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

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