1 // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -x c -emit-llvm %s -o - | FileCheck %s
2 // RUN: %clang_cc1 -fopenmp -x c -triple x86_64-apple-darwin10 -emit-pch -o %t %s
3 // RUN: %clang_cc1 -fopenmp -x c -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s
4 // expected-no-diagnostics
5 // REQUIRES: x86-registered-target
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 "0" is currently an invalid register for global register variables.
76 // Use "esp" instead of "0".
77 // register int rix __asm__("0");
78 register int rix __asm__("esp");
79 
80 int main() {
81 // CHECK: store atomic i32 1, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @civ, i32 0, i32 1) monotonic,
82 #pragma omp atomic write
83  __imag(civ) = 1;
84 // CHECK: load i8, i8*
85 // CHECK: store atomic i8
86 #pragma omp atomic write
87   bx = bv;
88 // CHECK: load i8, i8*
89 // CHECK: store atomic i8
90 #pragma omp atomic write
91   cx = cv;
92 // CHECK: load i8, i8*
93 // CHECK: store atomic i8
94 #pragma omp atomic write
95   ucx = ucv;
96 // CHECK: load i16, i16*
97 // CHECK: store atomic i16
98 #pragma omp atomic write
99   sx = sv;
100 // CHECK: load i16, i16*
101 // CHECK: store atomic i16
102 #pragma omp atomic write
103   usx = usv;
104 // CHECK: load i32, i32*
105 // CHECK: store atomic i32
106 #pragma omp atomic write
107   ix = iv;
108 // CHECK: load i32, i32*
109 // CHECK: store atomic i32
110 #pragma omp atomic write
111   uix = uiv;
112 // CHECK: load i64, i64*
113 // CHECK: store atomic i64
114 #pragma omp atomic write
115   lx = lv;
116 // CHECK: load i64, i64*
117 // CHECK: store atomic i64
118 #pragma omp atomic write
119   ulx = ulv;
120 // CHECK: load i64, i64*
121 // CHECK: store atomic i64
122 #pragma omp atomic write
123   llx = llv;
124 // CHECK: load i64, i64*
125 // CHECK: store atomic i64
126 #pragma omp atomic write
127   ullx = ullv;
128 // CHECK: load float, float*
129 // CHECK: bitcast float {{.*}} to i32
130 // CHECK: store atomic i32 {{.*}}, i32* bitcast (float*
131 #pragma omp atomic write
132   fx = fv;
133 // CHECK: load double, double*
134 // CHECK: bitcast double {{.*}} to i64
135 // CHECK: store atomic i64 {{.*}}, i64* bitcast (double*
136 #pragma omp atomic write
137   dx = dv;
138 // CHECK: [[LD:%.+]] = load x86_fp80, x86_fp80*
139 // CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[LDTEMP:%.*]] to i8*
140 // CHECK: call void @llvm.memset.p0i8.i64(i8* [[BITCAST]], i8 0, i64 16, i32 16, i1 false)
141 // CHECK: store x86_fp80 [[LD]], x86_fp80* [[LDTEMP]]
142 // CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[LDTEMP:%.*]] to i128*
143 // CHECK: [[LD:%.+]] = load i128, i128* [[BITCAST]]
144 // CHECK: store atomic i128 [[LD]], i128* bitcast (x86_fp80*
145 #pragma omp atomic write
146   ldx = ldv;
147 // CHECK: [[REAL_VAL:%.+]] = load i32, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.*}}, i32 0, i32 0)
148 // CHECK: [[IMG_VAL:%.+]] = load i32, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.*}}, i32 0, i32 1)
149 // CHECK: [[TEMP_REAL_REF:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP:%.+]], i32 0, i32 0
150 // CHECK: [[TEMP_IMG_REF:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1
151 // CHECK: store i32 [[REAL_VAL]], i32* [[TEMP_REAL_REF]]
152 // CHECK: store i32 [[IMG_VAL]], i32* [[TEMP_IMG_REF]]
153 // CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[TEMP]] to i8*
154 // CHECK: call void @__atomic_store(i64 8, i8* bitcast ({ i32, i32 }* @{{.*}} to i8*), i8* [[BITCAST]], i32 0)
155 #pragma omp atomic write
156   cix = civ;
157 // CHECK: [[REAL_VAL:%.+]] = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @{{.*}}, i32 0, i32 0)
158 // CHECK: [[IMG_VAL:%.+]] = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @{{.*}}, i32 0, i32 1)
159 // CHECK: [[TEMP_REAL_REF:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[TEMP:%.+]], i32 0, i32 0
160 // CHECK: [[TEMP_IMG_REF:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[TEMP]], i32 0, i32 1
161 // CHECK: store float [[REAL_VAL]], float* [[TEMP_REAL_REF]]
162 // CHECK: store float [[IMG_VAL]], float* [[TEMP_IMG_REF]]
163 // CHECK: [[BITCAST:%.+]] = bitcast { float, float }* [[TEMP]] to i8*
164 // CHECK: call void @__atomic_store(i64 8, i8* bitcast ({ float, float }* @{{.*}} to i8*), i8* [[BITCAST]], i32 0)
165 #pragma omp atomic write
166   cfx = cfv;
167 // CHECK: [[REAL_VAL:%.+]] = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @{{.*}}, i32 0, i32 0)
168 // CHECK: [[IMG_VAL:%.+]] = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @{{.*}}, i32 0, i32 1)
169 // CHECK: [[TEMP_REAL_REF:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[TEMP:%.+]], i32 0, i32 0
170 // CHECK: [[TEMP_IMG_REF:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[TEMP]], i32 0, i32 1
171 // CHECK: store double [[REAL_VAL]], double* [[TEMP_REAL_REF]]
172 // CHECK: store double [[IMG_VAL]], double* [[TEMP_IMG_REF]]
173 // CHECK: [[BITCAST:%.+]] = bitcast { double, double }* [[TEMP]] to i8*
174 // CHECK: call void @__atomic_store(i64 16, i8* bitcast ({ double, double }* @{{.*}} to i8*), i8* [[BITCAST]], i32 5)
175 // CHECK: call{{.*}} @__kmpc_flush(
176 #pragma omp atomic seq_cst write
177   cdx = cdv;
178 // CHECK: load i8, i8*
179 // CHECK: store atomic i64
180 #pragma omp atomic write
181   ulx = bv;
182 // CHECK: load i8, i8*
183 // CHECK: store atomic i8
184 #pragma omp atomic write
185   bx = cv;
186 // CHECK: load i8, i8*
187 // CHECK: store atomic i8
188 // CHECK: call{{.*}} @__kmpc_flush(
189 #pragma omp atomic write, seq_cst
190   cx = ucv;
191 // CHECK: load i16, i16*
192 // CHECK: store atomic i64
193 #pragma omp atomic write
194   ulx = sv;
195 // CHECK: load i16, i16*
196 // CHECK: store atomic i64
197 #pragma omp atomic write
198   lx = usv;
199 // CHECK: load i32, i32*
200 // CHECK: store atomic i32
201 // CHECK: call{{.*}} @__kmpc_flush(
202 #pragma omp atomic seq_cst, write
203   uix = iv;
204 // CHECK: load i32, i32*
205 // CHECK: store atomic i32
206 #pragma omp atomic write
207   ix = uiv;
208 // CHECK: load i64, i64*
209 // CHECK: [[VAL:%.+]] = trunc i64 %{{.*}} to i32
210 // CHECK: [[TEMP_REAL_REF:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP:%.+]], i32 0, i32 0
211 // CHECK: [[TEMP_IMG_REF:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1
212 // CHECK: store i32 [[VAL]], i32* [[TEMP_REAL_REF]]
213 // CHECK: store i32 0, i32* [[TEMP_IMG_REF]]
214 // CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[TEMP]] to i8*
215 // CHECK: call void @__atomic_store(i64 8, i8* bitcast ({ i32, i32 }* @{{.+}} to i8*), i8* [[BITCAST]], i32 0)
216 #pragma omp atomic write
217   cix = lv;
218 // CHECK: load i64, i64*
219 // CHECK: store atomic i32 %{{.+}}, i32* bitcast (float*
220 #pragma omp atomic write
221   fx = ulv;
222 // CHECK: load i64, i64*
223 // CHECK: store atomic i64 %{{.+}}, i64* bitcast (double*
224 #pragma omp atomic write
225   dx = llv;
226 // CHECK: load i64, i64*
227 // CHECK: [[VAL:%.+]] = uitofp i64 %{{.+}} to x86_fp80
228 // CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP:%.+]] to i8*
229 // CHECK: call void @llvm.memset.p0i8.i64(i8* [[BITCAST]], i8 0, i64 16, i32 16, i1 false)
230 // CHECK: store x86_fp80 [[VAL]], x86_fp80* [[TEMP]]
231 // CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP]] to i128*
232 // CHECK: [[VAL:%.+]] = load i128, i128* [[BITCAST]]
233 // CHECK: store atomic i128 [[VAL]], i128* bitcast (x86_fp80*
234 #pragma omp atomic write
235   ldx = ullv;
236 // CHECK: load float, float*
237 // CHECK: [[VAL:%.+]] = fptosi float %{{.*}} to i32
238 // CHECK: [[TEMP_REAL_REF:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP:%.+]], i32 0, i32 0
239 // CHECK: [[TEMP_IMG_REF:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1
240 // CHECK: store i32 [[VAL]], i32* [[TEMP_REAL_REF]]
241 // CHECK: store i32 0, i32* [[TEMP_IMG_REF]]
242 // CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[TEMP]] to i8*
243 // CHECK: call void @__atomic_store(i64 8, i8* bitcast ({ i32, i32 }* @{{.+}} to i8*), i8* [[BITCAST]], i32 0)
244 #pragma omp atomic write
245   cix = fv;
246 // CHECK: load double, double*
247 // CHECK: store atomic i16
248 #pragma omp atomic write
249   sx = dv;
250 // CHECK: load x86_fp80, x86_fp80*
251 // CHECK: store atomic i8
252 #pragma omp atomic write
253   bx = ldv;
254 // CHECK: load i32, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.+}}, i32 0, i32 0)
255 // CHECK: load i32, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.+}}, i32 0, i32 1)
256 // CHECK: icmp ne i32 %{{.+}}, 0
257 // CHECK: icmp ne i32 %{{.+}}, 0
258 // CHECK: or i1
259 // CHECK: store atomic i8
260 #pragma omp atomic write
261   bx = civ;
262 // CHECK: load float, float* getelementptr inbounds ({ float, float }, { float, float }* @{{.*}}, i32 0, i32 0)
263 // CHECK: store atomic i16
264 #pragma omp atomic write
265   usx = cfv;
266 // CHECK: load double, double* getelementptr inbounds ({ double, double }, { double, double }* @{{.+}}, i32 0, i32 0)
267 // CHECK: store atomic i64
268 #pragma omp atomic write
269   llx = cdv;
270 // CHECK-DAG: [[IDX:%.+]] = load i16, i16* @{{.+}}
271 // CHECK-DAG: load i8, i8*
272 // CHECK-DAG: [[VEC_ITEM_VAL:%.+]] = zext i1 %{{.+}} to i32
273 // CHECK: [[I128VAL:%.+]] = load atomic i128, i128* bitcast (<4 x i32>* [[DEST:@.+]] to i128*) monotonic
274 // CHECK: br label %[[CONT:.+]]
275 // CHECK: [[CONT]]
276 // CHECK: [[OLD_I128:%.+]] = phi i128 [ [[I128VAL]], %{{.+}} ], [ [[FAILED_I128_OLD_VAL:%.+]], %[[CONT]] ]
277 // CHECK: [[BITCAST:%.+]] = bitcast <4 x i32>* [[LDTEMP:%.+]] to i128*
278 // CHECK: store i128 [[OLD_I128]], i128* [[BITCAST]],
279 // CHECK: [[VEC_VAL:%.+]] = load <4 x i32>, <4 x i32>* [[LDTEMP]]
280 // CHECK: [[NEW_VEC_VAL:%.+]] = insertelement <4 x i32> [[VEC_VAL]], i32 [[VEC_ITEM_VAL]], i16 [[IDX]]
281 // CHECK: store <4 x i32> [[NEW_VEC_VAL]], <4 x i32>* [[LDTEMP]]
282 // CHECK: [[NEW_I128:%.+]] = load i128, i128* [[BITCAST]]
283 // CHECK: [[RES:%.+]] = cmpxchg i128* bitcast (<4 x i32>* [[DEST]] to i128*), i128 [[OLD_I128]], i128 [[NEW_I128]] monotonic monotonic
284 // CHECK: [[FAILED_I128_OLD_VAL:%.+]] = extractvalue { i128, i1 } [[RES]], 0
285 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i128, i1 } [[RES]], 1
286 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
287 // CHECK: [[EXIT]]
288 #pragma omp atomic write
289   int4x[sv] = bv;
290 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
291 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32
292 // CHECK: [[PREV_VALUE:%.+]] = load atomic i32, i32* bitcast (i8* getelementptr (i8, i8* bitcast (%struct.BitFields* @{{.+}} to i8*), i64 4) to i32*) monotonic
293 // CHECK: br label %[[CONT:.+]]
294 // CHECK: [[CONT]]
295 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i32 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
296 // CHECK: [[BF_VALUE:%.+]] = and i32 [[NEW_VAL]], 2147483647
297 // CHECK: [[BF_CLEAR:%.+]] = and i32 %{{.+}}, -2147483648
298 // CHECK: or i32 [[BF_CLEAR]], [[BF_VALUE]]
299 // CHECK: store i32 %{{.+}}, i32* [[LDTEMP:%.+]]
300 // CHECK: [[NEW_BF_VALUE:%.+]] = load i32, i32* [[LDTEMP]]
301 // 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
302 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i32, i1 } [[RES]], 0
303 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i32, i1 } [[RES]], 1
304 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
305 // CHECK: [[EXIT]]
306 #pragma omp atomic write
307   bfx.a = ldv;
308 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
309 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32
310 // CHECK: [[BITCAST:%.+]] = bitcast i32* [[LDTEMP:%.+]] to i8*
311 // CHECK: call void @__atomic_load(i64 4, i8* getelementptr (i8, i8* bitcast (%struct.BitFields_packed* @{{.+}} to i8*), i64 4), i8* [[BITCAST]], i32 0)
312 // CHECK: br label %[[CONT:.+]]
313 // CHECK: [[CONT]]
314 // CHECK: [[OLD_BF_VALUE:%.+]] = load i32, i32* [[LDTEMP]],
315 // CHECK: store i32 [[OLD_BF_VALUE]], i32* [[LDTEMP1:%.+]],
316 // CHECK: [[OLD_BF_VALUE:%.+]] = load i32, i32* [[LDTEMP1]],
317 // CHECK: [[BF_VALUE:%.+]] = and i32 [[NEW_VAL]], 2147483647
318 // CHECK: [[BF_CLEAR:%.+]] = and i32 [[OLD_BF_VALUE]], -2147483648
319 // CHECK: or i32 [[BF_CLEAR]], [[BF_VALUE]]
320 // CHECK: store i32 %{{.+}}, i32* [[LDTEMP1]]
321 // CHECK: [[BITCAST_TEMP_OLD_BF_ADDR:%.+]] = bitcast i32* [[LDTEMP]] to i8*
322 // CHECK: [[BITCAST_TEMP_NEW_BF_ADDR:%.+]] = bitcast i32* [[LDTEMP1]] to i8*
323 // 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)
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: br label %[[CONT:.+]]
392 // CHECK: [[CONT]]
393 // CHECK: [[OLD_VAL:%.+]] = load i24, i24* %{{.+}},
394 // CHECK: store i24 [[OLD_VAL]], i24* [[TEMP:%.+]],
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* [[TEMP]]
401 // CHECK: [[BITCAST_TEMP_OLD_BF_ADDR:%.+]] = bitcast i24* [[LDTEMP]] to i8*
402 // CHECK: [[BITCAST_TEMP_NEW_BF_ADDR:%.+]] = bitcast i24* [[TEMP]] to i8*
403 // 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)
404 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
405 // CHECK: [[EXIT]]
406 #pragma omp atomic write
407   bfx3_packed.a = ldv;
408 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
409 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32
410 // CHECK: [[PREV_VALUE:%.+]] = load atomic i64, i64* bitcast (%struct.BitFields4* @{{.+}} to i64*) monotonic
411 // CHECK: br label %[[CONT:.+]]
412 // CHECK: [[CONT]]
413 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i64 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
414 // CHECK: [[ZEXT:%.+]] = zext i32 [[NEW_VAL]] to i64
415 // CHECK: [[BF_AND:%.+]] = and i64 [[ZEXT]], 1
416 // CHECK: [[BF_VALUE:%.+]] = shl i64 [[BF_AND]], 16
417 // CHECK: [[BF_CLEAR:%.+]] = and i64 %{{.+}}, -65537
418 // CHECK: or i64 [[BF_CLEAR]], [[BF_VALUE]]
419 // CHECK: store i64 %{{.+}}, i64* [[LDTEMP:%.+]]
420 // CHECK: [[NEW_BF_VALUE:%.+]] = load i64, i64* [[LDTEMP]]
421 // CHECK: [[RES:%.+]] = cmpxchg i64* bitcast (%struct.BitFields4* @{{.+}} to i64*), i64 [[OLD_BF_VALUE]], i64 [[NEW_BF_VALUE]] monotonic monotonic
422 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i64, i1 } [[RES]], 0
423 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i64, i1 } [[RES]], 1
424 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
425 // CHECK: [[EXIT]]
426 #pragma omp atomic write
427   bfx4.a = ldv;
428 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
429 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32
430 // CHECK: [[PREV_VALUE:%.+]] = load atomic i8, i8* getelementptr inbounds (%struct.BitFields4_packed, %struct.BitFields4_packed* @{{.+}}, i32 0, i32 0, i64 2) monotonic
431 // CHECK: br label %[[CONT:.+]]
432 // CHECK: [[CONT]]
433 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i8 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
434 // CHECK: [[TRUNC:%.+]] = trunc i32 [[NEW_VAL]] to i8
435 // CHECK: [[BF_VALUE:%.+]] = and i8 [[TRUNC]], 1
436 // CHECK: [[BF_CLEAR:%.+]] = and i8 %{{.+}}, -2
437 // CHECK: or i8 [[BF_CLEAR]], [[BF_VALUE]]
438 // CHECK: store i8 %{{.+}}, i8* [[LDTEMP:%.+]]
439 // CHECK: [[NEW_BF_VALUE:%.+]] = load i8, i8* [[LDTEMP]]
440 // 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
441 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i8, i1 } [[RES]], 0
442 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i8, i1 } [[RES]], 1
443 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
444 // CHECK: [[EXIT]]
445 #pragma omp atomic write
446   bfx4_packed.a = ldv;
447 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
448 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i64
449 // CHECK: [[PREV_VALUE:%.+]] = load atomic i64, i64* bitcast (%struct.BitFields4* @{{.+}} to i64*) monotonic
450 // CHECK: br label %[[CONT:.+]]
451 // CHECK: [[CONT]]
452 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i64 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
453 // CHECK: [[BF_AND:%.+]] = and i64 [[NEW_VAL]], 127
454 // CHECK: [[BF_VALUE:%.+]] = shl i64 [[BF_AND]], 17
455 // CHECK: [[BF_CLEAR:%.+]] = and i64 %{{.+}}, -16646145
456 // CHECK: or i64 [[BF_CLEAR]], [[BF_VALUE]]
457 // CHECK: store i64 %{{.+}}, i64* [[LDTEMP:%.+]]
458 // CHECK: [[NEW_BF_VALUE:%.+]] = load i64, i64* [[LDTEMP]]
459 // CHECK: [[RES:%.+]] = cmpxchg i64* bitcast (%struct.BitFields4* @{{.+}} to i64*), i64 [[OLD_BF_VALUE]], i64 [[NEW_BF_VALUE]] monotonic monotonic
460 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i64, i1 } [[RES]], 0
461 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i64, i1 } [[RES]], 1
462 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
463 // CHECK: [[EXIT]]
464 #pragma omp atomic write
465   bfx4.b = ldv;
466 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
467 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i64
468 // CHECK: [[PREV_VALUE:%.+]] = load atomic i8, i8* getelementptr inbounds (%struct.BitFields4_packed, %struct.BitFields4_packed* @{{.+}}, i32 0, i32 0, i64 2) monotonic
469 // CHECK: br label %[[CONT:.+]]
470 // CHECK: [[CONT]]
471 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i8 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
472 // CHECK: [[TRUNC:%.+]] = trunc i64 [[NEW_VAL]] to i8
473 // CHECK: [[BF_AND:%.+]] = and i8 [[TRUNC]], 127
474 // CHECK: [[BF_VALUE:%.+]] = shl i8 [[BF_AND]], 1
475 // CHECK: [[BF_CLEAR:%.+]] = and i8 %{{.+}}, 1
476 // CHECK: or i8 [[BF_CLEAR]], [[BF_VALUE]]
477 // CHECK: store i8 %{{.+}}, i8* [[LDTEMP:%.+]]
478 // CHECK: [[NEW_BF_VALUE:%.+]] = load i8, i8* [[LDTEMP]]
479 // 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
480 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i8, i1 } [[RES]], 0
481 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i8, i1 } [[RES]], 1
482 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
483 // CHECK: [[EXIT]]
484 #pragma omp atomic write
485   bfx4_packed.b = ldv;
486 // CHECK: load i64, i64*
487 // CHECK: [[VEC_ITEM_VAL:%.+]] = uitofp i64 %{{.+}} to float
488 // CHECK: [[I64VAL:%.+]] = load atomic i64, i64* bitcast (<2 x float>* [[DEST:@.+]] to i64*) monotonic
489 // CHECK: br label %[[CONT:.+]]
490 // CHECK: [[CONT]]
491 // CHECK: [[OLD_I64:%.+]] = phi i64 [ [[I64VAL]], %{{.+}} ], [ [[FAILED_I64_OLD_VAL:%.+]], %[[CONT]] ]
492 // CHECK: [[BITCAST:%.+]] = bitcast <2 x float>* [[LDTEMP:%.+]] to i64*
493 // CHECK: store i64 [[OLD_I64]], i64* [[BITCAST]],
494 // CHECK: [[VEC_VAL:%.+]] = load <2 x float>, <2 x float>* [[LDTEMP]]
495 // CHECK: [[NEW_VEC_VAL:%.+]] = insertelement <2 x float> [[VEC_VAL]], float [[VEC_ITEM_VAL]], i64 0
496 // CHECK: store <2 x float> [[NEW_VEC_VAL]], <2 x float>* [[LDTEMP]]
497 // CHECK: [[NEW_I64:%.+]] = load i64, i64* [[BITCAST]]
498 // CHECK: [[RES:%.+]] = cmpxchg i64* bitcast (<2 x float>* [[DEST]] to i64*), i64 [[OLD_I64]], i64 [[NEW_I64]] monotonic monotonic
499 // CHECK: [[FAILED_I64_OLD_VAL:%.+]] = extractvalue { i64, i1 } [[RES]], 0
500 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i64, i1 } [[RES]], 1
501 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
502 // CHECK: [[EXIT]]
503 #pragma omp atomic write
504   float2x.x = ulv;
505 // CHECK: call i32 @llvm.read_register.i32(
506 // CHECK: sitofp i32 %{{.+}} to double
507 // CHECK: bitcast double %{{.+}} to i64
508 // CHECK: store atomic i64 %{{.+}}, i64* bitcast (double* @{{.+}} to i64*) seq_cst
509 // CHECK: call{{.*}} @__kmpc_flush(
510 #pragma omp atomic write seq_cst
511   dv = rix;
512   return 0;
513 }
514 
515 #endif
516