1 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs --replace-value-regex "__omp_offloading_[0-9a-z]+_[0-9a-z]+" "reduction_size[.].+[.]" "pl_cond[.].+[.|,]" --prefix-filecheck-ir-name _
2 // Test target codegen - host bc file has to be created first.
3 // RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc
4 // RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - -disable-llvm-optzns | FileCheck %s --check-prefix=CHECK1
5 // RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple i386-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm-bc %s -o %t-x86-host.bc
6 // RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - -disable-llvm-optzns | FileCheck %s --check-prefix=CHECK2
7 // RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -fexceptions -fcxx-exceptions -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - -disable-llvm-optzns | FileCheck %s --check-prefix=CHECK2
8 
9 // RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-cuda-mode -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc
10 // RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-cuda-mode -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - -disable-llvm-optzns | FileCheck %s --check-prefix=CHECK1
11 // RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-cuda-mode -x c++ -triple i386-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm-bc %s -o %t-x86-host.bc
12 // RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-cuda-mode -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - -disable-llvm-optzns | FileCheck %s --check-prefix=CHECK2
13 // RUN: %clang_cc1 -no-opaque-pointers -verify -fopenmp -fopenmp-cuda-mode -fexceptions -fcxx-exceptions -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - -disable-llvm-optzns | FileCheck %s --check-prefix=CHECK2
14 
15 // expected-no-diagnostics
16 #ifndef HEADER
17 #define HEADER
18 
19 template<typename tx>
20 tx ftemplate(int n) {
21   tx a = 0;
22   short aa = 0;
23   tx b[10];
24 
25   #pragma omp target parallel if(target: 0)
26   {
27     a += 1;
28   }
29 
30   #pragma omp target parallel map(tofrom: aa)
31   {
32     aa += 1;
33   }
34 
35   #pragma omp target parallel map(tofrom:a, aa, b) if(target: n>40)
36   {
37     a += 1;
38     aa += 1;
39     b[2] += 1;
40   }
41 
42   return a;
43 }
44 
45 int bar(int n){
46   int a = 0;
47 
48   a += ftemplate<int>(n);
49 
50   return a;
51 }
52 
53 #endif
54 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l30
55 // CHECK1-SAME: (i16* noundef nonnull align 2 dereferenceable(2) [[AA:%.*]]) #[[ATTR0:[0-9]+]] {
56 // CHECK1-NEXT:  entry:
57 // CHECK1-NEXT:    [[AA_ADDR:%.*]] = alloca i16*, align 8
58 // CHECK1-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x i8*], align 8
59 // CHECK1-NEXT:    store i16* [[AA]], i16** [[AA_ADDR]], align 8
60 // CHECK1-NEXT:    [[TMP0:%.*]] = load i16*, i16** [[AA_ADDR]], align 8
61 // CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i8 2, i1 false, i1 true)
62 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
63 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
64 // CHECK1:       user_code.entry:
65 // CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2:[0-9]+]])
66 // CHECK1-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
67 // CHECK1-NEXT:    [[TMP4:%.*]] = bitcast i16* [[TMP0]] to i8*
68 // CHECK1-NEXT:    store i8* [[TMP4]], i8** [[TMP3]], align 8
69 // CHECK1-NEXT:    [[TMP5:%.*]] = bitcast [1 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
70 // CHECK1-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i16*)* @__omp_outlined__ to i8*), i8* null, i8** [[TMP5]], i64 1)
71 // CHECK1-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 true)
72 // CHECK1-NEXT:    ret void
73 // CHECK1:       worker.exit:
74 // CHECK1-NEXT:    ret void
75 //
76 //
77 // CHECK1-LABEL: define {{[^@]+}}@__omp_outlined__
78 // CHECK1-SAME: (i32* noalias noundef [[DOTGLOBAL_TID_:%.*]], i32* noalias noundef [[DOTBOUND_TID_:%.*]], i16* noundef nonnull align 2 dereferenceable(2) [[AA:%.*]]) #[[ATTR1:[0-9]+]] {
79 // CHECK1-NEXT:  entry:
80 // CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
81 // CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
82 // CHECK1-NEXT:    [[AA_ADDR:%.*]] = alloca i16*, align 8
83 // CHECK1-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
84 // CHECK1-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
85 // CHECK1-NEXT:    store i16* [[AA]], i16** [[AA_ADDR]], align 8
86 // CHECK1-NEXT:    [[TMP0:%.*]] = load i16*, i16** [[AA_ADDR]], align 8
87 // CHECK1-NEXT:    [[TMP1:%.*]] = load i16, i16* [[TMP0]], align 2
88 // CHECK1-NEXT:    [[CONV:%.*]] = sext i16 [[TMP1]] to i32
89 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[CONV]], 1
90 // CHECK1-NEXT:    [[CONV1:%.*]] = trunc i32 [[ADD]] to i16
91 // CHECK1-NEXT:    store i16 [[CONV1]], i16* [[TMP0]], align 2
92 // CHECK1-NEXT:    ret void
93 //
94 //
95 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l35
96 // CHECK1-SAME: (i32* noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i16* noundef nonnull align 2 dereferenceable(2) [[AA:%.*]], [10 x i32]* noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
97 // CHECK1-NEXT:  entry:
98 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca i32*, align 8
99 // CHECK1-NEXT:    [[AA_ADDR:%.*]] = alloca i16*, align 8
100 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 8
101 // CHECK1-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x i8*], align 8
102 // CHECK1-NEXT:    store i32* [[A]], i32** [[A_ADDR]], align 8
103 // CHECK1-NEXT:    store i16* [[AA]], i16** [[AA_ADDR]], align 8
104 // CHECK1-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 8
105 // CHECK1-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 8
106 // CHECK1-NEXT:    [[TMP1:%.*]] = load i16*, i16** [[AA_ADDR]], align 8
107 // CHECK1-NEXT:    [[TMP2:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 8
108 // CHECK1-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false, i1 true)
109 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP3]], -1
110 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
111 // CHECK1:       user_code.entry:
112 // CHECK1-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
113 // CHECK1-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
114 // CHECK1-NEXT:    [[TMP6:%.*]] = bitcast i32* [[TMP0]] to i8*
115 // CHECK1-NEXT:    store i8* [[TMP6]], i8** [[TMP5]], align 8
116 // CHECK1-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
117 // CHECK1-NEXT:    [[TMP8:%.*]] = bitcast i16* [[TMP1]] to i8*
118 // CHECK1-NEXT:    store i8* [[TMP8]], i8** [[TMP7]], align 8
119 // CHECK1-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
120 // CHECK1-NEXT:    [[TMP10:%.*]] = bitcast [10 x i32]* [[TMP2]] to i8*
121 // CHECK1-NEXT:    store i8* [[TMP10]], i8** [[TMP9]], align 8
122 // CHECK1-NEXT:    [[TMP11:%.*]] = bitcast [3 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
123 // CHECK1-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32*, i16*, [10 x i32]*)* @__omp_outlined__1 to i8*), i8* null, i8** [[TMP11]], i64 3)
124 // CHECK1-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 true)
125 // CHECK1-NEXT:    ret void
126 // CHECK1:       worker.exit:
127 // CHECK1-NEXT:    ret void
128 //
129 //
130 // CHECK1-LABEL: define {{[^@]+}}@__omp_outlined__1
131 // CHECK1-SAME: (i32* noalias noundef [[DOTGLOBAL_TID_:%.*]], i32* noalias noundef [[DOTBOUND_TID_:%.*]], i32* noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i16* noundef nonnull align 2 dereferenceable(2) [[AA:%.*]], [10 x i32]* noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR1]] {
132 // CHECK1-NEXT:  entry:
133 // CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
134 // CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
135 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca i32*, align 8
136 // CHECK1-NEXT:    [[AA_ADDR:%.*]] = alloca i16*, align 8
137 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 8
138 // CHECK1-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
139 // CHECK1-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
140 // CHECK1-NEXT:    store i32* [[A]], i32** [[A_ADDR]], align 8
141 // CHECK1-NEXT:    store i16* [[AA]], i16** [[AA_ADDR]], align 8
142 // CHECK1-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 8
143 // CHECK1-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 8
144 // CHECK1-NEXT:    [[TMP1:%.*]] = load i16*, i16** [[AA_ADDR]], align 8
145 // CHECK1-NEXT:    [[TMP2:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 8
146 // CHECK1-NEXT:    [[TMP3:%.*]] = load i32, i32* [[TMP0]], align 4
147 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], 1
148 // CHECK1-NEXT:    store i32 [[ADD]], i32* [[TMP0]], align 4
149 // CHECK1-NEXT:    [[TMP4:%.*]] = load i16, i16* [[TMP1]], align 2
150 // CHECK1-NEXT:    [[CONV:%.*]] = sext i16 [[TMP4]] to i32
151 // CHECK1-NEXT:    [[ADD1:%.*]] = add nsw i32 [[CONV]], 1
152 // CHECK1-NEXT:    [[CONV2:%.*]] = trunc i32 [[ADD1]] to i16
153 // CHECK1-NEXT:    store i16 [[CONV2]], i16* [[TMP1]], align 2
154 // CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[TMP2]], i64 0, i64 2
155 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
156 // CHECK1-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP5]], 1
157 // CHECK1-NEXT:    store i32 [[ADD3]], i32* [[ARRAYIDX]], align 4
158 // CHECK1-NEXT:    ret void
159 //
160 //
161 // CHECK2-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l30
162 // CHECK2-SAME: (i16* noundef nonnull align 2 dereferenceable(2) [[AA:%.*]]) #[[ATTR0:[0-9]+]] {
163 // CHECK2-NEXT:  entry:
164 // CHECK2-NEXT:    [[AA_ADDR:%.*]] = alloca i16*, align 4
165 // CHECK2-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x i8*], align 4
166 // CHECK2-NEXT:    store i16* [[AA]], i16** [[AA_ADDR]], align 4
167 // CHECK2-NEXT:    [[TMP0:%.*]] = load i16*, i16** [[AA_ADDR]], align 4
168 // CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1:[0-9]+]], i8 2, i1 false, i1 true)
169 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
170 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
171 // CHECK2:       user_code.entry:
172 // CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2:[0-9]+]])
173 // CHECK2-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
174 // CHECK2-NEXT:    [[TMP4:%.*]] = bitcast i16* [[TMP0]] to i8*
175 // CHECK2-NEXT:    store i8* [[TMP4]], i8** [[TMP3]], align 4
176 // CHECK2-NEXT:    [[TMP5:%.*]] = bitcast [1 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
177 // CHECK2-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB2]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i16*)* @__omp_outlined__ to i8*), i8* null, i8** [[TMP5]], i32 1)
178 // CHECK2-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 true)
179 // CHECK2-NEXT:    ret void
180 // CHECK2:       worker.exit:
181 // CHECK2-NEXT:    ret void
182 //
183 //
184 // CHECK2-LABEL: define {{[^@]+}}@__omp_outlined__
185 // CHECK2-SAME: (i32* noalias noundef [[DOTGLOBAL_TID_:%.*]], i32* noalias noundef [[DOTBOUND_TID_:%.*]], i16* noundef nonnull align 2 dereferenceable(2) [[AA:%.*]]) #[[ATTR1:[0-9]+]] {
186 // CHECK2-NEXT:  entry:
187 // CHECK2-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
188 // CHECK2-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
189 // CHECK2-NEXT:    [[AA_ADDR:%.*]] = alloca i16*, align 4
190 // CHECK2-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
191 // CHECK2-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
192 // CHECK2-NEXT:    store i16* [[AA]], i16** [[AA_ADDR]], align 4
193 // CHECK2-NEXT:    [[TMP0:%.*]] = load i16*, i16** [[AA_ADDR]], align 4
194 // CHECK2-NEXT:    [[TMP1:%.*]] = load i16, i16* [[TMP0]], align 2
195 // CHECK2-NEXT:    [[CONV:%.*]] = sext i16 [[TMP1]] to i32
196 // CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 [[CONV]], 1
197 // CHECK2-NEXT:    [[CONV1:%.*]] = trunc i32 [[ADD]] to i16
198 // CHECK2-NEXT:    store i16 [[CONV1]], i16* [[TMP0]], align 2
199 // CHECK2-NEXT:    ret void
200 //
201 //
202 // CHECK2-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l35
203 // CHECK2-SAME: (i32* noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i16* noundef nonnull align 2 dereferenceable(2) [[AA:%.*]], [10 x i32]* noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
204 // CHECK2-NEXT:  entry:
205 // CHECK2-NEXT:    [[A_ADDR:%.*]] = alloca i32*, align 4
206 // CHECK2-NEXT:    [[AA_ADDR:%.*]] = alloca i16*, align 4
207 // CHECK2-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
208 // CHECK2-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x i8*], align 4
209 // CHECK2-NEXT:    store i32* [[A]], i32** [[A_ADDR]], align 4
210 // CHECK2-NEXT:    store i16* [[AA]], i16** [[AA_ADDR]], align 4
211 // CHECK2-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
212 // CHECK2-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 4
213 // CHECK2-NEXT:    [[TMP1:%.*]] = load i16*, i16** [[AA_ADDR]], align 4
214 // CHECK2-NEXT:    [[TMP2:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
215 // CHECK2-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i8 2, i1 false, i1 true)
216 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP3]], -1
217 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
218 // CHECK2:       user_code.entry:
219 // CHECK2-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2]])
220 // CHECK2-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
221 // CHECK2-NEXT:    [[TMP6:%.*]] = bitcast i32* [[TMP0]] to i8*
222 // CHECK2-NEXT:    store i8* [[TMP6]], i8** [[TMP5]], align 4
223 // CHECK2-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
224 // CHECK2-NEXT:    [[TMP8:%.*]] = bitcast i16* [[TMP1]] to i8*
225 // CHECK2-NEXT:    store i8* [[TMP8]], i8** [[TMP7]], align 4
226 // CHECK2-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
227 // CHECK2-NEXT:    [[TMP10:%.*]] = bitcast [10 x i32]* [[TMP2]] to i8*
228 // CHECK2-NEXT:    store i8* [[TMP10]], i8** [[TMP9]], align 4
229 // CHECK2-NEXT:    [[TMP11:%.*]] = bitcast [3 x i8*]* [[CAPTURED_VARS_ADDRS]] to i8**
230 // CHECK2-NEXT:    call void @__kmpc_parallel_51(%struct.ident_t* @[[GLOB2]], i32 [[TMP4]], i32 1, i32 -1, i32 -1, i8* bitcast (void (i32*, i32*, i32*, i16*, [10 x i32]*)* @__omp_outlined__1 to i8*), i8* null, i8** [[TMP11]], i32 3)
231 // CHECK2-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i8 2, i1 true)
232 // CHECK2-NEXT:    ret void
233 // CHECK2:       worker.exit:
234 // CHECK2-NEXT:    ret void
235 //
236 //
237 // CHECK2-LABEL: define {{[^@]+}}@__omp_outlined__1
238 // CHECK2-SAME: (i32* noalias noundef [[DOTGLOBAL_TID_:%.*]], i32* noalias noundef [[DOTBOUND_TID_:%.*]], i32* noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i16* noundef nonnull align 2 dereferenceable(2) [[AA:%.*]], [10 x i32]* noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR1]] {
239 // CHECK2-NEXT:  entry:
240 // CHECK2-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 4
241 // CHECK2-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 4
242 // CHECK2-NEXT:    [[A_ADDR:%.*]] = alloca i32*, align 4
243 // CHECK2-NEXT:    [[AA_ADDR:%.*]] = alloca i16*, align 4
244 // CHECK2-NEXT:    [[B_ADDR:%.*]] = alloca [10 x i32]*, align 4
245 // CHECK2-NEXT:    store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 4
246 // CHECK2-NEXT:    store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 4
247 // CHECK2-NEXT:    store i32* [[A]], i32** [[A_ADDR]], align 4
248 // CHECK2-NEXT:    store i16* [[AA]], i16** [[AA_ADDR]], align 4
249 // CHECK2-NEXT:    store [10 x i32]* [[B]], [10 x i32]** [[B_ADDR]], align 4
250 // CHECK2-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[A_ADDR]], align 4
251 // CHECK2-NEXT:    [[TMP1:%.*]] = load i16*, i16** [[AA_ADDR]], align 4
252 // CHECK2-NEXT:    [[TMP2:%.*]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align 4
253 // CHECK2-NEXT:    [[TMP3:%.*]] = load i32, i32* [[TMP0]], align 4
254 // CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], 1
255 // CHECK2-NEXT:    store i32 [[ADD]], i32* [[TMP0]], align 4
256 // CHECK2-NEXT:    [[TMP4:%.*]] = load i16, i16* [[TMP1]], align 2
257 // CHECK2-NEXT:    [[CONV:%.*]] = sext i16 [[TMP4]] to i32
258 // CHECK2-NEXT:    [[ADD1:%.*]] = add nsw i32 [[CONV]], 1
259 // CHECK2-NEXT:    [[CONV2:%.*]] = trunc i32 [[ADD1]] to i16
260 // CHECK2-NEXT:    store i16 [[CONV2]], i16* [[TMP1]], align 2
261 // CHECK2-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], [10 x i32]* [[TMP2]], i32 0, i32 2
262 // CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
263 // CHECK2-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP5]], 1
264 // CHECK2-NEXT:    store i32 [[ADD3]], i32* [[ARRAYIDX]], align 4
265 // CHECK2-NEXT:    ret void
266 //
267