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 #ifndef HEADER
6 #define HEADER
7 
8 _Bool bv, bx;
9 char cv, cx;
10 unsigned char ucv, ucx;
11 short sv, sx;
12 unsigned short usv, usx;
13 int iv, ix;
14 unsigned int uiv, uix;
15 long lv, lx;
16 unsigned long ulv, ulx;
17 long long llv, llx;
18 unsigned long long ullv, ullx;
19 float fv, fx;
20 double dv, dx;
21 long double ldv, ldx;
22 _Complex int civ, cix;
23 _Complex float cfv, cfx;
24 _Complex double cdv, cdx;
25 
26 typedef int int4 __attribute__((__vector_size__(16)));
27 int4 int4x;
28 
29 struct BitFields {
30   int : 32;
31   int a : 31;
32 } bfx;
33 
34 struct BitFields_packed {
35   int : 32;
36   int a : 31;
37 } __attribute__ ((__packed__)) bfx_packed;
38 
39 struct BitFields2 {
40   int : 31;
41   int a : 1;
42 } bfx2;
43 
44 struct BitFields2_packed {
45   int : 31;
46   int a : 1;
47 } __attribute__ ((__packed__)) bfx2_packed;
48 
49 struct BitFields3 {
50   int : 11;
51   int a : 14;
52 } bfx3;
53 
54 struct BitFields3_packed {
55   int : 11;
56   int a : 14;
57 } __attribute__ ((__packed__)) bfx3_packed;
58 
59 struct BitFields4 {
60   short : 16;
61   int a: 1;
62   long b : 7;
63 } bfx4;
64 
65 struct BitFields4_packed {
66   short : 16;
67   int a: 1;
68   long b : 7;
69 } __attribute__ ((__packed__)) bfx4_packed;
70 
71 typedef float float2 __attribute__((ext_vector_type(2)));
72 float2 float2x;
73 
74 // Register "0" is currently an invalid register for global register variables.
75 // Use "esp" instead of "0".
76 // register int rix __asm__("0");
77 register int rix __asm__("esp");
78 
79 int main() {
80 // CHECK: [[PREV:%.+]] = atomicrmw add i8* @{{.+}}, i8 1 monotonic
81 // CHECK: store i8 [[PREV]], i8* @{{.+}},
82 #pragma omp atomic capture
83   bv = bx++;
84 // CHECK: atomicrmw add i8* @{{.+}}, i8 1 monotonic
85 // CHECK: add nsw i32 %{{.+}}, 1
86 // CHECK: store i8 %{{.+}}, i8* @{{.+}},
87 #pragma omp atomic capture
88   cv = ++cx;
89 // CHECK: [[PREV:%.+]] = atomicrmw sub i8* @{{.+}}, i8 1 monotonic
90 // CHECK: store i8 [[PREV]], i8* @{{.+}},
91 #pragma omp atomic capture
92   ucv = ucx--;
93 // CHECK: atomicrmw sub i16* @{{.+}}, i16 1 monotonic
94 // CHECK: sub nsw i32 %{{.+}}, 1
95 // CHECK: store i16 %{{.+}}, i16* @{{.+}},
96 #pragma omp atomic capture
97   sv = --sx;
98 // CHECK: [[USV:%.+]] = load i16, i16* @{{.+}},
99 // CHECK: [[EXPR:%.+]] = zext i16 [[USV]] to i32
100 // CHECK: [[X:%.+]] = load atomic i16, i16* [[X_ADDR:@.+]] monotonic
101 // CHECK: br label %[[CONT:.+]]
102 // CHECK: [[CONT]]
103 // CHECK: [[EXPECTED:%.+]] = phi i16 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
104 // CHECK: [[CONV:%.+]] = zext i16 [[EXPECTED]] to i32
105 // CHECK: [[ADD:%.+]] = add nsw i32 [[CONV]], [[EXPR]]
106 // CHECK: [[DESIRED_CALC:%.+]] = trunc i32 [[ADD]] to i16
107 // CHECK: store i16 [[DESIRED_CALC]], i16* [[TEMP:%.+]],
108 // CHECK: [[DESIRED:%.+]] = load i16, i16* [[TEMP]],
109 // CHECK: [[RES:%.+]] = cmpxchg i16* [[X_ADDR]], i16 [[EXPECTED]], i16 [[DESIRED]] monotonic monotonic
110 // CHECK: [[OLD_X]] = extractvalue { i16, i1 } [[RES]], 0
111 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i16, i1 } [[RES]], 1
112 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
113 // CHECK: [[EXIT]]
114 // CHECK: store i16 [[DESIRED_CALC]], i16* @{{.+}},
115 #pragma omp atomic capture
116   sv = usx += usv;
117 // CHECK: [[EXPR:%.+]] = load i32, i32* @{{.+}},
118 // CHECK: [[X:%.+]] = load atomic i32, i32* [[X_ADDR:@.+]] monotonic
119 // CHECK: br label %[[CONT:.+]]
120 // CHECK: [[CONT]]
121 // CHECK: [[EXPECTED:%.+]] = phi i32 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
122 // CHECK: [[DESIRED_CALC:%.+]] = mul nsw i32 [[EXPECTED]], [[EXPR]]
123 // CHECK: store i32 [[DESIRED_CALC]], i32* [[TEMP:%.+]],
124 // CHECK: [[DESIRED:%.+]] = load i32, i32* [[TEMP]],
125 // CHECK: [[RES:%.+]] = cmpxchg i32* [[X_ADDR]], i32 [[EXPECTED]], i32 [[DESIRED]] monotonic monotonic
126 // CHECK: [[OLD_X]] = extractvalue { i32, i1 } [[RES]], 0
127 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1
128 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
129 // CHECK: [[EXIT]]
130 // CHECK: store i32 [[DESIRED_CALC]], i32* @{{.+}},
131 #pragma omp atomic capture
132   uiv = ix *= iv;
133 // CHECK: [[EXPR:%.+]] = load i32, i32* @{{.+}},
134 // CHECK: [[PREV:%.+]] = atomicrmw sub i32* @{{.+}}, i32 [[EXPR]] monotonic
135 // CHECK: store i32 [[PREV]], i32* @{{.+}},
136 #pragma omp atomic capture
137   {iv = uix; uix -= uiv;}
138 // CHECK: [[EXPR:%.+]] = load i32, i32* @{{.+}},
139 // CHECK: [[X:%.+]] = load atomic i32, i32* [[X_ADDR:@.+]] monotonic
140 // CHECK: br label %[[CONT:.+]]
141 // CHECK: [[CONT]]
142 // CHECK: [[EXPECTED:%.+]] = phi i32 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
143 // CHECK: [[DESIRED_CALC:%.+]] = shl i32 [[EXPECTED]], [[EXPR]]
144 // CHECK: store i32 [[DESIRED_CALC]], i32* [[TEMP:%.+]],
145 // CHECK: [[DESIRED:%.+]] = load i32, i32* [[TEMP]],
146 // CHECK: [[RES:%.+]] = cmpxchg i32* [[X_ADDR]], i32 [[EXPECTED]], i32 [[DESIRED]] monotonic monotonic
147 // CHECK: [[OLD_X]] = extractvalue { i32, i1 } [[RES]], 0
148 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1
149 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
150 // CHECK: [[EXIT]]
151 // CHECK: store i32 [[DESIRED_CALC]], i32* @{{.+}},
152 #pragma omp atomic capture
153   {ix <<= iv; uiv = ix;}
154 // CHECK: [[EXPR:%.+]] = load i32, i32* @{{.+}},
155 // CHECK: [[X:%.+]] = load atomic i32, i32* [[X_ADDR:@.+]] monotonic
156 // CHECK: br label %[[CONT:.+]]
157 // CHECK: [[CONT]]
158 // CHECK: [[EXPECTED:%.+]] = phi i32 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
159 // CHECK: [[DESIRED_CALC:%.+]] = lshr i32 [[EXPECTED]], [[EXPR]]
160 // CHECK: store i32 [[DESIRED_CALC]], i32* [[TEMP:%.+]],
161 // CHECK: [[DESIRED:%.+]] = load i32, i32* [[TEMP]],
162 // CHECK: [[RES:%.+]] = cmpxchg i32* [[X_ADDR]], i32 [[EXPECTED]], i32 [[DESIRED]] monotonic monotonic
163 // CHECK: [[OLD_X]] = extractvalue { i32, i1 } [[RES]], 0
164 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1
165 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
166 // CHECK: [[EXIT]]
167 // CHECK: store i32 [[DESIRED_CALC]], i32* @{{.+}},
168 #pragma omp atomic capture
169   iv = uix >>= uiv;
170 // CHECK: [[EXPR:%.+]] = load i64, i64* @{{.+}},
171 // CHECK: [[X:%.+]] = load atomic i64, i64* [[X_ADDR:@.+]] monotonic
172 // CHECK: br label %[[CONT:.+]]
173 // CHECK: [[CONT]]
174 // CHECK: [[EXPECTED:%.+]] = phi i64 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
175 // CHECK: [[DESIRED:%.+]] = sdiv i64 [[EXPECTED]], [[EXPR]]
176 // CHECK: store i64 [[DESIRED]], i64* [[TEMP:%.+]],
177 // CHECK: [[DESIRED:%.+]] = load i64, i64* [[TEMP]],
178 // CHECK: [[RES:%.+]] = cmpxchg i64* [[X_ADDR]], i64 [[EXPECTED]], i64 [[DESIRED]] monotonic monotonic
179 // CHECK: [[OLD_X]] = extractvalue { i64, i1 } [[RES]], 0
180 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i64, i1 } [[RES]], 1
181 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
182 // CHECK: [[EXIT]]
183 // CHECK: store i64 [[EXPECTED]], i64* @{{.+}},
184 #pragma omp atomic capture
185   {ulv = lx; lx /= lv;}
186 // CHECK: [[EXPR:%.+]] = load i64, i64* @{{.+}},
187 // CHECK: [[OLD:%.+]] = atomicrmw and i64* @{{.+}}, i64 [[EXPR]] monotonic
188 // CHECK: [[DESIRED:%.+]] = and i64 [[OLD]], [[EXPR]]
189 // CHECK:  store i64 [[DESIRED]], i64* @{{.+}},
190 #pragma omp atomic capture
191   {ulx &= ulv; lv = ulx;}
192 // CHECK: [[EXPR:%.+]] = load i64, i64* @{{.+}},
193 // CHECK: [[OLD:%.+]] = atomicrmw xor i64* @{{.+}}, i64 [[EXPR]] monotonic
194 // CHECK: [[DESIRED:%.+]] = xor i64 [[OLD]], [[EXPR]]
195 // CHECK:  store i64 [[DESIRED]], i64* @{{.+}},
196 #pragma omp atomic capture
197   ullv = llx ^= llv;
198 // CHECK: [[EXPR:%.+]] = load i64, i64* @{{.+}},
199 // CHECK: [[OLD:%.+]] = atomicrmw or i64* @{{.+}}, i64 [[EXPR]] monotonic
200 // CHECK: [[DESIRED:%.+]] = or i64 [[OLD]], [[EXPR]]
201 // CHECK:  store i64 [[DESIRED]], i64* @{{.+}},
202 #pragma omp atomic capture
203   llv = ullx |= ullv;
204 // CHECK: [[EXPR:%.+]] = load float, float* @{{.+}},
205 // CHECK: [[X:%.+]] = load atomic i32, i32*  bitcast (float* [[X_ADDR:@.+]] to i32*) monotonic
206 // CHECK: br label %[[CONT:.+]]
207 // CHECK: [[CONT]]
208 // CHECK: [[EXPECTED:%.+]] = phi i32 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
209 // CHECK: [[TEMP_I:%.+]] = bitcast float* [[TEMP:%.+]] to i32*
210 // CHECK: [[OLD:%.+]] = bitcast i32 [[EXPECTED]] to float
211 // CHECK: [[ADD:%.+]] = fadd float [[OLD]], [[EXPR]]
212 // CHECK: store float [[ADD]], float* [[TEMP]],
213 // CHECK: [[DESIRED:%.+]] = load i32, i32* [[TEMP_I]],
214 // CHECK: [[RES:%.+]] = cmpxchg i32* bitcast (float* [[X_ADDR]] to i32*), i32 [[EXPECTED]], i32 [[DESIRED]] monotonic monotonic
215 // CHECK: [[OLD_X:%.+]] = extractvalue { i32, i1 } [[RES]], 0
216 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1
217 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
218 // CHECK: [[EXIT]]
219 // CHECK: [[CAST:%.+]] = fpext float [[ADD]] to double
220 // CHECK: store double [[CAST]], double* @{{.+}},
221 #pragma omp atomic capture
222   dv = fx = fx + fv;
223 // CHECK: [[EXPR:%.+]] = load double, double* @{{.+}},
224 // CHECK: [[X:%.+]] = load atomic i64, i64*  bitcast (double* [[X_ADDR:@.+]] to i64*) monotonic
225 // CHECK: br label %[[CONT:.+]]
226 // CHECK: [[CONT]]
227 // CHECK: [[EXPECTED:%.+]] = phi i64 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
228 // CHECK: [[TEMP_I:%.+]] = bitcast double* [[TEMP:%.+]] to i64*
229 // CHECK: [[OLD:%.+]] = bitcast i64 [[EXPECTED]] to double
230 // CHECK: [[SUB:%.+]] = fsub double [[EXPR]], [[OLD]]
231 // CHECK: store double [[SUB]], double* [[TEMP]],
232 // CHECK: [[DESIRED:%.+]] = load i64, i64* [[TEMP_I]],
233 // CHECK: [[RES:%.+]] = cmpxchg i64* bitcast (double* [[X_ADDR]] to i64*), i64 [[EXPECTED]], i64 [[DESIRED]] monotonic monotonic
234 // CHECK: [[OLD_X:%.+]] = extractvalue { i64, i1 } [[RES]], 0
235 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i64, i1 } [[RES]], 1
236 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
237 // CHECK: [[EXIT]]
238 // CHECK: [[CAST:%.+]] = fptrunc double [[OLD]] to float
239 // CHECK: store float [[CAST]], float* @{{.+}},
240 #pragma omp atomic capture
241   {fv = dx; dx = dv - dx;}
242 // CHECK: [[EXPR:%.+]] = load x86_fp80, x86_fp80* @{{.+}},
243 // CHECK: [[X:%.+]] = load atomic i128, i128*  bitcast (x86_fp80* [[X_ADDR:@.+]] to i128*) monotonic
244 // CHECK: br label %[[CONT:.+]]
245 // CHECK: [[CONT]]
246 // CHECK: [[EXPECTED:%.+]] = phi i128 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
247 // CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP:%.+]] to i128*
248 // CHECK: store i128 [[EXPECTED]], i128* [[BITCAST]]
249 // CHECK: [[BITCAST1:%.+]] = bitcast x86_fp80* [[TEMP1:%.+]] to i128*
250 // CHECK: store i128 [[EXPECTED]], i128* [[BITCAST1]]
251 // CHECK: [[OLD:%.+]] = load x86_fp80, x86_fp80* [[TEMP1]]
252 // CHECK: [[MUL:%.+]] = fmul x86_fp80 [[OLD]], [[EXPR]]
253 // CHECK: store x86_fp80 [[MUL]], x86_fp80* [[TEMP]]
254 // CHECK: [[DESIRED:%.+]] = load i128, i128* [[BITCAST]]
255 // CHECK: [[RES:%.+]] = cmpxchg i128* bitcast (x86_fp80* [[X_ADDR]] to i128*), i128 [[EXPECTED]], i128 [[DESIRED]] monotonic monotonic
256 // CHECK: [[OLD_X:%.+]] = extractvalue { i128, i1 } [[RES]], 0
257 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i128, i1 } [[RES]], 1
258 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
259 // CHECK: [[EXIT]]
260 // CHECK: [[CAST:%.+]] = fptrunc x86_fp80 [[MUL]] to double
261 // CHECK: store double [[CAST]], double* @{{.+}},
262 #pragma omp atomic capture
263   {ldx = ldx * ldv; dv = ldx;}
264 // CHECK: [[EXPR_RE:%.+]] = load i32, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.+}}, i32 0, i32 0)
265 // CHECK: [[EXPR_IM:%.+]] = load i32, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.+}}, i32 0, i32 1)
266 // CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[EXPECTED_ADDR:%.+]] to i8*
267 // CHECK: call void @__atomic_load(i64 8, i8* bitcast ({ i32, i32 }* [[X_ADDR:@.+]] to i8*), i8* [[BITCAST]], i32 0)
268 // CHECK: br label %[[CONT:.+]]
269 // CHECK: [[CONT]]
270 // CHECK: [[LD_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[EXPECTED_ADDR]], i32 0, i32 0
271 // CHECK: [[LD_RE:%.+]] = load i32, i32* [[LD_RE_ADDR]]
272 // CHECK: [[LD_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[EXPECTED_ADDR]], i32 0, i32 1
273 // CHECK: [[LD_IM:%.+]] = load i32, i32* [[LD_IM_ADDR]]
274 // <Skip checks for complex calculations>
275 // CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[DESIRED_ADDR:%.+]], i32 0, i32 0
276 // CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[DESIRED_ADDR]], i32 0, i32 1
277 // CHECK: store i32 [[NEW_RE:%.+]], i32* [[X_RE_ADDR]]
278 // CHECK: store i32 [[NEW_IM:%.+]], i32* [[X_IM_ADDR]]
279 // CHECK: [[EXPECTED:%.+]] = bitcast { i32, i32 }* [[EXPECTED_ADDR]] to i8*
280 // CHECK: [[DESIRED:%.+]] = bitcast { i32, i32 }* [[DESIRED_ADDR]] to i8*
281 // CHECK: [[SUCCESS_FAIL:%.+]] = call zeroext i1 @__atomic_compare_exchange(i64 8, i8* bitcast ({ i32, i32 }* [[X_ADDR]] to i8*), i8* [[EXPECTED]], i8* [[DESIRED]], i32 0, i32 0)
282 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
283 // CHECK: [[EXIT]]
284 // CHECK: [[RE_CAST:%.+]] = sitofp i32 [[NEW_RE]] to float
285 // CHECK: [[IM_CAST:%.+]] = sitofp i32 [[NEW_IM]] to float
286 // CHECK: store float [[RE_CAST]], float* getelementptr inbounds ({ float, float }, { float, float }* @{{.+}}, i32 0, i32 0),
287 // CHECK: store float [[IM_CAST]], float* getelementptr inbounds ({ float, float }, { float, float }* @{{.+}}, i32 0, i32 1),
288 #pragma omp atomic capture
289   cfv = cix = civ / cix;
290 // CHECK: [[EXPR_RE:%.+]] = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @{{.+}}, i32 0, i32 0)
291 // CHECK: [[EXPR_IM:%.+]] = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @{{.+}}, i32 0, i32 1)
292 // CHECK: [[BITCAST:%.+]] = bitcast { float, float }* [[EXPECTED_ADDR:%.+]] to i8*
293 // CHECK: call void @__atomic_load(i64 8, i8* bitcast ({ float, float }* [[X_ADDR:@.+]] to i8*), i8* [[BITCAST]], i32 0)
294 // CHECK: br label %[[CONT:.+]]
295 // CHECK: [[CONT]]
296 // CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[EXPECTED_ADDR]], i32 0, i32 0
297 // CHECK: [[X_RE_OLD:%.+]] = load float, float* [[X_RE_ADDR]]
298 // CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[EXPECTED_ADDR]], i32 0, i32 1
299 // CHECK: [[X_IM_OLD:%.+]] = load float, float* [[X_IM_ADDR]]
300 // <Skip checks for complex calculations>
301 // CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[DESIRED_ADDR:%.+]], i32 0, i32 0
302 // CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[DESIRED_ADDR]], i32 0, i32 1
303 // CHECK: store float [[NEW_RE:%.+]], float* [[X_RE_ADDR]]
304 // CHECK: store float [[NEW_IM:%.+]], float* [[X_IM_ADDR]]
305 // CHECK: [[EXPECTED:%.+]] = bitcast { float, float }* [[EXPECTED_ADDR]] to i8*
306 // CHECK: [[DESIRED:%.+]] = bitcast { float, float }* [[DESIRED_ADDR]] to i8*
307 // CHECK: [[SUCCESS_FAIL:%.+]] = call zeroext i1 @__atomic_compare_exchange(i64 8, i8* bitcast ({ float, float }* [[X_ADDR]] to i8*), i8* [[EXPECTED]], i8* [[DESIRED]], i32 0, i32 0)
308 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
309 // CHECK: [[EXIT]]
310 // CHECK: [[RE_CAST:%.+]] = fptosi float [[X_RE_OLD]] to i32
311 // CHECK: [[IM_CAST:%.+]] = fptosi float [[X_IM_OLD]] to i32
312 // CHECK: store i32 [[RE_CAST]], i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.+}}, i32 0, i32 0),
313 // CHECK: store i32 [[IM_CAST]], i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.+}}, i32 0, i32 1),
314 #pragma omp atomic capture
315   {civ = cfx; cfx = cfv + cfx;}
316 // CHECK: [[EXPR_RE:%.+]] = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @{{.+}}, i32 0, i32 0)
317 // CHECK: [[EXPR_IM:%.+]] = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @{{.+}}, i32 0, i32 1)
318 // CHECK: [[BITCAST:%.+]] = bitcast { double, double }* [[EXPECTED_ADDR:%.+]] to i8*
319 // CHECK: call void @__atomic_load(i64 16, i8* bitcast ({ double, double }* [[X_ADDR:@.+]] to i8*), i8* [[BITCAST]], i32 5)
320 // CHECK: br label %[[CONT:.+]]
321 // CHECK: [[CONT]]
322 // CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[EXPECTED_ADDR]], i32 0, i32 0
323 // CHECK: [[X_RE:%.+]] = load double, double* [[X_RE_ADDR]]
324 // CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[EXPECTED_ADDR]], i32 0, i32 1
325 // CHECK: [[X_IM:%.+]] = load double, double* [[X_IM_ADDR]]
326 // <Skip checks for complex calculations>
327 // CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[DESIRED_ADDR:%.+]], i32 0, i32 0
328 // CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[DESIRED_ADDR]], i32 0, i32 1
329 // CHECK: store double [[NEW_RE:%.+]], double* [[X_RE_ADDR]]
330 // CHECK: store double [[NEW_IM:%.+]], double* [[X_IM_ADDR]]
331 // CHECK: [[EXPECTED:%.+]] = bitcast { double, double }* [[EXPECTED_ADDR]] to i8*
332 // CHECK: [[DESIRED:%.+]] = bitcast { double, double }* [[DESIRED_ADDR]] to i8*
333 // CHECK: [[SUCCESS_FAIL:%.+]] = call zeroext i1 @__atomic_compare_exchange(i64 16, i8* bitcast ({ double, double }* [[X_ADDR]] to i8*), i8* [[EXPECTED]], i8* [[DESIRED]], i32 5, i32 5)
334 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
335 // CHECK: [[EXIT]]
336 // CHECK: [[RE_CAST:%.+]] = fptrunc double [[NEW_RE]] to float
337 // CHECK: [[IM_CAST:%.+]] = fptrunc double [[NEW_IM]] to float
338 // CHECK: store float [[RE_CAST]], float* getelementptr inbounds ({ float, float }, { float, float }* @{{.+}}, i32 0, i32 0),
339 // CHECK: store float [[IM_CAST]], float* getelementptr inbounds ({ float, float }, { float, float }* @{{.+}}, i32 0, i32 1),
340 // CHECK: call{{.*}} @__kmpc_flush(
341 #pragma omp atomic capture seq_cst
342   {cdx = cdx - cdv; cfv = cdx;}
343 // CHECK: [[BV:%.+]] = load i8, i8* @{{.+}}
344 // CHECK: [[BOOL:%.+]] = trunc i8 [[BV]] to i1
345 // CHECK: [[EXPR:%.+]] = zext i1 [[BOOL]] to i64
346 // CHECK: [[OLD:%.+]] = atomicrmw and i64* @{{.+}}, i64 [[EXPR]] monotonic
347 // CHECK: [[DESIRED:%.+]] = and i64 [[OLD]], [[EXPR]]
348 // CHECK: store i64 [[DESIRED]], i64* @{{.+}},
349 #pragma omp atomic capture
350   ulv = ulx = ulx & bv;
351 // CHECK: [[CV:%.+]]  = load i8, i8* @{{.+}}, align 1
352 // CHECK: [[EXPR:%.+]] = sext i8 [[CV]] to i32
353 // CHECK: [[X:%.+]] = load atomic i8, i8* [[BX_ADDR:@.+]] monotonic
354 // CHECK: br label %[[CONT:.+]]
355 // CHECK: [[CONT]]
356 // CHECK: [[EXPECTED:%.+]] = phi i8 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
357 // CHECK: [[OLD_BOOL:%.+]] = trunc i8 [[EXPECTED]] to i1
358 // CHECK: [[X_RVAL:%.+]] = zext i1 [[OLD_BOOL]] to i32
359 // CHECK: [[AND:%.+]] = and i32 [[EXPR]], [[X_RVAL]]
360 // CHECK: [[CAST:%.+]] = icmp ne i32 [[AND]], 0
361 // CHECK: [[NEW:%.+]] = zext i1 [[CAST]] to i8
362 // CHECK: store i8 [[NEW]], i8* [[TEMP:%.+]],
363 // CHECK: [[DESIRED:%.+]] = load i8, i8* [[TEMP]],
364 // CHECK: [[RES:%.+]] = cmpxchg i8* [[BX_ADDR]], i8 [[EXPECTED]], i8 [[DESIRED]] monotonic monotonic
365 // CHECK: [[OLD:%.+]] = extractvalue { i8, i1 } [[RES]], 0
366 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i8, i1 } [[RES]], 1
367 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
368 // CHECK: [[EXIT]]
369 // CHECK: [[OLD_I8:%.+]] = zext i1 [[OLD_BOOL]] to i8
370 // CHECK: store i8 [[OLD_I8]], i8* @{{.+}},
371 #pragma omp atomic capture
372   {bv = bx; bx = cv & bx;}
373 // CHECK: [[UCV:%.+]]  = load i8, i8* @{{.+}},
374 // CHECK: [[EXPR:%.+]] = zext i8 [[UCV]] to i32
375 // CHECK: [[X:%.+]] = load atomic i8, i8* [[CX_ADDR:@.+]] seq_cst
376 // CHECK: br label %[[CONT:.+]]
377 // CHECK: [[CONT]]
378 // CHECK: [[EXPECTED:%.+]] = phi i8 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
379 // CHECK: [[X_RVAL:%.+]] = sext i8 [[EXPECTED]] to i32
380 // CHECK: [[ASHR:%.+]] = ashr i32 [[X_RVAL]], [[EXPR]]
381 // CHECK: [[NEW:%.+]] = trunc i32 [[ASHR]] to i8
382 // CHECK: store i8 [[NEW]], i8* [[TEMP:%.+]],
383 // CHECK: [[DESIRED:%.+]] = load i8, i8* [[TEMP]],
384 // CHECK: [[RES:%.+]] = cmpxchg i8* [[CX_ADDR]], i8 [[EXPECTED]], i8 [[DESIRED]] seq_cst seq_cst
385 // CHECK: [[OLD_X:%.+]] = extractvalue { i8, i1 } [[RES]], 0
386 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i8, i1 } [[RES]], 1
387 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
388 // CHECK: [[EXIT]]
389 // CHECK: store i8 [[NEW]], i8* @{{.+}},
390 // CHECK: call{{.*}} @__kmpc_flush(
391 #pragma omp atomic capture, seq_cst
392   {cx = cx >> ucv; cv = cx;}
393 // CHECK: [[SV:%.+]]  = load i16, i16* @{{.+}},
394 // CHECK: [[EXPR:%.+]] = sext i16 [[SV]] to i32
395 // CHECK: [[X:%.+]] = load atomic i64, i64* [[ULX_ADDR:@.+]] monotonic
396 // CHECK: br label %[[CONT:.+]]
397 // CHECK: [[CONT]]
398 // CHECK: [[EXPECTED:%.+]] = phi i64 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
399 // CHECK: [[X_RVAL:%.+]] = trunc i64 [[EXPECTED]] to i32
400 // CHECK: [[SHL:%.+]] = shl i32 [[EXPR]], [[X_RVAL]]
401 // CHECK: [[NEW:%.+]] = sext i32 [[SHL]] to i64
402 // CHECK: store i64 [[NEW]], i64* [[TEMP:%.+]],
403 // CHECK: [[DESIRED:%.+]] = load i64, i64* [[TEMP]],
404 // CHECK: [[RES:%.+]] = cmpxchg i64* [[ULX_ADDR]], i64 [[EXPECTED]], i64 [[DESIRED]] monotonic monotonic
405 // CHECK: [[OLD_X:%.+]] = extractvalue { i64, i1 } [[RES]], 0
406 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i64, i1 } [[RES]], 1
407 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
408 // CHECK: [[EXIT]]
409 // CHECK: store i64 [[NEW]], i64* @{{.+}},
410 #pragma omp atomic capture
411   ulv = ulx = sv << ulx;
412 // CHECK: [[USV:%.+]]  = load i16, i16* @{{.+}},
413 // CHECK: [[EXPR:%.+]] = zext i16 [[USV]] to i64
414 // CHECK: [[X:%.+]] = load atomic i64, i64* [[LX_ADDR:@.+]] monotonic
415 // CHECK: br label %[[CONT:.+]]
416 // CHECK: [[CONT]]
417 // CHECK: [[EXPECTED:%.+]] = phi i64 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
418 // CHECK: [[DESIRED:%.+]] = srem i64 [[EXPECTED]], [[EXPR]]
419 // CHECK: store i64 [[DESIRED]], i64* [[TEMP:%.+]],
420 // CHECK: [[DESIRED:%.+]] = load i64, i64* [[TEMP]],
421 // CHECK: [[RES:%.+]] = cmpxchg i64* [[LX_ADDR]], i64 [[EXPECTED]], i64 [[DESIRED]] monotonic monotonic
422 // CHECK: [[OLD_X:%.+]] = extractvalue { i64, i1 } [[RES]], 0
423 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i64, i1 } [[RES]], 1
424 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
425 // CHECK: [[EXIT]]
426 // CHECK: store i64 [[EXPECTED]], i64* @{{.+}},
427 #pragma omp atomic capture
428   {lv = lx; lx = lx % usv;}
429 // CHECK: [[EXPR:%.+]] = load i32, i32* @{{.+}}
430 // CHECK: [[OLD:%.+]] = atomicrmw or i32* @{{.+}}, i32 [[EXPR]] seq_cst
431 // CHECK: [[DESIRED:%.+]] = or i32 [[EXPR]], [[OLD]]
432 // CHECK: store i32 [[DESIRED]], i32* @{{.+}},
433 // CHECK: call{{.*}} @__kmpc_flush(
434 #pragma omp atomic seq_cst, capture
435   {uix = iv | uix; uiv = uix;}
436 // CHECK: [[EXPR:%.+]] = load i32, i32* @{{.+}}
437 // CHECK: [[OLD:%.+]] = atomicrmw and i32* @{{.+}}, i32 [[EXPR]] monotonic
438 // CHECK: [[DESIRED:%.+]] = and i32 [[OLD]], [[EXPR]]
439 // CHECK: store i32 [[DESIRED]], i32* @{{.+}},
440 #pragma omp atomic capture
441   iv = ix = ix & uiv;
442 // CHECK: [[EXPR:%.+]] = load i64, i64* @{{.+}},
443 // CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[EXPECTED_ADDR:%.+]] to i8*
444 // CHECK: call void @__atomic_load(i64 8, i8* bitcast ({ i32, i32 }* [[X_ADDR:@.+]] to i8*), i8* [[BITCAST]], i32 0)
445 // CHECK: br label %[[CONT:.+]]
446 // CHECK: [[CONT]]
447 // CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[EXPECTED_ADDR]], i32 0, i32 0
448 // CHECK: [[OLD_RE:%.+]] = load i32, i32* [[X_RE_ADDR]]
449 // CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[EXPECTED_ADDR]], i32 0, i32 1
450 // CHECK: [[OLD_IM:%.+]] = load i32, i32* [[X_IM_ADDR]]
451 // <Skip checks for complex calculations>
452 // CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[DESIRED_ADDR:%.+]], i32 0, i32 0
453 // CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[DESIRED_ADDR]], i32 0, i32 1
454 // CHECK: store i32 %{{.+}}, i32* [[X_RE_ADDR]]
455 // CHECK: store i32 %{{.+}}, i32* [[X_IM_ADDR]]
456 // CHECK: [[EXPECTED:%.+]] = bitcast { i32, i32 }* [[EXPECTED_ADDR]] to i8*
457 // CHECK: [[DESIRED:%.+]] = bitcast { i32, i32 }* [[DESIRED_ADDR]] to i8*
458 // CHECK: [[SUCCESS_FAIL:%.+]] = call zeroext i1 @__atomic_compare_exchange(i64 8, i8* bitcast ({ i32, i32 }* [[X_ADDR]] to i8*), i8* [[EXPECTED]], i8* [[DESIRED]], i32 0, i32 0)
459 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
460 // CHECK: [[EXIT]]
461 // CHECK: store i32 [[OLD_RE]], i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.+}}, i32 0, i32 0),
462 // CHECK: store i32 [[OLD_IM]], i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.+}}, i32 0, i32 1),
463 #pragma omp atomic capture
464   {civ = cix; cix = lv + cix;}
465 // CHECK: [[ULV:%.+]] = load i64, i64* @{{.+}},
466 // CHECK: [[EXPR:%.+]] = uitofp i64 [[ULV]] to float
467 // CHECK: [[X:%.+]] = load atomic i32, i32*  bitcast (float* [[X_ADDR:@.+]] to i32*) monotonic
468 // CHECK: br label %[[CONT:.+]]
469 // CHECK: [[CONT]]
470 // CHECK: [[EXPECTED:%.+]] = phi i32 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
471 // CHECK: [[TEMP_I:%.+]] = bitcast float* [[TEMP:%.+]] to i32*
472 // CHECK: [[OLD:%.+]] = bitcast i32 [[EXPECTED]] to float
473 // CHECK: [[MUL:%.+]] = fmul float [[OLD]], [[EXPR]]
474 // CHECK: store float [[MUL]], float* [[TEMP]],
475 // CHECK: [[DESIRED:%.+]] = load i32, i32* [[TEMP_I]],
476 // CHECK: [[RES:%.+]] = cmpxchg i32* bitcast (float* [[X_ADDR]] to i32*), i32 [[EXPECTED]], i32 [[DESIRED]] monotonic monotonic
477 // CHECK: [[OLD_X:%.+]] = extractvalue { i32, i1 } [[RES]], 0
478 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i32, i1 } [[RES]], 1
479 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
480 // CHECK: [[EXIT]]
481 // CHECK: store float [[MUL]], float* @{{.+}},
482 #pragma omp atomic capture
483   {fx = fx * ulv; fv = fx;}
484 // CHECK: [[LLV:%.+]] = load i64, i64* @{{.+}},
485 // CHECK: [[EXPR:%.+]] = sitofp i64 [[LLV]] to double
486 // CHECK: [[X:%.+]] = load atomic i64, i64*  bitcast (double* [[X_ADDR:@.+]] to i64*) monotonic
487 // CHECK: br label %[[CONT:.+]]
488 // CHECK: [[CONT]]
489 // CHECK: [[EXPECTED:%.+]] = phi i64 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
490 // CHECK: [[TEMP_I:%.+]] = bitcast double* [[TEMP:%.+]] to i64*
491 // CHECK: [[OLD:%.+]] = bitcast i64 [[EXPECTED]] to double
492 // CHECK: [[DIV:%.+]] = fdiv double [[OLD]], [[EXPR]]
493 // CHECK: store double [[DIV]], double* [[TEMP]],
494 // CHECK: [[DESIRED:%.+]] = load i64, i64* [[TEMP_I]],
495 // CHECK: [[RES:%.+]] = cmpxchg i64* bitcast (double* [[X_ADDR]] to i64*), i64 [[EXPECTED]], i64 [[DESIRED]] monotonic monotonic
496 // CHECK: [[OLD_X:%.+]] = extractvalue { i64, i1 } [[RES]], 0
497 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i64, i1 } [[RES]], 1
498 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
499 // CHECK: [[EXIT]]
500 // CHECK: store double [[DIV]], double* @{{.+}},
501 #pragma omp atomic capture
502   dv = dx /= llv;
503 // CHECK: [[ULLV:%.+]] = load i64, i64* @{{.+}},
504 // CHECK: [[EXPR:%.+]] = uitofp i64 [[ULLV]] to x86_fp80
505 // CHECK: [[X:%.+]] = load atomic i128, i128*  bitcast (x86_fp80* [[X_ADDR:@.+]] to i128*) monotonic
506 // CHECK: br label %[[CONT:.+]]
507 // CHECK: [[CONT]]
508 // CHECK: [[EXPECTED:%.+]] = phi i128 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
509 // CHECK: [[TEMP_I1:%.+]] = bitcast x86_fp80* [[TEMP1:%.+]] to i128*
510 // CHECK: store i128 [[EXPECTED]], i128* [[TEMP_I1]],
511 // CHECK: [[TEMP_I:%.+]] = bitcast x86_fp80* [[TEMP:%.+]] to i128*
512 // CHECK: store i128 [[EXPECTED]], i128* [[TEMP_I]],
513 // CHECK: [[OLD:%.+]] = load x86_fp80, x86_fp80* [[TEMP]],
514 // CHECK: [[SUB:%.+]] = fsub x86_fp80 [[OLD]], [[EXPR]]
515 // CHECK: store x86_fp80 [[SUB]], x86_fp80* [[TEMP1]]
516 // CHECK: [[DESIRED:%.+]] = load i128, i128* [[TEMP_I1]]
517 // CHECK: [[RES:%.+]] = cmpxchg i128* bitcast (x86_fp80* [[X_ADDR]] to i128*), i128 [[EXPECTED]], i128 [[DESIRED]] monotonic monotonic
518 // CHECK: [[OLD_X:%.+]] = extractvalue { i128, i1 } [[RES]], 0
519 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i128, i1 } [[RES]], 1
520 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
521 // CHECK: [[EXIT]]
522 // CHECK: store x86_fp80 [[OLD]], x86_fp80* @{{.+}},
523 #pragma omp atomic capture
524   {ldv = ldx; ldx -= ullv;}
525 // CHECK: [[EXPR:%.+]] = load float, float* @{{.+}},
526 // CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[EXPECTED_ADDR:%.+]] to i8*
527 // CHECK: call void @__atomic_load(i64 8, i8* bitcast ({ i32, i32 }* [[X_ADDR:@.+]] to i8*), i8* [[BITCAST]], i32 0)
528 // CHECK: br label %[[CONT:.+]]
529 // CHECK: [[CONT]]
530 // CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[EXPECTED_ADDR]], i32 0, i32 0
531 // CHECK: [[X_RE:%.+]] = load i32, i32* [[X_RE_ADDR]]
532 // CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[EXPECTED_ADDR]], i32 0, i32 1
533 // CHECK: [[X_IM:%.+]] = load i32, i32* [[X_IM_ADDR]]
534 // <Skip checks for complex calculations>
535 // CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[DESIRED_ADDR:%.+]], i32 0, i32 0
536 // CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[DESIRED_ADDR]], i32 0, i32 1
537 // CHECK: store i32 [[NEW_RE:%.+]], i32* [[X_RE_ADDR]]
538 // CHECK: store i32 [[NEW_IM:%.+]], i32* [[X_IM_ADDR]]
539 // CHECK: [[EXPECTED:%.+]] = bitcast { i32, i32 }* [[EXPECTED_ADDR]] to i8*
540 // CHECK: [[DESIRED:%.+]] = bitcast { i32, i32 }* [[DESIRED_ADDR]] to i8*
541 // CHECK: [[SUCCESS_FAIL:%.+]] = call zeroext i1 @__atomic_compare_exchange(i64 8, i8* bitcast ({ i32, i32 }* [[X_ADDR]] to i8*), i8* [[EXPECTED]], i8* [[DESIRED]], i32 0, i32 0)
542 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
543 // CHECK: [[EXIT]]
544 // CHECK: store i32 [[NEW_RE]], i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.+}}, i32 0, i32 0),
545 // CHECK: store i32 [[NEW_IM]], i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.+}}, i32 0, i32 1),
546 #pragma omp atomic capture
547   {cix = fv / cix; civ = cix;}
548 // CHECK: [[EXPR:%.+]] = load double, double* @{{.+}},
549 // CHECK: [[X:%.+]] = load atomic i16, i16* [[X_ADDR:@.+]] monotonic
550 // CHECK: br label %[[CONT:.+]]
551 // CHECK: [[CONT]]
552 // CHECK: [[EXPECTED:%.+]] = phi i16 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
553 // CHECK: [[CONV:%.+]] = sext i16 [[EXPECTED]] to i32
554 // CHECK: [[X_RVAL:%.+]] = sitofp i32 [[CONV]] to double
555 // CHECK: [[ADD:%.+]] = fadd double [[X_RVAL]], [[EXPR]]
556 // CHECK: [[NEW:%.+]] = fptosi double [[ADD]] to i16
557 // CHECK: store i16 [[NEW]], i16* [[TEMP:%.+]],
558 // CHECK: [[DESIRED:%.+]] = load i16, i16* [[TEMP]],
559 // CHECK: [[RES:%.+]] = cmpxchg i16* [[X_ADDR]], i16 [[EXPECTED]], i16 [[DESIRED]] monotonic monotonic
560 // CHECK: [[OLD_X]] = extractvalue { i16, i1 } [[RES]], 0
561 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i16, i1 } [[RES]], 1
562 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
563 // CHECK: [[EXIT]]
564 // CHECK: store i16 [[NEW]], i16* @{{.+}},
565 #pragma omp atomic capture
566   sv = sx = sx + dv;
567 // CHECK: [[EXPR:%.+]] = load x86_fp80, x86_fp80* @{{.+}},
568 // CHECK: [[XI8:%.+]] = load atomic i8, i8* [[X_ADDR:@.+]] monotonic
569 // CHECK: br label %[[CONT:.+]]
570 // CHECK: [[CONT]]
571 // CHECK: [[EXPECTED:%.+]] = phi i8 [ [[XI8]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
572 // CHECK: [[BOOL_EXPECTED:%.+]] = trunc i8 [[EXPECTED]] to i1
573 // CHECK: [[CONV:%.+]] = zext i1 [[BOOL_EXPECTED]] to i32
574 // CHECK: [[X_RVAL:%.+]] = sitofp i32 [[CONV]] to x86_fp80
575 // CHECK: [[MUL:%.+]] = fmul x86_fp80 [[EXPR]], [[X_RVAL]]
576 // CHECK: [[BOOL_DESIRED:%.+]] = fcmp une x86_fp80 [[MUL]], 0xK00000000000000000000
577 // CHECK: [[DESIRED:%.+]] = zext i1 [[BOOL_DESIRED]] to i8
578 // CHECK: store i8 [[DESIRED]], i8* [[TEMP:%.+]],
579 // CHECK: [[DESIRED:%.+]] = load i8, i8* [[TEMP]],
580 // CHECK: [[RES:%.+]] = cmpxchg i8* [[X_ADDR]], i8 [[EXPECTED]], i8 [[DESIRED]] monotonic monotonic
581 // CHECK: [[OLD_X:%.+]] = extractvalue { i8, i1 } [[RES]], 0
582 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i8, i1 } [[RES]], 1
583 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
584 // CHECK: [[EXIT]]
585 // CHECK: [[EXPECTED_I8:%.+]] = zext i1 [[BOOL_EXPECTED]] to i8
586 // CHECK: store i8 [[EXPECTED_I8]], i8* @{{.+}},
587 #pragma omp atomic capture
588   {bv = bx; bx = ldv * bx;}
589 // CHECK: [[EXPR_RE:%.+]] = load i32, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* [[CIV_ADDR:@.+]], i32 0, i32 0),
590 // CHECK: [[EXPR_IM:%.+]] = load i32, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* [[CIV_ADDR]], i32 0, i32 1),
591 // CHECK: [[XI8:%.+]] = load atomic i8, i8* [[X_ADDR:@.+]] monotonic
592 // CHECK: br label %[[CONT:.+]]
593 // CHECK: [[CONT]]
594 // CHECK: [[EXPECTED:%.+]] = phi i8 [ [[XI8]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
595 // CHECK: [[BOOL_EXPECTED:%.+]] = trunc i8 [[EXPECTED]] to i1
596 // CHECK: [[X_RVAL:%.+]] = zext i1 [[BOOL_EXPECTED]] to i32
597 // CHECK: [[SUB_RE:%.+]] = sub i32 [[EXPR_RE:%.+]], [[X_RVAL]]
598 // CHECK: [[SUB_IM:%.+]] = sub i32 [[EXPR_IM:%.+]], 0
599 // CHECK: icmp ne i32 [[SUB_RE]], 0
600 // CHECK: icmp ne i32 [[SUB_IM]], 0
601 // CHECK: [[BOOL_DESIRED:%.+]] = or i1
602 // CHECK: [[DESIRED:%.+]] = zext i1 [[BOOL_DESIRED]] to i8
603 // CHECK: store i8 [[DESIRED]], i8* [[TEMP:%.+]],
604 // CHECK: [[DESIRED:%.+]] = load i8, i8* [[TEMP]],
605 // CHECK: [[RES:%.+]] = cmpxchg i8* [[X_ADDR]], i8 [[EXPECTED]], i8 [[DESIRED]] monotonic monotonic
606 // CHECK: [[OLD_X:%.+]] = extractvalue { i8, i1 } [[RES]], 0
607 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i8, i1 } [[RES]], 1
608 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
609 // CHECK: [[EXIT]]
610 // CHECK: [[DESIRED_I8:%.+]] = zext i1 [[BOOL_DESIRED]] to i8
611 // CHECK: store i8 [[DESIRED_I8]], i8* @{{.+}},
612 #pragma omp atomic capture
613   {bx = civ - bx; bv = bx;}
614 // CHECK: [[EXPR_RE:%.+]] = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @{{.+}}, i32 0, i32 0)
615 // CHECK: [[EXPR_IM:%.+]] = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @{{.+}}, i32 0, i32 1)
616 // CHECK: [[X:%.+]] = load atomic i16, i16* [[X_ADDR:@.+]] monotonic
617 // CHECK: br label %[[CONT:.+]]
618 // CHECK: [[CONT]]
619 // CHECK: [[EXPECTED:%.+]] = phi i16 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
620 // CHECK: [[CONV:%.+]] = zext i16 [[EXPECTED]] to i32
621 // CHECK: [[X_RVAL:%.+]] = sitofp i32 [[CONV]] to float
622 // <Skip checks for complex calculations>
623 // CHECK: [[X_RE_ADDR:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[TEMP:%.+]], i32 0, i32 0
624 // CHECK: [[X_RE:%.+]] = load float, float* [[X_RE_ADDR]]
625 // CHECK: [[X_IM_ADDR:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[TEMP]], i32 0, i32 1
626 // CHECK: [[X_IM:%.+]] = load float, float* [[X_IM_ADDR]]
627 // CHECK: [[NEW:%.+]] = fptoui float [[X_RE]] to i16
628 // CHECK: store i16 [[NEW]], i16* [[TEMP:%.+]],
629 // CHECK: [[DESIRED:%.+]] = load i16, i16* [[TEMP]],
630 // CHECK: [[RES:%.+]] = cmpxchg i16* [[X_ADDR]], i16 [[EXPECTED]], i16 [[DESIRED]] monotonic monotonic
631 // CHECK: [[OLD_X]] = extractvalue { i16, i1 } [[RES]], 0
632 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i16, i1 } [[RES]], 1
633 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
634 // CHECK: [[EXIT]]
635 // CHECK: store i16 [[NEW]], i16* @{{.+}},
636 #pragma omp atomic capture
637   usv = usx /= cfv;
638 // CHECK: [[EXPR_RE:%.+]] = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @{{.+}}, i32 0, i32 0)
639 // CHECK: [[EXPR_IM:%.+]] = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @{{.+}}, i32 0, i32 1)
640 // CHECK: [[X:%.+]] = load atomic i64, i64* [[X_ADDR:@.+]] monotonic
641 // CHECK: br label %[[CONT:.+]]
642 // CHECK: [[CONT]]
643 // CHECK: [[EXPECTED:%.+]] = phi i64 [ [[X]], %{{.+}} ], [ [[OLD_X:%.+]], %[[CONT]] ]
644 // CHECK: [[X_RVAL:%.+]] = sitofp i64 [[EXPECTED]] to double
645 // CHECK: [[ADD_RE:%.+]] = fadd double [[X_RVAL]], [[EXPR_RE]]
646 // CHECK: [[ADD_IM:%.+]] = fadd double 0.000000e+00, [[EXPR_IM]]
647 // CHECK: [[DESIRED:%.+]] = fptosi double [[ADD_RE]] to i64
648 // CHECK: store i64 [[DESIRED]], i64* [[TEMP:%.+]],
649 // CHECK: [[DESIRED:%.+]] = load i64, i64* [[TEMP]],
650 // CHECK: [[RES:%.+]] = cmpxchg i64* [[X_ADDR]], i64 [[EXPECTED]], i64 [[DESIRED]] monotonic monotonic
651 // CHECK: [[OLD_X]] = extractvalue { i64, i1 } [[RES]], 0
652 // CHECK: [[SUCCESS_FAIL:%.+]] = extractvalue { i64, i1 } [[RES]], 1
653 // CHECK: br i1 [[SUCCESS_FAIL]], label %[[EXIT:.+]], label %[[CONT]]
654 // CHECK: [[EXIT]]
655 // CHECK: store i64 [[EXPECTED]], i64* @{{.+}},
656 #pragma omp atomic capture
657   {llv = llx; llx += cdv;}
658 // CHECK: [[IDX:%.+]] = load i16, i16* @{{.+}}
659 // CHECK: load i8, i8*
660 // CHECK: [[VEC_ITEM_VAL:%.+]] = zext i1 %{{.+}} to i32
661 // CHECK: [[I128VAL:%.+]] = load atomic i128, i128* bitcast (<4 x i32>* [[DEST:@.+]] to i128*) monotonic
662 // CHECK: br label %[[CONT:.+]]
663 // CHECK: [[CONT]]
664 // CHECK: [[OLD_I128:%.+]] = phi i128 [ [[I128VAL]], %{{.+}} ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
665 // CHECK: [[TEMP_I:%.+]] = bitcast <4 x i32>* [[TEMP:%.+]] to i128*
666 // CHECK: store i128 [[OLD_I128]], i128* [[TEMP_I]],
667 // CHECK: [[LD:%.+]] = bitcast i128 [[OLD_I128]] to <4 x i32>
668 // CHECK: store <4 x i32> [[LD]], <4 x i32>* [[TEMP1:%.+]],
669 // CHECK: [[VEC_VAL:%.+]] = load <4 x i32>, <4 x i32>* [[TEMP1]]
670 // CHECK: [[ITEM:%.+]] = extractelement <4 x i32> [[VEC_VAL]], i16 [[IDX]]
671 // CHECK: [[OR:%.+]] = or i32 [[ITEM]], [[VEC_ITEM_VAL]]
672 // CHECK: [[VEC_VAL:%.+]] = load <4 x i32>, <4 x i32>* [[TEMP]]
673 // CHECK: [[NEW_VEC_VAL:%.+]] = insertelement <4 x i32> [[VEC_VAL]], i32 [[OR]], i16 [[IDX]]
674 // CHECK: store <4 x i32> [[NEW_VEC_VAL]], <4 x i32>* [[TEMP]]
675 // CHECK: [[NEW_I128:%.+]] = load i128, i128* [[TEMP_I]],
676 // CHECK: [[RES:%.+]] = cmpxchg i128* bitcast (<4 x i32>* [[DEST]] to i128*), i128 [[OLD_I128]], i128 [[NEW_I128]] monotonic monotonic
677 // CHECK: [[FAILED_OLD_VAL:%.+]] = extractvalue { i128, i1 } [[RES]], 0
678 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i128, i1 } [[RES]], 1
679 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
680 // CHECK: [[EXIT]]
681 // CHECK: store i32 [[OR]], i32* @{{.+}},
682 #pragma omp atomic capture
683   {int4x[sv] |= bv; iv = int4x[sv];}
684 // CHECK: [[EXPR:%.+]] = load x86_fp80, x86_fp80* @{{.+}}
685 // CHECK: [[PREV_VALUE:%.+]] = load atomic i32, i32* bitcast (i8* getelementptr (i8, i8* bitcast (%struct.BitFields* @{{.+}} to i8*), i64 4) to i32*) monotonic
686 // CHECK: br label %[[CONT:.+]]
687 // CHECK: [[CONT]]
688 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i32 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
689 // CHECK: store i32 [[OLD_BF_VALUE]], i32* [[TEMP1:%.+]],
690 // CHECK: store i32 [[OLD_BF_VALUE]], i32* [[TEMP:%.+]],
691 // CHECK: [[A_LD:%.+]] = load i32, i32* [[TEMP]],
692 // CHECK: [[A_SHL:%.+]] = shl i32 [[A_LD]], 1
693 // CHECK: [[A_ASHR:%.+]] = ashr i32 [[A_SHL]], 1
694 // CHECK: [[X_RVAL:%.+]] = sitofp i32 [[A_ASHR]] to x86_fp80
695 // CHECK: [[SUB:%.+]] = fsub x86_fp80 [[X_RVAL]], [[EXPR]]
696 // CHECK: [[CONV:%.+]] = fptosi x86_fp80 [[SUB]] to i32
697 // CHECK: [[NEW_VAL:%.+]] = load i32, i32* [[TEMP1]],
698 // CHECK: [[BF_VALUE:%.+]] = and i32 [[CONV]], 2147483647
699 // CHECK: [[BF_CLEAR:%.+]] = and i32 [[NEW_VAL]], -2147483648
700 // CHECK: [[BF_SET:%.+]] = or i32 [[BF_CLEAR]], [[BF_VALUE]]
701 // CHECK: store i32 [[BF_SET]], i32* [[TEMP1]],
702 // CHECK: [[NEW_BF_VALUE:%.+]] = load i32, i32* [[TEMP1]],
703 // 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
704 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i32, i1 } [[RES]], 0
705 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i32, i1 } [[RES]], 1
706 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
707 // CHECK: [[EXIT]]
708 // CHECK: store i32 [[CONV]], i32* @{{.+}},
709 #pragma omp atomic capture
710   iv = bfx.a = bfx.a - ldv;
711 // CHECK: [[EXPR:%.+]] = load x86_fp80, x86_fp80* @{{.+}}
712 // CHECK: [[BITCAST:%.+]] = bitcast i32* [[LDTEMP:%.+]] to i8*
713 // CHECK: call void @__atomic_load(i64 4, i8* getelementptr (i8, i8* bitcast (%struct.BitFields_packed* @{{.+}} to i8*), i64 4), i8* [[BITCAST]], i32 0)
714 // CHECK: br label %[[CONT:.+]]
715 // CHECK: [[CONT]]
716 // CHECK: [[OLD:%.+]] = load i32, i32* [[LDTEMP]],
717 // CHECK: store i32 [[OLD]], i32* [[TEMP1:%.+]],
718 // CHECK: [[OLD:%.+]] = load i32, i32* [[LDTEMP]],
719 // CHECK: store i32 [[OLD]], i32* [[TEMP:%.+]],
720 // CHECK: [[A_LD:%.+]] = load i32, i32* [[TEMP]],
721 // CHECK: [[A_SHL:%.+]] = shl i32 [[A_LD]], 1
722 // CHECK: [[A_ASHR:%.+]] = ashr i32 [[A_SHL]], 1
723 // CHECK: [[X_RVAL:%.+]] = sitofp i32 [[A_ASHR]] to x86_fp80
724 // CHECK: [[MUL:%.+]] = fmul x86_fp80 [[X_RVAL]], [[EXPR]]
725 // CHECK: [[CONV:%.+]] = fptosi x86_fp80 [[MUL]] to i32
726 // CHECK: [[NEW_VAL:%.+]] = load i32, i32* [[TEMP1]],
727 // CHECK: [[BF_VALUE:%.+]] = and i32 [[CONV]], 2147483647
728 // CHECK: [[BF_CLEAR:%.+]] = and i32 [[NEW_VAL]], -2147483648
729 // CHECK: or i32 [[BF_CLEAR]], [[BF_VALUE]]
730 // CHECK: store i32 %{{.+}}, i32* [[TEMP1]]
731 // CHECK: [[BITCAST_TEMP_OLD_BF_ADDR:%.+]] = bitcast i32* [[LDTEMP]] to i8*
732 // CHECK: [[BITCAST_TEMP_NEW_BF_ADDR:%.+]] = bitcast i32* [[TEMP1]] to i8*
733 // 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)
734 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
735 // CHECK: [[EXIT]]
736 // CHECK: store i32 [[A_ASHR]], i32* @{{.+}},
737 #pragma omp atomic capture
738   {iv = bfx_packed.a; bfx_packed.a *= ldv;}
739 // CHECK: [[EXPR:%.+]] = load x86_fp80, x86_fp80* @{{.+}}
740 // CHECK: [[PREV_VALUE:%.+]] = load atomic i32, i32* getelementptr inbounds (%struct.BitFields2, %struct.BitFields2* @{{.+}}, i32 0, i32 0) monotonic
741 // CHECK: br label %[[CONT:.+]]
742 // CHECK: [[CONT]]
743 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i32 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
744 // CHECK: store i32 [[OLD_BF_VALUE]], i32* [[TEMP1:%.+]],
745 // CHECK: store i32 [[OLD_BF_VALUE]], i32* [[TEMP:%.+]],
746 // CHECK: [[A_LD:%.+]] = load i32, i32* [[TEMP]],
747 // CHECK: [[A_ASHR:%.+]] = ashr i32 [[A_LD]], 31
748 // CHECK: [[X_RVAL:%.+]] = sitofp i32 [[A_ASHR]] to x86_fp80
749 // CHECK: [[SUB:%.+]] = fsub x86_fp80 [[X_RVAL]], [[EXPR]]
750 // CHECK: [[CONV:%.+]] = fptosi x86_fp80 [[SUB]] to i32
751 // CHECK: [[NEW_VAL:%.+]] = load i32, i32* [[TEMP1]],
752 // CHECK: [[BF_AND:%.+]] = and i32 [[CONV]], 1
753 // CHECK: [[BF_VALUE:%.+]] = shl i32 [[BF_AND]], 31
754 // CHECK: [[BF_CLEAR:%.+]] = and i32 [[NEW_VAL]], 2147483647
755 // CHECK: or i32 [[BF_CLEAR]], [[BF_VALUE]]
756 // CHECK: store i32 %{{.+}}, i32* [[TEMP1]]
757 // CHECK: [[NEW_BF_VALUE:%.+]] = load i32, i32* [[TEMP1]]
758 // CHECK: [[RES:%.+]] = cmpxchg i32* getelementptr inbounds (%struct.BitFields2, %struct.BitFields2* @{{.+}}, i32 0, i32 0), i32 [[OLD_BF_VALUE]], i32 [[NEW_BF_VALUE]] monotonic monotonic
759 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i32, i1 } [[RES]], 0
760 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i32, i1 } [[RES]], 1
761 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
762 // CHECK: [[EXIT]]
763 // CHECK: store i32 [[CONV]], i32* @{{.+}},
764 #pragma omp atomic capture
765   {bfx2.a -= ldv; iv = bfx2.a;}
766 // CHECK: [[EXPR:%.+]] = load x86_fp80, x86_fp80* @{{.+}}
767 // CHECK: [[PREV_VALUE:%.+]] = load atomic i8, i8* getelementptr (i8, i8* bitcast (%struct.BitFields2_packed* @{{.+}} to i8*), i64 3) monotonic
768 // CHECK: br label %[[CONT:.+]]
769 // CHECK: [[CONT]]
770 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i8 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
771 // CHECK: [[BITCAST_NEW:%.+]] = bitcast i32* %{{.+}} to i8*
772 // CHECK: store i8 [[OLD_BF_VALUE]], i8* [[BITCAST_NEW]],
773 // CHECK: [[BITCAST:%.+]] = bitcast i32* %{{.+}} to i8*
774 // CHECK: store i8 [[OLD_BF_VALUE]], i8* [[BITCAST]],
775 // CHECK: [[A_LD:%.+]] = load i8, i8* [[BITCAST]],
776 // CHECK: [[A_ASHR:%.+]] = ashr i8 [[A_LD]], 7
777 // CHECK: [[CAST:%.+]] = sext i8 [[A_ASHR]] to i32
778 // CHECK: [[X_RVAL:%.+]] = sitofp i32 [[CAST]] to x86_fp80
779 // CHECK: [[DIV:%.+]] = fdiv x86_fp80 [[EXPR]], [[X_RVAL]]
780 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 [[DIV]] to i32
781 // CHECK: [[TRUNC:%.+]] = trunc i32 [[NEW_VAL]] to i8
782 // CHECK: [[BF_LD:%.+]] = load i8, i8* [[BITCAST_NEW]],
783 // CHECK: [[BF_AND:%.+]] = and i8 [[TRUNC]], 1
784 // CHECK: [[BF_VALUE:%.+]] = shl i8 [[BF_AND]], 7
785 // CHECK: [[BF_CLEAR:%.+]] = and i8 %{{.+}}, 127
786 // CHECK: or i8 [[BF_CLEAR]], [[BF_VALUE]]
787 // CHECK: store i8 %{{.+}}, i8* [[BITCAST_NEW]]
788 // CHECK: [[NEW_BF_VALUE:%.+]] = load i8, i8* [[BITCAST_NEW]]
789 // 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
790 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i8, i1 } [[RES]], 0
791 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i8, i1 } [[RES]], 1
792 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
793 // CHECK: [[EXIT]]
794 // CHECK: store i32 [[NEW_VAL]], i32* @{{.+}},
795 #pragma omp atomic capture
796   iv = bfx2_packed.a = ldv / bfx2_packed.a;
797 // CHECK: [[EXPR:%.+]] = load x86_fp80, x86_fp80* @{{.+}}
798 // CHECK: [[PREV_VALUE:%.+]] = load atomic i32, i32* getelementptr inbounds (%struct.BitFields3, %struct.BitFields3* @{{.+}}, i32 0, i32 0) monotonic
799 // CHECK: br label %[[CONT:.+]]
800 // CHECK: [[CONT]]
801 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i32 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
802 // CHECK: store i32 [[OLD_BF_VALUE]], i32* [[TEMP1:%.+]],
803 // CHECK: store i32 [[OLD_BF_VALUE]], i32* [[TEMP:%.+]],
804 // CHECK: [[A_LD:%.+]] = load i32, i32* [[TEMP]],
805 // CHECK: [[A_SHL:%.+]] = shl i32 [[A_LD]], 7
806 // CHECK: [[A_ASHR:%.+]] = ashr i32 [[A_SHL]], 18
807 // CHECK: [[X_RVAL:%.+]] = sitofp i32 [[A_ASHR]] to x86_fp80
808 // CHECK: [[DIV:%.+]] = fdiv x86_fp80 [[X_RVAL]], [[EXPR]]
809 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 [[DIV]] to i32
810 // CHECK: [[BF_LD:%.+]] = load i32, i32* [[TEMP1]],
811 // CHECK: [[BF_AND:%.+]] = and i32 [[NEW_VAL]], 16383
812 // CHECK: [[BF_VALUE:%.+]] = shl i32 [[BF_AND]], 11
813 // CHECK: [[BF_CLEAR:%.+]] = and i32 %{{.+}}, -33552385
814 // CHECK: or i32 [[BF_CLEAR]], [[BF_VALUE]]
815 // CHECK: store i32 %{{.+}}, i32* [[TEMP1]]
816 // CHECK: [[NEW_BF_VALUE:%.+]] = load i32, i32* [[TEMP1]]
817 // CHECK: [[RES:%.+]] = cmpxchg i32* getelementptr inbounds (%struct.BitFields3, %struct.BitFields3* @{{.+}}, i32 0, i32 0), i32 [[OLD_BF_VALUE]], i32 [[NEW_BF_VALUE]] monotonic monotonic
818 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i32, i1 } [[RES]], 0
819 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i32, i1 } [[RES]], 1
820 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
821 // CHECK: [[EXIT]]
822 // CHECK: store i32 [[A_ASHR]], i32* @{{.+}},
823 #pragma omp atomic capture
824   {iv = bfx3.a; bfx3.a /= ldv;}
825 // CHECK: [[EXPR:%.+]] = load x86_fp80, x86_fp80* @{{.+}}
826 // CHECK: [[LDTEMP:%.+]] = bitcast i32* %{{.+}} to i24*
827 // CHECK: [[BITCAST:%.+]] = bitcast i24* [[LDTEMP]] to i8*
828 // CHECK: call void @__atomic_load(i64 3, i8* getelementptr (i8, i8* bitcast (%struct.BitFields3_packed* @{{.+}} to i8*), i64 1), i8* [[BITCAST]], i32 0)
829 // CHECK: br label %[[CONT:.+]]
830 // CHECK: [[CONT]]
831 // CHECK: [[OLD:%.+]] = load i24, i24* [[LDTEMP]],
832 // CHECK: store i24 [[OLD]], i24* [[BITCAST2:%.+]],
833 // CHECK: [[OLD:%.+]] = load i24, i24* [[LDTEMP]],
834 // CHECK: store i24 [[OLD]], i24* [[BITCAST1:%.+]],
835 // CHECK: [[A_LD:%.+]] = load i24, i24* [[BITCAST1]],
836 // CHECK: [[A_SHL:%.+]] = shl i24 [[A_LD]], 7
837 // CHECK: [[A_ASHR:%.+]] = ashr i24 [[A_SHL]], 10
838 // CHECK: [[CAST:%.+]] = sext i24 [[A_ASHR]] to i32
839 // CHECK: [[X_RVAL:%.+]] = sitofp i32 [[CAST]] to x86_fp80
840 // CHECK: [[ADD:%.+]] = fadd x86_fp80 [[X_RVAL]], [[EXPR]]
841 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 [[ADD]] to i32
842 // CHECK: [[TRUNC:%.+]] = trunc i32 [[NEW_VAL]] to i24
843 // CHECK: [[BF_LD:%.+]] = load i24, i24* [[BITCAST2]],
844 // CHECK: [[BF_AND:%.+]] = and i24 [[TRUNC]], 16383
845 // CHECK: [[BF_VALUE:%.+]] = shl i24 [[BF_AND]], 3
846 // CHECK: [[BF_CLEAR:%.+]] = and i24 [[BF_LD]], -131065
847 // CHECK: or i24 [[BF_CLEAR]], [[BF_VALUE]]
848 // CHECK: store i24 %{{.+}}, i24* [[BITCAST2]]
849 // CHECK: [[BITCAST_TEMP_OLD_BF_ADDR:%.+]] = bitcast i24* [[LDTEMP]] to i8*
850 // CHECK: [[BITCAST_TEMP_NEW_BF_ADDR:%.+]] = bitcast i24* [[BITCAST2]] to i8*
851 // 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)
852 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
853 // CHECK: [[EXIT]]
854 // CHECK: store i32 [[NEW_VAL]], i32* @{{.+}},
855 #pragma omp atomic capture
856   {bfx3_packed.a += ldv; iv = bfx3_packed.a;}
857 // CHECK: [[EXPR:%.+]] = load x86_fp80, x86_fp80* @{{.+}}
858 // CHECK: [[PREV_VALUE:%.+]] = load atomic i64, i64* bitcast (%struct.BitFields4* @{{.+}} to i64*) monotonic
859 // CHECK: br label %[[CONT:.+]]
860 // CHECK: [[CONT]]
861 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i64 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
862 // CHECK: store i64 [[OLD_BF_VALUE]], i64* [[TEMP1:%.+]],
863 // CHECK: store i64 [[OLD_BF_VALUE]], i64* [[TEMP:%.+]],
864 // CHECK: [[A_LD:%.+]] = load i64, i64* [[TEMP]],
865 // CHECK: [[A_SHL:%.+]] = shl i64 [[A_LD]], 47
866 // CHECK: [[A_ASHR:%.+]] = ashr i64 [[A_SHL:%.+]], 63
867 // CHECK: [[A_CAST:%.+]] = trunc i64 [[A_ASHR:%.+]] to i32
868 // CHECK: [[X_RVAL:%.+]] = sitofp i32 [[CAST:%.+]] to x86_fp80
869 // CHECK: [[MUL:%.+]] = fmul x86_fp80 [[X_RVAL]], [[EXPR]]
870 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 [[MUL]] to i32
871 // CHECK: [[ZEXT:%.+]] = zext i32 [[NEW_VAL]] to i64
872 // CHECK: [[BF_LD:%.+]] = load i64, i64* [[TEMP1]],
873 // CHECK: [[BF_AND:%.+]] = and i64 [[ZEXT]], 1
874 // CHECK: [[BF_VALUE:%.+]] = shl i64 [[BF_AND]], 16
875 // CHECK: [[BF_CLEAR:%.+]] = and i64 [[BF_LD]], -65537
876 // CHECK: or i64 [[BF_CLEAR]], [[BF_VALUE]]
877 // CHECK: store i64 %{{.+}}, i64* [[TEMP1]]
878 // CHECK: [[NEW_BF_VALUE:%.+]] = load i64, i64* [[TEMP1]]
879 // CHECK: [[RES:%.+]] = cmpxchg i64* bitcast (%struct.BitFields4* @{{.+}} to i64*), i64 [[OLD_BF_VALUE]], i64 [[NEW_BF_VALUE]] monotonic monotonic
880 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i64, i1 } [[RES]], 0
881 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i64, i1 } [[RES]], 1
882 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
883 // CHECK: [[EXIT]]
884 // CHECK: store i32 [[NEW_VAL]], i32* @{{.+}},
885 #pragma omp atomic capture
886   iv = bfx4.a = bfx4.a * ldv;
887 // CHECK: [[EXPR:%.+]] = load x86_fp80, x86_fp80* @{{.+}}
888 // CHECK: [[PREV_VALUE:%.+]] = load atomic i8, i8* getelementptr inbounds (%struct.BitFields4_packed, %struct.BitFields4_packed* @{{.+}}, i32 0, i32 0, i64 2) monotonic
889 // CHECK: br label %[[CONT:.+]]
890 // CHECK: [[CONT]]
891 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i8 [ [[PREV_VALUE]], %{{.+}} ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
892 // CHECK: [[BITCAST1:%.+]] = bitcast i32* %{{.+}} to i8*
893 // CHECK: store i8 [[OLD_BF_VALUE]], i8* [[BITCAST1]],
894 // CHECK: [[BITCAST:%.+]] = bitcast i32* %{{.+}} to i8*
895 // CHECK: store i8 [[OLD_BF_VALUE]], i8* [[BITCAST]],
896 // CHECK: [[A_LD:%.+]] = load i8, i8* [[BITCAST]],
897 // CHECK: [[A_SHL:%.+]] = shl i8 [[A_LD]], 7
898 // CHECK: [[A_ASHR:%.+]] = ashr i8 [[A_SHL:%.+]], 7
899 // CHECK: [[CAST:%.+]] = sext i8 [[A_ASHR:%.+]] to i32
900 // CHECK: [[CONV:%.+]] = sitofp i32 [[CAST]] to x86_fp80
901 // CHECK: [[SUB: %.+]] = fsub x86_fp80 [[CONV]], [[EXPR]]
902 // CHECK: [[CONV:%.+]] = fptosi x86_fp80 [[SUB:%.+]] to i32
903 // CHECK: [[NEW_VAL:%.+]] = trunc i32 [[CONV]] to i8
904 // CHECK: [[BF_LD:%.+]] = load i8, i8* [[BITCAST1]],
905 // CHECK: [[BF_VALUE:%.+]] = and i8 [[NEW_VAL]], 1
906 // CHECK: [[BF_CLEAR:%.+]] = and i8 [[BF_LD]], -2
907 // CHECK: or i8 [[BF_CLEAR]], [[BF_VALUE]]
908 // CHECK: store i8 %{{.+}}, i8* [[BITCAST1]]
909 // CHECK: [[NEW_BF_VALUE:%.+]] = load i8, i8* [[BITCAST1]]
910 // 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
911 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i8, i1 } [[RES]], 0
912 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i8, i1 } [[RES]], 1
913 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
914 // CHECK: [[EXIT]]
915 // CHECK: store i32 [[CAST]], i32* @{{.+}},
916 #pragma omp atomic capture
917   {iv = bfx4_packed.a; bfx4_packed.a -= ldv;}
918 // CHECK: [[EXPR:%.+]] = load x86_fp80, x86_fp80* @{{.+}}
919 // CHECK: [[PREV_VALUE:%.+]] = load atomic i64, i64* bitcast (%struct.BitFields4* @{{.+}} to i64*) monotonic
920 // CHECK: br label %[[CONT:.+]]
921 // CHECK: [[CONT]]
922 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i64 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
923 // CHECK: store i64 [[OLD_BF_VALUE]], i64* [[TEMP1:%.+]],
924 // CHECK: store i64 [[OLD_BF_VALUE]], i64* [[TEMP:%.+]],
925 // CHECK: [[A_LD:%.+]] = load i64, i64* [[TEMP]],
926 // CHECK: [[A_SHL:%.+]] = shl i64 [[A_LD]], 40
927 // CHECK: [[A_ASHR:%.+]] = ashr i64 [[A_SHL:%.+]], 57
928 // CHECK: [[CONV:%.+]] = sitofp i64 [[A_ASHR]] to x86_fp80
929 // CHECK: [[DIV:%.+]] = fdiv x86_fp80 [[CONV]], [[EXPR]]
930 // CHECK: [[CONV:%.+]] = fptosi x86_fp80 [[DIV]] to i64
931 // CHECK: [[BF_LD:%.+]] = load i64, i64* [[TEMP1]],
932 // CHECK: [[BF_AND:%.+]] = and i64 [[CONV]], 127
933 // CHECK: [[BF_VALUE:%.+]] = shl i64 [[BF_AND:%.+]], 17
934 // CHECK: [[BF_CLEAR:%.+]] = and i64 [[BF_LD]], -16646145
935 // CHECK: [[VAL:%.+]] = or i64 [[BF_CLEAR]], [[BF_VALUE]]
936 // CHECK: store i64 [[VAL]], i64* [[TEMP1]]
937 // CHECK: [[NEW_BF_VALUE:%.+]] = load i64, i64* [[TEMP1]]
938 // CHECK: [[RES:%.+]] = cmpxchg i64* bitcast (%struct.BitFields4* @{{.+}} to i64*), i64 [[OLD_BF_VALUE]], i64 [[NEW_BF_VALUE]] monotonic monotonic
939 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i64, i1 } [[RES]], 0
940 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i64, i1 } [[RES]], 1
941 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
942 // CHECK: [[EXIT]]
943 // CHECK: [[NEW_VAL:%.+]] = trunc i64 [[CONV]] to i32
944 // CHECK: store i32 [[NEW_VAL]], i32* @{{.+}},
945 #pragma omp atomic capture
946   {bfx4.b /= ldv; iv = bfx4.b;}
947 // CHECK: [[EXPR:%.+]] = load x86_fp80, x86_fp80* @{{.+}}
948 // CHECK: [[PREV_VALUE:%.+]] = load atomic i8, i8* getelementptr inbounds (%struct.BitFields4_packed, %struct.BitFields4_packed* @{{.+}}, i32 0, i32 0, i64 2) monotonic
949 // CHECK: br label %[[CONT:.+]]
950 // CHECK: [[CONT]]
951 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i8 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
952 // CHECK: [[BITCAST1:%.+]] = bitcast i64* %{{.+}} to i8*
953 // CHECK: store i8 [[OLD_BF_VALUE]], i8* [[BITCAST1]],
954 // CHECK: [[BITCAST:%.+]] = bitcast i64* %{{.+}} to i8*
955 // CHECK: store i8 [[OLD_BF_VALUE]], i8* [[BITCAST]],
956 // CHECK: [[A_LD:%.+]] = load i8, i8* [[BITCAST]],
957 // CHECK: [[A_ASHR:%.+]] = ashr i8 [[A_LD]], 1
958 // CHECK: [[CAST:%.+]] = sext i8 [[A_ASHR]] to i64
959 // CHECK: [[CONV:%.+]] = sitofp i64 [[CAST]] to x86_fp80
960 // CHECK: [[ADD:%.+]] = fadd x86_fp80 [[CONV]], [[EXPR]]
961 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 [[ADD]] to i64
962 // CHECK: [[TRUNC:%.+]] = trunc i64 [[NEW_VAL]] to i8
963 // CHECK: [[BF_LD:%.+]] = load i8, i8* [[BITCAST1]],
964 // CHECK: [[BF_AND:%.+]] = and i8 [[TRUNC]], 127
965 // CHECK: [[BF_VALUE:%.+]] = shl i8 [[BF_AND]], 1
966 // CHECK: [[BF_CLEAR:%.+]] = and i8 [[BF_LD]], 1
967 // CHECK: or i8 [[BF_CLEAR]], [[BF_VALUE]]
968 // CHECK: store i8 %{{.+}}, i8* [[BITCAST1]]
969 // CHECK: [[NEW_BF_VALUE:%.+]] = load i8, i8* [[BITCAST1]]
970 // 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
971 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i8, i1 } [[RES]], 0
972 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i8, i1 } [[RES]], 1
973 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
974 // CHECK: [[EXIT]]
975 // CHECK: [[NEW_VAL_I32:%.+]] = trunc i64 [[NEW_VAL]] to i32
976 // CHECK: store i32 [[NEW_VAL_I32]], i32* @{{.+}},
977 #pragma omp atomic capture
978   iv = bfx4_packed.b += ldv;
979 // CHECK: load i64, i64*
980 // CHECK: [[EXPR:%.+]] = uitofp i64 %{{.+}} to float
981 // CHECK: [[I64VAL:%.+]] = load atomic i64, i64* bitcast (<2 x float>* [[DEST:@.+]] to i64*) monotonic
982 // CHECK: br label %[[CONT:.+]]
983 // CHECK: [[CONT]]
984 // CHECK: [[OLD_I64:%.+]] = phi i64 [ [[I64VAL]], %{{.+}} ], [ [[FAILED_I64_OLD_VAL:%.+]], %[[CONT]] ]
985 // CHECK: [[BITCAST:%.+]] = bitcast <2 x float>* [[LDTEMP1:%.+]] to i64*
986 // CHECK: store i64 [[OLD_I64]], i64* [[BITCAST]],
987 // CHECK: [[OLD_VEC_VAL:%.+]] = bitcast i64 [[OLD_I64]] to <2 x float>
988 // CHECK: store <2 x float> [[OLD_VEC_VAL]], <2 x float>* [[LDTEMP:%.+]],
989 // CHECK: [[VEC_VAL:%.+]] = load <2 x float>, <2 x float>* [[LDTEMP]]
990 // CHECK: [[X:%.+]] = extractelement <2 x float> [[VEC_VAL]], i64 0
991 // CHECK: [[VEC_ITEM_VAL:%.+]] = fsub float [[EXPR]], [[X]]
992 // CHECK: [[VEC_VAL:%.+]] = load <2 x float>, <2 x float>* [[LDTEMP1]],
993 // CHECK: [[NEW_VEC_VAL:%.+]] = insertelement <2 x float> [[VEC_VAL]], float [[VEC_ITEM_VAL]], i64 0
994 // CHECK: store <2 x float> [[NEW_VEC_VAL]], <2 x float>* [[LDTEMP1]]
995 // CHECK: [[NEW_I64:%.+]] = load i64, i64* [[BITCAST]]
996 // CHECK: [[RES:%.+]] = cmpxchg i64* bitcast (<2 x float>* [[DEST]] to i64*), i64 [[OLD_I64]], i64 [[NEW_I64]] monotonic monotonic
997 // CHECK: [[FAILED_I64_OLD_VAL:%.+]] = extractvalue { i64, i1 } [[RES]], 0
998 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i64, i1 } [[RES]], 1
999 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
1000 // CHECK: [[EXIT]]
1001 // CHECK: store float [[X]], float* @{{.+}},
1002 #pragma omp atomic capture
1003   {fv = float2x.x; float2x.x = ulv - float2x.x;}
1004 // CHECK: [[EXPR:%.+]] = load double, double* @{{.+}},
1005 // CHECK: [[OLD_VAL:%.+]] = call i32 @llvm.read_register.i32([[REG:metadata ![0-9]+]])
1006 // CHECK: [[X_RVAL:%.+]] = sitofp i32 [[OLD_VAL]] to double
1007 // CHECK: [[DIV:%.+]] = fdiv double [[EXPR]], [[X_RVAL]]
1008 // CHECK: [[NEW_VAL:%.+]] = fptosi double [[DIV]] to i32
1009 // CHECK: call void @llvm.write_register.i32([[REG]], i32 [[NEW_VAL]])
1010 // CHECK: store i32 [[NEW_VAL]], i32* @{{.+}},
1011 // CHECK: call{{.*}} @__kmpc_flush(
1012 #pragma omp atomic capture seq_cst
1013   {rix = dv / rix; iv = rix;}
1014 // CHECK: [[OLD_VAL:%.+]] = atomicrmw xchg i32* @{{.+}}, i32 5 monotonic
1015 // CHECK: call void @llvm.write_register.i32([[REG]], i32 [[OLD_VAL]])
1016 #pragma omp atomic capture
1017   {rix = ix; ix = 5;}
1018   return 0;
1019 }
1020 #endif
1021