Home | History | Annotate | Download | only in OpenMP
      1 // expected-no-diagnostics
      2 #ifndef HEADER
      3 #define HEADER
      4 
      5 ///==========================================================================///
      6 // RUN: %clang_cc1 -DCK1 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64
      7 // RUN: %clang_cc1 -DCK1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
      8 // RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-64
      9 // RUN: %clang_cc1 -DCK1 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
     10 // RUN: %clang_cc1 -DCK1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
     11 // RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK1 --check-prefix CK1-32
     12 #ifdef CK1
     13 
     14 // CK1: [[ST:%.+]] = type { i32, double* }
     15 template <typename T>
     16 struct ST {
     17   T a;
     18   double *b;
     19 };
     20 
     21 ST<int> gb;
     22 double gc[100];
     23 
     24 // CK1: [[SIZE00:@.+]] = {{.+}}constant [1 x i[[sz:64|32]]] [i{{64|32}} 800]
     25 // CK1: [[MTYPE00:@.+]] = {{.+}}constant [1 x i32] [i32 34]
     26 
     27 // CK1: [[SIZE02:@.+]] = {{.+}}constant [1 x i[[sz]]] [i[[sz]] 4]
     28 // CK1: [[MTYPE02:@.+]] = {{.+}}constant [1 x i32] [i32 33]
     29 
     30 // CK1: [[MTYPE03:@.+]] = {{.+}}constant [1 x i32] [i32 34]
     31 
     32 // CK1: [[SIZE04:@.+]] = {{.+}}constant [2 x i[[sz]]] [i[[sz]] {{8|4}}, i[[sz]] 24]
     33 // CK1: [[MTYPE04:@.+]] = {{.+}}constant [2 x i32] [i32 33, i32 17]
     34 
     35 // CK1-LABEL: _Z3fooi
     36 void foo(int arg) {
     37   int la;
     38   float lb[arg];
     39 
     40   // Region 00
     41   // CK1-DAG: call void @__tgt_target_data_update(i32 [[DEV:%[^,]+]], i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[SIZE00]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE00]]{{.+}})
     42   // CK1-DAG: [[DEV]] = load i32, i32* %{{[^,]+}},
     43   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
     44   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
     45 
     46   // CK1-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
     47   // CK1-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
     48   // CK1-DAG: store i8* bitcast ([100 x double]* @gc to i8*), i8** [[BP0]]
     49   // CK1-DAG: store i8* bitcast ([100 x double]* @gc to i8*), i8** [[P0]]
     50 
     51   // CK1: %{{.+}} = add nsw i32 %{{[^,]+}}, 1
     52   #pragma omp target update if(1+3-5) device(arg) from(gc)
     53   {++arg;}
     54 
     55   // Region 01
     56   // CK1: %{{.+}} = add nsw i32 %{{[^,]+}}, 1
     57   #pragma omp target update to(la) if(1+3-4)
     58   {++arg;}
     59 
     60   // Region 02
     61   // CK1: br i1 %{{[^,]+}}, label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]]
     62   // CK1: [[IFTHEN]]
     63   // CK1-DAG: call void @__tgt_target_data_update(i32 4, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[SIZE02]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE02]]{{.+}})
     64   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
     65   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
     66 
     67   // CK1-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
     68   // CK1-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
     69   // CK1-DAG: store i8* [[CBPVAL0:%[^,]+]], i8** [[BP0]]
     70   // CK1-DAG: store i8* [[CPVAL0:%[^,]+]], i8** [[P0]]
     71   // CK1-DAG: [[CBPVAL0]] = bitcast i32* [[VAR0:%.+]] to i8*
     72   // CK1-DAG: [[CPVAL0]] = bitcast i32* [[VAR0]] to i8*
     73   // CK1: br label %[[IFEND:[^,]+]]
     74 
     75   // CK1: [[IFELSE]]
     76   // CK1: br label %[[IFEND]]
     77   // CK1: [[IFEND]]
     78   // CK1: %{{.+}} = add nsw i32 %{{[^,]+}}, 1
     79   #pragma omp target update to(arg) if(arg) device(4)
     80   {++arg;}
     81 
     82   // CK1: %{{.+}} = add nsw i32 %{{[^,]+}}, 1
     83   {++arg;}
     84 
     85   // Region 03
     86   // CK1-DAG: call void @__tgt_target_data_update(i32 -1, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE03]]{{.+}})
     87   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
     88   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
     89   // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]]
     90 
     91   // CK1-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
     92   // CK1-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
     93   // CK1-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0
     94   // CK1-DAG: store i8* [[CBPVAL0:%[^,]+]], i8** [[BP0]]
     95   // CK1-DAG: store i8* [[CPVAL0:%[^,]+]], i8** [[P0]]
     96   // CK1-DAG: store i[[sz]] [[CSVAL0:%[^,]+]], i[[sz]]* [[S0]]
     97   // CK1-DAG: [[CBPVAL0]] = bitcast float* [[VAR0:%.+]] to i8*
     98   // CK1-DAG: [[CPVAL0]] = bitcast float* [[VAR0]] to i8*
     99   // CK1-DAG: [[CSVAL0]] = mul nuw i[[sz]] %{{[^,]+}}, 4
    100   // CK1: %{{.+}} = add nsw i32 %{{[^,]+}}, 1
    101   // CK1-NOT: __tgt_target_data_end
    102   #pragma omp target update from(lb)
    103   {++arg;}
    104 
    105   // CK1: %{{.+}} = add nsw i32 %{{[^,]+}}, 1
    106   {++arg;}
    107 
    108   // Region 04
    109   // CK1-DAG: call void @__tgt_target_data_update(i32 -1, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[SIZE04]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE04]]{{.+}})
    110   // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
    111   // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
    112 
    113   // CK1-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
    114   // CK1-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
    115   // CK1-DAG: store i8* bitcast ([[ST]]* @gb to i8*), i8** [[BP0]]
    116   // CK1-DAG: store i8* bitcast (double** getelementptr inbounds ([[ST]], [[ST]]* @gb, i32 0, i32 1) to i8*), i8** [[P0]]
    117 
    118 
    119   // CK1-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1
    120   // CK1-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1
    121   // CK1-DAG: store i8* bitcast (double** getelementptr inbounds ([[ST]], [[ST]]* @gb, i32 0, i32 1) to i8*), i8** [[BP1]]
    122   // CK1-DAG: store i8* [[CPVAL1:%[^,]+]], i8** [[P1]]
    123   // CK1-DAG: [[CPVAL1]] = bitcast double* [[SEC1:%.+]] to i8*
    124   // CK1-DAG: [[SEC1]] = getelementptr inbounds {{.+}}double* [[SEC11:%[^,]+]], i{{.+}} 0
    125   // CK1-DAG: [[SEC11]] = load double*, double** getelementptr inbounds ([[ST]], [[ST]]* @gb, i32 0, i32 1),
    126 
    127   // CK1: %{{.+}} = add nsw i32 %{{[^,]+}}, 1
    128   // CK1-NOT: __tgt_target_data_end
    129   #pragma omp target update to(gb.b[:3])
    130   {++arg;}
    131 }
    132 #endif
    133 ///==========================================================================///
    134 // RUN: %clang_cc1 -DCK2 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-64
    135 // RUN: %clang_cc1 -DCK2 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
    136 // RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK2 --check-prefix CK2-64
    137 // RUN: %clang_cc1 -DCK2 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK2 --check-prefix CK2-32
    138 // RUN: %clang_cc1 -DCK2 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
    139 // RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK2 --check-prefix CK2-32
    140 #ifdef CK2
    141 
    142 // CK2: [[ST:%.+]] = type { i32, double* }
    143 template <typename T>
    144 struct ST {
    145   T a;
    146   double *b;
    147 
    148   T foo(T arg) {
    149     // Region 00
    150     #pragma omp target update from(b[1:3]) if(a>123) device(arg)
    151     {arg++;}
    152     return arg;
    153   }
    154 };
    155 
    156 // CK2: [[SIZE00:@.+]] = {{.+}}constant [2 x i[[sz:64|32]]] [i{{64|32}} {{8|4}}, i{{64|32}} 24]
    157 // CK2: [[MTYPE00:@.+]] = {{.+}}constant [2 x i32] [i32 34, i32 18]
    158 
    159 // CK2-LABEL: _Z3bari
    160 int bar(int arg){
    161   ST<int> A;
    162   return A.foo(arg);
    163 }
    164 
    165 // Region 00
    166 // CK2: br i1 %{{[^,]+}}, label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]]
    167 // CK2: [[IFTHEN]]
    168 // CK2-DAG: call void @__tgt_target_data_update(i32 [[DEV:%[^,]+]], i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[SIZE00]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE00]]{{.+}})
    169 // CK2-DAG: [[DEV]] = load i32, i32* %{{[^,]+}},
    170 // CK2-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]]
    171 // CK2-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]]
    172 
    173 // CK2-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0
    174 // CK2-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0
    175 // CK2-DAG: store i8* [[CBPVAL0:%[^,]+]], i8** [[BP0]]
    176 // CK2-DAG: store i8* [[CPVAL0:%[^,]+]], i8** [[P0]]
    177 // CK2-DAG: [[CBPVAL0]] = bitcast [[ST]]* [[VAR0:%.+]] to i8*
    178 // CK2-DAG: [[CPVAL0]] = bitcast double** [[SEC0:%[^,]+]] to i8*
    179 // CK2-DAG: [[SEC0]] = getelementptr inbounds {{.*}}[[ST]]* [[VAR0]], i32 0, i32 1
    180 
    181 
    182 // CK2-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1
    183 // CK2-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1
    184 // CK2-DAG: store i8* [[CBPVAL1:%[^,]+]], i8** [[BP1]]
    185 // CK2-DAG: store i8* [[CPVAL1:%[^,]+]], i8** [[P1]]
    186 // CK2-DAG: [[CBPVAL1]] = bitcast double** [[SEC0]] to i8*
    187 // CK2-DAG: [[CPVAL1]] = bitcast double* [[SEC1:%[^,]+]] to i8*
    188 // CK2-DAG: [[SEC1]] = getelementptr inbounds {{.*}}double* [[SEC11:%[^,]+]], i{{.+}} 1
    189 // CK2-DAG: [[SEC11]] = load double*, double** [[SEC111:%[^,]+]],
    190 // CK2-DAG: [[SEC111]] = getelementptr inbounds {{.*}}[[ST]]* [[VAR0]], i32 0, i32 1
    191 
    192 // CK2: br label %[[IFEND:[^,]+]]
    193 
    194 // CK2: [[IFELSE]]
    195 // CK2: br label %[[IFEND]]
    196 // CK2: [[IFEND]]
    197 // CK2: %{{.+}} = add nsw i32 %{{[^,]+}}, 1
    198 #endif
    199 ///==========================================================================///
    200 // RUN: %clang_cc1 -DCK3 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-64
    201 // RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
    202 // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK3 --check-prefix CK3-64
    203 // RUN: %clang_cc1 -DCK3 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK3 --check-prefix CK3-32
    204 // RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
    205 // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK3 --check-prefix CK3-32
    206 #ifdef CK3
    207 
    208 // CK3-LABEL: no_target_devices
    209 void no_target_devices(int arg) {
    210   // CK3-NOT: tgt_target_data_update
    211   // CK3: %{{.+}} = add nsw i32 %{{[^,]+}}, 1
    212   // CK3: ret
    213   #pragma omp target update to(arg) if(arg) device(4)
    214   {++arg;}
    215 }
    216 #endif
    217 ///==========================================================================///
    218 // RUN: %clang_cc1 -DCK4 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-64
    219 // RUN: %clang_cc1 -DCK4 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s
    220 // RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK4 --check-prefix CK4-64
    221 // RUN: %clang_cc1 -DCK4 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s  --check-prefix CK4 --check-prefix CK4-32
    222 // RUN: %clang_cc1 -DCK4 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s
    223 // RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s  --check-prefix CK4 --check-prefix CK4-32
    224 
    225 // RUN: %clang_cc1 -DCK4 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
    226 // RUN: %clang_cc1 -DCK4 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCK4 --check-prefix TCK4-64
    227 // RUN: %clang_cc1 -DCK4 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
    228 // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCK4 --check-prefix TCK4-64
    229 // RUN: %clang_cc1 -DCK4 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
    230 // RUN: %clang_cc1 -DCK4 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCK4 --check-prefix TCK4-32
    231 // RUN: %clang_cc1 -DCK4 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
    232 // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCK4 --check-prefix TCK4-32
    233 #ifdef CK4
    234 
    235 // CK4-LABEL: device_side_scan
    236 void device_side_scan(int arg) {
    237   // CK4: tgt_target_data_update
    238   // CK4: %{{.+}} = add nsw i32 %{{[^,]+}}, 1
    239   // CK4: ret
    240   // TCK4-NOT: tgt_target_data_update
    241   #pragma omp target update from(arg) if(arg) device(4)
    242   {++arg;}
    243 }
    244 #endif
    245 #endif
    246