, i32 0, i32 0) to i8*), i{{64|32}} 84, i16 1, i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
 // CHECK: [[PTR:%.+]] = load i8*, i8* addrspace(3)* [[KERNEL_PTR]],
 // CHECK: [[STACK:%.+]] = bitcast i8* [[PTR]] to %struct._globalized_locals_ty*
 // CHECK: [[ARGC:%.+]] = load i32, i32* %{{.+}}, align
@ -48,7 +46,7 @@ int main(int argc, char **argv) {

 // CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* @

-// CHECK: call void @__kmpc_restore_team_static_memory(i16 0)
+// CHECK: call void @__kmpc_restore_team_static_memory(i16 1)

 // CHECK: define internal void [[PARALLEL]](
 // CHECK-NOT: call i8* @__kmpc_data_sharing_push_stack(
--- a/clang/test/OpenMP/nvptx_parallel_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_parallel_codegen.cpp
@ -72,10 +72,10 @@ int bar(int n){
 }

 // CHECK: [[MEM_TY:%.+]] = type { [4 x i8] }
-// CHECK-DAG: [[GLOBAL_RD:@.+]] = weak global [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]] zeroinitializer
-// CHECK-DAG: [[GLOBAL_RD_PTR:@.+]] = weak unnamed_addr constant i8* getelementptr inbounds ([{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]], [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]]* [[GLOBAL_RD]], i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0)
+// CHECK-DAG: [[SHARED_GLOBAL_RD:@.+]] = weak addrspace(3) global [[MEM_TY]] zeroinitializer
 // CHECK-DAG: [[KERNEL_PTR:@.+]] = internal addrspace(3) global i8* null
 // CHECK-DAG: [[KERNEL_SIZE:@.+]] = internal unnamed_addr constant i{{64|32}} 4
+// CHECK-DAG: [[KERNEL_SHARED:@.+]] = internal unnamed_addr constant i16 1

 // CHECK-NOT: define {{.*}}void {{@__omp_offloading_.+template.+l17}}_worker()

@ -324,9 +324,9 @@ int bar(int n){
 // CHECK-32: [[A_ADDR:%.+]] = alloca i32,
 // CHECK-64: [[A_ADDR:%.+]] = alloca i64,
 // CHECK-64: [[CONV:%.+]] = bitcast i64* [[A_ADDR]] to i32*
-// CHECK: [[GLOBAL_RD:%.+]] = load i8*, i8** [[GLOBAL_RD_PTR]],
+// CHECK: [[IS_SHARED:%.+]] = load i16, i16* [[KERNEL_SHARED]],
 // CHECK: [[SIZE:%.+]] = load i{{64|32}}, i{{64|32}}* [[KERNEL_SIZE]],
-// CHECK: call void @__kmpc_get_team_static_memory(i8* [[GLOBAL_RD]], i{{64|32}} [[SIZE]], i16 0, i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
+// CHECK: call void @__kmpc_get_team_static_memory(i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([[MEM_TY]], [[MEM_TY]] addrspace(3)* [[SHARED_GLOBAL_RD]], i32 0, i32 0, i32 0) to i8*), i{{64|32}} [[SIZE]], i16 [[IS_SHARED]], i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
 // CHECK: [[KERNEL_RD:%.+]] = load i8*, i8* addrspace(3)* [[KERNEL_PTR]],
 // CHECK: [[STACK:%.+]] = getelementptr inbounds i8, i8* [[KERNEL_RD]], i{{64|32}} 0
 // CHECK: [[BC:%.+]] = bitcast i8* [[STACK]] to %struct._globalized_locals_ty*
@ -334,7 +334,8 @@ int bar(int n){
 // CHECK-64: [[A:%.+]] = load i32, i32* [[CONV]],
 // CHECK: [[GLOBAL_A_ADDR:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[BC]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
 // CHECK: store i32 [[A]], i32* [[GLOBAL_A_ADDR]],
-// CHECK: call void @__kmpc_restore_team_static_memory(i16 0)
+// CHECK: [[IS_SHARED:%.+]] = load i16, i16* [[KERNEL_SHARED]],
+// CHECK: call void @__kmpc_restore_team_static_memory(i16 [[IS_SHARED]])

 // CHECK-LABEL: define internal void @{{.+}}(i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* dereferenceable{{.*}})
 // CHECK:  [[CC:%.+]] = alloca i32,
--- a/clang/test/OpenMP/nvptx_parallel_for_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_parallel_for_codegen.cpp
@ -31,10 +31,10 @@ int bar(int n){
 }

 // CHECK: [[MEM_TY:%.+]] = type { [4 x i8] }
-// CHECK-DAG: [[GLOBAL_RD:@.+]] = weak global [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]] zeroinitializer
-// CHECK-DAG: [[GLOBAL_RD_PTR:@.+]] = weak unnamed_addr constant i8* getelementptr inbounds ([{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]], [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]]* [[GLOBAL_RD]], i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0)
+// CHECK-DAG: [[SHARED_GLOBAL_RD:@.+]] = weak addrspace(3) global [[MEM_TY]] zeroinitializer
 // CHECK-DAG: [[KERNEL_PTR:@.+]] = internal addrspace(3) global i8* null
 // CHECK-DAG: [[KERNEL_SIZE:@.+]] = internal unnamed_addr constant i{{64|32}} 4
+// CHECK-DAG: [[KERNEL_SHARED:@.+]] = internal unnamed_addr constant i16 1

 // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l12}}_worker()
 // CHECK: call void @llvm.nvvm.barrier0()
@ -45,9 +45,9 @@ int bar(int n){
 // CHECK: call void @__omp_offloading_{{.*}}l12_worker()
 // CHECK: call void @__kmpc_kernel_init(
 // CHECK: call void @__kmpc_data_sharing_init_stack()
-// CHECK: [[GLOBAL_RD:%.+]] = load i8*, i8** [[GLOBAL_RD_PTR]],
+// CHECK: [[IS_SHARED:%.+]] = load i16, i16* [[KERNEL_SHARED]],
 // CHECK: [[SIZE:%.+]] = load i{{64|32}}, i{{64|32}}* [[KERNEL_SIZE]],
-// CHECK: call void @__kmpc_get_team_static_memory(i8* [[GLOBAL_RD]], i{{64|32}} [[SIZE]], i16 0, i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
+// CHECK: call void @__kmpc_get_team_static_memory(i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([[MEM_TY]], [[MEM_TY]] addrspace(3)* [[SHARED_GLOBAL_RD]], i32 0, i32 0, i32 0) to i8*), i64 %7, i16 %6, i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
 // CHECK: [[KERNEL_RD:%.+]] = load i8*, i8* addrspace(3)* [[KERNEL_PTR]],
 // CHECK: [[STACK:%.+]] = getelementptr inbounds i8, i8* [[KERNEL_RD]], i{{64|32}} 0
 // CHECK: call void @__kmpc_kernel_prepare_parallel(
@ -55,7 +55,8 @@ int bar(int n){
 // CHECK: call void @llvm.nvvm.barrier0()
 // CHECK: call void @llvm.nvvm.barrier0()
 // CHECK: call void @__kmpc_end_sharing_variables()
-// CHECK: call void @__kmpc_restore_team_static_memory(i16 0)
+// CHECK: [[IS_SHARED:%.+]] = load i16, i16* [[KERNEL_SHARED]],
+// CHECK: call void @__kmpc_restore_team_static_memory(i16 [[IS_SHARED]])
 // CHECK: call void @__kmpc_kernel_deinit(i16 1)

 // CHECK: define internal void @__omp_outlined__(
--- a/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_codegen.cpp
@ -68,23 +68,22 @@ int bar(int n){
 }

 // CHECK-DAG: [[MEM_TY:%.+]] = type { [4 x i8] }
-// CHECK-DAG: [[GLOBAL_RD:@.+]] = weak global [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]] zeroinitializer
-// CHECK-DAG: [[GLOBAL_RD_PTR:@.+]] = weak unnamed_addr constant i8* getelementptr inbounds ([{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]], [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]]* [[GLOBAL_RD]], i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0)
+// CHECK-DAG: [[SHARED_GLOBAL_RD:@.+]] = weak addrspace(3) global [[MEM_TY]] zeroinitializer
 // CHECK-DAG: [[KERNEL_PTR:@.+]] = internal addrspace(3) global i8* null
 // CHECK-DAG: [[KERNEL_SIZE:@.+]] = internal unnamed_addr constant i{{64|32}} 4
+// CHECK-DAG: [[KERNEL_SHARED:@.+]] = internal unnamed_addr constant i16 1

 // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}_l32(
 // CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
 // CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]], i16 0, i16 1)
-// CHECK: [[GLOBAL_RD:%.+]] = load i8*, i8** [[GLOBAL_RD_PTR]],
-// CHECK: call void @__kmpc_get_team_static_memory(i8* [[GLOBAL_RD]], i{{64|32}} 4, i16 0, i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
+// CHECK: call void @__kmpc_get_team_static_memory(i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([[MEM_TY]], [[MEM_TY]] addrspace(3)* [[SHARED_GLOBAL_RD]], i32 0, i32 0, i32 0) to i8*), i{{64|32}} 4, i16 1, i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
 // CHECK: [[TEAM_ALLOC:%.+]] = load i8*, i8* addrspace(3)* [[KERNEL_PTR]],
 // CHECK: [[BC:%.+]] = bitcast i8* [[TEAM_ALLOC]] to [[REC:%.+]]*
 // CHECK: getelementptr inbounds [[REC]], [[REC]]* [[BC]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
 // CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 91,
 // CHECK: {{call|invoke}} void [[OUTL1:@.+]](
 // CHECK: call void @__kmpc_for_static_fini(
-// CHECK: call void @__kmpc_restore_team_static_memory(i16 0)
+// CHECK: call void @__kmpc_restore_team_static_memory(i16 1)
 // CHECK: call void @__kmpc_spmd_kernel_deinit()
 // CHECK: ret void

--- a/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_simd_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_simd_codegen.cpp
@ -63,23 +63,22 @@ int bar(int n){
 }

 // CHECK-DAG: [[MEM_TY:%.+]] = type { [4 x i8] }
-// CHECK-DAG: [[GLOBAL_RD:@.+]] = weak global [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]] zeroinitializer
-// CHECK-DAG: [[GLOBAL_RD_PTR:@.+]] = weak unnamed_addr constant i8* getelementptr inbounds ([{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]], [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]]* [[GLOBAL_RD]], i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0)
+// CHECK-DAG: [[SHARED_GLOBAL_RD:@.+]] = weak addrspace(3) global [[MEM_TY]] zeroinitializer
 // CHECK-DAG: [[KERNEL_PTR:@.+]] = internal addrspace(3) global i8* null
 // CHECK-DAG: [[KERNEL_SIZE:@.+]] = internal unnamed_addr constant i{{64|32}} 4
+// CHECK-DAG: [[KERNEL_SHARED:@.+]] = internal unnamed_addr constant i16 1

 // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}_l30(
 // CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
 // CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]], i16 0, i16 1)
-// CHECK: [[GLOBAL_RD:%.+]] = load i8*, i8** [[GLOBAL_RD_PTR]],
-// CHECK: call void @__kmpc_get_team_static_memory(i8* [[GLOBAL_RD]], i{{64|32}} 4, i16 0, i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
+// CHECK: call void @__kmpc_get_team_static_memory(i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([[MEM_TY]], [[MEM_TY]] addrspace(3)* [[SHARED_GLOBAL_RD]], i32 0, i32 0, i32 0) to i8*), i{{64|32}} 4, i16 1, i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
 // CHECK: [[TEAM_ALLOC:%.+]] = load i8*, i8* addrspace(3)* [[KERNEL_PTR]],
 // CHECK: [[BC:%.+]] = bitcast i8* [[TEAM_ALLOC]] to [[REC:%.+]]*
 // CHECK: getelementptr inbounds [[REC]], [[REC]]* [[BC]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
 // CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 91,
 // CHECK: {{call|invoke}} void [[OUTL1:@.+]](
 // CHECK: call void @__kmpc_for_static_fini(
-// CHECK: call void @__kmpc_restore_team_static_memory(i16 0)
+// CHECK: call void @__kmpc_restore_team_static_memory(i16 1)
 // CHECK: call void @__kmpc_spmd_kernel_deinit()
 // CHECK: ret void

--- a/clang/test/OpenMP/nvptx_teams_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_teams_codegen.cpp
@ -28,11 +28,12 @@ int main (int argc, char **argv) {
 }

 // CK1: [[MEM_TY:%.+]] = type { [{{4|8}} x i8] }
-// CK1-DAG: [[GLOBAL_RD:@.+]] = weak global [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]] zeroinitializer
-// CK1-DAG: [[GLOBAL_RD_PTR:@.+]] = weak unnamed_addr constant i8* getelementptr inbounds ([{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]], [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]]* [[GLOBAL_RD]], i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0)
+// CK1-DAG: [[SHARED_GLOBAL_RD:@.+]] = weak addrspace(3) global [[MEM_TY]] zeroinitializer
 // CK1-DAG: [[KERNEL_PTR:@.+]] = internal addrspace(3) global i8* null
 // CK1-DAG: [[KERNEL_SIZE1:@.+]] = internal unnamed_addr constant i{{64|32}} 4
 // CK1-DAG: [[KERNEL_SIZE2:@.+]] = internal unnamed_addr constant i{{64|32}} {{8|4}}
+// CK1-DAG: [[KERNEL_SHARED1:@.+]] = internal unnamed_addr constant i16 1
+// CK1-DAG: [[KERNEL_SHARED2:@.+]] = internal unnamed_addr constant i16 1

 // only nvptx side: do not outline teams region and do not call fork_teams
 // CK1:  define {{.*}}void @{{[^,]+}}(i{{[0-9]+}} [[ARGC:%.+]])
@ -43,9 +44,9 @@ int main (int argc, char **argv) {
 // CK1:  store {{.+}} 0, {{.+}},
 // CK1:  store i{{[0-9]+}} [[ARGC]], i{{[0-9]+}}* [[ARGCADDR]],
 // CK1-64:  [[CONV:%.+]] = bitcast i{{[0-9]+}}* [[ARGCADDR]] to i{{[0-9]+}}*
-// CK1: [[GLOBAL_RD:%.+]] = load i8*, i8** [[GLOBAL_RD_PTR]],
+// CK1: [[IS_SHARED:%.+]] = load i16, i16* [[KERNEL_SHARED1]],
 // CK1: [[SIZE:%.+]] = load i{{64|32}}, i{{64|32}}* [[KERNEL_SIZE1]],
-// CK1: call void @__kmpc_get_team_static_memory(i8* [[GLOBAL_RD]], i{{64|32}} [[SIZE]], i16 0, i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
+// CK1: call void @__kmpc_get_team_static_memory(i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([[MEM_TY]], [[MEM_TY]] addrspace(3)* [[SHARED_GLOBAL_RD]], i32 0, i32 0, i32 0) to i8*), i{{64|32}} [[SIZE]], i16 [[IS_SHARED]], i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
 // CK1: [[KERNEL_RD:%.+]] = load i8*, i8* addrspace(3)* [[KERNEL_PTR]],
 // CK1: [[GLOBALSTACK:%.+]] = getelementptr inbounds i8, i8* [[KERNEL_RD]], i{{64|32}} 0
 // CK1-64:  [[ARG:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[CONV]]
@ -64,9 +65,9 @@ int main (int argc, char **argv) {
 // CK1: [[ARGCADDR_PTR:%.+]] = alloca i{{.+}}***,
 // CK1: [[ARGCADDR:%.+]] = alloca i{{.+}}**,
 // CK1: store i{{.+}}** [[ARGC]], i{{.+}}*** [[ARGCADDR]]
-// CK1: [[GLOBAL_RD:%.+]] = load i8*, i8** [[GLOBAL_RD_PTR]],
+// CK1: [[IS_SHARED:%.+]] = load i16, i16* [[KERNEL_SHARED2]],
 // CK1: [[SIZE:%.+]] = load i{{64|32}}, i{{64|32}}* [[KERNEL_SIZE2]],
-// CK1: call void @__kmpc_get_team_static_memory(i8* [[GLOBAL_RD]], i{{64|32}} [[SIZE]], i16 0, i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
+// CK1: call void @__kmpc_get_team_static_memory(i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([[MEM_TY]], [[MEM_TY]] addrspace(3)* [[SHARED_GLOBAL_RD]], i32 0, i32 0, i32 0) to i8*), i{{64|32}} [[SIZE]], i16 [[IS_SHARED]], i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
 // CK1: [[KERNEL_RD:%.+]] = load i8*, i8* addrspace(3)* [[KERNEL_PTR]],
 // CK1: [[GLOBALSTACK:%.+]] = getelementptr inbounds i8, i8* [[KERNEL_RD]], i{{64|32}} 0
 // CK1: [[ARG:%.+]] = load i{{[0-9]+}}**, i{{[0-9]+}}*** [[ARGCADDR]]
@ -114,11 +115,12 @@ int main (int argc, char **argv) {
 }

 // CK2: [[MEM_TY:%.+]] = type { [{{4|8}} x i8] }
-// CK2-DAG: [[GLOBAL_RD:@.+]] = weak global [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]] zeroinitializer
-// CK2-DAG: [[GLOBAL_RD_PTR:@.+]] = weak unnamed_addr constant i8* getelementptr inbounds ([{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]], [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]]* [[GLOBAL_RD]], i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0)
+// CK2-DAG: [[SHARED_GLOBAL_RD:@.+]] = weak addrspace(3) global [[MEM_TY]] zeroinitializer
 // CK2-DAG: [[KERNEL_PTR:@.+]] = internal addrspace(3) global i8* null
 // CK2-DAG: [[KERNEL_SIZE1:@.+]] = internal unnamed_addr constant i{{64|32}} 4
 // CK2-DAG: [[KERNEL_SIZE2:@.+]] = internal unnamed_addr constant i{{64|32}} {{8|4}}
+// CK2-DAG: [[KERNEL_SHARED1:@.+]] = internal unnamed_addr constant i16 1
+// CK2-DAG: [[KERNEL_SHARED2:@.+]] = internal unnamed_addr constant i16 1

 // CK2: define {{.*}}void @{{[^,]+}}(i{{[0-9]+}} [[A_IN:%.+]], i{{[0-9]+}} [[B_IN:%.+]], i{{[0-9]+}} [[ARGC_IN:.+]])
 // CK2: {{.}} = alloca i{{[0-9]+}}*,
@ -133,9 +135,9 @@ int main (int argc, char **argv) {
 // CK2-64: [[ACONV:%.+]] = bitcast i64* [[AADDR]] to i32*
 // CK2-64: [[BCONV:%.+]] = bitcast i64* [[BADDR]] to i32*
 // CK2-64: [[CONV:%.+]] = bitcast i64* [[ARGCADDR]] to i32*
-// CK2: [[GLOBAL_RD:%.+]] = load i8*, i8** [[GLOBAL_RD_PTR]],
+// CK2: [[IS_SHARED:%.+]] = load i16, i16* [[KERNEL_SHARED1]],
 // CK2: [[SIZE:%.+]] = load i{{64|32}}, i{{64|32}}* [[KERNEL_SIZE1]],
-// CK2: call void @__kmpc_get_team_static_memory(i8* [[GLOBAL_RD]], i{{64|32}} [[SIZE]], i16 0, i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
+// CK2: call void @__kmpc_get_team_static_memory(i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([[MEM_TY]], [[MEM_TY]] addrspace(3)* [[SHARED_GLOBAL_RD]], i32 0, i32 0, i32 0) to i8*), i{{64|32}} [[SIZE]], i16 [[IS_SHARED]], i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
 // CK2: [[KERNEL_RD:%.+]] = load i8*, i8* addrspace(3)* [[KERNEL_PTR]],
 // CK2: [[GLOBALSTACK:%.+]] = getelementptr inbounds i8, i8* [[KERNEL_RD]], i{{64|32}} 0
 // CK2-64:  [[ARG:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[CONV]]
@ -158,9 +160,9 @@ int main (int argc, char **argv) {
 // CK2: store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[AADDR]],
 // CK2: store i{{[0-9]+}} [[B_IN]], i{{[0-9]+}}* [[BADDR]],
 // CK2: store i{{[0-9]+}}** [[ARGC]], i{{[0-9]+}}*** [[ARGCADDR]],
-// CK2: [[GLOBAL_RD:%.+]] = load i8*, i8** [[GLOBAL_RD_PTR]],
+// CK2: [[IS_SHARED:%.+]] = load i16, i16* [[KERNEL_SHARED2]],
 // CK2: [[SIZE:%.+]] = load i{{64|32}}, i{{64|32}}* [[KERNEL_SIZE2]],
-// CK2: call void @__kmpc_get_team_static_memory(i8* [[GLOBAL_RD]], i{{64|32}} [[SIZE]], i16 0, i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
+// CK2: call void @__kmpc_get_team_static_memory(i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([[MEM_TY]], [[MEM_TY]] addrspace(3)* [[SHARED_GLOBAL_RD]], i32 0, i32 0, i32 0) to i8*), i{{64|32}} [[SIZE]], i16 [[IS_SHARED]], i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
 // CK2: [[KERNEL_RD:%.+]] = load i8*, i8* addrspace(3)* [[KERNEL_PTR]],
 // CK2: [[GLOBALSTACK:%.+]] = getelementptr inbounds i8, i8* [[KERNEL_RD]], i{{64|32}} 0
 // CK2: [[ARG:%.+]] = load i{{[0-9]+}}**, i{{[0-9]+}}*** [[ARGCADDR]]