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