1 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | 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 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 - | 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 - | 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=OMP50 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=OMP50 8 9 // RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | 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 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 - | 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 - | 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=OMP50 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=OMP50 16 // expected-no-diagnostics 17 #ifndef HEADER 18 #define HEADER 19 20 // CHECK: [[SS_TY:%.+]] = type { i32 } 21 22 long long get_val() { return 0; } 23 double *g_ptr; 24 25 // CHECK-LABEL: define {{.*void}} @{{.*}}simple{{.*}}(float* {{.+}}, float* {{.+}}, float* {{.+}}, float* {{.+}}) 26 void simple(float *a, float *b, float *c, float *d) { 27 #ifdef OMP5 28 #pragma omp simd if (simd: true) 29 #else 30 #pragma omp simd 31 #endif 32 // CHECK: store i32 0, i32* [[OMP_IV:%[^,]+]] 33 34 // CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group 35 // CHECK-NEXT: [[CMP:%.+]] = icmp slt i32 [[IV]], 6 36 // CHECK-NEXT: br i1 [[CMP]], label %[[SIMPLE_LOOP1_BODY:.+]], label %[[SIMPLE_LOOP1_END:[^,]+]] 37 for (int i = 3; i < 32; i += 5) { 38 // CHECK: [[SIMPLE_LOOP1_BODY]]: 39 // Start of body: calculate i from IV: 40 // CHECK: [[IV1_1:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group 41 // CHECK: [[CALC_I_1:%.+]] = mul nsw i32 [[IV1_1]], 5 42 // CHECK-NEXT: [[CALC_I_2:%.+]] = add nsw i32 3, [[CALC_I_1]] 43 // CHECK-NEXT: store i32 [[CALC_I_2]], i32* [[LC_I:.+]]{{.*}}!llvm.access.group 44 // ... loop body ... 45 // End of body: store into a[i]: 46 // CHECK: store float [[RESULT:%.+]], float* {{%.+}}{{.*}}!llvm.access.group 47 a[i] = b[i] * c[i] * d[i]; 48 // CHECK: [[IV1_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group 49 // CHECK-NEXT: [[ADD1_2:%.+]] = add nsw i32 [[IV1_2]], 1 50 // CHECK-NEXT: store i32 [[ADD1_2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group 51 // br label %{{.+}}, !llvm.loop !{{.+}} 52 } 53 // CHECK: [[SIMPLE_LOOP1_END]]: 54 55 long long k = get_val(); 56 57 #pragma omp simd linear(k : 3) 58 // CHECK: [[K0:%.+]] = call {{.*}}i64 @{{.*}}get_val 59 // CHECK-NEXT: store i64 [[K0]], i64* [[K_VAR:%[^,]+]] 60 // CHECK: store i32 0, i32* [[OMP_IV2:%[^,]+]] 61 // CHECK: [[K0LOAD:%.+]] = load i64, i64* [[K_VAR]] 62 // CHECK-NEXT: store i64 [[K0LOAD]], i64* [[LIN0:%[^,]+]] 63 64 // CHECK: [[IV2:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.access.group 65 // CHECK-NEXT: [[CMP2:%.+]] = icmp slt i32 [[IV2]], 9 66 // CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP2_BODY:.+]], label %[[SIMPLE_LOOP2_END:[^,]+]] 67 for (int i = 10; i > 1; i--) { 68 // CHECK: [[SIMPLE_LOOP2_BODY]]: 69 // Start of body: calculate i from IV: 70 // CHECK: [[IV2_0:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.access.group 71 // FIXME: It is interesting, why the following "mul 1" was not constant folded? 72 // CHECK-NEXT: [[IV2_1:%.+]] = mul nsw i32 [[IV2_0]], 1 73 // CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV2_1]] 74 // CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.access.group 75 // 76 // CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.access.group 77 // CHECK-NEXT: [[IV2_2:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.access.group 78 // CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV2_2]], 3 79 // CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64 80 // CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]] 81 // Update of the privatized version of linear variable! 82 // CHECK-NEXT: store i64 [[LIN_ADD1]], i64* [[K_PRIVATIZED:%[^,]+]] 83 a[k]++; 84 k = k + 3; 85 // CHECK: [[IV2_2:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.access.group 86 // CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV2_2]], 1 87 // CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV2]]{{.*}}!llvm.access.group 88 // br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP2_ID]] 89 } 90 // CHECK: [[SIMPLE_LOOP2_END]]: 91 // 92 // Update linear vars after loop, as the loop was operating on a private version. 93 // CHECK: [[LIN0_2:%.+]] = load i64, i64* [[LIN0]] 94 // CHECK-NEXT: [[LIN_ADD2:%.+]] = add nsw i64 [[LIN0_2]], 27 95 // CHECK-NEXT: store i64 [[LIN_ADD2]], i64* [[K_VAR]] 96 // 97 98 int lin = 12; 99 #pragma omp simd linear(lin : get_val()), linear(g_ptr) 100 101 // Init linear private var. 102 // CHECK: store i32 12, i32* [[LIN_VAR:%[^,]+]] 103 // CHECK: store i64 0, i64* [[OMP_IV3:%[^,]+]] 104 105 // CHECK: [[LIN_LOAD:%.+]] = load i32, i32* [[LIN_VAR]] 106 // CHECK-NEXT: store i32 [[LIN_LOAD]], i32* [[LIN_START:%[^,]+]] 107 // Remember linear step. 108 // CHECK: [[CALL_VAL:%.+]] = invoke 109 // CHECK: store i64 [[CALL_VAL]], i64* [[LIN_STEP:%[^,]+]] 110 111 // CHECK: [[GLIN_LOAD:%.+]] = load double*, double** [[GLIN_VAR:@[^,]+]] 112 // CHECK-NEXT: store double* [[GLIN_LOAD]], double** [[GLIN_START:%[^,]+]] 113 114 // CHECK: [[IV3:%.+]] = load i64, i64* [[OMP_IV3]]{{.*}}!llvm.access.group 115 // CHECK-NEXT: [[CMP3:%.+]] = icmp ult i64 [[IV3]], 4 116 // CHECK-NEXT: br i1 [[CMP3]], label %[[SIMPLE_LOOP3_BODY:.+]], label %[[SIMPLE_LOOP3_END:[^,]+]] 117 for (unsigned long long it = 2000; it >= 600; it-=400) { 118 // CHECK: [[SIMPLE_LOOP3_BODY]]: 119 // Start of body: calculate it from IV: 120 // CHECK: [[IV3_0:%.+]] = load i64, i64* [[OMP_IV3]]{{.*}}!llvm.access.group 121 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul i64 [[IV3_0]], 400 122 // CHECK-NEXT: [[LC_IT_2:%.+]] = sub i64 2000, [[LC_IT_1]] 123 // CHECK-NEXT: store i64 [[LC_IT_2]], i64* {{.+}}, !llvm.access.group 124 // 125 // Linear start and step are used to calculate current value of the linear variable. 126 // CHECK: [[LINSTART:.+]] = load i32, i32* [[LIN_START]]{{.*}}!llvm.access.group 127 // CHECK: [[LINSTEP:.+]] = load i64, i64* [[LIN_STEP]]{{.*}}!llvm.access.group 128 // CHECK-NOT: store i32 {{.+}}, i32* [[LIN_VAR]],{{.*}}!llvm.access.group 129 // CHECK: [[GLINSTART:.+]] = load double*, double** [[GLIN_START]]{{.*}}!llvm.access.group 130 // CHECK-NEXT: [[IV3_1:%.+]] = load i64, i64* [[OMP_IV3]]{{.*}}!llvm.access.group 131 // CHECK-NEXT: [[MUL:%.+]] = mul i64 [[IV3_1]], 1 132 // CHECK: [[GEP:%.+]] = getelementptr{{.*}}[[GLINSTART]] 133 // CHECK-NEXT: store double* [[GEP]], double** [[G_PTR_CUR:%[^,]+]]{{.*}}!llvm.access.group 134 *g_ptr++ = 0.0; 135 // CHECK: [[GEP_VAL:%.+]] = load double{{.*}}[[G_PTR_CUR]]{{.*}}!llvm.access.group 136 // CHECK: store double{{.*}}[[GEP_VAL]]{{.*}}!llvm.access.group 137 a[it + lin]++; 138 // CHECK: [[FLT_INC:%.+]] = fadd float 139 // CHECK-NEXT: store float [[FLT_INC]],{{.*}}!llvm.access.group 140 // CHECK: [[IV3_2:%.+]] = load i64, i64* [[OMP_IV3]]{{.*}}!llvm.access.group 141 // CHECK-NEXT: [[ADD3_2:%.+]] = add i64 [[IV3_2]], 1 142 // CHECK-NEXT: store i64 [[ADD3_2]], i64* [[OMP_IV3]]{{.*}}!llvm.access.group 143 } 144 // CHECK: [[SIMPLE_LOOP3_END]]: 145 // 146 // Linear start and step are used to calculate final value of the linear variables. 147 // CHECK: [[LINSTART:.+]] = load i32, i32* [[LIN_START]] 148 // CHECK: [[LINSTEP:.+]] = load i64, i64* [[LIN_STEP]] 149 // CHECK: store i32 {{.+}}, i32* [[LIN_VAR]], 150 // CHECK: [[GLINSTART:.+]] = load double*, double** [[GLIN_START]] 151 // CHECK: store double* {{.*}}[[GLIN_VAR]] 152 153 #pragma omp simd 154 // CHECK: store i32 0, i32* [[OMP_IV4:%[^,]+]] 155 156 // CHECK: [[IV4:%.+]] = load i32, i32* [[OMP_IV4]]{{.*}}!llvm.access.group 157 // CHECK-NEXT: [[CMP4:%.+]] = icmp slt i32 [[IV4]], 4 158 // CHECK-NEXT: br i1 [[CMP4]], label %[[SIMPLE_LOOP4_BODY:.+]], label %[[SIMPLE_LOOP4_END:[^,]+]] 159 for (short it = 6; it <= 20; it-=-4) { 160 // CHECK: [[SIMPLE_LOOP4_BODY]]: 161 // Start of body: calculate it from IV: 162 // CHECK: [[IV4_0:%.+]] = load i32, i32* [[OMP_IV4]]{{.*}}!llvm.access.group 163 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul nsw i32 [[IV4_0]], 4 164 // CHECK-NEXT: [[LC_IT_2:%.+]] = add nsw i32 6, [[LC_IT_1]] 165 // CHECK-NEXT: [[LC_IT_3:%.+]] = trunc i32 [[LC_IT_2]] to i16 166 // CHECK-NEXT: store i16 [[LC_IT_3]], i16* {{.+}}, !llvm.access.group 167 168 // CHECK: [[IV4_2:%.+]] = load i32, i32* [[OMP_IV4]]{{.*}}!llvm.access.group 169 // CHECK-NEXT: [[ADD4_2:%.+]] = add nsw i32 [[IV4_2]], 1 170 // CHECK-NEXT: store i32 [[ADD4_2]], i32* [[OMP_IV4]]{{.*}}!llvm.access.group 171 } 172 // CHECK: [[SIMPLE_LOOP4_END]]: 173 174 #pragma omp simd 175 // CHECK: store i32 0, i32* [[OMP_IV5:%[^,]+]] 176 177 // CHECK: [[IV5:%.+]] = load i32, i32* [[OMP_IV5]]{{.*}}!llvm.access.group 178 // CHECK-NEXT: [[CMP5:%.+]] = icmp slt i32 [[IV5]], 26 179 // CHECK-NEXT: br i1 [[CMP5]], label %[[SIMPLE_LOOP5_BODY:.+]], label %[[SIMPLE_LOOP5_END:[^,]+]] 180 for (unsigned char it = 'z'; it >= 'a'; it+=-1) { 181 // CHECK: [[SIMPLE_LOOP5_BODY]]: 182 // Start of body: calculate it from IV: 183 // CHECK: [[IV5_0:%.+]] = load i32, i32* [[OMP_IV5]]{{.*}}!llvm.access.group 184 // CHECK-NEXT: [[IV5_1:%.+]] = mul nsw i32 [[IV5_0]], 1 185 // CHECK-NEXT: [[LC_IT_1:%.+]] = sub nsw i32 122, [[IV5_1]] 186 // CHECK-NEXT: [[LC_IT_2:%.+]] = trunc i32 [[LC_IT_1]] to i8 187 // CHECK-NEXT: store i8 [[LC_IT_2]], i8* {{.+}}, !llvm.access.group 188 189 // CHECK: [[IV5_2:%.+]] = load i32, i32* [[OMP_IV5]]{{.*}}!llvm.access.group 190 // CHECK-NEXT: [[ADD5_2:%.+]] = add nsw i32 [[IV5_2]], 1 191 // CHECK-NEXT: store i32 [[ADD5_2]], i32* [[OMP_IV5]]{{.*}}!llvm.access.group 192 } 193 // CHECK: [[SIMPLE_LOOP5_END]]: 194 195 // CHECK-NOT: mul i32 %{{.+}}, 10 196 #pragma omp simd 197 for (unsigned i=100; i<10; i+=10) { 198 } 199 200 int A; 201 // CHECK: store i32 -1, i32* [[A:%.+]], 202 A = -1; 203 #pragma omp simd lastprivate(A) 204 // CHECK: store i64 0, i64* [[OMP_IV7:%[^,]+]] 205 // CHECK: br label %[[SIMD_LOOP7_COND:[^,]+]] 206 // CHECK: [[SIMD_LOOP7_COND]]: 207 // CHECK-NEXT: [[IV7:%.+]] = load i64, i64* [[OMP_IV7]]{{.*}}!llvm.access.group 208 // CHECK-NEXT: [[CMP7:%.+]] = icmp slt i64 [[IV7]], 7 209 // CHECK-NEXT: br i1 [[CMP7]], label %[[SIMPLE_LOOP7_BODY:.+]], label %[[SIMPLE_LOOP7_END:[^,]+]] 210 for (long long i = -10; i < 10; i += 3) { 211 // CHECK: [[SIMPLE_LOOP7_BODY]]: 212 // Start of body: calculate i from IV: 213 // CHECK: [[IV7_0:%.+]] = load i64, i64* [[OMP_IV7]]{{.*}}!llvm.access.group 214 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul nsw i64 [[IV7_0]], 3 215 // CHECK-NEXT: [[LC_IT_2:%.+]] = add nsw i64 -10, [[LC_IT_1]] 216 // CHECK-NEXT: store i64 [[LC_IT_2]], i64* [[LC:%[^,]+]],{{.+}}!llvm.access.group 217 // CHECK-NEXT: [[LC_VAL:%.+]] = load i64, i64* [[LC]]{{.+}}!llvm.access.group 218 // CHECK-NEXT: [[CONV:%.+]] = trunc i64 [[LC_VAL]] to i32 219 // CHECK-NEXT: store i32 [[CONV]], i32* [[A_PRIV:%[^,]+]],{{.+}}!llvm.access.group 220 A = i; 221 // CHECK: [[IV7_2:%.+]] = load i64, i64* [[OMP_IV7]]{{.*}}!llvm.access.group 222 // CHECK-NEXT: [[ADD7_2:%.+]] = add nsw i64 [[IV7_2]], 1 223 // CHECK-NEXT: store i64 [[ADD7_2]], i64* [[OMP_IV7]]{{.*}}!llvm.access.group 224 } 225 // CHECK: [[SIMPLE_LOOP7_END]]: 226 // CHECK-NEXT: store i64 11, i64* 227 // CHECK-NEXT: [[A_PRIV_VAL:%.+]] = load i32, i32* [[A_PRIV]], 228 // CHECK-NEXT: store i32 [[A_PRIV_VAL]], i32* [[A]], 229 int R; 230 // CHECK: store i32 -1, i32* [[R:%[^,]+]], 231 R = -1; 232 // CHECK: store i64 0, i64* [[OMP_IV8:%[^,]+]], 233 // CHECK: store i32 1, i32* [[R_PRIV:%[^,]+]], 234 #ifdef OMP5 235 #pragma omp simd reduction(*:R) if(A) 236 #else 237 #pragma omp simd reduction(*:R) 238 #endif 239 // OMP50: [[A_VAL:%.+]] = load i32, i32* [[A]], 240 // OMP50-NEXT: [[COND:%.+]] = icmp ne i32 [[A_VAL]], 0 241 // OMP50-NEXT: br i1 [[COND]], label {{%?}}[[THEN:[^,]+]], label {{%?}}[[ELSE:[^,]+]] 242 // OMP50: [[THEN]]: 243 244 // CHECK: br label %[[SIMD_LOOP8_COND:[^,]+]] 245 // CHECK: [[SIMD_LOOP8_COND]]: 246 // CHECK-NEXT: [[IV8:%.+]] = load i64, i64* [[OMP_IV8]]{{.*}}!llvm.access.group 247 // CHECK-NEXT: [[CMP8:%.+]] = icmp slt i64 [[IV8]], 7 248 // CHECK-NEXT: br i1 [[CMP8]], label %[[SIMPLE_LOOP8_BODY:.+]], label %[[SIMPLE_LOOP8_END:[^,]+]] 249 for (long long i = -10; i < 10; i += 3) { 250 // CHECK: [[SIMPLE_LOOP8_BODY]]: 251 // Start of body: calculate i from IV: 252 // CHECK: [[IV8_0:%.+]] = load i64, i64* [[OMP_IV8]]{{.*}}!llvm.access.group 253 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul nsw i64 [[IV8_0]], 3 254 // CHECK-NEXT: [[LC_IT_2:%.+]] = add nsw i64 -10, [[LC_IT_1]] 255 // CHECK-NEXT: store i64 [[LC_IT_2]], i64* [[LC:%[^,]+]],{{.+}}!llvm.access.group 256 // CHECK-NEXT: [[LC_VAL:%.+]] = load i64, i64* [[LC]]{{.+}}!llvm.access.group 257 // CHECK: store i32 %{{.+}}, i32* [[R_PRIV]],{{.+}}!llvm.access.group 258 R *= i; 259 // CHECK: [[IV8_2:%.+]] = load i64, i64* [[OMP_IV8]]{{.*}}!llvm.access.group 260 // CHECK-NEXT: [[ADD8_2:%.+]] = add nsw i64 [[IV8_2]], 1 261 // CHECK-NEXT: store i64 [[ADD8_2]], i64* [[OMP_IV8]]{{.*}}!llvm.access.group 262 } 263 // CHECK: [[SIMPLE_LOOP8_END]]: 264 // OMP50: br label {{%?}}[[EXIT:[^,]+]] 265 // OMP50: br label %[[SIMD_LOOP8_COND:[^,]+]] 266 // OMP50: [[SIMD_LOOP8_COND]]: 267 // OMP50-NEXT: [[IV8:%.+]] = load i64, i64* [[OMP_IV8]],{{[^!]*}} 268 // OMP50-NEXT: [[CMP8:%.+]] = icmp slt i64 [[IV8]], 7 269 // OMP50-NEXT: br i1 [[CMP8]], label %[[SIMPLE_LOOP8_BODY:.+]], label %[[SIMPLE_LOOP8_END:[^,]+]] 270 // OMP50: [[SIMPLE_LOOP8_BODY]]: 271 // Start of body: calculate i from IV: 272 // OMP50: [[IV8_0:%.+]] = load i64, i64* [[OMP_IV8]],{{[^!]*}} 273 // OMP50-NEXT: [[LC_IT_1:%.+]] = mul nsw i64 [[IV8_0]], 3 274 // OMP50-NEXT: [[LC_IT_2:%.+]] = add nsw i64 -10, [[LC_IT_1]] 275 // OMP50-NEXT: store i64 [[LC_IT_2]], i64* [[LC:%[^,]+]],{{[^!]*}} 276 // OMP50-NEXT: [[LC_VAL:%.+]] = load i64, i64* [[LC]],{{[^!]*}} 277 // OMP50: store i32 %{{.+}}, i32* [[R_PRIV]],{{[^!]*}} 278 // OMP50: [[IV8_2:%.+]] = load i64, i64* [[OMP_IV8]],{{[^!]*}} 279 // OMP50-NEXT: [[ADD8_2:%.+]] = add nsw i64 [[IV8_2]], 1 280 // OMP50-NEXT: store i64 [[ADD8_2]], i64* [[OMP_IV8]],{{[^!]*}} 281 // OMP50: br label {{%?}}[[SIMD_LOOP8_COND]], {{.*}}!llvm.loop ![[DISABLE_VECT:.+]] 282 // OMP50: [[SIMPLE_LOOP8_END]]: 283 // OMP50: br label {{%?}}[[EXIT]] 284 // OMP50: [[EXIT]]: 285 286 // CHECK-DAG: [[R_VAL:%.+]] = load i32, i32* [[R]], 287 // CHECK-DAG: [[R_PRIV_VAL:%.+]] = load i32, i32* [[R_PRIV]], 288 // CHECK: [[RED:%.+]] = mul nsw i32 [[R_VAL]], [[R_PRIV_VAL]] 289 // CHECK-NEXT: store i32 [[RED]], i32* [[R]], 290 // CHECK-NEXT: ret void 291 } 292 293 template <class T, unsigned K> T tfoo(T a) { return a + K; } 294 295 template <typename T, unsigned N> 296 int templ1(T a, T *z) { 297 #pragma omp simd collapse(N) 298 for (int i = 0; i < N * 2; i++) { 299 for (long long j = 0; j < (N + N + N + N); j += 2) { 300 z[i + j] = a + tfoo<T, N>(i + j); 301 } 302 } 303 return 0; 304 } 305 306 // Instatiation templ1<float,2> 307 // CHECK-LABEL: define {{.*i32}} @{{.*}}templ1{{.*}}(float {{.+}}, float* {{.+}}) 308 // CHECK: store i64 0, i64* [[T1_OMP_IV:[^,]+]] 309 // ... 310 // CHECK: [[IV:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group 311 // CHECK-NEXT: [[CMP1:%.+]] = icmp slt i64 [[IV]], 16 312 // CHECK-NEXT: br i1 [[CMP1]], label %[[T1_BODY:.+]], label %[[T1_END:[^,]+]] 313 // CHECK: [[T1_BODY]]: 314 // Loop counters i and j updates: 315 // CHECK: [[IV1:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group 316 // CHECK-NEXT: [[I_1:%.+]] = sdiv i64 [[IV1]], 4 317 // CHECK-NEXT: [[I_1_MUL1:%.+]] = mul nsw i64 [[I_1]], 1 318 // CHECK-NEXT: [[I_1_ADD0:%.+]] = add nsw i64 0, [[I_1_MUL1]] 319 // CHECK-NEXT: [[I_2:%.+]] = trunc i64 [[I_1_ADD0]] to i32 320 // CHECK-NEXT: store i32 [[I_2]], i32* {{%.+}}{{.*}}!llvm.access.group 321 // CHECK: [[IV2:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group 322 // CHECK: [[IV2_1:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group 323 // CHECK-NEXT: [[J_1_DIV1:%.+]] = sdiv i64 [[IV2_1]], 4 324 // CHECK-NEXT: [[J_1_MUL1:%.+]] = mul nsw i64 [[J_1_DIV1]], 4 325 // CHECK-NEXT: [[J_1_SUB0:%.+]] = sub nsw i64 [[IV2]], [[J_1_MUL1]] 326 // CHECK-NEXT: [[J_2:%.+]] = mul nsw i64 [[J_1_SUB0]], 2 327 // CHECK-NEXT: [[J_2_ADD0:%.+]] = add nsw i64 0, [[J_2]] 328 // CHECK-NEXT: store i64 [[J_2_ADD0]], i64* {{%.+}}{{.*}}!llvm.access.group 329 // simd.for.inc: 330 // CHECK: [[IV3:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group 331 // CHECK-NEXT: [[INC:%.+]] = add nsw i64 [[IV3]], 1 332 // CHECK-NEXT: store i64 [[INC]], i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group 333 // CHECK-NEXT: br label {{%.+}} 334 // CHECK: [[T1_END]]: 335 // CHECK: ret i32 0 336 // 337 void inst_templ1() { 338 float a; 339 float z[100]; 340 templ1<float,2> (a, z); 341 } 342 343 344 typedef int MyIdx; 345 346 class IterDouble { 347 double *Ptr; 348 public: 349 IterDouble operator++ () const { 350 IterDouble n; 351 n.Ptr = Ptr + 1; 352 return n; 353 } 354 bool operator < (const IterDouble &that) const { 355 return Ptr < that.Ptr; 356 } 357 double & operator *() const { 358 return *Ptr; 359 } 360 MyIdx operator - (const IterDouble &that) const { 361 return (MyIdx) (Ptr - that.Ptr); 362 } 363 IterDouble operator + (int Delta) { 364 IterDouble re; 365 re.Ptr = Ptr + Delta; 366 return re; 367 } 368 369 ///~IterDouble() {} 370 }; 371 372 // CHECK-LABEL: define {{.*void}} @{{.*}}iter_simple{{.*}} 373 void iter_simple(IterDouble ia, IterDouble ib, IterDouble ic) { 374 // 375 // Calculate number of iterations before the loop body. 376 // CHECK: [[DIFF1:%.+]] = invoke {{.*}}i32 @{{.*}}IterDouble{{.*}} 377 // CHECK: [[DIFF2:%.+]] = sub nsw i32 [[DIFF1]], 1 378 // CHECK-NEXT: [[DIFF3:%.+]] = add nsw i32 [[DIFF2]], 1 379 // CHECK-NEXT: [[DIFF4:%.+]] = sdiv i32 [[DIFF3]], 1 380 // CHECK-NEXT: [[DIFF5:%.+]] = sub nsw i32 [[DIFF4]], 1 381 // CHECK-NEXT: store i32 [[DIFF5]], i32* [[OMP_LAST_IT:%[^,]+]]{{.+}} 382 // CHECK: store i32 0, i32* [[IT_OMP_IV:%[^,]+]] 383 #pragma omp simd 384 385 // CHECK: [[IV:%.+]] = load i32, i32* [[IT_OMP_IV]]{{.+}} !llvm.access.group 386 // CHECK-NEXT: [[LAST_IT:%.+]] = load i32, i32* [[OMP_LAST_IT]]{{.+}}!llvm.access.group 387 // CHECK-NEXT: [[NUM_IT:%.+]] = add nsw i32 [[LAST_IT]], 1 388 // CHECK-NEXT: [[CMP:%.+]] = icmp slt i32 [[IV]], [[NUM_IT]] 389 // CHECK-NEXT: br i1 [[CMP]], label %[[IT_BODY:[^,]+]], label %[[IT_END:[^,]+]] 390 for (IterDouble i = ia; i < ib; ++i) { 391 // CHECK: [[IT_BODY]]: 392 // Start of body: calculate i from index: 393 // CHECK: [[IV1:%.+]] = load i32, i32* [[IT_OMP_IV]]{{.+}}!llvm.access.group 394 // Call of operator+ (i, IV). 395 // CHECK: {{%.+}} = invoke {{.+}} @{{.*}}IterDouble{{.*}} 396 // ... loop body ... 397 *i = *ic * 0.5; 398 // Float multiply and save result. 399 // CHECK: [[MULR:%.+]] = fmul double {{%.+}}, 5.000000e-01 400 // CHECK-NEXT: invoke {{.+}} @{{.*}}IterDouble{{.*}} 401 // CHECK: store double [[MULR:%.+]], double* [[RESULT_ADDR:%.+]], !llvm.access.group 402 ++ic; 403 // 404 // CHECK: [[IV2:%.+]] = load i32, i32* [[IT_OMP_IV]]{{.+}}!llvm.access.group 405 // CHECK-NEXT: [[ADD2:%.+]] = add nsw i32 [[IV2]], 1 406 // CHECK-NEXT: store i32 [[ADD2]], i32* [[IT_OMP_IV]]{{.+}}!llvm.access.group 407 // br label %{{.*}}, !llvm.loop ![[ITER_LOOP_ID]] 408 } 409 // CHECK: [[IT_END]]: 410 // CHECK: ret void 411 } 412 413 414 // CHECK-LABEL: define {{.*void}} @{{.*}}collapsed{{.*}} 415 void collapsed(float *a, float *b, float *c, float *d) { 416 int i; // outer loop counter 417 unsigned j; // middle loop couter, leads to unsigned icmp in loop header. 418 // k declared in the loop init below 419 short l; // inner loop counter 420 // CHECK: store i32 0, i32* [[OMP_IV:[^,]+]] 421 // 422 #pragma omp simd collapse(4) 423 424 // CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group 425 // CHECK-NEXT: [[CMP:%.+]] = icmp ult i32 [[IV]], 120 426 // CHECK-NEXT: br i1 [[CMP]], label %[[COLL1_BODY:[^,]+]], label %[[COLL1_END:[^,]+]] 427 for (i = 1; i < 3; i++) // 2 iterations 428 for (j = 2u; j < 5u; j++) //3 iterations 429 for (int k = 3; k <= 6; k++) // 4 iterations 430 for (l = 4; l < 9; ++l) // 5 iterations 431 { 432 // CHECK: [[COLL1_BODY]]: 433 // Start of body: calculate i from index: 434 // CHECK: [[IV1:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group 435 // Calculation of the loop counters values. 436 // CHECK: [[CALC_I_1:%.+]] = udiv i32 [[IV1]], 60 437 // CHECK-NEXT: [[CALC_I_1_MUL1:%.+]] = mul i32 [[CALC_I_1]], 1 438 // CHECK-NEXT: [[CALC_I_2:%.+]] = add i32 1, [[CALC_I_1_MUL1]] 439 // CHECK-NEXT: store i32 [[CALC_I_2]], i32* [[LC_I:.+]] 440 441 // CHECK: [[IV1_2:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group 442 // CHECK: [[IV1_2_1:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group 443 // CHECK-NEXT: [[CALC_J_1:%.+]] = udiv i32 [[IV1_2_1]], 60 444 // CHECK-NEXT: [[MUL_1:%.+]] = mul i32 [[CALC_J_1]], 60 445 // CHECK-NEXT: [[SUB_3:%.+]] = sub i32 [[IV1_2]], [[MUL_1]] 446 // CHECK-NEXT: [[CALC_J_2:%.+]] = udiv i32 [[SUB_3]], 20 447 // CHECK-NEXT: [[CALC_J_2_MUL1:%.+]] = mul i32 [[CALC_J_2]], 1 448 // CHECK-NEXT: [[CALC_J_3:%.+]] = add i32 2, [[CALC_J_2_MUL1]] 449 // CHECK-NEXT: store i32 [[CALC_J_3]], i32* [[LC_J:.+]] 450 451 // CHECK: [[IV1_3:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group 452 // CHECK: [[IV1_3_1:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group 453 // CHECK-NEXT: [[DIV_1:%.+]] = udiv i32 [[IV1_3_1]], 60 454 // CHECK-NEXT: [[MUL_2:%.+]] = mul i32 [[DIV_1]], 60 455 // CHECK-NEXT: [[ADD_3:%.+]] = sub i32 [[IV1_3]], [[MUL_2]] 456 457 // CHECK: [[IV1_4:%.+]] = load i32, i32* [[OMP_IV]] 458 // CHECK: [[IV1_4_1:%.+]] = load i32, i32* [[OMP_IV]] 459 // CHECK-NEXT: [[DIV_2:%.+]] = udiv i32 [[IV1_4_1]], 60 460 // CHECK-NEXT: [[MUL_3:%.+]] = mul i32 [[DIV_2]], 60 461 // CHECK-NEXT: [[SUB_6:%.+]] = sub i32 [[IV1_4]], [[MUL_3]] 462 // CHECK-NEXT: [[DIV_3:%.+]] = udiv i32 [[SUB_6]], 20 463 // CHECK-NEXT: [[MUL_4:%.+]] = mul i32 [[DIV_3]], 20 464 // CHECK-NEXT: [[ADD_5:%.+]] = sub i32 [[ADD_3]], [[MUL_4]] 465 // CHECK-NEXT: [[DIV_4:%.+]] = udiv i32 [[ADD_5]], 5 466 // CHECK-NEXT: [[MUL_5:%.+]] = mul i32 [[DIV_4]], 1 467 // CHECK-NEXT: [[ADD_6:%.+]] = add i32 3, [[MUL_5]] 468 // CHECK-NEXT: store i32 [[ADD_6]], i32* [[LC_K:.+]] 469 470 // CHECK: [[IV1_5:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group 471 // CHECK: [[IV1_5_1:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group 472 // CHECK-NEXT: [[DIV_5:%.+]] = udiv i32 [[IV1_5_1]], 60 473 // CHECK-NEXT: [[MUL_6:%.+]] = mul i32 [[DIV_5]], 60 474 // CHECK-NEXT: [[SUB_7:%.+]] = sub i32 [[IV1_5]], [[MUL_6]] 475 476 // CHECK: [[IV1_6:%.+]] = load i32, i32* [[OMP_IV]] 477 // CHECK: [[IV1_6_1:%.+]] = load i32, i32* [[OMP_IV]] 478 // CHECK-NEXT: [[DIV_6:%.+]] = udiv i32 [[IV1_6_1]], 60 479 // CHECK-NEXT: [[MUL_7:%.+]] = mul i32 [[DIV_6]], 60 480 // CHECK-NEXT: [[SUB_10:%.+]] = sub i32 [[IV1_6]], [[MUL_7]] 481 // CHECK-NEXT: [[DIV_7:%.+]] = udiv i32 [[SUB_10]], 20 482 // CHECK-NEXT: [[MUL_8:%.+]] = mul i32 [[DIV_7]], 20 483 // CHECK-NEXT: [[SUB_11:%.+]] = sub i32 [[SUB_7]], [[MUL_8]] 484 485 // CHECK: [[IV1_7:%.+]] = load i32, i32* [[OMP_IV]] 486 // CHECK: [[IV1_7_1:%.+]] = load i32, i32* [[OMP_IV]] 487 // CHECK-NEXT: [[DIV_8:%.+]] = udiv i32 [[IV1_7_1]], 60 488 // CHECK-NEXT: [[MUL_9:%.+]] = mul i32 [[DIV_8]], 60 489 // CHECK-NEXT: [[SUB_12:%.+]] = sub i32 [[IV1_7]], [[MUL_9]] 490 491 // CHECK: [[IV1_8:%.+]] = load i32, i32* [[OMP_IV]] 492 // CHECK: [[IV1_8_1:%.+]] = load i32, i32* [[OMP_IV]] 493 // CHECK-NEXT: [[DIV_3:%.+]] = udiv i32 [[IV1_8_1]], 60 494 // CHECK-NEXT: [[MUL_4:%.+]] = mul i32 [[DIV_3]], 60 495 // CHECK-NEXT: [[SUB_7:%.+]] = sub i32 [[IV1_8]], [[MUL_4]] 496 // CHECK-NEXT: [[DIV_4:%.+]] = udiv i32 [[SUB_7]], 20 497 // CHECK-NEXT: [[MUL_5:%.+]] = mul i32 [[DIV_4]], 20 498 // CHECK-NEXT: [[SUB_8:%.+]] = sub i32 [[SUB_12]], [[MUL_5]] 499 // CHECK-NEXT: [[DIV_5:%.+]] = udiv i32 [[SUB_8]], 5 500 // CHECK-NEXT: [[MUL_6:%.+]] = mul i32 [[DIV_5]], 5 501 // CHECK-NEXT: [[SUB_9:%.+]] = sub i32 [[SUB_11]], [[MUL_6]] 502 // CHECK-NEXT: [[MUL_6:%.+]] = mul i32 [[SUB_9]], 1 503 // CHECK-NEXT: [[CALC_L_2:%.+]] = add i32 4, [[MUL_6]] 504 // CHECK-NEXT: [[CALC_L_3:%.+]] = trunc i32 [[CALC_L_2]] to i16 505 // CHECK-NEXT: store i16 [[CALC_L_3]], i16* [[LC_L:.+]] 506 // ... loop body ... 507 // End of body: store into a[i]: 508 // CHECK: store float [[RESULT:%.+]], float* [[RESULT_ADDR:%.+]]{{.+}}!llvm.access.group 509 float res = b[j] * c[k]; 510 a[i] = res * d[l]; 511 // CHECK: [[IV2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group 512 // CHECK-NEXT: [[ADD2:%.+]] = add i32 [[IV2]], 1 513 // CHECK-NEXT: store i32 [[ADD2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group 514 // br label %{{[^,]+}}, !llvm.loop ![[COLL1_LOOP_ID]] 515 // CHECK: [[COLL1_END]]: 516 } 517 // i,j,l are updated; k is not updated. 518 // CHECK: store i32 3, i32* 519 // CHECK-NEXT: store i32 5, i32* 520 // CHECK-NEXT: store i32 7, i32* 521 // CHECK-NEXT: store i16 9, i16* 522 // CHECK: ret void 523 } 524 525 extern char foo(); 526 extern double globalfloat; 527 528 // CHECK-LABEL: define {{.*void}} @{{.*}}widened{{.*}} 529 void widened(float *a, float *b, float *c, float *d) { 530 int i; // outer loop counter 531 short j; // inner loop counter 532 globalfloat = 1.0; 533 int localint = 1; 534 // CHECK: store double {{.+}}, double* [[GLOBALFLOAT:@.+]] 535 // Counter is widened to 64 bits. 536 // CHECK: store i64 0, i64* [[OMP_IV:[^,]+]] 537 // 538 #pragma omp simd collapse(2) private(globalfloat, localint) 539 540 // CHECK: [[IV:%.+]] = load i64, i64* [[OMP_IV]]{{.+}}!llvm.access.group 541 // CHECK-NEXT: [[LI:%.+]] = load i64, i64* [[OMP_LI:%[^,]+]]{{.+}}!llvm.access.group 542 // CHECK-NEXT: [[NUMIT:%.+]] = add nsw i64 [[LI]], 1 543 // CHECK-NEXT: [[CMP:%.+]] = icmp slt i64 [[IV]], [[NUMIT]] 544 // CHECK-NEXT: br i1 [[CMP]], label %[[WIDE1_BODY:[^,]+]], label %[[WIDE1_END:[^,]+]] 545 for (i = 1; i < 3; i++) // 2 iterations 546 for (j = 0; j < foo(); j++) // foo() iterations 547 { 548 // CHECK: [[WIDE1_BODY]]: 549 // Start of body: calculate i from index: 550 // CHECK: [[IV1:%.+]] = load i64, i64* [[OMP_IV]]{{.+}}!llvm.access.group 551 // Calculation of the loop counters values... 552 // CHECK: store i32 {{[^,]+}}, i32* [[LC_I:.+]] 553 // CHECK: [[IV1_2:%.+]] = load i64, i64* [[OMP_IV]]{{.+}}!llvm.access.group 554 // CHECK: store i16 {{[^,]+}}, i16* [[LC_J:.+]] 555 // ... loop body ... 556 // 557 // Here we expect store into private double var, not global 558 // CHECK-NOT: store double {{.+}}, double* [[GLOBALFLOAT]] 559 globalfloat = (float)j/i; 560 float res = b[j] * c[j]; 561 // Store into a[i]: 562 // CHECK: store float [[RESULT:%.+]], float* [[RESULT_ADDR:%.+]]{{.+}}!llvm.access.group 563 a[i] = res * d[i]; 564 // Then there's a store into private var localint: 565 // CHECK: store i32 {{.+}}, i32* [[LOCALINT:%[^,]+]]{{.+}}!llvm.access.group 566 localint = (int)j; 567 // CHECK: [[IV2:%.+]] = load i64, i64* [[OMP_IV]]{{.*}}!llvm.access.group 568 // CHECK-NEXT: [[ADD2:%.+]] = add nsw i64 [[IV2]], 1 569 // CHECK-NEXT: store i64 [[ADD2]], i64* [[OMP_IV]]{{.*}}!llvm.access.group 570 // 571 // br label %{{[^,]+}}, !llvm.loop ![[WIDE1_LOOP_ID]] 572 // CHECK: [[WIDE1_END]]: 573 } 574 // i,j are updated. 575 // CHECK: store i32 3, i32* [[I:%[^,]+]] 576 // CHECK: store i16 577 // 578 // Here we expect store into original localint, not its privatized version. 579 // CHECK-NOT: store i32 {{.+}}, i32* [[LOCALINT]] 580 localint = (int)j; 581 // CHECK: ret void 582 } 583 584 // CHECK-LABEL: define {{.*void}} @{{.*}}linear{{.*}}(float* {{.+}}) 585 void linear(float *a) { 586 // CHECK: [[VAL_ADDR:%.+]] = alloca i64, 587 // CHECK: [[K_ADDR:%.+]] = alloca i64*, 588 long long val = 0; 589 long long &k = val; 590 591 #pragma omp simd linear(k : 3) 592 // CHECK: store i64* [[VAL_ADDR]], i64** [[K_ADDR]], 593 // CHECK: [[VAL_REF:%.+]] = load i64*, i64** [[K_ADDR]], 594 // CHECK: store i64* [[VAL_REF]], i64** [[K_ADDR_REF:%.+]], 595 // CHECK: store i32 0, i32* [[OMP_IV:%[^,]+]] 596 // CHECK: [[K_REF:%.+]] = load i64*, i64** [[K_ADDR_REF]], 597 // CHECK: [[K0LOAD:%.+]] = load i64, i64* [[K_REF]] 598 // CHECK-NEXT: store i64 [[K0LOAD]], i64* [[LIN0:%[^,]+]] 599 600 // CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group 601 // CHECK-NEXT: [[CMP2:%.+]] = icmp slt i32 [[IV]], 9 602 // CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP_BODY:.+]], label %[[SIMPLE_LOOP_END:[^,]+]] 603 for (int i = 10; i > 1; i--) { 604 // CHECK: [[SIMPLE_LOOP_BODY]]: 605 // Start of body: calculate i from IV: 606 // CHECK: [[IV_0:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group 607 // FIXME: It is interesting, why the following "mul 1" was not constant folded? 608 // CHECK-NEXT: [[IV_1:%.+]] = mul nsw i32 [[IV_0]], 1 609 // CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV_1]] 610 // CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.access.group 611 // 612 // CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.access.group 613 // CHECK-NEXT: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group 614 // CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV_2]], 3 615 // CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64 616 // CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]] 617 // Update of the privatized version of linear variable! 618 // CHECK-NEXT: store i64 [[LIN_ADD1]], i64* [[K_PRIVATIZED:%[^,]+]] 619 a[k]++; 620 k = k + 3; 621 // CHECK: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group 622 // CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV_2]], 1 623 // CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group 624 // br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP_ID]] 625 } 626 // CHECK: [[SIMPLE_LOOP_END]]: 627 // 628 // Update linear vars after loop, as the loop was operating on a private version. 629 // CHECK: [[K_REF:%.+]] = load i64*, i64** [[K_ADDR_REF]], 630 // CHECK: store i64* [[K_REF]], i64** [[K_PRIV_REF:%.+]], 631 // CHECK: [[LIN0_2:%.+]] = load i64, i64* [[LIN0]] 632 // CHECK-NEXT: [[LIN_ADD2:%.+]] = add nsw i64 [[LIN0_2]], 27 633 // CHECK-NEXT: [[K_REF:%.+]] = load i64*, i64** [[K_PRIV_REF]], 634 // CHECK-NEXT: store i64 [[LIN_ADD2]], i64* [[K_REF]] 635 // 636 637 #pragma omp simd linear(val(k) : 3) 638 // CHECK: [[VAL_REF:%.+]] = load i64*, i64** [[K_ADDR]], 639 // CHECK: store i64* [[VAL_REF]], i64** [[K_ADDR_REF:%.+]], 640 // CHECK: store i32 0, i32* [[OMP_IV:%[^,]+]] 641 // CHECK: [[K_REF:%.+]] = load i64*, i64** [[K_ADDR_REF]], 642 // CHECK: [[K0LOAD:%.+]] = load i64, i64* [[K_REF]] 643 // CHECK-NEXT: store i64 [[K0LOAD]], i64* [[LIN0:%[^,]+]] 644 645 // CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group 646 // CHECK-NEXT: [[CMP2:%.+]] = icmp slt i32 [[IV]], 9 647 // CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP_BODY:.+]], label %[[SIMPLE_LOOP_END:[^,]+]] 648 for (int i = 10; i > 1; i--) { 649 // CHECK: [[SIMPLE_LOOP_BODY]]: 650 // Start of body: calculate i from IV: 651 // CHECK: [[IV_0:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group 652 // FIXME: It is interesting, why the following "mul 1" was not constant folded? 653 // CHECK-NEXT: [[IV_1:%.+]] = mul nsw i32 [[IV_0]], 1 654 // CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV_1]] 655 // CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.access.group 656 // 657 // CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.access.group 658 // CHECK-NEXT: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group 659 // CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV_2]], 3 660 // CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64 661 // CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]] 662 // Update of the privatized version of linear variable! 663 // CHECK-NEXT: store i64 [[LIN_ADD1]], i64* [[K_PRIVATIZED:%[^,]+]] 664 a[k]++; 665 k = k + 3; 666 // CHECK: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group 667 // CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV_2]], 1 668 // CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group 669 // br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP_ID]] 670 } 671 // CHECK: [[SIMPLE_LOOP_END]]: 672 // 673 // Update linear vars after loop, as the loop was operating on a private version. 674 // CHECK: [[K_REF:%.+]] = load i64*, i64** [[K_ADDR_REF]], 675 // CHECK: store i64* [[K_REF]], i64** [[K_PRIV_REF:%.+]], 676 // CHECK: [[LIN0_2:%.+]] = load i64, i64* [[LIN0]] 677 // CHECK-NEXT: [[LIN_ADD2:%.+]] = add nsw i64 [[LIN0_2]], 27 678 // CHECK-NEXT: [[K_REF:%.+]] = load i64*, i64** [[K_PRIV_REF]], 679 // CHECK-NEXT: store i64 [[LIN_ADD2]], i64* [[K_REF]] 680 // 681 #pragma omp simd linear(uval(k) : 3) 682 // CHECK: store i32 0, i32* [[OMP_IV:%[^,]+]] 683 // CHECK: [[K0LOAD:%.+]] = load i64, i64* [[VAL_ADDR]] 684 // CHECK-NEXT: store i64 [[K0LOAD]], i64* [[LIN0:%[^,]+]] 685 686 // CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group 687 // CHECK-NEXT: [[CMP2:%.+]] = icmp slt i32 [[IV]], 9 688 // CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP_BODY:.+]], label %[[SIMPLE_LOOP_END:[^,]+]] 689 for (int i = 10; i > 1; i--) { 690 // CHECK: [[SIMPLE_LOOP_BODY]]: 691 // Start of body: calculate i from IV: 692 // CHECK: [[IV_0:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group 693 // FIXME: It is interesting, why the following "mul 1" was not constant folded? 694 // CHECK-NEXT: [[IV_1:%.+]] = mul nsw i32 [[IV_0]], 1 695 // CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV_1]] 696 // CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.access.group 697 // 698 // CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.access.group 699 // CHECK-NEXT: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group 700 // CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV_2]], 3 701 // CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64 702 // CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]] 703 // Update of the privatized version of linear variable! 704 // CHECK-NEXT: store i64 [[LIN_ADD1]], i64* [[K_PRIVATIZED:%[^,]+]] 705 a[k]++; 706 k = k + 3; 707 // CHECK: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group 708 // CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV_2]], 1 709 // CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group 710 // br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP_ID]] 711 } 712 // CHECK: [[SIMPLE_LOOP_END]]: 713 // 714 // Update linear vars after loop, as the loop was operating on a private version. 715 // CHECK: [[LIN0_2:%.+]] = load i64, i64* [[LIN0]] 716 // CHECK-NEXT: [[LIN_ADD2:%.+]] = add nsw i64 [[LIN0_2]], 27 717 // CHECK-NEXT: store i64 [[LIN_ADD2]], i64* [[VAL_ADDR]] 718 // 719 } 720 721 // TERM_DEBUG-LABEL: bar 722 int bar() {return 0;}; 723 724 // TERM_DEBUG-LABEL: parallel_simd 725 void parallel_simd(float *a) { 726 #pragma omp parallel 727 #pragma omp simd 728 // TERM_DEBUG-NOT: __kmpc_global_thread_num 729 // TERM_DEBUG: invoke i32 {{.*}}bar{{.*}}() 730 // TERM_DEBUG: unwind label %[[TERM_LPAD:[^,]+]], 731 // TERM_DEBUG-NOT: __kmpc_global_thread_num 732 // TERM_DEBUG: [[TERM_LPAD]] 733 // TERM_DEBUG: call void @__clang_call_terminate 734 // TERM_DEBUG: unreachable 735 for (unsigned i = 131071; i <= 2147483647; i += 127) 736 a[i] += bar(); 737 } 738 // TERM_DEBUG: !{{[0-9]+}} = !DILocation(line: [[@LINE-11]], 739 740 // CHECK-LABEL: S8 741 // CHECK-DAG: ptrtoint [[SS_TY]]* %{{.+}} to i64 742 // CHECK-DAG: ptrtoint [[SS_TY]]* %{{.+}} to i64 743 // CHECK-DAG: ptrtoint [[SS_TY]]* %{{.+}} to i64 744 // CHECK-DAG: ptrtoint [[SS_TY]]* %{{.+}} to i64 745 746 // CHECK-DAG: and i64 %{{.+}}, 15 747 // CHECK-DAG: icmp eq i64 %{{.+}}, 0 748 // CHECK-DAG: call void @llvm.assume(i1 749 750 // CHECK-DAG: and i64 %{{.+}}, 7 751 // CHECK-DAG: icmp eq i64 %{{.+}}, 0 752 // CHECK-DAG: call void @llvm.assume(i1 753 754 // CHECK-DAG: and i64 %{{.+}}, 15 755 // CHECK-DAG: icmp eq i64 %{{.+}}, 0 756 // CHECK-DAG: call void @llvm.assume(i1 757 758 // CHECK-DAG: and i64 %{{.+}}, 3 759 // CHECK-DAG: icmp eq i64 %{{.+}}, 0 760 // CHECK-DAG: call void @llvm.assume(i1 761 struct SS { 762 SS(): a(0) {} 763 SS(int v) : a(v) {} 764 int a; 765 typedef int type; 766 }; 767 768 template <typename T> 769 class S7 : public T { 770 protected: 771 T *a; 772 T b[2]; 773 S7() : a(0) {} 774 775 public: 776 S7(typename T::type &v) : a((T*)&v) { 777 #pragma omp simd aligned(a) 778 for (int k = 0; k < a->a; ++k) 779 ++this->a->a; 780 #pragma omp simd aligned(this->b : 8) 781 for (int k = 0; k < a->a; ++k) 782 ++a->a; 783 } 784 }; 785 786 class S8 : private IterDouble, public S7<SS> { 787 S8() {} 788 789 public: 790 S8(int v) : S7<SS>(v){ 791 #pragma omp parallel private(a) 792 #pragma omp simd aligned(S7<SS>::a) 793 for (int k = 0; k < a->a; ++k) 794 ++this->a->a; 795 #pragma omp parallel shared(b) 796 #pragma omp simd aligned(this->b: 4) 797 for (int k = 0; k < a->a; ++k) 798 ++a->a; 799 } 800 }; 801 S8 s8(0); 802 803 // TERM_DEBUG-NOT: line: 0, 804 // TERM_DEBUG: distinct !DISubprogram(linkageName: "_GLOBAL__sub_I_simd_codegen.cpp", 805 // OMP50-DAG: ![[NOVECT:.+]] = !{!"llvm.loop.vectorize.enable", i1 false} 806 // OMP50-DAG: ![[DISABLE_VECT]] = distinct !{{.*}}![[NOVECT]]{{[,}]}} 807 #endif // HEADER 808 809