1 // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp=libiomp5 -x c -emit-llvm %s -o - | FileCheck %s
2 // RUN: %clang_cc1 -fopenmp=libiomp5 -x c -triple x86_64-apple-darwin10 -emit-pch -o %t %s
3 // RUN: %clang_cc1 -fopenmp=libiomp5 -x c -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s
4 // expected-no-diagnostics
5 
6 #ifndef HEADER
7 #define HEADER
8 
9 _Bool bv, bx;
10 char cv, cx;
11 unsigned char ucv, ucx;
12 short sv, sx;
13 unsigned short usv, usx;
14 int iv, ix;
15 unsigned int uiv, uix;
16 long lv, lx;
17 unsigned long ulv, ulx;
18 long long llv, llx;
19 unsigned long long ullv, ullx;
20 float fv, fx;
21 double dv, dx;
22 long double ldv, ldx;
23 _Complex int civ, cix;
24 _Complex float cfv, cfx;
25 _Complex double cdv, cdx;
26 
27 typedef int int4 __attribute__((__vector_size__(16)));
28 int4 int4x;
29 
30 struct BitFields {
31   int : 32;
32   int a : 31;
33 } bfx;
34 
35 struct BitFields_packed {
36   int : 32;
37   int a : 31;
38 } __attribute__ ((__packed__)) bfx_packed;
39 
40 struct BitFields2 {
41   int : 31;
42   int a : 1;
43 } bfx2;
44 
45 struct BitFields2_packed {
46   int : 31;
47   int a : 1;
48 } __attribute__ ((__packed__)) bfx2_packed;
49 
50 struct BitFields3 {
51   int : 11;
52   int a : 14;
53 } bfx3;
54 
55 struct BitFields3_packed {
56   int : 11;
57   int a : 14;
58 } __attribute__ ((__packed__)) bfx3_packed;
59 
60 struct BitFields4 {
61   short : 16;
62   int a: 1;
63   long b : 7;
64 } bfx4;
65 
66 struct BitFields4_packed {
67   short : 16;
68   int a: 1;
69   long b : 7;
70 } __attribute__ ((__packed__)) bfx4_packed;
71 
72 typedef float float2 __attribute__((ext_vector_type(2)));
73 float2 float2x;
74 
75 register int rix __asm__("0");
76 
77 int main() {
78 // CHECK: load i8, i8*
79 // CHECK: store atomic i8
80 #pragma omp atomic write
81   bx = bv;
82 // CHECK: load i8, i8*
83 // CHECK: store atomic i8
84 #pragma omp atomic write
85   cx = cv;
86 // CHECK: load i8, i8*
87 // CHECK: store atomic i8
88 #pragma omp atomic write
89   ucx = ucv;
90 // CHECK: load i16, i16*
91 // CHECK: store atomic i16
92 #pragma omp atomic write
93   sx = sv;
94 // CHECK: load i16, i16*
95 // CHECK: store atomic i16
96 #pragma omp atomic write
97   usx = usv;
98 // CHECK: load i32, i32*
99 // CHECK: store atomic i32
100 #pragma omp atomic write
101   ix = iv;
102 // CHECK: load i32, i32*
103 // CHECK: store atomic i32
104 #pragma omp atomic write
105   uix = uiv;
106 // CHECK: load i64, i64*
107 // CHECK: store atomic i64
108 #pragma omp atomic write
109   lx = lv;
110 // CHECK: load i64, i64*
111 // CHECK: store atomic i64
112 #pragma omp atomic write
113   ulx = ulv;
114 // CHECK: load i64, i64*
115 // CHECK: store atomic i64
116 #pragma omp atomic write
117   llx = llv;
118 // CHECK: load i64, i64*
119 // CHECK: store atomic i64
120 #pragma omp atomic write
121   ullx = ullv;
122 // CHECK: load float, float*
123 // CHECK: bitcast float {{.*}} to i32
124 // CHECK: store atomic i32 {{.*}}, i32* bitcast (float*
125 #pragma omp atomic write
126   fx = fv;
127 // CHECK: load double, double*
128 // CHECK: bitcast double {{.*}} to i64
129 // CHECK: store atomic i64 {{.*}}, i64* bitcast (double*
130 #pragma omp atomic write
131   dx = dv;
132 // CHECK: [[LD:%.+]] = load x86_fp80, x86_fp80*
133 // CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[LDTEMP:%.*]] to i8*
134 // CHECK: call void @llvm.memset.p0i8.i64(i8* [[BITCAST]], i8 0, i64 16, i32 16, i1 false)
135 // CHECK: store x86_fp80 [[LD]], x86_fp80* [[LDTEMP]]
136 // CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[LDTEMP:%.*]] to i128*
137 // CHECK: [[LD:%.+]] = load i128, i128* [[BITCAST]]
138 // CHECK: store atomic i128 [[LD]], i128* bitcast (x86_fp80*
139 #pragma omp atomic write
140   ldx = ldv;
141 // CHECK: [[REAL_VAL:%.+]] = load i32, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.*}}, i32 0, i32 0)
142 // CHECK: [[IMG_VAL:%.+]] = load i32, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.*}}, i32 0, i32 1)
143 // CHECK: [[TEMP_REAL_REF:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP:%.+]], i32 0, i32 0
144 // CHECK: [[TEMP_IMG_REF:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1
145 // CHECK: store i32 [[REAL_VAL]], i32* [[TEMP_REAL_REF]]
146 // CHECK: store i32 [[IMG_VAL]], i32* [[TEMP_IMG_REF]]
147 // CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[TEMP]] to i8*
148 // CHECK: call void @__atomic_store(i64 8, i8* bitcast ({ i32, i32 }* @{{.*}} to i8*), i8* [[BITCAST]], i32 0)
149 #pragma omp atomic write
150   cix = civ;
151 // CHECK: [[REAL_VAL:%.+]] = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @{{.*}}, i32 0, i32 0)
152 // CHECK: [[IMG_VAL:%.+]] = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @{{.*}}, i32 0, i32 1)
153 // CHECK: [[TEMP_REAL_REF:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[TEMP:%.+]], i32 0, i32 0
154 // CHECK: [[TEMP_IMG_REF:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[TEMP]], i32 0, i32 1
155 // CHECK: store float [[REAL_VAL]], float* [[TEMP_REAL_REF]]
156 // CHECK: store float [[IMG_VAL]], float* [[TEMP_IMG_REF]]
157 // CHECK: [[BITCAST:%.+]] = bitcast { float, float }* [[TEMP]] to i8*
158 // CHECK: call void @__atomic_store(i64 8, i8* bitcast ({ float, float }* @{{.*}} to i8*), i8* [[BITCAST]], i32 0)
159 #pragma omp atomic write
160   cfx = cfv;
161 // CHECK: [[REAL_VAL:%.+]] = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @{{.*}}, i32 0, i32 0)
162 // CHECK: [[IMG_VAL:%.+]] = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @{{.*}}, i32 0, i32 1)
163 // CHECK: [[TEMP_REAL_REF:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[TEMP:%.+]], i32 0, i32 0
164 // CHECK: [[TEMP_IMG_REF:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[TEMP]], i32 0, i32 1
165 // CHECK: store double [[REAL_VAL]], double* [[TEMP_REAL_REF]]
166 // CHECK: store double [[IMG_VAL]], double* [[TEMP_IMG_REF]]
167 // CHECK: [[BITCAST:%.+]] = bitcast { double, double }* [[TEMP]] to i8*
168 // CHECK: call void @__atomic_store(i64 16, i8* bitcast ({ double, double }* @{{.*}} to i8*), i8* [[BITCAST]], i32 5)
169 // CHECK: call{{.*}} @__kmpc_flush(
170 #pragma omp atomic seq_cst write
171   cdx = cdv;
172 // CHECK: load i8, i8*
173 // CHECK: store atomic i64
174 #pragma omp atomic write
175   ulx = bv;
176 // CHECK: load i8, i8*
177 // CHECK: store atomic i8
178 #pragma omp atomic write
179   bx = cv;
180 // CHECK: load i8, i8*
181 // CHECK: store atomic i8
182 // CHECK: call{{.*}} @__kmpc_flush(
183 #pragma omp atomic write, seq_cst
184   cx = ucv;
185 // CHECK: load i16, i16*
186 // CHECK: store atomic i64
187 #pragma omp atomic write
188   ulx = sv;
189 // CHECK: load i16, i16*
190 // CHECK: store atomic i64
191 #pragma omp atomic write
192   lx = usv;
193 // CHECK: load i32, i32*
194 // CHECK: store atomic i32
195 // CHECK: call{{.*}} @__kmpc_flush(
196 #pragma omp atomic seq_cst, write
197   uix = iv;
198 // CHECK: load i32, i32*
199 // CHECK: store atomic i32
200 #pragma omp atomic write
201   ix = uiv;
202 // CHECK: load i64, i64*
203 // CHECK: [[VAL:%.+]] = trunc i64 %{{.*}} to i32
204 // CHECK: [[TEMP_REAL_REF:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP:%.+]], i32 0, i32 0
205 // CHECK: [[TEMP_IMG_REF:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1
206 // CHECK: store i32 [[VAL]], i32* [[TEMP_REAL_REF]]
207 // CHECK: store i32 0, i32* [[TEMP_IMG_REF]]
208 // CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[TEMP]] to i8*
209 // CHECK: call void @__atomic_store(i64 8, i8* bitcast ({ i32, i32 }* @{{.+}} to i8*), i8* [[BITCAST]], i32 0)
210 #pragma omp atomic write
211   cix = lv;
212 // CHECK: load i64, i64*
213 // CHECK: store atomic i32 %{{.+}}, i32* bitcast (float*
214 #pragma omp atomic write
215   fx = ulv;
216 // CHECK: load i64, i64*
217 // CHECK: store atomic i64 %{{.+}}, i64* bitcast (double*
218 #pragma omp atomic write
219   dx = llv;
220 // CHECK: load i64, i64*
221 // CHECK: [[VAL:%.+]] = uitofp i64 %{{.+}} to x86_fp80
222 // CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP:%.+]] to i8*
223 // CHECK: call void @llvm.memset.p0i8.i64(i8* [[BITCAST]], i8 0, i64 16, i32 16, i1 false)
224 // CHECK: store x86_fp80 [[VAL]], x86_fp80* [[TEMP]]
225 // CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP]] to i128*
226 // CHECK: [[VAL:%.+]] = load i128, i128* [[BITCAST]]
227 // CHECK: store atomic i128 [[VAL]], i128* bitcast (x86_fp80*
228 #pragma omp atomic write
229   ldx = ullv;
230 // CHECK: load float, float*
231 // CHECK: [[VAL:%.+]] = fptosi float %{{.*}} to i32
232 // CHECK: [[TEMP_REAL_REF:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP:%.+]], i32 0, i32 0
233 // CHECK: [[TEMP_IMG_REF:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1
234 // CHECK: store i32 [[VAL]], i32* [[TEMP_REAL_REF]]
235 // CHECK: store i32 0, i32* [[TEMP_IMG_REF]]
236 // CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[TEMP]] to i8*
237 // CHECK: call void @__atomic_store(i64 8, i8* bitcast ({ i32, i32 }* @{{.+}} to i8*), i8* [[BITCAST]], i32 0)
238 #pragma omp atomic write
239   cix = fv;
240 // CHECK: load double, double*
241 // CHECK: store atomic i16
242 #pragma omp atomic write
243   sx = dv;
244 // CHECK: load x86_fp80, x86_fp80*
245 // CHECK: store atomic i8
246 #pragma omp atomic write
247   bx = ldv;
248 // CHECK: load i32, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.+}}, i32 0, i32 0)
249 // CHECK: load i32, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.+}}, i32 0, i32 1)
250 // CHECK: icmp ne i32 %{{.+}}, 0
251 // CHECK: icmp ne i32 %{{.+}}, 0
252 // CHECK: or i1
253 // CHECK: store atomic i8
254 #pragma omp atomic write
255   bx = civ;
256 // CHECK: load float, float* getelementptr inbounds ({ float, float }, { float, float }* @{{.*}}, i32 0, i32 0)
257 // CHECK: store atomic i16
258 #pragma omp atomic write
259   usx = cfv;
260 // CHECK: load double, double* getelementptr inbounds ({ double, double }, { double, double }* @{{.+}}, i32 0, i32 0)
261 // CHECK: store atomic i64
262 #pragma omp atomic write
263   llx = cdv;
264 // CHECK-DAG: [[IDX:%.+]] = load i16, i16* @{{.+}}
265 // CHECK-DAG: load i8, i8*
266 // CHECK-DAG: [[VEC_ITEM_VAL:%.+]] = zext i1 %{{.+}} to i32
267 // CHECK: [[I128VAL:%.+]] = load atomic i128, i128* bitcast (<4 x i32>* [[DEST:@.+]] to i128*) monotonic
268 // CHECK: [[LD:%.+]] = bitcast i128 [[I128VAL]] to <4 x i32>
269 // CHECK: br label %[[CONT:.+]]
270 // CHECK: [[CONT]]
271 // CHECK: [[OLD_VEC_VAL:%.+]] = phi <4 x i32> [ [[LD]], %{{.+}} ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
272 // CHECK: store <4 x i32> [[OLD_VEC_VAL]], <4 x i32>* [[LDTEMP:%.+]],
273 // CHECK: [[VEC_VAL:%.+]] = load <4 x i32>, <4 x i32>* [[LDTEMP]]
274 // CHECK: [[NEW_VEC_VAL:%.+]] = insertelement <4 x i32> [[VEC_VAL]], i32 [[VEC_ITEM_VAL]], i16 [[IDX]]
275 // CHECK: store <4 x i32> [[NEW_VEC_VAL]], <4 x i32>* [[LDTEMP]]
276 // CHECK: [[NEW_VEC_VAL:%.+]] = load <4 x i32>, <4 x i32>* [[LDTEMP]]
277 // CHECK: [[OLD_I128:%.+]] = bitcast <4 x i32> [[OLD_VEC_VAL]] to i128
278 // CHECK: [[NEW_I128:%.+]] = bitcast <4 x i32> [[NEW_VEC_VAL]] to i128
279 // CHECK: [[RES:%.+]] = cmpxchg i128* bitcast (<4 x i32>* [[DEST]] to i128*), i128 [[OLD_I128]], i128 [[NEW_I128]] monotonic monotonic
280 // CHECK: [[FAILED_I128_OLD_VAL:%.+]] = extractvalue { i128, i1 } [[RES]], 0
281 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i128, i1 } [[RES]], 1
282 // CHECK: [[FAILED_OLD_VAL]] = bitcast i128 [[FAILED_I128_OLD_VAL]] to <4 x i32>
283 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
284 // CHECK: [[EXIT]]
285 #pragma omp atomic write
286   int4x[sv] = bv;
287 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
288 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32
289 // CHECK: [[PREV_VALUE:%.+]] = load atomic i32, i32* bitcast (i8* getelementptr (i8, i8* bitcast (%struct.BitFields* @{{.+}} to i8*), i64 4) to i32*) monotonic
290 // CHECK: br label %[[CONT:.+]]
291 // CHECK: [[CONT]]
292 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i32 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
293 // CHECK: [[BF_VALUE:%.+]] = and i32 [[NEW_VAL]], 2147483647
294 // CHECK: [[BF_CLEAR:%.+]] = and i32 %{{.+}}, -2147483648
295 // CHECK: or i32 [[BF_CLEAR]], [[BF_VALUE]]
296 // CHECK: store i32 %{{.+}}, i32* [[LDTEMP:%.+]]
297 // CHECK: [[NEW_BF_VALUE:%.+]] = load i32, i32* [[LDTEMP]]
298 // CHECK: [[RES:%.+]] = cmpxchg i32* bitcast (i8* getelementptr (i8, i8* bitcast (%struct.BitFields* @{{.+}} to i8*), i64 4) to i32*), i32 [[OLD_BF_VALUE]], i32 [[NEW_BF_VALUE]] monotonic monotonic
299 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i32, i1 } [[RES]], 0
300 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i32, i1 } [[RES]], 1
301 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
302 // CHECK: [[EXIT]]
303 #pragma omp atomic write
304   bfx.a = ldv;
305 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
306 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32
307 // CHECK: [[BITCAST:%.+]] = bitcast i32* [[LDTEMP:%.+]] to i8*
308 // CHECK: call void @__atomic_load(i64 4, i8* getelementptr (i8, i8* bitcast (%struct.BitFields_packed* @{{.+}} to i8*), i64 4), i8* [[BITCAST]], i32 0)
309 // CHECK: [[PREV_VALUE:%.+]] = load i32, i32* [[LDTEMP]]
310 // CHECK: br label %[[CONT:.+]]
311 // CHECK: [[CONT]]
312 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i32 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
313 // CHECK: [[BF_VALUE:%.+]] = and i32 [[NEW_VAL]], 2147483647
314 // CHECK: [[BF_CLEAR:%.+]] = and i32 %{{.+}}, -2147483648
315 // CHECK: or i32 [[BF_CLEAR]], [[BF_VALUE]]
316 // CHECK: store i32 %{{.+}}, i32* [[LDTEMP:%.+]]
317 // CHECK: [[NEW_BF_VALUE:%.+]] = load i32, i32* [[LDTEMP]]
318 // CHECK: store i32 [[OLD_BF_VALUE]], i32* [[TEMP_OLD_BF_ADDR:%.+]],
319 // CHECK: store i32 [[NEW_BF_VALUE]], i32* [[TEMP_NEW_BF_ADDR:%.+]],
320 // CHECK: [[BITCAST_TEMP_OLD_BF_ADDR:%.+]] = bitcast i32* [[TEMP_OLD_BF_ADDR]] to i8*
321 // CHECK: [[BITCAST_TEMP_NEW_BF_ADDR:%.+]] = bitcast i32* [[TEMP_NEW_BF_ADDR]] to i8*
322 // CHECK: [[FAIL_SUCCESS:%.+]] = call zeroext i1 @__atomic_compare_exchange(i64 4, i8* getelementptr (i8, i8* bitcast (%struct.BitFields_packed* @{{.+}} to i8*), i64 4), i8* [[BITCAST_TEMP_OLD_BF_ADDR]], i8* [[BITCAST_TEMP_NEW_BF_ADDR]], i32 0, i32 0)
323 // CHECK: [[FAILED_OLD_VAL]] = load i32, i32* [[TEMP_OLD_BF_ADDR]]
324 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
325 // CHECK: [[EXIT]]
326 #pragma omp atomic write
327   bfx_packed.a = ldv;
328 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
329 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32
330 // CHECK: [[PREV_VALUE:%.+]] = load atomic i32, i32* getelementptr inbounds (%struct.BitFields2, %struct.BitFields2* @{{.+}}, i32 0, i32 0) monotonic
331 // CHECK: br label %[[CONT:.+]]
332 // CHECK: [[CONT]]
333 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i32 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
334 // CHECK: [[BF_AND:%.+]] = and i32 [[NEW_VAL]], 1
335 // CHECK: [[BF_VALUE:%.+]] = shl i32 [[BF_AND]], 31
336 // CHECK: [[BF_CLEAR:%.+]] = and i32 %{{.+}}, 2147483647
337 // CHECK: or i32 [[BF_CLEAR]], [[BF_VALUE]]
338 // CHECK: store i32 %{{.+}}, i32* [[LDTEMP:%.+]]
339 // CHECK: [[NEW_BF_VALUE:%.+]] = load i32, i32* [[LDTEMP]]
340 // CHECK: [[RES:%.+]] = cmpxchg i32* getelementptr inbounds (%struct.BitFields2, %struct.BitFields2* @{{.+}}, i32 0, i32 0), i32 [[OLD_BF_VALUE]], i32 [[NEW_BF_VALUE]] monotonic monotonic
341 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i32, i1 } [[RES]], 0
342 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i32, i1 } [[RES]], 1
343 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
344 // CHECK: [[EXIT]]
345 #pragma omp atomic write
346   bfx2.a = ldv;
347 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
348 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32
349 // CHECK: [[PREV_VALUE:%.+]] = load atomic i8, i8* getelementptr (i8, i8* bitcast (%struct.BitFields2_packed* @{{.+}} to i8*), i64 3) monotonic
350 // CHECK: br label %[[CONT:.+]]
351 // CHECK: [[CONT]]
352 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i8 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
353 // CHECK: [[TRUNC:%.+]] = trunc i32 [[NEW_VAL]] to i8
354 // CHECK: [[BF_AND:%.+]] = and i8 [[TRUNC]], 1
355 // CHECK: [[BF_VALUE:%.+]] = shl i8 [[BF_AND]], 7
356 // CHECK: [[BF_CLEAR:%.+]] = and i8 %{{.+}}, 127
357 // CHECK: or i8 [[BF_CLEAR]], [[BF_VALUE]]
358 // CHECK: store i8 %{{.+}}, i8* [[LDTEMP:%.+]]
359 // CHECK: [[NEW_BF_VALUE:%.+]] = load i8, i8* [[LDTEMP]]
360 // CHECK: [[RES:%.+]] = cmpxchg i8* getelementptr (i8, i8* bitcast (%struct.BitFields2_packed* @{{.+}} to i8*), i64 3), i8 [[OLD_BF_VALUE]], i8 [[NEW_BF_VALUE]] monotonic monotonic
361 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i8, i1 } [[RES]], 0
362 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i8, i1 } [[RES]], 1
363 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
364 // CHECK: [[EXIT]]
365 #pragma omp atomic write
366   bfx2_packed.a = ldv;
367 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
368 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32
369 // CHECK: [[PREV_VALUE:%.+]] = load atomic i32, i32* getelementptr inbounds (%struct.BitFields3, %struct.BitFields3* @{{.+}}, i32 0, i32 0) monotonic
370 // CHECK: br label %[[CONT:.+]]
371 // CHECK: [[CONT]]
372 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i32 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
373 // CHECK: [[BF_AND:%.+]] = and i32 [[NEW_VAL]], 16383
374 // CHECK: [[BF_VALUE:%.+]] = shl i32 [[BF_AND]], 11
375 // CHECK: [[BF_CLEAR:%.+]] = and i32 %{{.+}}, -33552385
376 // CHECK: or i32 [[BF_CLEAR]], [[BF_VALUE]]
377 // CHECK: store i32 %{{.+}}, i32* [[LDTEMP:%.+]]
378 // CHECK: [[NEW_BF_VALUE:%.+]] = load i32, i32* [[LDTEMP]]
379 // CHECK: [[RES:%.+]] = cmpxchg i32* getelementptr inbounds (%struct.BitFields3, %struct.BitFields3* @{{.+}}, i32 0, i32 0), i32 [[OLD_BF_VALUE]], i32 [[NEW_BF_VALUE]] monotonic monotonic
380 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i32, i1 } [[RES]], 0
381 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i32, i1 } [[RES]], 1
382 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
383 // CHECK: [[EXIT]]
384 #pragma omp atomic write
385   bfx3.a = ldv;
386 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
387 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32
388 // CHECK: [[LDTEMP:%.+]] = bitcast i32* %{{.+}} to i24*
389 // CHECK: [[BITCAST:%.+]] = bitcast i24* %{{.+}} to i8*
390 // CHECK: call void @__atomic_load(i64 3, i8* getelementptr (i8, i8* bitcast (%struct.BitFields3_packed* @{{.+}} to i8*), i64 1), i8* [[BITCAST]], i32 0)
391 // CHECK: [[PREV_VALUE:%.+]] = load i24, i24* [[LDTEMP]]
392 // CHECK: br label %[[CONT:.+]]
393 // CHECK: [[CONT]]
394 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i24 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
395 // CHECK: [[TRUNC:%.+]] = trunc i32 [[NEW_VAL]] to i24
396 // CHECK: [[BF_AND:%.+]] = and i24 [[TRUNC]], 16383
397 // CHECK: [[BF_VALUE:%.+]] = shl i24 [[BF_AND]], 3
398 // CHECK: [[BF_CLEAR:%.+]] = and i24 %{{.+}}, -131065
399 // CHECK: or i24 [[BF_CLEAR]], [[BF_VALUE]]
400 // CHECK: store i24 %{{.+}}, i24* [[LDTEMP:%.+]]
401 // CHECK: [[NEW_BF_VALUE:%.+]] = load i24, i24* [[LDTEMP]]
402 // CHECK: [[TEMP_OLD_BF_ADDR:%.+]] = bitcast i32* %{{.+}} to i24*
403 // CHECK: store i24 [[OLD_BF_VALUE]], i24* [[TEMP_OLD_BF_ADDR]]
404 // CHECK: [[TEMP_NEW_BF_ADDR:%.+]] = bitcast i32* %{{.+}} to i24*
405 // CHECK: store i24 [[NEW_BF_VALUE]], i24* [[TEMP_NEW_BF_ADDR]]
406 // CHECK: [[BITCAST_TEMP_OLD_BF_ADDR:%.+]] = bitcast i24* [[TEMP_OLD_BF_ADDR]] to i8*
407 // CHECK: [[BITCAST_TEMP_NEW_BF_ADDR:%.+]] = bitcast i24* [[TEMP_NEW_BF_ADDR]] to i8*
408 // CHECK: [[FAIL_SUCCESS:%.+]] = call zeroext i1 @__atomic_compare_exchange(i64 3, i8* getelementptr (i8, i8* bitcast (%struct.BitFields3_packed* @{{.+}} to i8*), i64 1), i8* [[BITCAST_TEMP_OLD_BF_ADDR]], i8* [[BITCAST_TEMP_NEW_BF_ADDR]], i32 0, i32 0)
409 // CHECK: [[FAILED_OLD_VAL]] = load i24, i24* [[TEMP_OLD_BF_ADDR]]
410 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
411 // CHECK: [[EXIT]]
412 #pragma omp atomic write
413   bfx3_packed.a = ldv;
414 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
415 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32
416 // CHECK: [[PREV_VALUE:%.+]] = load atomic i64, i64* bitcast (%struct.BitFields4* @{{.+}} to i64*) monotonic
417 // CHECK: br label %[[CONT:.+]]
418 // CHECK: [[CONT]]
419 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i64 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
420 // CHECK: [[ZEXT:%.+]] = zext i32 [[NEW_VAL]] to i64
421 // CHECK: [[BF_AND:%.+]] = and i64 [[ZEXT]], 1
422 // CHECK: [[BF_VALUE:%.+]] = shl i64 [[BF_AND]], 16
423 // CHECK: [[BF_CLEAR:%.+]] = and i64 %{{.+}}, -65537
424 // CHECK: or i64 [[BF_CLEAR]], [[BF_VALUE]]
425 // CHECK: store i64 %{{.+}}, i64* [[LDTEMP:%.+]]
426 // CHECK: [[NEW_BF_VALUE:%.+]] = load i64, i64* [[LDTEMP]]
427 // CHECK: [[RES:%.+]] = cmpxchg i64* bitcast (%struct.BitFields4* @{{.+}} to i64*), i64 [[OLD_BF_VALUE]], i64 [[NEW_BF_VALUE]] monotonic monotonic
428 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i64, i1 } [[RES]], 0
429 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i64, i1 } [[RES]], 1
430 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
431 // CHECK: [[EXIT]]
432 #pragma omp atomic write
433   bfx4.a = ldv;
434 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
435 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32
436 // CHECK: [[PREV_VALUE:%.+]] = load atomic i8, i8* getelementptr inbounds (%struct.BitFields4_packed, %struct.BitFields4_packed* @{{.+}}, i32 0, i32 0, i64 2) monotonic
437 // CHECK: br label %[[CONT:.+]]
438 // CHECK: [[CONT]]
439 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i8 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
440 // CHECK: [[TRUNC:%.+]] = trunc i32 [[NEW_VAL]] to i8
441 // CHECK: [[BF_VALUE:%.+]] = and i8 [[TRUNC]], 1
442 // CHECK: [[BF_CLEAR:%.+]] = and i8 %{{.+}}, -2
443 // CHECK: or i8 [[BF_CLEAR]], [[BF_VALUE]]
444 // CHECK: store i8 %{{.+}}, i8* [[LDTEMP:%.+]]
445 // CHECK: [[NEW_BF_VALUE:%.+]] = load i8, i8* [[LDTEMP]]
446 // CHECK: [[RES:%.+]] = cmpxchg i8* getelementptr inbounds (%struct.BitFields4_packed, %struct.BitFields4_packed* @{{.+}}, i32 0, i32 0, i64 2), i8 [[OLD_BF_VALUE]], i8 [[NEW_BF_VALUE]] monotonic monotonic
447 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i8, i1 } [[RES]], 0
448 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i8, i1 } [[RES]], 1
449 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
450 // CHECK: [[EXIT]]
451 #pragma omp atomic write
452   bfx4_packed.a = ldv;
453 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
454 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i64
455 // CHECK: [[PREV_VALUE:%.+]] = load atomic i64, i64* bitcast (%struct.BitFields4* @{{.+}} to i64*) monotonic
456 // CHECK: br label %[[CONT:.+]]
457 // CHECK: [[CONT]]
458 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i64 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
459 // CHECK: [[BF_AND:%.+]] = and i64 [[NEW_VAL]], 127
460 // CHECK: [[BF_VALUE:%.+]] = shl i64 [[BF_AND]], 17
461 // CHECK: [[BF_CLEAR:%.+]] = and i64 %{{.+}}, -16646145
462 // CHECK: or i64 [[BF_CLEAR]], [[BF_VALUE]]
463 // CHECK: store i64 %{{.+}}, i64* [[LDTEMP:%.+]]
464 // CHECK: [[NEW_BF_VALUE:%.+]] = load i64, i64* [[LDTEMP]]
465 // CHECK: [[RES:%.+]] = cmpxchg i64* bitcast (%struct.BitFields4* @{{.+}} to i64*), i64 [[OLD_BF_VALUE]], i64 [[NEW_BF_VALUE]] monotonic monotonic
466 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i64, i1 } [[RES]], 0
467 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i64, i1 } [[RES]], 1
468 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
469 // CHECK: [[EXIT]]
470 #pragma omp atomic write
471   bfx4.b = ldv;
472 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
473 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i64
474 // CHECK: [[PREV_VALUE:%.+]] = load atomic i8, i8* getelementptr inbounds (%struct.BitFields4_packed, %struct.BitFields4_packed* @{{.+}}, i32 0, i32 0, i64 2) monotonic
475 // CHECK: br label %[[CONT:.+]]
476 // CHECK: [[CONT]]
477 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i8 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
478 // CHECK: [[TRUNC:%.+]] = trunc i64 [[NEW_VAL]] to i8
479 // CHECK: [[BF_AND:%.+]] = and i8 [[TRUNC]], 127
480 // CHECK: [[BF_VALUE:%.+]] = shl i8 [[BF_AND]], 1
481 // CHECK: [[BF_CLEAR:%.+]] = and i8 %{{.+}}, 1
482 // CHECK: or i8 [[BF_CLEAR]], [[BF_VALUE]]
483 // CHECK: store i8 %{{.+}}, i8* [[LDTEMP:%.+]]
484 // CHECK: [[NEW_BF_VALUE:%.+]] = load i8, i8* [[LDTEMP]]
485 // CHECK: [[RES:%.+]] = cmpxchg i8* getelementptr inbounds (%struct.BitFields4_packed, %struct.BitFields4_packed* @{{.+}}, i32 0, i32 0, i64 2), i8 [[OLD_BF_VALUE]], i8 [[NEW_BF_VALUE]] monotonic monotonic
486 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i8, i1 } [[RES]], 0
487 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i8, i1 } [[RES]], 1
488 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
489 // CHECK: [[EXIT]]
490 #pragma omp atomic write
491   bfx4_packed.b = ldv;
492 // CHECK: load i64, i64*
493 // CHECK: [[VEC_ITEM_VAL:%.+]] = uitofp i64 %{{.+}} to float
494 // CHECK: [[I64VAL:%.+]] = load atomic i64, i64* bitcast (<2 x float>* [[DEST:@.+]] to i64*) monotonic
495 // CHECK: [[LD:%.+]] = bitcast i64 [[I64VAL]] to <2 x float>
496 // CHECK: br label %[[CONT:.+]]
497 // CHECK: [[CONT]]
498 // CHECK: [[OLD_VEC_VAL:%.+]] = phi <2 x float> [ [[LD]], %{{.+}} ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
499 // CHECK: store <2 x float> [[OLD_VEC_VAL]], <2 x float>* [[LDTEMP:%.+]],
500 // CHECK: [[VEC_VAL:%.+]] = load <2 x float>, <2 x float>* [[LDTEMP]]
501 // CHECK: [[NEW_VEC_VAL:%.+]] = insertelement <2 x float> [[VEC_VAL]], float [[VEC_ITEM_VAL]], i64 0
502 // CHECK: store <2 x float> [[NEW_VEC_VAL]], <2 x float>* [[LDTEMP]]
503 // CHECK: [[NEW_VEC_VAL:%.+]] = load <2 x float>, <2 x float>* [[LDTEMP]]
504 // CHECK: [[OLD_I64:%.+]] = bitcast <2 x float> [[OLD_VEC_VAL]] to i64
505 // CHECK: [[NEW_I64:%.+]] = bitcast <2 x float> [[NEW_VEC_VAL]] to i64
506 // CHECK: [[RES:%.+]] = cmpxchg i64* bitcast (<2 x float>* [[DEST]] to i64*), i64 [[OLD_I64]], i64 [[NEW_I64]] monotonic monotonic
507 // CHECK: [[FAILED_I64_OLD_VAL:%.+]] = extractvalue { i64, i1 } [[RES]], 0
508 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i64, i1 } [[RES]], 1
509 // CHECK: [[FAILED_OLD_VAL]] = bitcast i64 [[FAILED_I64_OLD_VAL]] to <2 x float>
510 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
511 // CHECK: [[EXIT]]
512 #pragma omp atomic write
513   float2x.x = ulv;
514 // CHECK: call i32 @llvm.read_register.i32(
515 // CHECK: sitofp i32 %{{.+}} to double
516 // CHECK: bitcast double %{{.+}} to i64
517 // CHECK: store atomic i64 %{{.+}}, i64* bitcast (double* @{{.+}} to i64*) seq_cst
518 // CHECK: call{{.*}} @__kmpc_flush(
519 #pragma omp atomic write seq_cst
520   dv = rix;
521   return 0;
522 }
523 
524 #endif
525