[OpenMP] Initialize data sharing stack for SPMD case

[gtbercea]

Summary

In the SPMD case, we need to initialize the data sharing and globalization infrastructure. This covers the case when an SPMD region calls a function in a different compilation unit.

[ABataev]

Tests?

[gtbercea]

Fix test.

[ABataev]

What about the check for the new function call?

[Hahnfeld]

lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp

84–85 ↗

(On Diff #155002)

In case I'm not missing something obvious this function doesn't exist (yet) in libomptarget-nvptx?

1110–1114 ↗

(On Diff #155002)

Why is this call being removed? There's no mention in the summary AFAICS

[gtbercea]

Add test for spmd stack init function.

[gtbercea]

lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp

84–85 ↗

(On Diff #155002)

Correct. There's a libomptarget patch coming up as well.

1110–1114 ↗

(On Diff #155002)

Not needed on workers after libomptarget patch. I could perhaps put this into a separate patch along with the test fix.

[ABataev]

test/OpenMP/nvptx_data_sharing.cpp

33 ↗

(On Diff #155037)

It is better to check that this call is not emitted, like this CK1-NOT: call void @__kmpc_data_sharing_init_stack

test/OpenMP/nvptx_data_sharing_spmd.cpp

1–24 ↗

(On Diff #155037)

I think we already have some tests with SPMD construct, you can just modify the existing tests instead of adding another one (e.g. nvptx_target_parallel_codegen.cpp)

[gtbercea]

Fix tests.

[ABataev]

LG

cfe/trunk/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp

enum OpenMPRTLFunctionNVPTX {
   /// int32_t index, int32_t width),
   /// void (*kmp_LoadReduceFctPtr)(void *reduce_data, void * scratchpad, int32_t
   /// index, int32_t width, int32_t reduce))
   OMPRTL_NVPTX__kmpc_teams_reduce_nowait,
   /// Call to __kmpc_nvptx_end_reduce_nowait(int32_t global_tid);
   OMPRTL_NVPTX__kmpc_end_reduce_nowait,
   /// Call to void __kmpc_data_sharing_init_stack();
   OMPRTL_NVPTX__kmpc_data_sharing_init_stack,
+  /// Call to void __kmpc_data_sharing_init_stack_spmd();
+  OMPRTL_NVPTX__kmpc_data_sharing_init_stack_spmd,
   /// Call to void* __kmpc_data_sharing_push_stack(size_t size,
   /// int16_t UseSharedMemory);
   OMPRTL_NVPTX__kmpc_data_sharing_push_stack,
   /// Call to void __kmpc_data_sharing_pop_stack(void *a);
   OMPRTL_NVPTX__kmpc_data_sharing_pop_stack,
   /// Call to void __kmpc_begin_sharing_variables(void ***args,
   /// size_t n_args);
   OMPRTL_NVPTX__kmpc_begin_sharing_variables,
void CGOpenMPRuntimeNVPTX::emitSPMDEntryHeader(
   // Initialize the OMP state in the runtime; called by all active threads.
   // TODO: Set RequiresOMPRuntime and RequiresDataSharing parameters
   // based on code analysis of the target region.
   llvm::Value *Args[] = {getThreadLimit(CGF, /*IsInSPMDExecutionMode=*/true),
                          /*RequiresOMPRuntime=*/Bld.getInt16(1),
                          /*RequiresDataSharing=*/Bld.getInt16(1)};
   CGF.EmitRuntimeCall(
       createNVPTXRuntimeFunction(OMPRTL_NVPTX__kmpc_spmd_kernel_init), Args);
+
+  // For data sharing, we need to initialize the stack.
+  CGF.EmitRuntimeCall(
+      createNVPTXRuntimeFunction(
+          OMPRTL_NVPTX__kmpc_data_sharing_init_stack_spmd));
+
   CGF.EmitBranch(ExecuteBB);
 
   CGF.EmitBlock(ExecuteBB);
 
   IsInTargetMasterThreadRegion = true;
 }
 
 void CGOpenMPRuntimeNVPTX::emitSPMDEntryFooter(CodeGenFunction &CGF,
void CGOpenMPRuntimeNVPTX::emitWorkerLoop(CodeGenFunction &CGF,
 
   CGF.EmitBranch(AwaitBB);
 
   // Workers wait for work from master.
   CGF.EmitBlock(AwaitBB);
   // Wait for parallel work
   syncCTAThreads(CGF);
 
-  // For data sharing, we need to initialize the stack for workers.
-  CGF.EmitRuntimeCall(
-      createNVPTXRuntimeFunction(
-          OMPRTL_NVPTX__kmpc_data_sharing_init_stack));
-
   Address WorkFn =
       CGF.CreateDefaultAlignTempAlloca(CGF.Int8PtrTy, /*Name=*/"work_fn");
   Address ExecStatus =
       CGF.CreateDefaultAlignTempAlloca(CGF.Int8Ty, /*Name=*/"exec_status");
   CGF.InitTempAlloca(ExecStatus, Bld.getInt8(/*C=*/0));
   CGF.InitTempAlloca(WorkFn, llvm::Constant::getNullValue(CGF.Int8PtrTy));
 
   // TODO: Optimize runtime initialization and pass in correct value.
CGOpenMPRuntimeNVPTX::createNVPTXRuntimeFunction(unsigned Function) {
   }
   case OMPRTL_NVPTX__kmpc_data_sharing_init_stack: {
     /// Build void __kmpc_data_sharing_init_stack();
     auto *FnTy =
         llvm::FunctionType::get(CGM.VoidTy, llvm::None, /*isVarArg*/ false);
     RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_data_sharing_init_stack");
     break;
   }
+  case OMPRTL_NVPTX__kmpc_data_sharing_init_stack_spmd: {
+    /// Build void __kmpc_data_sharing_init_stack_spmd();
+    auto *FnTy =
+        llvm::FunctionType::get(CGM.VoidTy, llvm::None, /*isVarArg*/ false);
+    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_data_sharing_init_stack_spmd");
+    break;
+  }
   case OMPRTL_NVPTX__kmpc_data_sharing_push_stack: {
     // Build void *__kmpc_data_sharing_push_stack(size_t size,
     // int16_t UseSharedMemory);
     llvm::Type *TypeParams[] = {CGM.SizeTy, CGM.Int16Ty};
     auto *FnTy =
         llvm::FunctionType::get(CGM.VoidPtrTy, TypeParams, /*isVarArg=*/false);
     RTLFn = CGM.CreateRuntimeFunction(
         FnTy, /*Name=*/"__kmpc_data_sharing_push_stack");

cfe/trunk/test/OpenMP/nvptx_data_sharing.cpp

#pragma omp parallel private(c)
       b = a + 10000;
     }
   }
 }
 
 /// ========= In the worker function ========= ///
 // CK1: {{.*}}define internal void @__omp_offloading{{.*}}test_ds{{.*}}_worker()
 // CK1: call void @llvm.nvvm.barrier0()
-// CK1: call void @__kmpc_data_sharing_init_stack
+// CK1-NOT: call void @__kmpc_data_sharing_init_stack
 
 /// ========= In the kernel function ========= ///
 
 // CK1: {{.*}}define weak void @__omp_offloading{{.*}}test_ds{{.*}}()
 // CK1: [[SHAREDARGS1:%.+]] = alloca i8**
 // CK1: [[SHAREDARGS2:%.+]] = alloca i8**
 // CK1: call void @__kmpc_kernel_init
 // CK1: call void @__kmpc_data_sharing_init_stack

cfe/trunk/test/OpenMP/nvptx_target_parallel_codegen.cpp

int bar(int n){
 
   // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l26}}(
   // CHECK: [[AA_ADDR:%.+]] = alloca i16*, align
   // CHECK-NOT: call i8* @__kmpc_data_sharing_push_stack
   // CHECK: store i16* {{%.+}}, i16** [[AA_ADDR]], align
   // CHECK: [[AA:%.+]] = load i16*, i16** [[AA_ADDR]], align
   // CHECK: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
   // CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]],
+  // CHECK: call void @__kmpc_data_sharing_init_stack_spmd
   // CHECK: br label {{%?}}[[EXEC:.+]]
   //
   // CHECK: [[EXEC]]
   // CHECK: {{call|invoke}} void [[OP1:@.+]]({{.+}}, {{.+}}, i16* [[AA]])
   // CHECK: br label {{%?}}[[DONE:.+]]
   //
   // CHECK: [[DONE]]
   // CHECK: call void @__kmpc_spmd_kernel_deinit()
int bar(int n){
   // CHECK: store i32* {{%.+}}, i32** [[A_ADDR]], align
   // CHECK: store i16* {{%.+}}, i16** [[AA_ADDR]], align
   // CHECK: store [10 x i32]* {{%.+}}, [10 x i32]** [[B_ADDR]], align
   // CHECK: [[A:%.+]] = load i32*, i32** [[A_ADDR]], align
   // CHECK: [[AA:%.+]] = load i16*, i16** [[AA_ADDR]], align
   // CHECK: [[B:%.+]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align
   // CHECK: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
   // CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]],
+  // CHECK: call void @__kmpc_data_sharing_init_stack_spmd
   // CHECK: br label {{%?}}[[EXEC:.+]]
   //
   // CHECK: [[EXEC]]
   // CHECK: {{call|invoke}} void [[OP2:@.+]]({{.+}}, {{.+}}, i32* [[A]], i16* [[AA]], [10 x i32]* [[B]])
   // CHECK: br label {{%?}}[[DONE:.+]]
   //
   // CHECK: [[DONE]]
   // CHECK: call void @__kmpc_spmd_kernel_deinit()

cfe/trunk/test/OpenMP/nvptx_target_parallel_proc_bind_codegen.cpp

int bar(int n){
 
   a += ftemplate<int>(n);
 
   return a;
 }
 
   // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l22}}(
   // CHECK: call void @__kmpc_spmd_kernel_init(
+  // CHECK: call void @__kmpc_data_sharing_init_stack_spmd
   // CHECK: br label {{%?}}[[EXEC:.+]]
   //
   // CHECK: [[EXEC]]
   // CHECK-NOT: call void @__kmpc_push_proc_bind
   // CHECK: {{call|invoke}} void [[OP1:@.+]](
   // CHECK: br label {{%?}}[[DONE:.+]]
   //
   // CHECK: [[DONE]]
   // CHECK: call void @__kmpc_spmd_kernel_deinit()
   // CHECK: br label {{%?}}[[EXIT:.+]]
   //
   // CHECK: [[EXIT]]
   // CHECK: ret void
   // CHECK: }
 
 
 
 
 
   // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l26}}(
   // CHECK: call void @__kmpc_spmd_kernel_init(
+  // CHECK: call void @__kmpc_data_sharing_init_stack_spmd
   // CHECK: br label {{%?}}[[EXEC:.+]]
   //
   // CHECK: [[EXEC]]
   // CHECK-NOT: call void @__kmpc_push_proc_bind
   // CHECK: {{call|invoke}} void [[OP1:@.+]](
   // CHECK: br label {{%?}}[[DONE:.+]]
   //
   // CHECK: [[DONE]]
   // CHECK: call void @__kmpc_spmd_kernel_deinit()
   // CHECK: br label {{%?}}[[EXIT:.+]]
   //
   // CHECK: [[EXIT]]
   // CHECK: ret void
   // CHECK: }
 
 
 
 
   // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l31}}(
   // CHECK: call void @__kmpc_spmd_kernel_init(
+  // CHECK: call void @__kmpc_data_sharing_init_stack_spmd
   // CHECK: br label {{%?}}[[EXEC:.+]]
   //
   // CHECK: [[EXEC]]
   // CHECK-NOT: call void @__kmpc_push_proc_bind
   // CHECK: {{call|invoke}} void [[OP1:@.+]](
   // CHECK: br label {{%?}}[[DONE:.+]]
   //
   // CHECK: [[DONE]]
   // CHECK: call void @__kmpc_spmd_kernel_deinit()
   // CHECK: br label {{%?}}[[EXIT:.+]]
   //
   // CHECK: [[EXIT]]
   // CHECK: ret void
   // CHECK: }
 #endif

cfe/trunk/test/OpenMP/nvptx_target_parallel_reduction_codegen.cpp

int bar(int n){
   a += ftemplate<char>(n);
 
   return a;
 }
 
   // CHECK: define {{.*}}void {{@__omp_offloading_.+template.+l27}}(
   //
   // CHECK: call void @__kmpc_spmd_kernel_init(
+  // CHECK: call void @__kmpc_data_sharing_init_stack_spmd
   // CHECK: br label {{%?}}[[EXECUTE:.+]]
   //
   // CHECK: [[EXECUTE]]
   // CHECK: {{call|invoke}} void [[PFN:@.+]](i32*
   // CHECK: call void @__kmpc_spmd_kernel_deinit()
   //
   //
   // define internal void [[PFN]](
int bar(int n){
 
 
 
 
 
   // CHECK: define {{.*}}void {{@__omp_offloading_.+template.+l32}}(
   //
   // CHECK: call void @__kmpc_spmd_kernel_init(
+  // CHECK: call void @__kmpc_data_sharing_init_stack_spmd
   // CHECK: br label {{%?}}[[EXECUTE:.+]]
   //
   // CHECK: [[EXECUTE]]
   // CHECK: {{call|invoke}} void [[PFN1:@.+]](i32*
   // CHECK: call void @__kmpc_spmd_kernel_deinit()
   //
   //
   // define internal void [[PFN1]](
int bar(int n){
 
 
 
 
 
   // CHECK: define {{.*}}void {{@__omp_offloading_.+template.+l38}}(
   //
   // CHECK: call void @__kmpc_spmd_kernel_init(
+  // CHECK: call void @__kmpc_data_sharing_init_stack_spmd
   // CHECK: br label {{%?}}[[EXECUTE:.+]]
   //
   // CHECK: [[EXECUTE]]
   // CHECK: {{call|invoke}} void [[PFN2:@.+]](i32*
   // CHECK: call void @__kmpc_spmd_kernel_deinit()
   //
   //
   // define internal void [[PFN2]](

cfe/trunk/test/OpenMP/nvptx_target_teams_codegen.cpp

int bar(int n){
   // CHECK: call void @llvm.nvvm.barrier0()
   // CHECK: br label {{%?}}[[EXIT]]
   //
   // CHECK: [[EXIT]]
   // CHECK: ret void
 
 // CHECK: define weak void @__omp_offloading_{{.*}}ftemplate{{.*}}_l37(
 // CHECK: call void @__kmpc_spmd_kernel_init(
+// CHECK: call void @__kmpc_data_sharing_init_stack_spmd
 // CHECK: call i8* @__kmpc_data_sharing_push_stack(
 // CHECK-NOT: call void @__kmpc_serialized_parallel(
 // CHECK: call void [[L0:@.+]](i32* %{{.+}}, i32* %{{.+}}, i16* %{{.*}})
 // CHECK-NOT: call void @__kmpc_end_serialized_parallel(
 // CHECK: call void @__kmpc_data_sharing_pop_stack(
 // CHECK: call void @__kmpc_spmd_kernel_deinit()
 // CHECK: ret
 

cfe/trunk/test/OpenMP/nvptx_target_teams_distribute_parallel_for_codegen.cpp

int bar(int n){
   a += ftemplate<int>(n);
 
   return a;
 }
 
 // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}(
 // CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
 // CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]],
+// CHECK: call void @__kmpc_data_sharing_init_stack_spmd
 // CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 91,
 // CHECK: {{call|invoke}} void [[OUTL1:@.+]](
 // CHECK: call void @__kmpc_for_static_fini(
 // CHECK: call void @__kmpc_spmd_kernel_deinit()
 // CHECK: ret void
 
 // CHECK: define internal void [[OUTL1]](
 // CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 33,
 // CHECK: call void @__kmpc_for_static_fini(
 // CHECK: ret void
 
 // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}(
 // CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
 // CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]],
+// CHECK: call void @__kmpc_data_sharing_init_stack_spmd
 // CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 92,
 // CHECK: {{call|invoke}} void [[OUTL2:@.+]](
 // CHECK: call void @__kmpc_for_static_fini(
 // CHECK: call void @__kmpc_spmd_kernel_deinit()
 // CHECK: ret void
 
 // CHECK: define internal void [[OUTL2]](
 // CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 34,
 // CHECK: call void @__kmpc_for_static_fini(
 // CHECK: ret void
 
 // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}(
 // CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
 // CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]],
+// CHECK: call void @__kmpc_data_sharing_init_stack_spmd
 // CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 92,
 // CHECK: {{call|invoke}} void [[OUTL3:@.+]](
 // CHECK: call void @__kmpc_for_static_fini(
 // CHECK: call void @__kmpc_spmd_kernel_deinit()
 // CHECK: ret void
 
 // CHECK: define internal void [[OUTL3]](
 // CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 34,
 // CHECK: call void @__kmpc_for_static_fini(
 // CHECK: ret void
 
 // CHECK: define {{.*}}void {{@__omp_offloading_.+}}({{.+}}, i{{32|64}} [[F_IN:%.+]])
 // CHECK: store {{.+}} [[F_IN]], {{.+}}* {{.+}},
 // CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
 // CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]],
+// CHECK: call void @__kmpc_data_sharing_init_stack_spmd
 // CHECK: store {{.+}} 99, {{.+}}* [[COMB_UB:%.+]], align
 // CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 92, {{.+}}, {{.+}}, {{.+}}* [[COMB_UB]],
 // CHECK: {{call|invoke}} void [[OUTL4:@.+]](
 // CHECK: call void @__kmpc_for_static_fini(
 // CHECK: call void @__kmpc_spmd_kernel_deinit()
 // CHECK: ret void
 
 // CHECK: define internal void [[OUTL4]](

cfe/trunk/test/OpenMP/nvptx_target_teams_distribute_parallel_for_simd_codegen.cpp

 #define N 1000
 #define M 10
 
 template<typename tx>
 tx ftemplate(int n) {
   tx a[N];
   short aa[N];
   tx b[10];
-  tx c[M][M];  
+  tx c[M][M];
   tx f = n;
   tx l;
   int k;
 
 #pragma omp target teams distribute parallel for simd lastprivate(l) dist_schedule(static,128) schedule(static,32)
   for(int i = 0; i < n; i++) {
     a[i] = 1;
     l = i;
   }
 
   #pragma omp target teams distribute parallel for simd map(tofrom: aa) num_teams(M) thread_limit(64)
   for(int i = 0; i < n; i++) {
     aa[i] += 1;
   }
 
 #pragma omp target teams distribute parallel for simd map(tofrom:a, aa, b) if(target: n>40) proc_bind(spread)
   for(int i = 0; i < 10; i++) {
     b[i] += 1;
   }
 
 #pragma omp target teams distribute parallel for simd collapse(2) firstprivate(f) private(k)
   for(int i = 0; i < M; i++) {
     for(int j = 0; j < M; j++) {
       k = M;
-      c[i][j] = i+j*f+k;      
+      c[i][j] = i+j*f+k;
     }
   }
 
   return a[0];
 }
 
 int bar(int n){
   int a = 0;
 
   a += ftemplate<int>(n);
 
   return a;
 }
 
 // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}(
 // CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
 // CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]],
+// CHECK: call void @__kmpc_data_sharing_init_stack_spmd
 // CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 91,
 // CHECK: {{call|invoke}} void [[OUTL1:@.+]](
 // CHECK: call void @__kmpc_for_static_fini(
 // CHECK: call void @__kmpc_spmd_kernel_deinit()
 // CHECK: ret void
 
 // CHECK: define internal void [[OUTL1]](
 // CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 33,
 // CHECK: call void @__kmpc_for_static_fini(
 // CHECK: ret void
 
 // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}(
 // CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
 // CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]],
+// CHECK: call void @__kmpc_data_sharing_init_stack_spmd
 // CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 92,
 // CHECK: {{call|invoke}} void [[OUTL2:@.+]](
 // CHECK: call void @__kmpc_for_static_fini(
 // CHECK: call void @__kmpc_spmd_kernel_deinit()
 // CHECK: ret void
 
 // CHECK: define internal void [[OUTL2]](
 // CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 34,
 // CHECK: call void @__kmpc_for_static_fini(
 // CHECK: ret void
 
 // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}(
 // CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
 // CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]],
+// CHECK: call void @__kmpc_data_sharing_init_stack_spmd
 // CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 92,
 // CHECK: {{call|invoke}} void [[OUTL3:@.+]](
 // CHECK: call void @__kmpc_for_static_fini(
 // CHECK: call void @__kmpc_spmd_kernel_deinit()
 // CHECK: ret void
 
 // CHECK: define internal void [[OUTL3]](
 // CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 34,
 // CHECK: call void @__kmpc_for_static_fini(
 // CHECK: ret void
 
 // CHECK: define {{.*}}void {{@__omp_offloading_.+}}({{.+}}, i{{32|64}} [[F_IN:%.+]])
 // CHECK: store {{.+}} [[F_IN]], {{.+}}* {{.+}},
 // CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
 // CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]],
+// CHECK: call void @__kmpc_data_sharing_init_stack_spmd
 // CHECK: store {{.+}} 99, {{.+}}* [[COMB_UB:%.+]], align
 // CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 92, {{.+}}, {{.+}}, {{.+}}* [[COMB_UB]],
 // CHECK: {{call|invoke}} void [[OUTL4:@.+]](
 // CHECK: call void @__kmpc_for_static_fini(
 // CHECK: call void @__kmpc_spmd_kernel_deinit()
 // CHECK: ret void
 
 // CHECK: define internal void [[OUTL4]](
 // CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 34,
 // CHECK: call void @__kmpc_for_static_fini(
 // CHECK: ret void
 
 #endif