clang/test/OpenMP/ordered_doacross_codegen.cpp

   1 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s --check-prefixes=CHECK,CHECK-NORMAL
   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 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefixes=CHECK,CHECK-NORMAL
   4
   5 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-enable-irbuilder -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s --check-prefixes=CHECK,CHECK-IRBUILDER
   6 // RUN: %clang_cc1 -fopenmp -fopenmp-enable-irbuilder -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s
   7 // RUN: %clang_cc1 -fopenmp -fopenmp-enable-irbuilder -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefixes=CHECK,CHECK-IRBUILDER
   8
   9 // RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %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 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
  12 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
  13 // expected-no-diagnostics
  14
  15 #ifndef HEADER
  16 #define HEADER
  17
  18 // CHECK: [[KMP_DIM:%.+]] = type { i64, i64, i64 }
  19 extern int n;
  20 int a[10], b[10], c[10], d[10];
  21 void foo();
  22
  23 // CHECK-LABEL:bar
  24 void bar() {
  25   int i,j;
  26 // CHECK: call void @__kmpc_doacross_init(
  27 // CHECK: call void @__kmpc_doacross_fini(
  28 #pragma omp parallel for ordered(2)
  29   for (i = 0; i < n; ++i)
  30   for (j = 0; j < n; ++j)
  31     a[i] = b[i] + 1;
  32 // CHECK: call void @__kmpc_doacross_init(
  33 // CHECK: call void @__kmpc_doacross_fini(
  34  #pragma omp for collapse(2) ordered(2)
  35   for (int i = 0; i < n; i++)
  36     for (int j = 0; j < n; j++)
  37       ;
  38 }
  39
  40 // CHECK-LABEL: @main()
  41 int main() {
  42   int i;
  43 // CHECK: [[DIMS:%.+]] = alloca [1 x [[KMP_DIM]]],
  44 // CHECK-NORMAL: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(ptr [[IDENT:@.+]])
  45 // CHECK: icmp
  46 // CHECK-NEXT: br i1 %
  47 // CHECK: call void @llvm.memset.p0.i64(ptr align 8 [[DIMS]], i8 0, i64 24, i1 false)
  48 // CHECK: [[DIM:%.+]] = getelementptr inbounds [1 x [[KMP_DIM]]], ptr [[DIMS]], i64 0, i64 0
  49 // CHECK: getelementptr inbounds [[KMP_DIM]], ptr [[DIM]], i32 0, i32 1
  50 // CHECK: store i64 %{{.+}}, ptr %
  51 // CHECK: getelementptr inbounds [[KMP_DIM]], ptr [[DIM]], i32 0, i32 2
  52 // CHECK: store i64 1, ptr %
  53 // CHECK: [[DIM:%.+]] = getelementptr inbounds [1 x [[KMP_DIM]]], ptr [[DIMS]], i64 0, i64 0
  54 // CHECK-NORMAL: call void @__kmpc_doacross_init(ptr [[IDENT]], i32 [[GTID]], i32 1, ptr [[DIM]])
  55 // CHECK-NORMAL: call void @__kmpc_for_static_init_4(ptr @{{.+}}, i32 [[GTID]], i32 33, ptr %{{.+}}, ptr %{{.+}}, ptr %{{.+}}, ptr %{{.+}}, i32 1, i32 1)
  56 #pragma omp for ordered(1)
  57   for (int i = 0; i < n; ++i) {
  58     a[i] = b[i] + 1;
  59     foo();
  60 // CHECK: invoke void [[FOO:.+]](
  61 // CHECK: load i32, ptr [[I:%.+]],
  62 // CHECK-NEXT: sub nsw i32 %{{.+}}, 0
  63 // CHECK-NEXT: sdiv i32 %{{.+}}, 1
  64 // CHECK-NEXT: sext i32 %{{.+}} to i64
  65 // CHECK-NEXT: [[TMP:%.+]] = getelementptr inbounds [1 x i64], ptr [[CNT:%.+]], i64 0, i64 0
  66 // CHECK-NEXT: store i64 %{{.+}}, ptr [[TMP]],
  67 // CHECK-NEXT: [[TMP:%.+]] = getelementptr inbounds [1 x i64], ptr [[CNT]], i64 0, i64 0
  68 // CHECK-NORMAL-NEXT: call void @__kmpc_doacross_post(ptr [[IDENT]], i32 [[GTID]], ptr [[TMP]])
  69 // CHECK-IRBUILDER-NEXT: [[GTID18:%.+]] = call i32 @__kmpc_global_thread_num(ptr [[IDENT:@.+]])
  70 // CHECK-IRBUILDER-NEXT: call void @__kmpc_doacross_post(ptr [[IDENT]], i32 [[GTID18]], ptr [[TMP]])
  71 #pragma omp ordered depend(source)
  72     c[i] = c[i] + 1;
  73     foo();
  74 // CHECK: invoke void [[FOO]]
  75 // CHECK: load i32, ptr [[I]],
  76 // CHECK-NEXT: sub nsw i32 %{{.+}}, 2
  77 // CHECK-NEXT: sub nsw i32 %{{.+}}, 0
  78 // CHECK-NEXT: sdiv i32 %{{.+}}, 1
  79 // CHECK-NEXT: sext i32 %{{.+}} to i64
  80 // CHECK-NEXT: [[TMP:%.+]] = getelementptr inbounds [1 x i64], ptr [[CNT:%.+]], i64 0, i64 0
  81 // CHECK-NEXT: store i64 %{{.+}}, ptr [[TMP]],
  82 // CHECK-NEXT: [[TMP:%.+]] = getelementptr inbounds [1 x i64], ptr [[CNT]], i64 0, i64 0
  83 // CHECK-NORMAL-NEXT: call void @__kmpc_doacross_wait(ptr [[IDENT]], i32 [[GTID]], ptr [[TMP]])
  84 // CHECK-IRBUILDER-NEXT: [[GTID30:%.+]] = call i32 @__kmpc_global_thread_num(ptr [[IDENT:@.+]])
  85 // CHECK-IRBUILDER-NEXT: call void @__kmpc_doacross_wait(ptr [[IDENT]], i32 [[GTID30]], ptr [[TMP]])
  86 #pragma omp ordered depend(sink : i - 2)
  87     d[i] = a[i - 2];
  88   }
  89
  90   // CHECK: call void @__kmpc_for_static_fini(
  91   // CHECK-NORMAL: call void @__kmpc_doacross_fini(ptr [[IDENT]], i32 [[GTID]])
  92   // CHECK: ret i32 0
  93   return 0;
  94 }
  95
  96 // CHECK-LABEL: main1
  97 int main1() {
  98 // CHECK: [[DIMS:%.+]] = alloca [1 x [[KMP_DIM]]],
  99 // CHECK-NORMAL: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(ptr [[IDENT:@.+]])
 100 // CHECK: icmp
 101 // CHECK-NEXT: br i1 %
 102 // CHECK: call void @llvm.memset.p0.i64(ptr align 8 [[DIMS]], i8 0, i64 24, i1 false)
 103 // CHECK: [[DIM:%.+]] = getelementptr inbounds [1 x [[KMP_DIM]]], ptr [[DIMS]], i64 0, i64 0
 104 // CHECK: getelementptr inbounds [[KMP_DIM]], ptr [[DIM]], i32 0, i32 1
 105 // CHECK: store i64 %{{.+}}, ptr %
 106 // CHECK: getelementptr inbounds [[KMP_DIM]], ptr [[DIM]], i32 0, i32 2
 107 // CHECK: store i64 1, ptr %
 108 // CHECK: [[DIM:%.+]] = getelementptr inbounds [1 x [[KMP_DIM]]], ptr [[DIMS]], i64 0, i64 0
 109 // CHECK-NORMAL: call void @__kmpc_doacross_init(ptr [[IDENT]], i32 [[GTID]], i32 1, ptr [[DIM]])
 110 // CHECK-NORMAL: call void @__kmpc_for_static_init_4(ptr @{{.+}}, i32 [[GTID]], i32 33, ptr %{{.+}}, ptr %{{.+}}, ptr %{{.+}}, ptr %{{.+}}, i32 1, i32 1)
 111 #pragma omp for ordered(1)
 112   for (int i = n; i > 0; --i) {
 113     a[i] = b[i] + 1;
 114     foo();
 115 // CHECK: invoke void [[FOO:.+]](
 116 // CHECK: [[UB_VAL:%.+]] = load i32, ptr [[UB:%.+]],
 117 // CHECK-NEXT: [[I_VAL:%.+]] = load i32, ptr [[I:%.+]],
 118 // CHECK-NEXT: sub i32 [[UB_VAL]], [[I_VAL]]
 119 // CHECK-NEXT: udiv i32 %{{.+}}, 1
 120 // CHECK-NEXT: zext i32 %{{.+}} to i64
 121 // CHECK-NEXT: [[TMP:%.+]] = getelementptr inbounds [1 x i64], ptr [[CNT:%.+]], i64 0, i64 0
 122 // CHECK-NEXT: store i64 %{{.+}}, ptr [[TMP]],
 123 // CHECK-NEXT: [[TMP:%.+]] = getelementptr inbounds [1 x i64], ptr [[CNT]], i64 0, i64 0
 124 // CHECK-NORMAL-NEXT: call void @__kmpc_doacross_post(ptr [[IDENT]], i32 [[GTID]], ptr [[TMP]])
 125 // CHECK-IRBUILDER-NEXT: [[GTID17:%.+]] = call i32 @__kmpc_global_thread_num(ptr [[IDENT:@.+]])
 126 // CHECK-IRBUILDER-NEXT: call void @__kmpc_doacross_post(ptr [[IDENT]], i32 [[GTID17]], ptr [[TMP]])
 127 #pragma omp ordered depend(source)
 128     c[i] = c[i] + 1;
 129     foo();
 130 // CHECK: invoke void [[FOO]]
 131 // CHECK: [[UB_VAL:%.+]] = load i32, ptr [[UB]],
 132 // CHECK-NEXT: [[I_VAL:%.+]] = load i32, ptr [[I]],
 133 // CHECK-NEXT: [[SUB:%.+]] = sub nsw i32 [[I_VAL]], 2
 134 // CHECK-NEXT: sub i32 [[UB_VAL]], [[SUB]]
 135 // CHECK-NEXT: udiv i32 %{{.+}}, 1
 136 // CHECK-NEXT: zext i32 %{{.+}} to i64
 137 // CHECK-NEXT: [[TMP:%.+]] = getelementptr inbounds [1 x i64], ptr [[CNT:%.+]], i64 0, i64 0
 138 // CHECK-NEXT: store i64 %{{.+}}, ptr [[TMP]],
 139 // CHECK-NEXT: [[TMP:%.+]] = getelementptr inbounds [1 x i64], ptr [[CNT]], i64 0, i64 0
 140 // CHECK-NORMAL-NEXT: call void @__kmpc_doacross_wait(ptr [[IDENT]], i32 [[GTID]], ptr [[TMP]])
 141 // CHECK-IRBUILDER-NEXT: [[GTID29:%.+]] = call i32 @__kmpc_global_thread_num(ptr [[IDENT:@.+]])
 142 // CHECK-IRBUILDER-NEXT: call void @__kmpc_doacross_wait(ptr [[IDENT]], i32 [[GTID29]], ptr [[TMP]])
 143 #pragma omp ordered depend(sink : i - 2)
 144     d[i] = a[i - 2];
 145   }
 146   // CHECK: call void @__kmpc_for_static_fini(
 147   // CHECK-NORMAL: call void @__kmpc_doacross_fini(ptr [[IDENT]], i32 [[GTID]])
 148   // CHECK: ret i32 0
 149   return 0;
 150 }
 151
 152 // CHECK: define {{.+}}TestStruct
 153 template <typename T>
 154 struct TestStruct {
 155   static const int M = 10;
 156   static const int N = 20;
 157   T i;
 158   T a[N][M];
 159   T b[N][M];
 160   T foo(T, T);
 161   T bar(T, T, T);
 162   void baz(T, T);
 163   TestStruct() {
 164 // CHECK: [[DIMS:%.+]] = alloca [2 x [[KMP_DIM]]],
 165 // CHECK-NORMAL: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(ptr [[IDENT:@.+]])
 166 // CHECK: call void @llvm.memset.p0.i64(ptr align 8 [[DIMS]], i8 0, i64 48, i1 false)
 167 // CHECK: [[DIM:%.+]] = getelementptr inbounds [2 x [[KMP_DIM]]], ptr [[DIMS]], i64 0, i64 0
 168 // CHECK: getelementptr inbounds [[KMP_DIM]], ptr [[DIM]], i32 0, i32 1
 169 // CHECK: store i64 10, ptr %
 170 // CHECK: getelementptr inbounds [[KMP_DIM]], ptr [[DIM]], i32 0, i32 2
 171 // CHECK: store i64 1, ptr %
 172 // CHECK: [[DIM:%.+]] = getelementptr inbounds [2 x [[KMP_DIM]]], ptr [[DIMS]], i64 0, i64 1
 173 // CHECK: getelementptr inbounds [[KMP_DIM]], ptr [[DIM]], i32 0, i32 1
 174 // CHECK: store i64 %{{.+}}, ptr %
 175 // CHECK: getelementptr inbounds [[KMP_DIM]], ptr [[DIM]], i32 0, i32 2
 176 // CHECK: store i64 1, ptr %
 177 // CHECK: [[DIM:%.+]] = getelementptr inbounds [2 x [[KMP_DIM]]], ptr [[DIMS]], i64 0, i64 0
 178 // CHECK-NORMAL: call void @__kmpc_doacross_init(ptr [[IDENT]], i32 [[GTID]], i32 2, ptr [[DIM]])
 179 // CHECK-NORMAL: call void @__kmpc_for_static_init_4(ptr @{{.+}}, i32 [[GTID]], i32 33, ptr %{{.+}}, ptr %{{.+}}, ptr %{{.+}}, ptr %{{.+}}, i32 1, i32 1)
 180 #pragma omp for ordered(2)
 181     for (T j = 0; j < M; j++)
 182       for (i = 0; i < n; i += 2) {
 183         a[i][j] = foo(i, j);
 184 // CHECK: invoke {{.+TestStruct.+foo}}
 185 // CHECK: load ptr, ptr %
 186 // CHECK: load i32, ptr %
 187 // CHECK: load i32, ptr %
 188 // CHECK: load i32, ptr [[J:%.+]],
 189 // CHECK-NEXT: sub nsw i32 %{{.+}}, 0
 190 // CHECK-NEXT: sdiv i32 %{{.+}}, 1
 191 // CHECK-NEXT: sext i32 %{{.+}} to i64
 192 // CHECK-NORMAL-NEXT: [[TMP:%.+]] = getelementptr inbounds [2 x i64], ptr [[CNT:%.+]], i64 0, i64 0
 193 // CHECK-NORMAL-NEXT: store i64 %{{.+}}, ptr [[TMP:%.+]],
 194 // CHECK-NEXT: [[I:%.+]] = load ptr, ptr [[I_REF:%.+]],
 195 // CHECK-NEXT: load i32, ptr [[I]],
 196 // CHECK-NEXT: sub nsw i32 %{{.+}}, 2
 197 // CHECK-NEXT: sub nsw i32 %{{.+}}, 0
 198 // CHECK-NEXT: sdiv i32 %{{.+}}, 2
 199 // CHECK-NEXT: sext i32 %{{.+}} to i64
 200 // CHECK-IRBUILDER-NEXT: [[TMP:%.+]] = getelementptr inbounds [2 x i64], ptr [[CNT:%.+]], i64 0, i64 0
 201 // CHECK-IRBUILDER-NEXT: store i64 %{{.+}}, ptr [[TMP:%.+]],
 202 // CHECK-NEXT: [[TMP:%.+]] = getelementptr inbounds [2 x i64], ptr [[CNT]], i64 0, i64 1
 203 // CHECK-NEXT: store i64 %{{.+}}, ptr [[TMP]],
 204 // CHECK-NEXT: [[TMP:%.+]] = getelementptr inbounds [2 x i64], ptr [[CNT]], i64 0, i64 0
 205 // CHECK-NORMAL-NEXT: call void @__kmpc_doacross_wait(ptr [[IDENT]], i32 [[GTID]], ptr [[TMP]])
 206 // CHECK-IRBUILDER-NEXT: [[GTID18:%.+]] = call i32 @__kmpc_global_thread_num(ptr [[IDENT:@.+]])
 207 // CHECK-IRBUILDER-NEXT: call void @__kmpc_doacross_wait(ptr [[IDENT]], i32 [[GTID18]], ptr [[TMP]])
 208 // CHECK-NEXT: load i32, ptr [[J:%.+]],
 209 // CHECK-NEXT: sub nsw i32 %{{.+}}, 1
 210 // CHECK-NEXT: sub nsw i32 %{{.+}}, 0
 211 // CHECK-NEXT: sdiv i32 %{{.+}}, 1
 212 // CHECK-NEXT: sext i32 %{{.+}} to i64
 213 // CHECK-NORMAL-NEXT: [[TMP:%.+]] = getelementptr inbounds [2 x i64], ptr [[CNT:%.+]], i64 0, i64 0
 214 // CHECK-NORMAL-NEXT: store i64 %{{.+}}, ptr [[TMP]],
 215 // CHECK-NEXT: [[I:%.+]] = load ptr, ptr [[I_REF]],
 216 // CHECK-NEXT: load i32, ptr [[I]],
 217 // CHECK-NEXT: sub nsw i32 %{{.+}}, 0
 218 // CHECK-NEXT: sdiv i32 %{{.+}}, 2
 219 // CHECK-NEXT: sext i32 %{{.+}} to i64
 220 // CHECK-IRBUILDER-NEXT: [[TMP:%.+]] = getelementptr inbounds [2 x i64], ptr [[CNT:%.+]], i64 0, i64 0
 221 // CHECK-IRBUILDER-NEXT: store i64 %{{.+}}, ptr [[TMP]],
 222 // CHECK-NEXT: [[TMP:%.+]] = getelementptr inbounds [2 x i64], ptr [[CNT]], i64 0, i64 1
 223 // CHECK-NEXT: store i64 %{{.+}}, ptr [[TMP]],
 224 // CHECK-NEXT: [[TMP:%.+]] = getelementptr inbounds [2 x i64], ptr [[CNT]], i64 0, i64 0
 225 // CHECK-NORMAL-NEXT: call void @__kmpc_doacross_wait(ptr [[IDENT]], i32 [[GTID]], ptr [[TMP]])
 226 // CHECK-IRBUILDER-NEXT: [[GTID27:%.+]] = call i32 @__kmpc_global_thread_num(ptr [[IDENT:@.+]])
 227 // CHECK-IRBUILDER-NEXT: call void @__kmpc_doacross_wait(ptr [[IDENT]], i32 [[GTID27]], ptr [[TMP]])
 228 #pragma omp ordered depend(sink : j, i - 2) depend(sink : j - 1, i)
 229         b[i][j] = bar(a[i][j], b[i - 1][j], b[i][j - 1]);
 230 // CHECK: invoke {{.+TestStruct.+bar}}
 231 // CHECK: load ptr, ptr %
 232 // CHECK: load i32, ptr %
 233 // CHECK: load i32, ptr %
 234 // CHECK: load i32, ptr [[J]],
 235 // CHECK-NEXT: sub nsw i32 %{{.+}}, 0
 236 // CHECK-NEXT: sdiv i32 %{{.+}}, 1
 237 // CHECK-NEXT: sext i32 %{{.+}} to i64
 238 // CHECK-NORMAL-NEXT: [[TMP:%.+]] = getelementptr inbounds [2 x i64], ptr [[CNT:%.+]], i64 0, i64 0
 239 // CHECK-NORMAL-NEXT: store i64 %{{.+}}, ptr [[TMP]],
 240 // CHECK-NEXT: [[I:%.+]] = load ptr, ptr [[I_REF]],
 241 // CHECK-NEXT: load i32, ptr [[I]],
 242 // CHECK-NEXT: sub nsw i32 %{{.+}}, 0
 243 // CHECK-NEXT: sdiv i32 %{{.+}}, 2
 244 // CHECK-NEXT: sext i32 %{{.+}} to i64
 245 // CHECK-IRBUILDER-NEXT: [[TMP:%.+]] = getelementptr inbounds [2 x i64], ptr [[CNT:%.+]], i64 0, i64 0
 246 // CHECK-IRBUILDER-NEXT: store i64 %{{.+}}, ptr [[TMP]],
 247 // CHECK-NEXT: [[TMP:%.+]] = getelementptr inbounds [2 x i64], ptr [[CNT]], i64 0, i64 1
 248 // CHECK-NEXT: store i64 %{{.+}}, ptr [[TMP]],
 249 // CHECK-NEXT: [[TMP:%.+]] = getelementptr inbounds [2 x i64], ptr [[CNT]], i64 0, i64 0
 250 // CHECK-NORMAL-NEXT: call void @__kmpc_doacross_post(ptr [[IDENT]], i32 [[GTID]], ptr [[TMP]])
 251 // CHECK-IRBUILDER-NEXT: [[GTID58:%.+]] = call i32 @__kmpc_global_thread_num(ptr [[IDENT:@.+]])
 252 // CHECK-IRBUILDER-NEXT: call void @__kmpc_doacross_post(ptr [[IDENT]], i32 [[GTID58]], ptr [[TMP]])
 253 #pragma omp ordered depend(source)
 254         baz(a[i][j], b[i][j]);
 255       }
 256   }
 257   // CHECK: call void @__kmpc_for_static_fini(
 258   // CHECK-NORMAL: call void @__kmpc_doacross_fini(ptr [[IDENT]], i32 [[GTID]])
 259   // CHECK: ret
 260 };
 261
 262 TestStruct<int> s;
 263 #endif // HEADER