1 /* 2 * kmp_os.h -- KPTS runtime header file. 3 */ 4 5 //===----------------------------------------------------------------------===// 6 // 7 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 8 // See https://llvm.org/LICENSE.txt for license information. 9 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 10 // 11 //===----------------------------------------------------------------------===// 12 13 #ifndef KMP_OS_H 14 #define KMP_OS_H 15 16 #include "kmp_config.h" 17 #include <atomic> 18 #include <stdarg.h> 19 #include <stdlib.h> 20 21 #define KMP_FTN_PLAIN 1 22 #define KMP_FTN_APPEND 2 23 #define KMP_FTN_UPPER 3 24 /* 25 #define KMP_FTN_PREPEND 4 26 #define KMP_FTN_UAPPEND 5 27 */ 28 29 #define KMP_PTR_SKIP (sizeof(void *)) 30 31 /* -------------------------- Compiler variations ------------------------ */ 32 33 #define KMP_OFF 0 34 #define KMP_ON 1 35 36 #define KMP_MEM_CONS_VOLATILE 0 37 #define KMP_MEM_CONS_FENCE 1 38 39 #ifndef KMP_MEM_CONS_MODEL 40 #define KMP_MEM_CONS_MODEL KMP_MEM_CONS_VOLATILE 41 #endif 42 43 #ifndef __has_cpp_attribute 44 #define __has_cpp_attribute(x) 0 45 #endif 46 47 #ifndef __has_attribute 48 #define __has_attribute(x) 0 49 #endif 50 51 /* ------------------------- Compiler recognition ---------------------- */ 52 #define KMP_COMPILER_ICC 0 53 #define KMP_COMPILER_GCC 0 54 #define KMP_COMPILER_CLANG 0 55 #define KMP_COMPILER_MSVC 0 56 57 #if defined(__INTEL_COMPILER) 58 #undef KMP_COMPILER_ICC 59 #define KMP_COMPILER_ICC 1 60 #elif defined(__clang__) 61 #undef KMP_COMPILER_CLANG 62 #define KMP_COMPILER_CLANG 1 63 #elif defined(__GNUC__) 64 #undef KMP_COMPILER_GCC 65 #define KMP_COMPILER_GCC 1 66 #elif defined(_MSC_VER) 67 #undef KMP_COMPILER_MSVC 68 #define KMP_COMPILER_MSVC 1 69 #else 70 #error Unknown compiler 71 #endif 72 73 #if (KMP_OS_LINUX || KMP_OS_WINDOWS || KMP_OS_FREEBSD) && !KMP_OS_CNK 74 #define KMP_AFFINITY_SUPPORTED 1 75 #if KMP_OS_WINDOWS && KMP_ARCH_X86_64 76 #define KMP_GROUP_AFFINITY 1 77 #else 78 #define KMP_GROUP_AFFINITY 0 79 #endif 80 #else 81 #define KMP_AFFINITY_SUPPORTED 0 82 #define KMP_GROUP_AFFINITY 0 83 #endif 84 85 /* Check for quad-precision extension. */ 86 #define KMP_HAVE_QUAD 0 87 #if KMP_ARCH_X86 || KMP_ARCH_X86_64 88 #if KMP_COMPILER_ICC 89 /* _Quad is already defined for icc */ 90 #undef KMP_HAVE_QUAD 91 #define KMP_HAVE_QUAD 1 92 #elif KMP_COMPILER_CLANG 93 /* Clang doesn't support a software-implemented 94 128-bit extended precision type yet */ 95 typedef long double _Quad; 96 #elif KMP_COMPILER_GCC 97 /* GCC on NetBSD lacks __multc3/__divtc3 builtins needed for quad */ 98 #if !KMP_OS_NETBSD 99 typedef __float128 _Quad; 100 #undef KMP_HAVE_QUAD 101 #define KMP_HAVE_QUAD 1 102 #endif 103 #elif KMP_COMPILER_MSVC 104 typedef long double _Quad; 105 #endif 106 #else 107 #if __LDBL_MAX_EXP__ >= 16384 && KMP_COMPILER_GCC 108 typedef long double _Quad; 109 #undef KMP_HAVE_QUAD 110 #define KMP_HAVE_QUAD 1 111 #endif 112 #endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */ 113 114 #define KMP_USE_X87CONTROL 0 115 #if KMP_OS_WINDOWS 116 #define KMP_END_OF_LINE "\r\n" 117 typedef char kmp_int8; 118 typedef unsigned char kmp_uint8; 119 typedef short kmp_int16; 120 typedef unsigned short kmp_uint16; 121 typedef int kmp_int32; 122 typedef unsigned int kmp_uint32; 123 #define KMP_INT32_SPEC "d" 124 #define KMP_UINT32_SPEC "u" 125 #ifndef KMP_STRUCT64 126 typedef __int64 kmp_int64; 127 typedef unsigned __int64 kmp_uint64; 128 #define KMP_INT64_SPEC "I64d" 129 #define KMP_UINT64_SPEC "I64u" 130 #else 131 struct kmp_struct64 { 132 kmp_int32 a, b; 133 }; 134 typedef struct kmp_struct64 kmp_int64; 135 typedef struct kmp_struct64 kmp_uint64; 136 /* Not sure what to use for KMP_[U]INT64_SPEC here */ 137 #endif 138 #if KMP_ARCH_X86 && KMP_MSVC_COMPAT 139 #undef KMP_USE_X87CONTROL 140 #define KMP_USE_X87CONTROL 1 141 #endif 142 #if KMP_ARCH_X86_64 143 #define KMP_INTPTR 1 144 typedef __int64 kmp_intptr_t; 145 typedef unsigned __int64 kmp_uintptr_t; 146 #define KMP_INTPTR_SPEC "I64d" 147 #define KMP_UINTPTR_SPEC "I64u" 148 #endif 149 #endif /* KMP_OS_WINDOWS */ 150 151 #if KMP_OS_UNIX 152 #define KMP_END_OF_LINE "\n" 153 typedef char kmp_int8; 154 typedef unsigned char kmp_uint8; 155 typedef short kmp_int16; 156 typedef unsigned short kmp_uint16; 157 typedef int kmp_int32; 158 typedef unsigned int kmp_uint32; 159 typedef long long kmp_int64; 160 typedef unsigned long long kmp_uint64; 161 #define KMP_INT32_SPEC "d" 162 #define KMP_UINT32_SPEC "u" 163 #define KMP_INT64_SPEC "lld" 164 #define KMP_UINT64_SPEC "llu" 165 #endif /* KMP_OS_UNIX */ 166 167 #if KMP_ARCH_X86 || KMP_ARCH_ARM || KMP_ARCH_MIPS 168 #define KMP_SIZE_T_SPEC KMP_UINT32_SPEC 169 #elif KMP_ARCH_X86_64 || KMP_ARCH_PPC64 || KMP_ARCH_AARCH64 || \ 170 KMP_ARCH_MIPS64 || KMP_ARCH_RISCV64 171 #define KMP_SIZE_T_SPEC KMP_UINT64_SPEC 172 #else 173 #error "Can't determine size_t printf format specifier." 174 #endif 175 176 #if KMP_ARCH_X86 177 #define KMP_SIZE_T_MAX (0xFFFFFFFF) 178 #else 179 #define KMP_SIZE_T_MAX (0xFFFFFFFFFFFFFFFF) 180 #endif 181 182 typedef size_t kmp_size_t; 183 typedef float kmp_real32; 184 typedef double kmp_real64; 185 186 #ifndef KMP_INTPTR 187 #define KMP_INTPTR 1 188 typedef long kmp_intptr_t; 189 typedef unsigned long kmp_uintptr_t; 190 #define KMP_INTPTR_SPEC "ld" 191 #define KMP_UINTPTR_SPEC "lu" 192 #endif 193 194 #ifdef BUILD_I8 195 typedef kmp_int64 kmp_int; 196 typedef kmp_uint64 kmp_uint; 197 #else 198 typedef kmp_int32 kmp_int; 199 typedef kmp_uint32 kmp_uint; 200 #endif /* BUILD_I8 */ 201 #define KMP_INT_MAX ((kmp_int32)0x7FFFFFFF) 202 #define KMP_INT_MIN ((kmp_int32)0x80000000) 203 204 // stdarg handling 205 #if (KMP_ARCH_ARM || KMP_ARCH_X86_64 || KMP_ARCH_AARCH64) && \ 206 (KMP_OS_FREEBSD || KMP_OS_LINUX) 207 typedef va_list *kmp_va_list; 208 #define kmp_va_deref(ap) (*(ap)) 209 #define kmp_va_addr_of(ap) (&(ap)) 210 #else 211 typedef va_list kmp_va_list; 212 #define kmp_va_deref(ap) (ap) 213 #define kmp_va_addr_of(ap) (ap) 214 #endif 215 216 #ifdef __cplusplus 217 // macros to cast out qualifiers and to re-interpret types 218 #define CCAST(type, var) const_cast<type>(var) 219 #define RCAST(type, var) reinterpret_cast<type>(var) 220 //------------------------------------------------------------------------- 221 // template for debug prints specification ( d, u, lld, llu ), and to obtain 222 // signed/unsigned flavors of a type 223 template <typename T> struct traits_t {}; 224 // int 225 template <> struct traits_t<signed int> { 226 typedef signed int signed_t; 227 typedef unsigned int unsigned_t; 228 typedef double floating_t; 229 static char const *spec; 230 static const signed_t max_value = 0x7fffffff; 231 static const signed_t min_value = 0x80000000; 232 static const int type_size = sizeof(signed_t); 233 }; 234 // unsigned int 235 template <> struct traits_t<unsigned int> { 236 typedef signed int signed_t; 237 typedef unsigned int unsigned_t; 238 typedef double floating_t; 239 static char const *spec; 240 static const unsigned_t max_value = 0xffffffff; 241 static const unsigned_t min_value = 0x00000000; 242 static const int type_size = sizeof(unsigned_t); 243 }; 244 // long 245 template <> struct traits_t<signed long> { 246 typedef signed long signed_t; 247 typedef unsigned long unsigned_t; 248 typedef long double floating_t; 249 static char const *spec; 250 static const int type_size = sizeof(signed_t); 251 }; 252 // long long 253 template <> struct traits_t<signed long long> { 254 typedef signed long long signed_t; 255 typedef unsigned long long unsigned_t; 256 typedef long double floating_t; 257 static char const *spec; 258 static const signed_t max_value = 0x7fffffffffffffffLL; 259 static const signed_t min_value = 0x8000000000000000LL; 260 static const int type_size = sizeof(signed_t); 261 }; 262 // unsigned long long 263 template <> struct traits_t<unsigned long long> { 264 typedef signed long long signed_t; 265 typedef unsigned long long unsigned_t; 266 typedef long double floating_t; 267 static char const *spec; 268 static const unsigned_t max_value = 0xffffffffffffffffLL; 269 static const unsigned_t min_value = 0x0000000000000000LL; 270 static const int type_size = sizeof(unsigned_t); 271 }; 272 //------------------------------------------------------------------------- 273 #else 274 #define CCAST(type, var) (type)(var) 275 #define RCAST(type, var) (type)(var) 276 #endif // __cplusplus 277 278 #define KMP_EXPORT extern /* export declaration in guide libraries */ 279 280 #if __GNUC__ >= 4 && !defined(__MINGW32__) 281 #define __forceinline __inline 282 #endif 283 284 #if KMP_OS_WINDOWS 285 #include <windows.h> 286 287 static inline int KMP_GET_PAGE_SIZE(void) { 288 SYSTEM_INFO si; 289 GetSystemInfo(&si); 290 return si.dwPageSize; 291 } 292 #else 293 #define KMP_GET_PAGE_SIZE() getpagesize() 294 #endif 295 296 #define PAGE_ALIGNED(_addr) \ 297 (!((size_t)_addr & (size_t)(KMP_GET_PAGE_SIZE() - 1))) 298 #define ALIGN_TO_PAGE(x) \ 299 (void *)(((size_t)(x)) & ~((size_t)(KMP_GET_PAGE_SIZE() - 1))) 300 301 /* ---------- Support for cache alignment, padding, etc. ----------------*/ 302 303 #ifdef __cplusplus 304 extern "C" { 305 #endif // __cplusplus 306 307 #define INTERNODE_CACHE_LINE 4096 /* for multi-node systems */ 308 309 /* Define the default size of the cache line */ 310 #ifndef CACHE_LINE 311 #define CACHE_LINE 128 /* cache line size in bytes */ 312 #else 313 #if (CACHE_LINE < 64) && !defined(KMP_OS_DARWIN) 314 // 2006-02-13: This produces too many warnings on OS X*. Disable for now 315 #warning CACHE_LINE is too small. 316 #endif 317 #endif /* CACHE_LINE */ 318 319 #define KMP_CACHE_PREFETCH(ADDR) /* nothing */ 320 321 // Define attribute that indicates that the fall through from the previous 322 // case label is intentional and should not be diagnosed by a compiler 323 // Code from libcxx/include/__config 324 // Use a function like macro to imply that it must be followed by a semicolon 325 #if __cplusplus > 201402L && __has_cpp_attribute(fallthrough) 326 # define KMP_FALLTHROUGH() [[fallthrough]] 327 #elif __has_cpp_attribute(clang::fallthrough) 328 # define KMP_FALLTHROUGH() [[clang::fallthrough]] 329 #elif __has_attribute(fallthrough) || __GNUC__ >= 7 330 # define KMP_FALLTHROUGH() __attribute__((__fallthrough__)) 331 #else 332 # define KMP_FALLTHROUGH() ((void)0) 333 #endif 334 335 // Define attribute that indicates a function does not return 336 #if __cplusplus >= 201103L 337 #define KMP_NORETURN [[noreturn]] 338 #elif KMP_OS_WINDOWS 339 #define KMP_NORETURN __declspec(noreturn) 340 #else 341 #define KMP_NORETURN __attribute__((noreturn)) 342 #endif 343 344 #if KMP_OS_WINDOWS && KMP_MSVC_COMPAT 345 #define KMP_ALIGN(bytes) __declspec(align(bytes)) 346 #define KMP_THREAD_LOCAL __declspec(thread) 347 #define KMP_ALIAS /* Nothing */ 348 #else 349 #define KMP_ALIGN(bytes) __attribute__((aligned(bytes))) 350 #define KMP_THREAD_LOCAL __thread 351 #define KMP_ALIAS(alias_of) __attribute__((alias(alias_of))) 352 #endif 353 354 #if KMP_HAVE_WEAK_ATTRIBUTE && !KMP_DYNAMIC_LIB 355 #define KMP_WEAK_ATTRIBUTE_EXTERNAL __attribute__((weak)) 356 #else 357 #define KMP_WEAK_ATTRIBUTE_EXTERNAL /* Nothing */ 358 #endif 359 360 #if KMP_HAVE_WEAK_ATTRIBUTE 361 #define KMP_WEAK_ATTRIBUTE_INTERNAL __attribute__((weak)) 362 #else 363 #define KMP_WEAK_ATTRIBUTE_INTERNAL /* Nothing */ 364 #endif 365 366 // Define KMP_VERSION_SYMBOL and KMP_EXPAND_NAME 367 #ifndef KMP_STR 368 #define KMP_STR(x) _KMP_STR(x) 369 #define _KMP_STR(x) #x 370 #endif 371 372 #ifdef KMP_USE_VERSION_SYMBOLS 373 // If using versioned symbols, KMP_EXPAND_NAME prepends 374 // __kmp_api_ to the real API name 375 #define KMP_EXPAND_NAME(api_name) _KMP_EXPAND_NAME(api_name) 376 #define _KMP_EXPAND_NAME(api_name) __kmp_api_##api_name 377 #define KMP_VERSION_SYMBOL(api_name, ver_num, ver_str) \ 378 _KMP_VERSION_SYMBOL(api_name, ver_num, ver_str, "VERSION") 379 #define _KMP_VERSION_SYMBOL(api_name, ver_num, ver_str, default_ver) \ 380 __typeof__(__kmp_api_##api_name) __kmp_api_##api_name##_##ver_num##_alias \ 381 __attribute__((alias(KMP_STR(__kmp_api_##api_name)))); \ 382 __asm__( \ 383 ".symver " KMP_STR(__kmp_api_##api_name##_##ver_num##_alias) "," KMP_STR( \ 384 api_name) "@" ver_str "\n\t"); \ 385 __asm__(".symver " KMP_STR(__kmp_api_##api_name) "," KMP_STR( \ 386 api_name) "@@" default_ver "\n\t") 387 #else // KMP_USE_VERSION_SYMBOLS 388 #define KMP_EXPAND_NAME(api_name) api_name 389 #define KMP_VERSION_SYMBOL(api_name, ver_num, ver_str) /* Nothing */ 390 #endif // KMP_USE_VERSION_SYMBOLS 391 392 /* Temporary note: if performance testing of this passes, we can remove 393 all references to KMP_DO_ALIGN and replace with KMP_ALIGN. */ 394 #define KMP_DO_ALIGN(bytes) KMP_ALIGN(bytes) 395 #define KMP_ALIGN_CACHE KMP_ALIGN(CACHE_LINE) 396 #define KMP_ALIGN_CACHE_INTERNODE KMP_ALIGN(INTERNODE_CACHE_LINE) 397 398 /* General purpose fence types for memory operations */ 399 enum kmp_mem_fence_type { 400 kmp_no_fence, /* No memory fence */ 401 kmp_acquire_fence, /* Acquire (read) memory fence */ 402 kmp_release_fence, /* Release (write) memory fence */ 403 kmp_full_fence /* Full (read+write) memory fence */ 404 }; 405 406 // Synchronization primitives 407 408 #if KMP_ASM_INTRINS && KMP_OS_WINDOWS 409 410 #if KMP_MSVC_COMPAT && !KMP_COMPILER_CLANG 411 #pragma intrinsic(InterlockedExchangeAdd) 412 #pragma intrinsic(InterlockedCompareExchange) 413 #pragma intrinsic(InterlockedExchange) 414 #pragma intrinsic(InterlockedExchange64) 415 #endif 416 417 // Using InterlockedIncrement / InterlockedDecrement causes a library loading 418 // ordering problem, so we use InterlockedExchangeAdd instead. 419 #define KMP_TEST_THEN_INC32(p) InterlockedExchangeAdd((volatile long *)(p), 1) 420 #define KMP_TEST_THEN_INC_ACQ32(p) \ 421 InterlockedExchangeAdd((volatile long *)(p), 1) 422 #define KMP_TEST_THEN_ADD4_32(p) InterlockedExchangeAdd((volatile long *)(p), 4) 423 #define KMP_TEST_THEN_ADD4_ACQ32(p) \ 424 InterlockedExchangeAdd((volatile long *)(p), 4) 425 #define KMP_TEST_THEN_DEC32(p) InterlockedExchangeAdd((volatile long *)(p), -1) 426 #define KMP_TEST_THEN_DEC_ACQ32(p) \ 427 InterlockedExchangeAdd((volatile long *)(p), -1) 428 #define KMP_TEST_THEN_ADD32(p, v) \ 429 InterlockedExchangeAdd((volatile long *)(p), (v)) 430 431 #define KMP_COMPARE_AND_STORE_RET32(p, cv, sv) \ 432 InterlockedCompareExchange((volatile long *)(p), (long)(sv), (long)(cv)) 433 434 #define KMP_XCHG_FIXED32(p, v) \ 435 InterlockedExchange((volatile long *)(p), (long)(v)) 436 #define KMP_XCHG_FIXED64(p, v) \ 437 InterlockedExchange64((volatile kmp_int64 *)(p), (kmp_int64)(v)) 438 439 inline kmp_real32 KMP_XCHG_REAL32(volatile kmp_real32 *p, kmp_real32 v) { 440 kmp_int32 tmp = InterlockedExchange((volatile long *)p, *(long *)&v); 441 return *(kmp_real32 *)&tmp; 442 } 443 444 // Routines that we still need to implement in assembly. 445 extern kmp_int8 __kmp_test_then_add8(volatile kmp_int8 *p, kmp_int8 v); 446 extern kmp_int8 __kmp_test_then_or8(volatile kmp_int8 *p, kmp_int8 v); 447 extern kmp_int8 __kmp_test_then_and8(volatile kmp_int8 *p, kmp_int8 v); 448 extern kmp_int32 __kmp_test_then_add32(volatile kmp_int32 *p, kmp_int32 v); 449 extern kmp_uint32 __kmp_test_then_or32(volatile kmp_uint32 *p, kmp_uint32 v); 450 extern kmp_uint32 __kmp_test_then_and32(volatile kmp_uint32 *p, kmp_uint32 v); 451 extern kmp_int64 __kmp_test_then_add64(volatile kmp_int64 *p, kmp_int64 v); 452 extern kmp_uint64 __kmp_test_then_or64(volatile kmp_uint64 *p, kmp_uint64 v); 453 extern kmp_uint64 __kmp_test_then_and64(volatile kmp_uint64 *p, kmp_uint64 v); 454 455 extern kmp_int8 __kmp_compare_and_store8(volatile kmp_int8 *p, kmp_int8 cv, 456 kmp_int8 sv); 457 extern kmp_int16 __kmp_compare_and_store16(volatile kmp_int16 *p, kmp_int16 cv, 458 kmp_int16 sv); 459 extern kmp_int32 __kmp_compare_and_store32(volatile kmp_int32 *p, kmp_int32 cv, 460 kmp_int32 sv); 461 extern kmp_int32 __kmp_compare_and_store64(volatile kmp_int64 *p, kmp_int64 cv, 462 kmp_int64 sv); 463 extern kmp_int8 __kmp_compare_and_store_ret8(volatile kmp_int8 *p, kmp_int8 cv, 464 kmp_int8 sv); 465 extern kmp_int16 __kmp_compare_and_store_ret16(volatile kmp_int16 *p, 466 kmp_int16 cv, kmp_int16 sv); 467 extern kmp_int32 __kmp_compare_and_store_ret32(volatile kmp_int32 *p, 468 kmp_int32 cv, kmp_int32 sv); 469 extern kmp_int64 __kmp_compare_and_store_ret64(volatile kmp_int64 *p, 470 kmp_int64 cv, kmp_int64 sv); 471 472 extern kmp_int8 __kmp_xchg_fixed8(volatile kmp_int8 *p, kmp_int8 v); 473 extern kmp_int16 __kmp_xchg_fixed16(volatile kmp_int16 *p, kmp_int16 v); 474 extern kmp_int32 __kmp_xchg_fixed32(volatile kmp_int32 *p, kmp_int32 v); 475 extern kmp_int64 __kmp_xchg_fixed64(volatile kmp_int64 *p, kmp_int64 v); 476 extern kmp_real32 __kmp_xchg_real32(volatile kmp_real32 *p, kmp_real32 v); 477 extern kmp_real64 __kmp_xchg_real64(volatile kmp_real64 *p, kmp_real64 v); 478 479 //#define KMP_TEST_THEN_INC32(p) __kmp_test_then_add32((p), 1) 480 //#define KMP_TEST_THEN_INC_ACQ32(p) __kmp_test_then_add32((p), 1) 481 #define KMP_TEST_THEN_INC64(p) __kmp_test_then_add64((p), 1LL) 482 #define KMP_TEST_THEN_INC_ACQ64(p) __kmp_test_then_add64((p), 1LL) 483 //#define KMP_TEST_THEN_ADD4_32(p) __kmp_test_then_add32((p), 4) 484 //#define KMP_TEST_THEN_ADD4_ACQ32(p) __kmp_test_then_add32((p), 4) 485 #define KMP_TEST_THEN_ADD4_64(p) __kmp_test_then_add64((p), 4LL) 486 #define KMP_TEST_THEN_ADD4_ACQ64(p) __kmp_test_then_add64((p), 4LL) 487 //#define KMP_TEST_THEN_DEC32(p) __kmp_test_then_add32((p), -1) 488 //#define KMP_TEST_THEN_DEC_ACQ32(p) __kmp_test_then_add32((p), -1) 489 #define KMP_TEST_THEN_DEC64(p) __kmp_test_then_add64((p), -1LL) 490 #define KMP_TEST_THEN_DEC_ACQ64(p) __kmp_test_then_add64((p), -1LL) 491 //#define KMP_TEST_THEN_ADD32(p, v) __kmp_test_then_add32((p), (v)) 492 #define KMP_TEST_THEN_ADD8(p, v) __kmp_test_then_add8((p), (v)) 493 #define KMP_TEST_THEN_ADD64(p, v) __kmp_test_then_add64((p), (v)) 494 495 #define KMP_TEST_THEN_OR8(p, v) __kmp_test_then_or8((p), (v)) 496 #define KMP_TEST_THEN_AND8(p, v) __kmp_test_then_and8((p), (v)) 497 #define KMP_TEST_THEN_OR32(p, v) __kmp_test_then_or32((p), (v)) 498 #define KMP_TEST_THEN_AND32(p, v) __kmp_test_then_and32((p), (v)) 499 #define KMP_TEST_THEN_OR64(p, v) __kmp_test_then_or64((p), (v)) 500 #define KMP_TEST_THEN_AND64(p, v) __kmp_test_then_and64((p), (v)) 501 502 #define KMP_COMPARE_AND_STORE_ACQ8(p, cv, sv) \ 503 __kmp_compare_and_store8((p), (cv), (sv)) 504 #define KMP_COMPARE_AND_STORE_REL8(p, cv, sv) \ 505 __kmp_compare_and_store8((p), (cv), (sv)) 506 #define KMP_COMPARE_AND_STORE_ACQ16(p, cv, sv) \ 507 __kmp_compare_and_store16((p), (cv), (sv)) 508 #define KMP_COMPARE_AND_STORE_REL16(p, cv, sv) \ 509 __kmp_compare_and_store16((p), (cv), (sv)) 510 #define KMP_COMPARE_AND_STORE_ACQ32(p, cv, sv) \ 511 __kmp_compare_and_store32((volatile kmp_int32 *)(p), (kmp_int32)(cv), \ 512 (kmp_int32)(sv)) 513 #define KMP_COMPARE_AND_STORE_REL32(p, cv, sv) \ 514 __kmp_compare_and_store32((volatile kmp_int32 *)(p), (kmp_int32)(cv), \ 515 (kmp_int32)(sv)) 516 #define KMP_COMPARE_AND_STORE_ACQ64(p, cv, sv) \ 517 __kmp_compare_and_store64((volatile kmp_int64 *)(p), (kmp_int64)(cv), \ 518 (kmp_int64)(sv)) 519 #define KMP_COMPARE_AND_STORE_REL64(p, cv, sv) \ 520 __kmp_compare_and_store64((volatile kmp_int64 *)(p), (kmp_int64)(cv), \ 521 (kmp_int64)(sv)) 522 523 #if KMP_ARCH_X86 524 #define KMP_COMPARE_AND_STORE_PTR(p, cv, sv) \ 525 __kmp_compare_and_store32((volatile kmp_int32 *)(p), (kmp_int32)(cv), \ 526 (kmp_int32)(sv)) 527 #else /* 64 bit pointers */ 528 #define KMP_COMPARE_AND_STORE_PTR(p, cv, sv) \ 529 __kmp_compare_and_store64((volatile kmp_int64 *)(p), (kmp_int64)(cv), \ 530 (kmp_int64)(sv)) 531 #endif /* KMP_ARCH_X86 */ 532 533 #define KMP_COMPARE_AND_STORE_RET8(p, cv, sv) \ 534 __kmp_compare_and_store_ret8((p), (cv), (sv)) 535 #define KMP_COMPARE_AND_STORE_RET16(p, cv, sv) \ 536 __kmp_compare_and_store_ret16((p), (cv), (sv)) 537 #define KMP_COMPARE_AND_STORE_RET64(p, cv, sv) \ 538 __kmp_compare_and_store_ret64((volatile kmp_int64 *)(p), (kmp_int64)(cv), \ 539 (kmp_int64)(sv)) 540 541 #define KMP_XCHG_FIXED8(p, v) \ 542 __kmp_xchg_fixed8((volatile kmp_int8 *)(p), (kmp_int8)(v)); 543 #define KMP_XCHG_FIXED16(p, v) __kmp_xchg_fixed16((p), (v)); 544 //#define KMP_XCHG_FIXED32(p, v) __kmp_xchg_fixed32((p), (v)); 545 //#define KMP_XCHG_FIXED64(p, v) __kmp_xchg_fixed64((p), (v)); 546 //#define KMP_XCHG_REAL32(p, v) __kmp_xchg_real32((p), (v)); 547 #define KMP_XCHG_REAL64(p, v) __kmp_xchg_real64((p), (v)); 548 549 #elif (KMP_ASM_INTRINS && KMP_OS_UNIX) || !(KMP_ARCH_X86 || KMP_ARCH_X86_64) 550 551 /* cast p to correct type so that proper intrinsic will be used */ 552 #define KMP_TEST_THEN_INC32(p) \ 553 __sync_fetch_and_add((volatile kmp_int32 *)(p), 1) 554 #define KMP_TEST_THEN_INC_ACQ32(p) \ 555 __sync_fetch_and_add((volatile kmp_int32 *)(p), 1) 556 #if KMP_ARCH_MIPS 557 #define KMP_TEST_THEN_INC64(p) \ 558 __atomic_fetch_add((volatile kmp_int64 *)(p), 1LL, __ATOMIC_SEQ_CST) 559 #define KMP_TEST_THEN_INC_ACQ64(p) \ 560 __atomic_fetch_add((volatile kmp_int64 *)(p), 1LL, __ATOMIC_SEQ_CST) 561 #else 562 #define KMP_TEST_THEN_INC64(p) \ 563 __sync_fetch_and_add((volatile kmp_int64 *)(p), 1LL) 564 #define KMP_TEST_THEN_INC_ACQ64(p) \ 565 __sync_fetch_and_add((volatile kmp_int64 *)(p), 1LL) 566 #endif 567 #define KMP_TEST_THEN_ADD4_32(p) \ 568 __sync_fetch_and_add((volatile kmp_int32 *)(p), 4) 569 #define KMP_TEST_THEN_ADD4_ACQ32(p) \ 570 __sync_fetch_and_add((volatile kmp_int32 *)(p), 4) 571 #if KMP_ARCH_MIPS 572 #define KMP_TEST_THEN_ADD4_64(p) \ 573 __atomic_fetch_add((volatile kmp_int64 *)(p), 4LL, __ATOMIC_SEQ_CST) 574 #define KMP_TEST_THEN_ADD4_ACQ64(p) \ 575 __atomic_fetch_add((volatile kmp_int64 *)(p), 4LL, __ATOMIC_SEQ_CST) 576 #define KMP_TEST_THEN_DEC64(p) \ 577 __atomic_fetch_sub((volatile kmp_int64 *)(p), 1LL, __ATOMIC_SEQ_CST) 578 #define KMP_TEST_THEN_DEC_ACQ64(p) \ 579 __atomic_fetch_sub((volatile kmp_int64 *)(p), 1LL, __ATOMIC_SEQ_CST) 580 #else 581 #define KMP_TEST_THEN_ADD4_64(p) \ 582 __sync_fetch_and_add((volatile kmp_int64 *)(p), 4LL) 583 #define KMP_TEST_THEN_ADD4_ACQ64(p) \ 584 __sync_fetch_and_add((volatile kmp_int64 *)(p), 4LL) 585 #define KMP_TEST_THEN_DEC64(p) \ 586 __sync_fetch_and_sub((volatile kmp_int64 *)(p), 1LL) 587 #define KMP_TEST_THEN_DEC_ACQ64(p) \ 588 __sync_fetch_and_sub((volatile kmp_int64 *)(p), 1LL) 589 #endif 590 #define KMP_TEST_THEN_DEC32(p) \ 591 __sync_fetch_and_sub((volatile kmp_int32 *)(p), 1) 592 #define KMP_TEST_THEN_DEC_ACQ32(p) \ 593 __sync_fetch_and_sub((volatile kmp_int32 *)(p), 1) 594 #define KMP_TEST_THEN_ADD8(p, v) \ 595 __sync_fetch_and_add((volatile kmp_int8 *)(p), (kmp_int8)(v)) 596 #define KMP_TEST_THEN_ADD32(p, v) \ 597 __sync_fetch_and_add((volatile kmp_int32 *)(p), (kmp_int32)(v)) 598 #if KMP_ARCH_MIPS 599 #define KMP_TEST_THEN_ADD64(p, v) \ 600 __atomic_fetch_add((volatile kmp_uint64 *)(p), (kmp_uint64)(v), \ 601 __ATOMIC_SEQ_CST) 602 #else 603 #define KMP_TEST_THEN_ADD64(p, v) \ 604 __sync_fetch_and_add((volatile kmp_int64 *)(p), (kmp_int64)(v)) 605 #endif 606 607 #define KMP_TEST_THEN_OR8(p, v) \ 608 __sync_fetch_and_or((volatile kmp_int8 *)(p), (kmp_int8)(v)) 609 #define KMP_TEST_THEN_AND8(p, v) \ 610 __sync_fetch_and_and((volatile kmp_int8 *)(p), (kmp_int8)(v)) 611 #define KMP_TEST_THEN_OR32(p, v) \ 612 __sync_fetch_and_or((volatile kmp_uint32 *)(p), (kmp_uint32)(v)) 613 #define KMP_TEST_THEN_AND32(p, v) \ 614 __sync_fetch_and_and((volatile kmp_uint32 *)(p), (kmp_uint32)(v)) 615 #if KMP_ARCH_MIPS 616 #define KMP_TEST_THEN_OR64(p, v) \ 617 __atomic_fetch_or((volatile kmp_uint64 *)(p), (kmp_uint64)(v), \ 618 __ATOMIC_SEQ_CST) 619 #define KMP_TEST_THEN_AND64(p, v) \ 620 __atomic_fetch_and((volatile kmp_uint64 *)(p), (kmp_uint64)(v), \ 621 __ATOMIC_SEQ_CST) 622 #else 623 #define KMP_TEST_THEN_OR64(p, v) \ 624 __sync_fetch_and_or((volatile kmp_uint64 *)(p), (kmp_uint64)(v)) 625 #define KMP_TEST_THEN_AND64(p, v) \ 626 __sync_fetch_and_and((volatile kmp_uint64 *)(p), (kmp_uint64)(v)) 627 #endif 628 629 #define KMP_COMPARE_AND_STORE_ACQ8(p, cv, sv) \ 630 __sync_bool_compare_and_swap((volatile kmp_uint8 *)(p), (kmp_uint8)(cv), \ 631 (kmp_uint8)(sv)) 632 #define KMP_COMPARE_AND_STORE_REL8(p, cv, sv) \ 633 __sync_bool_compare_and_swap((volatile kmp_uint8 *)(p), (kmp_uint8)(cv), \ 634 (kmp_uint8)(sv)) 635 #define KMP_COMPARE_AND_STORE_ACQ16(p, cv, sv) \ 636 __sync_bool_compare_and_swap((volatile kmp_uint16 *)(p), (kmp_uint16)(cv), \ 637 (kmp_uint16)(sv)) 638 #define KMP_COMPARE_AND_STORE_REL16(p, cv, sv) \ 639 __sync_bool_compare_and_swap((volatile kmp_uint16 *)(p), (kmp_uint16)(cv), \ 640 (kmp_uint16)(sv)) 641 #define KMP_COMPARE_AND_STORE_ACQ32(p, cv, sv) \ 642 __sync_bool_compare_and_swap((volatile kmp_uint32 *)(p), (kmp_uint32)(cv), \ 643 (kmp_uint32)(sv)) 644 #define KMP_COMPARE_AND_STORE_REL32(p, cv, sv) \ 645 __sync_bool_compare_and_swap((volatile kmp_uint32 *)(p), (kmp_uint32)(cv), \ 646 (kmp_uint32)(sv)) 647 #define KMP_COMPARE_AND_STORE_PTR(p, cv, sv) \ 648 __sync_bool_compare_and_swap((void *volatile *)(p), (void *)(cv), \ 649 (void *)(sv)) 650 651 #define KMP_COMPARE_AND_STORE_RET8(p, cv, sv) \ 652 __sync_val_compare_and_swap((volatile kmp_uint8 *)(p), (kmp_uint8)(cv), \ 653 (kmp_uint8)(sv)) 654 #define KMP_COMPARE_AND_STORE_RET16(p, cv, sv) \ 655 __sync_val_compare_and_swap((volatile kmp_uint16 *)(p), (kmp_uint16)(cv), \ 656 (kmp_uint16)(sv)) 657 #define KMP_COMPARE_AND_STORE_RET32(p, cv, sv) \ 658 __sync_val_compare_and_swap((volatile kmp_uint32 *)(p), (kmp_uint32)(cv), \ 659 (kmp_uint32)(sv)) 660 #if KMP_ARCH_MIPS 661 static inline bool mips_sync_bool_compare_and_swap( 662 volatile kmp_uint64 *p, kmp_uint64 cv, kmp_uint64 sv) { 663 return __atomic_compare_exchange(p, &cv, &sv, false, __ATOMIC_SEQ_CST, 664 __ATOMIC_SEQ_CST); 665 } 666 static inline bool mips_sync_val_compare_and_swap( 667 volatile kmp_uint64 *p, kmp_uint64 cv, kmp_uint64 sv) { 668 __atomic_compare_exchange(p, &cv, &sv, false, __ATOMIC_SEQ_CST, 669 __ATOMIC_SEQ_CST); 670 return cv; 671 } 672 #define KMP_COMPARE_AND_STORE_ACQ64(p, cv, sv) \ 673 mips_sync_bool_compare_and_swap((volatile kmp_uint64 *)(p), (kmp_uint64)(cv),\ 674 (kmp_uint64)(sv)) 675 #define KMP_COMPARE_AND_STORE_REL64(p, cv, sv) \ 676 mips_sync_bool_compare_and_swap((volatile kmp_uint64 *)(p), (kmp_uint64)(cv),\ 677 (kmp_uint64)(sv)) 678 #define KMP_COMPARE_AND_STORE_RET64(p, cv, sv) \ 679 mips_sync_val_compare_and_swap((volatile kmp_uint64 *)(p), (kmp_uint64)(cv), \ 680 (kmp_uint64)(sv)) 681 #else 682 #define KMP_COMPARE_AND_STORE_ACQ64(p, cv, sv) \ 683 __sync_bool_compare_and_swap((volatile kmp_uint64 *)(p), (kmp_uint64)(cv), \ 684 (kmp_uint64)(sv)) 685 #define KMP_COMPARE_AND_STORE_REL64(p, cv, sv) \ 686 __sync_bool_compare_and_swap((volatile kmp_uint64 *)(p), (kmp_uint64)(cv), \ 687 (kmp_uint64)(sv)) 688 #define KMP_COMPARE_AND_STORE_RET64(p, cv, sv) \ 689 __sync_val_compare_and_swap((volatile kmp_uint64 *)(p), (kmp_uint64)(cv), \ 690 (kmp_uint64)(sv)) 691 #endif 692 693 #define KMP_XCHG_FIXED8(p, v) \ 694 __sync_lock_test_and_set((volatile kmp_uint8 *)(p), (kmp_uint8)(v)) 695 #define KMP_XCHG_FIXED16(p, v) \ 696 __sync_lock_test_and_set((volatile kmp_uint16 *)(p), (kmp_uint16)(v)) 697 #define KMP_XCHG_FIXED32(p, v) \ 698 __sync_lock_test_and_set((volatile kmp_uint32 *)(p), (kmp_uint32)(v)) 699 #define KMP_XCHG_FIXED64(p, v) \ 700 __sync_lock_test_and_set((volatile kmp_uint64 *)(p), (kmp_uint64)(v)) 701 702 inline kmp_real32 KMP_XCHG_REAL32(volatile kmp_real32 *p, kmp_real32 v) { 703 kmp_int32 tmp = 704 __sync_lock_test_and_set((volatile kmp_uint32 *)(p), *(kmp_uint32 *)&v); 705 return *(kmp_real32 *)&tmp; 706 } 707 708 inline kmp_real64 KMP_XCHG_REAL64(volatile kmp_real64 *p, kmp_real64 v) { 709 kmp_int64 tmp = 710 __sync_lock_test_and_set((volatile kmp_uint64 *)(p), *(kmp_uint64 *)&v); 711 return *(kmp_real64 *)&tmp; 712 } 713 714 #else 715 716 extern kmp_int8 __kmp_test_then_add8(volatile kmp_int8 *p, kmp_int8 v); 717 extern kmp_int8 __kmp_test_then_or8(volatile kmp_int8 *p, kmp_int8 v); 718 extern kmp_int8 __kmp_test_then_and8(volatile kmp_int8 *p, kmp_int8 v); 719 extern kmp_int32 __kmp_test_then_add32(volatile kmp_int32 *p, kmp_int32 v); 720 extern kmp_uint32 __kmp_test_then_or32(volatile kmp_uint32 *p, kmp_uint32 v); 721 extern kmp_uint32 __kmp_test_then_and32(volatile kmp_uint32 *p, kmp_uint32 v); 722 extern kmp_int64 __kmp_test_then_add64(volatile kmp_int64 *p, kmp_int64 v); 723 extern kmp_uint64 __kmp_test_then_or64(volatile kmp_uint64 *p, kmp_uint64 v); 724 extern kmp_uint64 __kmp_test_then_and64(volatile kmp_uint64 *p, kmp_uint64 v); 725 726 extern kmp_int8 __kmp_compare_and_store8(volatile kmp_int8 *p, kmp_int8 cv, 727 kmp_int8 sv); 728 extern kmp_int16 __kmp_compare_and_store16(volatile kmp_int16 *p, kmp_int16 cv, 729 kmp_int16 sv); 730 extern kmp_int32 __kmp_compare_and_store32(volatile kmp_int32 *p, kmp_int32 cv, 731 kmp_int32 sv); 732 extern kmp_int32 __kmp_compare_and_store64(volatile kmp_int64 *p, kmp_int64 cv, 733 kmp_int64 sv); 734 extern kmp_int8 __kmp_compare_and_store_ret8(volatile kmp_int8 *p, kmp_int8 cv, 735 kmp_int8 sv); 736 extern kmp_int16 __kmp_compare_and_store_ret16(volatile kmp_int16 *p, 737 kmp_int16 cv, kmp_int16 sv); 738 extern kmp_int32 __kmp_compare_and_store_ret32(volatile kmp_int32 *p, 739 kmp_int32 cv, kmp_int32 sv); 740 extern kmp_int64 __kmp_compare_and_store_ret64(volatile kmp_int64 *p, 741 kmp_int64 cv, kmp_int64 sv); 742 743 extern kmp_int8 __kmp_xchg_fixed8(volatile kmp_int8 *p, kmp_int8 v); 744 extern kmp_int16 __kmp_xchg_fixed16(volatile kmp_int16 *p, kmp_int16 v); 745 extern kmp_int32 __kmp_xchg_fixed32(volatile kmp_int32 *p, kmp_int32 v); 746 extern kmp_int64 __kmp_xchg_fixed64(volatile kmp_int64 *p, kmp_int64 v); 747 extern kmp_real32 __kmp_xchg_real32(volatile kmp_real32 *p, kmp_real32 v); 748 extern kmp_real64 __kmp_xchg_real64(volatile kmp_real64 *p, kmp_real64 v); 749 750 #define KMP_TEST_THEN_INC32(p) \ 751 __kmp_test_then_add32((volatile kmp_int32 *)(p), 1) 752 #define KMP_TEST_THEN_INC_ACQ32(p) \ 753 __kmp_test_then_add32((volatile kmp_int32 *)(p), 1) 754 #define KMP_TEST_THEN_INC64(p) \ 755 __kmp_test_then_add64((volatile kmp_int64 *)(p), 1LL) 756 #define KMP_TEST_THEN_INC_ACQ64(p) \ 757 __kmp_test_then_add64((volatile kmp_int64 *)(p), 1LL) 758 #define KMP_TEST_THEN_ADD4_32(p) \ 759 __kmp_test_then_add32((volatile kmp_int32 *)(p), 4) 760 #define KMP_TEST_THEN_ADD4_ACQ32(p) \ 761 __kmp_test_then_add32((volatile kmp_int32 *)(p), 4) 762 #define KMP_TEST_THEN_ADD4_64(p) \ 763 __kmp_test_then_add64((volatile kmp_int64 *)(p), 4LL) 764 #define KMP_TEST_THEN_ADD4_ACQ64(p) \ 765 __kmp_test_then_add64((volatile kmp_int64 *)(p), 4LL) 766 #define KMP_TEST_THEN_DEC32(p) \ 767 __kmp_test_then_add32((volatile kmp_int32 *)(p), -1) 768 #define KMP_TEST_THEN_DEC_ACQ32(p) \ 769 __kmp_test_then_add32((volatile kmp_int32 *)(p), -1) 770 #define KMP_TEST_THEN_DEC64(p) \ 771 __kmp_test_then_add64((volatile kmp_int64 *)(p), -1LL) 772 #define KMP_TEST_THEN_DEC_ACQ64(p) \ 773 __kmp_test_then_add64((volatile kmp_int64 *)(p), -1LL) 774 #define KMP_TEST_THEN_ADD8(p, v) \ 775 __kmp_test_then_add8((volatile kmp_int8 *)(p), (kmp_int8)(v)) 776 #define KMP_TEST_THEN_ADD32(p, v) \ 777 __kmp_test_then_add32((volatile kmp_int32 *)(p), (kmp_int32)(v)) 778 #define KMP_TEST_THEN_ADD64(p, v) \ 779 __kmp_test_then_add64((volatile kmp_int64 *)(p), (kmp_int64)(v)) 780 781 #define KMP_TEST_THEN_OR8(p, v) \ 782 __kmp_test_then_or8((volatile kmp_int8 *)(p), (kmp_int8)(v)) 783 #define KMP_TEST_THEN_AND8(p, v) \ 784 __kmp_test_then_and8((volatile kmp_int8 *)(p), (kmp_int8)(v)) 785 #define KMP_TEST_THEN_OR32(p, v) \ 786 __kmp_test_then_or32((volatile kmp_uint32 *)(p), (kmp_uint32)(v)) 787 #define KMP_TEST_THEN_AND32(p, v) \ 788 __kmp_test_then_and32((volatile kmp_uint32 *)(p), (kmp_uint32)(v)) 789 #define KMP_TEST_THEN_OR64(p, v) \ 790 __kmp_test_then_or64((volatile kmp_uint64 *)(p), (kmp_uint64)(v)) 791 #define KMP_TEST_THEN_AND64(p, v) \ 792 __kmp_test_then_and64((volatile kmp_uint64 *)(p), (kmp_uint64)(v)) 793 794 #define KMP_COMPARE_AND_STORE_ACQ8(p, cv, sv) \ 795 __kmp_compare_and_store8((volatile kmp_int8 *)(p), (kmp_int8)(cv), \ 796 (kmp_int8)(sv)) 797 #define KMP_COMPARE_AND_STORE_REL8(p, cv, sv) \ 798 __kmp_compare_and_store8((volatile kmp_int8 *)(p), (kmp_int8)(cv), \ 799 (kmp_int8)(sv)) 800 #define KMP_COMPARE_AND_STORE_ACQ16(p, cv, sv) \ 801 __kmp_compare_and_store16((volatile kmp_int16 *)(p), (kmp_int16)(cv), \ 802 (kmp_int16)(sv)) 803 #define KMP_COMPARE_AND_STORE_REL16(p, cv, sv) \ 804 __kmp_compare_and_store16((volatile kmp_int16 *)(p), (kmp_int16)(cv), \ 805 (kmp_int16)(sv)) 806 #define KMP_COMPARE_AND_STORE_ACQ32(p, cv, sv) \ 807 __kmp_compare_and_store32((volatile kmp_int32 *)(p), (kmp_int32)(cv), \ 808 (kmp_int32)(sv)) 809 #define KMP_COMPARE_AND_STORE_REL32(p, cv, sv) \ 810 __kmp_compare_and_store32((volatile kmp_int32 *)(p), (kmp_int32)(cv), \ 811 (kmp_int32)(sv)) 812 #define KMP_COMPARE_AND_STORE_ACQ64(p, cv, sv) \ 813 __kmp_compare_and_store64((volatile kmp_int64 *)(p), (kmp_int64)(cv), \ 814 (kmp_int64)(sv)) 815 #define KMP_COMPARE_AND_STORE_REL64(p, cv, sv) \ 816 __kmp_compare_and_store64((volatile kmp_int64 *)(p), (kmp_int64)(cv), \ 817 (kmp_int64)(sv)) 818 819 #if KMP_ARCH_X86 820 #define KMP_COMPARE_AND_STORE_PTR(p, cv, sv) \ 821 __kmp_compare_and_store32((volatile kmp_int32 *)(p), (kmp_int32)(cv), \ 822 (kmp_int32)(sv)) 823 #else /* 64 bit pointers */ 824 #define KMP_COMPARE_AND_STORE_PTR(p, cv, sv) \ 825 __kmp_compare_and_store64((volatile kmp_int64 *)(p), (kmp_int64)(cv), \ 826 (kmp_int64)(sv)) 827 #endif /* KMP_ARCH_X86 */ 828 829 #define KMP_COMPARE_AND_STORE_RET8(p, cv, sv) \ 830 __kmp_compare_and_store_ret8((p), (cv), (sv)) 831 #define KMP_COMPARE_AND_STORE_RET16(p, cv, sv) \ 832 __kmp_compare_and_store_ret16((p), (cv), (sv)) 833 #define KMP_COMPARE_AND_STORE_RET32(p, cv, sv) \ 834 __kmp_compare_and_store_ret32((volatile kmp_int32 *)(p), (kmp_int32)(cv), \ 835 (kmp_int32)(sv)) 836 #define KMP_COMPARE_AND_STORE_RET64(p, cv, sv) \ 837 __kmp_compare_and_store_ret64((volatile kmp_int64 *)(p), (kmp_int64)(cv), \ 838 (kmp_int64)(sv)) 839 840 #define KMP_XCHG_FIXED8(p, v) \ 841 __kmp_xchg_fixed8((volatile kmp_int8 *)(p), (kmp_int8)(v)); 842 #define KMP_XCHG_FIXED16(p, v) __kmp_xchg_fixed16((p), (v)); 843 #define KMP_XCHG_FIXED32(p, v) __kmp_xchg_fixed32((p), (v)); 844 #define KMP_XCHG_FIXED64(p, v) __kmp_xchg_fixed64((p), (v)); 845 #define KMP_XCHG_REAL32(p, v) __kmp_xchg_real32((p), (v)); 846 #define KMP_XCHG_REAL64(p, v) __kmp_xchg_real64((p), (v)); 847 848 #endif /* KMP_ASM_INTRINS */ 849 850 /* ------------- relaxed consistency memory model stuff ------------------ */ 851 852 #if KMP_OS_WINDOWS 853 #ifdef __ABSOFT_WIN 854 #define KMP_MB() asm("nop") 855 #define KMP_IMB() asm("nop") 856 #else 857 #define KMP_MB() /* _asm{ nop } */ 858 #define KMP_IMB() /* _asm{ nop } */ 859 #endif 860 #endif /* KMP_OS_WINDOWS */ 861 862 #if KMP_ARCH_PPC64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64 || KMP_ARCH_MIPS || \ 863 KMP_ARCH_MIPS64 || KMP_ARCH_RISCV64 864 #define KMP_MB() __sync_synchronize() 865 #endif 866 867 #ifndef KMP_MB 868 #define KMP_MB() /* nothing to do */ 869 #endif 870 871 #ifndef KMP_IMB 872 #define KMP_IMB() /* nothing to do */ 873 #endif 874 875 #ifndef KMP_ST_REL32 876 #define KMP_ST_REL32(A, D) (*(A) = (D)) 877 #endif 878 879 #ifndef KMP_ST_REL64 880 #define KMP_ST_REL64(A, D) (*(A) = (D)) 881 #endif 882 883 #ifndef KMP_LD_ACQ32 884 #define KMP_LD_ACQ32(A) (*(A)) 885 #endif 886 887 #ifndef KMP_LD_ACQ64 888 #define KMP_LD_ACQ64(A) (*(A)) 889 #endif 890 891 /* ------------------------------------------------------------------------ */ 892 // FIXME - maybe this should this be 893 // 894 // #define TCR_4(a) (*(volatile kmp_int32 *)(&a)) 895 // #define TCW_4(a,b) (a) = (*(volatile kmp_int32 *)&(b)) 896 // 897 // #define TCR_8(a) (*(volatile kmp_int64 *)(a)) 898 // #define TCW_8(a,b) (a) = (*(volatile kmp_int64 *)(&b)) 899 // 900 // I'm fairly certain this is the correct thing to do, but I'm afraid 901 // of performance regressions. 902 903 #define TCR_1(a) (a) 904 #define TCW_1(a, b) (a) = (b) 905 #define TCR_4(a) (a) 906 #define TCW_4(a, b) (a) = (b) 907 #define TCI_4(a) (++(a)) 908 #define TCD_4(a) (--(a)) 909 #define TCR_8(a) (a) 910 #define TCW_8(a, b) (a) = (b) 911 #define TCI_8(a) (++(a)) 912 #define TCD_8(a) (--(a)) 913 #define TCR_SYNC_4(a) (a) 914 #define TCW_SYNC_4(a, b) (a) = (b) 915 #define TCX_SYNC_4(a, b, c) \ 916 KMP_COMPARE_AND_STORE_REL32((volatile kmp_int32 *)(volatile void *)&(a), \ 917 (kmp_int32)(b), (kmp_int32)(c)) 918 #define TCR_SYNC_8(a) (a) 919 #define TCW_SYNC_8(a, b) (a) = (b) 920 #define TCX_SYNC_8(a, b, c) \ 921 KMP_COMPARE_AND_STORE_REL64((volatile kmp_int64 *)(volatile void *)&(a), \ 922 (kmp_int64)(b), (kmp_int64)(c)) 923 924 #if KMP_ARCH_X86 || KMP_ARCH_MIPS 925 // What about ARM? 926 #define TCR_PTR(a) ((void *)TCR_4(a)) 927 #define TCW_PTR(a, b) TCW_4((a), (b)) 928 #define TCR_SYNC_PTR(a) ((void *)TCR_SYNC_4(a)) 929 #define TCW_SYNC_PTR(a, b) TCW_SYNC_4((a), (b)) 930 #define TCX_SYNC_PTR(a, b, c) ((void *)TCX_SYNC_4((a), (b), (c))) 931 932 #else /* 64 bit pointers */ 933 934 #define TCR_PTR(a) ((void *)TCR_8(a)) 935 #define TCW_PTR(a, b) TCW_8((a), (b)) 936 #define TCR_SYNC_PTR(a) ((void *)TCR_SYNC_8(a)) 937 #define TCW_SYNC_PTR(a, b) TCW_SYNC_8((a), (b)) 938 #define TCX_SYNC_PTR(a, b, c) ((void *)TCX_SYNC_8((a), (b), (c))) 939 940 #endif /* KMP_ARCH_X86 */ 941 942 /* If these FTN_{TRUE,FALSE} values change, may need to change several places 943 where they are used to check that language is Fortran, not C. */ 944 945 #ifndef FTN_TRUE 946 #define FTN_TRUE TRUE 947 #endif 948 949 #ifndef FTN_FALSE 950 #define FTN_FALSE FALSE 951 #endif 952 953 typedef void (*microtask_t)(int *gtid, int *npr, ...); 954 955 #ifdef USE_VOLATILE_CAST 956 #define VOLATILE_CAST(x) (volatile x) 957 #else 958 #define VOLATILE_CAST(x) (x) 959 #endif 960 961 #define KMP_WAIT __kmp_wait_4 962 #define KMP_WAIT_PTR __kmp_wait_4_ptr 963 #define KMP_EQ __kmp_eq_4 964 #define KMP_NEQ __kmp_neq_4 965 #define KMP_LT __kmp_lt_4 966 #define KMP_GE __kmp_ge_4 967 #define KMP_LE __kmp_le_4 968 969 /* Workaround for Intel(R) 64 code gen bug when taking address of static array 970 * (Intel(R) 64 Tracker #138) */ 971 #if (KMP_ARCH_X86_64 || KMP_ARCH_PPC64) && KMP_OS_LINUX 972 #define STATIC_EFI2_WORKAROUND 973 #else 974 #define STATIC_EFI2_WORKAROUND static 975 #endif 976 977 // Support of BGET usage 978 #ifndef KMP_USE_BGET 979 #define KMP_USE_BGET 1 980 #endif 981 982 // Switches for OSS builds 983 #ifndef USE_CMPXCHG_FIX 984 #define USE_CMPXCHG_FIX 1 985 #endif 986 987 // Enable dynamic user lock 988 #define KMP_USE_DYNAMIC_LOCK 1 989 990 // Enable Intel(R) Transactional Synchronization Extensions (Intel(R) TSX) if 991 // dynamic user lock is turned on 992 #if KMP_USE_DYNAMIC_LOCK 993 // Visual studio can't handle the asm sections in this code 994 #define KMP_USE_TSX (KMP_ARCH_X86 || KMP_ARCH_X86_64) && !KMP_COMPILER_MSVC 995 #ifdef KMP_USE_ADAPTIVE_LOCKS 996 #undef KMP_USE_ADAPTIVE_LOCKS 997 #endif 998 #define KMP_USE_ADAPTIVE_LOCKS KMP_USE_TSX 999 #endif 1000 1001 // Enable tick time conversion of ticks to seconds 1002 #if KMP_STATS_ENABLED 1003 #define KMP_HAVE_TICK_TIME \ 1004 (KMP_OS_LINUX && (KMP_MIC || KMP_ARCH_X86 || KMP_ARCH_X86_64)) 1005 #endif 1006 1007 // Warning levels 1008 enum kmp_warnings_level { 1009 kmp_warnings_off = 0, /* No warnings */ 1010 kmp_warnings_low, /* Minimal warnings (default) */ 1011 kmp_warnings_explicit = 6, /* Explicitly set to ON - more warnings */ 1012 kmp_warnings_verbose /* reserved */ 1013 }; 1014 1015 #ifdef __cplusplus 1016 } // extern "C" 1017 #endif // __cplusplus 1018 1019 // Macros for C++11 atomic functions 1020 #define KMP_ATOMIC_LD(p, order) (p)->load(std::memory_order_##order) 1021 #define KMP_ATOMIC_OP(op, p, v, order) (p)->op(v, std::memory_order_##order) 1022 1023 // For non-default load/store 1024 #define KMP_ATOMIC_LD_ACQ(p) KMP_ATOMIC_LD(p, acquire) 1025 #define KMP_ATOMIC_LD_RLX(p) KMP_ATOMIC_LD(p, relaxed) 1026 #define KMP_ATOMIC_ST_REL(p, v) KMP_ATOMIC_OP(store, p, v, release) 1027 #define KMP_ATOMIC_ST_RLX(p, v) KMP_ATOMIC_OP(store, p, v, relaxed) 1028 1029 // For non-default fetch_<op> 1030 #define KMP_ATOMIC_ADD(p, v) KMP_ATOMIC_OP(fetch_add, p, v, acq_rel) 1031 #define KMP_ATOMIC_SUB(p, v) KMP_ATOMIC_OP(fetch_sub, p, v, acq_rel) 1032 #define KMP_ATOMIC_AND(p, v) KMP_ATOMIC_OP(fetch_and, p, v, acq_rel) 1033 #define KMP_ATOMIC_OR(p, v) KMP_ATOMIC_OP(fetch_or, p, v, acq_rel) 1034 #define KMP_ATOMIC_INC(p) KMP_ATOMIC_OP(fetch_add, p, 1, acq_rel) 1035 #define KMP_ATOMIC_DEC(p) KMP_ATOMIC_OP(fetch_sub, p, 1, acq_rel) 1036 #define KMP_ATOMIC_ADD_RLX(p, v) KMP_ATOMIC_OP(fetch_add, p, v, relaxed) 1037 #define KMP_ATOMIC_INC_RLX(p) KMP_ATOMIC_OP(fetch_add, p, 1, relaxed) 1038 1039 // Callers of the following functions cannot see the side effect on "expected". 1040 template <typename T> 1041 bool __kmp_atomic_compare_store(std::atomic<T> *p, T expected, T desired) { 1042 return p->compare_exchange_strong( 1043 expected, desired, std::memory_order_acq_rel, std::memory_order_relaxed); 1044 } 1045 1046 template <typename T> 1047 bool __kmp_atomic_compare_store_acq(std::atomic<T> *p, T expected, T desired) { 1048 return p->compare_exchange_strong( 1049 expected, desired, std::memory_order_acquire, std::memory_order_relaxed); 1050 } 1051 1052 template <typename T> 1053 bool __kmp_atomic_compare_store_rel(std::atomic<T> *p, T expected, T desired) { 1054 return p->compare_exchange_strong( 1055 expected, desired, std::memory_order_release, std::memory_order_relaxed); 1056 } 1057 1058 #endif /* KMP_OS_H */ 1059 // Safe C API 1060 #include "kmp_safe_c_api.h" 1061