[openmp] Fix 51647, corrupt bitcode on amdgpu

Patch by @dpalermo

The corrupt bitcode reported in https://bugs.llvm.org/show_bug.cgi?id=51647 seems to be a result of a later pass changing the workfn variable to addrspace(5) (thread private, on the stack). That seems reasonable for an alloca without an address space so it's an open question why that can crash the bitcode reader.

This change puts it in the thread private address space to begin with which means whatever misfired further down the pipeline does not break it. That matches the codegen from clang where stack variables are always annotated (5) and then addrspace cast prior to following use.

This therefore patches around whatever unsuccessfully moved the alloca variable to addrspace(5). That solves the problem of openmp opt producing code that crashes the bitcode reader. It should be possible to create a minimal repro for the underlying bug based on some handwritten IR that uses an alloca in a generic address space.

Reviewed By: ronlieb, jdoerfert, dpalermo-phab

Differential Revision: https://reviews.llvm.org/D109500

diff --git a/llvm/lib/Transforms/IPO/OpenMPOpt.cpp b/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
index 456c940..00e3f1b 100644 (file)
--- a/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
+++ b/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
@@ -2041,7 +2041,8 @@ bool OpenMPOpt::rewriteDeviceCodeStateMachine() {
         UndefValue::get(Int8Ty), F->getName() + ".ID");
 
     for (Use *U : ToBeReplacedStateMachineUses)
-      U->set(ConstantExpr::getBitCast(ID, U->get()->getType()));
+      U->set(ConstantExpr::getPointerBitCastOrAddrSpaceCast(
+          ID, U->get()->getType()));
 
     ++NumOpenMPParallelRegionsReplacedInGPUStateMachine;
 
@@ -3455,10 +3456,14 @@ struct AAKernelInfoFunction : AAKernelInfo {
     IsWorker->setDebugLoc(DLoc);
     BranchInst::Create(StateMachineBeginBB, UserCodeEntryBB, IsWorker, InitBB);
 
+    Module &M = *Kernel->getParent();
+
     // Create local storage for the work function pointer.
+    const DataLayout &DL = M.getDataLayout();
     Type *VoidPtrTy = Type::getInt8PtrTy(Ctx);
-    AllocaInst *WorkFnAI = new AllocaInst(VoidPtrTy, 0, "worker.work_fn.addr",
-                                          &Kernel->getEntryBlock().front());
+    Instruction *WorkFnAI =
+        new AllocaInst(VoidPtrTy, DL.getAllocaAddrSpace(), nullptr,
+                       "worker.work_fn.addr", &Kernel->getEntryBlock().front());
     WorkFnAI->setDebugLoc(DLoc);
 
     auto &OMPInfoCache = static_cast<OMPInformationCache &>(A.getInfoCache());
@@ -3471,13 +3476,23 @@ struct AAKernelInfoFunction : AAKernelInfo {
     Value *Ident = KernelInitCB->getArgOperand(0);
     Value *GTid = KernelInitCB;
 
-    Module &M = *Kernel->getParent();
     FunctionCallee BarrierFn =
         OMPInfoCache.OMPBuilder.getOrCreateRuntimeFunction(
             M, OMPRTL___kmpc_barrier_simple_spmd);
     CallInst::Create(BarrierFn, {Ident, GTid}, "", StateMachineBeginBB)
         ->setDebugLoc(DLoc);
 
+    if (WorkFnAI->getType()->getPointerAddressSpace() !=
+        (unsigned int)AddressSpace::Generic) {
+      WorkFnAI = new AddrSpaceCastInst(
+          WorkFnAI,
+          PointerType::getWithSamePointeeType(
+              cast<PointerType>(WorkFnAI->getType()),
+              (unsigned int)AddressSpace::Generic),
+          WorkFnAI->getName() + ".generic", StateMachineBeginBB);
+      WorkFnAI->setDebugLoc(DLoc);
+    }
+
     FunctionCallee KernelParallelFn =
         OMPInfoCache.OMPBuilder.getOrCreateRuntimeFunction(
             M, OMPRTL___kmpc_kernel_parallel);

diff --git a/llvm/test/Transforms/OpenMP/custom_state_machines.ll b/llvm/test/Transforms/OpenMP/custom_state_machines.ll
index 8f57cac..e95a5f6 100644 (file)
--- a/llvm/test/Transforms/OpenMP/custom_state_machines.ll
+++ b/llvm/test/Transforms/OpenMP/custom_state_machines.ll
@@ -1,6 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function-signature --check-attributes --include-generated-funcs
-; RUN: opt -S -passes=openmp-opt < %s | FileCheck %s
-; RUN: opt -openmp-opt-disable-state-machine-rewrite -S -passes=openmp-opt < %s | FileCheck %s --check-prefix=CHECK-DISABLED
+; RUN: opt --mtriple=amdgcn-amd-amdhsa --data-layout=A5 -S -passes=openmp-opt < %s | FileCheck %s --check-prefixes=AMDGPU
+; RUN: opt --mtriple=nvptx64--         -S -passes=openmp-opt < %s | FileCheck %s --check-prefixes=NVPTX
+; RUN: opt --mtriple=amdgcn-amd-amdhsa --data-layout=A5 -openmp-opt-disable-state-machine-rewrite -S -passes=openmp-opt < %s | FileCheck %s --check-prefixes=AMDGPU-DISABLED
+; RUN: opt --mtriple=nvptx64--         -openmp-opt-disable-state-machine-rewrite -S -passes=openmp-opt < %s | FileCheck %s --check-prefixes=NVPTX-DISABLED
 
 ;; void p0(void);
 ;; void p1(void);
@@ -117,8 +119,6 @@
 ;;   { weak_callee_empty(); }
 ;; }
 
-target triple = "nvptx64"
-
 %struct.ident_t = type { i32, i32, i32, i32, i8* }
 
 @0 = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00", align 1
@@ -841,1696 +841,3397 @@ attributes #9 = { convergent nounwind readonly willreturn }
 !16 = !{i32 1, !"wchar_size", i32 4}
 !17 = !{i32 7, !"openmp", i32 50}
 !18 = !{i32 7, !"openmp-device", i32 50}
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_no_state_machine_needed_l14
-; CHECK-SAME: () #[[ATTR0:[0-9]+]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i1 false, i1 false, i1 true)
-; 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(%struct.ident_t* @[[GLOB1]]) #[[ATTR3:[0-9]+]]
-; CHECK-NEXT:    call void @__omp_outlined__(i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
-; CHECK-NEXT:    ret void
-; CHECK:       worker.exit:
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__
-; CHECK-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    call void @no_parallel_region_in_here.internalized() #[[ATTR7:[0-9]+]]
-; CHECK-NEXT:    call void @unknown_no_openmp() #[[ATTR8:[0-9]+]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline nounwind
-; CHECK-LABEL: define {{[^@]+}}@no_parallel_region_in_here.internalized
-; CHECK-SAME: () #[[ATTR1:[0-9]+]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef @[[GLOB2:[0-9]+]]) #[[ATTR3]]
-; CHECK-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_single(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]]) #[[ATTR3]]
-; CHECK-NEXT:    [[TMP2:%.*]] = icmp ne i32 [[TMP1]], 0
-; CHECK-NEXT:    br i1 [[TMP2]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
-; CHECK:       omp_if.then:
-; CHECK-NEXT:    store i32 0, i32* @G, align 4
-; CHECK-NEXT:    call void @__kmpc_end_single(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]]) #[[ATTR3]]
-; CHECK-NEXT:    br label [[OMP_IF_END]]
-; CHECK:       omp_if.end:
-; CHECK-NEXT:    call void @__kmpc_barrier(%struct.ident_t* noundef @[[GLOB3:[0-9]+]], i32 [[TMP0]]) #[[ATTR3]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline nounwind
-; CHECK-LABEL: define {{[^@]+}}@no_parallel_region_in_here
-; CHECK-SAME: () #[[ATTR1]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
-; CHECK-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
-; CHECK-NEXT:    [[TMP2:%.*]] = icmp ne i32 [[TMP1]], 0
-; CHECK-NEXT:    br i1 [[TMP2]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
-; CHECK:       omp_if.then:
-; CHECK-NEXT:    store i32 0, i32* @G, align 4
-; CHECK-NEXT:    call void @__kmpc_end_single(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
-; CHECK-NEXT:    br label [[OMP_IF_END]]
-; CHECK:       omp_if.end:
-; CHECK-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB3]], i32 [[TMP0]])
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_l22
-; CHECK-SAME: () #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
-; CHECK-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
-; CHECK-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
-; CHECK:       worker_state_machine.begin:
-; CHECK-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
-; CHECK-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
-; CHECK-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
-; CHECK-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
-; CHECK-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
-; CHECK:       worker_state_machine.finished:
-; CHECK-NEXT:    ret void
-; CHECK:       worker_state_machine.is_active.check:
-; CHECK-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
-; CHECK:       worker_state_machine.parallel_region.check:
-; CHECK-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__2_wrapper.ID to void (i16, i32)*)
-; CHECK-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK1:%.*]]
-; CHECK:       worker_state_machine.parallel_region.execute:
-; CHECK-NEXT:    call void @__omp_outlined__2_wrapper(i16 0, i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
-; CHECK:       worker_state_machine.parallel_region.check1:
-; CHECK-NEXT:    br i1 true, label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE2:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK3:%.*]]
-; CHECK:       worker_state_machine.parallel_region.execute2:
-; CHECK-NEXT:    call void @__omp_outlined__3_wrapper(i16 0, i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
-; CHECK:       worker_state_machine.parallel_region.check3:
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
-; CHECK:       worker_state_machine.parallel_region.end:
-; CHECK-NEXT:    call void @__kmpc_kernel_end_parallel()
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
-; CHECK:       worker_state_machine.done.barrier:
-; CHECK-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
-; CHECK:       thread.user_code.check:
-; 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(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
-; CHECK-NEXT:    call void @__omp_outlined__1(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
-; CHECK-NEXT:    ret void
-; CHECK:       worker.exit:
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__1
-; CHECK-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x i8*], align 8
-; CHECK-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
-; CHECK-NEXT:    call void @unknown_no_openmp() #[[ATTR8]]
-; CHECK-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
-; CHECK-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__2 to i8*), i8* noundef @__omp_outlined__2_wrapper.ID, i8** noundef [[TMP1]], i64 noundef 0)
-; CHECK-NEXT:    call void @no_parallel_region_in_here.internalized() #[[ATTR7]]
-; CHECK-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS1]] to i8**
-; CHECK-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__3 to i8*), i8* noundef @__omp_outlined__3_wrapper.ID, i8** noundef [[TMP2]], i64 noundef 0)
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__2
-; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    call void @p0() #[[ATTR9:[0-9]+]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__2_wrapper
-; CHECK-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-NEXT:    call void @__omp_outlined__2(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__3
-; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    call void @p1() #[[ATTR9]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__3_wrapper
-; CHECK-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-NEXT:    call void @__omp_outlined__3(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_interprocedural_l39
-; CHECK-SAME: () #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
-; CHECK-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
-; CHECK-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
-; CHECK:       worker_state_machine.begin:
-; CHECK-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
-; CHECK-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
-; CHECK-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
-; CHECK-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
-; CHECK-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
-; CHECK:       worker_state_machine.finished:
-; CHECK-NEXT:    ret void
-; CHECK:       worker_state_machine.is_active.check:
-; CHECK-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
-; CHECK:       worker_state_machine.parallel_region.check:
-; CHECK-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], @__omp_outlined__17_wrapper
-; CHECK-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK1:%.*]]
-; CHECK:       worker_state_machine.parallel_region.execute:
-; CHECK-NEXT:    call void @__omp_outlined__17_wrapper(i16 0, i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
-; CHECK:       worker_state_machine.parallel_region.check1:
-; CHECK-NEXT:    [[WORKER_CHECK_PARALLEL_REGION4:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__5_wrapper.ID to void (i16, i32)*)
-; CHECK-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION4]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE2:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK3:%.*]]
-; CHECK:       worker_state_machine.parallel_region.execute2:
-; CHECK-NEXT:    call void @__omp_outlined__5_wrapper(i16 0, i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
-; CHECK:       worker_state_machine.parallel_region.check3:
-; CHECK-NEXT:    br i1 true, label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE5:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK6:%.*]]
-; CHECK:       worker_state_machine.parallel_region.execute5:
-; CHECK-NEXT:    call void @__omp_outlined__18_wrapper(i16 0, i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
-; CHECK:       worker_state_machine.parallel_region.check6:
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
-; CHECK:       worker_state_machine.parallel_region.end:
-; CHECK-NEXT:    call void @__kmpc_kernel_end_parallel()
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
-; CHECK:       worker_state_machine.done.barrier:
-; CHECK-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
-; CHECK:       thread.user_code.check:
-; 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(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
-; CHECK-NEXT:    call void @__omp_outlined__4(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
-; CHECK-NEXT:    ret void
-; CHECK:       worker.exit:
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__4
-; CHECK-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
-; CHECK-NEXT:    call void @unknown_no_openmp() #[[ATTR8]]
-; CHECK-NEXT:    call void @simple_state_machine_interprocedural_before.internalized() #[[ATTR7]]
-; CHECK-NEXT:    call void @no_parallel_region_in_here.internalized() #[[ATTR7]]
-; CHECK-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
-; CHECK-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__5 to i8*), i8* noundef @__omp_outlined__5_wrapper.ID, i8** noundef [[TMP1]], i64 noundef 0)
-; CHECK-NEXT:    call void @simple_state_machine_interprocedural_after.internalized() #[[ATTR7]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline nounwind
-; CHECK-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_before.internalized
-; CHECK-SAME: () #[[ATTR1]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef @[[GLOB2]]) #[[ATTR3]]
-; CHECK-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__17 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__17_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline nounwind
-; CHECK-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_before
-; CHECK-SAME: () #[[ATTR1]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
-; CHECK-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__17 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__17_wrapper to i8*), i8** [[TMP1]], i64 0)
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__5
-; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    call void @p1() #[[ATTR9]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__5_wrapper
-; CHECK-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-NEXT:    call void @__omp_outlined__5(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline nounwind
-; CHECK-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_after.internalized
-; CHECK-SAME: () #[[ATTR1]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef @[[GLOB2]]) #[[ATTR3]]
-; CHECK-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__18 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__18_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline nounwind
-; CHECK-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_after
-; CHECK-SAME: () #[[ATTR1]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
-; CHECK-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__18 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__18_wrapper to i8*), i8** [[TMP1]], i64 0)
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_with_fallback_l55
-; CHECK-SAME: () #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
-; CHECK-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
-; CHECK-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
-; CHECK:       worker_state_machine.begin:
-; CHECK-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
-; CHECK-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
-; CHECK-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
-; CHECK-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
-; CHECK-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
-; CHECK:       worker_state_machine.finished:
-; CHECK-NEXT:    ret void
-; CHECK:       worker_state_machine.is_active.check:
-; CHECK-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
-; CHECK:       worker_state_machine.parallel_region.check:
-; CHECK-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__7_wrapper.ID to void (i16, i32)*)
-; CHECK-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK1:%.*]]
-; CHECK:       worker_state_machine.parallel_region.execute:
-; CHECK-NEXT:    call void @__omp_outlined__7_wrapper(i16 0, i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
-; CHECK:       worker_state_machine.parallel_region.check1:
-; CHECK-NEXT:    [[WORKER_CHECK_PARALLEL_REGION4:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__8_wrapper.ID to void (i16, i32)*)
-; CHECK-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION4]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE2:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]]
-; CHECK:       worker_state_machine.parallel_region.execute2:
-; CHECK-NEXT:    call void @__omp_outlined__8_wrapper(i16 0, i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
-; CHECK:       worker_state_machine.parallel_region.fallback.execute:
-; CHECK-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
-; CHECK:       worker_state_machine.parallel_region.end:
-; CHECK-NEXT:    call void @__kmpc_kernel_end_parallel()
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
-; CHECK:       worker_state_machine.done.barrier:
-; CHECK-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
-; CHECK:       thread.user_code.check:
-; 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(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
-; CHECK-NEXT:    call void @__omp_outlined__6(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
-; CHECK-NEXT:    ret void
-; CHECK:       worker.exit:
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__6
-; CHECK-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x i8*], align 8
-; CHECK-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
-; CHECK-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
-; CHECK-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__7 to i8*), i8* noundef @__omp_outlined__7_wrapper.ID, i8** noundef [[TMP1]], i64 noundef 0)
-; CHECK-NEXT:    [[CALL:%.*]] = call i32 @unknown() #[[ATTR9]]
-; CHECK-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS1]] to i8**
-; CHECK-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__8 to i8*), i8* noundef @__omp_outlined__8_wrapper.ID, i8** noundef [[TMP2]], i64 noundef 0)
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__7
-; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    call void @p0() #[[ATTR9]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__7_wrapper
-; CHECK-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-NEXT:    call void @__omp_outlined__7(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__8
-; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    call void @p1() #[[ATTR9]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__8_wrapper
-; CHECK-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-NEXT:    call void @__omp_outlined__8(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_no_openmp_attr_l66
-; CHECK-SAME: () #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
-; CHECK-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
-; CHECK-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
-; CHECK:       worker_state_machine.begin:
-; CHECK-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
-; CHECK-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
-; CHECK-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
-; CHECK-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
-; CHECK-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
-; CHECK:       worker_state_machine.finished:
-; CHECK-NEXT:    ret void
-; CHECK:       worker_state_machine.is_active.check:
-; CHECK-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
-; CHECK:       worker_state_machine.parallel_region.check:
-; CHECK-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__10_wrapper.ID to void (i16, i32)*)
-; CHECK-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK1:%.*]]
-; CHECK:       worker_state_machine.parallel_region.execute:
-; CHECK-NEXT:    call void @__omp_outlined__10_wrapper(i16 0, i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
-; CHECK:       worker_state_machine.parallel_region.check1:
-; CHECK-NEXT:    br i1 true, label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE2:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK3:%.*]]
-; CHECK:       worker_state_machine.parallel_region.execute2:
-; CHECK-NEXT:    call void @__omp_outlined__11_wrapper(i16 0, i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
-; CHECK:       worker_state_machine.parallel_region.check3:
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
-; CHECK:       worker_state_machine.parallel_region.end:
-; CHECK-NEXT:    call void @__kmpc_kernel_end_parallel()
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
-; CHECK:       worker_state_machine.done.barrier:
-; CHECK-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
-; CHECK:       thread.user_code.check:
-; 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(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
-; CHECK-NEXT:    call void @__omp_outlined__9(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
-; CHECK-NEXT:    ret void
-; CHECK:       worker.exit:
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__9
-; CHECK-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x i8*], align 8
-; CHECK-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
-; CHECK-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
-; CHECK-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__10 to i8*), i8* noundef @__omp_outlined__10_wrapper.ID, i8** noundef [[TMP1]], i64 noundef 0)
-; CHECK-NEXT:    call void @unknown_no_openmp() #[[ATTR8]]
-; CHECK-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS1]] to i8**
-; CHECK-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__11 to i8*), i8* noundef @__omp_outlined__11_wrapper.ID, i8** noundef [[TMP2]], i64 noundef 0)
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__10
-; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    call void @p0() #[[ATTR9]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__10_wrapper
-; CHECK-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-NEXT:    call void @__omp_outlined__10(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__11
-; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    call void @p1() #[[ATTR9]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__11_wrapper
-; CHECK-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-NEXT:    call void @__omp_outlined__11(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_pure_l77
-; CHECK-SAME: () #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
-; CHECK-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
-; CHECK-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
-; CHECK:       worker_state_machine.begin:
-; CHECK-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
-; CHECK-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
-; CHECK-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
-; CHECK-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
-; CHECK-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
-; CHECK:       worker_state_machine.finished:
-; CHECK-NEXT:    ret void
-; CHECK:       worker_state_machine.is_active.check:
-; CHECK-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
-; CHECK:       worker_state_machine.parallel_region.check:
-; CHECK-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__13_wrapper.ID to void (i16, i32)*)
-; CHECK-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK1:%.*]]
-; CHECK:       worker_state_machine.parallel_region.execute:
-; CHECK-NEXT:    call void @__omp_outlined__13_wrapper(i16 0, i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
-; CHECK:       worker_state_machine.parallel_region.check1:
-; CHECK-NEXT:    br i1 true, label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE2:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK3:%.*]]
-; CHECK:       worker_state_machine.parallel_region.execute2:
-; CHECK-NEXT:    call void @__omp_outlined__14_wrapper(i16 0, i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
-; CHECK:       worker_state_machine.parallel_region.check3:
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
-; CHECK:       worker_state_machine.parallel_region.end:
-; CHECK-NEXT:    call void @__kmpc_kernel_end_parallel()
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
-; CHECK:       worker_state_machine.done.barrier:
-; CHECK-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
-; CHECK:       thread.user_code.check:
-; 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(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
-; CHECK-NEXT:    call void @__omp_outlined__12(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
-; CHECK-NEXT:    ret void
-; CHECK:       worker.exit:
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__12
-; CHECK-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x i8*], align 8
-; CHECK-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
-; CHECK-NEXT:    call void @unknown_no_openmp() #[[ATTR8]]
-; CHECK-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
-; CHECK-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__13 to i8*), i8* noundef @__omp_outlined__13_wrapper.ID, i8** noundef [[TMP1]], i64 noundef 0)
-; CHECK-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS1]] to i8**
-; CHECK-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__14 to i8*), i8* noundef @__omp_outlined__14_wrapper.ID, i8** noundef [[TMP2]], i64 noundef 0)
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__13
-; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    call void @p0() #[[ATTR9]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__13_wrapper
-; CHECK-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-NEXT:    call void @__omp_outlined__13(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__14
-; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    call void @p1() #[[ATTR9]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__14_wrapper
-; CHECK-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-NEXT:    call void @__omp_outlined__14(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_interprocedural_nested_recursive_l92
-; CHECK-SAME: () #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
-; CHECK-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
-; CHECK-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
-; CHECK:       worker_state_machine.begin:
-; CHECK-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
-; CHECK-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
-; CHECK-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
-; CHECK-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
-; CHECK-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
-; CHECK:       worker_state_machine.finished:
-; CHECK-NEXT:    ret void
-; CHECK:       worker_state_machine.is_active.check:
-; CHECK-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
-; CHECK:       worker_state_machine.parallel_region.check:
-; CHECK-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], @__omp_outlined__19_wrapper
-; CHECK-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]]
-; CHECK:       worker_state_machine.parallel_region.execute:
-; CHECK-NEXT:    call void @__omp_outlined__19_wrapper(i16 0, i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
-; CHECK:       worker_state_machine.parallel_region.fallback.execute:
-; CHECK-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
-; CHECK:       worker_state_machine.parallel_region.end:
-; CHECK-NEXT:    call void @__kmpc_kernel_end_parallel()
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
-; CHECK:       worker_state_machine.done.barrier:
-; CHECK-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
-; CHECK:       thread.user_code.check:
-; 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(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
-; CHECK-NEXT:    call void @__omp_outlined__15(i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
-; CHECK-NEXT:    ret void
-; CHECK:       worker.exit:
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__15
-; CHECK-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[CALL:%.*]] = call i32 bitcast (i32 (...)* @omp_get_thread_num to i32 ()*)() #[[ATTR9]]
-; CHECK-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after.internalized(i32 [[CALL]]) #[[ATTR7]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline nounwind
-; CHECK-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_nested_recursive_after.internalized
-; CHECK-SAME: (i32 [[A:%.*]]) #[[ATTR1]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
-; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 [[A]], 0
-; CHECK-NEXT:    br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
-; CHECK:       if.then:
-; CHECK-NEXT:    br label [[RETURN:%.*]]
-; CHECK:       if.end:
-; CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[A]], 1
-; CHECK-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after.internalized(i32 [[SUB]]) #[[ATTR7]]
-; CHECK-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after_after.internalized() #[[ATTR7]]
-; CHECK-NEXT:    br label [[RETURN]]
-; CHECK:       return:
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline nounwind
-; CHECK-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_nested_recursive_after
-; CHECK-SAME: (i32 [[A:%.*]]) #[[ATTR1]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
-; CHECK-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4
-; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 [[TMP0]], 0
-; CHECK-NEXT:    br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
-; CHECK:       if.then:
-; CHECK-NEXT:    br label [[RETURN:%.*]]
-; CHECK:       if.end:
-; CHECK-NEXT:    [[TMP1:%.*]] = load i32, i32* [[A_ADDR]], align 4
-; CHECK-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP1]], 1
-; CHECK-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after(i32 [[SUB]]) #[[ATTR9]]
-; CHECK-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after_after() #[[ATTR9]]
-; CHECK-NEXT:    br label [[RETURN]]
-; CHECK:       return:
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_no_state_machine_weak_callee_l112
-; CHECK-SAME: () #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
-; CHECK-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
-; CHECK-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
-; CHECK:       worker_state_machine.begin:
-; CHECK-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
-; CHECK-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
-; CHECK-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
-; CHECK-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
-; CHECK-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
-; CHECK:       worker_state_machine.finished:
-; CHECK-NEXT:    ret void
-; CHECK:       worker_state_machine.is_active.check:
-; CHECK-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
-; CHECK:       worker_state_machine.parallel_region.fallback.execute:
-; CHECK-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
-; CHECK:       worker_state_machine.parallel_region.end:
-; CHECK-NEXT:    call void @__kmpc_kernel_end_parallel()
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
-; CHECK:       worker_state_machine.done.barrier:
-; CHECK-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
-; CHECK:       thread.user_code.check:
-; 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(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
-; CHECK-NEXT:    call void @__omp_outlined__16(i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
-; CHECK-NEXT:    ret void
-; CHECK:       worker.exit:
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__16
-; CHECK-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    call void @weak_callee_empty() #[[ATTR7]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline nounwind
-; CHECK-LABEL: define {{[^@]+}}@weak_callee_empty
-; CHECK-SAME: () #[[ATTR1]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__17
-; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    call void @p0() #[[ATTR9]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__17_wrapper
-; CHECK-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-NEXT:    call void @__omp_outlined__17(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__18
-; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    call void @p0() #[[ATTR9]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__18_wrapper
-; CHECK-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-NEXT:    call void @__omp_outlined__18(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline nounwind
-; CHECK-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_nested_recursive_after_after.internalized
-; CHECK-SAME: () #[[ATTR1]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef @[[GLOB2]]) #[[ATTR3]]
-; CHECK-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__19 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__19_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline nounwind
-; CHECK-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_nested_recursive_after_after
-; CHECK-SAME: () #[[ATTR1]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
-; CHECK-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__19 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__19_wrapper to i8*), i8** [[TMP1]], i64 0)
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__19
-; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    call void @p0() #[[ATTR9]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__19_wrapper
-; CHECK-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-NEXT:    call void @__omp_outlined__19(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_no_state_machine_needed_l14
-; CHECK-DISABLED-SAME: () #[[ATTR0:[0-9]+]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i1 false, i1 true, i1 true)
-; CHECK-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
-; CHECK-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
-; CHECK-DISABLED:       user_code.entry:
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3:[0-9]+]]
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__(i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
-; CHECK-DISABLED-NEXT:    ret void
-; CHECK-DISABLED:       worker.exit:
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    call void @no_parallel_region_in_here.internalized() #[[ATTR7:[0-9]+]]
-; CHECK-DISABLED-NEXT:    call void @unknown_no_openmp() #[[ATTR8:[0-9]+]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@no_parallel_region_in_here.internalized
-; CHECK-DISABLED-SAME: () #[[ATTR1:[0-9]+]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef @[[GLOB2:[0-9]+]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_single(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    [[TMP2:%.*]] = icmp ne i32 [[TMP1]], 0
-; CHECK-DISABLED-NEXT:    br i1 [[TMP2]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
-; CHECK-DISABLED:       omp_if.then:
-; CHECK-DISABLED-NEXT:    store i32 0, i32* @G, align 4
-; CHECK-DISABLED-NEXT:    call void @__kmpc_end_single(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    br label [[OMP_IF_END]]
-; CHECK-DISABLED:       omp_if.end:
-; CHECK-DISABLED-NEXT:    call void @__kmpc_barrier(%struct.ident_t* noundef @[[GLOB3:[0-9]+]], i32 [[TMP0]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@no_parallel_region_in_here
-; CHECK-DISABLED-SAME: () #[[ATTR1]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
-; CHECK-DISABLED-NEXT:    [[TMP2:%.*]] = icmp ne i32 [[TMP1]], 0
-; CHECK-DISABLED-NEXT:    br i1 [[TMP2]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
-; CHECK-DISABLED:       omp_if.then:
-; CHECK-DISABLED-NEXT:    store i32 0, i32* @G, align 4
-; CHECK-DISABLED-NEXT:    call void @__kmpc_end_single(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
-; CHECK-DISABLED-NEXT:    br label [[OMP_IF_END]]
-; CHECK-DISABLED:       omp_if.end:
-; CHECK-DISABLED-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB3]], i32 [[TMP0]])
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_l22
-; CHECK-DISABLED-SAME: () #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 true, i1 true)
-; CHECK-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
-; CHECK-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
-; CHECK-DISABLED:       user_code.entry:
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__1(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
-; CHECK-DISABLED-NEXT:    ret void
-; CHECK-DISABLED:       worker.exit:
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__1
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x i8*], align 8
-; CHECK-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
-; CHECK-DISABLED-NEXT:    call void @unknown_no_openmp() #[[ATTR8]]
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__2 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__2_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
-; CHECK-DISABLED-NEXT:    call void @no_parallel_region_in_here.internalized() #[[ATTR7]]
-; CHECK-DISABLED-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS1]] to i8**
-; CHECK-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__3 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__3_wrapper to i8*), i8** noundef [[TMP2]], i64 noundef 0)
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__2
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    call void @p0() #[[ATTR9:[0-9]+]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__2_wrapper
-; CHECK-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__2(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__3
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    call void @p1() #[[ATTR9]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__3_wrapper
-; CHECK-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__3(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_interprocedural_l39
-; CHECK-DISABLED-SAME: () #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 true, i1 true)
-; CHECK-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
-; CHECK-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
-; CHECK-DISABLED:       user_code.entry:
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__4(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
-; CHECK-DISABLED-NEXT:    ret void
-; CHECK-DISABLED:       worker.exit:
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__4
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
-; CHECK-DISABLED-NEXT:    call void @unknown_no_openmp() #[[ATTR8]]
-; CHECK-DISABLED-NEXT:    call void @simple_state_machine_interprocedural_before.internalized() #[[ATTR7]]
-; CHECK-DISABLED-NEXT:    call void @no_parallel_region_in_here.internalized() #[[ATTR7]]
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__5 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__5_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
-; CHECK-DISABLED-NEXT:    call void @simple_state_machine_interprocedural_after.internalized() #[[ATTR7]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_before.internalized
-; CHECK-DISABLED-SAME: () #[[ATTR1]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef @[[GLOB2]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__17 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__17_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_before
-; CHECK-DISABLED-SAME: () #[[ATTR1]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__17 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__17_wrapper to i8*), i8** [[TMP1]], i64 0)
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__5
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    call void @p1() #[[ATTR9]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__5_wrapper
-; CHECK-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__5(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_after.internalized
-; CHECK-DISABLED-SAME: () #[[ATTR1]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef @[[GLOB2]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__18 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__18_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_after
-; CHECK-DISABLED-SAME: () #[[ATTR1]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__18 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__18_wrapper to i8*), i8** [[TMP1]], i64 0)
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_with_fallback_l55
-; CHECK-DISABLED-SAME: () #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 true, i1 true)
-; CHECK-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
-; CHECK-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
-; CHECK-DISABLED:       user_code.entry:
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__6(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
-; CHECK-DISABLED-NEXT:    ret void
-; CHECK-DISABLED:       worker.exit:
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__6
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x i8*], align 8
-; CHECK-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__7 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__7_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
-; CHECK-DISABLED-NEXT:    [[CALL:%.*]] = call i32 @unknown() #[[ATTR9]]
-; CHECK-DISABLED-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS1]] to i8**
-; CHECK-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__8 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__8_wrapper to i8*), i8** noundef [[TMP2]], i64 noundef 0)
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__7
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    call void @p0() #[[ATTR9]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__7_wrapper
-; CHECK-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__7(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__8
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    call void @p1() #[[ATTR9]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__8_wrapper
-; CHECK-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__8(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_no_openmp_attr_l66
-; CHECK-DISABLED-SAME: () #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 true, i1 true)
-; CHECK-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
-; CHECK-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
-; CHECK-DISABLED:       user_code.entry:
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__9(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
-; CHECK-DISABLED-NEXT:    ret void
-; CHECK-DISABLED:       worker.exit:
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__9
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x i8*], align 8
-; CHECK-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__10 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__10_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
-; CHECK-DISABLED-NEXT:    call void @unknown_no_openmp() #[[ATTR8]]
-; CHECK-DISABLED-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS1]] to i8**
-; CHECK-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__11 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__11_wrapper to i8*), i8** noundef [[TMP2]], i64 noundef 0)
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__10
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    call void @p0() #[[ATTR9]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__10_wrapper
-; CHECK-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__10(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__11
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    call void @p1() #[[ATTR9]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__11_wrapper
-; CHECK-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__11(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_pure_l77
-; CHECK-DISABLED-SAME: () #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 true, i1 true)
-; CHECK-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
-; CHECK-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
-; CHECK-DISABLED:       user_code.entry:
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__12(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
-; CHECK-DISABLED-NEXT:    ret void
-; CHECK-DISABLED:       worker.exit:
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__12
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x i8*], align 8
-; CHECK-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
-; CHECK-DISABLED-NEXT:    call void @unknown_no_openmp() #[[ATTR8]]
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__13 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__13_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
-; CHECK-DISABLED-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS1]] to i8**
-; CHECK-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__14 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__14_wrapper to i8*), i8** noundef [[TMP2]], i64 noundef 0)
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__13
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    call void @p0() #[[ATTR9]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__13_wrapper
-; CHECK-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__13(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__14
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    call void @p1() #[[ATTR9]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__14_wrapper
-; CHECK-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__14(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_interprocedural_nested_recursive_l92
-; CHECK-DISABLED-SAME: () #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 true, i1 true)
-; CHECK-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
-; CHECK-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
-; CHECK-DISABLED:       user_code.entry:
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__15(i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
-; CHECK-DISABLED-NEXT:    ret void
-; CHECK-DISABLED:       worker.exit:
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__15
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[CALL:%.*]] = call i32 bitcast (i32 (...)* @omp_get_thread_num to i32 ()*)() #[[ATTR9]]
-; CHECK-DISABLED-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after.internalized(i32 [[CALL]]) #[[ATTR7]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_nested_recursive_after.internalized
-; CHECK-DISABLED-SAME: (i32 [[A:%.*]]) #[[ATTR1]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    [[CMP:%.*]] = icmp eq i32 [[A]], 0
-; CHECK-DISABLED-NEXT:    br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
-; CHECK-DISABLED:       if.then:
-; CHECK-DISABLED-NEXT:    br label [[RETURN:%.*]]
-; CHECK-DISABLED:       if.end:
-; CHECK-DISABLED-NEXT:    [[SUB:%.*]] = sub nsw i32 [[A]], 1
-; CHECK-DISABLED-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after.internalized(i32 [[SUB]]) #[[ATTR7]]
-; CHECK-DISABLED-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after_after.internalized() #[[ATTR7]]
-; CHECK-DISABLED-NEXT:    br label [[RETURN]]
-; CHECK-DISABLED:       return:
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_nested_recursive_after
-; CHECK-DISABLED-SAME: (i32 [[A:%.*]]) #[[ATTR1]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    [[CMP:%.*]] = icmp eq i32 [[TMP0]], 0
-; CHECK-DISABLED-NEXT:    br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
-; CHECK-DISABLED:       if.then:
-; CHECK-DISABLED-NEXT:    br label [[RETURN:%.*]]
-; CHECK-DISABLED:       if.end:
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = load i32, i32* [[A_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP1]], 1
-; CHECK-DISABLED-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after(i32 [[SUB]]) #[[ATTR9]]
-; CHECK-DISABLED-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after_after() #[[ATTR9]]
-; CHECK-DISABLED-NEXT:    br label [[RETURN]]
-; CHECK-DISABLED:       return:
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_no_state_machine_weak_callee_l112
-; CHECK-DISABLED-SAME: () #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 true, i1 true)
-; CHECK-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
-; CHECK-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
-; CHECK-DISABLED:       user_code.entry:
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__16(i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
-; CHECK-DISABLED-NEXT:    ret void
-; CHECK-DISABLED:       worker.exit:
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__16
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    call void @weak_callee_empty() #[[ATTR7]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@weak_callee_empty
-; CHECK-DISABLED-SAME: () #[[ATTR1]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__17
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    call void @p0() #[[ATTR9]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__17_wrapper
-; CHECK-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__17(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__18
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    call void @p0() #[[ATTR9]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__18_wrapper
-; CHECK-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__18(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_nested_recursive_after_after.internalized
-; CHECK-DISABLED-SAME: () #[[ATTR1]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef @[[GLOB2]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__19 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__19_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_nested_recursive_after_after
-; CHECK-DISABLED-SAME: () #[[ATTR1]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__19 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__19_wrapper to i8*), i8** [[TMP1]], i64 0)
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__19
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    call void @p0() #[[ATTR9]]
-; CHECK-DISABLED-NEXT:    ret void
-;
-;
-; CHECK-DISABLED: Function Attrs: convergent noinline norecurse nounwind
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__19_wrapper
-; CHECK-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__19(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    ret void
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_no_state_machine_needed_l14
+; AMDGPU-SAME: () #[[ATTR0:[0-9]+]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i1 false, i1 false, i1 true)
+; AMDGPU-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU:       user_code.entry:
+; AMDGPU-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3:[0-9]+]]
+; AMDGPU-NEXT:    call void @__omp_outlined__(i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; AMDGPU-NEXT:    ret void
+; AMDGPU:       worker.exit:
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__
+; AMDGPU-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    call void @no_parallel_region_in_here.internalized() #[[ATTR8:[0-9]+]]
+; AMDGPU-NEXT:    call void @unknown_no_openmp() #[[ATTR9:[0-9]+]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@no_parallel_region_in_here.internalized
+; AMDGPU-SAME: () #[[ATTR1:[0-9]+]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef @[[GLOB2:[0-9]+]]) #[[ATTR3]]
+; AMDGPU-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_single(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]]) #[[ATTR3]]
+; AMDGPU-NEXT:    [[TMP2:%.*]] = icmp ne i32 [[TMP1]], 0
+; AMDGPU-NEXT:    br i1 [[TMP2]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
+; AMDGPU:       omp_if.then:
+; AMDGPU-NEXT:    store i32 0, i32* @G, align 4
+; AMDGPU-NEXT:    call void @__kmpc_end_single(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]]) #[[ATTR3]]
+; AMDGPU-NEXT:    br label [[OMP_IF_END]]
+; AMDGPU:       omp_if.end:
+; AMDGPU-NEXT:    call void @__kmpc_barrier(%struct.ident_t* noundef @[[GLOB3:[0-9]+]], i32 [[TMP0]]) #[[ATTR3]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@no_parallel_region_in_here
+; AMDGPU-SAME: () #[[ATTR1]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
+; AMDGPU-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
+; AMDGPU-NEXT:    [[TMP2:%.*]] = icmp ne i32 [[TMP1]], 0
+; AMDGPU-NEXT:    br i1 [[TMP2]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
+; AMDGPU:       omp_if.then:
+; AMDGPU-NEXT:    store i32 0, i32* @G, align 4
+; AMDGPU-NEXT:    call void @__kmpc_end_single(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[OMP_IF_END]]
+; AMDGPU:       omp_if.end:
+; AMDGPU-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB3]], i32 [[TMP0]])
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_l22
+; AMDGPU-SAME: () #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8, addrspace(5)
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; AMDGPU-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; AMDGPU-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; AMDGPU:       worker_state_machine.begin:
+; AMDGPU-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR_GENERIC:%.*]] = addrspacecast i8* addrspace(5)* [[WORKER_WORK_FN_ADDR]] to i8**
+; AMDGPU-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR_GENERIC]])
+; AMDGPU-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR_GENERIC]], align 8
+; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; AMDGPU-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; AMDGPU-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; AMDGPU:       worker_state_machine.finished:
+; AMDGPU-NEXT:    ret void
+; AMDGPU:       worker_state_machine.is_active.check:
+; AMDGPU-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.check:
+; AMDGPU-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__2_wrapper.ID to void (i16, i32)*)
+; AMDGPU-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK1:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.execute:
+; AMDGPU-NEXT:    call void @__omp_outlined__2_wrapper(i16 0, i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.check1:
+; AMDGPU-NEXT:    br i1 true, label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE2:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK3:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.execute2:
+; AMDGPU-NEXT:    call void @__omp_outlined__3_wrapper(i16 0, i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; AMDGPU:       worker_state_machine.parallel_region.check3:
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; AMDGPU:       worker_state_machine.parallel_region.end:
+; AMDGPU-NEXT:    call void @__kmpc_kernel_end_parallel()
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; AMDGPU:       worker_state_machine.done.barrier:
+; AMDGPU-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; AMDGPU:       thread.user_code.check:
+; AMDGPU-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU:       user_code.entry:
+; AMDGPU-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; AMDGPU-NEXT:    call void @__omp_outlined__1(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; AMDGPU-NEXT:    ret void
+; AMDGPU:       worker.exit:
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__1
+; AMDGPU-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; AMDGPU-NEXT:    call void @unknown_no_openmp() #[[ATTR9]]
+; AMDGPU-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; AMDGPU-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__2 to i8*), i8* noundef @__omp_outlined__2_wrapper.ID, i8** noundef [[TMP1]], i64 noundef 0)
+; AMDGPU-NEXT:    call void @no_parallel_region_in_here.internalized() #[[ATTR8]]
+; AMDGPU-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS1]] to i8**
+; AMDGPU-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__3 to i8*), i8* noundef @__omp_outlined__3_wrapper.ID, i8** noundef [[TMP2]], i64 noundef 0)
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__2
+; AMDGPU-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    call void @p0() #[[ATTR10:[0-9]+]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__2_wrapper
+; AMDGPU-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-NEXT:    call void @__omp_outlined__2(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__3
+; AMDGPU-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    call void @p1() #[[ATTR10]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__3_wrapper
+; AMDGPU-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-NEXT:    call void @__omp_outlined__3(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_interprocedural_l39
+; AMDGPU-SAME: () #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8, addrspace(5)
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; AMDGPU-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; AMDGPU-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; AMDGPU:       worker_state_machine.begin:
+; AMDGPU-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR_GENERIC:%.*]] = addrspacecast i8* addrspace(5)* [[WORKER_WORK_FN_ADDR]] to i8**
+; AMDGPU-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR_GENERIC]])
+; AMDGPU-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR_GENERIC]], align 8
+; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; AMDGPU-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; AMDGPU-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; AMDGPU:       worker_state_machine.finished:
+; AMDGPU-NEXT:    ret void
+; AMDGPU:       worker_state_machine.is_active.check:
+; AMDGPU-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.check:
+; AMDGPU-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], @__omp_outlined__17_wrapper
+; AMDGPU-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK1:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.execute:
+; AMDGPU-NEXT:    call void @__omp_outlined__17_wrapper(i16 0, i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.check1:
+; AMDGPU-NEXT:    [[WORKER_CHECK_PARALLEL_REGION4:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__5_wrapper.ID to void (i16, i32)*)
+; AMDGPU-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION4]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE2:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK3:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.execute2:
+; AMDGPU-NEXT:    call void @__omp_outlined__5_wrapper(i16 0, i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; AMDGPU:       worker_state_machine.parallel_region.check3:
+; AMDGPU-NEXT:    br i1 true, label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE5:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK6:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.execute5:
+; AMDGPU-NEXT:    call void @__omp_outlined__18_wrapper(i16 0, i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; AMDGPU:       worker_state_machine.parallel_region.check6:
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; AMDGPU:       worker_state_machine.parallel_region.end:
+; AMDGPU-NEXT:    call void @__kmpc_kernel_end_parallel()
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; AMDGPU:       worker_state_machine.done.barrier:
+; AMDGPU-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; AMDGPU:       thread.user_code.check:
+; AMDGPU-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU:       user_code.entry:
+; AMDGPU-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; AMDGPU-NEXT:    call void @__omp_outlined__4(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; AMDGPU-NEXT:    ret void
+; AMDGPU:       worker.exit:
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__4
+; AMDGPU-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; AMDGPU-NEXT:    call void @unknown_no_openmp() #[[ATTR9]]
+; AMDGPU-NEXT:    call void @simple_state_machine_interprocedural_before.internalized() #[[ATTR8]]
+; AMDGPU-NEXT:    call void @no_parallel_region_in_here.internalized() #[[ATTR8]]
+; AMDGPU-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; AMDGPU-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__5 to i8*), i8* noundef @__omp_outlined__5_wrapper.ID, i8** noundef [[TMP1]], i64 noundef 0)
+; AMDGPU-NEXT:    call void @simple_state_machine_interprocedural_after.internalized() #[[ATTR8]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_before.internalized
+; AMDGPU-SAME: () #[[ATTR1]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef @[[GLOB2]]) #[[ATTR3]]
+; AMDGPU-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__17 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__17_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_before
+; AMDGPU-SAME: () #[[ATTR1]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
+; AMDGPU-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__17 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__17_wrapper to i8*), i8** [[TMP1]], i64 0)
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__5
+; AMDGPU-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    call void @p1() #[[ATTR10]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__5_wrapper
+; AMDGPU-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-NEXT:    call void @__omp_outlined__5(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_after.internalized
+; AMDGPU-SAME: () #[[ATTR1]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef @[[GLOB2]]) #[[ATTR3]]
+; AMDGPU-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__18 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__18_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_after
+; AMDGPU-SAME: () #[[ATTR1]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
+; AMDGPU-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__18 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__18_wrapper to i8*), i8** [[TMP1]], i64 0)
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_with_fallback_l55
+; AMDGPU-SAME: () #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8, addrspace(5)
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; AMDGPU-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; AMDGPU-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; AMDGPU:       worker_state_machine.begin:
+; AMDGPU-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR_GENERIC:%.*]] = addrspacecast i8* addrspace(5)* [[WORKER_WORK_FN_ADDR]] to i8**
+; AMDGPU-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR_GENERIC]])
+; AMDGPU-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR_GENERIC]], align 8
+; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; AMDGPU-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; AMDGPU-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; AMDGPU:       worker_state_machine.finished:
+; AMDGPU-NEXT:    ret void
+; AMDGPU:       worker_state_machine.is_active.check:
+; AMDGPU-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.check:
+; AMDGPU-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__7_wrapper.ID to void (i16, i32)*)
+; AMDGPU-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK1:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.execute:
+; AMDGPU-NEXT:    call void @__omp_outlined__7_wrapper(i16 0, i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.check1:
+; AMDGPU-NEXT:    [[WORKER_CHECK_PARALLEL_REGION4:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__8_wrapper.ID to void (i16, i32)*)
+; AMDGPU-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION4]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE2:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.execute2:
+; AMDGPU-NEXT:    call void @__omp_outlined__8_wrapper(i16 0, i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; AMDGPU:       worker_state_machine.parallel_region.fallback.execute:
+; AMDGPU-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; AMDGPU:       worker_state_machine.parallel_region.end:
+; AMDGPU-NEXT:    call void @__kmpc_kernel_end_parallel()
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; AMDGPU:       worker_state_machine.done.barrier:
+; AMDGPU-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; AMDGPU:       thread.user_code.check:
+; AMDGPU-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU:       user_code.entry:
+; AMDGPU-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; AMDGPU-NEXT:    call void @__omp_outlined__6(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; AMDGPU-NEXT:    ret void
+; AMDGPU:       worker.exit:
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__6
+; AMDGPU-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; AMDGPU-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; AMDGPU-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__7 to i8*), i8* noundef @__omp_outlined__7_wrapper.ID, i8** noundef [[TMP1]], i64 noundef 0)
+; AMDGPU-NEXT:    [[CALL:%.*]] = call i32 @unknown() #[[ATTR10]]
+; AMDGPU-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS1]] to i8**
+; AMDGPU-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__8 to i8*), i8* noundef @__omp_outlined__8_wrapper.ID, i8** noundef [[TMP2]], i64 noundef 0)
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__7
+; AMDGPU-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    call void @p0() #[[ATTR10]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__7_wrapper
+; AMDGPU-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-NEXT:    call void @__omp_outlined__7(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__8
+; AMDGPU-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    call void @p1() #[[ATTR10]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__8_wrapper
+; AMDGPU-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-NEXT:    call void @__omp_outlined__8(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_no_openmp_attr_l66
+; AMDGPU-SAME: () #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8, addrspace(5)
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; AMDGPU-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; AMDGPU-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; AMDGPU:       worker_state_machine.begin:
+; AMDGPU-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR_GENERIC:%.*]] = addrspacecast i8* addrspace(5)* [[WORKER_WORK_FN_ADDR]] to i8**
+; AMDGPU-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR_GENERIC]])
+; AMDGPU-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR_GENERIC]], align 8
+; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; AMDGPU-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; AMDGPU-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; AMDGPU:       worker_state_machine.finished:
+; AMDGPU-NEXT:    ret void
+; AMDGPU:       worker_state_machine.is_active.check:
+; AMDGPU-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.check:
+; AMDGPU-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__10_wrapper.ID to void (i16, i32)*)
+; AMDGPU-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK1:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.execute:
+; AMDGPU-NEXT:    call void @__omp_outlined__10_wrapper(i16 0, i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.check1:
+; AMDGPU-NEXT:    br i1 true, label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE2:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK3:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.execute2:
+; AMDGPU-NEXT:    call void @__omp_outlined__11_wrapper(i16 0, i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; AMDGPU:       worker_state_machine.parallel_region.check3:
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; AMDGPU:       worker_state_machine.parallel_region.end:
+; AMDGPU-NEXT:    call void @__kmpc_kernel_end_parallel()
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; AMDGPU:       worker_state_machine.done.barrier:
+; AMDGPU-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; AMDGPU:       thread.user_code.check:
+; AMDGPU-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU:       user_code.entry:
+; AMDGPU-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; AMDGPU-NEXT:    call void @__omp_outlined__9(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; AMDGPU-NEXT:    ret void
+; AMDGPU:       worker.exit:
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__9
+; AMDGPU-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; AMDGPU-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; AMDGPU-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__10 to i8*), i8* noundef @__omp_outlined__10_wrapper.ID, i8** noundef [[TMP1]], i64 noundef 0)
+; AMDGPU-NEXT:    call void @unknown_no_openmp() #[[ATTR9]]
+; AMDGPU-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS1]] to i8**
+; AMDGPU-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__11 to i8*), i8* noundef @__omp_outlined__11_wrapper.ID, i8** noundef [[TMP2]], i64 noundef 0)
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__10
+; AMDGPU-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    call void @p0() #[[ATTR10]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__10_wrapper
+; AMDGPU-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-NEXT:    call void @__omp_outlined__10(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__11
+; AMDGPU-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    call void @p1() #[[ATTR10]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__11_wrapper
+; AMDGPU-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-NEXT:    call void @__omp_outlined__11(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_pure_l77
+; AMDGPU-SAME: () #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8, addrspace(5)
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; AMDGPU-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; AMDGPU-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; AMDGPU:       worker_state_machine.begin:
+; AMDGPU-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR_GENERIC:%.*]] = addrspacecast i8* addrspace(5)* [[WORKER_WORK_FN_ADDR]] to i8**
+; AMDGPU-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR_GENERIC]])
+; AMDGPU-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR_GENERIC]], align 8
+; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; AMDGPU-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; AMDGPU-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; AMDGPU:       worker_state_machine.finished:
+; AMDGPU-NEXT:    ret void
+; AMDGPU:       worker_state_machine.is_active.check:
+; AMDGPU-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.check:
+; AMDGPU-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__13_wrapper.ID to void (i16, i32)*)
+; AMDGPU-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK1:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.execute:
+; AMDGPU-NEXT:    call void @__omp_outlined__13_wrapper(i16 0, i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.check1:
+; AMDGPU-NEXT:    br i1 true, label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE2:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK3:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.execute2:
+; AMDGPU-NEXT:    call void @__omp_outlined__14_wrapper(i16 0, i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; AMDGPU:       worker_state_machine.parallel_region.check3:
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; AMDGPU:       worker_state_machine.parallel_region.end:
+; AMDGPU-NEXT:    call void @__kmpc_kernel_end_parallel()
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; AMDGPU:       worker_state_machine.done.barrier:
+; AMDGPU-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; AMDGPU:       thread.user_code.check:
+; AMDGPU-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU:       user_code.entry:
+; AMDGPU-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; AMDGPU-NEXT:    call void @__omp_outlined__12(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; AMDGPU-NEXT:    ret void
+; AMDGPU:       worker.exit:
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__12
+; AMDGPU-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; AMDGPU-NEXT:    call void @unknown_no_openmp() #[[ATTR9]]
+; AMDGPU-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; AMDGPU-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__13 to i8*), i8* noundef @__omp_outlined__13_wrapper.ID, i8** noundef [[TMP1]], i64 noundef 0)
+; AMDGPU-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS1]] to i8**
+; AMDGPU-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__14 to i8*), i8* noundef @__omp_outlined__14_wrapper.ID, i8** noundef [[TMP2]], i64 noundef 0)
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__13
+; AMDGPU-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    call void @p0() #[[ATTR10]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__13_wrapper
+; AMDGPU-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-NEXT:    call void @__omp_outlined__13(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__14
+; AMDGPU-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    call void @p1() #[[ATTR10]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__14_wrapper
+; AMDGPU-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-NEXT:    call void @__omp_outlined__14(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_interprocedural_nested_recursive_l92
+; AMDGPU-SAME: () #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8, addrspace(5)
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; AMDGPU-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; AMDGPU-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; AMDGPU:       worker_state_machine.begin:
+; AMDGPU-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR_GENERIC:%.*]] = addrspacecast i8* addrspace(5)* [[WORKER_WORK_FN_ADDR]] to i8**
+; AMDGPU-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR_GENERIC]])
+; AMDGPU-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR_GENERIC]], align 8
+; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; AMDGPU-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; AMDGPU-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; AMDGPU:       worker_state_machine.finished:
+; AMDGPU-NEXT:    ret void
+; AMDGPU:       worker_state_machine.is_active.check:
+; AMDGPU-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.check:
+; AMDGPU-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], @__omp_outlined__19_wrapper
+; AMDGPU-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.execute:
+; AMDGPU-NEXT:    call void @__omp_outlined__19_wrapper(i16 0, i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.fallback.execute:
+; AMDGPU-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; AMDGPU:       worker_state_machine.parallel_region.end:
+; AMDGPU-NEXT:    call void @__kmpc_kernel_end_parallel()
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; AMDGPU:       worker_state_machine.done.barrier:
+; AMDGPU-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; AMDGPU:       thread.user_code.check:
+; AMDGPU-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU:       user_code.entry:
+; AMDGPU-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; AMDGPU-NEXT:    call void @__omp_outlined__15(i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; AMDGPU-NEXT:    ret void
+; AMDGPU:       worker.exit:
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__15
+; AMDGPU-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[CALL:%.*]] = call i32 bitcast (i32 (...)* @omp_get_thread_num to i32 ()*)() #[[ATTR10]]
+; AMDGPU-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after.internalized(i32 [[CALL]]) #[[ATTR8]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_nested_recursive_after.internalized
+; AMDGPU-SAME: (i32 [[A:%.*]]) #[[ATTR1]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
+; AMDGPU-NEXT:    [[CMP:%.*]] = icmp eq i32 [[A]], 0
+; AMDGPU-NEXT:    br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
+; AMDGPU:       if.then:
+; AMDGPU-NEXT:    br label [[RETURN:%.*]]
+; AMDGPU:       if.end:
+; AMDGPU-NEXT:    [[SUB:%.*]] = sub nsw i32 [[A]], 1
+; AMDGPU-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after.internalized(i32 [[SUB]]) #[[ATTR8]]
+; AMDGPU-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after_after.internalized() #[[ATTR8]]
+; AMDGPU-NEXT:    br label [[RETURN]]
+; AMDGPU:       return:
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_nested_recursive_after
+; AMDGPU-SAME: (i32 [[A:%.*]]) #[[ATTR1]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
+; AMDGPU-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4
+; AMDGPU-NEXT:    [[CMP:%.*]] = icmp eq i32 [[TMP0]], 0
+; AMDGPU-NEXT:    br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
+; AMDGPU:       if.then:
+; AMDGPU-NEXT:    br label [[RETURN:%.*]]
+; AMDGPU:       if.end:
+; AMDGPU-NEXT:    [[TMP1:%.*]] = load i32, i32* [[A_ADDR]], align 4
+; AMDGPU-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP1]], 1
+; AMDGPU-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after(i32 [[SUB]]) #[[ATTR10]]
+; AMDGPU-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after_after() #[[ATTR10]]
+; AMDGPU-NEXT:    br label [[RETURN]]
+; AMDGPU:       return:
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_no_state_machine_weak_callee_l112
+; AMDGPU-SAME: () #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8, addrspace(5)
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; AMDGPU-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; AMDGPU-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; AMDGPU:       worker_state_machine.begin:
+; AMDGPU-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR_GENERIC:%.*]] = addrspacecast i8* addrspace(5)* [[WORKER_WORK_FN_ADDR]] to i8**
+; AMDGPU-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR_GENERIC]])
+; AMDGPU-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR_GENERIC]], align 8
+; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; AMDGPU-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; AMDGPU-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; AMDGPU:       worker_state_machine.finished:
+; AMDGPU-NEXT:    ret void
+; AMDGPU:       worker_state_machine.is_active.check:
+; AMDGPU-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.fallback.execute:
+; AMDGPU-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.end:
+; AMDGPU-NEXT:    call void @__kmpc_kernel_end_parallel()
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; AMDGPU:       worker_state_machine.done.barrier:
+; AMDGPU-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; AMDGPU:       thread.user_code.check:
+; AMDGPU-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU:       user_code.entry:
+; AMDGPU-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; AMDGPU-NEXT:    call void @__omp_outlined__16(i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; AMDGPU-NEXT:    ret void
+; AMDGPU:       worker.exit:
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__16
+; AMDGPU-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    call void @weak_callee_empty() #[[ATTR8]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@weak_callee_empty
+; AMDGPU-SAME: () #[[ATTR1]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__17
+; AMDGPU-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    call void @p0() #[[ATTR10]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__17_wrapper
+; AMDGPU-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-NEXT:    call void @__omp_outlined__17(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__18
+; AMDGPU-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    call void @p0() #[[ATTR10]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__18_wrapper
+; AMDGPU-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-NEXT:    call void @__omp_outlined__18(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_nested_recursive_after_after.internalized
+; AMDGPU-SAME: () #[[ATTR1]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef @[[GLOB2]]) #[[ATTR3]]
+; AMDGPU-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__19 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__19_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_nested_recursive_after_after
+; AMDGPU-SAME: () #[[ATTR1]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
+; AMDGPU-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__19 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__19_wrapper to i8*), i8** [[TMP1]], i64 0)
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__19
+; AMDGPU-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    call void @p0() #[[ATTR10]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; AMDGPU: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__19_wrapper
+; AMDGPU-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-NEXT:    call void @__omp_outlined__19(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_no_state_machine_needed_l14
+; NVPTX-SAME: () #[[ATTR0:[0-9]+]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i1 false, i1 false, i1 true)
+; NVPTX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX:       user_code.entry:
+; NVPTX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3:[0-9]+]]
+; NVPTX-NEXT:    call void @__omp_outlined__(i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; NVPTX-NEXT:    ret void
+; NVPTX:       worker.exit:
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__
+; NVPTX-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    call void @no_parallel_region_in_here.internalized() #[[ATTR8:[0-9]+]]
+; NVPTX-NEXT:    call void @unknown_no_openmp() #[[ATTR9:[0-9]+]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline nounwind
+; NVPTX-LABEL: define {{[^@]+}}@no_parallel_region_in_here.internalized
+; NVPTX-SAME: () #[[ATTR1:[0-9]+]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef @[[GLOB2:[0-9]+]]) #[[ATTR3]]
+; NVPTX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_single(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]]) #[[ATTR3]]
+; NVPTX-NEXT:    [[TMP2:%.*]] = icmp ne i32 [[TMP1]], 0
+; NVPTX-NEXT:    br i1 [[TMP2]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
+; NVPTX:       omp_if.then:
+; NVPTX-NEXT:    store i32 0, i32* @G, align 4
+; NVPTX-NEXT:    call void @__kmpc_end_single(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]]) #[[ATTR3]]
+; NVPTX-NEXT:    br label [[OMP_IF_END]]
+; NVPTX:       omp_if.end:
+; NVPTX-NEXT:    call void @__kmpc_barrier(%struct.ident_t* noundef @[[GLOB3:[0-9]+]], i32 [[TMP0]]) #[[ATTR3]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline nounwind
+; NVPTX-LABEL: define {{[^@]+}}@no_parallel_region_in_here
+; NVPTX-SAME: () #[[ATTR1]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
+; NVPTX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
+; NVPTX-NEXT:    [[TMP2:%.*]] = icmp ne i32 [[TMP1]], 0
+; NVPTX-NEXT:    br i1 [[TMP2]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
+; NVPTX:       omp_if.then:
+; NVPTX-NEXT:    store i32 0, i32* @G, align 4
+; NVPTX-NEXT:    call void @__kmpc_end_single(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[OMP_IF_END]]
+; NVPTX:       omp_if.end:
+; NVPTX-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB3]], i32 [[TMP0]])
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_l22
+; NVPTX-SAME: () #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; NVPTX-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; NVPTX-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; NVPTX:       worker_state_machine.begin:
+; NVPTX-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
+; NVPTX-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
+; NVPTX-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; NVPTX-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; NVPTX-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; NVPTX:       worker_state_machine.finished:
+; NVPTX-NEXT:    ret void
+; NVPTX:       worker_state_machine.is_active.check:
+; NVPTX-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.check:
+; NVPTX-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__2_wrapper.ID to void (i16, i32)*)
+; NVPTX-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK1:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.execute:
+; NVPTX-NEXT:    call void @__omp_outlined__2_wrapper(i16 0, i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.check1:
+; NVPTX-NEXT:    br i1 true, label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE2:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK3:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.execute2:
+; NVPTX-NEXT:    call void @__omp_outlined__3_wrapper(i16 0, i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; NVPTX:       worker_state_machine.parallel_region.check3:
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; NVPTX:       worker_state_machine.parallel_region.end:
+; NVPTX-NEXT:    call void @__kmpc_kernel_end_parallel()
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; NVPTX:       worker_state_machine.done.barrier:
+; NVPTX-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; NVPTX:       thread.user_code.check:
+; NVPTX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX:       user_code.entry:
+; NVPTX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; NVPTX-NEXT:    call void @__omp_outlined__1(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; NVPTX-NEXT:    ret void
+; NVPTX:       worker.exit:
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__1
+; NVPTX-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; NVPTX-NEXT:    call void @unknown_no_openmp() #[[ATTR9]]
+; NVPTX-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; NVPTX-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__2 to i8*), i8* noundef @__omp_outlined__2_wrapper.ID, i8** noundef [[TMP1]], i64 noundef 0)
+; NVPTX-NEXT:    call void @no_parallel_region_in_here.internalized() #[[ATTR8]]
+; NVPTX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS1]] to i8**
+; NVPTX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__3 to i8*), i8* noundef @__omp_outlined__3_wrapper.ID, i8** noundef [[TMP2]], i64 noundef 0)
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__2
+; NVPTX-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    call void @p0() #[[ATTR10:[0-9]+]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__2_wrapper
+; NVPTX-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-NEXT:    call void @__omp_outlined__2(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__3
+; NVPTX-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    call void @p1() #[[ATTR10]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__3_wrapper
+; NVPTX-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-NEXT:    call void @__omp_outlined__3(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_interprocedural_l39
+; NVPTX-SAME: () #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; NVPTX-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; NVPTX-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; NVPTX:       worker_state_machine.begin:
+; NVPTX-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
+; NVPTX-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
+; NVPTX-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; NVPTX-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; NVPTX-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; NVPTX:       worker_state_machine.finished:
+; NVPTX-NEXT:    ret void
+; NVPTX:       worker_state_machine.is_active.check:
+; NVPTX-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.check:
+; NVPTX-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], @__omp_outlined__17_wrapper
+; NVPTX-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK1:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.execute:
+; NVPTX-NEXT:    call void @__omp_outlined__17_wrapper(i16 0, i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.check1:
+; NVPTX-NEXT:    [[WORKER_CHECK_PARALLEL_REGION4:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__5_wrapper.ID to void (i16, i32)*)
+; NVPTX-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION4]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE2:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK3:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.execute2:
+; NVPTX-NEXT:    call void @__omp_outlined__5_wrapper(i16 0, i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; NVPTX:       worker_state_machine.parallel_region.check3:
+; NVPTX-NEXT:    br i1 true, label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE5:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK6:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.execute5:
+; NVPTX-NEXT:    call void @__omp_outlined__18_wrapper(i16 0, i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; NVPTX:       worker_state_machine.parallel_region.check6:
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; NVPTX:       worker_state_machine.parallel_region.end:
+; NVPTX-NEXT:    call void @__kmpc_kernel_end_parallel()
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; NVPTX:       worker_state_machine.done.barrier:
+; NVPTX-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; NVPTX:       thread.user_code.check:
+; NVPTX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX:       user_code.entry:
+; NVPTX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; NVPTX-NEXT:    call void @__omp_outlined__4(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; NVPTX-NEXT:    ret void
+; NVPTX:       worker.exit:
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__4
+; NVPTX-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; NVPTX-NEXT:    call void @unknown_no_openmp() #[[ATTR9]]
+; NVPTX-NEXT:    call void @simple_state_machine_interprocedural_before.internalized() #[[ATTR8]]
+; NVPTX-NEXT:    call void @no_parallel_region_in_here.internalized() #[[ATTR8]]
+; NVPTX-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; NVPTX-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__5 to i8*), i8* noundef @__omp_outlined__5_wrapper.ID, i8** noundef [[TMP1]], i64 noundef 0)
+; NVPTX-NEXT:    call void @simple_state_machine_interprocedural_after.internalized() #[[ATTR8]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline nounwind
+; NVPTX-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_before.internalized
+; NVPTX-SAME: () #[[ATTR1]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef @[[GLOB2]]) #[[ATTR3]]
+; NVPTX-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__17 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__17_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline nounwind
+; NVPTX-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_before
+; NVPTX-SAME: () #[[ATTR1]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
+; NVPTX-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__17 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__17_wrapper to i8*), i8** [[TMP1]], i64 0)
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__5
+; NVPTX-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    call void @p1() #[[ATTR10]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__5_wrapper
+; NVPTX-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-NEXT:    call void @__omp_outlined__5(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline nounwind
+; NVPTX-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_after.internalized
+; NVPTX-SAME: () #[[ATTR1]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef @[[GLOB2]]) #[[ATTR3]]
+; NVPTX-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__18 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__18_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline nounwind
+; NVPTX-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_after
+; NVPTX-SAME: () #[[ATTR1]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
+; NVPTX-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__18 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__18_wrapper to i8*), i8** [[TMP1]], i64 0)
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_with_fallback_l55
+; NVPTX-SAME: () #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; NVPTX-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; NVPTX-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; NVPTX:       worker_state_machine.begin:
+; NVPTX-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
+; NVPTX-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
+; NVPTX-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; NVPTX-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; NVPTX-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; NVPTX:       worker_state_machine.finished:
+; NVPTX-NEXT:    ret void
+; NVPTX:       worker_state_machine.is_active.check:
+; NVPTX-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.check:
+; NVPTX-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__7_wrapper.ID to void (i16, i32)*)
+; NVPTX-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK1:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.execute:
+; NVPTX-NEXT:    call void @__omp_outlined__7_wrapper(i16 0, i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.check1:
+; NVPTX-NEXT:    [[WORKER_CHECK_PARALLEL_REGION4:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__8_wrapper.ID to void (i16, i32)*)
+; NVPTX-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION4]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE2:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.execute2:
+; NVPTX-NEXT:    call void @__omp_outlined__8_wrapper(i16 0, i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; NVPTX:       worker_state_machine.parallel_region.fallback.execute:
+; NVPTX-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; NVPTX:       worker_state_machine.parallel_region.end:
+; NVPTX-NEXT:    call void @__kmpc_kernel_end_parallel()
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; NVPTX:       worker_state_machine.done.barrier:
+; NVPTX-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; NVPTX:       thread.user_code.check:
+; NVPTX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX:       user_code.entry:
+; NVPTX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; NVPTX-NEXT:    call void @__omp_outlined__6(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; NVPTX-NEXT:    ret void
+; NVPTX:       worker.exit:
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__6
+; NVPTX-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; NVPTX-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; NVPTX-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__7 to i8*), i8* noundef @__omp_outlined__7_wrapper.ID, i8** noundef [[TMP1]], i64 noundef 0)
+; NVPTX-NEXT:    [[CALL:%.*]] = call i32 @unknown() #[[ATTR10]]
+; NVPTX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS1]] to i8**
+; NVPTX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__8 to i8*), i8* noundef @__omp_outlined__8_wrapper.ID, i8** noundef [[TMP2]], i64 noundef 0)
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__7
+; NVPTX-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    call void @p0() #[[ATTR10]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__7_wrapper
+; NVPTX-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-NEXT:    call void @__omp_outlined__7(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__8
+; NVPTX-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    call void @p1() #[[ATTR10]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__8_wrapper
+; NVPTX-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-NEXT:    call void @__omp_outlined__8(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_no_openmp_attr_l66
+; NVPTX-SAME: () #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; NVPTX-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; NVPTX-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; NVPTX:       worker_state_machine.begin:
+; NVPTX-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
+; NVPTX-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
+; NVPTX-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; NVPTX-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; NVPTX-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; NVPTX:       worker_state_machine.finished:
+; NVPTX-NEXT:    ret void
+; NVPTX:       worker_state_machine.is_active.check:
+; NVPTX-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.check:
+; NVPTX-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__10_wrapper.ID to void (i16, i32)*)
+; NVPTX-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK1:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.execute:
+; NVPTX-NEXT:    call void @__omp_outlined__10_wrapper(i16 0, i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.check1:
+; NVPTX-NEXT:    br i1 true, label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE2:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK3:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.execute2:
+; NVPTX-NEXT:    call void @__omp_outlined__11_wrapper(i16 0, i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; NVPTX:       worker_state_machine.parallel_region.check3:
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; NVPTX:       worker_state_machine.parallel_region.end:
+; NVPTX-NEXT:    call void @__kmpc_kernel_end_parallel()
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; NVPTX:       worker_state_machine.done.barrier:
+; NVPTX-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; NVPTX:       thread.user_code.check:
+; NVPTX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX:       user_code.entry:
+; NVPTX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; NVPTX-NEXT:    call void @__omp_outlined__9(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; NVPTX-NEXT:    ret void
+; NVPTX:       worker.exit:
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__9
+; NVPTX-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; NVPTX-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; NVPTX-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__10 to i8*), i8* noundef @__omp_outlined__10_wrapper.ID, i8** noundef [[TMP1]], i64 noundef 0)
+; NVPTX-NEXT:    call void @unknown_no_openmp() #[[ATTR9]]
+; NVPTX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS1]] to i8**
+; NVPTX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__11 to i8*), i8* noundef @__omp_outlined__11_wrapper.ID, i8** noundef [[TMP2]], i64 noundef 0)
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__10
+; NVPTX-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    call void @p0() #[[ATTR10]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__10_wrapper
+; NVPTX-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-NEXT:    call void @__omp_outlined__10(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__11
+; NVPTX-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    call void @p1() #[[ATTR10]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__11_wrapper
+; NVPTX-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-NEXT:    call void @__omp_outlined__11(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_pure_l77
+; NVPTX-SAME: () #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; NVPTX-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; NVPTX-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; NVPTX:       worker_state_machine.begin:
+; NVPTX-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
+; NVPTX-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
+; NVPTX-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; NVPTX-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; NVPTX-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; NVPTX:       worker_state_machine.finished:
+; NVPTX-NEXT:    ret void
+; NVPTX:       worker_state_machine.is_active.check:
+; NVPTX-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.check:
+; NVPTX-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__13_wrapper.ID to void (i16, i32)*)
+; NVPTX-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK1:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.execute:
+; NVPTX-NEXT:    call void @__omp_outlined__13_wrapper(i16 0, i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.check1:
+; NVPTX-NEXT:    br i1 true, label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE2:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK3:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.execute2:
+; NVPTX-NEXT:    call void @__omp_outlined__14_wrapper(i16 0, i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; NVPTX:       worker_state_machine.parallel_region.check3:
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; NVPTX:       worker_state_machine.parallel_region.end:
+; NVPTX-NEXT:    call void @__kmpc_kernel_end_parallel()
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; NVPTX:       worker_state_machine.done.barrier:
+; NVPTX-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; NVPTX:       thread.user_code.check:
+; NVPTX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX:       user_code.entry:
+; NVPTX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; NVPTX-NEXT:    call void @__omp_outlined__12(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; NVPTX-NEXT:    ret void
+; NVPTX:       worker.exit:
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__12
+; NVPTX-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; NVPTX-NEXT:    call void @unknown_no_openmp() #[[ATTR9]]
+; NVPTX-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; NVPTX-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__13 to i8*), i8* noundef @__omp_outlined__13_wrapper.ID, i8** noundef [[TMP1]], i64 noundef 0)
+; NVPTX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS1]] to i8**
+; NVPTX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__14 to i8*), i8* noundef @__omp_outlined__14_wrapper.ID, i8** noundef [[TMP2]], i64 noundef 0)
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__13
+; NVPTX-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    call void @p0() #[[ATTR10]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__13_wrapper
+; NVPTX-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-NEXT:    call void @__omp_outlined__13(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__14
+; NVPTX-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    call void @p1() #[[ATTR10]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__14_wrapper
+; NVPTX-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-NEXT:    call void @__omp_outlined__14(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_interprocedural_nested_recursive_l92
+; NVPTX-SAME: () #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; NVPTX-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; NVPTX-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; NVPTX:       worker_state_machine.begin:
+; NVPTX-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
+; NVPTX-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
+; NVPTX-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; NVPTX-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; NVPTX-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; NVPTX:       worker_state_machine.finished:
+; NVPTX-NEXT:    ret void
+; NVPTX:       worker_state_machine.is_active.check:
+; NVPTX-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.check:
+; NVPTX-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], @__omp_outlined__19_wrapper
+; NVPTX-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.execute:
+; NVPTX-NEXT:    call void @__omp_outlined__19_wrapper(i16 0, i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.fallback.execute:
+; NVPTX-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; NVPTX:       worker_state_machine.parallel_region.end:
+; NVPTX-NEXT:    call void @__kmpc_kernel_end_parallel()
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; NVPTX:       worker_state_machine.done.barrier:
+; NVPTX-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; NVPTX:       thread.user_code.check:
+; NVPTX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX:       user_code.entry:
+; NVPTX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; NVPTX-NEXT:    call void @__omp_outlined__15(i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; NVPTX-NEXT:    ret void
+; NVPTX:       worker.exit:
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__15
+; NVPTX-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[CALL:%.*]] = call i32 bitcast (i32 (...)* @omp_get_thread_num to i32 ()*)() #[[ATTR10]]
+; NVPTX-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after.internalized(i32 [[CALL]]) #[[ATTR8]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline nounwind
+; NVPTX-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_nested_recursive_after.internalized
+; NVPTX-SAME: (i32 [[A:%.*]]) #[[ATTR1]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
+; NVPTX-NEXT:    [[CMP:%.*]] = icmp eq i32 [[A]], 0
+; NVPTX-NEXT:    br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
+; NVPTX:       if.then:
+; NVPTX-NEXT:    br label [[RETURN:%.*]]
+; NVPTX:       if.end:
+; NVPTX-NEXT:    [[SUB:%.*]] = sub nsw i32 [[A]], 1
+; NVPTX-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after.internalized(i32 [[SUB]]) #[[ATTR8]]
+; NVPTX-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after_after.internalized() #[[ATTR8]]
+; NVPTX-NEXT:    br label [[RETURN]]
+; NVPTX:       return:
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline nounwind
+; NVPTX-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_nested_recursive_after
+; NVPTX-SAME: (i32 [[A:%.*]]) #[[ATTR1]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
+; NVPTX-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4
+; NVPTX-NEXT:    [[CMP:%.*]] = icmp eq i32 [[TMP0]], 0
+; NVPTX-NEXT:    br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
+; NVPTX:       if.then:
+; NVPTX-NEXT:    br label [[RETURN:%.*]]
+; NVPTX:       if.end:
+; NVPTX-NEXT:    [[TMP1:%.*]] = load i32, i32* [[A_ADDR]], align 4
+; NVPTX-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP1]], 1
+; NVPTX-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after(i32 [[SUB]]) #[[ATTR10]]
+; NVPTX-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after_after() #[[ATTR10]]
+; NVPTX-NEXT:    br label [[RETURN]]
+; NVPTX:       return:
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_no_state_machine_weak_callee_l112
+; NVPTX-SAME: () #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; NVPTX-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; NVPTX-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; NVPTX:       worker_state_machine.begin:
+; NVPTX-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
+; NVPTX-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
+; NVPTX-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; NVPTX-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; NVPTX-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; NVPTX:       worker_state_machine.finished:
+; NVPTX-NEXT:    ret void
+; NVPTX:       worker_state_machine.is_active.check:
+; NVPTX-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.fallback.execute:
+; NVPTX-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.end:
+; NVPTX-NEXT:    call void @__kmpc_kernel_end_parallel()
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; NVPTX:       worker_state_machine.done.barrier:
+; NVPTX-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; NVPTX:       thread.user_code.check:
+; NVPTX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX:       user_code.entry:
+; NVPTX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; NVPTX-NEXT:    call void @__omp_outlined__16(i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; NVPTX-NEXT:    ret void
+; NVPTX:       worker.exit:
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__16
+; NVPTX-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    call void @weak_callee_empty() #[[ATTR8]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline nounwind
+; NVPTX-LABEL: define {{[^@]+}}@weak_callee_empty
+; NVPTX-SAME: () #[[ATTR1]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__17
+; NVPTX-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    call void @p0() #[[ATTR10]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__17_wrapper
+; NVPTX-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-NEXT:    call void @__omp_outlined__17(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__18
+; NVPTX-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    call void @p0() #[[ATTR10]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__18_wrapper
+; NVPTX-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-NEXT:    call void @__omp_outlined__18(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline nounwind
+; NVPTX-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_nested_recursive_after_after.internalized
+; NVPTX-SAME: () #[[ATTR1]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef @[[GLOB2]]) #[[ATTR3]]
+; NVPTX-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__19 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__19_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline nounwind
+; NVPTX-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_nested_recursive_after_after
+; NVPTX-SAME: () #[[ATTR1]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
+; NVPTX-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__19 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__19_wrapper to i8*), i8** [[TMP1]], i64 0)
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__19
+; NVPTX-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    call void @p0() #[[ATTR10]]
+; NVPTX-NEXT:    ret void
+;
+;
+; NVPTX: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__19_wrapper
+; NVPTX-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-NEXT:    call void @__omp_outlined__19(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_no_state_machine_needed_l14
+; AMDGPU-DISABLED-SAME: () #[[ATTR0:[0-9]+]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i1 false, i1 true, i1 true)
+; AMDGPU-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU-DISABLED:       user_code.entry:
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3:[0-9]+]]
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__(i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; AMDGPU-DISABLED-NEXT:    ret void
+; AMDGPU-DISABLED:       worker.exit:
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    call void @no_parallel_region_in_here.internalized() #[[ATTR8:[0-9]+]]
+; AMDGPU-DISABLED-NEXT:    call void @unknown_no_openmp() #[[ATTR9:[0-9]+]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@no_parallel_region_in_here.internalized
+; AMDGPU-DISABLED-SAME: () #[[ATTR1:[0-9]+]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef @[[GLOB2:[0-9]+]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_single(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    [[TMP2:%.*]] = icmp ne i32 [[TMP1]], 0
+; AMDGPU-DISABLED-NEXT:    br i1 [[TMP2]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
+; AMDGPU-DISABLED:       omp_if.then:
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* @G, align 4
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_end_single(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    br label [[OMP_IF_END]]
+; AMDGPU-DISABLED:       omp_if.end:
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_barrier(%struct.ident_t* noundef @[[GLOB3:[0-9]+]], i32 [[TMP0]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@no_parallel_region_in_here
+; AMDGPU-DISABLED-SAME: () #[[ATTR1]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
+; AMDGPU-DISABLED-NEXT:    [[TMP2:%.*]] = icmp ne i32 [[TMP1]], 0
+; AMDGPU-DISABLED-NEXT:    br i1 [[TMP2]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
+; AMDGPU-DISABLED:       omp_if.then:
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* @G, align 4
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_end_single(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
+; AMDGPU-DISABLED-NEXT:    br label [[OMP_IF_END]]
+; AMDGPU-DISABLED:       omp_if.end:
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB3]], i32 [[TMP0]])
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_l22
+; AMDGPU-DISABLED-SAME: () #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 true, i1 true)
+; AMDGPU-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU-DISABLED:       user_code.entry:
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__1(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; AMDGPU-DISABLED-NEXT:    ret void
+; AMDGPU-DISABLED:       worker.exit:
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__1
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; AMDGPU-DISABLED-NEXT:    call void @unknown_no_openmp() #[[ATTR9]]
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__2 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__2_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
+; AMDGPU-DISABLED-NEXT:    call void @no_parallel_region_in_here.internalized() #[[ATTR8]]
+; AMDGPU-DISABLED-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS1]] to i8**
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__3 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__3_wrapper to i8*), i8** noundef [[TMP2]], i64 noundef 0)
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__2
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    call void @p0() #[[ATTR10:[0-9]+]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__2_wrapper
+; AMDGPU-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__2(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__3
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    call void @p1() #[[ATTR10]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__3_wrapper
+; AMDGPU-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__3(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_interprocedural_l39
+; AMDGPU-DISABLED-SAME: () #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 true, i1 true)
+; AMDGPU-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU-DISABLED:       user_code.entry:
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__4(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; AMDGPU-DISABLED-NEXT:    ret void
+; AMDGPU-DISABLED:       worker.exit:
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__4
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; AMDGPU-DISABLED-NEXT:    call void @unknown_no_openmp() #[[ATTR9]]
+; AMDGPU-DISABLED-NEXT:    call void @simple_state_machine_interprocedural_before.internalized() #[[ATTR8]]
+; AMDGPU-DISABLED-NEXT:    call void @no_parallel_region_in_here.internalized() #[[ATTR8]]
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__5 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__5_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
+; AMDGPU-DISABLED-NEXT:    call void @simple_state_machine_interprocedural_after.internalized() #[[ATTR8]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_before.internalized
+; AMDGPU-DISABLED-SAME: () #[[ATTR1]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef @[[GLOB2]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__17 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__17_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_before
+; AMDGPU-DISABLED-SAME: () #[[ATTR1]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__17 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__17_wrapper to i8*), i8** [[TMP1]], i64 0)
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__5
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    call void @p1() #[[ATTR10]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__5_wrapper
+; AMDGPU-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__5(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_after.internalized
+; AMDGPU-DISABLED-SAME: () #[[ATTR1]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef @[[GLOB2]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__18 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__18_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_after
+; AMDGPU-DISABLED-SAME: () #[[ATTR1]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__18 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__18_wrapper to i8*), i8** [[TMP1]], i64 0)
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_with_fallback_l55
+; AMDGPU-DISABLED-SAME: () #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 true, i1 true)
+; AMDGPU-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU-DISABLED:       user_code.entry:
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__6(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; AMDGPU-DISABLED-NEXT:    ret void
+; AMDGPU-DISABLED:       worker.exit:
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__6
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__7 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__7_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
+; AMDGPU-DISABLED-NEXT:    [[CALL:%.*]] = call i32 @unknown() #[[ATTR10]]
+; AMDGPU-DISABLED-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS1]] to i8**
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__8 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__8_wrapper to i8*), i8** noundef [[TMP2]], i64 noundef 0)
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__7
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    call void @p0() #[[ATTR10]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__7_wrapper
+; AMDGPU-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__7(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__8
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    call void @p1() #[[ATTR10]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__8_wrapper
+; AMDGPU-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__8(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_no_openmp_attr_l66
+; AMDGPU-DISABLED-SAME: () #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 true, i1 true)
+; AMDGPU-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU-DISABLED:       user_code.entry:
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__9(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; AMDGPU-DISABLED-NEXT:    ret void
+; AMDGPU-DISABLED:       worker.exit:
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__9
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__10 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__10_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
+; AMDGPU-DISABLED-NEXT:    call void @unknown_no_openmp() #[[ATTR9]]
+; AMDGPU-DISABLED-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS1]] to i8**
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__11 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__11_wrapper to i8*), i8** noundef [[TMP2]], i64 noundef 0)
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__10
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    call void @p0() #[[ATTR10]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__10_wrapper
+; AMDGPU-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__10(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__11
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    call void @p1() #[[ATTR10]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__11_wrapper
+; AMDGPU-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__11(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_pure_l77
+; AMDGPU-DISABLED-SAME: () #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 true, i1 true)
+; AMDGPU-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU-DISABLED:       user_code.entry:
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__12(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; AMDGPU-DISABLED-NEXT:    ret void
+; AMDGPU-DISABLED:       worker.exit:
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__12
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; AMDGPU-DISABLED-NEXT:    call void @unknown_no_openmp() #[[ATTR9]]
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__13 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__13_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
+; AMDGPU-DISABLED-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS1]] to i8**
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__14 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__14_wrapper to i8*), i8** noundef [[TMP2]], i64 noundef 0)
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__13
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    call void @p0() #[[ATTR10]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__13_wrapper
+; AMDGPU-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__13(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__14
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    call void @p1() #[[ATTR10]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__14_wrapper
+; AMDGPU-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__14(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_interprocedural_nested_recursive_l92
+; AMDGPU-DISABLED-SAME: () #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 true, i1 true)
+; AMDGPU-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU-DISABLED:       user_code.entry:
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__15(i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; AMDGPU-DISABLED-NEXT:    ret void
+; AMDGPU-DISABLED:       worker.exit:
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__15
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[CALL:%.*]] = call i32 bitcast (i32 (...)* @omp_get_thread_num to i32 ()*)() #[[ATTR10]]
+; AMDGPU-DISABLED-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after.internalized(i32 [[CALL]]) #[[ATTR8]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_nested_recursive_after.internalized
+; AMDGPU-DISABLED-SAME: (i32 [[A:%.*]]) #[[ATTR1]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    [[CMP:%.*]] = icmp eq i32 [[A]], 0
+; AMDGPU-DISABLED-NEXT:    br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
+; AMDGPU-DISABLED:       if.then:
+; AMDGPU-DISABLED-NEXT:    br label [[RETURN:%.*]]
+; AMDGPU-DISABLED:       if.end:
+; AMDGPU-DISABLED-NEXT:    [[SUB:%.*]] = sub nsw i32 [[A]], 1
+; AMDGPU-DISABLED-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after.internalized(i32 [[SUB]]) #[[ATTR8]]
+; AMDGPU-DISABLED-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after_after.internalized() #[[ATTR8]]
+; AMDGPU-DISABLED-NEXT:    br label [[RETURN]]
+; AMDGPU-DISABLED:       return:
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_nested_recursive_after
+; AMDGPU-DISABLED-SAME: (i32 [[A:%.*]]) #[[ATTR1]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    [[CMP:%.*]] = icmp eq i32 [[TMP0]], 0
+; AMDGPU-DISABLED-NEXT:    br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
+; AMDGPU-DISABLED:       if.then:
+; AMDGPU-DISABLED-NEXT:    br label [[RETURN:%.*]]
+; AMDGPU-DISABLED:       if.end:
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = load i32, i32* [[A_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP1]], 1
+; AMDGPU-DISABLED-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after(i32 [[SUB]]) #[[ATTR10]]
+; AMDGPU-DISABLED-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after_after() #[[ATTR10]]
+; AMDGPU-DISABLED-NEXT:    br label [[RETURN]]
+; AMDGPU-DISABLED:       return:
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_no_state_machine_weak_callee_l112
+; AMDGPU-DISABLED-SAME: () #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 true, i1 true)
+; AMDGPU-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU-DISABLED:       user_code.entry:
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__16(i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; AMDGPU-DISABLED-NEXT:    ret void
+; AMDGPU-DISABLED:       worker.exit:
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__16
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    call void @weak_callee_empty() #[[ATTR8]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@weak_callee_empty
+; AMDGPU-DISABLED-SAME: () #[[ATTR1]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__17
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    call void @p0() #[[ATTR10]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__17_wrapper
+; AMDGPU-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__17(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__18
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    call void @p0() #[[ATTR10]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__18_wrapper
+; AMDGPU-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__18(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_nested_recursive_after_after.internalized
+; AMDGPU-DISABLED-SAME: () #[[ATTR1]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef @[[GLOB2]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__19 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__19_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_nested_recursive_after_after
+; AMDGPU-DISABLED-SAME: () #[[ATTR1]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__19 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__19_wrapper to i8*), i8** [[TMP1]], i64 0)
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__19
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    call void @p0() #[[ATTR10]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__19_wrapper
+; AMDGPU-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__19(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_no_state_machine_needed_l14
+; NVPTX-DISABLED-SAME: () #[[ATTR0:[0-9]+]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i1 false, i1 true, i1 true)
+; NVPTX-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX-DISABLED:       user_code.entry:
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3:[0-9]+]]
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__(i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; NVPTX-DISABLED-NEXT:    ret void
+; NVPTX-DISABLED:       worker.exit:
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    call void @no_parallel_region_in_here.internalized() #[[ATTR8:[0-9]+]]
+; NVPTX-DISABLED-NEXT:    call void @unknown_no_openmp() #[[ATTR9:[0-9]+]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@no_parallel_region_in_here.internalized
+; NVPTX-DISABLED-SAME: () #[[ATTR1:[0-9]+]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef @[[GLOB2:[0-9]+]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_single(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    [[TMP2:%.*]] = icmp ne i32 [[TMP1]], 0
+; NVPTX-DISABLED-NEXT:    br i1 [[TMP2]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
+; NVPTX-DISABLED:       omp_if.then:
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* @G, align 4
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_end_single(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    br label [[OMP_IF_END]]
+; NVPTX-DISABLED:       omp_if.end:
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_barrier(%struct.ident_t* noundef @[[GLOB3:[0-9]+]], i32 [[TMP0]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@no_parallel_region_in_here
+; NVPTX-DISABLED-SAME: () #[[ATTR1]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
+; NVPTX-DISABLED-NEXT:    [[TMP2:%.*]] = icmp ne i32 [[TMP1]], 0
+; NVPTX-DISABLED-NEXT:    br i1 [[TMP2]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
+; NVPTX-DISABLED:       omp_if.then:
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* @G, align 4
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_end_single(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
+; NVPTX-DISABLED-NEXT:    br label [[OMP_IF_END]]
+; NVPTX-DISABLED:       omp_if.end:
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB3]], i32 [[TMP0]])
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_l22
+; NVPTX-DISABLED-SAME: () #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 true, i1 true)
+; NVPTX-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX-DISABLED:       user_code.entry:
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__1(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; NVPTX-DISABLED-NEXT:    ret void
+; NVPTX-DISABLED:       worker.exit:
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__1
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; NVPTX-DISABLED-NEXT:    call void @unknown_no_openmp() #[[ATTR9]]
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__2 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__2_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
+; NVPTX-DISABLED-NEXT:    call void @no_parallel_region_in_here.internalized() #[[ATTR8]]
+; NVPTX-DISABLED-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS1]] to i8**
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__3 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__3_wrapper to i8*), i8** noundef [[TMP2]], i64 noundef 0)
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__2
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    call void @p0() #[[ATTR10:[0-9]+]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__2_wrapper
+; NVPTX-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__2(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__3
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    call void @p1() #[[ATTR10]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__3_wrapper
+; NVPTX-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__3(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_interprocedural_l39
+; NVPTX-DISABLED-SAME: () #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 true, i1 true)
+; NVPTX-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX-DISABLED:       user_code.entry:
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__4(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; NVPTX-DISABLED-NEXT:    ret void
+; NVPTX-DISABLED:       worker.exit:
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__4
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; NVPTX-DISABLED-NEXT:    call void @unknown_no_openmp() #[[ATTR9]]
+; NVPTX-DISABLED-NEXT:    call void @simple_state_machine_interprocedural_before.internalized() #[[ATTR8]]
+; NVPTX-DISABLED-NEXT:    call void @no_parallel_region_in_here.internalized() #[[ATTR8]]
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__5 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__5_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
+; NVPTX-DISABLED-NEXT:    call void @simple_state_machine_interprocedural_after.internalized() #[[ATTR8]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_before.internalized
+; NVPTX-DISABLED-SAME: () #[[ATTR1]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef @[[GLOB2]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__17 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__17_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_before
+; NVPTX-DISABLED-SAME: () #[[ATTR1]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__17 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__17_wrapper to i8*), i8** [[TMP1]], i64 0)
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__5
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    call void @p1() #[[ATTR10]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__5_wrapper
+; NVPTX-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__5(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_after.internalized
+; NVPTX-DISABLED-SAME: () #[[ATTR1]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef @[[GLOB2]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__18 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__18_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_after
+; NVPTX-DISABLED-SAME: () #[[ATTR1]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__18 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__18_wrapper to i8*), i8** [[TMP1]], i64 0)
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_with_fallback_l55
+; NVPTX-DISABLED-SAME: () #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 true, i1 true)
+; NVPTX-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX-DISABLED:       user_code.entry:
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__6(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; NVPTX-DISABLED-NEXT:    ret void
+; NVPTX-DISABLED:       worker.exit:
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__6
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__7 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__7_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
+; NVPTX-DISABLED-NEXT:    [[CALL:%.*]] = call i32 @unknown() #[[ATTR10]]
+; NVPTX-DISABLED-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS1]] to i8**
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__8 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__8_wrapper to i8*), i8** noundef [[TMP2]], i64 noundef 0)
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__7
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    call void @p0() #[[ATTR10]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__7_wrapper
+; NVPTX-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__7(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__8
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    call void @p1() #[[ATTR10]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__8_wrapper
+; NVPTX-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__8(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_no_openmp_attr_l66
+; NVPTX-DISABLED-SAME: () #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 true, i1 true)
+; NVPTX-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX-DISABLED:       user_code.entry:
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__9(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; NVPTX-DISABLED-NEXT:    ret void
+; NVPTX-DISABLED:       worker.exit:
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__9
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__10 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__10_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
+; NVPTX-DISABLED-NEXT:    call void @unknown_no_openmp() #[[ATTR9]]
+; NVPTX-DISABLED-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS1]] to i8**
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__11 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__11_wrapper to i8*), i8** noundef [[TMP2]], i64 noundef 0)
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__10
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    call void @p0() #[[ATTR10]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__10_wrapper
+; NVPTX-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__10(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__11
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    call void @p1() #[[ATTR10]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__11_wrapper
+; NVPTX-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__11(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_pure_l77
+; NVPTX-DISABLED-SAME: () #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 true, i1 true)
+; NVPTX-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX-DISABLED:       user_code.entry:
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__12(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; NVPTX-DISABLED-NEXT:    ret void
+; NVPTX-DISABLED:       worker.exit:
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__12
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS1:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; NVPTX-DISABLED-NEXT:    call void @unknown_no_openmp() #[[ATTR9]]
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__13 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__13_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
+; NVPTX-DISABLED-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS1]] to i8**
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__14 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__14_wrapper to i8*), i8** noundef [[TMP2]], i64 noundef 0)
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__13
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    call void @p0() #[[ATTR10]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__13_wrapper
+; NVPTX-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__13(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__14
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    call void @p1() #[[ATTR10]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__14_wrapper
+; NVPTX-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__14(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_simple_state_machine_interprocedural_nested_recursive_l92
+; NVPTX-DISABLED-SAME: () #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 true, i1 true)
+; NVPTX-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX-DISABLED:       user_code.entry:
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__15(i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; NVPTX-DISABLED-NEXT:    ret void
+; NVPTX-DISABLED:       worker.exit:
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__15
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[CALL:%.*]] = call i32 bitcast (i32 (...)* @omp_get_thread_num to i32 ()*)() #[[ATTR10]]
+; NVPTX-DISABLED-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after.internalized(i32 [[CALL]]) #[[ATTR8]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_nested_recursive_after.internalized
+; NVPTX-DISABLED-SAME: (i32 [[A:%.*]]) #[[ATTR1]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    [[CMP:%.*]] = icmp eq i32 [[A]], 0
+; NVPTX-DISABLED-NEXT:    br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
+; NVPTX-DISABLED:       if.then:
+; NVPTX-DISABLED-NEXT:    br label [[RETURN:%.*]]
+; NVPTX-DISABLED:       if.end:
+; NVPTX-DISABLED-NEXT:    [[SUB:%.*]] = sub nsw i32 [[A]], 1
+; NVPTX-DISABLED-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after.internalized(i32 [[SUB]]) #[[ATTR8]]
+; NVPTX-DISABLED-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after_after.internalized() #[[ATTR8]]
+; NVPTX-DISABLED-NEXT:    br label [[RETURN]]
+; NVPTX-DISABLED:       return:
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_nested_recursive_after
+; NVPTX-DISABLED-SAME: (i32 [[A:%.*]]) #[[ATTR1]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    store i32 [[A]], i32* [[A_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    [[CMP:%.*]] = icmp eq i32 [[TMP0]], 0
+; NVPTX-DISABLED-NEXT:    br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_END:%.*]]
+; NVPTX-DISABLED:       if.then:
+; NVPTX-DISABLED-NEXT:    br label [[RETURN:%.*]]
+; NVPTX-DISABLED:       if.end:
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = load i32, i32* [[A_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP1]], 1
+; NVPTX-DISABLED-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after(i32 [[SUB]]) #[[ATTR10]]
+; NVPTX-DISABLED-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after_after() #[[ATTR10]]
+; NVPTX-DISABLED-NEXT:    br label [[RETURN]]
+; NVPTX-DISABLED:       return:
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a36502b_no_state_machine_weak_callee_l112
+; NVPTX-DISABLED-SAME: () #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 true, i1 true)
+; NVPTX-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX-DISABLED:       user_code.entry:
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__16(i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; NVPTX-DISABLED-NEXT:    ret void
+; NVPTX-DISABLED:       worker.exit:
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__16
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    call void @weak_callee_empty() #[[ATTR8]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@weak_callee_empty
+; NVPTX-DISABLED-SAME: () #[[ATTR1]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__17
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    call void @p0() #[[ATTR10]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__17_wrapper
+; NVPTX-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__17(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__18
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    call void @p0() #[[ATTR10]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__18_wrapper
+; NVPTX-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__18(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_nested_recursive_after_after.internalized
+; NVPTX-DISABLED-SAME: () #[[ATTR1]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* noundef @[[GLOB2]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB2]], i32 [[TMP0]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__19 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__19_wrapper to i8*), i8** noundef [[TMP1]], i64 noundef 0)
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@simple_state_machine_interprocedural_nested_recursive_after_after
+; NVPTX-DISABLED-SAME: () #[[ATTR1]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__19 to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined__19_wrapper to i8*), i8** [[TMP1]], i64 0)
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__19
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    call void @p0() #[[ATTR10]]
+; NVPTX-DISABLED-NEXT:    ret void
+;
+;
+; NVPTX-DISABLED: Function Attrs: convergent noinline norecurse nounwind
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__19_wrapper
+; NVPTX-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__19(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
+; NVPTX-DISABLED-NEXT:    ret void
 ;

diff --git a/llvm/test/Transforms/OpenMP/spmdization.ll b/llvm/test/Transforms/OpenMP/spmdization.ll
index a9b8525..64ea133 100644 (file)
--- a/llvm/test/Transforms/OpenMP/spmdization.ll
+++ b/llvm/test/Transforms/OpenMP/spmdization.ll
@@ -1,6 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function-signature --check-globals
-; RUN: opt -S -passes=openmp-opt < %s | FileCheck %s
-; RUN: opt -S -passes=openmp-opt -openmp-opt-disable-spmdization < %s | FileCheck %s --check-prefix=CHECK-DISABLED
+; RUN: opt --mtriple=amdgcn-amd-amdhsa --data-layout=A5 -S -passes=openmp-opt < %s | FileCheck %s --check-prefixes=AMDGPU
+; RUN: opt --mtriple=nvptx64-- -S -passes=openmp-opt < %s | FileCheck %s --check-prefixes=NVPTX
+; RUN: opt --mtriple=amdgcn-amd-amdhsa --data-layout=A5 -S -passes=openmp-opt -openmp-opt-disable-spmdization < %s | FileCheck %s --check-prefix=AMDGPU-DISABLED
+; RUN: opt --mtriple=nvptx64-- -S -passes=openmp-opt -openmp-opt-disable-spmdization < %s | FileCheck %s --check-prefix=NVPTX-DISABLED
 
 ;; void unknown(void);
 ;; void spmd_amenable(void) __attribute__((assume("ompx_spmd_amenable")));
@@ -74,8 +76,6 @@
 ;;     }
 ;; }
 
-target triple = "nvptx64"
-
 %struct.ident_t = type { i32, i32, i32, i32, i8* }
 
 @0 = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00", align 1
@@ -87,98 +87,191 @@ target triple = "nvptx64"
 @__omp_offloading_14_a34ca11_do_not_spmdize_target_l65_exec_mode = weak constant i8 1
 @llvm.compiler.used = appending global [5 x i8*] [i8* @__omp_offloading_14_a34ca11_sequential_loop_l5_exec_mode, i8* @__omp_offloading_14_a34ca11_sequential_loop_to_stack_var_l20_exec_mode, i8* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_l35_exec_mode, i8* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_guarded_l50_exec_mode, i8* @__omp_offloading_14_a34ca11_do_not_spmdize_target_l65_exec_mode], section "llvm.metadata"
 
+
+;.
+; AMDGPU: @[[GLOB0:[0-9]+]] = private unnamed_addr constant [23 x i8] c"
+; AMDGPU: @[[GLOB1:[0-9]+]] = private unnamed_addr constant [[STRUCT_IDENT_T:%.*]] { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* @[[GLOB0]], i32 0, i32 0) }, align 8
+; AMDGPU: @[[__OMP_OFFLOADING_14_A34CA11_SEQUENTIAL_LOOP_L5_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 2
+; AMDGPU: @[[__OMP_OFFLOADING_14_A34CA11_SEQUENTIAL_LOOP_TO_STACK_VAR_L20_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 2
+; AMDGPU: @[[__OMP_OFFLOADING_14_A34CA11_SEQUENTIAL_LOOP_TO_SHARED_VAR_L35_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 2
+; AMDGPU: @[[__OMP_OFFLOADING_14_A34CA11_SEQUENTIAL_LOOP_TO_SHARED_VAR_GUARDED_L50_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 2
+; AMDGPU: @[[__OMP_OFFLOADING_14_A34CA11_DO_NOT_SPMDIZE_TARGET_L65_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 1
+; AMDGPU: @[[LLVM_COMPILER_USED:[a-zA-Z0-9_$"\\.-]+]] = appending global [5 x i8*] [i8* @__omp_offloading_14_a34ca11_sequential_loop_l5_exec_mode, i8* @__omp_offloading_14_a34ca11_sequential_loop_to_stack_var_l20_exec_mode, i8* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_l35_exec_mode, i8* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_guarded_l50_exec_mode, i8* @__omp_offloading_14_a34ca11_do_not_spmdize_target_l65_exec_mode], section "llvm.metadata"
+; AMDGPU: @[[X:[a-zA-Z0-9_$"\\.-]+]] = internal addrspace(3) global [4 x i8] undef, align 32
+; AMDGPU: @[[X_1:[a-zA-Z0-9_$"\\.-]+]] = internal addrspace(3) global [4 x i8] undef, align 32
 ;.
-; CHECK: @[[GLOB0:[0-9]+]] = private unnamed_addr constant [23 x i8] c"
-; CHECK: @[[GLOB1:[0-9]+]] = private unnamed_addr constant [[STRUCT_IDENT_T:%.*]] { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* @[[GLOB0]], i32 0, i32 0) }, align 8
-; CHECK: @[[__OMP_OFFLOADING_14_A34CA11_SEQUENTIAL_LOOP_L5_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 2
-; CHECK: @[[__OMP_OFFLOADING_14_A34CA11_SEQUENTIAL_LOOP_TO_STACK_VAR_L20_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 2
-; CHECK: @[[__OMP_OFFLOADING_14_A34CA11_SEQUENTIAL_LOOP_TO_SHARED_VAR_L35_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 2
-; CHECK: @[[__OMP_OFFLOADING_14_A34CA11_SEQUENTIAL_LOOP_TO_SHARED_VAR_GUARDED_L50_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 2
-; CHECK: @[[__OMP_OFFLOADING_14_A34CA11_DO_NOT_SPMDIZE_TARGET_L65_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 1
-; CHECK: @[[LLVM_COMPILER_USED:[a-zA-Z0-9_$"\\.-]+]] = appending global [5 x i8*] [i8* @__omp_offloading_14_a34ca11_sequential_loop_l5_exec_mode, i8* @__omp_offloading_14_a34ca11_sequential_loop_to_stack_var_l20_exec_mode, i8* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_l35_exec_mode, i8* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_guarded_l50_exec_mode, i8* @__omp_offloading_14_a34ca11_do_not_spmdize_target_l65_exec_mode], section "llvm.metadata"
-; CHECK: @[[X:[a-zA-Z0-9_$"\\.-]+]] = internal addrspace(3) global [4 x i8] undef, align 32
-; CHECK: @[[X1:[a-zA-Z0-9_$"\\.-]+]] = internal addrspace(3) global [4 x i8] undef, align 32
+; NVPTX: @[[GLOB0:[0-9]+]] = private unnamed_addr constant [23 x i8] c"
+; NVPTX: @[[GLOB1:[0-9]+]] = private unnamed_addr constant [[STRUCT_IDENT_T:%.*]] { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* @[[GLOB0]], i32 0, i32 0) }, align 8
+; NVPTX: @[[__OMP_OFFLOADING_14_A34CA11_SEQUENTIAL_LOOP_L5_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 2
+; NVPTX: @[[__OMP_OFFLOADING_14_A34CA11_SEQUENTIAL_LOOP_TO_STACK_VAR_L20_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 2
+; NVPTX: @[[__OMP_OFFLOADING_14_A34CA11_SEQUENTIAL_LOOP_TO_SHARED_VAR_L35_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 2
+; NVPTX: @[[__OMP_OFFLOADING_14_A34CA11_SEQUENTIAL_LOOP_TO_SHARED_VAR_GUARDED_L50_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 2
+; NVPTX: @[[__OMP_OFFLOADING_14_A34CA11_DO_NOT_SPMDIZE_TARGET_L65_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 1
+; NVPTX: @[[LLVM_COMPILER_USED:[a-zA-Z0-9_$"\\.-]+]] = appending global [5 x i8*] [i8* @__omp_offloading_14_a34ca11_sequential_loop_l5_exec_mode, i8* @__omp_offloading_14_a34ca11_sequential_loop_to_stack_var_l20_exec_mode, i8* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_l35_exec_mode, i8* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_guarded_l50_exec_mode, i8* @__omp_offloading_14_a34ca11_do_not_spmdize_target_l65_exec_mode], section "llvm.metadata"
+; NVPTX: @[[X:[a-zA-Z0-9_$"\\.-]+]] = internal addrspace(3) global [4 x i8] undef, align 32
+; NVPTX: @[[X1:[a-zA-Z0-9_$"\\.-]+]] = internal addrspace(3) global [4 x i8] undef, align 32
 ;.
-; CHECK-DISABLED: @[[GLOB0:[0-9]+]] = private unnamed_addr constant [23 x i8] c"
-; CHECK-DISABLED: @[[GLOB1:[0-9]+]] = private unnamed_addr constant [[STRUCT_IDENT_T:%.*]] { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* @[[GLOB0]], i32 0, i32 0) }, align 8
-; CHECK-DISABLED: @[[__OMP_OFFLOADING_14_A34CA11_SEQUENTIAL_LOOP_L5_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 1
-; CHECK-DISABLED: @[[__OMP_OFFLOADING_14_A34CA11_SEQUENTIAL_LOOP_TO_STACK_VAR_L20_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 1
-; CHECK-DISABLED: @[[__OMP_OFFLOADING_14_A34CA11_SEQUENTIAL_LOOP_TO_SHARED_VAR_L35_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 1
-; CHECK-DISABLED: @[[__OMP_OFFLOADING_14_A34CA11_SEQUENTIAL_LOOP_TO_SHARED_VAR_GUARDED_L50_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 1
-; CHECK-DISABLED: @[[__OMP_OFFLOADING_14_A34CA11_DO_NOT_SPMDIZE_TARGET_L65_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 1
-; CHECK-DISABLED: @[[LLVM_COMPILER_USED:[a-zA-Z0-9_$"\\.-]+]] = appending global [5 x i8*] [i8* @__omp_offloading_14_a34ca11_sequential_loop_l5_exec_mode, i8* @__omp_offloading_14_a34ca11_sequential_loop_to_stack_var_l20_exec_mode, i8* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_l35_exec_mode, i8* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_guarded_l50_exec_mode, i8* @__omp_offloading_14_a34ca11_do_not_spmdize_target_l65_exec_mode], section "llvm.metadata"
-; CHECK-DISABLED: @[[X:[a-zA-Z0-9_$"\\.-]+]] = internal addrspace(3) global [4 x i8] undef, align 32
-; CHECK-DISABLED: @[[X1:[a-zA-Z0-9_$"\\.-]+]] = internal addrspace(3) global [4 x i8] undef, align 32
-; CHECK-DISABLED: @[[__OMP_OUTLINED__1_WRAPPER_ID:[a-zA-Z0-9_$"\\.-]+]] = private constant i8 undef
-; CHECK-DISABLED: @[[__OMP_OUTLINED__3_WRAPPER_ID:[a-zA-Z0-9_$"\\.-]+]] = private constant i8 undef
-; CHECK-DISABLED: @[[__OMP_OUTLINED__5_WRAPPER_ID:[a-zA-Z0-9_$"\\.-]+]] = private constant i8 undef
-; CHECK-DISABLED: @[[__OMP_OUTLINED__7_WRAPPER_ID:[a-zA-Z0-9_$"\\.-]+]] = private constant i8 undef
+; AMDGPU-DISABLED: @[[GLOB0:[0-9]+]] = private unnamed_addr constant [23 x i8] c"
+; AMDGPU-DISABLED: @[[GLOB1:[0-9]+]] = private unnamed_addr constant [[STRUCT_IDENT_T:%.*]] { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* @[[GLOB0]], i32 0, i32 0) }, align 8
+; AMDGPU-DISABLED: @[[__OMP_OFFLOADING_14_A34CA11_SEQUENTIAL_LOOP_L5_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 1
+; AMDGPU-DISABLED: @[[__OMP_OFFLOADING_14_A34CA11_SEQUENTIAL_LOOP_TO_STACK_VAR_L20_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 1
+; AMDGPU-DISABLED: @[[__OMP_OFFLOADING_14_A34CA11_SEQUENTIAL_LOOP_TO_SHARED_VAR_L35_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 1
+; AMDGPU-DISABLED: @[[__OMP_OFFLOADING_14_A34CA11_SEQUENTIAL_LOOP_TO_SHARED_VAR_GUARDED_L50_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 1
+; AMDGPU-DISABLED: @[[__OMP_OFFLOADING_14_A34CA11_DO_NOT_SPMDIZE_TARGET_L65_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 1
+; AMDGPU-DISABLED: @[[LLVM_COMPILER_USED:[a-zA-Z0-9_$"\\.-]+]] = appending global [5 x i8*] [i8* @__omp_offloading_14_a34ca11_sequential_loop_l5_exec_mode, i8* @__omp_offloading_14_a34ca11_sequential_loop_to_stack_var_l20_exec_mode, i8* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_l35_exec_mode, i8* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_guarded_l50_exec_mode, i8* @__omp_offloading_14_a34ca11_do_not_spmdize_target_l65_exec_mode], section "llvm.metadata"
+; AMDGPU-DISABLED: @[[X:[a-zA-Z0-9_$"\\.-]+]] = internal addrspace(3) global [4 x i8] undef, align 32
+; AMDGPU-DISABLED: @[[X_1:[a-zA-Z0-9_$"\\.-]+]] = internal addrspace(3) global [4 x i8] undef, align 32
+; AMDGPU-DISABLED: @[[__OMP_OUTLINED__1_WRAPPER_ID:[a-zA-Z0-9_$"\\.-]+]] = private constant i8 undef
+; AMDGPU-DISABLED: @[[__OMP_OUTLINED__3_WRAPPER_ID:[a-zA-Z0-9_$"\\.-]+]] = private constant i8 undef
+; AMDGPU-DISABLED: @[[__OMP_OUTLINED__5_WRAPPER_ID:[a-zA-Z0-9_$"\\.-]+]] = private constant i8 undef
+; AMDGPU-DISABLED: @[[__OMP_OUTLINED__7_WRAPPER_ID:[a-zA-Z0-9_$"\\.-]+]] = private constant i8 undef
+;.
+; NVPTX-DISABLED: @[[GLOB0:[0-9]+]] = private unnamed_addr constant [23 x i8] c"
+; NVPTX-DISABLED: @[[GLOB1:[0-9]+]] = private unnamed_addr constant [[STRUCT_IDENT_T:%.*]] { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* @[[GLOB0]], i32 0, i32 0) }, align 8
+; NVPTX-DISABLED: @[[__OMP_OFFLOADING_14_A34CA11_SEQUENTIAL_LOOP_L5_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 1
+; NVPTX-DISABLED: @[[__OMP_OFFLOADING_14_A34CA11_SEQUENTIAL_LOOP_TO_STACK_VAR_L20_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 1
+; NVPTX-DISABLED: @[[__OMP_OFFLOADING_14_A34CA11_SEQUENTIAL_LOOP_TO_SHARED_VAR_L35_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 1
+; NVPTX-DISABLED: @[[__OMP_OFFLOADING_14_A34CA11_SEQUENTIAL_LOOP_TO_SHARED_VAR_GUARDED_L50_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 1
+; NVPTX-DISABLED: @[[__OMP_OFFLOADING_14_A34CA11_DO_NOT_SPMDIZE_TARGET_L65_EXEC_MODE:[a-zA-Z0-9_$"\\.-]+]] = weak constant i8 1
+; NVPTX-DISABLED: @[[LLVM_COMPILER_USED:[a-zA-Z0-9_$"\\.-]+]] = appending global [5 x i8*] [i8* @__omp_offloading_14_a34ca11_sequential_loop_l5_exec_mode, i8* @__omp_offloading_14_a34ca11_sequential_loop_to_stack_var_l20_exec_mode, i8* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_l35_exec_mode, i8* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_guarded_l50_exec_mode, i8* @__omp_offloading_14_a34ca11_do_not_spmdize_target_l65_exec_mode], section "llvm.metadata"
+; NVPTX-DISABLED: @[[X:[a-zA-Z0-9_$"\\.-]+]] = internal addrspace(3) global [4 x i8] undef, align 32
+; NVPTX-DISABLED: @[[X1:[a-zA-Z0-9_$"\\.-]+]] = internal addrspace(3) global [4 x i8] undef, align 32
+; NVPTX-DISABLED: @[[__OMP_OUTLINED__1_WRAPPER_ID:[a-zA-Z0-9_$"\\.-]+]] = private constant i8 undef
+; NVPTX-DISABLED: @[[__OMP_OUTLINED__3_WRAPPER_ID:[a-zA-Z0-9_$"\\.-]+]] = private constant i8 undef
+; NVPTX-DISABLED: @[[__OMP_OUTLINED__5_WRAPPER_ID:[a-zA-Z0-9_$"\\.-]+]] = private constant i8 undef
+; NVPTX-DISABLED: @[[__OMP_OUTLINED__7_WRAPPER_ID:[a-zA-Z0-9_$"\\.-]+]] = private constant i8 undef
 ;.
 define weak void @__omp_offloading_14_a34ca11_sequential_loop_l5() #0 {
-; CHECK-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_sequential_loop_l5
-; CHECK-SAME: () #[[ATTR0:[0-9]+]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 true, i1 false, 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(%struct.ident_t* @[[GLOB1]]) #[[ATTR3:[0-9]+]]
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
-; CHECK-NEXT:    call void @__omp_outlined__(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 true, i1 false)
-; CHECK-NEXT:    ret void
-; CHECK:       worker.exit:
-; CHECK-NEXT:    ret void
-;
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_sequential_loop_l5
-; CHECK-DISABLED-SAME: () #[[ATTR0:[0-9]+]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
-; CHECK-DISABLED-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
-; CHECK-DISABLED-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
-; CHECK-DISABLED:       worker_state_machine.begin:
-; CHECK-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-DISABLED-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
-; CHECK-DISABLED-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
-; CHECK-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
-; CHECK-DISABLED-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
-; CHECK-DISABLED-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
-; CHECK-DISABLED:       worker_state_machine.finished:
-; CHECK-DISABLED-NEXT:    ret void
-; CHECK-DISABLED:       worker_state_machine.is_active.check:
-; CHECK-DISABLED-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
-; CHECK-DISABLED:       worker_state_machine.parallel_region.check:
-; CHECK-DISABLED-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__1_wrapper.ID to void (i16, i32)*)
-; CHECK-DISABLED-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]]
-; CHECK-DISABLED:       worker_state_machine.parallel_region.execute:
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__1_wrapper(i16 0, i32 [[TMP0]])
-; CHECK-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
-; CHECK-DISABLED:       worker_state_machine.parallel_region.fallback.execute:
-; CHECK-DISABLED-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
-; CHECK-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
-; CHECK-DISABLED:       worker_state_machine.parallel_region.end:
-; CHECK-DISABLED-NEXT:    call void @__kmpc_kernel_end_parallel()
-; CHECK-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
-; CHECK-DISABLED:       worker_state_machine.done.barrier:
-; CHECK-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
-; CHECK-DISABLED:       thread.user_code.check:
-; CHECK-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
-; CHECK-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
-; CHECK-DISABLED:       user_code.entry:
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3:[0-9]+]]
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
-; CHECK-DISABLED-NEXT:    ret void
-; CHECK-DISABLED:       worker.exit:
-; CHECK-DISABLED-NEXT:    ret void
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_sequential_loop_l5
+; AMDGPU-SAME: () #[[ATTR0:[0-9]+]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 true, i1 false, i1 false)
+; AMDGPU-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU:       user_code.entry:
+; AMDGPU-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR4:[0-9]+]]
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; AMDGPU-NEXT:    call void @__omp_outlined__(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR4]]
+; AMDGPU-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 true, i1 false)
+; AMDGPU-NEXT:    ret void
+; AMDGPU:       worker.exit:
+; AMDGPU-NEXT:    ret void
+;
+; NVPTX-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_sequential_loop_l5
+; NVPTX-SAME: () #[[ATTR0:[0-9]+]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 true, i1 false, i1 false)
+; NVPTX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX:       user_code.entry:
+; NVPTX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR4:[0-9]+]]
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; NVPTX-NEXT:    call void @__omp_outlined__(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR4]]
+; NVPTX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 true, i1 false)
+; NVPTX-NEXT:    ret void
+; NVPTX:       worker.exit:
+; NVPTX-NEXT:    ret void
+;
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_sequential_loop_l5
+; AMDGPU-DISABLED-SAME: () #[[ATTR0:[0-9]+]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8, addrspace(5)
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; AMDGPU-DISABLED-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; AMDGPU-DISABLED-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; AMDGPU-DISABLED:       worker_state_machine.begin:
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR_GENERIC:%.*]] = addrspacecast i8* addrspace(5)* [[WORKER_WORK_FN_ADDR]] to i8**
+; AMDGPU-DISABLED-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR_GENERIC]])
+; AMDGPU-DISABLED-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR_GENERIC]], align 8
+; AMDGPU-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; AMDGPU-DISABLED-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; AMDGPU-DISABLED-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; AMDGPU-DISABLED:       worker_state_machine.finished:
+; AMDGPU-DISABLED-NEXT:    ret void
+; AMDGPU-DISABLED:       worker_state_machine.is_active.check:
+; AMDGPU-DISABLED-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; AMDGPU-DISABLED:       worker_state_machine.parallel_region.check:
+; AMDGPU-DISABLED-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__1_wrapper.ID to void (i16, i32)*)
+; AMDGPU-DISABLED-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]]
+; AMDGPU-DISABLED:       worker_state_machine.parallel_region.execute:
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__1_wrapper(i16 0, i32 [[TMP0]])
+; AMDGPU-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; AMDGPU-DISABLED:       worker_state_machine.parallel_region.fallback.execute:
+; AMDGPU-DISABLED-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; AMDGPU-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; AMDGPU-DISABLED:       worker_state_machine.parallel_region.end:
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_kernel_end_parallel()
+; AMDGPU-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; AMDGPU-DISABLED:       worker_state_machine.done.barrier:
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; AMDGPU-DISABLED:       thread.user_code.check:
+; AMDGPU-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU-DISABLED:       user_code.entry:
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR4:[0-9]+]]
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR4]]
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; AMDGPU-DISABLED-NEXT:    ret void
+; AMDGPU-DISABLED:       worker.exit:
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_sequential_loop_l5
+; NVPTX-DISABLED-SAME: () #[[ATTR0:[0-9]+]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; NVPTX-DISABLED-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; NVPTX-DISABLED-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; NVPTX-DISABLED:       worker_state_machine.begin:
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-DISABLED-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
+; NVPTX-DISABLED-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
+; NVPTX-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; NVPTX-DISABLED-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; NVPTX-DISABLED-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; NVPTX-DISABLED:       worker_state_machine.finished:
+; NVPTX-DISABLED-NEXT:    ret void
+; NVPTX-DISABLED:       worker_state_machine.is_active.check:
+; NVPTX-DISABLED-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; NVPTX-DISABLED:       worker_state_machine.parallel_region.check:
+; NVPTX-DISABLED-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__1_wrapper.ID to void (i16, i32)*)
+; NVPTX-DISABLED-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]]
+; NVPTX-DISABLED:       worker_state_machine.parallel_region.execute:
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__1_wrapper(i16 0, i32 [[TMP0]])
+; NVPTX-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; NVPTX-DISABLED:       worker_state_machine.parallel_region.fallback.execute:
+; NVPTX-DISABLED-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; NVPTX-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; NVPTX-DISABLED:       worker_state_machine.parallel_region.end:
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_kernel_end_parallel()
+; NVPTX-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; NVPTX-DISABLED:       worker_state_machine.done.barrier:
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; NVPTX-DISABLED:       thread.user_code.check:
+; NVPTX-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX-DISABLED:       user_code.entry:
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR4:[0-9]+]]
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR4]]
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; NVPTX-DISABLED-NEXT:    ret void
+; NVPTX-DISABLED:       worker.exit:
+; NVPTX-DISABLED-NEXT:    ret void
 ;
 entry:
   %.zero.addr = alloca i32, align 4
@@ -202,61 +295,119 @@ worker.exit:                                      ; preds = %entry
 declare i32 @__kmpc_target_init(%struct.ident_t*, i1, i1, i1)
 
 define internal void @__omp_outlined__(i32* noalias %.global_tid., i32* noalias %.bound_tid.) #0 {
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__
-; CHECK-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[I:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
-; CHECK-NEXT:    store i32 0, i32* [[I]], align 4
-; CHECK-NEXT:    br label [[FOR_COND:%.*]]
-; CHECK:       for.cond:
-; CHECK-NEXT:    [[TMP0:%.*]] = load i32, i32* [[I]], align 4
-; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP0]], 100
-; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
-; CHECK:       for.body:
-; CHECK-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
-; CHECK-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP1]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__1 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__1_wrapper to i8*), i8** noundef [[TMP2]], i64 noundef 0)
-; CHECK-NEXT:    br label [[FOR_INC:%.*]]
-; CHECK:       for.inc:
-; CHECK-NEXT:    [[TMP3:%.*]] = load i32, i32* [[I]], align 4
-; CHECK-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP3]], 1
-; CHECK-NEXT:    store i32 [[INC]], i32* [[I]], align 4
-; CHECK-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP13:![0-9]+]]
-; CHECK:       for.end:
-; CHECK-NEXT:    call void @spmd_amenable() #[[ATTR5:[0-9]+]]
-; CHECK-NEXT:    ret void
-;
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[I:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[I]], align 4
-; CHECK-DISABLED-NEXT:    br label [[FOR_COND:%.*]]
-; CHECK-DISABLED:       for.cond:
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[I]], align 4
-; CHECK-DISABLED-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP0]], 100
-; CHECK-DISABLED-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
-; CHECK-DISABLED:       for.body:
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
-; CHECK-DISABLED-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP1]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__1 to i8*), i8* noundef @__omp_outlined__1_wrapper.ID, i8** noundef [[TMP2]], i64 noundef 0)
-; CHECK-DISABLED-NEXT:    br label [[FOR_INC:%.*]]
-; CHECK-DISABLED:       for.inc:
-; CHECK-DISABLED-NEXT:    [[TMP3:%.*]] = load i32, i32* [[I]], align 4
-; CHECK-DISABLED-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP3]], 1
-; CHECK-DISABLED-NEXT:    store i32 [[INC]], i32* [[I]], align 4
-; CHECK-DISABLED-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP13:![0-9]+]]
-; CHECK-DISABLED:       for.end:
-; CHECK-DISABLED-NEXT:    call void @spmd_amenable() #[[ATTR5:[0-9]+]]
-; CHECK-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__
+; AMDGPU-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[I:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; AMDGPU-NEXT:    store i32 0, i32* [[I]], align 4
+; AMDGPU-NEXT:    br label [[FOR_COND:%.*]]
+; AMDGPU:       for.cond:
+; AMDGPU-NEXT:    [[TMP0:%.*]] = load i32, i32* [[I]], align 4
+; AMDGPU-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP0]], 100
+; AMDGPU-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+; AMDGPU:       for.body:
+; AMDGPU-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; AMDGPU-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP1]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__1 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__1_wrapper to i8*), i8** noundef [[TMP2]], i64 noundef 0)
+; AMDGPU-NEXT:    br label [[FOR_INC:%.*]]
+; AMDGPU:       for.inc:
+; AMDGPU-NEXT:    [[TMP3:%.*]] = load i32, i32* [[I]], align 4
+; AMDGPU-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP3]], 1
+; AMDGPU-NEXT:    store i32 [[INC]], i32* [[I]], align 4
+; AMDGPU-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP13:![0-9]+]]
+; AMDGPU:       for.end:
+; AMDGPU-NEXT:    call void @spmd_amenable() #[[ATTR6:[0-9]+]]
+; AMDGPU-NEXT:    ret void
+;
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__
+; NVPTX-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[I:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; NVPTX-NEXT:    store i32 0, i32* [[I]], align 4
+; NVPTX-NEXT:    br label [[FOR_COND:%.*]]
+; NVPTX:       for.cond:
+; NVPTX-NEXT:    [[TMP0:%.*]] = load i32, i32* [[I]], align 4
+; NVPTX-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP0]], 100
+; NVPTX-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+; NVPTX:       for.body:
+; NVPTX-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; NVPTX-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP1]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__1 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__1_wrapper to i8*), i8** noundef [[TMP2]], i64 noundef 0)
+; NVPTX-NEXT:    br label [[FOR_INC:%.*]]
+; NVPTX:       for.inc:
+; NVPTX-NEXT:    [[TMP3:%.*]] = load i32, i32* [[I]], align 4
+; NVPTX-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP3]], 1
+; NVPTX-NEXT:    store i32 [[INC]], i32* [[I]], align 4
+; NVPTX-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP13:![0-9]+]]
+; NVPTX:       for.end:
+; NVPTX-NEXT:    call void @spmd_amenable() #[[ATTR6:[0-9]+]]
+; NVPTX-NEXT:    ret void
+;
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[I:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[I]], align 4
+; AMDGPU-DISABLED-NEXT:    br label [[FOR_COND:%.*]]
+; AMDGPU-DISABLED:       for.cond:
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[I]], align 4
+; AMDGPU-DISABLED-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP0]], 100
+; AMDGPU-DISABLED-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+; AMDGPU-DISABLED:       for.body:
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; AMDGPU-DISABLED-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP1]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__1 to i8*), i8* noundef @__omp_outlined__1_wrapper.ID, i8** noundef [[TMP2]], i64 noundef 0)
+; AMDGPU-DISABLED-NEXT:    br label [[FOR_INC:%.*]]
+; AMDGPU-DISABLED:       for.inc:
+; AMDGPU-DISABLED-NEXT:    [[TMP3:%.*]] = load i32, i32* [[I]], align 4
+; AMDGPU-DISABLED-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP3]], 1
+; AMDGPU-DISABLED-NEXT:    store i32 [[INC]], i32* [[I]], align 4
+; AMDGPU-DISABLED-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP13:![0-9]+]]
+; AMDGPU-DISABLED:       for.end:
+; AMDGPU-DISABLED-NEXT:    call void @spmd_amenable() #[[ATTR6:[0-9]+]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[I:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[I]], align 4
+; NVPTX-DISABLED-NEXT:    br label [[FOR_COND:%.*]]
+; NVPTX-DISABLED:       for.cond:
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[I]], align 4
+; NVPTX-DISABLED-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP0]], 100
+; NVPTX-DISABLED-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+; NVPTX-DISABLED:       for.body:
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; NVPTX-DISABLED-NEXT:    [[TMP2:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP1]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__1 to i8*), i8* noundef @__omp_outlined__1_wrapper.ID, i8** noundef [[TMP2]], i64 noundef 0)
+; NVPTX-DISABLED-NEXT:    br label [[FOR_INC:%.*]]
+; NVPTX-DISABLED:       for.inc:
+; NVPTX-DISABLED-NEXT:    [[TMP3:%.*]] = load i32, i32* [[I]], align 4
+; NVPTX-DISABLED-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP3]], 1
+; NVPTX-DISABLED-NEXT:    store i32 [[INC]], i32* [[I]], align 4
+; NVPTX-DISABLED-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP13:![0-9]+]]
+; NVPTX-DISABLED:       for.end:
+; NVPTX-DISABLED-NEXT:    call void @spmd_amenable() #[[ATTR6:[0-9]+]]
+; NVPTX-DISABLED-NEXT:    ret void
 ;
 entry:
   %.global_tid..addr = alloca i32*, align 8
@@ -292,21 +443,39 @@ for.end:                                          ; preds = %for.cond
 }
 
 define internal void @__omp_outlined__1(i32* noalias %.global_tid., i32* noalias %.bound_tid.) #0 {
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__1
-; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    call void @unknown() #[[ATTR6:[0-9]+]]
-; CHECK-NEXT:    ret void
-;
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__1
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    call void @unknown() #[[ATTR6:[0-9]+]]
-; CHECK-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__1
+; AMDGPU-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    call void @unknown() #[[ATTR7:[0-9]+]]
+; AMDGPU-NEXT:    ret void
+;
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__1
+; NVPTX-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    call void @unknown() #[[ATTR7:[0-9]+]]
+; NVPTX-NEXT:    ret void
+;
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__1
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    call void @unknown() #[[ATTR7:[0-9]+]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__1
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    call void @unknown() #[[ATTR7:[0-9]+]]
+; NVPTX-DISABLED-NEXT:    ret void
 ;
 entry:
   %.global_tid..addr = alloca i32*, align 8
@@ -320,33 +489,63 @@ entry:
 declare void @unknown() #1
 
 define internal void @__omp_outlined__1_wrapper(i16 zeroext %0, i32 %1) #0 {
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__1_wrapper
-; CHECK-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-NEXT:    call void @__omp_outlined__1(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    ret void
-;
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__1_wrapper
-; CHECK-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__1(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__1_wrapper
+; AMDGPU-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-NEXT:    call void @__omp_outlined__1(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR4]]
+; AMDGPU-NEXT:    ret void
+;
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__1_wrapper
+; NVPTX-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-NEXT:    call void @__omp_outlined__1(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR4]]
+; NVPTX-NEXT:    ret void
+;
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__1_wrapper
+; AMDGPU-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__1(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR4]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__1_wrapper
+; NVPTX-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__1(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR4]]
+; NVPTX-DISABLED-NEXT:    ret void
 ;
 entry:
   %.addr = alloca i16, align 2
@@ -372,70 +571,138 @@ declare i32 @__kmpc_global_thread_num(%struct.ident_t*) #3
 declare void @__kmpc_target_deinit(%struct.ident_t*, i1, i1)
 
 define weak void @__omp_offloading_14_a34ca11_sequential_loop_to_stack_var_l20() #0 {
-; CHECK-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_sequential_loop_to_stack_var_l20
-; CHECK-SAME: () #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 true, i1 false, 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(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
-; CHECK-NEXT:    call void @__omp_outlined__2(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 true, i1 false)
-; CHECK-NEXT:    ret void
-; CHECK:       worker.exit:
-; CHECK-NEXT:    ret void
-;
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_sequential_loop_to_stack_var_l20
-; CHECK-DISABLED-SAME: () #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
-; CHECK-DISABLED-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
-; CHECK-DISABLED-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
-; CHECK-DISABLED:       worker_state_machine.begin:
-; CHECK-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-DISABLED-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
-; CHECK-DISABLED-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
-; CHECK-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
-; CHECK-DISABLED-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
-; CHECK-DISABLED-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
-; CHECK-DISABLED:       worker_state_machine.finished:
-; CHECK-DISABLED-NEXT:    ret void
-; CHECK-DISABLED:       worker_state_machine.is_active.check:
-; CHECK-DISABLED-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
-; CHECK-DISABLED:       worker_state_machine.parallel_region.check:
-; CHECK-DISABLED-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__3_wrapper.ID to void (i16, i32)*)
-; CHECK-DISABLED-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]]
-; CHECK-DISABLED:       worker_state_machine.parallel_region.execute:
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__3_wrapper(i16 0, i32 [[TMP0]])
-; CHECK-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
-; CHECK-DISABLED:       worker_state_machine.parallel_region.fallback.execute:
-; CHECK-DISABLED-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
-; CHECK-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
-; CHECK-DISABLED:       worker_state_machine.parallel_region.end:
-; CHECK-DISABLED-NEXT:    call void @__kmpc_kernel_end_parallel()
-; CHECK-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
-; CHECK-DISABLED:       worker_state_machine.done.barrier:
-; CHECK-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
-; CHECK-DISABLED:       thread.user_code.check:
-; CHECK-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
-; CHECK-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
-; CHECK-DISABLED:       user_code.entry:
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__2(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
-; CHECK-DISABLED-NEXT:    ret void
-; CHECK-DISABLED:       worker.exit:
-; CHECK-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_sequential_loop_to_stack_var_l20
+; AMDGPU-SAME: () #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 true, i1 false, i1 false)
+; AMDGPU-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU:       user_code.entry:
+; AMDGPU-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR4]]
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; AMDGPU-NEXT:    call void @__omp_outlined__2(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR4]]
+; AMDGPU-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 true, i1 false)
+; AMDGPU-NEXT:    ret void
+; AMDGPU:       worker.exit:
+; AMDGPU-NEXT:    ret void
+;
+; NVPTX-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_sequential_loop_to_stack_var_l20
+; NVPTX-SAME: () #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 true, i1 false, i1 false)
+; NVPTX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX:       user_code.entry:
+; NVPTX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR4]]
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; NVPTX-NEXT:    call void @__omp_outlined__2(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR4]]
+; NVPTX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 true, i1 false)
+; NVPTX-NEXT:    ret void
+; NVPTX:       worker.exit:
+; NVPTX-NEXT:    ret void
+;
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_sequential_loop_to_stack_var_l20
+; AMDGPU-DISABLED-SAME: () #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8, addrspace(5)
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; AMDGPU-DISABLED-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; AMDGPU-DISABLED-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; AMDGPU-DISABLED:       worker_state_machine.begin:
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR_GENERIC:%.*]] = addrspacecast i8* addrspace(5)* [[WORKER_WORK_FN_ADDR]] to i8**
+; AMDGPU-DISABLED-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR_GENERIC]])
+; AMDGPU-DISABLED-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR_GENERIC]], align 8
+; AMDGPU-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; AMDGPU-DISABLED-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; AMDGPU-DISABLED-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; AMDGPU-DISABLED:       worker_state_machine.finished:
+; AMDGPU-DISABLED-NEXT:    ret void
+; AMDGPU-DISABLED:       worker_state_machine.is_active.check:
+; AMDGPU-DISABLED-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; AMDGPU-DISABLED:       worker_state_machine.parallel_region.check:
+; AMDGPU-DISABLED-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__3_wrapper.ID to void (i16, i32)*)
+; AMDGPU-DISABLED-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]]
+; AMDGPU-DISABLED:       worker_state_machine.parallel_region.execute:
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__3_wrapper(i16 0, i32 [[TMP0]])
+; AMDGPU-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; AMDGPU-DISABLED:       worker_state_machine.parallel_region.fallback.execute:
+; AMDGPU-DISABLED-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; AMDGPU-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; AMDGPU-DISABLED:       worker_state_machine.parallel_region.end:
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_kernel_end_parallel()
+; AMDGPU-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; AMDGPU-DISABLED:       worker_state_machine.done.barrier:
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; AMDGPU-DISABLED:       thread.user_code.check:
+; AMDGPU-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU-DISABLED:       user_code.entry:
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR4]]
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__2(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR4]]
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; AMDGPU-DISABLED-NEXT:    ret void
+; AMDGPU-DISABLED:       worker.exit:
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_sequential_loop_to_stack_var_l20
+; NVPTX-DISABLED-SAME: () #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; NVPTX-DISABLED-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; NVPTX-DISABLED-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; NVPTX-DISABLED:       worker_state_machine.begin:
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-DISABLED-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
+; NVPTX-DISABLED-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
+; NVPTX-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; NVPTX-DISABLED-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; NVPTX-DISABLED-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; NVPTX-DISABLED:       worker_state_machine.finished:
+; NVPTX-DISABLED-NEXT:    ret void
+; NVPTX-DISABLED:       worker_state_machine.is_active.check:
+; NVPTX-DISABLED-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; NVPTX-DISABLED:       worker_state_machine.parallel_region.check:
+; NVPTX-DISABLED-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__3_wrapper.ID to void (i16, i32)*)
+; NVPTX-DISABLED-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]]
+; NVPTX-DISABLED:       worker_state_machine.parallel_region.execute:
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__3_wrapper(i16 0, i32 [[TMP0]])
+; NVPTX-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; NVPTX-DISABLED:       worker_state_machine.parallel_region.fallback.execute:
+; NVPTX-DISABLED-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; NVPTX-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; NVPTX-DISABLED:       worker_state_machine.parallel_region.end:
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_kernel_end_parallel()
+; NVPTX-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; NVPTX-DISABLED:       worker_state_machine.done.barrier:
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; NVPTX-DISABLED:       thread.user_code.check:
+; NVPTX-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX-DISABLED:       user_code.entry:
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR4]]
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__2(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR4]]
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; NVPTX-DISABLED-NEXT:    ret void
+; NVPTX-DISABLED:       worker.exit:
+; NVPTX-DISABLED-NEXT:    ret void
 ;
 entry:
   %.zero.addr = alloca i32, align 4
@@ -457,67 +724,129 @@ worker.exit:                                      ; preds = %entry
 }
 
 define internal void @__omp_outlined__2(i32* noalias %.global_tid., i32* noalias %.bound_tid.) #0 {
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__2
-; CHECK-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[I:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
-; CHECK-NEXT:    [[TMP0:%.*]] = alloca i8, i64 4, align 1
-; CHECK-NEXT:    [[X_ON_STACK:%.*]] = bitcast i8* [[TMP0]] to i32*
-; CHECK-NEXT:    call void @use(i32* nocapture [[X_ON_STACK]]) #[[ATTR5]]
-; CHECK-NEXT:    store i32 0, i32* [[I]], align 4
-; CHECK-NEXT:    br label [[FOR_COND:%.*]]
-; CHECK:       for.cond:
-; CHECK-NEXT:    [[TMP1:%.*]] = load i32, i32* [[I]], align 4
-; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP1]], 100
-; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
-; CHECK:       for.body:
-; CHECK-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
-; CHECK-NEXT:    [[TMP3:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP2]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__3 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__3_wrapper to i8*), i8** noundef [[TMP3]], i64 noundef 0)
-; CHECK-NEXT:    br label [[FOR_INC:%.*]]
-; CHECK:       for.inc:
-; CHECK-NEXT:    [[TMP4:%.*]] = load i32, i32* [[I]], align 4
-; CHECK-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
-; CHECK-NEXT:    store i32 [[INC]], i32* [[I]], align 4
-; CHECK-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP15:![0-9]+]]
-; CHECK:       for.end:
-; CHECK-NEXT:    call void @spmd_amenable() #[[ATTR5]]
-; CHECK-NEXT:    ret void
-;
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__2
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[I:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
-; CHECK-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = alloca i8, i64 4, align 1
-; CHECK-DISABLED-NEXT:    [[X_ON_STACK:%.*]] = bitcast i8* [[TMP0]] to i32*
-; CHECK-DISABLED-NEXT:    call void @use(i32* nocapture [[X_ON_STACK]]) #[[ATTR5]]
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[I]], align 4
-; CHECK-DISABLED-NEXT:    br label [[FOR_COND:%.*]]
-; CHECK-DISABLED:       for.cond:
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = load i32, i32* [[I]], align 4
-; CHECK-DISABLED-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP1]], 100
-; CHECK-DISABLED-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
-; CHECK-DISABLED:       for.body:
-; CHECK-DISABLED-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
-; CHECK-DISABLED-NEXT:    [[TMP3:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP2]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__3 to i8*), i8* noundef @__omp_outlined__3_wrapper.ID, i8** noundef [[TMP3]], i64 noundef 0)
-; CHECK-DISABLED-NEXT:    br label [[FOR_INC:%.*]]
-; CHECK-DISABLED:       for.inc:
-; CHECK-DISABLED-NEXT:    [[TMP4:%.*]] = load i32, i32* [[I]], align 4
-; CHECK-DISABLED-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
-; CHECK-DISABLED-NEXT:    store i32 [[INC]], i32* [[I]], align 4
-; CHECK-DISABLED-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP15:![0-9]+]]
-; CHECK-DISABLED:       for.end:
-; CHECK-DISABLED-NEXT:    call void @spmd_amenable() #[[ATTR5]]
-; CHECK-DISABLED-NEXT:    ret void
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__2
+; AMDGPU-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[I:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; AMDGPU-NEXT:    [[TMP0:%.*]] = alloca i8, i64 4, align 1
+; AMDGPU-NEXT:    [[X_ON_STACK:%.*]] = bitcast i8* [[TMP0]] to i32*
+; AMDGPU-NEXT:    call void @use(i32* nocapture [[X_ON_STACK]]) #[[ATTR6]]
+; AMDGPU-NEXT:    store i32 0, i32* [[I]], align 4
+; AMDGPU-NEXT:    br label [[FOR_COND:%.*]]
+; AMDGPU:       for.cond:
+; AMDGPU-NEXT:    [[TMP1:%.*]] = load i32, i32* [[I]], align 4
+; AMDGPU-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP1]], 100
+; AMDGPU-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+; AMDGPU:       for.body:
+; AMDGPU-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; AMDGPU-NEXT:    [[TMP3:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP2]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__3 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__3_wrapper to i8*), i8** noundef [[TMP3]], i64 noundef 0)
+; AMDGPU-NEXT:    br label [[FOR_INC:%.*]]
+; AMDGPU:       for.inc:
+; AMDGPU-NEXT:    [[TMP4:%.*]] = load i32, i32* [[I]], align 4
+; AMDGPU-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
+; AMDGPU-NEXT:    store i32 [[INC]], i32* [[I]], align 4
+; AMDGPU-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP15:![0-9]+]]
+; AMDGPU:       for.end:
+; AMDGPU-NEXT:    call void @spmd_amenable() #[[ATTR6]]
+; AMDGPU-NEXT:    ret void
+;
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__2
+; NVPTX-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[I:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; NVPTX-NEXT:    [[TMP0:%.*]] = alloca i8, i64 4, align 1
+; NVPTX-NEXT:    [[X_ON_STACK:%.*]] = bitcast i8* [[TMP0]] to i32*
+; NVPTX-NEXT:    call void @use(i32* nocapture [[X_ON_STACK]]) #[[ATTR6]]
+; NVPTX-NEXT:    store i32 0, i32* [[I]], align 4
+; NVPTX-NEXT:    br label [[FOR_COND:%.*]]
+; NVPTX:       for.cond:
+; NVPTX-NEXT:    [[TMP1:%.*]] = load i32, i32* [[I]], align 4
+; NVPTX-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP1]], 100
+; NVPTX-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+; NVPTX:       for.body:
+; NVPTX-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; NVPTX-NEXT:    [[TMP3:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP2]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__3 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__3_wrapper to i8*), i8** noundef [[TMP3]], i64 noundef 0)
+; NVPTX-NEXT:    br label [[FOR_INC:%.*]]
+; NVPTX:       for.inc:
+; NVPTX-NEXT:    [[TMP4:%.*]] = load i32, i32* [[I]], align 4
+; NVPTX-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
+; NVPTX-NEXT:    store i32 [[INC]], i32* [[I]], align 4
+; NVPTX-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP15:![0-9]+]]
+; NVPTX:       for.end:
+; NVPTX-NEXT:    call void @spmd_amenable() #[[ATTR6]]
+; NVPTX-NEXT:    ret void
+;
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__2
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[I:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; AMDGPU-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = alloca i8, i64 4, align 1
+; AMDGPU-DISABLED-NEXT:    [[X_ON_STACK:%.*]] = bitcast i8* [[TMP0]] to i32*
+; AMDGPU-DISABLED-NEXT:    call void @use(i32* nocapture [[X_ON_STACK]]) #[[ATTR6]]
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[I]], align 4
+; AMDGPU-DISABLED-NEXT:    br label [[FOR_COND:%.*]]
+; AMDGPU-DISABLED:       for.cond:
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = load i32, i32* [[I]], align 4
+; AMDGPU-DISABLED-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP1]], 100
+; AMDGPU-DISABLED-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+; AMDGPU-DISABLED:       for.body:
+; AMDGPU-DISABLED-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; AMDGPU-DISABLED-NEXT:    [[TMP3:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP2]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__3 to i8*), i8* noundef @__omp_outlined__3_wrapper.ID, i8** noundef [[TMP3]], i64 noundef 0)
+; AMDGPU-DISABLED-NEXT:    br label [[FOR_INC:%.*]]
+; AMDGPU-DISABLED:       for.inc:
+; AMDGPU-DISABLED-NEXT:    [[TMP4:%.*]] = load i32, i32* [[I]], align 4
+; AMDGPU-DISABLED-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
+; AMDGPU-DISABLED-NEXT:    store i32 [[INC]], i32* [[I]], align 4
+; AMDGPU-DISABLED-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP15:![0-9]+]]
+; AMDGPU-DISABLED:       for.end:
+; AMDGPU-DISABLED-NEXT:    call void @spmd_amenable() #[[ATTR6]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__2
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[I:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
+; NVPTX-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = alloca i8, i64 4, align 1
+; NVPTX-DISABLED-NEXT:    [[X_ON_STACK:%.*]] = bitcast i8* [[TMP0]] to i32*
+; NVPTX-DISABLED-NEXT:    call void @use(i32* nocapture [[X_ON_STACK]]) #[[ATTR6]]
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[I]], align 4
+; NVPTX-DISABLED-NEXT:    br label [[FOR_COND:%.*]]
+; NVPTX-DISABLED:       for.cond:
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = load i32, i32* [[I]], align 4
+; NVPTX-DISABLED-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP1]], 100
+; NVPTX-DISABLED-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+; NVPTX-DISABLED:       for.body:
+; NVPTX-DISABLED-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; NVPTX-DISABLED-NEXT:    [[TMP3:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP2]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*)* @__omp_outlined__3 to i8*), i8* noundef @__omp_outlined__3_wrapper.ID, i8** noundef [[TMP3]], i64 noundef 0)
+; NVPTX-DISABLED-NEXT:    br label [[FOR_INC:%.*]]
+; NVPTX-DISABLED:       for.inc:
+; NVPTX-DISABLED-NEXT:    [[TMP4:%.*]] = load i32, i32* [[I]], align 4
+; NVPTX-DISABLED-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
+; NVPTX-DISABLED-NEXT:    store i32 [[INC]], i32* [[I]], align 4
+; NVPTX-DISABLED-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP15:![0-9]+]]
+; NVPTX-DISABLED:       for.end:
+; NVPTX-DISABLED-NEXT:    call void @spmd_amenable() #[[ATTR6]]
+; NVPTX-DISABLED-NEXT:    ret void
 ;
 entry:
   %.global_tid..addr = alloca i32*, align 8
@@ -561,21 +890,39 @@ declare i8* @__kmpc_alloc_shared(i64) #3
 declare void @use(i32* nocapture) #2
 
 define internal void @__omp_outlined__3(i32* noalias %.global_tid., i32* noalias %.bound_tid.) #0 {
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__3
-; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    call void @unknown() #[[ATTR6]]
-; CHECK-NEXT:    ret void
-;
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__3
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    call void @unknown() #[[ATTR6]]
-; CHECK-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__3
+; AMDGPU-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    call void @unknown() #[[ATTR7]]
+; AMDGPU-NEXT:    ret void
+;
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__3
+; NVPTX-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    call void @unknown() #[[ATTR7]]
+; NVPTX-NEXT:    ret void
+;
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__3
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    call void @unknown() #[[ATTR7]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__3
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    call void @unknown() #[[ATTR7]]
+; NVPTX-DISABLED-NEXT:    ret void
 ;
 entry:
   %.global_tid..addr = alloca i32*, align 8
@@ -587,33 +934,63 @@ entry:
 }
 
 define internal void @__omp_outlined__3_wrapper(i16 zeroext %0, i32 %1) #0 {
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__3_wrapper
-; CHECK-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-NEXT:    call void @__omp_outlined__3(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    ret void
-;
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__3_wrapper
-; CHECK-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__3(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__3_wrapper
+; AMDGPU-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-NEXT:    call void @__omp_outlined__3(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR4]]
+; AMDGPU-NEXT:    ret void
+;
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__3_wrapper
+; NVPTX-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-NEXT:    call void @__omp_outlined__3(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR4]]
+; NVPTX-NEXT:    ret void
+;
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__3_wrapper
+; AMDGPU-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__3(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR4]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__3_wrapper
+; NVPTX-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__3(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]]) #[[ATTR4]]
+; NVPTX-DISABLED-NEXT:    ret void
 ;
 entry:
   %.addr = alloca i16, align 2
@@ -631,70 +1008,138 @@ entry:
 declare void @__kmpc_free_shared(i8* nocapture, i64) #3
 
 define weak void @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_l35() #0 {
-; CHECK-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_l35
-; CHECK-SAME: () #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 true, i1 false, 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(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
-; CHECK-NEXT:    call void @__omp_outlined__4(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 true, i1 false)
-; CHECK-NEXT:    ret void
-; CHECK:       worker.exit:
-; CHECK-NEXT:    ret void
-;
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_l35
-; CHECK-DISABLED-SAME: () #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
-; CHECK-DISABLED-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
-; CHECK-DISABLED-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
-; CHECK-DISABLED:       worker_state_machine.begin:
-; CHECK-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-DISABLED-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
-; CHECK-DISABLED-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
-; CHECK-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
-; CHECK-DISABLED-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
-; CHECK-DISABLED-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
-; CHECK-DISABLED:       worker_state_machine.finished:
-; CHECK-DISABLED-NEXT:    ret void
-; CHECK-DISABLED:       worker_state_machine.is_active.check:
-; CHECK-DISABLED-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
-; CHECK-DISABLED:       worker_state_machine.parallel_region.check:
-; CHECK-DISABLED-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__5_wrapper.ID to void (i16, i32)*)
-; CHECK-DISABLED-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]]
-; CHECK-DISABLED:       worker_state_machine.parallel_region.execute:
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__5_wrapper(i16 0, i32 [[TMP0]])
-; CHECK-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
-; CHECK-DISABLED:       worker_state_machine.parallel_region.fallback.execute:
-; CHECK-DISABLED-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
-; CHECK-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
-; CHECK-DISABLED:       worker_state_machine.parallel_region.end:
-; CHECK-DISABLED-NEXT:    call void @__kmpc_kernel_end_parallel()
-; CHECK-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
-; CHECK-DISABLED:       worker_state_machine.done.barrier:
-; CHECK-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
-; CHECK-DISABLED:       thread.user_code.check:
-; CHECK-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
-; CHECK-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
-; CHECK-DISABLED:       user_code.entry:
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__4(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
-; CHECK-DISABLED-NEXT:    ret void
-; CHECK-DISABLED:       worker.exit:
-; CHECK-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_l35
+; AMDGPU-SAME: () #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 true, i1 false, i1 false)
+; AMDGPU-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU:       user_code.entry:
+; AMDGPU-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR4]]
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; AMDGPU-NEXT:    call void @__omp_outlined__4(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR4]]
+; AMDGPU-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 true, i1 false)
+; AMDGPU-NEXT:    ret void
+; AMDGPU:       worker.exit:
+; AMDGPU-NEXT:    ret void
+;
+; NVPTX-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_l35
+; NVPTX-SAME: () #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 true, i1 false, i1 false)
+; NVPTX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX:       user_code.entry:
+; NVPTX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR4]]
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; NVPTX-NEXT:    call void @__omp_outlined__4(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR4]]
+; NVPTX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 true, i1 false)
+; NVPTX-NEXT:    ret void
+; NVPTX:       worker.exit:
+; NVPTX-NEXT:    ret void
+;
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_l35
+; AMDGPU-DISABLED-SAME: () #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8, addrspace(5)
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; AMDGPU-DISABLED-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; AMDGPU-DISABLED-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; AMDGPU-DISABLED:       worker_state_machine.begin:
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR_GENERIC:%.*]] = addrspacecast i8* addrspace(5)* [[WORKER_WORK_FN_ADDR]] to i8**
+; AMDGPU-DISABLED-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR_GENERIC]])
+; AMDGPU-DISABLED-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR_GENERIC]], align 8
+; AMDGPU-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; AMDGPU-DISABLED-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; AMDGPU-DISABLED-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; AMDGPU-DISABLED:       worker_state_machine.finished:
+; AMDGPU-DISABLED-NEXT:    ret void
+; AMDGPU-DISABLED:       worker_state_machine.is_active.check:
+; AMDGPU-DISABLED-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; AMDGPU-DISABLED:       worker_state_machine.parallel_region.check:
+; AMDGPU-DISABLED-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__5_wrapper.ID to void (i16, i32)*)
+; AMDGPU-DISABLED-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]]
+; AMDGPU-DISABLED:       worker_state_machine.parallel_region.execute:
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__5_wrapper(i16 0, i32 [[TMP0]])
+; AMDGPU-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; AMDGPU-DISABLED:       worker_state_machine.parallel_region.fallback.execute:
+; AMDGPU-DISABLED-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; AMDGPU-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; AMDGPU-DISABLED:       worker_state_machine.parallel_region.end:
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_kernel_end_parallel()
+; AMDGPU-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; AMDGPU-DISABLED:       worker_state_machine.done.barrier:
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; AMDGPU-DISABLED:       thread.user_code.check:
+; AMDGPU-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU-DISABLED:       user_code.entry:
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR4]]
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__4(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR4]]
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; AMDGPU-DISABLED-NEXT:    ret void
+; AMDGPU-DISABLED:       worker.exit:
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_l35
+; NVPTX-DISABLED-SAME: () #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; NVPTX-DISABLED-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; NVPTX-DISABLED-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; NVPTX-DISABLED:       worker_state_machine.begin:
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-DISABLED-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
+; NVPTX-DISABLED-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
+; NVPTX-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; NVPTX-DISABLED-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; NVPTX-DISABLED-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; NVPTX-DISABLED:       worker_state_machine.finished:
+; NVPTX-DISABLED-NEXT:    ret void
+; NVPTX-DISABLED:       worker_state_machine.is_active.check:
+; NVPTX-DISABLED-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; NVPTX-DISABLED:       worker_state_machine.parallel_region.check:
+; NVPTX-DISABLED-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__5_wrapper.ID to void (i16, i32)*)
+; NVPTX-DISABLED-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]]
+; NVPTX-DISABLED:       worker_state_machine.parallel_region.execute:
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__5_wrapper(i16 0, i32 [[TMP0]])
+; NVPTX-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; NVPTX-DISABLED:       worker_state_machine.parallel_region.fallback.execute:
+; NVPTX-DISABLED-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; NVPTX-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; NVPTX-DISABLED:       worker_state_machine.parallel_region.end:
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_kernel_end_parallel()
+; NVPTX-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; NVPTX-DISABLED:       worker_state_machine.done.barrier:
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; NVPTX-DISABLED:       thread.user_code.check:
+; NVPTX-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX-DISABLED:       user_code.entry:
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR4]]
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__4(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR4]]
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; NVPTX-DISABLED-NEXT:    ret void
+; NVPTX-DISABLED:       worker.exit:
+; NVPTX-DISABLED-NEXT:    ret void
 ;
 entry:
   %.zero.addr = alloca i32, align 4
@@ -716,65 +1161,127 @@ worker.exit:                                      ; preds = %entry
 }
 
 define internal void @__omp_outlined__4(i32* noalias %.global_tid., i32* noalias %.bound_tid.) #0 {
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__4
-; CHECK-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[I:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x i8*], align 8
-; CHECK-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
-; CHECK-NEXT:    store i32 0, i32* [[I]], align 4
-; CHECK-NEXT:    br label [[FOR_COND:%.*]]
-; CHECK:       for.cond:
-; CHECK-NEXT:    [[TMP0:%.*]] = load i32, i32* [[I]], align 4
-; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP0]], 100
-; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
-; CHECK:       for.body:
-; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
-; CHECK-NEXT:    store i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([4 x i8], [4 x i8] addrspace(3)* @x, i32 0, i32 0) to i8*), i8** [[TMP1]], align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
-; CHECK-NEXT:    [[TMP3:%.*]] = bitcast [1 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP2]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*, i32*)* @__omp_outlined__5 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__5_wrapper to i8*), i8** noundef [[TMP3]], i64 noundef 1)
-; CHECK-NEXT:    br label [[FOR_INC:%.*]]
-; CHECK:       for.inc:
-; CHECK-NEXT:    [[TMP4:%.*]] = load i32, i32* [[I]], align 4
-; CHECK-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
-; CHECK-NEXT:    store i32 [[INC]], i32* [[I]], align 4
-; CHECK-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP16:![0-9]+]]
-; CHECK:       for.end:
-; CHECK-NEXT:    call void @spmd_amenable() #[[ATTR5]]
-; CHECK-NEXT:    ret void
-;
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__4
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[I:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x i8*], align 8
-; CHECK-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[I]], align 4
-; CHECK-DISABLED-NEXT:    br label [[FOR_COND:%.*]]
-; CHECK-DISABLED:       for.cond:
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[I]], align 4
-; CHECK-DISABLED-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP0]], 100
-; CHECK-DISABLED-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
-; CHECK-DISABLED:       for.body:
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
-; CHECK-DISABLED-NEXT:    store i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([4 x i8], [4 x i8] addrspace(3)* @x, i32 0, i32 0) to i8*), i8** [[TMP1]], align 8
-; CHECK-DISABLED-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
-; CHECK-DISABLED-NEXT:    [[TMP3:%.*]] = bitcast [1 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP2]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*, i32*)* @__omp_outlined__5 to i8*), i8* noundef @__omp_outlined__5_wrapper.ID, i8** noundef [[TMP3]], i64 noundef 1)
-; CHECK-DISABLED-NEXT:    br label [[FOR_INC:%.*]]
-; CHECK-DISABLED:       for.inc:
-; CHECK-DISABLED-NEXT:    [[TMP4:%.*]] = load i32, i32* [[I]], align 4
-; CHECK-DISABLED-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
-; CHECK-DISABLED-NEXT:    store i32 [[INC]], i32* [[I]], align 4
-; CHECK-DISABLED-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP16:![0-9]+]]
-; CHECK-DISABLED:       for.end:
-; CHECK-DISABLED-NEXT:    call void @spmd_amenable() #[[ATTR5]]
-; CHECK-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__4
+; AMDGPU-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[I:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x i8*], align 8
+; AMDGPU-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; AMDGPU-NEXT:    store i32 0, i32* [[I]], align 4
+; AMDGPU-NEXT:    br label [[FOR_COND:%.*]]
+; AMDGPU:       for.cond:
+; AMDGPU-NEXT:    [[TMP0:%.*]] = load i32, i32* [[I]], align 4
+; AMDGPU-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP0]], 100
+; AMDGPU-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+; AMDGPU:       for.body:
+; AMDGPU-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
+; AMDGPU-NEXT:    store i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([4 x i8], [4 x i8] addrspace(3)* @x, i32 0, i32 0) to i8*), i8** [[TMP1]], align 8
+; AMDGPU-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; AMDGPU-NEXT:    [[TMP3:%.*]] = bitcast [1 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP2]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*, i32*)* @__omp_outlined__5 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__5_wrapper to i8*), i8** noundef [[TMP3]], i64 noundef 1)
+; AMDGPU-NEXT:    br label [[FOR_INC:%.*]]
+; AMDGPU:       for.inc:
+; AMDGPU-NEXT:    [[TMP4:%.*]] = load i32, i32* [[I]], align 4
+; AMDGPU-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
+; AMDGPU-NEXT:    store i32 [[INC]], i32* [[I]], align 4
+; AMDGPU-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP16:![0-9]+]]
+; AMDGPU:       for.end:
+; AMDGPU-NEXT:    call void @spmd_amenable() #[[ATTR6]]
+; AMDGPU-NEXT:    ret void
+;
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__4
+; NVPTX-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[I:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x i8*], align 8
+; NVPTX-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; NVPTX-NEXT:    store i32 0, i32* [[I]], align 4
+; NVPTX-NEXT:    br label [[FOR_COND:%.*]]
+; NVPTX:       for.cond:
+; NVPTX-NEXT:    [[TMP0:%.*]] = load i32, i32* [[I]], align 4
+; NVPTX-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP0]], 100
+; NVPTX-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+; NVPTX:       for.body:
+; NVPTX-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
+; NVPTX-NEXT:    store i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([4 x i8], [4 x i8] addrspace(3)* @x, i32 0, i32 0) to i8*), i8** [[TMP1]], align 8
+; NVPTX-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; NVPTX-NEXT:    [[TMP3:%.*]] = bitcast [1 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP2]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*, i32*)* @__omp_outlined__5 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__5_wrapper to i8*), i8** noundef [[TMP3]], i64 noundef 1)
+; NVPTX-NEXT:    br label [[FOR_INC:%.*]]
+; NVPTX:       for.inc:
+; NVPTX-NEXT:    [[TMP4:%.*]] = load i32, i32* [[I]], align 4
+; NVPTX-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
+; NVPTX-NEXT:    store i32 [[INC]], i32* [[I]], align 4
+; NVPTX-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP16:![0-9]+]]
+; NVPTX:       for.end:
+; NVPTX-NEXT:    call void @spmd_amenable() #[[ATTR6]]
+; NVPTX-NEXT:    ret void
+;
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__4
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[I:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x i8*], align 8
+; AMDGPU-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[I]], align 4
+; AMDGPU-DISABLED-NEXT:    br label [[FOR_COND:%.*]]
+; AMDGPU-DISABLED:       for.cond:
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[I]], align 4
+; AMDGPU-DISABLED-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP0]], 100
+; AMDGPU-DISABLED-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+; AMDGPU-DISABLED:       for.body:
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
+; AMDGPU-DISABLED-NEXT:    store i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([4 x i8], [4 x i8] addrspace(3)* @x, i32 0, i32 0) to i8*), i8** [[TMP1]], align 8
+; AMDGPU-DISABLED-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; AMDGPU-DISABLED-NEXT:    [[TMP3:%.*]] = bitcast [1 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP2]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*, i32*)* @__omp_outlined__5 to i8*), i8* noundef @__omp_outlined__5_wrapper.ID, i8** noundef [[TMP3]], i64 noundef 1)
+; AMDGPU-DISABLED-NEXT:    br label [[FOR_INC:%.*]]
+; AMDGPU-DISABLED:       for.inc:
+; AMDGPU-DISABLED-NEXT:    [[TMP4:%.*]] = load i32, i32* [[I]], align 4
+; AMDGPU-DISABLED-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
+; AMDGPU-DISABLED-NEXT:    store i32 [[INC]], i32* [[I]], align 4
+; AMDGPU-DISABLED-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP16:![0-9]+]]
+; AMDGPU-DISABLED:       for.end:
+; AMDGPU-DISABLED-NEXT:    call void @spmd_amenable() #[[ATTR6]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__4
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[I:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x i8*], align 8
+; NVPTX-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[I]], align 4
+; NVPTX-DISABLED-NEXT:    br label [[FOR_COND:%.*]]
+; NVPTX-DISABLED:       for.cond:
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[I]], align 4
+; NVPTX-DISABLED-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP0]], 100
+; NVPTX-DISABLED-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+; NVPTX-DISABLED:       for.body:
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
+; NVPTX-DISABLED-NEXT:    store i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([4 x i8], [4 x i8] addrspace(3)* @x, i32 0, i32 0) to i8*), i8** [[TMP1]], align 8
+; NVPTX-DISABLED-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; NVPTX-DISABLED-NEXT:    [[TMP3:%.*]] = bitcast [1 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP2]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*, i32*)* @__omp_outlined__5 to i8*), i8* noundef @__omp_outlined__5_wrapper.ID, i8** noundef [[TMP3]], i64 noundef 1)
+; NVPTX-DISABLED-NEXT:    br label [[FOR_INC:%.*]]
+; NVPTX-DISABLED:       for.inc:
+; NVPTX-DISABLED-NEXT:    [[TMP4:%.*]] = load i32, i32* [[I]], align 4
+; NVPTX-DISABLED-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
+; NVPTX-DISABLED-NEXT:    store i32 [[INC]], i32* [[I]], align 4
+; NVPTX-DISABLED-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP16:![0-9]+]]
+; NVPTX-DISABLED:       for.end:
+; NVPTX-DISABLED-NEXT:    call void @spmd_amenable() #[[ATTR6]]
+; NVPTX-DISABLED-NEXT:    ret void
 ;
 entry:
   %.global_tid..addr = alloca i32*, align 8
@@ -816,31 +1323,59 @@ for.end:                                          ; preds = %for.cond
 }
 
 define internal void @__omp_outlined__5(i32* noalias %.global_tid., i32* noalias %.bound_tid., i32* nonnull align 4 dereferenceable(4) %x) #0 {
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__5
-; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]], i32* nocapture nofree nonnull align 4 dereferenceable(4) [[X:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[X_ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    store i32* [[X]], i32** [[X_ADDR]], align 8
-; CHECK-NEXT:    [[TMP0:%.*]] = load i32, i32* [[X]], align 4
-; CHECK-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP0]], 1
-; CHECK-NEXT:    store i32 [[INC]], i32* [[X]], align 4
-; CHECK-NEXT:    call void @unknown() #[[ATTR6]]
-; CHECK-NEXT:    ret void
-;
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__5
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]], i32* nocapture nofree nonnull align 4 dereferenceable(4) [[X:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[X_ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    store i32* [[X]], i32** [[X_ADDR]], align 8
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[X]], align 4
-; CHECK-DISABLED-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP0]], 1
-; CHECK-DISABLED-NEXT:    store i32 [[INC]], i32* [[X]], align 4
-; CHECK-DISABLED-NEXT:    call void @unknown() #[[ATTR6]]
-; CHECK-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__5
+; AMDGPU-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]], i32* nocapture nofree nonnull align 4 dereferenceable(4) [[X:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[X_ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    store i32* [[X]], i32** [[X_ADDR]], align 8
+; AMDGPU-NEXT:    [[TMP0:%.*]] = load i32, i32* [[X]], align 4
+; AMDGPU-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP0]], 1
+; AMDGPU-NEXT:    store i32 [[INC]], i32* [[X]], align 4
+; AMDGPU-NEXT:    call void @unknown() #[[ATTR7]]
+; AMDGPU-NEXT:    ret void
+;
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__5
+; NVPTX-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]], i32* nocapture nofree nonnull align 4 dereferenceable(4) [[X:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[X_ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    store i32* [[X]], i32** [[X_ADDR]], align 8
+; NVPTX-NEXT:    [[TMP0:%.*]] = load i32, i32* [[X]], align 4
+; NVPTX-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP0]], 1
+; NVPTX-NEXT:    store i32 [[INC]], i32* [[X]], align 4
+; NVPTX-NEXT:    call void @unknown() #[[ATTR7]]
+; NVPTX-NEXT:    ret void
+;
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__5
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]], i32* nocapture nofree nonnull align 4 dereferenceable(4) [[X:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[X_ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    store i32* [[X]], i32** [[X_ADDR]], align 8
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[X]], align 4
+; AMDGPU-DISABLED-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP0]], 1
+; AMDGPU-DISABLED-NEXT:    store i32 [[INC]], i32* [[X]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @unknown() #[[ATTR7]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__5
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]], i32* nocapture nofree nonnull align 4 dereferenceable(4) [[X:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[X_ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    store i32* [[X]], i32** [[X_ADDR]], align 8
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[X]], align 4
+; NVPTX-DISABLED-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP0]], 1
+; NVPTX-DISABLED-NEXT:    store i32 [[INC]], i32* [[X]], align 4
+; NVPTX-DISABLED-NEXT:    call void @unknown() #[[ATTR7]]
+; NVPTX-DISABLED-NEXT:    ret void
 ;
 entry:
   %.global_tid..addr = alloca i32*, align 8
@@ -858,41 +1393,79 @@ entry:
 }
 
 define internal void @__omp_outlined__5_wrapper(i16 zeroext %0, i32 %1) #0 {
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__5_wrapper
-; CHECK-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-NEXT:    [[TMP2:%.*]] = load i8**, i8*** [[GLOBAL_ARGS]], align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 0
-; CHECK-NEXT:    [[TMP4:%.*]] = bitcast i8** [[TMP3]] to i32**
-; CHECK-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[TMP4]], align 8
-; CHECK-NEXT:    call void @__omp_outlined__5(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]], i32* [[TMP5]]) #[[ATTR3]]
-; CHECK-NEXT:    ret void
-;
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__5_wrapper
-; CHECK-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-DISABLED-NEXT:    [[TMP2:%.*]] = load i8**, i8*** [[GLOBAL_ARGS]], align 8
-; CHECK-DISABLED-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 0
-; CHECK-DISABLED-NEXT:    [[TMP4:%.*]] = bitcast i8** [[TMP3]] to i32**
-; CHECK-DISABLED-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[TMP4]], align 8
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__5(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]], i32* [[TMP5]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__5_wrapper
+; AMDGPU-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-NEXT:    [[TMP2:%.*]] = load i8**, i8*** [[GLOBAL_ARGS]], align 8
+; AMDGPU-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 0
+; AMDGPU-NEXT:    [[TMP4:%.*]] = bitcast i8** [[TMP3]] to i32**
+; AMDGPU-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[TMP4]], align 8
+; AMDGPU-NEXT:    call void @__omp_outlined__5(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]], i32* [[TMP5]]) #[[ATTR4]]
+; AMDGPU-NEXT:    ret void
+;
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__5_wrapper
+; NVPTX-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-NEXT:    [[TMP2:%.*]] = load i8**, i8*** [[GLOBAL_ARGS]], align 8
+; NVPTX-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 0
+; NVPTX-NEXT:    [[TMP4:%.*]] = bitcast i8** [[TMP3]] to i32**
+; NVPTX-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[TMP4]], align 8
+; NVPTX-NEXT:    call void @__omp_outlined__5(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]], i32* [[TMP5]]) #[[ATTR4]]
+; NVPTX-NEXT:    ret void
+;
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__5_wrapper
+; AMDGPU-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-DISABLED-NEXT:    [[TMP2:%.*]] = load i8**, i8*** [[GLOBAL_ARGS]], align 8
+; AMDGPU-DISABLED-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 0
+; AMDGPU-DISABLED-NEXT:    [[TMP4:%.*]] = bitcast i8** [[TMP3]] to i32**
+; AMDGPU-DISABLED-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[TMP4]], align 8
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__5(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]], i32* [[TMP5]]) #[[ATTR4]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__5_wrapper
+; NVPTX-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-DISABLED-NEXT:    [[TMP2:%.*]] = load i8**, i8*** [[GLOBAL_ARGS]], align 8
+; NVPTX-DISABLED-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 0
+; NVPTX-DISABLED-NEXT:    [[TMP4:%.*]] = bitcast i8** [[TMP3]] to i32**
+; NVPTX-DISABLED-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[TMP4]], align 8
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__5(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]], i32* [[TMP5]]) #[[ATTR4]]
+; NVPTX-DISABLED-NEXT:    ret void
 ;
 entry:
   %.addr = alloca i16, align 2
@@ -912,70 +1485,138 @@ entry:
 }
 
 define weak void @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_guarded_l50() #0 {
-; CHECK-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_guarded_l50
-; CHECK-SAME: () #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 true, i1 false, 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(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
-; CHECK-NEXT:    call void @__omp_outlined__6(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 true, i1 false)
-; CHECK-NEXT:    ret void
-; CHECK:       worker.exit:
-; CHECK-NEXT:    ret void
-;
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_guarded_l50
-; CHECK-DISABLED-SAME: () #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
-; CHECK-DISABLED-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
-; CHECK-DISABLED-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
-; CHECK-DISABLED:       worker_state_machine.begin:
-; CHECK-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-DISABLED-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
-; CHECK-DISABLED-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
-; CHECK-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
-; CHECK-DISABLED-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
-; CHECK-DISABLED-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
-; CHECK-DISABLED:       worker_state_machine.finished:
-; CHECK-DISABLED-NEXT:    ret void
-; CHECK-DISABLED:       worker_state_machine.is_active.check:
-; CHECK-DISABLED-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
-; CHECK-DISABLED:       worker_state_machine.parallel_region.check:
-; CHECK-DISABLED-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__7_wrapper.ID to void (i16, i32)*)
-; CHECK-DISABLED-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]]
-; CHECK-DISABLED:       worker_state_machine.parallel_region.execute:
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__7_wrapper(i16 0, i32 [[TMP0]])
-; CHECK-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
-; CHECK-DISABLED:       worker_state_machine.parallel_region.fallback.execute:
-; CHECK-DISABLED-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
-; CHECK-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
-; CHECK-DISABLED:       worker_state_machine.parallel_region.end:
-; CHECK-DISABLED-NEXT:    call void @__kmpc_kernel_end_parallel()
-; CHECK-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
-; CHECK-DISABLED:       worker_state_machine.done.barrier:
-; CHECK-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
-; CHECK-DISABLED:       thread.user_code.check:
-; CHECK-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
-; CHECK-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
-; CHECK-DISABLED:       user_code.entry:
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__6(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
-; CHECK-DISABLED-NEXT:    ret void
-; CHECK-DISABLED:       worker.exit:
-; CHECK-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_guarded_l50
+; AMDGPU-SAME: () #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 true, i1 false, i1 false)
+; AMDGPU-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU:       user_code.entry:
+; AMDGPU-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR4]]
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; AMDGPU-NEXT:    call void @__omp_outlined__6(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR4]]
+; AMDGPU-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 true, i1 false)
+; AMDGPU-NEXT:    ret void
+; AMDGPU:       worker.exit:
+; AMDGPU-NEXT:    ret void
+;
+; NVPTX-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_guarded_l50
+; NVPTX-SAME: () #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 true, i1 false, i1 false)
+; NVPTX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX:       user_code.entry:
+; NVPTX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR4]]
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; NVPTX-NEXT:    call void @__omp_outlined__6(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR4]]
+; NVPTX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 true, i1 false)
+; NVPTX-NEXT:    ret void
+; NVPTX:       worker.exit:
+; NVPTX-NEXT:    ret void
+;
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_guarded_l50
+; AMDGPU-DISABLED-SAME: () #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8, addrspace(5)
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; AMDGPU-DISABLED-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; AMDGPU-DISABLED-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; AMDGPU-DISABLED:       worker_state_machine.begin:
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR_GENERIC:%.*]] = addrspacecast i8* addrspace(5)* [[WORKER_WORK_FN_ADDR]] to i8**
+; AMDGPU-DISABLED-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR_GENERIC]])
+; AMDGPU-DISABLED-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR_GENERIC]], align 8
+; AMDGPU-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; AMDGPU-DISABLED-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; AMDGPU-DISABLED-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; AMDGPU-DISABLED:       worker_state_machine.finished:
+; AMDGPU-DISABLED-NEXT:    ret void
+; AMDGPU-DISABLED:       worker_state_machine.is_active.check:
+; AMDGPU-DISABLED-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; AMDGPU-DISABLED:       worker_state_machine.parallel_region.check:
+; AMDGPU-DISABLED-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__7_wrapper.ID to void (i16, i32)*)
+; AMDGPU-DISABLED-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]]
+; AMDGPU-DISABLED:       worker_state_machine.parallel_region.execute:
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__7_wrapper(i16 0, i32 [[TMP0]])
+; AMDGPU-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; AMDGPU-DISABLED:       worker_state_machine.parallel_region.fallback.execute:
+; AMDGPU-DISABLED-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; AMDGPU-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; AMDGPU-DISABLED:       worker_state_machine.parallel_region.end:
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_kernel_end_parallel()
+; AMDGPU-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; AMDGPU-DISABLED:       worker_state_machine.done.barrier:
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; AMDGPU-DISABLED:       thread.user_code.check:
+; AMDGPU-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU-DISABLED:       user_code.entry:
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR4]]
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__6(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR4]]
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; AMDGPU-DISABLED-NEXT:    ret void
+; AMDGPU-DISABLED:       worker.exit:
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_guarded_l50
+; NVPTX-DISABLED-SAME: () #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; NVPTX-DISABLED-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; NVPTX-DISABLED-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; NVPTX-DISABLED:       worker_state_machine.begin:
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-DISABLED-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
+; NVPTX-DISABLED-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
+; NVPTX-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; NVPTX-DISABLED-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; NVPTX-DISABLED-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; NVPTX-DISABLED:       worker_state_machine.finished:
+; NVPTX-DISABLED-NEXT:    ret void
+; NVPTX-DISABLED:       worker_state_machine.is_active.check:
+; NVPTX-DISABLED-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; NVPTX-DISABLED:       worker_state_machine.parallel_region.check:
+; NVPTX-DISABLED-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq void (i16, i32)* [[WORKER_WORK_FN_ADDR_CAST]], bitcast (i8* @__omp_outlined__7_wrapper.ID to void (i16, i32)*)
+; NVPTX-DISABLED-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]]
+; NVPTX-DISABLED:       worker_state_machine.parallel_region.execute:
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__7_wrapper(i16 0, i32 [[TMP0]])
+; NVPTX-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; NVPTX-DISABLED:       worker_state_machine.parallel_region.fallback.execute:
+; NVPTX-DISABLED-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; NVPTX-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
+; NVPTX-DISABLED:       worker_state_machine.parallel_region.end:
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_kernel_end_parallel()
+; NVPTX-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; NVPTX-DISABLED:       worker_state_machine.done.barrier:
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; NVPTX-DISABLED:       thread.user_code.check:
+; NVPTX-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX-DISABLED:       user_code.entry:
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR4]]
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTTHREADID_TEMP_]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__6(i32* noalias nocapture noundef nonnull readonly align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR4]]
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; NVPTX-DISABLED-NEXT:    ret void
+; NVPTX-DISABLED:       worker.exit:
+; NVPTX-DISABLED-NEXT:    ret void
 ;
 entry:
   %.zero.addr = alloca i32, align 4
@@ -997,82 +1638,161 @@ worker.exit:                                      ; preds = %entry
 }
 
 define internal void @__omp_outlined__6(i32* noalias %.global_tid., i32* noalias %.bound_tid.) #0 {
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__6
-; CHECK-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[I:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x i8*], align 8
-; CHECK-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
-; CHECK-NEXT:    [[X_ON_STACK:%.*]] = bitcast i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([4 x i8], [4 x i8] addrspace(3)* @x1, i32 0, i32 0) to i8*) to i32*
-; CHECK-NEXT:    br label [[REGION_CHECK_TID:%.*]]
-; CHECK:       region.check.tid:
-; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
-; CHECK-NEXT:    [[TMP1:%.*]] = icmp eq i32 [[TMP0]], 0
-; CHECK-NEXT:    br i1 [[TMP1]], label [[REGION_GUARDED:%.*]], label [[REGION_BARRIER:%.*]]
-; CHECK:       region.guarded:
-; CHECK-NEXT:    store i32 42, i32* [[X_ON_STACK]], align 4
-; CHECK-NEXT:    br label [[REGION_GUARDED_END:%.*]]
-; CHECK:       region.guarded.end:
-; CHECK-NEXT:    br label [[REGION_BARRIER]]
-; CHECK:       region.barrier:
-; CHECK-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-NEXT:    br label [[REGION_EXIT:%.*]]
-; CHECK:       region.exit:
-; CHECK-NEXT:    store i32 0, i32* [[I]], align 4
-; CHECK-NEXT:    br label [[FOR_COND:%.*]]
-; CHECK:       for.cond:
-; CHECK-NEXT:    [[TMP2:%.*]] = load i32, i32* [[I]], align 4
-; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP2]], 100
-; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
-; CHECK:       for.body:
-; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
-; CHECK-NEXT:    store i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([4 x i8], [4 x i8] addrspace(3)* @x1, i32 0, i32 0) to i8*), i8** [[TMP3]], align 8
-; CHECK-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
-; CHECK-NEXT:    [[TMP5:%.*]] = bitcast [1 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP4]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*, i32*)* @__omp_outlined__7 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__7_wrapper to i8*), i8** noundef [[TMP5]], i64 noundef 1)
-; CHECK-NEXT:    br label [[FOR_INC:%.*]]
-; CHECK:       for.inc:
-; CHECK-NEXT:    [[TMP6:%.*]] = load i32, i32* [[I]], align 4
-; CHECK-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP6]], 1
-; CHECK-NEXT:    store i32 [[INC]], i32* [[I]], align 4
-; CHECK-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP17:![0-9]+]]
-; CHECK:       for.end:
-; CHECK-NEXT:    call void @spmd_amenable() #[[ATTR5]]
-; CHECK-NEXT:    ret void
-;
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__6
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[I:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x i8*], align 8
-; CHECK-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
-; CHECK-DISABLED-NEXT:    [[X_ON_STACK:%.*]] = bitcast i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([4 x i8], [4 x i8] addrspace(3)* @x1, i32 0, i32 0) to i8*) to i32*
-; CHECK-DISABLED-NEXT:    store i32 42, i32* [[X_ON_STACK]], align 4
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[I]], align 4
-; CHECK-DISABLED-NEXT:    br label [[FOR_COND:%.*]]
-; CHECK-DISABLED:       for.cond:
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[I]], align 4
-; CHECK-DISABLED-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP0]], 100
-; CHECK-DISABLED-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
-; CHECK-DISABLED:       for.body:
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
-; CHECK-DISABLED-NEXT:    store i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([4 x i8], [4 x i8] addrspace(3)* @x1, i32 0, i32 0) to i8*), i8** [[TMP1]], align 8
-; CHECK-DISABLED-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
-; CHECK-DISABLED-NEXT:    [[TMP3:%.*]] = bitcast [1 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
-; CHECK-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP2]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*, i32*)* @__omp_outlined__7 to i8*), i8* noundef @__omp_outlined__7_wrapper.ID, i8** noundef [[TMP3]], i64 noundef 1)
-; CHECK-DISABLED-NEXT:    br label [[FOR_INC:%.*]]
-; CHECK-DISABLED:       for.inc:
-; CHECK-DISABLED-NEXT:    [[TMP4:%.*]] = load i32, i32* [[I]], align 4
-; CHECK-DISABLED-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
-; CHECK-DISABLED-NEXT:    store i32 [[INC]], i32* [[I]], align 4
-; CHECK-DISABLED-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP17:![0-9]+]]
-; CHECK-DISABLED:       for.end:
-; CHECK-DISABLED-NEXT:    call void @spmd_amenable() #[[ATTR5]]
-; CHECK-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__6
+; AMDGPU-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[I:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x i8*], align 8
+; AMDGPU-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; AMDGPU-NEXT:    [[X_ON_STACK:%.*]] = bitcast i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([4 x i8], [4 x i8] addrspace(3)* @x.1, i32 0, i32 0) to i8*) to i32*
+; AMDGPU-NEXT:    br label [[REGION_CHECK_TID:%.*]]
+; AMDGPU:       region.check.tid:
+; AMDGPU-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
+; AMDGPU-NEXT:    [[TMP1:%.*]] = icmp eq i32 [[TMP0]], 0
+; AMDGPU-NEXT:    br i1 [[TMP1]], label [[REGION_GUARDED:%.*]], label [[REGION_BARRIER:%.*]]
+; AMDGPU:       region.guarded:
+; AMDGPU-NEXT:    store i32 42, i32* [[X_ON_STACK]], align 4
+; AMDGPU-NEXT:    br label [[REGION_GUARDED_END:%.*]]
+; AMDGPU:       region.guarded.end:
+; AMDGPU-NEXT:    br label [[REGION_BARRIER]]
+; AMDGPU:       region.barrier:
+; AMDGPU-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[REGION_EXIT:%.*]]
+; AMDGPU:       region.exit:
+; AMDGPU-NEXT:    store i32 0, i32* [[I]], align 4
+; AMDGPU-NEXT:    br label [[FOR_COND:%.*]]
+; AMDGPU:       for.cond:
+; AMDGPU-NEXT:    [[TMP2:%.*]] = load i32, i32* [[I]], align 4
+; AMDGPU-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP2]], 100
+; AMDGPU-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+; AMDGPU:       for.body:
+; AMDGPU-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
+; AMDGPU-NEXT:    store i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([4 x i8], [4 x i8] addrspace(3)* @x.1, i32 0, i32 0) to i8*), i8** [[TMP3]], align 8
+; AMDGPU-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; AMDGPU-NEXT:    [[TMP5:%.*]] = bitcast [1 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP4]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*, i32*)* @__omp_outlined__7 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__7_wrapper to i8*), i8** noundef [[TMP5]], i64 noundef 1)
+; AMDGPU-NEXT:    br label [[FOR_INC:%.*]]
+; AMDGPU:       for.inc:
+; AMDGPU-NEXT:    [[TMP6:%.*]] = load i32, i32* [[I]], align 4
+; AMDGPU-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP6]], 1
+; AMDGPU-NEXT:    store i32 [[INC]], i32* [[I]], align 4
+; AMDGPU-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP17:![0-9]+]]
+; AMDGPU:       for.end:
+; AMDGPU-NEXT:    call void @spmd_amenable() #[[ATTR6]]
+; AMDGPU-NEXT:    ret void
+;
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__6
+; NVPTX-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[I:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x i8*], align 8
+; NVPTX-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; NVPTX-NEXT:    [[X_ON_STACK:%.*]] = bitcast i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([4 x i8], [4 x i8] addrspace(3)* @x1, i32 0, i32 0) to i8*) to i32*
+; NVPTX-NEXT:    br label [[REGION_CHECK_TID:%.*]]
+; NVPTX:       region.check.tid:
+; NVPTX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
+; NVPTX-NEXT:    [[TMP1:%.*]] = icmp eq i32 [[TMP0]], 0
+; NVPTX-NEXT:    br i1 [[TMP1]], label [[REGION_GUARDED:%.*]], label [[REGION_BARRIER:%.*]]
+; NVPTX:       region.guarded:
+; NVPTX-NEXT:    store i32 42, i32* [[X_ON_STACK]], align 4
+; NVPTX-NEXT:    br label [[REGION_GUARDED_END:%.*]]
+; NVPTX:       region.guarded.end:
+; NVPTX-NEXT:    br label [[REGION_BARRIER]]
+; NVPTX:       region.barrier:
+; NVPTX-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[REGION_EXIT:%.*]]
+; NVPTX:       region.exit:
+; NVPTX-NEXT:    store i32 0, i32* [[I]], align 4
+; NVPTX-NEXT:    br label [[FOR_COND:%.*]]
+; NVPTX:       for.cond:
+; NVPTX-NEXT:    [[TMP2:%.*]] = load i32, i32* [[I]], align 4
+; NVPTX-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP2]], 100
+; NVPTX-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+; NVPTX:       for.body:
+; NVPTX-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
+; NVPTX-NEXT:    store i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([4 x i8], [4 x i8] addrspace(3)* @x1, i32 0, i32 0) to i8*), i8** [[TMP3]], align 8
+; NVPTX-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; NVPTX-NEXT:    [[TMP5:%.*]] = bitcast [1 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP4]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*, i32*)* @__omp_outlined__7 to i8*), i8* noundef bitcast (void (i16, i32)* @__omp_outlined__7_wrapper to i8*), i8** noundef [[TMP5]], i64 noundef 1)
+; NVPTX-NEXT:    br label [[FOR_INC:%.*]]
+; NVPTX:       for.inc:
+; NVPTX-NEXT:    [[TMP6:%.*]] = load i32, i32* [[I]], align 4
+; NVPTX-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP6]], 1
+; NVPTX-NEXT:    store i32 [[INC]], i32* [[I]], align 4
+; NVPTX-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP17:![0-9]+]]
+; NVPTX:       for.end:
+; NVPTX-NEXT:    call void @spmd_amenable() #[[ATTR6]]
+; NVPTX-NEXT:    ret void
+;
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__6
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[I:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x i8*], align 8
+; AMDGPU-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; AMDGPU-DISABLED-NEXT:    [[X_ON_STACK:%.*]] = bitcast i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([4 x i8], [4 x i8] addrspace(3)* @x.1, i32 0, i32 0) to i8*) to i32*
+; AMDGPU-DISABLED-NEXT:    store i32 42, i32* [[X_ON_STACK]], align 4
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[I]], align 4
+; AMDGPU-DISABLED-NEXT:    br label [[FOR_COND:%.*]]
+; AMDGPU-DISABLED:       for.cond:
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[I]], align 4
+; AMDGPU-DISABLED-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP0]], 100
+; AMDGPU-DISABLED-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+; AMDGPU-DISABLED:       for.body:
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
+; AMDGPU-DISABLED-NEXT:    store i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([4 x i8], [4 x i8] addrspace(3)* @x.1, i32 0, i32 0) to i8*), i8** [[TMP1]], align 8
+; AMDGPU-DISABLED-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; AMDGPU-DISABLED-NEXT:    [[TMP3:%.*]] = bitcast [1 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP2]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*, i32*)* @__omp_outlined__7 to i8*), i8* noundef @__omp_outlined__7_wrapper.ID, i8** noundef [[TMP3]], i64 noundef 1)
+; AMDGPU-DISABLED-NEXT:    br label [[FOR_INC:%.*]]
+; AMDGPU-DISABLED:       for.inc:
+; AMDGPU-DISABLED-NEXT:    [[TMP4:%.*]] = load i32, i32* [[I]], align 4
+; AMDGPU-DISABLED-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
+; AMDGPU-DISABLED-NEXT:    store i32 [[INC]], i32* [[I]], align 4
+; AMDGPU-DISABLED-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP17:![0-9]+]]
+; AMDGPU-DISABLED:       for.end:
+; AMDGPU-DISABLED-NEXT:    call void @spmd_amenable() #[[ATTR6]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__6
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[I:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x i8*], align 8
+; NVPTX-DISABLED-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
+; NVPTX-DISABLED-NEXT:    [[X_ON_STACK:%.*]] = bitcast i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([4 x i8], [4 x i8] addrspace(3)* @x1, i32 0, i32 0) to i8*) to i32*
+; NVPTX-DISABLED-NEXT:    store i32 42, i32* [[X_ON_STACK]], align 4
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[I]], align 4
+; NVPTX-DISABLED-NEXT:    br label [[FOR_COND:%.*]]
+; NVPTX-DISABLED:       for.cond:
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[I]], align 4
+; NVPTX-DISABLED-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP0]], 100
+; NVPTX-DISABLED-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+; NVPTX-DISABLED:       for.body:
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
+; NVPTX-DISABLED-NEXT:    store i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([4 x i8], [4 x i8] addrspace(3)* @x1, i32 0, i32 0) to i8*), i8** [[TMP1]], align 8
+; NVPTX-DISABLED-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTGLOBAL_TID_]], align 4
+; NVPTX-DISABLED-NEXT:    [[TMP3:%.*]] = bitcast [1 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* noundef @[[GLOB1]], i32 [[TMP2]], i32 noundef 1, i32 noundef -1, i32 noundef -1, i8* noundef bitcast (void (i32*, i32*, i32*)* @__omp_outlined__7 to i8*), i8* noundef @__omp_outlined__7_wrapper.ID, i8** noundef [[TMP3]], i64 noundef 1)
+; NVPTX-DISABLED-NEXT:    br label [[FOR_INC:%.*]]
+; NVPTX-DISABLED:       for.inc:
+; NVPTX-DISABLED-NEXT:    [[TMP4:%.*]] = load i32, i32* [[I]], align 4
+; NVPTX-DISABLED-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
+; NVPTX-DISABLED-NEXT:    store i32 [[INC]], i32* [[I]], align 4
+; NVPTX-DISABLED-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP17:![0-9]+]]
+; NVPTX-DISABLED:       for.end:
+; NVPTX-DISABLED-NEXT:    call void @spmd_amenable() #[[ATTR6]]
+; NVPTX-DISABLED-NEXT:    ret void
 ;
 entry:
   %.global_tid..addr = alloca i32*, align 8
@@ -1115,31 +1835,59 @@ for.end:                                          ; preds = %for.cond
 }
 
 define internal void @__omp_outlined__7(i32* noalias %.global_tid., i32* noalias %.bound_tid., i32* nonnull align 4 dereferenceable(4) %x) #0 {
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__7
-; CHECK-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]], i32* nocapture nofree nonnull align 4 dereferenceable(4) [[X:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[X_ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    store i32* [[X]], i32** [[X_ADDR]], align 8
-; CHECK-NEXT:    [[TMP0:%.*]] = load i32, i32* [[X]], align 4
-; CHECK-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP0]], 1
-; CHECK-NEXT:    store i32 [[INC]], i32* [[X]], align 4
-; CHECK-NEXT:    call void @unknown() #[[ATTR6]]
-; CHECK-NEXT:    ret void
-;
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__7
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]], i32* nocapture nofree nonnull align 4 dereferenceable(4) [[X:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[X_ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    store i32* [[X]], i32** [[X_ADDR]], align 8
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[X]], align 4
-; CHECK-DISABLED-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP0]], 1
-; CHECK-DISABLED-NEXT:    store i32 [[INC]], i32* [[X]], align 4
-; CHECK-DISABLED-NEXT:    call void @unknown() #[[ATTR6]]
-; CHECK-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__7
+; AMDGPU-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]], i32* nocapture nofree nonnull align 4 dereferenceable(4) [[X:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[X_ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    store i32* [[X]], i32** [[X_ADDR]], align 8
+; AMDGPU-NEXT:    [[TMP0:%.*]] = load i32, i32* [[X]], align 4
+; AMDGPU-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP0]], 1
+; AMDGPU-NEXT:    store i32 [[INC]], i32* [[X]], align 4
+; AMDGPU-NEXT:    call void @unknown() #[[ATTR7]]
+; AMDGPU-NEXT:    ret void
+;
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__7
+; NVPTX-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]], i32* nocapture nofree nonnull align 4 dereferenceable(4) [[X:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[X_ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    store i32* [[X]], i32** [[X_ADDR]], align 8
+; NVPTX-NEXT:    [[TMP0:%.*]] = load i32, i32* [[X]], align 4
+; NVPTX-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP0]], 1
+; NVPTX-NEXT:    store i32 [[INC]], i32* [[X]], align 4
+; NVPTX-NEXT:    call void @unknown() #[[ATTR7]]
+; NVPTX-NEXT:    ret void
+;
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__7
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]], i32* nocapture nofree nonnull align 4 dereferenceable(4) [[X:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[X_ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    store i32* [[X]], i32** [[X_ADDR]], align 8
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[X]], align 4
+; AMDGPU-DISABLED-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP0]], 1
+; AMDGPU-DISABLED-NEXT:    store i32 [[INC]], i32* [[X]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @unknown() #[[ATTR7]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__7
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree readnone [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree readnone [[DOTBOUND_TID_:%.*]], i32* nocapture nofree nonnull align 4 dereferenceable(4) [[X:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[X_ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    store i32* [[X]], i32** [[X_ADDR]], align 8
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[X]], align 4
+; NVPTX-DISABLED-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP0]], 1
+; NVPTX-DISABLED-NEXT:    store i32 [[INC]], i32* [[X]], align 4
+; NVPTX-DISABLED-NEXT:    call void @unknown() #[[ATTR7]]
+; NVPTX-DISABLED-NEXT:    ret void
 ;
 entry:
   %.global_tid..addr = alloca i32*, align 8
@@ -1157,41 +1905,79 @@ entry:
 }
 
 define internal void @__omp_outlined__7_wrapper(i16 zeroext %0, i32 %1) #0 {
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__7_wrapper
-; CHECK-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-NEXT:    [[TMP2:%.*]] = load i8**, i8*** [[GLOBAL_ARGS]], align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 0
-; CHECK-NEXT:    [[TMP4:%.*]] = bitcast i8** [[TMP3]] to i32**
-; CHECK-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[TMP4]], align 8
-; CHECK-NEXT:    call void @__omp_outlined__7(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]], i32* [[TMP5]]) #[[ATTR3]]
-; CHECK-NEXT:    ret void
-;
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__7_wrapper
-; CHECK-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
-; CHECK-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
-; CHECK-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
-; CHECK-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
-; CHECK-DISABLED-NEXT:    [[TMP2:%.*]] = load i8**, i8*** [[GLOBAL_ARGS]], align 8
-; CHECK-DISABLED-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 0
-; CHECK-DISABLED-NEXT:    [[TMP4:%.*]] = bitcast i8** [[TMP3]] to i32**
-; CHECK-DISABLED-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[TMP4]], align 8
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__7(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]], i32* [[TMP5]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__7_wrapper
+; AMDGPU-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-NEXT:    [[TMP2:%.*]] = load i8**, i8*** [[GLOBAL_ARGS]], align 8
+; AMDGPU-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 0
+; AMDGPU-NEXT:    [[TMP4:%.*]] = bitcast i8** [[TMP3]] to i32**
+; AMDGPU-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[TMP4]], align 8
+; AMDGPU-NEXT:    call void @__omp_outlined__7(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]], i32* [[TMP5]]) #[[ATTR4]]
+; AMDGPU-NEXT:    ret void
+;
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__7_wrapper
+; NVPTX-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-NEXT:    [[TMP2:%.*]] = load i8**, i8*** [[GLOBAL_ARGS]], align 8
+; NVPTX-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 0
+; NVPTX-NEXT:    [[TMP4:%.*]] = bitcast i8** [[TMP3]] to i32**
+; NVPTX-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[TMP4]], align 8
+; NVPTX-NEXT:    call void @__omp_outlined__7(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]], i32* [[TMP5]]) #[[ATTR4]]
+; NVPTX-NEXT:    ret void
+;
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__7_wrapper
+; AMDGPU-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; AMDGPU-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; AMDGPU-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; AMDGPU-DISABLED-NEXT:    [[TMP2:%.*]] = load i8**, i8*** [[GLOBAL_ARGS]], align 8
+; AMDGPU-DISABLED-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 0
+; AMDGPU-DISABLED-NEXT:    [[TMP4:%.*]] = bitcast i8** [[TMP3]] to i32**
+; AMDGPU-DISABLED-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[TMP4]], align 8
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__7(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]], i32* [[TMP5]]) #[[ATTR4]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__7_wrapper
+; NVPTX-DISABLED-SAME: (i16 zeroext [[TMP0:%.*]], i32 [[TMP1:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTADDR:%.*]] = alloca i16, align 2
+; NVPTX-DISABLED-NEXT:    [[DOTADDR1:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[GLOBAL_ARGS:%.*]] = alloca i8**, align 8
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    store i16 [[TMP0]], i16* [[DOTADDR]], align 2
+; NVPTX-DISABLED-NEXT:    store i32 [[TMP1]], i32* [[DOTADDR1]], align 4
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_get_shared_variables(i8*** [[GLOBAL_ARGS]])
+; NVPTX-DISABLED-NEXT:    [[TMP2:%.*]] = load i8**, i8*** [[GLOBAL_ARGS]], align 8
+; NVPTX-DISABLED-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8*, i8** [[TMP2]], i64 0
+; NVPTX-DISABLED-NEXT:    [[TMP4:%.*]] = bitcast i8** [[TMP3]] to i32**
+; NVPTX-DISABLED-NEXT:    [[TMP5:%.*]] = load i32*, i32** [[TMP4]], align 8
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__7(i32* [[DOTADDR1]], i32* [[DOTZERO_ADDR]], i32* [[TMP5]]) #[[ATTR4]]
+; NVPTX-DISABLED-NEXT:    ret void
 ;
 entry:
   %.addr = alloca i16, align 2
@@ -1211,86 +1997,171 @@ entry:
 }
 
 define weak void @__omp_offloading_14_a34ca11_do_not_spmdize_target_l65() #0 {
-; CHECK-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_do_not_spmdize_target_l65
-; CHECK-SAME: () #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
-; CHECK-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
-; CHECK-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
-; CHECK-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
-; CHECK:       worker_state_machine.begin:
-; CHECK-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
-; CHECK-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
-; CHECK-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
-; CHECK-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
-; CHECK-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
-; CHECK:       worker_state_machine.finished:
-; CHECK-NEXT:    ret void
-; CHECK:       worker_state_machine.is_active.check:
-; CHECK-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
-; CHECK:       worker_state_machine.parallel_region.fallback.execute:
-; CHECK-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
-; CHECK:       worker_state_machine.parallel_region.end:
-; CHECK-NEXT:    call void @__kmpc_kernel_end_parallel()
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
-; CHECK:       worker_state_machine.done.barrier:
-; CHECK-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
-; CHECK:       thread.user_code.check:
-; 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(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
-; CHECK-NEXT:    call void @__omp_outlined__8(i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
-; CHECK-NEXT:    ret void
-; CHECK:       worker.exit:
-; CHECK-NEXT:    ret void
-;
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_do_not_spmdize_target_l65
-; CHECK-DISABLED-SAME: () #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
-; CHECK-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
-; CHECK-DISABLED-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
-; CHECK-DISABLED-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
-; CHECK-DISABLED:       worker_state_machine.begin:
-; CHECK-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-DISABLED-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
-; CHECK-DISABLED-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
-; CHECK-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
-; CHECK-DISABLED-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
-; CHECK-DISABLED-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
-; CHECK-DISABLED:       worker_state_machine.finished:
-; CHECK-DISABLED-NEXT:    ret void
-; CHECK-DISABLED:       worker_state_machine.is_active.check:
-; CHECK-DISABLED-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
-; CHECK-DISABLED:       worker_state_machine.parallel_region.fallback.execute:
-; CHECK-DISABLED-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
-; CHECK-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
-; CHECK-DISABLED:       worker_state_machine.parallel_region.end:
-; CHECK-DISABLED-NEXT:    call void @__kmpc_kernel_end_parallel()
-; CHECK-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
-; CHECK-DISABLED:       worker_state_machine.done.barrier:
-; CHECK-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
-; CHECK-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
-; CHECK-DISABLED:       thread.user_code.check:
-; CHECK-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
-; CHECK-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
-; CHECK-DISABLED:       user_code.entry:
-; CHECK-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    call void @__omp_outlined__8(i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR3]]
-; CHECK-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
-; CHECK-DISABLED-NEXT:    ret void
-; CHECK-DISABLED:       worker.exit:
-; CHECK-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_do_not_spmdize_target_l65
+; AMDGPU-SAME: () #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8, addrspace(5)
+; AMDGPU-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; AMDGPU-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; AMDGPU-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; AMDGPU:       worker_state_machine.begin:
+; AMDGPU-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR_GENERIC:%.*]] = addrspacecast i8* addrspace(5)* [[WORKER_WORK_FN_ADDR]] to i8**
+; AMDGPU-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR_GENERIC]])
+; AMDGPU-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR_GENERIC]], align 8
+; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; AMDGPU-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; AMDGPU-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; AMDGPU:       worker_state_machine.finished:
+; AMDGPU-NEXT:    ret void
+; AMDGPU:       worker_state_machine.is_active.check:
+; AMDGPU-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.fallback.execute:
+; AMDGPU-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; AMDGPU:       worker_state_machine.parallel_region.end:
+; AMDGPU-NEXT:    call void @__kmpc_kernel_end_parallel()
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; AMDGPU:       worker_state_machine.done.barrier:
+; AMDGPU-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; AMDGPU:       thread.user_code.check:
+; AMDGPU-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU:       user_code.entry:
+; AMDGPU-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR4]]
+; AMDGPU-NEXT:    call void @__omp_outlined__8(i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR4]]
+; AMDGPU-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; AMDGPU-NEXT:    ret void
+; AMDGPU:       worker.exit:
+; AMDGPU-NEXT:    ret void
+;
+; NVPTX-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_do_not_spmdize_target_l65
+; NVPTX-SAME: () #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
+; NVPTX-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; NVPTX-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; NVPTX-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; NVPTX:       worker_state_machine.begin:
+; NVPTX-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
+; NVPTX-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
+; NVPTX-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; NVPTX-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; NVPTX-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; NVPTX:       worker_state_machine.finished:
+; NVPTX-NEXT:    ret void
+; NVPTX:       worker_state_machine.is_active.check:
+; NVPTX-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.fallback.execute:
+; NVPTX-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; NVPTX:       worker_state_machine.parallel_region.end:
+; NVPTX-NEXT:    call void @__kmpc_kernel_end_parallel()
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; NVPTX:       worker_state_machine.done.barrier:
+; NVPTX-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; NVPTX:       thread.user_code.check:
+; NVPTX-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX:       user_code.entry:
+; NVPTX-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR4]]
+; NVPTX-NEXT:    call void @__omp_outlined__8(i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR4]]
+; NVPTX-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; NVPTX-NEXT:    ret void
+; NVPTX:       worker.exit:
+; NVPTX-NEXT:    ret void
+;
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_do_not_spmdize_target_l65
+; AMDGPU-DISABLED-SAME: () #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8, addrspace(5)
+; AMDGPU-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; AMDGPU-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; AMDGPU-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; AMDGPU-DISABLED-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; AMDGPU-DISABLED-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; AMDGPU-DISABLED:       worker_state_machine.begin:
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR_GENERIC:%.*]] = addrspacecast i8* addrspace(5)* [[WORKER_WORK_FN_ADDR]] to i8**
+; AMDGPU-DISABLED-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR_GENERIC]])
+; AMDGPU-DISABLED-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR_GENERIC]], align 8
+; AMDGPU-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; AMDGPU-DISABLED-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; AMDGPU-DISABLED-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; AMDGPU-DISABLED:       worker_state_machine.finished:
+; AMDGPU-DISABLED-NEXT:    ret void
+; AMDGPU-DISABLED:       worker_state_machine.is_active.check:
+; AMDGPU-DISABLED-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; AMDGPU-DISABLED:       worker_state_machine.parallel_region.fallback.execute:
+; AMDGPU-DISABLED-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; AMDGPU-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; AMDGPU-DISABLED:       worker_state_machine.parallel_region.end:
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_kernel_end_parallel()
+; AMDGPU-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; AMDGPU-DISABLED:       worker_state_machine.done.barrier:
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; AMDGPU-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; AMDGPU-DISABLED:       thread.user_code.check:
+; AMDGPU-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; AMDGPU-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; AMDGPU-DISABLED:       user_code.entry:
+; AMDGPU-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR4]]
+; AMDGPU-DISABLED-NEXT:    call void @__omp_outlined__8(i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR4]]
+; AMDGPU-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; AMDGPU-DISABLED-NEXT:    ret void
+; AMDGPU-DISABLED:       worker.exit:
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_offloading_14_a34ca11_do_not_spmdize_target_l65
+; NVPTX-DISABLED-SAME: () #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR:%.*]] = alloca i8*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
+; NVPTX-DISABLED-NEXT:    store i32 0, i32* [[DOTZERO_ADDR]], align 4
+; NVPTX-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
+; NVPTX-DISABLED-NEXT:    [[THREAD_IS_WORKER:%.*]] = icmp ne i32 [[TMP0]], -1
+; NVPTX-DISABLED-NEXT:    br i1 [[THREAD_IS_WORKER]], label [[WORKER_STATE_MACHINE_BEGIN:%.*]], label [[THREAD_USER_CODE_CHECK:%.*]]
+; NVPTX-DISABLED:       worker_state_machine.begin:
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-DISABLED-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(i8** [[WORKER_WORK_FN_ADDR]])
+; NVPTX-DISABLED-NEXT:    [[WORKER_WORK_FN:%.*]] = load i8*, i8** [[WORKER_WORK_FN_ADDR]], align 8
+; NVPTX-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast i8* [[WORKER_WORK_FN]] to void (i16, i32)*
+; NVPTX-DISABLED-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq i8* [[WORKER_WORK_FN]], null
+; NVPTX-DISABLED-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED:%.*]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
+; NVPTX-DISABLED:       worker_state_machine.finished:
+; NVPTX-DISABLED-NEXT:    ret void
+; NVPTX-DISABLED:       worker_state_machine.is_active.check:
+; NVPTX-DISABLED-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
+; NVPTX-DISABLED:       worker_state_machine.parallel_region.fallback.execute:
+; NVPTX-DISABLED-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; NVPTX-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
+; NVPTX-DISABLED:       worker_state_machine.parallel_region.end:
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_kernel_end_parallel()
+; NVPTX-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_DONE_BARRIER]]
+; NVPTX-DISABLED:       worker_state_machine.done.barrier:
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_barrier_simple_spmd(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
+; NVPTX-DISABLED-NEXT:    br label [[WORKER_STATE_MACHINE_BEGIN]]
+; NVPTX-DISABLED:       thread.user_code.check:
+; NVPTX-DISABLED-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
+; NVPTX-DISABLED-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
+; NVPTX-DISABLED:       user_code.entry:
+; NVPTX-DISABLED-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]]) #[[ATTR4]]
+; NVPTX-DISABLED-NEXT:    call void @__omp_outlined__8(i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTTHREADID_TEMP_]], i32* noalias nocapture noundef nonnull readnone align 4 dereferenceable(4) [[DOTZERO_ADDR]]) #[[ATTR4]]
+; NVPTX-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
+; NVPTX-DISABLED-NEXT:    ret void
+; NVPTX-DISABLED:       worker.exit:
+; NVPTX-DISABLED-NEXT:    ret void
 ;
 entry:
   %.zero.addr = alloca i32, align 4
@@ -1312,21 +2183,39 @@ worker.exit:                                      ; preds = %entry
 }
 
 define internal void @__omp_outlined__8(i32* noalias %.global_tid., i32* noalias %.bound_tid.) #0 {
-; CHECK-LABEL: define {{[^@]+}}@__omp_outlined__8
-; CHECK-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-NEXT:    call void @unknown() #[[ATTR6]]
-; CHECK-NEXT:    ret void
-;
-; CHECK-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__8
-; CHECK-DISABLED-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
-; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
-; CHECK-DISABLED-NEXT:    call void @unknown() #[[ATTR6]]
-; CHECK-DISABLED-NEXT:    ret void
+;
+;
+; AMDGPU-LABEL: define {{[^@]+}}@__omp_outlined__8
+; AMDGPU-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-NEXT:  entry:
+; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-NEXT:    call void @unknown() #[[ATTR7]]
+; AMDGPU-NEXT:    ret void
+;
+; NVPTX-LABEL: define {{[^@]+}}@__omp_outlined__8
+; NVPTX-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-NEXT:  entry:
+; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-NEXT:    call void @unknown() #[[ATTR7]]
+; NVPTX-NEXT:    ret void
+;
+; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__8
+; AMDGPU-DISABLED-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-NEXT:  entry:
+; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; AMDGPU-DISABLED-NEXT:    call void @unknown() #[[ATTR7]]
+; AMDGPU-DISABLED-NEXT:    ret void
+;
+; NVPTX-DISABLED-LABEL: define {{[^@]+}}@__omp_outlined__8
+; NVPTX-DISABLED-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-NEXT:  entry:
+; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
+; NVPTX-DISABLED-NEXT:    call void @unknown() #[[ATTR7]]
+; NVPTX-DISABLED-NEXT:    ret void
 ;
 entry:
   %.global_tid..addr = alloca i32*, align 8
@@ -1367,57 +2256,115 @@ attributes #5 = { convergent }
 !16 = distinct !{!16, !14}
 !17 = distinct !{!17, !14}
 ;.
-; CHECK: attributes #[[ATTR0]] = { convergent noinline norecurse nounwind "frame-pointer"="none" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ptx32,+sm_20" }
-; CHECK: attributes #[[ATTR1:[0-9]+]] = { convergent "frame-pointer"="none" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ptx32,+sm_20" }
-; CHECK: attributes #[[ATTR2:[0-9]+]] = { convergent "frame-pointer"="none" "llvm.assume"="ompx_spmd_amenable" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ptx32,+sm_20" }
-; CHECK: attributes #[[ATTR3]] = { nounwind }
-; CHECK: attributes #[[ATTR4:[0-9]+]] = { convergent nounwind }
-; CHECK: attributes #[[ATTR5]] = { convergent "llvm.assume"="ompx_spmd_amenable" }
-; CHECK: attributes #[[ATTR6]] = { convergent }
+; AMDGPU: attributes #[[ATTR0]] = { convergent noinline norecurse nounwind "frame-pointer"="none" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ptx32,+sm_20" }
+; AMDGPU: attributes #[[ATTR1:[0-9]+]] = { convergent "frame-pointer"="none" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ptx32,+sm_20" }
+; AMDGPU: attributes #[[ATTR2:[0-9]+]] = { alwaysinline }
+; AMDGPU: attributes #[[ATTR3:[0-9]+]] = { convergent "frame-pointer"="none" "llvm.assume"="ompx_spmd_amenable" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ptx32,+sm_20" }
+; AMDGPU: attributes #[[ATTR4]] = { nounwind }
+; AMDGPU: attributes #[[ATTR5:[0-9]+]] = { convergent nounwind }
+; AMDGPU: attributes #[[ATTR6]] = { convergent "llvm.assume"="ompx_spmd_amenable" }
+; AMDGPU: attributes #[[ATTR7]] = { convergent }
+;.
+; NVPTX: attributes #[[ATTR0]] = { convergent noinline norecurse nounwind "frame-pointer"="none" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ptx32,+sm_20" }
+; NVPTX: attributes #[[ATTR1:[0-9]+]] = { convergent "frame-pointer"="none" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ptx32,+sm_20" }
+; NVPTX: attributes #[[ATTR2:[0-9]+]] = { alwaysinline }
+; NVPTX: attributes #[[ATTR3:[0-9]+]] = { convergent "frame-pointer"="none" "llvm.assume"="ompx_spmd_amenable" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ptx32,+sm_20" }
+; NVPTX: attributes #[[ATTR4]] = { nounwind }
+; NVPTX: attributes #[[ATTR5:[0-9]+]] = { convergent nounwind }
+; NVPTX: attributes #[[ATTR6]] = { convergent "llvm.assume"="ompx_spmd_amenable" }
+; NVPTX: attributes #[[ATTR7]] = { convergent }
+;.
+; AMDGPU-DISABLED: attributes #[[ATTR0]] = { convergent noinline norecurse nounwind "frame-pointer"="none" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ptx32,+sm_20" }
+; AMDGPU-DISABLED: attributes #[[ATTR1:[0-9]+]] = { convergent "frame-pointer"="none" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ptx32,+sm_20" }
+; AMDGPU-DISABLED: attributes #[[ATTR2:[0-9]+]] = { alwaysinline }
+; AMDGPU-DISABLED: attributes #[[ATTR3:[0-9]+]] = { convergent "frame-pointer"="none" "llvm.assume"="ompx_spmd_amenable" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ptx32,+sm_20" }
+; AMDGPU-DISABLED: attributes #[[ATTR4]] = { nounwind }
+; AMDGPU-DISABLED: attributes #[[ATTR5:[0-9]+]] = { convergent nounwind }
+; AMDGPU-DISABLED: attributes #[[ATTR6]] = { convergent "llvm.assume"="ompx_spmd_amenable" }
+; AMDGPU-DISABLED: attributes #[[ATTR7]] = { convergent }
+;.
+; NVPTX-DISABLED: attributes #[[ATTR0]] = { convergent noinline norecurse nounwind "frame-pointer"="none" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ptx32,+sm_20" }
+; NVPTX-DISABLED: attributes #[[ATTR1:[0-9]+]] = { convergent "frame-pointer"="none" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ptx32,+sm_20" }
+; NVPTX-DISABLED: attributes #[[ATTR2:[0-9]+]] = { alwaysinline }
+; NVPTX-DISABLED: attributes #[[ATTR3:[0-9]+]] = { convergent "frame-pointer"="none" "llvm.assume"="ompx_spmd_amenable" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ptx32,+sm_20" }
+; NVPTX-DISABLED: attributes #[[ATTR4]] = { nounwind }
+; NVPTX-DISABLED: attributes #[[ATTR5:[0-9]+]] = { convergent nounwind }
+; NVPTX-DISABLED: attributes #[[ATTR6]] = { convergent "llvm.assume"="ompx_spmd_amenable" }
+; NVPTX-DISABLED: attributes #[[ATTR7]] = { convergent }
+;.
+; AMDGPU: [[META0:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"sequential_loop_to_stack_var", i32 20, i32 1}
+; AMDGPU: [[META1:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"sequential_loop", i32 5, i32 0}
+; AMDGPU: [[META2:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"sequential_loop_to_shared_var", i32 35, i32 2}
+; AMDGPU: [[META3:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"do_not_spmdize_target", i32 65, i32 4}
+; AMDGPU: [[META4:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"sequential_loop_to_shared_var_guarded", i32 50, i32 3}
+; AMDGPU: [[META5:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_sequential_loop_l5, !"kernel", i32 1}
+; AMDGPU: [[META6:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_sequential_loop_to_stack_var_l20, !"kernel", i32 1}
+; AMDGPU: [[META7:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_l35, !"kernel", i32 1}
+; AMDGPU: [[META8:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_guarded_l50, !"kernel", i32 1}
+; AMDGPU: [[META9:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_do_not_spmdize_target_l65, !"kernel", i32 1}
+; AMDGPU: [[META10:![0-9]+]] = !{i32 1, !"wchar_size", i32 4}
+; AMDGPU: [[META11:![0-9]+]] = !{i32 7, !"openmp", i32 50}
+; AMDGPU: [[META12:![0-9]+]] = !{i32 7, !"openmp-device", i32 50}
+; AMDGPU: [[LOOP13]] = distinct !{!13, !14}
+; AMDGPU: [[META14:![0-9]+]] = !{!"llvm.loop.mustprogress"}
+; AMDGPU: [[LOOP15]] = distinct !{!15, !14}
+; AMDGPU: [[LOOP16]] = distinct !{!16, !14}
+; AMDGPU: [[LOOP17]] = distinct !{!17, !14}
 ;.
-; CHECK-DISABLED: attributes #[[ATTR0]] = { convergent noinline norecurse nounwind "frame-pointer"="none" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ptx32,+sm_20" }
-; CHECK-DISABLED: attributes #[[ATTR1:[0-9]+]] = { convergent "frame-pointer"="none" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ptx32,+sm_20" }
-; CHECK-DISABLED: attributes #[[ATTR2:[0-9]+]] = { convergent "frame-pointer"="none" "llvm.assume"="ompx_spmd_amenable" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+ptx32,+sm_20" }
-; CHECK-DISABLED: attributes #[[ATTR3]] = { nounwind }
-; CHECK-DISABLED: attributes #[[ATTR4:[0-9]+]] = { convergent nounwind }
-; CHECK-DISABLED: attributes #[[ATTR5]] = { convergent "llvm.assume"="ompx_spmd_amenable" }
-; CHECK-DISABLED: attributes #[[ATTR6]] = { convergent }
+; NVPTX: [[META0:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"sequential_loop_to_stack_var", i32 20, i32 1}
+; NVPTX: [[META1:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"sequential_loop", i32 5, i32 0}
+; NVPTX: [[META2:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"sequential_loop_to_shared_var", i32 35, i32 2}
+; NVPTX: [[META3:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"do_not_spmdize_target", i32 65, i32 4}
+; NVPTX: [[META4:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"sequential_loop_to_shared_var_guarded", i32 50, i32 3}
+; NVPTX: [[META5:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_sequential_loop_l5, !"kernel", i32 1}
+; NVPTX: [[META6:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_sequential_loop_to_stack_var_l20, !"kernel", i32 1}
+; NVPTX: [[META7:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_l35, !"kernel", i32 1}
+; NVPTX: [[META8:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_guarded_l50, !"kernel", i32 1}
+; NVPTX: [[META9:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_do_not_spmdize_target_l65, !"kernel", i32 1}
+; NVPTX: [[META10:![0-9]+]] = !{i32 1, !"wchar_size", i32 4}
+; NVPTX: [[META11:![0-9]+]] = !{i32 7, !"openmp", i32 50}
+; NVPTX: [[META12:![0-9]+]] = !{i32 7, !"openmp-device", i32 50}
+; NVPTX: [[LOOP13]] = distinct !{!13, !14}
+; NVPTX: [[META14:![0-9]+]] = !{!"llvm.loop.mustprogress"}
+; NVPTX: [[LOOP15]] = distinct !{!15, !14}
+; NVPTX: [[LOOP16]] = distinct !{!16, !14}
+; NVPTX: [[LOOP17]] = distinct !{!17, !14}
 ;.
-; CHECK: [[META0:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"sequential_loop_to_stack_var", i32 20, i32 1}
-; CHECK: [[META1:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"sequential_loop", i32 5, i32 0}
-; CHECK: [[META2:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"sequential_loop_to_shared_var", i32 35, i32 2}
-; CHECK: [[META3:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"do_not_spmdize_target", i32 65, i32 4}
-; CHECK: [[META4:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"sequential_loop_to_shared_var_guarded", i32 50, i32 3}
-; CHECK: [[META5:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_sequential_loop_l5, !"kernel", i32 1}
-; CHECK: [[META6:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_sequential_loop_to_stack_var_l20, !"kernel", i32 1}
-; CHECK: [[META7:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_l35, !"kernel", i32 1}
-; CHECK: [[META8:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_guarded_l50, !"kernel", i32 1}
-; CHECK: [[META9:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_do_not_spmdize_target_l65, !"kernel", i32 1}
-; CHECK: [[META10:![0-9]+]] = !{i32 1, !"wchar_size", i32 4}
-; CHECK: [[META11:![0-9]+]] = !{i32 7, !"openmp", i32 50}
-; CHECK: [[META12:![0-9]+]] = !{i32 7, !"openmp-device", i32 50}
-; CHECK: [[LOOP13]] = distinct !{!13, !14}
-; CHECK: [[META14:![0-9]+]] = !{!"llvm.loop.mustprogress"}
-; CHECK: [[LOOP15]] = distinct !{!15, !14}
-; CHECK: [[LOOP16]] = distinct !{!16, !14}
-; CHECK: [[LOOP17]] = distinct !{!17, !14}
+; AMDGPU-DISABLED: [[META0:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"sequential_loop_to_stack_var", i32 20, i32 1}
+; AMDGPU-DISABLED: [[META1:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"sequential_loop", i32 5, i32 0}
+; AMDGPU-DISABLED: [[META2:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"sequential_loop_to_shared_var", i32 35, i32 2}
+; AMDGPU-DISABLED: [[META3:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"do_not_spmdize_target", i32 65, i32 4}
+; AMDGPU-DISABLED: [[META4:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"sequential_loop_to_shared_var_guarded", i32 50, i32 3}
+; AMDGPU-DISABLED: [[META5:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_sequential_loop_l5, !"kernel", i32 1}
+; AMDGPU-DISABLED: [[META6:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_sequential_loop_to_stack_var_l20, !"kernel", i32 1}
+; AMDGPU-DISABLED: [[META7:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_l35, !"kernel", i32 1}
+; AMDGPU-DISABLED: [[META8:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_guarded_l50, !"kernel", i32 1}
+; AMDGPU-DISABLED: [[META9:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_do_not_spmdize_target_l65, !"kernel", i32 1}
+; AMDGPU-DISABLED: [[META10:![0-9]+]] = !{i32 1, !"wchar_size", i32 4}
+; AMDGPU-DISABLED: [[META11:![0-9]+]] = !{i32 7, !"openmp", i32 50}
+; AMDGPU-DISABLED: [[META12:![0-9]+]] = !{i32 7, !"openmp-device", i32 50}
+; AMDGPU-DISABLED: [[LOOP13]] = distinct !{!13, !14}
+; AMDGPU-DISABLED: [[META14:![0-9]+]] = !{!"llvm.loop.mustprogress"}
+; AMDGPU-DISABLED: [[LOOP15]] = distinct !{!15, !14}
+; AMDGPU-DISABLED: [[LOOP16]] = distinct !{!16, !14}
+; AMDGPU-DISABLED: [[LOOP17]] = distinct !{!17, !14}
 ;.
-; CHECK-DISABLED: [[META0:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"sequential_loop_to_stack_var", i32 20, i32 1}
-; CHECK-DISABLED: [[META1:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"sequential_loop", i32 5, i32 0}
-; CHECK-DISABLED: [[META2:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"sequential_loop_to_shared_var", i32 35, i32 2}
-; CHECK-DISABLED: [[META3:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"do_not_spmdize_target", i32 65, i32 4}
-; CHECK-DISABLED: [[META4:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"sequential_loop_to_shared_var_guarded", i32 50, i32 3}
-; CHECK-DISABLED: [[META5:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_sequential_loop_l5, !"kernel", i32 1}
-; CHECK-DISABLED: [[META6:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_sequential_loop_to_stack_var_l20, !"kernel", i32 1}
-; CHECK-DISABLED: [[META7:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_l35, !"kernel", i32 1}
-; CHECK-DISABLED: [[META8:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_guarded_l50, !"kernel", i32 1}
-; CHECK-DISABLED: [[META9:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_do_not_spmdize_target_l65, !"kernel", i32 1}
-; CHECK-DISABLED: [[META10:![0-9]+]] = !{i32 1, !"wchar_size", i32 4}
-; CHECK-DISABLED: [[META11:![0-9]+]] = !{i32 7, !"openmp", i32 50}
-; CHECK-DISABLED: [[META12:![0-9]+]] = !{i32 7, !"openmp-device", i32 50}
-; CHECK-DISABLED: [[LOOP13]] = distinct !{!13, !14}
-; CHECK-DISABLED: [[META14:![0-9]+]] = !{!"llvm.loop.mustprogress"}
-; CHECK-DISABLED: [[LOOP15]] = distinct !{!15, !14}
-; CHECK-DISABLED: [[LOOP16]] = distinct !{!16, !14}
-; CHECK-DISABLED: [[LOOP17]] = distinct !{!17, !14}
+; NVPTX-DISABLED: [[META0:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"sequential_loop_to_stack_var", i32 20, i32 1}
+; NVPTX-DISABLED: [[META1:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"sequential_loop", i32 5, i32 0}
+; NVPTX-DISABLED: [[META2:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"sequential_loop_to_shared_var", i32 35, i32 2}
+; NVPTX-DISABLED: [[META3:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"do_not_spmdize_target", i32 65, i32 4}
+; NVPTX-DISABLED: [[META4:![0-9]+]] = !{i32 0, i32 20, i32 171231761, !"sequential_loop_to_shared_var_guarded", i32 50, i32 3}
+; NVPTX-DISABLED: [[META5:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_sequential_loop_l5, !"kernel", i32 1}
+; NVPTX-DISABLED: [[META6:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_sequential_loop_to_stack_var_l20, !"kernel", i32 1}
+; NVPTX-DISABLED: [[META7:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_l35, !"kernel", i32 1}
+; NVPTX-DISABLED: [[META8:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_sequential_loop_to_shared_var_guarded_l50, !"kernel", i32 1}
+; NVPTX-DISABLED: [[META9:![0-9]+]] = !{void ()* @__omp_offloading_14_a34ca11_do_not_spmdize_target_l65, !"kernel", i32 1}
+; NVPTX-DISABLED: [[META10:![0-9]+]] = !{i32 1, !"wchar_size", i32 4}
+; NVPTX-DISABLED: [[META11:![0-9]+]] = !{i32 7, !"openmp", i32 50}
+; NVPTX-DISABLED: [[META12:![0-9]+]] = !{i32 7, !"openmp-device", i32 50}
+; NVPTX-DISABLED: [[LOOP13]] = distinct !{!13, !14}
+; NVPTX-DISABLED: [[META14:![0-9]+]] = !{!"llvm.loop.mustprogress"}
+; NVPTX-DISABLED: [[LOOP15]] = distinct !{!15, !14}
+; NVPTX-DISABLED: [[LOOP16]] = distinct !{!16, !14}
+; NVPTX-DISABLED: [[LOOP17]] = distinct !{!17, !14}
 ;.