[OPENMP][NVPTX]Improve emission of the globalized variables for

target/teams/distribute regions.

Target/teams/distribute regions exist for all the time the kernel is
executed. Thus, if the variable is declared in their context and then
escape it, we can allocate global memory statically instead of
allocating it dynamically.
Patch captures all the globalized variables in target/teams/distribute
contexts, merges them into the records, one per each target region.
Those records are then joined into the union, one per compilation unit
(to save the global memory). Those units are organized into
2 x dimensional arrays, where the first dimension is
the number of blocks per SM and the second one is the number of SMs.
Runtime functions manage this global memory space between the executing
teams.

llvm-svn: 345978

7 files changed, 99 insertions, 14 deletions

diff --git a/clang/test/OpenMP/nvptx_data_sharing.cpp b/clang/test/OpenMP/nvptx_data_sharing.cpp
index 0acb1199152..2f251a7dd06 100644
--- a/clang/test/OpenMP/nvptx_data_sharing.cpp
+++ b/clang/test/OpenMP/nvptx_data_sharing.cpp
@@ -26,6 +26,11 @@ void test_ds(){
     }
   }
 }
+// CK1: [[MEM_TY:%.+]] = type { [8 x i8] }
+// CK1-DAG: [[GLOBAL_RD:@.+]] = weak global [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]] zeroinitializer
+// CK1-DAG: [[GLOBAL_RD_PTR:@.+]] = weak unnamed_addr constant i8* getelementptr inbounds ([{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]], [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]]* [[GLOBAL_RD]], i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0)
+// CK1-DAG: [[KERNEL_PTR:@.+]] = internal addrspace(3) global i8* null
+// CK1-DAG: [[KERNEL_SIZE:@.+]] = internal unnamed_addr constant i64 8
 
 /// ========= In the worker function ========= ///
 // CK1: {{.*}}define internal void @__omp_offloading{{.*}}test_ds{{.*}}_worker()
@@ -39,7 +44,11 @@ void test_ds(){
 // CK1: [[SHAREDARGS2:%.+]] = alloca i8**
 // CK1: call void @__kmpc_kernel_init
 // CK1: call void @__kmpc_data_sharing_init_stack
-// CK1: [[GLOBALSTACK:%.+]] = call i8* @__kmpc_data_sharing_push_stack(i64 8, i16 0)
+// CK1: [[GLOBAL_RD:%.+]] = load i8*, i8** [[GLOBAL_RD_PTR]],
+// CK1: [[SIZE:%.+]] = load i64, i64* [[KERNEL_SIZE]],
+// CK1: call void @__kmpc_get_team_static_memory(i8* [[GLOBAL_RD]], i64 [[SIZE]], i16 0, i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
+// CK1: [[KERNEL_RD:%.+]] = load i8*, i8* addrspace(3)* [[KERNEL_PTR]],
+// CK1: [[GLOBALSTACK:%.+]] = getelementptr inbounds i8, i8* [[KERNEL_RD]], i64 0
 // CK1: [[GLOBALSTACK2:%.+]] = bitcast i8* [[GLOBALSTACK]] to %struct._globalized_locals_ty*
 // CK1: [[A:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[GLOBALSTACK2]], i32 0, i32 0
 // CK1: [[B:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[GLOBALSTACK2]], i32 0, i32 1
@@ -66,7 +75,7 @@ void test_ds(){
 // CK1: call void @llvm.nvvm.barrier0()
 // CK1: call void @llvm.nvvm.barrier0()
 // CK1: call void @__kmpc_end_sharing_variables()
-// CK1: call void @__kmpc_data_sharing_pop_stack(i8* [[GLOBALSTACK]])
+// CK1: call void @__kmpc_restore_team_static_memory(i16 0)
 // CK1: call void @__kmpc_kernel_deinit(i16 1)
 
 /// ========= In the data sharing wrapper function ========= ///
diff --git a/clang/test/OpenMP/nvptx_distribute_parallel_generic_mode_codegen.cpp b/clang/test/OpenMP/nvptx_distribute_parallel_generic_mode_codegen.cpp
index 2b18f6d3f9b..c41964517bc 100644
--- a/clang/test/OpenMP/nvptx_distribute_parallel_generic_mode_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_distribute_parallel_generic_mode_codegen.cpp
@@ -21,12 +21,19 @@ int main(int argc, char **argv) {
   return 0;
 }
 
-// CHECK: @__omp_offloading_{{.*}}_main_l17_exec_mode = weak constant i8 1
+// CHECK: [[MEM_TY:%.+]] = type { [84 x i8] }
+// CHECK-DAG: [[GLOBAL_RD:@.+]] = weak global [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]] zeroinitializer
+// CHECK-DAG: [[GLOBAL_RD_PTR:@.+]] = weak unnamed_addr constant i8* getelementptr inbounds ([{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]], [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]]* [[GLOBAL_RD]], i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0)
+// CHECK-DAG: [[KERNEL_PTR:@.+]] = internal addrspace(3) global i8* null
+// CHECK-DAG: [[KERNEL_SIZE:@.+]] = internal unnamed_addr constant i{{64|32}} 84
+// CHECK-DAG: @__omp_offloading_{{.*}}_main_l17_exec_mode = weak constant i8 1
 
 // CHECK-LABEL: define internal void @__omp_offloading_{{.*}}_main_l17_worker(
 
 // CHECK: define weak void @__omp_offloading_{{.*}}_main_l17([10 x i32]* dereferenceable(40) %{{.+}}, [10 x i32]* dereferenceable(40) %{{.+}}, i32* dereferenceable(4) %{{.+}}, i{{64|32}} %{{.+}}, [10 x i32]* dereferenceable(40) %{{.+}})
-// CHECK: [[PTR:%.+]] = call i8* @__kmpc_data_sharing_push_stack(i{{64|32}} 84, i16 0)
+// CHECK: [[GLOBAL_RD:%.+]] = load i8*, i8** [[GLOBAL_RD_PTR]],
+// CHECK: call void @__kmpc_get_team_static_memory(i8* [[GLOBAL_RD]], i{{64|32}} 84, i16 0, i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
+// CHECK: [[PTR:%.+]] = load i8*, i8* addrspace(3)* [[KERNEL_PTR]],
 // CHECK: [[STACK:%.+]] = bitcast i8* [[PTR]] to %struct._globalized_locals_ty*
 // CHECK: [[ARGC:%.+]] = load i32, i32* %{{.+}}, align
 // CHECK: [[ARGC_ADDR:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[STACK]], i{{32|64}} 0, i{{32|64}} 0
@@ -41,7 +48,7 @@ int main(int argc, char **argv) {
 
 // CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* @
 
-// CHECK: call void @__kmpc_data_sharing_pop_stack(i8* [[PTR]])
+// CHECK: call void @__kmpc_restore_team_static_memory(i16 0)
 
 // CHECK: define internal void [[PARALLEL]](
 // CHECK-NOT: call i8* @__kmpc_data_sharing_push_stack(
diff --git a/clang/test/OpenMP/nvptx_parallel_codegen.cpp b/clang/test/OpenMP/nvptx_parallel_codegen.cpp
index d1a3104407d..5f0f09af870 100644
--- a/clang/test/OpenMP/nvptx_parallel_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_parallel_codegen.cpp
@@ -71,6 +71,12 @@ int bar(int n){
   return a;
 }
 
+// CHECK: [[MEM_TY:%.+]] = type { [4 x i8] }
+// CHECK-DAG: [[GLOBAL_RD:@.+]] = weak global [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]] zeroinitializer
+// CHECK-DAG: [[GLOBAL_RD_PTR:@.+]] = weak unnamed_addr constant i8* getelementptr inbounds ([{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]], [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]]* [[GLOBAL_RD]], i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0)
+// CHECK-DAG: [[KERNEL_PTR:@.+]] = internal addrspace(3) global i8* null
+// CHECK-DAG: [[KERNEL_SIZE:@.+]] = internal unnamed_addr constant i{{64|32}} 4
+
 // CHECK-NOT: define {{.*}}void {{@__omp_offloading_.+template.+l17}}_worker()
 
 // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l26}}_worker()
@@ -318,12 +324,17 @@ int bar(int n){
 // CHECK-32: [[A_ADDR:%.+]] = alloca i32,
 // CHECK-64: [[A_ADDR:%.+]] = alloca i64,
 // CHECK-64: [[CONV:%.+]] = bitcast i64* [[A_ADDR]] to i32*
-// CHECK: [[STACK:%.+]] = call i8* @__kmpc_data_sharing_push_stack(i{{64|32}} 4, i16 0)
+// CHECK: [[GLOBAL_RD:%.+]] = load i8*, i8** [[GLOBAL_RD_PTR]],
+// CHECK: [[SIZE:%.+]] = load i{{64|32}}, i{{64|32}}* [[KERNEL_SIZE]],
+// CHECK: call void @__kmpc_get_team_static_memory(i8* [[GLOBAL_RD]], i{{64|32}} [[SIZE]], i16 0, i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
+// CHECK: [[KERNEL_RD:%.+]] = load i8*, i8* addrspace(3)* [[KERNEL_PTR]],
+// CHECK: [[STACK:%.+]] = getelementptr inbounds i8, i8* [[KERNEL_RD]], i{{64|32}} 0
 // CHECK: [[BC:%.+]] = bitcast i8* [[STACK]] to %struct._globalized_locals_ty*
 // CHECK-32: [[A:%.+]] = load i32, i32* [[A_ADDR]],
 // CHECK-64: [[A:%.+]] = load i32, i32* [[CONV]],
 // CHECK: [[GLOBAL_A_ADDR:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[BC]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
 // CHECK: store i32 [[A]], i32* [[GLOBAL_A_ADDR]],
+// CHECK: call void @__kmpc_restore_team_static_memory(i16 0)
 
 // CHECK-LABEL: define internal void @{{.+}}(i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* dereferenceable{{.*}})
 // CHECK:  [[CC:%.+]] = alloca i32,
diff --git a/clang/test/OpenMP/nvptx_parallel_for_codegen.cpp b/clang/test/OpenMP/nvptx_parallel_for_codegen.cpp
index c61ba91bbfd..75379c08b4b 100644
--- a/clang/test/OpenMP/nvptx_parallel_for_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_parallel_for_codegen.cpp
@@ -30,6 +30,12 @@ int bar(int n){
   return a;
 }
 
+// CHECK: [[MEM_TY:%.+]] = type { [4 x i8] }
+// CHECK-DAG: [[GLOBAL_RD:@.+]] = weak global [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]] zeroinitializer
+// CHECK-DAG: [[GLOBAL_RD_PTR:@.+]] = weak unnamed_addr constant i8* getelementptr inbounds ([{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]], [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]]* [[GLOBAL_RD]], i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0)
+// CHECK-DAG: [[KERNEL_PTR:@.+]] = internal addrspace(3) global i8* null
+// CHECK-DAG: [[KERNEL_SIZE:@.+]] = internal unnamed_addr constant i{{64|32}} 4
+
 // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l12}}_worker()
 // CHECK: call void @llvm.nvvm.barrier0()
 // CHECK: call i1 @__kmpc_kernel_parallel(
@@ -39,13 +45,17 @@ int bar(int n){
 // CHECK: call void @__omp_offloading_{{.*}}l12_worker()
 // CHECK: call void @__kmpc_kernel_init(
 // CHECK: call void @__kmpc_data_sharing_init_stack()
-// CHECK: call i8* @__kmpc_data_sharing_push_stack(i64 4, i16 0)
+// CHECK: [[GLOBAL_RD:%.+]] = load i8*, i8** [[GLOBAL_RD_PTR]],
+// CHECK: [[SIZE:%.+]] = load i{{64|32}}, i{{64|32}}* [[KERNEL_SIZE]],
+// CHECK: call void @__kmpc_get_team_static_memory(i8* [[GLOBAL_RD]], i{{64|32}} [[SIZE]], i16 0, i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
+// CHECK: [[KERNEL_RD:%.+]] = load i8*, i8* addrspace(3)* [[KERNEL_PTR]],
+// CHECK: [[STACK:%.+]] = getelementptr inbounds i8, i8* [[KERNEL_RD]], i{{64|32}} 0
 // CHECK: call void @__kmpc_kernel_prepare_parallel(
 // CHECK: call void @__kmpc_begin_sharing_variables({{.*}}, i64 2)
 // CHECK: call void @llvm.nvvm.barrier0()
 // CHECK: call void @llvm.nvvm.barrier0()
 // CHECK: call void @__kmpc_end_sharing_variables()
-// CHECK: call void @__kmpc_data_sharing_pop_stack(
+// CHECK: call void @__kmpc_restore_team_static_memory(i16 0)
 // CHECK: call void @__kmpc_kernel_deinit(i16 1)
 
 // CHECK: define internal void @__omp_outlined__(
diff --git a/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_codegen.cpp b/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_codegen.cpp
index 7cded3c5a3e..3a46886d8cd 100644
--- a/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_codegen.cpp
@@ -67,15 +67,24 @@ int bar(int n){
   return a;
 }
 
+// CHECK-DAG: [[MEM_TY:%.+]] = type { [4 x i8] }
+// CHECK-DAG: [[GLOBAL_RD:@.+]] = weak global [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]] zeroinitializer
+// CHECK-DAG: [[GLOBAL_RD_PTR:@.+]] = weak unnamed_addr constant i8* getelementptr inbounds ([{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]], [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]]* [[GLOBAL_RD]], i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0)
+// CHECK-DAG: [[KERNEL_PTR:@.+]] = internal addrspace(3) global i8* null
+// CHECK-DAG: [[KERNEL_SIZE:@.+]] = internal unnamed_addr constant i{{64|32}} 4
+
 // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}_l32(
 // CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
 // CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]], i16 0, i16 1)
-// CHECK: [[TEAM_ALLOC:%.+]] = call i8* @__kmpc_data_sharing_push_stack(i{{[0-9]+}} 4, i16 0)
+// CHECK: [[GLOBAL_RD:%.+]] = load i8*, i8** [[GLOBAL_RD_PTR]],
+// CHECK: call void @__kmpc_get_team_static_memory(i8* [[GLOBAL_RD]], i{{64|32}} 4, i16 0, i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
+// CHECK: [[TEAM_ALLOC:%.+]] = load i8*, i8* addrspace(3)* [[KERNEL_PTR]],
 // CHECK: [[BC:%.+]] = bitcast i8* [[TEAM_ALLOC]] to [[REC:%.+]]*
 // CHECK: getelementptr inbounds [[REC]], [[REC]]* [[BC]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
 // CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 91,
 // CHECK: {{call|invoke}} void [[OUTL1:@.+]](
 // CHECK: call void @__kmpc_for_static_fini(
+// CHECK: call void @__kmpc_restore_team_static_memory(i16 0)
 // CHECK: call void @__kmpc_spmd_kernel_deinit()
 // CHECK: ret void
 
diff --git a/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_simd_codegen.cpp b/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_simd_codegen.cpp
index 1555073cd0b..f7a3aaecc32 100644
--- a/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_simd_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_simd_codegen.cpp
@@ -62,15 +62,24 @@ int bar(int n){
   return a;
 }
 
+// CHECK-DAG: [[MEM_TY:%.+]] = type { [4 x i8] }
+// CHECK-DAG: [[GLOBAL_RD:@.+]] = weak global [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]] zeroinitializer
+// CHECK-DAG: [[GLOBAL_RD_PTR:@.+]] = weak unnamed_addr constant i8* getelementptr inbounds ([{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]], [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]]* [[GLOBAL_RD]], i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0)
+// CHECK-DAG: [[KERNEL_PTR:@.+]] = internal addrspace(3) global i8* null
+// CHECK-DAG: [[KERNEL_SIZE:@.+]] = internal unnamed_addr constant i{{64|32}} 4
+
 // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}_l30(
 // CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
 // CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]], i16 0, i16 1)
-// CHECK: [[TEAM_ALLOC:%.+]] = call i8* @__kmpc_data_sharing_push_stack(i{{[0-9]+}} 4, i16 0)
+// CHECK: [[GLOBAL_RD:%.+]] = load i8*, i8** [[GLOBAL_RD_PTR]],
+// CHECK: call void @__kmpc_get_team_static_memory(i8* [[GLOBAL_RD]], i{{64|32}} 4, i16 0, i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
+// CHECK: [[TEAM_ALLOC:%.+]] = load i8*, i8* addrspace(3)* [[KERNEL_PTR]],
 // CHECK: [[BC:%.+]] = bitcast i8* [[TEAM_ALLOC]] to [[REC:%.+]]*
 // CHECK: getelementptr inbounds [[REC]], [[REC]]* [[BC]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
 // CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 91,
 // CHECK: {{call|invoke}} void [[OUTL1:@.+]](
 // CHECK: call void @__kmpc_for_static_fini(
+// CHECK: call void @__kmpc_restore_team_static_memory(i16 0)
 // CHECK: call void @__kmpc_spmd_kernel_deinit()
 // CHECK: ret void
 
diff --git a/clang/test/OpenMP/nvptx_teams_codegen.cpp b/clang/test/OpenMP/nvptx_teams_codegen.cpp
index 4e3f2674fc5..af1ae40cdea 100644
--- a/clang/test/OpenMP/nvptx_teams_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_teams_codegen.cpp
@@ -27,6 +27,13 @@ int main (int argc, char **argv) {
   return tmain(argv);
 }
 
+// CK1: [[MEM_TY:%.+]] = type { [{{4|8}} x i8] }
+// CK1-DAG: [[GLOBAL_RD:@.+]] = weak global [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]] zeroinitializer
+// CK1-DAG: [[GLOBAL_RD_PTR:@.+]] = weak unnamed_addr constant i8* getelementptr inbounds ([{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]], [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]]* [[GLOBAL_RD]], i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0)
+// CK1-DAG: [[KERNEL_PTR:@.+]] = internal addrspace(3) global i8* null
+// CK1-DAG: [[KERNEL_SIZE1:@.+]] = internal unnamed_addr constant i{{64|32}} 4
+// CK1-DAG: [[KERNEL_SIZE2:@.+]] = internal unnamed_addr constant i{{64|32}} {{8|4}}
+
 // only nvptx side: do not outline teams region and do not call fork_teams
 // CK1:  define {{.*}}void @{{[^,]+}}(i{{[0-9]+}} [[ARGC:%.+]])
 // CK1:  {{.+}} = alloca i{{[0-9]+}}*,
@@ -36,7 +43,11 @@ int main (int argc, char **argv) {
 // CK1:  store {{.+}} 0, {{.+}},
 // CK1:  store i{{[0-9]+}} [[ARGC]], i{{[0-9]+}}* [[ARGCADDR]],
 // CK1-64:  [[CONV:%.+]] = bitcast i{{[0-9]+}}* [[ARGCADDR]] to i{{[0-9]+}}*
-// CK1:  call i8* @__kmpc_data_sharing_push_stack(i{{[0-9]+}} 4, i16 0)
+// CK1: [[GLOBAL_RD:%.+]] = load i8*, i8** [[GLOBAL_RD_PTR]],
+// CK1: [[SIZE:%.+]] = load i{{64|32}}, i{{64|32}}* [[KERNEL_SIZE1]],
+// CK1: call void @__kmpc_get_team_static_memory(i8* [[GLOBAL_RD]], i{{64|32}} [[SIZE]], i16 0, i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
+// CK1: [[KERNEL_RD:%.+]] = load i8*, i8* addrspace(3)* [[KERNEL_PTR]],
+// CK1: [[GLOBALSTACK:%.+]] = getelementptr inbounds i8, i8* [[KERNEL_RD]], i{{64|32}} 0
 // CK1-64:  [[ARG:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[CONV]]
 // CK1-32:  [[ARG:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[ARGCADDR]]
 // CK1:  [[ARGCADDR:%.+]] = getelementptr inbounds %struct.{{.*}}, %struct.{{.*}}* %{{.*}}, i{{[0-9]+}} 0, i{{[0-9]+}} 0
@@ -53,7 +64,11 @@ int main (int argc, char **argv) {
 // CK1: [[ARGCADDR_PTR:%.+]] = alloca i{{.+}}***,
 // CK1: [[ARGCADDR:%.+]] = alloca i{{.+}}**,
 // CK1: store i{{.+}}** [[ARGC]], i{{.+}}*** [[ARGCADDR]]
-// CK1: call i8* @__kmpc_data_sharing_push_stack(i{{[0-9]+}} {{4|8}}, i16 0)
+// CK1: [[GLOBAL_RD:%.+]] = load i8*, i8** [[GLOBAL_RD_PTR]],
+// CK1: [[SIZE:%.+]] = load i{{64|32}}, i{{64|32}}* [[KERNEL_SIZE2]],
+// CK1: call void @__kmpc_get_team_static_memory(i8* [[GLOBAL_RD]], i{{64|32}} [[SIZE]], i16 0, i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
+// CK1: [[KERNEL_RD:%.+]] = load i8*, i8* addrspace(3)* [[KERNEL_PTR]],
+// CK1: [[GLOBALSTACK:%.+]] = getelementptr inbounds i8, i8* [[KERNEL_RD]], i{{64|32}} 0
 // CK1: [[ARG:%.+]] = load i{{[0-9]+}}**, i{{[0-9]+}}*** [[ARGCADDR]]
 // CK1: [[ARGCADDR:%.+]] = getelementptr inbounds %struct.{{.*}}, %struct.{{.*}}* %{{.*}}, i{{[0-9]+}} 0, i{{[0-9]+}} 0
 // CK1: store i{{[0-9]+}}** [[ARG]], i{{[0-9]+}}*** [[ARGCADDR]],
@@ -98,6 +113,13 @@ int main (int argc, char **argv) {
   return tmain(argv);
 }
 
+// CK2: [[MEM_TY:%.+]] = type { [{{4|8}} x i8] }
+// CK2-DAG: [[GLOBAL_RD:@.+]] = weak global [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]] zeroinitializer
+// CK2-DAG: [[GLOBAL_RD_PTR:@.+]] = weak unnamed_addr constant i8* getelementptr inbounds ([{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]], [{{[0-9]+}} x [{{[0-9]+}} x [[MEM_TY]]]]* [[GLOBAL_RD]], i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0, i{{[0-9]+}} 0)
+// CK2-DAG: [[KERNEL_PTR:@.+]] = internal addrspace(3) global i8* null
+// CK2-DAG: [[KERNEL_SIZE1:@.+]] = internal unnamed_addr constant i{{64|32}} 4
+// CK2-DAG: [[KERNEL_SIZE2:@.+]] = internal unnamed_addr constant i{{64|32}} {{8|4}}
+
 // CK2: define {{.*}}void @{{[^,]+}}(i{{[0-9]+}} [[A_IN:%.+]], i{{[0-9]+}} [[B_IN:%.+]], i{{[0-9]+}} [[ARGC_IN:.+]])
 // CK2: {{.}} = alloca i{{[0-9]+}}*,
 // CK2: {{.}} = alloca i{{[0-9]+}}*,
@@ -111,7 +133,11 @@ int main (int argc, char **argv) {
 // CK2-64: [[ACONV:%.+]] = bitcast i64* [[AADDR]] to i32*
 // CK2-64: [[BCONV:%.+]] = bitcast i64* [[BADDR]] to i32*
 // CK2-64: [[CONV:%.+]] = bitcast i64* [[ARGCADDR]] to i32*
-// CK2:  call i8* @__kmpc_data_sharing_push_stack(i{{[0-9]+}} 4, i16 0)
+// CK2: [[GLOBAL_RD:%.+]] = load i8*, i8** [[GLOBAL_RD_PTR]],
+// CK2: [[SIZE:%.+]] = load i{{64|32}}, i{{64|32}}* [[KERNEL_SIZE1]],
+// CK2: call void @__kmpc_get_team_static_memory(i8* [[GLOBAL_RD]], i{{64|32}} [[SIZE]], i16 0, i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
+// CK2: [[KERNEL_RD:%.+]] = load i8*, i8* addrspace(3)* [[KERNEL_PTR]],
+// CK2: [[GLOBALSTACK:%.+]] = getelementptr inbounds i8, i8* [[KERNEL_RD]], i{{64|32}} 0
 // CK2-64:  [[ARG:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[CONV]]
 // CK2-32:  [[ARG:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[ARGCADDR]]
 // CK2:  [[ARGCADDR:%.+]] = getelementptr inbounds %struct.{{.*}}, %struct.{{.*}}* %{{.*}}, i{{[0-9]+}} 0, i{{[0-9]+}} 0
@@ -132,7 +158,11 @@ int main (int argc, char **argv) {
 // CK2: store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[AADDR]],
 // CK2: store i{{[0-9]+}} [[B_IN]], i{{[0-9]+}}* [[BADDR]],
 // CK2: store i{{[0-9]+}}** [[ARGC]], i{{[0-9]+}}*** [[ARGCADDR]],
-// CK2: call i8* @__kmpc_data_sharing_push_stack(i{{[0-9]+}} {{4|8}}, i16 0)
+// CK2: [[GLOBAL_RD:%.+]] = load i8*, i8** [[GLOBAL_RD_PTR]],
+// CK2: [[SIZE:%.+]] = load i{{64|32}}, i{{64|32}}* [[KERNEL_SIZE2]],
+// CK2: call void @__kmpc_get_team_static_memory(i8* [[GLOBAL_RD]], i{{64|32}} [[SIZE]], i16 0, i8** addrspacecast (i8* addrspace(3)* [[KERNEL_PTR]] to i8**))
+// CK2: [[KERNEL_RD:%.+]] = load i8*, i8* addrspace(3)* [[KERNEL_PTR]],
+// CK2: [[GLOBALSTACK:%.+]] = getelementptr inbounds i8, i8* [[KERNEL_RD]], i{{64|32}} 0
 // CK2: [[ARG:%.+]] = load i{{[0-9]+}}**, i{{[0-9]+}}*** [[ARGCADDR]]
 // CK2: [[ARGCADDR:%.+]] = getelementptr inbounds %struct.{{.*}}, %struct.{{.*}}* %{{.*}}, i{{[0-9]+}} 0, i{{[0-9]+}} 0
 // CK2: store i{{[0-9]+}}** [[ARG]], i{{[0-9]+}}*** [[ARGCADDR]],