1 | // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -target-cpu core2 -fopenmp -x c -emit-llvm %s -o - | FileCheck %s |
2 | // RUN: %clang_cc1 -fopenmp -x c -triple x86_64-apple-darwin10 -target-cpu core2 -emit-pch -o %t %s |
3 | // RUN: %clang_cc1 -fopenmp -x c -triple x86_64-apple-darwin10 -target-cpu core2 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s |
4 | |
5 | // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -target-cpu core2 -fopenmp-simd -x c -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s |
6 | // RUN: %clang_cc1 -fopenmp-simd -x c -triple x86_64-apple-darwin10 -target-cpu core2 -emit-pch -o %t %s |
7 | // RUN: %clang_cc1 -fopenmp-simd -x c -triple x86_64-apple-darwin10 -target-cpu core2 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s |
8 | // SIMD-ONLY0-NOT: {{__kmpc|__tgt}} |
9 | // expected-no-diagnostics |
10 | // REQUIRES: x86-registered-target |
11 | #ifndef HEADER |
12 | #define HEADER |
13 | |
14 | _Bool bv, bx; |
15 | char cv, cx; |
16 | unsigned char ucv, ucx; |
17 | short sv, sx; |
18 | unsigned short usv, usx; |
19 | int iv, ix; |
20 | unsigned int uiv, uix; |
21 | long lv, lx; |
22 | unsigned long ulv, ulx; |
23 | long long llv, llx; |
24 | unsigned long long ullv, ullx; |
25 | float fv, fx; |
26 | double dv, dx; |
27 | long double ldv, ldx; |
28 | _Complex int civ, cix; |
29 | _Complex float cfv, cfx; |
30 | _Complex double cdv, cdx; |
31 | |
32 | typedef int int4 __attribute__((__vector_size__(16))); |
33 | int4 int4x; |
34 | |
35 | struct BitFields { |
36 | int : 32; |
37 | int a : 31; |
38 | } bfx; |
39 | |
40 | struct BitFields_packed { |
41 | int : 32; |
42 | int a : 31; |
43 | } __attribute__ ((__packed__)) bfx_packed; |
44 | |
45 | struct BitFields2 { |
46 | int : 31; |
47 | int a : 1; |
48 | } bfx2; |
49 | |
50 | struct BitFields2_packed { |
51 | int : 31; |
52 | int a : 1; |
53 | } __attribute__ ((__packed__)) bfx2_packed; |
54 | |
55 | struct BitFields3 { |
56 | int : 11; |
57 | int a : 14; |
58 | } bfx3; |
59 | |
60 | struct BitFields3_packed { |
61 | int : 11; |
62 | int a : 14; |
63 | } __attribute__ ((__packed__)) bfx3_packed; |
64 | |
65 | struct BitFields4 { |
66 | short : 16; |
67 | int a: 1; |
68 | long b : 7; |
69 | } bfx4; |
70 | |
71 | struct BitFields4_packed { |
72 | short : 16; |
73 | int a: 1; |
74 | long b : 7; |
75 | } __attribute__ ((__packed__)) bfx4_packed; |
76 | |
77 | typedef float float2 __attribute__((ext_vector_type(2))); |
78 | float2 float2x; |
79 | |
80 | // Register "0" is currently an invalid register for global register variables. |
81 | // Use "esp" instead of "0". |
82 | // register int rix __asm__("0"); |
83 | register int rix __asm__("esp"); |
84 | |
85 | int main() { |
86 | // CHECK: store atomic i32 1, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @civ, i32 0, i32 1) monotonic, |
87 | #pragma omp atomic write |
88 | __imag(civ) = 1; |
89 | // CHECK: load i8, i8* |
90 | // CHECK: store atomic i8 |
91 | #pragma omp atomic write |
92 | bx = bv; |
93 | // CHECK: load i8, i8* |
94 | // CHECK: store atomic i8 |
95 | #pragma omp atomic write |
96 | cx = cv; |
97 | // CHECK: load i8, i8* |
98 | // CHECK: store atomic i8 |
99 | #pragma omp atomic write |
100 | ucx = ucv; |
101 | // CHECK: load i16, i16* |
102 | // CHECK: store atomic i16 |
103 | #pragma omp atomic write |
104 | sx = sv; |
105 | // CHECK: load i16, i16* |
106 | // CHECK: store atomic i16 |
107 | #pragma omp atomic write |
108 | usx = usv; |
109 | // CHECK: load i32, i32* |
110 | // CHECK: store atomic i32 |
111 | #pragma omp atomic write |
112 | ix = iv; |
113 | // CHECK: load i32, i32* |
114 | // CHECK: store atomic i32 |
115 | #pragma omp atomic write |
116 | uix = uiv; |
117 | // CHECK: load i64, i64* |
118 | // CHECK: store atomic i64 |
119 | #pragma omp atomic write |
120 | lx = lv; |
121 | // CHECK: load i64, i64* |
122 | // CHECK: store atomic i64 |
123 | #pragma omp atomic write |
124 | ulx = ulv; |
125 | // CHECK: load i64, i64* |
126 | // CHECK: store atomic i64 |
127 | #pragma omp atomic write |
128 | llx = llv; |
129 | // CHECK: load i64, i64* |
130 | // CHECK: store atomic i64 |
131 | #pragma omp atomic write |
132 | ullx = ullv; |
133 | // CHECK: load float, float* |
134 | // CHECK: bitcast float {{.*}} to i32 |
135 | // CHECK: store atomic i32 {{.*}}, i32* bitcast (float* |
136 | #pragma omp atomic write |
137 | fx = fv; |
138 | // CHECK: load double, double* |
139 | // CHECK: bitcast double {{.*}} to i64 |
140 | // CHECK: store atomic i64 {{.*}}, i64* bitcast (double* |
141 | #pragma omp atomic write |
142 | dx = dv; |
143 | // CHECK: [[LD:%.+]] = load x86_fp80, x86_fp80* |
144 | // CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[LDTEMP:%.*]] to i8* |
145 | // CHECK: call void @llvm.memset.p0i8.i64(i8* align 16 [[BITCAST]], i8 0, i64 16, i1 false) |
146 | // CHECK: store x86_fp80 [[LD]], x86_fp80* [[LDTEMP]] |
147 | // CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[LDTEMP:%.*]] to i128* |
148 | // CHECK: [[LD:%.+]] = load i128, i128* [[BITCAST]] |
149 | // CHECK: store atomic i128 [[LD]], i128* bitcast (x86_fp80* |
150 | #pragma omp atomic write |
151 | ldx = ldv; |
152 | // CHECK: [[REAL_VAL:%.+]] = load i32, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.*}}, i32 0, i32 0) |
153 | // CHECK: [[IMG_VAL:%.+]] = load i32, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.*}}, i32 0, i32 1) |
154 | // CHECK: [[TEMP_REAL_REF:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP:%.+]], i32 0, i32 0 |
155 | // CHECK: [[TEMP_IMG_REF:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1 |
156 | // CHECK: store i32 [[REAL_VAL]], i32* [[TEMP_REAL_REF]] |
157 | // CHECK: store i32 [[IMG_VAL]], i32* [[TEMP_IMG_REF]] |
158 | // CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[TEMP]] to i8* |
159 | // CHECK: call void @__atomic_store(i64 8, i8* bitcast ({ i32, i32 }* @{{.*}} to i8*), i8* [[BITCAST]], i32 0) |
160 | #pragma omp atomic write |
161 | cix = civ; |
162 | // CHECK: [[REAL_VAL:%.+]] = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @{{.*}}, i32 0, i32 0) |
163 | // CHECK: [[IMG_VAL:%.+]] = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @{{.*}}, i32 0, i32 1) |
164 | // CHECK: [[TEMP_REAL_REF:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[TEMP:%.+]], i32 0, i32 0 |
165 | // CHECK: [[TEMP_IMG_REF:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[TEMP]], i32 0, i32 1 |
166 | // CHECK: store float [[REAL_VAL]], float* [[TEMP_REAL_REF]] |
167 | // CHECK: store float [[IMG_VAL]], float* [[TEMP_IMG_REF]] |
168 | // CHECK: [[BITCAST:%.+]] = bitcast { float, float }* [[TEMP]] to i8* |
169 | // CHECK: call void @__atomic_store(i64 8, i8* bitcast ({ float, float }* @{{.*}} to i8*), i8* [[BITCAST]], i32 0) |
170 | #pragma omp atomic write |
171 | cfx = cfv; |
172 | // CHECK: [[REAL_VAL:%.+]] = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @{{.*}}, i32 0, i32 0) |
173 | // CHECK: [[IMG_VAL:%.+]] = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @{{.*}}, i32 0, i32 1) |
174 | // CHECK: [[TEMP_REAL_REF:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[TEMP:%.+]], i32 0, i32 0 |
175 | // CHECK: [[TEMP_IMG_REF:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[TEMP]], i32 0, i32 1 |
176 | // CHECK: store double [[REAL_VAL]], double* [[TEMP_REAL_REF]] |
177 | // CHECK: store double [[IMG_VAL]], double* [[TEMP_IMG_REF]] |
178 | // CHECK: [[BITCAST:%.+]] = bitcast { double, double }* [[TEMP]] to i8* |
179 | // CHECK: call void @__atomic_store(i64 16, i8* bitcast ({ double, double }* @{{.*}} to i8*), i8* [[BITCAST]], i32 5) |
180 | // CHECK: call{{.*}} @__kmpc_flush( |
181 | #pragma omp atomic seq_cst write |
182 | cdx = cdv; |
183 | // CHECK: load i8, i8* |
184 | // CHECK: store atomic i64 |
185 | #pragma omp atomic write |
186 | ulx = bv; |
187 | // CHECK: load i8, i8* |
188 | // CHECK: store atomic i8 |
189 | #pragma omp atomic write |
190 | bx = cv; |
191 | // CHECK: load i8, i8* |
192 | // CHECK: store atomic i8 |
193 | // CHECK: call{{.*}} @__kmpc_flush( |
194 | #pragma omp atomic write, seq_cst |
195 | cx = ucv; |
196 | // CHECK: load i16, i16* |
197 | // CHECK: store atomic i64 |
198 | #pragma omp atomic write |
199 | ulx = sv; |
200 | // CHECK: load i16, i16* |
201 | // CHECK: store atomic i64 |
202 | #pragma omp atomic write |
203 | lx = usv; |
204 | // CHECK: load i32, i32* |
205 | // CHECK: store atomic i32 |
206 | // CHECK: call{{.*}} @__kmpc_flush( |
207 | #pragma omp atomic seq_cst, write |
208 | uix = iv; |
209 | // CHECK: load i32, i32* |
210 | // CHECK: store atomic i32 |
211 | #pragma omp atomic write |
212 | ix = uiv; |
213 | // CHECK: load i64, i64* |
214 | // CHECK: [[VAL:%.+]] = trunc i64 %{{.*}} to i32 |
215 | // CHECK: [[TEMP_REAL_REF:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP:%.+]], i32 0, i32 0 |
216 | // CHECK: [[TEMP_IMG_REF:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1 |
217 | // CHECK: store i32 [[VAL]], i32* [[TEMP_REAL_REF]] |
218 | // CHECK: store i32 0, i32* [[TEMP_IMG_REF]] |
219 | // CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[TEMP]] to i8* |
220 | // CHECK: call void @__atomic_store(i64 8, i8* bitcast ({ i32, i32 }* @{{.+}} to i8*), i8* [[BITCAST]], i32 0) |
221 | #pragma omp atomic write |
222 | cix = lv; |
223 | // CHECK: load i64, i64* |
224 | // CHECK: store atomic i32 %{{.+}}, i32* bitcast (float* |
225 | #pragma omp atomic write |
226 | fx = ulv; |
227 | // CHECK: load i64, i64* |
228 | // CHECK: store atomic i64 %{{.+}}, i64* bitcast (double* |
229 | #pragma omp atomic write |
230 | dx = llv; |
231 | // CHECK: load i64, i64* |
232 | // CHECK: [[VAL:%.+]] = uitofp i64 %{{.+}} to x86_fp80 |
233 | // CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP:%.+]] to i8* |
234 | // CHECK: call void @llvm.memset.p0i8.i64(i8* align 16 [[BITCAST]], i8 0, i64 16, i1 false) |
235 | // CHECK: store x86_fp80 [[VAL]], x86_fp80* [[TEMP]] |
236 | // CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP]] to i128* |
237 | // CHECK: [[VAL:%.+]] = load i128, i128* [[BITCAST]] |
238 | // CHECK: store atomic i128 [[VAL]], i128* bitcast (x86_fp80* |
239 | #pragma omp atomic write |
240 | ldx = ullv; |
241 | // CHECK: load float, float* |
242 | // CHECK: [[VAL:%.+]] = fptosi float %{{.*}} to i32 |
243 | // CHECK: [[TEMP_REAL_REF:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP:%.+]], i32 0, i32 0 |
244 | // CHECK: [[TEMP_IMG_REF:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1 |
245 | // CHECK: store i32 [[VAL]], i32* [[TEMP_REAL_REF]] |
246 | // CHECK: store i32 0, i32* [[TEMP_IMG_REF]] |
247 | // CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[TEMP]] to i8* |
248 | // CHECK: call void @__atomic_store(i64 8, i8* bitcast ({ i32, i32 }* @{{.+}} to i8*), i8* [[BITCAST]], i32 0) |
249 | #pragma omp atomic write |
250 | cix = fv; |
251 | // CHECK: load double, double* |
252 | // CHECK: store atomic i16 |
253 | #pragma omp atomic write |
254 | sx = dv; |
255 | // CHECK: load x86_fp80, x86_fp80* |
256 | // CHECK: store atomic i8 |
257 | #pragma omp atomic write |
258 | bx = ldv; |
259 | // CHECK: load i32, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.+}}, i32 0, i32 0) |
260 | // CHECK: load i32, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.+}}, i32 0, i32 1) |
261 | // CHECK: icmp ne i32 %{{.+}}, 0 |
262 | // CHECK: icmp ne i32 %{{.+}}, 0 |
263 | // CHECK: or i1 |
264 | // CHECK: store atomic i8 |
265 | #pragma omp atomic write |
266 | bx = civ; |
267 | // CHECK: load float, float* getelementptr inbounds ({ float, float }, { float, float }* @{{.*}}, i32 0, i32 0) |
268 | // CHECK: store atomic i16 |
269 | #pragma omp atomic write |
270 | usx = cfv; |
271 | // CHECK: load double, double* getelementptr inbounds ({ double, double }, { double, double }* @{{.+}}, i32 0, i32 0) |
272 | // CHECK: store atomic i64 |
273 | #pragma omp atomic write |
274 | llx = cdv; |
275 | // CHECK-DAG: [[IDX:%.+]] = load i16, i16* @{{.+}} |
276 | // CHECK-DAG: load i8, i8* |
277 | // CHECK-DAG: [[VEC_ITEM_VAL:%.+]] = zext i1 %{{.+}} to i32 |
278 | // CHECK: [[I128VAL:%.+]] = load atomic i128, i128* bitcast (<4 x i32>* [[DEST:@.+]] to i128*) monotonic |
279 | // CHECK: br label %[[CONT:.+]] |
280 | // CHECK: [[CONT]] |
281 | // CHECK: [[OLD_I128:%.+]] = phi i128 [ [[I128VAL]], %{{.+}} ], [ [[FAILED_I128_OLD_VAL:%.+]], %[[CONT]] ] |
282 | // CHECK: [[BITCAST:%.+]] = bitcast <4 x i32>* [[LDTEMP:%.+]] to i128* |
283 | // CHECK: store i128 [[OLD_I128]], i128* [[BITCAST]], |
284 | // CHECK: [[VEC_VAL:%.+]] = load <4 x i32>, <4 x i32>* [[LDTEMP]] |
285 | // CHECK: [[NEW_VEC_VAL:%.+]] = insertelement <4 x i32> [[VEC_VAL]], i32 [[VEC_ITEM_VAL]], i16 [[IDX]] |
286 | // CHECK: store <4 x i32> [[NEW_VEC_VAL]], <4 x i32>* [[LDTEMP]] |
287 | // CHECK: [[NEW_I128:%.+]] = load i128, i128* [[BITCAST]] |
288 | // CHECK: [[RES:%.+]] = cmpxchg i128* bitcast (<4 x i32>* [[DEST]] to i128*), i128 [[OLD_I128]], i128 [[NEW_I128]] monotonic monotonic |
289 | // CHECK: [[FAILED_I128_OLD_VAL:%.+]] = extractvalue { i128, i1 } [[RES]], 0 |
290 | // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i128, i1 } [[RES]], 1 |
291 | // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]] |
292 | // CHECK: [[EXIT]] |
293 | #pragma omp atomic write |
294 | int4x[sv] = bv; |
295 | // CHECK: load x86_fp80, x86_fp80* @{{.+}} |
296 | // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32 |
297 | // CHECK: [[PREV_VALUE:%.+]] = load atomic i32, i32* bitcast (i8* getelementptr (i8, i8* bitcast (%struct.BitFields* @{{.+}} to i8*), i64 4) to i32*) monotonic |
298 | // CHECK: br label %[[CONT:.+]] |
299 | // CHECK: [[CONT]] |
300 | // CHECK: [[OLD_BF_VALUE:%.+]] = phi i32 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ] |
301 | // CHECK: [[BF_VALUE:%.+]] = and i32 [[NEW_VAL]], 2147483647 |
302 | // CHECK: [[BF_CLEAR:%.+]] = and i32 %{{.+}}, -2147483648 |
303 | // CHECK: or i32 [[BF_CLEAR]], [[BF_VALUE]] |
304 | // CHECK: store i32 %{{.+}}, i32* [[LDTEMP:%.+]] |
305 | // CHECK: [[NEW_BF_VALUE:%.+]] = load i32, i32* [[LDTEMP]] |
306 | // 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 |
307 | // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i32, i1 } [[RES]], 0 |
308 | // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i32, i1 } [[RES]], 1 |
309 | // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]] |
310 | // CHECK: [[EXIT]] |
311 | #pragma omp atomic write |
312 | bfx.a = ldv; |
313 | // CHECK: load x86_fp80, x86_fp80* @{{.+}} |
314 | // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32 |
315 | // CHECK: [[BITCAST:%.+]] = bitcast i32* [[LDTEMP:%.+]] to i8* |
316 | // CHECK: call void @__atomic_load(i64 4, i8* getelementptr (i8, i8* bitcast (%struct.BitFields_packed* @{{.+}} to i8*), i64 4), i8* [[BITCAST]], i32 0) |
317 | // CHECK: br label %[[CONT:.+]] |
318 | // CHECK: [[CONT]] |
319 | // CHECK: [[OLD_BF_VALUE:%.+]] = load i32, i32* [[LDTEMP]], |
320 | // CHECK: store i32 [[OLD_BF_VALUE]], i32* [[LDTEMP1:%.+]], |
321 | // CHECK: [[OLD_BF_VALUE:%.+]] = load i32, i32* [[LDTEMP1]], |
322 | // CHECK: [[BF_VALUE:%.+]] = and i32 [[NEW_VAL]], 2147483647 |
323 | // CHECK: [[BF_CLEAR:%.+]] = and i32 [[OLD_BF_VALUE]], -2147483648 |
324 | // CHECK: or i32 [[BF_CLEAR]], [[BF_VALUE]] |
325 | // CHECK: store i32 %{{.+}}, i32* [[LDTEMP1]] |
326 | // CHECK: [[BITCAST_TEMP_OLD_BF_ADDR:%.+]] = bitcast i32* [[LDTEMP]] to i8* |
327 | // CHECK: [[BITCAST_TEMP_NEW_BF_ADDR:%.+]] = bitcast i32* [[LDTEMP1]] to i8* |
328 | // 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) |
329 | // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]] |
330 | // CHECK: [[EXIT]] |
331 | #pragma omp atomic write |
332 | bfx_packed.a = ldv; |
333 | // CHECK: load x86_fp80, x86_fp80* @{{.+}} |
334 | // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32 |
335 | // CHECK: [[PREV_VALUE:%.+]] = load atomic i32, i32* getelementptr inbounds (%struct.BitFields2, %struct.BitFields2* @{{.+}}, i32 0, i32 0) monotonic |
336 | // CHECK: br label %[[CONT:.+]] |
337 | // CHECK: [[CONT]] |
338 | // CHECK: [[OLD_BF_VALUE:%.+]] = phi i32 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ] |
339 | // CHECK: [[BF_AND:%.+]] = and i32 [[NEW_VAL]], 1 |
340 | // CHECK: [[BF_VALUE:%.+]] = shl i32 [[BF_AND]], 31 |
341 | // CHECK: [[BF_CLEAR:%.+]] = and i32 %{{.+}}, 2147483647 |
342 | // CHECK: or i32 [[BF_CLEAR]], [[BF_VALUE]] |
343 | // CHECK: store i32 %{{.+}}, i32* [[LDTEMP:%.+]] |
344 | // CHECK: [[NEW_BF_VALUE:%.+]] = load i32, i32* [[LDTEMP]] |
345 | // CHECK: [[RES:%.+]] = cmpxchg i32* getelementptr inbounds (%struct.BitFields2, %struct.BitFields2* @{{.+}}, i32 0, i32 0), i32 [[OLD_BF_VALUE]], i32 [[NEW_BF_VALUE]] monotonic monotonic |
346 | // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i32, i1 } [[RES]], 0 |
347 | // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i32, i1 } [[RES]], 1 |
348 | // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]] |
349 | // CHECK: [[EXIT]] |
350 | #pragma omp atomic write |
351 | bfx2.a = ldv; |
352 | // CHECK: load x86_fp80, x86_fp80* @{{.+}} |
353 | // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32 |
354 | // CHECK: [[PREV_VALUE:%.+]] = load atomic i8, i8* getelementptr (i8, i8* bitcast (%struct.BitFields2_packed* @{{.+}} to i8*), i64 3) monotonic |
355 | // CHECK: br label %[[CONT:.+]] |
356 | // CHECK: [[CONT]] |
357 | // CHECK: [[OLD_BF_VALUE:%.+]] = phi i8 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ] |
358 | // CHECK: [[TRUNC:%.+]] = trunc i32 [[NEW_VAL]] to i8 |
359 | // CHECK: [[BF_AND:%.+]] = and i8 [[TRUNC]], 1 |
360 | // CHECK: [[BF_VALUE:%.+]] = shl i8 [[BF_AND]], 7 |
361 | // CHECK: [[BF_CLEAR:%.+]] = and i8 %{{.+}}, 127 |
362 | // CHECK: or i8 [[BF_CLEAR]], [[BF_VALUE]] |
363 | // CHECK: store i8 %{{.+}}, i8* [[LDTEMP:%.+]] |
364 | // CHECK: [[NEW_BF_VALUE:%.+]] = load i8, i8* [[LDTEMP]] |
365 | // 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 |
366 | // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i8, i1 } [[RES]], 0 |
367 | // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i8, i1 } [[RES]], 1 |
368 | // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]] |
369 | // CHECK: [[EXIT]] |
370 | #pragma omp atomic write |
371 | bfx2_packed.a = ldv; |
372 | // CHECK: load x86_fp80, x86_fp80* @{{.+}} |
373 | // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32 |
374 | // CHECK: [[PREV_VALUE:%.+]] = load atomic i32, i32* getelementptr inbounds (%struct.BitFields3, %struct.BitFields3* @{{.+}}, i32 0, i32 0) monotonic |
375 | // CHECK: br label %[[CONT:.+]] |
376 | // CHECK: [[CONT]] |
377 | // CHECK: [[OLD_BF_VALUE:%.+]] = phi i32 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ] |
378 | // CHECK: [[BF_AND:%.+]] = and i32 [[NEW_VAL]], 16383 |
379 | // CHECK: [[BF_VALUE:%.+]] = shl i32 [[BF_AND]], 11 |
380 | // CHECK: [[BF_CLEAR:%.+]] = and i32 %{{.+}}, -33552385 |
381 | // CHECK: or i32 [[BF_CLEAR]], [[BF_VALUE]] |
382 | // CHECK: store i32 %{{.+}}, i32* [[LDTEMP:%.+]] |
383 | // CHECK: [[NEW_BF_VALUE:%.+]] = load i32, i32* [[LDTEMP]] |
384 | // CHECK: [[RES:%.+]] = cmpxchg i32* getelementptr inbounds (%struct.BitFields3, %struct.BitFields3* @{{.+}}, i32 0, i32 0), i32 [[OLD_BF_VALUE]], i32 [[NEW_BF_VALUE]] monotonic monotonic |
385 | // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i32, i1 } [[RES]], 0 |
386 | // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i32, i1 } [[RES]], 1 |
387 | // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]] |
388 | // CHECK: [[EXIT]] |
389 | #pragma omp atomic write |
390 | bfx3.a = ldv; |
391 | // CHECK: load x86_fp80, x86_fp80* @{{.+}} |
392 | // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32 |
393 | // CHECK: [[LDTEMP:%.+]] = bitcast i32* %{{.+}} to i24* |
394 | // CHECK: [[BITCAST:%.+]] = bitcast i24* %{{.+}} to i8* |
395 | // CHECK: call void @__atomic_load(i64 3, i8* getelementptr (i8, i8* bitcast (%struct.BitFields3_packed* @{{.+}} to i8*), i64 1), i8* [[BITCAST]], i32 0) |
396 | // CHECK: br label %[[CONT:.+]] |
397 | // CHECK: [[CONT]] |
398 | // CHECK: [[OLD_VAL:%.+]] = load i24, i24* %{{.+}}, |
399 | // CHECK: store i24 [[OLD_VAL]], i24* [[TEMP:%.+]], |
400 | // CHECK: [[TRUNC:%.+]] = trunc i32 [[NEW_VAL]] to i24 |
401 | // CHECK: [[BF_AND:%.+]] = and i24 [[TRUNC]], 16383 |
402 | // CHECK: [[BF_VALUE:%.+]] = shl i24 [[BF_AND]], 3 |
403 | // CHECK: [[BF_CLEAR:%.+]] = and i24 %{{.+}}, -131065 |
404 | // CHECK: or i24 [[BF_CLEAR]], [[BF_VALUE]] |
405 | // CHECK: store i24 %{{.+}}, i24* [[TEMP]] |
406 | // CHECK: [[BITCAST_TEMP_OLD_BF_ADDR:%.+]] = bitcast i24* [[LDTEMP]] to i8* |
407 | // CHECK: [[BITCAST_TEMP_NEW_BF_ADDR:%.+]] = bitcast i24* [[TEMP]] 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: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]] |
410 | // CHECK: [[EXIT]] |
411 | #pragma omp atomic write |
412 | bfx3_packed.a = ldv; |
413 | // CHECK: load x86_fp80, x86_fp80* @{{.+}} |
414 | // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32 |
415 | // CHECK: [[PREV_VALUE:%.+]] = load atomic i64, i64* bitcast (%struct.BitFields4* @{{.+}} to i64*) monotonic |
416 | // CHECK: br label %[[CONT:.+]] |
417 | // CHECK: [[CONT]] |
418 | // CHECK: [[OLD_BF_VALUE:%.+]] = phi i64 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ] |
419 | // CHECK: [[ZEXT:%.+]] = zext i32 [[NEW_VAL]] to i64 |
420 | // CHECK: [[BF_AND:%.+]] = and i64 [[ZEXT]], 1 |
421 | // CHECK: [[BF_VALUE:%.+]] = shl i64 [[BF_AND]], 16 |
422 | // CHECK: [[BF_CLEAR:%.+]] = and i64 %{{.+}}, -65537 |
423 | // CHECK: or i64 [[BF_CLEAR]], [[BF_VALUE]] |
424 | // CHECK: store i64 %{{.+}}, i64* [[LDTEMP:%.+]] |
425 | // CHECK: [[NEW_BF_VALUE:%.+]] = load i64, i64* [[LDTEMP]] |
426 | // CHECK: [[RES:%.+]] = cmpxchg i64* bitcast (%struct.BitFields4* @{{.+}} to i64*), i64 [[OLD_BF_VALUE]], i64 [[NEW_BF_VALUE]] monotonic monotonic |
427 | // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i64, i1 } [[RES]], 0 |
428 | // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i64, i1 } [[RES]], 1 |
429 | // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]] |
430 | // CHECK: [[EXIT]] |
431 | #pragma omp atomic write |
432 | bfx4.a = ldv; |
433 | // CHECK: load x86_fp80, x86_fp80* @{{.+}} |
434 | // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32 |
435 | // CHECK: [[PREV_VALUE:%.+]] = load atomic i8, i8* getelementptr inbounds (%struct.BitFields4_packed, %struct.BitFields4_packed* @{{.+}}, i32 0, i32 0, i64 2) monotonic |
436 | // CHECK: br label %[[CONT:.+]] |
437 | // CHECK: [[CONT]] |
438 | // CHECK: [[OLD_BF_VALUE:%.+]] = phi i8 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ] |
439 | // CHECK: [[TRUNC:%.+]] = trunc i32 [[NEW_VAL]] to i8 |
440 | // CHECK: [[BF_VALUE:%.+]] = and i8 [[TRUNC]], 1 |
441 | // CHECK: [[BF_CLEAR:%.+]] = and i8 %{{.+}}, -2 |
442 | // CHECK: or i8 [[BF_CLEAR]], [[BF_VALUE]] |
443 | // CHECK: store i8 %{{.+}}, i8* [[LDTEMP:%.+]] |
444 | // CHECK: [[NEW_BF_VALUE:%.+]] = load i8, i8* [[LDTEMP]] |
445 | // 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 |
446 | // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i8, i1 } [[RES]], 0 |
447 | // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i8, i1 } [[RES]], 1 |
448 | // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]] |
449 | // CHECK: [[EXIT]] |
450 | #pragma omp atomic write |
451 | bfx4_packed.a = ldv; |
452 | // CHECK: load x86_fp80, x86_fp80* @{{.+}} |
453 | // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i64 |
454 | // CHECK: [[PREV_VALUE:%.+]] = load atomic i64, i64* bitcast (%struct.BitFields4* @{{.+}} to i64*) monotonic |
455 | // CHECK: br label %[[CONT:.+]] |
456 | // CHECK: [[CONT]] |
457 | // CHECK: [[OLD_BF_VALUE:%.+]] = phi i64 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ] |
458 | // CHECK: [[BF_AND:%.+]] = and i64 [[NEW_VAL]], 127 |
459 | // CHECK: [[BF_VALUE:%.+]] = shl i64 [[BF_AND]], 17 |
460 | // CHECK: [[BF_CLEAR:%.+]] = and i64 %{{.+}}, -16646145 |
461 | // CHECK: or i64 [[BF_CLEAR]], [[BF_VALUE]] |
462 | // CHECK: store i64 %{{.+}}, i64* [[LDTEMP:%.+]] |
463 | // CHECK: [[NEW_BF_VALUE:%.+]] = load i64, i64* [[LDTEMP]] |
464 | // CHECK: [[RES:%.+]] = cmpxchg i64* bitcast (%struct.BitFields4* @{{.+}} to i64*), i64 [[OLD_BF_VALUE]], i64 [[NEW_BF_VALUE]] monotonic monotonic |
465 | // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i64, i1 } [[RES]], 0 |
466 | // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i64, i1 } [[RES]], 1 |
467 | // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]] |
468 | // CHECK: [[EXIT]] |
469 | #pragma omp atomic write |
470 | bfx4.b = ldv; |
471 | // CHECK: load x86_fp80, x86_fp80* @{{.+}} |
472 | // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i64 |
473 | // CHECK: [[PREV_VALUE:%.+]] = load atomic i8, i8* getelementptr inbounds (%struct.BitFields4_packed, %struct.BitFields4_packed* @{{.+}}, i32 0, i32 0, i64 2) monotonic |
474 | // CHECK: br label %[[CONT:.+]] |
475 | // CHECK: [[CONT]] |
476 | // CHECK: [[OLD_BF_VALUE:%.+]] = phi i8 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ] |
477 | // CHECK: [[TRUNC:%.+]] = trunc i64 [[NEW_VAL]] to i8 |
478 | // CHECK: [[BF_AND:%.+]] = and i8 [[TRUNC]], 127 |
479 | // CHECK: [[BF_VALUE:%.+]] = shl i8 [[BF_AND]], 1 |
480 | // CHECK: [[BF_CLEAR:%.+]] = and i8 %{{.+}}, 1 |
481 | // CHECK: or i8 [[BF_CLEAR]], [[BF_VALUE]] |
482 | // CHECK: store i8 %{{.+}}, i8* [[LDTEMP:%.+]] |
483 | // CHECK: [[NEW_BF_VALUE:%.+]] = load i8, i8* [[LDTEMP]] |
484 | // 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 |
485 | // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i8, i1 } [[RES]], 0 |
486 | // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i8, i1 } [[RES]], 1 |
487 | // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]] |
488 | // CHECK: [[EXIT]] |
489 | #pragma omp atomic write |
490 | bfx4_packed.b = ldv; |
491 | // CHECK: load i64, i64* |
492 | // CHECK: [[VEC_ITEM_VAL:%.+]] = uitofp i64 %{{.+}} to float |
493 | // CHECK: [[I64VAL:%.+]] = load atomic i64, i64* bitcast (<2 x float>* [[DEST:@.+]] to i64*) monotonic |
494 | // CHECK: br label %[[CONT:.+]] |
495 | // CHECK: [[CONT]] |
496 | // CHECK: [[OLD_I64:%.+]] = phi i64 [ [[I64VAL]], %{{.+}} ], [ [[FAILED_I64_OLD_VAL:%.+]], %[[CONT]] ] |
497 | // CHECK: [[BITCAST:%.+]] = bitcast <2 x float>* [[LDTEMP:%.+]] to i64* |
498 | // CHECK: store i64 [[OLD_I64]], i64* [[BITCAST]], |
499 | // CHECK: [[VEC_VAL:%.+]] = load <2 x float>, <2 x float>* [[LDTEMP]] |
500 | // CHECK: [[NEW_VEC_VAL:%.+]] = insertelement <2 x float> [[VEC_VAL]], float [[VEC_ITEM_VAL]], i64 0 |
501 | // CHECK: store <2 x float> [[NEW_VEC_VAL]], <2 x float>* [[LDTEMP]] |
502 | // CHECK: [[NEW_I64:%.+]] = load i64, i64* [[BITCAST]] |
503 | // CHECK: [[RES:%.+]] = cmpxchg i64* bitcast (<2 x float>* [[DEST]] to i64*), i64 [[OLD_I64]], i64 [[NEW_I64]] monotonic monotonic |
504 | // CHECK: [[FAILED_I64_OLD_VAL:%.+]] = extractvalue { i64, i1 } [[RES]], 0 |
505 | // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i64, i1 } [[RES]], 1 |
506 | // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]] |
507 | // CHECK: [[EXIT]] |
508 | #pragma omp atomic write |
509 | float2x.x = ulv; |
510 | // CHECK: call i32 @llvm.read_register.i32( |
511 | // CHECK: sitofp i32 %{{.+}} to double |
512 | // CHECK: bitcast double %{{.+}} to i64 |
513 | // CHECK: store atomic i64 %{{.+}}, i64* bitcast (double* @{{.+}} to i64*) seq_cst |
514 | // CHECK: call{{.*}} @__kmpc_flush( |
515 | #pragma omp atomic write seq_cst |
516 | dv = rix; |
517 | return 0; |
518 | } |
519 | |
520 | #endif |
521 | |