clang/test/OpenMP/simd_codegen.cpp

   1 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - -fopenmp-version=45 | FileCheck %s
   2 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s -fopenmp-version=45
   3 // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -fopenmp-version=45 | FileCheck %s
   4 // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - -fopenmp-version=45 | FileCheck %s --check-prefix=TERM_DEBUG
   5 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - -fopenmp-version=50 -DOMP5 | FileCheck %s --check-prefix=CHECK --check-prefix=OMP5 --check-prefix=OMP5RT
   6 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s -fopenmp-version=50 -DOMP5
   7 // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -fopenmp-version=50 -DOMP5 | FileCheck %s --check-prefix=CHECK --check-prefix=OMP5 --check-prefix=OMP5RT
   8
   9 // RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - -fopenmp-version=45 | FileCheck %s
  10 // RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s -fopenmp-version=45
  11 // RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -fopenmp-version=45 | FileCheck %s
  12 // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - -fopenmp-version=45 | FileCheck --check-prefix=TERM_DEBUG %s
  13 // RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - -fopenmp-version=50 -DOMP5 | FileCheck %s --check-prefix=CHECK --check-prefix=OMP5
  14 // RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s -fopenmp-version=50 -DOMP5
  15 // RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -fopenmp-version=50 -DOMP5 | FileCheck %s --check-prefix=CHECK --check-prefix=OMP5
  16 // expected-no-diagnostics
  17  #ifndef HEADER
  18  #define HEADER
  19
  20 #ifdef OMP5
  21 #define CONDITIONAL conditional :
  22 #else
  23 #define CONDITIONAL
  24 #endif //OMP5
  25 // CHECK: [[SS_TY:%.+]] = type { i32 }
  26 // OMP5-DAG: [[LAST_IV:@.+]] = {{.*}}common global i64 0
  27 // OMP5-DAG: [[LAST_A:@.+]] = {{.*}}common global i32 0
  28
  29 long long get_val() { extern void mayThrow(); mayThrow(); return 0; }
  30 double *g_ptr;
  31
  32 struct S {
  33   int a, b;
  34 };
  35
  36 // CHECK-LABEL: define {{.*void}} @{{.*}}simple{{.*}}(ptr noundef {{.+}}, ptr noundef {{.+}}, ptr noundef {{.+}}, ptr noundef {{.+}})
  37 void simple(float *a, float *b, float *c, float *d) {
  38   S s, *p;
  39 #ifdef OMP5
  40   #pragma omp simd if (simd: true) nontemporal(a, b, c, d, s)
  41 #else
  42   #pragma omp simd
  43 #endif
  44   // CHECK: store i32 0, ptr [[OMP_IV:%[^,]+]]
  45
  46 // CHECK: [[IV:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
  47 // CHECK-NEXT: [[CMP:%.+]] = icmp slt i32 [[IV]], 6
  48 // CHECK-NEXT: br i1 [[CMP]], label %[[SIMPLE_LOOP1_BODY:.+]], label %[[SIMPLE_LOOP1_END:[^,]+]]
  49   for (int i = 3; i < 32; i += 5) {
  50 // CHECK: [[SIMPLE_LOOP1_BODY]]:
  51 // Start of body: calculate i from IV:
  52 // CHECK: [[IV1_1:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
  53 // CHECK: [[CALC_I_1:%.+]] = mul nsw i32 [[IV1_1]], 5
  54 // CHECK-NEXT: [[CALC_I_2:%.+]] = add nsw i32 3, [[CALC_I_1]]
  55 // CHECK-NEXT: store i32 [[CALC_I_2]], ptr [[LC_I:.+]]{{.*}}!llvm.access.group
  56 // ... loop body ...
  57 // End of body: store into a[i]:
  58 // OMP45-NOT: load ptr,{{.*}}!nontemporal
  59 // CHECK-NOT: load float,{{.*}}!nontemporal
  60 // OMP5: load ptr,{{.*}}!nontemporal
  61 // OMP5: load ptr,{{.*}}!nontemporal
  62 // OMP5: load ptr,{{.*}}!nontemporal
  63 // OMP5: load i32,{{.*}}!nontemporal
  64 // OMP5-NOT: load i32,{{.*}}!nontemporal
  65 // OMP5: load ptr,{{.*}}!nontemporal
  66 // CHECK-NOT: load float,{{.*}}!nontemporal
  67 // CHECK: store float [[RESULT:%.+]], ptr {{%.+}}{{.*}}!llvm.access.group
  68     a[i] = b[i] * c[i] * d[i] + s.a + p->a;
  69 // CHECK: [[IV1_2:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
  70 // CHECK-NEXT: [[ADD1_2:%.+]] = add nsw i32 [[IV1_2]], 1
  71 // CHECK-NEXT: store i32 [[ADD1_2]], ptr [[OMP_IV]]{{.*}}!llvm.access.group
  72 // br label %{{.+}}, !llvm.loop !{{.+}}
  73   }
  74 // CHECK: [[SIMPLE_LOOP1_END]]:
  75
  76   long long k = get_val();
  77
  78   #pragma omp simd linear(k : 3)
  79 // CHECK: [[K0:%.+]] = call {{.*}}i64 @{{.*}}get_val
  80 // CHECK-NEXT: store i64 [[K0]], ptr [[K_VAR:%[^,]+]]
  81 // CHECK: store i32 0, ptr [[OMP_IV2:%[^,]+]]
  82 // CHECK: [[K0LOAD:%.+]] = load i64, ptr [[K_VAR]]
  83 // CHECK-NEXT: store i64 [[K0LOAD]], ptr [[LIN0:%[^,]+]]
  84
  85 // CHECK: [[IV2:%.+]] = load i32, ptr [[OMP_IV2]]{{.*}}!llvm.access.group
  86 // CHECK-NEXT: [[CMP2:%.+]] = icmp slt i32 [[IV2]], 9
  87 // CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP2_BODY:.+]], label %[[SIMPLE_LOOP2_END:[^,]+]]
  88   for (int i = 10; i > 1; i--) {
  89 // CHECK: [[SIMPLE_LOOP2_BODY]]:
  90 // Start of body: calculate i from IV:
  91 // CHECK: [[IV2_0:%.+]] = load i32, ptr [[OMP_IV2]]{{.*}}!llvm.access.group
  92 // FIXME: It is interesting, why the following "mul 1" was not constant folded?
  93 // CHECK-NEXT: [[IV2_1:%.+]] = mul nsw i32 [[IV2_0]], 1
  94 // CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV2_1]]
  95 // CHECK-NEXT: store i32 [[LC_I_1]], ptr {{.+}}, !llvm.access.group
  96 //
  97 // CHECK-NEXT: [[LIN0_1:%.+]] = load i64, ptr [[LIN0]]{{.*}}!llvm.access.group
  98 // CHECK-NEXT: [[IV2_2:%.+]] = load i32, ptr [[OMP_IV2]]{{.*}}!llvm.access.group
  99 // CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV2_2]], 3
 100 // CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64
 101 // CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]]
 102 // Update of the privatized version of linear variable!
 103 // CHECK-NEXT: store i64 [[LIN_ADD1]], ptr [[K_PRIVATIZED:%[^,]+]]
 104     a[k]++;
 105     k = k + 3;
 106 // CHECK: [[IV2_2:%.+]] = load i32, ptr [[OMP_IV2]]{{.*}}!llvm.access.group
 107 // CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV2_2]], 1
 108 // CHECK-NEXT: store i32 [[ADD2_2]], ptr [[OMP_IV2]]{{.*}}!llvm.access.group
 109 // br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP2_ID]]
 110   }
 111 // CHECK: [[SIMPLE_LOOP2_END]]:
 112 //
 113 // Update linear vars after loop, as the loop was operating on a private version.
 114 // CHECK: [[LIN0_2:%.+]] = load i64, ptr [[K_PRIVATIZED]],
 115 // CHECK-NEXT: store i64 [[LIN0_2]], ptr [[K_VAR]]
 116 //
 117
 118   int lin = 12;
 119   #pragma omp simd linear(lin : get_val()), linear(g_ptr)
 120
 121 // Init linear private var.
 122 // CHECK: store i32 12, ptr [[LIN_VAR:%[^,]+]]
 123 // CHECK: store i64 0, ptr [[OMP_IV3:%[^,]+]]
 124
 125 // CHECK: [[LIN_LOAD:%.+]] = load i32, ptr [[LIN_VAR]]
 126 // CHECK-NEXT: store i32 [[LIN_LOAD]], ptr [[LIN_START:%[^,]+]]
 127 // Remember linear step.
 128 // CHECK: [[CALL_VAL:%.+]] = invoke
 129 // CHECK: store i64 [[CALL_VAL]], ptr [[LIN_STEP:%[^,]+]]
 130
 131 // CHECK: [[GLIN_LOAD:%.+]] = load ptr, ptr [[GLIN_VAR:@[^,]+]]
 132 // CHECK-NEXT: store ptr [[GLIN_LOAD]], ptr [[GLIN_START:%[^,]+]]
 133
 134 // CHECK: [[IV3:%.+]] = load i64, ptr [[OMP_IV3]]{{.*}}!llvm.access.group
 135 // CHECK-NEXT: [[CMP3:%.+]] = icmp ult i64 [[IV3]], 4
 136 // CHECK-NEXT: br i1 [[CMP3]], label %[[SIMPLE_LOOP3_BODY:.+]], label %[[SIMPLE_LOOP3_END:[^,]+]]
 137   for (unsigned long long it = 2000; it >= 600; it-=400) {
 138 // CHECK: [[SIMPLE_LOOP3_BODY]]:
 139 // Start of body: calculate it from IV:
 140 // CHECK: [[IV3_0:%.+]] = load i64, ptr [[OMP_IV3]]{{.*}}!llvm.access.group
 141 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul i64 [[IV3_0]], 400
 142 // CHECK-NEXT: [[LC_IT_2:%.+]] = sub i64 2000, [[LC_IT_1]]
 143 // CHECK-NEXT: store i64 [[LC_IT_2]], ptr {{.+}}, !llvm.access.group
 144 //
 145 // Linear start and step are used to calculate current value of the linear variable.
 146 // CHECK: [[LINSTART:.+]] = load i32, ptr [[LIN_START]]{{.*}}!llvm.access.group
 147 // CHECK: [[LINSTEP:.+]] = load i64, ptr [[LIN_STEP]]{{.*}}!llvm.access.group
 148 // CHECK-NOT: store i32 {{.+}}, ptr [[LIN_VAR]],{{.*}}!llvm.access.group
 149 // CHECK: store i32 {{.+}}, ptr [[LIN_PRIV:%.+]],{{.*}}!llvm.access.group
 150 // CHECK: [[GLINSTART:.+]] = load ptr, ptr [[GLIN_START]]{{.*}}!llvm.access.group
 151 // CHECK-NEXT: [[IV3_1:%.+]] = load i64, ptr [[OMP_IV3]]{{.*}}!llvm.access.group
 152 // CHECK-NEXT: [[MUL:%.+]] = mul i64 [[IV3_1]], 1
 153 // CHECK: [[GEP:%.+]] = getelementptr{{.*}}[[GLINSTART]]
 154 // CHECK-NEXT: store ptr [[GEP]], ptr [[G_PTR_CUR:%[^,]+]]{{.*}}!llvm.access.group
 155     *g_ptr++ = 0.0;
 156 // CHECK: [[GEP_VAL:%.+]] = load ptr, ptr [[G_PTR_CUR]]{{.*}}!llvm.access.group
 157 // CHECK: store double{{.*}}[[GEP_VAL]]{{.*}}!llvm.access.group
 158     a[it + lin]++;
 159 // CHECK: [[FLT_INC:%.+]] = fadd float
 160 // CHECK-NEXT: store float [[FLT_INC]],{{.*}}!llvm.access.group
 161 // CHECK: [[IV3_2:%.+]] = load i64, ptr [[OMP_IV3]]{{.*}}!llvm.access.group
 162 // CHECK-NEXT: [[ADD3_2:%.+]] = add i64 [[IV3_2]], 1
 163 // CHECK-NEXT: store i64 [[ADD3_2]], ptr [[OMP_IV3]]{{.*}}!llvm.access.group
 164   }
 165 // CHECK: [[SIMPLE_LOOP3_END]]:
 166 //
 167 // Linear start and step are used to calculate final value of the linear variables.
 168 // CHECK: [[LIN:%.+]] = load i32, ptr [[LIN_PRIV]]
 169 // CHECK-NEXT: store i32 [[LIN]], ptr [[LIN_VAR]],
 170 // CHECK: [[GLIN:%.+]] = load ptr, ptr [[G_PTR_CUR]]
 171 // CHECK-NEXT: store ptr [[GLIN]], ptr [[GLIN_VAR]]
 172
 173   #pragma omp simd
 174 // CHECK: store i32 0, ptr [[OMP_IV4:%[^,]+]]
 175
 176 // CHECK: [[IV4:%.+]] = load i32, ptr [[OMP_IV4]]{{.*}}!llvm.access.group
 177 // CHECK-NEXT: [[CMP4:%.+]] = icmp slt i32 [[IV4]], 4
 178 // CHECK-NEXT: br i1 [[CMP4]], label %[[SIMPLE_LOOP4_BODY:.+]], label %[[SIMPLE_LOOP4_END:[^,]+]]
 179   for (short it = 6; it <= 20; it-=-4) {
 180 // CHECK: [[SIMPLE_LOOP4_BODY]]:
 181 // Start of body: calculate it from IV:
 182 // CHECK: [[IV4_0:%.+]] = load i32, ptr [[OMP_IV4]]{{.*}}!llvm.access.group
 183 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul nsw i32 [[IV4_0]], 4
 184 // CHECK-NEXT: [[LC_IT_2:%.+]] = add nsw i32 6, [[LC_IT_1]]
 185 // CHECK-NEXT: [[LC_IT_3:%.+]] = trunc i32 [[LC_IT_2]] to i16
 186 // CHECK-NEXT: store i16 [[LC_IT_3]], ptr {{.+}}, !llvm.access.group
 187
 188 // CHECK: [[IV4_2:%.+]] = load i32, ptr [[OMP_IV4]]{{.*}}!llvm.access.group
 189 // CHECK-NEXT: [[ADD4_2:%.+]] = add nsw i32 [[IV4_2]], 1
 190 // CHECK-NEXT: store i32 [[ADD4_2]], ptr [[OMP_IV4]]{{.*}}!llvm.access.group
 191   }
 192 // CHECK: [[SIMPLE_LOOP4_END]]:
 193
 194   #pragma omp simd
 195 // CHECK: store i32 0, ptr [[OMP_IV5:%[^,]+]]
 196
 197 // CHECK: [[IV5:%.+]] = load i32, ptr [[OMP_IV5]]{{.*}}!llvm.access.group
 198 // CHECK-NEXT: [[CMP5:%.+]] = icmp slt i32 [[IV5]], 26
 199 // CHECK-NEXT: br i1 [[CMP5]], label %[[SIMPLE_LOOP5_BODY:.+]], label %[[SIMPLE_LOOP5_END:[^,]+]]
 200   for (unsigned char it = 'z'; it >= 'a'; it+=-1) {
 201 // CHECK: [[SIMPLE_LOOP5_BODY]]:
 202 // Start of body: calculate it from IV:
 203 // CHECK: [[IV5_0:%.+]] = load i32, ptr [[OMP_IV5]]{{.*}}!llvm.access.group
 204 // CHECK-NEXT: [[IV5_1:%.+]] = mul nsw i32 [[IV5_0]], 1
 205 // CHECK-NEXT: [[LC_IT_1:%.+]] = sub nsw i32 122, [[IV5_1]]
 206 // CHECK-NEXT: [[LC_IT_2:%.+]] = trunc i32 [[LC_IT_1]] to i8
 207 // CHECK-NEXT: store i8 [[LC_IT_2]], ptr {{.+}}, !llvm.access.group
 208
 209 // CHECK: [[IV5_2:%.+]] = load i32, ptr [[OMP_IV5]]{{.*}}!llvm.access.group
 210 // CHECK-NEXT: [[ADD5_2:%.+]] = add nsw i32 [[IV5_2]], 1
 211 // CHECK-NEXT: store i32 [[ADD5_2]], ptr [[OMP_IV5]]{{.*}}!llvm.access.group
 212   }
 213 // CHECK: [[SIMPLE_LOOP5_END]]:
 214
 215 // CHECK-NOT: mul i32 %{{.+}}, 10
 216   #pragma omp simd
 217   for (unsigned i=100; i<10; i+=10) {
 218   }
 219
 220   int A;
 221   // CHECK: store i32 -1, ptr [[A:%.+]],
 222   A = -1;
 223   #pragma omp simd lastprivate(CONDITIONAL A)
 224 // CHECK: store i64 0, ptr [[OMP_IV7:%[^,]+]]
 225 // CHECK: br label %[[SIMD_LOOP7_COND:[^,]+]]
 226 // CHECK: [[SIMD_LOOP7_COND]]:
 227 // CHECK-NEXT: [[IV7:%.+]] = load i64, ptr [[OMP_IV7]]{{.*}}!llvm.access.group
 228 // CHECK-NEXT: [[CMP7:%.+]] = icmp slt i64 [[IV7]], 7
 229 // CHECK-NEXT: br i1 [[CMP7]], label %[[SIMPLE_LOOP7_BODY:.+]], label %[[SIMPLE_LOOP7_END:[^,]+]]
 230   for (long long i = -10; i < 10; i += 3) {
 231 // CHECK: [[SIMPLE_LOOP7_BODY]]:
 232 // Start of body: calculate i from IV:
 233 // CHECK: [[IV7_0:%.+]] = load i64, ptr [[OMP_IV7]]{{.*}}!llvm.access.group
 234 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul nsw i64 [[IV7_0]], 3
 235 // CHECK-NEXT: [[LC_IT_2:%.+]] = add nsw i64 -10, [[LC_IT_1]]
 236 // CHECK-NEXT: store i64 [[LC_IT_2]], ptr [[LC:%[^,]+]],{{.+}}!llvm.access.group
 237 // CHECK-NEXT: [[LC_VAL:%.+]] = load i64, ptr [[LC]]{{.+}}!llvm.access.group
 238 // CHECK-NEXT: [[A_VAL:%.+]] = load i32, ptr [[A_PRIV:%[^,]+]],{{.+}}!llvm.access.group
 239 // CHECK-NEXT: [[CAST:%.+]] = sext i32 [[A_VAL]] to i64
 240 // CHECK-NEXT: [[ADD:%.+]] = add nsw i64 [[CAST]], [[LC_VAL]]
 241 // CHECK-NEXT: [[CONV:%.+]] = trunc i64 [[ADD]] to i32
 242 // CHECK-NEXT: store i32 [[CONV]], ptr [[A_PRIV]],{{.+}}!llvm.access.group
 243 // OMP5-NEXT: [[IV:%.+]] = load i64, ptr [[OMP_IV7]],{{.+}}!llvm.access.group
 244 // OMP5RT:    call void @__kmpc_critical(ptr {{.+}}, i32 [[GTID:%.+]], ptr [[A_REGION:@.+]]),{{.+}}!llvm.access.group
 245 // OMP5-NEXT: [[LAST_IV_VAL:%.+]] = load i64, ptr [[LAST_IV]],{{.+}}!llvm.access.group
 246 // OMP5-NEXT: [[CMP:%.+]] = icmp sle i64 [[LAST_IV_VAL]], [[IV]]
 247 // OMP5-NEXT: br i1 [[CMP]], label %[[LP_THEN:.+]], label %[[LP_DONE:[^,]+]]
 248 // OMP5:      [[LP_THEN]]:
 249 // OMP5-NEXT: store i64 [[IV]], ptr [[LAST_IV]],{{.+}}!llvm.access.group
 250 // OMP5-NEXT: [[A_VAL:%.+]] = load i32, ptr [[A_PRIV]],{{.+}}!llvm.access.group
 251 // OMP5-NEXT: store i32 [[A_VAL]], ptr [[LAST_A]],{{.+}}!llvm.access.group
 252 // OMP5-NEXT: br label %[[LP_DONE]]
 253 // OMP5:      [[LP_DONE]]:
 254 // OMP5RT-NEXT: call void @__kmpc_end_critical(ptr {{.+}}, i32 [[GTID]], ptr [[A_REGION]]),{{.+}}!llvm.access.group
 255     A += i;
 256 // CHECK: [[IV7_2:%.+]] = load i64, ptr [[OMP_IV7]]{{.*}}!llvm.access.group
 257 // CHECK-NEXT: [[ADD7_2:%.+]] = add nsw i64 [[IV7_2]], 1
 258 // CHECK-NEXT: store i64 [[ADD7_2]], ptr [[OMP_IV7]]{{.*}}!llvm.access.group
 259   }
 260 // CHECK: [[SIMPLE_LOOP7_END]]:
 261 // CHECK-NEXT: store i64 11, ptr
 262 // OMP5-NEXT: [[LAST_A_VAL:%.+]] = load i32, ptr [[LAST_A]],
 263 // OMP5-NEXT: store i32 [[LAST_A_VAL]], ptr [[A_PRIV]],
 264 // CHECK-NEXT: [[A_PRIV_VAL:%.+]] = load i32, ptr [[A_PRIV]],
 265 // CHECK-NEXT: store i32 [[A_PRIV_VAL]], ptr [[A]],
 266   int R;
 267   // CHECK: store i32 -1, ptr [[R:%[^,]+]],
 268   R = -1;
 269 // CHECK: store i64 0, ptr [[OMP_IV8:%[^,]+]],
 270 // CHECK: store i32 1, ptr [[R_PRIV:%[^,]+]],
 271 #ifdef OMP5
 272   #pragma omp simd reduction(*:R) if(A)
 273 #else
 274   #pragma omp simd reduction(*:R)
 275 #endif
 276 // OMP5:      [[A_VAL:%.+]] = load i32, ptr [[A]],
 277 // OMP5-NEXT: [[COND:%.+]] = icmp ne i32 [[A_VAL]], 0
 278 // OMP5-NEXT: br i1 [[COND]], label {{%?}}[[THEN:[^,]+]], label {{%?}}[[ELSE:[^,]+]]
 279 // OMP5:      [[THEN]]:
 280
 281 // CHECK: br label %[[SIMD_LOOP8_COND:[^,]+]]
 282 // CHECK: [[SIMD_LOOP8_COND]]:
 283 // CHECK-NEXT: [[IV8:%.+]] = load i64, ptr [[OMP_IV8]]{{.*}}!llvm.access.group
 284 // CHECK-NEXT: [[CMP8:%.+]] = icmp slt i64 [[IV8]], 7
 285 // CHECK-NEXT: br i1 [[CMP8]], label %[[SIMPLE_LOOP8_BODY:.+]], label %[[SIMPLE_LOOP8_END:[^,]+]]
 286   for (long long i = -10; i < 10; i += 3) {
 287 // CHECK: [[SIMPLE_LOOP8_BODY]]:
 288 // Start of body: calculate i from IV:
 289 // CHECK: [[IV8_0:%.+]] = load i64, ptr [[OMP_IV8]]{{.*}}!llvm.access.group
 290 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul nsw i64 [[IV8_0]], 3
 291 // CHECK-NEXT: [[LC_IT_2:%.+]] = add nsw i64 -10, [[LC_IT_1]]
 292 // CHECK-NEXT: store i64 [[LC_IT_2]], ptr [[LC:%[^,]+]],{{.+}}!llvm.access.group
 293 // CHECK-NEXT: [[LC_VAL:%.+]] = load i64, ptr [[LC]]{{.+}}!llvm.access.group
 294 // CHECK: store i32 %{{.+}}, ptr [[R_PRIV]],{{.+}}!llvm.access.group
 295     R *= i;
 296 // CHECK: [[IV8_2:%.+]] = load i64, ptr [[OMP_IV8]]{{.*}}!llvm.access.group
 297 // CHECK-NEXT: [[ADD8_2:%.+]] = add nsw i64 [[IV8_2]], 1
 298 // CHECK-NEXT: store i64 [[ADD8_2]], ptr [[OMP_IV8]]{{.*}}!llvm.access.group
 299   }
 300 // CHECK: [[SIMPLE_LOOP8_END]]:
 301 // OMP5: br label {{%?}}[[EXIT:[^,]+]]
 302 // OMP5: br label %[[SIMD_LOOP8_COND:[^,]+]]
 303 // OMP5: [[SIMD_LOOP8_COND]]:
 304 // OMP5-NEXT: [[IV8:%.+]] = load i64, ptr [[OMP_IV8]],{{[^!]*}}
 305 // OMP5-NEXT: [[CMP8:%.+]] = icmp slt i64 [[IV8]], 7
 306 // OMP5-NEXT: br i1 [[CMP8]], label %[[SIMPLE_LOOP8_BODY:.+]], label %[[SIMPLE_LOOP8_END:[^,]+]]
 307 // OMP5: [[SIMPLE_LOOP8_BODY]]:
 308 // Start of body: calculate i from IV:
 309 // OMP5: [[IV8_0:%.+]] = load i64, ptr [[OMP_IV8]],{{[^!]*}}
 310 // OMP5-NEXT: [[LC_IT_1:%.+]] = mul nsw i64 [[IV8_0]], 3
 311 // OMP5-NEXT: [[LC_IT_2:%.+]] = add nsw i64 -10, [[LC_IT_1]]
 312 // OMP5-NEXT: store i64 [[LC_IT_2]], ptr [[LC:%[^,]+]],{{[^!]*}}
 313 // OMP5-NEXT: [[LC_VAL:%.+]] = load i64, ptr [[LC]],{{[^!]*}}
 314 // OMP5: store i32 %{{.+}}, ptr [[R_PRIV]],{{[^!]*}}
 315 // OMP5: [[IV8_2:%.+]] = load i64, ptr [[OMP_IV8]],{{[^!]*}}
 316 // OMP5-NEXT: [[ADD8_2:%.+]] = add nsw i64 [[IV8_2]], 1
 317 // OMP5-NEXT: store i64 [[ADD8_2]], ptr [[OMP_IV8]],{{[^!]*}}
 318 // OMP5:      br label {{%?}}[[SIMD_LOOP8_COND]], {{.*}}!llvm.loop ![[DISABLE_VECT:.+]]
 319 // OMP5: [[SIMPLE_LOOP8_END]]:
 320 // OMP5: br label {{%?}}[[EXIT]]
 321 // OMP5: [[EXIT]]:
 322
 323 // CHECK-DAG: [[R_VAL:%.+]] = load i32, ptr [[R]],
 324 // CHECK-DAG: [[R_PRIV_VAL:%.+]] = load i32, ptr [[R_PRIV]],
 325 // CHECK: [[RED:%.+]] = mul nsw i32 [[R_VAL]], [[R_PRIV_VAL]]
 326 // CHECK-NEXT: store i32 [[RED]], ptr [[R]],
 327 // CHECK-NEXT: ret void
 328 }
 329
 330 template <class T, unsigned K> T tfoo(T a) { return a + K; }
 331
 332 template <typename T, unsigned N>
 333 int templ1(T a, T *z) {
 334   #pragma omp simd collapse(N)
 335   for (int i = 0; i < N * 2; i++) {
 336     for (long long j = 0; j < (N + N + N + N); j += 2) {
 337       z[i + j] = a + tfoo<T, N>(i + j);
 338     }
 339   }
 340   return 0;
 341 }
 342
 343 // Instatiation templ1<float,2>
 344 // CHECK-LABEL: define {{.*i32}} @{{.*}}templ1{{.*}}(float noundef {{.+}}, ptr noundef {{.+}})
 345 // CHECK: store i64 0, ptr [[T1_OMP_IV:[^,]+]]
 346 // ...
 347 // CHECK: [[IV:%.+]] = load i64, ptr [[T1_OMP_IV]]{{.*}}!llvm.access.group
 348 // CHECK-NEXT: [[CMP1:%.+]] = icmp slt i64 [[IV]], 16
 349 // CHECK-NEXT: br i1 [[CMP1]], label %[[T1_BODY:.+]], label %[[T1_END:[^,]+]]
 350 // CHECK: [[T1_BODY]]:
 351 // Loop counters i and j updates:
 352 // CHECK: [[IV1:%.+]] = load i64, ptr [[T1_OMP_IV]]{{.*}}!llvm.access.group
 353 // CHECK-NEXT: [[I_1:%.+]] = sdiv i64 [[IV1]], 4
 354 // CHECK-NEXT: [[I_1_MUL1:%.+]] = mul nsw i64 [[I_1]], 1
 355 // CHECK-NEXT: [[I_1_ADD0:%.+]] = add nsw i64 0, [[I_1_MUL1]]
 356 // CHECK-NEXT: [[I_2:%.+]] = trunc i64 [[I_1_ADD0]] to i32
 357 // CHECK-NEXT: store i32 [[I_2]], ptr {{%.+}}{{.*}}!llvm.access.group
 358 // CHECK: [[IV2:%.+]] = load i64, ptr [[T1_OMP_IV]]{{.*}}!llvm.access.group
 359 // CHECK: [[IV2_1:%.+]] = load i64, ptr [[T1_OMP_IV]]{{.*}}!llvm.access.group
 360 // CHECK-NEXT: [[J_1_DIV1:%.+]] = sdiv i64 [[IV2_1]], 4
 361 // CHECK-NEXT: [[J_1_MUL1:%.+]] = mul nsw i64 [[J_1_DIV1]], 4
 362 // CHECK-NEXT: [[J_1_SUB0:%.+]] = sub nsw i64 [[IV2]], [[J_1_MUL1]]
 363 // CHECK-NEXT: [[J_2:%.+]] = mul nsw i64 [[J_1_SUB0]], 2
 364 // CHECK-NEXT: [[J_2_ADD0:%.+]] = add nsw i64 0, [[J_2]]
 365 // CHECK-NEXT: store i64 [[J_2_ADD0]], ptr {{%.+}}{{.*}}!llvm.access.group
 366 // simd.for.inc:
 367 // CHECK: [[IV3:%.+]] = load i64, ptr [[T1_OMP_IV]]{{.*}}!llvm.access.group
 368 // CHECK-NEXT: [[INC:%.+]] = add nsw i64 [[IV3]], 1
 369 // CHECK-NEXT: store i64 [[INC]], ptr [[T1_OMP_IV]]{{.*}}!llvm.access.group
 370 // CHECK-NEXT: br label {{%.+}}
 371 // CHECK: [[T1_END]]:
 372 // CHECK: ret i32 0
 373 //
 374 void inst_templ1() {
 375   float a;
 376   float z[100];
 377   templ1<float,2> (a, z);
 378 }
 379
 380
 381 typedef int MyIdx;
 382
 383 class IterDouble {
 384   double *Ptr;
 385 public:
 386   IterDouble operator++ () const {
 387     IterDouble n;
 388     n.Ptr = Ptr + 1;
 389     return n;
 390   }
 391   bool operator < (const IterDouble &that) const {
 392     return Ptr < that.Ptr;
 393   }
 394   double & operator *() const {
 395     return *Ptr;
 396   }
 397   MyIdx operator - (const IterDouble &that) const {
 398     return (MyIdx) (Ptr - that.Ptr);
 399   }
 400   IterDouble operator + (int Delta) {
 401     IterDouble re;
 402     re.Ptr = Ptr + Delta;
 403     return re;
 404   }
 405
 406   ///~IterDouble() {}
 407 };
 408
 409 // CHECK-LABEL: define {{.*void}} @{{.*}}iter_simple{{.*}}
 410 void iter_simple(IterDouble ia, IterDouble ib, IterDouble ic) {
 411 //
 412 // Calculate number of iterations before the loop body.
 413 // CHECK: [[DIFF1:%.+]] = invoke {{.*}}i32 @{{.*}}IterDouble{{.*}}
 414 // CHECK: [[DIFF2:%.+]] = sub nsw i32 [[DIFF1]], 1
 415 // CHECK-NEXT: [[DIFF3:%.+]] = add nsw i32 [[DIFF2]], 1
 416 // CHECK-NEXT: [[DIFF4:%.+]] = sdiv i32 [[DIFF3]], 1
 417 // CHECK-NEXT: [[DIFF5:%.+]] = sub nsw i32 [[DIFF4]], 1
 418 // CHECK-NEXT: store i32 [[DIFF5]], ptr [[OMP_LAST_IT:%[^,]+]]{{.+}}
 419 // CHECK: store i32 0, ptr [[IT_OMP_IV:%[^,]+]]
 420   #pragma omp simd
 421
 422 // CHECK: [[IV:%.+]] = load i32, ptr [[IT_OMP_IV]]{{.+}} !llvm.access.group
 423 // CHECK-NEXT: [[LAST_IT:%.+]] = load i32, ptr [[OMP_LAST_IT]]{{.+}}!llvm.access.group
 424 // CHECK-NEXT: [[NUM_IT:%.+]] = add nsw i32 [[LAST_IT]], 1
 425 // CHECK-NEXT: [[CMP:%.+]] = icmp slt i32 [[IV]], [[NUM_IT]]
 426 // CHECK-NEXT: br i1 [[CMP]], label %[[IT_BODY:[^,]+]], label %[[IT_END:[^,]+]]
 427   for (IterDouble i = ia; i < ib; ++i) {
 428 // CHECK: [[IT_BODY]]:
 429 // Start of body: calculate i from index:
 430 // CHECK: [[IV1:%.+]] = load i32, ptr [[IT_OMP_IV]]{{.+}}!llvm.access.group
 431 // Call of operator+ (i, IV).
 432 // CHECK: {{%.+}} = invoke {{.+}} @{{.*}}IterDouble{{.*}}
 433 // ... loop body ...
 434    *i = *ic * 0.5;
 435 // Float multiply and save result.
 436 // CHECK: [[MULR:%.+]] = fmul double {{%.+}}, 5.000000e-01
 437 // CHECK-NEXT: invoke {{.+}} @{{.*}}IterDouble{{.*}}
 438 // CHECK: store double [[MULR:%.+]], ptr [[RESULT_ADDR:%.+]], !llvm.access.group
 439    ++ic;
 440 //
 441 // CHECK: [[IV2:%.+]] = load i32, ptr [[IT_OMP_IV]]{{.+}}!llvm.access.group
 442 // CHECK-NEXT: [[ADD2:%.+]] = add nsw i32 [[IV2]], 1
 443 // CHECK-NEXT: store i32 [[ADD2]], ptr [[IT_OMP_IV]]{{.+}}!llvm.access.group
 444 // br label %{{.*}}, !llvm.loop ![[ITER_LOOP_ID]]
 445   }
 446 // CHECK: [[IT_END]]:
 447 // CHECK: ret void
 448 }
 449
 450
 451 // CHECK-LABEL: define {{.*void}} @{{.*}}collapsed{{.*}}
 452 void collapsed(float *a, float *b, float *c, float *d) {
 453   int i; // outer loop counter
 454   unsigned j; // middle loop couter, leads to unsigned icmp in loop header.
 455   // k declared in the loop init below
 456   short l; // inner loop counter
 457 // CHECK: store i32 0, ptr [[OMP_IV:[^,]+]]
 458 //
 459   #pragma omp simd collapse(4)
 460
 461 // CHECK: [[IV:%.+]] = load i32, ptr [[OMP_IV]]{{.+}}!llvm.access.group
 462 // CHECK-NEXT: [[CMP:%.+]] = icmp ult i32 [[IV]], 120
 463 // CHECK-NEXT: br i1 [[CMP]], label %[[COLL1_BODY:[^,]+]], label %[[COLL1_END:[^,]+]]
 464   for (i = 1; i < 3; i++) // 2 iterations
 465     for (j = 2u; j < 5u; j++) //3 iterations
 466       for (int k = 3; k <= 6; k++) // 4 iterations
 467         for (l = 4; l < 9; ++l) // 5 iterations
 468         {
 469 // CHECK: [[COLL1_BODY]]:
 470 // Start of body: calculate i from index:
 471 // CHECK: [[IV1:%.+]] = load i32, ptr [[OMP_IV]]{{.+}}!llvm.access.group
 472 // Calculation of the loop counters values.
 473 // CHECK: [[CALC_I_1:%.+]] = udiv i32 [[IV1]], 60
 474 // CHECK-NEXT: [[CALC_I_1_MUL1:%.+]] = mul i32 [[CALC_I_1]], 1
 475 // CHECK-NEXT: [[CALC_I_2:%.+]] = add i32 1, [[CALC_I_1_MUL1]]
 476 // CHECK-NEXT: store i32 [[CALC_I_2]], ptr [[LC_I:.+]]
 477
 478 // CHECK: [[IV1_2:%.+]] = load i32, ptr [[OMP_IV]]{{.+}}!llvm.access.group
 479 // CHECK: [[IV1_2_1:%.+]] = load i32, ptr [[OMP_IV]]{{.+}}!llvm.access.group
 480 // CHECK-NEXT: [[CALC_J_1:%.+]] = udiv i32 [[IV1_2_1]], 60
 481 // CHECK-NEXT: [[MUL_1:%.+]] = mul i32 [[CALC_J_1]], 60
 482 // CHECK-NEXT: [[SUB_3:%.+]] = sub i32 [[IV1_2]], [[MUL_1]]
 483 // CHECK-NEXT: [[CALC_J_2:%.+]] = udiv i32 [[SUB_3]], 20
 484 // CHECK-NEXT: [[CALC_J_2_MUL1:%.+]] = mul i32 [[CALC_J_2]], 1
 485 // CHECK-NEXT: [[CALC_J_3:%.+]] = add i32 2, [[CALC_J_2_MUL1]]
 486 // CHECK-NEXT: store i32 [[CALC_J_3]], ptr [[LC_J:.+]]
 487
 488 // CHECK: [[IV1_3:%.+]] = load i32, ptr [[OMP_IV]]{{.+}}!llvm.access.group
 489 // CHECK: [[IV1_3_1:%.+]] = load i32, ptr [[OMP_IV]]{{.+}}!llvm.access.group
 490 // CHECK-NEXT: [[DIV_1:%.+]] = udiv i32 [[IV1_3_1]], 60
 491 // CHECK-NEXT: [[MUL_2:%.+]] = mul i32 [[DIV_1]], 60
 492 // CHECK-NEXT: [[ADD_3:%.+]] = sub i32 [[IV1_3]], [[MUL_2]]
 493
 494 // CHECK: [[IV1_4:%.+]] = load i32, ptr [[OMP_IV]]
 495 // CHECK: [[IV1_4_1:%.+]] = load i32, ptr [[OMP_IV]]
 496 // CHECK-NEXT: [[DIV_2:%.+]] = udiv i32 [[IV1_4_1]], 60
 497 // CHECK-NEXT: [[MUL_3:%.+]] = mul i32 [[DIV_2]], 60
 498 // CHECK-NEXT: [[SUB_6:%.+]] = sub i32 [[IV1_4]], [[MUL_3]]
 499 // CHECK-NEXT: [[DIV_3:%.+]] = udiv i32 [[SUB_6]], 20
 500 // CHECK-NEXT: [[MUL_4:%.+]] = mul i32 [[DIV_3]], 20
 501 // CHECK-NEXT: [[ADD_5:%.+]] = sub i32 [[ADD_3]], [[MUL_4]]
 502 // CHECK-NEXT: [[DIV_4:%.+]] = udiv i32 [[ADD_5]], 5
 503 // CHECK-NEXT: [[MUL_5:%.+]] = mul i32 [[DIV_4]], 1
 504 // CHECK-NEXT: [[ADD_6:%.+]] = add i32 3, [[MUL_5]]
 505 // CHECK-NEXT: store i32 [[ADD_6]], ptr [[LC_K:.+]]
 506
 507 // CHECK: [[IV1_5:%.+]] = load i32, ptr [[OMP_IV]]{{.+}}!llvm.access.group
 508 // CHECK: [[IV1_5_1:%.+]] = load i32, ptr [[OMP_IV]]{{.+}}!llvm.access.group
 509 // CHECK-NEXT: [[DIV_5:%.+]] = udiv i32 [[IV1_5_1]], 60
 510 // CHECK-NEXT: [[MUL_6:%.+]] = mul i32 [[DIV_5]], 60
 511 // CHECK-NEXT: [[SUB_7:%.+]] = sub i32 [[IV1_5]], [[MUL_6]]
 512
 513 // CHECK: [[IV1_6:%.+]] = load i32, ptr [[OMP_IV]]
 514 // CHECK: [[IV1_6_1:%.+]] = load i32, ptr [[OMP_IV]]
 515 // CHECK-NEXT: [[DIV_6:%.+]] = udiv i32 [[IV1_6_1]], 60
 516 // CHECK-NEXT: [[MUL_7:%.+]] = mul i32 [[DIV_6]], 60
 517 // CHECK-NEXT: [[SUB_10:%.+]] = sub i32 [[IV1_6]], [[MUL_7]]
 518 // CHECK-NEXT: [[DIV_7:%.+]] = udiv i32 [[SUB_10]], 20
 519 // CHECK-NEXT: [[MUL_8:%.+]] = mul i32 [[DIV_7]], 20
 520 // CHECK-NEXT: [[SUB_11:%.+]] = sub i32 [[SUB_7]], [[MUL_8]]
 521
 522 // CHECK: [[IV1_7:%.+]] = load i32, ptr [[OMP_IV]]
 523 // CHECK: [[IV1_7_1:%.+]] = load i32, ptr [[OMP_IV]]
 524 // CHECK-NEXT: [[DIV_8:%.+]] = udiv i32 [[IV1_7_1]], 60
 525 // CHECK-NEXT: [[MUL_9:%.+]] = mul i32 [[DIV_8]], 60
 526 // CHECK-NEXT: [[SUB_12:%.+]] = sub i32 [[IV1_7]], [[MUL_9]]
 527
 528 // CHECK: [[IV1_8:%.+]] = load i32, ptr [[OMP_IV]]
 529 // CHECK: [[IV1_8_1:%.+]] = load i32, ptr [[OMP_IV]]
 530 // CHECK-NEXT: [[DIV_3:%.+]] = udiv i32 [[IV1_8_1]], 60
 531 // CHECK-NEXT: [[MUL_4:%.+]] = mul i32 [[DIV_3]], 60
 532 // CHECK-NEXT: [[SUB_7:%.+]] = sub i32 [[IV1_8]], [[MUL_4]]
 533 // CHECK-NEXT: [[DIV_4:%.+]] = udiv i32 [[SUB_7]], 20
 534 // CHECK-NEXT: [[MUL_5:%.+]] = mul i32 [[DIV_4]], 20
 535 // CHECK-NEXT: [[SUB_8:%.+]] = sub i32 [[SUB_12]], [[MUL_5]]
 536 // CHECK-NEXT: [[DIV_5:%.+]] = udiv i32 [[SUB_8]], 5
 537 // CHECK-NEXT: [[MUL_6:%.+]] = mul i32 [[DIV_5]], 5
 538 // CHECK-NEXT: [[SUB_9:%.+]] = sub i32 [[SUB_11]], [[MUL_6]]
 539 // CHECK-NEXT: [[MUL_6:%.+]] = mul i32 [[SUB_9]], 1
 540 // CHECK-NEXT: [[CALC_L_2:%.+]] = add i32 4, [[MUL_6]]
 541 // CHECK-NEXT: [[CALC_L_3:%.+]] = trunc i32 [[CALC_L_2]] to i16
 542 // CHECK-NEXT: store i16 [[CALC_L_3]], ptr [[LC_L:.+]]
 543 // ... loop body ...
 544 // End of body: store into a[i]:
 545 // CHECK: store float [[RESULT:%.+]], ptr [[RESULT_ADDR:%.+]]{{.+}}!llvm.access.group
 546     float res = b[j] * c[k];
 547     a[i] = res * d[l];
 548 // CHECK: [[IV2:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 549 // CHECK-NEXT: [[ADD2:%.+]] = add i32 [[IV2]], 1
 550 // CHECK-NEXT: store i32 [[ADD2]], ptr [[OMP_IV]]{{.*}}!llvm.access.group
 551 // br label %{{[^,]+}}, !llvm.loop ![[COLL1_LOOP_ID]]
 552 // CHECK: [[COLL1_END]]:
 553   }
 554 // i,j,l are updated; k is not updated.
 555 // CHECK: store i32 3, ptr
 556 // CHECK-NEXT: store i32 5, ptr
 557 // CHECK-NEXT: store i32 7, ptr
 558 // CHECK-NEXT: store i16 9, ptr
 559 // CHECK: ret void
 560 }
 561
 562 extern char foo();
 563 extern double globalfloat;
 564
 565 // CHECK-LABEL: define {{.*void}} @{{.*}}widened{{.*}}
 566 void widened(float *a, float *b, float *c, float *d) {
 567   int i; // outer loop counter
 568   short j; // inner loop counter
 569   globalfloat = 1.0;
 570   int localint = 1;
 571 // CHECK: store double {{.+}}, ptr [[GLOBALFLOAT:@.+]]
 572 // Counter is widened to 64 bits.
 573 // CHECK: store i64 0, ptr [[OMP_IV:[^,]+]]
 574 //
 575   #pragma omp simd collapse(2) private(globalfloat, localint)
 576
 577 // CHECK: [[IV:%.+]] = load i64, ptr [[OMP_IV]]{{.+}}!llvm.access.group
 578 // CHECK-NEXT: [[LI:%.+]] = load i64, ptr [[OMP_LI:%[^,]+]]{{.+}}!llvm.access.group
 579 // CHECK-NEXT: [[NUMIT:%.+]] = add nsw i64 [[LI]], 1
 580 // CHECK-NEXT: [[CMP:%.+]] = icmp slt i64 [[IV]], [[NUMIT]]
 581 // CHECK-NEXT: br i1 [[CMP]], label %[[WIDE1_BODY:[^,]+]], label %[[WIDE1_END:[^,]+]]
 582   for (i = 1; i < 3; i++) // 2 iterations
 583     for (j = 0; j < foo(); j++) // foo() iterations
 584   {
 585 // CHECK: [[WIDE1_BODY]]:
 586 // Start of body: calculate i from index:
 587 // CHECK: [[IV1:%.+]] = load i64, ptr [[OMP_IV]]{{.+}}!llvm.access.group
 588 // Calculation of the loop counters values...
 589 // CHECK: store i32 {{[^,]+}}, ptr [[LC_I:.+]]
 590 // CHECK: [[IV1_2:%.+]] = load i64, ptr [[OMP_IV]]{{.+}}!llvm.access.group
 591 // CHECK: store i16 {{[^,]+}}, ptr [[LC_J:.+]]
 592 // ... loop body ...
 593 //
 594 // Here we expect store into private double var, not global
 595 // CHECK-NOT: store double {{.+}}, ptr [[GLOBALFLOAT]]
 596     globalfloat = (float)j/i;
 597     float res = b[j] * c[j];
 598 // Store into a[i]:
 599 // CHECK: store float [[RESULT:%.+]], ptr [[RESULT_ADDR:%.+]]{{.+}}!llvm.access.group
 600     a[i] = res * d[i];
 601 // Then there's a store into private var localint:
 602 // CHECK: store i32 {{.+}}, ptr [[LOCALINT:%[^,]+]]{{.+}}!llvm.access.group
 603     localint = (int)j;
 604 // CHECK: [[IV2:%.+]] = load i64, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 605 // CHECK-NEXT: [[ADD2:%.+]] = add nsw i64 [[IV2]], 1
 606 // CHECK-NEXT: store i64 [[ADD2]], ptr [[OMP_IV]]{{.*}}!llvm.access.group
 607 //
 608 // br label %{{[^,]+}}, !llvm.loop ![[WIDE1_LOOP_ID]]
 609 // CHECK: [[WIDE1_END]]:
 610   }
 611 // i,j are updated.
 612 // CHECK: store i32 3, ptr [[I:%[^,]+]]
 613 // CHECK: store i16
 614 //
 615 // Here we expect store into original localint, not its privatized version.
 616 // CHECK-NOT: store i32 {{.+}}, ptr [[LOCALINT]]
 617   localint = (int)j;
 618 // CHECK: ret void
 619 }
 620
 621 // CHECK-LABEL: define {{.*void}} @{{.*}}linear{{.*}}(ptr noundef {{.+}})
 622 void linear(float *a) {
 623   // CHECK: [[VAL_ADDR:%.+]] = alloca i64,
 624   // CHECK: [[K_ADDR:%.+]] = alloca ptr,
 625   long long val = 0;
 626   long long &k = val;
 627
 628   #pragma omp simd linear(k : 3)
 629 // CHECK: store ptr [[VAL_ADDR]], ptr [[K_ADDR]],
 630 // CHECK: [[VAL_REF:%.+]] = load ptr, ptr [[K_ADDR]],
 631 // CHECK: store ptr [[VAL_REF]], ptr [[K_ADDR_REF:%.+]],
 632 // CHECK: store i32 0, ptr [[OMP_IV:%[^,]+]]
 633 // CHECK: [[K_REF:%.+]] = load ptr, ptr [[K_ADDR_REF]],
 634 // CHECK: [[K0LOAD:%.+]] = load i64, ptr [[K_REF]]
 635 // CHECK-NEXT: store i64 [[K0LOAD]], ptr [[LIN0:%[^,]+]]
 636
 637 // CHECK: [[IV:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 638 // CHECK-NEXT: [[CMP2:%.+]] = icmp slt i32 [[IV]], 9
 639 // CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP_BODY:.+]], label %[[SIMPLE_LOOP_END:[^,]+]]
 640   for (int i = 10; i > 1; i--) {
 641 // CHECK: [[SIMPLE_LOOP_BODY]]:
 642 // Start of body: calculate i from IV:
 643 // CHECK: [[IV_0:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 644 // FIXME: It is interesting, why the following "mul 1" was not constant folded?
 645 // CHECK-NEXT: [[IV_1:%.+]] = mul nsw i32 [[IV_0]], 1
 646 // CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV_1]]
 647 // CHECK-NEXT: store i32 [[LC_I_1]], ptr {{.+}}, !llvm.access.group
 648 //
 649 // CHECK-NEXT: [[LIN0_1:%.+]] = load i64, ptr [[LIN0]]{{.*}}!llvm.access.group
 650 // CHECK-NEXT: [[IV_2:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 651 // CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV_2]], 3
 652 // CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64
 653 // CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]]
 654 // Update of the privatized version of linear variable!
 655 // CHECK-NEXT: store i64 [[LIN_ADD1]], ptr [[K_PRIVATIZED:%[^,]+]]
 656     a[k]++;
 657     k = k + 3;
 658 // CHECK: [[IV_2:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 659 // CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV_2]], 1
 660 // CHECK-NEXT: store i32 [[ADD2_2]], ptr [[OMP_IV]]{{.*}}!llvm.access.group
 661 // br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP_ID]]
 662   }
 663 // CHECK: [[SIMPLE_LOOP_END]]:
 664 //
 665 // Update linear vars after loop, as the loop was operating on a private version.
 666 // CHECK: [[K_REF:%.+]] = load ptr, ptr [[K_ADDR_REF]],
 667 // CHECK: store ptr [[K_REF]], ptr [[K_PRIV_REF:%.+]],
 668 // CHECK: [[LIN0_2:%.+]] = load i64, ptr [[K_PRIVATIZED]]
 669 // CHECK-NEXT: [[K_REF:%.+]] = load ptr, ptr [[K_PRIV_REF]],
 670 // CHECK-NEXT: store i64 [[LIN0_2]], ptr [[K_REF]]
 671 //
 672
 673   #pragma omp simd linear(val(k) : 3)
 674 // CHECK: [[VAL_REF:%.+]] = load ptr, ptr [[K_ADDR]],
 675 // CHECK: store ptr [[VAL_REF]], ptr [[K_ADDR_REF:%.+]],
 676 // CHECK: store i32 0, ptr [[OMP_IV:%[^,]+]]
 677 // CHECK: [[K_REF:%.+]] = load ptr, ptr [[K_ADDR_REF]],
 678 // CHECK: [[K0LOAD:%.+]] = load i64, ptr [[K_REF]]
 679 // CHECK-NEXT: store i64 [[K0LOAD]], ptr [[LIN0:%[^,]+]]
 680
 681 // CHECK: [[IV:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 682 // CHECK-NEXT: [[CMP2:%.+]] = icmp slt i32 [[IV]], 9
 683 // CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP_BODY:.+]], label %[[SIMPLE_LOOP_END:[^,]+]]
 684   for (int i = 10; i > 1; i--) {
 685 // CHECK: [[SIMPLE_LOOP_BODY]]:
 686 // Start of body: calculate i from IV:
 687 // CHECK: [[IV_0:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 688 // FIXME: It is interesting, why the following "mul 1" was not constant folded?
 689 // CHECK-NEXT: [[IV_1:%.+]] = mul nsw i32 [[IV_0]], 1
 690 // CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV_1]]
 691 // CHECK-NEXT: store i32 [[LC_I_1]], ptr {{.+}}, !llvm.access.group
 692 //
 693 // CHECK-NEXT: [[LIN0_1:%.+]] = load i64, ptr [[LIN0]]{{.*}}!llvm.access.group
 694 // CHECK-NEXT: [[IV_2:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 695 // CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV_2]], 3
 696 // CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64
 697 // CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]]
 698 // Update of the privatized version of linear variable!
 699 // CHECK-NEXT: store i64 [[LIN_ADD1]], ptr [[K_PRIVATIZED:%[^,]+]]
 700     a[k]++;
 701     k = k + 3;
 702 // CHECK: [[IV_2:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 703 // CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV_2]], 1
 704 // CHECK-NEXT: store i32 [[ADD2_2]], ptr [[OMP_IV]]{{.*}}!llvm.access.group
 705 // br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP_ID]]
 706   }
 707 // CHECK: [[SIMPLE_LOOP_END]]:
 708 //
 709 // Update linear vars after loop, as the loop was operating on a private version.
 710 // CHECK: [[K_REF:%.+]] = load ptr, ptr [[K_ADDR_REF]],
 711 // CHECK: store ptr [[K_REF]], ptr [[K_PRIV_REF:%.+]],
 712 // CHECK: [[LIN0_2:%.+]] = load i64, ptr [[K_PRIVATIZED]]
 713 // CHECK-NEXT: [[K_REF:%.+]] = load ptr, ptr [[K_PRIV_REF]],
 714 // CHECK-NEXT: store i64 [[LIN0_2]], ptr [[K_REF]]
 715 //
 716   #pragma omp simd linear(uval(k) : 3)
 717 // CHECK: store i32 0, ptr [[OMP_IV:%[^,]+]]
 718 // CHECK: [[K0LOAD:%.+]] = load i64, ptr [[VAL_ADDR]]
 719 // CHECK-NEXT: store i64 [[K0LOAD]], ptr [[LIN0:%[^,]+]]
 720
 721 // CHECK: [[IV:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 722 // CHECK-NEXT: [[CMP2:%.+]] = icmp slt i32 [[IV]], 9
 723 // CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP_BODY:.+]], label %[[SIMPLE_LOOP_END:[^,]+]]
 724   for (int i = 10; i > 1; i--) {
 725 // CHECK: [[SIMPLE_LOOP_BODY]]:
 726 // Start of body: calculate i from IV:
 727 // CHECK: [[IV_0:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 728 // FIXME: It is interesting, why the following "mul 1" was not constant folded?
 729 // CHECK-NEXT: [[IV_1:%.+]] = mul nsw i32 [[IV_0]], 1
 730 // CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV_1]]
 731 // CHECK-NEXT: store i32 [[LC_I_1]], ptr {{.+}}, !llvm.access.group
 732 //
 733 // CHECK-NEXT: [[LIN0_1:%.+]] = load i64, ptr [[LIN0]]{{.*}}!llvm.access.group
 734 // CHECK-NEXT: [[IV_2:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 735 // CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV_2]], 3
 736 // CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64
 737 // CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]]
 738 // Update of the privatized version of linear variable!
 739 // CHECK-NEXT: store i64 [[LIN_ADD1]], ptr [[K_PRIVATIZED:%[^,]+]]
 740     a[k]++;
 741     k = k + 3;
 742 // CHECK: [[IV_2:%.+]] = load i32, ptr [[OMP_IV]]{{.*}}!llvm.access.group
 743 // CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV_2]], 1
 744 // CHECK-NEXT: store i32 [[ADD2_2]], ptr [[OMP_IV]]{{.*}}!llvm.access.group
 745 // br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP_ID]]
 746   }
 747 // CHECK: [[SIMPLE_LOOP_END]]:
 748 //
 749 // Update linear vars after loop, as the loop was operating on a private version.
 750 // CHECK: [[LIN0_2:%.+]] = load i64, ptr [[K_PRIVATIZED]]
 751 // CHECK-NEXT: store i64 [[LIN0_2]], ptr [[VAL_ADDR]]
 752 //
 753 }
 754
 755 #ifdef OMP5
 756 // OMP5-LABEL: inner_simd
 757 void inner_simd() {
 758   double a, b;
 759 #pragma omp simd nontemporal(a)
 760   for (int i = 0; i < 10; ++i) {
 761 #pragma omp simd nontemporal(b)
 762     for (int k = 0; k < 10; ++k) {
 763       // OMP5: load double,{{.*}}!nontemporal
 764       // OMP5: store double{{.*}}!nontemporal
 765       a = b;
 766     }
 767     // OMP5-NOT: load double,{{.*}}!nontemporal
 768     // OMP5: load double,
 769     // OMP5: store double{{.*}}!nontemporal
 770     a = b;
 771   }
 772 }
 773
 774 extern struct T t;
 775 struct Base {
 776   float a;
 777 };
 778 struct T : public Base {
 779   void foo() {
 780 #pragma omp simd nontemporal(Base::a)
 781     for (int i = 0; i < 10; ++i) {
 782     // OMP5: store float{{.*}}!nontemporal
 783     // OMP5-NOT: nontemporal
 784     // OMP5-NEXT: store float
 785       Base::a = 0;
 786       t.a = 0;
 787     }
 788   }
 789 } t;
 790
 791 void bartfoo() {
 792   t.foo();
 793 }
 794
 795 #endif // OMP5
 796 // TERM_DEBUG-LABEL: bar
 797 int bar() { extern void mayThrow(); mayThrow(); return 0; };
 798
 799 // TERM_DEBUG-LABEL: parallel_simd
 800 void parallel_simd(float *a) {
 801 #pragma omp parallel
 802 #pragma omp simd
 803   // TERM_DEBUG-NOT: __kmpc_global_thread_num
 804   // TERM_DEBUG:     invoke noundef i32 {{.*}}bar{{.*}}()
 805   // TERM_DEBUG:     unwind label %[[TERM_LPAD:[^,]+]],
 806   // TERM_DEBUG-NOT: __kmpc_global_thread_num
 807   // TERM_DEBUG:     [[TERM_LPAD]]
 808   // TERM_DEBUG:     call void @__clang_call_terminate
 809   // TERM_DEBUG:     unreachable
 810   for (unsigned i = 131071; i <= 2147483647; i += 127)
 811     a[i] += bar();
 812 }
 813 // TERM_DEBUG: !{{[0-9]+}} = !DILocation(line: [[@LINE-11]],
 814
 815 // CHECK-LABEL: S8
 816 // CHECK-DAG: call void @llvm.assume(i1
 817 // CHECK-DAG: call void @llvm.assume(i1
 818 // CHECK-DAG: call void @llvm.assume(i1
 819 // CHECK-DAG: call void @llvm.assume(i1
 820 struct SS {
 821   SS(): a(0) {}
 822   SS(int v) : a(v) {}
 823   int a;
 824   typedef int type;
 825 };
 826
 827 template <typename T>
 828 class S7 : public T {
 829 protected:
 830   T *a;
 831   T b[2];
 832   S7() : a(0) {}
 833
 834 public:
 835   S7(typename T::type &v) : a((T*)&v) {
 836 #pragma omp simd aligned(a)
 837     for (int k = 0; k < a->a; ++k)
 838       ++this->a->a;
 839 #pragma omp simd aligned(this->b : 8)
 840     for (int k = 0; k < a->a; ++k)
 841       ++a->a;
 842   }
 843 };
 844
 845 class S8 : private IterDouble, public S7<SS> {
 846   S8() {}
 847
 848 public:
 849   S8(int v) : S7<SS>(v){
 850 #pragma omp parallel private(a)
 851 #pragma omp simd aligned(S7<SS>::a)
 852     for (int k = 0; k < a->a; ++k)
 853       ++this->a->a;
 854 #pragma omp parallel shared(b)
 855 #pragma omp simd aligned(this->b: 4)
 856     for (int k = 0; k < a->a; ++k)
 857       ++a->a;
 858   }
 859 };
 860 S8 s8(0);
 861
 862 // TERM_DEBUG-NOT: line: 0,
 863 // TERM_DEBUG: distinct !DISubprogram(linkageName: "_GLOBAL__sub_I_simd_codegen.cpp",
 864 // OMP5-DAG: ![[NOVECT:.+]] = !{!"llvm.loop.vectorize.enable", i1 false}
 865 // OMP5-DAG: ![[DISABLE_VECT]] = distinct !{{.*}}![[NOVECT]]{{[,}]}}
 866 #endif // HEADER
 867