clang/test/OpenMP/target_offload_mandatory_codegen.cpp

   1 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs --replace-value-regex "__omp_offloading_[0-9a-z]+_[0-9a-z]+"
   2 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -fopenmp-offload-mandatory -emit-llvm %s -o - | FileCheck %s --check-prefix=MANDATORY
   3 // expected-no-diagnostics
   4
   5 void foo() {}
   6 #pragma omp declare target(foo)
   7
   8 void bar() {}
   9 #pragma omp declare target device_type(nohost) to(bar)
  10
  11 void host() {
  12 #pragma omp target
  13   { bar(); }
  14 }
  15
  16 void host_if(bool cond) {
  17 #pragma omp target if(cond)
  18   { bar(); }
  19 }
  20
  21 void host_dev(int device) {
  22 #pragma omp target device(device)
  23   { bar(); }
  24 }
  25 // MANDATORY-LABEL: define {{[^@]+}}@_Z3foov
  26 // MANDATORY-SAME: () #[[ATTR0:[0-9]+]] {
  27 // MANDATORY-NEXT:  entry:
  28 // MANDATORY-NEXT:    ret void
  29 //
  30 //
  31 // MANDATORY-LABEL: define {{[^@]+}}@_Z4hostv
  32 // MANDATORY-SAME: () #[[ATTR0]] {
  33 // MANDATORY-NEXT:  entry:
  34 // MANDATORY-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
  35 // MANDATORY-NEXT:    [[TMP0:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
  36 // MANDATORY-NEXT:    store i32 2, ptr [[TMP0]], align 4
  37 // MANDATORY-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
  38 // MANDATORY-NEXT:    store i32 0, ptr [[TMP1]], align 4
  39 // MANDATORY-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
  40 // MANDATORY-NEXT:    store ptr null, ptr [[TMP2]], align 8
  41 // MANDATORY-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
  42 // MANDATORY-NEXT:    store ptr null, ptr [[TMP3]], align 8
  43 // MANDATORY-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
  44 // MANDATORY-NEXT:    store ptr null, ptr [[TMP4]], align 8
  45 // MANDATORY-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
  46 // MANDATORY-NEXT:    store ptr null, ptr [[TMP5]], align 8
  47 // MANDATORY-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
  48 // MANDATORY-NEXT:    store ptr null, ptr [[TMP6]], align 8
  49 // MANDATORY-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
  50 // MANDATORY-NEXT:    store ptr null, ptr [[TMP7]], align 8
  51 // MANDATORY-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
  52 // MANDATORY-NEXT:    store i64 0, ptr [[TMP8]], align 8
  53 // MANDATORY-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
  54 // MANDATORY-NEXT:    store i64 0, ptr [[TMP9]], align 8
  55 // MANDATORY-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
  56 // MANDATORY-NEXT:    store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP10]], align 4
  57 // MANDATORY-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
  58 // MANDATORY-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP11]], align 4
  59 // MANDATORY-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
  60 // MANDATORY-NEXT:    store i32 0, ptr [[TMP12]], align 4
  61 // MANDATORY-NEXT:    [[TMP13:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z4hostv_l12.region_id, ptr [[KERNEL_ARGS]])
  62 // MANDATORY-NEXT:    [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0
  63 // MANDATORY-NEXT:    br i1 [[TMP14]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
  64 // MANDATORY:       omp_offload.failed:
  65 // MANDATORY-NEXT:    unreachable
  66 // MANDATORY:       omp_offload.cont:
  67 // MANDATORY-NEXT:    ret void
  68 //
  69 //
  70 // MANDATORY-LABEL: define {{[^@]+}}@_Z7host_ifb
  71 // MANDATORY-SAME: (i1 noundef zeroext [[COND:%.*]]) #[[ATTR0]] {
  72 // MANDATORY-NEXT:  entry:
  73 // MANDATORY-NEXT:    [[COND_ADDR:%.*]] = alloca i8, align 1
  74 // MANDATORY-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
  75 // MANDATORY-NEXT:    [[FROMBOOL:%.*]] = zext i1 [[COND]] to i8
  76 // MANDATORY-NEXT:    store i8 [[FROMBOOL]], ptr [[COND_ADDR]], align 1
  77 // MANDATORY-NEXT:    [[TMP0:%.*]] = load i8, ptr [[COND_ADDR]], align 1
  78 // MANDATORY-NEXT:    [[TOBOOL:%.*]] = trunc i8 [[TMP0]] to i1
  79 // MANDATORY-NEXT:    br i1 [[TOBOOL]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
  80 // MANDATORY:       omp_if.then:
  81 // MANDATORY-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
  82 // MANDATORY-NEXT:    store i32 2, ptr [[TMP1]], align 4
  83 // MANDATORY-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
  84 // MANDATORY-NEXT:    store i32 0, ptr [[TMP2]], align 4
  85 // MANDATORY-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
  86 // MANDATORY-NEXT:    store ptr null, ptr [[TMP3]], align 8
  87 // MANDATORY-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
  88 // MANDATORY-NEXT:    store ptr null, ptr [[TMP4]], align 8
  89 // MANDATORY-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
  90 // MANDATORY-NEXT:    store ptr null, ptr [[TMP5]], align 8
  91 // MANDATORY-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
  92 // MANDATORY-NEXT:    store ptr null, ptr [[TMP6]], align 8
  93 // MANDATORY-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
  94 // MANDATORY-NEXT:    store ptr null, ptr [[TMP7]], align 8
  95 // MANDATORY-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
  96 // MANDATORY-NEXT:    store ptr null, ptr [[TMP8]], align 8
  97 // MANDATORY-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
  98 // MANDATORY-NEXT:    store i64 0, ptr [[TMP9]], align 8
  99 // MANDATORY-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
 100 // MANDATORY-NEXT:    store i64 0, ptr [[TMP10]], align 8
 101 // MANDATORY-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
 102 // MANDATORY-NEXT:    store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP11]], align 4
 103 // MANDATORY-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
 104 // MANDATORY-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP12]], align 4
 105 // MANDATORY-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
 106 // MANDATORY-NEXT:    store i32 0, ptr [[TMP13]], align 4
 107 // MANDATORY-NEXT:    [[TMP14:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z7host_ifb_l17.region_id, ptr [[KERNEL_ARGS]])
 108 // MANDATORY-NEXT:    [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0
 109 // MANDATORY-NEXT:    br i1 [[TMP15]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
 110 // MANDATORY:       omp_offload.failed:
 111 // MANDATORY-NEXT:    unreachable
 112 // MANDATORY:       omp_offload.cont:
 113 // MANDATORY-NEXT:    br label [[OMP_IF_END:%.*]]
 114 // MANDATORY:       omp_if.else:
 115 // MANDATORY-NEXT:    unreachable
 116 // MANDATORY:       omp_if.end:
 117 // MANDATORY-NEXT:    ret void
 118 //
 119 //
 120 // MANDATORY-LABEL: define {{[^@]+}}@_Z8host_devi
 121 // MANDATORY-SAME: (i32 noundef signext [[DEVICE:%.*]]) #[[ATTR0]] {
 122 // MANDATORY-NEXT:  entry:
 123 // MANDATORY-NEXT:    [[DEVICE_ADDR:%.*]] = alloca i32, align 4
 124 // MANDATORY-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 125 // MANDATORY-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
 126 // MANDATORY-NEXT:    store i32 [[DEVICE]], ptr [[DEVICE_ADDR]], align 4
 127 // MANDATORY-NEXT:    [[TMP0:%.*]] = load i32, ptr [[DEVICE_ADDR]], align 4
 128 // MANDATORY-NEXT:    store i32 [[TMP0]], ptr [[DOTCAPTURE_EXPR_]], align 4
 129 // MANDATORY-NEXT:    [[TMP1:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 130 // MANDATORY-NEXT:    [[TMP2:%.*]] = sext i32 [[TMP1]] to i64
 131 // MANDATORY-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
 132 // MANDATORY-NEXT:    store i32 2, ptr [[TMP3]], align 4
 133 // MANDATORY-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
 134 // MANDATORY-NEXT:    store i32 0, ptr [[TMP4]], align 4
 135 // MANDATORY-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
 136 // MANDATORY-NEXT:    store ptr null, ptr [[TMP5]], align 8
 137 // MANDATORY-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
 138 // MANDATORY-NEXT:    store ptr null, ptr [[TMP6]], align 8
 139 // MANDATORY-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
 140 // MANDATORY-NEXT:    store ptr null, ptr [[TMP7]], align 8
 141 // MANDATORY-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
 142 // MANDATORY-NEXT:    store ptr null, ptr [[TMP8]], align 8
 143 // MANDATORY-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
 144 // MANDATORY-NEXT:    store ptr null, ptr [[TMP9]], align 8
 145 // MANDATORY-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
 146 // MANDATORY-NEXT:    store ptr null, ptr [[TMP10]], align 8
 147 // MANDATORY-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
 148 // MANDATORY-NEXT:    store i64 0, ptr [[TMP11]], align 8
 149 // MANDATORY-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
 150 // MANDATORY-NEXT:    store i64 0, ptr [[TMP12]], align 8
 151 // MANDATORY-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
 152 // MANDATORY-NEXT:    store [3 x i32] [i32 -1, i32 0, i32 0], ptr [[TMP13]], align 4
 153 // MANDATORY-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
 154 // MANDATORY-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP14]], align 4
 155 // MANDATORY-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
 156 // MANDATORY-NEXT:    store i32 0, ptr [[TMP15]], align 4
 157 // MANDATORY-NEXT:    [[TMP16:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 [[TMP2]], i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z8host_devi_l22.region_id, ptr [[KERNEL_ARGS]])
 158 // MANDATORY-NEXT:    [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0
 159 // MANDATORY-NEXT:    br i1 [[TMP17]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
 160 // MANDATORY:       omp_offload.failed:
 161 // MANDATORY-NEXT:    unreachable
 162 // MANDATORY:       omp_offload.cont:
 163 // MANDATORY-NEXT:    ret void
 164 //
 165 //
 166 // MANDATORY-LABEL: define {{[^@]+}}@.omp_offloading.requires_reg
 167 // MANDATORY-SAME: () #[[ATTR3:[0-9]+]] {
 168 // MANDATORY-NEXT:  entry:
 169 // MANDATORY-NEXT:    call void @__tgt_register_requires(i64 1)
 170 // MANDATORY-NEXT:    ret void
 171 //