diff --git a/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp b/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
--- a/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
@@ -781,7 +781,7 @@
     emitGenericVarsEpilog(CGF);
 
   CGBuilderTy &Bld = CGF.Builder;
-  OMPBuilder.createTargetDeinit(Bld, IsSPMD);
+  OMPBuilder.createTargetDeinit(Bld);
 }
 
 void CGOpenMPRuntimeGPU::emitSPMDKernel(const OMPExecutableDirective &D,
@@ -819,24 +819,6 @@
   IsInTTDRegion = false;
 }
 
-// Create a unique global variable to indicate the execution mode of this target
-// region. The execution mode is either 'generic', or 'spmd' depending on the
-// target directive. This variable is picked up by the offload library to setup
-// the device appropriately before kernel launch. If the execution mode is
-// 'generic', the runtime reserves one warp for the master, otherwise, all
-// warps participate in parallel work.
-static void setPropertyExecutionMode(CodeGenModule &CGM, StringRef Name,
-                                     bool Mode) {
-  auto *GVMode = new llvm::GlobalVariable(
-      CGM.getModule(), CGM.Int8Ty, /*isConstant=*/true,
-      llvm::GlobalValue::WeakAnyLinkage,
-      llvm::ConstantInt::get(CGM.Int8Ty, Mode ? OMP_TGT_EXEC_MODE_SPMD
-                                              : OMP_TGT_EXEC_MODE_GENERIC),
-      Twine(Name, "_exec_mode"));
-  GVMode->setVisibility(llvm::GlobalVariable::ProtectedVisibility);
-  CGM.addCompilerUsedGlobal(GVMode);
-}
-
 void CGOpenMPRuntimeGPU::emitTargetOutlinedFunction(
     const OMPExecutableDirective &D, StringRef ParentName,
     llvm::Function *&OutlinedFn, llvm::Constant *&OutlinedFnID,
@@ -853,8 +835,6 @@
   else
     emitNonSPMDKernel(D, ParentName, OutlinedFn, OutlinedFnID, IsOffloadEntry,
                       CodeGen);
-
-  setPropertyExecutionMode(CGM, OutlinedFn->getName(), Mode);
 }
 
 CGOpenMPRuntimeGPU::CGOpenMPRuntimeGPU(CodeGenModule &CGM)
diff --git a/clang/test/Headers/amdgcn_openmp_device_math_c.c b/clang/test/Headers/amdgcn_openmp_device_math_c.c
--- a/clang/test/Headers/amdgcn_openmp_device_math_c.c
+++ b/clang/test/Headers/amdgcn_openmp_device_math_c.c
@@ -36,7 +36,7 @@
 // CHECK-NEXT:    [[X_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[X_ADDR]] to ptr
 // CHECK-NEXT:    [[L1_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L1]] to ptr
 // CHECK-NEXT:    store i64 [[X]], ptr [[X_ADDR_ASCAST]], align 8
-// CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr addrspacecast (ptr addrspace(1) @[[GLOB1:[0-9]+]] to ptr), i8 1, i1 true)
+// CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr addrspacecast (ptr addrspace(1) @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_test_math_int_l9_kernel_environment to ptr))
 // CHECK-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK:       user_code.entry:
@@ -54,7 +54,7 @@
 // CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr [[__SGN_ASCAST_I]], align 4
 // CHECK-NEXT:    [[SUB_I:%.*]] = sub nsw i32 [[XOR_I]], [[TMP5]]
 // CHECK-NEXT:    store i32 [[SUB_I]], ptr [[L1_ASCAST]], align 4
-// CHECK-NEXT:    call void @__kmpc_target_deinit(ptr addrspacecast (ptr addrspace(1) @[[GLOB1]] to ptr), i8 1)
+// CHECK-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-NEXT:    ret void
 // CHECK:       worker.exit:
 // CHECK-NEXT:    ret void
@@ -71,7 +71,7 @@
 // CHECK-NEXT:    [[X_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[X_ADDR]] to ptr
 // CHECK-NEXT:    [[L1_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L1]] to ptr
 // CHECK-NEXT:    store i64 [[X]], ptr [[X_ADDR_ASCAST]], align 8
-// CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr addrspacecast (ptr addrspace(1) @[[GLOB1]] to ptr), i8 1, i1 true)
+// CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr addrspacecast (ptr addrspace(1) @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_test_math_long_l16_kernel_environment to ptr))
 // CHECK-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK:       user_code.entry:
@@ -89,7 +89,7 @@
 // CHECK-NEXT:    [[TMP5:%.*]] = load i64, ptr [[__SGN_ASCAST_I]], align 8
 // CHECK-NEXT:    [[SUB_I:%.*]] = sub nsw i64 [[XOR_I]], [[TMP5]]
 // CHECK-NEXT:    store i64 [[SUB_I]], ptr [[L1_ASCAST]], align 8
-// CHECK-NEXT:    call void @__kmpc_target_deinit(ptr addrspacecast (ptr addrspace(1) @[[GLOB1]] to ptr), i8 1)
+// CHECK-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-NEXT:    ret void
 // CHECK:       worker.exit:
 // CHECK-NEXT:    ret void
@@ -106,7 +106,7 @@
 // CHECK-NEXT:    [[X_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[X_ADDR]] to ptr
 // CHECK-NEXT:    [[L1_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L1]] to ptr
 // CHECK-NEXT:    store i64 [[X]], ptr [[X_ADDR_ASCAST]], align 8
-// CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr addrspacecast (ptr addrspace(1) @[[GLOB1]] to ptr), i8 1, i1 true)
+// CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr addrspacecast (ptr addrspace(1) @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_test_math_long_long_l23_kernel_environment to ptr))
 // CHECK-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK:       user_code.entry:
@@ -124,7 +124,7 @@
 // CHECK-NEXT:    [[TMP5:%.*]] = load i64, ptr [[__SGN_ASCAST_I]], align 8
 // CHECK-NEXT:    [[SUB_I:%.*]] = sub nsw i64 [[XOR_I]], [[TMP5]]
 // CHECK-NEXT:    store i64 [[SUB_I]], ptr [[L1_ASCAST]], align 8
-// CHECK-NEXT:    call void @__kmpc_target_deinit(ptr addrspacecast (ptr addrspace(1) @[[GLOB1]] to ptr), i8 1)
+// CHECK-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-NEXT:    ret void
 // CHECK:       worker.exit:
 // CHECK-NEXT:    ret void
diff --git a/clang/test/OpenMP/amdgcn_target_codegen.cpp b/clang/test/OpenMP/amdgcn_target_codegen.cpp
--- a/clang/test/OpenMP/amdgcn_target_codegen.cpp
+++ b/clang/test/OpenMP/amdgcn_target_codegen.cpp
@@ -37,7 +37,7 @@
 // CHECK-NEXT:    [[I_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[I]] to ptr
 // CHECK-NEXT:    store ptr [[ARR]], ptr [[ARR_ADDR_ASCAST]], align 8
 // CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[ARR_ADDR_ASCAST]], align 8
-// CHECK-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr addrspacecast (ptr addrspace(1) @[[GLOB1:[0-9]+]] to ptr), i8 1, i1 true)
+// CHECK-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr addrspacecast (ptr addrspace(1) @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z30test_amdgcn_target_tid_threadsv_l14_kernel_environment to ptr))
 // CHECK-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK:       user_code.entry:
@@ -61,7 +61,7 @@
 // CHECK:       worker.exit:
 // CHECK-NEXT:    ret void
 // CHECK:       for.end:
-// CHECK-NEXT:    call void @__kmpc_target_deinit(ptr addrspacecast (ptr addrspace(1) @[[GLOB1]] to ptr), i8 1)
+// CHECK-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-NEXT:    ret void
 //
 //
@@ -78,7 +78,7 @@
 // CHECK-NEXT:    [[I_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[I]] to ptr
 // CHECK-NEXT:    store ptr [[ARR]], ptr [[ARR_ADDR_ASCAST]], align 8
 // CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[ARR_ADDR_ASCAST]], align 8
-// CHECK-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr addrspacecast (ptr addrspace(1) @[[GLOB1]] to ptr), i8 2, i1 false)
+// CHECK-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr addrspacecast (ptr addrspace(1) @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z35test_amdgcn_target_tid_threads_simdv_l23_kernel_environment to ptr))
 // CHECK-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK:       user_code.entry:
@@ -109,6 +109,6 @@
 // CHECK-NEXT:    ret void
 // CHECK:       omp.inner.for.end:
 // CHECK-NEXT:    store i32 1000, ptr [[I_ASCAST]], align 4
-// CHECK-NEXT:    call void @__kmpc_target_deinit(ptr addrspacecast (ptr addrspace(1) @[[GLOB1]] to ptr), i8 2)
+// CHECK-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-NEXT:    ret void
 //
diff --git a/clang/test/OpenMP/declare_target_codegen_globalization.cpp b/clang/test/OpenMP/declare_target_codegen_globalization.cpp
--- a/clang/test/OpenMP/declare_target_codegen_globalization.cpp
+++ b/clang/test/OpenMP/declare_target_codegen_globalization.cpp
@@ -31,15 +31,15 @@
 // CHECK1-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 8
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z6maini1v_l16_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
-// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK1-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[TMP3]], align 8
 // CHECK1-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 1)
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
diff --git a/clang/test/OpenMP/nvptx_SPMD_codegen.cpp b/clang/test/OpenMP/nvptx_SPMD_codegen.cpp
--- a/clang/test/OpenMP/nvptx_SPMD_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_SPMD_codegen.cpp
@@ -281,11 +281,11 @@
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    store i64 [[DOTCAPTURE_EXPR_]], i64* [[DOTCAPTURE_EXPR__ADDR]], align 8
 // CHECK-64-NEXT:    [[CONV:%.*]] = bitcast i64* [[DOTCAPTURE_EXPR__ADDR]] to i8*
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l15_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
-// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
+// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = load i8, i8* [[CONV]], align 1
 // CHECK-64-NEXT:    [[TOBOOL:%.*]] = trunc i8 [[TMP2]] to i1
 // CHECK-64-NEXT:    [[CONV1:%.*]] = bitcast i64* [[DOTCAPTURE_EXPR__CASTED]] to i8*
@@ -295,7 +295,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP3]]) #[[ATTR2:[0-9]+]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -703,7 +703,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l18_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -711,7 +711,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__3(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -895,7 +895,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l21_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -903,7 +903,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__5(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -1078,7 +1078,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l24_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -1086,7 +1086,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__7(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -1268,7 +1268,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l27_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -1276,7 +1276,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__9(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -1458,7 +1458,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l30_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -1466,7 +1466,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__11(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -1648,7 +1648,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l33_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -1656,7 +1656,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__13(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -1842,7 +1842,7 @@
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    store i64 [[A]], i64* [[A_ADDR]], align 8
 // CHECK-64-NEXT:    [[CONV:%.*]] = bitcast i64* [[A_ADDR]] to i32*
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l37_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -1854,7 +1854,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__15(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i64 [[TMP3]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -2051,7 +2051,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l40_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -2059,7 +2059,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__17(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -2229,7 +2229,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l43_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -2237,7 +2237,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__19(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -2398,7 +2398,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l46_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -2406,7 +2406,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__21(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -2574,7 +2574,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l49_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -2582,7 +2582,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__23(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -2750,7 +2750,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l52_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -2758,7 +2758,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__25(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -2926,7 +2926,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l55_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -2934,7 +2934,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__27(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -3102,7 +3102,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l58_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -3110,7 +3110,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__29(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -3286,7 +3286,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l66_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -3294,7 +3294,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__31(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -3481,7 +3481,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 1, i1 true)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l73_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -3489,7 +3489,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__33(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 1)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -3672,7 +3672,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l81_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -3680,7 +3680,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__35(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -3862,7 +3862,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l85_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -3870,7 +3870,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__37(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -4052,7 +4052,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l89_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -4060,7 +4060,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__39(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -4242,7 +4242,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l93_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -4250,7 +4250,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__41(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -4432,7 +4432,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l97_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -4440,7 +4440,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__43(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -4601,7 +4601,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l101_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -4609,7 +4609,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__45(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -4779,7 +4779,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l105_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -4787,7 +4787,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__47(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -4948,7 +4948,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l109_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -4956,7 +4956,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__49(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -5124,7 +5124,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l113_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -5132,7 +5132,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__51(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -5300,7 +5300,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l117_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -5308,7 +5308,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__53(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -5476,7 +5476,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l121_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -5484,7 +5484,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__55(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -5652,7 +5652,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l125_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -5660,7 +5660,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__57(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -5821,7 +5821,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l130_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -5829,7 +5829,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__59(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -5999,7 +5999,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l135_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -6007,7 +6007,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__61(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -6168,7 +6168,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l140_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -6176,7 +6176,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__63(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -6344,7 +6344,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l145_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -6352,7 +6352,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__65(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -6520,7 +6520,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l150_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -6528,7 +6528,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__67(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -6696,7 +6696,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l155_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -6704,7 +6704,7 @@
 // CHECK-64-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__69(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -6874,7 +6874,7 @@
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
 // CHECK-64-NEXT:    store i64 [[DOTCAPTURE_EXPR_]], i64* [[DOTCAPTURE_EXPR__ADDR]], align 8
 // CHECK-64-NEXT:    [[CONV:%.*]] = bitcast i64* [[DOTCAPTURE_EXPR__ADDR]] to i8*
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l160_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -6884,7 +6884,7 @@
 // CHECK-64-NEXT:    [[TMP3:%.*]] = zext i1 [[TOBOOL]] to i32
 // CHECK-64-NEXT:    [[TMP4:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 [[TMP3]], i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__71 to i8*), i8* null, i8** [[TMP4]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -6971,14 +6971,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l163_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__72 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -7048,14 +7048,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l166_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__73 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -7142,14 +7142,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l169_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__74 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -7215,14 +7215,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l172_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__75 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -7288,14 +7288,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l175_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__76 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -7361,14 +7361,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l178_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__77 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -7437,7 +7437,7 @@
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
 // CHECK-64-NEXT:    store i64 [[DOTCAPTURE_EXPR_]], i64* [[DOTCAPTURE_EXPR__ADDR]], align 8
 // CHECK-64-NEXT:    [[CONV:%.*]] = bitcast i64* [[DOTCAPTURE_EXPR__ADDR]] to i8*
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l181_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -7447,7 +7447,7 @@
 // CHECK-64-NEXT:    [[TMP3:%.*]] = zext i1 [[TOBOOL]] to i32
 // CHECK-64-NEXT:    [[TMP4:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 [[TMP3]], i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__78 to i8*), i8* null, i8** [[TMP4]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -7542,14 +7542,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l185_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__79 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -7627,14 +7627,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l189_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__80 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -7729,14 +7729,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l193_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__81 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -7810,14 +7810,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l197_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__82 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -7891,14 +7891,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l201_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__83 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -7972,14 +7972,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l205_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__84 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -8053,14 +8053,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l209_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__85 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -8135,14 +8135,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l214_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__86 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -8220,14 +8220,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l219_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__87 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -8322,14 +8322,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l224_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__88 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -8403,14 +8403,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l229_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__89 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -8484,14 +8484,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l234_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__90 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -8565,14 +8565,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l239_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__91 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -8646,14 +8646,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l244_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__92 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -8740,14 +8740,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l248_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__93 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -8817,14 +8817,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l252_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__94 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -8911,14 +8911,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l256_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__95 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -8984,14 +8984,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l260_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__96 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -9057,14 +9057,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l264_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__97 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -9130,14 +9130,14 @@
 // CHECK-64-SAME: () #[[ATTR8]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l268_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-64-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__98 to i8*), i8* null, i8** [[TMP2]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -9208,11 +9208,11 @@
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    store i32 [[DOTCAPTURE_EXPR_]], i32* [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK-32-NEXT:    [[CONV:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR__ADDR]] to i8*
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l15_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
-// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
+// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = load i8, i8* [[CONV]], align 1
 // CHECK-32-NEXT:    [[TOBOOL:%.*]] = trunc i8 [[TMP2]] to i1
 // CHECK-32-NEXT:    [[CONV1:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR__CASTED]] to i8*
@@ -9222,7 +9222,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]]) #[[ATTR2:[0-9]+]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -9618,7 +9618,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l18_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -9626,7 +9626,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__3(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -9806,7 +9806,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l21_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -9814,7 +9814,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__5(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -9984,7 +9984,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l24_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -9992,7 +9992,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__7(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -10170,7 +10170,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l27_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -10178,7 +10178,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__9(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -10356,7 +10356,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l30_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -10364,7 +10364,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__11(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -10542,7 +10542,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l33_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -10550,7 +10550,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__13(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -10731,7 +10731,7 @@
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l37_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -10742,7 +10742,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__15(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -10931,7 +10931,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l40_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -10939,7 +10939,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__17(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -11105,7 +11105,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l43_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -11113,7 +11113,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__19(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -11269,7 +11269,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l46_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -11277,7 +11277,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__21(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -11441,7 +11441,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l49_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -11449,7 +11449,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__23(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -11613,7 +11613,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l52_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -11621,7 +11621,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__25(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -11785,7 +11785,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l55_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -11793,7 +11793,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__27(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -11957,7 +11957,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l58_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -11965,7 +11965,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__29(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -12136,7 +12136,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l66_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -12144,7 +12144,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__31(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -12327,7 +12327,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 1, i1 true)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l73_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -12335,7 +12335,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__33(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 1)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -12513,7 +12513,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l81_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -12521,7 +12521,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__35(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -12699,7 +12699,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l85_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -12707,7 +12707,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__37(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -12885,7 +12885,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l89_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -12893,7 +12893,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__39(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -13071,7 +13071,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l93_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -13079,7 +13079,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__41(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -13257,7 +13257,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l97_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -13265,7 +13265,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__43(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -13421,7 +13421,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l101_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -13429,7 +13429,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__45(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -13595,7 +13595,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l105_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -13603,7 +13603,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__47(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -13759,7 +13759,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l109_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -13767,7 +13767,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__49(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -13931,7 +13931,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l113_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -13939,7 +13939,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__51(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -14103,7 +14103,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l117_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -14111,7 +14111,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__53(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -14275,7 +14275,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l121_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -14283,7 +14283,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__55(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -14447,7 +14447,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l125_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -14455,7 +14455,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__57(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -14611,7 +14611,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l130_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -14619,7 +14619,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__59(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -14785,7 +14785,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l135_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -14793,7 +14793,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__61(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -14949,7 +14949,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l140_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -14957,7 +14957,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__63(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -15121,7 +15121,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l145_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -15129,7 +15129,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__65(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -15293,7 +15293,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l150_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -15301,7 +15301,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__67(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -15465,7 +15465,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l155_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -15473,7 +15473,7 @@
 // CHECK-32-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__69(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -15639,7 +15639,7 @@
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
 // CHECK-32-NEXT:    store i32 [[DOTCAPTURE_EXPR_]], i32* [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK-32-NEXT:    [[CONV:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR__ADDR]] to i8*
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l160_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -15649,7 +15649,7 @@
 // CHECK-32-NEXT:    [[TMP3:%.*]] = zext i1 [[TOBOOL]] to i32
 // CHECK-32-NEXT:    [[TMP4:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 [[TMP3]], i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__71 to i8*), i8* null, i8** [[TMP4]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -15736,14 +15736,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l163_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__72 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -15813,14 +15813,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l166_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__73 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -15907,14 +15907,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l169_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__74 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -15980,14 +15980,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l172_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__75 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -16053,14 +16053,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l175_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__76 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -16126,14 +16126,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l178_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__77 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -16202,7 +16202,7 @@
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
 // CHECK-32-NEXT:    store i32 [[DOTCAPTURE_EXPR_]], i32* [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK-32-NEXT:    [[CONV:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR__ADDR]] to i8*
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l181_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -16212,7 +16212,7 @@
 // CHECK-32-NEXT:    [[TMP3:%.*]] = zext i1 [[TOBOOL]] to i32
 // CHECK-32-NEXT:    [[TMP4:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 [[TMP3]], i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__78 to i8*), i8* null, i8** [[TMP4]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -16307,14 +16307,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l185_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__79 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -16392,14 +16392,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l189_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__80 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -16494,14 +16494,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l193_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__81 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -16575,14 +16575,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l197_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__82 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -16656,14 +16656,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l201_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__83 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -16737,14 +16737,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l205_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__84 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -16818,14 +16818,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l209_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__85 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -16900,14 +16900,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l214_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__86 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -16985,14 +16985,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l219_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__87 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -17087,14 +17087,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l224_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__88 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -17168,14 +17168,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l229_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__89 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -17249,14 +17249,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l234_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__90 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -17330,14 +17330,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l239_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__91 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -17411,14 +17411,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l244_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__92 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -17505,14 +17505,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l248_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__93 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -17582,14 +17582,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l252_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__94 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -17676,14 +17676,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l256_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__95 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -17749,14 +17749,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l260_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__96 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -17822,14 +17822,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l264_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__97 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -17895,14 +17895,14 @@
 // CHECK-32-SAME: () #[[ATTR8]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l268_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__98 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -17973,11 +17973,11 @@
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    store i32 [[DOTCAPTURE_EXPR_]], i32* [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK-32-EX-NEXT:    [[CONV:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR__ADDR]] to i8*
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l15_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
-// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
+// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = load i8, i8* [[CONV]], align 1
 // CHECK-32-EX-NEXT:    [[TOBOOL:%.*]] = trunc i8 [[TMP2]] to i1
 // CHECK-32-EX-NEXT:    [[CONV1:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR__CASTED]] to i8*
@@ -17987,7 +17987,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]]) #[[ATTR2:[0-9]+]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -18383,7 +18383,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l18_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -18391,7 +18391,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__3(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -18571,7 +18571,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l21_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -18579,7 +18579,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__5(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -18749,7 +18749,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l24_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -18757,7 +18757,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__7(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -18935,7 +18935,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l27_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -18943,7 +18943,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__9(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -19121,7 +19121,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l30_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -19129,7 +19129,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__11(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -19307,7 +19307,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l33_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -19315,7 +19315,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__13(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -19496,7 +19496,7 @@
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l37_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -19507,7 +19507,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__15(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]], i32 [[TMP3]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -19696,7 +19696,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l40_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -19704,7 +19704,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__17(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -19870,7 +19870,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l43_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -19878,7 +19878,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__19(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -20034,7 +20034,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l46_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -20042,7 +20042,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__21(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -20206,7 +20206,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l49_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -20214,7 +20214,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__23(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -20378,7 +20378,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l52_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -20386,7 +20386,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__25(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -20550,7 +20550,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l55_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -20558,7 +20558,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__27(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -20722,7 +20722,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l58_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -20730,7 +20730,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__29(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -20901,7 +20901,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l66_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -20909,7 +20909,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__31(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -21092,7 +21092,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 1, i1 true)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l73_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -21100,7 +21100,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__33(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 1)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -21278,7 +21278,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l81_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -21286,7 +21286,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__35(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -21464,7 +21464,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l85_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -21472,7 +21472,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__37(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -21650,7 +21650,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l89_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -21658,7 +21658,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__39(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -21836,7 +21836,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l93_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -21844,7 +21844,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__41(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -22022,7 +22022,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l97_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -22030,7 +22030,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__43(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -22186,7 +22186,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l101_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -22194,7 +22194,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__45(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -22360,7 +22360,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l105_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -22368,7 +22368,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__47(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -22524,7 +22524,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l109_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -22532,7 +22532,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__49(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -22696,7 +22696,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l113_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -22704,7 +22704,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__51(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -22868,7 +22868,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l117_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -22876,7 +22876,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__53(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -23040,7 +23040,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l121_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -23048,7 +23048,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__55(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -23212,7 +23212,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l125_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -23220,7 +23220,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__57(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -23376,7 +23376,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l130_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -23384,7 +23384,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__59(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -23550,7 +23550,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l135_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -23558,7 +23558,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__61(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -23714,7 +23714,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l140_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -23722,7 +23722,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__63(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -23886,7 +23886,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l145_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -23894,7 +23894,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__65(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -24058,7 +24058,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l150_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -24066,7 +24066,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__67(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -24230,7 +24230,7 @@
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l155_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -24238,7 +24238,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__69(i32* [[DOTTHREADID_TEMP_]], i32* [[DOTZERO_ADDR]]) #[[ATTR2]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -24404,7 +24404,7 @@
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
 // CHECK-32-EX-NEXT:    store i32 [[DOTCAPTURE_EXPR_]], i32* [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK-32-EX-NEXT:    [[CONV:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR__ADDR]] to i8*
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l160_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -24414,7 +24414,7 @@
 // CHECK-32-EX-NEXT:    [[TMP3:%.*]] = zext i1 [[TOBOOL]] to i32
 // CHECK-32-EX-NEXT:    [[TMP4:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 [[TMP3]], i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__71 to i8*), i8* null, i8** [[TMP4]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -24501,14 +24501,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l163_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__72 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -24578,14 +24578,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l166_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__73 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -24672,14 +24672,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l169_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__74 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -24745,14 +24745,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l172_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__75 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -24818,14 +24818,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l175_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__76 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -24891,14 +24891,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l178_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__77 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -24967,7 +24967,7 @@
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
 // CHECK-32-EX-NEXT:    store i32 [[DOTCAPTURE_EXPR_]], i32* [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK-32-EX-NEXT:    [[CONV:%.*]] = bitcast i32* [[DOTCAPTURE_EXPR__ADDR]] to i8*
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l181_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -24977,7 +24977,7 @@
 // CHECK-32-EX-NEXT:    [[TMP3:%.*]] = zext i1 [[TOBOOL]] to i32
 // CHECK-32-EX-NEXT:    [[TMP4:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 [[TMP3]], i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__78 to i8*), i8* null, i8** [[TMP4]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -25072,14 +25072,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l185_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__79 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -25157,14 +25157,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l189_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__80 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -25259,14 +25259,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l193_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__81 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -25340,14 +25340,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l197_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__82 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -25421,14 +25421,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l201_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__83 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -25502,14 +25502,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l205_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__84 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -25583,14 +25583,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l209_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__85 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -25665,14 +25665,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l214_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__86 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -25750,14 +25750,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l219_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__87 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -25852,14 +25852,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l224_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__88 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -25933,14 +25933,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l229_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__89 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -26014,14 +26014,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l234_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__90 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -26095,14 +26095,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l239_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__91 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -26176,14 +26176,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l244_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__92 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -26270,14 +26270,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l248_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__93 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -26347,14 +26347,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l252_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__94 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -26441,14 +26441,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l256_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__95 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -26514,14 +26514,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l260_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__96 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -26587,14 +26587,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l264_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__97 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -26660,14 +26660,14 @@
 // CHECK-32-EX-SAME: () #[[ATTR8]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3foov_l268_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
 // CHECK-32-EX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__98 to i8*), i8* null, i8** [[TMP2]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
diff --git a/clang/test/OpenMP/nvptx_data_sharing.cpp b/clang/test/OpenMP/nvptx_data_sharing.cpp
--- a/clang/test/OpenMP/nvptx_data_sharing.cpp
+++ b/clang/test/OpenMP/nvptx_data_sharing.cpp
@@ -36,13 +36,13 @@
 // CHECK-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 8
 // CHECK-NEXT:    [[C:%.*]] = alloca i32, align 4
 // CHECK-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [2 x ptr], align 8
-// CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true)
+// CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z7test_dsv_l14_kernel_environment)
 // CHECK-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK:       user_code.entry:
 // CHECK-NEXT:    [[A:%.*]] = call align 8 ptr @__kmpc_alloc_shared(i64 4)
 // CHECK-NEXT:    [[B:%.*]] = call align 8 ptr @__kmpc_alloc_shared(i64 4)
-// CHECK-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK-NEXT:    store i32 10, ptr [[A]], align 4
 // CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [1 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
 // CHECK-NEXT:    store ptr [[A]], ptr [[TMP2]], align 8
@@ -56,7 +56,7 @@
 // CHECK-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__1, ptr @__omp_outlined__1_wrapper, ptr [[CAPTURED_VARS_ADDRS1]], i64 2)
 // CHECK-NEXT:    call void @__kmpc_free_shared(ptr [[B]], i64 4)
 // CHECK-NEXT:    call void @__kmpc_free_shared(ptr [[A]], i64 4)
-// CHECK-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-NEXT:    ret void
 // CHECK:       worker.exit:
 // CHECK-NEXT:    ret void
diff --git a/clang/test/OpenMP/nvptx_distribute_parallel_generic_mode_codegen.cpp b/clang/test/OpenMP/nvptx_distribute_parallel_generic_mode_codegen.cpp
--- a/clang/test/OpenMP/nvptx_distribute_parallel_generic_mode_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_distribute_parallel_generic_mode_codegen.cpp
@@ -44,18 +44,18 @@
 // CHECK4-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK4-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK4-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[D_ADDR]], align 8
-// CHECK4-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK4-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l19_kernel_environment)
 // CHECK4-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP4]], -1
 // CHECK4-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK4:       user_code.entry:
-// CHECK4-NEXT:    [[TMP5:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK4-NEXT:    [[TMP5:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK4-NEXT:    [[TMP6:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
 // CHECK4-NEXT:    store i32 [[TMP6]], ptr [[ARGC_CASTED]], align 4
 // CHECK4-NEXT:    [[TMP7:%.*]] = load i64, ptr [[ARGC_CASTED]], align 8
 // CHECK4-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK4-NEXT:    store i32 [[TMP5]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK4-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], i64 [[TMP7]], ptr [[TMP3]]) #[[ATTR5:[0-9]+]]
-// CHECK4-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK4-NEXT:    call void @__kmpc_target_deinit()
 // CHECK4-NEXT:    ret void
 // CHECK4:       worker.exit:
 // CHECK4-NEXT:    ret void
@@ -360,18 +360,18 @@
 // CHECK5-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK5-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK5-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[D_ADDR]], align 4
-// CHECK5-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK5-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l19_kernel_environment)
 // CHECK5-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP4]], -1
 // CHECK5-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK5:       user_code.entry:
-// CHECK5-NEXT:    [[TMP5:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK5-NEXT:    [[TMP5:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK5-NEXT:    [[TMP6:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
 // CHECK5-NEXT:    store i32 [[TMP6]], ptr [[ARGC_CASTED]], align 4
 // CHECK5-NEXT:    [[TMP7:%.*]] = load i32, ptr [[ARGC_CASTED]], align 4
 // CHECK5-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK5-NEXT:    store i32 [[TMP5]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK5-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], i32 [[TMP7]], ptr [[TMP3]]) #[[ATTR5:[0-9]+]]
-// CHECK5-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK5-NEXT:    call void @__kmpc_target_deinit()
 // CHECK5-NEXT:    ret void
 // CHECK5:       worker.exit:
 // CHECK5-NEXT:    ret void
diff --git a/clang/test/OpenMP/nvptx_multi_target_parallel_codegen.cpp b/clang/test/OpenMP/nvptx_multi_target_parallel_codegen.cpp
--- a/clang/test/OpenMP/nvptx_multi_target_parallel_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_multi_target_parallel_codegen.cpp
@@ -29,13 +29,13 @@
 // CHECK1-SAME: () #[[ATTR0:[0-9]+]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x ptr], align 8
-// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l21_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
-// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK1-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 0)
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -64,12 +64,12 @@
 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l23
 // CHECK1-SAME: () #[[ATTR5:[0-9]+]] {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l23_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
 // CHECK1-NEXT:    call void @_Z3usev() #[[ATTR8]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -105,13 +105,13 @@
 // CHECK2-SAME: () #[[ATTR0:[0-9]+]] {
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x ptr], align 4
-// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l21_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
-// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK2-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 0)
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -140,12 +140,12 @@
 // CHECK2-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l23
 // CHECK2-SAME: () #[[ATTR5:[0-9]+]] {
 // CHECK2-NEXT:  entry:
-// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l23_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
 // CHECK2-NEXT:    call void @_Z3usev() #[[ATTR8]]
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
diff --git a/clang/test/OpenMP/nvptx_nested_parallel_codegen.cpp b/clang/test/OpenMP/nvptx_nested_parallel_codegen.cpp
--- a/clang/test/OpenMP/nvptx_nested_parallel_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_nested_parallel_codegen.cpp
@@ -40,16 +40,16 @@
 // CHECK1-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 8
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true)
+// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l25_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
-// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK1-NEXT:    call void @_Z3usePi(ptr noundef [[TMP0]]) #[[ATTR7:[0-9]+]]
 // CHECK1-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[TMP3]], align 8
 // CHECK1-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP2]], i32 1, i32 2, i32 -1, ptr @__omp_outlined__, ptr @__omp_outlined___wrapper, ptr [[CAPTURED_VARS_ADDRS]], i64 1)
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -160,16 +160,16 @@
 // CHECK2-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 4
 // CHECK2-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true)
+// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l25_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
-// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK2-NEXT:    call void @_Z3usePi(ptr noundef [[TMP0]]) #[[ATTR7:[0-9]+]]
 // CHECK2-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
 // CHECK2-NEXT:    store ptr [[TMP0]], ptr [[TMP3]], align 4
 // CHECK2-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP2]], i32 1, i32 2, i32 -1, ptr @__omp_outlined__, ptr @__omp_outlined___wrapper, ptr [[CAPTURED_VARS_ADDRS]], i32 1)
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
diff --git a/clang/test/OpenMP/nvptx_parallel_codegen.cpp b/clang/test/OpenMP/nvptx_parallel_codegen.cpp
--- a/clang/test/OpenMP/nvptx_parallel_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_parallel_codegen.cpp
@@ -81,18 +81,18 @@
 // CHECK1-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x ptr], align 8
 // CHECK1-NEXT:    [[CAPTURED_VARS_ADDRS2:%.*]] = alloca [0 x ptr], align 8
 // CHECK1-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true)
+// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l26_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
-// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK1-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__, ptr @__omp_outlined___wrapper, ptr [[CAPTURED_VARS_ADDRS]], i64 0)
 // CHECK1-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP1]], i32 0, i32 -1, i32 -1, ptr @__omp_outlined__1, ptr @__omp_outlined__1_wrapper, ptr [[CAPTURED_VARS_ADDRS1]], i64 0)
 // CHECK1-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__2, ptr @__omp_outlined__2_wrapper, ptr [[CAPTURED_VARS_ADDRS2]], i64 0)
 // CHECK1-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A_ADDR]], align 4
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 1
 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[A_ADDR]], align 4
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -192,7 +192,7 @@
 // CHECK1-NEXT:    store i64 [[AA]], ptr [[AA_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l43_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
@@ -213,7 +213,7 @@
 // CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 // CHECK1-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP7]], 1
 // CHECK1-NEXT:    store i32 [[ADD3]], ptr [[ARRAYIDX]], align 4
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -255,7 +255,7 @@
 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
 // CHECK1-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 8
 // CHECK1-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l55_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
@@ -270,7 +270,7 @@
 // CHECK1-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
 // CHECK1-NEXT:    store i32 [[INC]], ptr [[A1]], align 4
 // CHECK1-NEXT:    call void @__kmpc_free_shared(ptr [[A1]], i64 4)
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -344,18 +344,18 @@
 // CHECK2-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x ptr], align 4
 // CHECK2-NEXT:    [[CAPTURED_VARS_ADDRS2:%.*]] = alloca [0 x ptr], align 4
 // CHECK2-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true)
+// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l26_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
-// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK2-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__, ptr @__omp_outlined___wrapper, ptr [[CAPTURED_VARS_ADDRS]], i32 0)
 // CHECK2-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP1]], i32 0, i32 -1, i32 -1, ptr @__omp_outlined__1, ptr @__omp_outlined__1_wrapper, ptr [[CAPTURED_VARS_ADDRS1]], i32 0)
 // CHECK2-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__2, ptr @__omp_outlined__2_wrapper, ptr [[CAPTURED_VARS_ADDRS2]], i32 0)
 // CHECK2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A_ADDR]], align 4
 // CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 1
 // CHECK2-NEXT:    store i32 [[ADD]], ptr [[A_ADDR]], align 4
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -455,7 +455,7 @@
 // CHECK2-NEXT:    store i32 [[AA]], ptr [[AA_ADDR]], align 4
 // CHECK2-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l43_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
@@ -476,7 +476,7 @@
 // CHECK2-NEXT:    [[TMP7:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 // CHECK2-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP7]], 1
 // CHECK2-NEXT:    store i32 [[ADD3]], ptr [[ARRAYIDX]], align 4
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -518,7 +518,7 @@
 // CHECK2-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 4
 // CHECK2-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l55_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
@@ -533,7 +533,7 @@
 // CHECK2-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
 // CHECK2-NEXT:    store i32 [[INC]], ptr [[A1]], align 4
 // CHECK2-NEXT:    call void @__kmpc_free_shared(ptr [[A1]], i32 4)
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
diff --git a/clang/test/OpenMP/nvptx_parallel_for_codegen.cpp b/clang/test/OpenMP/nvptx_parallel_for_codegen.cpp
--- a/clang/test/OpenMP/nvptx_parallel_for_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_parallel_for_codegen.cpp
@@ -41,12 +41,12 @@
 // CHECK-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
-// CHECK-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true)
+// CHECK-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l13_kernel_environment)
 // CHECK-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK:       user_code.entry:
 // CHECK-NEXT:    [[D:%.*]] = call align 8 ptr @__kmpc_alloc_shared(i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
 // CHECK-NEXT:    store i32 [[TMP3]], ptr [[D]], align 4
 // CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [2 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
@@ -59,7 +59,7 @@
 // CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP6]], 1
 // CHECK-NEXT:    store i32 [[ADD]], ptr [[ARRAYIDX]], align 4
 // CHECK-NEXT:    call void @__kmpc_free_shared(ptr [[D]], i64 4)
-// CHECK-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-NEXT:    ret void
 // CHECK:       worker.exit:
 // CHECK-NEXT:    ret void
diff --git a/clang/test/OpenMP/nvptx_target_codegen.cpp b/clang/test/OpenMP/nvptx_target_codegen.cpp
--- a/clang/test/OpenMP/nvptx_target_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_target_codegen.cpp
@@ -153,17 +153,17 @@
 // CHECK1-NEXT:    store ptr [[PTR1]], ptr [[PTR1_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[PTR2]], ptr [[PTR2_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[PTR2_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9targetBarPiS__l25_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
-// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK1-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [2 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
 // CHECK1-NEXT:    store ptr [[PTR1_ADDR]], ptr [[TMP3]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [2 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[TMP4]], align 8
 // CHECK1-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP2]], i32 1, i32 2, i32 -1, ptr @__omp_outlined__, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 2)
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -192,11 +192,11 @@
 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l39
 // CHECK1-SAME: () #[[ATTR4:[0-9]+]] {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l39_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -207,7 +207,7 @@
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[AA_ADDR:%.*]] = alloca i64, align 8
 // CHECK1-NEXT:    store i64 [[AA]], ptr [[AA_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l47_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
@@ -221,7 +221,7 @@
 // CHECK1-NEXT:    [[ADD3:%.*]] = add nsw i32 [[CONV2]], 2
 // CHECK1-NEXT:    [[CONV4:%.*]] = trunc i32 [[ADD3]] to i16
 // CHECK1-NEXT:    store i16 [[CONV4]], ptr [[AA_ADDR]], align 2
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -256,7 +256,7 @@
 // CHECK1-NEXT:    [[TMP5:%.*]] = load i64, ptr [[VLA_ADDR4]], align 8
 // CHECK1-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[CN_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[D_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP8:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK1-NEXT:    [[TMP8:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l53_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP8]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
@@ -300,7 +300,7 @@
 // CHECK1-NEXT:    [[TMP17:%.*]] = load i64, ptr [[CALL]], align 8
 // CHECK1-NEXT:    [[ADD21:%.*]] = add nsw i64 [[TMP17]], 1
 // CHECK1-NEXT:    store i64 [[ADD21]], ptr [[CALL]], align 8
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -330,7 +330,7 @@
 // CHECK1-NEXT:    store i64 [[AAA]], ptr [[AAA_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l90_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
@@ -351,7 +351,7 @@
 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 // CHECK1-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP5]], 1
 // CHECK1-NEXT:    store i32 [[ADD6]], ptr [[ARRAYIDX]], align 4
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -374,7 +374,7 @@
 // CHECK1-NEXT:    [[TMP1:%.*]] = load i64, ptr [[VLA_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load i64, ptr [[VLA_ADDR2]], align 8
 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK1-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l108_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP4]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
@@ -397,7 +397,7 @@
 // CHECK1-NEXT:    [[CONV7:%.*]] = fptosi double [[TMP8]] to i32
 // CHECK1-NEXT:    [[A8:%.*]] = getelementptr inbounds [[STRUCT_S1]], ptr [[TMP0]], i32 0, i32 0
 // CHECK1-NEXT:    [[CALL:%.*]] = call i32 @_Z3baziRd(i32 [[CONV7]], ptr nonnull align 8 dereferenceable(8) [[A8]]) #[[ATTR10]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -426,7 +426,7 @@
 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16unreachable_callv_l142
 // CHECK1-SAME: () #[[ATTR4]] {
 // CHECK1-NEXT:  entry:
-// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16unreachable_callv_l142_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
@@ -435,7 +435,7 @@
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
 // CHECK1:       1:
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 //
 //
@@ -449,7 +449,7 @@
 // CHECK1-NEXT:    store i64 [[AA]], ptr [[AA_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l74_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
@@ -465,7 +465,7 @@
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 // CHECK1-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP4]], 1
 // CHECK1-NEXT:    store i32 [[ADD3]], ptr [[ARRAYIDX]], align 4
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -523,17 +523,17 @@
 // CHECK2-NEXT:    store ptr [[PTR1]], ptr [[PTR1_ADDR]], align 4
 // CHECK2-NEXT:    store ptr [[PTR2]], ptr [[PTR2_ADDR]], align 4
 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[PTR2_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9targetBarPiS__l25_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
-// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK2-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [2 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
 // CHECK2-NEXT:    store ptr [[PTR1_ADDR]], ptr [[TMP3]], align 4
 // CHECK2-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [2 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
 // CHECK2-NEXT:    store ptr [[TMP0]], ptr [[TMP4]], align 4
 // CHECK2-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP2]], i32 1, i32 2, i32 -1, ptr @__omp_outlined__, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 2)
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -562,11 +562,11 @@
 // CHECK2-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l39
 // CHECK2-SAME: () #[[ATTR4:[0-9]+]] {
 // CHECK2-NEXT:  entry:
-// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l39_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -577,7 +577,7 @@
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[AA_ADDR:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    store i32 [[AA]], ptr [[AA_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l47_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
@@ -591,7 +591,7 @@
 // CHECK2-NEXT:    [[ADD3:%.*]] = add nsw i32 [[CONV2]], 2
 // CHECK2-NEXT:    [[CONV4:%.*]] = trunc i32 [[ADD3]] to i16
 // CHECK2-NEXT:    store i16 [[CONV4]], ptr [[AA_ADDR]], align 2
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -626,7 +626,7 @@
 // CHECK2-NEXT:    [[TMP5:%.*]] = load i32, ptr [[VLA_ADDR4]], align 4
 // CHECK2-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[CN_ADDR]], align 4
 // CHECK2-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[D_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP8:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK2-NEXT:    [[TMP8:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l53_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP8]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
@@ -670,7 +670,7 @@
 // CHECK2-NEXT:    [[TMP17:%.*]] = load i64, ptr [[CALL]], align 8
 // CHECK2-NEXT:    [[ADD21:%.*]] = add nsw i64 [[TMP17]], 1
 // CHECK2-NEXT:    store i64 [[ADD21]], ptr [[CALL]], align 8
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -700,7 +700,7 @@
 // CHECK2-NEXT:    store i32 [[AAA]], ptr [[AAA_ADDR]], align 4
 // CHECK2-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l90_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
@@ -721,7 +721,7 @@
 // CHECK2-NEXT:    [[TMP5:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 // CHECK2-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP5]], 1
 // CHECK2-NEXT:    store i32 [[ADD6]], ptr [[ARRAYIDX]], align 4
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -744,7 +744,7 @@
 // CHECK2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[VLA_ADDR]], align 4
 // CHECK2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[VLA_ADDR2]], align 4
 // CHECK2-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK2-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l108_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP4]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
@@ -767,7 +767,7 @@
 // CHECK2-NEXT:    [[CONV7:%.*]] = fptosi double [[TMP8]] to i32
 // CHECK2-NEXT:    [[A8:%.*]] = getelementptr inbounds [[STRUCT_S1]], ptr [[TMP0]], i32 0, i32 0
 // CHECK2-NEXT:    [[CALL:%.*]] = call i32 @_Z3baziRd(i32 [[CONV7]], ptr nonnull align 8 dereferenceable(8) [[A8]]) #[[ATTR10]]
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -796,7 +796,7 @@
 // CHECK2-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16unreachable_callv_l142
 // CHECK2-SAME: () #[[ATTR4]] {
 // CHECK2-NEXT:  entry:
-// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z16unreachable_callv_l142_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
@@ -805,7 +805,7 @@
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
 // CHECK2:       1:
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 //
 //
@@ -819,7 +819,7 @@
 // CHECK2-NEXT:    store i32 [[AA]], ptr [[AA_ADDR]], align 4
 // CHECK2-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l74_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
@@ -835,7 +835,7 @@
 // CHECK2-NEXT:    [[TMP4:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 // CHECK2-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP4]], 1
 // CHECK2-NEXT:    store i32 [[ADD3]], ptr [[ARRAYIDX]], align 4
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
diff --git a/clang/test/OpenMP/nvptx_target_parallel_codegen.cpp b/clang/test/OpenMP/nvptx_target_parallel_codegen.cpp
--- a/clang/test/OpenMP/nvptx_target_parallel_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_target_parallel_codegen.cpp
@@ -58,15 +58,15 @@
 // CHECK1-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 8
 // CHECK1-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l30_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
-// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK1-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[TMP3]], align 8
 // CHECK1-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 1)
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -103,7 +103,7 @@
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[AA_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK1-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l35_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP3]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
@@ -115,7 +115,7 @@
 // CHECK1-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [3 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
 // CHECK1-NEXT:    store ptr [[TMP2]], ptr [[TMP7]], align 8
 // CHECK1-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP4]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__1, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 3)
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -159,15 +159,15 @@
 // CHECK2-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 4
 // CHECK2-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 4
 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l30_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
-// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK2-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
 // CHECK2-NEXT:    store ptr [[TMP0]], ptr [[TMP3]], align 4
 // CHECK2-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 1)
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -204,7 +204,7 @@
 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK2-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
 // CHECK2-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK2-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l35_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP3]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
@@ -216,7 +216,7 @@
 // CHECK2-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [3 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
 // CHECK2-NEXT:    store ptr [[TMP2]], ptr [[TMP7]], align 4
 // CHECK2-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP4]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__1, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 3)
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
diff --git a/clang/test/OpenMP/nvptx_target_parallel_num_threads_codegen.cpp b/clang/test/OpenMP/nvptx_target_parallel_num_threads_codegen.cpp
--- a/clang/test/OpenMP/nvptx_target_parallel_num_threads_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_target_parallel_num_threads_codegen.cpp
@@ -53,15 +53,15 @@
 // CHECK1-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 8
 // CHECK1-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l25_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
-// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK1-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[TMP3]], align 8
 // CHECK1-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP2]], i32 1, i32 1024, i32 -1, ptr @__omp_outlined__, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 1)
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -100,7 +100,7 @@
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[AA_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK1-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l30_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP3]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
@@ -113,7 +113,7 @@
 // CHECK1-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [3 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
 // CHECK1-NEXT:    store ptr [[TMP2]], ptr [[TMP8]], align 8
 // CHECK1-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP4]], i32 1, i32 [[TMP5]], i32 -1, ptr @__omp_outlined__1, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 3)
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -157,15 +157,15 @@
 // CHECK2-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 4
 // CHECK2-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 4
 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l25_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
-// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK2-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
 // CHECK2-NEXT:    store ptr [[TMP0]], ptr [[TMP3]], align 4
 // CHECK2-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP2]], i32 1, i32 1024, i32 -1, ptr @__omp_outlined__, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 1)
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -204,7 +204,7 @@
 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK2-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
 // CHECK2-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK2-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l30_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP3]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
@@ -217,7 +217,7 @@
 // CHECK2-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [3 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
 // CHECK2-NEXT:    store ptr [[TMP2]], ptr [[TMP8]], align 4
 // CHECK2-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP4]], i32 1, i32 [[TMP5]], i32 -1, ptr @__omp_outlined__1, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 3)
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
diff --git a/clang/test/OpenMP/nvptx_target_parallel_proc_bind_codegen.cpp b/clang/test/OpenMP/nvptx_target_parallel_proc_bind_codegen.cpp
--- a/clang/test/OpenMP/nvptx_target_parallel_proc_bind_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_target_parallel_proc_bind_codegen.cpp
@@ -56,13 +56,13 @@
 // CHECK45-64-SAME: () #[[ATTR0:[0-9]+]] {
 // CHECK45-64-NEXT:  entry:
 // CHECK45-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x ptr], align 8
-// CHECK45-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK45-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l27_kernel_environment)
 // CHECK45-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK45-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-64:       user_code.entry:
-// CHECK45-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK45-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK45-64-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 0)
-// CHECK45-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-64-NEXT:    ret void
 // CHECK45-64:       worker.exit:
 // CHECK45-64-NEXT:    ret void
@@ -85,7 +85,7 @@
 // CHECK45-64-NEXT:    [[AA_CASTED:%.*]] = alloca i64, align 8
 // CHECK45-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 8
 // CHECK45-64-NEXT:    store i64 [[AA]], ptr [[AA_ADDR]], align 8
-// CHECK45-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK45-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l31_kernel_environment)
 // CHECK45-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK45-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-64:       user_code.entry:
@@ -97,7 +97,7 @@
 // CHECK45-64-NEXT:    [[TMP5:%.*]] = inttoptr i64 [[TMP3]] to ptr
 // CHECK45-64-NEXT:    store ptr [[TMP5]], ptr [[TMP4]], align 8
 // CHECK45-64-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__1, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 1)
-// CHECK45-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-64-NEXT:    ret void
 // CHECK45-64:       worker.exit:
 // CHECK45-64-NEXT:    ret void
@@ -133,7 +133,7 @@
 // CHECK45-64-NEXT:    store i64 [[AA]], ptr [[AA_ADDR]], align 8
 // CHECK45-64-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK45-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
-// CHECK45-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK45-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l36_kernel_environment)
 // CHECK45-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK45-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-64:       user_code.entry:
@@ -153,7 +153,7 @@
 // CHECK45-64-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [3 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
 // CHECK45-64-NEXT:    store ptr [[TMP0]], ptr [[TMP11]], align 8
 // CHECK45-64-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__2, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 3)
-// CHECK45-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-64-NEXT:    ret void
 // CHECK45-64:       worker.exit:
 // CHECK45-64-NEXT:    ret void
@@ -192,13 +192,13 @@
 // CHECK45-32-SAME: () #[[ATTR0:[0-9]+]] {
 // CHECK45-32-NEXT:  entry:
 // CHECK45-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x ptr], align 4
-// CHECK45-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK45-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l27_kernel_environment)
 // CHECK45-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK45-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-32:       user_code.entry:
-// CHECK45-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK45-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK45-32-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 0)
-// CHECK45-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-32-NEXT:    ret void
 // CHECK45-32:       worker.exit:
 // CHECK45-32-NEXT:    ret void
@@ -221,7 +221,7 @@
 // CHECK45-32-NEXT:    [[AA_CASTED:%.*]] = alloca i32, align 4
 // CHECK45-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 4
 // CHECK45-32-NEXT:    store i32 [[AA]], ptr [[AA_ADDR]], align 4
-// CHECK45-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK45-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l31_kernel_environment)
 // CHECK45-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK45-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-32:       user_code.entry:
@@ -233,7 +233,7 @@
 // CHECK45-32-NEXT:    [[TMP5:%.*]] = inttoptr i32 [[TMP3]] to ptr
 // CHECK45-32-NEXT:    store ptr [[TMP5]], ptr [[TMP4]], align 4
 // CHECK45-32-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__1, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 1)
-// CHECK45-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-32-NEXT:    ret void
 // CHECK45-32:       worker.exit:
 // CHECK45-32-NEXT:    ret void
@@ -269,7 +269,7 @@
 // CHECK45-32-NEXT:    store i32 [[AA]], ptr [[AA_ADDR]], align 4
 // CHECK45-32-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK45-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
-// CHECK45-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK45-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l36_kernel_environment)
 // CHECK45-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK45-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-32:       user_code.entry:
@@ -289,7 +289,7 @@
 // CHECK45-32-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [3 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
 // CHECK45-32-NEXT:    store ptr [[TMP0]], ptr [[TMP11]], align 4
 // CHECK45-32-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__2, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 3)
-// CHECK45-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-32-NEXT:    ret void
 // CHECK45-32:       worker.exit:
 // CHECK45-32-NEXT:    ret void
@@ -328,13 +328,13 @@
 // CHECK45-32-EX-SAME: () #[[ATTR0:[0-9]+]] {
 // CHECK45-32-EX-NEXT:  entry:
 // CHECK45-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x ptr], align 4
-// CHECK45-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK45-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l27_kernel_environment)
 // CHECK45-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK45-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-32-EX:       user_code.entry:
-// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK45-32-EX-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 0)
-// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-32-EX-NEXT:    ret void
 // CHECK45-32-EX:       worker.exit:
 // CHECK45-32-EX-NEXT:    ret void
@@ -357,7 +357,7 @@
 // CHECK45-32-EX-NEXT:    [[AA_CASTED:%.*]] = alloca i32, align 4
 // CHECK45-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 4
 // CHECK45-32-EX-NEXT:    store i32 [[AA]], ptr [[AA_ADDR]], align 4
-// CHECK45-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK45-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l31_kernel_environment)
 // CHECK45-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK45-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-32-EX:       user_code.entry:
@@ -369,7 +369,7 @@
 // CHECK45-32-EX-NEXT:    [[TMP5:%.*]] = inttoptr i32 [[TMP3]] to ptr
 // CHECK45-32-EX-NEXT:    store ptr [[TMP5]], ptr [[TMP4]], align 4
 // CHECK45-32-EX-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__1, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 1)
-// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-32-EX-NEXT:    ret void
 // CHECK45-32-EX:       worker.exit:
 // CHECK45-32-EX-NEXT:    ret void
@@ -405,7 +405,7 @@
 // CHECK45-32-EX-NEXT:    store i32 [[AA]], ptr [[AA_ADDR]], align 4
 // CHECK45-32-EX-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK45-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
-// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l36_kernel_environment)
 // CHECK45-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK45-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-32-EX:       user_code.entry:
@@ -425,7 +425,7 @@
 // CHECK45-32-EX-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [3 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
 // CHECK45-32-EX-NEXT:    store ptr [[TMP0]], ptr [[TMP11]], align 4
 // CHECK45-32-EX-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__2, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 3)
-// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-32-EX-NEXT:    ret void
 // CHECK45-32-EX:       worker.exit:
 // CHECK45-32-EX-NEXT:    ret void
@@ -464,13 +464,13 @@
 // CHECK-64-SAME: () #[[ATTR0:[0-9]+]] {
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x ptr], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l27_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
-// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -493,7 +493,7 @@
 // CHECK-64-NEXT:    [[AA_CASTED:%.*]] = alloca i64, align 8
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 8
 // CHECK-64-NEXT:    store i64 [[AA]], ptr [[AA_ADDR]], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l31_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -505,7 +505,7 @@
 // CHECK-64-NEXT:    [[TMP5:%.*]] = inttoptr i64 [[TMP3]] to ptr
 // CHECK-64-NEXT:    store ptr [[TMP5]], ptr [[TMP4]], align 8
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__1, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 1)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -541,7 +541,7 @@
 // CHECK-64-NEXT:    store i64 [[AA]], ptr [[AA_ADDR]], align 8
 // CHECK-64-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
-// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l36_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -561,7 +561,7 @@
 // CHECK-64-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [3 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
 // CHECK-64-NEXT:    store ptr [[TMP0]], ptr [[TMP11]], align 8
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__2, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 3)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -596,153 +596,17 @@
 // CHECK-64-NEXT:    ret void
 //
 //
-// CHECK-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l27
-// CHECK-SAME: () #[[ATTR0:[0-9]+]] {
-// CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x ptr], align 4
-// CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
-// CHECK-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
-// CHECK-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
-// CHECK:       user_code.entry:
-// CHECK-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
-// CHECK-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 0)
-// CHECK-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
-// CHECK-NEXT:    ret void
-// CHECK:       worker.exit:
-// CHECK-NEXT:    ret void
-//
-//
-// CHECK-LABEL: define {{[^@]+}}@__omp_outlined__
-// CHECK-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]]) #[[ATTR1:[0-9]+]] {
-// CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
-// CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
-// CHECK-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
-// CHECK-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
-// CHECK-NEXT:    ret void
-//
-//
-// CHECK-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l31
-// CHECK-SAME: (i32 noundef [[AA:%.*]]) #[[ATTR0]] {
-// CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[AA_ADDR:%.*]] = alloca i32, align 4
-// CHECK-NEXT:    [[AA_CASTED:%.*]] = alloca i32, align 4
-// CHECK-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 4
-// CHECK-NEXT:    store i32 [[AA]], ptr [[AA_ADDR]], align 4
-// CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
-// CHECK-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
-// CHECK-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
-// CHECK:       user_code.entry:
-// CHECK-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
-// CHECK-NEXT:    [[TMP2:%.*]] = load i16, ptr [[AA_ADDR]], align 2
-// CHECK-NEXT:    store i16 [[TMP2]], ptr [[AA_CASTED]], align 2
-// CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[AA_CASTED]], align 4
-// CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
-// CHECK-NEXT:    [[TMP5:%.*]] = inttoptr i32 [[TMP3]] to ptr
-// CHECK-NEXT:    store ptr [[TMP5]], ptr [[TMP4]], align 4
-// CHECK-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__1, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 1)
-// CHECK-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
-// CHECK-NEXT:    ret void
-// CHECK:       worker.exit:
-// CHECK-NEXT:    ret void
-//
-//
-// CHECK-LABEL: define {{[^@]+}}@__omp_outlined__1
-// CHECK-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[AA:%.*]]) #[[ATTR1]] {
-// CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
-// CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
-// CHECK-NEXT:    [[AA_ADDR:%.*]] = alloca i32, align 4
-// CHECK-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
-// CHECK-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
-// CHECK-NEXT:    store i32 [[AA]], ptr [[AA_ADDR]], align 4
-// CHECK-NEXT:    [[TMP0:%.*]] = load i16, ptr [[AA_ADDR]], align 2
-// CHECK-NEXT:    [[CONV:%.*]] = sext i16 [[TMP0]] to i32
-// CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 [[CONV]], 1
-// CHECK-NEXT:    [[CONV1:%.*]] = trunc i32 [[ADD]] to i16
-// CHECK-NEXT:    store i16 [[CONV1]], ptr [[AA_ADDR]], align 2
-// CHECK-NEXT:    ret void
-//
-//
-// CHECK-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l36
-// CHECK-SAME: (i32 noundef [[A:%.*]], i32 noundef [[AA:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
-// CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
-// CHECK-NEXT:    [[AA_ADDR:%.*]] = alloca i32, align 4
-// CHECK-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
-// CHECK-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
-// CHECK-NEXT:    [[AA_CASTED:%.*]] = alloca i32, align 4
-// CHECK-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x ptr], align 4
-// CHECK-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
-// CHECK-NEXT:    store i32 [[AA]], ptr [[AA_ADDR]], align 4
-// CHECK-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
-// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
-// CHECK-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
-// CHECK-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
-// CHECK-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
-// CHECK:       user_code.entry:
-// CHECK-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A_ADDR]], align 4
-// CHECK-NEXT:    store i32 [[TMP3]], ptr [[A_CASTED]], align 4
-// CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A_CASTED]], align 4
-// CHECK-NEXT:    [[TMP5:%.*]] = load i16, ptr [[AA_ADDR]], align 2
-// CHECK-NEXT:    store i16 [[TMP5]], ptr [[AA_CASTED]], align 2
-// CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[AA_CASTED]], align 4
-// CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [3 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
-// CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i32 [[TMP4]] to ptr
-// CHECK-NEXT:    store ptr [[TMP8]], ptr [[TMP7]], align 4
-// CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [3 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
-// CHECK-NEXT:    [[TMP10:%.*]] = inttoptr i32 [[TMP6]] to ptr
-// CHECK-NEXT:    store ptr [[TMP10]], ptr [[TMP9]], align 4
-// CHECK-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [3 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
-// CHECK-NEXT:    store ptr [[TMP0]], ptr [[TMP11]], align 4
-// CHECK-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__2, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 3)
-// CHECK-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
-// CHECK-NEXT:    ret void
-// CHECK:       worker.exit:
-// CHECK-NEXT:    ret void
-//
-//
-// CHECK-LABEL: define {{[^@]+}}@__omp_outlined__2
-// CHECK-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[A:%.*]], i32 noundef [[AA:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR1]] {
-// CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
-// CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
-// CHECK-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
-// CHECK-NEXT:    [[AA_ADDR:%.*]] = alloca i32, align 4
-// CHECK-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
-// CHECK-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
-// CHECK-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
-// CHECK-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
-// CHECK-NEXT:    store i32 [[AA]], ptr [[AA_ADDR]], align 4
-// CHECK-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
-// CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
-// CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[A_ADDR]], align 4
-// CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP1]], 1
-// CHECK-NEXT:    store i32 [[ADD]], ptr [[A_ADDR]], align 4
-// CHECK-NEXT:    [[TMP2:%.*]] = load i16, ptr [[AA_ADDR]], align 2
-// CHECK-NEXT:    [[CONV:%.*]] = sext i16 [[TMP2]] to i32
-// CHECK-NEXT:    [[ADD1:%.*]] = add nsw i32 [[CONV]], 1
-// CHECK-NEXT:    [[CONV2:%.*]] = trunc i32 [[ADD1]] to i16
-// CHECK-NEXT:    store i16 [[CONV2]], ptr [[AA_ADDR]], align 2
-// CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i32 0, i32 2
-// CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-// CHECK-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP3]], 1
-// CHECK-NEXT:    store i32 [[ADD3]], ptr [[ARRAYIDX]], align 4
-// CHECK-NEXT:    ret void
-//
-//
 // CHECK-32-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l27
 // CHECK-32-SAME: () #[[ATTR0:[0-9]+]] {
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x ptr], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l27_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
-// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -765,7 +629,7 @@
 // CHECK-32-NEXT:    [[AA_CASTED:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 4
 // CHECK-32-NEXT:    store i32 [[AA]], ptr [[AA_ADDR]], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l31_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -777,7 +641,7 @@
 // CHECK-32-NEXT:    [[TMP5:%.*]] = inttoptr i32 [[TMP3]] to ptr
 // CHECK-32-NEXT:    store ptr [[TMP5]], ptr [[TMP4]], align 4
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__1, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 1)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -813,7 +677,7 @@
 // CHECK-32-NEXT:    store i32 [[AA]], ptr [[AA_ADDR]], align 4
 // CHECK-32-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
-// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l36_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -833,7 +697,7 @@
 // CHECK-32-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [3 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
 // CHECK-32-NEXT:    store ptr [[TMP0]], ptr [[TMP11]], align 4
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__2, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 3)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -872,13 +736,13 @@
 // CHECK-32-EX-SAME: () #[[ATTR0:[0-9]+]] {
 // CHECK-32-EX-NEXT:  entry:
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x ptr], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l27_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
-// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 0)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -901,7 +765,7 @@
 // CHECK-32-EX-NEXT:    [[AA_CASTED:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 4
 // CHECK-32-EX-NEXT:    store i32 [[AA]], ptr [[AA_ADDR]], align 4
-// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l31_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -913,7 +777,7 @@
 // CHECK-32-EX-NEXT:    [[TMP5:%.*]] = inttoptr i32 [[TMP3]] to ptr
 // CHECK-32-EX-NEXT:    store ptr [[TMP5]], ptr [[TMP4]], align 4
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__1, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 1)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -949,7 +813,7 @@
 // CHECK-32-EX-NEXT:    store i32 [[AA]], ptr [[AA_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
-// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l36_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -969,7 +833,7 @@
 // CHECK-32-EX-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [3 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
 // CHECK-32-EX-NEXT:    store ptr [[TMP0]], ptr [[TMP11]], align 4
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__2, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 3)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
diff --git a/clang/test/OpenMP/nvptx_target_parallel_reduction_codegen.cpp b/clang/test/OpenMP/nvptx_target_parallel_reduction_codegen.cpp
--- a/clang/test/OpenMP/nvptx_target_parallel_reduction_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_target_parallel_reduction_codegen.cpp
@@ -104,17 +104,17 @@
 // CHECK-64-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x i8*], align 8
 // CHECK-64-NEXT:    store double* [[E]], double** [[E_ADDR]], align 8
 // CHECK-64-NEXT:    [[TMP0:%.*]] = load double*, double** [[E_ADDR]], align 8
-// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l24_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
-// CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
+// CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]])
 // CHECK-64-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
 // CHECK-64-NEXT:    [[TMP4:%.*]] = bitcast double* [[TMP0]] to i8*
 // CHECK-64-NEXT:    store i8* [[TMP4]], i8** [[TMP3]], align 8
 // CHECK-64-NEXT:    [[TMP5:%.*]] = bitcast [1 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, double*)* @__omp_outlined__ to i8*), i8* null, i8** [[TMP5]], i64 1)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -303,7 +303,7 @@
 // CHECK-64-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
 // CHECK-64-NEXT:    [[TMP0:%.*]] = load i8*, i8** [[C_ADDR]], align 8
 // CHECK-64-NEXT:    [[TMP1:%.*]] = load float*, float** [[D_ADDR]], align 8
-// CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP2]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -315,7 +315,7 @@
 // CHECK-64-NEXT:    store i8* [[TMP6]], i8** [[TMP5]], align 8
 // CHECK-64-NEXT:    [[TMP7:%.*]] = bitcast [2 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i8*, float*)* @__omp_outlined__1 to i8*), i8* null, i8** [[TMP7]], i64 2)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -559,7 +559,7 @@
 // CHECK-64-NEXT:    store i16* [[B]], i16** [[B_ADDR]], align 8
 // CHECK-64-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 8
 // CHECK-64-NEXT:    [[TMP1:%.*]] = load i16*, i16** [[B_ADDR]], align 8
-// CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l35_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP2]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -572,7 +572,7 @@
 // CHECK-64-NEXT:    store i8* [[TMP7]], i8** [[TMP6]], align 8
 // CHECK-64-NEXT:    [[TMP8:%.*]] = bitcast [2 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-64-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32*, i16*)* @__omp_outlined__5 to i8*), i8* null, i8** [[TMP8]], i64 2)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -836,17 +836,17 @@
 // CHECK-32-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x i8*], align 4
 // CHECK-32-NEXT:    store double* [[E]], double** [[E_ADDR]], align 4
 // CHECK-32-NEXT:    [[TMP0:%.*]] = load double*, double** [[E_ADDR]], align 4
-// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l24_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
-// CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
+// CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]])
 // CHECK-32-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
 // CHECK-32-NEXT:    [[TMP4:%.*]] = bitcast double* [[TMP0]] to i8*
 // CHECK-32-NEXT:    store i8* [[TMP4]], i8** [[TMP3]], align 4
 // CHECK-32-NEXT:    [[TMP5:%.*]] = bitcast [1 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, double*)* @__omp_outlined__ to i8*), i8* null, i8** [[TMP5]], i32 1)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -1035,7 +1035,7 @@
 // CHECK-32-NEXT:    store float* [[D]], float** [[D_ADDR]], align 4
 // CHECK-32-NEXT:    [[TMP0:%.*]] = load i8*, i8** [[C_ADDR]], align 4
 // CHECK-32-NEXT:    [[TMP1:%.*]] = load float*, float** [[D_ADDR]], align 4
-// CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP2]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -1047,7 +1047,7 @@
 // CHECK-32-NEXT:    store i8* [[TMP6]], i8** [[TMP5]], align 4
 // CHECK-32-NEXT:    [[TMP7:%.*]] = bitcast [2 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i8*, float*)* @__omp_outlined__1 to i8*), i8* null, i8** [[TMP7]], i32 2)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -1291,7 +1291,7 @@
 // CHECK-32-NEXT:    store i16* [[B]], i16** [[B_ADDR]], align 4
 // CHECK-32-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 4
 // CHECK-32-NEXT:    [[TMP1:%.*]] = load i16*, i16** [[B_ADDR]], align 4
-// CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l35_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP2]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -1304,7 +1304,7 @@
 // CHECK-32-NEXT:    store i8* [[TMP7]], i8** [[TMP6]], align 4
 // CHECK-32-NEXT:    [[TMP8:%.*]] = bitcast [2 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32*, i16*)* @__omp_outlined__5 to i8*), i8* null, i8** [[TMP8]], i32 2)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -1568,17 +1568,17 @@
 // CHECK-32-EX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x i8*], align 4
 // CHECK-32-EX-NEXT:    store double* [[E]], double** [[E_ADDR]], align 4
 // CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load double*, double** [[E_ADDR]], align 4
-// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l24_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
-// CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
+// CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]])
 // CHECK-32-EX-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
 // CHECK-32-EX-NEXT:    [[TMP4:%.*]] = bitcast double* [[TMP0]] to i8*
 // CHECK-32-EX-NEXT:    store i8* [[TMP4]], i8** [[TMP3]], align 4
 // CHECK-32-EX-NEXT:    [[TMP5:%.*]] = bitcast [1 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, double*)* @__omp_outlined__ to i8*), i8* null, i8** [[TMP5]], i32 1)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -1767,7 +1767,7 @@
 // CHECK-32-EX-NEXT:    store float* [[D]], float** [[D_ADDR]], align 4
 // CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load i8*, i8** [[C_ADDR]], align 4
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = load float*, float** [[D_ADDR]], align 4
-// CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l29_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP2]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -1779,7 +1779,7 @@
 // CHECK-32-EX-NEXT:    store i8* [[TMP6]], i8** [[TMP5]], align 4
 // CHECK-32-EX-NEXT:    [[TMP7:%.*]] = bitcast [2 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i8*, float*)* @__omp_outlined__1 to i8*), i8* null, i8** [[TMP7]], i32 2)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -2023,7 +2023,7 @@
 // CHECK-32-EX-NEXT:    store i16* [[B]], i16** [[B_ADDR]], align 4
 // CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 4
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = load i16*, i16** [[B_ADDR]], align 4
-// CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(%struct.KernelEnvironmentTy* @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l35_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP2]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -2036,7 +2036,7 @@
 // CHECK-32-EX-NEXT:    store i8* [[TMP7]], i8** [[TMP6]], align 4
 // CHECK-32-EX-NEXT:    [[TMP8:%.*]] = bitcast [2 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
 // CHECK-32-EX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32*, i16*)* @__omp_outlined__5 to i8*), i8* null, i8** [[TMP8]], i32 2)
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
diff --git a/clang/test/OpenMP/nvptx_target_parallel_reduction_codegen_tbaa_PR46146.cpp b/clang/test/OpenMP/nvptx_target_parallel_reduction_codegen_tbaa_PR46146.cpp
--- a/clang/test/OpenMP/nvptx_target_parallel_reduction_codegen_tbaa_PR46146.cpp
+++ b/clang/test/OpenMP/nvptx_target_parallel_reduction_codegen_tbaa_PR46146.cpp
@@ -35,15 +35,15 @@
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true)
+// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z17complex_reductionIfEvv_l16_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
-// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK1-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK1-NEXT:    store i32 [[TMP1]], ptr [[DOTTHREADID_TEMP_]], align 4, !tbaa [[TBAA8:![0-9]+]]
 // CHECK1-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]]) #[[ATTR6:[0-9]+]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -434,7 +434,7 @@
 // CHECK1-NEXT:    [[TMP34:%.*]] = load ptr, ptr [[TMP33]], align 8
 // CHECK1-NEXT:    [[TMP35:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP4]], i64 0, i64 0
 // CHECK1-NEXT:    [[TMP36:%.*]] = load ptr, ptr [[TMP35]], align 8
-// CHECK1-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 4 [[TMP36]], ptr align 4 [[TMP34]], i64 8, i1 false), !tbaa.struct !21
+// CHECK1-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 4 [[TMP36]], ptr align 4 [[TMP34]], i64 8, i1 false), !tbaa.struct [[TBAA_STRUCT21:![0-9]+]]
 // CHECK1-NEXT:    br label [[IFCONT6:%.*]]
 // CHECK1:       else5:
 // CHECK1-NEXT:    br label [[IFCONT6]]
@@ -527,7 +527,7 @@
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z17complex_reductionIdEvv_l16_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
@@ -535,7 +535,7 @@
 // CHECK1-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK1-NEXT:    store i32 [[TMP1]], ptr [[DOTTHREADID_TEMP_]], align 4, !tbaa [[TBAA8]]
 // CHECK1-NEXT:    call void @__omp_outlined__2(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]]) #[[ATTR6]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -939,7 +939,7 @@
 // CHECK1-NEXT:    [[TMP41:%.*]] = load ptr, ptr [[TMP40]], align 8
 // CHECK1-NEXT:    [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[TMP4]], i64 0, i64 0
 // CHECK1-NEXT:    [[TMP43:%.*]] = load ptr, ptr [[TMP42]], align 8
-// CHECK1-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[TMP43]], ptr align 8 [[TMP41]], i64 16, i1 false), !tbaa.struct !27
+// CHECK1-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[TMP43]], ptr align 8 [[TMP41]], i64 16, i1 false), !tbaa.struct [[TBAA_STRUCT27:![0-9]+]]
 // CHECK1-NEXT:    br label [[IFCONT6:%.*]]
 // CHECK1:       else5:
 // CHECK1-NEXT:    br label [[IFCONT6]]
diff --git a/clang/test/OpenMP/nvptx_target_printf_codegen.c b/clang/test/OpenMP/nvptx_target_printf_codegen.c
--- a/clang/test/OpenMP/nvptx_target_printf_codegen.c
+++ b/clang/test/OpenMP/nvptx_target_printf_codegen.c
@@ -45,7 +45,7 @@
 // CHECK-64-NEXT:  entry:
 // CHECK-64-NEXT:    [[FMT:%.*]] = alloca ptr, align 8
 // CHECK-64-NEXT:    [[TMP:%.*]] = alloca [[PRINTF_ARGS:%.*]], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_CheckSimple_l13_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -58,7 +58,7 @@
 // CHECK-64-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[PRINTF_ARGS]], ptr [[TMP]], i32 0, i32 2
 // CHECK-64-NEXT:    store double 3.000000e+00, ptr [[TMP4]], align 8
 // CHECK-64-NEXT:    [[TMP5:%.*]] = call i32 @__llvm_omp_vprintf(ptr [[TMP1]], ptr [[TMP]], i32 24)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -67,12 +67,12 @@
 // CHECK-64-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_CheckNoArgs_l25
 // CHECK-64-SAME: () #[[ATTR0]] {
 // CHECK-64-NEXT:  entry:
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_CheckNoArgs_l25_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
 // CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__llvm_omp_vprintf(ptr @.str1, ptr null, i32 0)
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -84,7 +84,7 @@
 // CHECK-64-NEXT:    [[FOO_ADDR:%.*]] = alloca i64, align 8
 // CHECK-64-NEXT:    [[TMP:%.*]] = alloca [[PRINTF_ARGS_0:%.*]], align 8
 // CHECK-64-NEXT:    store i64 [[FOO]], ptr [[FOO_ADDR]], align 8
-// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_CheckAllocaIsInEntryBlock_l36_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -99,7 +99,7 @@
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       if.end:
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 //
 //
@@ -108,7 +108,7 @@
 // CHECK-32-NEXT:  entry:
 // CHECK-32-NEXT:    [[FMT:%.*]] = alloca ptr, align 4
 // CHECK-32-NEXT:    [[TMP:%.*]] = alloca [[PRINTF_ARGS:%.*]], align 8
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_CheckSimple_l13_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -121,7 +121,7 @@
 // CHECK-32-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[PRINTF_ARGS]], ptr [[TMP]], i32 0, i32 2
 // CHECK-32-NEXT:    store double 3.000000e+00, ptr [[TMP4]], align 8
 // CHECK-32-NEXT:    [[TMP5:%.*]] = call i32 @__llvm_omp_vprintf(ptr [[TMP1]], ptr [[TMP]], i32 24)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -130,12 +130,12 @@
 // CHECK-32-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_CheckNoArgs_l25
 // CHECK-32-SAME: () #[[ATTR0]] {
 // CHECK-32-NEXT:  entry:
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_CheckNoArgs_l25_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
 // CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__llvm_omp_vprintf(ptr @.str1, ptr null, i32 0)
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -147,7 +147,7 @@
 // CHECK-32-NEXT:    [[FOO_ADDR:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    [[TMP:%.*]] = alloca [[PRINTF_ARGS_0:%.*]], align 8
 // CHECK-32-NEXT:    store i32 [[FOO]], ptr [[FOO_ADDR]], align 4
-// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_CheckAllocaIsInEntryBlock_l36_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -162,6 +162,6 @@
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       if.end:
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 //
diff --git a/clang/test/OpenMP/nvptx_target_simd_codegen.cpp b/clang/test/OpenMP/nvptx_target_simd_codegen.cpp
--- a/clang/test/OpenMP/nvptx_target_simd_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_target_simd_codegen.cpp
@@ -72,7 +72,7 @@
 // CHECK45-64-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK45-64-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK45-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
-// CHECK45-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK45-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l29_kernel_environment)
 // CHECK45-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK45-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-64:       user_code.entry:
@@ -124,7 +124,7 @@
 // CHECK45-64-NEXT:    store i32 [[ADD10]], ptr [[I3]], align 4
 // CHECK45-64-NEXT:    br label [[SIMD_IF_END]]
 // CHECK45-64:       simd.if.end:
-// CHECK45-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-64-NEXT:    ret void
 //
 //
@@ -142,7 +142,7 @@
 // CHECK45-64-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK45-64-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 8
 // CHECK45-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 8
-// CHECK45-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK45-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_kernel_environment)
 // CHECK45-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK45-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-64:       user_code.entry:
@@ -198,7 +198,7 @@
 // CHECK45-64-NEXT:    store i32 [[ADD12]], ptr [[I3]], align 4
 // CHECK45-64-NEXT:    br label [[SIMD_IF_END]]
 // CHECK45-64:       simd.if.end:
-// CHECK45-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-64-NEXT:    ret void
 //
 //
@@ -211,7 +211,7 @@
 // CHECK45-64-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK45-64-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK45-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
-// CHECK45-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK45-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l39_kernel_environment)
 // CHECK45-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK45-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-64:       user_code.entry:
@@ -244,7 +244,7 @@
 // CHECK45-64-NEXT:    ret void
 // CHECK45-64:       omp.inner.for.end:
 // CHECK45-64-NEXT:    store i32 10, ptr [[I]], align 4
-// CHECK45-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-64-NEXT:    ret void
 //
 //
@@ -261,7 +261,7 @@
 // CHECK45-64-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK45-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK45-64-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[N_ADDR]], align 8
-// CHECK45-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK45-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l44_kernel_environment)
 // CHECK45-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP2]], -1
 // CHECK45-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-64:       user_code.entry:
@@ -299,7 +299,7 @@
 // CHECK45-64-NEXT:    [[TMP9:%.*]] = load i32, ptr [[N1]], align 4
 // CHECK45-64-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP8]], [[TMP9]]
 // CHECK45-64-NEXT:    store i32 [[ADD4]], ptr [[TMP1]], align 4
-// CHECK45-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-64-NEXT:    ret void
 //
 //
@@ -317,7 +317,7 @@
 // CHECK45-32-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK45-32-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK45-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
-// CHECK45-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK45-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l29_kernel_environment)
 // CHECK45-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK45-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-32:       user_code.entry:
@@ -368,7 +368,7 @@
 // CHECK45-32-NEXT:    store i32 [[ADD10]], ptr [[I3]], align 4
 // CHECK45-32-NEXT:    br label [[SIMD_IF_END]]
 // CHECK45-32:       simd.if.end:
-// CHECK45-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-32-NEXT:    ret void
 //
 //
@@ -386,7 +386,7 @@
 // CHECK45-32-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK45-32-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 4
 // CHECK45-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
-// CHECK45-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK45-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_kernel_environment)
 // CHECK45-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK45-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-32:       user_code.entry:
@@ -441,7 +441,7 @@
 // CHECK45-32-NEXT:    store i32 [[ADD12]], ptr [[I3]], align 4
 // CHECK45-32-NEXT:    br label [[SIMD_IF_END]]
 // CHECK45-32:       simd.if.end:
-// CHECK45-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-32-NEXT:    ret void
 //
 //
@@ -454,7 +454,7 @@
 // CHECK45-32-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK45-32-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK45-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
-// CHECK45-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK45-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l39_kernel_environment)
 // CHECK45-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK45-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-32:       user_code.entry:
@@ -486,7 +486,7 @@
 // CHECK45-32-NEXT:    ret void
 // CHECK45-32:       omp.inner.for.end:
 // CHECK45-32-NEXT:    store i32 10, ptr [[I]], align 4
-// CHECK45-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-32-NEXT:    ret void
 //
 //
@@ -503,7 +503,7 @@
 // CHECK45-32-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK45-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK45-32-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[N_ADDR]], align 4
-// CHECK45-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK45-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l44_kernel_environment)
 // CHECK45-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP2]], -1
 // CHECK45-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-32:       user_code.entry:
@@ -540,7 +540,7 @@
 // CHECK45-32-NEXT:    [[TMP9:%.*]] = load i32, ptr [[N1]], align 4
 // CHECK45-32-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP8]], [[TMP9]]
 // CHECK45-32-NEXT:    store i32 [[ADD4]], ptr [[TMP1]], align 4
-// CHECK45-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-32-NEXT:    ret void
 //
 //
@@ -558,7 +558,7 @@
 // CHECK45-32-EX-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK45-32-EX-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK45-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
-// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l29_kernel_environment)
 // CHECK45-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK45-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-32-EX:       user_code.entry:
@@ -609,7 +609,7 @@
 // CHECK45-32-EX-NEXT:    store i32 [[ADD10]], ptr [[I3]], align 4
 // CHECK45-32-EX-NEXT:    br label [[SIMD_IF_END]]
 // CHECK45-32-EX:       simd.if.end:
-// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-32-EX-NEXT:    ret void
 //
 //
@@ -627,7 +627,7 @@
 // CHECK45-32-EX-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK45-32-EX-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 4
 // CHECK45-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
-// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_kernel_environment)
 // CHECK45-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK45-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-32-EX:       user_code.entry:
@@ -682,7 +682,7 @@
 // CHECK45-32-EX-NEXT:    store i32 [[ADD12]], ptr [[I3]], align 4
 // CHECK45-32-EX-NEXT:    br label [[SIMD_IF_END]]
 // CHECK45-32-EX:       simd.if.end:
-// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-32-EX-NEXT:    ret void
 //
 //
@@ -695,7 +695,7 @@
 // CHECK45-32-EX-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK45-32-EX-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK45-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
-// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l39_kernel_environment)
 // CHECK45-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK45-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-32-EX:       user_code.entry:
@@ -727,7 +727,7 @@
 // CHECK45-32-EX-NEXT:    ret void
 // CHECK45-32-EX:       omp.inner.for.end:
 // CHECK45-32-EX-NEXT:    store i32 10, ptr [[I]], align 4
-// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-32-EX-NEXT:    ret void
 //
 //
@@ -744,7 +744,7 @@
 // CHECK45-32-EX-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK45-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[N_ADDR]], align 4
-// CHECK45-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK45-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l44_kernel_environment)
 // CHECK45-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP2]], -1
 // CHECK45-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-32-EX:       user_code.entry:
@@ -781,7 +781,7 @@
 // CHECK45-32-EX-NEXT:    [[TMP9:%.*]] = load i32, ptr [[N1]], align 4
 // CHECK45-32-EX-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP8]], [[TMP9]]
 // CHECK45-32-EX-NEXT:    store i32 [[ADD4]], ptr [[TMP1]], align 4
-// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-32-EX-NEXT:    ret void
 //
 //
@@ -799,7 +799,7 @@
 // CHECK-64-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK-64-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
-// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l29_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -851,7 +851,7 @@
 // CHECK-64-NEXT:    store i32 [[ADD10]], ptr [[I3]], align 4
 // CHECK-64-NEXT:    br label [[SIMD_IF_END]]
 // CHECK-64:       simd.if.end:
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 //
 //
@@ -869,7 +869,7 @@
 // CHECK-64-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK-64-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 8
 // CHECK-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 8
-// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -925,7 +925,7 @@
 // CHECK-64-NEXT:    store i32 [[ADD12]], ptr [[I3]], align 4
 // CHECK-64-NEXT:    br label [[SIMD_IF_END]]
 // CHECK-64:       simd.if.end:
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 //
 //
@@ -938,7 +938,7 @@
 // CHECK-64-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
-// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l39_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -971,7 +971,7 @@
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       omp.inner.for.end:
 // CHECK-64-NEXT:    store i32 10, ptr [[I]], align 4
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 //
 //
@@ -988,7 +988,7 @@
 // CHECK-64-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK-64-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[N_ADDR]], align 8
-// CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l44_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP2]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -1026,7 +1026,7 @@
 // CHECK-64-NEXT:    [[TMP9:%.*]] = load i32, ptr [[N1]], align 4
 // CHECK-64-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP8]], [[TMP9]]
 // CHECK-64-NEXT:    store i32 [[ADD4]], ptr [[TMP1]], align 4
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 //
 //
@@ -1044,7 +1044,7 @@
 // CHECK-32-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK-32-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
-// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l29_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -1095,7 +1095,7 @@
 // CHECK-32-NEXT:    store i32 [[ADD10]], ptr [[I3]], align 4
 // CHECK-32-NEXT:    br label [[SIMD_IF_END]]
 // CHECK-32:       simd.if.end:
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 //
 //
@@ -1113,7 +1113,7 @@
 // CHECK-32-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK-32-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 4
 // CHECK-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
-// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -1168,7 +1168,7 @@
 // CHECK-32-NEXT:    store i32 [[ADD12]], ptr [[I3]], align 4
 // CHECK-32-NEXT:    br label [[SIMD_IF_END]]
 // CHECK-32:       simd.if.end:
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 //
 //
@@ -1181,7 +1181,7 @@
 // CHECK-32-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
-// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l39_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -1213,7 +1213,7 @@
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       omp.inner.for.end:
 // CHECK-32-NEXT:    store i32 10, ptr [[I]], align 4
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 //
 //
@@ -1230,7 +1230,7 @@
 // CHECK-32-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK-32-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[N_ADDR]], align 4
-// CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l44_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP2]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -1267,7 +1267,7 @@
 // CHECK-32-NEXT:    [[TMP9:%.*]] = load i32, ptr [[N1]], align 4
 // CHECK-32-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP8]], [[TMP9]]
 // CHECK-32-NEXT:    store i32 [[ADD4]], ptr [[TMP1]], align 4
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 //
 //
@@ -1285,7 +1285,7 @@
 // CHECK-32-EX-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
-// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l29_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -1336,7 +1336,7 @@
 // CHECK-32-EX-NEXT:    store i32 [[ADD10]], ptr [[I3]], align 4
 // CHECK-32-EX-NEXT:    br label [[SIMD_IF_END]]
 // CHECK-32-EX:       simd.if.end:
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 //
 //
@@ -1354,7 +1354,7 @@
 // CHECK-32-EX-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 4
 // CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
-// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -1409,7 +1409,7 @@
 // CHECK-32-EX-NEXT:    store i32 [[ADD12]], ptr [[I3]], align 4
 // CHECK-32-EX-NEXT:    br label [[SIMD_IF_END]]
 // CHECK-32-EX:       simd.if.end:
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 //
 //
@@ -1422,7 +1422,7 @@
 // CHECK-32-EX-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
-// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l39_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -1454,7 +1454,7 @@
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       omp.inner.for.end:
 // CHECK-32-EX-NEXT:    store i32 10, ptr [[I]], align 4
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 //
 //
@@ -1471,7 +1471,7 @@
 // CHECK-32-EX-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK-32-EX-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[N_ADDR]], align 4
-// CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l44_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP2]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -1508,6 +1508,6 @@
 // CHECK-32-EX-NEXT:    [[TMP9:%.*]] = load i32, ptr [[N1]], align 4
 // CHECK-32-EX-NEXT:    [[ADD4:%.*]] = add nsw i32 [[TMP8]], [[TMP9]]
 // CHECK-32-EX-NEXT:    store i32 [[ADD4]], ptr [[TMP1]], align 4
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 //
diff --git a/clang/test/OpenMP/nvptx_target_teams_codegen.cpp b/clang/test/OpenMP/nvptx_target_teams_codegen.cpp
--- a/clang/test/OpenMP/nvptx_target_teams_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_target_teams_codegen.cpp
@@ -57,18 +57,18 @@
 // CHECK1-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true)
+// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l23_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
-// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK1-NEXT:    [[TMP2:%.*]] = load i8, ptr [[A_ADDR]], align 1
 // CHECK1-NEXT:    store i8 [[TMP2]], ptr [[A_CASTED]], align 1
 // CHECK1-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
 // CHECK1-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK1-NEXT:    store i32 [[TMP1]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK1-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i64 [[TMP3]]) #[[ATTR2:[0-9]+]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -95,7 +95,7 @@
 // CHECK1-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    store i64 [[AA]], ptr [[AA_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l28_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
@@ -106,7 +106,7 @@
 // CHECK1-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK1-NEXT:    store i32 [[TMP1]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK1-NEXT:    call void @__omp_outlined__1(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i64 [[TMP3]]) #[[ATTR2]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -133,7 +133,7 @@
 // CHECK1-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    store i64 [[AA]], ptr [[AA_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l33_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
@@ -144,7 +144,7 @@
 // CHECK1-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK1-NEXT:    store i32 [[TMP1]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK1-NEXT:    call void @__omp_outlined__2(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i64 [[TMP3]]) #[[ATTR2]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -209,18 +209,18 @@
 // CHECK2-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true)
+// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l23_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
-// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK2-NEXT:    [[TMP2:%.*]] = load i8, ptr [[A_ADDR]], align 1
 // CHECK2-NEXT:    store i8 [[TMP2]], ptr [[A_CASTED]], align 1
 // CHECK2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A_CASTED]], align 4
 // CHECK2-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[TMP1]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK2-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP3]]) #[[ATTR2:[0-9]+]]
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -247,7 +247,7 @@
 // CHECK2-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    store i32 [[AA]], ptr [[AA_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l28_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
@@ -258,7 +258,7 @@
 // CHECK2-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[TMP1]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK2-NEXT:    call void @__omp_outlined__1(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP3]]) #[[ATTR2]]
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -285,7 +285,7 @@
 // CHECK2-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    store i32 [[AA]], ptr [[AA_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l33_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
@@ -296,7 +296,7 @@
 // CHECK2-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[TMP1]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK2-NEXT:    call void @__omp_outlined__2(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP3]]) #[[ATTR2]]
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
diff --git a/clang/test/OpenMP/nvptx_target_teams_distribute_codegen.cpp b/clang/test/OpenMP/nvptx_target_teams_distribute_codegen.cpp
--- a/clang/test/OpenMP/nvptx_target_teams_distribute_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_target_teams_distribute_codegen.cpp
@@ -37,15 +37,15 @@
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true)
+// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l16_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
-// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK1-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK1-NEXT:    store i32 [[TMP1]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK1-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]]) #[[ATTR5:[0-9]+]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -155,15 +155,15 @@
 // CHECK2-NEXT:  entry:
 // CHECK2-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true)
+// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l16_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
-// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK2-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[TMP1]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK2-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]]) #[[ATTR5:[0-9]+]]
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
diff --git a/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_codegen.cpp b/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_codegen.cpp
--- a/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_codegen.cpp
@@ -85,11 +85,11 @@
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store i64 [[L]], ptr [[L_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l28_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
-// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK1-NEXT:    [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
 // CHECK1-NEXT:    store i32 [[TMP3]], ptr [[N_CASTED]], align 4
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i64, ptr [[N_CASTED]], align 8
@@ -99,7 +99,7 @@
 // CHECK1-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK1-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK1-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i64 [[TMP4]], ptr [[TMP0]], i64 [[TMP6]]) #[[ATTR3:[0-9]+]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -396,7 +396,7 @@
 // CHECK1-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
@@ -407,7 +407,7 @@
 // CHECK1-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK1-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK1-NEXT:    call void @__omp_outlined__2(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i64 [[TMP4]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -645,7 +645,7 @@
 // CHECK1-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l39_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
@@ -653,7 +653,7 @@
 // CHECK1-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK1-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK1-NEXT:    call void @__omp_outlined__4(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -833,7 +833,7 @@
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    store i64 [[F]], ptr [[F_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l44_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
@@ -844,7 +844,7 @@
 // CHECK1-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK1-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK1-NEXT:    call void @__omp_outlined__6(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]], i64 [[TMP4]]) #[[ATTR3]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -1059,7 +1059,7 @@
 // CHECK1-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l52_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
@@ -1070,7 +1070,7 @@
 // CHECK1-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK1-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK1-NEXT:    call void @__omp_outlined__8(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i64 [[TMP4]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -1371,7 +1371,7 @@
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[V]], ptr [[V_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l59_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
@@ -1383,7 +1383,7 @@
 // CHECK1-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK1-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK1-NEXT:    call void @__omp_outlined__10(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i64 [[TMP4]], ptr [[TMP0]], ptr [[TMP5]]) #[[ATTR3]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -1635,11 +1635,11 @@
 // CHECK2-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK2-NEXT:    store i64 [[L]], ptr [[L_ADDR]], align 8
 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
-// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l28_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
-// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[TMP3]], ptr [[N_CASTED]], align 4
 // CHECK2-NEXT:    [[TMP4:%.*]] = load i64, ptr [[N_CASTED]], align 8
@@ -1649,7 +1649,7 @@
 // CHECK2-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK2-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i64 [[TMP4]], ptr [[TMP0]], i64 [[TMP6]]) #[[ATTR3:[0-9]+]]
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -1946,7 +1946,7 @@
 // CHECK2-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK2-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 8
 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 8
-// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
@@ -1957,7 +1957,7 @@
 // CHECK2-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK2-NEXT:    call void @__omp_outlined__2(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i64 [[TMP4]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -2195,7 +2195,7 @@
 // CHECK2-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
-// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l39_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
@@ -2203,7 +2203,7 @@
 // CHECK2-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK2-NEXT:    call void @__omp_outlined__4(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -2383,7 +2383,7 @@
 // CHECK2-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK2-NEXT:    store i64 [[F]], ptr [[F_ADDR]], align 8
 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 8
-// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l44_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
@@ -2394,7 +2394,7 @@
 // CHECK2-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK2-NEXT:    call void @__omp_outlined__6(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]], i64 [[TMP4]]) #[[ATTR3]]
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -2609,7 +2609,7 @@
 // CHECK2-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK2-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 8
-// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l52_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
@@ -2620,7 +2620,7 @@
 // CHECK2-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK2-NEXT:    call void @__omp_outlined__8(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i64 [[TMP4]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -2916,7 +2916,7 @@
 // CHECK2-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK2-NEXT:    store ptr [[V]], ptr [[V_ADDR]], align 8
 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
-// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l59_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
@@ -2928,7 +2928,7 @@
 // CHECK2-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK2-NEXT:    call void @__omp_outlined__10(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i64 [[TMP4]], ptr [[TMP0]], ptr [[TMP5]]) #[[ATTR3]]
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -3180,11 +3180,11 @@
 // CHECK3-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[L]], ptr [[L_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
-// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l28_kernel_environment)
 // CHECK3-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK3-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK3:       user_code.entry:
-// CHECK3-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK3-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK3-NEXT:    [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[TMP3]], ptr [[N_CASTED]], align 4
 // CHECK3-NEXT:    [[TMP4:%.*]] = load i32, ptr [[N_CASTED]], align 4
@@ -3194,7 +3194,7 @@
 // CHECK3-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK3-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP4]], ptr [[TMP0]], i32 [[TMP6]]) #[[ATTR3:[0-9]+]]
-// CHECK3-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK3-NEXT:    call void @__kmpc_target_deinit()
 // CHECK3-NEXT:    ret void
 // CHECK3:       worker.exit:
 // CHECK3-NEXT:    ret void
@@ -3484,7 +3484,7 @@
 // CHECK3-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
-// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_kernel_environment)
 // CHECK3-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK3-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK3:       user_code.entry:
@@ -3495,7 +3495,7 @@
 // CHECK3-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK3-NEXT:    call void @__omp_outlined__2(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP4]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK3-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK3-NEXT:    call void @__kmpc_target_deinit()
 // CHECK3-NEXT:    ret void
 // CHECK3:       worker.exit:
 // CHECK3-NEXT:    ret void
@@ -3727,7 +3727,7 @@
 // CHECK3-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
-// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l39_kernel_environment)
 // CHECK3-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK3-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK3:       user_code.entry:
@@ -3735,7 +3735,7 @@
 // CHECK3-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK3-NEXT:    call void @__omp_outlined__4(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK3-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK3-NEXT:    call void @__kmpc_target_deinit()
 // CHECK3-NEXT:    ret void
 // CHECK3:       worker.exit:
 // CHECK3-NEXT:    ret void
@@ -3909,7 +3909,7 @@
 // CHECK3-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[F]], ptr [[F_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 4
-// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l44_kernel_environment)
 // CHECK3-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK3-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK3:       user_code.entry:
@@ -3920,7 +3920,7 @@
 // CHECK3-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK3-NEXT:    call void @__omp_outlined__6(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]], i32 [[TMP4]]) #[[ATTR3]]
-// CHECK3-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK3-NEXT:    call void @__kmpc_target_deinit()
 // CHECK3-NEXT:    ret void
 // CHECK3:       worker.exit:
 // CHECK3-NEXT:    ret void
@@ -4128,7 +4128,7 @@
 // CHECK3-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 4
-// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l52_kernel_environment)
 // CHECK3-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK3-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK3:       user_code.entry:
@@ -4139,7 +4139,7 @@
 // CHECK3-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK3-NEXT:    call void @__omp_outlined__8(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP4]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK3-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK3-NEXT:    call void @__kmpc_target_deinit()
 // CHECK3-NEXT:    ret void
 // CHECK3:       worker.exit:
 // CHECK3-NEXT:    ret void
@@ -4443,7 +4443,7 @@
 // CHECK3-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[V]], ptr [[V_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
-// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l59_kernel_environment)
 // CHECK3-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK3-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK3:       user_code.entry:
@@ -4455,7 +4455,7 @@
 // CHECK3-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK3-NEXT:    call void @__omp_outlined__10(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP4]], ptr [[TMP0]], ptr [[TMP5]]) #[[ATTR3]]
-// CHECK3-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK3-NEXT:    call void @__kmpc_target_deinit()
 // CHECK3-NEXT:    ret void
 // CHECK3:       worker.exit:
 // CHECK3-NEXT:    ret void
diff --git a/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_generic_mode_codegen.cpp b/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_generic_mode_codegen.cpp
--- a/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_generic_mode_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_generic_mode_codegen.cpp
@@ -43,11 +43,11 @@
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l24_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
-// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK1-NEXT:    [[TMP3:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
 // CHECK1-NEXT:    store i32 [[TMP3]], ptr [[ARGC_CASTED]], align 4
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i64, ptr [[ARGC_CASTED]], align 8
@@ -57,7 +57,7 @@
 // CHECK1-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK1-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK1-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i64 [[TMP4]], ptr [[TMP0]], i64 [[TMP6]]) #[[ATTR3:[0-9]+]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -341,11 +341,11 @@
 // CHECK2-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l24_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
-// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[TMP3]], ptr [[ARGC_CASTED]], align 4
 // CHECK2-NEXT:    [[TMP4:%.*]] = load i32, ptr [[ARGC_CASTED]], align 4
@@ -355,7 +355,7 @@
 // CHECK2-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK2-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP4]], ptr [[TMP0]], i32 [[TMP6]]) #[[ATTR3:[0-9]+]]
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
diff --git 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
--- 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
@@ -73,11 +73,11 @@
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store i64 [[L]], ptr [[L_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l26_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
-// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK1-NEXT:    [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
 // CHECK1-NEXT:    store i32 [[TMP3]], ptr [[N_CASTED]], align 4
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i64, ptr [[N_CASTED]], align 8
@@ -87,7 +87,7 @@
 // CHECK1-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK1-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK1-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i64 [[TMP4]], ptr [[TMP0]], i64 [[TMP6]]) #[[ATTR3:[0-9]+]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -408,7 +408,7 @@
 // CHECK1-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l32_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
@@ -419,7 +419,7 @@
 // CHECK1-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK1-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK1-NEXT:    call void @__omp_outlined__2(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i64 [[TMP4]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -681,7 +681,7 @@
 // CHECK1-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l37_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
@@ -689,7 +689,7 @@
 // CHECK1-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK1-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK1-NEXT:    call void @__omp_outlined__4(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -883,7 +883,7 @@
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    store i64 [[F]], ptr [[F_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l42_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
@@ -894,7 +894,7 @@
 // CHECK1-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK1-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK1-NEXT:    call void @__omp_outlined__6(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]], i64 [[TMP4]]) #[[ATTR3]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -1128,11 +1128,11 @@
 // CHECK2-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[L]], ptr [[L_ADDR]], align 4
 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l26_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
-// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[TMP3]], ptr [[N_CASTED]], align 4
 // CHECK2-NEXT:    [[TMP4:%.*]] = load i32, ptr [[N_CASTED]], align 4
@@ -1142,7 +1142,7 @@
 // CHECK2-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK2-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP4]], ptr [[TMP0]], i32 [[TMP6]]) #[[ATTR3:[0-9]+]]
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -1456,7 +1456,7 @@
 // CHECK2-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK2-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 4
 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l32_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
@@ -1467,7 +1467,7 @@
 // CHECK2-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK2-NEXT:    call void @__omp_outlined__2(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP4]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -1723,7 +1723,7 @@
 // CHECK2-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l37_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
@@ -1731,7 +1731,7 @@
 // CHECK2-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK2-NEXT:    call void @__omp_outlined__4(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -1919,7 +1919,7 @@
 // CHECK2-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[F]], ptr [[F_ADDR]], align 4
 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l42_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
@@ -1930,7 +1930,7 @@
 // CHECK2-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK2-NEXT:    call void @__omp_outlined__6(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]], i32 [[TMP4]]) #[[ATTR3]]
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
diff --git a/clang/test/OpenMP/nvptx_target_teams_distribute_simd_codegen.cpp b/clang/test/OpenMP/nvptx_target_teams_distribute_simd_codegen.cpp
--- a/clang/test/OpenMP/nvptx_target_teams_distribute_simd_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_target_teams_distribute_simd_codegen.cpp
@@ -81,11 +81,11 @@
 // CHECK45-64-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK45-64-NEXT:    store i64 [[L]], ptr [[L_ADDR]], align 8
 // CHECK45-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
-// CHECK45-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK45-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_kernel_environment)
 // CHECK45-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK45-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-64:       user_code.entry:
-// CHECK45-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK45-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK45-64-NEXT:    [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
 // CHECK45-64-NEXT:    store i32 [[TMP3]], ptr [[N_CASTED]], align 4
 // CHECK45-64-NEXT:    [[TMP4:%.*]] = load i64, ptr [[N_CASTED]], align 8
@@ -95,7 +95,7 @@
 // CHECK45-64-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK45-64-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK45-64-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i64 [[TMP4]], ptr [[TMP0]], i64 [[TMP6]]) #[[ATTR3:[0-9]+]]
-// CHECK45-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-64-NEXT:    ret void
 // CHECK45-64:       worker.exit:
 // CHECK45-64-NEXT:    ret void
@@ -246,7 +246,7 @@
 // CHECK45-64-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK45-64-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 8
 // CHECK45-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 8
-// CHECK45-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK45-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l40_kernel_environment)
 // CHECK45-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK45-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-64:       user_code.entry:
@@ -257,7 +257,7 @@
 // CHECK45-64-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK45-64-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK45-64-NEXT:    call void @__omp_outlined__1(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i64 [[TMP4]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK45-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-64-NEXT:    ret void
 // CHECK45-64:       worker.exit:
 // CHECK45-64-NEXT:    ret void
@@ -396,7 +396,7 @@
 // CHECK45-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK45-64-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK45-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
-// CHECK45-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK45-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l45_kernel_environment)
 // CHECK45-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK45-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-64:       user_code.entry:
@@ -404,7 +404,7 @@
 // CHECK45-64-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK45-64-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK45-64-NEXT:    call void @__omp_outlined__2(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK45-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-64-NEXT:    ret void
 // CHECK45-64:       worker.exit:
 // CHECK45-64-NEXT:    ret void
@@ -515,7 +515,7 @@
 // CHECK45-64-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK45-64-NEXT:    store i64 [[F]], ptr [[F_ADDR]], align 8
 // CHECK45-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 8
-// CHECK45-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK45-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l50_kernel_environment)
 // CHECK45-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK45-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-64:       user_code.entry:
@@ -526,7 +526,7 @@
 // CHECK45-64-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK45-64-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK45-64-NEXT:    call void @__omp_outlined__3(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]], i64 [[TMP4]]) #[[ATTR3]]
-// CHECK45-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-64-NEXT:    ret void
 // CHECK45-64:       worker.exit:
 // CHECK45-64-NEXT:    ret void
@@ -664,11 +664,11 @@
 // CHECK45-32-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK45-32-NEXT:    store i32 [[L]], ptr [[L_ADDR]], align 4
 // CHECK45-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
-// CHECK45-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK45-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_kernel_environment)
 // CHECK45-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK45-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-32:       user_code.entry:
-// CHECK45-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK45-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK45-32-NEXT:    [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
 // CHECK45-32-NEXT:    store i32 [[TMP3]], ptr [[N_CASTED]], align 4
 // CHECK45-32-NEXT:    [[TMP4:%.*]] = load i32, ptr [[N_CASTED]], align 4
@@ -678,7 +678,7 @@
 // CHECK45-32-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK45-32-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK45-32-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP4]], ptr [[TMP0]], i32 [[TMP6]]) #[[ATTR3:[0-9]+]]
-// CHECK45-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-32-NEXT:    ret void
 // CHECK45-32:       worker.exit:
 // CHECK45-32-NEXT:    ret void
@@ -828,7 +828,7 @@
 // CHECK45-32-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK45-32-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 4
 // CHECK45-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
-// CHECK45-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK45-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l40_kernel_environment)
 // CHECK45-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK45-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-32:       user_code.entry:
@@ -839,7 +839,7 @@
 // CHECK45-32-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK45-32-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK45-32-NEXT:    call void @__omp_outlined__1(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP4]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK45-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-32-NEXT:    ret void
 // CHECK45-32:       worker.exit:
 // CHECK45-32-NEXT:    ret void
@@ -977,7 +977,7 @@
 // CHECK45-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK45-32-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK45-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
-// CHECK45-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK45-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l45_kernel_environment)
 // CHECK45-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK45-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-32:       user_code.entry:
@@ -985,7 +985,7 @@
 // CHECK45-32-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK45-32-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK45-32-NEXT:    call void @__omp_outlined__2(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK45-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-32-NEXT:    ret void
 // CHECK45-32:       worker.exit:
 // CHECK45-32-NEXT:    ret void
@@ -1095,7 +1095,7 @@
 // CHECK45-32-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK45-32-NEXT:    store i32 [[F]], ptr [[F_ADDR]], align 4
 // CHECK45-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 4
-// CHECK45-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK45-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l50_kernel_environment)
 // CHECK45-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK45-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-32:       user_code.entry:
@@ -1106,7 +1106,7 @@
 // CHECK45-32-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK45-32-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK45-32-NEXT:    call void @__omp_outlined__3(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]], i32 [[TMP4]]) #[[ATTR3]]
-// CHECK45-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-32-NEXT:    ret void
 // CHECK45-32:       worker.exit:
 // CHECK45-32-NEXT:    ret void
@@ -1242,11 +1242,11 @@
 // CHECK45-32-EX-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK45-32-EX-NEXT:    store i32 [[L]], ptr [[L_ADDR]], align 4
 // CHECK45-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
-// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_kernel_environment)
 // CHECK45-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK45-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-32-EX:       user_code.entry:
-// CHECK45-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK45-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK45-32-EX-NEXT:    [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
 // CHECK45-32-EX-NEXT:    store i32 [[TMP3]], ptr [[N_CASTED]], align 4
 // CHECK45-32-EX-NEXT:    [[TMP4:%.*]] = load i32, ptr [[N_CASTED]], align 4
@@ -1256,7 +1256,7 @@
 // CHECK45-32-EX-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK45-32-EX-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK45-32-EX-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP4]], ptr [[TMP0]], i32 [[TMP6]]) #[[ATTR3:[0-9]+]]
-// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-32-EX-NEXT:    ret void
 // CHECK45-32-EX:       worker.exit:
 // CHECK45-32-EX-NEXT:    ret void
@@ -1406,7 +1406,7 @@
 // CHECK45-32-EX-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK45-32-EX-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 4
 // CHECK45-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
-// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l40_kernel_environment)
 // CHECK45-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK45-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-32-EX:       user_code.entry:
@@ -1417,7 +1417,7 @@
 // CHECK45-32-EX-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK45-32-EX-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK45-32-EX-NEXT:    call void @__omp_outlined__1(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP4]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-32-EX-NEXT:    ret void
 // CHECK45-32-EX:       worker.exit:
 // CHECK45-32-EX-NEXT:    ret void
@@ -1555,7 +1555,7 @@
 // CHECK45-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK45-32-EX-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK45-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
-// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l45_kernel_environment)
 // CHECK45-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK45-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-32-EX:       user_code.entry:
@@ -1563,7 +1563,7 @@
 // CHECK45-32-EX-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK45-32-EX-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK45-32-EX-NEXT:    call void @__omp_outlined__2(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-32-EX-NEXT:    ret void
 // CHECK45-32-EX:       worker.exit:
 // CHECK45-32-EX-NEXT:    ret void
@@ -1673,7 +1673,7 @@
 // CHECK45-32-EX-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK45-32-EX-NEXT:    store i32 [[F]], ptr [[F_ADDR]], align 4
 // CHECK45-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 4
-// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK45-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l50_kernel_environment)
 // CHECK45-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK45-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK45-32-EX:       user_code.entry:
@@ -1684,7 +1684,7 @@
 // CHECK45-32-EX-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK45-32-EX-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK45-32-EX-NEXT:    call void @__omp_outlined__3(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]], i32 [[TMP4]]) #[[ATTR3]]
-// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK45-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK45-32-EX-NEXT:    ret void
 // CHECK45-32-EX:       worker.exit:
 // CHECK45-32-EX-NEXT:    ret void
@@ -1820,11 +1820,11 @@
 // CHECK-64-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK-64-NEXT:    store i64 [[L]], ptr [[L_ADDR]], align 8
 // CHECK-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
-// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
-// CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK-64-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK-64-NEXT:    [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP3]], ptr [[N_CASTED]], align 4
 // CHECK-64-NEXT:    [[TMP4:%.*]] = load i64, ptr [[N_CASTED]], align 8
@@ -1834,7 +1834,7 @@
 // CHECK-64-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i64 [[TMP4]], ptr [[TMP0]], i64 [[TMP6]]) #[[ATTR3:[0-9]+]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -1985,7 +1985,7 @@
 // CHECK-64-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK-64-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 8
 // CHECK-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 8
-// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l40_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -1996,7 +1996,7 @@
 // CHECK-64-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__1(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i64 [[TMP4]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -2135,7 +2135,7 @@
 // CHECK-64-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK-64-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
-// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l45_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -2143,7 +2143,7 @@
 // CHECK-64-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__2(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -2254,7 +2254,7 @@
 // CHECK-64-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK-64-NEXT:    store i64 [[F]], ptr [[F_ADDR]], align 8
 // CHECK-64-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 8
-// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK-64-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l50_kernel_environment)
 // CHECK-64-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-64-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-64:       user_code.entry:
@@ -2265,7 +2265,7 @@
 // CHECK-64-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK-64-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK-64-NEXT:    call void @__omp_outlined__3(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]], i64 [[TMP4]]) #[[ATTR3]]
-// CHECK-64-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-64-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-64-NEXT:    ret void
 // CHECK-64:       worker.exit:
 // CHECK-64-NEXT:    ret void
@@ -2403,11 +2403,11 @@
 // CHECK-32-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[L]], ptr [[L_ADDR]], align 4
 // CHECK-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
-// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
-// CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK-32-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK-32-NEXT:    [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP3]], ptr [[N_CASTED]], align 4
 // CHECK-32-NEXT:    [[TMP4:%.*]] = load i32, ptr [[N_CASTED]], align 4
@@ -2417,7 +2417,7 @@
 // CHECK-32-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP4]], ptr [[TMP0]], i32 [[TMP6]]) #[[ATTR3:[0-9]+]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -2567,7 +2567,7 @@
 // CHECK-32-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK-32-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 4
 // CHECK-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
-// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l40_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -2578,7 +2578,7 @@
 // CHECK-32-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__1(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP4]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -2716,7 +2716,7 @@
 // CHECK-32-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK-32-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
-// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l45_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -2724,7 +2724,7 @@
 // CHECK-32-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__2(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -2834,7 +2834,7 @@
 // CHECK-32-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[F]], ptr [[F_ADDR]], align 4
 // CHECK-32-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 4
-// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l50_kernel_environment)
 // CHECK-32-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-32-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32:       user_code.entry:
@@ -2845,7 +2845,7 @@
 // CHECK-32-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK-32-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-NEXT:    call void @__omp_outlined__3(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]], i32 [[TMP4]]) #[[ATTR3]]
-// CHECK-32-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-32-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-NEXT:    ret void
 // CHECK-32:       worker.exit:
 // CHECK-32-NEXT:    ret void
@@ -2981,11 +2981,11 @@
 // CHECK-32-EX-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[L]], ptr [[L_ADDR]], align 4
 // CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
-// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l34_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
-// CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK-32-EX-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK-32-EX-NEXT:    [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP3]], ptr [[N_CASTED]], align 4
 // CHECK-32-EX-NEXT:    [[TMP4:%.*]] = load i32, ptr [[N_CASTED]], align 4
@@ -2995,7 +2995,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP4]], ptr [[TMP0]], i32 [[TMP6]]) #[[ATTR3:[0-9]+]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -3145,7 +3145,7 @@
 // CHECK-32-EX-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 4
 // CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
-// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l40_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -3156,7 +3156,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__1(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP4]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -3294,7 +3294,7 @@
 // CHECK-32-EX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK-32-EX-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
-// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l45_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -3302,7 +3302,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__2(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]]) #[[ATTR3]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
@@ -3412,7 +3412,7 @@
 // CHECK-32-EX-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[F]], ptr [[F_ADDR]], align 4
 // CHECK-32-EX-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[C_ADDR]], align 4
-// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK-32-EX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l50_kernel_environment)
 // CHECK-32-EX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK-32-EX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK-32-EX:       user_code.entry:
@@ -3423,7 +3423,7 @@
 // CHECK-32-EX-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK-32-EX-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK-32-EX-NEXT:    call void @__omp_outlined__3(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[TMP0]], i32 [[TMP4]]) #[[ATTR3]]
-// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-32-EX-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-32-EX-NEXT:    ret void
 // CHECK-32-EX:       worker.exit:
 // CHECK-32-EX-NEXT:    ret void
diff --git a/clang/test/OpenMP/nvptx_teams_codegen.cpp b/clang/test/OpenMP/nvptx_teams_codegen.cpp
--- a/clang/test/OpenMP/nvptx_teams_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_teams_codegen.cpp
@@ -83,19 +83,19 @@
 // CHECK1-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    store i64 [[ARGC]], ptr [[ARGC_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true)
+// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l23_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
 // CHECK1-NEXT:    [[TMP1:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
 // CHECK1-NEXT:    [[ARGC1:%.*]] = call align 8 ptr @__kmpc_alloc_shared(i64 4)
 // CHECK1-NEXT:    store i32 [[TMP1]], ptr [[ARGC1]], align 4
-// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK1-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK1-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK1-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[ARGC1]]) #[[ATTR3:[0-9]+]]
 // CHECK1-NEXT:    call void @__kmpc_free_shared(ptr [[ARGC1]], i64 4)
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -122,7 +122,7 @@
 // CHECK1-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    store ptr [[ARGC]], ptr [[ARGC_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIPPcEiT__l15_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
@@ -134,7 +134,7 @@
 // CHECK1-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK1-NEXT:    call void @__omp_outlined__1(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[ARGC1]]) #[[ATTR3]]
 // CHECK1-NEXT:    call void @__kmpc_free_shared(ptr [[ARGC1]], i64 8)
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -161,19 +161,19 @@
 // CHECK2-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    store i32 [[ARGC]], ptr [[ARGC_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true)
+// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l23_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
 // CHECK2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
 // CHECK2-NEXT:    [[ARGC1:%.*]] = call align 8 ptr @__kmpc_alloc_shared(i32 4)
 // CHECK2-NEXT:    store i32 [[TMP1]], ptr [[ARGC1]], align 4
-// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK2-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK2-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[ARGC1]]) #[[ATTR3:[0-9]+]]
 // CHECK2-NEXT:    call void @__kmpc_free_shared(ptr [[ARGC1]], i32 4)
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -200,7 +200,7 @@
 // CHECK2-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    store ptr [[ARGC]], ptr [[ARGC_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIPPcEiT__l15_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
@@ -212,7 +212,7 @@
 // CHECK2-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK2-NEXT:    call void @__omp_outlined__1(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[ARGC1]]) #[[ATTR3]]
 // CHECK2-NEXT:    call void @__kmpc_free_shared(ptr [[ARGC1]], i32 4)
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -243,19 +243,19 @@
 // CHECK3-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
 // CHECK3-NEXT:    store i64 [[B]], ptr [[B_ADDR]], align 8
 // CHECK3-NEXT:    store i64 [[ARGC]], ptr [[ARGC_ADDR]], align 8
-// CHECK3-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true)
+// CHECK3-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l64_kernel_environment)
 // CHECK3-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK3-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK3:       user_code.entry:
 // CHECK3-NEXT:    [[TMP1:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
 // CHECK3-NEXT:    [[ARGC1:%.*]] = call align 8 ptr @__kmpc_alloc_shared(i64 4)
 // CHECK3-NEXT:    store i32 [[TMP1]], ptr [[ARGC1]], align 4
-// CHECK3-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK3-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK3-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK3-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[ARGC1]]) #[[ATTR3:[0-9]+]]
 // CHECK3-NEXT:    call void @__kmpc_free_shared(ptr [[ARGC1]], i64 4)
-// CHECK3-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK3-NEXT:    call void @__kmpc_target_deinit()
 // CHECK3-NEXT:    ret void
 // CHECK3:       worker.exit:
 // CHECK3-NEXT:    ret void
@@ -286,7 +286,7 @@
 // CHECK3-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
 // CHECK3-NEXT:    store i64 [[B]], ptr [[B_ADDR]], align 8
 // CHECK3-NEXT:    store ptr [[ARGC]], ptr [[ARGC_ADDR]], align 8
-// CHECK3-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK3-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIPPcEiT__l53_kernel_environment)
 // CHECK3-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK3-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK3:       user_code.entry:
@@ -298,7 +298,7 @@
 // CHECK3-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK3-NEXT:    call void @__omp_outlined__1(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[ARGC1]]) #[[ATTR3]]
 // CHECK3-NEXT:    call void @__kmpc_free_shared(ptr [[ARGC1]], i64 8)
-// CHECK3-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK3-NEXT:    call void @__kmpc_target_deinit()
 // CHECK3-NEXT:    ret void
 // CHECK3:       worker.exit:
 // CHECK3-NEXT:    ret void
@@ -329,19 +329,19 @@
 // CHECK4-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
 // CHECK4-NEXT:    store i32 [[B]], ptr [[B_ADDR]], align 4
 // CHECK4-NEXT:    store i32 [[ARGC]], ptr [[ARGC_ADDR]], align 4
-// CHECK4-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true)
+// CHECK4-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l64_kernel_environment)
 // CHECK4-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK4-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK4:       user_code.entry:
 // CHECK4-NEXT:    [[TMP1:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
 // CHECK4-NEXT:    [[ARGC1:%.*]] = call align 8 ptr @__kmpc_alloc_shared(i32 4)
 // CHECK4-NEXT:    store i32 [[TMP1]], ptr [[ARGC1]], align 4
-// CHECK4-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK4-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK4-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK4-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK4-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[ARGC1]]) #[[ATTR3:[0-9]+]]
 // CHECK4-NEXT:    call void @__kmpc_free_shared(ptr [[ARGC1]], i32 4)
-// CHECK4-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK4-NEXT:    call void @__kmpc_target_deinit()
 // CHECK4-NEXT:    ret void
 // CHECK4:       worker.exit:
 // CHECK4-NEXT:    ret void
@@ -372,7 +372,7 @@
 // CHECK4-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
 // CHECK4-NEXT:    store i32 [[B]], ptr [[B_ADDR]], align 4
 // CHECK4-NEXT:    store ptr [[ARGC]], ptr [[ARGC_ADDR]], align 4
-// CHECK4-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK4-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIPPcEiT__l53_kernel_environment)
 // CHECK4-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK4-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK4:       user_code.entry:
@@ -384,7 +384,7 @@
 // CHECK4-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK4-NEXT:    call void @__omp_outlined__1(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[ARGC1]]) #[[ATTR3]]
 // CHECK4-NEXT:    call void @__kmpc_free_shared(ptr [[ARGC1]], i32 4)
-// CHECK4-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK4-NEXT:    call void @__kmpc_target_deinit()
 // CHECK4-NEXT:    ret void
 // CHECK4:       worker.exit:
 // CHECK4-NEXT:    ret void
diff --git a/clang/test/OpenMP/nvptx_teams_reduction_codegen.cpp b/clang/test/OpenMP/nvptx_teams_reduction_codegen.cpp
--- a/clang/test/OpenMP/nvptx_teams_reduction_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_teams_reduction_codegen.cpp
@@ -57,19 +57,19 @@
 // CHECK1-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    store i64 [[E]], ptr [[E_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true)
+// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l20_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
 // CHECK1-NEXT:    [[TMP1:%.*]] = load double, ptr [[E_ADDR]], align 8
 // CHECK1-NEXT:    [[E1:%.*]] = call align 8 ptr @__kmpc_alloc_shared(i64 8)
 // CHECK1-NEXT:    store double [[TMP1]], ptr [[E1]], align 8
-// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK1-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK1-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK1-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[E1]]) #[[ATTR4:[0-9]+]]
 // CHECK1-NEXT:    call void @__kmpc_free_shared(ptr [[E1]], i64 8)
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -328,7 +328,7 @@
 // CHECK1-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    store i64 [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    store i64 [[D]], ptr [[D_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l26_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
@@ -344,7 +344,7 @@
 // CHECK1-NEXT:    call void @__omp_outlined__1(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[C1]], ptr [[D2]]) #[[ATTR4]]
 // CHECK1-NEXT:    call void @__kmpc_free_shared(ptr [[D2]], i64 4)
 // CHECK1-NEXT:    call void @__kmpc_free_shared(ptr [[C1]], i64 1)
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -677,7 +677,7 @@
 // CHECK1-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store i64 [[B]], ptr [[B_ADDR]], align 8
-// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK1-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l33_kernel_environment)
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK1:       user_code.entry:
@@ -685,7 +685,7 @@
 // CHECK1-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK1-NEXT:    store i32 [[TMP1]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK1-NEXT:    call void @__omp_outlined__9(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]]) #[[ATTR4]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK1-NEXT:    call void @__kmpc_target_deinit()
 // CHECK1-NEXT:    ret void
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void
@@ -1250,17 +1250,17 @@
 // CHECK2-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    store ptr [[E]], ptr [[E_ADDR]], align 4
 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[E_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true)
+// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l20_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
-// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK2-NEXT:    [[TMP3:%.*]] = load double, ptr [[TMP0]], align 8
 // CHECK2-NEXT:    store double [[TMP3]], ptr [[E1]], align 8
 // CHECK2-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK2-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[E1]]) #[[ATTR4:[0-9]+]]
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -1519,7 +1519,7 @@
 // CHECK2-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    store i32 [[C]], ptr [[C_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[D]], ptr [[D_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l26_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
@@ -1535,7 +1535,7 @@
 // CHECK2-NEXT:    call void @__omp_outlined__1(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[C1]], ptr [[D2]]) #[[ATTR4]]
 // CHECK2-NEXT:    call void @__kmpc_free_shared(ptr [[D2]], i32 4)
 // CHECK2-NEXT:    call void @__kmpc_free_shared(ptr [[C1]], i32 1)
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -1868,7 +1868,7 @@
 // CHECK2-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK2-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[B]], ptr [[B_ADDR]], align 4
-// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK2-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l33_kernel_environment)
 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK2:       user_code.entry:
@@ -1876,7 +1876,7 @@
 // CHECK2-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK2-NEXT:    store i32 [[TMP1]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK2-NEXT:    call void @__omp_outlined__9(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]]) #[[ATTR4]]
-// CHECK2-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK2-NEXT:    call void @__kmpc_target_deinit()
 // CHECK2-NEXT:    ret void
 // CHECK2:       worker.exit:
 // CHECK2-NEXT:    ret void
@@ -2441,17 +2441,17 @@
 // CHECK3-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    store ptr [[E]], ptr [[E_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[E_ADDR]], align 4
-// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true)
+// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l20_kernel_environment)
 // CHECK3-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 // CHECK3-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK3:       user_code.entry:
-// CHECK3-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK3-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK3-NEXT:    [[TMP3:%.*]] = load double, ptr [[TMP0]], align 8
 // CHECK3-NEXT:    store double [[TMP3]], ptr [[E1]], align 8
 // CHECK3-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK3-NEXT:    call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[E1]]) #[[ATTR4:[0-9]+]]
-// CHECK3-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK3-NEXT:    call void @__kmpc_target_deinit()
 // CHECK3-NEXT:    ret void
 // CHECK3:       worker.exit:
 // CHECK3-NEXT:    ret void
@@ -2710,7 +2710,7 @@
 // CHECK3-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    store i32 [[C]], ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[D]], ptr [[D_ADDR]], align 4
-// CHECK3-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
+// CHECK3-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l26_kernel_environment)
 // CHECK3-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK3-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK3:       user_code.entry:
@@ -2726,7 +2726,7 @@
 // CHECK3-NEXT:    call void @__omp_outlined__1(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[C1]], ptr [[D2]]) #[[ATTR4]]
 // CHECK3-NEXT:    call void @__kmpc_free_shared(ptr [[D2]], i32 4)
 // CHECK3-NEXT:    call void @__kmpc_free_shared(ptr [[C1]], i32 1)
-// CHECK3-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
+// CHECK3-NEXT:    call void @__kmpc_target_deinit()
 // CHECK3-NEXT:    ret void
 // CHECK3:       worker.exit:
 // CHECK3-NEXT:    ret void
@@ -3059,7 +3059,7 @@
 // CHECK3-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[B]], ptr [[B_ADDR]], align 4
-// CHECK3-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 2, i1 false)
+// CHECK3-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIcET_i_l33_kernel_environment)
 // CHECK3-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK3-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK3:       user_code.entry:
@@ -3067,7 +3067,7 @@
 // CHECK3-NEXT:    store i32 0, ptr [[DOTZERO_ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[TMP1]], ptr [[DOTTHREADID_TEMP_]], align 4
 // CHECK3-NEXT:    call void @__omp_outlined__9(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]]) #[[ATTR4]]
-// CHECK3-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK3-NEXT:    call void @__kmpc_target_deinit()
 // CHECK3-NEXT:    ret void
 // CHECK3:       worker.exit:
 // CHECK3-NEXT:    ret void
diff --git a/clang/test/OpenMP/reduction_implicit_map.cpp b/clang/test/OpenMP/reduction_implicit_map.cpp
--- a/clang/test/OpenMP/reduction_implicit_map.cpp
+++ b/clang/test/OpenMP/reduction_implicit_map.cpp
@@ -103,16 +103,16 @@
 // CHECK-NEXT:    [[E_ADDR:%.*]] = alloca ptr, align 8
 // CHECK-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 8
 // CHECK-NEXT:    store ptr [[E]], ptr [[E_ADDR]], align 8
-// CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
+// CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l32_kernel_environment)
 // CHECK-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
 // CHECK-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 // CHECK:       user_code.entry:
-// CHECK-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
+// CHECK-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 // CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[E_ADDR]], align 8
 // CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
 // CHECK-NEXT:    store ptr [[TMP2]], ptr [[TMP3]], align 8
 // CHECK-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 1)
-// CHECK-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
+// CHECK-NEXT:    call void @__kmpc_target_deinit()
 // CHECK-NEXT:    ret void
 // CHECK:       worker.exit:
 // CHECK-NEXT:    ret void
diff --git a/clang/test/OpenMP/remarks_parallel_in_multiple_target_state_machines.c b/clang/test/OpenMP/remarks_parallel_in_multiple_target_state_machines.c
--- a/clang/test/OpenMP/remarks_parallel_in_multiple_target_state_machines.c
+++ b/clang/test/OpenMP/remarks_parallel_in_multiple_target_state_machines.c
@@ -84,9 +84,9 @@
 }
 
 #pragma omp begin declare target device_type(nohost)
-__attribute__((weak)) 
-extern "C" int __kmpc_target_init(void *Ident, char Mode,
-                       bool UseGenericStateMachine) { // all-remark {{Could not internalize function. Some optimizations may not be possible. [OMP140]}}
+struct KernelEnvironmentTy;
+__attribute__((weak))
+extern "C" int __kmpc_target_init(struct KernelEnvironmentTy *) { // all-remark {{Could not internalize function. Some optimizations may not be possible. [OMP140]}}
   return 0;
 }
 #pragma omp end declare target
diff --git a/clang/test/OpenMP/remarks_parallel_in_target_state_machine.c b/clang/test/OpenMP/remarks_parallel_in_target_state_machine.c
--- a/clang/test/OpenMP/remarks_parallel_in_target_state_machine.c
+++ b/clang/test/OpenMP/remarks_parallel_in_target_state_machine.c
@@ -40,9 +40,9 @@
 }
 
 #pragma omp begin declare target device_type(nohost)
-__attribute__((weak)) 
-extern "C" int __kmpc_target_init(void *Ident, char Mode,
-                       bool UseGenericStateMachine) { // expected-remark {{Could not internalize function. Some optimizations may not be possible. [OMP140]}}
+struct KernelEnvironmentTy;
+__attribute__((weak))
+extern "C" int __kmpc_target_init(struct KernelEnvironmentTy *) { // expected-remark {{Could not internalize function. Some optimizations may not be possible. [OMP140]}}
   return 0;
 }
 #pragma omp end declare target
diff --git a/clang/test/OpenMP/target_parallel_debug_codegen.cpp b/clang/test/OpenMP/target_parallel_debug_codegen.cpp
--- a/clang/test/OpenMP/target_parallel_debug_codegen.cpp
+++ b/clang/test/OpenMP/target_parallel_debug_codegen.cpp
@@ -95,7 +95,7 @@
 // CHECK1-NEXT:    [[TMP6:%.*]] = addrspacecast ptr addrspace(1) [[TMP5]] to ptr, !dbg [[DBG47]]
 // CHECK1-NEXT:    store ptr [[TMP6]], ptr [[_TMP2]], align 8, !dbg [[DBG47]]
 // CHECK1-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[_TMP2]], align 8, !dbg [[DBG47]]
-// CHECK1-NEXT:    [[TMP8:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false), !dbg [[DBG47]]
+// CHECK1-NEXT:    [[TMP8:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l23_kernel_environment), !dbg [[DBG47]]
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP8]], -1, !dbg [[DBG47]]
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]], !dbg [[DBG47]]
 // CHECK1:       user_code.entry:
@@ -113,7 +113,7 @@
 // CHECK1-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 3, !dbg [[DBG48]]
 // CHECK1-NEXT:    store ptr [[TMP7]], ptr [[TMP16]], align 8, !dbg [[DBG48]]
 // CHECK1-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB3]], i32 [[TMP9]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 4), !dbg [[DBG48]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB5:[0-9]+]], i8 2), !dbg [[DBG49:![0-9]+]]
+// CHECK1-NEXT:    call void @__kmpc_target_deinit(), !dbg [[DBG49:![0-9]+]]
 // CHECK1-NEXT:    ret void, !dbg [[DBG51:![0-9]+]]
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void, !dbg [[DBG47]]
@@ -308,11 +308,11 @@
 // CHECK1-NEXT:    [[TMP7:%.*]] = addrspacecast ptr addrspace(1) [[TMP6]] to ptr, !dbg [[DBG137]]
 // CHECK1-NEXT:    store ptr [[TMP7]], ptr [[_TMP2]], align 8, !dbg [[DBG137]]
 // CHECK1-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[_TMP2]], align 8, !dbg [[DBG137]]
-// CHECK1-NEXT:    [[TMP9:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB7:[0-9]+]], i8 2, i1 false), !dbg [[DBG137]]
+// CHECK1-NEXT:    [[TMP9:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l37_kernel_environment), !dbg [[DBG137]]
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP9]], -1, !dbg [[DBG137]]
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]], !dbg [[DBG137]]
 // CHECK1:       user_code.entry:
-// CHECK1-NEXT:    [[TMP10:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB9:[0-9]+]])
+// CHECK1-NEXT:    [[TMP10:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB7:[0-9]+]])
 // CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG138:![0-9]+]]
 // CHECK1-NEXT:    store i32 [[TMP11]], ptr [[A_CASTED]], align 4, !dbg [[DBG138]]
 // CHECK1-NEXT:    [[TMP12:%.*]] = load i64, ptr [[A_CASTED]], align 8, !dbg [[DBG138]]
@@ -325,8 +325,8 @@
 // CHECK1-NEXT:    store ptr [[TMP5]], ptr [[TMP16]], align 8, !dbg [[DBG138]]
 // CHECK1-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 3, !dbg [[DBG138]]
 // CHECK1-NEXT:    store ptr [[TMP8]], ptr [[TMP17]], align 8, !dbg [[DBG138]]
-// CHECK1-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB9]], i32 [[TMP10]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__2, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 4), !dbg [[DBG138]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB11:[0-9]+]], i8 2), !dbg [[DBG139:![0-9]+]]
+// CHECK1-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB7]], i32 [[TMP10]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__2, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 4), !dbg [[DBG138]]
+// CHECK1-NEXT:    call void @__kmpc_target_deinit(), !dbg [[DBG139:![0-9]+]]
 // CHECK1-NEXT:    ret void, !dbg [[DBG141:![0-9]+]]
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void, !dbg [[DBG137]]
@@ -517,11 +517,11 @@
 // CHECK1-NEXT:    [[TMP10:%.*]] = addrspacecast ptr addrspace(1) [[TMP9]] to ptr, !dbg [[DBG212]]
 // CHECK1-NEXT:    store ptr [[TMP10]], ptr [[_TMP3]], align 8, !dbg [[DBG212]]
 // CHECK1-NEXT:    [[TMP11:%.*]] = load ptr, ptr [[_TMP3]], align 8, !dbg [[DBG212]]
-// CHECK1-NEXT:    [[TMP12:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB13:[0-9]+]], i8 2, i1 false), !dbg [[DBG212]]
+// CHECK1-NEXT:    [[TMP12:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l51_kernel_environment), !dbg [[DBG212]]
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP12]], -1, !dbg [[DBG212]]
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]], !dbg [[DBG212]]
 // CHECK1:       user_code.entry:
-// CHECK1-NEXT:    [[TMP13:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB15:[0-9]+]])
+// CHECK1-NEXT:    [[TMP13:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB11:[0-9]+]])
 // CHECK1-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 0, !dbg [[DBG213:![0-9]+]]
 // CHECK1-NEXT:    store ptr [[TMP2]], ptr [[TMP14]], align 8, !dbg [[DBG213]]
 // CHECK1-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 1, !dbg [[DBG213]]
@@ -530,8 +530,8 @@
 // CHECK1-NEXT:    store ptr [[TMP8]], ptr [[TMP16]], align 8, !dbg [[DBG213]]
 // CHECK1-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 3, !dbg [[DBG213]]
 // CHECK1-NEXT:    store ptr [[TMP11]], ptr [[TMP17]], align 8, !dbg [[DBG213]]
-// CHECK1-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB15]], i32 [[TMP13]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__4, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 4), !dbg [[DBG213]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB17:[0-9]+]], i8 2), !dbg [[DBG214:![0-9]+]]
+// CHECK1-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB11]], i32 [[TMP13]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__4, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 4), !dbg [[DBG213]]
+// CHECK1-NEXT:    call void @__kmpc_target_deinit(), !dbg [[DBG214:![0-9]+]]
 // CHECK1-NEXT:    ret void, !dbg [[DBG216:![0-9]+]]
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void, !dbg [[DBG212]]
diff --git a/clang/test/OpenMP/target_parallel_for_debug_codegen.cpp b/clang/test/OpenMP/target_parallel_for_debug_codegen.cpp
--- a/clang/test/OpenMP/target_parallel_for_debug_codegen.cpp
+++ b/clang/test/OpenMP/target_parallel_for_debug_codegen.cpp
@@ -89,7 +89,7 @@
 // CHECK1-NEXT:    [[TMP6:%.*]] = addrspacecast ptr addrspace(1) [[TMP5]] to ptr, !dbg [[DBG41]]
 // CHECK1-NEXT:    store ptr [[TMP6]], ptr [[_TMP2]], align 8, !dbg [[DBG41]]
 // CHECK1-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[_TMP2]], align 8, !dbg [[DBG41]]
-// CHECK1-NEXT:    [[TMP8:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false), !dbg [[DBG41]]
+// CHECK1-NEXT:    [[TMP8:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l13_kernel_environment), !dbg [[DBG41]]
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP8]], -1, !dbg [[DBG41]]
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]], !dbg [[DBG41]]
 // CHECK1:       user_code.entry:
@@ -110,7 +110,7 @@
 // CHECK1-NEXT:    [[TOBOOL:%.*]] = trunc i8 [[TMP17]] to i1, !dbg [[DBG43]]
 // CHECK1-NEXT:    [[TMP18:%.*]] = zext i1 [[TOBOOL]] to i32, !dbg [[DBG42]]
 // CHECK1-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB6]], i32 [[TMP9]], i32 [[TMP18]], i32 -1, i32 -1, ptr @__omp_outlined__, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 4), !dbg [[DBG42]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB8:[0-9]+]], i8 2), !dbg [[DBG45:![0-9]+]]
+// CHECK1-NEXT:    call void @__kmpc_target_deinit(), !dbg [[DBG45:![0-9]+]]
 // CHECK1-NEXT:    ret void, !dbg [[DBG46:![0-9]+]]
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void, !dbg [[DBG41]]
@@ -383,11 +383,11 @@
 // CHECK1-NEXT:    [[TMP7:%.*]] = addrspacecast ptr addrspace(1) [[TMP6]] to ptr, !dbg [[DBG146]]
 // CHECK1-NEXT:    store ptr [[TMP7]], ptr [[_TMP2]], align 8, !dbg [[DBG146]]
 // CHECK1-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[_TMP2]], align 8, !dbg [[DBG146]]
-// CHECK1-NEXT:    [[TMP9:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB10:[0-9]+]], i8 2, i1 false), !dbg [[DBG146]]
+// CHECK1-NEXT:    [[TMP9:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l27_kernel_environment), !dbg [[DBG146]]
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP9]], -1, !dbg [[DBG146]]
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]], !dbg [[DBG146]]
 // CHECK1:       user_code.entry:
-// CHECK1-NEXT:    [[TMP10:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB15:[0-9]+]])
+// CHECK1-NEXT:    [[TMP10:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB13:[0-9]+]])
 // CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG147:![0-9]+]]
 // CHECK1-NEXT:    store i32 [[TMP11]], ptr [[A_CASTED]], align 4, !dbg [[DBG147]]
 // CHECK1-NEXT:    [[TMP12:%.*]] = load i64, ptr [[A_CASTED]], align 8, !dbg [[DBG147]]
@@ -400,8 +400,8 @@
 // CHECK1-NEXT:    store ptr [[TMP5]], ptr [[TMP16]], align 8, !dbg [[DBG147]]
 // CHECK1-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 3, !dbg [[DBG147]]
 // CHECK1-NEXT:    store ptr [[TMP8]], ptr [[TMP17]], align 8, !dbg [[DBG147]]
-// CHECK1-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB15]], i32 [[TMP10]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__2, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 4), !dbg [[DBG147]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB17:[0-9]+]], i8 2), !dbg [[DBG148:![0-9]+]]
+// CHECK1-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB13]], i32 [[TMP10]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__2, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 4), !dbg [[DBG147]]
+// CHECK1-NEXT:    call void @__kmpc_target_deinit(), !dbg [[DBG148:![0-9]+]]
 // CHECK1-NEXT:    ret void, !dbg [[DBG150:![0-9]+]]
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void, !dbg [[DBG146]]
@@ -466,7 +466,7 @@
 // CHECK1-NEXT:    call void @llvm.dbg.declare(metadata ptr [[I]], metadata [[META172:![0-9]+]], metadata !DIExpression()), !dbg [[DBG155]]
 // CHECK1-NEXT:    [[TMP9:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8, !dbg [[DBG165]]
 // CHECK1-NEXT:    [[TMP10:%.*]] = load i32, ptr [[TMP9]], align 4, !dbg [[DBG165]]
-// CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB12:[0-9]+]], i32 [[TMP10]], i32 33, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1), !dbg [[DBG173:![0-9]+]]
+// CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB10:[0-9]+]], i32 [[TMP10]], i32 33, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1), !dbg [[DBG173:![0-9]+]]
 // CHECK1-NEXT:    br label [[OMP_DISPATCH_COND:%.*]], !dbg [[DBG165]]
 // CHECK1:       omp.dispatch.cond:
 // CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !dbg [[DBG168]]
@@ -559,7 +559,7 @@
 // CHECK1-NEXT:    store i32 [[ADD25]], ptr [[DOTOMP_UB]], align 4, !dbg [[DBG165]]
 // CHECK1-NEXT:    br label [[OMP_DISPATCH_COND]], !dbg [[DBG173]], !llvm.loop [[LOOP203:![0-9]+]]
 // CHECK1:       omp.dispatch.end:
-// CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB14:[0-9]+]], i32 [[TMP10]]), !dbg [[DBG202:![0-9]+]]
+// CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB12:[0-9]+]], i32 [[TMP10]]), !dbg [[DBG202:![0-9]+]]
 // CHECK1-NEXT:    ret void, !dbg [[DBG204:![0-9]+]]
 //
 //
@@ -665,11 +665,11 @@
 // CHECK1-NEXT:    [[TMP10:%.*]] = addrspacecast ptr addrspace(1) [[TMP9]] to ptr, !dbg [[DBG236]]
 // CHECK1-NEXT:    store ptr [[TMP10]], ptr [[_TMP3]], align 8, !dbg [[DBG236]]
 // CHECK1-NEXT:    [[TMP11:%.*]] = load ptr, ptr [[_TMP3]], align 8, !dbg [[DBG236]]
-// CHECK1-NEXT:    [[TMP12:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB19:[0-9]+]], i8 2, i1 false), !dbg [[DBG236]]
+// CHECK1-NEXT:    [[TMP12:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l41_kernel_environment), !dbg [[DBG236]]
 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP12]], -1, !dbg [[DBG236]]
 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]], !dbg [[DBG236]]
 // CHECK1:       user_code.entry:
-// CHECK1-NEXT:    [[TMP13:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB24:[0-9]+]])
+// CHECK1-NEXT:    [[TMP13:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB20:[0-9]+]])
 // CHECK1-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 0, !dbg [[DBG237:![0-9]+]]
 // CHECK1-NEXT:    store ptr [[TMP2]], ptr [[TMP14]], align 8, !dbg [[DBG237]]
 // CHECK1-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 1, !dbg [[DBG237]]
@@ -678,8 +678,8 @@
 // CHECK1-NEXT:    store ptr [[TMP8]], ptr [[TMP16]], align 8, !dbg [[DBG237]]
 // CHECK1-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 3, !dbg [[DBG237]]
 // CHECK1-NEXT:    store ptr [[TMP11]], ptr [[TMP17]], align 8, !dbg [[DBG237]]
-// CHECK1-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB24]], i32 [[TMP13]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__4, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 4), !dbg [[DBG237]]
-// CHECK1-NEXT:    call void @__kmpc_target_deinit(ptr @[[GLOB26:[0-9]+]], i8 2), !dbg [[DBG238:![0-9]+]]
+// CHECK1-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB20]], i32 [[TMP13]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__4, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 4), !dbg [[DBG237]]
+// CHECK1-NEXT:    call void @__kmpc_target_deinit(), !dbg [[DBG238:![0-9]+]]
 // CHECK1-NEXT:    ret void, !dbg [[DBG240:![0-9]+]]
 // CHECK1:       worker.exit:
 // CHECK1-NEXT:    ret void, !dbg [[DBG236]]
@@ -749,7 +749,7 @@
 // CHECK1-NEXT:    call void @llvm.dbg.declare(metadata ptr [[I]], metadata [[META262:![0-9]+]], metadata !DIExpression()), !dbg [[DBG245]]
 // CHECK1-NEXT:    [[TMP12:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8, !dbg [[DBG255]]
 // CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[TMP12]], align 4, !dbg [[DBG255]]
-// CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB21:[0-9]+]], i32 [[TMP13]], i32 33, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1), !dbg [[DBG263:![0-9]+]]
+// CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB17:[0-9]+]], i32 [[TMP13]], i32 33, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1), !dbg [[DBG263:![0-9]+]]
 // CHECK1-NEXT:    br label [[OMP_DISPATCH_COND:%.*]], !dbg [[DBG255]]
 // CHECK1:       omp.dispatch.cond:
 // CHECK1-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !dbg [[DBG258]]
@@ -846,7 +846,7 @@
 // CHECK1-NEXT:    store i32 [[ADD29]], ptr [[DOTOMP_UB]], align 4, !dbg [[DBG255]]
 // CHECK1-NEXT:    br label [[OMP_DISPATCH_COND]], !dbg [[DBG263]], !llvm.loop [[LOOP294:![0-9]+]]
 // CHECK1:       omp.dispatch.end:
-// CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB23:[0-9]+]], i32 [[TMP13]]), !dbg [[DBG293:![0-9]+]]
+// CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB19:[0-9]+]], i32 [[TMP13]]), !dbg [[DBG293:![0-9]+]]
 // CHECK1-NEXT:    ret void, !dbg [[DBG295:![0-9]+]]
 //
 //
diff --git a/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h b/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
--- a/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
+++ b/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
@@ -1468,8 +1468,7 @@
   /// Create a runtime call for kmpc_target_deinit
   ///
   /// \param Loc The insert and source location description.
-  /// \param IsSPMD Flag to indicate if the kernel is an SPMD kernel or not.
-  void createTargetDeinit(const LocationDescription &Loc, bool IsSPMD);
+  void createTargetDeinit(const LocationDescription &Loc);
 
   ///}
 
diff --git a/llvm/include/llvm/Frontend/OpenMP/OMPKinds.def b/llvm/include/llvm/Frontend/OpenMP/OMPKinds.def
--- a/llvm/include/llvm/Frontend/OpenMP/OMPKinds.def
+++ b/llvm/include/llvm/Frontend/OpenMP/OMPKinds.def
@@ -96,6 +96,11 @@
 		  Int64, Int64, Int32Arr3Ty, Int32Arr3Ty, Int32)
 __OMP_STRUCT_TYPE(AsyncInfo, __tgt_async_info, false, Int8Ptr)
 __OMP_STRUCT_TYPE(DependInfo, kmp_dep_info, false, SizeTy, SizeTy, Int8)
+__OMP_STRUCT_TYPE(ConfigurationEnvironment, ConfigurationEnvironmentTy, false,
+                  Int8, Int8, Int8)
+__OMP_STRUCT_TYPE(DynamicEnvironment, DynamicEnvironmentTy, false, Int16)
+__OMP_STRUCT_TYPE(KernelEnvironment, KernelEnvironmentTy, false,
+                  ConfigurationEnvironment, IdentPtr, DynamicEnvironmentPtr)
 
 #undef __OMP_STRUCT_TYPE
 #undef OMP_STRUCT_TYPE
@@ -452,8 +457,8 @@
           /* Int */ Int32, /* kmp_task_t */ VoidPtr)
 
 /// OpenMP Device runtime functions
-__OMP_RTL(__kmpc_target_init, false, Int32, IdentPtr, Int8, Int1)
-__OMP_RTL(__kmpc_target_deinit, false, Void, IdentPtr, Int8)
+__OMP_RTL(__kmpc_target_init, false, Int32, KernelEnvironmentPtr)
+__OMP_RTL(__kmpc_target_deinit, false, Void,)
 __OMP_RTL(__kmpc_kernel_prepare_parallel, false, Void, VoidPtr)
 __OMP_RTL(__kmpc_parallel_51, false, Void, IdentPtr, Int32, Int32, Int32, Int32,
           VoidPtr, VoidPtr, VoidPtrPtr, SizeTy)
@@ -1012,9 +1017,9 @@
                 ReturnPtrAttrs, ParamAttrs(ReadOnlyPtrAttrs, SExt))
 
 __OMP_RTL_ATTRS(__kmpc_target_init, AttributeSet(), SExt,
-                ParamAttrs(AttributeSet(), SExt, SExt))
+                ParamAttrs(AttributeSet()))
 __OMP_RTL_ATTRS(__kmpc_target_deinit, AttributeSet(), AttributeSet(),
-                ParamAttrs(AttributeSet(), SExt))
+                ParamAttrs())
 __OMP_RTL_ATTRS(__kmpc_parallel_51, AlwaysInlineAttrs, AttributeSet(),
                 ParamAttrs(AttributeSet(), SExt, SExt, SExt, SExt,
                            AttributeSet(), AttributeSet(), AttributeSet(),
diff --git a/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp b/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
--- a/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
+++ b/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
@@ -22,6 +22,7 @@
 #include "llvm/Analysis/ScalarEvolution.h"
 #include "llvm/Analysis/TargetLibraryInfo.h"
 #include "llvm/IR/CFG.h"
+#include "llvm/IR/Constant.h"
 #include "llvm/IR/Constants.h"
 #include "llvm/IR/DebugInfoMetadata.h"
 #include "llvm/IR/DerivedTypes.h"
@@ -3895,14 +3896,60 @@
   ConstantInt *IsSPMDVal = ConstantInt::getSigned(
       IntegerType::getInt8Ty(Int8->getContext()),
       IsSPMD ? OMP_TGT_EXEC_MODE_SPMD : OMP_TGT_EXEC_MODE_GENERIC);
-  ConstantInt *UseGenericStateMachine =
-      ConstantInt::getBool(Int32->getContext(), !IsSPMD);
+  ConstantInt *UseGenericStateMachineVal = ConstantInt::getSigned(
+      IntegerType::getInt8Ty(Int8->getContext()), !IsSPMD);
+  ConstantInt *MayUseNestedParallelismVal =
+      ConstantInt::getSigned(IntegerType::getInt8Ty(Int8->getContext()), true);
+  ConstantInt *DebugIndentionLevelVal =
+      ConstantInt::getSigned(IntegerType::getInt16Ty(Int8->getContext()), 0);
+
+  // We need to strip the debug prefix to get the correct kernel name.
+  Function *Kernel = Builder.GetInsertBlock()->getParent();
+  StringRef KernelName = Kernel->getName();
+  const std::string DebugPrefix = "_debug__";
+  if (KernelName.ends_with(DebugPrefix))
+    KernelName = KernelName.drop_back(DebugPrefix.length());
 
   Function *Fn = getOrCreateRuntimeFunctionPtr(
       omp::RuntimeFunction::OMPRTL___kmpc_target_init);
-
-  CallInst *ThreadKind = Builder.CreateCall(
-      Fn, {Ident, IsSPMDVal, UseGenericStateMachine});
+  const DataLayout &DL = Fn->getParent()->getDataLayout();
+
+  Twine DynamicEnvironmentName = KernelName + "_dynamic_environment";
+  Constant *DynamicEnvironmentInitializer =
+      ConstantStruct::get(DynamicEnvironment, {DebugIndentionLevelVal});
+  Constant *DynamicEnvironmentGV = new GlobalVariable(
+      M, DynamicEnvironment, /* IsConstant */ false,
+      GlobalValue::InternalLinkage, DynamicEnvironmentInitializer,
+      DynamicEnvironmentName,
+      /* InsertBefore */ nullptr, llvm::GlobalValue::NotThreadLocal,
+      DL.getDefaultGlobalsAddressSpace());
+  if (DynamicEnvironmentGV->getType() != DynamicEnvironmentPtr)
+    DynamicEnvironmentGV = ConstantExpr::getAddrSpaceCast(
+        DynamicEnvironmentGV, DynamicEnvironmentPtr);
+
+  Constant *ConfigurationEnvironmentInitializer = ConstantStruct::get(
+      ConfigurationEnvironment, {
+                                    UseGenericStateMachineVal,
+                                    MayUseNestedParallelismVal,
+                                    IsSPMDVal,
+                                });
+  Constant *KernelEnvironmentInitializer = ConstantStruct::get(
+      KernelEnvironment, {
+                             ConfigurationEnvironmentInitializer,
+                             Ident,
+                             DynamicEnvironmentGV,
+                         });
+  Twine KernelEnvironmentName = KernelName + "_kernel_environment";
+  Constant *KernelEnvironmentGV = new GlobalVariable(
+      M, KernelEnvironment, /* IsConstant */ true, GlobalValue::ExternalLinkage,
+      KernelEnvironmentInitializer, KernelEnvironmentName,
+      /* InsertBefore */ nullptr, llvm::GlobalValue::NotThreadLocal,
+      DL.getDefaultGlobalsAddressSpace());
+  if (KernelEnvironmentGV->getType() != KernelEnvironmentPtr)
+    KernelEnvironmentGV = ConstantExpr::getAddrSpaceCast(KernelEnvironmentGV,
+                                                         KernelEnvironmentPtr);
+
+  CallInst *ThreadKind = Builder.CreateCall(Fn, {KernelEnvironmentGV});
 
   Value *ExecUserCode = Builder.CreateICmpEQ(
       ThreadKind, ConstantInt::get(ThreadKind->getType(), -1),
@@ -3935,22 +3982,14 @@
   return InsertPointTy(UserCodeEntryBB, UserCodeEntryBB->getFirstInsertionPt());
 }
 
-void OpenMPIRBuilder::createTargetDeinit(const LocationDescription &Loc,
-                                         bool IsSPMD) {
+void OpenMPIRBuilder::createTargetDeinit(const LocationDescription &Loc) {
   if (!updateToLocation(Loc))
     return;
 
-  uint32_t SrcLocStrSize;
-  Constant *SrcLocStr = getOrCreateSrcLocStr(Loc, SrcLocStrSize);
-  Value *Ident = getOrCreateIdent(SrcLocStr, SrcLocStrSize);
-  ConstantInt *IsSPMDVal = ConstantInt::getSigned(
-      IntegerType::getInt8Ty(Int8->getContext()),
-      IsSPMD ? OMP_TGT_EXEC_MODE_SPMD : OMP_TGT_EXEC_MODE_GENERIC);
-
   Function *Fn = getOrCreateRuntimeFunctionPtr(
       omp::RuntimeFunction::OMPRTL___kmpc_target_deinit);
 
-  Builder.CreateCall(Fn, {Ident, IsSPMDVal});
+  Builder.CreateCall(Fn, {});
 }
 
 void OpenMPIRBuilder::setOutlinedTargetRegionFunctionAttributes(
diff --git a/llvm/lib/Transforms/IPO/OpenMPOpt.cpp b/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
--- a/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
+++ b/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
@@ -31,6 +31,7 @@
 #include "llvm/Analysis/OptimizationRemarkEmitter.h"
 #include "llvm/Analysis/ValueTracking.h"
 #include "llvm/Frontend/OpenMP/OMPConstants.h"
+#include "llvm/Frontend/OpenMP/OMPDeviceConstants.h"
 #include "llvm/Frontend/OpenMP/OMPIRBuilder.h"
 #include "llvm/IR/Assumptions.h"
 #include "llvm/IR/BasicBlock.h"
@@ -44,6 +45,7 @@
 #include "llvm/IR/IntrinsicsAMDGPU.h"
 #include "llvm/IR/IntrinsicsNVPTX.h"
 #include "llvm/IR/LLVMContext.h"
+#include "llvm/IR/Value.h"
 #include "llvm/InitializePasses.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
@@ -177,6 +179,80 @@
 static constexpr auto TAG = "[" DEBUG_TYPE "]";
 #endif
 
+namespace KernelInfo {
+
+// struct ConfigurationEnvironmentTy {
+//   uint8_t UseGenericStateMachine;
+//   uint8_t MayUseNestedParallelism;
+//   llvm::omp::OMPTgtExecModeFlags ExecMode;
+// };
+
+// struct DynamicEnvironmentTy {
+//   uint16_t DebugIndentionLevel;
+// };
+
+// struct KernelEnvironmentTy {
+//   ConfigurationEnvironmentTy Configuration;
+//   IdentTy *Ident;
+//   DynamicEnvironmentTy *DynamicEnv;
+// };
+
+#define KERNEL_ENVIRONMENT_IDX(MEMBER, IDX)                                    \
+  constexpr const unsigned MEMBER##Idx = IDX;
+
+KERNEL_ENVIRONMENT_IDX(Configuration, 0)
+KERNEL_ENVIRONMENT_IDX(Ident, 1)
+
+#undef KERNEL_ENVIRONMENT_IDX
+
+#define KERNEL_ENVIRONMENT_CONFIGURATION_IDX(MEMBER, IDX)                      \
+  constexpr const unsigned MEMBER##Idx = IDX;
+
+KERNEL_ENVIRONMENT_CONFIGURATION_IDX(UseGenericStateMachine, 0)
+KERNEL_ENVIRONMENT_CONFIGURATION_IDX(MayUseNestedParallelism, 1)
+KERNEL_ENVIRONMENT_CONFIGURATION_IDX(ExecMode, 2)
+
+#undef KERNEL_ENVIRONMENT_CONFIGURATION_IDX
+
+#define KERNEL_ENVIRONMENT_GETTER(MEMBER, RETURNTYPE)                          \
+  RETURNTYPE *get##MEMBER##FromKernelEnvironment(ConstantStruct *KernelEnvC) { \
+    return cast<RETURNTYPE>(KernelEnvC->getAggregateElement(MEMBER##Idx));     \
+  }
+
+KERNEL_ENVIRONMENT_GETTER(Ident, Constant)
+KERNEL_ENVIRONMENT_GETTER(Configuration, ConstantStruct)
+
+#undef KERNEL_ENVIRONMENT_GETTER
+
+#define KERNEL_ENVIRONMENT_CONFIGURATION_GETTER(MEMBER)                        \
+  ConstantInt *get##MEMBER##FromKernelEnvironment(                             \
+      ConstantStruct *KernelEnvC) {                                            \
+    ConstantStruct *ConfigC =                                                  \
+        getConfigurationFromKernelEnvironment(KernelEnvC);                     \
+    return dyn_cast<ConstantInt>(ConfigC->getAggregateElement(MEMBER##Idx));   \
+  }
+
+KERNEL_ENVIRONMENT_CONFIGURATION_GETTER(UseGenericStateMachine)
+KERNEL_ENVIRONMENT_CONFIGURATION_GETTER(MayUseNestedParallelism)
+KERNEL_ENVIRONMENT_CONFIGURATION_GETTER(ExecMode)
+
+#undef KERNEL_ENVIRONMENT_CONFIGURATION_GETTER
+
+GlobalVariable *
+getKernelEnvironementGVFromKernelInitCB(CallBase *KernelInitCB) {
+  constexpr const int InitKernelEnvironmentArgNo = 0;
+  return cast<GlobalVariable>(
+      KernelInitCB->getArgOperand(InitKernelEnvironmentArgNo)
+          ->stripPointerCasts());
+}
+
+ConstantStruct *getKernelEnvironementFromKernelInitCB(CallBase *KernelInitCB) {
+  GlobalVariable *KernelEnvGV =
+      getKernelEnvironementGVFromKernelInitCB(KernelInitCB);
+  return cast<ConstantStruct>(KernelEnvGV->getInitializer());
+}
+} // namespace KernelInfo
+
 namespace {
 
 struct AAHeapToShared;
@@ -609,6 +685,10 @@
   /// one we abort as the kernel is malformed.
   CallBase *KernelInitCB = nullptr;
 
+  /// The constant kernel environement as taken from and passed to
+  /// __kmpc_target_init.
+  ConstantStruct *KernelEnvC = nullptr;
+
   /// The __kmpc_target_deinit call in this kernel, if any. If we find more than
   /// one we abort as the kernel is malformed.
   CallBase *KernelDeinitCB = nullptr;
@@ -713,6 +793,12 @@
                          "assumptions.");
       KernelDeinitCB = KIS.KernelDeinitCB;
     }
+    if (KIS.KernelEnvC) {
+      if (KernelEnvC && KernelEnvC != KIS.KernelEnvC)
+        llvm_unreachable("Kernel that calls another kernel violates OpenMP-Opt "
+                         "assumptions.");
+      KernelEnvC = KIS.KernelEnvC;
+    }
     SPMDCompatibilityTracker ^= KIS.SPMDCompatibilityTracker;
     ReachedKnownParallelRegions ^= KIS.ReachedKnownParallelRegions;
     ReachedUnknownParallelRegions ^= KIS.ReachedUnknownParallelRegions;
@@ -2790,9 +2876,11 @@
       CB = CB ? OpenMPOpt::getCallIfRegularCall(*CB, &RFI) : nullptr;
       if (!CB)
         return false;
-      const int InitModeArgNo = 1;
-      auto *ModeCI = dyn_cast<ConstantInt>(CB->getOperand(InitModeArgNo));
-      return ModeCI && (ModeCI->getSExtValue() & OMP_TGT_EXEC_MODE_GENERIC);
+      ConstantStruct *KernelEnvC =
+          KernelInfo::getKernelEnvironementFromKernelInitCB(CB);
+      ConstantInt *ExecModeC =
+          KernelInfo::getExecModeFromKernelEnvironment(KernelEnvC);
+      return ExecModeC->getSExtValue() & OMP_TGT_EXEC_MODE_GENERIC;
     }
 
     if (C->isZero()) {
@@ -3434,6 +3522,29 @@
     return GuardedInstructions;
   }
 
+  void setConfigurationOfKernelEnvironment(ConstantStruct *ConfigC) {
+    Constant *NewKernelEnvC = ConstantFoldInsertValueInstruction(
+        KernelEnvC, ConfigC, {KernelInfo::ConfigurationIdx});
+    assert(NewKernelEnvC && "Failed to create new kernel environment");
+    KernelEnvC = cast<ConstantStruct>(NewKernelEnvC);
+  }
+
+#define KERNEL_ENVIRONMENT_CONFIGURATION_SETTER(MEMBER)                        \
+  void set##MEMBER##OfKernelEnvironment(ConstantInt *NewVal) {                 \
+    ConstantStruct *ConfigC =                                                  \
+        KernelInfo::getConfigurationFromKernelEnvironment(KernelEnvC);         \
+    Constant *NewConfigC = ConstantFoldInsertValueInstruction(                 \
+        ConfigC, NewVal, {KernelInfo::MEMBER##Idx});                           \
+    assert(NewConfigC && "Failed to create new configuration environment");    \
+    setConfigurationOfKernelEnvironment(cast<ConstantStruct>(NewConfigC));     \
+  }
+
+  KERNEL_ENVIRONMENT_CONFIGURATION_SETTER(UseGenericStateMachine)
+  KERNEL_ENVIRONMENT_CONFIGURATION_SETTER(MayUseNestedParallelism)
+  KERNEL_ENVIRONMENT_CONFIGURATION_SETTER(ExecMode)
+
+#undef KERNEL_ENVIRONMENT_CONFIGURATION_SETTER
+
   /// See AbstractAttribute::initialize(...).
   void initialize(Attributor &A) override {
     // This is a high-level transform that might change the constant arguments
@@ -3482,76 +3593,39 @@
     ReachingKernelEntries.insert(Fn);
     IsKernelEntry = true;
 
-    // For kernels we might need to initialize/finalize the IsSPMD state and
-    // we need to register a simplification callback so that the Attributor
-    // knows the constant arguments to __kmpc_target_init and
-    // __kmpc_target_deinit might actually change.
+    KernelEnvC =
+        KernelInfo::getKernelEnvironementFromKernelInitCB(KernelInitCB);
 
-    Attributor::SimplifictionCallbackTy StateMachineSimplifyCB =
+    Attributor::SimplifictionCallbackTy KernelConfigurationSimplifyCB =
         [&](const IRPosition &IRP, const AbstractAttribute *AA,
             bool &UsedAssumedInformation) -> std::optional<Value *> {
-      // IRP represents the "use generic state machine" argument of an
-      // __kmpc_target_init call. We will answer this one with the internal
-      // state. As long as we are not in an invalid state, we will create a
-      // custom state machine so the value should be a `i1 false`. If we are
-      // in an invalid state, we won't change the value that is in the IR.
-      if (!ReachedKnownParallelRegions.isValidState())
-        return nullptr;
-      // If we have disabled state machine rewrites, don't make a custom one.
-      if (DisableOpenMPOptStateMachineRewrite)
-        return nullptr;
-      if (AA)
-        A.recordDependence(*this, *AA, DepClassTy::OPTIONAL);
-      UsedAssumedInformation = !isAtFixpoint();
-      auto *FalseVal =
-          ConstantInt::getBool(IRP.getAnchorValue().getContext(), false);
-      return FalseVal;
+      return KernelEnvC;
     };
 
-    Attributor::SimplifictionCallbackTy ModeSimplifyCB =
-        [&](const IRPosition &IRP, const AbstractAttribute *AA,
-            bool &UsedAssumedInformation) -> std::optional<Value *> {
-      // IRP represents the "SPMDCompatibilityTracker" argument of an
-      // __kmpc_target_init or
-      // __kmpc_target_deinit call. We will answer this one with the internal
-      // state.
-      if (!SPMDCompatibilityTracker.isValidState())
-        return nullptr;
-      if (!SPMDCompatibilityTracker.isAtFixpoint()) {
-        if (AA)
-          A.recordDependence(*this, *AA, DepClassTy::OPTIONAL);
-        UsedAssumedInformation = true;
-      } else {
-        UsedAssumedInformation = false;
-      }
-      auto *Val = ConstantInt::getSigned(
-          IntegerType::getInt8Ty(IRP.getAnchorValue().getContext()),
-          SPMDCompatibilityTracker.isAssumed() ? OMP_TGT_EXEC_MODE_SPMD
-                                               : OMP_TGT_EXEC_MODE_GENERIC);
-      return Val;
-    };
-
-    constexpr const int InitModeArgNo = 1;
-    constexpr const int DeinitModeArgNo = 1;
-    constexpr const int InitUseStateMachineArgNo = 2;
-    A.registerSimplificationCallback(
-        IRPosition::callsite_argument(*KernelInitCB, InitUseStateMachineArgNo),
-        StateMachineSimplifyCB);
-    A.registerSimplificationCallback(
-        IRPosition::callsite_argument(*KernelInitCB, InitModeArgNo),
-        ModeSimplifyCB);
-    A.registerSimplificationCallback(
-        IRPosition::callsite_argument(*KernelDeinitCB, DeinitModeArgNo),
-        ModeSimplifyCB);
+    A.registerSimplificationCallback(IRPosition::value(*KernelEnvC),
+                                     KernelConfigurationSimplifyCB);
 
     // Check if we know we are in SPMD-mode already.
-    ConstantInt *ModeArg =
-        dyn_cast<ConstantInt>(KernelInitCB->getArgOperand(InitModeArgNo));
-    if (ModeArg && (ModeArg->getSExtValue() & OMP_TGT_EXEC_MODE_SPMD))
+    ConstantInt *ExecModeC =
+        KernelInfo::getExecModeFromKernelEnvironment(KernelEnvC);
+    ConstantInt *AssumedExecModeC = ConstantInt::get(
+        ExecModeC->getType(),
+        ExecModeC->getSExtValue() | OMP_TGT_EXEC_MODE_GENERIC_SPMD);
+    if (ExecModeC->getSExtValue() & OMP_TGT_EXEC_MODE_SPMD)
       SPMDCompatibilityTracker.indicateOptimisticFixpoint();
-    // This is a generic region but SPMDization is disabled so stop tracking.
     else if (DisableOpenMPOptSPMDization)
+      // This is a generic region but SPMDization is disabled so stop
+      // tracking.
       SPMDCompatibilityTracker.indicatePessimisticFixpoint();
+    else
+      setExecModeOfKernelEnvironment(AssumedExecModeC);
+
+    ConstantInt *MayUseNestedParallelismC =
+        KernelInfo::getMayUseNestedParallelismFromKernelEnvironment(KernelEnvC);
+    ConstantInt *AssumedMayUseNestedParallelismC = ConstantInt::get(
+        MayUseNestedParallelismC->getType(), NestedParallelism);
+    setMayUseNestedParallelismOfKernelEnvironment(
+        AssumedMayUseNestedParallelismC);
 
     // Register virtual uses of functions we might need to preserve.
     auto RegisterVirtualUse = [&](RuntimeFunction RFKind,
@@ -3652,21 +3726,21 @@
     if (!KernelInitCB || !KernelDeinitCB)
       return ChangeStatus::UNCHANGED;
 
-    /// Insert nested Parallelism global variable
-    Function *Kernel = getAnchorScope();
-    Module &M = *Kernel->getParent();
-    Type *Int8Ty = Type::getInt8Ty(M.getContext());
-    new GlobalVariable(M, Int8Ty, /* isConstant */ true,
-                       GlobalValue::WeakAnyLinkage,
-                       ConstantInt::get(Int8Ty, NestedParallelism ? 1 : 0),
-                       Kernel->getName() + "_nested_parallelism");
+    ChangeStatus Changed = ChangeStatus::UNCHANGED;
 
     // If we can we change the execution mode to SPMD-mode otherwise we build a
     // custom state machine.
-    ChangeStatus Changed = ChangeStatus::UNCHANGED;
     if (!changeToSPMDMode(A, Changed)) {
       if (!KernelInitCB->getCalledFunction()->isDeclaration())
-        return buildCustomStateMachine(A);
+        Changed |= buildCustomStateMachine(A);
+    }
+
+    // At last, update the KernelEnvc
+    GlobalVariable *KernelEnvGV =
+        KernelInfo::getKernelEnvironementGVFromKernelInitCB(KernelInitCB);
+    if (KernelEnvGV->getInitializer() != KernelEnvC) {
+      KernelEnvGV->setInitializer(KernelEnvC);
+      Changed = ChangeStatus::CHANGED;
     }
 
     return Changed;
@@ -3736,14 +3810,14 @@
       // Find escaping outputs from the guarded region to outside users and
       // broadcast their values to them.
       for (Instruction &I : *RegionStartBB) {
-        SmallPtrSet<Instruction *, 4> OutsideUsers;
-        for (User *Usr : I.users()) {
-          Instruction &UsrI = *cast<Instruction>(Usr);
+        SmallVector<Use *, 4> OutsideUses;
+        for (Use &U : I.uses()) {
+          Instruction &UsrI = *cast<Instruction>(U.getUser());
           if (UsrI.getParent() != RegionStartBB)
-            OutsideUsers.insert(&UsrI);
+            OutsideUses.push_back(&U);
         }
 
-        if (OutsideUsers.empty())
+        if (OutsideUses.empty())
           continue;
 
         HasBroadcastValues = true;
@@ -3766,8 +3840,8 @@
                                        RegionBarrierBB->getTerminator());
 
         // Emit a load instruction and replace uses of the output value.
-        for (Instruction *UsrI : OutsideUsers)
-          UsrI->replaceUsesOfWith(&I, LoadI);
+        for (Use *U : OutsideUses)
+          A.changeUseAfterManifest(*U, *LoadI);
       }
 
       auto &OMPInfoCache = static_cast<OMPInformationCache &>(A.getInfoCache());
@@ -3994,16 +4068,11 @@
     assert(OMPInfoCache.Kernels.count(Kernel) && "Expected kernel function!");
 
     // Check if the kernel is already in SPMD mode, if so, return success.
-    GlobalVariable *ExecMode = Kernel->getParent()->getGlobalVariable(
-        (Kernel->getName() + "_exec_mode").str());
-    assert(ExecMode && "Kernel without exec mode?");
-    assert(ExecMode->getInitializer() && "ExecMode doesn't have initializer!");
-
-    // Set the global exec mode flag to indicate SPMD-Generic mode.
-    assert(isa<ConstantInt>(ExecMode->getInitializer()) &&
-           "ExecMode is not an integer!");
-    const int8_t ExecModeVal =
-        cast<ConstantInt>(ExecMode->getInitializer())->getSExtValue();
+    ConstantStruct *ExistingKernelEnvC =
+        KernelInfo::getKernelEnvironementFromKernelInitCB(KernelInitCB);
+    auto *ExecModeC =
+        KernelInfo::getExecModeFromKernelEnvironment(ExistingKernelEnvC);
+    const int8_t ExecModeVal = ExecModeC->getSExtValue();
     if (ExecModeVal != OMP_TGT_EXEC_MODE_GENERIC)
       return true;
 
@@ -4021,27 +4090,8 @@
     // kernel is executed in.
     assert(ExecModeVal == OMP_TGT_EXEC_MODE_GENERIC &&
            "Initially non-SPMD kernel has SPMD exec mode!");
-    ExecMode->setInitializer(
-        ConstantInt::get(ExecMode->getInitializer()->getType(),
-                         ExecModeVal | OMP_TGT_EXEC_MODE_GENERIC_SPMD));
-
-    // Next rewrite the init and deinit calls to indicate we use SPMD-mode now.
-    const int InitModeArgNo = 1;
-    const int DeinitModeArgNo = 1;
-    const int InitUseStateMachineArgNo = 2;
-
-    auto &Ctx = getAnchorValue().getContext();
-    A.changeUseAfterManifest(
-        KernelInitCB->getArgOperandUse(InitModeArgNo),
-        *ConstantInt::getSigned(IntegerType::getInt8Ty(Ctx),
-                                OMP_TGT_EXEC_MODE_SPMD));
-    A.changeUseAfterManifest(
-        KernelInitCB->getArgOperandUse(InitUseStateMachineArgNo),
-        *ConstantInt::getBool(Ctx, false));
-    A.changeUseAfterManifest(
-        KernelDeinitCB->getArgOperandUse(DeinitModeArgNo),
-        *ConstantInt::getSigned(IntegerType::getInt8Ty(Ctx),
-                                OMP_TGT_EXEC_MODE_SPMD));
+    setExecModeOfKernelEnvironment(ConstantInt::get(
+        ExecModeC->getType(), ExecModeVal | OMP_TGT_EXEC_MODE_GENERIC_SPMD));
 
     ++NumOpenMPTargetRegionKernelsSPMD;
 
@@ -4068,30 +4118,29 @@
              OMPRTL___kmpc_kernel_parallel, OMPRTL___kmpc_kernel_end_parallel}))
       return ChangeStatus::UNCHANGED;
 
-    const int InitModeArgNo = 1;
-    const int InitUseStateMachineArgNo = 2;
+    ConstantStruct *ExistingKernelEnvC =
+        KernelInfo::getKernelEnvironementFromKernelInitCB(KernelInitCB);
 
     // Check if the current configuration is non-SPMD and generic state machine.
     // If we already have SPMD mode or a custom state machine we do not need to
     // go any further. If it is anything but a constant something is weird and
     // we give up.
-    ConstantInt *UseStateMachine = dyn_cast<ConstantInt>(
-        KernelInitCB->getArgOperand(InitUseStateMachineArgNo));
-    ConstantInt *Mode =
-        dyn_cast<ConstantInt>(KernelInitCB->getArgOperand(InitModeArgNo));
+    ConstantInt *UseStateMachineC =
+        KernelInfo::getUseGenericStateMachineFromKernelEnvironment(
+            ExistingKernelEnvC);
+    ConstantInt *ModeC =
+        KernelInfo::getExecModeFromKernelEnvironment(ExistingKernelEnvC);
 
     // If we are stuck with generic mode, try to create a custom device (=GPU)
     // state machine which is specialized for the parallel regions that are
     // reachable by the kernel.
-    if (!UseStateMachine || UseStateMachine->isZero() || !Mode ||
-        (Mode->getSExtValue() & OMP_TGT_EXEC_MODE_SPMD))
+    if (UseStateMachineC->isZero() ||
+        (ModeC->getSExtValue() & OMP_TGT_EXEC_MODE_SPMD))
       return ChangeStatus::UNCHANGED;
 
     // If not SPMD mode, indicate we use a custom state machine now.
-    auto &Ctx = getAnchorValue().getContext();
-    auto *FalseVal = ConstantInt::getBool(Ctx, false);
-    A.changeUseAfterManifest(
-        KernelInitCB->getArgOperandUse(InitUseStateMachineArgNo), *FalseVal);
+    setUseGenericStateMachineOfKernelEnvironment(
+        ConstantInt::get(UseStateMachineC->getType(), false));
 
     // If we don't actually need a state machine we are done here. This can
     // happen if there simply are no parallel regions. In the resulting kernel
@@ -4170,6 +4219,7 @@
     // UserCodeEntryBB:      // user code
     //                       __kmpc_target_deinit(...)
     //
+    auto &Ctx = getAnchorValue().getContext();
     Function *Kernel = getAssociatedFunction();
     assert(Kernel && "Expected an associated function!");
 
@@ -4252,7 +4302,7 @@
                                      StateMachineBeginBB->end()),
             DLoc));
 
-    Value *Ident = KernelInitCB->getArgOperand(0);
+    Value *Ident = KernelInfo::getIdentFromKernelEnvironment(KernelEnvC);
     Value *GTid = KernelInitCB;
 
     FunctionCallee BarrierFn =
@@ -4382,6 +4432,46 @@
   ChangeStatus updateImpl(Attributor &A) override {
     KernelInfoState StateBefore = getState();
 
+    // When we leave this function this RAII will make sure the member
+    // KernelEnvC is updated properly depending on the state. That member is
+    // used for simplification of values and needs to be up to date at all
+    // times.
+    struct UpdateKernelEnvCRAII {
+      AAKernelInfoFunction &AA;
+
+      UpdateKernelEnvCRAII(AAKernelInfoFunction &AA) : AA(AA) {}
+
+      ~UpdateKernelEnvCRAII() {
+        if (!AA.KernelEnvC)
+          return;
+
+        ConstantStruct *ExistingKernelEnvC =
+            KernelInfo::getKernelEnvironementFromKernelInitCB(AA.KernelInitCB);
+
+        if (!AA.isValidState()) {
+          AA.KernelEnvC = ExistingKernelEnvC;
+          return;
+        }
+
+        if (!AA.ReachedKnownParallelRegions.isValidState())
+          AA.setUseGenericStateMachineOfKernelEnvironment(
+              KernelInfo::getUseGenericStateMachineFromKernelEnvironment(
+                  ExistingKernelEnvC));
+
+        if (!AA.SPMDCompatibilityTracker.isValidState())
+          AA.setExecModeOfKernelEnvironment(
+              KernelInfo::getExecModeFromKernelEnvironment(ExistingKernelEnvC));
+
+        ConstantInt *MayUseNestedParallelismC =
+            KernelInfo::getMayUseNestedParallelismFromKernelEnvironment(
+                AA.KernelEnvC);
+        ConstantInt *NewMayUseNestedParallelismC = ConstantInt::get(
+            MayUseNestedParallelismC->getType(), AA.NestedParallelism);
+        AA.setMayUseNestedParallelismOfKernelEnvironment(
+            NewMayUseNestedParallelismC);
+      }
+    } RAII(*this);
+
     // Callback to check a read/write instruction.
     auto CheckRWInst = [&](Instruction &I) {
       // We handle calls later.
diff --git a/openmp/libomptarget/DeviceRTL/CMakeLists.txt b/openmp/libomptarget/DeviceRTL/CMakeLists.txt
--- a/openmp/libomptarget/DeviceRTL/CMakeLists.txt
+++ b/openmp/libomptarget/DeviceRTL/CMakeLists.txt
@@ -128,6 +128,7 @@
              -nocudalib -nogpulib -nostdinc
              -fopenmp -fopenmp-cuda-mode
              -Wno-unknown-cuda-version
+             -DOMPTARGET_DEVICE_RUNTIME
              -I${include_directory}
              -I${devicertl_base_directory}/../include
              ${LIBOMPTARGET_LLVM_INCLUDE_DIRS_DEVICERTL}
diff --git a/openmp/libomptarget/DeviceRTL/include/Debug.h b/openmp/libomptarget/DeviceRTL/include/Debug.h
--- a/openmp/libomptarget/DeviceRTL/include/Debug.h
+++ b/openmp/libomptarget/DeviceRTL/include/Debug.h
@@ -50,8 +50,6 @@
 struct DebugEntryRAII {
   DebugEntryRAII(const char *File, const unsigned Line, const char *Function);
   ~DebugEntryRAII();
-
-  static void init();
 };
 
 #endif
diff --git a/openmp/libomptarget/DeviceRTL/include/Interface.h b/openmp/libomptarget/DeviceRTL/include/Interface.h
--- a/openmp/libomptarget/DeviceRTL/include/Interface.h
+++ b/openmp/libomptarget/DeviceRTL/include/Interface.h
@@ -214,12 +214,14 @@
 /// Kernel
 ///
 ///{
+// Forward declaration
+struct KernelEnvironmentTy;
+
 int8_t __kmpc_is_spmd_exec_mode();
 
-int32_t __kmpc_target_init(IdentTy *Ident, int8_t Mode,
-                           bool UseGenericStateMachine);
+int32_t __kmpc_target_init(KernelEnvironmentTy &KernelEnvironment);
 
-void __kmpc_target_deinit(IdentTy *Ident, int8_t Mode);
+void __kmpc_target_deinit();
 
 ///}
 
diff --git a/openmp/libomptarget/DeviceRTL/include/State.h b/openmp/libomptarget/DeviceRTL/include/State.h
--- a/openmp/libomptarget/DeviceRTL/include/State.h
+++ b/openmp/libomptarget/DeviceRTL/include/State.h
@@ -17,6 +17,9 @@
 #include "Types.h"
 #include "Utils.h"
 
+// Forward declaration.
+struct KernelEnvironmentTy;
+
 #pragma omp begin declare target device_type(nohost)
 
 namespace ompx {
@@ -113,7 +116,10 @@
 #pragma omp allocate(ThreadStates) allocator(omp_pteam_mem_alloc)
 
 /// Initialize the state machinery. Must be called by all threads.
-void init(bool IsSPMD);
+void init(bool IsSPMD, KernelEnvironmentTy &KernelEnvironment);
+
+/// Return the kernel environment associated with the current kernel.
+KernelEnvironmentTy &getKernelEnvironment();
 
 /// TODO
 enum ValueKind {
diff --git a/openmp/libomptarget/DeviceRTL/src/Configuration.cpp b/openmp/libomptarget/DeviceRTL/src/Configuration.cpp
--- a/openmp/libomptarget/DeviceRTL/src/Configuration.cpp
+++ b/openmp/libomptarget/DeviceRTL/src/Configuration.cpp
@@ -12,7 +12,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "Configuration.h"
-#include "DeviceEnvironment.h"
+#include "Environment.h"
 #include "State.h"
 #include "Types.h"
 
@@ -53,7 +53,9 @@
 bool config::mayUseThreadStates() { return !__omp_rtl_assume_no_thread_state; }
 
 bool config::mayUseNestedParallelism() {
-  return !__omp_rtl_assume_no_nested_parallelism;
+  if (__omp_rtl_assume_no_nested_parallelism)
+    return false;
+  return state::getKernelEnvironment().Configuration.MayUseNestedParallelism;
 }
 
 #pragma omp end declare target
diff --git a/openmp/libomptarget/DeviceRTL/src/Debug.cpp b/openmp/libomptarget/DeviceRTL/src/Debug.cpp
--- a/openmp/libomptarget/DeviceRTL/src/Debug.cpp
+++ b/openmp/libomptarget/DeviceRTL/src/Debug.cpp
@@ -12,8 +12,10 @@
 
 #include "Debug.h"
 #include "Configuration.h"
+#include "Environment.h"
 #include "Interface.h"
 #include "Mapping.h"
+#include "State.h"
 #include "Types.h"
 
 using namespace ompx;
@@ -31,15 +33,14 @@
 }
 }
 
-/// Current indentation level for the function trace. Only accessed by thread 0.
-__attribute__((loader_uninitialized)) static uint32_t Level;
-#pragma omp allocate(Level) allocator(omp_pteam_mem_alloc)
-
 DebugEntryRAII::DebugEntryRAII(const char *File, const unsigned Line,
                                const char *Function) {
   if (config::isDebugMode(config::DebugKind::FunctionTracing) &&
       mapping::getThreadIdInBlock() == 0 && mapping::getBlockId() == 0) {
 
+    uint16_t &Level =
+        state::getKernelEnvironment().DynamicEnv->DebugIndentionLevel;
+
     for (int I = 0; I < Level; ++I)
       PRINTF("%s", "  ");
 
@@ -51,10 +52,11 @@
 
 DebugEntryRAII::~DebugEntryRAII() {
   if (config::isDebugMode(config::DebugKind::FunctionTracing) &&
-      mapping::getThreadIdInBlock() == 0 && mapping::getBlockId() == 0)
+      mapping::getThreadIdInBlock() == 0 && mapping::getBlockId() == 0) {
+    uint16_t &Level =
+        state::getKernelEnvironment().DynamicEnv->DebugIndentionLevel;
     Level--;
+  }
 }
 
-void DebugEntryRAII::init() { Level = 0; }
-
 #pragma omp end declare target
diff --git a/openmp/libomptarget/DeviceRTL/src/Kernel.cpp b/openmp/libomptarget/DeviceRTL/src/Kernel.cpp
--- a/openmp/libomptarget/DeviceRTL/src/Kernel.cpp
+++ b/openmp/libomptarget/DeviceRTL/src/Kernel.cpp
@@ -11,6 +11,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "Debug.h"
+#include "Environment.h"
 #include "Interface.h"
 #include "Mapping.h"
 #include "State.h"
@@ -23,11 +24,12 @@
 
 #pragma omp begin declare target device_type(nohost)
 
-static void inititializeRuntime(bool IsSPMD) {
+static void inititializeRuntime(bool IsSPMD,
+                                KernelEnvironmentTy &KernelEnvironment) {
   // Order is important here.
   synchronize::init(IsSPMD);
   mapping::init(IsSPMD);
-  state::init(IsSPMD);
+  state::init(IsSPMD, KernelEnvironment);
 }
 
 /// Simple generic state machine for worker threads.
@@ -67,16 +69,17 @@
 ///
 /// \param Ident               Source location identification, can be NULL.
 ///
-int32_t __kmpc_target_init(IdentTy *Ident, int8_t Mode,
-                           bool UseGenericStateMachine) {
+int32_t __kmpc_target_init(KernelEnvironmentTy &KernelEnvironment) {
   FunctionTracingRAII();
-  const bool IsSPMD =
-      Mode & llvm::omp::OMPTgtExecModeFlags::OMP_TGT_EXEC_MODE_SPMD;
+  ConfigurationEnvironmentTy &Configuration = KernelEnvironment.Configuration;
+  bool IsSPMD = Configuration.ExecMode &
+                llvm::omp::OMPTgtExecModeFlags::OMP_TGT_EXEC_MODE_SPMD;
+  bool UseGenericStateMachine = Configuration.UseGenericStateMachine;
   if (IsSPMD) {
-    inititializeRuntime(/* IsSPMD */ true);
+    inititializeRuntime(/* IsSPMD */ true, KernelEnvironment);
     synchronize::threadsAligned();
   } else {
-    inititializeRuntime(/* IsSPMD */ false);
+    inititializeRuntime(/* IsSPMD */ false, KernelEnvironment);
     // No need to wait since only the main threads will execute user
     // code and workers will run into a barrier right away.
   }
@@ -104,7 +107,7 @@
   // thread's warp, so none of its threads can ever be active worker threads.
   if (UseGenericStateMachine &&
       mapping::getThreadIdInBlock() < mapping::getBlockSize(IsSPMD)) {
-    genericStateMachine(Ident);
+    genericStateMachine(KernelEnvironment.Ident);
   } else {
     // Retrieve the work function just to ensure we always call
     // __kmpc_kernel_parallel even if a custom state machine is used.
@@ -128,10 +131,9 @@
 ///
 /// \param Ident Source location identification, can be NULL.
 ///
-void __kmpc_target_deinit(IdentTy *Ident, int8_t Mode) {
+void __kmpc_target_deinit() {
   FunctionTracingRAII();
-  const bool IsSPMD =
-      Mode & llvm::omp::OMPTgtExecModeFlags::OMP_TGT_EXEC_MODE_SPMD;
+  bool IsSPMD = mapping::isSPMDMode();
   state::assumeInitialState(IsSPMD);
   if (IsSPMD)
     return;
diff --git a/openmp/libomptarget/DeviceRTL/src/State.cpp b/openmp/libomptarget/DeviceRTL/src/State.cpp
--- a/openmp/libomptarget/DeviceRTL/src/State.cpp
+++ b/openmp/libomptarget/DeviceRTL/src/State.cpp
@@ -9,8 +9,8 @@
 //===----------------------------------------------------------------------===//
 
 #include "State.h"
-#include "Configuration.h"
 #include "Debug.h"
+#include "Environment.h"
 #include "Interface.h"
 #include "Mapping.h"
 #include "Synchronization.h"
@@ -34,6 +34,9 @@
 extern unsigned char DynamicSharedBuffer[] __attribute__((aligned(Alignment)));
 #pragma omp allocate(DynamicSharedBuffer) allocator(omp_pteam_mem_alloc)
 
+/// The kernel environment passed to the init method by the compiler.
+static KernelEnvironmentTy *SHARED(KernelEnvironmentPtr);
+
 namespace {
 
 /// Fallback implementations are missing to trigger a link time error.
@@ -241,15 +244,19 @@
 
 } // namespace
 
-void state::init(bool IsSPMD) {
+void state::init(bool IsSPMD, KernelEnvironmentTy &KernelEnvironment) {
   SharedMemorySmartStack.init(IsSPMD);
   if (mapping::isInitialThreadInLevel0(IsSPMD)) {
     TeamState.init(IsSPMD);
-    DebugEntryRAII::init();
     ThreadStates = nullptr;
+    KernelEnvironmentPtr = &KernelEnvironment;
   }
 }
 
+KernelEnvironmentTy &state::getKernelEnvironment() {
+  return *KernelEnvironmentPtr;
+}
+
 void state::enterDataEnvironment(IdentTy *Ident) {
   ASSERT(config::mayUseThreadStates() &&
          "Thread state modified while explicitly disabled!");
diff --git a/openmp/libomptarget/include/DeviceEnvironment.h b/openmp/libomptarget/include/DeviceEnvironment.h
deleted file mode 100644
--- a/openmp/libomptarget/include/DeviceEnvironment.h
+++ /dev/null
@@ -1,25 +0,0 @@
-//===---- device_environment.h - OpenMP GPU device environment ---- C++ -*-===//
-//
-// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
-// See https://llvm.org/LICENSE.txt for license information.
-// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
-//
-//===----------------------------------------------------------------------===//
-//
-// Global device environment
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef _OMPTARGET_DEVICE_ENVIRONMENT_H_
-#define _OMPTARGET_DEVICE_ENVIRONMENT_H_
-
-// deviceRTL uses <stdint> and DeviceRTL uses explicit definitions
-
-struct DeviceEnvironmentTy {
-  uint32_t DebugKind;
-  uint32_t NumDevices;
-  uint32_t DeviceNum;
-  uint32_t DynamicMemSize;
-};
-
-#endif
diff --git a/openmp/libomptarget/include/Environment.h b/openmp/libomptarget/include/Environment.h
new file mode 100644
--- /dev/null
+++ b/openmp/libomptarget/include/Environment.h
@@ -0,0 +1,61 @@
+//===------------ Environment.h - OpenMP GPU environments --------- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// Environments shared between host and device.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef _OMPTARGET_ENVIRONMENT_H_
+#define _OMPTARGET_ENVIRONMENT_H_
+
+#ifdef OMPTARGET_DEVICE_RUNTIME
+#include "Types.h"
+#else
+#include "SourceInfo.h"
+
+#include <cstdint>
+
+using IdentTy = ident_t;
+#endif
+
+#include "llvm/Frontend/OpenMP/OMPDeviceConstants.h"
+
+struct DeviceEnvironmentTy {
+  uint32_t DebugKind;
+  uint32_t NumDevices;
+  uint32_t DeviceNum;
+  uint32_t DynamicMemSize;
+};
+
+// NOTE: Please don't change the order of those members as their indices are
+// used in the middle end. Always add the new data member at the end.
+// Different from KernelEnvironmentTy below, this structure contains members
+// that might be modified at runtime.
+struct DynamicEnvironmentTy {
+  /// Current indentation level for the function trace. Only accessed by thread
+  /// 0.
+  uint16_t DebugIndentionLevel;
+};
+
+// NOTE: Please don't change the order of those members as their indices are
+// used in the middle end. Always add the new data member at the end.
+struct ConfigurationEnvironmentTy {
+  uint8_t UseGenericStateMachine;
+  uint8_t MayUseNestedParallelism;
+  llvm::omp::OMPTgtExecModeFlags ExecMode;
+};
+
+// NOTE: Please don't change the order of those members as their indices are
+// used in the middle end. Always add the new data member at the end.
+struct KernelEnvironmentTy {
+  ConfigurationEnvironmentTy Configuration;
+  IdentTy *Ident;
+  DynamicEnvironmentTy *DynamicEnv;
+};
+
+#endif // _OMPTARGET_ENVIRONMENT_H_
diff --git a/openmp/libomptarget/plugins-nextgen/amdgpu/src/rtl.cpp b/openmp/libomptarget/plugins-nextgen/amdgpu/src/rtl.cpp
--- a/openmp/libomptarget/plugins-nextgen/amdgpu/src/rtl.cpp
+++ b/openmp/libomptarget/plugins-nextgen/amdgpu/src/rtl.cpp
@@ -21,7 +21,7 @@
 #include <unordered_map>
 
 #include "Debug.h"
-#include "DeviceEnvironment.h"
+#include "Environment.h"
 #include "GlobalHandler.h"
 #include "PluginInterface.h"
 #include "Utilities.h"
diff --git a/openmp/libomptarget/plugins-nextgen/common/PluginInterface/PluginInterface.h b/openmp/libomptarget/plugins-nextgen/common/PluginInterface/PluginInterface.h
--- a/openmp/libomptarget/plugins-nextgen/common/PluginInterface/PluginInterface.h
+++ b/openmp/libomptarget/plugins-nextgen/common/PluginInterface/PluginInterface.h
@@ -19,7 +19,7 @@
 #include <vector>
 
 #include "Debug.h"
-#include "DeviceEnvironment.h"
+#include "Environment.h"
 #include "GlobalHandler.h"
 #include "JIT.h"
 #include "MemoryManager.h"
@@ -732,6 +732,11 @@
 
   /// Map of host pinned allocations used for optimize device transfers.
   PinnedAllocationMapTy PinnedAllocs;
+
+private:
+  /// Return the kernel environment object for kernel \p Name.
+  Expected<KernelEnvironmentTy>
+  getKernelEnvironmentForKernel(StringRef Name, DeviceImageTy &Image);
 };
 
 /// Class implementing common functionalities of offload plugins. Each plugin
diff --git a/openmp/libomptarget/plugins-nextgen/common/PluginInterface/PluginInterface.cpp b/openmp/libomptarget/plugins-nextgen/common/PluginInterface/PluginInterface.cpp
--- a/openmp/libomptarget/plugins-nextgen/common/PluginInterface/PluginInterface.cpp
+++ b/openmp/libomptarget/plugins-nextgen/common/PluginInterface/PluginInterface.cpp
@@ -554,32 +554,45 @@
   return Plugin::success();
 }
 
-Expected<OMPTgtExecModeFlags>
-GenericDeviceTy::getExecutionModeForKernel(StringRef Name,
-                                           DeviceImageTy &Image) {
-  // Create a metadata object for the exec mode global (auto-generated).
-  StaticGlobalTy<llvm::omp::OMPTgtExecModeFlags> ExecModeGlobal(Name.data(),
-                                                                "_exec_mode");
-
-  // Retrieve execution mode for the kernel. This may fail since some kernels
-  // may not have an execution mode.
+Expected<KernelEnvironmentTy>
+GenericDeviceTy::getKernelEnvironmentForKernel(StringRef Name,
+                                               DeviceImageTy &Image) {
+  // Create a metadata object for the kernel environment object.
+  StaticGlobalTy<KernelEnvironmentTy> KernelEnv(Name.data(),
+                                                "_kernel_environment");
+
+  // Retrieve kernel environment object for the kernel.
   GenericGlobalHandlerTy &GHandler = Plugin::get().getGlobalHandler();
-  if (auto Err = GHandler.readGlobalFromImage(*this, Image, ExecModeGlobal)) {
+  if (auto Err = GHandler.readGlobalFromImage(*this, Image, KernelEnv)) {
     // Consume the error since it is acceptable to fail.
     [[maybe_unused]] std::string ErrStr = toString(std::move(Err));
-    DP("Failed to read execution mode for '%s': %s\n"
-       "Using default SPMD (2) execution mode\n",
-       Name.data(), ErrStr.data());
+    DP("Failed to read kernel environment object for '%s': %s\n", Name.data(),
+       ErrStr.data());
+
+    return createStringError(inconvertibleErrorCode(), ErrStr);
+  }
 
+  return KernelEnv.getValue();
+}
+
+Expected<OMPTgtExecModeFlags>
+GenericDeviceTy::getExecutionModeForKernel(StringRef Name,
+                                           DeviceImageTy &Image) {
+  auto KernelEnvOrError = getKernelEnvironmentForKernel(Name, Image);
+  if (!KernelEnvOrError) {
+    (void)KernelEnvOrError.takeError();
     return OMP_TGT_EXEC_MODE_SPMD;
   }
 
+  auto &KernelEnv = *KernelEnvOrError;
+  auto ExecMode = KernelEnv.Configuration.ExecMode;
+
   // Check that the retrieved execution mode is valid.
-  if (!GenericKernelTy::isValidExecutionMode(ExecModeGlobal.getValue()))
-    return Plugin::error("Invalid execution mode %d for '%s'",
-                         ExecModeGlobal.getValue(), Name.data());
+  if (!GenericKernelTy::isValidExecutionMode(ExecMode))
+    return Plugin::error("Invalid execution mode %d for '%s'", ExecMode,
+                         Name.data());
 
-  return ExecModeGlobal.getValue();
+  return ExecMode;
 }
 
 Error PinnedAllocationMapTy::insertEntry(void *HstPtr, void *DevAccessiblePtr,
diff --git a/openmp/libomptarget/plugins-nextgen/cuda/src/rtl.cpp b/openmp/libomptarget/plugins-nextgen/cuda/src/rtl.cpp
--- a/openmp/libomptarget/plugins-nextgen/cuda/src/rtl.cpp
+++ b/openmp/libomptarget/plugins-nextgen/cuda/src/rtl.cpp
@@ -17,7 +17,7 @@
 #include <unordered_map>
 
 #include "Debug.h"
-#include "DeviceEnvironment.h"
+#include "Environment.h"
 #include "GlobalHandler.h"
 #include "PluginInterface.h"
 
diff --git a/openmp/libomptarget/plugins-nextgen/generic-elf-64bit/src/rtl.cpp b/openmp/libomptarget/plugins-nextgen/generic-elf-64bit/src/rtl.cpp
--- a/openmp/libomptarget/plugins-nextgen/generic-elf-64bit/src/rtl.cpp
+++ b/openmp/libomptarget/plugins-nextgen/generic-elf-64bit/src/rtl.cpp
@@ -17,7 +17,7 @@
 #include <unordered_map>
 
 #include "Debug.h"
-#include "DeviceEnvironment.h"
+#include "Environment.h"
 #include "GlobalHandler.h"
 #include "PluginInterface.h"
 #include "omptarget.h"
diff --git a/openmp/libomptarget/plugins/amdgpu/src/rtl.cpp b/openmp/libomptarget/plugins/amdgpu/src/rtl.cpp
--- a/openmp/libomptarget/plugins/amdgpu/src/rtl.cpp
+++ b/openmp/libomptarget/plugins/amdgpu/src/rtl.cpp
@@ -37,7 +37,7 @@
 #include "internal.h"
 #include "rt.h"
 
-#include "DeviceEnvironment.h"
+#include "Environment.h"
 #include "get_elf_mach_gfx_name.h"
 #include "omptargetplugin.h"
 #include "print_tracing.h"
diff --git a/openmp/libomptarget/plugins/cuda/src/rtl.cpp b/openmp/libomptarget/plugins/cuda/src/rtl.cpp
--- a/openmp/libomptarget/plugins/cuda/src/rtl.cpp
+++ b/openmp/libomptarget/plugins/cuda/src/rtl.cpp
@@ -23,7 +23,7 @@
 #include <vector>
 
 #include "Debug.h"
-#include "DeviceEnvironment.h"
+#include "Environment.h"
 #include "omptarget.h"
 #include "omptargetplugin.h"