1 //===----- opencl-c-base.h - OpenCL C language base definitions -----------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 9 #ifndef _OPENCL_BASE_H_ 10 #define _OPENCL_BASE_H_ 11 12 // Define extension macros 13 14 #if (defined(__OPENCL_CPP_VERSION__) || __OPENCL_C_VERSION__ >= 200) 15 // For SPIR and SPIR-V all extensions are supported. 16 #if defined(__SPIR__) || defined(__SPIRV__) 17 #define cl_khr_subgroup_extended_types 1 18 #define cl_khr_subgroup_non_uniform_vote 1 19 #define cl_khr_subgroup_ballot 1 20 #define cl_khr_subgroup_non_uniform_arithmetic 1 21 #define cl_khr_subgroup_shuffle 1 22 #define cl_khr_subgroup_shuffle_relative 1 23 #define cl_khr_subgroup_clustered_reduce 1 24 #define cl_khr_subgroup_rotate 1 25 #define cl_khr_extended_bit_ops 1 26 #define cl_khr_integer_dot_product 1 27 #define __opencl_c_integer_dot_product_input_4x8bit 1 28 #define __opencl_c_integer_dot_product_input_4x8bit_packed 1 29 #define cl_ext_float_atomics 1 30 #ifdef cl_khr_fp16 31 #define __opencl_c_ext_fp16_global_atomic_load_store 1 32 #define __opencl_c_ext_fp16_local_atomic_load_store 1 33 #define __opencl_c_ext_fp16_global_atomic_add 1 34 #define __opencl_c_ext_fp16_local_atomic_add 1 35 #define __opencl_c_ext_fp16_global_atomic_min_max 1 36 #define __opencl_c_ext_fp16_local_atomic_min_max 1 37 #endif 38 #ifdef cl_khr_fp64 39 #define __opencl_c_ext_fp64_global_atomic_add 1 40 #define __opencl_c_ext_fp64_local_atomic_add 1 41 #define __opencl_c_ext_fp64_global_atomic_min_max 1 42 #define __opencl_c_ext_fp64_local_atomic_min_max 1 43 #endif 44 #define __opencl_c_ext_fp32_global_atomic_add 1 45 #define __opencl_c_ext_fp32_local_atomic_add 1 46 #define __opencl_c_ext_fp32_global_atomic_min_max 1 47 #define __opencl_c_ext_fp32_local_atomic_min_max 1 48 49 #endif // defined(__SPIR__) || defined(__SPIRV__) 50 #endif // (defined(__OPENCL_CPP_VERSION__) || __OPENCL_C_VERSION__ >= 200) 51 52 // Define feature macros for OpenCL C 2.0 53 #if (__OPENCL_CPP_VERSION__ == 100 || __OPENCL_C_VERSION__ == 200) 54 #define __opencl_c_pipes 1 55 #define __opencl_c_generic_address_space 1 56 #define __opencl_c_work_group_collective_functions 1 57 #define __opencl_c_atomic_order_acq_rel 1 58 #define __opencl_c_atomic_order_seq_cst 1 59 #define __opencl_c_atomic_scope_device 1 60 #define __opencl_c_atomic_scope_all_devices 1 61 #define __opencl_c_device_enqueue 1 62 #define __opencl_c_read_write_images 1 63 #define __opencl_c_program_scope_global_variables 1 64 #define __opencl_c_images 1 65 #endif 66 67 // Define header-only feature macros for OpenCL C 3.0. 68 #if (__OPENCL_CPP_VERSION__ == 202100 || __OPENCL_C_VERSION__ == 300) 69 // For the SPIR and SPIR-V target all features are supported. 70 #if defined(__SPIR__) || defined(__SPIRV__) 71 #define __opencl_c_work_group_collective_functions 1 72 #define __opencl_c_atomic_order_seq_cst 1 73 #define __opencl_c_atomic_scope_device 1 74 #define __opencl_c_atomic_scope_all_devices 1 75 #define __opencl_c_read_write_images 1 76 #endif // defined(__SPIR__) 77 #endif // (__OPENCL_CPP_VERSION__ == 202100 || __OPENCL_C_VERSION__ == 300) 78 79 #if !defined(__opencl_c_generic_address_space) 80 // Internal feature macro to provide named (global, local, private) address 81 // space overloads for builtin functions that take a pointer argument. 82 #define __opencl_c_named_address_space_builtins 1 83 #endif // !defined(__opencl_c_generic_address_space) 84 85 #if defined(cl_intel_subgroups) || defined(cl_khr_subgroups) || defined(__opencl_c_subgroups) 86 // Internal feature macro to provide subgroup builtins. 87 #define __opencl_subgroup_builtins 1 88 #endif 89 90 // built-in scalar data types: 91 92 /** 93 * An unsigned 8-bit integer. 94 */ 95 typedef unsigned char uchar; 96 97 /** 98 * An unsigned 16-bit integer. 99 */ 100 typedef unsigned short ushort; 101 102 /** 103 * An unsigned 32-bit integer. 104 */ 105 typedef unsigned int uint; 106 107 /** 108 * An unsigned 64-bit integer. 109 */ 110 typedef unsigned long ulong; 111 112 /** 113 * The unsigned integer type of the result of the sizeof operator. This 114 * is a 32-bit unsigned integer if CL_DEVICE_ADDRESS_BITS 115 * defined in table 4.3 is 32-bits and is a 64-bit unsigned integer if 116 * CL_DEVICE_ADDRESS_BITS is 64-bits. 117 */ 118 typedef __SIZE_TYPE__ size_t; 119 120 /** 121 * A signed integer type that is the result of subtracting two pointers. 122 * This is a 32-bit signed integer if CL_DEVICE_ADDRESS_BITS 123 * defined in table 4.3 is 32-bits and is a 64-bit signed integer if 124 * CL_DEVICE_ADDRESS_BITS is 64-bits. 125 */ 126 typedef __PTRDIFF_TYPE__ ptrdiff_t; 127 128 /** 129 * A signed integer type with the property that any valid pointer to 130 * void can be converted to this type, then converted back to pointer 131 * to void, and the result will compare equal to the original pointer. 132 */ 133 typedef __INTPTR_TYPE__ intptr_t; 134 135 /** 136 * An unsigned integer type with the property that any valid pointer to 137 * void can be converted to this type, then converted back to pointer 138 * to void, and the result will compare equal to the original pointer. 139 */ 140 typedef __UINTPTR_TYPE__ uintptr_t; 141 142 // built-in vector data types: 143 typedef char char2 __attribute__((ext_vector_type(2))); 144 typedef char char3 __attribute__((ext_vector_type(3))); 145 typedef char char4 __attribute__((ext_vector_type(4))); 146 typedef char char8 __attribute__((ext_vector_type(8))); 147 typedef char char16 __attribute__((ext_vector_type(16))); 148 typedef uchar uchar2 __attribute__((ext_vector_type(2))); 149 typedef uchar uchar3 __attribute__((ext_vector_type(3))); 150 typedef uchar uchar4 __attribute__((ext_vector_type(4))); 151 typedef uchar uchar8 __attribute__((ext_vector_type(8))); 152 typedef uchar uchar16 __attribute__((ext_vector_type(16))); 153 typedef short short2 __attribute__((ext_vector_type(2))); 154 typedef short short3 __attribute__((ext_vector_type(3))); 155 typedef short short4 __attribute__((ext_vector_type(4))); 156 typedef short short8 __attribute__((ext_vector_type(8))); 157 typedef short short16 __attribute__((ext_vector_type(16))); 158 typedef ushort ushort2 __attribute__((ext_vector_type(2))); 159 typedef ushort ushort3 __attribute__((ext_vector_type(3))); 160 typedef ushort ushort4 __attribute__((ext_vector_type(4))); 161 typedef ushort ushort8 __attribute__((ext_vector_type(8))); 162 typedef ushort ushort16 __attribute__((ext_vector_type(16))); 163 typedef int int2 __attribute__((ext_vector_type(2))); 164 typedef int int3 __attribute__((ext_vector_type(3))); 165 typedef int int4 __attribute__((ext_vector_type(4))); 166 typedef int int8 __attribute__((ext_vector_type(8))); 167 typedef int int16 __attribute__((ext_vector_type(16))); 168 typedef uint uint2 __attribute__((ext_vector_type(2))); 169 typedef uint uint3 __attribute__((ext_vector_type(3))); 170 typedef uint uint4 __attribute__((ext_vector_type(4))); 171 typedef uint uint8 __attribute__((ext_vector_type(8))); 172 typedef uint uint16 __attribute__((ext_vector_type(16))); 173 typedef long long2 __attribute__((ext_vector_type(2))); 174 typedef long long3 __attribute__((ext_vector_type(3))); 175 typedef long long4 __attribute__((ext_vector_type(4))); 176 typedef long long8 __attribute__((ext_vector_type(8))); 177 typedef long long16 __attribute__((ext_vector_type(16))); 178 typedef ulong ulong2 __attribute__((ext_vector_type(2))); 179 typedef ulong ulong3 __attribute__((ext_vector_type(3))); 180 typedef ulong ulong4 __attribute__((ext_vector_type(4))); 181 typedef ulong ulong8 __attribute__((ext_vector_type(8))); 182 typedef ulong ulong16 __attribute__((ext_vector_type(16))); 183 typedef float float2 __attribute__((ext_vector_type(2))); 184 typedef float float3 __attribute__((ext_vector_type(3))); 185 typedef float float4 __attribute__((ext_vector_type(4))); 186 typedef float float8 __attribute__((ext_vector_type(8))); 187 typedef float float16 __attribute__((ext_vector_type(16))); 188 #ifdef cl_khr_fp16 189 #pragma OPENCL EXTENSION cl_khr_fp16 : enable 190 typedef half half2 __attribute__((ext_vector_type(2))); 191 typedef half half3 __attribute__((ext_vector_type(3))); 192 typedef half half4 __attribute__((ext_vector_type(4))); 193 typedef half half8 __attribute__((ext_vector_type(8))); 194 typedef half half16 __attribute__((ext_vector_type(16))); 195 #endif 196 #ifdef cl_khr_fp64 197 #if __OPENCL_C_VERSION__ < CL_VERSION_1_2 198 #pragma OPENCL EXTENSION cl_khr_fp64 : enable 199 #endif 200 typedef double double2 __attribute__((ext_vector_type(2))); 201 typedef double double3 __attribute__((ext_vector_type(3))); 202 typedef double double4 __attribute__((ext_vector_type(4))); 203 typedef double double8 __attribute__((ext_vector_type(8))); 204 typedef double double16 __attribute__((ext_vector_type(16))); 205 #endif 206 207 // An internal alias for half, for use by OpenCLBuiltins.td. 208 #define __half half 209 210 #if defined(__OPENCL_CPP_VERSION__) 211 #define NULL nullptr 212 #elif defined(__OPENCL_C_VERSION__) 213 #define NULL ((void*)0) 214 #endif 215 216 /** 217 * Value of maximum non-infinite single-precision floating-point 218 * number. 219 */ 220 #define MAXFLOAT 0x1.fffffep127f 221 222 /** 223 * A positive float constant expression. HUGE_VALF evaluates 224 * to +infinity. Used as an error value returned by the built-in 225 * math functions. 226 */ 227 #define HUGE_VALF (__builtin_huge_valf()) 228 229 /** 230 * A positive double constant expression. HUGE_VAL evaluates 231 * to +infinity. Used as an error value returned by the built-in 232 * math functions. 233 */ 234 #define HUGE_VAL (__builtin_huge_val()) 235 236 /** 237 * A constant expression of type float representing positive or 238 * unsigned infinity. 239 */ 240 #define INFINITY (__builtin_inff()) 241 242 /** 243 * A constant expression of type float representing a quiet NaN. 244 */ 245 #define NAN as_float(INT_MAX) 246 247 #define FP_ILOGB0 INT_MIN 248 #define FP_ILOGBNAN INT_MAX 249 250 #define FLT_DIG 6 251 #define FLT_MANT_DIG 24 252 #define FLT_MAX_10_EXP +38 253 #define FLT_MAX_EXP +128 254 #define FLT_MIN_10_EXP -37 255 #define FLT_MIN_EXP -125 256 #define FLT_RADIX 2 257 #define FLT_MAX 0x1.fffffep127f 258 #define FLT_MIN 0x1.0p-126f 259 #define FLT_EPSILON 0x1.0p-23f 260 261 #define M_E_F 2.71828182845904523536028747135266250f 262 #define M_LOG2E_F 1.44269504088896340735992468100189214f 263 #define M_LOG10E_F 0.434294481903251827651128918916605082f 264 #define M_LN2_F 0.693147180559945309417232121458176568f 265 #define M_LN10_F 2.30258509299404568401799145468436421f 266 #define M_PI_F 3.14159265358979323846264338327950288f 267 #define M_PI_2_F 1.57079632679489661923132169163975144f 268 #define M_PI_4_F 0.785398163397448309615660845819875721f 269 #define M_1_PI_F 0.318309886183790671537767526745028724f 270 #define M_2_PI_F 0.636619772367581343075535053490057448f 271 #define M_2_SQRTPI_F 1.12837916709551257389615890312154517f 272 #define M_SQRT2_F 1.41421356237309504880168872420969808f 273 #define M_SQRT1_2_F 0.707106781186547524400844362104849039f 274 275 #define DBL_DIG 15 276 #define DBL_MANT_DIG 53 277 #define DBL_MAX_10_EXP +308 278 #define DBL_MAX_EXP +1024 279 #define DBL_MIN_10_EXP -307 280 #define DBL_MIN_EXP -1021 281 #define DBL_RADIX 2 282 #define DBL_MAX 0x1.fffffffffffffp1023 283 #define DBL_MIN 0x1.0p-1022 284 #define DBL_EPSILON 0x1.0p-52 285 286 #define M_E 0x1.5bf0a8b145769p+1 287 #define M_LOG2E 0x1.71547652b82fep+0 288 #define M_LOG10E 0x1.bcb7b1526e50ep-2 289 #define M_LN2 0x1.62e42fefa39efp-1 290 #define M_LN10 0x1.26bb1bbb55516p+1 291 #define M_PI 0x1.921fb54442d18p+1 292 #define M_PI_2 0x1.921fb54442d18p+0 293 #define M_PI_4 0x1.921fb54442d18p-1 294 #define M_1_PI 0x1.45f306dc9c883p-2 295 #define M_2_PI 0x1.45f306dc9c883p-1 296 #define M_2_SQRTPI 0x1.20dd750429b6dp+0 297 #define M_SQRT2 0x1.6a09e667f3bcdp+0 298 #define M_SQRT1_2 0x1.6a09e667f3bcdp-1 299 300 #ifdef cl_khr_fp16 301 302 #define HALF_DIG 3 303 #define HALF_MANT_DIG 11 304 #define HALF_MAX_10_EXP +4 305 #define HALF_MAX_EXP +16 306 #define HALF_MIN_10_EXP -4 307 #define HALF_MIN_EXP -13 308 #define HALF_RADIX 2 309 #define HALF_MAX ((0x1.ffcp15h)) 310 #define HALF_MIN ((0x1.0p-14h)) 311 #define HALF_EPSILON ((0x1.0p-10h)) 312 313 #define M_E_H 2.71828182845904523536028747135266250h 314 #define M_LOG2E_H 1.44269504088896340735992468100189214h 315 #define M_LOG10E_H 0.434294481903251827651128918916605082h 316 #define M_LN2_H 0.693147180559945309417232121458176568h 317 #define M_LN10_H 2.30258509299404568401799145468436421h 318 #define M_PI_H 3.14159265358979323846264338327950288h 319 #define M_PI_2_H 1.57079632679489661923132169163975144h 320 #define M_PI_4_H 0.785398163397448309615660845819875721h 321 #define M_1_PI_H 0.318309886183790671537767526745028724h 322 #define M_2_PI_H 0.636619772367581343075535053490057448h 323 #define M_2_SQRTPI_H 1.12837916709551257389615890312154517h 324 #define M_SQRT2_H 1.41421356237309504880168872420969808h 325 #define M_SQRT1_2_H 0.707106781186547524400844362104849039h 326 327 #endif //cl_khr_fp16 328 329 #define CHAR_BIT 8 330 #define SCHAR_MAX 127 331 #define SCHAR_MIN (-128) 332 #define UCHAR_MAX 255 333 #define CHAR_MAX SCHAR_MAX 334 #define CHAR_MIN SCHAR_MIN 335 #define USHRT_MAX 65535 336 #define SHRT_MAX 32767 337 #define SHRT_MIN (-32768) 338 #define UINT_MAX 0xffffffff 339 #define INT_MAX 2147483647 340 #define INT_MIN (-2147483647-1) 341 #define ULONG_MAX 0xffffffffffffffffUL 342 #define LONG_MAX 0x7fffffffffffffffL 343 #define LONG_MIN (-0x7fffffffffffffffL-1) 344 345 // OpenCL v1.1 s6.11.8, v1.2 s6.12.8, v2.0 s6.13.8 - Synchronization Functions 346 347 // Flag type and values for barrier, mem_fence, read_mem_fence, write_mem_fence 348 typedef uint cl_mem_fence_flags; 349 350 /** 351 * Queue a memory fence to ensure correct 352 * ordering of memory operations to local memory 353 */ 354 #define CLK_LOCAL_MEM_FENCE 0x01 355 356 /** 357 * Queue a memory fence to ensure correct 358 * ordering of memory operations to global memory 359 */ 360 #define CLK_GLOBAL_MEM_FENCE 0x02 361 362 #if defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0) 363 364 typedef enum memory_scope { 365 memory_scope_work_item = __OPENCL_MEMORY_SCOPE_WORK_ITEM, 366 memory_scope_work_group = __OPENCL_MEMORY_SCOPE_WORK_GROUP, 367 memory_scope_device = __OPENCL_MEMORY_SCOPE_DEVICE, 368 #if defined(__opencl_c_atomic_scope_all_devices) 369 memory_scope_all_svm_devices = __OPENCL_MEMORY_SCOPE_ALL_SVM_DEVICES, 370 #if (__OPENCL_C_VERSION__ >= CL_VERSION_3_0 || __OPENCL_CPP_VERSION__ >= 202100) 371 memory_scope_all_devices = memory_scope_all_svm_devices, 372 #endif // (__OPENCL_C_VERSION__ >= CL_VERSION_3_0 || __OPENCL_CPP_VERSION__ >= 202100) 373 #endif // defined(__opencl_c_atomic_scope_all_devices) 374 /** 375 * Subgroups have different requirements on forward progress, so just test 376 * all the relevant macros. 377 * CL 3.0 sub-groups "they are not guaranteed to make independent forward progress" 378 * KHR subgroups "Subgroups within a workgroup are independent, make forward progress with respect to each other" 379 */ 380 #if defined(cl_intel_subgroups) || defined(cl_khr_subgroups) || defined(__opencl_c_subgroups) 381 memory_scope_sub_group = __OPENCL_MEMORY_SCOPE_SUB_GROUP 382 #endif 383 } memory_scope; 384 385 /** 386 * Queue a memory fence to ensure correct ordering of memory 387 * operations between work-items of a work-group to 388 * image memory. 389 */ 390 #define CLK_IMAGE_MEM_FENCE 0x04 391 392 #ifndef ATOMIC_VAR_INIT 393 #define ATOMIC_VAR_INIT(x) (x) 394 #endif //ATOMIC_VAR_INIT 395 #define ATOMIC_FLAG_INIT 0 396 397 // enum values aligned with what clang uses in EmitAtomicExpr() 398 typedef enum memory_order 399 { 400 memory_order_relaxed = __ATOMIC_RELAXED, 401 memory_order_acquire = __ATOMIC_ACQUIRE, 402 memory_order_release = __ATOMIC_RELEASE, 403 memory_order_acq_rel = __ATOMIC_ACQ_REL, 404 #if defined(__opencl_c_atomic_order_seq_cst) 405 memory_order_seq_cst = __ATOMIC_SEQ_CST 406 #endif 407 } memory_order; 408 409 #endif // defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0) 410 411 // OpenCL v1.1 s6.11.3, v1.2 s6.12.14, v2.0 s6.13.14 - Image Read and Write Functions 412 413 // These values need to match the runtime equivalent 414 // 415 // Addressing Mode. 416 // 417 #define CLK_ADDRESS_NONE 0 418 #define CLK_ADDRESS_CLAMP_TO_EDGE 2 419 #define CLK_ADDRESS_CLAMP 4 420 #define CLK_ADDRESS_REPEAT 6 421 #define CLK_ADDRESS_MIRRORED_REPEAT 8 422 423 // 424 // Coordination Normalization 425 // 426 #define CLK_NORMALIZED_COORDS_FALSE 0 427 #define CLK_NORMALIZED_COORDS_TRUE 1 428 429 // 430 // Filtering Mode. 431 // 432 #define CLK_FILTER_NEAREST 0x10 433 #define CLK_FILTER_LINEAR 0x20 434 435 #ifdef cl_khr_gl_msaa_sharing 436 #pragma OPENCL EXTENSION cl_khr_gl_msaa_sharing : enable 437 #endif //cl_khr_gl_msaa_sharing 438 439 // 440 // Channel Datatype. 441 // 442 #define CLK_SNORM_INT8 0x10D0 443 #define CLK_SNORM_INT16 0x10D1 444 #define CLK_UNORM_INT8 0x10D2 445 #define CLK_UNORM_INT16 0x10D3 446 #define CLK_UNORM_SHORT_565 0x10D4 447 #define CLK_UNORM_SHORT_555 0x10D5 448 #define CLK_UNORM_INT_101010 0x10D6 449 #define CLK_SIGNED_INT8 0x10D7 450 #define CLK_SIGNED_INT16 0x10D8 451 #define CLK_SIGNED_INT32 0x10D9 452 #define CLK_UNSIGNED_INT8 0x10DA 453 #define CLK_UNSIGNED_INT16 0x10DB 454 #define CLK_UNSIGNED_INT32 0x10DC 455 #define CLK_HALF_FLOAT 0x10DD 456 #define CLK_FLOAT 0x10DE 457 #define CLK_UNORM_INT24 0x10DF 458 459 // Channel order, numbering must be aligned with cl_channel_order in cl.h 460 // 461 #define CLK_R 0x10B0 462 #define CLK_A 0x10B1 463 #define CLK_RG 0x10B2 464 #define CLK_RA 0x10B3 465 #define CLK_RGB 0x10B4 466 #define CLK_RGBA 0x10B5 467 #define CLK_BGRA 0x10B6 468 #define CLK_ARGB 0x10B7 469 #define CLK_INTENSITY 0x10B8 470 #define CLK_LUMINANCE 0x10B9 471 #define CLK_Rx 0x10BA 472 #define CLK_RGx 0x10BB 473 #define CLK_RGBx 0x10BC 474 #define CLK_DEPTH 0x10BD 475 #define CLK_DEPTH_STENCIL 0x10BE 476 #if __OPENCL_C_VERSION__ >= CL_VERSION_2_0 477 #define CLK_sRGB 0x10BF 478 #define CLK_sRGBx 0x10C0 479 #define CLK_sRGBA 0x10C1 480 #define CLK_sBGRA 0x10C2 481 #define CLK_ABGR 0x10C3 482 #endif //__OPENCL_C_VERSION__ >= CL_VERSION_2_0 483 484 // OpenCL v2.0 s6.13.16 - Pipe Functions 485 #if defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0) 486 #define CLK_NULL_RESERVE_ID (__builtin_astype(((void*)(__SIZE_MAX__)), reserve_id_t)) 487 488 // OpenCL v2.0 s6.13.17 - Enqueue Kernels 489 #define CL_COMPLETE 0x0 490 #define CL_RUNNING 0x1 491 #define CL_SUBMITTED 0x2 492 #define CL_QUEUED 0x3 493 494 #define CLK_SUCCESS 0 495 #define CLK_ENQUEUE_FAILURE -101 496 #define CLK_INVALID_QUEUE -102 497 #define CLK_INVALID_NDRANGE -160 498 #define CLK_INVALID_EVENT_WAIT_LIST -57 499 #define CLK_DEVICE_QUEUE_FULL -161 500 #define CLK_INVALID_ARG_SIZE -51 501 #define CLK_EVENT_ALLOCATION_FAILURE -100 502 #define CLK_OUT_OF_RESOURCES -5 503 504 #define CLK_NULL_QUEUE 0 505 #define CLK_NULL_EVENT (__builtin_astype(((__SIZE_MAX__)), clk_event_t)) 506 507 // execution model related definitions 508 #define CLK_ENQUEUE_FLAGS_NO_WAIT 0x0 509 #define CLK_ENQUEUE_FLAGS_WAIT_KERNEL 0x1 510 #define CLK_ENQUEUE_FLAGS_WAIT_WORK_GROUP 0x2 511 512 typedef int kernel_enqueue_flags_t; 513 typedef int clk_profiling_info; 514 515 // Profiling info name (see capture_event_profiling_info) 516 #define CLK_PROFILING_COMMAND_EXEC_TIME 0x1 517 518 #define MAX_WORK_DIM 3 519 520 #ifdef __opencl_c_device_enqueue 521 typedef struct { 522 unsigned int workDimension; 523 size_t globalWorkOffset[MAX_WORK_DIM]; 524 size_t globalWorkSize[MAX_WORK_DIM]; 525 size_t localWorkSize[MAX_WORK_DIM]; 526 } ndrange_t; 527 #endif // __opencl_c_device_enqueue 528 529 #endif // defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0) 530 531 /** 532 * OpenCL v1.1/1.2/2.0 s6.2.4.2 - as_type operators 533 * Reinterprets a data type as another data type of the same size 534 */ 535 #define as_char(x) __builtin_astype((x), char) 536 #define as_char2(x) __builtin_astype((x), char2) 537 #define as_char3(x) __builtin_astype((x), char3) 538 #define as_char4(x) __builtin_astype((x), char4) 539 #define as_char8(x) __builtin_astype((x), char8) 540 #define as_char16(x) __builtin_astype((x), char16) 541 542 #define as_uchar(x) __builtin_astype((x), uchar) 543 #define as_uchar2(x) __builtin_astype((x), uchar2) 544 #define as_uchar3(x) __builtin_astype((x), uchar3) 545 #define as_uchar4(x) __builtin_astype((x), uchar4) 546 #define as_uchar8(x) __builtin_astype((x), uchar8) 547 #define as_uchar16(x) __builtin_astype((x), uchar16) 548 549 #define as_short(x) __builtin_astype((x), short) 550 #define as_short2(x) __builtin_astype((x), short2) 551 #define as_short3(x) __builtin_astype((x), short3) 552 #define as_short4(x) __builtin_astype((x), short4) 553 #define as_short8(x) __builtin_astype((x), short8) 554 #define as_short16(x) __builtin_astype((x), short16) 555 556 #define as_ushort(x) __builtin_astype((x), ushort) 557 #define as_ushort2(x) __builtin_astype((x), ushort2) 558 #define as_ushort3(x) __builtin_astype((x), ushort3) 559 #define as_ushort4(x) __builtin_astype((x), ushort4) 560 #define as_ushort8(x) __builtin_astype((x), ushort8) 561 #define as_ushort16(x) __builtin_astype((x), ushort16) 562 563 #define as_int(x) __builtin_astype((x), int) 564 #define as_int2(x) __builtin_astype((x), int2) 565 #define as_int3(x) __builtin_astype((x), int3) 566 #define as_int4(x) __builtin_astype((x), int4) 567 #define as_int8(x) __builtin_astype((x), int8) 568 #define as_int16(x) __builtin_astype((x), int16) 569 570 #define as_uint(x) __builtin_astype((x), uint) 571 #define as_uint2(x) __builtin_astype((x), uint2) 572 #define as_uint3(x) __builtin_astype((x), uint3) 573 #define as_uint4(x) __builtin_astype((x), uint4) 574 #define as_uint8(x) __builtin_astype((x), uint8) 575 #define as_uint16(x) __builtin_astype((x), uint16) 576 577 #define as_long(x) __builtin_astype((x), long) 578 #define as_long2(x) __builtin_astype((x), long2) 579 #define as_long3(x) __builtin_astype((x), long3) 580 #define as_long4(x) __builtin_astype((x), long4) 581 #define as_long8(x) __builtin_astype((x), long8) 582 #define as_long16(x) __builtin_astype((x), long16) 583 584 #define as_ulong(x) __builtin_astype((x), ulong) 585 #define as_ulong2(x) __builtin_astype((x), ulong2) 586 #define as_ulong3(x) __builtin_astype((x), ulong3) 587 #define as_ulong4(x) __builtin_astype((x), ulong4) 588 #define as_ulong8(x) __builtin_astype((x), ulong8) 589 #define as_ulong16(x) __builtin_astype((x), ulong16) 590 591 #define as_float(x) __builtin_astype((x), float) 592 #define as_float2(x) __builtin_astype((x), float2) 593 #define as_float3(x) __builtin_astype((x), float3) 594 #define as_float4(x) __builtin_astype((x), float4) 595 #define as_float8(x) __builtin_astype((x), float8) 596 #define as_float16(x) __builtin_astype((x), float16) 597 598 #ifdef cl_khr_fp64 599 #define as_double(x) __builtin_astype((x), double) 600 #define as_double2(x) __builtin_astype((x), double2) 601 #define as_double3(x) __builtin_astype((x), double3) 602 #define as_double4(x) __builtin_astype((x), double4) 603 #define as_double8(x) __builtin_astype((x), double8) 604 #define as_double16(x) __builtin_astype((x), double16) 605 #endif // cl_khr_fp64 606 607 #ifdef cl_khr_fp16 608 #define as_half(x) __builtin_astype((x), half) 609 #define as_half2(x) __builtin_astype((x), half2) 610 #define as_half3(x) __builtin_astype((x), half3) 611 #define as_half4(x) __builtin_astype((x), half4) 612 #define as_half8(x) __builtin_astype((x), half8) 613 #define as_half16(x) __builtin_astype((x), half16) 614 #endif // cl_khr_fp16 615 616 #define as_size_t(x) __builtin_astype((x), size_t) 617 #define as_ptrdiff_t(x) __builtin_astype((x), ptrdiff_t) 618 #define as_intptr_t(x) __builtin_astype((x), intptr_t) 619 #define as_uintptr_t(x) __builtin_astype((x), uintptr_t) 620 621 // C++ for OpenCL - __remove_address_space 622 #if defined(__OPENCL_CPP_VERSION__) 623 template <typename _Tp> struct __remove_address_space { using type = _Tp; }; 624 #if defined(__opencl_c_generic_address_space) 625 template <typename _Tp> struct __remove_address_space<__generic _Tp> { 626 using type = _Tp; 627 }; 628 #endif 629 template <typename _Tp> struct __remove_address_space<__global _Tp> { 630 using type = _Tp; 631 }; 632 template <typename _Tp> struct __remove_address_space<__private _Tp> { 633 using type = _Tp; 634 }; 635 template <typename _Tp> struct __remove_address_space<__local _Tp> { 636 using type = _Tp; 637 }; 638 template <typename _Tp> struct __remove_address_space<__constant _Tp> { 639 using type = _Tp; 640 }; 641 #endif 642 643 // OpenCL v1.1 s6.9, v1.2/2.0 s6.10 - Function qualifiers 644 645 #define __kernel_exec(X, typen) __kernel \ 646 __attribute__((work_group_size_hint(X, 1, 1))) \ 647 __attribute__((vec_type_hint(typen))) 648 649 #define kernel_exec(X, typen) __kernel \ 650 __attribute__((work_group_size_hint(X, 1, 1))) \ 651 __attribute__((vec_type_hint(typen))) 652 653 #if defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_1_2) 654 // OpenCL v1.2 s6.12.13, v2.0 s6.13.13 - printf 655 656 int printf(__constant const char* st, ...) __attribute__((format(printf, 1, 2))); 657 #endif 658 659 #ifdef cl_intel_device_side_avc_motion_estimation 660 661 #define CLK_AVC_ME_MAJOR_16x16_INTEL 0x0 662 #define CLK_AVC_ME_MAJOR_16x8_INTEL 0x1 663 #define CLK_AVC_ME_MAJOR_8x16_INTEL 0x2 664 #define CLK_AVC_ME_MAJOR_8x8_INTEL 0x3 665 666 #define CLK_AVC_ME_MINOR_8x8_INTEL 0x0 667 #define CLK_AVC_ME_MINOR_8x4_INTEL 0x1 668 #define CLK_AVC_ME_MINOR_4x8_INTEL 0x2 669 #define CLK_AVC_ME_MINOR_4x4_INTEL 0x3 670 671 #define CLK_AVC_ME_MAJOR_FORWARD_INTEL 0x0 672 #define CLK_AVC_ME_MAJOR_BACKWARD_INTEL 0x1 673 #define CLK_AVC_ME_MAJOR_BIDIRECTIONAL_INTEL 0x2 674 675 #define CLK_AVC_ME_PARTITION_MASK_ALL_INTEL 0x0 676 #define CLK_AVC_ME_PARTITION_MASK_16x16_INTEL 0x7E 677 #define CLK_AVC_ME_PARTITION_MASK_16x8_INTEL 0x7D 678 #define CLK_AVC_ME_PARTITION_MASK_8x16_INTEL 0x7B 679 #define CLK_AVC_ME_PARTITION_MASK_8x8_INTEL 0x77 680 #define CLK_AVC_ME_PARTITION_MASK_8x4_INTEL 0x6F 681 #define CLK_AVC_ME_PARTITION_MASK_4x8_INTEL 0x5F 682 #define CLK_AVC_ME_PARTITION_MASK_4x4_INTEL 0x3F 683 684 #define CLK_AVC_ME_SLICE_TYPE_PRED_INTEL 0x0 685 #define CLK_AVC_ME_SLICE_TYPE_BPRED_INTEL 0x1 686 #define CLK_AVC_ME_SLICE_TYPE_INTRA_INTEL 0x2 687 688 #define CLK_AVC_ME_SEARCH_WINDOW_EXHAUSTIVE_INTEL 0x0 689 #define CLK_AVC_ME_SEARCH_WINDOW_SMALL_INTEL 0x1 690 #define CLK_AVC_ME_SEARCH_WINDOW_TINY_INTEL 0x2 691 #define CLK_AVC_ME_SEARCH_WINDOW_EXTRA_TINY_INTEL 0x3 692 #define CLK_AVC_ME_SEARCH_WINDOW_DIAMOND_INTEL 0x4 693 #define CLK_AVC_ME_SEARCH_WINDOW_LARGE_DIAMOND_INTEL 0x5 694 #define CLK_AVC_ME_SEARCH_WINDOW_RESERVED0_INTEL 0x6 695 #define CLK_AVC_ME_SEARCH_WINDOW_RESERVED1_INTEL 0x7 696 #define CLK_AVC_ME_SEARCH_WINDOW_CUSTOM_INTEL 0x8 697 698 #define CLK_AVC_ME_SAD_ADJUST_MODE_NONE_INTEL 0x0 699 #define CLK_AVC_ME_SAD_ADJUST_MODE_HAAR_INTEL 0x2 700 701 #define CLK_AVC_ME_SUBPIXEL_MODE_INTEGER_INTEL 0x0 702 #define CLK_AVC_ME_SUBPIXEL_MODE_HPEL_INTEL 0x1 703 #define CLK_AVC_ME_SUBPIXEL_MODE_QPEL_INTEL 0x3 704 705 #define CLK_AVC_ME_COST_PRECISION_QPEL_INTEL 0x0 706 #define CLK_AVC_ME_COST_PRECISION_HPEL_INTEL 0x1 707 #define CLK_AVC_ME_COST_PRECISION_PEL_INTEL 0x2 708 #define CLK_AVC_ME_COST_PRECISION_DPEL_INTEL 0x3 709 710 #define CLK_AVC_ME_BIDIR_WEIGHT_QUARTER_INTEL 0x10 711 #define CLK_AVC_ME_BIDIR_WEIGHT_THIRD_INTEL 0x15 712 #define CLK_AVC_ME_BIDIR_WEIGHT_HALF_INTEL 0x20 713 #define CLK_AVC_ME_BIDIR_WEIGHT_TWO_THIRD_INTEL 0x2B 714 #define CLK_AVC_ME_BIDIR_WEIGHT_THREE_QUARTER_INTEL 0x30 715 716 #define CLK_AVC_ME_BORDER_REACHED_LEFT_INTEL 0x0 717 #define CLK_AVC_ME_BORDER_REACHED_RIGHT_INTEL 0x2 718 #define CLK_AVC_ME_BORDER_REACHED_TOP_INTEL 0x4 719 #define CLK_AVC_ME_BORDER_REACHED_BOTTOM_INTEL 0x8 720 721 #define CLK_AVC_ME_INTRA_16x16_INTEL 0x0 722 #define CLK_AVC_ME_INTRA_8x8_INTEL 0x1 723 #define CLK_AVC_ME_INTRA_4x4_INTEL 0x2 724 725 #define CLK_AVC_ME_SKIP_BLOCK_PARTITION_16x16_INTEL 0x0 726 #define CLK_AVC_ME_SKIP_BLOCK_PARTITION_8x8_INTEL 0x4000 727 728 #define CLK_AVC_ME_SKIP_BLOCK_16x16_FORWARD_ENABLE_INTEL (0x1 << 24) 729 #define CLK_AVC_ME_SKIP_BLOCK_16x16_BACKWARD_ENABLE_INTEL (0x2 << 24) 730 #define CLK_AVC_ME_SKIP_BLOCK_16x16_DUAL_ENABLE_INTEL (0x3 << 24) 731 #define CLK_AVC_ME_SKIP_BLOCK_8x8_FORWARD_ENABLE_INTEL (0x55 << 24) 732 #define CLK_AVC_ME_SKIP_BLOCK_8x8_BACKWARD_ENABLE_INTEL (0xAA << 24) 733 #define CLK_AVC_ME_SKIP_BLOCK_8x8_DUAL_ENABLE_INTEL (0xFF << 24) 734 #define CLK_AVC_ME_SKIP_BLOCK_8x8_0_FORWARD_ENABLE_INTEL (0x1 << 24) 735 #define CLK_AVC_ME_SKIP_BLOCK_8x8_0_BACKWARD_ENABLE_INTEL (0x2 << 24) 736 #define CLK_AVC_ME_SKIP_BLOCK_8x8_1_FORWARD_ENABLE_INTEL (0x1 << 26) 737 #define CLK_AVC_ME_SKIP_BLOCK_8x8_1_BACKWARD_ENABLE_INTEL (0x2 << 26) 738 #define CLK_AVC_ME_SKIP_BLOCK_8x8_2_FORWARD_ENABLE_INTEL (0x1 << 28) 739 #define CLK_AVC_ME_SKIP_BLOCK_8x8_2_BACKWARD_ENABLE_INTEL (0x2 << 28) 740 #define CLK_AVC_ME_SKIP_BLOCK_8x8_3_FORWARD_ENABLE_INTEL (0x1 << 30) 741 #define CLK_AVC_ME_SKIP_BLOCK_8x8_3_BACKWARD_ENABLE_INTEL (0x2 << 30) 742 743 #define CLK_AVC_ME_BLOCK_BASED_SKIP_4x4_INTEL 0x00 744 #define CLK_AVC_ME_BLOCK_BASED_SKIP_8x8_INTEL 0x80 745 746 #define CLK_AVC_ME_INTRA_LUMA_PARTITION_MASK_ALL_INTEL 0x0 747 #define CLK_AVC_ME_INTRA_LUMA_PARTITION_MASK_16x16_INTEL 0x6 748 #define CLK_AVC_ME_INTRA_LUMA_PARTITION_MASK_8x8_INTEL 0x5 749 #define CLK_AVC_ME_INTRA_LUMA_PARTITION_MASK_4x4_INTEL 0x3 750 751 #define CLK_AVC_ME_INTRA_NEIGHBOR_LEFT_MASK_ENABLE_INTEL 0x60 752 #define CLK_AVC_ME_INTRA_NEIGHBOR_UPPER_MASK_ENABLE_INTEL 0x10 753 #define CLK_AVC_ME_INTRA_NEIGHBOR_UPPER_RIGHT_MASK_ENABLE_INTEL 0x8 754 #define CLK_AVC_ME_INTRA_NEIGHBOR_UPPER_LEFT_MASK_ENABLE_INTEL 0x4 755 756 #define CLK_AVC_ME_LUMA_PREDICTOR_MODE_VERTICAL_INTEL 0x0 757 #define CLK_AVC_ME_LUMA_PREDICTOR_MODE_HORIZONTAL_INTEL 0x1 758 #define CLK_AVC_ME_LUMA_PREDICTOR_MODE_DC_INTEL 0x2 759 #define CLK_AVC_ME_LUMA_PREDICTOR_MODE_DIAGONAL_DOWN_LEFT_INTEL 0x3 760 #define CLK_AVC_ME_LUMA_PREDICTOR_MODE_DIAGONAL_DOWN_RIGHT_INTEL 0x4 761 #define CLK_AVC_ME_LUMA_PREDICTOR_MODE_PLANE_INTEL 0x4 762 #define CLK_AVC_ME_LUMA_PREDICTOR_MODE_VERTICAL_RIGHT_INTEL 0x5 763 #define CLK_AVC_ME_LUMA_PREDICTOR_MODE_HORIZONTAL_DOWN_INTEL 0x6 764 #define CLK_AVC_ME_LUMA_PREDICTOR_MODE_VERTICAL_LEFT_INTEL 0x7 765 #define CLK_AVC_ME_LUMA_PREDICTOR_MODE_HORIZONTAL_UP_INTEL 0x8 766 #define CLK_AVC_ME_CHROMA_PREDICTOR_MODE_DC_INTEL 0x0 767 #define CLK_AVC_ME_CHROMA_PREDICTOR_MODE_HORIZONTAL_INTEL 0x1 768 #define CLK_AVC_ME_CHROMA_PREDICTOR_MODE_VERTICAL_INTEL 0x2 769 #define CLK_AVC_ME_CHROMA_PREDICTOR_MODE_PLANE_INTEL 0x3 770 771 #define CLK_AVC_ME_FRAME_FORWARD_INTEL 0x1 772 #define CLK_AVC_ME_FRAME_BACKWARD_INTEL 0x2 773 #define CLK_AVC_ME_FRAME_DUAL_INTEL 0x3 774 775 #define CLK_AVC_ME_INTERLACED_SCAN_TOP_FIELD_INTEL 0x0 776 #define CLK_AVC_ME_INTERLACED_SCAN_BOTTOM_FIELD_INTEL 0x1 777 778 #define CLK_AVC_ME_INITIALIZE_INTEL 0x0 779 780 #define CLK_AVC_IME_PAYLOAD_INITIALIZE_INTEL 0x0 781 #define CLK_AVC_REF_PAYLOAD_INITIALIZE_INTEL 0x0 782 #define CLK_AVC_SIC_PAYLOAD_INITIALIZE_INTEL 0x0 783 784 #define CLK_AVC_IME_RESULT_INITIALIZE_INTEL 0x0 785 #define CLK_AVC_REF_RESULT_INITIALIZE_INTEL 0x0 786 #define CLK_AVC_SIC_RESULT_INITIALIZE_INTEL 0x0 787 788 #define CLK_AVC_IME_RESULT_SINGLE_REFERENCE_STREAMOUT_INITIALIZE_INTEL 0x0 789 #define CLK_AVC_IME_RESULT_SINGLE_REFERENCE_STREAMIN_INITIALIZE_INTEL 0x0 790 #define CLK_AVC_IME_RESULT_DUAL_REFERENCE_STREAMOUT_INITIALIZE_INTEL 0x0 791 #define CLK_AVC_IME_RESULT_DUAL_REFERENCE_STREAMIN_INITIALIZE_INTEL 0x0 792 793 #endif // cl_intel_device_side_avc_motion_estimation 794 795 // Disable any extensions we may have enabled previously. 796 #pragma OPENCL EXTENSION all : disable 797 798 #endif //_OPENCL_BASE_H_ 799