clang/test/Headers/opencl-c-header.cl

   1 // RUN: %clang_cc1 -O0 -triple spir-unknown-unknown -internal-isystem ../../lib/Headers -include opencl-c.h -emit-llvm -o - %s -verify | FileCheck %s
   2 // RUN: %clang_cc1 -O0 -triple spir-unknown-unknown -internal-isystem ../../lib/Headers -include opencl-c.h -emit-llvm -o - %s -verify -cl-std=CL1.1 | FileCheck %s
   3 // RUN: %clang_cc1 -O0 -triple spir-unknown-unknown -internal-isystem ../../lib/Headers -include opencl-c.h -emit-llvm -o - %s -verify -cl-std=CL1.2 | FileCheck %s
   4 // RUN: %clang_cc1 -O0 -triple spir-unknown-unknown -internal-isystem ../../lib/Headers -include opencl-c.h -emit-llvm -o - %s -verify -cl-std=clc++1.0 | FileCheck %s --check-prefix=CHECK20
   5 // RUN: %clang_cc1 -O0 -triple spir-unknown-unknown -internal-isystem ../../lib/Headers -include opencl-c.h -emit-llvm -o - %s -verify -cl-std=CL3.0 | FileCheck %s
   6 // RUN: %clang_cc1 -O0 -triple spir-unknown-unknown -internal-isystem ../../lib/Headers -include opencl-c.h -emit-llvm -o - %s -verify -cl-std=clc++2021 | FileCheck %s
   7
   8 // RUN: %clang_cc1 -O0 -triple spirv32-unknown-unknown -internal-isystem ../../lib/Headers -include opencl-c.h -emit-llvm -o - %s -verify | FileCheck %s
   9
  10
  11 // Test including the default header as a module.
  12 // The module should be compiled only once and loaded from cache afterwards.
  13 // Change the directory mode to read only to make sure no new modules are created.
  14 // Check time report to make sure module is used.
  15 // Check that some builtins occur in the generated IR when called.
  16
  17 // ===
  18 // Clear current directory.
  19 // RUN: rm -rf %t
  20 // RUN: mkdir -p %t
  21
  22 // ===
  23 // Compile for OpenCL 1.0 for the first time. A module should be generated.
  24 // RUN: %clang_cc1 -triple spir-unknown-unknown -emit-llvm -o - -finclude-default-header -fmodules -fimplicit-module-maps -fmodules-cache-path=%t -fdisable-module-hash -ftime-report %s 2>&1 | FileCheck --check-prefix=CHECK --check-prefix=CHECK-MOD %s
  25 // RUN: chmod u-w %t/opencl_c.pcm
  26
  27 // ===
  28 // Compile for OpenCL 1.0 for the second time. The module should not be re-created.
  29 // RUN: %clang_cc1 -triple spir-unknown-unknown -emit-llvm -o - -finclude-default-header -fmodules -fimplicit-module-maps -fmodules-cache-path=%t -fdisable-module-hash -ftime-report %s 2>&1 | FileCheck --check-prefix=CHECK --check-prefix=CHECK-MOD %s
  30 // RUN: chmod u+w %t/opencl_c.pcm
  31 // RUN: mv %t/opencl_c.pcm %t/1_0.pcm
  32
  33 // ===
  34 // Compile for OpenCL 2.0 for the first time. The module should change.
  35 // RUN: %clang_cc1 -triple spir-unknown-unknown -O0 -emit-llvm -o - -cl-std=CL2.0 -finclude-default-header -fmodules -fimplicit-module-maps -fmodules-cache-path=%t -fdisable-module-hash -ftime-report %s 2>&1 | FileCheck --check-prefix=CHECK20 --check-prefix=CHECK-MOD %s
  36 // RUN: not diff %t/1_0.pcm %t/opencl_c.pcm
  37 // RUN: chmod u-w %t/opencl_c.pcm
  38
  39 // ===
  40 // Compile for OpenCL 2.0 for the second time. The module should not change.
  41 // RUN: %clang_cc1 -triple spir-unknown-unknown -O0 -emit-llvm -o - -cl-std=CL2.0 -finclude-default-header -fmodules -fimplicit-module-maps -fmodules-cache-path=%t -fdisable-module-hash -ftime-report %s 2>&1 | FileCheck --check-prefix=CHECK20 --check-prefix=CHECK-MOD %s
  42
  43 // Check cached module works for different OpenCL versions.
  44 // RUN: rm -rf %t
  45 // RUN: mkdir -p %t
  46 // RUN: %clang_cc1 -triple spir64-unknown-unknown -emit-llvm -o - -cl-std=CL1.2 -finclude-default-header -fmodules -fimplicit-module-maps -fmodules-cache-path=%t -ftime-report %s 2>&1 | FileCheck --check-prefix=CHECK --check-prefix=CHECK-MOD %s
  47 // RUN: %clang_cc1 -triple amdgcn--amdhsa -O0 -emit-llvm -o - -cl-std=CL2.0 -finclude-default-header -fmodules -fimplicit-module-maps -fmodules-cache-path=%t -ftime-report %s 2>&1 | FileCheck --check-prefix=CHECK20 --check-prefix=CHECK-MOD %s
  48 // RUN: chmod u-w %t
  49 // RUN: %clang_cc1 -triple spir64-unknown-unknown -emit-llvm -o - -cl-std=CL1.2 -finclude-default-header -fmodules -fimplicit-module-maps -fmodules-cache-path=%t -ftime-report %s 2>&1 | FileCheck --check-prefix=CHECK --check-prefix=CHECK-MOD %s
  50 // RUN: %clang_cc1 -triple amdgcn--amdhsa -O0 -emit-llvm -o - -cl-std=CL2.0 -finclude-default-header -fmodules -fimplicit-module-maps -fmodules-cache-path=%t -ftime-report %s 2>&1 | FileCheck --check-prefix=CHECK20 --check-prefix=CHECK-MOD %s
  51 // RUN: chmod u+w %t
  52
  53 // Verify that called builtins occur in the generated IR.
  54
  55 // CHECK-NOT: intel_sub_group_avc_mce_get_default_inter_base_multi_reference_penalty
  56 // CHECK-NOT: ndrange_t
  57 // CHECK20: ndrange_t
  58 // CHECK: _Z16convert_char_rtec
  59 // CHECK-NOT: _Z3ctzc
  60 // CHECK20: _Z3ctzc
  61 // CHECK20: _Z16convert_char_rtec
  62 char f(char x) {
  63 // Check functionality from OpenCL 2.0 onwards
  64 #if (__OPENCL_CPP_VERSION__ == 100) || (__OPENCL_C_VERSION__ == CL_VERSION_2_0)
  65   ndrange_t t;
  66   x = ctz(x);
  67 #endif //__OPENCL_C_VERSION__
  68   return convert_char_rte(x);
  69 }
  70
  71 // Verify that a builtin using a write_only image3d_t type is available
  72 // from OpenCL 2.0 onwards.
  73
  74 // CHECK20: _Z12write_imagef14ocl_image3d_wo
  75 #if defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
  76 void test_image3dwo(write_only image3d_t img) {
  77   write_imagef(img, (0), (0.0f));
  78 }
  79 #endif //__OPENCL_C_VERSION__
  80
  81 #if defined(__OPENCL_CPP_VERSION__)
  82 // Test old atomic overloaded with generic addr space.
  83 void test_atomics(__generic volatile unsigned int* a) {
  84   atomic_add(a, 1);
  85 }
  86 #endif
  87
  88 // Verify that ATOMIC_VAR_INIT is defined.
  89 #if (__OPENCL_CPP_VERSION__ == 100) || (__OPENCL_C_VERSION__ == CL_VERSION_2_0)
  90 global atomic_int z = ATOMIC_VAR_INIT(99);
  91 #endif //__OPENCL_C_VERSION__
  92 // CHECK-MOD: Reading modules
  93
  94 // Check that extension macros are defined correctly.
  95
  96 // For SPIR all extensions are supported.
  97 #if defined(__SPIR__) || defined(__SPIRV__)
  98
  99 // Verify that cl_intel_planar_yuv extension is defined from OpenCL 1.2 onwards.
 100 #if defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_1_2)
 101 // expected-no-diagnostics
 102 #else //__OPENCL_C_VERSION__
 103 // expected-warning@+2{{OpenCL extension 'cl_intel_planar_yuv' unknown or does not require pragma - ignoring}}
 104 #endif //__OPENCL_C_VERSION__
 105 #pragma OPENCL EXTENSION cl_intel_planar_yuv : enable
 106
 107 #if (defined(__OPENCL_CPP_VERSION__) || __OPENCL_C_VERSION__ >= 200)
 108
 109 #if cl_khr_subgroup_extended_types != 1
 110 #error "Incorrectly defined cl_khr_subgroup_extended_types"
 111 #endif
 112 #if cl_khr_subgroup_non_uniform_vote != 1
 113 #error "Incorrectly defined cl_khr_subgroup_non_uniform_vote"
 114 #endif
 115 #if cl_khr_subgroup_ballot != 1
 116 #error "Incorrectly defined cl_khr_subgroup_ballot"
 117 #endif
 118 #if cl_khr_subgroup_non_uniform_arithmetic != 1
 119 #error "Incorrectly defined cl_khr_subgroup_non_uniform_arithmetic"
 120 #endif
 121 #if cl_khr_subgroup_shuffle != 1
 122 #error "Incorrectly defined cl_khr_subgroup_shuffle"
 123 #endif
 124 #if cl_khr_subgroup_shuffle_relative != 1
 125 #error "Incorrectly defined cl_khr_subgroup_shuffle_relative"
 126 #endif
 127 #if cl_khr_subgroup_clustered_reduce != 1
 128 #error "Incorrectly defined cl_khr_subgroup_clustered_reduce"
 129 #endif
 130 #if cl_khr_subgroup_rotate != 1
 131 #error "Incorrectly defined cl_khr_subgroup_rotate"
 132 #endif
 133 #if cl_khr_extended_bit_ops != 1
 134 #error "Incorrectly defined cl_khr_extended_bit_ops"
 135 #endif
 136 #if cl_khr_integer_dot_product != 1
 137 #error "Incorrectly defined cl_khr_integer_dot_product"
 138 #endif
 139 #if __opencl_c_integer_dot_product_input_4x8bit != 1
 140 #error "Incorrectly defined __opencl_c_integer_dot_product_input_4x8bit"
 141 #endif
 142 #if __opencl_c_integer_dot_product_input_4x8bit_packed != 1
 143 #error "Incorrectly defined __opencl_c_integer_dot_product_input_4x8bit_packed"
 144 #endif
 145 #if cl_ext_float_atomics != 1
 146 #error "Incorrectly defined cl_ext_float_atomics"
 147 #endif
 148 #if __opencl_c_ext_fp16_global_atomic_load_store != 1
 149 #error "Incorrectly defined __opencl_c_ext_fp16_global_atomic_load_store"
 150 #endif
 151 #if __opencl_c_ext_fp16_local_atomic_load_store != 1
 152 #error "Incorrectly defined __opencl_c_ext_fp16_local_atomic_load_store"
 153 #endif
 154 #if __opencl_c_ext_fp16_global_atomic_add != 1
 155 #error "Incorrectly defined __opencl_c_ext_fp16_global_atomic_add"
 156 #endif
 157 #if __opencl_c_ext_fp32_global_atomic_add != 1
 158 #error "Incorrectly defined __opencl_c_ext_fp32_global_atomic_add"
 159 #endif
 160 #if __opencl_c_ext_fp64_global_atomic_add != 1
 161 #error "Incorrectly defined __opencl_c_ext_fp64_global_atomic_add"
 162 #endif
 163 #if __opencl_c_ext_fp16_local_atomic_add != 1
 164 #error "Incorrectly defined __opencl_c_ext_fp16_local_atomic_add"
 165 #endif
 166 #if __opencl_c_ext_fp32_local_atomic_add != 1
 167 #error "Incorrectly defined __opencl_c_ext_fp32_local_atomic_add"
 168 #endif
 169 #if __opencl_c_ext_fp64_local_atomic_add != 1
 170 #error "Incorrectly defined __opencl_c_ext_fp64_local_atomic_add"
 171 #endif
 172 #if __opencl_c_ext_fp16_global_atomic_min_max != 1
 173 #error "Incorrectly defined __opencl_c_ext_fp16_global_atomic_min_max"
 174 #endif
 175 #if __opencl_c_ext_fp32_global_atomic_min_max != 1
 176 #error "Incorrectly defined __opencl_c_ext_fp32_global_atomic_min_max"
 177 #endif
 178 #if __opencl_c_ext_fp64_global_atomic_min_max != 1
 179 #error "Incorrectly defined __opencl_c_ext_fp64_global_atomic_min_max"
 180 #endif
 181 #if __opencl_c_ext_fp16_local_atomic_min_max != 1
 182 #error "Incorrectly defined __opencl_c_ext_fp16_local_atomic_min_max"
 183 #endif
 184 #if __opencl_c_ext_fp32_local_atomic_min_max != 1
 185 #error "Incorrectly defined __opencl_c_ext_fp32_local_atomic_min_max"
 186 #endif
 187 #if __opencl_c_ext_fp64_local_atomic_min_max != 1
 188 #error "Incorrectly defined __opencl_c_ext_fp64_local_atomic_min_max"
 189 #endif
 190 #if __opencl_c_ext_image_raw10_raw12 != 1
 191 #error "Incorrectly defined __opencl_c_ext_image_raw10_raw12"
 192 #endif
 193 #if __opencl_c_ext_image_unorm_int_2_101010 != 1
 194 #error "Incorrectly defined __opencl_c_ext_image_unorm_int_2_101010"
 195 #endif
 196
 197 #else
 198
 199 #ifdef cl_khr_subgroup_extended_types
 200 #error "Incorrect cl_khr_subgroup_extended_types define"
 201 #endif
 202 #ifdef cl_khr_subgroup_non_uniform_vote
 203 #error "Incorrect cl_khr_subgroup_non_uniform_vote define"
 204 #endif
 205 #ifdef cl_khr_subgroup_ballot
 206 #error "Incorrect cl_khr_subgroup_ballot define"
 207 #endif
 208 #ifdef cl_khr_subgroup_non_uniform_arithmetic
 209 #error "Incorrect cl_khr_subgroup_non_uniform_arithmetic define"
 210 #endif
 211 #ifdef cl_khr_subgroup_shuffle
 212 #error "Incorrect cl_khr_subgroup_shuffle define"
 213 #endif
 214 #ifdef cl_khr_subgroup_shuffle_relative
 215 #error "Incorrect cl_khr_subgroup_shuffle_relative define"
 216 #endif
 217 #ifdef cl_khr_subgroup_clustered_reduce
 218 #error "Incorrect cl_khr_subgroup_clustered_reduce define"
 219 #endif
 220 #ifdef cl_khr_subgroup_rotate
 221 #error "Incorrect cl_khr_subgroup_rotate define"
 222 #endif
 223 #ifdef cl_khr_extended_bit_ops
 224 #error "Incorrect cl_khr_extended_bit_ops define"
 225 #endif
 226 #ifdef cl_khr_integer_dot_product
 227 #error "Incorrect cl_khr_integer_dot_product define"
 228 #endif
 229 #ifdef __opencl_c_integer_dot_product_input_4x8bit
 230 #error "Incorrect __opencl_c_integer_dot_product_input_4x8bit define"
 231 #endif
 232 #ifdef __opencl_c_integer_dot_product_input_4x8bit_packed
 233 #error "Incorrect __opencl_c_integer_dot_product_input_4x8bit_packed define"
 234 #endif
 235 #ifdef cl_ext_float_atomics
 236 #error "Incorrect cl_ext_float_atomics define"
 237 #endif
 238 #ifdef __opencl_c_ext_fp16_global_atomic_load_store
 239 #error "Incorrect __opencl_c_ext_fp16_global_atomic_load_store define"
 240 #endif
 241 #ifdef __opencl_c_ext_fp16_local_atomic_load_store
 242 #error "Incorrect __opencl_c_ext_fp16_local_atomic_load_store define"
 243 #endif
 244 #ifdef __opencl_c_ext_fp16_global_atomic_add
 245 #error "Incorrect __opencl_c_ext_fp16_global_atomic_add define"
 246 #endif
 247 #ifdef __opencl_c_ext_fp32_global_atomic_add
 248 #error "Incorrect __opencl_c_ext_fp32_global_atomic_add define"
 249 #endif
 250 #ifdef __opencl_c_ext_fp64_global_atomic_add
 251 #error "Incorrect __opencl_c_ext_fp64_global_atomic_add define"
 252 #endif
 253 #ifdef __opencl_c_ext_fp16_local_atomic_add
 254 #error "Incorrect __opencl_c_ext_fp16_local_atomic_add define"
 255 #endif
 256 #ifdef __opencl_c_ext_fp32_local_atomic_add
 257 #error "Incorrect __opencl_c_ext_fp32_local_atomic_add define"
 258 #endif
 259 #ifdef __opencl_c_ext_fp64_local_atomic_add
 260 #error "Incorrect __opencl_c_ext_fp64_local_atomic_add define"
 261 #endif
 262 #ifdef __opencl_c_ext_fp16_global_atomic_min_max
 263 #error "Incorrect __opencl_c_ext_fp16_global_atomic_min_max define"
 264 #endif
 265 #ifdef __opencl_c_ext_fp32_global_atomic_min_max
 266 #error "Incorrect __opencl_c_ext_fp32_global_atomic_min_max define"
 267 #endif
 268 #ifdef __opencl_c_ext_fp64_global_atomic_min_max
 269 #error "Incorrect __opencl_c_ext_fp64_global_atomic_min_max define"
 270 #endif
 271 #ifdef __opencl_c_ext_fp16_local_atomic_min_max
 272 #error "Incorrect __opencl_c_ext_fp16_local_atomic_min_max define"
 273 #endif
 274 #ifdef __opencl_c_ext_fp32_local_atomic_min_max
 275 #error "Incorrect __opencl_c_ext_fp32_local_atomic_min_max define"
 276 #endif
 277 #ifdef __opencl_c_ext_fp64_local_atomic_min_max
 278 #error "Incorrect __opencl_c_ext_fp64_local_atomic_min_max define"
 279 #endif
 280 #ifdef __opencl_c_ext_image_raw10_raw12
 281 #error "Incorrect __opencl_c_ext_image_raw10_raw12 define"
 282 #endif
 283 #ifdef __opencl_c_ext_image_unorm_int_2_101010
 284 #error "Incorrect __opencl_c_ext_image_unorm_int_2_101010 define"
 285 #endif
 286
 287 #endif //(defined(__OPENCL_CPP_VERSION__) || __OPENCL_C_VERSION__ >= 200)
 288
 289 // OpenCL C features.
 290 #if (__OPENCL_CPP_VERSION__ == 202100 || __OPENCL_C_VERSION__ == 300)
 291
 292 #if __opencl_c_atomic_scope_all_devices != 1
 293 #error "Incorrectly defined feature macro __opencl_c_atomic_scope_all_devices"
 294 #endif
 295
 296 #elif (__OPENCL_CPP_VERSION__ == 100 || __OPENCL_C_VERSION__ == 200)
 297
 298 #ifndef  __opencl_c_pipes
 299 #error "Feature macro __opencl_c_pipes should be defined"
 300 #endif
 301 #ifndef __opencl_c_generic_address_space
 302 #error "Feature macro __opencl_c_generic_address_space should be defined"
 303 #endif
 304 #ifndef __opencl_c_work_group_collective_functions
 305 #error "Feature macro __opencl_c_work_group_collective_functions should be defined"
 306 #endif
 307 #ifndef __opencl_c_atomic_order_acq_rel
 308 #error "Feature macro __opencl_c_atomic_order_acq_rel should be defined"
 309 #endif
 310 #ifndef __opencl_c_atomic_order_seq_cst
 311 #error "Feature macro __opencl_c_atomic_order_seq_cst should be defined"
 312 #endif
 313 #ifndef __opencl_c_atomic_scope_device
 314 #error "Feature macro __opencl_c_atomic_scope_device should be defined"
 315 #endif
 316 #ifndef __opencl_c_atomic_scope_all_devices
 317 #error "Feature macro __opencl_c_atomic_scope_all_devices should be defined"
 318 #endif
 319 #ifndef __opencl_c_device_enqueue
 320 #error "Feature macro __opencl_c_device_enqueue should be defined"
 321 #endif
 322 #ifndef __opencl_c_read_write_images
 323 #error "Feature macro __opencl_c_read_write_images should be defined"
 324 #endif
 325 #ifndef __opencl_c_program_scope_global_variables
 326 #error "Feature macro __opencl_c_program_scope_global_variables should be defined"
 327 #endif
 328 #ifndef __opencl_c_images
 329 #error "Feature macro __opencl_c_images should be defined"
 330 #endif
 331
 332 #elif (__OPENCL_C_VERSION__ < 200)
 333
 334 #ifdef  __opencl_c_pipes
 335 #error "Incorrect feature macro __opencl_c_pipes define"
 336 #endif
 337 #ifdef __opencl_c_generic_address_space
 338 #error "Incorrect feature macro __opencl_c_generic_address_space define"
 339 #endif
 340 #ifdef __opencl_c_work_group_collective_functions
 341 #error "Incorrect feature macro __opencl_c_work_group_collective_functions define"
 342 #endif
 343 #ifdef __opencl_c_atomic_order_acq_rel
 344 #error "Incorrect feature macro __opencl_c_atomic_order_acq_rel define"
 345 #endif
 346 #ifdef __opencl_c_atomic_order_seq_cst
 347 #error "Incorrect feature macro __opencl_c_atomic_order_seq_cst define"
 348 #endif
 349 #ifdef __opencl_c_atomic_scope_device
 350 #error "Incorrect feature macro __opencl_c_atomic_scope_device define"
 351 #endif
 352 #ifdef __opencl_c_atomic_scope_all_devices
 353 #error "Incorrect feature macro __opencl_c_atomic_scope_all_devices define"
 354 #endif
 355 #ifdef __opencl_c_device_enqueue
 356 #error "Incorrect feature macro __opencl_c_device_enqueue define"
 357 #endif
 358 #ifdef __opencl_c_read_write_images
 359 #error "Incorrect feature macro __opencl_c_read_write_images define"
 360 #endif
 361 #ifdef __opencl_c_program_scope_global_variables
 362 #error "Incorrect feature macro __opencl_c_program_scope_global_variables define"
 363 #endif
 364 #ifdef __opencl_c_images
 365 #error "Incorrect feature macro __opencl_c_images define"
 366 #endif
 367 #ifdef __opencl_c_3d_image_writes
 368 #error "Incorrect feature macro __opencl_c_3d_image_writes define"
 369 #endif
 370 #ifdef __opencl_c_fp64
 371 #error "Incorrect feature macro __opencl_c_fp64 define"
 372 #endif
 373 #ifdef __opencl_c_subgroups
 374 #error "Incorrect feature macro __opencl_c_subgroups define"
 375 #endif
 376
 377 #endif // (__OPENCL_CPP_VERSION__ == 202100 || __OPENCL_C_VERSION__ == 300)
 378
 379 #endif // defined(__SPIR__)