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