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