1// RUN: %clang_cc1 %s -emit-llvm -o - -triple=armv7-apple-ios -std=c11 | FileCheck %s
2
3// There isn't really anything special about iOS; it just happens to
4// only deploy on processors with native atomics support, so it's a good
5// way to test those code-paths.
6
7// This work was done in pursuit of <rdar://13338582>.
8
9// CHECK-LABEL: define void @testFloat(float*
10void testFloat(_Atomic(float) *fp) {
11// CHECK:      [[FP:%.*]] = alloca float*
12// CHECK-NEXT: [[X:%.*]] = alloca float
13// CHECK-NEXT: [[F:%.*]] = alloca float
14// CHECK-NEXT: store float* {{%.*}}, float** [[FP]]
15
16// CHECK-NEXT: [[T0:%.*]] = load float*, float** [[FP]]
17// CHECK-NEXT: store float 1.000000e+00, float* [[T0]], align 4
18  __c11_atomic_init(fp, 1.0f);
19
20// CHECK-NEXT: store float 2.000000e+00, float* [[X]], align 4
21  _Atomic(float) x = 2.0f;
22
23// CHECK-NEXT: [[T0:%.*]] = load float*, float** [[FP]]
24// CHECK-NEXT: [[T1:%.*]] = bitcast float* [[T0]] to i32*
25// CHECK-NEXT: [[T2:%.*]] = load atomic i32, i32* [[T1]] seq_cst, align 4
26// CHECK-NEXT: [[T3:%.*]] = bitcast i32 [[T2]] to float
27// CHECK-NEXT: store float [[T3]], float* [[F]]
28  float f = *fp;
29
30// CHECK-NEXT: [[T0:%.*]] = load float, float* [[F]], align 4
31// CHECK-NEXT: [[T1:%.*]] = load float*, float** [[FP]], align 4
32// CHECK-NEXT: [[T2:%.*]] = bitcast float [[T0]] to i32
33// CHECK-NEXT: [[T3:%.*]] = bitcast float* [[T1]] to i32*
34// CHECK-NEXT: store atomic i32 [[T2]], i32* [[T3]] seq_cst, align 4
35  *fp = f;
36
37// CHECK-NEXT: ret void
38}
39
40// CHECK: define void @testComplexFloat([[CF:{ float, float }]]*
41void testComplexFloat(_Atomic(_Complex float) *fp) {
42// CHECK:      [[FP:%.*]] = alloca [[CF]]*, align 4
43// CHECK-NEXT: [[X:%.*]] = alloca [[CF]], align 8
44// CHECK-NEXT: [[F:%.*]] = alloca [[CF]], align 4
45// CHECK-NEXT: [[TMP0:%.*]] = alloca [[CF]], align 8
46// CHECK-NEXT: [[TMP1:%.*]] = alloca [[CF]], align 8
47// CHECK-NEXT: store [[CF]]*
48
49// CHECK-NEXT: [[P:%.*]] = load [[CF]]*, [[CF]]** [[FP]]
50// CHECK-NEXT: [[T0:%.*]] = getelementptr inbounds [[CF]], [[CF]]* [[P]], i32 0, i32 0
51// CHECK-NEXT: [[T1:%.*]] = getelementptr inbounds [[CF]], [[CF]]* [[P]], i32 0, i32 1
52// CHECK-NEXT: store float 1.000000e+00, float* [[T0]]
53// CHECK-NEXT: store float 0.000000e+00, float* [[T1]]
54  __c11_atomic_init(fp, 1.0f);
55
56// CHECK-NEXT: [[T0:%.*]] = getelementptr inbounds [[CF]], [[CF]]* [[X]], i32 0, i32 0
57// CHECK-NEXT: [[T1:%.*]] = getelementptr inbounds [[CF]], [[CF]]* [[X]], i32 0, i32 1
58// CHECK-NEXT: store float 2.000000e+00, float* [[T0]]
59// CHECK-NEXT: store float 0.000000e+00, float* [[T1]]
60  _Atomic(_Complex float) x = 2.0f;
61
62// CHECK-NEXT: [[T0:%.*]] = load [[CF]]*, [[CF]]** [[FP]]
63// CHECK-NEXT: [[T1:%.*]] = bitcast [[CF]]* [[T0]] to i64*
64// CHECK-NEXT: [[T2:%.*]] = load atomic i64, i64* [[T1]] seq_cst, align 8
65// CHECK-NEXT: [[T3:%.*]] = bitcast [[CF]]* [[TMP0]] to i64*
66// CHECK-NEXT: store i64 [[T2]], i64* [[T3]], align 8
67// CHECK-NEXT: [[T0:%.*]] = getelementptr inbounds [[CF]], [[CF]]* [[TMP0]], i32 0, i32 0
68// CHECK-NEXT: [[R:%.*]] = load float, float* [[T0]]
69// CHECK-NEXT: [[T0:%.*]] = getelementptr inbounds [[CF]], [[CF]]* [[TMP0]], i32 0, i32 1
70// CHECK-NEXT: [[I:%.*]] = load float, float* [[T0]]
71// CHECK-NEXT: [[T0:%.*]] = getelementptr inbounds [[CF]], [[CF]]* [[F]], i32 0, i32 0
72// CHECK-NEXT: [[T1:%.*]] = getelementptr inbounds [[CF]], [[CF]]* [[F]], i32 0, i32 1
73// CHECK-NEXT: store float [[R]], float* [[T0]]
74// CHECK-NEXT: store float [[I]], float* [[T1]]
75  _Complex float f = *fp;
76
77// CHECK-NEXT: [[T0:%.*]] = getelementptr inbounds [[CF]], [[CF]]* [[F]], i32 0, i32 0
78// CHECK-NEXT: [[R:%.*]] = load float, float* [[T0]]
79// CHECK-NEXT: [[T0:%.*]] = getelementptr inbounds [[CF]], [[CF]]* [[F]], i32 0, i32 1
80// CHECK-NEXT: [[I:%.*]] = load float, float* [[T0]]
81// CHECK-NEXT: [[DEST:%.*]] = load [[CF]]*, [[CF]]** [[FP]], align 4
82// CHECK-NEXT: [[T0:%.*]] = getelementptr inbounds [[CF]], [[CF]]* [[TMP1]], i32 0, i32 0
83// CHECK-NEXT: [[T1:%.*]] = getelementptr inbounds [[CF]], [[CF]]* [[TMP1]], i32 0, i32 1
84// CHECK-NEXT: store float [[R]], float* [[T0]]
85// CHECK-NEXT: store float [[I]], float* [[T1]]
86// CHECK-NEXT: [[T0:%.*]] = bitcast [[CF]]* [[TMP1]] to i64*
87// CHECK-NEXT: [[T1:%.*]] = load i64, i64* [[T0]], align 8
88// CHECK-NEXT: [[T2:%.*]] = bitcast [[CF]]* [[DEST]] to i64*
89// CHECK-NEXT: store atomic i64 [[T1]], i64* [[T2]] seq_cst, align 8
90  *fp = f;
91
92// CHECK-NEXT: ret void
93}
94
95typedef struct { short x, y, z, w; } S;
96// CHECK: define void @testStruct([[S:.*]]*
97void testStruct(_Atomic(S) *fp) {
98// CHECK:      [[FP:%.*]] = alloca [[S]]*, align 4
99// CHECK-NEXT: [[X:%.*]] = alloca [[S]], align 8
100// CHECK-NEXT: [[F:%.*]] = alloca [[S:%.*]], align 2
101// CHECK-NEXT: [[TMP0:%.*]] = alloca [[S]], align 8
102// CHECK-NEXT: store [[S]]*
103
104// CHECK-NEXT: [[P:%.*]] = load [[S]]*, [[S]]** [[FP]]
105// CHECK-NEXT: [[T0:%.*]] = getelementptr inbounds [[S]], [[S]]* [[P]], i32 0, i32 0
106// CHECK-NEXT: store i16 1, i16* [[T0]], align 8
107// CHECK-NEXT: [[T0:%.*]] = getelementptr inbounds [[S]], [[S]]* [[P]], i32 0, i32 1
108// CHECK-NEXT: store i16 2, i16* [[T0]], align 2
109// CHECK-NEXT: [[T0:%.*]] = getelementptr inbounds [[S]], [[S]]* [[P]], i32 0, i32 2
110// CHECK-NEXT: store i16 3, i16* [[T0]], align 4
111// CHECK-NEXT: [[T0:%.*]] = getelementptr inbounds [[S]], [[S]]* [[P]], i32 0, i32 3
112// CHECK-NEXT: store i16 4, i16* [[T0]], align 2
113  __c11_atomic_init(fp, (S){1,2,3,4});
114
115// CHECK-NEXT: [[T0:%.*]] = getelementptr inbounds [[S]], [[S]]* [[X]], i32 0, i32 0
116// CHECK-NEXT: store i16 1, i16* [[T0]], align 8
117// CHECK-NEXT: [[T0:%.*]] = getelementptr inbounds [[S]], [[S]]* [[X]], i32 0, i32 1
118// CHECK-NEXT: store i16 2, i16* [[T0]], align 2
119// CHECK-NEXT: [[T0:%.*]] = getelementptr inbounds [[S]], [[S]]* [[X]], i32 0, i32 2
120// CHECK-NEXT: store i16 3, i16* [[T0]], align 4
121// CHECK-NEXT: [[T0:%.*]] = getelementptr inbounds [[S]], [[S]]* [[X]], i32 0, i32 3
122// CHECK-NEXT: store i16 4, i16* [[T0]], align 2
123  _Atomic(S) x = (S){1,2,3,4};
124
125// CHECK-NEXT: [[T0:%.*]] = load [[S]]*, [[S]]** [[FP]]
126// CHECK-NEXT: [[T1:%.*]] = bitcast [[S]]* [[T0]] to i64*
127// CHECK-NEXT: [[T2:%.*]] = load atomic i64, i64* [[T1]] seq_cst, align 8
128// CHECK-NEXT: [[T3:%.*]] = bitcast [[S]]* [[F]] to i64*
129// CHECK-NEXT: store i64 [[T2]], i64* [[T3]], align 2
130  S f = *fp;
131
132// CHECK-NEXT: [[T0:%.*]] = load [[S]]*, [[S]]** [[FP]]
133// CHECK-NEXT: [[T1:%.*]] = bitcast [[S]]* [[TMP0]] to i8*
134// CHECK-NEXT: [[T2:%.*]] = bitcast [[S]]* [[F]] to i8*
135// CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* [[T1]], i8* [[T2]], i32 8, i32 2, i1 false)
136// CHECK-NEXT: [[T3:%.*]] = bitcast [[S]]* [[TMP0]] to i64*
137// CHECK-NEXT: [[T4:%.*]] = load i64, i64* [[T3]], align 8
138// CHECK-NEXT: [[T5:%.*]] = bitcast [[S]]* [[T0]] to i64*
139// CHECK-NEXT: store atomic i64 [[T4]], i64* [[T5]] seq_cst, align 8
140  *fp = f;
141
142// CHECK-NEXT: ret void
143}
144
145typedef struct { short x, y, z; } PS;
146// CHECK: define void @testPromotedStruct([[APS:.*]]*
147void testPromotedStruct(_Atomic(PS) *fp) {
148// CHECK:      [[FP:%.*]] = alloca [[APS]]*, align 4
149// CHECK-NEXT: [[X:%.*]] = alloca [[APS]], align 8
150// CHECK-NEXT: [[F:%.*]] = alloca [[PS:%.*]], align 2
151// CHECK-NEXT: [[TMP0:%.*]] = alloca [[APS]], align 8
152// CHECK-NEXT: [[TMP1:%.*]] = alloca [[APS]], align 8
153// CHECK-NEXT: store [[APS]]*
154
155// CHECK-NEXT: [[P:%.*]] = load [[APS]]*, [[APS]]** [[FP]]
156// CHECK-NEXT: [[T0:%.*]] = bitcast [[APS]]* [[P]] to i8*
157// CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* [[T0]], i8 0, i64 8, i32 8, i1 false)
158// CHECK-NEXT: [[T0:%.*]] = getelementptr inbounds [[APS]], [[APS]]* [[P]], i32 0, i32 0
159// CHECK-NEXT: [[T1:%.*]] = getelementptr inbounds [[PS]], [[PS]]* [[T0]], i32 0, i32 0
160// CHECK-NEXT: store i16 1, i16* [[T1]], align 8
161// CHECK-NEXT: [[T1:%.*]] = getelementptr inbounds [[PS]], [[PS]]* [[T0]], i32 0, i32 1
162// CHECK-NEXT: store i16 2, i16* [[T1]], align 2
163// CHECK-NEXT: [[T1:%.*]] = getelementptr inbounds [[PS]], [[PS]]* [[T0]], i32 0, i32 2
164// CHECK-NEXT: store i16 3, i16* [[T1]], align 4
165  __c11_atomic_init(fp, (PS){1,2,3});
166
167// CHECK-NEXT: [[T0:%.*]] = bitcast [[APS]]* [[X]] to i8*
168// CHECK-NEXT: call void @llvm.memset.p0i8.i32(i8* [[T0]], i8 0, i32 8, i32 8, i1 false)
169// CHECK-NEXT: [[T0:%.*]] = getelementptr inbounds [[APS]], [[APS]]* [[X]], i32 0, i32 0
170// CHECK-NEXT: [[T1:%.*]] = getelementptr inbounds [[PS]], [[PS]]* [[T0]], i32 0, i32 0
171// CHECK-NEXT: store i16 1, i16* [[T1]], align 8
172// CHECK-NEXT: [[T1:%.*]] = getelementptr inbounds [[PS]], [[PS]]* [[T0]], i32 0, i32 1
173// CHECK-NEXT: store i16 2, i16* [[T1]], align 2
174// CHECK-NEXT: [[T1:%.*]] = getelementptr inbounds [[PS]], [[PS]]* [[T0]], i32 0, i32 2
175// CHECK-NEXT: store i16 3, i16* [[T1]], align 4
176  _Atomic(PS) x = (PS){1,2,3};
177
178// CHECK-NEXT: [[T0:%.*]] = load [[APS]]*, [[APS]]** [[FP]]
179// CHECK-NEXT: [[T1:%.*]] = bitcast [[APS]]* [[T0]] to i64*
180// CHECK-NEXT: [[T2:%.*]] = load atomic i64, i64* [[T1]] seq_cst, align 8
181// CHECK-NEXT: [[T3:%.*]] = bitcast [[APS]]* [[TMP0]] to i64*
182// CHECK-NEXT: store i64 [[T2]], i64* [[T3]], align 8
183// CHECK-NEXT: [[T0:%.*]] = getelementptr inbounds [[APS]], [[APS]]* [[TMP0]], i32 0, i32 0
184// CHECK-NEXT: [[T1:%.*]] = bitcast [[PS]]* [[F]] to i8*
185// CHECK-NEXT: [[T2:%.*]] = bitcast [[PS]]* [[T0]] to i8*
186// CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* [[T1]], i8* [[T2]], i32 6, i32 2, i1 false)
187  PS f = *fp;
188
189// CHECK-NEXT: [[T0:%.*]] = load [[APS]]*, [[APS]]** [[FP]]
190// CHECK-NEXT: [[T1:%.*]] = bitcast { %struct.PS, [2 x i8] }* [[TMP1]] to i8*
191// CHECK-NEXT: call void @llvm.memset.p0i8.i32(i8* [[T1]], i8 0, i32 8, i32 8, i1 false)
192// CHECK-NEXT: [[T1:%.*]] = getelementptr inbounds [[APS]], [[APS]]* [[TMP1]], i32 0, i32 0
193// CHECK-NEXT: [[T2:%.*]] = bitcast [[PS]]* [[T1]] to i8*
194// CHECK-NEXT: [[T3:%.*]] = bitcast [[PS]]* [[F]] to i8*
195// CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* [[T2]], i8* [[T3]], i32 6, i32 2, i1 false)
196// CHECK-NEXT: [[T4:%.*]] = bitcast [[APS]]* [[TMP1]] to i64*
197// CHECK-NEXT: [[T5:%.*]] = load i64, i64* [[T4]], align 8
198// CHECK-NEXT: [[T6:%.*]] = bitcast [[APS]]* [[T0]] to i64*
199// CHECK-NEXT: store atomic i64 [[T5]], i64* [[T6]] seq_cst, align 8
200  *fp = f;
201
202// CHECK-NEXT: ret void
203}
204
205PS test_promoted_load(_Atomic(PS) *addr) {
206  // CHECK-LABEL: @test_promoted_load(%struct.PS* noalias sret %agg.result, { %struct.PS, [2 x i8] }* %addr)
207  // CHECK:   [[ADDR_ARG:%.*]] = alloca { %struct.PS, [2 x i8] }*, align 4
208  // CHECK:   [[ATOMIC_RES:%.*]] = alloca { %struct.PS, [2 x i8] }, align 8
209  // CHECK:   store { %struct.PS, [2 x i8] }* %addr, { %struct.PS, [2 x i8] }** [[ADDR_ARG]], align 4
210  // CHECK:   [[ADDR:%.*]] = load { %struct.PS, [2 x i8] }*, { %struct.PS, [2 x i8] }** [[ADDR_ARG]], align 4
211  // CHECK:   [[ADDR64:%.*]] = bitcast { %struct.PS, [2 x i8] }* [[ADDR]] to i64*
212  // CHECK:   [[ATOMIC_RES64:%.*]] = bitcast { %struct.PS, [2 x i8] }* [[ATOMIC_RES]] to i64*
213  // CHECK:   [[VAL:%.*]] = load atomic i64, i64* [[ADDR64]] seq_cst, align 8
214  // CHECK:   store i64 [[VAL]], i64* [[ATOMIC_RES64]], align 8
215  // CHECK:   [[ATOMIC_RES_STRUCT:%.*]] = bitcast i64* [[ATOMIC_RES64]] to %struct.PS*
216  // CHECK:   [[AGG_RESULT8:%.*]] = bitcast %struct.PS* %agg.result to i8*
217  // CHECK:   [[ATOMIC_RES8:%.*]] = bitcast %struct.PS* [[ATOMIC_RES_STRUCT]] to i8*
218  // CHECK:   call void @llvm.memcpy.p0i8.p0i8.i32(i8* [[AGG_RESULT8]], i8* [[ATOMIC_RES8]], i32 6, i32 2, i1 false)
219
220  return __c11_atomic_load(addr, 5);
221}
222
223void test_promoted_store(_Atomic(PS) *addr, PS *val) {
224  // CHECK-LABEL: @test_promoted_store({ %struct.PS, [2 x i8] }* %addr, %struct.PS* %val)
225  // CHECK:   [[ADDR_ARG:%.*]] = alloca { %struct.PS, [2 x i8] }*, align 4
226  // CHECK:   [[VAL_ARG:%.*]] = alloca %struct.PS*, align 4
227  // CHECK:   [[NONATOMIC_TMP:%.*]] = alloca %struct.PS, align 2
228  // CHECK:   [[ATOMIC_VAL:%.*]] = alloca { %struct.PS, [2 x i8] }, align 8
229  // CHECK:   store { %struct.PS, [2 x i8] }* %addr, { %struct.PS, [2 x i8] }** [[ADDR_ARG]], align 4
230  // CHECK:   store %struct.PS* %val, %struct.PS** [[VAL_ARG]], align 4
231  // CHECK:   [[ADDR:%.*]] = load { %struct.PS, [2 x i8] }*, { %struct.PS, [2 x i8] }** [[ADDR_ARG]], align 4
232  // CHECK:   [[VAL:%.*]] = load %struct.PS*, %struct.PS** [[VAL_ARG]], align 4
233  // CHECK:   [[NONATOMIC_TMP8:%.*]] = bitcast %struct.PS* [[NONATOMIC_TMP]] to i8*
234  // CHECK:   [[VAL8:%.*]] = bitcast %struct.PS* [[VAL]] to i8*
235  // CHECK:   call void @llvm.memcpy.p0i8.p0i8.i32(i8* [[NONATOMIC_TMP8]], i8* [[VAL8]], i32 6, i32 2, i1 false)
236  // CHECK:   [[ADDR64:%.*]] = bitcast { %struct.PS, [2 x i8] }* [[ADDR]] to i64*
237  // CHECK:   [[ATOMIC_VAL8:%.*]] = bitcast { %struct.PS, [2 x i8] }* [[ATOMIC_VAL]] to i8*
238  // CHECK:   [[NONATOMIC_TMP8:%.*]] = bitcast %struct.PS* [[NONATOMIC_TMP]] to i8*
239  // CHECK:   call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[ATOMIC_VAL8]], i8* [[NONATOMIC_TMP8]], i64 6, i32 2, i1 false)
240  // CHECK:   [[ATOMIC_VAL64:%.*]] = bitcast { %struct.PS, [2 x i8] }* [[ATOMIC_VAL]] to i64*
241  // CHECK:   [[VAL64:%.*]] = load i64, i64* [[ATOMIC_VAL64]], align 8
242  // CHECK:   store atomic i64 [[VAL64]], i64* [[ADDR64]] seq_cst, align 8
243
244  __c11_atomic_store(addr, *val, 5);
245}
246
247PS test_promoted_exchange(_Atomic(PS) *addr, PS *val) {
248  // CHECK-LABEL: @test_promoted_exchange(%struct.PS* noalias sret %agg.result, { %struct.PS, [2 x i8] }* %addr, %struct.PS* %val)
249  // CHECK:   [[ADDR_ARG:%.*]] = alloca { %struct.PS, [2 x i8] }*, align 4
250  // CHECK:   [[VAL_ARG:%.*]] = alloca %struct.PS*, align 4
251  // CHECK:   [[NONATOMIC_TMP:%.*]] = alloca %struct.PS, align 2
252  // CHECK:   [[ATOMIC_VAL:%.*]] = alloca { %struct.PS, [2 x i8] }, align 8
253  // CHECK:   [[ATOMIC_RES:%.*]] = alloca { %struct.PS, [2 x i8] }, align 8
254  // CHECK:   store { %struct.PS, [2 x i8] }* %addr, { %struct.PS, [2 x i8] }** [[ADDR_ARG]], align 4
255  // CHECK:   store %struct.PS* %val, %struct.PS** [[VAL_ARG]], align 4
256  // CHECK:   [[ADDR:%.*]] = load { %struct.PS, [2 x i8] }*, { %struct.PS, [2 x i8] }** [[ADDR_ARG]], align 4
257  // CHECK:   [[VAL:%.*]] = load %struct.PS*, %struct.PS** [[VAL_ARG]], align 4
258  // CHECK:   [[NONATOMIC_TMP8:%.*]] = bitcast %struct.PS* [[NONATOMIC_TMP]] to i8*
259  // CHECK:   [[VAL8:%.*]] = bitcast %struct.PS* [[VAL]] to i8*
260  // CHECK:   call void @llvm.memcpy.p0i8.p0i8.i32(i8* [[NONATOMIC_TMP8]], i8* [[VAL8]], i32 6, i32 2, i1 false)
261  // CHECK:   [[ADDR64:%.*]] = bitcast { %struct.PS, [2 x i8] }* [[ADDR]] to i64*
262  // CHECK:   [[ATOMIC_VAL8:%.*]] = bitcast { %struct.PS, [2 x i8] }* [[ATOMIC_VAL]] to i8*
263  // CHECK:   [[NONATOMIC_TMP8:%.*]] = bitcast %struct.PS* [[NONATOMIC_TMP]] to i8*
264  // CHECK:   call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[ATOMIC_VAL8]], i8* [[NONATOMIC_TMP8]], i64 6, i32 2, i1 false)
265  // CHECK:   [[ATOMIC_VAL64:%.*]] = bitcast { %struct.PS, [2 x i8] }* [[ATOMIC_VAL]] to i64*
266  // CHECK:   [[ATOMIC_RES64:%.*]] = bitcast { %struct.PS, [2 x i8] }* [[ATOMIC_RES]] to i64*
267  // CHECK:   [[VAL64:%.*]] = load i64, i64* [[ATOMIC_VAL64]], align 8
268  // CHECK:   [[RES:%.*]] = atomicrmw xchg i64* [[ADDR64]], i64 [[VAL64]] seq_cst
269  // CHECK:   store i64 [[RES]], i64* [[ATOMIC_RES64]], align 8
270  // CHECK:   [[ATOMIC_RES_STRUCT:%.*]] = bitcast i64* [[ATOMIC_RES64]] to %struct.PS*
271  // CHECK:   [[AGG_RESULT8:%.*]] = bitcast %struct.PS* %agg.result to i8*
272  // CHECK:   [[ATOMIC_RES8:%.*]] = bitcast %struct.PS* [[ATOMIC_RES_STRUCT]] to i8*
273  // CHECK:   call void @llvm.memcpy.p0i8.p0i8.i32(i8* [[AGG_RESULT8]], i8* [[ATOMIC_RES8]], i32 6, i32 2, i1 false)
274  return __c11_atomic_exchange(addr, *val, 5);
275}
276
277_Bool test_promoted_cmpxchg(_Atomic(PS) *addr, PS *desired, PS *new) {
278  // CHECK:   define zeroext i1 @test_promoted_cmpxchg({ %struct.PS, [2 x i8] }* %addr, %struct.PS* %desired, %struct.PS* %new) #0 {
279  // CHECK:   [[ADDR_ARG:%.*]] = alloca { %struct.PS, [2 x i8] }*, align 4
280  // CHECK:   [[DESIRED_ARG:%.*]] = alloca %struct.PS*, align 4
281  // CHECK:   [[NEW_ARG:%.*]] = alloca %struct.PS*, align 4
282  // CHECK:   [[NONATOMIC_TMP:%.*]] = alloca %struct.PS, align 2
283  // CHECK:   [[ATOMIC_DESIRED:%.*]] = alloca { %struct.PS, [2 x i8] }, align 8
284  // CHECK:   [[ATOMIC_NEW:%.*]] = alloca { %struct.PS, [2 x i8] }, align 8
285  // CHECK:   [[RES_ADDR:%.*]] = alloca i8, align 1
286  // CHECK:   store { %struct.PS, [2 x i8] }* %addr, { %struct.PS, [2 x i8] }** [[ADDR_ARG]], align 4
287  // CHECK:   store %struct.PS* %desired, %struct.PS** [[DESIRED_ARG]], align 4
288  // CHECK:   store %struct.PS* %new, %struct.PS** [[NEW_ARG]], align 4
289  // CHECK:   [[ADDR:%.*]] = load { %struct.PS, [2 x i8] }*, { %struct.PS, [2 x i8] }** [[ADDR_ARG]], align 4
290  // CHECK:   [[DESIRED:%.*]] = load %struct.PS*, %struct.PS** [[DESIRED_ARG]], align 4
291  // CHECK:   [[NEW:%.*]] = load %struct.PS*, %struct.PS** [[NEW_ARG]], align 4
292  // CHECK:   [[NONATOMIC_TMP8:%.*]] = bitcast %struct.PS* [[NONATOMIC_TMP]] to i8*
293  // CHECK:   [[NEW8:%.*]] = bitcast %struct.PS* [[NEW]] to i8*
294  // CHECK:   call void @llvm.memcpy.p0i8.p0i8.i32(i8* [[NONATOMIC_TMP8]], i8* [[NEW8]], i32 6, i32 2, i1 false)
295  // CHECK:   [[ADDR64:%.*]] = bitcast { %struct.PS, [2 x i8] }* [[ADDR]] to i64*
296  // CHECK:   [[ATOMIC_DESIRED8:%.*]] = bitcast { %struct.PS, [2 x i8] }* [[ATOMIC_DESIRED:%.*]] to i8*
297  // CHECK:   [[DESIRED8:%.*]] = bitcast %struct.PS* [[DESIRED]] to i8*
298  // CHECK:   call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[ATOMIC_DESIRED8]], i8* [[DESIRED8]], i64 6, i32 2, i1 false)
299  // CHECK:   [[ATOMIC_DESIRED64:%.*]] = bitcast { %struct.PS, [2 x i8] }* [[ATOMIC_DESIRED:%.*]] to i64*
300  // CHECK:   [[ATOMIC_NEW8:%.*]] = bitcast { %struct.PS, [2 x i8] }* [[ATOMIC_NEW]] to i8*
301  // CHECK:   [[NONATOMIC_TMP8:%.*]] = bitcast %struct.PS* [[NONATOMIC_TMP]] to i8*
302  // CHECK:   call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[ATOMIC_NEW8]], i8* [[NONATOMIC_TMP8]], i64 6, i32 2, i1 false)
303  // CHECK:   [[ATOMIC_NEW64:%.*]] = bitcast { %struct.PS, [2 x i8] }* [[ATOMIC_NEW]] to i64*
304  // CHECK:   [[ATOMIC_DESIRED_VAL64:%.*]] = load i64, i64* [[ATOMIC_DESIRED64]], align 8
305  // CHECK:   [[ATOMIC_NEW_VAL64:%.*]] = load i64, i64* [[ATOMIC_NEW64]], align 8
306  // CHECK:   [[RES:%.*]] = cmpxchg i64* [[ADDR64]], i64 [[ATOMIC_DESIRED_VAL64]], i64 [[ATOMIC_NEW_VAL64]] seq_cst seq_cst
307  // CHECK:   [[RES_VAL64:%.*]] = extractvalue { i64, i1 } [[RES]], 0
308  // CHECK:   [[RES_BOOL:%.*]] = extractvalue { i64, i1 } [[RES]], 1
309  // CHECK:   br i1 [[RES_BOOL]], label {{%.*}}, label {{%.*}}
310
311  // CHECK:   store i64 [[RES_VAL64]], i64* [[ATOMIC_DESIRED64]], align 8
312  // CHECK:   br label {{%.*}}
313
314  // CHECK:   [[RES_BOOL8:%.*]] = zext i1 [[RES_BOOL]] to i8
315  // CHECK:   store i8 [[RES_BOOL8]], i8* [[RES_ADDR]], align 1
316  // CHECK:   [[RES_BOOL8:%.*]] = load i8, i8* [[RES_ADDR]], align 1
317  // CHECK:   [[RETVAL:%.*]] = trunc i8 [[RES_BOOL8]] to i1
318  // CHECK:   ret i1 [[RETVAL]]
319
320  return __c11_atomic_compare_exchange_strong(addr, desired, *new, 5, 5);
321}
322