clang/test/Headers/amdgcn_openmp_device_math.c

   1 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
   2 // RUN: %clang_cc1 -internal-isystem %S/Inputs/include -x c -fopenmp -triple x86_64-unknown-unknown -fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm-bc %s -o %t-host.bc
   3 // RUN: %clang_cc1 -internal-isystem %S/../../lib/Headers/openmp_wrappers -include __clang_openmp_device_functions.h -internal-isystem %S/../../lib/Headers/openmp_wrappers -internal-isystem %S/Inputs/include -x c -fopenmp -triple amdgcn-amd-amdhsa -aux-triple x86_64-unknown-unknown -fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-host.bc -o - | FileCheck %s --check-prefixes=CHECK,CHECK-C
   4 // RUN: %clang_cc1 -internal-isystem %S/Inputs/include -x c++ -fopenmp -triple x86_64-unknown-unknown -fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm-bc %s -o %t-host.bc
   5 // RUN: %clang_cc1 -internal-isystem %S/../../lib/Headers/openmp_wrappers -include __clang_openmp_device_functions.h -internal-isystem %S/../../lib/Headers/openmp_wrappers -internal-isystem %S/Inputs/include -x c++ -fopenmp -triple amdgcn-amd-amdhsa -aux-triple x86_64-unknown-unknown -fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-host.bc -o - | FileCheck %s --check-prefixes=CHECK,CHECK-CPP
   6
   7 #ifdef __cplusplus
   8 #include <cmath>
   9 #else
  10 #include <math.h>
  11 #endif
  12
  13 #pragma omp begin declare target
  14
  15 // CHECK-C-LABEL: @test_math_f64(
  16 // CHECK-C-NEXT:  entry:
  17 // CHECK-C-NEXT:    [[RETVAL_I13:%.*]] = alloca double, align 8, addrspace(5)
  18 // CHECK-C-NEXT:    [[__X_ADDR_I14:%.*]] = alloca double, align 8, addrspace(5)
  19 // CHECK-C-NEXT:    [[__Y_ADDR_I:%.*]] = alloca double, align 8, addrspace(5)
  20 // CHECK-C-NEXT:    [[__Z_ADDR_I:%.*]] = alloca double, align 8, addrspace(5)
  21 // CHECK-C-NEXT:    [[RETVAL_I9:%.*]] = alloca double, align 8, addrspace(5)
  22 // CHECK-C-NEXT:    [[__X_ADDR_I10:%.*]] = alloca double, align 8, addrspace(5)
  23 // CHECK-C-NEXT:    [[RETVAL_I4:%.*]] = alloca double, align 8, addrspace(5)
  24 // CHECK-C-NEXT:    [[__X_ADDR_I5:%.*]] = alloca double, align 8, addrspace(5)
  25 // CHECK-C-NEXT:    [[RETVAL_I:%.*]] = alloca double, align 8, addrspace(5)
  26 // CHECK-C-NEXT:    [[__X_ADDR_I:%.*]] = alloca double, align 8, addrspace(5)
  27 // CHECK-C-NEXT:    [[X_ADDR:%.*]] = alloca double, align 8, addrspace(5)
  28 // CHECK-C-NEXT:    [[Y_ADDR:%.*]] = alloca double, align 8, addrspace(5)
  29 // CHECK-C-NEXT:    [[Z_ADDR:%.*]] = alloca double, align 8, addrspace(5)
  30 // CHECK-C-NEXT:    [[L1:%.*]] = alloca double, align 8, addrspace(5)
  31 // CHECK-C-NEXT:    [[L2:%.*]] = alloca double, align 8, addrspace(5)
  32 // CHECK-C-NEXT:    [[L3:%.*]] = alloca double, align 8, addrspace(5)
  33 // CHECK-C-NEXT:    [[L4:%.*]] = alloca double, align 8, addrspace(5)
  34 // CHECK-C-NEXT:    [[X_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[X_ADDR]] to ptr
  35 // CHECK-C-NEXT:    [[Y_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[Y_ADDR]] to ptr
  36 // CHECK-C-NEXT:    [[Z_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[Z_ADDR]] to ptr
  37 // CHECK-C-NEXT:    [[L1_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L1]] to ptr
  38 // CHECK-C-NEXT:    [[L2_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L2]] to ptr
  39 // CHECK-C-NEXT:    [[L3_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L3]] to ptr
  40 // CHECK-C-NEXT:    [[L4_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L4]] to ptr
  41 // CHECK-C-NEXT:    store double [[X:%.*]], ptr [[X_ADDR_ASCAST]], align 8
  42 // CHECK-C-NEXT:    store double [[Y:%.*]], ptr [[Y_ADDR_ASCAST]], align 8
  43 // CHECK-C-NEXT:    store double [[Z:%.*]], ptr [[Z_ADDR_ASCAST]], align 8
  44 // CHECK-C-NEXT:    [[TMP0:%.*]] = load double, ptr [[X_ADDR_ASCAST]], align 8
  45 // CHECK-C-NEXT:    [[RETVAL_ASCAST_I:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I]] to ptr
  46 // CHECK-C-NEXT:    [[__X_ADDR_ASCAST_I:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I]] to ptr
  47 // CHECK-C-NEXT:    store double [[TMP0]], ptr [[__X_ADDR_ASCAST_I]], align 8
  48 // CHECK-C-NEXT:    [[TMP1:%.*]] = load double, ptr [[__X_ADDR_ASCAST_I]], align 8
  49 // CHECK-C-NEXT:    [[CALL_I:%.*]] = call double @__ocml_sin_f64(double noundef [[TMP1]]) #[[ATTR3:[0-9]+]]
  50 // CHECK-C-NEXT:    store double [[CALL_I]], ptr [[L1_ASCAST]], align 8
  51 // CHECK-C-NEXT:    [[TMP2:%.*]] = load double, ptr [[X_ADDR_ASCAST]], align 8
  52 // CHECK-C-NEXT:    [[RETVAL_ASCAST_I6:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I4]] to ptr
  53 // CHECK-C-NEXT:    [[__X_ADDR_ASCAST_I7:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I5]] to ptr
  54 // CHECK-C-NEXT:    store double [[TMP2]], ptr [[__X_ADDR_ASCAST_I7]], align 8
  55 // CHECK-C-NEXT:    [[TMP3:%.*]] = load double, ptr [[__X_ADDR_ASCAST_I7]], align 8
  56 // CHECK-C-NEXT:    [[CALL_I8:%.*]] = call double @__ocml_cos_f64(double noundef [[TMP3]]) #[[ATTR3]]
  57 // CHECK-C-NEXT:    store double [[CALL_I8]], ptr [[L2_ASCAST]], align 8
  58 // CHECK-C-NEXT:    [[TMP4:%.*]] = load double, ptr [[X_ADDR_ASCAST]], align 8
  59 // CHECK-C-NEXT:    [[RETVAL_ASCAST_I11:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I9]] to ptr
  60 // CHECK-C-NEXT:    [[__X_ADDR_ASCAST_I12:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I10]] to ptr
  61 // CHECK-C-NEXT:    store double [[TMP4]], ptr [[__X_ADDR_ASCAST_I12]], align 8
  62 // CHECK-C-NEXT:    [[TMP5:%.*]] = load double, ptr [[__X_ADDR_ASCAST_I12]], align 8
  63 // CHECK-C-NEXT:    [[TMP6:%.*]] = call double @llvm.fabs.f64(double [[TMP5]])
  64 // CHECK-C-NEXT:    store double [[TMP6]], ptr [[L3_ASCAST]], align 8
  65 // CHECK-C-NEXT:    [[TMP7:%.*]] = load double, ptr [[X_ADDR_ASCAST]], align 8
  66 // CHECK-C-NEXT:    [[TMP8:%.*]] = load double, ptr [[Y_ADDR_ASCAST]], align 8
  67 // CHECK-C-NEXT:    [[TMP9:%.*]] = load double, ptr [[Z_ADDR_ASCAST]], align 8
  68 // CHECK-C-NEXT:    [[RETVAL_ASCAST_I15:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I13]] to ptr
  69 // CHECK-C-NEXT:    [[__X_ADDR_ASCAST_I16:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I14]] to ptr
  70 // CHECK-C-NEXT:    [[__Y_ADDR_ASCAST_I:%.*]] = addrspacecast ptr addrspace(5) [[__Y_ADDR_I]] to ptr
  71 // CHECK-C-NEXT:    [[__Z_ADDR_ASCAST_I:%.*]] = addrspacecast ptr addrspace(5) [[__Z_ADDR_I]] to ptr
  72 // CHECK-C-NEXT:    store double [[TMP7]], ptr [[__X_ADDR_ASCAST_I16]], align 8
  73 // CHECK-C-NEXT:    store double [[TMP8]], ptr [[__Y_ADDR_ASCAST_I]], align 8
  74 // CHECK-C-NEXT:    store double [[TMP9]], ptr [[__Z_ADDR_ASCAST_I]], align 8
  75 // CHECK-C-NEXT:    [[TMP10:%.*]] = load double, ptr [[__X_ADDR_ASCAST_I16]], align 8
  76 // CHECK-C-NEXT:    [[TMP11:%.*]] = load double, ptr [[__Y_ADDR_ASCAST_I]], align 8
  77 // CHECK-C-NEXT:    [[TMP12:%.*]] = load double, ptr [[__Z_ADDR_ASCAST_I]], align 8
  78 // CHECK-C-NEXT:    [[TMP13:%.*]] = call double @llvm.fma.f64(double [[TMP10]], double [[TMP11]], double [[TMP12]])
  79 // CHECK-C-NEXT:    store double [[TMP13]], ptr [[L4_ASCAST]], align 8
  80 // CHECK-C-NEXT:    ret void
  81 //
  82 // CHECK-CPP-LABEL: @_Z13test_math_f64ddd(
  83 // CHECK-CPP-NEXT:  entry:
  84 // CHECK-CPP-NEXT:    [[RETVAL_I13:%.*]] = alloca double, align 8, addrspace(5)
  85 // CHECK-CPP-NEXT:    [[__X_ADDR_I14:%.*]] = alloca double, align 8, addrspace(5)
  86 // CHECK-CPP-NEXT:    [[__Y_ADDR_I:%.*]] = alloca double, align 8, addrspace(5)
  87 // CHECK-CPP-NEXT:    [[__Z_ADDR_I:%.*]] = alloca double, align 8, addrspace(5)
  88 // CHECK-CPP-NEXT:    [[RETVAL_I9:%.*]] = alloca double, align 8, addrspace(5)
  89 // CHECK-CPP-NEXT:    [[__X_ADDR_I10:%.*]] = alloca double, align 8, addrspace(5)
  90 // CHECK-CPP-NEXT:    [[RETVAL_I4:%.*]] = alloca double, align 8, addrspace(5)
  91 // CHECK-CPP-NEXT:    [[__X_ADDR_I5:%.*]] = alloca double, align 8, addrspace(5)
  92 // CHECK-CPP-NEXT:    [[RETVAL_I:%.*]] = alloca double, align 8, addrspace(5)
  93 // CHECK-CPP-NEXT:    [[__X_ADDR_I:%.*]] = alloca double, align 8, addrspace(5)
  94 // CHECK-CPP-NEXT:    [[X_ADDR:%.*]] = alloca double, align 8, addrspace(5)
  95 // CHECK-CPP-NEXT:    [[Y_ADDR:%.*]] = alloca double, align 8, addrspace(5)
  96 // CHECK-CPP-NEXT:    [[Z_ADDR:%.*]] = alloca double, align 8, addrspace(5)
  97 // CHECK-CPP-NEXT:    [[L1:%.*]] = alloca double, align 8, addrspace(5)
  98 // CHECK-CPP-NEXT:    [[L2:%.*]] = alloca double, align 8, addrspace(5)
  99 // CHECK-CPP-NEXT:    [[L3:%.*]] = alloca double, align 8, addrspace(5)
 100 // CHECK-CPP-NEXT:    [[L4:%.*]] = alloca double, align 8, addrspace(5)
 101 // CHECK-CPP-NEXT:    [[X_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[X_ADDR]] to ptr
 102 // CHECK-CPP-NEXT:    [[Y_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[Y_ADDR]] to ptr
 103 // CHECK-CPP-NEXT:    [[Z_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[Z_ADDR]] to ptr
 104 // CHECK-CPP-NEXT:    [[L1_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L1]] to ptr
 105 // CHECK-CPP-NEXT:    [[L2_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L2]] to ptr
 106 // CHECK-CPP-NEXT:    [[L3_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L3]] to ptr
 107 // CHECK-CPP-NEXT:    [[L4_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L4]] to ptr
 108 // CHECK-CPP-NEXT:    store double [[X:%.*]], ptr [[X_ADDR_ASCAST]], align 8
 109 // CHECK-CPP-NEXT:    store double [[Y:%.*]], ptr [[Y_ADDR_ASCAST]], align 8
 110 // CHECK-CPP-NEXT:    store double [[Z:%.*]], ptr [[Z_ADDR_ASCAST]], align 8
 111 // CHECK-CPP-NEXT:    [[TMP0:%.*]] = load double, ptr [[X_ADDR_ASCAST]], align 8
 112 // CHECK-CPP-NEXT:    [[RETVAL_ASCAST_I:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I]] to ptr
 113 // CHECK-CPP-NEXT:    [[__X_ADDR_ASCAST_I:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I]] to ptr
 114 // CHECK-CPP-NEXT:    store double [[TMP0]], ptr [[__X_ADDR_ASCAST_I]], align 8
 115 // CHECK-CPP-NEXT:    [[TMP1:%.*]] = load double, ptr [[__X_ADDR_ASCAST_I]], align 8
 116 // CHECK-CPP-NEXT:    [[CALL_I:%.*]] = call noundef double @__ocml_sin_f64(double noundef [[TMP1]]) #[[ATTR3:[0-9]+]]
 117 // CHECK-CPP-NEXT:    store double [[CALL_I]], ptr [[L1_ASCAST]], align 8
 118 // CHECK-CPP-NEXT:    [[TMP2:%.*]] = load double, ptr [[X_ADDR_ASCAST]], align 8
 119 // CHECK-CPP-NEXT:    [[RETVAL_ASCAST_I6:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I4]] to ptr
 120 // CHECK-CPP-NEXT:    [[__X_ADDR_ASCAST_I7:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I5]] to ptr
 121 // CHECK-CPP-NEXT:    store double [[TMP2]], ptr [[__X_ADDR_ASCAST_I7]], align 8
 122 // CHECK-CPP-NEXT:    [[TMP3:%.*]] = load double, ptr [[__X_ADDR_ASCAST_I7]], align 8
 123 // CHECK-CPP-NEXT:    [[CALL_I8:%.*]] = call noundef double @__ocml_cos_f64(double noundef [[TMP3]]) #[[ATTR3]]
 124 // CHECK-CPP-NEXT:    store double [[CALL_I8]], ptr [[L2_ASCAST]], align 8
 125 // CHECK-CPP-NEXT:    [[TMP4:%.*]] = load double, ptr [[X_ADDR_ASCAST]], align 8
 126 // CHECK-CPP-NEXT:    [[RETVAL_ASCAST_I11:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I9]] to ptr
 127 // CHECK-CPP-NEXT:    [[__X_ADDR_ASCAST_I12:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I10]] to ptr
 128 // CHECK-CPP-NEXT:    store double [[TMP4]], ptr [[__X_ADDR_ASCAST_I12]], align 8
 129 // CHECK-CPP-NEXT:    [[TMP5:%.*]] = load double, ptr [[__X_ADDR_ASCAST_I12]], align 8
 130 // CHECK-CPP-NEXT:    [[TMP6:%.*]] = call noundef double @llvm.fabs.f64(double [[TMP5]])
 131 // CHECK-CPP-NEXT:    store double [[TMP6]], ptr [[L3_ASCAST]], align 8
 132 // CHECK-CPP-NEXT:    [[TMP7:%.*]] = load double, ptr [[X_ADDR_ASCAST]], align 8
 133 // CHECK-CPP-NEXT:    [[TMP8:%.*]] = load double, ptr [[Y_ADDR_ASCAST]], align 8
 134 // CHECK-CPP-NEXT:    [[TMP9:%.*]] = load double, ptr [[Z_ADDR_ASCAST]], align 8
 135 // CHECK-CPP-NEXT:    [[RETVAL_ASCAST_I15:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I13]] to ptr
 136 // CHECK-CPP-NEXT:    [[__X_ADDR_ASCAST_I16:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I14]] to ptr
 137 // CHECK-CPP-NEXT:    [[__Y_ADDR_ASCAST_I:%.*]] = addrspacecast ptr addrspace(5) [[__Y_ADDR_I]] to ptr
 138 // CHECK-CPP-NEXT:    [[__Z_ADDR_ASCAST_I:%.*]] = addrspacecast ptr addrspace(5) [[__Z_ADDR_I]] to ptr
 139 // CHECK-CPP-NEXT:    store double [[TMP7]], ptr [[__X_ADDR_ASCAST_I16]], align 8
 140 // CHECK-CPP-NEXT:    store double [[TMP8]], ptr [[__Y_ADDR_ASCAST_I]], align 8
 141 // CHECK-CPP-NEXT:    store double [[TMP9]], ptr [[__Z_ADDR_ASCAST_I]], align 8
 142 // CHECK-CPP-NEXT:    [[TMP10:%.*]] = load double, ptr [[__X_ADDR_ASCAST_I16]], align 8
 143 // CHECK-CPP-NEXT:    [[TMP11:%.*]] = load double, ptr [[__Y_ADDR_ASCAST_I]], align 8
 144 // CHECK-CPP-NEXT:    [[TMP12:%.*]] = load double, ptr [[__Z_ADDR_ASCAST_I]], align 8
 145 // CHECK-CPP-NEXT:    [[TMP13:%.*]] = call noundef double @llvm.fma.f64(double [[TMP10]], double [[TMP11]], double [[TMP12]])
 146 // CHECK-CPP-NEXT:    store double [[TMP13]], ptr [[L4_ASCAST]], align 8
 147 // CHECK-CPP-NEXT:    ret void
 148 //
 149 void test_math_f64(double x, double y, double z) {
 150   double l1 = sin(x);
 151   double l2 = cos(x);
 152   double l3 = fabs(x);
 153   double l4 = fma(x, y, z);
 154 }
 155
 156 // CHECK-C-LABEL: @test_math_f32(
 157 // CHECK-C-NEXT:  entry:
 158 // CHECK-C-NEXT:    [[RETVAL_I22:%.*]] = alloca double, align 8, addrspace(5)
 159 // CHECK-C-NEXT:    [[__X_ADDR_I23:%.*]] = alloca double, align 8, addrspace(5)
 160 // CHECK-C-NEXT:    [[__Y_ADDR_I:%.*]] = alloca double, align 8, addrspace(5)
 161 // CHECK-C-NEXT:    [[__Z_ADDR_I:%.*]] = alloca double, align 8, addrspace(5)
 162 // CHECK-C-NEXT:    [[RETVAL_I18:%.*]] = alloca double, align 8, addrspace(5)
 163 // CHECK-C-NEXT:    [[__X_ADDR_I19:%.*]] = alloca double, align 8, addrspace(5)
 164 // CHECK-C-NEXT:    [[RETVAL_I13:%.*]] = alloca double, align 8, addrspace(5)
 165 // CHECK-C-NEXT:    [[__X_ADDR_I14:%.*]] = alloca double, align 8, addrspace(5)
 166 // CHECK-C-NEXT:    [[RETVAL_I:%.*]] = alloca double, align 8, addrspace(5)
 167 // CHECK-C-NEXT:    [[__X_ADDR_I:%.*]] = alloca double, align 8, addrspace(5)
 168 // CHECK-C-NEXT:    [[X_ADDR:%.*]] = alloca float, align 4, addrspace(5)
 169 // CHECK-C-NEXT:    [[Y_ADDR:%.*]] = alloca float, align 4, addrspace(5)
 170 // CHECK-C-NEXT:    [[Z_ADDR:%.*]] = alloca float, align 4, addrspace(5)
 171 // CHECK-C-NEXT:    [[L1:%.*]] = alloca float, align 4, addrspace(5)
 172 // CHECK-C-NEXT:    [[L2:%.*]] = alloca float, align 4, addrspace(5)
 173 // CHECK-C-NEXT:    [[L3:%.*]] = alloca float, align 4, addrspace(5)
 174 // CHECK-C-NEXT:    [[L4:%.*]] = alloca float, align 4, addrspace(5)
 175 // CHECK-C-NEXT:    [[X_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[X_ADDR]] to ptr
 176 // CHECK-C-NEXT:    [[Y_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[Y_ADDR]] to ptr
 177 // CHECK-C-NEXT:    [[Z_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[Z_ADDR]] to ptr
 178 // CHECK-C-NEXT:    [[L1_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L1]] to ptr
 179 // CHECK-C-NEXT:    [[L2_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L2]] to ptr
 180 // CHECK-C-NEXT:    [[L3_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L3]] to ptr
 181 // CHECK-C-NEXT:    [[L4_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L4]] to ptr
 182 // CHECK-C-NEXT:    store float [[X:%.*]], ptr [[X_ADDR_ASCAST]], align 4
 183 // CHECK-C-NEXT:    store float [[Y:%.*]], ptr [[Y_ADDR_ASCAST]], align 4
 184 // CHECK-C-NEXT:    store float [[Z:%.*]], ptr [[Z_ADDR_ASCAST]], align 4
 185 // CHECK-C-NEXT:    [[TMP0:%.*]] = load float, ptr [[X_ADDR_ASCAST]], align 4
 186 // CHECK-C-NEXT:    [[CONV:%.*]] = fpext float [[TMP0]] to double
 187 // CHECK-C-NEXT:    [[RETVAL_ASCAST_I:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I]] to ptr
 188 // CHECK-C-NEXT:    [[__X_ADDR_ASCAST_I:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I]] to ptr
 189 // CHECK-C-NEXT:    store double [[CONV]], ptr [[__X_ADDR_ASCAST_I]], align 8
 190 // CHECK-C-NEXT:    [[TMP1:%.*]] = load double, ptr [[__X_ADDR_ASCAST_I]], align 8
 191 // CHECK-C-NEXT:    [[CALL_I:%.*]] = call double @__ocml_sin_f64(double noundef [[TMP1]]) #[[ATTR3]]
 192 // CHECK-C-NEXT:    [[CONV1:%.*]] = fptrunc double [[CALL_I]] to float
 193 // CHECK-C-NEXT:    store float [[CONV1]], ptr [[L1_ASCAST]], align 4
 194 // CHECK-C-NEXT:    [[TMP2:%.*]] = load float, ptr [[X_ADDR_ASCAST]], align 4
 195 // CHECK-C-NEXT:    [[CONV2:%.*]] = fpext float [[TMP2]] to double
 196 // CHECK-C-NEXT:    [[RETVAL_ASCAST_I15:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I13]] to ptr
 197 // CHECK-C-NEXT:    [[__X_ADDR_ASCAST_I16:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I14]] to ptr
 198 // CHECK-C-NEXT:    store double [[CONV2]], ptr [[__X_ADDR_ASCAST_I16]], align 8
 199 // CHECK-C-NEXT:    [[TMP3:%.*]] = load double, ptr [[__X_ADDR_ASCAST_I16]], align 8
 200 // CHECK-C-NEXT:    [[CALL_I17:%.*]] = call double @__ocml_cos_f64(double noundef [[TMP3]]) #[[ATTR3]]
 201 // CHECK-C-NEXT:    [[CONV4:%.*]] = fptrunc double [[CALL_I17]] to float
 202 // CHECK-C-NEXT:    store float [[CONV4]], ptr [[L2_ASCAST]], align 4
 203 // CHECK-C-NEXT:    [[TMP4:%.*]] = load float, ptr [[X_ADDR_ASCAST]], align 4
 204 // CHECK-C-NEXT:    [[CONV5:%.*]] = fpext float [[TMP4]] to double
 205 // CHECK-C-NEXT:    [[RETVAL_ASCAST_I20:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I18]] to ptr
 206 // CHECK-C-NEXT:    [[__X_ADDR_ASCAST_I21:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I19]] to ptr
 207 // CHECK-C-NEXT:    store double [[CONV5]], ptr [[__X_ADDR_ASCAST_I21]], align 8
 208 // CHECK-C-NEXT:    [[TMP5:%.*]] = load double, ptr [[__X_ADDR_ASCAST_I21]], align 8
 209 // CHECK-C-NEXT:    [[TMP6:%.*]] = call double @llvm.fabs.f64(double [[TMP5]])
 210 // CHECK-C-NEXT:    [[CONV7:%.*]] = fptrunc double [[TMP6]] to float
 211 // CHECK-C-NEXT:    store float [[CONV7]], ptr [[L3_ASCAST]], align 4
 212 // CHECK-C-NEXT:    [[TMP7:%.*]] = load float, ptr [[X_ADDR_ASCAST]], align 4
 213 // CHECK-C-NEXT:    [[CONV8:%.*]] = fpext float [[TMP7]] to double
 214 // CHECK-C-NEXT:    [[TMP8:%.*]] = load float, ptr [[Y_ADDR_ASCAST]], align 4
 215 // CHECK-C-NEXT:    [[CONV9:%.*]] = fpext float [[TMP8]] to double
 216 // CHECK-C-NEXT:    [[TMP9:%.*]] = load float, ptr [[Z_ADDR_ASCAST]], align 4
 217 // CHECK-C-NEXT:    [[CONV10:%.*]] = fpext float [[TMP9]] to double
 218 // CHECK-C-NEXT:    [[RETVAL_ASCAST_I24:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I22]] to ptr
 219 // CHECK-C-NEXT:    [[__X_ADDR_ASCAST_I25:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I23]] to ptr
 220 // CHECK-C-NEXT:    [[__Y_ADDR_ASCAST_I:%.*]] = addrspacecast ptr addrspace(5) [[__Y_ADDR_I]] to ptr
 221 // CHECK-C-NEXT:    [[__Z_ADDR_ASCAST_I:%.*]] = addrspacecast ptr addrspace(5) [[__Z_ADDR_I]] to ptr
 222 // CHECK-C-NEXT:    store double [[CONV8]], ptr [[__X_ADDR_ASCAST_I25]], align 8
 223 // CHECK-C-NEXT:    store double [[CONV9]], ptr [[__Y_ADDR_ASCAST_I]], align 8
 224 // CHECK-C-NEXT:    store double [[CONV10]], ptr [[__Z_ADDR_ASCAST_I]], align 8
 225 // CHECK-C-NEXT:    [[TMP10:%.*]] = load double, ptr [[__X_ADDR_ASCAST_I25]], align 8
 226 // CHECK-C-NEXT:    [[TMP11:%.*]] = load double, ptr [[__Y_ADDR_ASCAST_I]], align 8
 227 // CHECK-C-NEXT:    [[TMP12:%.*]] = load double, ptr [[__Z_ADDR_ASCAST_I]], align 8
 228 // CHECK-C-NEXT:    [[TMP13:%.*]] = call double @llvm.fma.f64(double [[TMP10]], double [[TMP11]], double [[TMP12]])
 229 // CHECK-C-NEXT:    [[CONV12:%.*]] = fptrunc double [[TMP13]] to float
 230 // CHECK-C-NEXT:    store float [[CONV12]], ptr [[L4_ASCAST]], align 4
 231 // CHECK-C-NEXT:    ret void
 232 //
 233 // CHECK-CPP-LABEL: @_Z13test_math_f32fff(
 234 // CHECK-CPP-NEXT:  entry:
 235 // CHECK-CPP-NEXT:    [[RETVAL_I32:%.*]] = alloca float, align 4, addrspace(5)
 236 // CHECK-CPP-NEXT:    [[__X_ADDR_I33:%.*]] = alloca float, align 4, addrspace(5)
 237 // CHECK-CPP-NEXT:    [[__Y_ADDR_I34:%.*]] = alloca float, align 4, addrspace(5)
 238 // CHECK-CPP-NEXT:    [[__Z_ADDR_I35:%.*]] = alloca float, align 4, addrspace(5)
 239 // CHECK-CPP-NEXT:    [[RETVAL_I28:%.*]] = alloca float, align 4, addrspace(5)
 240 // CHECK-CPP-NEXT:    [[__X_ADDR_I29:%.*]] = alloca float, align 4, addrspace(5)
 241 // CHECK-CPP-NEXT:    [[RETVAL_I24:%.*]] = alloca float, align 4, addrspace(5)
 242 // CHECK-CPP-NEXT:    [[__X_ADDR_I25:%.*]] = alloca float, align 4, addrspace(5)
 243 // CHECK-CPP-NEXT:    [[RETVAL_I19:%.*]] = alloca float, align 4, addrspace(5)
 244 // CHECK-CPP-NEXT:    [[__X_ADDR_I20:%.*]] = alloca float, align 4, addrspace(5)
 245 // CHECK-CPP-NEXT:    [[RETVAL_I14:%.*]] = alloca float, align 4, addrspace(5)
 246 // CHECK-CPP-NEXT:    [[__X_ADDR_I15:%.*]] = alloca float, align 4, addrspace(5)
 247 // CHECK-CPP-NEXT:    [[__Y_ADDR_I:%.*]] = alloca float, align 4, addrspace(5)
 248 // CHECK-CPP-NEXT:    [[__Z_ADDR_I:%.*]] = alloca float, align 4, addrspace(5)
 249 // CHECK-CPP-NEXT:    [[RETVAL_I9:%.*]] = alloca float, align 4, addrspace(5)
 250 // CHECK-CPP-NEXT:    [[__X_ADDR_I10:%.*]] = alloca float, align 4, addrspace(5)
 251 // CHECK-CPP-NEXT:    [[RETVAL_I4:%.*]] = alloca float, align 4, addrspace(5)
 252 // CHECK-CPP-NEXT:    [[__X_ADDR_I5:%.*]] = alloca float, align 4, addrspace(5)
 253 // CHECK-CPP-NEXT:    [[RETVAL_I:%.*]] = alloca float, align 4, addrspace(5)
 254 // CHECK-CPP-NEXT:    [[__X_ADDR_I:%.*]] = alloca float, align 4, addrspace(5)
 255 // CHECK-CPP-NEXT:    [[X_ADDR:%.*]] = alloca float, align 4, addrspace(5)
 256 // CHECK-CPP-NEXT:    [[Y_ADDR:%.*]] = alloca float, align 4, addrspace(5)
 257 // CHECK-CPP-NEXT:    [[Z_ADDR:%.*]] = alloca float, align 4, addrspace(5)
 258 // CHECK-CPP-NEXT:    [[L1:%.*]] = alloca float, align 4, addrspace(5)
 259 // CHECK-CPP-NEXT:    [[L2:%.*]] = alloca float, align 4, addrspace(5)
 260 // CHECK-CPP-NEXT:    [[L3:%.*]] = alloca float, align 4, addrspace(5)
 261 // CHECK-CPP-NEXT:    [[L4:%.*]] = alloca float, align 4, addrspace(5)
 262 // CHECK-CPP-NEXT:    [[X_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[X_ADDR]] to ptr
 263 // CHECK-CPP-NEXT:    [[Y_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[Y_ADDR]] to ptr
 264 // CHECK-CPP-NEXT:    [[Z_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[Z_ADDR]] to ptr
 265 // CHECK-CPP-NEXT:    [[L1_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L1]] to ptr
 266 // CHECK-CPP-NEXT:    [[L2_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L2]] to ptr
 267 // CHECK-CPP-NEXT:    [[L3_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L3]] to ptr
 268 // CHECK-CPP-NEXT:    [[L4_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L4]] to ptr
 269 // CHECK-CPP-NEXT:    store float [[X:%.*]], ptr [[X_ADDR_ASCAST]], align 4
 270 // CHECK-CPP-NEXT:    store float [[Y:%.*]], ptr [[Y_ADDR_ASCAST]], align 4
 271 // CHECK-CPP-NEXT:    store float [[Z:%.*]], ptr [[Z_ADDR_ASCAST]], align 4
 272 // CHECK-CPP-NEXT:    [[TMP0:%.*]] = load float, ptr [[X_ADDR_ASCAST]], align 4
 273 // CHECK-CPP-NEXT:    [[RETVAL_ASCAST_I:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I]] to ptr
 274 // CHECK-CPP-NEXT:    [[__X_ADDR_ASCAST_I:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I]] to ptr
 275 // CHECK-CPP-NEXT:    store float [[TMP0]], ptr [[__X_ADDR_ASCAST_I]], align 4
 276 // CHECK-CPP-NEXT:    [[TMP1:%.*]] = load float, ptr [[__X_ADDR_ASCAST_I]], align 4
 277 // CHECK-CPP-NEXT:    [[RETVAL_ASCAST_I21:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I19]] to ptr
 278 // CHECK-CPP-NEXT:    [[__X_ADDR_ASCAST_I22:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I20]] to ptr
 279 // CHECK-CPP-NEXT:    store float [[TMP1]], ptr [[__X_ADDR_ASCAST_I22]], align 4
 280 // CHECK-CPP-NEXT:    [[TMP2:%.*]] = load float, ptr [[__X_ADDR_ASCAST_I22]], align 4
 281 // CHECK-CPP-NEXT:    [[CALL_I23:%.*]] = call noundef float @__ocml_sin_f32(float noundef [[TMP2]]) #[[ATTR3]]
 282 // CHECK-CPP-NEXT:    store float [[CALL_I23]], ptr [[L1_ASCAST]], align 4
 283 // CHECK-CPP-NEXT:    [[TMP3:%.*]] = load float, ptr [[X_ADDR_ASCAST]], align 4
 284 // CHECK-CPP-NEXT:    [[RETVAL_ASCAST_I6:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I4]] to ptr
 285 // CHECK-CPP-NEXT:    [[__X_ADDR_ASCAST_I7:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I5]] to ptr
 286 // CHECK-CPP-NEXT:    store float [[TMP3]], ptr [[__X_ADDR_ASCAST_I7]], align 4
 287 // CHECK-CPP-NEXT:    [[TMP4:%.*]] = load float, ptr [[__X_ADDR_ASCAST_I7]], align 4
 288 // CHECK-CPP-NEXT:    [[RETVAL_ASCAST_I26:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I24]] to ptr
 289 // CHECK-CPP-NEXT:    [[__X_ADDR_ASCAST_I27:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I25]] to ptr
 290 // CHECK-CPP-NEXT:    store float [[TMP4]], ptr [[__X_ADDR_ASCAST_I27]], align 4
 291 // CHECK-CPP-NEXT:    [[TMP5:%.*]] = load float, ptr [[__X_ADDR_ASCAST_I27]], align 4
 292 // CHECK-CPP-NEXT:    [[CALL_I:%.*]] = call noundef float @__ocml_cos_f32(float noundef [[TMP5]]) #[[ATTR3]]
 293 // CHECK-CPP-NEXT:    store float [[CALL_I]], ptr [[L2_ASCAST]], align 4
 294 // CHECK-CPP-NEXT:    [[TMP6:%.*]] = load float, ptr [[X_ADDR_ASCAST]], align 4
 295 // CHECK-CPP-NEXT:    [[RETVAL_ASCAST_I11:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I9]] to ptr
 296 // CHECK-CPP-NEXT:    [[__X_ADDR_ASCAST_I12:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I10]] to ptr
 297 // CHECK-CPP-NEXT:    store float [[TMP6]], ptr [[__X_ADDR_ASCAST_I12]], align 4
 298 // CHECK-CPP-NEXT:    [[TMP7:%.*]] = load float, ptr [[__X_ADDR_ASCAST_I12]], align 4
 299 // CHECK-CPP-NEXT:    [[RETVAL_ASCAST_I30:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I28]] to ptr
 300 // CHECK-CPP-NEXT:    [[__X_ADDR_ASCAST_I31:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I29]] to ptr
 301 // CHECK-CPP-NEXT:    store float [[TMP7]], ptr [[__X_ADDR_ASCAST_I31]], align 4
 302 // CHECK-CPP-NEXT:    [[TMP8:%.*]] = load float, ptr [[__X_ADDR_ASCAST_I31]], align 4
 303 // CHECK-CPP-NEXT:    [[TMP9:%.*]] = call noundef float @llvm.fabs.f32(float [[TMP8]])
 304 // CHECK-CPP-NEXT:    store float [[TMP9]], ptr [[L3_ASCAST]], align 4
 305 // CHECK-CPP-NEXT:    [[TMP10:%.*]] = load float, ptr [[X_ADDR_ASCAST]], align 4
 306 // CHECK-CPP-NEXT:    [[TMP11:%.*]] = load float, ptr [[Y_ADDR_ASCAST]], align 4
 307 // CHECK-CPP-NEXT:    [[TMP12:%.*]] = load float, ptr [[Z_ADDR_ASCAST]], align 4
 308 // CHECK-CPP-NEXT:    [[RETVAL_ASCAST_I16:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I14]] to ptr
 309 // CHECK-CPP-NEXT:    [[__X_ADDR_ASCAST_I17:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I15]] to ptr
 310 // CHECK-CPP-NEXT:    [[__Y_ADDR_ASCAST_I:%.*]] = addrspacecast ptr addrspace(5) [[__Y_ADDR_I]] to ptr
 311 // CHECK-CPP-NEXT:    [[__Z_ADDR_ASCAST_I:%.*]] = addrspacecast ptr addrspace(5) [[__Z_ADDR_I]] to ptr
 312 // CHECK-CPP-NEXT:    store float [[TMP10]], ptr [[__X_ADDR_ASCAST_I17]], align 4
 313 // CHECK-CPP-NEXT:    store float [[TMP11]], ptr [[__Y_ADDR_ASCAST_I]], align 4
 314 // CHECK-CPP-NEXT:    store float [[TMP12]], ptr [[__Z_ADDR_ASCAST_I]], align 4
 315 // CHECK-CPP-NEXT:    [[TMP13:%.*]] = load float, ptr [[__X_ADDR_ASCAST_I17]], align 4
 316 // CHECK-CPP-NEXT:    [[TMP14:%.*]] = load float, ptr [[__Y_ADDR_ASCAST_I]], align 4
 317 // CHECK-CPP-NEXT:    [[TMP15:%.*]] = load float, ptr [[__Z_ADDR_ASCAST_I]], align 4
 318 // CHECK-CPP-NEXT:    [[RETVAL_ASCAST_I36:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I32]] to ptr
 319 // CHECK-CPP-NEXT:    [[__X_ADDR_ASCAST_I37:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I33]] to ptr
 320 // CHECK-CPP-NEXT:    [[__Y_ADDR_ASCAST_I38:%.*]] = addrspacecast ptr addrspace(5) [[__Y_ADDR_I34]] to ptr
 321 // CHECK-CPP-NEXT:    [[__Z_ADDR_ASCAST_I39:%.*]] = addrspacecast ptr addrspace(5) [[__Z_ADDR_I35]] to ptr
 322 // CHECK-CPP-NEXT:    store float [[TMP13]], ptr [[__X_ADDR_ASCAST_I37]], align 4
 323 // CHECK-CPP-NEXT:    store float [[TMP14]], ptr [[__Y_ADDR_ASCAST_I38]], align 4
 324 // CHECK-CPP-NEXT:    store float [[TMP15]], ptr [[__Z_ADDR_ASCAST_I39]], align 4
 325 // CHECK-CPP-NEXT:    [[TMP16:%.*]] = load float, ptr [[__X_ADDR_ASCAST_I37]], align 4
 326 // CHECK-CPP-NEXT:    [[TMP17:%.*]] = load float, ptr [[__Y_ADDR_ASCAST_I38]], align 4
 327 // CHECK-CPP-NEXT:    [[TMP18:%.*]] = load float, ptr [[__Z_ADDR_ASCAST_I39]], align 4
 328 // CHECK-CPP-NEXT:    [[TMP19:%.*]] = call noundef float @llvm.fma.f32(float [[TMP16]], float [[TMP17]], float [[TMP18]])
 329 // CHECK-CPP-NEXT:    store float [[TMP19]], ptr [[L4_ASCAST]], align 4
 330 // CHECK-CPP-NEXT:    ret void
 331 //
 332 void test_math_f32(float x, float y, float z) {
 333   float l1 = sin(x);
 334   float l2 = cos(x);
 335   float l3 = fabs(x);
 336   float l4 = fma(x, y, z);
 337 }
 338
 339 // CHECK-C-LABEL: @test_math_f32_suffix(
 340 // CHECK-C-NEXT:  entry:
 341 // CHECK-C-NEXT:    [[RETVAL_I13:%.*]] = alloca float, align 4, addrspace(5)
 342 // CHECK-C-NEXT:    [[__X_ADDR_I14:%.*]] = alloca float, align 4, addrspace(5)
 343 // CHECK-C-NEXT:    [[__Y_ADDR_I:%.*]] = alloca float, align 4, addrspace(5)
 344 // CHECK-C-NEXT:    [[__Z_ADDR_I:%.*]] = alloca float, align 4, addrspace(5)
 345 // CHECK-C-NEXT:    [[RETVAL_I9:%.*]] = alloca float, align 4, addrspace(5)
 346 // CHECK-C-NEXT:    [[__X_ADDR_I10:%.*]] = alloca float, align 4, addrspace(5)
 347 // CHECK-C-NEXT:    [[RETVAL_I4:%.*]] = alloca float, align 4, addrspace(5)
 348 // CHECK-C-NEXT:    [[__X_ADDR_I5:%.*]] = alloca float, align 4, addrspace(5)
 349 // CHECK-C-NEXT:    [[RETVAL_I:%.*]] = alloca float, align 4, addrspace(5)
 350 // CHECK-C-NEXT:    [[__X_ADDR_I:%.*]] = alloca float, align 4, addrspace(5)
 351 // CHECK-C-NEXT:    [[X_ADDR:%.*]] = alloca float, align 4, addrspace(5)
 352 // CHECK-C-NEXT:    [[Y_ADDR:%.*]] = alloca float, align 4, addrspace(5)
 353 // CHECK-C-NEXT:    [[Z_ADDR:%.*]] = alloca float, align 4, addrspace(5)
 354 // CHECK-C-NEXT:    [[L1:%.*]] = alloca float, align 4, addrspace(5)
 355 // CHECK-C-NEXT:    [[L2:%.*]] = alloca float, align 4, addrspace(5)
 356 // CHECK-C-NEXT:    [[L3:%.*]] = alloca float, align 4, addrspace(5)
 357 // CHECK-C-NEXT:    [[L4:%.*]] = alloca float, align 4, addrspace(5)
 358 // CHECK-C-NEXT:    [[X_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[X_ADDR]] to ptr
 359 // CHECK-C-NEXT:    [[Y_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[Y_ADDR]] to ptr
 360 // CHECK-C-NEXT:    [[Z_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[Z_ADDR]] to ptr
 361 // CHECK-C-NEXT:    [[L1_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L1]] to ptr
 362 // CHECK-C-NEXT:    [[L2_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L2]] to ptr
 363 // CHECK-C-NEXT:    [[L3_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L3]] to ptr
 364 // CHECK-C-NEXT:    [[L4_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L4]] to ptr
 365 // CHECK-C-NEXT:    store float [[X:%.*]], ptr [[X_ADDR_ASCAST]], align 4
 366 // CHECK-C-NEXT:    store float [[Y:%.*]], ptr [[Y_ADDR_ASCAST]], align 4
 367 // CHECK-C-NEXT:    store float [[Z:%.*]], ptr [[Z_ADDR_ASCAST]], align 4
 368 // CHECK-C-NEXT:    [[TMP0:%.*]] = load float, ptr [[X_ADDR_ASCAST]], align 4
 369 // CHECK-C-NEXT:    [[RETVAL_ASCAST_I:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I]] to ptr
 370 // CHECK-C-NEXT:    [[__X_ADDR_ASCAST_I:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I]] to ptr
 371 // CHECK-C-NEXT:    store float [[TMP0]], ptr [[__X_ADDR_ASCAST_I]], align 4
 372 // CHECK-C-NEXT:    [[TMP1:%.*]] = load float, ptr [[__X_ADDR_ASCAST_I]], align 4
 373 // CHECK-C-NEXT:    [[CALL_I:%.*]] = call float @__ocml_sin_f32(float noundef [[TMP1]]) #[[ATTR3]]
 374 // CHECK-C-NEXT:    store float [[CALL_I]], ptr [[L1_ASCAST]], align 4
 375 // CHECK-C-NEXT:    [[TMP2:%.*]] = load float, ptr [[X_ADDR_ASCAST]], align 4
 376 // CHECK-C-NEXT:    [[RETVAL_ASCAST_I6:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I4]] to ptr
 377 // CHECK-C-NEXT:    [[__X_ADDR_ASCAST_I7:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I5]] to ptr
 378 // CHECK-C-NEXT:    store float [[TMP2]], ptr [[__X_ADDR_ASCAST_I7]], align 4
 379 // CHECK-C-NEXT:    [[TMP3:%.*]] = load float, ptr [[__X_ADDR_ASCAST_I7]], align 4
 380 // CHECK-C-NEXT:    [[CALL_I8:%.*]] = call float @__ocml_cos_f32(float noundef [[TMP3]]) #[[ATTR3]]
 381 // CHECK-C-NEXT:    store float [[CALL_I8]], ptr [[L2_ASCAST]], align 4
 382 // CHECK-C-NEXT:    [[TMP4:%.*]] = load float, ptr [[X_ADDR_ASCAST]], align 4
 383 // CHECK-C-NEXT:    [[RETVAL_ASCAST_I11:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I9]] to ptr
 384 // CHECK-C-NEXT:    [[__X_ADDR_ASCAST_I12:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I10]] to ptr
 385 // CHECK-C-NEXT:    store float [[TMP4]], ptr [[__X_ADDR_ASCAST_I12]], align 4
 386 // CHECK-C-NEXT:    [[TMP5:%.*]] = load float, ptr [[__X_ADDR_ASCAST_I12]], align 4
 387 // CHECK-C-NEXT:    [[TMP6:%.*]] = call float @llvm.fabs.f32(float [[TMP5]])
 388 // CHECK-C-NEXT:    store float [[TMP6]], ptr [[L3_ASCAST]], align 4
 389 // CHECK-C-NEXT:    [[TMP7:%.*]] = load float, ptr [[X_ADDR_ASCAST]], align 4
 390 // CHECK-C-NEXT:    [[TMP8:%.*]] = load float, ptr [[Y_ADDR_ASCAST]], align 4
 391 // CHECK-C-NEXT:    [[TMP9:%.*]] = load float, ptr [[Z_ADDR_ASCAST]], align 4
 392 // CHECK-C-NEXT:    [[RETVAL_ASCAST_I15:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I13]] to ptr
 393 // CHECK-C-NEXT:    [[__X_ADDR_ASCAST_I16:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I14]] to ptr
 394 // CHECK-C-NEXT:    [[__Y_ADDR_ASCAST_I:%.*]] = addrspacecast ptr addrspace(5) [[__Y_ADDR_I]] to ptr
 395 // CHECK-C-NEXT:    [[__Z_ADDR_ASCAST_I:%.*]] = addrspacecast ptr addrspace(5) [[__Z_ADDR_I]] to ptr
 396 // CHECK-C-NEXT:    store float [[TMP7]], ptr [[__X_ADDR_ASCAST_I16]], align 4
 397 // CHECK-C-NEXT:    store float [[TMP8]], ptr [[__Y_ADDR_ASCAST_I]], align 4
 398 // CHECK-C-NEXT:    store float [[TMP9]], ptr [[__Z_ADDR_ASCAST_I]], align 4
 399 // CHECK-C-NEXT:    [[TMP10:%.*]] = load float, ptr [[__X_ADDR_ASCAST_I16]], align 4
 400 // CHECK-C-NEXT:    [[TMP11:%.*]] = load float, ptr [[__Y_ADDR_ASCAST_I]], align 4
 401 // CHECK-C-NEXT:    [[TMP12:%.*]] = load float, ptr [[__Z_ADDR_ASCAST_I]], align 4
 402 // CHECK-C-NEXT:    [[TMP13:%.*]] = call float @llvm.fma.f32(float [[TMP10]], float [[TMP11]], float [[TMP12]])
 403 // CHECK-C-NEXT:    store float [[TMP13]], ptr [[L4_ASCAST]], align 4
 404 // CHECK-C-NEXT:    ret void
 405 //
 406 // CHECK-CPP-LABEL: @_Z20test_math_f32_suffixfff(
 407 // CHECK-CPP-NEXT:  entry:
 408 // CHECK-CPP-NEXT:    [[RETVAL_I13:%.*]] = alloca float, align 4, addrspace(5)
 409 // CHECK-CPP-NEXT:    [[__X_ADDR_I14:%.*]] = alloca float, align 4, addrspace(5)
 410 // CHECK-CPP-NEXT:    [[__Y_ADDR_I:%.*]] = alloca float, align 4, addrspace(5)
 411 // CHECK-CPP-NEXT:    [[__Z_ADDR_I:%.*]] = alloca float, align 4, addrspace(5)
 412 // CHECK-CPP-NEXT:    [[RETVAL_I9:%.*]] = alloca float, align 4, addrspace(5)
 413 // CHECK-CPP-NEXT:    [[__X_ADDR_I10:%.*]] = alloca float, align 4, addrspace(5)
 414 // CHECK-CPP-NEXT:    [[RETVAL_I4:%.*]] = alloca float, align 4, addrspace(5)
 415 // CHECK-CPP-NEXT:    [[__X_ADDR_I5:%.*]] = alloca float, align 4, addrspace(5)
 416 // CHECK-CPP-NEXT:    [[RETVAL_I:%.*]] = alloca float, align 4, addrspace(5)
 417 // CHECK-CPP-NEXT:    [[__X_ADDR_I:%.*]] = alloca float, align 4, addrspace(5)
 418 // CHECK-CPP-NEXT:    [[X_ADDR:%.*]] = alloca float, align 4, addrspace(5)
 419 // CHECK-CPP-NEXT:    [[Y_ADDR:%.*]] = alloca float, align 4, addrspace(5)
 420 // CHECK-CPP-NEXT:    [[Z_ADDR:%.*]] = alloca float, align 4, addrspace(5)
 421 // CHECK-CPP-NEXT:    [[L1:%.*]] = alloca float, align 4, addrspace(5)
 422 // CHECK-CPP-NEXT:    [[L2:%.*]] = alloca float, align 4, addrspace(5)
 423 // CHECK-CPP-NEXT:    [[L3:%.*]] = alloca float, align 4, addrspace(5)
 424 // CHECK-CPP-NEXT:    [[L4:%.*]] = alloca float, align 4, addrspace(5)
 425 // CHECK-CPP-NEXT:    [[X_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[X_ADDR]] to ptr
 426 // CHECK-CPP-NEXT:    [[Y_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[Y_ADDR]] to ptr
 427 // CHECK-CPP-NEXT:    [[Z_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[Z_ADDR]] to ptr
 428 // CHECK-CPP-NEXT:    [[L1_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L1]] to ptr
 429 // CHECK-CPP-NEXT:    [[L2_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L2]] to ptr
 430 // CHECK-CPP-NEXT:    [[L3_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L3]] to ptr
 431 // CHECK-CPP-NEXT:    [[L4_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[L4]] to ptr
 432 // CHECK-CPP-NEXT:    store float [[X:%.*]], ptr [[X_ADDR_ASCAST]], align 4
 433 // CHECK-CPP-NEXT:    store float [[Y:%.*]], ptr [[Y_ADDR_ASCAST]], align 4
 434 // CHECK-CPP-NEXT:    store float [[Z:%.*]], ptr [[Z_ADDR_ASCAST]], align 4
 435 // CHECK-CPP-NEXT:    [[TMP0:%.*]] = load float, ptr [[X_ADDR_ASCAST]], align 4
 436 // CHECK-CPP-NEXT:    [[RETVAL_ASCAST_I:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I]] to ptr
 437 // CHECK-CPP-NEXT:    [[__X_ADDR_ASCAST_I:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I]] to ptr
 438 // CHECK-CPP-NEXT:    store float [[TMP0]], ptr [[__X_ADDR_ASCAST_I]], align 4
 439 // CHECK-CPP-NEXT:    [[TMP1:%.*]] = load float, ptr [[__X_ADDR_ASCAST_I]], align 4
 440 // CHECK-CPP-NEXT:    [[CALL_I:%.*]] = call noundef float @__ocml_sin_f32(float noundef [[TMP1]]) #[[ATTR3]]
 441 // CHECK-CPP-NEXT:    store float [[CALL_I]], ptr [[L1_ASCAST]], align 4
 442 // CHECK-CPP-NEXT:    [[TMP2:%.*]] = load float, ptr [[X_ADDR_ASCAST]], align 4
 443 // CHECK-CPP-NEXT:    [[RETVAL_ASCAST_I6:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I4]] to ptr
 444 // CHECK-CPP-NEXT:    [[__X_ADDR_ASCAST_I7:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I5]] to ptr
 445 // CHECK-CPP-NEXT:    store float [[TMP2]], ptr [[__X_ADDR_ASCAST_I7]], align 4
 446 // CHECK-CPP-NEXT:    [[TMP3:%.*]] = load float, ptr [[__X_ADDR_ASCAST_I7]], align 4
 447 // CHECK-CPP-NEXT:    [[CALL_I8:%.*]] = call noundef float @__ocml_cos_f32(float noundef [[TMP3]]) #[[ATTR3]]
 448 // CHECK-CPP-NEXT:    store float [[CALL_I8]], ptr [[L2_ASCAST]], align 4
 449 // CHECK-CPP-NEXT:    [[TMP4:%.*]] = load float, ptr [[X_ADDR_ASCAST]], align 4
 450 // CHECK-CPP-NEXT:    [[RETVAL_ASCAST_I11:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I9]] to ptr
 451 // CHECK-CPP-NEXT:    [[__X_ADDR_ASCAST_I12:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I10]] to ptr
 452 // CHECK-CPP-NEXT:    store float [[TMP4]], ptr [[__X_ADDR_ASCAST_I12]], align 4
 453 // CHECK-CPP-NEXT:    [[TMP5:%.*]] = load float, ptr [[__X_ADDR_ASCAST_I12]], align 4
 454 // CHECK-CPP-NEXT:    [[TMP6:%.*]] = call noundef float @llvm.fabs.f32(float [[TMP5]])
 455 // CHECK-CPP-NEXT:    store float [[TMP6]], ptr [[L3_ASCAST]], align 4
 456 // CHECK-CPP-NEXT:    [[TMP7:%.*]] = load float, ptr [[X_ADDR_ASCAST]], align 4
 457 // CHECK-CPP-NEXT:    [[TMP8:%.*]] = load float, ptr [[Y_ADDR_ASCAST]], align 4
 458 // CHECK-CPP-NEXT:    [[TMP9:%.*]] = load float, ptr [[Z_ADDR_ASCAST]], align 4
 459 // CHECK-CPP-NEXT:    [[RETVAL_ASCAST_I15:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL_I13]] to ptr
 460 // CHECK-CPP-NEXT:    [[__X_ADDR_ASCAST_I16:%.*]] = addrspacecast ptr addrspace(5) [[__X_ADDR_I14]] to ptr
 461 // CHECK-CPP-NEXT:    [[__Y_ADDR_ASCAST_I:%.*]] = addrspacecast ptr addrspace(5) [[__Y_ADDR_I]] to ptr
 462 // CHECK-CPP-NEXT:    [[__Z_ADDR_ASCAST_I:%.*]] = addrspacecast ptr addrspace(5) [[__Z_ADDR_I]] to ptr
 463 // CHECK-CPP-NEXT:    store float [[TMP7]], ptr [[__X_ADDR_ASCAST_I16]], align 4
 464 // CHECK-CPP-NEXT:    store float [[TMP8]], ptr [[__Y_ADDR_ASCAST_I]], align 4
 465 // CHECK-CPP-NEXT:    store float [[TMP9]], ptr [[__Z_ADDR_ASCAST_I]], align 4
 466 // CHECK-CPP-NEXT:    [[TMP10:%.*]] = load float, ptr [[__X_ADDR_ASCAST_I16]], align 4
 467 // CHECK-CPP-NEXT:    [[TMP11:%.*]] = load float, ptr [[__Y_ADDR_ASCAST_I]], align 4
 468 // CHECK-CPP-NEXT:    [[TMP12:%.*]] = load float, ptr [[__Z_ADDR_ASCAST_I]], align 4
 469 // CHECK-CPP-NEXT:    [[TMP13:%.*]] = call noundef float @llvm.fma.f32(float [[TMP10]], float [[TMP11]], float [[TMP12]])
 470 // CHECK-CPP-NEXT:    store float [[TMP13]], ptr [[L4_ASCAST]], align 4
 471 // CHECK-CPP-NEXT:    ret void
 472 //
 473 void test_math_f32_suffix(float x, float y, float z) {
 474   float l1 = sinf(x);
 475   float l2 = cosf(x);
 476   float l3 = fabsf(x);
 477   float l4 = fmaf(x, y, z);
 478 }
 479
 480 #pragma omp end declare target
 481 //// NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
 482 // CHECK: {{.*}}