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#include <string.h>
21
22#define KMP_FTN_PLAIN 1
23#define KMP_FTN_APPEND 2
24#define KMP_FTN_UPPER 3
25/*
26#define KMP_FTN_PREPEND 4
27#define KMP_FTN_UAPPEND 5
28*/
29
30#define KMP_PTR_SKIP (sizeof(void *))
31
32/* -------------------------- Compiler variations ------------------------ */
33
34#define KMP_OFF 0
35#define KMP_ON 1
36
37#define KMP_MEM_CONS_VOLATILE 0
38#define KMP_MEM_CONS_FENCE 1
39
40#ifndef KMP_MEM_CONS_MODEL
41#define KMP_MEM_CONS_MODEL KMP_MEM_CONS_VOLATILE
42#endif
43
44#ifndef __has_cpp_attribute
45#define __has_cpp_attribute(x) 0
46#endif
47
48#ifndef __has_attribute
49#define __has_attribute(x) 0
50#endif
51
52/* ------------------------- Compiler recognition ---------------------- */
53#define KMP_COMPILER_ICC 0
54#define KMP_COMPILER_GCC 0
55#define KMP_COMPILER_CLANG 0
56#define KMP_COMPILER_MSVC 0
57#define KMP_COMPILER_ICX 0
58
59#if __INTEL_CLANG_COMPILER
60#undef KMP_COMPILER_ICX
61#define KMP_COMPILER_ICX 1
62#elif defined(__INTEL_COMPILER)
63#undef KMP_COMPILER_ICC
64#define KMP_COMPILER_ICC 1
65#elif defined(__clang__)
66#undef KMP_COMPILER_CLANG
67#define KMP_COMPILER_CLANG 1
68#elif defined(__GNUC__)
69#undef KMP_COMPILER_GCC
70#define KMP_COMPILER_GCC 1
71#elif defined(_MSC_VER)
72#undef KMP_COMPILER_MSVC
73#define KMP_COMPILER_MSVC 1
74#else
75#error Unknown compiler
76#endif
77
78#if (KMP_OS_LINUX || KMP_OS_WINDOWS || KMP_OS_FREEBSD || KMP_OS_NETBSD || \
79 KMP_OS_DRAGONFLY || KMP_OS_AIX) && \
80 !KMP_OS_WASI && !KMP_OS_EMSCRIPTEN
81#define KMP_AFFINITY_SUPPORTED 1
82#if KMP_OS_WINDOWS && KMP_ARCH_X86_64
83#define KMP_GROUP_AFFINITY 1
84#else
85#define KMP_GROUP_AFFINITY 0
86#endif
87#else
88#define KMP_AFFINITY_SUPPORTED 0
89#define KMP_GROUP_AFFINITY 0
90#endif
91
92#if (KMP_OS_LINUX || (KMP_OS_FREEBSD && __FreeBSD_version >= 1301000))
93#define KMP_HAVE_SCHED_GETCPU 1
94#else
95#define KMP_HAVE_SCHED_GETCPU 0
96#endif
97
98/* Check for quad-precision extension. */
99#define KMP_HAVE_QUAD 0
100#if KMP_ARCH_X86 || KMP_ARCH_X86_64
101#if KMP_COMPILER_ICC || KMP_COMPILER_ICX
102/* _Quad is already defined for icc */
103#undef KMP_HAVE_QUAD
104#define KMP_HAVE_QUAD 1
105#elif KMP_COMPILER_CLANG
106/* Clang doesn't support a software-implemented
107 128-bit extended precision type yet */
108typedef long double _Quad;
109#elif KMP_COMPILER_GCC
110/* GCC on NetBSD lacks __multc3/__divtc3 builtins needed for quad until
111 NetBSD 10.0 which ships with GCC 10.5 */
112#if (!KMP_OS_NETBSD || __GNUC__ >= 10)
113typedef __float128 _Quad;
114#undef KMP_HAVE_QUAD
115#define KMP_HAVE_QUAD 1
116#endif
117#elif KMP_COMPILER_MSVC
118typedef long double _Quad;
119#endif
120#else
121#if __LDBL_MAX_EXP__ >= 16384 && KMP_COMPILER_GCC
122typedef long double _Quad;
123#undef KMP_HAVE_QUAD
124#define KMP_HAVE_QUAD 1
125#endif
126#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
127
128#define KMP_USE_X87CONTROL 0
129#if KMP_OS_WINDOWS
130#define KMP_END_OF_LINE "\r\n"
131typedef char kmp_int8;
132typedef unsigned char kmp_uint8;
133typedef short kmp_int16;
134typedef unsigned short kmp_uint16;
135typedef int kmp_int32;
136typedef unsigned int kmp_uint32;
137#define KMP_INT32_SPEC "d"
138#define KMP_UINT32_SPEC "u"
139#ifndef KMP_STRUCT64
140typedef __int64 kmp_int64;
141typedef unsigned __int64 kmp_uint64;
142#define KMP_INT64_SPEC "I64d"
143#define KMP_UINT64_SPEC "I64u"
144#else
145struct kmp_struct64 {
146 kmp_int32 a, b;
147};
148typedef struct kmp_struct64 kmp_int64;
149typedef struct kmp_struct64 kmp_uint64;
150/* Not sure what to use for KMP_[U]INT64_SPEC here */
151#endif
152#if KMP_ARCH_X86 && KMP_MSVC_COMPAT
153#undef KMP_USE_X87CONTROL
154#define KMP_USE_X87CONTROL 1
155#endif
156#if KMP_ARCH_X86_64 || KMP_ARCH_AARCH64
157#define KMP_INTPTR 1
158typedef __int64 kmp_intptr_t;
159typedef unsigned __int64 kmp_uintptr_t;
160#define KMP_INTPTR_SPEC "I64d"
161#define KMP_UINTPTR_SPEC "I64u"
162#endif
163#endif /* KMP_OS_WINDOWS */
164
165#if KMP_OS_UNIX
166#define KMP_END_OF_LINE "\n"
167typedef char kmp_int8;
168typedef unsigned char kmp_uint8;
169typedef short kmp_int16;
170typedef unsigned short kmp_uint16;
171typedef int kmp_int32;
172typedef unsigned int kmp_uint32;
173typedef long long kmp_int64;
174typedef unsigned long long kmp_uint64;
175#define KMP_INT32_SPEC "d"
176#define KMP_UINT32_SPEC "u"
177#define KMP_INT64_SPEC "lld"
178#define KMP_UINT64_SPEC "llu"
179#endif /* KMP_OS_UNIX */
180
181#if KMP_ARCH_X86 || KMP_ARCH_ARM || KMP_ARCH_MIPS || KMP_ARCH_WASM || \
182 KMP_ARCH_PPC || KMP_ARCH_AARCH64_32 || KMP_ARCH_SPARC32
183#define KMP_SIZE_T_SPEC KMP_UINT32_SPEC
184#elif KMP_ARCH_X86_64 || KMP_ARCH_PPC64 || KMP_ARCH_AARCH64 || \
185 KMP_ARCH_MIPS64 || KMP_ARCH_RISCV64 || KMP_ARCH_LOONGARCH64 || \
186 KMP_ARCH_VE || KMP_ARCH_S390X || KMP_ARCH_SPARC64
187#define KMP_SIZE_T_SPEC KMP_UINT64_SPEC
188#else
189#error "Can't determine size_t printf format specifier."
190#endif
191
192#if KMP_ARCH_X86 || KMP_ARCH_ARM || KMP_ARCH_WASM || KMP_ARCH_PPC
193#define KMP_SIZE_T_MAX (0xFFFFFFFF)
194#else
195#define KMP_SIZE_T_MAX (0xFFFFFFFFFFFFFFFF)
196#endif
197
198typedef size_t kmp_size_t;
199typedef float kmp_real32;
200typedef double kmp_real64;
201
202#ifndef KMP_INTPTR
203#define KMP_INTPTR 1
204typedef long kmp_intptr_t;
205typedef unsigned long kmp_uintptr_t;
206#define KMP_INTPTR_SPEC "ld"
207#define KMP_UINTPTR_SPEC "lu"
208#endif
209
210#ifdef BUILD_I8
211typedef kmp_int64 kmp_int;
212typedef kmp_uint64 kmp_uint;
213#else
214typedef kmp_int32 kmp_int;
215typedef kmp_uint32 kmp_uint;
216#endif /* BUILD_I8 */
217#define KMP_INT_MAX ((kmp_int32)0x7FFFFFFF)
218#define KMP_INT_MIN ((kmp_int32)0x80000000)
219
220// stdarg handling
221#if (KMP_ARCH_ARM || KMP_ARCH_X86_64 || KMP_ARCH_AARCH64 || KMP_ARCH_WASM) && \
222 (KMP_OS_FREEBSD || KMP_OS_NETBSD || KMP_OS_OPENBSD || KMP_OS_DRAGONFLY || \
223 KMP_OS_LINUX || KMP_OS_WASI)
224typedef va_list *kmp_va_list;
225#define kmp_va_deref(ap) (*(ap))
226#define kmp_va_addr_of(ap) (&(ap))
227#else
228typedef va_list kmp_va_list;
229#define kmp_va_deref(ap) (ap)
230#define kmp_va_addr_of(ap) (ap)
231#endif
232
233#ifdef __cplusplus
234// macros to cast out qualifiers and to re-interpret types
235#define CCAST(type, var) const_cast<type>(var)
236#define RCAST(type, var) reinterpret_cast<type>(var)
237//-------------------------------------------------------------------------
238// template for debug prints specification ( d, u, lld, llu ), and to obtain
239// signed/unsigned flavors of a type
240template <typename T> struct traits_t {};
241// int
242template <> struct traits_t<signed int> {
243 typedef signed int signed_t;
244 typedef unsigned int unsigned_t;
245 typedef double floating_t;
246 static char const *spec;
247 static const signed_t max_value = 0x7fffffff;
248 static const signed_t min_value = 0x80000000;
249 static const int type_size = sizeof(signed_t);
250};
251// unsigned int
252template <> struct traits_t<unsigned int> {
253 typedef signed int signed_t;
254 typedef unsigned int unsigned_t;
255 typedef double floating_t;
256 static char const *spec;
257 static const unsigned_t max_value = 0xffffffff;
258 static const unsigned_t min_value = 0x00000000;
259 static const int type_size = sizeof(unsigned_t);
260};
261// long
262template <> struct traits_t<signed long> {
263 typedef signed long signed_t;
264 typedef unsigned long unsigned_t;
265 typedef long double floating_t;
266 static char const *spec;
267 static const int type_size = sizeof(signed_t);
268};
269// long long
270template <> struct traits_t<signed long long> {
271 typedef signed long long signed_t;
272 typedef unsigned long long unsigned_t;
273 typedef long double floating_t;
274 static char const *spec;
275 static const signed_t max_value = 0x7fffffffffffffffLL;
276 static const signed_t min_value = 0x8000000000000000LL;
277 static const int type_size = sizeof(signed_t);
278};
279// unsigned long long
280template <> struct traits_t<unsigned long long> {
281 typedef signed long long signed_t;
282 typedef unsigned long long unsigned_t;
283 typedef long double floating_t;
284 static char const *spec;
285 static const unsigned_t max_value = 0xffffffffffffffffLL;
286 static const unsigned_t min_value = 0x0000000000000000LL;
287 static const int type_size = sizeof(unsigned_t);
288};
289//-------------------------------------------------------------------------
290#else
291#define CCAST(type, var) (type)(var)
292#define RCAST(type, var) (type)(var)
293#endif // __cplusplus
294
295#define KMP_EXPORT extern /* export declaration in guide libraries */
296
297#if __GNUC__ >= 4 && !defined(__MINGW32__)
298#define __forceinline __inline
299#endif
300
301/* Check if the OS/arch can support user-level mwait */
302// All mwait code tests for UMWAIT first, so it should only fall back to ring3
303// MWAIT for KNL.
304#define KMP_HAVE_MWAIT \
305 ((KMP_ARCH_X86 || KMP_ARCH_X86_64) && (KMP_OS_LINUX || KMP_OS_WINDOWS) && \
306 !KMP_MIC2)
307#define KMP_HAVE_UMWAIT \
308 ((KMP_ARCH_X86 || KMP_ARCH_X86_64) && (KMP_OS_LINUX || KMP_OS_WINDOWS) && \
309 !KMP_MIC)
310
311#if KMP_OS_WINDOWS
312// Don't include everything related to NT status code, we'll do that explicitly
313#define WIN32_NO_STATUS
314#include <windows.h>
315
316static inline int KMP_GET_PAGE_SIZE(void) {
317 SYSTEM_INFO si;
318 GetSystemInfo(&si);
319 return si.dwPageSize;
320}
321#else
322#define KMP_GET_PAGE_SIZE() getpagesize()
323#endif
324
325#define PAGE_ALIGNED(_addr) \
326 (!((size_t)_addr & (size_t)(KMP_GET_PAGE_SIZE() - 1)))
327#define ALIGN_TO_PAGE(x) \
328 (void *)(((size_t)(x)) & ~((size_t)(KMP_GET_PAGE_SIZE() - 1)))
329
330/* ---------- Support for cache alignment, padding, etc. ----------------*/
331
332#ifdef __cplusplus
333extern "C" {
334#endif // __cplusplus
335
336#define INTERNODE_CACHE_LINE 4096 /* for multi-node systems */
337
338/* Define the default size of the cache line */
339#ifndef CACHE_LINE
340#define CACHE_LINE 128 /* cache line size in bytes */
341#else
342#if (CACHE_LINE < 64) && !defined(KMP_OS_DARWIN)
343// 2006-02-13: This produces too many warnings on OS X*. Disable for now
344#warning CACHE_LINE is too small.
345#endif
346#endif /* CACHE_LINE */
347
348#define KMP_CACHE_PREFETCH(ADDR) /* nothing */
349
350// Define attribute that indicates that the fall through from the previous
351// case label is intentional and should not be diagnosed by a compiler
352// Code from libcxx/include/__config
353// Use a function like macro to imply that it must be followed by a semicolon
354#if __cplusplus > 201402L && __has_cpp_attribute(fallthrough)
355#define KMP_FALLTHROUGH() [[fallthrough]]
356// icc cannot properly tell this attribute is absent so force off
357#elif KMP_COMPILER_ICC
358#define KMP_FALLTHROUGH() ((void)0)
359#elif __has_cpp_attribute(clang::fallthrough)
360#define KMP_FALLTHROUGH() [[clang::fallthrough]]
361#elif __has_attribute(fallthrough) || __GNUC__ >= 7
362#define KMP_FALLTHROUGH() __attribute__((__fallthrough__))
363#else
364#define KMP_FALLTHROUGH() ((void)0)
365#endif
366
367#if KMP_HAVE_ATTRIBUTE_WAITPKG
368#define KMP_ATTRIBUTE_TARGET_WAITPKG __attribute__((target("waitpkg")))
369#else
370#define KMP_ATTRIBUTE_TARGET_WAITPKG /* Nothing */
371#endif
372
373#if KMP_HAVE_ATTRIBUTE_RTM
374#define KMP_ATTRIBUTE_TARGET_RTM __attribute__((target("rtm")))
375#else
376#define KMP_ATTRIBUTE_TARGET_RTM /* Nothing */
377#endif
378
379// Define attribute that indicates a function does not return
380#if __cplusplus >= 201103L
381#define KMP_NORETURN [[noreturn]]
382#elif KMP_OS_WINDOWS
383#define KMP_NORETURN __declspec(noreturn)
384#else
385#define KMP_NORETURN __attribute__((noreturn))
386#endif
387
388#if KMP_OS_WINDOWS && KMP_MSVC_COMPAT
389#define KMP_ALIGN(bytes) __declspec(align(bytes))
390#define KMP_THREAD_LOCAL __declspec(thread)
391#define KMP_ALIAS /* Nothing */
392#else
393#define KMP_ALIGN(bytes) __attribute__((aligned(bytes)))
394#define KMP_THREAD_LOCAL __thread
395#define KMP_ALIAS(alias_of) __attribute__((alias(alias_of)))
396#endif
397
398#if KMP_HAVE_WEAK_ATTRIBUTE && !KMP_DYNAMIC_LIB
399#define KMP_WEAK_ATTRIBUTE_EXTERNAL __attribute__((weak))
400#else
401#define KMP_WEAK_ATTRIBUTE_EXTERNAL /* Nothing */
402#endif
403
404#if KMP_HAVE_WEAK_ATTRIBUTE
405#define KMP_WEAK_ATTRIBUTE_INTERNAL __attribute__((weak))
406#else
407#define KMP_WEAK_ATTRIBUTE_INTERNAL /* Nothing */
408#endif
409
410// Define KMP_VERSION_SYMBOL and KMP_EXPAND_NAME
411#ifndef KMP_STR
412#define KMP_STR(x) _KMP_STR(x)
413#define _KMP_STR(x) #x
414#endif
415
416#ifdef KMP_USE_VERSION_SYMBOLS
417// If using versioned symbols, KMP_EXPAND_NAME prepends
418// __kmp_api_ to the real API name
419#define KMP_EXPAND_NAME(api_name) _KMP_EXPAND_NAME(api_name)
420#define _KMP_EXPAND_NAME(api_name) __kmp_api_##api_name
421#define KMP_VERSION_SYMBOL(api_name, ver_num, ver_str) \
422 _KMP_VERSION_SYMBOL(api_name, ver_num, ver_str, "VERSION")
423#define _KMP_VERSION_SYMBOL(api_name, ver_num, ver_str, default_ver) \
424 __typeof__(__kmp_api_##api_name) __kmp_api_##api_name##_##ver_num##_alias \
425 __attribute__((alias(KMP_STR(__kmp_api_##api_name)))); \
426 __asm__( \
427 ".symver " KMP_STR(__kmp_api_##api_name##_##ver_num##_alias) "," KMP_STR( \
428 api_name) "@" ver_str "\n\t"); \
429 __asm__(".symver " KMP_STR(__kmp_api_##api_name) "," KMP_STR( \
430 api_name) "@@" default_ver "\n\t")
431
432#define KMP_VERSION_OMPC_SYMBOL(apic_name, api_name, ver_num, ver_str) \
433 _KMP_VERSION_OMPC_SYMBOL(apic_name, api_name, ver_num, ver_str, "VERSION")
434#define _KMP_VERSION_OMPC_SYMBOL(apic_name, api_name, ver_num, ver_str, \
435 default_ver) \
436 __typeof__(__kmp_api_##apic_name) __kmp_api_##apic_name##_##ver_num##_alias \
437 __attribute__((alias(KMP_STR(__kmp_api_##apic_name)))); \
438 __asm__(".symver " KMP_STR(__kmp_api_##apic_name) "," KMP_STR( \
439 apic_name) "@@" default_ver "\n\t"); \
440 __asm__( \
441 ".symver " KMP_STR(__kmp_api_##apic_name##_##ver_num##_alias) "," KMP_STR( \
442 api_name) "@" ver_str "\n\t")
443
444#else // KMP_USE_VERSION_SYMBOLS
445#define KMP_EXPAND_NAME(api_name) api_name
446#define KMP_VERSION_SYMBOL(api_name, ver_num, ver_str) /* Nothing */
447#define KMP_VERSION_OMPC_SYMBOL(apic_name, api_name, ver_num, \
448 ver_str) /* Nothing */
449#endif // KMP_USE_VERSION_SYMBOLS
450
451/* Temporary note: if performance testing of this passes, we can remove
452 all references to KMP_DO_ALIGN and replace with KMP_ALIGN. */
453#define KMP_DO_ALIGN(bytes) KMP_ALIGN(bytes)
454#define KMP_ALIGN_CACHE KMP_ALIGN(CACHE_LINE)
455#define KMP_ALIGN_CACHE_INTERNODE KMP_ALIGN(INTERNODE_CACHE_LINE)
456
457/* General purpose fence types for memory operations */
458enum kmp_mem_fence_type {
459 kmp_no_fence, /* No memory fence */
460 kmp_acquire_fence, /* Acquire (read) memory fence */
461 kmp_release_fence, /* Release (write) memory fence */
462 kmp_full_fence /* Full (read+write) memory fence */
463};
464
465// Synchronization primitives
466
467#if KMP_ASM_INTRINS && KMP_OS_WINDOWS && !((KMP_ARCH_AARCH64 || KMP_ARCH_ARM) && (KMP_COMPILER_CLANG || KMP_COMPILER_GCC))
468
469#if KMP_MSVC_COMPAT && !KMP_COMPILER_CLANG
470#pragma intrinsic(InterlockedExchangeAdd)
471#pragma intrinsic(InterlockedCompareExchange)
472#pragma intrinsic(InterlockedExchange)
473#if !KMP_32_BIT_ARCH
474#pragma intrinsic(InterlockedExchange64)
475#endif
476#endif
477
478// Using InterlockedIncrement / InterlockedDecrement causes a library loading
479// ordering problem, so we use InterlockedExchangeAdd instead.
480#define KMP_TEST_THEN_INC32(p) InterlockedExchangeAdd((volatile long *)(p), 1)
481#define KMP_TEST_THEN_INC_ACQ32(p) \
482 InterlockedExchangeAdd((volatile long *)(p), 1)
483#define KMP_TEST_THEN_ADD4_32(p) InterlockedExchangeAdd((volatile long *)(p), 4)
484#define KMP_TEST_THEN_ADD4_ACQ32(p) \
485 InterlockedExchangeAdd((volatile long *)(p), 4)
486#define KMP_TEST_THEN_DEC32(p) InterlockedExchangeAdd((volatile long *)(p), -1)
487#define KMP_TEST_THEN_DEC_ACQ32(p) \
488 InterlockedExchangeAdd((volatile long *)(p), -1)
489#define KMP_TEST_THEN_ADD32(p, v) \
490 InterlockedExchangeAdd((volatile long *)(p), (v))
491
492#define KMP_COMPARE_AND_STORE_RET32(p, cv, sv) \
493 InterlockedCompareExchange((volatile long *)(p), (long)(sv), (long)(cv))
494
495#define KMP_XCHG_FIXED32(p, v) \
496 InterlockedExchange((volatile long *)(p), (long)(v))
497#define KMP_XCHG_FIXED64(p, v) \
498 InterlockedExchange64((volatile kmp_int64 *)(p), (kmp_int64)(v))
499
500inline kmp_real32 KMP_XCHG_REAL32(volatile kmp_real32 *p, kmp_real32 v) {
501 kmp_int32 tmp = InterlockedExchange((volatile long *)p, *(long *)&v);
502 return *(kmp_real32 *)&tmp;
503}
504
505#define KMP_TEST_THEN_OR8(p, v) __kmp_test_then_or8((p), (v))
506#define KMP_TEST_THEN_AND8(p, v) __kmp_test_then_and8((p), (v))
507#define KMP_TEST_THEN_OR32(p, v) __kmp_test_then_or32((p), (v))
508#define KMP_TEST_THEN_AND32(p, v) __kmp_test_then_and32((p), (v))
509#define KMP_TEST_THEN_OR64(p, v) __kmp_test_then_or64((p), (v))
510#define KMP_TEST_THEN_AND64(p, v) __kmp_test_then_and64((p), (v))
511
512extern kmp_int8 __kmp_test_then_or8(volatile kmp_int8 *p, kmp_int8 v);
513extern kmp_int8 __kmp_test_then_and8(volatile kmp_int8 *p, kmp_int8 v);
514extern kmp_int32 __kmp_test_then_add32(volatile kmp_int32 *p, kmp_int32 v);
515extern kmp_uint32 __kmp_test_then_or32(volatile kmp_uint32 *p, kmp_uint32 v);
516extern kmp_uint32 __kmp_test_then_and32(volatile kmp_uint32 *p, kmp_uint32 v);
517extern kmp_int64 __kmp_test_then_add64(volatile kmp_int64 *p, kmp_int64 v);
518extern kmp_uint64 __kmp_test_then_or64(volatile kmp_uint64 *p, kmp_uint64 v);
519extern kmp_uint64 __kmp_test_then_and64(volatile kmp_uint64 *p, kmp_uint64 v);
520
521#if KMP_ARCH_AARCH64 && KMP_COMPILER_MSVC && !KMP_COMPILER_CLANG
522#define KMP_TEST_THEN_INC64(p) _InterlockedExchangeAdd64((p), 1LL)
523#define KMP_TEST_THEN_INC_ACQ64(p) _InterlockedExchangeAdd64_acq((p), 1LL)
524#define KMP_TEST_THEN_ADD4_64(p) _InterlockedExchangeAdd64((p), 4LL)
525// #define KMP_TEST_THEN_ADD4_ACQ64(p) _InterlockedExchangeAdd64_acq((p), 4LL)
526// #define KMP_TEST_THEN_DEC64(p) _InterlockedExchangeAdd64((p), -1LL)
527// #define KMP_TEST_THEN_DEC_ACQ64(p) _InterlockedExchangeAdd64_acq((p), -1LL)
528// #define KMP_TEST_THEN_ADD8(p, v) _InterlockedExchangeAdd8((p), (v))
529#define KMP_TEST_THEN_ADD64(p, v) _InterlockedExchangeAdd64((p), (v))
530
531#define KMP_COMPARE_AND_STORE_ACQ8(p, cv, sv) \
532 __kmp_compare_and_store_acq8((p), (cv), (sv))
533#define KMP_COMPARE_AND_STORE_REL8(p, cv, sv) \
534 __kmp_compare_and_store_rel8((p), (cv), (sv))
535#define KMP_COMPARE_AND_STORE_ACQ16(p, cv, sv) \
536 __kmp_compare_and_store_acq16((p), (cv), (sv))
537/*
538#define KMP_COMPARE_AND_STORE_REL16(p, cv, sv) \
539 __kmp_compare_and_store_rel16((p), (cv), (sv))
540*/
541#define KMP_COMPARE_AND_STORE_ACQ32(p, cv, sv) \
542 __kmp_compare_and_store_acq32((volatile kmp_int32 *)(p), (kmp_int32)(cv), \
543 (kmp_int32)(sv))
544#define KMP_COMPARE_AND_STORE_REL32(p, cv, sv) \
545 __kmp_compare_and_store_rel32((volatile kmp_int32 *)(p), (kmp_int32)(cv), \
546 (kmp_int32)(sv))
547#define KMP_COMPARE_AND_STORE_ACQ64(p, cv, sv) \
548 __kmp_compare_and_store_acq64((volatile kmp_int64 *)(p), (kmp_int64)(cv), \
549 (kmp_int64)(sv))
550#define KMP_COMPARE_AND_STORE_REL64(p, cv, sv) \
551 __kmp_compare_and_store_rel64((volatile kmp_int64 *)(p), (kmp_int64)(cv), \
552 (kmp_int64)(sv))
553#define KMP_COMPARE_AND_STORE_PTR(p, cv, sv) \
554 __kmp_compare_and_store_ptr((void *volatile *)(p), (void *)(cv), (void *)(sv))
555
556// KMP_COMPARE_AND_STORE expects this order: pointer, compare, exchange
557// _InterlockedCompareExchange expects this order: pointer, exchange, compare
558// KMP_COMPARE_AND_STORE also returns a bool indicating a successful write. A
559// write is successful if the return value of _InterlockedCompareExchange is the
560// same as the compare value.
561inline kmp_int8 __kmp_compare_and_store_acq8(volatile kmp_int8 *p, kmp_int8 cv,
562 kmp_int8 sv) {
563 return _InterlockedCompareExchange8_acq(p, sv, cv) == cv;
564}
565
566inline kmp_int8 __kmp_compare_and_store_rel8(volatile kmp_int8 *p, kmp_int8 cv,
567 kmp_int8 sv) {
568 return _InterlockedCompareExchange8_rel(p, sv, cv) == cv;
569}
570
571inline kmp_int16 __kmp_compare_and_store_acq16(volatile kmp_int16 *p,
572 kmp_int16 cv, kmp_int16 sv) {
573 return _InterlockedCompareExchange16_acq(p, sv, cv) == cv;
574}
575
576inline kmp_int16 __kmp_compare_and_store_rel16(volatile kmp_int16 *p,
577 kmp_int16 cv, kmp_int16 sv) {
578 return _InterlockedCompareExchange16_rel(p, sv, cv) == cv;
579}
580
581inline kmp_int32 __kmp_compare_and_store_acq32(volatile kmp_int32 *p,
582 kmp_int32 cv, kmp_int32 sv) {
583 return _InterlockedCompareExchange_acq((volatile long *)p, sv, cv) == cv;
584}
585
586inline kmp_int32 __kmp_compare_and_store_rel32(volatile kmp_int32 *p,
587 kmp_int32 cv, kmp_int32 sv) {
588 return _InterlockedCompareExchange_rel((volatile long *)p, sv, cv) == cv;
589}
590
591inline kmp_int32 __kmp_compare_and_store_acq64(volatile kmp_int64 *p,
592 kmp_int64 cv, kmp_int64 sv) {
593 return _InterlockedCompareExchange64_acq(p, sv, cv) == cv;
594}
595
596inline kmp_int32 __kmp_compare_and_store_rel64(volatile kmp_int64 *p,
597 kmp_int64 cv, kmp_int64 sv) {
598 return _InterlockedCompareExchange64_rel(p, sv, cv) == cv;
599}
600
601inline kmp_int32 __kmp_compare_and_store_ptr(void *volatile *p, void *cv,
602 void *sv) {
603 return _InterlockedCompareExchangePointer(p, sv, cv) == cv;
604}
605
606// The _RET versions return the value instead of a bool
607
608#define KMP_COMPARE_AND_STORE_RET8(p, cv, sv) \
609 _InterlockedCompareExchange8((p), (sv), (cv))
610#define KMP_COMPARE_AND_STORE_RET16(p, cv, sv) \
611 _InterlockedCompareExchange16((p), (sv), (cv))
612
613#define KMP_COMPARE_AND_STORE_RET64(p, cv, sv) \
614 _InterlockedCompareExchange64((volatile kmp_int64 *)(p), (kmp_int64)(sv), \
615 (kmp_int64)(cv))
616
617
618#define KMP_XCHG_FIXED8(p, v) \
619 _InterlockedExchange8((volatile kmp_int8 *)(p), (kmp_int8)(v));
620#define KMP_XCHG_FIXED16(p, v) _InterlockedExchange16((p), (v));
621#define KMP_XCHG_REAL64(p, v) __kmp_xchg_real64((p), (v));
622
623inline kmp_real64 __kmp_xchg_real64(volatile kmp_real64 *p, kmp_real64 v) {
624 kmp_int64 tmp = _InterlockedExchange64((volatile kmp_int64 *)p, *(kmp_int64
625 *)&v); return *(kmp_real64 *)&tmp;
626}
627
628#else // !KMP_ARCH_AARCH64
629
630// Routines that we still need to implement in assembly.
631extern kmp_int8 __kmp_test_then_add8(volatile kmp_int8 *p, kmp_int8 v);
632
633extern kmp_int8 __kmp_compare_and_store8(volatile kmp_int8 *p, kmp_int8 cv,
634 kmp_int8 sv);
635extern kmp_int16 __kmp_compare_and_store16(volatile kmp_int16 *p, kmp_int16 cv,
636 kmp_int16 sv);
637extern kmp_int32 __kmp_compare_and_store32(volatile kmp_int32 *p, kmp_int32 cv,
638 kmp_int32 sv);
639extern kmp_int32 __kmp_compare_and_store64(volatile kmp_int64 *p, kmp_int64 cv,
640 kmp_int64 sv);
641extern kmp_int8 __kmp_compare_and_store_ret8(volatile kmp_int8 *p, kmp_int8 cv,
642 kmp_int8 sv);
643extern kmp_int16 __kmp_compare_and_store_ret16(volatile kmp_int16 *p,
644 kmp_int16 cv, kmp_int16 sv);
645extern kmp_int32 __kmp_compare_and_store_ret32(volatile kmp_int32 *p,
646 kmp_int32 cv, kmp_int32 sv);
647extern kmp_int64 __kmp_compare_and_store_ret64(volatile kmp_int64 *p,
648 kmp_int64 cv, kmp_int64 sv);
649
650extern kmp_int8 __kmp_xchg_fixed8(volatile kmp_int8 *p, kmp_int8 v);
651extern kmp_int16 __kmp_xchg_fixed16(volatile kmp_int16 *p, kmp_int16 v);
652extern kmp_int32 __kmp_xchg_fixed32(volatile kmp_int32 *p, kmp_int32 v);
653extern kmp_int64 __kmp_xchg_fixed64(volatile kmp_int64 *p, kmp_int64 v);
654extern kmp_real32 __kmp_xchg_real32(volatile kmp_real32 *p, kmp_real32 v);
655extern kmp_real64 __kmp_xchg_real64(volatile kmp_real64 *p, kmp_real64 v);
656
657//#define KMP_TEST_THEN_INC32(p) __kmp_test_then_add32((p), 1)
658//#define KMP_TEST_THEN_INC_ACQ32(p) __kmp_test_then_add32((p), 1)
659#define KMP_TEST_THEN_INC64(p) __kmp_test_then_add64((p), 1LL)
660#define KMP_TEST_THEN_INC_ACQ64(p) __kmp_test_then_add64((p), 1LL)
661//#define KMP_TEST_THEN_ADD4_32(p) __kmp_test_then_add32((p), 4)
662//#define KMP_TEST_THEN_ADD4_ACQ32(p) __kmp_test_then_add32((p), 4)
663#define KMP_TEST_THEN_ADD4_64(p) __kmp_test_then_add64((p), 4LL)
664#define KMP_TEST_THEN_ADD4_ACQ64(p) __kmp_test_then_add64((p), 4LL)
665//#define KMP_TEST_THEN_DEC32(p) __kmp_test_then_add32((p), -1)
666//#define KMP_TEST_THEN_DEC_ACQ32(p) __kmp_test_then_add32((p), -1)
667#define KMP_TEST_THEN_DEC64(p) __kmp_test_then_add64((p), -1LL)
668#define KMP_TEST_THEN_DEC_ACQ64(p) __kmp_test_then_add64((p), -1LL)
669//#define KMP_TEST_THEN_ADD32(p, v) __kmp_test_then_add32((p), (v))
670#define KMP_TEST_THEN_ADD8(p, v) __kmp_test_then_add8((p), (v))
671#define KMP_TEST_THEN_ADD64(p, v) __kmp_test_then_add64((p), (v))
672
673
674#define KMP_COMPARE_AND_STORE_ACQ8(p, cv, sv) \
675 __kmp_compare_and_store8((p), (cv), (sv))
676#define KMP_COMPARE_AND_STORE_REL8(p, cv, sv) \
677 __kmp_compare_and_store8((p), (cv), (sv))
678#define KMP_COMPARE_AND_STORE_ACQ16(p, cv, sv) \
679 __kmp_compare_and_store16((p), (cv), (sv))
680#define KMP_COMPARE_AND_STORE_REL16(p, cv, sv) \
681 __kmp_compare_and_store16((p), (cv), (sv))
682#define KMP_COMPARE_AND_STORE_ACQ32(p, cv, sv) \
683 __kmp_compare_and_store32((volatile kmp_int32 *)(p), (kmp_int32)(cv), \
684 (kmp_int32)(sv))
685#define KMP_COMPARE_AND_STORE_REL32(p, cv, sv) \
686 __kmp_compare_and_store32((volatile kmp_int32 *)(p), (kmp_int32)(cv), \
687 (kmp_int32)(sv))
688#define KMP_COMPARE_AND_STORE_ACQ64(p, cv, sv) \
689 __kmp_compare_and_store64((volatile kmp_int64 *)(p), (kmp_int64)(cv), \
690 (kmp_int64)(sv))
691#define KMP_COMPARE_AND_STORE_REL64(p, cv, sv) \
692 __kmp_compare_and_store64((volatile kmp_int64 *)(p), (kmp_int64)(cv), \
693 (kmp_int64)(sv))
694
695#if KMP_ARCH_X86
696#define KMP_COMPARE_AND_STORE_PTR(p, cv, sv) \
697 __kmp_compare_and_store32((volatile kmp_int32 *)(p), (kmp_int32)(cv), \
698 (kmp_int32)(sv))
699#else /* 64 bit pointers */
700#define KMP_COMPARE_AND_STORE_PTR(p, cv, sv) \
701 __kmp_compare_and_store64((volatile kmp_int64 *)(p), (kmp_int64)(cv), \
702 (kmp_int64)(sv))
703#endif /* KMP_ARCH_X86 */
704
705#define KMP_COMPARE_AND_STORE_RET8(p, cv, sv) \
706 __kmp_compare_and_store_ret8((p), (cv), (sv))
707#define KMP_COMPARE_AND_STORE_RET16(p, cv, sv) \
708 __kmp_compare_and_store_ret16((p), (cv), (sv))
709#define KMP_COMPARE_AND_STORE_RET64(p, cv, sv) \
710 __kmp_compare_and_store_ret64((volatile kmp_int64 *)(p), (kmp_int64)(cv), \
711 (kmp_int64)(sv))
712
713#define KMP_XCHG_FIXED8(p, v) \
714 __kmp_xchg_fixed8((volatile kmp_int8 *)(p), (kmp_int8)(v));
715#define KMP_XCHG_FIXED16(p, v) __kmp_xchg_fixed16((p), (v));
716//#define KMP_XCHG_FIXED32(p, v) __kmp_xchg_fixed32((p), (v));
717//#define KMP_XCHG_FIXED64(p, v) __kmp_xchg_fixed64((p), (v));
718//#define KMP_XCHG_REAL32(p, v) __kmp_xchg_real32((p), (v));
719#define KMP_XCHG_REAL64(p, v) __kmp_xchg_real64((p), (v));
720#endif
721
722#elif (KMP_ASM_INTRINS && KMP_OS_UNIX) || !(KMP_ARCH_X86 || KMP_ARCH_X86_64)
723
724/* cast p to correct type so that proper intrinsic will be used */
725#define KMP_TEST_THEN_INC32(p) \
726 __sync_fetch_and_add((volatile kmp_int32 *)(p), 1)
727#define KMP_TEST_THEN_INC_ACQ32(p) \
728 __sync_fetch_and_add((volatile kmp_int32 *)(p), 1)
729#if KMP_ARCH_MIPS
730#define KMP_TEST_THEN_INC64(p) \
731 __atomic_fetch_add((volatile kmp_int64 *)(p), 1LL, __ATOMIC_SEQ_CST)
732#define KMP_TEST_THEN_INC_ACQ64(p) \
733 __atomic_fetch_add((volatile kmp_int64 *)(p), 1LL, __ATOMIC_SEQ_CST)
734#else
735#define KMP_TEST_THEN_INC64(p) \
736 __sync_fetch_and_add((volatile kmp_int64 *)(p), 1LL)
737#define KMP_TEST_THEN_INC_ACQ64(p) \
738 __sync_fetch_and_add((volatile kmp_int64 *)(p), 1LL)
739#endif
740#define KMP_TEST_THEN_ADD4_32(p) \
741 __sync_fetch_and_add((volatile kmp_int32 *)(p), 4)
742#define KMP_TEST_THEN_ADD4_ACQ32(p) \
743 __sync_fetch_and_add((volatile kmp_int32 *)(p), 4)
744#if KMP_ARCH_MIPS
745#define KMP_TEST_THEN_ADD4_64(p) \
746 __atomic_fetch_add((volatile kmp_int64 *)(p), 4LL, __ATOMIC_SEQ_CST)
747#define KMP_TEST_THEN_ADD4_ACQ64(p) \
748 __atomic_fetch_add((volatile kmp_int64 *)(p), 4LL, __ATOMIC_SEQ_CST)
749#define KMP_TEST_THEN_DEC64(p) \
750 __atomic_fetch_sub((volatile kmp_int64 *)(p), 1LL, __ATOMIC_SEQ_CST)
751#define KMP_TEST_THEN_DEC_ACQ64(p) \
752 __atomic_fetch_sub((volatile kmp_int64 *)(p), 1LL, __ATOMIC_SEQ_CST)
753#else
754#define KMP_TEST_THEN_ADD4_64(p) \
755 __sync_fetch_and_add((volatile kmp_int64 *)(p), 4LL)
756#define KMP_TEST_THEN_ADD4_ACQ64(p) \
757 __sync_fetch_and_add((volatile kmp_int64 *)(p), 4LL)
758#define KMP_TEST_THEN_DEC64(p) \
759 __sync_fetch_and_sub((volatile kmp_int64 *)(p), 1LL)
760#define KMP_TEST_THEN_DEC_ACQ64(p) \
761 __sync_fetch_and_sub((volatile kmp_int64 *)(p), 1LL)
762#endif
763#define KMP_TEST_THEN_DEC32(p) \
764 __sync_fetch_and_sub((volatile kmp_int32 *)(p), 1)
765#define KMP_TEST_THEN_DEC_ACQ32(p) \
766 __sync_fetch_and_sub((volatile kmp_int32 *)(p), 1)
767#define KMP_TEST_THEN_ADD8(p, v) \
768 __sync_fetch_and_add((volatile kmp_int8 *)(p), (kmp_int8)(v))
769#define KMP_TEST_THEN_ADD32(p, v) \
770 __sync_fetch_and_add((volatile kmp_int32 *)(p), (kmp_int32)(v))
771#if KMP_ARCH_MIPS
772#define KMP_TEST_THEN_ADD64(p, v) \
773 __atomic_fetch_add((volatile kmp_uint64 *)(p), (kmp_uint64)(v), \
774 __ATOMIC_SEQ_CST)
775#else
776#define KMP_TEST_THEN_ADD64(p, v) \
777 __sync_fetch_and_add((volatile kmp_int64 *)(p), (kmp_int64)(v))
778#endif
779
780#define KMP_TEST_THEN_OR8(p, v) \
781 __sync_fetch_and_or((volatile kmp_int8 *)(p), (kmp_int8)(v))
782#define KMP_TEST_THEN_AND8(p, v) \
783 __sync_fetch_and_and((volatile kmp_int8 *)(p), (kmp_int8)(v))
784#define KMP_TEST_THEN_OR32(p, v) \
785 __sync_fetch_and_or((volatile kmp_uint32 *)(p), (kmp_uint32)(v))
786#define KMP_TEST_THEN_AND32(p, v) \
787 __sync_fetch_and_and((volatile kmp_uint32 *)(p), (kmp_uint32)(v))
788#if KMP_ARCH_MIPS
789#define KMP_TEST_THEN_OR64(p, v) \
790 __atomic_fetch_or((volatile kmp_uint64 *)(p), (kmp_uint64)(v), \
791 __ATOMIC_SEQ_CST)
792#define KMP_TEST_THEN_AND64(p, v) \
793 __atomic_fetch_and((volatile kmp_uint64 *)(p), (kmp_uint64)(v), \
794 __ATOMIC_SEQ_CST)
795#else
796#define KMP_TEST_THEN_OR64(p, v) \
797 __sync_fetch_and_or((volatile kmp_uint64 *)(p), (kmp_uint64)(v))
798#define KMP_TEST_THEN_AND64(p, v) \
799 __sync_fetch_and_and((volatile kmp_uint64 *)(p), (kmp_uint64)(v))
800#endif
801
802#define KMP_COMPARE_AND_STORE_ACQ8(p, cv, sv) \
803 __sync_bool_compare_and_swap((volatile kmp_uint8 *)(p), (kmp_uint8)(cv), \
804 (kmp_uint8)(sv))
805#define KMP_COMPARE_AND_STORE_REL8(p, cv, sv) \
806 __sync_bool_compare_and_swap((volatile kmp_uint8 *)(p), (kmp_uint8)(cv), \
807 (kmp_uint8)(sv))
808#define KMP_COMPARE_AND_STORE_ACQ16(p, cv, sv) \
809 __sync_bool_compare_and_swap((volatile kmp_uint16 *)(p), (kmp_uint16)(cv), \
810 (kmp_uint16)(sv))
811#define KMP_COMPARE_AND_STORE_REL16(p, cv, sv) \
812 __sync_bool_compare_and_swap((volatile kmp_uint16 *)(p), (kmp_uint16)(cv), \
813 (kmp_uint16)(sv))
814#define KMP_COMPARE_AND_STORE_ACQ32(p, cv, sv) \
815 __sync_bool_compare_and_swap((volatile kmp_uint32 *)(p), (kmp_uint32)(cv), \
816 (kmp_uint32)(sv))
817#define KMP_COMPARE_AND_STORE_REL32(p, cv, sv) \
818 __sync_bool_compare_and_swap((volatile kmp_uint32 *)(p), (kmp_uint32)(cv), \
819 (kmp_uint32)(sv))
820#define KMP_COMPARE_AND_STORE_PTR(p, cv, sv) \
821 __sync_bool_compare_and_swap((void *volatile *)(p), (void *)(cv), \
822 (void *)(sv))
823
824#define KMP_COMPARE_AND_STORE_RET8(p, cv, sv) \
825 __sync_val_compare_and_swap((volatile kmp_uint8 *)(p), (kmp_uint8)(cv), \
826 (kmp_uint8)(sv))
827#define KMP_COMPARE_AND_STORE_RET16(p, cv, sv) \
828 __sync_val_compare_and_swap((volatile kmp_uint16 *)(p), (kmp_uint16)(cv), \
829 (kmp_uint16)(sv))
830#define KMP_COMPARE_AND_STORE_RET32(p, cv, sv) \
831 __sync_val_compare_and_swap((volatile kmp_uint32 *)(p), (kmp_uint32)(cv), \
832 (kmp_uint32)(sv))
833#if KMP_ARCH_MIPS
834static inline bool mips_sync_bool_compare_and_swap(volatile kmp_uint64 *p,
835 kmp_uint64 cv,
836 kmp_uint64 sv) {
837 return __atomic_compare_exchange(p, &cv, &sv, false, __ATOMIC_SEQ_CST,
838 __ATOMIC_SEQ_CST);
839}
840static inline bool mips_sync_val_compare_and_swap(volatile kmp_uint64 *p,
841 kmp_uint64 cv,
842 kmp_uint64 sv) {
843 __atomic_compare_exchange(p, &cv, &sv, false, __ATOMIC_SEQ_CST,
844 __ATOMIC_SEQ_CST);
845 return cv;
846}
847#define KMP_COMPARE_AND_STORE_ACQ64(p, cv, sv) \
848 mips_sync_bool_compare_and_swap((volatile kmp_uint64 *)(p), \
849 (kmp_uint64)(cv), (kmp_uint64)(sv))
850#define KMP_COMPARE_AND_STORE_REL64(p, cv, sv) \
851 mips_sync_bool_compare_and_swap((volatile kmp_uint64 *)(p), \
852 (kmp_uint64)(cv), (kmp_uint64)(sv))
853#define KMP_COMPARE_AND_STORE_RET64(p, cv, sv) \
854 mips_sync_val_compare_and_swap((volatile kmp_uint64 *)(p), (kmp_uint64)(cv), \
855 (kmp_uint64)(sv))
856#else
857#define KMP_COMPARE_AND_STORE_ACQ64(p, cv, sv) \
858 __sync_bool_compare_and_swap((volatile kmp_uint64 *)(p), (kmp_uint64)(cv), \
859 (kmp_uint64)(sv))
860#define KMP_COMPARE_AND_STORE_REL64(p, cv, sv) \
861 __sync_bool_compare_and_swap((volatile kmp_uint64 *)(p), (kmp_uint64)(cv), \
862 (kmp_uint64)(sv))
863#define KMP_COMPARE_AND_STORE_RET64(p, cv, sv) \
864 __sync_val_compare_and_swap((volatile kmp_uint64 *)(p), (kmp_uint64)(cv), \
865 (kmp_uint64)(sv))
866#endif
867
868#if KMP_OS_DARWIN && defined(__INTEL_COMPILER) && __INTEL_COMPILER >= 1800
869#define KMP_XCHG_FIXED8(p, v) \
870 __atomic_exchange_1((volatile kmp_uint8 *)(p), (kmp_uint8)(v), \
871 __ATOMIC_SEQ_CST)
872#else
873#define KMP_XCHG_FIXED8(p, v) \
874 __sync_lock_test_and_set((volatile kmp_uint8 *)(p), (kmp_uint8)(v))
875#endif
876#define KMP_XCHG_FIXED16(p, v) \
877 __sync_lock_test_and_set((volatile kmp_uint16 *)(p), (kmp_uint16)(v))
878#define KMP_XCHG_FIXED32(p, v) \
879 __sync_lock_test_and_set((volatile kmp_uint32 *)(p), (kmp_uint32)(v))
880#define KMP_XCHG_FIXED64(p, v) \
881 __sync_lock_test_and_set((volatile kmp_uint64 *)(p), (kmp_uint64)(v))
882
883inline kmp_real32 KMP_XCHG_REAL32(volatile kmp_real32 *p, kmp_real32 v) {
884 volatile kmp_uint32 *up;
885 kmp_uint32 uv;
886 memcpy(dest: &up, src: &p, n: sizeof(up));
887 memcpy(dest: &uv, src: &v, n: sizeof(uv));
888 kmp_int32 tmp = __sync_lock_test_and_set(up, uv);
889 kmp_real32 ftmp;
890 memcpy(dest: &ftmp, src: &tmp, n: sizeof(tmp));
891 return ftmp;
892}
893
894inline kmp_real64 KMP_XCHG_REAL64(volatile kmp_real64 *p, kmp_real64 v) {
895 volatile kmp_uint64 *up;
896 kmp_uint64 uv;
897 memcpy(dest: &up, src: &p, n: sizeof(up));
898 memcpy(dest: &uv, src: &v, n: sizeof(uv));
899 kmp_int64 tmp = __sync_lock_test_and_set(up, uv);
900 kmp_real64 dtmp;
901 memcpy(dest: &dtmp, src: &tmp, n: sizeof(tmp));
902 return dtmp;
903}
904
905#else
906
907extern kmp_int8 __kmp_test_then_add8(volatile kmp_int8 *p, kmp_int8 v);
908extern kmp_int8 __kmp_test_then_or8(volatile kmp_int8 *p, kmp_int8 v);
909extern kmp_int8 __kmp_test_then_and8(volatile kmp_int8 *p, kmp_int8 v);
910extern kmp_int32 __kmp_test_then_add32(volatile kmp_int32 *p, kmp_int32 v);
911extern kmp_uint32 __kmp_test_then_or32(volatile kmp_uint32 *p, kmp_uint32 v);
912extern kmp_uint32 __kmp_test_then_and32(volatile kmp_uint32 *p, kmp_uint32 v);
913extern kmp_int64 __kmp_test_then_add64(volatile kmp_int64 *p, kmp_int64 v);
914extern kmp_uint64 __kmp_test_then_or64(volatile kmp_uint64 *p, kmp_uint64 v);
915extern kmp_uint64 __kmp_test_then_and64(volatile kmp_uint64 *p, kmp_uint64 v);
916
917extern kmp_int8 __kmp_compare_and_store8(volatile kmp_int8 *p, kmp_int8 cv,
918 kmp_int8 sv);
919extern kmp_int16 __kmp_compare_and_store16(volatile kmp_int16 *p, kmp_int16 cv,
920 kmp_int16 sv);
921extern kmp_int32 __kmp_compare_and_store32(volatile kmp_int32 *p, kmp_int32 cv,
922 kmp_int32 sv);
923extern kmp_int32 __kmp_compare_and_store64(volatile kmp_int64 *p, kmp_int64 cv,
924 kmp_int64 sv);
925extern kmp_int8 __kmp_compare_and_store_ret8(volatile kmp_int8 *p, kmp_int8 cv,
926 kmp_int8 sv);
927extern kmp_int16 __kmp_compare_and_store_ret16(volatile kmp_int16 *p,
928 kmp_int16 cv, kmp_int16 sv);
929extern kmp_int32 __kmp_compare_and_store_ret32(volatile kmp_int32 *p,
930 kmp_int32 cv, kmp_int32 sv);
931extern kmp_int64 __kmp_compare_and_store_ret64(volatile kmp_int64 *p,
932 kmp_int64 cv, kmp_int64 sv);
933
934extern kmp_int8 __kmp_xchg_fixed8(volatile kmp_int8 *p, kmp_int8 v);
935extern kmp_int16 __kmp_xchg_fixed16(volatile kmp_int16 *p, kmp_int16 v);
936extern kmp_int32 __kmp_xchg_fixed32(volatile kmp_int32 *p, kmp_int32 v);
937extern kmp_int64 __kmp_xchg_fixed64(volatile kmp_int64 *p, kmp_int64 v);
938extern kmp_real32 __kmp_xchg_real32(volatile kmp_real32 *p, kmp_real32 v);
939extern kmp_real64 __kmp_xchg_real64(volatile kmp_real64 *p, kmp_real64 v);
940
941#define KMP_TEST_THEN_INC32(p) \
942 __kmp_test_then_add32((volatile kmp_int32 *)(p), 1)
943#define KMP_TEST_THEN_INC_ACQ32(p) \
944 __kmp_test_then_add32((volatile kmp_int32 *)(p), 1)
945#define KMP_TEST_THEN_INC64(p) \
946 __kmp_test_then_add64((volatile kmp_int64 *)(p), 1LL)
947#define KMP_TEST_THEN_INC_ACQ64(p) \
948 __kmp_test_then_add64((volatile kmp_int64 *)(p), 1LL)
949#define KMP_TEST_THEN_ADD4_32(p) \
950 __kmp_test_then_add32((volatile kmp_int32 *)(p), 4)
951#define KMP_TEST_THEN_ADD4_ACQ32(p) \
952 __kmp_test_then_add32((volatile kmp_int32 *)(p), 4)
953#define KMP_TEST_THEN_ADD4_64(p) \
954 __kmp_test_then_add64((volatile kmp_int64 *)(p), 4LL)
955#define KMP_TEST_THEN_ADD4_ACQ64(p) \
956 __kmp_test_then_add64((volatile kmp_int64 *)(p), 4LL)
957#define KMP_TEST_THEN_DEC32(p) \
958 __kmp_test_then_add32((volatile kmp_int32 *)(p), -1)
959#define KMP_TEST_THEN_DEC_ACQ32(p) \
960 __kmp_test_then_add32((volatile kmp_int32 *)(p), -1)
961#define KMP_TEST_THEN_DEC64(p) \
962 __kmp_test_then_add64((volatile kmp_int64 *)(p), -1LL)
963#define KMP_TEST_THEN_DEC_ACQ64(p) \
964 __kmp_test_then_add64((volatile kmp_int64 *)(p), -1LL)
965#define KMP_TEST_THEN_ADD8(p, v) \
966 __kmp_test_then_add8((volatile kmp_int8 *)(p), (kmp_int8)(v))
967#define KMP_TEST_THEN_ADD32(p, v) \
968 __kmp_test_then_add32((volatile kmp_int32 *)(p), (kmp_int32)(v))
969#define KMP_TEST_THEN_ADD64(p, v) \
970 __kmp_test_then_add64((volatile kmp_int64 *)(p), (kmp_int64)(v))
971
972#define KMP_TEST_THEN_OR8(p, v) \
973 __kmp_test_then_or8((volatile kmp_int8 *)(p), (kmp_int8)(v))
974#define KMP_TEST_THEN_AND8(p, v) \
975 __kmp_test_then_and8((volatile kmp_int8 *)(p), (kmp_int8)(v))
976#define KMP_TEST_THEN_OR32(p, v) \
977 __kmp_test_then_or32((volatile kmp_uint32 *)(p), (kmp_uint32)(v))
978#define KMP_TEST_THEN_AND32(p, v) \
979 __kmp_test_then_and32((volatile kmp_uint32 *)(p), (kmp_uint32)(v))
980#define KMP_TEST_THEN_OR64(p, v) \
981 __kmp_test_then_or64((volatile kmp_uint64 *)(p), (kmp_uint64)(v))
982#define KMP_TEST_THEN_AND64(p, v) \
983 __kmp_test_then_and64((volatile kmp_uint64 *)(p), (kmp_uint64)(v))
984
985#define KMP_COMPARE_AND_STORE_ACQ8(p, cv, sv) \
986 __kmp_compare_and_store8((volatile kmp_int8 *)(p), (kmp_int8)(cv), \
987 (kmp_int8)(sv))
988#define KMP_COMPARE_AND_STORE_REL8(p, cv, sv) \
989 __kmp_compare_and_store8((volatile kmp_int8 *)(p), (kmp_int8)(cv), \
990 (kmp_int8)(sv))
991#define KMP_COMPARE_AND_STORE_ACQ16(p, cv, sv) \
992 __kmp_compare_and_store16((volatile kmp_int16 *)(p), (kmp_int16)(cv), \
993 (kmp_int16)(sv))
994#define KMP_COMPARE_AND_STORE_REL16(p, cv, sv) \
995 __kmp_compare_and_store16((volatile kmp_int16 *)(p), (kmp_int16)(cv), \
996 (kmp_int16)(sv))
997#define KMP_COMPARE_AND_STORE_ACQ32(p, cv, sv) \
998 __kmp_compare_and_store32((volatile kmp_int32 *)(p), (kmp_int32)(cv), \
999 (kmp_int32)(sv))
1000#define KMP_COMPARE_AND_STORE_REL32(p, cv, sv) \
1001 __kmp_compare_and_store32((volatile kmp_int32 *)(p), (kmp_int32)(cv), \
1002 (kmp_int32)(sv))
1003#define KMP_COMPARE_AND_STORE_ACQ64(p, cv, sv) \
1004 __kmp_compare_and_store64((volatile kmp_int64 *)(p), (kmp_int64)(cv), \
1005 (kmp_int64)(sv))
1006#define KMP_COMPARE_AND_STORE_REL64(p, cv, sv) \
1007 __kmp_compare_and_store64((volatile kmp_int64 *)(p), (kmp_int64)(cv), \
1008 (kmp_int64)(sv))
1009
1010#if KMP_ARCH_X86
1011#define KMP_COMPARE_AND_STORE_PTR(p, cv, sv) \
1012 __kmp_compare_and_store32((volatile kmp_int32 *)(p), (kmp_int32)(cv), \
1013 (kmp_int32)(sv))
1014#else /* 64 bit pointers */
1015#define KMP_COMPARE_AND_STORE_PTR(p, cv, sv) \
1016 __kmp_compare_and_store64((volatile kmp_int64 *)(p), (kmp_int64)(cv), \
1017 (kmp_int64)(sv))
1018#endif /* KMP_ARCH_X86 */
1019
1020#define KMP_COMPARE_AND_STORE_RET8(p, cv, sv) \
1021 __kmp_compare_and_store_ret8((p), (cv), (sv))
1022#define KMP_COMPARE_AND_STORE_RET16(p, cv, sv) \
1023 __kmp_compare_and_store_ret16((p), (cv), (sv))
1024#define KMP_COMPARE_AND_STORE_RET32(p, cv, sv) \
1025 __kmp_compare_and_store_ret32((volatile kmp_int32 *)(p), (kmp_int32)(cv), \
1026 (kmp_int32)(sv))
1027#define KMP_COMPARE_AND_STORE_RET64(p, cv, sv) \
1028 __kmp_compare_and_store_ret64((volatile kmp_int64 *)(p), (kmp_int64)(cv), \
1029 (kmp_int64)(sv))
1030
1031#define KMP_XCHG_FIXED8(p, v) \
1032 __kmp_xchg_fixed8((volatile kmp_int8 *)(p), (kmp_int8)(v));
1033#define KMP_XCHG_FIXED16(p, v) __kmp_xchg_fixed16((p), (v));
1034#define KMP_XCHG_FIXED32(p, v) __kmp_xchg_fixed32((p), (v));
1035#define KMP_XCHG_FIXED64(p, v) __kmp_xchg_fixed64((p), (v));
1036#define KMP_XCHG_REAL32(p, v) __kmp_xchg_real32((p), (v));
1037#define KMP_XCHG_REAL64(p, v) __kmp_xchg_real64((p), (v));
1038
1039#endif /* KMP_ASM_INTRINS */
1040
1041/* ------------- relaxed consistency memory model stuff ------------------ */
1042
1043#if KMP_OS_WINDOWS
1044#ifdef __ABSOFT_WIN
1045#define KMP_MB() asm("nop")
1046#define KMP_IMB() asm("nop")
1047#else
1048#define KMP_MB() /* _asm{ nop } */
1049#define KMP_IMB() /* _asm{ nop } */
1050#endif
1051#endif /* KMP_OS_WINDOWS */
1052
1053#if KMP_ARCH_PPC64 || KMP_ARCH_ARM || KMP_ARCH_AARCH64 || KMP_ARCH_MIPS || \
1054 KMP_ARCH_MIPS64 || KMP_ARCH_RISCV64 || KMP_ARCH_LOONGARCH64 || \
1055 KMP_ARCH_VE || KMP_ARCH_S390X || KMP_ARCH_PPC || KMP_ARCH_AARCH64_32 || \
1056 KMP_ARCH_SPARC
1057#if KMP_OS_WINDOWS
1058#undef KMP_MB
1059#define KMP_MB() std::atomic_thread_fence(std::memory_order_seq_cst)
1060#else /* !KMP_OS_WINDOWS */
1061#define KMP_MB() __sync_synchronize()
1062#endif
1063#endif
1064
1065#ifndef KMP_MB
1066#define KMP_MB() /* nothing to do */
1067#endif
1068
1069#if KMP_ARCH_X86 || KMP_ARCH_X86_64
1070#if KMP_MIC
1071// fence-style instructions do not exist, but lock; xaddl $0,(%rsp) can be used.
1072// We shouldn't need it, though, since the ABI rules require that
1073// * If the compiler generates NGO stores it also generates the fence
1074// * If users hand-code NGO stores they should insert the fence
1075// therefore no incomplete unordered stores should be visible.
1076#define KMP_MFENCE() /* Nothing */
1077#define KMP_SFENCE() /* Nothing */
1078#else
1079#if KMP_COMPILER_ICC || KMP_COMPILER_ICX
1080#define KMP_MFENCE_() _mm_mfence()
1081#define KMP_SFENCE_() _mm_sfence()
1082#elif KMP_COMPILER_MSVC
1083#define KMP_MFENCE_() MemoryBarrier()
1084#define KMP_SFENCE_() MemoryBarrier()
1085#else
1086#define KMP_MFENCE_() __sync_synchronize()
1087#define KMP_SFENCE_() __sync_synchronize()
1088#endif
1089#define KMP_MFENCE() \
1090 if (UNLIKELY(!__kmp_cpuinfo.initialized)) { \
1091 __kmp_query_cpuid(&__kmp_cpuinfo); \
1092 } \
1093 if (__kmp_cpuinfo.flags.sse2) { \
1094 KMP_MFENCE_(); \
1095 }
1096#define KMP_SFENCE() KMP_SFENCE_()
1097#endif
1098#else
1099#define KMP_MFENCE() KMP_MB()
1100#define KMP_SFENCE() KMP_MB()
1101#endif
1102
1103#ifndef KMP_IMB
1104#define KMP_IMB() /* nothing to do */
1105#endif
1106
1107#ifndef KMP_ST_REL32
1108#define KMP_ST_REL32(A, D) (*(A) = (D))
1109#endif
1110
1111#ifndef KMP_ST_REL64
1112#define KMP_ST_REL64(A, D) (*(A) = (D))
1113#endif
1114
1115#ifndef KMP_LD_ACQ32
1116#define KMP_LD_ACQ32(A) (*(A))
1117#endif
1118
1119#ifndef KMP_LD_ACQ64
1120#define KMP_LD_ACQ64(A) (*(A))
1121#endif
1122
1123/* ------------------------------------------------------------------------ */
1124// FIXME - maybe this should this be
1125//
1126// #define TCR_4(a) (*(volatile kmp_int32 *)(&a))
1127// #define TCW_4(a,b) (a) = (*(volatile kmp_int32 *)&(b))
1128//
1129// #define TCR_8(a) (*(volatile kmp_int64 *)(a))
1130// #define TCW_8(a,b) (a) = (*(volatile kmp_int64 *)(&b))
1131//
1132// I'm fairly certain this is the correct thing to do, but I'm afraid
1133// of performance regressions.
1134
1135#define TCR_1(a) (a)
1136#define TCW_1(a, b) (a) = (b)
1137#define TCR_4(a) (a)
1138#define TCW_4(a, b) (a) = (b)
1139#define TCI_4(a) (++(a))
1140#define TCD_4(a) (--(a))
1141#define TCR_8(a) (a)
1142#define TCW_8(a, b) (a) = (b)
1143#define TCI_8(a) (++(a))
1144#define TCD_8(a) (--(a))
1145#define TCR_SYNC_4(a) (a)
1146#define TCW_SYNC_4(a, b) (a) = (b)
1147#define TCX_SYNC_4(a, b, c) \
1148 KMP_COMPARE_AND_STORE_REL32((volatile kmp_int32 *)(volatile void *)&(a), \
1149 (kmp_int32)(b), (kmp_int32)(c))
1150#define TCR_SYNC_8(a) (a)
1151#define TCW_SYNC_8(a, b) (a) = (b)
1152#define TCX_SYNC_8(a, b, c) \
1153 KMP_COMPARE_AND_STORE_REL64((volatile kmp_int64 *)(volatile void *)&(a), \
1154 (kmp_int64)(b), (kmp_int64)(c))
1155
1156#if KMP_ARCH_X86 || KMP_ARCH_MIPS || KMP_ARCH_WASM || KMP_ARCH_PPC
1157// What about ARM?
1158#define TCR_PTR(a) ((void *)TCR_4(a))
1159#define TCW_PTR(a, b) TCW_4((a), (b))
1160#define TCR_SYNC_PTR(a) ((void *)TCR_SYNC_4(a))
1161#define TCW_SYNC_PTR(a, b) TCW_SYNC_4((a), (b))
1162#define TCX_SYNC_PTR(a, b, c) ((void *)TCX_SYNC_4((a), (b), (c)))
1163
1164#else /* 64 bit pointers */
1165
1166#define TCR_PTR(a) ((void *)TCR_8(a))
1167#define TCW_PTR(a, b) TCW_8((a), (b))
1168#define TCR_SYNC_PTR(a) ((void *)TCR_SYNC_8(a))
1169#define TCW_SYNC_PTR(a, b) TCW_SYNC_8((a), (b))
1170#define TCX_SYNC_PTR(a, b, c) ((void *)TCX_SYNC_8((a), (b), (c)))
1171
1172#endif /* KMP_ARCH_X86 */
1173
1174/* If these FTN_{TRUE,FALSE} values change, may need to change several places
1175 where they are used to check that language is Fortran, not C. */
1176
1177#ifndef FTN_TRUE
1178#define FTN_TRUE TRUE
1179#endif
1180
1181#ifndef FTN_FALSE
1182#define FTN_FALSE FALSE
1183#endif
1184
1185typedef void (*microtask_t)(int *gtid, int *npr, ...);
1186
1187#ifdef USE_VOLATILE_CAST
1188#define VOLATILE_CAST(x) (volatile x)
1189#else
1190#define VOLATILE_CAST(x) (x)
1191#endif
1192
1193#define KMP_WAIT __kmp_wait_4
1194#define KMP_WAIT_PTR __kmp_wait_4_ptr
1195#define KMP_EQ __kmp_eq_4
1196#define KMP_NEQ __kmp_neq_4
1197#define KMP_LT __kmp_lt_4
1198#define KMP_GE __kmp_ge_4
1199#define KMP_LE __kmp_le_4
1200
1201/* Workaround for Intel(R) 64 code gen bug when taking address of static array
1202 * (Intel(R) 64 Tracker #138) */
1203#if (KMP_ARCH_X86_64 || KMP_ARCH_PPC64) && KMP_OS_LINUX
1204#define STATIC_EFI2_WORKAROUND
1205#else
1206#define STATIC_EFI2_WORKAROUND static
1207#endif
1208
1209// Support of BGET usage
1210#ifndef KMP_USE_BGET
1211#define KMP_USE_BGET 1
1212#endif
1213
1214// Switches for OSS builds
1215#ifndef USE_CMPXCHG_FIX
1216#define USE_CMPXCHG_FIX 1
1217#endif
1218
1219// Enable dynamic user lock
1220#define KMP_USE_DYNAMIC_LOCK 1
1221
1222// Enable Intel(R) Transactional Synchronization Extensions (Intel(R) TSX) if
1223// dynamic user lock is turned on
1224#if KMP_USE_DYNAMIC_LOCK
1225// Visual studio can't handle the asm sections in this code
1226#define KMP_USE_TSX (KMP_ARCH_X86 || KMP_ARCH_X86_64) && !KMP_COMPILER_MSVC
1227#ifdef KMP_USE_ADAPTIVE_LOCKS
1228#undef KMP_USE_ADAPTIVE_LOCKS
1229#endif
1230#define KMP_USE_ADAPTIVE_LOCKS KMP_USE_TSX
1231#endif
1232
1233// Enable tick time conversion of ticks to seconds
1234#if KMP_STATS_ENABLED
1235#define KMP_HAVE_TICK_TIME \
1236 (KMP_OS_LINUX && (KMP_MIC || KMP_ARCH_X86 || KMP_ARCH_X86_64))
1237#endif
1238
1239// Warning levels
1240enum kmp_warnings_level {
1241 kmp_warnings_off = 0, /* No warnings */
1242 kmp_warnings_low, /* Minimal warnings (default) */
1243 kmp_warnings_explicit = 6, /* Explicitly set to ON - more warnings */
1244 kmp_warnings_verbose /* reserved */
1245};
1246
1247#ifdef __cplusplus
1248} // extern "C"
1249#endif // __cplusplus
1250
1251// Safe C API
1252#include "kmp_safe_c_api.h"
1253
1254// Macros for C++11 atomic functions
1255#define KMP_ATOMIC_LD(p, order) (p)->load(std::memory_order_##order)
1256#define KMP_ATOMIC_OP(op, p, v, order) (p)->op(v, std::memory_order_##order)
1257
1258// For non-default load/store
1259#define KMP_ATOMIC_LD_ACQ(p) KMP_ATOMIC_LD(p, acquire)
1260#define KMP_ATOMIC_LD_RLX(p) KMP_ATOMIC_LD(p, relaxed)
1261#define KMP_ATOMIC_ST_REL(p, v) KMP_ATOMIC_OP(store, p, v, release)
1262#define KMP_ATOMIC_ST_RLX(p, v) KMP_ATOMIC_OP(store, p, v, relaxed)
1263
1264// For non-default fetch_<op>
1265#define KMP_ATOMIC_ADD(p, v) KMP_ATOMIC_OP(fetch_add, p, v, acq_rel)
1266#define KMP_ATOMIC_SUB(p, v) KMP_ATOMIC_OP(fetch_sub, p, v, acq_rel)
1267#define KMP_ATOMIC_AND(p, v) KMP_ATOMIC_OP(fetch_and, p, v, acq_rel)
1268#define KMP_ATOMIC_OR(p, v) KMP_ATOMIC_OP(fetch_or, p, v, acq_rel)
1269#define KMP_ATOMIC_INC(p) KMP_ATOMIC_OP(fetch_add, p, 1, acq_rel)
1270#define KMP_ATOMIC_DEC(p) KMP_ATOMIC_OP(fetch_sub, p, 1, acq_rel)
1271#define KMP_ATOMIC_ADD_RLX(p, v) KMP_ATOMIC_OP(fetch_add, p, v, relaxed)
1272#define KMP_ATOMIC_INC_RLX(p) KMP_ATOMIC_OP(fetch_add, p, 1, relaxed)
1273
1274// Callers of the following functions cannot see the side effect on "expected".
1275template <typename T>
1276bool __kmp_atomic_compare_store(std::atomic<T> *p, T expected, T desired) {
1277 return p->compare_exchange_strong(
1278 expected, desired, std::memory_order_acq_rel, std::memory_order_relaxed);
1279}
1280
1281template <typename T>
1282bool __kmp_atomic_compare_store_acq(std::atomic<T> *p, T expected, T desired) {
1283 return p->compare_exchange_strong(
1284 expected, desired, std::memory_order_acquire, std::memory_order_relaxed);
1285}
1286
1287template <typename T>
1288bool __kmp_atomic_compare_store_rel(std::atomic<T> *p, T expected, T desired) {
1289 return p->compare_exchange_strong(
1290 expected, desired, std::memory_order_release, std::memory_order_relaxed);
1291}
1292
1293// Symbol lookup on Linux/Windows
1294#if KMP_OS_WINDOWS
1295extern void *__kmp_lookup_symbol(const char *name, bool next = false);
1296#define KMP_DLSYM(name) __kmp_lookup_symbol(name)
1297#define KMP_DLSYM_NEXT(name) __kmp_lookup_symbol(name, true)
1298#elif KMP_OS_WASI || KMP_OS_EMSCRIPTEN
1299#define KMP_DLSYM(name) nullptr
1300#define KMP_DLSYM_NEXT(name) nullptr
1301#else
1302#define KMP_DLSYM(name) dlsym(RTLD_DEFAULT, name)
1303#define KMP_DLSYM_NEXT(name) dlsym(RTLD_NEXT, name)
1304#endif
1305
1306// MSVC doesn't have this, but clang/clang-cl does.
1307#ifndef __has_builtin
1308#define __has_builtin(x) 0
1309#endif
1310
1311// Same as LLVM_BUILTIN_UNREACHABLE. States that it is UB to reach this point.
1312#if __has_builtin(__builtin_unreachable) || defined(__GNUC__)
1313#define KMP_BUILTIN_UNREACHABLE __builtin_unreachable()
1314#elif defined(_MSC_VER)
1315#define KMP_BUILTIN_UNREACHABLE __assume(false)
1316#else
1317#define KMP_BUILTIN_UNREACHABLE
1318#endif
1319
1320#endif /* KMP_OS_H */
1321

source code of openmp/runtime/src/kmp_os.h