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 78 // Undefine any feature macros that have been explicitly disabled using 79 // an __undef_<feature> macro. 80 #ifdef __undef___opencl_c_work_group_collective_functions 81 #undef __opencl_c_work_group_collective_functions 82 #endif 83 #ifdef __undef___opencl_c_atomic_order_seq_cst 84 #undef __opencl_c_atomic_order_seq_cst 85 #endif 86 #ifdef __undef___opencl_c_atomic_scope_device 87 #undef __opencl_c_atomic_scope_device 88 #endif 89 #ifdef __undef___opencl_c_atomic_scope_all_devices 90 #undef __opencl_c_atomic_scope_all_devices 91 #endif 92 #ifdef __undef___opencl_c_read_write_images 93 #undef __opencl_c_read_write_images 94 #endif 95 96 #endif // (__OPENCL_CPP_VERSION__ == 202100 || __OPENCL_C_VERSION__ == 300) 97 98 #if !defined(__opencl_c_generic_address_space) 99 // Internal feature macro to provide named (global, local, private) address 100 // space overloads for builtin functions that take a pointer argument. 101 #define __opencl_c_named_address_space_builtins 1 102 #endif // !defined(__opencl_c_generic_address_space) 103 104 #if defined(cl_intel_subgroups) || defined(cl_khr_subgroups) || defined(__opencl_c_subgroups) 105 // Internal feature macro to provide subgroup builtins. 106 #define __opencl_subgroup_builtins 1 107 #endif 108 109 // built-in scalar data types: 110 111 /** 112 * An unsigned 8-bit integer. 113 */ 114 typedef unsigned char uchar; 115 116 /** 117 * An unsigned 16-bit integer. 118 */ 119 typedef unsigned short ushort; 120 121 /** 122 * An unsigned 32-bit integer. 123 */ 124 typedef unsigned int uint; 125 126 /** 127 * An unsigned 64-bit integer. 128 */ 129 typedef unsigned long ulong; 130 131 /** 132 * The unsigned integer type of the result of the sizeof operator. This 133 * is a 32-bit unsigned integer if CL_DEVICE_ADDRESS_BITS 134 * defined in table 4.3 is 32-bits and is a 64-bit unsigned integer if 135 * CL_DEVICE_ADDRESS_BITS is 64-bits. 136 */ 137 typedef __SIZE_TYPE__ size_t; 138 139 /** 140 * A signed integer type that is the result of subtracting two pointers. 141 * This is a 32-bit signed integer if CL_DEVICE_ADDRESS_BITS 142 * defined in table 4.3 is 32-bits and is a 64-bit signed integer if 143 * CL_DEVICE_ADDRESS_BITS is 64-bits. 144 */ 145 typedef __PTRDIFF_TYPE__ ptrdiff_t; 146 147 /** 148 * A signed integer type with the property that any valid pointer to 149 * void can be converted to this type, then converted back to pointer 150 * to void, and the result will compare equal to the original pointer. 151 */ 152 typedef __INTPTR_TYPE__ intptr_t; 153 154 /** 155 * An unsigned integer type with the property that any valid pointer to 156 * void can be converted to this type, then converted back to pointer 157 * to void, and the result will compare equal to the original pointer. 158 */ 159 typedef __UINTPTR_TYPE__ uintptr_t; 160 161 // built-in vector data types: 162 typedef char char2 __attribute__((ext_vector_type(2))); 163 typedef char char3 __attribute__((ext_vector_type(3))); 164 typedef char char4 __attribute__((ext_vector_type(4))); 165 typedef char char8 __attribute__((ext_vector_type(8))); 166 typedef char char16 __attribute__((ext_vector_type(16))); 167 typedef uchar uchar2 __attribute__((ext_vector_type(2))); 168 typedef uchar uchar3 __attribute__((ext_vector_type(3))); 169 typedef uchar uchar4 __attribute__((ext_vector_type(4))); 170 typedef uchar uchar8 __attribute__((ext_vector_type(8))); 171 typedef uchar uchar16 __attribute__((ext_vector_type(16))); 172 typedef short short2 __attribute__((ext_vector_type(2))); 173 typedef short short3 __attribute__((ext_vector_type(3))); 174 typedef short short4 __attribute__((ext_vector_type(4))); 175 typedef short short8 __attribute__((ext_vector_type(8))); 176 typedef short short16 __attribute__((ext_vector_type(16))); 177 typedef ushort ushort2 __attribute__((ext_vector_type(2))); 178 typedef ushort ushort3 __attribute__((ext_vector_type(3))); 179 typedef ushort ushort4 __attribute__((ext_vector_type(4))); 180 typedef ushort ushort8 __attribute__((ext_vector_type(8))); 181 typedef ushort ushort16 __attribute__((ext_vector_type(16))); 182 typedef int int2 __attribute__((ext_vector_type(2))); 183 typedef int int3 __attribute__((ext_vector_type(3))); 184 typedef int int4 __attribute__((ext_vector_type(4))); 185 typedef int int8 __attribute__((ext_vector_type(8))); 186 typedef int int16 __attribute__((ext_vector_type(16))); 187 typedef uint uint2 __attribute__((ext_vector_type(2))); 188 typedef uint uint3 __attribute__((ext_vector_type(3))); 189 typedef uint uint4 __attribute__((ext_vector_type(4))); 190 typedef uint uint8 __attribute__((ext_vector_type(8))); 191 typedef uint uint16 __attribute__((ext_vector_type(16))); 192 typedef long long2 __attribute__((ext_vector_type(2))); 193 typedef long long3 __attribute__((ext_vector_type(3))); 194 typedef long long4 __attribute__((ext_vector_type(4))); 195 typedef long long8 __attribute__((ext_vector_type(8))); 196 typedef long long16 __attribute__((ext_vector_type(16))); 197 typedef ulong ulong2 __attribute__((ext_vector_type(2))); 198 typedef ulong ulong3 __attribute__((ext_vector_type(3))); 199 typedef ulong ulong4 __attribute__((ext_vector_type(4))); 200 typedef ulong ulong8 __attribute__((ext_vector_type(8))); 201 typedef ulong ulong16 __attribute__((ext_vector_type(16))); 202 typedef float float2 __attribute__((ext_vector_type(2))); 203 typedef float float3 __attribute__((ext_vector_type(3))); 204 typedef float float4 __attribute__((ext_vector_type(4))); 205 typedef float float8 __attribute__((ext_vector_type(8))); 206 typedef float float16 __attribute__((ext_vector_type(16))); 207 #ifdef cl_khr_fp16 208 #pragma OPENCL EXTENSION cl_khr_fp16 : enable 209 typedef half half2 __attribute__((ext_vector_type(2))); 210 typedef half half3 __attribute__((ext_vector_type(3))); 211 typedef half half4 __attribute__((ext_vector_type(4))); 212 typedef half half8 __attribute__((ext_vector_type(8))); 213 typedef half half16 __attribute__((ext_vector_type(16))); 214 #endif 215 #ifdef cl_khr_fp64 216 #if __OPENCL_C_VERSION__ < CL_VERSION_1_2 217 #pragma OPENCL EXTENSION cl_khr_fp64 : enable 218 #endif 219 typedef double double2 __attribute__((ext_vector_type(2))); 220 typedef double double3 __attribute__((ext_vector_type(3))); 221 typedef double double4 __attribute__((ext_vector_type(4))); 222 typedef double double8 __attribute__((ext_vector_type(8))); 223 typedef double double16 __attribute__((ext_vector_type(16))); 224 #endif 225 226 // An internal alias for half, for use by OpenCLBuiltins.td. 227 #define __half half 228 229 #if defined(__OPENCL_CPP_VERSION__) 230 #define NULL nullptr 231 #elif defined(__OPENCL_C_VERSION__) 232 #define NULL ((void*)0) 233 #endif 234 235 /** 236 * Value of maximum non-infinite single-precision floating-point 237 * number. 238 */ 239 #define MAXFLOAT 0x1.fffffep127f 240 241 /** 242 * A positive float constant expression. HUGE_VALF evaluates 243 * to +infinity. Used as an error value returned by the built-in 244 * math functions. 245 */ 246 #define HUGE_VALF (__builtin_huge_valf()) 247 248 /** 249 * A positive double constant expression. HUGE_VAL evaluates 250 * to +infinity. Used as an error value returned by the built-in 251 * math functions. 252 */ 253 #define HUGE_VAL (__builtin_huge_val()) 254 255 /** 256 * A constant expression of type float representing positive or 257 * unsigned infinity. 258 */ 259 #define INFINITY (__builtin_inff()) 260 261 /** 262 * A constant expression of type float representing a quiet NaN. 263 */ 264 #define NAN as_float(INT_MAX) 265 266 #define FP_ILOGB0 INT_MIN 267 #define FP_ILOGBNAN INT_MAX 268 269 #define FLT_DIG 6 270 #define FLT_MANT_DIG 24 271 #define FLT_MAX_10_EXP +38 272 #define FLT_MAX_EXP +128 273 #define FLT_MIN_10_EXP -37 274 #define FLT_MIN_EXP -125 275 #define FLT_RADIX 2 276 #define FLT_MAX 0x1.fffffep127f 277 #define FLT_MIN 0x1.0p-126f 278 #define FLT_EPSILON 0x1.0p-23f 279 280 #define M_E_F 2.71828182845904523536028747135266250f 281 #define M_LOG2E_F 1.44269504088896340735992468100189214f 282 #define M_LOG10E_F 0.434294481903251827651128918916605082f 283 #define M_LN2_F 0.693147180559945309417232121458176568f 284 #define M_LN10_F 2.30258509299404568401799145468436421f 285 #define M_PI_F 3.14159265358979323846264338327950288f 286 #define M_PI_2_F 1.57079632679489661923132169163975144f 287 #define M_PI_4_F 0.785398163397448309615660845819875721f 288 #define M_1_PI_F 0.318309886183790671537767526745028724f 289 #define M_2_PI_F 0.636619772367581343075535053490057448f 290 #define M_2_SQRTPI_F 1.12837916709551257389615890312154517f 291 #define M_SQRT2_F 1.41421356237309504880168872420969808f 292 #define M_SQRT1_2_F 0.707106781186547524400844362104849039f 293 294 #define DBL_DIG 15 295 #define DBL_MANT_DIG 53 296 #define DBL_MAX_10_EXP +308 297 #define DBL_MAX_EXP +1024 298 #define DBL_MIN_10_EXP -307 299 #define DBL_MIN_EXP -1021 300 #define DBL_RADIX 2 301 #define DBL_MAX 0x1.fffffffffffffp1023 302 #define DBL_MIN 0x1.0p-1022 303 #define DBL_EPSILON 0x1.0p-52 304 305 #define M_E 0x1.5bf0a8b145769p+1 306 #define M_LOG2E 0x1.71547652b82fep+0 307 #define M_LOG10E 0x1.bcb7b1526e50ep-2 308 #define M_LN2 0x1.62e42fefa39efp-1 309 #define M_LN10 0x1.26bb1bbb55516p+1 310 #define M_PI 0x1.921fb54442d18p+1 311 #define M_PI_2 0x1.921fb54442d18p+0 312 #define M_PI_4 0x1.921fb54442d18p-1 313 #define M_1_PI 0x1.45f306dc9c883p-2 314 #define M_2_PI 0x1.45f306dc9c883p-1 315 #define M_2_SQRTPI 0x1.20dd750429b6dp+0 316 #define M_SQRT2 0x1.6a09e667f3bcdp+0 317 #define M_SQRT1_2 0x1.6a09e667f3bcdp-1 318 319 #ifdef cl_khr_fp16 320 321 #define HALF_DIG 3 322 #define HALF_MANT_DIG 11 323 #define HALF_MAX_10_EXP +4 324 #define HALF_MAX_EXP +16 325 #define HALF_MIN_10_EXP -4 326 #define HALF_MIN_EXP -13 327 #define HALF_RADIX 2 328 #define HALF_MAX ((0x1.ffcp15h)) 329 #define HALF_MIN ((0x1.0p-14h)) 330 #define HALF_EPSILON ((0x1.0p-10h)) 331 332 #define M_E_H 2.71828182845904523536028747135266250h 333 #define M_LOG2E_H 1.44269504088896340735992468100189214h 334 #define M_LOG10E_H 0.434294481903251827651128918916605082h 335 #define M_LN2_H 0.693147180559945309417232121458176568h 336 #define M_LN10_H 2.30258509299404568401799145468436421h 337 #define M_PI_H 3.14159265358979323846264338327950288h 338 #define M_PI_2_H 1.57079632679489661923132169163975144h 339 #define M_PI_4_H 0.785398163397448309615660845819875721h 340 #define M_1_PI_H 0.318309886183790671537767526745028724h 341 #define M_2_PI_H 0.636619772367581343075535053490057448h 342 #define M_2_SQRTPI_H 1.12837916709551257389615890312154517h 343 #define M_SQRT2_H 1.41421356237309504880168872420969808h 344 #define M_SQRT1_2_H 0.707106781186547524400844362104849039h 345 346 #endif //cl_khr_fp16 347 348 #define CHAR_BIT 8 349 #define SCHAR_MAX 127 350 #define SCHAR_MIN (-128) 351 #define UCHAR_MAX 255 352 #define CHAR_MAX SCHAR_MAX 353 #define CHAR_MIN SCHAR_MIN 354 #define USHRT_MAX 65535 355 #define SHRT_MAX 32767 356 #define SHRT_MIN (-32768) 357 #define UINT_MAX 0xffffffff 358 #define INT_MAX 2147483647 359 #define INT_MIN (-2147483647-1) 360 #define ULONG_MAX 0xffffffffffffffffUL 361 #define LONG_MAX 0x7fffffffffffffffL 362 #define LONG_MIN (-0x7fffffffffffffffL-1) 363 364 // OpenCL v1.1 s6.11.8, v1.2 s6.12.8, v2.0 s6.13.8 - Synchronization Functions 365 366 // Flag type and values for barrier, mem_fence, read_mem_fence, write_mem_fence 367 typedef uint cl_mem_fence_flags; 368 369 /** 370 * Queue a memory fence to ensure correct 371 * ordering of memory operations to local memory 372 */ 373 #define CLK_LOCAL_MEM_FENCE 0x01 374 375 /** 376 * Queue a memory fence to ensure correct 377 * ordering of memory operations to global memory 378 */ 379 #define CLK_GLOBAL_MEM_FENCE 0x02 380 381 #if defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0) 382 383 typedef enum memory_scope { 384 memory_scope_work_item = __OPENCL_MEMORY_SCOPE_WORK_ITEM, 385 memory_scope_work_group = __OPENCL_MEMORY_SCOPE_WORK_GROUP, 386 memory_scope_device = __OPENCL_MEMORY_SCOPE_DEVICE, 387 #if defined(__opencl_c_atomic_scope_all_devices) 388 memory_scope_all_svm_devices = __OPENCL_MEMORY_SCOPE_ALL_SVM_DEVICES, 389 #if (__OPENCL_C_VERSION__ >= CL_VERSION_3_0 || __OPENCL_CPP_VERSION__ >= 202100) 390 memory_scope_all_devices = memory_scope_all_svm_devices, 391 #endif // (__OPENCL_C_VERSION__ >= CL_VERSION_3_0 || __OPENCL_CPP_VERSION__ >= 202100) 392 #endif // defined(__opencl_c_atomic_scope_all_devices) 393 /** 394 * Subgroups have different requirements on forward progress, so just test 395 * all the relevant macros. 396 * CL 3.0 sub-groups "they are not guaranteed to make independent forward progress" 397 * KHR subgroups "Subgroups within a workgroup are independent, make forward progress with respect to each other" 398 */ 399 #if defined(cl_intel_subgroups) || defined(cl_khr_subgroups) || defined(__opencl_c_subgroups) 400 memory_scope_sub_group = __OPENCL_MEMORY_SCOPE_SUB_GROUP 401 #endif 402 } memory_scope; 403 404 /** 405 * Queue a memory fence to ensure correct ordering of memory 406 * operations between work-items of a work-group to 407 * image memory. 408 */ 409 #define CLK_IMAGE_MEM_FENCE 0x04 410 411 #ifndef ATOMIC_VAR_INIT 412 #define ATOMIC_VAR_INIT(x) (x) 413 #endif //ATOMIC_VAR_INIT 414 #define ATOMIC_FLAG_INIT 0 415 416 // enum values aligned with what clang uses in EmitAtomicExpr() 417 typedef enum memory_order 418 { 419 memory_order_relaxed = __ATOMIC_RELAXED, 420 memory_order_acquire = __ATOMIC_ACQUIRE, 421 memory_order_release = __ATOMIC_RELEASE, 422 memory_order_acq_rel = __ATOMIC_ACQ_REL, 423 #if defined(__opencl_c_atomic_order_seq_cst) 424 memory_order_seq_cst = __ATOMIC_SEQ_CST 425 #endif 426 } memory_order; 427 428 #endif // defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0) 429 430 // OpenCL v1.1 s6.11.3, v1.2 s6.12.14, v2.0 s6.13.14 - Image Read and Write Functions 431 432 // These values need to match the runtime equivalent 433 // 434 // Addressing Mode. 435 // 436 #define CLK_ADDRESS_NONE 0 437 #define CLK_ADDRESS_CLAMP_TO_EDGE 2 438 #define CLK_ADDRESS_CLAMP 4 439 #define CLK_ADDRESS_REPEAT 6 440 #define CLK_ADDRESS_MIRRORED_REPEAT 8 441 442 // 443 // Coordination Normalization 444 // 445 #define CLK_NORMALIZED_COORDS_FALSE 0 446 #define CLK_NORMALIZED_COORDS_TRUE 1 447 448 // 449 // Filtering Mode. 450 // 451 #define CLK_FILTER_NEAREST 0x10 452 #define CLK_FILTER_LINEAR 0x20 453 454 #ifdef cl_khr_gl_msaa_sharing 455 #pragma OPENCL EXTENSION cl_khr_gl_msaa_sharing : enable 456 #endif //cl_khr_gl_msaa_sharing 457 458 // 459 // Channel Datatype. 460 // 461 #define CLK_SNORM_INT8 0x10D0 462 #define CLK_SNORM_INT16 0x10D1 463 #define CLK_UNORM_INT8 0x10D2 464 #define CLK_UNORM_INT16 0x10D3 465 #define CLK_UNORM_SHORT_565 0x10D4 466 #define CLK_UNORM_SHORT_555 0x10D5 467 #define CLK_UNORM_INT_101010 0x10D6 468 #define CLK_SIGNED_INT8 0x10D7 469 #define CLK_SIGNED_INT16 0x10D8 470 #define CLK_SIGNED_INT32 0x10D9 471 #define CLK_UNSIGNED_INT8 0x10DA 472 #define CLK_UNSIGNED_INT16 0x10DB 473 #define CLK_UNSIGNED_INT32 0x10DC 474 #define CLK_HALF_FLOAT 0x10DD 475 #define CLK_FLOAT 0x10DE 476 #define CLK_UNORM_INT24 0x10DF 477 478 // Channel order, numbering must be aligned with cl_channel_order in cl.h 479 // 480 #define CLK_R 0x10B0 481 #define CLK_A 0x10B1 482 #define CLK_RG 0x10B2 483 #define CLK_RA 0x10B3 484 #define CLK_RGB 0x10B4 485 #define CLK_RGBA 0x10B5 486 #define CLK_BGRA 0x10B6 487 #define CLK_ARGB 0x10B7 488 #define CLK_INTENSITY 0x10B8 489 #define CLK_LUMINANCE 0x10B9 490 #define CLK_Rx 0x10BA 491 #define CLK_RGx 0x10BB 492 #define CLK_RGBx 0x10BC 493 #define CLK_DEPTH 0x10BD 494 #define CLK_DEPTH_STENCIL 0x10BE 495 #if __OPENCL_C_VERSION__ >= CL_VERSION_2_0 496 #define CLK_sRGB 0x10BF 497 #define CLK_sRGBx 0x10C0 498 #define CLK_sRGBA 0x10C1 499 #define CLK_sBGRA 0x10C2 500 #define CLK_ABGR 0x10C3 501 #endif //__OPENCL_C_VERSION__ >= CL_VERSION_2_0 502 503 // OpenCL v2.0 s6.13.16 - Pipe Functions 504 #if defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0) 505 #define CLK_NULL_RESERVE_ID (__builtin_astype(((void*)(__SIZE_MAX__)), reserve_id_t)) 506 507 // OpenCL v2.0 s6.13.17 - Enqueue Kernels 508 #define CL_COMPLETE 0x0 509 #define CL_RUNNING 0x1 510 #define CL_SUBMITTED 0x2 511 #define CL_QUEUED 0x3 512 513 #define CLK_SUCCESS 0 514 #define CLK_ENQUEUE_FAILURE -101 515 #define CLK_INVALID_QUEUE -102 516 #define CLK_INVALID_NDRANGE -160 517 #define CLK_INVALID_EVENT_WAIT_LIST -57 518 #define CLK_DEVICE_QUEUE_FULL -161 519 #define CLK_INVALID_ARG_SIZE -51 520 #define CLK_EVENT_ALLOCATION_FAILURE -100 521 #define CLK_OUT_OF_RESOURCES -5 522 523 #define CLK_NULL_QUEUE 0 524 #define CLK_NULL_EVENT (__builtin_astype(((__SIZE_MAX__)), clk_event_t)) 525 526 // execution model related definitions 527 #define CLK_ENQUEUE_FLAGS_NO_WAIT 0x0 528 #define CLK_ENQUEUE_FLAGS_WAIT_KERNEL 0x1 529 #define CLK_ENQUEUE_FLAGS_WAIT_WORK_GROUP 0x2 530 531 typedef int kernel_enqueue_flags_t; 532 typedef int clk_profiling_info; 533 534 // Profiling info name (see capture_event_profiling_info) 535 #define CLK_PROFILING_COMMAND_EXEC_TIME 0x1 536 537 #define MAX_WORK_DIM 3 538 539 #ifdef __opencl_c_device_enqueue 540 typedef struct { 541 unsigned int workDimension; 542 size_t globalWorkOffset[MAX_WORK_DIM]; 543 size_t globalWorkSize[MAX_WORK_DIM]; 544 size_t localWorkSize[MAX_WORK_DIM]; 545 } ndrange_t; 546 #endif // __opencl_c_device_enqueue 547 548 #endif // defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0) 549 550 /** 551 * OpenCL v1.1/1.2/2.0 s6.2.4.2 - as_type operators 552 * Reinterprets a data type as another data type of the same size 553 */ 554 #define as_char(x) __builtin_astype((x), char) 555 #define as_char2(x) __builtin_astype((x), char2) 556 #define as_char3(x) __builtin_astype((x), char3) 557 #define as_char4(x) __builtin_astype((x), char4) 558 #define as_char8(x) __builtin_astype((x), char8) 559 #define as_char16(x) __builtin_astype((x), char16) 560 561 #define as_uchar(x) __builtin_astype((x), uchar) 562 #define as_uchar2(x) __builtin_astype((x), uchar2) 563 #define as_uchar3(x) __builtin_astype((x), uchar3) 564 #define as_uchar4(x) __builtin_astype((x), uchar4) 565 #define as_uchar8(x) __builtin_astype((x), uchar8) 566 #define as_uchar16(x) __builtin_astype((x), uchar16) 567 568 #define as_short(x) __builtin_astype((x), short) 569 #define as_short2(x) __builtin_astype((x), short2) 570 #define as_short3(x) __builtin_astype((x), short3) 571 #define as_short4(x) __builtin_astype((x), short4) 572 #define as_short8(x) __builtin_astype((x), short8) 573 #define as_short16(x) __builtin_astype((x), short16) 574 575 #define as_ushort(x) __builtin_astype((x), ushort) 576 #define as_ushort2(x) __builtin_astype((x), ushort2) 577 #define as_ushort3(x) __builtin_astype((x), ushort3) 578 #define as_ushort4(x) __builtin_astype((x), ushort4) 579 #define as_ushort8(x) __builtin_astype((x), ushort8) 580 #define as_ushort16(x) __builtin_astype((x), ushort16) 581 582 #define as_int(x) __builtin_astype((x), int) 583 #define as_int2(x) __builtin_astype((x), int2) 584 #define as_int3(x) __builtin_astype((x), int3) 585 #define as_int4(x) __builtin_astype((x), int4) 586 #define as_int8(x) __builtin_astype((x), int8) 587 #define as_int16(x) __builtin_astype((x), int16) 588 589 #define as_uint(x) __builtin_astype((x), uint) 590 #define as_uint2(x) __builtin_astype((x), uint2) 591 #define as_uint3(x) __builtin_astype((x), uint3) 592 #define as_uint4(x) __builtin_astype((x), uint4) 593 #define as_uint8(x) __builtin_astype((x), uint8) 594 #define as_uint16(x) __builtin_astype((x), uint16) 595 596 #define as_long(x) __builtin_astype((x), long) 597 #define as_long2(x) __builtin_astype((x), long2) 598 #define as_long3(x) __builtin_astype((x), long3) 599 #define as_long4(x) __builtin_astype((x), long4) 600 #define as_long8(x) __builtin_astype((x), long8) 601 #define as_long16(x) __builtin_astype((x), long16) 602 603 #define as_ulong(x) __builtin_astype((x), ulong) 604 #define as_ulong2(x) __builtin_astype((x), ulong2) 605 #define as_ulong3(x) __builtin_astype((x), ulong3) 606 #define as_ulong4(x) __builtin_astype((x), ulong4) 607 #define as_ulong8(x) __builtin_astype((x), ulong8) 608 #define as_ulong16(x) __builtin_astype((x), ulong16) 609 610 #define as_float(x) __builtin_astype((x), float) 611 #define as_float2(x) __builtin_astype((x), float2) 612 #define as_float3(x) __builtin_astype((x), float3) 613 #define as_float4(x) __builtin_astype((x), float4) 614 #define as_float8(x) __builtin_astype((x), float8) 615 #define as_float16(x) __builtin_astype((x), float16) 616 617 #ifdef cl_khr_fp64 618 #define as_double(x) __builtin_astype((x), double) 619 #define as_double2(x) __builtin_astype((x), double2) 620 #define as_double3(x) __builtin_astype((x), double3) 621 #define as_double4(x) __builtin_astype((x), double4) 622 #define as_double8(x) __builtin_astype((x), double8) 623 #define as_double16(x) __builtin_astype((x), double16) 624 #endif // cl_khr_fp64 625 626 #ifdef cl_khr_fp16 627 #define as_half(x) __builtin_astype((x), half) 628 #define as_half2(x) __builtin_astype((x), half2) 629 #define as_half3(x) __builtin_astype((x), half3) 630 #define as_half4(x) __builtin_astype((x), half4) 631 #define as_half8(x) __builtin_astype((x), half8) 632 #define as_half16(x) __builtin_astype((x), half16) 633 #endif // cl_khr_fp16 634 635 #define as_size_t(x) __builtin_astype((x), size_t) 636 #define as_ptrdiff_t(x) __builtin_astype((x), ptrdiff_t) 637 #define as_intptr_t(x) __builtin_astype((x), intptr_t) 638 #define as_uintptr_t(x) __builtin_astype((x), uintptr_t) 639 640 // C++ for OpenCL - __remove_address_space 641 #if defined(__OPENCL_CPP_VERSION__) 642 template <typename _Tp> struct __remove_address_space { using type = _Tp; }; 643 #if defined(__opencl_c_generic_address_space) 644 template <typename _Tp> struct __remove_address_space<__generic _Tp> { 645 using type = _Tp; 646 }; 647 #endif 648 template <typename _Tp> struct __remove_address_space<__global _Tp> { 649 using type = _Tp; 650 }; 651 template <typename _Tp> struct __remove_address_space<__private _Tp> { 652 using type = _Tp; 653 }; 654 template <typename _Tp> struct __remove_address_space<__local _Tp> { 655 using type = _Tp; 656 }; 657 template <typename _Tp> struct __remove_address_space<__constant _Tp> { 658 using type = _Tp; 659 }; 660 #endif 661 662 // OpenCL v1.1 s6.9, v1.2/2.0 s6.10 - Function qualifiers 663 664 #define __kernel_exec(X, typen) __kernel \ 665 __attribute__((work_group_size_hint(X, 1, 1))) \ 666 __attribute__((vec_type_hint(typen))) 667 668 #define kernel_exec(X, typen) __kernel \ 669 __attribute__((work_group_size_hint(X, 1, 1))) \ 670 __attribute__((vec_type_hint(typen))) 671 672 #if defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_1_2) 673 // OpenCL v1.2 s6.12.13, v2.0 s6.13.13 - printf 674 675 int printf(__constant const char* st, ...) __attribute__((format(printf, 1, 2))); 676 #endif 677 678 #ifdef cl_intel_device_side_avc_motion_estimation 679 680 #define CLK_AVC_ME_MAJOR_16x16_INTEL 0x0 681 #define CLK_AVC_ME_MAJOR_16x8_INTEL 0x1 682 #define CLK_AVC_ME_MAJOR_8x16_INTEL 0x2 683 #define CLK_AVC_ME_MAJOR_8x8_INTEL 0x3 684 685 #define CLK_AVC_ME_MINOR_8x8_INTEL 0x0 686 #define CLK_AVC_ME_MINOR_8x4_INTEL 0x1 687 #define CLK_AVC_ME_MINOR_4x8_INTEL 0x2 688 #define CLK_AVC_ME_MINOR_4x4_INTEL 0x3 689 690 #define CLK_AVC_ME_MAJOR_FORWARD_INTEL 0x0 691 #define CLK_AVC_ME_MAJOR_BACKWARD_INTEL 0x1 692 #define CLK_AVC_ME_MAJOR_BIDIRECTIONAL_INTEL 0x2 693 694 #define CLK_AVC_ME_PARTITION_MASK_ALL_INTEL 0x0 695 #define CLK_AVC_ME_PARTITION_MASK_16x16_INTEL 0x7E 696 #define CLK_AVC_ME_PARTITION_MASK_16x8_INTEL 0x7D 697 #define CLK_AVC_ME_PARTITION_MASK_8x16_INTEL 0x7B 698 #define CLK_AVC_ME_PARTITION_MASK_8x8_INTEL 0x77 699 #define CLK_AVC_ME_PARTITION_MASK_8x4_INTEL 0x6F 700 #define CLK_AVC_ME_PARTITION_MASK_4x8_INTEL 0x5F 701 #define CLK_AVC_ME_PARTITION_MASK_4x4_INTEL 0x3F 702 703 #define CLK_AVC_ME_SLICE_TYPE_PRED_INTEL 0x0 704 #define CLK_AVC_ME_SLICE_TYPE_BPRED_INTEL 0x1 705 #define CLK_AVC_ME_SLICE_TYPE_INTRA_INTEL 0x2 706 707 #define CLK_AVC_ME_SEARCH_WINDOW_EXHAUSTIVE_INTEL 0x0 708 #define CLK_AVC_ME_SEARCH_WINDOW_SMALL_INTEL 0x1 709 #define CLK_AVC_ME_SEARCH_WINDOW_TINY_INTEL 0x2 710 #define CLK_AVC_ME_SEARCH_WINDOW_EXTRA_TINY_INTEL 0x3 711 #define CLK_AVC_ME_SEARCH_WINDOW_DIAMOND_INTEL 0x4 712 #define CLK_AVC_ME_SEARCH_WINDOW_LARGE_DIAMOND_INTEL 0x5 713 #define CLK_AVC_ME_SEARCH_WINDOW_RESERVED0_INTEL 0x6 714 #define CLK_AVC_ME_SEARCH_WINDOW_RESERVED1_INTEL 0x7 715 #define CLK_AVC_ME_SEARCH_WINDOW_CUSTOM_INTEL 0x8 716 717 #define CLK_AVC_ME_SAD_ADJUST_MODE_NONE_INTEL 0x0 718 #define CLK_AVC_ME_SAD_ADJUST_MODE_HAAR_INTEL 0x2 719 720 #define CLK_AVC_ME_SUBPIXEL_MODE_INTEGER_INTEL 0x0 721 #define CLK_AVC_ME_SUBPIXEL_MODE_HPEL_INTEL 0x1 722 #define CLK_AVC_ME_SUBPIXEL_MODE_QPEL_INTEL 0x3 723 724 #define CLK_AVC_ME_COST_PRECISION_QPEL_INTEL 0x0 725 #define CLK_AVC_ME_COST_PRECISION_HPEL_INTEL 0x1 726 #define CLK_AVC_ME_COST_PRECISION_PEL_INTEL 0x2 727 #define CLK_AVC_ME_COST_PRECISION_DPEL_INTEL 0x3 728 729 #define CLK_AVC_ME_BIDIR_WEIGHT_QUARTER_INTEL 0x10 730 #define CLK_AVC_ME_BIDIR_WEIGHT_THIRD_INTEL 0x15 731 #define CLK_AVC_ME_BIDIR_WEIGHT_HALF_INTEL 0x20 732 #define CLK_AVC_ME_BIDIR_WEIGHT_TWO_THIRD_INTEL 0x2B 733 #define CLK_AVC_ME_BIDIR_WEIGHT_THREE_QUARTER_INTEL 0x30 734 735 #define CLK_AVC_ME_BORDER_REACHED_LEFT_INTEL 0x0 736 #define CLK_AVC_ME_BORDER_REACHED_RIGHT_INTEL 0x2 737 #define CLK_AVC_ME_BORDER_REACHED_TOP_INTEL 0x4 738 #define CLK_AVC_ME_BORDER_REACHED_BOTTOM_INTEL 0x8 739 740 #define CLK_AVC_ME_INTRA_16x16_INTEL 0x0 741 #define CLK_AVC_ME_INTRA_8x8_INTEL 0x1 742 #define CLK_AVC_ME_INTRA_4x4_INTEL 0x2 743 744 #define CLK_AVC_ME_SKIP_BLOCK_PARTITION_16x16_INTEL 0x0 745 #define CLK_AVC_ME_SKIP_BLOCK_PARTITION_8x8_INTEL 0x4000 746 747 #define CLK_AVC_ME_SKIP_BLOCK_16x16_FORWARD_ENABLE_INTEL (0x1 << 24) 748 #define CLK_AVC_ME_SKIP_BLOCK_16x16_BACKWARD_ENABLE_INTEL (0x2 << 24) 749 #define CLK_AVC_ME_SKIP_BLOCK_16x16_DUAL_ENABLE_INTEL (0x3 << 24) 750 #define CLK_AVC_ME_SKIP_BLOCK_8x8_FORWARD_ENABLE_INTEL (0x55 << 24) 751 #define CLK_AVC_ME_SKIP_BLOCK_8x8_BACKWARD_ENABLE_INTEL (0xAA << 24) 752 #define CLK_AVC_ME_SKIP_BLOCK_8x8_DUAL_ENABLE_INTEL (0xFF << 24) 753 #define CLK_AVC_ME_SKIP_BLOCK_8x8_0_FORWARD_ENABLE_INTEL (0x1 << 24) 754 #define CLK_AVC_ME_SKIP_BLOCK_8x8_0_BACKWARD_ENABLE_INTEL (0x2 << 24) 755 #define CLK_AVC_ME_SKIP_BLOCK_8x8_1_FORWARD_ENABLE_INTEL (0x1 << 26) 756 #define CLK_AVC_ME_SKIP_BLOCK_8x8_1_BACKWARD_ENABLE_INTEL (0x2 << 26) 757 #define CLK_AVC_ME_SKIP_BLOCK_8x8_2_FORWARD_ENABLE_INTEL (0x1 << 28) 758 #define CLK_AVC_ME_SKIP_BLOCK_8x8_2_BACKWARD_ENABLE_INTEL (0x2 << 28) 759 #define CLK_AVC_ME_SKIP_BLOCK_8x8_3_FORWARD_ENABLE_INTEL (0x1 << 30) 760 #define CLK_AVC_ME_SKIP_BLOCK_8x8_3_BACKWARD_ENABLE_INTEL (0x2 << 30) 761 762 #define CLK_AVC_ME_BLOCK_BASED_SKIP_4x4_INTEL 0x00 763 #define CLK_AVC_ME_BLOCK_BASED_SKIP_8x8_INTEL 0x80 764 765 #define CLK_AVC_ME_INTRA_LUMA_PARTITION_MASK_ALL_INTEL 0x0 766 #define CLK_AVC_ME_INTRA_LUMA_PARTITION_MASK_16x16_INTEL 0x6 767 #define CLK_AVC_ME_INTRA_LUMA_PARTITION_MASK_8x8_INTEL 0x5 768 #define CLK_AVC_ME_INTRA_LUMA_PARTITION_MASK_4x4_INTEL 0x3 769 770 #define CLK_AVC_ME_INTRA_NEIGHBOR_LEFT_MASK_ENABLE_INTEL 0x60 771 #define CLK_AVC_ME_INTRA_NEIGHBOR_UPPER_MASK_ENABLE_INTEL 0x10 772 #define CLK_AVC_ME_INTRA_NEIGHBOR_UPPER_RIGHT_MASK_ENABLE_INTEL 0x8 773 #define CLK_AVC_ME_INTRA_NEIGHBOR_UPPER_LEFT_MASK_ENABLE_INTEL 0x4 774 775 #define CLK_AVC_ME_LUMA_PREDICTOR_MODE_VERTICAL_INTEL 0x0 776 #define CLK_AVC_ME_LUMA_PREDICTOR_MODE_HORIZONTAL_INTEL 0x1 777 #define CLK_AVC_ME_LUMA_PREDICTOR_MODE_DC_INTEL 0x2 778 #define CLK_AVC_ME_LUMA_PREDICTOR_MODE_DIAGONAL_DOWN_LEFT_INTEL 0x3 779 #define CLK_AVC_ME_LUMA_PREDICTOR_MODE_DIAGONAL_DOWN_RIGHT_INTEL 0x4 780 #define CLK_AVC_ME_LUMA_PREDICTOR_MODE_PLANE_INTEL 0x4 781 #define CLK_AVC_ME_LUMA_PREDICTOR_MODE_VERTICAL_RIGHT_INTEL 0x5 782 #define CLK_AVC_ME_LUMA_PREDICTOR_MODE_HORIZONTAL_DOWN_INTEL 0x6 783 #define CLK_AVC_ME_LUMA_PREDICTOR_MODE_VERTICAL_LEFT_INTEL 0x7 784 #define CLK_AVC_ME_LUMA_PREDICTOR_MODE_HORIZONTAL_UP_INTEL 0x8 785 #define CLK_AVC_ME_CHROMA_PREDICTOR_MODE_DC_INTEL 0x0 786 #define CLK_AVC_ME_CHROMA_PREDICTOR_MODE_HORIZONTAL_INTEL 0x1 787 #define CLK_AVC_ME_CHROMA_PREDICTOR_MODE_VERTICAL_INTEL 0x2 788 #define CLK_AVC_ME_CHROMA_PREDICTOR_MODE_PLANE_INTEL 0x3 789 790 #define CLK_AVC_ME_FRAME_FORWARD_INTEL 0x1 791 #define CLK_AVC_ME_FRAME_BACKWARD_INTEL 0x2 792 #define CLK_AVC_ME_FRAME_DUAL_INTEL 0x3 793 794 #define CLK_AVC_ME_INTERLACED_SCAN_TOP_FIELD_INTEL 0x0 795 #define CLK_AVC_ME_INTERLACED_SCAN_BOTTOM_FIELD_INTEL 0x1 796 797 #define CLK_AVC_ME_INITIALIZE_INTEL 0x0 798 799 #define CLK_AVC_IME_PAYLOAD_INITIALIZE_INTEL 0x0 800 #define CLK_AVC_REF_PAYLOAD_INITIALIZE_INTEL 0x0 801 #define CLK_AVC_SIC_PAYLOAD_INITIALIZE_INTEL 0x0 802 803 #define CLK_AVC_IME_RESULT_INITIALIZE_INTEL 0x0 804 #define CLK_AVC_REF_RESULT_INITIALIZE_INTEL 0x0 805 #define CLK_AVC_SIC_RESULT_INITIALIZE_INTEL 0x0 806 807 #define CLK_AVC_IME_RESULT_SINGLE_REFERENCE_STREAMOUT_INITIALIZE_INTEL 0x0 808 #define CLK_AVC_IME_RESULT_SINGLE_REFERENCE_STREAMIN_INITIALIZE_INTEL 0x0 809 #define CLK_AVC_IME_RESULT_DUAL_REFERENCE_STREAMOUT_INITIALIZE_INTEL 0x0 810 #define CLK_AVC_IME_RESULT_DUAL_REFERENCE_STREAMIN_INITIALIZE_INTEL 0x0 811 812 #endif // cl_intel_device_side_avc_motion_estimation 813 814 // Disable any extensions we may have enabled previously. 815 #pragma OPENCL EXTENSION all : disable 816 817 #endif //_OPENCL_BASE_H_ 818