Home | History | Annotate | Download | only in OpenMP 
      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: load i8, i8*
     82 // CHECK: store atomic i8
     83 #pragma omp atomic write
     84   bx = bv;
     85 // CHECK: load i8, i8*
     86 // CHECK: store atomic i8
     87 #pragma omp atomic write
     88   cx = cv;
     89 // CHECK: load i8, i8*
     90 // CHECK: store atomic i8
     91 #pragma omp atomic write
     92   ucx = ucv;
     93 // CHECK: load i16, i16*
     94 // CHECK: store atomic i16
     95 #pragma omp atomic write
     96   sx = sv;
     97 // CHECK: load i16, i16*
     98 // CHECK: store atomic i16
     99 #pragma omp atomic write
    100   usx = usv;
    101 // CHECK: load i32, i32*
    102 // CHECK: store atomic i32
    103 #pragma omp atomic write
    104   ix = iv;
    105 // CHECK: load i32, i32*
    106 // CHECK: store atomic i32
    107 #pragma omp atomic write
    108   uix = uiv;
    109 // CHECK: load i64, i64*
    110 // CHECK: store atomic i64
    111 #pragma omp atomic write
    112   lx = lv;
    113 // CHECK: load i64, i64*
    114 // CHECK: store atomic i64
    115 #pragma omp atomic write
    116   ulx = ulv;
    117 // CHECK: load i64, i64*
    118 // CHECK: store atomic i64
    119 #pragma omp atomic write
    120   llx = llv;
    121 // CHECK: load i64, i64*
    122 // CHECK: store atomic i64
    123 #pragma omp atomic write
    124   ullx = ullv;
    125 // CHECK: load float, float*
    126 // CHECK: bitcast float {{.*}} to i32
    127 // CHECK: store atomic i32 {{.*}}, i32* bitcast (float*
    128 #pragma omp atomic write
    129   fx = fv;
    130 // CHECK: load double, double*
    131 // CHECK: bitcast double {{.*}} to i64
    132 // CHECK: store atomic i64 {{.*}}, i64* bitcast (double*
    133 #pragma omp atomic write
    134   dx = dv;
    135 // CHECK: [[LD:%.+]] = load x86_fp80, x86_fp80*
    136 // CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[LDTEMP:%.*]] to i8*
    137 // CHECK: call void @llvm.memset.p0i8.i64(i8* [[BITCAST]], i8 0, i64 16, i32 16, i1 false)
    138 // CHECK: store x86_fp80 [[LD]], x86_fp80* [[LDTEMP]]
    139 // CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[LDTEMP:%.*]] to i128*
    140 // CHECK: [[LD:%.+]] = load i128, i128* [[BITCAST]]
    141 // CHECK: store atomic i128 [[LD]], i128* bitcast (x86_fp80*
    142 #pragma omp atomic write
    143   ldx = ldv;
    144 // CHECK: [[REAL_VAL:%.+]] = load i32, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.*}}, i32 0, i32 0)
    145 // CHECK: [[IMG_VAL:%.+]] = load i32, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.*}}, i32 0, i32 1)
    146 // CHECK: [[TEMP_REAL_REF:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP:%.+]], i32 0, i32 0
    147 // CHECK: [[TEMP_IMG_REF:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1
    148 // CHECK: store i32 [[REAL_VAL]], i32* [[TEMP_REAL_REF]]
    149 // CHECK: store i32 [[IMG_VAL]], i32* [[TEMP_IMG_REF]]
    150 // CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[TEMP]] to i8*
    151 // CHECK: call void @__atomic_store(i64 8, i8* bitcast ({ i32, i32 }* @{{.*}} to i8*), i8* [[BITCAST]], i32 0)
    152 #pragma omp atomic write
    153   cix = civ;
    154 // CHECK: [[REAL_VAL:%.+]] = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @{{.*}}, i32 0, i32 0)
    155 // CHECK: [[IMG_VAL:%.+]] = load float, float* getelementptr inbounds ({ float, float }, { float, float }* @{{.*}}, i32 0, i32 1)
    156 // CHECK: [[TEMP_REAL_REF:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[TEMP:%.+]], i32 0, i32 0
    157 // CHECK: [[TEMP_IMG_REF:%.+]] = getelementptr inbounds { float, float }, { float, float }* [[TEMP]], i32 0, i32 1
    158 // CHECK: store float [[REAL_VAL]], float* [[TEMP_REAL_REF]]
    159 // CHECK: store float [[IMG_VAL]], float* [[TEMP_IMG_REF]]
    160 // CHECK: [[BITCAST:%.+]] = bitcast { float, float }* [[TEMP]] to i8*
    161 // CHECK: call void @__atomic_store(i64 8, i8* bitcast ({ float, float }* @{{.*}} to i8*), i8* [[BITCAST]], i32 0)
    162 #pragma omp atomic write
    163   cfx = cfv;
    164 // CHECK: [[REAL_VAL:%.+]] = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @{{.*}}, i32 0, i32 0)
    165 // CHECK: [[IMG_VAL:%.+]] = load double, double* getelementptr inbounds ({ double, double }, { double, double }* @{{.*}}, i32 0, i32 1)
    166 // CHECK: [[TEMP_REAL_REF:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[TEMP:%.+]], i32 0, i32 0
    167 // CHECK: [[TEMP_IMG_REF:%.+]] = getelementptr inbounds { double, double }, { double, double }* [[TEMP]], i32 0, i32 1
    168 // CHECK: store double [[REAL_VAL]], double* [[TEMP_REAL_REF]]
    169 // CHECK: store double [[IMG_VAL]], double* [[TEMP_IMG_REF]]
    170 // CHECK: [[BITCAST:%.+]] = bitcast { double, double }* [[TEMP]] to i8*
    171 // CHECK: call void @__atomic_store(i64 16, i8* bitcast ({ double, double }* @{{.*}} to i8*), i8* [[BITCAST]], i32 5)
    172 // CHECK: call{{.*}} @__kmpc_flush(
    173 #pragma omp atomic seq_cst write
    174   cdx = cdv;
    175 // CHECK: load i8, i8*
    176 // CHECK: store atomic i64
    177 #pragma omp atomic write
    178   ulx = bv;
    179 // CHECK: load i8, i8*
    180 // CHECK: store atomic i8
    181 #pragma omp atomic write
    182   bx = cv;
    183 // CHECK: load i8, i8*
    184 // CHECK: store atomic i8
    185 // CHECK: call{{.*}} @__kmpc_flush(
    186 #pragma omp atomic write, seq_cst
    187   cx = ucv;
    188 // CHECK: load i16, i16*
    189 // CHECK: store atomic i64
    190 #pragma omp atomic write
    191   ulx = sv;
    192 // CHECK: load i16, i16*
    193 // CHECK: store atomic i64
    194 #pragma omp atomic write
    195   lx = usv;
    196 // CHECK: load i32, i32*
    197 // CHECK: store atomic i32
    198 // CHECK: call{{.*}} @__kmpc_flush(
    199 #pragma omp atomic seq_cst, write
    200   uix = iv;
    201 // CHECK: load i32, i32*
    202 // CHECK: store atomic i32
    203 #pragma omp atomic write
    204   ix = uiv;
    205 // CHECK: load i64, i64*
    206 // CHECK: [[VAL:%.+]] = trunc i64 %{{.*}} to i32
    207 // CHECK: [[TEMP_REAL_REF:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP:%.+]], i32 0, i32 0
    208 // CHECK: [[TEMP_IMG_REF:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1
    209 // CHECK: store i32 [[VAL]], i32* [[TEMP_REAL_REF]]
    210 // CHECK: store i32 0, i32* [[TEMP_IMG_REF]]
    211 // CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[TEMP]] to i8*
    212 // CHECK: call void @__atomic_store(i64 8, i8* bitcast ({ i32, i32 }* @{{.+}} to i8*), i8* [[BITCAST]], i32 0)
    213 #pragma omp atomic write
    214   cix = lv;
    215 // CHECK: load i64, i64*
    216 // CHECK: store atomic i32 %{{.+}}, i32* bitcast (float*
    217 #pragma omp atomic write
    218   fx = ulv;
    219 // CHECK: load i64, i64*
    220 // CHECK: store atomic i64 %{{.+}}, i64* bitcast (double*
    221 #pragma omp atomic write
    222   dx = llv;
    223 // CHECK: load i64, i64*
    224 // CHECK: [[VAL:%.+]] = uitofp i64 %{{.+}} to x86_fp80
    225 // CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP:%.+]] to i8*
    226 // CHECK: call void @llvm.memset.p0i8.i64(i8* [[BITCAST]], i8 0, i64 16, i32 16, i1 false)
    227 // CHECK: store x86_fp80 [[VAL]], x86_fp80* [[TEMP]]
    228 // CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP]] to i128*
    229 // CHECK: [[VAL:%.+]] = load i128, i128* [[BITCAST]]
    230 // CHECK: store atomic i128 [[VAL]], i128* bitcast (x86_fp80*
    231 #pragma omp atomic write
    232   ldx = ullv;
    233 // CHECK: load float, float*
    234 // CHECK: [[VAL:%.+]] = fptosi float %{{.*}} to i32
    235 // CHECK: [[TEMP_REAL_REF:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP:%.+]], i32 0, i32 0
    236 // CHECK: [[TEMP_IMG_REF:%.+]] = getelementptr inbounds { i32, i32 }, { i32, i32 }* [[TEMP]], i32 0, i32 1
    237 // CHECK: store i32 [[VAL]], i32* [[TEMP_REAL_REF]]
    238 // CHECK: store i32 0, i32* [[TEMP_IMG_REF]]
    239 // CHECK: [[BITCAST:%.+]] = bitcast { i32, i32 }* [[TEMP]] to i8*
    240 // CHECK: call void @__atomic_store(i64 8, i8* bitcast ({ i32, i32 }* @{{.+}} to i8*), i8* [[BITCAST]], i32 0)
    241 #pragma omp atomic write
    242   cix = fv;
    243 // CHECK: load double, double*
    244 // CHECK: store atomic i16
    245 #pragma omp atomic write
    246   sx = dv;
    247 // CHECK: load x86_fp80, x86_fp80*
    248 // CHECK: store atomic i8
    249 #pragma omp atomic write
    250   bx = ldv;
    251 // CHECK: load i32, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.+}}, i32 0, i32 0)
    252 // CHECK: load i32, i32* getelementptr inbounds ({ i32, i32 }, { i32, i32 }* @{{.+}}, i32 0, i32 1)
    253 // CHECK: icmp ne i32 %{{.+}}, 0
    254 // CHECK: icmp ne i32 %{{.+}}, 0
    255 // CHECK: or i1
    256 // CHECK: store atomic i8
    257 #pragma omp atomic write
    258   bx = civ;
    259 // CHECK: load float, float* getelementptr inbounds ({ float, float }, { float, float }* @{{.*}}, i32 0, i32 0)
    260 // CHECK: store atomic i16
    261 #pragma omp atomic write
    262   usx = cfv;
    263 // CHECK: load double, double* getelementptr inbounds ({ double, double }, { double, double }* @{{.+}}, i32 0, i32 0)
    264 // CHECK: store atomic i64
    265 #pragma omp atomic write
    266   llx = cdv;
    267 // CHECK-DAG: [[IDX:%.+]] = load i16, i16* @{{.+}}
    268 // CHECK-DAG: load i8, i8*
    269 // CHECK-DAG: [[VEC_ITEM_VAL:%.+]] = zext i1 %{{.+}} to i32
    270 // CHECK: [[I128VAL:%.+]] = load atomic i128, i128* bitcast (<4 x i32>* [[DEST:@.+]] to i128*) monotonic
    271 // CHECK: br label %[[CONT:.+]]
    272 // CHECK: [[CONT]]
    273 // CHECK: [[OLD_I128:%.+]] = phi i128 [ [[I128VAL]], %{{.+}} ], [ [[FAILED_I128_OLD_VAL:%.+]], %[[CONT]] ]
    274 // CHECK: [[BITCAST:%.+]] = bitcast <4 x i32>* [[LDTEMP:%.+]] to i128*
    275 // CHECK: store i128 [[OLD_I128]], i128* [[BITCAST]],
    276 // CHECK: [[VEC_VAL:%.+]] = load <4 x i32>, <4 x i32>* [[LDTEMP]]
    277 // CHECK: [[NEW_VEC_VAL:%.+]] = insertelement <4 x i32> [[VEC_VAL]], i32 [[VEC_ITEM_VAL]], i16 [[IDX]]
    278 // CHECK: store <4 x i32> [[NEW_VEC_VAL]], <4 x i32>* [[LDTEMP]]
    279 // CHECK: [[NEW_I128:%.+]] = load i128, i128* [[BITCAST]]
    280 // CHECK: [[RES:%.+]] = cmpxchg i128* bitcast (<4 x i32>* [[DEST]] to i128*), i128 [[OLD_I128]], i128 [[NEW_I128]] monotonic monotonic
    281 // CHECK: [[FAILED_I128_OLD_VAL:%.+]] = extractvalue { i128, i1 } [[RES]], 0
    282 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i128, i1 } [[RES]], 1
    283 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
    284 // CHECK: [[EXIT]]
    285 #pragma omp atomic write
    286   int4x[sv] = bv;
    287 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
    288 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32
    289 // CHECK: [[PREV_VALUE:%.+]] = load atomic i32, i32* bitcast (i8* getelementptr (i8, i8* bitcast (%struct.BitFields* @{{.+}} to i8*), i64 4) to i32*) monotonic
    290 // CHECK: br label %[[CONT:.+]]
    291 // CHECK: [[CONT]]
    292 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i32 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
    293 // CHECK: [[BF_VALUE:%.+]] = and i32 [[NEW_VAL]], 2147483647
    294 // CHECK: [[BF_CLEAR:%.+]] = and i32 %{{.+}}, -2147483648
    295 // CHECK: or i32 [[BF_CLEAR]], [[BF_VALUE]]
    296 // CHECK: store i32 %{{.+}}, i32* [[LDTEMP:%.+]]
    297 // CHECK: [[NEW_BF_VALUE:%.+]] = load i32, i32* [[LDTEMP]]
    298 // CHECK: [[RES:%.+]] = cmpxchg i32* bitcast (i8* getelementptr (i8, i8* bitcast (%struct.BitFields* @{{.+}} to i8*), i64 4) to i32*), i32 [[OLD_BF_VALUE]], i32 [[NEW_BF_VALUE]] monotonic monotonic
    299 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i32, i1 } [[RES]], 0
    300 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i32, i1 } [[RES]], 1
    301 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
    302 // CHECK: [[EXIT]]
    303 #pragma omp atomic write
    304   bfx.a = ldv;
    305 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
    306 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32
    307 // CHECK: [[BITCAST:%.+]] = bitcast i32* [[LDTEMP:%.+]] to i8*
    308 // CHECK: call void @__atomic_load(i64 4, i8* getelementptr (i8, i8* bitcast (%struct.BitFields_packed* @{{.+}} to i8*), i64 4), i8* [[BITCAST]], i32 0)
    309 // CHECK: br label %[[CONT:.+]]
    310 // CHECK: [[CONT]]
    311 // CHECK: [[OLD_BF_VALUE:%.+]] = load i32, i32* [[LDTEMP]],
    312 // CHECK: store i32 [[OLD_BF_VALUE]], i32* [[LDTEMP1:%.+]],
    313 // CHECK: [[OLD_BF_VALUE:%.+]] = load i32, i32* [[LDTEMP1]],
    314 // CHECK: [[BF_VALUE:%.+]] = and i32 [[NEW_VAL]], 2147483647
    315 // CHECK: [[BF_CLEAR:%.+]] = and i32 [[OLD_BF_VALUE]], -2147483648
    316 // CHECK: or i32 [[BF_CLEAR]], [[BF_VALUE]]
    317 // CHECK: store i32 %{{.+}}, i32* [[LDTEMP1]]
    318 // CHECK: [[BITCAST_TEMP_OLD_BF_ADDR:%.+]] = bitcast i32* [[LDTEMP]] to i8*
    319 // CHECK: [[BITCAST_TEMP_NEW_BF_ADDR:%.+]] = bitcast i32* [[LDTEMP1]] to i8*
    320 // 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)
    321 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
    322 // CHECK: [[EXIT]]
    323 #pragma omp atomic write
    324   bfx_packed.a = ldv;
    325 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
    326 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32
    327 // CHECK: [[PREV_VALUE:%.+]] = load atomic i32, i32* getelementptr inbounds (%struct.BitFields2, %struct.BitFields2* @{{.+}}, i32 0, i32 0) monotonic
    328 // CHECK: br label %[[CONT:.+]]
    329 // CHECK: [[CONT]]
    330 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i32 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
    331 // CHECK: [[BF_AND:%.+]] = and i32 [[NEW_VAL]], 1
    332 // CHECK: [[BF_VALUE:%.+]] = shl i32 [[BF_AND]], 31
    333 // CHECK: [[BF_CLEAR:%.+]] = and i32 %{{.+}}, 2147483647
    334 // CHECK: or i32 [[BF_CLEAR]], [[BF_VALUE]]
    335 // CHECK: store i32 %{{.+}}, i32* [[LDTEMP:%.+]]
    336 // CHECK: [[NEW_BF_VALUE:%.+]] = load i32, i32* [[LDTEMP]]
    337 // CHECK: [[RES:%.+]] = cmpxchg i32* getelementptr inbounds (%struct.BitFields2, %struct.BitFields2* @{{.+}}, i32 0, i32 0), i32 [[OLD_BF_VALUE]], i32 [[NEW_BF_VALUE]] monotonic monotonic
    338 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i32, i1 } [[RES]], 0
    339 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i32, i1 } [[RES]], 1
    340 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
    341 // CHECK: [[EXIT]]
    342 #pragma omp atomic write
    343   bfx2.a = ldv;
    344 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
    345 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32
    346 // CHECK: [[PREV_VALUE:%.+]] = load atomic i8, i8* getelementptr (i8, i8* bitcast (%struct.BitFields2_packed* @{{.+}} to i8*), i64 3) monotonic
    347 // CHECK: br label %[[CONT:.+]]
    348 // CHECK: [[CONT]]
    349 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i8 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
    350 // CHECK: [[TRUNC:%.+]] = trunc i32 [[NEW_VAL]] to i8
    351 // CHECK: [[BF_AND:%.+]] = and i8 [[TRUNC]], 1
    352 // CHECK: [[BF_VALUE:%.+]] = shl i8 [[BF_AND]], 7
    353 // CHECK: [[BF_CLEAR:%.+]] = and i8 %{{.+}}, 127
    354 // CHECK: or i8 [[BF_CLEAR]], [[BF_VALUE]]
    355 // CHECK: store i8 %{{.+}}, i8* [[LDTEMP:%.+]]
    356 // CHECK: [[NEW_BF_VALUE:%.+]] = load i8, i8* [[LDTEMP]]
    357 // 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
    358 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i8, i1 } [[RES]], 0
    359 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i8, i1 } [[RES]], 1
    360 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
    361 // CHECK: [[EXIT]]
    362 #pragma omp atomic write
    363   bfx2_packed.a = ldv;
    364 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
    365 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32
    366 // CHECK: [[PREV_VALUE:%.+]] = load atomic i32, i32* getelementptr inbounds (%struct.BitFields3, %struct.BitFields3* @{{.+}}, i32 0, i32 0) monotonic
    367 // CHECK: br label %[[CONT:.+]]
    368 // CHECK: [[CONT]]
    369 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i32 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
    370 // CHECK: [[BF_AND:%.+]] = and i32 [[NEW_VAL]], 16383
    371 // CHECK: [[BF_VALUE:%.+]] = shl i32 [[BF_AND]], 11
    372 // CHECK: [[BF_CLEAR:%.+]] = and i32 %{{.+}}, -33552385
    373 // CHECK: or i32 [[BF_CLEAR]], [[BF_VALUE]]
    374 // CHECK: store i32 %{{.+}}, i32* [[LDTEMP:%.+]]
    375 // CHECK: [[NEW_BF_VALUE:%.+]] = load i32, i32* [[LDTEMP]]
    376 // CHECK: [[RES:%.+]] = cmpxchg i32* getelementptr inbounds (%struct.BitFields3, %struct.BitFields3* @{{.+}}, i32 0, i32 0), i32 [[OLD_BF_VALUE]], i32 [[NEW_BF_VALUE]] monotonic monotonic
    377 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i32, i1 } [[RES]], 0
    378 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i32, i1 } [[RES]], 1
    379 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
    380 // CHECK: [[EXIT]]
    381 #pragma omp atomic write
    382   bfx3.a = ldv;
    383 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
    384 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32
    385 // CHECK: [[LDTEMP:%.+]] = bitcast i32* %{{.+}} to i24*
    386 // CHECK: [[BITCAST:%.+]] = bitcast i24* %{{.+}} to i8*
    387 // CHECK: call void @__atomic_load(i64 3, i8* getelementptr (i8, i8* bitcast (%struct.BitFields3_packed* @{{.+}} to i8*), i64 1), i8* [[BITCAST]], i32 0)
    388 // CHECK: br label %[[CONT:.+]]
    389 // CHECK: [[CONT]]
    390 // CHECK: [[OLD_VAL:%.+]] = load i24, i24* %{{.+}},
    391 // CHECK: store i24 [[OLD_VAL]], i24* [[TEMP:%.+]],
    392 // CHECK: [[TRUNC:%.+]] = trunc i32 [[NEW_VAL]] to i24
    393 // CHECK: [[BF_AND:%.+]] = and i24 [[TRUNC]], 16383
    394 // CHECK: [[BF_VALUE:%.+]] = shl i24 [[BF_AND]], 3
    395 // CHECK: [[BF_CLEAR:%.+]] = and i24 %{{.+}}, -131065
    396 // CHECK: or i24 [[BF_CLEAR]], [[BF_VALUE]]
    397 // CHECK: store i24 %{{.+}}, i24* [[TEMP]]
    398 // CHECK: [[BITCAST_TEMP_OLD_BF_ADDR:%.+]] = bitcast i24* [[LDTEMP]] to i8*
    399 // CHECK: [[BITCAST_TEMP_NEW_BF_ADDR:%.+]] = bitcast i24* [[TEMP]] to i8*
    400 // 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)
    401 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
    402 // CHECK: [[EXIT]]
    403 #pragma omp atomic write
    404   bfx3_packed.a = ldv;
    405 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
    406 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32
    407 // CHECK: [[PREV_VALUE:%.+]] = load atomic i64, i64* bitcast (%struct.BitFields4* @{{.+}} to i64*) monotonic
    408 // CHECK: br label %[[CONT:.+]]
    409 // CHECK: [[CONT]]
    410 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i64 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
    411 // CHECK: [[ZEXT:%.+]] = zext i32 [[NEW_VAL]] to i64
    412 // CHECK: [[BF_AND:%.+]] = and i64 [[ZEXT]], 1
    413 // CHECK: [[BF_VALUE:%.+]] = shl i64 [[BF_AND]], 16
    414 // CHECK: [[BF_CLEAR:%.+]] = and i64 %{{.+}}, -65537
    415 // CHECK: or i64 [[BF_CLEAR]], [[BF_VALUE]]
    416 // CHECK: store i64 %{{.+}}, i64* [[LDTEMP:%.+]]
    417 // CHECK: [[NEW_BF_VALUE:%.+]] = load i64, i64* [[LDTEMP]]
    418 // CHECK: [[RES:%.+]] = cmpxchg i64* bitcast (%struct.BitFields4* @{{.+}} to i64*), i64 [[OLD_BF_VALUE]], i64 [[NEW_BF_VALUE]] monotonic monotonic
    419 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i64, i1 } [[RES]], 0
    420 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i64, i1 } [[RES]], 1
    421 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
    422 // CHECK: [[EXIT]]
    423 #pragma omp atomic write
    424   bfx4.a = ldv;
    425 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
    426 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i32
    427 // CHECK: [[PREV_VALUE:%.+]] = load atomic i8, i8* getelementptr inbounds (%struct.BitFields4_packed, %struct.BitFields4_packed* @{{.+}}, i32 0, i32 0, i64 2) monotonic
    428 // CHECK: br label %[[CONT:.+]]
    429 // CHECK: [[CONT]]
    430 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i8 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
    431 // CHECK: [[TRUNC:%.+]] = trunc i32 [[NEW_VAL]] to i8
    432 // CHECK: [[BF_VALUE:%.+]] = and i8 [[TRUNC]], 1
    433 // CHECK: [[BF_CLEAR:%.+]] = and i8 %{{.+}}, -2
    434 // CHECK: or i8 [[BF_CLEAR]], [[BF_VALUE]]
    435 // CHECK: store i8 %{{.+}}, i8* [[LDTEMP:%.+]]
    436 // CHECK: [[NEW_BF_VALUE:%.+]] = load i8, i8* [[LDTEMP]]
    437 // 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
    438 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i8, i1 } [[RES]], 0
    439 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i8, i1 } [[RES]], 1
    440 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
    441 // CHECK: [[EXIT]]
    442 #pragma omp atomic write
    443   bfx4_packed.a = ldv;
    444 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
    445 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i64
    446 // CHECK: [[PREV_VALUE:%.+]] = load atomic i64, i64* bitcast (%struct.BitFields4* @{{.+}} to i64*) monotonic
    447 // CHECK: br label %[[CONT:.+]]
    448 // CHECK: [[CONT]]
    449 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i64 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
    450 // CHECK: [[BF_AND:%.+]] = and i64 [[NEW_VAL]], 127
    451 // CHECK: [[BF_VALUE:%.+]] = shl i64 [[BF_AND]], 17
    452 // CHECK: [[BF_CLEAR:%.+]] = and i64 %{{.+}}, -16646145
    453 // CHECK: or i64 [[BF_CLEAR]], [[BF_VALUE]]
    454 // CHECK: store i64 %{{.+}}, i64* [[LDTEMP:%.+]]
    455 // CHECK: [[NEW_BF_VALUE:%.+]] = load i64, i64* [[LDTEMP]]
    456 // CHECK: [[RES:%.+]] = cmpxchg i64* bitcast (%struct.BitFields4* @{{.+}} to i64*), i64 [[OLD_BF_VALUE]], i64 [[NEW_BF_VALUE]] monotonic monotonic
    457 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i64, i1 } [[RES]], 0
    458 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i64, i1 } [[RES]], 1
    459 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
    460 // CHECK: [[EXIT]]
    461 #pragma omp atomic write
    462   bfx4.b = ldv;
    463 // CHECK: load x86_fp80, x86_fp80* @{{.+}}
    464 // CHECK: [[NEW_VAL:%.+]] = fptosi x86_fp80 %{{.+}} to i64
    465 // CHECK: [[PREV_VALUE:%.+]] = load atomic i8, i8* getelementptr inbounds (%struct.BitFields4_packed, %struct.BitFields4_packed* @{{.+}}, i32 0, i32 0, i64 2) monotonic
    466 // CHECK: br label %[[CONT:.+]]
    467 // CHECK: [[CONT]]
    468 // CHECK: [[OLD_BF_VALUE:%.+]] = phi i8 [ [[PREV_VALUE]], %[[EXIT]] ], [ [[FAILED_OLD_VAL:%.+]], %[[CONT]] ]
    469 // CHECK: [[TRUNC:%.+]] = trunc i64 [[NEW_VAL]] to i8
    470 // CHECK: [[BF_AND:%.+]] = and i8 [[TRUNC]], 127
    471 // CHECK: [[BF_VALUE:%.+]] = shl i8 [[BF_AND]], 1
    472 // CHECK: [[BF_CLEAR:%.+]] = and i8 %{{.+}}, 1
    473 // CHECK: or i8 [[BF_CLEAR]], [[BF_VALUE]]
    474 // CHECK: store i8 %{{.+}}, i8* [[LDTEMP:%.+]]
    475 // CHECK: [[NEW_BF_VALUE:%.+]] = load i8, i8* [[LDTEMP]]
    476 // 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
    477 // CHECK: [[FAILED_OLD_VAL]] = extractvalue { i8, i1 } [[RES]], 0
    478 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i8, i1 } [[RES]], 1
    479 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
    480 // CHECK: [[EXIT]]
    481 #pragma omp atomic write
    482   bfx4_packed.b = ldv;
    483 // CHECK: load i64, i64*
    484 // CHECK: [[VEC_ITEM_VAL:%.+]] = uitofp i64 %{{.+}} to float
    485 // CHECK: [[I64VAL:%.+]] = load atomic i64, i64* bitcast (<2 x float>* [[DEST:@.+]] to i64*) monotonic
    486 // CHECK: br label %[[CONT:.+]]
    487 // CHECK: [[CONT]]
    488 // CHECK: [[OLD_I64:%.+]] = phi i64 [ [[I64VAL]], %{{.+}} ], [ [[FAILED_I64_OLD_VAL:%.+]], %[[CONT]] ]
    489 // CHECK: [[BITCAST:%.+]] = bitcast <2 x float>* [[LDTEMP:%.+]] to i64*
    490 // CHECK: store i64 [[OLD_I64]], i64* [[BITCAST]],
    491 // CHECK: [[VEC_VAL:%.+]] = load <2 x float>, <2 x float>* [[LDTEMP]]
    492 // CHECK: [[NEW_VEC_VAL:%.+]] = insertelement <2 x float> [[VEC_VAL]], float [[VEC_ITEM_VAL]], i64 0
    493 // CHECK: store <2 x float> [[NEW_VEC_VAL]], <2 x float>* [[LDTEMP]]
    494 // CHECK: [[NEW_I64:%.+]] = load i64, i64* [[BITCAST]]
    495 // CHECK: [[RES:%.+]] = cmpxchg i64* bitcast (<2 x float>* [[DEST]] to i64*), i64 [[OLD_I64]], i64 [[NEW_I64]] monotonic monotonic
    496 // CHECK: [[FAILED_I64_OLD_VAL:%.+]] = extractvalue { i64, i1 } [[RES]], 0
    497 // CHECK: [[FAIL_SUCCESS:%.+]] = extractvalue { i64, i1 } [[RES]], 1
    498 // CHECK: br i1 [[FAIL_SUCCESS]], label %[[EXIT:.+]], label %[[CONT]]
    499 // CHECK: [[EXIT]]
    500 #pragma omp atomic write
    501   float2x.x = ulv;
    502 // CHECK: call i32 @llvm.read_register.i32(
    503 // CHECK: sitofp i32 %{{.+}} to double
    504 // CHECK: bitcast double %{{.+}} to i64
    505 // CHECK: store atomic i64 %{{.+}}, i64* bitcast (double* @{{.+}} to i64*) seq_cst
    506 // CHECK: call{{.*}} @__kmpc_flush(
    507 #pragma omp atomic write seq_cst
    508   dv = rix;
    509   return 0;
    510 }
    511 
    512 #endif
    513