144 changes: 72 additions & 72 deletions clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_codegen.cpp
View file

Large diffs are not rendered by default.

16 changes: 8 additions & 8 deletions clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_generic_mode_codegen.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -43,11 +43,11 @@ int main(int argc, char **argv) {
// CHECK1-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8
// CHECK1-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
// CHECK1-NEXT: [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false, i1 true)
// CHECK1-NEXT: [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
// CHECK1-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK1-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK1: user_code.entry:
// CHECK1-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB4:[0-9]+]])
// CHECK1-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK1-NEXT: [[TMP3:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
// CHECK1-NEXT: store i32 [[TMP3]], ptr [[ARGC_CASTED]], align 4
// CHECK1-NEXT: [[TMP4:%.*]] = load i64, ptr [[ARGC_CASTED]], align 8
Expand All @@ -57,7 +57,7 @@ int main(int argc, char **argv) {
// CHECK1-NEXT: store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK1-NEXT: store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK1-NEXT: call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i64 [[TMP4]], ptr [[TMP0]], i64 [[TMP6]]) #[[ATTR3:[0-9]+]]
// CHECK1-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2, i1 true)
// CHECK1-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK1-NEXT: ret void
// CHECK1: worker.exit:
// CHECK1-NEXT: ret void
Expand Down Expand Up @@ -160,7 +160,7 @@ int main(int argc, char **argv) {
// CHECK1-NEXT: store ptr [[TMP30]], ptr [[TMP29]], align 8
// CHECK1-NEXT: [[TMP31:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT: [[TMP32:%.*]] = load i32, ptr [[TMP31]], align 4
// CHECK1-NEXT: call void @__kmpc_parallel_51(ptr @[[GLOB4]], i32 [[TMP32]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__1, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 5)
// CHECK1-NEXT: call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP32]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__1, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 5)
// CHECK1-NEXT: br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK1: omp.inner.for.inc:
// CHECK1-NEXT: [[TMP33:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
Expand Down Expand Up @@ -341,11 +341,11 @@ int main(int argc, char **argv) {
// CHECK2-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4
// CHECK2-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4
// CHECK2-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
// CHECK2-NEXT: [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false, i1 true)
// CHECK2-NEXT: [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
// CHECK2-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
// CHECK2-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK2: user_code.entry:
// CHECK2-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB4:[0-9]+]])
// CHECK2-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK2-NEXT: [[TMP3:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
// CHECK2-NEXT: store i32 [[TMP3]], ptr [[ARGC_CASTED]], align 4
// CHECK2-NEXT: [[TMP4:%.*]] = load i32, ptr [[ARGC_CASTED]], align 4
Expand All @@ -355,7 +355,7 @@ int main(int argc, char **argv) {
// CHECK2-NEXT: store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK2-NEXT: store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK2-NEXT: call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], i32 [[TMP4]], ptr [[TMP0]], i32 [[TMP6]]) #[[ATTR3:[0-9]+]]
// CHECK2-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2, i1 true)
// CHECK2-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK2-NEXT: ret void
// CHECK2: worker.exit:
// CHECK2-NEXT: ret void
Expand Down Expand Up @@ -456,7 +456,7 @@ int main(int argc, char **argv) {
// CHECK2-NEXT: store ptr [[TMP28]], ptr [[TMP27]], align 4
// CHECK2-NEXT: [[TMP29:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK2-NEXT: [[TMP30:%.*]] = load i32, ptr [[TMP29]], align 4
// CHECK2-NEXT: call void @__kmpc_parallel_51(ptr @[[GLOB4]], i32 [[TMP30]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__1, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 5)
// CHECK2-NEXT: call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP30]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__1, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 5)
// CHECK2-NEXT: br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK2: omp.inner.for.inc:
// CHECK2-NEXT: [[TMP31:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
Expand Down
64 changes: 32 additions & 32 deletions clang/test/OpenMP/nvptx_target_teams_distribute_parallel_for_simd_codegen.cpp
View file

Large diffs are not rendered by default.

16 changes: 8 additions & 8 deletions clang/test/OpenMP/nvptx_target_teams_distribute_simd_codegen.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -70,33 +70,33 @@ int bar(int n){
}

// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}_l37(
// CHECK: call i32 @__kmpc_target_init({{.*}}, i8 2, i1 false, i1 true)
// CHECK: call void @__kmpc_target_deinit({{.*}}, i8 2, i1 true)
// CHECK: call i32 @__kmpc_target_init({{.*}}, i8 2, i1 false)
// CHECK: call void @__kmpc_target_deinit({{.*}}, i8 2)

// CHECK: call void @__kmpc_distribute_static_init_4({{.+}}, {{.+}}, {{.+}} 91,
// CHECK: call void @__kmpc_distribute_static_fini(
// CHECK: ret void

// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}_l43(
// CHECK: call i32 @__kmpc_target_init({{.*}}, i8 2, i1 false, i1 true)
// CHECK: call void @__kmpc_target_deinit({{.*}}, i8 2, i1 true)
// CHECK: call i32 @__kmpc_target_init({{.*}}, i8 2, i1 false)
// CHECK: call void @__kmpc_target_deinit({{.*}}, i8 2)

// CHECK: call void @__kmpc_distribute_static_init_4({{.+}}, {{.+}}, {{.+}} 91,
// CHECK: call void @__kmpc_distribute_static_fini(
// CHECK: ret void

// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}_l48(
// CHECK: call i32 @__kmpc_target_init({{.*}}, i8 2, i1 false, i1 true)
// CHECK: call void @__kmpc_target_deinit({{.*}}, i8 2, i1 true)
// CHECK: call i32 @__kmpc_target_init({{.*}}, i8 2, i1 false)
// CHECK: call void @__kmpc_target_deinit({{.*}}, i8 2)

// CHECK: call void @__kmpc_distribute_static_init_4({{.+}}, {{.+}}, {{.+}} 91,
// CHECK: call void @__kmpc_distribute_static_fini(
// CHECK: ret void

// CHECK: define {{.*}}void {{@__omp_offloading_.+}}_l53({{.+}}, i{{32|64}} [[F_IN:%.+]])
// CHECK: store {{.+}} [[F_IN]], ptr {{.+}},
// CHECK: call i32 @__kmpc_target_init({{.*}}, i8 2, i1 false, i1 true)
// CHECK: call void @__kmpc_target_deinit({{.*}}, i8 2, i1 true)
// CHECK: call i32 @__kmpc_target_init({{.*}}, i8 2, i1 false)
// CHECK: call void @__kmpc_target_deinit({{.*}}, i8 2)

// CHECK: store {{.+}} 99, ptr [[COMB_UB:%.+]], align
// CHECK: call void @__kmpc_distribute_static_init_4({{.+}}, {{.+}}, {{.+}} 91, {{.+}}, {{.+}}, ptr [[COMB_UB]],
Expand Down
48 changes: 24 additions & 24 deletions clang/test/OpenMP/nvptx_teams_codegen.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -83,7 +83,7 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK1-NEXT: store i64 [[ARGC]], ptr [[ARGC_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true, i1 true)
// CHECK1-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true)
// CHECK1-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
// CHECK1-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK1: user_code.entry:
Expand All @@ -95,7 +95,7 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK1-NEXT: call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[ARGC1]]) #[[ATTR3:[0-9]+]]
// CHECK1-NEXT: call void @__kmpc_free_shared(ptr [[ARGC1]], i64 4)
// CHECK1-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1, i1 true)
// CHECK1-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
// CHECK1-NEXT: ret void
// CHECK1: worker.exit:
// CHECK1-NEXT: ret void
Expand All @@ -122,7 +122,7 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK1-NEXT: store ptr [[ARGC]], ptr [[ARGC_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true, i1 true)
// CHECK1-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
// CHECK1-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
// CHECK1-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK1: user_code.entry:
Expand All @@ -134,7 +134,7 @@ int main (int argc, char **argv) {
// CHECK1-NEXT: store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK1-NEXT: call void @__omp_outlined__1(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[ARGC1]]) #[[ATTR3]]
// CHECK1-NEXT: call void @__kmpc_free_shared(ptr [[ARGC1]], i64 8)
// CHECK1-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1, i1 true)
// CHECK1-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
// CHECK1-NEXT: ret void
// CHECK1: worker.exit:
// CHECK1-NEXT: ret void
Expand All @@ -161,7 +161,7 @@ int main (int argc, char **argv) {
// CHECK2-NEXT: [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK2-NEXT: store i32 [[ARGC]], ptr [[ARGC_ADDR]], align 4
// CHECK2-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true, i1 true)
// CHECK2-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true)
// CHECK2-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
// CHECK2-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK2: user_code.entry:
Expand All @@ -173,7 +173,7 @@ int main (int argc, char **argv) {
// CHECK2-NEXT: store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK2-NEXT: call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[ARGC1]]) #[[ATTR3:[0-9]+]]
// CHECK2-NEXT: call void @__kmpc_free_shared(ptr [[ARGC1]], i32 4)
// CHECK2-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1, i1 true)
// CHECK2-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
// CHECK2-NEXT: ret void
// CHECK2: worker.exit:
// CHECK2-NEXT: ret void
Expand All @@ -200,7 +200,7 @@ int main (int argc, char **argv) {
// CHECK2-NEXT: [[DOTZERO_ADDR:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[DOTTHREADID_TEMP_:%.*]] = alloca i32, align 4
// CHECK2-NEXT: store ptr [[ARGC]], ptr [[ARGC_ADDR]], align 4
// CHECK2-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true, i1 true)
// CHECK2-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
// CHECK2-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
// CHECK2-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK2: user_code.entry:
Expand All @@ -212,7 +212,7 @@ int main (int argc, char **argv) {
// CHECK2-NEXT: store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK2-NEXT: call void @__omp_outlined__1(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[ARGC1]]) #[[ATTR3]]
// CHECK2-NEXT: call void @__kmpc_free_shared(ptr [[ARGC1]], i32 4)
// CHECK2-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1, i1 true)
// CHECK2-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
// CHECK2-NEXT: ret void
// CHECK2: worker.exit:
// CHECK2-NEXT: ret void
Expand Down Expand Up @@ -243,19 +243,19 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK3-NEXT: store i64 [[B]], ptr [[B_ADDR]], align 8
// CHECK3-NEXT: store i64 [[ARGC]], ptr [[ARGC_ADDR]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true, i1 true)
// CHECK3-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true)
// CHECK3-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
// CHECK3-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK3: user_code.entry:
// CHECK3-NEXT: [[TMP1:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4
// CHECK3-NEXT: [[ARGC3:%.*]] = call align 8 ptr @__kmpc_alloc_shared(i64 4)
// CHECK3-NEXT: store i32 [[TMP1]], ptr [[ARGC3]], align 4
// CHECK3-NEXT: [[ARGC1:%.*]] = call align 8 ptr @__kmpc_alloc_shared(i64 4)
// CHECK3-NEXT: store i32 [[TMP1]], ptr [[ARGC1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK3-NEXT: store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK3-NEXT: store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK3-NEXT: call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[ARGC3]]) #[[ATTR3:[0-9]+]]
// CHECK3-NEXT: call void @__kmpc_free_shared(ptr [[ARGC3]], i64 4)
// CHECK3-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1, i1 true)
// CHECK3-NEXT: call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[ARGC1]]) #[[ATTR3:[0-9]+]]
// CHECK3-NEXT: call void @__kmpc_free_shared(ptr [[ARGC1]], i64 4)
// CHECK3-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
// CHECK3-NEXT: ret void
// CHECK3: worker.exit:
// CHECK3-NEXT: ret void
Expand Down Expand Up @@ -286,19 +286,19 @@ int main (int argc, char **argv) {
// CHECK3-NEXT: store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK3-NEXT: store i64 [[B]], ptr [[B_ADDR]], align 8
// CHECK3-NEXT: store ptr [[ARGC]], ptr [[ARGC_ADDR]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true, i1 true)
// CHECK3-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
// CHECK3-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
// CHECK3-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK3: user_code.entry:
// CHECK3-NEXT: [[TMP1:%.*]] = load ptr, ptr [[ARGC_ADDR]], align 8
// CHECK3-NEXT: [[ARGC2:%.*]] = call align 8 ptr @__kmpc_alloc_shared(i64 8)
// CHECK3-NEXT: store ptr [[TMP1]], ptr [[ARGC2]], align 8
// CHECK3-NEXT: [[ARGC1:%.*]] = call align 8 ptr @__kmpc_alloc_shared(i64 8)
// CHECK3-NEXT: store ptr [[TMP1]], ptr [[ARGC1]], align 8
// CHECK3-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
// CHECK3-NEXT: store i32 0, ptr [[DOTZERO_ADDR]], align 4
// CHECK3-NEXT: store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK3-NEXT: call void @__omp_outlined__1(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[ARGC2]]) #[[ATTR3]]
// CHECK3-NEXT: call void @__kmpc_free_shared(ptr [[ARGC2]], i64 8)
// CHECK3-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1, i1 true)
// CHECK3-NEXT: call void @__omp_outlined__1(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[ARGC1]]) #[[ATTR3]]
// CHECK3-NEXT: call void @__kmpc_free_shared(ptr [[ARGC1]], i64 8)
// CHECK3-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
// CHECK3-NEXT: ret void
// CHECK3: worker.exit:
// CHECK3-NEXT: ret void
Expand Down Expand Up @@ -329,7 +329,7 @@ int main (int argc, char **argv) {
// CHECK4-NEXT: store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK4-NEXT: store i32 [[B]], ptr [[B_ADDR]], align 4
// CHECK4-NEXT: store i32 [[ARGC]], ptr [[ARGC_ADDR]], align 4
// CHECK4-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true, i1 true)
// CHECK4-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 1, i1 true)
// CHECK4-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
// CHECK4-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK4: user_code.entry:
Expand All @@ -341,7 +341,7 @@ int main (int argc, char **argv) {
// CHECK4-NEXT: store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK4-NEXT: call void @__omp_outlined__(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[ARGC1]]) #[[ATTR3:[0-9]+]]
// CHECK4-NEXT: call void @__kmpc_free_shared(ptr [[ARGC1]], i32 4)
// CHECK4-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1, i1 true)
// CHECK4-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
// CHECK4-NEXT: ret void
// CHECK4: worker.exit:
// CHECK4-NEXT: ret void
Expand Down Expand Up @@ -372,7 +372,7 @@ int main (int argc, char **argv) {
// CHECK4-NEXT: store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK4-NEXT: store i32 [[B]], ptr [[B_ADDR]], align 4
// CHECK4-NEXT: store ptr [[ARGC]], ptr [[ARGC_ADDR]], align 4
// CHECK4-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true, i1 true)
// CHECK4-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1]], i8 1, i1 true)
// CHECK4-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
// CHECK4-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK4: user_code.entry:
Expand All @@ -384,7 +384,7 @@ int main (int argc, char **argv) {
// CHECK4-NEXT: store i32 [[TMP2]], ptr [[DOTTHREADID_TEMP_]], align 4
// CHECK4-NEXT: call void @__omp_outlined__1(ptr [[DOTTHREADID_TEMP_]], ptr [[DOTZERO_ADDR]], ptr [[ARGC1]]) #[[ATTR3]]
// CHECK4-NEXT: call void @__kmpc_free_shared(ptr [[ARGC1]], i32 4)
// CHECK4-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1, i1 true)
// CHECK4-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 1)
// CHECK4-NEXT: ret void
// CHECK4: worker.exit:
// CHECK4-NEXT: ret void
Expand Down
3,012 changes: 1,506 additions & 1,506 deletions clang/test/OpenMP/nvptx_teams_reduction_codegen.cpp
View file

Large diffs are not rendered by default.

4 changes: 2 additions & 2 deletions clang/test/OpenMP/reduction_implicit_map.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -103,7 +103,7 @@ int main()
// CHECK-NEXT: [[E_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT: [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 8
// CHECK-NEXT: store ptr [[E]], ptr [[E_ADDR]], align 8
// CHECK-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false, i1 true)
// CHECK-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
// CHECK-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
// CHECK-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
// CHECK: user_code.entry:
Expand All @@ -112,7 +112,7 @@ int main()
// CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
// CHECK-NEXT: store ptr [[TMP2]], ptr [[TMP3]], align 8
// CHECK-NEXT: call void @__kmpc_parallel_51(ptr @[[GLOB2]], i32 [[TMP1]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 1)
// CHECK-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2, i1 true)
// CHECK-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
// CHECK-NEXT: ret void
// CHECK: worker.exit:
// CHECK-NEXT: ret void
Expand Down
2 changes: 1 addition & 1 deletion clang/test/OpenMP/remarks_parallel_in_multiple_target_state_machines.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -86,7 +86,7 @@ void spmd(void) {
#pragma omp begin declare target device_type(nohost)
__attribute__((weak))
extern "C" int __kmpc_target_init(void *Ident, char Mode,
bool UseGenericStateMachine, bool) { // all-remark {{Could not internalize function. Some optimizations may not be possible. [OMP140]}}
bool UseGenericStateMachine) { // all-remark {{Could not internalize function. Some optimizations may not be possible. [OMP140]}}
return 0;
}
#pragma omp end declare target
Expand Down
2 changes: 1 addition & 1 deletion clang/test/OpenMP/remarks_parallel_in_target_state_machine.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -42,7 +42,7 @@ void spmd(void) {
#pragma omp begin declare target device_type(nohost)
__attribute__((weak))
extern "C" int __kmpc_target_init(void *Ident, char Mode,
bool UseGenericStateMachine, bool) { // expected-remark {{Could not internalize function. Some optimizations may not be possible. [OMP140]}}
bool UseGenericStateMachine) { // expected-remark {{Could not internalize function. Some optimizations may not be possible. [OMP140]}}
return 0;
}
#pragma omp end declare target
Expand Down
590 changes: 295 additions & 295 deletions clang/test/OpenMP/target_has_device_addr_codegen.cpp
View file

Large diffs are not rendered by default.

84 changes: 42 additions & 42 deletions clang/test/OpenMP/target_parallel_debug_codegen.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -95,7 +95,7 @@ int main() {
// CHECK1-NEXT: [[TMP6:%.*]] = addrspacecast ptr addrspace(1) [[TMP5]] to ptr, !dbg [[DBG47]]
// CHECK1-NEXT: store ptr [[TMP6]], ptr [[_TMP2]], align 8, !dbg [[DBG47]]
// CHECK1-NEXT: [[TMP7:%.*]] = load ptr, ptr [[_TMP2]], align 8, !dbg [[DBG47]]
// CHECK1-NEXT: [[TMP8:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false, i1 true), !dbg [[DBG47]]
// CHECK1-NEXT: [[TMP8:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false), !dbg [[DBG47]]
// CHECK1-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP8]], -1, !dbg [[DBG47]]
// CHECK1-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]], !dbg [[DBG47]]
// CHECK1: user_code.entry:
Expand All @@ -105,15 +105,15 @@ int main() {
// CHECK1-NEXT: [[TMP11:%.*]] = load i64, ptr [[A_CASTED]], align 8, !dbg [[DBG48]]
// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 0, !dbg [[DBG48]]
// CHECK1-NEXT: store ptr [[TMP2]], ptr [[TMP12]], align 8, !dbg [[DBG48]]
// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 1, !dbg [[DBG48]]
// CHECK1-NEXT: [[TMP15:%.*]] = inttoptr i64 [[TMP11]] to ptr, !dbg [[DBG48]]
// CHECK1-NEXT: store ptr [[TMP15]], ptr [[TMP14]], align 8, !dbg [[DBG48]]
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 2, !dbg [[DBG48]]
// CHECK1-NEXT: store ptr [[TMP4]], ptr [[TMP16]], align 8, !dbg [[DBG48]]
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 3, !dbg [[DBG48]]
// CHECK1-NEXT: store ptr [[TMP7]], ptr [[TMP18]], align 8, !dbg [[DBG48]]
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 1, !dbg [[DBG48]]
// CHECK1-NEXT: [[TMP14:%.*]] = inttoptr i64 [[TMP11]] to ptr, !dbg [[DBG48]]
// CHECK1-NEXT: store ptr [[TMP14]], ptr [[TMP13]], align 8, !dbg [[DBG48]]
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 2, !dbg [[DBG48]]
// CHECK1-NEXT: store ptr [[TMP4]], ptr [[TMP15]], align 8, !dbg [[DBG48]]
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 3, !dbg [[DBG48]]
// CHECK1-NEXT: store ptr [[TMP7]], ptr [[TMP16]], align 8, !dbg [[DBG48]]
// CHECK1-NEXT: call void @__kmpc_parallel_51(ptr @[[GLOB3]], i32 [[TMP9]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 4), !dbg [[DBG48]]
// CHECK1-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB5:[0-9]+]], i8 2, i1 true), !dbg [[DBG49:![0-9]+]]
// CHECK1-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB5:[0-9]+]], i8 2), !dbg [[DBG49:![0-9]+]]
// CHECK1-NEXT: ret void, !dbg [[DBG51:![0-9]+]]
// CHECK1: worker.exit:
// CHECK1-NEXT: ret void, !dbg [[DBG47]]
Expand Down Expand Up @@ -176,36 +176,36 @@ int main() {
// CHECK1-NEXT: store i32 15, ptr [[D]], align 4, !dbg [[DBG83]]
// CHECK1-NEXT: store i32 5, ptr [[A_ADDR]], align 4, !dbg [[DBG84:![0-9]+]]
// CHECK1-NEXT: [[ARRAYIDX8:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[B3]], i64 0, i64 0, !dbg [[DBG85:![0-9]+]]
// CHECK1-NEXT: [[TMP10:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG86:![0-9]+]]
// CHECK1-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP10]] to i64, !dbg [[DBG85]]
// CHECK1-NEXT: [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG86:![0-9]+]]
// CHECK1-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP8]] to i64, !dbg [[DBG85]]
// CHECK1-NEXT: [[ARRAYIDX9:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX8]], i64 0, i64 [[IDXPROM]], !dbg [[DBG85]]
// CHECK1-NEXT: store i32 10, ptr [[ARRAYIDX9]], align 4, !dbg [[DBG87:![0-9]+]]
// CHECK1-NEXT: [[ARRAYIDX10:%.*]] = getelementptr inbounds [10 x [10 x [10 x i32]]], ptr [[TMP2]], i64 0, i64 0, !dbg [[DBG88:![0-9]+]]
// CHECK1-NEXT: [[ARRAYIDX11:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[ARRAYIDX10]], i64 0, i64 0, !dbg [[DBG88]]
// CHECK1-NEXT: [[TMP11:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG89:![0-9]+]]
// CHECK1-NEXT: [[IDXPROM12:%.*]] = sext i32 [[TMP11]] to i64, !dbg [[DBG88]]
// CHECK1-NEXT: [[TMP9:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG89:![0-9]+]]
// CHECK1-NEXT: [[IDXPROM12:%.*]] = sext i32 [[TMP9]] to i64, !dbg [[DBG88]]
// CHECK1-NEXT: [[ARRAYIDX13:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX11]], i64 0, i64 [[IDXPROM12]], !dbg [[DBG88]]
// CHECK1-NEXT: store i32 11, ptr [[ARRAYIDX13]], align 4, !dbg [[DBG90:![0-9]+]]
// CHECK1-NEXT: [[ARRAYIDX14:%.*]] = getelementptr inbounds [10 x [10 x [10 x i32]]], ptr [[TMP2]], i64 0, i64 0, !dbg [[DBG91:![0-9]+]]
// CHECK1-NEXT: [[ARRAYIDX15:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[ARRAYIDX14]], i64 0, i64 0, !dbg [[DBG91]]
// CHECK1-NEXT: [[TMP12:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG92:![0-9]+]]
// CHECK1-NEXT: [[IDXPROM16:%.*]] = sext i32 [[TMP12]] to i64, !dbg [[DBG91]]
// CHECK1-NEXT: [[TMP10:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG92:![0-9]+]]
// CHECK1-NEXT: [[IDXPROM16:%.*]] = sext i32 [[TMP10]] to i64, !dbg [[DBG91]]
// CHECK1-NEXT: [[ARRAYIDX17:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX15]], i64 0, i64 [[IDXPROM16]], !dbg [[DBG91]]
// CHECK1-NEXT: [[TMP13:%.*]] = load i32, ptr [[ARRAYIDX17]], align 4, !dbg [[DBG91]]
// CHECK1-NEXT: [[TMP11:%.*]] = load i32, ptr [[ARRAYIDX17]], align 4, !dbg [[DBG91]]
// CHECK1-NEXT: [[ARRAYIDX18:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[B3]], i64 0, i64 0, !dbg [[DBG93:![0-9]+]]
// CHECK1-NEXT: [[TMP14:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG94:![0-9]+]]
// CHECK1-NEXT: [[IDXPROM19:%.*]] = sext i32 [[TMP14]] to i64, !dbg [[DBG93]]
// CHECK1-NEXT: [[TMP12:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG94:![0-9]+]]
// CHECK1-NEXT: [[IDXPROM19:%.*]] = sext i32 [[TMP12]] to i64, !dbg [[DBG93]]
// CHECK1-NEXT: [[ARRAYIDX20:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX18]], i64 0, i64 [[IDXPROM19]], !dbg [[DBG93]]
// CHECK1-NEXT: store i32 [[TMP13]], ptr [[ARRAYIDX20]], align 4, !dbg [[DBG95:![0-9]+]]
// CHECK1-NEXT: store i32 [[TMP11]], ptr [[ARRAYIDX20]], align 4, !dbg [[DBG95:![0-9]+]]
// CHECK1-NEXT: [[ARRAYIDX21:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr [[B3]], i64 0, i64 0, !dbg [[DBG96:![0-9]+]]
// CHECK1-NEXT: [[TMP15:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG97:![0-9]+]]
// CHECK1-NEXT: [[IDXPROM22:%.*]] = sext i32 [[TMP15]] to i64, !dbg [[DBG96]]
// CHECK1-NEXT: [[TMP13:%.*]] = load i32, ptr [[A_ADDR]], align 4, !dbg [[DBG97:![0-9]+]]
// CHECK1-NEXT: [[IDXPROM22:%.*]] = sext i32 [[TMP13]] to i64, !dbg [[DBG96]]
// CHECK1-NEXT: [[ARRAYIDX23:%.*]] = getelementptr inbounds [10 x i32], ptr [[ARRAYIDX21]], i64 0, i64 [[IDXPROM22]], !dbg [[DBG96]]
// CHECK1-NEXT: [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX23]], align 4, !dbg [[DBG96]]
// CHECK1-NEXT: [[TMP17:%.*]] = load i8, ptr [[TMP7]], align 1, !dbg [[DBG98:![0-9]+]]
// CHECK1-NEXT: [[TOBOOL:%.*]] = trunc i8 [[TMP17]] to i1, !dbg [[DBG98]]
// CHECK1-NEXT: [[TMP14:%.*]] = load i32, ptr [[ARRAYIDX23]], align 4, !dbg [[DBG96]]
// CHECK1-NEXT: [[TMP15:%.*]] = load i8, ptr [[TMP7]], align 1, !dbg [[DBG98:![0-9]+]]
// CHECK1-NEXT: [[TOBOOL:%.*]] = trunc i8 [[TMP15]] to i1, !dbg [[DBG98]]
// CHECK1-NEXT: [[CONV:%.*]] = zext i1 [[TOBOOL]] to i32, !dbg [[DBG98]]
// CHECK1-NEXT: [[OR:%.*]] = or i32 [[CONV]], [[TMP16]], !dbg [[DBG98]]
// CHECK1-NEXT: [[OR:%.*]] = or i32 [[CONV]], [[TMP14]], !dbg [[DBG98]]
// CHECK1-NEXT: [[TOBOOL24:%.*]] = icmp ne i32 [[OR]], 0, !dbg [[DBG98]]
// CHECK1-NEXT: [[FROMBOOL:%.*]] = zext i1 [[TOBOOL24]] to i8, !dbg [[DBG98]]
// CHECK1-NEXT: store i8 [[FROMBOOL]], ptr [[TMP7]], align 1, !dbg [[DBG98]]
Expand Down Expand Up @@ -308,7 +308,7 @@ int main() {
// CHECK1-NEXT: [[TMP7:%.*]] = addrspacecast ptr addrspace(1) [[TMP6]] to ptr, !dbg [[DBG137]]
// CHECK1-NEXT: store ptr [[TMP7]], ptr [[_TMP2]], align 8, !dbg [[DBG137]]
// CHECK1-NEXT: [[TMP8:%.*]] = load ptr, ptr [[_TMP2]], align 8, !dbg [[DBG137]]
// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB7:[0-9]+]], i8 2, i1 false, i1 true), !dbg [[DBG137]]
// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB7:[0-9]+]], i8 2, i1 false), !dbg [[DBG137]]
// CHECK1-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP9]], -1, !dbg [[DBG137]]
// CHECK1-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]], !dbg [[DBG137]]
// CHECK1: user_code.entry:
Expand All @@ -318,15 +318,15 @@ int main() {
// CHECK1-NEXT: [[TMP12:%.*]] = load i64, ptr [[A_CASTED]], align 8, !dbg [[DBG138]]
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 0, !dbg [[DBG138]]
// CHECK1-NEXT: store ptr [[TMP2]], ptr [[TMP13]], align 8, !dbg [[DBG138]]
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 1, !dbg [[DBG138]]
// CHECK1-NEXT: [[TMP16:%.*]] = inttoptr i64 [[TMP12]] to ptr, !dbg [[DBG138]]
// CHECK1-NEXT: store ptr [[TMP16]], ptr [[TMP15]], align 8, !dbg [[DBG138]]
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 2, !dbg [[DBG138]]
// CHECK1-NEXT: store ptr [[TMP5]], ptr [[TMP17]], align 8, !dbg [[DBG138]]
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 3, !dbg [[DBG138]]
// CHECK1-NEXT: store ptr [[TMP8]], ptr [[TMP19]], align 8, !dbg [[DBG138]]
// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 1, !dbg [[DBG138]]
// CHECK1-NEXT: [[TMP15:%.*]] = inttoptr i64 [[TMP12]] to ptr, !dbg [[DBG138]]
// CHECK1-NEXT: store ptr [[TMP15]], ptr [[TMP14]], align 8, !dbg [[DBG138]]
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 2, !dbg [[DBG138]]
// CHECK1-NEXT: store ptr [[TMP5]], ptr [[TMP16]], align 8, !dbg [[DBG138]]
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 3, !dbg [[DBG138]]
// CHECK1-NEXT: store ptr [[TMP8]], ptr [[TMP17]], align 8, !dbg [[DBG138]]
// CHECK1-NEXT: call void @__kmpc_parallel_51(ptr @[[GLOB9]], i32 [[TMP10]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__2, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 4), !dbg [[DBG138]]
// CHECK1-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB11:[0-9]+]], i8 2, i1 true), !dbg [[DBG139:![0-9]+]]
// CHECK1-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB11:[0-9]+]], i8 2), !dbg [[DBG139:![0-9]+]]
// CHECK1-NEXT: ret void, !dbg [[DBG141:![0-9]+]]
// CHECK1: worker.exit:
// CHECK1-NEXT: ret void, !dbg [[DBG137]]
Expand Down Expand Up @@ -517,21 +517,21 @@ int main() {
// CHECK1-NEXT: [[TMP10:%.*]] = addrspacecast ptr addrspace(1) [[TMP9]] to ptr, !dbg [[DBG212]]
// CHECK1-NEXT: store ptr [[TMP10]], ptr [[_TMP3]], align 8, !dbg [[DBG212]]
// CHECK1-NEXT: [[TMP11:%.*]] = load ptr, ptr [[_TMP3]], align 8, !dbg [[DBG212]]
// CHECK1-NEXT: [[TMP12:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB13:[0-9]+]], i8 2, i1 false, i1 true), !dbg [[DBG212]]
// CHECK1-NEXT: [[TMP12:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB13:[0-9]+]], i8 2, i1 false), !dbg [[DBG212]]
// CHECK1-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP12]], -1, !dbg [[DBG212]]
// CHECK1-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]], !dbg [[DBG212]]
// CHECK1: user_code.entry:
// CHECK1-NEXT: [[TMP13:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB15:[0-9]+]])
// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 0, !dbg [[DBG213:![0-9]+]]
// CHECK1-NEXT: store ptr [[TMP2]], ptr [[TMP14]], align 8, !dbg [[DBG213]]
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 1, !dbg [[DBG213]]
// CHECK1-NEXT: store ptr [[TMP5]], ptr [[TMP16]], align 8, !dbg [[DBG213]]
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 2, !dbg [[DBG213]]
// CHECK1-NEXT: store ptr [[TMP8]], ptr [[TMP18]], align 8, !dbg [[DBG213]]
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 3, !dbg [[DBG213]]
// CHECK1-NEXT: store ptr [[TMP11]], ptr [[TMP20]], align 8, !dbg [[DBG213]]
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 1, !dbg [[DBG213]]
// CHECK1-NEXT: store ptr [[TMP5]], ptr [[TMP15]], align 8, !dbg [[DBG213]]
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 2, !dbg [[DBG213]]
// CHECK1-NEXT: store ptr [[TMP8]], ptr [[TMP16]], align 8, !dbg [[DBG213]]
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 3, !dbg [[DBG213]]
// CHECK1-NEXT: store ptr [[TMP11]], ptr [[TMP17]], align 8, !dbg [[DBG213]]
// CHECK1-NEXT: call void @__kmpc_parallel_51(ptr @[[GLOB15]], i32 [[TMP13]], i32 1, i32 -1, i32 -1, ptr @__omp_outlined__4, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 4), !dbg [[DBG213]]
// CHECK1-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB17:[0-9]+]], i8 2, i1 true), !dbg [[DBG214:![0-9]+]]
// CHECK1-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB17:[0-9]+]], i8 2), !dbg [[DBG214:![0-9]+]]
// CHECK1-NEXT: ret void, !dbg [[DBG216:![0-9]+]]
// CHECK1: worker.exit:
// CHECK1-NEXT: ret void, !dbg [[DBG212]]
Expand Down
144 changes: 72 additions & 72 deletions clang/test/OpenMP/target_parallel_for_debug_codegen.cpp
View file

Large diffs are not rendered by default.

404 changes: 202 additions & 202 deletions clang/test/OpenMP/target_parallel_for_reduction_task_codegen.cpp
View file

Large diffs are not rendered by default.

370 changes: 185 additions & 185 deletions clang/test/OpenMP/target_parallel_reduction_task_codegen.cpp
View file

Large diffs are not rendered by default.

634 changes: 317 additions & 317 deletions clang/test/OpenMP/target_teams_distribute_collapse_codegen.cpp
View file

Large diffs are not rendered by default.

682 changes: 341 additions & 341 deletions clang/test/OpenMP/target_teams_distribute_parallel_for_collapse_codegen.cpp
View file

Large diffs are not rendered by default.

124 changes: 62 additions & 62 deletions clang/test/OpenMP/target_teams_distribute_parallel_for_if_codegen.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -499,37 +499,37 @@ int main() {
// CHECK1: omp_if.then:
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK1-NEXT: store i64 [[TMP13]], ptr [[TMP15]], align 8
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: store i64 [[TMP13]], ptr [[TMP17]], align 8
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
// CHECK1-NEXT: store ptr null, ptr [[TMP19]], align 8
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[TMP22:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
// CHECK1-NEXT: [[TOBOOL4:%.*]] = trunc i8 [[TMP22]] to i1
// CHECK1-NEXT: [[TMP23:%.*]] = select i1 [[TOBOOL4]], i32 0, i32 1
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: store i64 [[TMP13]], ptr [[TMP16]], align 8
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
// CHECK1-NEXT: store ptr null, ptr [[TMP17]], align 8
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[TMP20:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
// CHECK1-NEXT: [[TOBOOL4:%.*]] = trunc i8 [[TMP20]] to i1
// CHECK1-NEXT: [[TMP21:%.*]] = select i1 [[TOBOOL4]], i32 0, i32 1
// CHECK1-NEXT: [[KERNEL_ARGS6:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
// CHECK1-NEXT: store i32 1, ptr [[TMP24]], align 4
// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP25]], align 4
// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP26]], align 8
// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
// CHECK1-NEXT: store ptr [[TMP21]], ptr [[TMP27]], align 8
// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP28]], align 8
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP29]], align 8
// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
// CHECK1-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
// CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8
// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP32]], align 8
// CHECK1-NEXT: [[TMP33:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP23]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l90.region_id, ptr [[KERNEL_ARGS6]])
// CHECK1-NEXT: [[TMP34:%.*]] = icmp ne i32 [[TMP33]], 0
// CHECK1-NEXT: br i1 [[TMP34]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 0
// CHECK1-NEXT: store i32 1, ptr [[TMP22]], align 4
// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP23]], align 4
// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 2
// CHECK1-NEXT: store ptr [[TMP18]], ptr [[TMP24]], align 8
// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 3
// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP25]], align 8
// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 4
// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP26]], align 8
// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 5
// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP27]], align 8
// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 6
// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 7
// CHECK1-NEXT: store ptr null, ptr [[TMP29]], align 8
// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS6]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP30]], align 8
// CHECK1-NEXT: [[TMP31:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP21]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l90.region_id, ptr [[KERNEL_ARGS6]])
// CHECK1-NEXT: [[TMP32:%.*]] = icmp ne i32 [[TMP31]], 0
// CHECK1-NEXT: br i1 [[TMP32]], label [[OMP_OFFLOAD_FAILED7:%.*]], label [[OMP_OFFLOAD_CONT8:%.*]]
// CHECK1: omp_offload.failed7:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l90(i64 [[TMP13]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT8]]
Expand All @@ -539,8 +539,8 @@ int main() {
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l90(i64 [[TMP13]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
// CHECK1-NEXT: [[TMP35:%.*]] = load i32, ptr @Arg, align 4
// CHECK1-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP35]])
// CHECK1-NEXT: [[TMP33:%.*]] = load i32, ptr @Arg, align 4
// CHECK1-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiEiT_(i32 noundef [[TMP33]])
// CHECK1-NEXT: ret i32 [[CALL]]
//
//
Expand Down Expand Up @@ -1037,37 +1037,37 @@ int main() {
// CHECK1-NEXT: [[TMP13:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK1-NEXT: store i64 [[TMP13]], ptr [[TMP14]], align 8
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: store i64 [[TMP13]], ptr [[TMP16]], align 8
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[TMP21:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
// CHECK1-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP21]] to i1
// CHECK1-NEXT: [[TMP22:%.*]] = select i1 [[TOBOOL3]], i32 0, i32 1
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: store i64 [[TMP13]], ptr [[TMP15]], align 8
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[TMP19:%.*]] = load i8, ptr [[DOTCAPTURE_EXPR_]], align 1
// CHECK1-NEXT: [[TOBOOL3:%.*]] = trunc i8 [[TMP19]] to i1
// CHECK1-NEXT: [[TMP20:%.*]] = select i1 [[TOBOOL3]], i32 0, i32 1
// CHECK1-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
// CHECK1-NEXT: store i32 1, ptr [[TMP23]], align 4
// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP24]], align 4
// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP25]], align 8
// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP26]], align 8
// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
// CHECK1-NEXT: store ptr @.offload_sizes.16, ptr [[TMP27]], align 8
// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
// CHECK1-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP28]], align 8
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
// CHECK1-NEXT: store ptr null, ptr [[TMP29]], align 8
// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
// CHECK1-NEXT: store ptr null, ptr [[TMP30]], align 8
// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP31]], align 8
// CHECK1-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP22]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l68.region_id, ptr [[KERNEL_ARGS5]])
// CHECK1-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
// CHECK1-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0
// CHECK1-NEXT: store i32 1, ptr [[TMP21]], align 4
// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1
// CHECK1-NEXT: store i32 1, ptr [[TMP22]], align 4
// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2
// CHECK1-NEXT: store ptr [[TMP17]], ptr [[TMP23]], align 8
// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3
// CHECK1-NEXT: store ptr [[TMP18]], ptr [[TMP24]], align 8
// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4
// CHECK1-NEXT: store ptr @.offload_sizes.16, ptr [[TMP25]], align 8
// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5
// CHECK1-NEXT: store ptr @.offload_maptypes.17, ptr [[TMP26]], align 8
// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6
// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7
// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8
// CHECK1-NEXT: store i64 100, ptr [[TMP29]], align 8
// CHECK1-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 [[TMP20]], ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l68.region_id, ptr [[KERNEL_ARGS5]])
// CHECK1-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
// CHECK1-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]]
// CHECK1: omp_offload.failed6:
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiEiT__l68(i64 [[TMP13]]) #[[ATTR2]]
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT7]]
Expand Down
12 changes: 6 additions & 6 deletions clang/test/OpenMP/target_teams_distribute_parallel_for_order_codegen.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -167,22 +167,22 @@ void gtid_test() {
// CHECK1-NEXT: store i32 [[TMP6]], ptr [[DOTOMP_IV]], align 4
// CHECK1-NEXT: br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK1: omp.inner.for.cond:
// CHECK1-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group !4
// CHECK1-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group !4
// CHECK1-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP4:![0-9]+]]
// CHECK1-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP4]]
// CHECK1-NEXT: [[CMP2:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]
// CHECK1-NEXT: br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK1: omp.inner.for.body:
// CHECK1-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group !4
// CHECK1-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP4]]
// CHECK1-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1
// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK1-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group !4
// CHECK1-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP4]]
// CHECK1-NEXT: br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK1: omp.body.continue:
// CHECK1-NEXT: br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK1: omp.inner.for.inc:
// CHECK1-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group !4
// CHECK1-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP4]]
// CHECK1-NEXT: [[ADD3:%.*]] = add nsw i32 [[TMP10]], 1
// CHECK1-NEXT: store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group !4
// CHECK1-NEXT: store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP4]]
// CHECK1-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
// CHECK1: omp.inner.for.end:
// CHECK1-NEXT: br label [[OMP_LOOP_EXIT:%.*]]
Expand Down
520 changes: 260 additions & 260 deletions clang/test/OpenMP/target_teams_distribute_parallel_for_reduction_codegen.cpp
View file

Large diffs are not rendered by default.

694 changes: 347 additions & 347 deletions clang/test/OpenMP/target_teams_distribute_parallel_for_reduction_task_codegen.cpp
View file

Large diffs are not rendered by default.

1,134 changes: 567 additions & 567 deletions clang/test/OpenMP/target_teams_distribute_parallel_for_simd_collapse_codegen.cpp
View file

Large diffs are not rendered by default.

126 changes: 63 additions & 63 deletions clang/test/OpenMP/target_teams_distribute_parallel_for_simd_proc_bind_codegen.cpp
View file

Large diffs are not rendered by default.

802 changes: 401 additions & 401 deletions clang/test/OpenMP/target_teams_distribute_parallel_for_simd_reduction_codegen.cpp
View file

Large diffs are not rendered by default.

370 changes: 185 additions & 185 deletions clang/test/OpenMP/target_teams_distribute_reduction_codegen.cpp
View file

Large diffs are not rendered by default.

950 changes: 475 additions & 475 deletions clang/test/OpenMP/target_teams_distribute_simd_collapse_codegen.cpp
View file

Large diffs are not rendered by default.

572 changes: 286 additions & 286 deletions clang/test/OpenMP/target_teams_distribute_simd_reduction_codegen.cpp
View file

Large diffs are not rendered by default.

2,200 changes: 1,100 additions & 1,100 deletions clang/test/OpenMP/target_teams_map_codegen.cpp
View file

Large diffs are not rendered by default.

8 changes: 2 additions & 6 deletions llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1458,17 +1458,13 @@ class OpenMPIRBuilder {
///
/// \param Loc The insert and source location description.
/// \param IsSPMD Flag to indicate if the kernel is an SPMD kernel or not.
/// \param RequiresFullRuntime Indicate if a full device runtime is necessary.
InsertPointTy createTargetInit(const LocationDescription &Loc, bool IsSPMD,
bool RequiresFullRuntime);
InsertPointTy createTargetInit(const LocationDescription &Loc, bool IsSPMD);

/// Create a runtime call for kmpc_target_deinit
///
/// \param Loc The insert and source location description.
/// \param IsSPMD Flag to indicate if the kernel is an SPMD kernel or not.
/// \param RequiresFullRuntime Indicate if a full device runtime is necessary.
void createTargetDeinit(const LocationDescription &Loc, bool IsSPMD,
bool RequiresFullRuntime);
void createTargetDeinit(const LocationDescription &Loc, bool IsSPMD);

///}

Expand Down
4 changes: 2 additions & 2 deletions llvm/include/llvm/Frontend/OpenMP/OMPKinds.def
View file
Original file line number Diff line number Diff line change
Expand Up @@ -447,8 +447,8 @@ __OMP_RTL(__kmpc_task_allow_completion_event, false, VoidPtr, IdentPtr,
/* Int */ Int32, /* kmp_task_t */ VoidPtr)

/// OpenMP Device runtime functions
__OMP_RTL(__kmpc_target_init, false, Int32, IdentPtr, Int8, Int1, Int1)
__OMP_RTL(__kmpc_target_deinit, false, Void, IdentPtr, Int8, Int1)
__OMP_RTL(__kmpc_target_init, false, Int32, IdentPtr, Int8, Int1)
__OMP_RTL(__kmpc_target_deinit, false, Void, IdentPtr, Int8)
__OMP_RTL(__kmpc_kernel_prepare_parallel, false, Void, VoidPtr)
__OMP_RTL(__kmpc_parallel_51, false, Void, IdentPtr, Int32, Int32, Int32, Int32,
VoidPtr, VoidPtr, VoidPtrPtr, SizeTy)
Expand Down
14 changes: 4 additions & 10 deletions llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -3854,8 +3854,7 @@ CallInst *OpenMPIRBuilder::createCachedThreadPrivate(
}

OpenMPIRBuilder::InsertPointTy
OpenMPIRBuilder::createTargetInit(const LocationDescription &Loc, bool IsSPMD,
bool RequiresFullRuntime) {
OpenMPIRBuilder::createTargetInit(const LocationDescription &Loc, bool IsSPMD) {
if (!updateToLocation(Loc))
return Loc.IP;

Expand All @@ -3867,14 +3866,12 @@ OpenMPIRBuilder::createTargetInit(const LocationDescription &Loc, bool IsSPMD,
IsSPMD ? OMP_TGT_EXEC_MODE_SPMD : OMP_TGT_EXEC_MODE_GENERIC);
ConstantInt *UseGenericStateMachine =
ConstantInt::getBool(Int32->getContext(), !IsSPMD);
ConstantInt *RequiresFullRuntimeVal =
ConstantInt::getBool(Int32->getContext(), RequiresFullRuntime);

Function *Fn = getOrCreateRuntimeFunctionPtr(
omp::RuntimeFunction::OMPRTL___kmpc_target_init);

CallInst *ThreadKind = Builder.CreateCall(
Fn, {Ident, IsSPMDVal, UseGenericStateMachine, RequiresFullRuntimeVal});
Fn, {Ident, IsSPMDVal, UseGenericStateMachine});

Value *ExecUserCode = Builder.CreateICmpEQ(
ThreadKind, ConstantInt::get(ThreadKind->getType(), -1),
Expand Down Expand Up @@ -3908,8 +3905,7 @@ OpenMPIRBuilder::createTargetInit(const LocationDescription &Loc, bool IsSPMD,
}

void OpenMPIRBuilder::createTargetDeinit(const LocationDescription &Loc,
bool IsSPMD,
bool RequiresFullRuntime) {
bool IsSPMD) {
if (!updateToLocation(Loc))
return;

Expand All @@ -3919,13 +3915,11 @@ void OpenMPIRBuilder::createTargetDeinit(const LocationDescription &Loc,
ConstantInt *IsSPMDVal = ConstantInt::getSigned(
IntegerType::getInt8Ty(Int8->getContext()),
IsSPMD ? OMP_TGT_EXEC_MODE_SPMD : OMP_TGT_EXEC_MODE_GENERIC);
ConstantInt *RequiresFullRuntimeVal =
ConstantInt::getBool(Int32->getContext(), RequiresFullRuntime);

Function *Fn = getOrCreateRuntimeFunctionPtr(
omp::RuntimeFunction::OMPRTL___kmpc_target_deinit);

Builder.CreateCall(Fn, {Ident, IsSPMDVal, RequiresFullRuntimeVal});
Builder.CreateCall(Fn, {Ident, IsSPMDVal});
}

void OpenMPIRBuilder::setOutlinedTargetRegionFunctionAttributes(
Expand Down
21 changes: 2 additions & 19 deletions llvm/lib/Transforms/IPO/OpenMPOpt.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1863,7 +1863,8 @@ struct OpenMPOpt {
if (!UV || UV->size() + (ReplVal != nullptr) < 2)
return false;

LLVM_DEBUG(
F.dump();
(
dbgs() << TAG << "Deduplicate " << UV->size() << " uses of " << RFI.Name
<< (ReplVal ? " with an existing value\n" : "\n") << "\n");

Expand Down Expand Up @@ -3290,8 +3291,6 @@ struct AAKernelInfoFunction : AAKernelInfo {
constexpr const int InitModeArgNo = 1;
constexpr const int DeinitModeArgNo = 1;
constexpr const int InitUseStateMachineArgNo = 2;
constexpr const int InitRequiresFullRuntimeArgNo = 3;
constexpr const int DeinitRequiresFullRuntimeArgNo = 2;
A.registerSimplificationCallback(
IRPosition::callsite_argument(*KernelInitCB, InitUseStateMachineArgNo),
StateMachineSimplifyCB);
Expand All @@ -3301,14 +3300,6 @@ struct AAKernelInfoFunction : AAKernelInfo {
A.registerSimplificationCallback(
IRPosition::callsite_argument(*KernelDeinitCB, DeinitModeArgNo),
ModeSimplifyCB);
A.registerSimplificationCallback(
IRPosition::callsite_argument(*KernelInitCB,
InitRequiresFullRuntimeArgNo),
IsGenericModeSimplifyCB);
A.registerSimplificationCallback(
IRPosition::callsite_argument(*KernelDeinitCB,
DeinitRequiresFullRuntimeArgNo),
IsGenericModeSimplifyCB);

// Check if we know we are in SPMD-mode already.
ConstantInt *ModeArg =
Expand Down Expand Up @@ -3702,8 +3693,6 @@ struct AAKernelInfoFunction : AAKernelInfo {
const int InitModeArgNo = 1;
const int DeinitModeArgNo = 1;
const int InitUseStateMachineArgNo = 2;
const int InitRequiresFullRuntimeArgNo = 3;
const int DeinitRequiresFullRuntimeArgNo = 2;

auto &Ctx = getAnchorValue().getContext();
A.changeUseAfterManifest(
Expand All @@ -3717,12 +3706,6 @@ struct AAKernelInfoFunction : AAKernelInfo {
KernelDeinitCB->getArgOperandUse(DeinitModeArgNo),
*ConstantInt::getSigned(IntegerType::getInt8Ty(Ctx),
OMP_TGT_EXEC_MODE_SPMD));
A.changeUseAfterManifest(
KernelInitCB->getArgOperandUse(InitRequiresFullRuntimeArgNo),
*ConstantInt::getBool(Ctx, false));
A.changeUseAfterManifest(
KernelDeinitCB->getArgOperandUse(DeinitRequiresFullRuntimeArgNo),
*ConstantInt::getBool(Ctx, false));

++NumOpenMPTargetRegionKernelsSPMD;

Expand Down
12 changes: 6 additions & 6 deletions llvm/test/Transforms/OpenMP/always_inline_device.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -13,7 +13,7 @@ define weak void @__omp_offloading_fd02_c0934fc2_foo_l4() #0 {
; CHECK: Function Attrs: convergent norecurse nounwind
; CHECK-LABEL: @__omp_offloading_fd02_c0934fc2_foo_l4(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i8 2, i1 false, i1 false)
; CHECK-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i8 2, i1 false)
; CHECK-NEXT: [[THREAD_ID_IN_BLOCK:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
; CHECK-NEXT: [[THREAD_IS_MAIN:%.*]] = icmp ne i32 [[THREAD_ID_IN_BLOCK]], 0
; CHECK-NEXT: br i1 [[THREAD_IS_MAIN]], label [[EXIT_THREADS:%.*]], label [[MAIN_THREAD_USER_CODE:%.*]]
Expand All @@ -24,13 +24,13 @@ define weak void @__omp_offloading_fd02_c0934fc2_foo_l4() #0 {
; CHECK-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
; CHECK: user_code.entry:
; CHECK-NEXT: store i8 1, i8* @G, align 1
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 false)
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
; CHECK-NEXT: ret void
; CHECK: worker.exit:
; CHECK-NEXT: ret void
;
entry:
%0 = call i32 @__kmpc_target_init(%struct.ident_t* @1, i8 1, i1 true, i1 true)
%0 = call i32 @__kmpc_target_init(%struct.ident_t* @1, i8 1, i1 true)
%exec_user_code = icmp eq i32 %0, -1
br i1 %exec_user_code, label %user_code.entry, label %worker.exit

Expand All @@ -42,7 +42,7 @@ user_code.entry: ; preds = %entry
%isSPMD = call i8 @__kmpc_is_spmd_exec_mode()
store i8 %isSPMD, i8* @G
call void @bar() #2
call void @__kmpc_target_deinit(%struct.ident_t* @1, i8 1, i1 true)
call void @__kmpc_target_deinit(%struct.ident_t* @1, i8 1)
ret void

worker.exit: ; preds = %entry
Expand All @@ -51,9 +51,9 @@ worker.exit: ; preds = %entry

declare i8 @__kmpc_is_spmd_exec_mode()

declare i32 @__kmpc_target_init(%struct.ident_t*, i8, i1, i1)
declare i32 @__kmpc_target_init(%struct.ident_t*, i8, i1)

declare void @__kmpc_target_deinit(%struct.ident_t*, i8, i1)
declare void @__kmpc_target_deinit(%struct.ident_t*, i8)

; Function Attrs: convergent nounwind
define hidden void @bar() #1 {
Expand Down
1 change: 1 addition & 0 deletions llvm/test/Transforms/OpenMP/barrier_removal.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -249,6 +249,7 @@ define void @pos_multiple() {
; CHECK: attributes #[[ATTR0:[0-9]+]] = { "llvm.assume"="ompx_aligned_barrier" }
; CHECK: attributes #[[ATTR1:[0-9]+]] = { convergent nocallback nounwind }
; CHECK: attributes #[[ATTR2:[0-9]+]] = { convergent nocallback nofree nounwind willreturn }
; CHECK: attributes #[[ATTR3:[0-9]+]] = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: readwrite) }
;.
; CHECK: [[META0:![0-9]+]] = !{i32 7, !"openmp", i32 50}
; CHECK: [[META1:![0-9]+]] = !{i32 7, !"openmp-device", i32 50}
Expand Down
184 changes: 102 additions & 82 deletions llvm/test/Transforms/OpenMP/custom_state_machines.ll
View file

Large diffs are not rendered by default.

100 changes: 50 additions & 50 deletions llvm/test/Transforms/OpenMP/custom_state_machines_pre_lto.ll
View file

Large diffs are not rendered by default.

12 changes: 6 additions & 6 deletions llvm/test/Transforms/OpenMP/custom_state_machines_remarks.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -60,7 +60,7 @@ target triple = "nvptx64"
define weak void @__omp_offloading_2a_d80d3d_test_fallback_l11() local_unnamed_addr #0 !dbg !15 {
entry:
%captured_vars_addrs.i.i = alloca [0 x i8*], align 8
%0 = call i32 @__kmpc_target_init(%struct.ident_t* nonnull @1, i8 1, i1 true, i1 true) #3, !dbg !18
%0 = call i32 @__kmpc_target_init(%struct.ident_t* nonnull @1, i8 1, i1 true) #3, !dbg !18
%exec_user_code = icmp eq i32 %0, -1, !dbg !18
br i1 %exec_user_code, label %user_code.entry, label %common.ret, !dbg !18

Expand All @@ -77,12 +77,12 @@ user_code.entry: ; preds = %entry
call void @__kmpc_parallel_51(%struct.ident_t* noundef nonnull @13, i32 %3, 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 nonnull %4, i64 noundef 0) #3, !dbg !23
call void @llvm.lifetime.end.p0i8(i64 0, i8* nonnull %2) #3, !dbg !26
call void @unknown() #6, !dbg !27
call void @__kmpc_target_deinit(%struct.ident_t* nonnull @5, i8 1, i1 true) #3, !dbg !28
call void @__kmpc_target_deinit(%struct.ident_t* nonnull @5, i8 1) #3, !dbg !28
br label %common.ret
}

; Make it a weak definition so we will apply custom state machine rewriting but can't use the body in the reasoning.
define weak i32 @__kmpc_target_init(%struct.ident_t*, i8, i1, i1) {
define weak i32 @__kmpc_target_init(%struct.ident_t*, i8, i1) {
ret i32 0
}

Expand All @@ -102,13 +102,13 @@ entry:
; Function Attrs: nounwind
declare i32 @__kmpc_global_thread_num(%struct.ident_t*) local_unnamed_addr #3

declare void @__kmpc_target_deinit(%struct.ident_t*, i8, i1) local_unnamed_addr
declare void @__kmpc_target_deinit(%struct.ident_t*, i8) local_unnamed_addr

; Function Attrs: norecurse nounwind
define weak void @__omp_offloading_2a_d80d3d_test_no_fallback_l20() local_unnamed_addr #4 !dbg !32 {
entry:
%captured_vars_addrs.i2.i = alloca [0 x i8*], align 8
%0 = call i32 @__kmpc_target_init(%struct.ident_t* nonnull @7, i8 1, i1 true, i1 true) #3, !dbg !33
%0 = call i32 @__kmpc_target_init(%struct.ident_t* nonnull @7, i8 1, i1 true) #3, !dbg !33
%exec_user_code = icmp eq i32 %0, -1, !dbg !33
br i1 %exec_user_code, label %user_code.entry, label %common.ret, !dbg !33

Expand All @@ -133,7 +133,7 @@ user_code.entry: ; preds = %entry
call void @llvm.lifetime.end.p0i8(i64 0, i8* nonnull %2) #3, !dbg !45
call void @no_openmp()
call void @no_parallelism()
call void @__kmpc_target_deinit(%struct.ident_t* nonnull @11, i8 1, i1 true) #3, !dbg !46
call void @__kmpc_target_deinit(%struct.ident_t* nonnull @11, i8 1) #3, !dbg !46
br label %common.ret
}

Expand Down
12 changes: 6 additions & 6 deletions llvm/test/Transforms/OpenMP/deduplication_target.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -19,37 +19,37 @@ define weak void @__omp_offloading_50_a3e09bf8_foo_l2() #0 {
; CHECK-SAME: () #[[ATTR0:[0-9]+]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
; CHECK-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i8 2, i1 false, i1 true)
; CHECK-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i8 2, i1 false)
; CHECK-NEXT: [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2:[0-9]+]])
; 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: call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 true)
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
; CHECK-NEXT: ret void
; CHECK: worker.exit:
; CHECK-NEXT: ret void
;
entry:
%captured_vars_addrs = alloca [0 x i8*], align 8
%0 = call i32 @__kmpc_target_init(%struct.ident_t* @1, i8 2, i1 false, i1 true)
%0 = call i32 @__kmpc_target_init(%struct.ident_t* @1, i8 2, i1 false)
%exec_user_code = icmp eq i32 %0, -1
br i1 %exec_user_code, label %user_code.entry, label %worker.exit

user_code.entry: ; preds = %entry
%1 = call i32 @__kmpc_global_thread_num(%struct.ident_t* @2)
%2 = call i32 @__kmpc_global_thread_num(%struct.ident_t* @2)
call void @__kmpc_target_deinit(%struct.ident_t* @1, i8 2, i1 true)
call void @__kmpc_target_deinit(%struct.ident_t* @1, i8 2)
ret void

worker.exit: ; preds = %entry
ret void
}

declare i32 @__kmpc_target_init(%struct.ident_t*, i8, i1, i1)
declare i32 @__kmpc_target_init(%struct.ident_t*, i8, i1)

declare i32 @__kmpc_global_thread_num(%struct.ident_t*) #1

declare void @__kmpc_target_deinit(%struct.ident_t*, i8, i1)
declare void @__kmpc_target_deinit(%struct.ident_t*, i8)

attributes #0 = { 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" }
attributes #1 = { nounwind }
Expand Down
16 changes: 8 additions & 8 deletions llvm/test/Transforms/OpenMP/fold_generic_main_thread.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -10,7 +10,7 @@

define void @kernel() {
; CHECK-LABEL: define {{[^@]+}}@kernel() {
; CHECK-NEXT: [[CALL:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* nonnull @[[GLOB1:[0-9]+]], i8 1, i1 false, i1 false)
; CHECK-NEXT: [[CALL:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* nonnull @[[GLOB1:[0-9]+]], i8 1, i1 false)
; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[CALL]], -1
; CHECK-NEXT: br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_ELSE:%.*]]
; CHECK: if.then:
Expand All @@ -20,11 +20,11 @@ define void @kernel() {
; CHECK-NEXT: call void @bar()
; CHECK-NEXT: br label [[IF_END]]
; CHECK: if.end:
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1, i1 true)
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1)
; CHECK-NEXT: ret void
;
; CHECK-DISABLED-LABEL: define {{[^@]+}}@kernel() {
; CHECK-DISABLED-NEXT: [[CALL:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* nonnull @[[GLOB1:[0-9]+]], i8 1, i1 false, i1 false)
; CHECK-DISABLED-NEXT: [[CALL:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* nonnull @[[GLOB1:[0-9]+]], i8 1, i1 false)
; CHECK-DISABLED-NEXT: [[CMP:%.*]] = icmp eq i32 [[CALL]], -1
; CHECK-DISABLED-NEXT: br i1 [[CMP]], label [[IF_THEN:%.*]], label [[IF_ELSE:%.*]]
; CHECK-DISABLED: if.then:
Expand All @@ -34,10 +34,10 @@ define void @kernel() {
; CHECK-DISABLED-NEXT: call void @bar()
; CHECK-DISABLED-NEXT: br label [[IF_END]]
; CHECK-DISABLED: if.end:
; CHECK-DISABLED-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1, i1 true)
; CHECK-DISABLED-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1)
; CHECK-DISABLED-NEXT: ret void
;
%call = call i32 @__kmpc_target_init(%struct.ident_t* nonnull @1, i8 1, i1 false, i1 false)
%call = call i32 @__kmpc_target_init(%struct.ident_t* nonnull @1, i8 1, i1 false)
%cmp = icmp eq i32 %call, -1
br i1 %cmp, label %if.then, label %if.else
if.then:
Expand All @@ -47,7 +47,7 @@ if.else:
call void @bar()
br label %if.end
if.end:
call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1, i1 true)
call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1)
ret void
}

Expand Down Expand Up @@ -135,9 +135,9 @@ declare i8 @__kmpc_is_generic_main_thread_id(i32)

declare i32 @__kmpc_get_hardware_thread_id()

declare i32 @__kmpc_target_init(%struct.ident_t*, i8, i1, i1)
declare i32 @__kmpc_target_init(%struct.ident_t*, i8, i1)

declare void @__kmpc_target_deinit(%struct.ident_t*, i8, i1)
declare void @__kmpc_target_deinit(%struct.ident_t*, i8)

!llvm.dbg.cu = !{!0}
!llvm.module.flags = !{!3, !4, !5, !6}
Expand Down
28 changes: 14 additions & 14 deletions llvm/test/Transforms/OpenMP/get_hardware_num_threads_in_block_fold.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -18,7 +18,7 @@ target triple = "nvptx64"
define weak void @kernel0() #0 {
; CHECK-LABEL: define {{[^@]+}}@kernel0
; CHECK-SAME: () #[[ATTR0:[0-9]+]] {
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 2, i1 false, i1 false)
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 2, i1 false)
; CHECK-NEXT: [[THREAD_ID_IN_BLOCK:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
; CHECK-NEXT: [[THREAD_IS_MAIN:%.*]] = icmp ne i32 [[THREAD_ID_IN_BLOCK]], 0
; CHECK-NEXT: br i1 [[THREAD_IS_MAIN]], label [[EXIT_THREADS:%.*]], label [[MAIN_THREAD_USER_CODE:%.*]]
Expand All @@ -28,15 +28,15 @@ define weak void @kernel0() #0 {
; CHECK-NEXT: call void @helper0() #[[ATTR1:[0-9]+]]
; CHECK-NEXT: call void @helper1() #[[ATTR1]]
; CHECK-NEXT: call void @helper2() #[[ATTR1]]
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i8 2, i1 false)
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i8 2)
; CHECK-NEXT: ret void
;

%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 1, i1 false, i1 false)
%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 1, i1 false)
call void @helper0()
call void @helper1()
call void @helper2()
call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1, i1 false)
call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1)
ret void
}

Expand All @@ -45,21 +45,21 @@ define weak void @kernel0() #0 {
define weak void @kernel1() #0 {
; CHECK-LABEL: define {{[^@]+}}@kernel1
; CHECK-SAME: () #[[ATTR0]] {
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 2, i1 false, i1 false)
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 2, i1 false)
; CHECK-NEXT: [[THREAD_ID_IN_BLOCK:%.*]] = call i32 @__kmpc_get_hardware_thread_id_in_block()
; CHECK-NEXT: [[THREAD_IS_MAIN:%.*]] = icmp ne i32 [[THREAD_ID_IN_BLOCK]], 0
; CHECK-NEXT: br i1 [[THREAD_IS_MAIN]], label [[EXIT_THREADS:%.*]], label [[MAIN_THREAD_USER_CODE:%.*]]
; CHECK: exit.threads:
; CHECK-NEXT: ret void
; CHECK: main.thread.user_code:
; CHECK-NEXT: call void @helper1() #[[ATTR1]]
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i8 2, i1 false)
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i8 2)
; CHECK-NEXT: ret void
;

%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 1, i1 false, i1 false)
%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 1, i1 false)
call void @helper1()
call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1, i1 false)
call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1)
ret void
}

Expand All @@ -70,7 +70,7 @@ define weak void @kernel2() #0 {
; CHECK-SAME: () #[[ATTR0]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 2, i1 false, i1 false)
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 2, i1 false)
; CHECK-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[I]], -1
; CHECK-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[COMMON_RET:%.*]]
; CHECK: common.ret:
Expand All @@ -82,12 +82,12 @@ define weak void @kernel2() #0 {
; CHECK-NEXT: call void @helper1() #[[ATTR1]]
; CHECK-NEXT: call void @helper2() #[[ATTR1]]
; CHECK-NEXT: call void @__kmpc_parallel_51(%struct.ident_t* null, i32 [[TMP0]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__ to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined___wrapper to i8*), i8** [[TMP1]], i64 0)
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i8 2, i1 false)
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i8 2)
; CHECK-NEXT: ret void
;
entry:
%captured_vars_addrs = alloca [0 x i8*], align 8
%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 1, i1 true, i1 true)
%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 1, i1 true)
%exec_user_code = icmp eq i32 %i, -1
br i1 %exec_user_code, label %user_code.entry, label %common.ret

Expand All @@ -101,7 +101,7 @@ user_code.entry:
call void @helper1()
call void @helper2()
call void @__kmpc_parallel_51(%struct.ident_t* null, i32 %0, i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__ to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined___wrapper to i8*), i8** %1, i64 0)
call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1, i1 true)
call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1)
ret void
}

Expand Down Expand Up @@ -202,8 +202,8 @@ define internal i32 @__kmpc_get_hardware_num_threads_in_block() {
ret i32 %ret
}
declare i32 @__kmpc_get_hardware_num_threads_in_block_dummy()
declare i32 @__kmpc_target_init(%struct.ident_t*, i8, i1 zeroext, i1 zeroext) #1
declare void @__kmpc_target_deinit(%struct.ident_t* nocapture readnone, i8, i1 zeroext) #1
declare i32 @__kmpc_target_init(%struct.ident_t*, i8, i1 zeroext) #1
declare void @__kmpc_target_deinit(%struct.ident_t* nocapture readnone, i8) #1
declare void @__kmpc_parallel_51(%struct.ident_t*, i32, i32, i32, i32, i8*, i8*, i8**, i64)
declare i32 @__kmpc_global_thread_num(%struct.ident_t*)

Expand Down
28 changes: 14 additions & 14 deletions llvm/test/Transforms/OpenMP/get_hardware_num_threads_in_block_fold_optnone.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -16,18 +16,18 @@ target triple = "nvptx64"
define weak void @kernel0() #0 {
; CHECK-LABEL: define {{[^@]+}}@kernel0
; CHECK-SAME: () #[[ATTR0:[0-9]+]] {
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i1 true, i1 false, i1 false)
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i1 true, i1 false)
; CHECK-NEXT: call void @helper0()
; CHECK-NEXT: call void @helper1()
; CHECK-NEXT: call void @helper2()
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i1 true, i1 false)
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i1 true)
; CHECK-NEXT: ret void
;
%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i1 true, i1 false, i1 false)
%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i1 true, i1 false)
call void @helper0()
call void @helper1()
call void @helper2()
call void @__kmpc_target_deinit(%struct.ident_t* null, i1 true, i1 false)
call void @__kmpc_target_deinit(%struct.ident_t* null, i1 true)
ret void
}

Expand All @@ -36,14 +36,14 @@ define weak void @kernel0() #0 {
define weak void @kernel1() #0 {
; CHECK-LABEL: define {{[^@]+}}@kernel1
; CHECK-SAME: () #[[ATTR0]] {
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i1 true, i1 false, i1 false)
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i1 true, i1 false)
; CHECK-NEXT: call void @helper1()
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i1 false, i1 false)
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i1 false)
; CHECK-NEXT: ret void
;
%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i1 true, i1 false, i1 false)
%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i1 true, i1 false)
call void @helper1()
call void @__kmpc_target_deinit(%struct.ident_t* null, i1 false, i1 false)
call void @__kmpc_target_deinit(%struct.ident_t* null, i1 false)
ret void
}

Expand All @@ -52,18 +52,18 @@ define weak void @kernel1() #0 {
define weak void @kernel2() #0 {
; CHECK-LABEL: define {{[^@]+}}@kernel2
; CHECK-SAME: () #[[ATTR0]] {
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i1 false, i1 false, i1 false)
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i1 false, i1 false)
; CHECK-NEXT: call void @helper0()
; CHECK-NEXT: call void @helper1()
; CHECK-NEXT: call void @helper2()
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i1 false, i1 false)
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i1 false)
; CHECK-NEXT: ret void
;
%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i1 false, i1 false, i1 false)
%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i1 false, i1 false)
call void @helper0()
call void @helper1()
call void @helper2()
call void @__kmpc_target_deinit(%struct.ident_t* null, i1 false, i1 false)
call void @__kmpc_target_deinit(%struct.ident_t* null, i1 false)
ret void
}

Expand Down Expand Up @@ -112,8 +112,8 @@ define internal void @helper2() {
}

declare i32 @__kmpc_get_hardware_num_threads_in_block()
declare i32 @__kmpc_target_init(%struct.ident_t*, i1 zeroext, i1 zeroext, i1 zeroext) #1
declare void @__kmpc_target_deinit(%struct.ident_t* nocapture readnone, i1 zeroext, i1 zeroext) #1
declare i32 @__kmpc_target_init(%struct.ident_t*, i1 zeroext, i1 zeroext) #1
declare void @__kmpc_target_deinit(%struct.ident_t* nocapture readnone, i1 zeroext) #1


!llvm.module.flags = !{!0, !1}
Expand Down
27 changes: 19 additions & 8 deletions llvm/test/Transforms/OpenMP/global_constructor.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -10,7 +10,7 @@

define weak void @__omp_offloading_fd02_85283c04_main_l11(double* nonnull align 8 dereferenceable(8) %X) local_unnamed_addr {
entry:
%0 = tail call i32 @__kmpc_target_init(%struct.ident_t* nonnull @1, i8 2, i1 false, i1 false) #0
%0 = tail call i32 @__kmpc_target_init(%struct.ident_t* nonnull @1, i8 2, i1 false) #0
%exec_user_code = icmp eq i32 %0, -1
br i1 %exec_user_code, label %user_code.entry, label %common.ret

Expand All @@ -29,13 +29,13 @@ region.guarded:

region.barrier:
tail call void @__kmpc_barrier_simple_spmd(%struct.ident_t* nonnull @1, i32 %2)
tail call void @__kmpc_target_deinit(%struct.ident_t* nonnull @1, i8 2, i1 false) #0
tail call void @__kmpc_target_deinit(%struct.ident_t* nonnull @1, i8 2) #0
br label %common.ret
}

declare i32 @__kmpc_target_init(%struct.ident_t*, i8, i1, i1) local_unnamed_addr
declare i32 @__kmpc_target_init(%struct.ident_t*, i8, i1) local_unnamed_addr

declare void @__kmpc_target_deinit(%struct.ident_t*, i8, i1) local_unnamed_addr
declare void @__kmpc_target_deinit(%struct.ident_t*, i8) local_unnamed_addr

define internal void @__omp_offloading__fd02_85283c04_Device_l6_ctor() {
entry:
Expand Down Expand Up @@ -78,21 +78,32 @@ attributes #1 = { convergent nounwind }
; CHECK-LABEL: define {{[^@]+}}@__omp_offloading_fd02_85283c04_main_l11
; CHECK-SAME: (double* nonnull align 8 dereferenceable(8) [[X:%.*]]) local_unnamed_addr {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = tail call i32 @__kmpc_target_init(%struct.ident_t* nonnull @[[GLOB1:[0-9]+]], i8 2, i1 false, i1 false) #[[ATTR1:[0-9]+]]
; CHECK-NEXT: [[TMP0:%.*]] = tail call i32 @__kmpc_target_init(%struct.ident_t* nonnull @[[GLOB1:[0-9]+]], i8 2, i1 false) #[[ATTR2:[0-9]+]]
; CHECK-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
; CHECK-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[COMMON_RET:%.*]]
; CHECK: common.ret:
; CHECK-NEXT: ret void
; CHECK: user_code.entry:
; CHECK-NEXT: [[TMP1:%.*]] = load double, double* @_ZL6Device, align 8, !tbaa [[TBAA11:![0-9]+]]
; CHECK-NEXT: [[TMP2:%.*]] = tail call i32 @__kmpc_get_hardware_thread_id_in_block() #[[ATTR1]]
; CHECK-NEXT: [[TMP2:%.*]] = tail call i32 @__kmpc_get_hardware_thread_id_in_block() #[[ATTR2]]
; CHECK-NEXT: [[TMP3:%.*]] = icmp eq i32 [[TMP2]], 0
; CHECK-NEXT: br i1 [[TMP3]], label [[REGION_GUARDED:%.*]], label [[REGION_BARRIER:%.*]]
; CHECK: region.guarded:
; CHECK-NEXT: store double [[TMP1]], double* [[X]], align 8, !tbaa [[TBAA11]]
; CHECK-NEXT: br label [[REGION_BARRIER]]
; CHECK: region.barrier:
; CHECK-NEXT: tail call void @__kmpc_barrier_simple_spmd(%struct.ident_t* nonnull @[[GLOB1]], i32 [[TMP2]]) #[[ATTR1]]
; CHECK-NEXT: tail call void @__kmpc_target_deinit(%struct.ident_t* nonnull @[[GLOB1]], i8 2, i1 false) #[[ATTR1]]
; CHECK-NEXT: tail call void @__kmpc_barrier_simple_spmd(%struct.ident_t* nonnull @[[GLOB1]], i32 [[TMP2]]) #[[ATTR2]]
; CHECK-NEXT: tail call void @__kmpc_target_deinit(%struct.ident_t* nonnull @[[GLOB1]], i8 2) #[[ATTR2]]
; CHECK-NEXT: br label [[COMMON_RET]]
;
;
; CHECK: Function Attrs: norecurse
; CHECK-LABEL: define {{[^@]+}}@__omp_offloading__fd02_85283c04_Device_l6_ctor
; CHECK-SAME: () #[[ATTR0:[0-9]+]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CALL_I:%.*]] = tail call double @__nv_log(double noundef 2.000000e+00) #[[ATTR1:[0-9]+]]
; CHECK-NEXT: [[CALL_I2:%.*]] = tail call double @__nv_log(double noundef 2.000000e+00) #[[ATTR1]]
; CHECK-NEXT: [[DIV:%.*]] = fdiv double [[CALL_I]], [[CALL_I2]]
; CHECK-NEXT: store double [[DIV]], double* @_ZL6Device, align 8, !tbaa [[TBAA11]]
; CHECK-NEXT: ret void
;
8 changes: 4 additions & 4 deletions llvm/test/Transforms/OpenMP/globalization_remarks.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -13,13 +13,13 @@ target triple = "nvptx64"

define void @foo() {
entry:
%c = call i32 @__kmpc_target_init(%struct.ident_t* null, i1 false, i1 true, i1 true)
%c = call i32 @__kmpc_target_init(%struct.ident_t* null, i1 false, i1 true)
%0 = call i8* @__kmpc_alloc_shared(i64 4), !dbg !10
%x_on_stack = bitcast i8* %0 to i32*
%1 = bitcast i32* %x_on_stack to i8*
call void @share(i8* %1), !dbg !10
call void @__kmpc_free_shared(i8* %0)
call void @__kmpc_target_deinit(%struct.ident_t* null, i1 false, i1 true)
call void @__kmpc_target_deinit(%struct.ident_t* null, i1 false)
ret void
}

Expand All @@ -33,9 +33,9 @@ declare i8* @__kmpc_alloc_shared(i64)

declare void @__kmpc_free_shared(i8* nocapture)

declare i32 @__kmpc_target_init(%struct.ident_t*, i1, i1, i1);
declare i32 @__kmpc_target_init(%struct.ident_t*, i1, i1);

declare void @__kmpc_target_deinit(%struct.ident_t*, i1, i1)
declare void @__kmpc_target_deinit(%struct.ident_t*, i1)

!llvm.dbg.cu = !{!0}
!llvm.module.flags = !{!3, !4, !5, !6}
Expand Down
8 changes: 4 additions & 4 deletions llvm/test/Transforms/OpenMP/gpu_state_machine_function_ptr_replacement.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -48,22 +48,22 @@ entry:
%.zero.addr = alloca i32, align 4
%.threadid_temp. = alloca i32, align 4
store i32 0, i32* %.zero.addr, align 4
%0 = call i32 @__kmpc_target_init(%struct.ident_t* @1, i8 1, i1 true, i1 true)
%0 = call i32 @__kmpc_target_init(%struct.ident_t* @1, i8 1, i1 true)
%exec_user_code = icmp eq i32 %0, -1
br i1 %exec_user_code, label %user_code.entry, label %worker.exit

user_code.entry: ; preds = %entry
%1 = call i32 @__kmpc_global_thread_num(%struct.ident_t* @1)
store i32 %1, i32* %.threadid_temp., align 4
call void @__omp_outlined__(i32* %.threadid_temp., i32* %.zero.addr)
call void @__kmpc_target_deinit(%struct.ident_t* @1, i8 1, i1 true)
call void @__kmpc_target_deinit(%struct.ident_t* @1, i8 1)
ret void

worker.exit: ; preds = %entry
ret void
}

define weak i32 @__kmpc_target_init(%struct.ident_t*, i8, i1, i1) {
define weak i32 @__kmpc_target_init(%struct.ident_t*, i8, i1) {
ret i32 0
}

Expand Down Expand Up @@ -149,7 +149,7 @@ entry:

declare i32 @__kmpc_global_thread_num(%struct.ident_t*)

declare void @__kmpc_target_deinit(%struct.ident_t*, i8, i1)
declare void @__kmpc_target_deinit(%struct.ident_t*, i8)

define internal void @__omp_outlined__3(i32* noalias %.global_tid., i32* noalias %.bound_tid.) {
entry:
Expand Down
36 changes: 18 additions & 18 deletions llvm/test/Transforms/OpenMP/is_spmd_exec_mode_fold.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -21,26 +21,26 @@ target triple = "nvptx64"
;.
define weak void @is_spmd() {
; CHECK-LABEL: define {{[^@]+}}@is_spmd() {
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 2, i1 false, i1 false)
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 2, i1 false)
; CHECK-NEXT: call void @is_spmd_helper1()
; CHECK-NEXT: call void @is_spmd_helper2()
; CHECK-NEXT: call void @is_mixed_helper()
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i8 2, i1 false)
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i8 2)
; CHECK-NEXT: ret void
;
%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 2, i1 false, i1 false)
%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 2, i1 false)
call void @is_spmd_helper1()
call void @is_spmd_helper2()
call void @is_mixed_helper()
call void @__kmpc_target_deinit(%struct.ident_t* null, i8 2, i1 false)
call void @__kmpc_target_deinit(%struct.ident_t* null, i8 2)
ret void
}

define weak void @will_be_spmd() {
; CHECK-LABEL: define {{[^@]+}}@will_be_spmd() {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CAPTURED_VARS_ADDRS:%.*]] = alloca [0 x i8*], align 8
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 2, i1 false, i1 false)
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 2, i1 false)
; CHECK-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[I]], -1
; CHECK-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[COMMON_RET:%.*]]
; CHECK: common.ret:
Expand All @@ -50,12 +50,12 @@ define weak void @will_be_spmd() {
; CHECK-NEXT: [[TMP1:%.*]] = bitcast [0 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
; CHECK-NEXT: call void @is_spmd_helper2()
; CHECK-NEXT: call void @__kmpc_parallel_51(%struct.ident_t* null, i32 [[TMP0]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__ to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined___wrapper to i8*), i8** [[TMP1]], i64 0)
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i8 2, i1 false)
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i8 2)
; CHECK-NEXT: ret void
;
entry:
%captured_vars_addrs = alloca [0 x i8*], align 8
%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 1, i1 true, i1 true)
%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 1, i1 true)
%exec_user_code = icmp eq i32 %i, -1
br i1 %exec_user_code, label %user_code.entry, label %common.ret

Expand All @@ -67,41 +67,41 @@ user_code.entry:
%1 = bitcast [0 x i8*]* %captured_vars_addrs to i8**
call void @is_spmd_helper2()
call void @__kmpc_parallel_51(%struct.ident_t* null, i32 %0, i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*)* @__omp_outlined__ to i8*), i8* bitcast (void (i16, i32)* @__omp_outlined___wrapper to i8*), i8** %1, i64 0)
call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1, i1 true)
call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1)
ret void
}

define weak void @non_spmd() {
; CHECK-LABEL: define {{[^@]+}}@non_spmd() {
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 1, i1 false, i1 false)
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 1, i1 false)
; CHECK-NEXT: call void @is_generic_helper1()
; CHECK-NEXT: call void @is_generic_helper2()
; CHECK-NEXT: call void @is_mixed_helper()
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1, i1 false)
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1)
; CHECK-NEXT: ret void
;
%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 1, i1 false, i1 false)
%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 1, i1 false)
call void @is_generic_helper1()
call void @is_generic_helper2()
call void @is_mixed_helper()
call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1, i1 false)
call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1)
ret void
}

define weak void @will_not_be_spmd() {
; CHECK-LABEL: define {{[^@]+}}@will_not_be_spmd() {
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 1, i1 false, i1 false)
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 1, i1 false)
; CHECK-NEXT: call void @is_generic_helper1()
; CHECK-NEXT: call void @is_generic_helper2()
; CHECK-NEXT: call void @is_mixed_helper()
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1, i1 false)
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1)
; CHECK-NEXT: ret void
;
%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 1, i1 false, i1 false)
%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 1, i1 false)
call void @is_generic_helper1()
call void @is_generic_helper2()
call void @is_mixed_helper()
call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1, i1 false)
call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1)
ret void
}

Expand Down Expand Up @@ -198,8 +198,8 @@ entry:

declare void @spmd_compatible() "llvm.assume"="ompx_spmd_amenable"
declare i8 @__kmpc_is_spmd_exec_mode()
declare i32 @__kmpc_target_init(%struct.ident_t*, i8, i1 zeroext, i1 zeroext)
declare void @__kmpc_target_deinit(%struct.ident_t* nocapture readnone, i8, i1 zeroext)
declare i32 @__kmpc_target_init(%struct.ident_t*, i8, i1 zeroext)
declare void @__kmpc_target_deinit(%struct.ident_t* nocapture readnone, i8)
declare void @__kmpc_parallel_51(%struct.ident_t*, i32, i32, i32, i32, i8*, i8*, i8**, i64)
declare i32 @__kmpc_global_thread_num(%struct.ident_t*)
declare void @foo()
Expand Down
28 changes: 14 additions & 14 deletions llvm/test/Transforms/OpenMP/parallel_level_fold.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -19,46 +19,46 @@ target triple = "nvptx64"
;.
define weak void @none_spmd() {
; CHECK-LABEL: define {{[^@]+}}@none_spmd() {
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 1, i1 false, i1 false)
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 1, i1 false)
; CHECK-NEXT: call void @none_spmd_helper()
; CHECK-NEXT: call void @mixed_helper()
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1, i1 false)
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1)
; CHECK-NEXT: ret void
;
%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 1, i1 false, i1 false)
%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 1, i1 false)
call void @none_spmd_helper()
call void @mixed_helper()
call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1, i1 false)
call void @__kmpc_target_deinit(%struct.ident_t* null, i8 1)
ret void
}

define weak void @spmd() {
; CHECK-LABEL: define {{[^@]+}}@spmd() {
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 2, i1 false, i1 false)
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 2, i1 false)
; CHECK-NEXT: call void @spmd_helper()
; CHECK-NEXT: call void @mixed_helper()
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i8 2, i1 false)
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i8 2)
; CHECK-NEXT: ret void
;
%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 2, i1 false, i1 false)
%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 2, i1 false)
call void @spmd_helper()
call void @mixed_helper()
call void @__kmpc_target_deinit(%struct.ident_t* null, i8 2, i1 false)
call void @__kmpc_target_deinit(%struct.ident_t* null, i8 2)
ret void
}

define weak void @parallel() {
; CHECK-LABEL: define {{[^@]+}}@parallel() {
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 2, i1 false, i1 false)
; CHECK-NEXT: [[I:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 2, i1 false)
; CHECK-NEXT: call void @spmd_helper()
; CHECK-NEXT: call void @__kmpc_parallel_51(%struct.ident_t* null, i32 0, i32 0, i32 0, i32 0, i8* null, i8* null, i8** null, i64 0)
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i8 2, i1 false)
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* null, i8 2)
; CHECK-NEXT: ret void
;
%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 2, i1 false, i1 false)
%i = call i32 @__kmpc_target_init(%struct.ident_t* null, i8 2, i1 false)
call void @spmd_helper()
call void @__kmpc_parallel_51(%struct.ident_t* null, i32 0, i32 0, i32 0, i32 0, i8* null, i8* null, i8** null, i64 0)
call void @__kmpc_target_deinit(%struct.ident_t* null, i8 2, i1 false)
call void @__kmpc_target_deinit(%struct.ident_t* null, i8 2)
ret void
}

Expand Down Expand Up @@ -130,8 +130,8 @@ define internal void @parallel_helper() {
declare void @foo()
declare void @bar()
declare i8 @__kmpc_parallel_level()
declare i32 @__kmpc_target_init(%struct.ident_t*, i8 zeroext, i1 zeroext, i1 zeroext) #1
declare void @__kmpc_target_deinit(%struct.ident_t* nocapture readnone, i8 zeroext, i1 zeroext) #1
declare i32 @__kmpc_target_init(%struct.ident_t*, i8 zeroext, i1 zeroext) #1
declare void @__kmpc_target_deinit(%struct.ident_t* nocapture readnone, i8 zeroext) #1

!llvm.module.flags = !{!0, !1}
!nvvm.annotations = !{!2, !3, !4}
Expand Down
32 changes: 21 additions & 11 deletions llvm/test/Transforms/OpenMP/remove_globalization.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -17,44 +17,52 @@ target triple = "nvptx64"
%struct.ident_t = type { i32, i32, i32, i32, i8* }

; Make it a weak definition so we will apply custom state machine rewriting but can't use the body in the reasoning.
define weak i32 @__kmpc_target_init(%struct.ident_t*, i8, i1, i1) {
ret i32 0
}
declare void @__kmpc_target_deinit(%struct.ident_t*, i8, i1)

;.
; CHECK: @[[S:[a-zA-Z0-9_$"\\.-]+]] = external local_unnamed_addr global i8*
;.
; CHECK-DISABLED: @[[S:[a-zA-Z0-9_$"\\.-]+]] = external local_unnamed_addr global i8*
;.
define weak i32 @__kmpc_target_init(%struct.ident_t*, i8, i1) {
; CHECK-LABEL: define {{[^@]+}}@__kmpc_target_init
; CHECK-SAME: (%struct.ident_t* [[TMP0:%.*]], i8 [[TMP1:%.*]], i1 [[TMP2:%.*]]) {
; CHECK-NEXT: ret i32 0
;
; CHECK-DISABLED-LABEL: define {{[^@]+}}@__kmpc_target_init
; CHECK-DISABLED-SAME: (%struct.ident_t* [[TMP0:%.*]], i8 [[TMP1:%.*]], i1 [[TMP2:%.*]]) {
; CHECK-DISABLED-NEXT: ret i32 0
;
ret i32 0
}
declare void @__kmpc_target_deinit(%struct.ident_t*, i8)

define void @kernel() {
; CHECK-LABEL: define {{[^@]+}}@kernel() {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* nonnull null, i8 1, i1 false, i1 true)
; CHECK-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* nonnull null, i8 1, i1 false)
; CHECK-NEXT: call void @foo() #[[ATTR0:[0-9]+]]
; CHECK-NEXT: call void @bar() #[[ATTR0]]
; CHECK-NEXT: call void @convert_and_move_alloca() #[[ATTR0]]
; CHECK-NEXT: call void @unknown_no_openmp()
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* nonnull null, i8 1, i1 true)
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* nonnull null, i8 1)
; CHECK-NEXT: ret void
;
; CHECK-DISABLED-LABEL: define {{[^@]+}}@kernel() {
; CHECK-DISABLED-NEXT: entry:
; CHECK-DISABLED-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* nonnull null, i8 1, i1 false, i1 true)
; CHECK-DISABLED-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* nonnull null, i8 1, i1 false)
; CHECK-DISABLED-NEXT: call void @foo() #[[ATTR0:[0-9]+]]
; CHECK-DISABLED-NEXT: call void @bar() #[[ATTR0]]
; CHECK-DISABLED-NEXT: call void @convert_and_move_alloca() #[[ATTR0]]
; CHECK-DISABLED-NEXT: call void @unknown_no_openmp()
; CHECK-DISABLED-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* nonnull null, i8 1, i1 true)
; CHECK-DISABLED-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* nonnull null, i8 1)
; CHECK-DISABLED-NEXT: ret void
;
entry:
%0 = call i32 @__kmpc_target_init(%struct.ident_t* nonnull null, i8 1, i1 true, i1 true)
%0 = call i32 @__kmpc_target_init(%struct.ident_t* nonnull null, i8 1, i1 true)
call void @foo()
call void @bar()
call void @convert_and_move_alloca()
call void @unknown_no_openmp()
call void @__kmpc_target_deinit(%struct.ident_t* nonnull null, i8 1, i1 true)
call void @__kmpc_target_deinit(%struct.ident_t* nonnull null, i8 1)
ret void
}

Expand Down Expand Up @@ -294,3 +302,5 @@ declare void @unknown_no_openmp() "llvm.assume"="omp_no_openmp"
; CHECK-DISABLED: [[META10:![0-9]+]] = !DISubroutineType(types: !2)
; CHECK-DISABLED: [[DBG11]] = !DILocation(line: 6, column: 2, scope: !9)
;.
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
; CHECK-REMARKS: {{.*}}
36 changes: 22 additions & 14 deletions llvm/test/Transforms/OpenMP/replace_globalization.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -25,19 +25,19 @@ target triple = "nvptx64"

define dso_local void @foo() "kernel" {
entry:
%c = call i32 @__kmpc_target_init(%struct.ident_t* @1, i8 1, i1 true, i1 true)
%c = call i32 @__kmpc_target_init(%struct.ident_t* @1, i8 1, i1 true)
%x = call align 4 i8* @__kmpc_alloc_shared(i64 4)
call void @unknown_no_openmp()
%x_on_stack = bitcast i8* %x to i32*
%0 = bitcast i32* %x_on_stack to i8*
call void @use(i8* %0)
call void @__kmpc_free_shared(i8* %x, i64 4)
call void @__kmpc_target_deinit(%struct.ident_t* @1, i8 1, i1 true)
call void @__kmpc_target_deinit(%struct.ident_t* @1, i8 1)
ret void
}

define void @bar() "kernel" {
%c = call i32 @__kmpc_target_init(%struct.ident_t* @1, i8 1, i1 true, i1 true)
%c = call i32 @__kmpc_target_init(%struct.ident_t* @1, i8 1, i1 true)
call void @unknown_no_openmp()
%cmp = icmp eq i32 %c, -1
br i1 %cmp, label %master1, label %exit
Expand All @@ -60,12 +60,12 @@ master2:
call void @__kmpc_free_shared(i8* %y, i64 4)
br label %exit
exit:
call void @__kmpc_target_deinit(%struct.ident_t* @1, i8 1, i1 true)
call void @__kmpc_target_deinit(%struct.ident_t* @1, i8 1)
ret void
}

define void @baz_spmd() "kernel" {
%c = call i32 @__kmpc_target_init(%struct.ident_t* @1, i8 2, i1 true, i1 true)
%c = call i32 @__kmpc_target_init(%struct.ident_t* @1, i8 2, i1 true)
call void @unknown_no_openmp()
%c0 = icmp eq i32 %c, -1
br i1 %c0, label %master3, label %exit
Expand All @@ -77,7 +77,7 @@ master3:
call void @__kmpc_free_shared(i8* %z, i64 24)
br label %exit
exit:
call void @__kmpc_target_deinit(%struct.ident_t* @1, i8 2, i1 true)
call void @__kmpc_target_deinit(%struct.ident_t* @1, i8 2)
ret void
}

Expand Down Expand Up @@ -106,11 +106,11 @@ declare i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
declare i32 @llvm.nvvm.read.ptx.sreg.warpsize()

; Make it a weak definition so we will apply custom state machine rewriting but can't use the body in the reasoning.
define weak i32 @__kmpc_target_init(%struct.ident_t*, i8, i1, i1) {
define weak i32 @__kmpc_target_init(%struct.ident_t*, i8, i1) {
ret i32 0
}

declare void @__kmpc_target_deinit(%struct.ident_t*, i8, i1)
declare void @__kmpc_target_deinit(%struct.ident_t*, i8)

declare void @unknown_no_openmp() "llvm.assume"="omp_no_openmp"

Expand Down Expand Up @@ -147,18 +147,18 @@ declare void @unknown_no_openmp() "llvm.assume"="omp_no_openmp"
; CHECK-LABEL: define {{[^@]+}}@foo
; CHECK-SAME: () #[[ATTR0:[0-9]+]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[C:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 1, i1 false, i1 true)
; CHECK-NEXT: [[C:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 1, i1 false)
; CHECK-NEXT: [[X:%.*]] = call align 4 i8* @__kmpc_alloc_shared(i64 4) #[[ATTR6:[0-9]+]]
; CHECK-NEXT: call void @unknown_no_openmp()
; CHECK-NEXT: call void @use.internalized(i8* nofree [[X]]) #[[ATTR6]]
; CHECK-NEXT: call void @__kmpc_free_shared(i8* [[X]], i64 4) #[[ATTR6]]
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 1, i1 true)
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 1)
; CHECK-NEXT: ret void
;
;
; CHECK-LABEL: define {{[^@]+}}@bar
; CHECK-SAME: () #[[ATTR0]] {
; CHECK-NEXT: [[C:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 1, i1 false, i1 true)
; CHECK-NEXT: [[C:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 1, i1 false)
; CHECK-NEXT: call void @unknown_no_openmp()
; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[C]], -1
; CHECK-NEXT: br i1 [[CMP]], label [[MASTER1:%.*]], label [[EXIT:%.*]]
Expand All @@ -173,13 +173,13 @@ declare void @unknown_no_openmp() "llvm.assume"="omp_no_openmp"
; CHECK-NEXT: call void @use.internalized(i8* nofree addrspacecast (i8 addrspace(3)* getelementptr inbounds ([4 x i8], [4 x i8] addrspace(3)* @y_shared, i32 0, i32 0) to i8*)) #[[ATTR6]]
; CHECK-NEXT: br label [[EXIT]]
; CHECK: exit:
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 1, i1 true)
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 1)
; CHECK-NEXT: ret void
;
;
; CHECK-LABEL: define {{[^@]+}}@baz_spmd
; CHECK-SAME: () #[[ATTR0]] {
; CHECK-NEXT: [[C:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 true, i1 true)
; CHECK-NEXT: [[C:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 true)
; CHECK-NEXT: call void @unknown_no_openmp()
; CHECK-NEXT: [[C0:%.*]] = icmp eq i32 [[C]], -1
; CHECK-NEXT: br i1 [[C0]], label [[MASTER3:%.*]], label [[EXIT:%.*]]
Expand All @@ -189,7 +189,7 @@ declare void @unknown_no_openmp() "llvm.assume"="omp_no_openmp"
; CHECK-NEXT: call void @__kmpc_free_shared(i8* [[Z]], i64 24) #[[ATTR6]]
; CHECK-NEXT: br label [[EXIT]]
; CHECK: exit:
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 true)
; CHECK-NEXT: call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2)
; CHECK-NEXT: ret void
;
;
Expand All @@ -215,6 +215,11 @@ declare void @unknown_no_openmp() "llvm.assume"="omp_no_openmp"
; CHECK-NEXT: [[GEP:%.*]] = getelementptr i8, i8* addrspacecast (i8 addrspace(3)* getelementptr inbounds ([1024 x i8], [1024 x i8] addrspace(3)* @stack, i32 0, i32 0) to i8*), i32 [[L]]
; CHECK-NEXT: ret i8* [[GEP]]
;
;
; CHECK-LABEL: define {{[^@]+}}@__kmpc_target_init
; CHECK-SAME: (%struct.ident_t* [[TMP0:%.*]], i8 [[TMP1:%.*]], i1 [[TMP2:%.*]]) {
; CHECK-NEXT: ret i32 0
;
;.
; CHECK: attributes #[[ATTR0]] = { "kernel" }
; CHECK: attributes #[[ATTR1]] = { nofree norecurse nounwind memory(write) }
Expand All @@ -238,3 +243,6 @@ declare void @unknown_no_openmp() "llvm.assume"="omp_no_openmp"
; CHECK: [[META11:![0-9]+]] = distinct !DISubprogram(name: "bar", scope: !1, file: !1, line: 1, type: !12, scopeLine: 1, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0, retainedNodes: !2)
; CHECK: [[META12:![0-9]+]] = !DISubroutineType(types: !2)
;.
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
; CHECK-LIMIT: {{.*}}
; CHECK-REMARKS: {{.*}}
Loading