z_Linux_util.cpp

Go to the documentation of this file.

/*
 * z_Linux_util.cpp -- platform specific routines.
 */
 
//===----------------------------------------------------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
 
#include "kmp.h"
#include "kmp_affinity.h"
#include "kmp_i18n.h"
#include "kmp_io.h"
#include "kmp_itt.h"
#include "kmp_lock.h"
#include "kmp_stats.h"
#include "kmp_str.h"
#include "kmp_wait_release.h"
#include "kmp_wrapper_getpid.h"
 
#if !KMP_OS_DRAGONFLY && !KMP_OS_FREEBSD && !KMP_OS_NETBSD && !KMP_OS_OPENBSD
#include <alloca.h>
#endif
#include <math.h> // HUGE_VAL.
#if KMP_OS_LINUX
#include <semaphore.h>
#endif // KMP_OS_LINUX
#include <sys/resource.h>
#if KMP_OS_AIX
#include <sys/ldr.h>
#include <libperfstat.h>
#elif !KMP_OS_HAIKU
#include <sys/syscall.h>
#endif
#include <sys/time.h>
#include <sys/times.h>
#include <unistd.h>
 
#if KMP_OS_LINUX
#include <sys/sysinfo.h>
#if KMP_USE_FUTEX
// We should really include <futex.h>, but that causes compatibility problems on
// different Linux* OS distributions that either require that you include (or
// break when you try to include) <pci/types.h>. Since all we need is the two
// macros below (which are part of the kernel ABI, so can't change) we just
// define the constants here and don't include <futex.h>
#ifndef FUTEX_WAIT
#define FUTEX_WAIT 0
#endif
#ifndef FUTEX_WAKE
#define FUTEX_WAKE 1
#endif
#endif
#elif KMP_OS_DARWIN
#include <mach/mach.h>
#include <sys/sysctl.h>
#elif KMP_OS_DRAGONFLY || KMP_OS_FREEBSD
#include <sys/types.h>
#include <sys/sysctl.h>
#include <sys/user.h>
#include <pthread_np.h>
#if KMP_OS_DRAGONFLY
#include <kvm.h>
#endif
#elif KMP_OS_NETBSD || KMP_OS_OPENBSD
#include <sys/types.h>
#include <sys/sysctl.h>
#if KMP_OS_NETBSD
#include <sched.h>
#endif
#if KMP_OS_OPENBSD
#include <pthread_np.h>
#endif
#elif KMP_OS_SOLARIS
#include <procfs.h>
#include <thread.h>
#include <sys/loadavg.h>
#endif
 
#include <ctype.h>
#include <dirent.h>
#include <fcntl.h>
 
struct kmp_sys_timer {
  struct timespec start;
};
struct kmp_sys_timer {…};
 
#ifndef TIMEVAL_TO_TIMESPEC
// Convert timeval to timespec.
#define TIMEVAL_TO_TIMESPEC(tv, ts)                                            \
  do {                                                                         \
    (ts)->tv_sec = (tv)->tv_sec;                                               \
    (ts)->tv_nsec = (tv)->tv_usec * 1000;                                      \
  } while (0)
#define TIMEVAL_TO_TIMESPEC(tv, ts)                                            \ …
#endif
 
// Convert timespec to nanoseconds.
#define TS2NS(timespec)                                                        \
  (((timespec).tv_sec * (long int)1e9) + (timespec).tv_nsec)
#define TS2NS(timespec)                                                        \ …
 
static struct kmp_sys_timer __kmp_sys_timer_data;
 
#if KMP_HANDLE_SIGNALS
typedef void (*sig_func_t)(int);
STATIC_EFI2_WORKAROUND struct sigaction __kmp_sighldrs[NSIG];
static sigset_t __kmp_sigset;
#endif
 
static int __kmp_init_runtime = FALSE;
 
static int __kmp_fork_count = 0;
 
static pthread_condattr_t __kmp_suspend_cond_attr;
static pthread_mutexattr_t __kmp_suspend_mutex_attr;
 
static kmp_cond_align_t __kmp_wait_cv;
static kmp_mutex_align_t __kmp_wait_mx;
 
kmp_uint64 __kmp_ticks_per_msec = 1000000;
kmp_uint64 __kmp_ticks_per_usec = 1000;
 
#ifdef DEBUG_SUSPEND
static void __kmp_print_cond(char *buffer, kmp_cond_align_t *cond) {
  KMP_SNPRINTF(buffer, 128, "(cond (lock (%ld, %d)), (descr (%p)))",
               cond->c_cond.__c_lock.__status, cond->c_cond.__c_lock.__spinlock,
               cond->c_cond.__c_waiting);
}
#endif
 
#if ((KMP_OS_LINUX || KMP_OS_FREEBSD || KMP_OS_NETBSD || KMP_OS_DRAGONFLY ||   \
      KMP_OS_AIX) &&                                                           \
     KMP_AFFINITY_SUPPORTED)
 
/* Affinity support */
 
void __kmp_affinity_bind_thread(int which) {
  KMP_ASSERT2(KMP_AFFINITY_CAPABLE(),
              "Illegal set affinity operation when not capable");
 
  kmp_affin_mask_t *mask;
  KMP_CPU_ALLOC_ON_STACK(mask);
  KMP_CPU_ZERO(mask);
  KMP_CPU_SET(which, mask);
  __kmp_set_system_affinity(mask, TRUE);
  KMP_CPU_FREE_FROM_STACK(mask);
}
 
#if KMP_OS_AIX
void __kmp_affinity_determine_capable(const char *env_var) {
  // All versions of AIX support bindprocessor().
 
  size_t mask_size = __kmp_xproc / CHAR_BIT;
  // Round up to byte boundary.
  if (__kmp_xproc % CHAR_BIT)
    ++mask_size;
 
  // Round up to the mask_size_type boundary.
  if (mask_size % sizeof(__kmp_affin_mask_size))
    mask_size += sizeof(__kmp_affin_mask_size) -
                 mask_size % sizeof(__kmp_affin_mask_size);
  KMP_AFFINITY_ENABLE(mask_size);
  KA_TRACE(10,
           ("__kmp_affinity_determine_capable: "
            "AIX OS affinity interface bindprocessor functional (mask size = "
            "%" KMP_SIZE_T_SPEC ").\n",
            __kmp_affin_mask_size));
}
 
#else // !KMP_OS_AIX
 
/* Determine if we can access affinity functionality on this version of
 * Linux* OS by checking __NR_sched_{get,set}affinity system calls, and set
 * __kmp_affin_mask_size to the appropriate value (0 means not capable). */
void __kmp_affinity_determine_capable(const char *env_var) {
  // Check and see if the OS supports thread affinity.
 
#if KMP_OS_LINUX
#define KMP_CPU_SET_SIZE_LIMIT (1024 * 1024)
#define KMP_CPU_SET_TRY_SIZE CACHE_LINE
#elif KMP_OS_FREEBSD || KMP_OS_DRAGONFLY
#define KMP_CPU_SET_SIZE_LIMIT (sizeof(cpuset_t))
#elif KMP_OS_NETBSD
#define KMP_CPU_SET_SIZE_LIMIT (256)
#endif
 
  int verbose = __kmp_affinity.flags.verbose;
  int warnings = __kmp_affinity.flags.warnings;
  enum affinity_type type = __kmp_affinity.type;
 
#if KMP_OS_LINUX
  long gCode;
  unsigned char *buf;
  buf = (unsigned char *)KMP_INTERNAL_MALLOC(KMP_CPU_SET_SIZE_LIMIT);
 
  // If the syscall returns a suggestion for the size,
  // then we don't have to search for an appropriate size.
  gCode = syscall(__NR_sched_getaffinity, 0, KMP_CPU_SET_TRY_SIZE, buf);
  KA_TRACE(30, ("__kmp_affinity_determine_capable: "
                "initial getaffinity call returned %ld errno = %d\n",
                gCode, errno));
 
  if (gCode < 0 && errno != EINVAL) {
    // System call not supported
    if (verbose ||
        (warnings && (type != affinity_none) && (type != affinity_default) &&
         (type != affinity_disabled))) {
      int error = errno;
      kmp_msg_t err_code = KMP_ERR(error);
      __kmp_msg(kmp_ms_warning, KMP_MSG(GetAffSysCallNotSupported, env_var),
                err_code, __kmp_msg_null);
      if (__kmp_generate_warnings == kmp_warnings_off) {
        __kmp_str_free(&err_code.str);
      }
    }
    KMP_AFFINITY_DISABLE();
    KMP_INTERNAL_FREE(buf);
    return;
  } else if (gCode > 0) {
    // The optimal situation: the OS returns the size of the buffer it expects.
    KMP_AFFINITY_ENABLE(gCode);
    KA_TRACE(10, ("__kmp_affinity_determine_capable: "
                  "affinity supported (mask size %d)\n",
                  (int)__kmp_affin_mask_size));
    KMP_INTERNAL_FREE(buf);
    return;
  }
 
  // Call the getaffinity system call repeatedly with increasing set sizes
  // until we succeed, or reach an upper bound on the search.
  KA_TRACE(30, ("__kmp_affinity_determine_capable: "
                "searching for proper set size\n"));
  int size;
  for (size = 1; size <= KMP_CPU_SET_SIZE_LIMIT; size *= 2) {
    gCode = syscall(__NR_sched_getaffinity, 0, size, buf);
    KA_TRACE(30, ("__kmp_affinity_determine_capable: "
                  "getaffinity for mask size %ld returned %ld errno = %d\n",
                  size, gCode, errno));
 
    if (gCode < 0) {
      if (errno == ENOSYS) {
        // We shouldn't get here
        KA_TRACE(30, ("__kmp_affinity_determine_capable: "
                      "inconsistent OS call behavior: errno == ENOSYS for mask "
                      "size %d\n",
                      size));
        if (verbose ||
            (warnings && (type != affinity_none) &&
             (type != affinity_default) && (type != affinity_disabled))) {
          int error = errno;
          kmp_msg_t err_code = KMP_ERR(error);
          __kmp_msg(kmp_ms_warning, KMP_MSG(GetAffSysCallNotSupported, env_var),
                    err_code, __kmp_msg_null);
          if (__kmp_generate_warnings == kmp_warnings_off) {
            __kmp_str_free(&err_code.str);
          }
        }
        KMP_AFFINITY_DISABLE();
        KMP_INTERNAL_FREE(buf);
        return;
      }
      continue;
    }
 
    KMP_AFFINITY_ENABLE(gCode);
    KA_TRACE(10, ("__kmp_affinity_determine_capable: "
                  "affinity supported (mask size %d)\n",
                  (int)__kmp_affin_mask_size));
    KMP_INTERNAL_FREE(buf);
    return;
  }
#elif KMP_OS_FREEBSD || KMP_OS_NETBSD || KMP_OS_DRAGONFLY
  long gCode;
  unsigned char *buf;
  buf = (unsigned char *)KMP_INTERNAL_MALLOC(KMP_CPU_SET_SIZE_LIMIT);
  gCode = pthread_getaffinity_np(pthread_self(), KMP_CPU_SET_SIZE_LIMIT,
                                 reinterpret_cast<cpuset_t *>(buf));
  KA_TRACE(30, ("__kmp_affinity_determine_capable: "
                "initial getaffinity call returned %d errno = %d\n",
                gCode, errno));
  if (gCode == 0) {
    KMP_AFFINITY_ENABLE(KMP_CPU_SET_SIZE_LIMIT);
    KA_TRACE(10, ("__kmp_affinity_determine_capable: "
                  "affinity supported (mask size %d)\n",
                  (int)__kmp_affin_mask_size));
    KMP_INTERNAL_FREE(buf);
    return;
  }
#endif
  KMP_INTERNAL_FREE(buf);
 
  // Affinity is not supported
  KMP_AFFINITY_DISABLE();
  KA_TRACE(10, ("__kmp_affinity_determine_capable: "
                "cannot determine mask size - affinity not supported\n"));
  if (verbose || (warnings && (type != affinity_none) &&
                  (type != affinity_default) && (type != affinity_disabled))) {
    KMP_WARNING(AffCantGetMaskSize, env_var);
  }
}
#endif // KMP_OS_AIX
#endif // (KMP_OS_LINUX || KMP_OS_FREEBSD || KMP_OS_NETBSD ||                  \
           KMP_OS_DRAGONFLY || KMP_OS_AIX) && KMP_AFFINITY_SUPPORTED
 
#if KMP_USE_FUTEX
 
int __kmp_futex_determine_capable() {
  int loc = 0;
  long rc = syscall(__NR_futex, &loc, FUTEX_WAKE, 1, NULL, NULL, 0);
  int retval = (rc == 0) || (errno != ENOSYS);
 
  KA_TRACE(10,
           ("__kmp_futex_determine_capable: rc = %d errno = %d\n", rc, errno));
  KA_TRACE(10, ("__kmp_futex_determine_capable: futex syscall%s supported\n",
                retval ? "" : " not"));
 
  return retval;
}
 
#endif // KMP_USE_FUTEX
 
#if (KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_ARCH_WASM) && (!KMP_ASM_INTRINS)
/* Only 32-bit "add-exchange" instruction on IA-32 architecture causes us to
   use compare_and_store for these routines */
 
kmp_int8 __kmp_test_then_or8(volatile kmp_int8 *p, kmp_int8 d) {
  kmp_int8 old_value, new_value;
 
  old_value = TCR_1(*p);
  new_value = old_value | d;
 
  while (!KMP_COMPARE_AND_STORE_REL8(p, old_value, new_value)) {
    KMP_CPU_PAUSE();
    old_value = TCR_1(*p);
    new_value = old_value | d;
  }
  return old_value;
}
 
kmp_int8 __kmp_test_then_and8(volatile kmp_int8 *p, kmp_int8 d) {
  kmp_int8 old_value, new_value;
 
  old_value = TCR_1(*p);
  new_value = old_value & d;
 
  while (!KMP_COMPARE_AND_STORE_REL8(p, old_value, new_value)) {
    KMP_CPU_PAUSE();
    old_value = TCR_1(*p);
    new_value = old_value & d;
  }
  return old_value;
}
 
kmp_uint32 __kmp_test_then_or32(volatile kmp_uint32 *p, kmp_uint32 d) {
  kmp_uint32 old_value, new_value;
 
  old_value = TCR_4(*p);
  new_value = old_value | d;
 
  while (!KMP_COMPARE_AND_STORE_REL32(p, old_value, new_value)) {
    KMP_CPU_PAUSE();
    old_value = TCR_4(*p);
    new_value = old_value | d;
  }
  return old_value;
}
 
kmp_uint32 __kmp_test_then_and32(volatile kmp_uint32 *p, kmp_uint32 d) {
  kmp_uint32 old_value, new_value;
 
  old_value = TCR_4(*p);
  new_value = old_value & d;
 
  while (!KMP_COMPARE_AND_STORE_REL32(p, old_value, new_value)) {
    KMP_CPU_PAUSE();
    old_value = TCR_4(*p);
    new_value = old_value & d;
  }
  return old_value;
}
 
#if KMP_ARCH_X86 || KMP_ARCH_WASM
kmp_int8 __kmp_test_then_add8(volatile kmp_int8 *p, kmp_int8 d) {
  kmp_int8 old_value, new_value;
 
  old_value = TCR_1(*p);
  new_value = old_value + d;
 
  while (!KMP_COMPARE_AND_STORE_REL8(p, old_value, new_value)) {
    KMP_CPU_PAUSE();
    old_value = TCR_1(*p);
    new_value = old_value + d;
  }
  return old_value;
}
 
kmp_int64 __kmp_test_then_add64(volatile kmp_int64 *p, kmp_int64 d) {
  kmp_int64 old_value, new_value;
 
  old_value = TCR_8(*p);
  new_value = old_value + d;
 
  while (!KMP_COMPARE_AND_STORE_REL64(p, old_value, new_value)) {
    KMP_CPU_PAUSE();
    old_value = TCR_8(*p);
    new_value = old_value + d;
  }
  return old_value;
}
#endif /* KMP_ARCH_X86 */
 
kmp_uint64 __kmp_test_then_or64(volatile kmp_uint64 *p, kmp_uint64 d) {
  kmp_uint64 old_value, new_value;
 
  old_value = TCR_8(*p);
  new_value = old_value | d;
  while (!KMP_COMPARE_AND_STORE_REL64(p, old_value, new_value)) {
    KMP_CPU_PAUSE();
    old_value = TCR_8(*p);
    new_value = old_value | d;
  }
  return old_value;
}
 
kmp_uint64 __kmp_test_then_and64(volatile kmp_uint64 *p, kmp_uint64 d) {
  kmp_uint64 old_value, new_value;
 
  old_value = TCR_8(*p);
  new_value = old_value & d;
  while (!KMP_COMPARE_AND_STORE_REL64(p, old_value, new_value)) {
    KMP_CPU_PAUSE();
    old_value = TCR_8(*p);
    new_value = old_value & d;
  }
  return old_value;
}
 
#endif /* (KMP_ARCH_X86 || KMP_ARCH_X86_64) && (! KMP_ASM_INTRINS) */
 
void __kmp_terminate_thread(int gtid) {
  int status;
  kmp_info_t *th = __kmp_threads[gtid];
 
  if (!th)
    return;
 
#ifdef KMP_CANCEL_THREADS
  KA_TRACE(10, ("__kmp_terminate_thread: kill (%d)\n", gtid));
  status = pthread_cancel(th->th.th_info.ds.ds_thread);
  if (status != 0 && status != ESRCH) {
    __kmp_fatal(KMP_MSG(CantTerminateWorkerThread), KMP_ERR(status),
                __kmp_msg_null);
  }
#endif
  KMP_YIELD(TRUE);
} //
void __kmp_terminate_thread(int gtid) {…}
 
/* Set thread stack info.
   If values are unreasonable, assume call failed and use incremental stack
   refinement method instead. Returns TRUE if the stack parameters could be
   determined exactly, FALSE if incremental refinement is necessary. */
static kmp_int32 __kmp_set_stack_info(int gtid, kmp_info_t *th) {
  int stack_data;
#if KMP_OS_LINUX || KMP_OS_DRAGONFLY || KMP_OS_FREEBSD || KMP_OS_NETBSD ||     \
    KMP_OS_HAIKU || KMP_OS_HURD || KMP_OS_SOLARIS || KMP_OS_AIX
  int status;
  size_t size = 0;
  void *addr = 0;
 
  /* Always do incremental stack refinement for ubermaster threads since the
     initial thread stack range can be reduced by sibling thread creation so
     pthread_attr_getstack may cause thread gtid aliasing */
  if (!KMP_UBER_GTID(gtid)) {
 
#if KMP_OS_SOLARIS
    stack_t s;
    if ((status = thr_stksegment(&s)) < 0) {
      KMP_CHECK_SYSFAIL("thr_stksegment", status);
    }
 
    addr = s.ss_sp;
    size = s.ss_size;
    KA_TRACE(60, ("__kmp_set_stack_info: T#%d thr_stksegment returned size:"
                  " %lu, low addr: %p\n",
                  gtid, size, addr));
#else
    pthread_attr_t attr;
    /* Fetch the real thread attributes */
    status = pthread_attr_init(&attr);
    KMP_CHECK_SYSFAIL("pthread_attr_init", status);
#if KMP_OS_DRAGONFLY || KMP_OS_FREEBSD || KMP_OS_NETBSD
    status = pthread_attr_get_np(pthread_self(), &attr);
    KMP_CHECK_SYSFAIL("pthread_attr_get_np", status);
#else
    status = pthread_getattr_np(pthread_self(), &attr);
    KMP_CHECK_SYSFAIL("pthread_getattr_np", status);
#endif
    status = pthread_attr_getstack(&attr, &addr, &size);
    KMP_CHECK_SYSFAIL("pthread_attr_getstack", status);
    KA_TRACE(60,
             ("__kmp_set_stack_info: T#%d pthread_attr_getstack returned size:"
              " %lu, low addr: %p\n",
              gtid, size, addr));
    status = pthread_attr_destroy(&attr);
    KMP_CHECK_SYSFAIL("pthread_attr_destroy", status);
#endif
  }
 
  if (size != 0 && addr != 0) { // was stack parameter determination successful?
    /* Store the correct base and size */
    TCW_PTR(th->th.th_info.ds.ds_stackbase, (((char *)addr) + size));
    TCW_PTR(th->th.th_info.ds.ds_stacksize, size);
    TCW_4(th->th.th_info.ds.ds_stackgrow, FALSE);
    return TRUE;
  }
#endif /* KMP_OS_LINUX || KMP_OS_DRAGONFLY || KMP_OS_FREEBSD || KMP_OS_NETBSD  \
          || KMP_OS_HAIKU || KMP_OS_HURD || KMP_OS_SOLARIS */
  /* Use incremental refinement starting from initial conservative estimate */
  TCW_PTR(th->th.th_info.ds.ds_stacksize, 0);
  TCW_PTR(th->th.th_info.ds.ds_stackbase, &stack_data);
  TCW_4(th->th.th_info.ds.ds_stackgrow, TRUE);
  return FALSE;
}
static kmp_int32 __kmp_set_stack_info(int gtid, kmp_info_t *th) {…}
 
static void *__kmp_launch_worker(void *thr) {
  int status, old_type, old_state;
#ifdef KMP_BLOCK_SIGNALS
  sigset_t new_set, old_set;
#endif /* KMP_BLOCK_SIGNALS */
  void *exit_val;
#if KMP_OS_LINUX || KMP_OS_DRAGONFLY || KMP_OS_FREEBSD || KMP_OS_NETBSD ||     \
    KMP_OS_OPENBSD || KMP_OS_HAIKU || KMP_OS_HURD || KMP_OS_SOLARIS ||         \
    KMP_OS_AIX
  void *volatile padding = 0;
#endif
  int gtid;
 
  gtid = ((kmp_info_t *)thr)->th.th_info.ds.ds_gtid;
  __kmp_gtid_set_specific(gtid);
#ifdef KMP_TDATA_GTID
  __kmp_gtid = gtid;
#endif
#if KMP_STATS_ENABLED
  // set thread local index to point to thread-specific stats
  __kmp_stats_thread_ptr = ((kmp_info_t *)thr)->th.th_stats;
  __kmp_stats_thread_ptr->startLife();
  KMP_SET_THREAD_STATE(IDLE);
  KMP_INIT_PARTITIONED_TIMERS(OMP_idle);
#endif
 
#if USE_ITT_BUILD
  __kmp_itt_thread_name(gtid);
#endif /* USE_ITT_BUILD */
 
#if KMP_AFFINITY_SUPPORTED
  __kmp_affinity_bind_init_mask(gtid);
#endif
 
#ifdef KMP_CANCEL_THREADS
  status = pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS, &old_type);
  KMP_CHECK_SYSFAIL("pthread_setcanceltype", status);
  // josh todo: isn't PTHREAD_CANCEL_ENABLE default for newly-created threads?
  status = pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &old_state);
  KMP_CHECK_SYSFAIL("pthread_setcancelstate", status);
#endif
 
#if KMP_ARCH_X86 || KMP_ARCH_X86_64
  // Set FP control regs to be a copy of the parallel initialization thread's.
  __kmp_clear_x87_fpu_status_word();
  __kmp_load_x87_fpu_control_word(&__kmp_init_x87_fpu_control_word);
  __kmp_load_mxcsr(&__kmp_init_mxcsr);
#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
 
#ifdef KMP_BLOCK_SIGNALS
  status = sigfillset(&new_set);
  KMP_CHECK_SYSFAIL_ERRNO("sigfillset", status);
  status = pthread_sigmask(SIG_BLOCK, &new_set, &old_set);
  KMP_CHECK_SYSFAIL("pthread_sigmask", status);
#endif /* KMP_BLOCK_SIGNALS */
 
#if KMP_OS_LINUX || KMP_OS_DRAGONFLY || KMP_OS_FREEBSD || KMP_OS_NETBSD ||     \
    KMP_OS_OPENBSD || KMP_OS_HAIKU || KMP_OS_HURD || KMP_OS_SOLARIS ||         \
    KMP_OS_AIX
  if (__kmp_stkoffset > 0 && gtid > 0) {
    padding = KMP_ALLOCA(gtid * __kmp_stkoffset);
    (void)padding;
  }
#endif
 
  KMP_MB();
  __kmp_set_stack_info(gtid, (kmp_info_t *)thr);
 
  __kmp_check_stack_overlap((kmp_info_t *)thr);
 
  exit_val = __kmp_launch_thread((kmp_info_t *)thr);
 
#ifdef KMP_BLOCK_SIGNALS
  status = pthread_sigmask(SIG_SETMASK, &old_set, NULL);
  KMP_CHECK_SYSFAIL("pthread_sigmask", status);
#endif /* KMP_BLOCK_SIGNALS */
 
  return exit_val;
}
static void *__kmp_launch_worker(void *thr) {…}
 
#if KMP_USE_MONITOR
/* The monitor thread controls all of the threads in the complex */
 
static void *__kmp_launch_monitor(void *thr) {
  int status, old_type, old_state;
#ifdef KMP_BLOCK_SIGNALS
  sigset_t new_set;
#endif /* KMP_BLOCK_SIGNALS */
  struct timespec interval;
 
  KMP_MB(); /* Flush all pending memory write invalidates.  */
 
  KA_TRACE(10, ("__kmp_launch_monitor: #1 launched\n"));
 
  /* register us as the monitor thread */
  __kmp_gtid_set_specific(KMP_GTID_MONITOR);
#ifdef KMP_TDATA_GTID
  __kmp_gtid = KMP_GTID_MONITOR;
#endif
 
  KMP_MB();
 
#if USE_ITT_BUILD
  // Instruct Intel(R) Threading Tools to ignore monitor thread.
  __kmp_itt_thread_ignore();
#endif /* USE_ITT_BUILD */
 
  __kmp_set_stack_info(((kmp_info_t *)thr)->th.th_info.ds.ds_gtid,
                       (kmp_info_t *)thr);
 
  __kmp_check_stack_overlap((kmp_info_t *)thr);
 
#ifdef KMP_CANCEL_THREADS
  status = pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS, &old_type);
  KMP_CHECK_SYSFAIL("pthread_setcanceltype", status);
  // josh todo: isn't PTHREAD_CANCEL_ENABLE default for newly-created threads?
  status = pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &old_state);
  KMP_CHECK_SYSFAIL("pthread_setcancelstate", status);
#endif
 
#if KMP_REAL_TIME_FIX
  // This is a potential fix which allows application with real-time scheduling
  // policy work. However, decision about the fix is not made yet, so it is
  // disabled by default.
  { // Are program started with real-time scheduling policy?
    int sched = sched_getscheduler(0);
    if (sched == SCHED_FIFO || sched == SCHED_RR) {
      // Yes, we are a part of real-time application. Try to increase the
      // priority of the monitor.
      struct sched_param param;
      int max_priority = sched_get_priority_max(sched);
      int rc;
      KMP_WARNING(RealTimeSchedNotSupported);
      sched_getparam(0, &param);
      if (param.sched_priority < max_priority) {
        param.sched_priority += 1;
        rc = sched_setscheduler(0, sched, &param);
        if (rc != 0) {
          int error = errno;
          kmp_msg_t err_code = KMP_ERR(error);
          __kmp_msg(kmp_ms_warning, KMP_MSG(CantChangeMonitorPriority),
                    err_code, KMP_MSG(MonitorWillStarve), __kmp_msg_null);
          if (__kmp_generate_warnings == kmp_warnings_off) {
            __kmp_str_free(&err_code.str);
          }
        }
      } else {
        // We cannot abort here, because number of CPUs may be enough for all
        // the threads, including the monitor thread, so application could
        // potentially work...
        __kmp_msg(kmp_ms_warning, KMP_MSG(RunningAtMaxPriority),
                  KMP_MSG(MonitorWillStarve), KMP_HNT(RunningAtMaxPriority),
                  __kmp_msg_null);
      }
    }
    // AC: free thread that waits for monitor started
    TCW_4(__kmp_global.g.g_time.dt.t_value, 0);
  }
#endif // KMP_REAL_TIME_FIX
 
  KMP_MB(); /* Flush all pending memory write invalidates.  */
 
  if (__kmp_monitor_wakeups == 1) {
    interval.tv_sec = 1;
    interval.tv_nsec = 0;
  } else {
    interval.tv_sec = 0;
    interval.tv_nsec = (KMP_NSEC_PER_SEC / __kmp_monitor_wakeups);
  }
 
  KA_TRACE(10, ("__kmp_launch_monitor: #2 monitor\n"));
 
  while (!TCR_4(__kmp_global.g.g_done)) {
    struct timespec now;
    struct timeval tval;
 
    /*  This thread monitors the state of the system */
 
    KA_TRACE(15, ("__kmp_launch_monitor: update\n"));
 
    status = gettimeofday(&tval, NULL);
    KMP_CHECK_SYSFAIL_ERRNO("gettimeofday", status);
    TIMEVAL_TO_TIMESPEC(&tval, &now);
 
    now.tv_sec += interval.tv_sec;
    now.tv_nsec += interval.tv_nsec;
 
    if (now.tv_nsec >= KMP_NSEC_PER_SEC) {
      now.tv_sec += 1;
      now.tv_nsec -= KMP_NSEC_PER_SEC;
    }
 
    status = pthread_mutex_lock(&__kmp_wait_mx.m_mutex);
    KMP_CHECK_SYSFAIL("pthread_mutex_lock", status);
    // AC: the monitor should not fall asleep if g_done has been set
    if (!TCR_4(__kmp_global.g.g_done)) { // check once more under mutex
      status = pthread_cond_timedwait(&__kmp_wait_cv.c_cond,
                                      &__kmp_wait_mx.m_mutex, &now);
      if (status != 0) {
        if (status != ETIMEDOUT && status != EINTR) {
          KMP_SYSFAIL("pthread_cond_timedwait", status);
        }
      }
    }
    status = pthread_mutex_unlock(&__kmp_wait_mx.m_mutex);
    KMP_CHECK_SYSFAIL("pthread_mutex_unlock", status);
 
    TCW_4(__kmp_global.g.g_time.dt.t_value,
          TCR_4(__kmp_global.g.g_time.dt.t_value) + 1);
 
    KMP_MB(); /* Flush all pending memory write invalidates.  */
  }
 
  KA_TRACE(10, ("__kmp_launch_monitor: #3 cleanup\n"));
 
#ifdef KMP_BLOCK_SIGNALS
  status = sigfillset(&new_set);
  KMP_CHECK_SYSFAIL_ERRNO("sigfillset", status);
  status = pthread_sigmask(SIG_UNBLOCK, &new_set, NULL);
  KMP_CHECK_SYSFAIL("pthread_sigmask", status);
#endif /* KMP_BLOCK_SIGNALS */
 
  KA_TRACE(10, ("__kmp_launch_monitor: #4 finished\n"));
 
  if (__kmp_global.g.g_abort != 0) {
    /* now we need to terminate the worker threads  */
    /* the value of t_abort is the signal we caught */
 
    int gtid;
 
    KA_TRACE(10, ("__kmp_launch_monitor: #5 terminate sig=%d\n",
                  __kmp_global.g.g_abort));
 
    /* terminate the OpenMP worker threads */
    /* TODO this is not valid for sibling threads!!
     * the uber master might not be 0 anymore.. */
    for (gtid = 1; gtid < __kmp_threads_capacity; ++gtid)
      __kmp_terminate_thread(gtid);
 
    __kmp_cleanup();
 
    KA_TRACE(10, ("__kmp_launch_monitor: #6 raise sig=%d\n",
                  __kmp_global.g.g_abort));
 
    if (__kmp_global.g.g_abort > 0)
      raise(__kmp_global.g.g_abort);
  }
 
  KA_TRACE(10, ("__kmp_launch_monitor: #7 exit\n"));
 
  return thr;
}
#endif // KMP_USE_MONITOR
 
void __kmp_create_worker(int gtid, kmp_info_t *th, size_t stack_size) {
  pthread_t handle;
  pthread_attr_t thread_attr;
  int status;
 
  th->th.th_info.ds.ds_gtid = gtid;
 
#if KMP_STATS_ENABLED
  // sets up worker thread stats
  __kmp_acquire_tas_lock(&__kmp_stats_lock, gtid);
 
  // th->th.th_stats is used to transfer thread-specific stats-pointer to
  // __kmp_launch_worker. So when thread is created (goes into
  // __kmp_launch_worker) it will set its thread local pointer to
  // th->th.th_stats
  if (!KMP_UBER_GTID(gtid)) {
    th->th.th_stats = __kmp_stats_list->push_back(gtid);
  } else {
    // For root threads, __kmp_stats_thread_ptr is set in __kmp_register_root(),
    // so set the th->th.th_stats field to it.
    th->th.th_stats = __kmp_stats_thread_ptr;
  }
  __kmp_release_tas_lock(&__kmp_stats_lock, gtid);
 
#endif // KMP_STATS_ENABLED
 
  if (KMP_UBER_GTID(gtid)) {
    KA_TRACE(10, ("__kmp_create_worker: uber thread (%d)\n", gtid));
    th->th.th_info.ds.ds_thread = pthread_self();
    __kmp_set_stack_info(gtid, th);
    __kmp_check_stack_overlap(th);
    return;
  }
 
  KA_TRACE(10, ("__kmp_create_worker: try to create thread (%d)\n", gtid));
 
  KMP_MB(); /* Flush all pending memory write invalidates.  */
 
#ifdef KMP_THREAD_ATTR
  status = pthread_attr_init(&thread_attr);
  if (status != 0) {
    __kmp_fatal(KMP_MSG(CantInitThreadAttrs), KMP_ERR(status), __kmp_msg_null);
  }
  status = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_JOINABLE);
  if (status != 0) {
    __kmp_fatal(KMP_MSG(CantSetWorkerState), KMP_ERR(status), __kmp_msg_null);
  }
 
  /* Set stack size for this thread now.
     The multiple of 2 is there because on some machines, requesting an unusual
     stacksize causes the thread to have an offset before the dummy alloca()
     takes place to create the offset.  Since we want the user to have a
     sufficient stacksize AND support a stack offset, we alloca() twice the
     offset so that the upcoming alloca() does not eliminate any premade offset,
     and also gives the user the stack space they requested for all threads */
  stack_size += gtid * __kmp_stkoffset * 2;
 
  KA_TRACE(10, ("__kmp_create_worker: T#%d, default stacksize = %lu bytes, "
                "__kmp_stksize = %lu bytes, final stacksize = %lu bytes\n",
                gtid, KMP_DEFAULT_STKSIZE, __kmp_stksize, stack_size));
 
#ifdef _POSIX_THREAD_ATTR_STACKSIZE
  status = pthread_attr_setstacksize(&thread_attr, stack_size);
#ifdef KMP_BACKUP_STKSIZE
  if (status != 0) {
    if (!__kmp_env_stksize) {
      stack_size = KMP_BACKUP_STKSIZE + gtid * __kmp_stkoffset;
      __kmp_stksize = KMP_BACKUP_STKSIZE;
      KA_TRACE(10, ("__kmp_create_worker: T#%d, default stacksize = %lu bytes, "
                    "__kmp_stksize = %lu bytes, (backup) final stacksize = %lu "
                    "bytes\n",
                    gtid, KMP_DEFAULT_STKSIZE, __kmp_stksize, stack_size));
      status = pthread_attr_setstacksize(&thread_attr, stack_size);
    }
  }
#endif /* KMP_BACKUP_STKSIZE */
  if (status != 0) {
    __kmp_fatal(KMP_MSG(CantSetWorkerStackSize, stack_size), KMP_ERR(status),
                KMP_HNT(ChangeWorkerStackSize), __kmp_msg_null);
  }
#endif /* _POSIX_THREAD_ATTR_STACKSIZE */
 
#endif /* KMP_THREAD_ATTR */
 
  status =
      pthread_create(&handle, &thread_attr, __kmp_launch_worker, (void *)th);
  if (status != 0 || !handle) { // ??? Why do we check handle??
#ifdef _POSIX_THREAD_ATTR_STACKSIZE
    if (status == EINVAL) {
      __kmp_fatal(KMP_MSG(CantSetWorkerStackSize, stack_size), KMP_ERR(status),
                  KMP_HNT(IncreaseWorkerStackSize), __kmp_msg_null);
    }
    if (status == ENOMEM) {
      __kmp_fatal(KMP_MSG(CantSetWorkerStackSize, stack_size), KMP_ERR(status),
                  KMP_HNT(DecreaseWorkerStackSize), __kmp_msg_null);
    }
#endif /* _POSIX_THREAD_ATTR_STACKSIZE */
    if (status == EAGAIN) {
      __kmp_fatal(KMP_MSG(NoResourcesForWorkerThread), KMP_ERR(status),
                  KMP_HNT(Decrease_NUM_THREADS), __kmp_msg_null);
    }
    KMP_SYSFAIL("pthread_create", status);
  }
 
  // Rename worker threads for improved debuggability
  if (!KMP_UBER_GTID(gtid)) {
#if defined(LIBOMP_HAVE_PTHREAD_SET_NAME_NP)
    pthread_set_name_np(handle, "openmp_worker");
#elif defined(LIBOMP_HAVE_PTHREAD_SETNAME_NP) && !KMP_OS_DARWIN
#if KMP_OS_NETBSD
    pthread_setname_np(handle, "%s", const_cast<char *>("openmp_worker"));
#else
    pthread_setname_np(handle, "openmp_worker");
#endif
#endif
  }
 
  th->th.th_info.ds.ds_thread = handle;
 
#ifdef KMP_THREAD_ATTR
  status = pthread_attr_destroy(&thread_attr);
  if (status) {
    kmp_msg_t err_code = KMP_ERR(status);
    __kmp_msg(kmp_ms_warning, KMP_MSG(CantDestroyThreadAttrs), err_code,
              __kmp_msg_null);
    if (__kmp_generate_warnings == kmp_warnings_off) {
      __kmp_str_free(&err_code.str);
    }
  }
#endif /* KMP_THREAD_ATTR */
 
  KMP_MB(); /* Flush all pending memory write invalidates.  */
 
  KA_TRACE(10, ("__kmp_create_worker: done creating thread (%d)\n", gtid));
 
} // __kmp_create_worker
void __kmp_create_worker(int gtid, kmp_info_t *th, size_t stack_size) {…}
 
#if KMP_USE_MONITOR
void __kmp_create_monitor(kmp_info_t *th) {
  pthread_t handle;
  pthread_attr_t thread_attr;
  size_t size;
  int status;
  int auto_adj_size = FALSE;
 
  if (__kmp_dflt_blocktime == KMP_MAX_BLOCKTIME) {
    // We don't need monitor thread in case of MAX_BLOCKTIME
    KA_TRACE(10, ("__kmp_create_monitor: skipping monitor thread because of "
                  "MAX blocktime\n"));
    th->th.th_info.ds.ds_tid = 0; // this makes reap_monitor no-op
    th->th.th_info.ds.ds_gtid = 0;
    return;
  }
  KA_TRACE(10, ("__kmp_create_monitor: try to create monitor\n"));
 
  KMP_MB(); /* Flush all pending memory write invalidates.  */
 
  th->th.th_info.ds.ds_tid = KMP_GTID_MONITOR;
  th->th.th_info.ds.ds_gtid = KMP_GTID_MONITOR;
#if KMP_REAL_TIME_FIX
  TCW_4(__kmp_global.g.g_time.dt.t_value,
        -1); // Will use it for synchronization a bit later.
#else
  TCW_4(__kmp_global.g.g_time.dt.t_value, 0);
#endif // KMP_REAL_TIME_FIX
 
#ifdef KMP_THREAD_ATTR
  if (__kmp_monitor_stksize == 0) {
    __kmp_monitor_stksize = KMP_DEFAULT_MONITOR_STKSIZE;
    auto_adj_size = TRUE;
  }
  status = pthread_attr_init(&thread_attr);
  if (status != 0) {
    __kmp_fatal(KMP_MSG(CantInitThreadAttrs), KMP_ERR(status), __kmp_msg_null);
  }
  status = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_JOINABLE);
  if (status != 0) {
    __kmp_fatal(KMP_MSG(CantSetMonitorState), KMP_ERR(status), __kmp_msg_null);
  }
 
#ifdef _POSIX_THREAD_ATTR_STACKSIZE
  status = pthread_attr_getstacksize(&thread_attr, &size);
  KMP_CHECK_SYSFAIL("pthread_attr_getstacksize", status);
#else
  size = __kmp_sys_min_stksize;
#endif /* _POSIX_THREAD_ATTR_STACKSIZE */
#endif /* KMP_THREAD_ATTR */
 
  if (__kmp_monitor_stksize == 0) {
    __kmp_monitor_stksize = KMP_DEFAULT_MONITOR_STKSIZE;
  }
  if (__kmp_monitor_stksize < __kmp_sys_min_stksize) {
    __kmp_monitor_stksize = __kmp_sys_min_stksize;
  }
 
  KA_TRACE(10, ("__kmp_create_monitor: default stacksize = %lu bytes,"
                "requested stacksize = %lu bytes\n",
                size, __kmp_monitor_stksize));
 
retry:
 
/* Set stack size for this thread now. */
#ifdef _POSIX_THREAD_ATTR_STACKSIZE
  KA_TRACE(10, ("__kmp_create_monitor: setting stacksize = %lu bytes,",
                __kmp_monitor_stksize));
  status = pthread_attr_setstacksize(&thread_attr, __kmp_monitor_stksize);
  if (status != 0) {
    if (auto_adj_size) {
      __kmp_monitor_stksize *= 2;
      goto retry;
    }
    kmp_msg_t err_code = KMP_ERR(status);
    __kmp_msg(kmp_ms_warning, // should this be fatal?  BB
              KMP_MSG(CantSetMonitorStackSize, (long int)__kmp_monitor_stksize),
              err_code, KMP_HNT(ChangeMonitorStackSize), __kmp_msg_null);
    if (__kmp_generate_warnings == kmp_warnings_off) {
      __kmp_str_free(&err_code.str);
    }
  }
#endif /* _POSIX_THREAD_ATTR_STACKSIZE */
 
  status =
      pthread_create(&handle, &thread_attr, __kmp_launch_monitor, (void *)th);
 
  if (status != 0) {
#ifdef _POSIX_THREAD_ATTR_STACKSIZE
    if (status == EINVAL) {
      if (auto_adj_size && (__kmp_monitor_stksize < (size_t)0x40000000)) {
        __kmp_monitor_stksize *= 2;
        goto retry;
      }
      __kmp_fatal(KMP_MSG(CantSetMonitorStackSize, __kmp_monitor_stksize),
                  KMP_ERR(status), KMP_HNT(IncreaseMonitorStackSize),
                  __kmp_msg_null);
    }
    if (status == ENOMEM) {
      __kmp_fatal(KMP_MSG(CantSetMonitorStackSize, __kmp_monitor_stksize),
                  KMP_ERR(status), KMP_HNT(DecreaseMonitorStackSize),
                  __kmp_msg_null);
    }
#endif /* _POSIX_THREAD_ATTR_STACKSIZE */
    if (status == EAGAIN) {
      __kmp_fatal(KMP_MSG(NoResourcesForMonitorThread), KMP_ERR(status),
                  KMP_HNT(DecreaseNumberOfThreadsInUse), __kmp_msg_null);
    }
    KMP_SYSFAIL("pthread_create", status);
  }
 
  th->th.th_info.ds.ds_thread = handle;
 
#if KMP_REAL_TIME_FIX
  // Wait for the monitor thread is really started and set its *priority*.
  KMP_DEBUG_ASSERT(sizeof(kmp_uint32) ==
                   sizeof(__kmp_global.g.g_time.dt.t_value));
  __kmp_wait_4((kmp_uint32 volatile *)&__kmp_global.g.g_time.dt.t_value, -1,
               &__kmp_neq_4, NULL);
#endif // KMP_REAL_TIME_FIX
 
#ifdef KMP_THREAD_ATTR
  status = pthread_attr_destroy(&thread_attr);
  if (status != 0) {
    kmp_msg_t err_code = KMP_ERR(status);
    __kmp_msg(kmp_ms_warning, KMP_MSG(CantDestroyThreadAttrs), err_code,
              __kmp_msg_null);
    if (__kmp_generate_warnings == kmp_warnings_off) {
      __kmp_str_free(&err_code.str);
    }
  }
#endif
 
  KMP_MB(); /* Flush all pending memory write invalidates.  */
 
  KA_TRACE(10, ("__kmp_create_monitor: monitor created %#.8lx\n",
                th->th.th_info.ds.ds_thread));
 
} // __kmp_create_monitor
#endif // KMP_USE_MONITOR
 
void __kmp_exit_thread(int exit_status) {
#if KMP_OS_WASI
// TODO: the wasm32-wasi-threads target does not yet support pthread_exit.
#else
  pthread_exit((void *)(intptr_t)exit_status);
#endif
} // __kmp_exit_thread
void __kmp_exit_thread(int exit_status) {…}
 
#if KMP_USE_MONITOR
void __kmp_resume_monitor();
 
extern "C" void __kmp_reap_monitor(kmp_info_t *th) {
  int status;
  void *exit_val;
 
  KA_TRACE(10, ("__kmp_reap_monitor: try to reap monitor thread with handle"
                " %#.8lx\n",
                th->th.th_info.ds.ds_thread));
 
  // If monitor has been created, its tid and gtid should be KMP_GTID_MONITOR.
  // If both tid and gtid are 0, it means the monitor did not ever start.
  // If both tid and gtid are KMP_GTID_DNE, the monitor has been shut down.
  KMP_DEBUG_ASSERT(th->th.th_info.ds.ds_tid == th->th.th_info.ds.ds_gtid);
  if (th->th.th_info.ds.ds_gtid != KMP_GTID_MONITOR) {
    KA_TRACE(10, ("__kmp_reap_monitor: monitor did not start, returning\n"));
    return;
  }
 
  KMP_MB(); /* Flush all pending memory write invalidates.  */
 
  /* First, check to see whether the monitor thread exists to wake it up. This
     is to avoid performance problem when the monitor sleeps during
     blocktime-size interval */
 
  status = pthread_kill(th->th.th_info.ds.ds_thread, 0);
  if (status != ESRCH) {
    __kmp_resume_monitor(); // Wake up the monitor thread
  }
  KA_TRACE(10, ("__kmp_reap_monitor: try to join with monitor\n"));
  status = pthread_join(th->th.th_info.ds.ds_thread, &exit_val);
  if (exit_val != th) {
    __kmp_fatal(KMP_MSG(ReapMonitorError), KMP_ERR(status), __kmp_msg_null);
  }
 
  th->th.th_info.ds.ds_tid = KMP_GTID_DNE;
  th->th.th_info.ds.ds_gtid = KMP_GTID_DNE;
 
  KA_TRACE(10, ("__kmp_reap_monitor: done reaping monitor thread with handle"
                " %#.8lx\n",
                th->th.th_info.ds.ds_thread));
 
  KMP_MB(); /* Flush all pending memory write invalidates.  */
}
#else
// Empty symbol to export (see exports_so.txt) when
// monitor thread feature is disabled
extern "C" void __kmp_reap_monitor(kmp_info_t *th) { (void)th; }
#endif // KMP_USE_MONITOR
 
void __kmp_reap_worker(kmp_info_t *th) {
  int status;
  void *exit_val;
 
  KMP_MB(); /* Flush all pending memory write invalidates.  */
 
  KA_TRACE(
      10, ("__kmp_reap_worker: try to reap T#%d\n", th->th.th_info.ds.ds_gtid));
 
  status = pthread_join(th->th.th_info.ds.ds_thread, &exit_val);
#ifdef KMP_DEBUG
  /* Don't expose these to the user until we understand when they trigger */
  if (status != 0) {
    __kmp_fatal(KMP_MSG(ReapWorkerError), KMP_ERR(status), __kmp_msg_null);
  }
  if (exit_val != th) {
    KA_TRACE(10, ("__kmp_reap_worker: worker T#%d did not reap properly, "
                  "exit_val = %p\n",
                  th->th.th_info.ds.ds_gtid, exit_val));
  }
#else
  (void)status; // unused variable
#endif /* KMP_DEBUG */
 
  KA_TRACE(10, ("__kmp_reap_worker: done reaping T#%d\n",
                th->th.th_info.ds.ds_gtid));
 
  KMP_MB(); /* Flush all pending memory write invalidates.  */
}
void __kmp_reap_worker(kmp_info_t *th) {…}
 
#if KMP_HANDLE_SIGNALS
 
static void __kmp_null_handler(int signo) {
  //  Do nothing, for doing SIG_IGN-type actions.
} // __kmp_null_handler
 
static void __kmp_team_handler(int signo) {
  if (__kmp_global.g.g_abort == 0) {
/* Stage 1 signal handler, let's shut down all of the threads */
#ifdef KMP_DEBUG
    __kmp_debug_printf("__kmp_team_handler: caught signal = %d\n", signo);
#endif
    switch (signo) {
    case SIGHUP:
    case SIGINT:
    case SIGQUIT:
    case SIGILL:
    case SIGABRT:
    case SIGFPE:
    case SIGBUS:
    case SIGSEGV:
#ifdef SIGSYS
    case SIGSYS:
#endif
    case SIGTERM:
      if (__kmp_debug_buf) {
        __kmp_dump_debug_buffer();
      }
      __kmp_unregister_library(); // cleanup shared memory
      KMP_MB(); // Flush all pending memory write invalidates.
      TCW_4(__kmp_global.g.g_abort, signo);
      KMP_MB(); // Flush all pending memory write invalidates.
      TCW_4(__kmp_global.g.g_done, TRUE);
      KMP_MB(); // Flush all pending memory write invalidates.
      break;
    default:
#ifdef KMP_DEBUG
      __kmp_debug_printf("__kmp_team_handler: unknown signal type");
#endif
      break;
    }
  }
} // __kmp_team_handler
 
static void __kmp_sigaction(int signum, const struct sigaction *act,
                            struct sigaction *oldact) {
  int rc = sigaction(signum, act, oldact);
  KMP_CHECK_SYSFAIL_ERRNO("sigaction", rc);
}
 
static void __kmp_install_one_handler(int sig, sig_func_t handler_func,
                                      int parallel_init) {
  KMP_MB(); // Flush all pending memory write invalidates.
  KB_TRACE(60,
           ("__kmp_install_one_handler( %d, ..., %d )\n", sig, parallel_init));
  if (parallel_init) {
    struct sigaction new_action;
    struct sigaction old_action;
    new_action.sa_handler = handler_func;
    new_action.sa_flags = 0;
    sigfillset(&new_action.sa_mask);
    __kmp_sigaction(sig, &new_action, &old_action);
    if (old_action.sa_handler == __kmp_sighldrs[sig].sa_handler) {
      sigaddset(&__kmp_sigset, sig);
    } else {
      // Restore/keep user's handler if one previously installed.
      __kmp_sigaction(sig, &old_action, NULL);
    }
  } else {
    // Save initial/system signal handlers to see if user handlers installed.
    __kmp_sigaction(sig, NULL, &__kmp_sighldrs[sig]);
  }
  KMP_MB(); // Flush all pending memory write invalidates.
} // __kmp_install_one_handler
 
static void __kmp_remove_one_handler(int sig) {
  KB_TRACE(60, ("__kmp_remove_one_handler( %d )\n", sig));
  if (sigismember(&__kmp_sigset, sig)) {
    struct sigaction old;
    KMP_MB(); // Flush all pending memory write invalidates.
    __kmp_sigaction(sig, &__kmp_sighldrs[sig], &old);
    if ((old.sa_handler != __kmp_team_handler) &&
        (old.sa_handler != __kmp_null_handler)) {
      // Restore the users signal handler.
      KB_TRACE(10, ("__kmp_remove_one_handler: oops, not our handler, "
                    "restoring: sig=%d\n",
                    sig));
      __kmp_sigaction(sig, &old, NULL);
    }
    sigdelset(&__kmp_sigset, sig);
    KMP_MB(); // Flush all pending memory write invalidates.
  }
} // __kmp_remove_one_handler
 
void __kmp_install_signals(int parallel_init) {
  KB_TRACE(10, ("__kmp_install_signals( %d )\n", parallel_init));
  if (__kmp_handle_signals || !parallel_init) {
    // If ! parallel_init, we do not install handlers, just save original
    // handlers. Let us do it even __handle_signals is 0.
    sigemptyset(&__kmp_sigset);
    __kmp_install_one_handler(SIGHUP, __kmp_team_handler, parallel_init);
    __kmp_install_one_handler(SIGINT, __kmp_team_handler, parallel_init);
    __kmp_install_one_handler(SIGQUIT, __kmp_team_handler, parallel_init);
    __kmp_install_one_handler(SIGILL, __kmp_team_handler, parallel_init);
    __kmp_install_one_handler(SIGABRT, __kmp_team_handler, parallel_init);
    __kmp_install_one_handler(SIGFPE, __kmp_team_handler, parallel_init);
    __kmp_install_one_handler(SIGBUS, __kmp_team_handler, parallel_init);
    __kmp_install_one_handler(SIGSEGV, __kmp_team_handler, parallel_init);
#ifdef SIGSYS
    __kmp_install_one_handler(SIGSYS, __kmp_team_handler, parallel_init);
#endif // SIGSYS
    __kmp_install_one_handler(SIGTERM, __kmp_team_handler, parallel_init);
#ifdef SIGPIPE
    __kmp_install_one_handler(SIGPIPE, __kmp_team_handler, parallel_init);
#endif // SIGPIPE
  }
} // __kmp_install_signals
 
void __kmp_remove_signals(void) {
  int sig;
  KB_TRACE(10, ("__kmp_remove_signals()\n"));
  for (sig = 1; sig < NSIG; ++sig) {
    __kmp_remove_one_handler(sig);
  }
} // __kmp_remove_signals
 
#endif // KMP_HANDLE_SIGNALS
 
void __kmp_enable(int new_state) {
#ifdef KMP_CANCEL_THREADS
  int status, old_state;
  status = pthread_setcancelstate(new_state, &old_state);
  KMP_CHECK_SYSFAIL("pthread_setcancelstate", status);
  KMP_DEBUG_ASSERT(old_state == PTHREAD_CANCEL_DISABLE);
#endif
}
void __kmp_enable(int new_state) {…}
 
void __kmp_disable(int *old_state) {
#ifdef KMP_CANCEL_THREADS
  int status;
  status = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, old_state);
  KMP_CHECK_SYSFAIL("pthread_setcancelstate", status);
#endif
}
void __kmp_disable(int *old_state) {…}
 
static void __kmp_atfork_prepare(void) {
  __kmp_acquire_bootstrap_lock(&__kmp_initz_lock);
  __kmp_acquire_bootstrap_lock(&__kmp_forkjoin_lock);
}
static void __kmp_atfork_prepare(void) {…}
 
static void __kmp_atfork_parent(void) {
  __kmp_release_bootstrap_lock(&__kmp_forkjoin_lock);
  __kmp_release_bootstrap_lock(&__kmp_initz_lock);
}
static void __kmp_atfork_parent(void) {…}
 
/* Reset the library so execution in the child starts "all over again" with
   clean data structures in initial states.  Don't worry about freeing memory
   allocated by parent, just abandon it to be safe. */
static void __kmp_atfork_child(void) {
  __kmp_release_bootstrap_lock(&__kmp_forkjoin_lock);
  __kmp_release_bootstrap_lock(&__kmp_initz_lock);
  /* TODO make sure this is done right for nested/sibling */
  // ATT:  Memory leaks are here? TODO: Check it and fix.
  /* KMP_ASSERT( 0 ); */
 
  ++__kmp_fork_count;
 
#if KMP_AFFINITY_SUPPORTED
#if KMP_OS_LINUX || KMP_OS_FREEBSD || KMP_OS_NETBSD || KMP_OS_DRAGONFLY ||     \
    KMP_OS_AIX
  // reset the affinity in the child to the initial thread
  // affinity in the parent
  kmp_set_thread_affinity_mask_initial();
#endif
  // Set default not to bind threads tightly in the child (we're expecting
  // over-subscription after the fork and this can improve things for
  // scripting languages that use OpenMP inside process-parallel code).
  if (__kmp_nested_proc_bind.bind_types != NULL) {
    __kmp_nested_proc_bind.bind_types[0] = proc_bind_false;
  }
  for (kmp_affinity_t *affinity : __kmp_affinities)
    *affinity = KMP_AFFINITY_INIT(affinity->env_var);
  __kmp_affin_fullMask = nullptr;
  __kmp_affin_origMask = nullptr;
  __kmp_topology = nullptr;
#endif // KMP_AFFINITY_SUPPORTED
 
#if KMP_USE_MONITOR
  __kmp_init_monitor = 0;
#endif
  __kmp_init_parallel = FALSE;
  __kmp_init_middle = FALSE;
  __kmp_init_serial = FALSE;
  TCW_4(__kmp_init_gtid, FALSE);
  __kmp_init_common = FALSE;
 
  TCW_4(__kmp_init_user_locks, FALSE);
#if !KMP_USE_DYNAMIC_LOCK
  __kmp_user_lock_table.used = 1;
  __kmp_user_lock_table.allocated = 0;
  __kmp_user_lock_table.table = NULL;
  __kmp_lock_blocks = NULL;
#endif
 
  __kmp_all_nth = 0;
  TCW_4(__kmp_nth, 0);
 
  __kmp_thread_pool = NULL;
  __kmp_thread_pool_insert_pt = NULL;
  __kmp_team_pool = NULL;
 
  /* Must actually zero all the *cache arguments passed to __kmpc_threadprivate
     here so threadprivate doesn't use stale data */
  KA_TRACE(10, ("__kmp_atfork_child: checking cache address list %p\n",
                __kmp_threadpriv_cache_list));
 
  while (__kmp_threadpriv_cache_list != NULL) {
 
    if (*__kmp_threadpriv_cache_list->addr != NULL) {
      KC_TRACE(50, ("__kmp_atfork_child: zeroing cache at address %p\n",
                    &(*__kmp_threadpriv_cache_list->addr)));
 
      *__kmp_threadpriv_cache_list->addr = NULL;
    }
    __kmp_threadpriv_cache_list = __kmp_threadpriv_cache_list->next;
  }
 
  __kmp_init_runtime = FALSE;
 
  /* reset statically initialized locks */
  __kmp_init_bootstrap_lock(&__kmp_initz_lock);
  __kmp_init_bootstrap_lock(&__kmp_stdio_lock);
  __kmp_init_bootstrap_lock(&__kmp_console_lock);
  __kmp_init_bootstrap_lock(&__kmp_task_team_lock);
 
#if USE_ITT_BUILD
  __kmp_itt_reset(); // reset ITT's global state
#endif /* USE_ITT_BUILD */
 
  {
    // Child process often get terminated without any use of OpenMP. That might
    // cause mapped shared memory file to be left unattended. Thus we postpone
    // library registration till middle initialization in the child process.
    __kmp_need_register_serial = FALSE;
    __kmp_serial_initialize();
  }
 
  /* This is necessary to make sure no stale data is left around */
  /* AC: customers complain that we use unsafe routines in the atfork
     handler. Mathworks: dlsym() is unsafe. We call dlsym and dlopen
     in dynamic_link when check the presence of shared tbbmalloc library.
     Suggestion is to make the library initialization lazier, similar
     to what done for __kmpc_begin(). */
  // TODO: synchronize all static initializations with regular library
  //       startup; look at kmp_global.cpp and etc.
  //__kmp_internal_begin ();
}
static void __kmp_atfork_child(void) {…}
 
void __kmp_register_atfork(void) {
  if (__kmp_need_register_atfork) {
#if !KMP_OS_WASI
    int status = pthread_atfork(__kmp_atfork_prepare, __kmp_atfork_parent,
                                __kmp_atfork_child);
    KMP_CHECK_SYSFAIL("pthread_atfork", status);
#endif
    __kmp_need_register_atfork = FALSE;
  }
}
void __kmp_register_atfork(void) {…}
 
void __kmp_suspend_initialize(void) {
  int status;
  status = pthread_mutexattr_init(&__kmp_suspend_mutex_attr);
  KMP_CHECK_SYSFAIL("pthread_mutexattr_init", status);
  status = pthread_condattr_init(&__kmp_suspend_cond_attr);
  KMP_CHECK_SYSFAIL("pthread_condattr_init", status);
}
void __kmp_suspend_initialize(void) {…}
 
void __kmp_suspend_initialize_thread(kmp_info_t *th) {
  int old_value = KMP_ATOMIC_LD_RLX(&th->th.th_suspend_init_count);
  int new_value = __kmp_fork_count + 1;
  // Return if already initialized
  if (old_value == new_value)
    return;
  // Wait, then return if being initialized
  if (old_value == -1 || !__kmp_atomic_compare_store(
                             &th->th.th_suspend_init_count, old_value, -1)) {
    while (KMP_ATOMIC_LD_ACQ(&th->th.th_suspend_init_count) != new_value) {
      KMP_CPU_PAUSE();
    }
  } else {
    // Claim to be the initializer and do initializations
    int status;
    status = pthread_cond_init(&th->th.th_suspend_cv.c_cond,
                               &__kmp_suspend_cond_attr);
    KMP_CHECK_SYSFAIL("pthread_cond_init", status);
    status = pthread_mutex_init(&th->th.th_suspend_mx.m_mutex,
                                &__kmp_suspend_mutex_attr);
    KMP_CHECK_SYSFAIL("pthread_mutex_init", status);
    KMP_ATOMIC_ST_REL(&th->th.th_suspend_init_count, new_value);
  }
}
void __kmp_suspend_initialize_thread(kmp_info_t *th) {…}
 
void __kmp_suspend_uninitialize_thread(kmp_info_t *th) {
  if (KMP_ATOMIC_LD_ACQ(&th->th.th_suspend_init_count) > __kmp_fork_count) {
    /* this means we have initialize the suspension pthread objects for this
       thread in this instance of the process */
    int status;
 
    status = pthread_cond_destroy(&th->th.th_suspend_cv.c_cond);
    if (status != 0 && status != EBUSY) {
      KMP_SYSFAIL("pthread_cond_destroy", status);
    }
    status = pthread_mutex_destroy(&th->th.th_suspend_mx.m_mutex);
    if (status != 0 && status != EBUSY) {
      KMP_SYSFAIL("pthread_mutex_destroy", status);
    }
    --th->th.th_suspend_init_count;
    KMP_DEBUG_ASSERT(KMP_ATOMIC_LD_RLX(&th->th.th_suspend_init_count) ==
                     __kmp_fork_count);
  }
}
void __kmp_suspend_uninitialize_thread(kmp_info_t *th) {…}
 
// return true if lock obtained, false otherwise
int __kmp_try_suspend_mx(kmp_info_t *th) {
  return (pthread_mutex_trylock(&th->th.th_suspend_mx.m_mutex) == 0);
}
int __kmp_try_suspend_mx(kmp_info_t *th) {…}
 
void __kmp_lock_suspend_mx(kmp_info_t *th) {
  int status = pthread_mutex_lock(&th->th.th_suspend_mx.m_mutex);
  KMP_CHECK_SYSFAIL("pthread_mutex_lock", status);
}
void __kmp_lock_suspend_mx(kmp_info_t *th) {…}
 
void __kmp_unlock_suspend_mx(kmp_info_t *th) {
  int status = pthread_mutex_unlock(&th->th.th_suspend_mx.m_mutex);
  KMP_CHECK_SYSFAIL("pthread_mutex_unlock", status);
}
void __kmp_unlock_suspend_mx(kmp_info_t *th) {…}
 
/* This routine puts the calling thread to sleep after setting the
   sleep bit for the indicated flag variable to true. */
template <class C>
static inline void __kmp_suspend_template(int th_gtid, C *flag) {
  KMP_TIME_DEVELOPER_PARTITIONED_BLOCK(USER_suspend);
  kmp_info_t *th = __kmp_threads[th_gtid];
  int status;
  typename C::flag_t old_spin;
 
  KF_TRACE(30, ("__kmp_suspend_template: T#%d enter for flag = %p\n", th_gtid,
                flag->get()));
 
  __kmp_suspend_initialize_thread(th);
 
  __kmp_lock_suspend_mx(th);
 
  KF_TRACE(10, ("__kmp_suspend_template: T#%d setting sleep bit for spin(%p)\n",
                th_gtid, flag->get()));
 
  /* TODO: shouldn't this use release semantics to ensure that
     __kmp_suspend_initialize_thread gets called first? */
  old_spin = flag->set_sleeping();
  TCW_PTR(th->th.th_sleep_loc, (void *)flag);
  th->th.th_sleep_loc_type = flag->get_type();
  if (__kmp_dflt_blocktime == KMP_MAX_BLOCKTIME &&
      __kmp_pause_status != kmp_soft_paused) {
    flag->unset_sleeping();
    TCW_PTR(th->th.th_sleep_loc, NULL);
    th->th.th_sleep_loc_type = flag_unset;
    __kmp_unlock_suspend_mx(th);
    return;
  }
  KF_TRACE(5, ("__kmp_suspend_template: T#%d set sleep bit for spin(%p)==%x,"
               " was %x\n",
               th_gtid, flag->get(), flag->load(), old_spin));
 
  if (flag->done_check_val(old_spin) || flag->done_check()) {
    flag->unset_sleeping();
    TCW_PTR(th->th.th_sleep_loc, NULL);
    th->th.th_sleep_loc_type = flag_unset;
    KF_TRACE(5, ("__kmp_suspend_template: T#%d false alarm, reset sleep bit "
                 "for spin(%p)\n",
                 th_gtid, flag->get()));
  } else {
    /* Encapsulate in a loop as the documentation states that this may
       "with low probability" return when the condition variable has
       not been signaled or broadcast */
    int deactivated = FALSE;
 
    while (flag->is_sleeping()) {
#ifdef DEBUG_SUSPEND
      char buffer[128];
      __kmp_suspend_count++;
      __kmp_print_cond(buffer, &th->th.th_suspend_cv);
      __kmp_printf("__kmp_suspend_template: suspending T#%d: %s\n", th_gtid,
                   buffer);
#endif
      // Mark the thread as no longer active (only in the first iteration of the
      // loop).
      if (!deactivated) {
        th->th.th_active = FALSE;
        if (th->th.th_active_in_pool) {
          th->th.th_active_in_pool = FALSE;
          KMP_ATOMIC_DEC(&__kmp_thread_pool_active_nth);
          KMP_DEBUG_ASSERT(TCR_4(__kmp_thread_pool_active_nth) >= 0);
        }
        deactivated = TRUE;
      }
 
      KMP_DEBUG_ASSERT(th->th.th_sleep_loc);
      KMP_DEBUG_ASSERT(flag->get_type() == th->th.th_sleep_loc_type);
 
#if USE_SUSPEND_TIMEOUT
      struct timespec now;
      struct timeval tval;
      int msecs;
 
      status = gettimeofday(&tval, NULL);
      KMP_CHECK_SYSFAIL_ERRNO("gettimeofday", status);
      TIMEVAL_TO_TIMESPEC(&tval, &now);
 
      msecs = (4 * __kmp_dflt_blocktime) + 200;
      now.tv_sec += msecs / 1000;
      now.tv_nsec += (msecs % 1000) * 1000;
 
      KF_TRACE(15, ("__kmp_suspend_template: T#%d about to perform "
                    "pthread_cond_timedwait\n",
                    th_gtid));
      status = pthread_cond_timedwait(&th->th.th_suspend_cv.c_cond,
                                      &th->th.th_suspend_mx.m_mutex, &now);
#else
      KF_TRACE(15, ("__kmp_suspend_template: T#%d about to perform"
                    " pthread_cond_wait\n",
                    th_gtid));
      status = pthread_cond_wait(&th->th.th_suspend_cv.c_cond,
                                 &th->th.th_suspend_mx.m_mutex);
#endif // USE_SUSPEND_TIMEOUT
 
      if ((status != 0) && (status != EINTR) && (status != ETIMEDOUT)) {
        KMP_SYSFAIL("pthread_cond_wait", status);
      }
 
      KMP_DEBUG_ASSERT(flag->get_type() == flag->get_ptr_type());
 
      if (!flag->is_sleeping() &&
          ((status == EINTR) || (status == ETIMEDOUT))) {
        // if interrupt or timeout, and thread is no longer sleeping, we need to
        // make sure sleep_loc gets reset; however, this shouldn't be needed if
        // we woke up with resume
        flag->unset_sleeping();
        TCW_PTR(th->th.th_sleep_loc, NULL);
        th->th.th_sleep_loc_type = flag_unset;
      }
#ifdef KMP_DEBUG
      if (status == ETIMEDOUT) {
        if (flag->is_sleeping()) {
          KF_TRACE(100,
                   ("__kmp_suspend_template: T#%d timeout wakeup\n", th_gtid));
        } else {
          KF_TRACE(2, ("__kmp_suspend_template: T#%d timeout wakeup, sleep bit "
                       "not set!\n",
                       th_gtid));
          TCW_PTR(th->th.th_sleep_loc, NULL);
          th->th.th_sleep_loc_type = flag_unset;
        }
      } else if (flag->is_sleeping()) {
        KF_TRACE(100,
                 ("__kmp_suspend_template: T#%d spurious wakeup\n", th_gtid));
      }
#endif
    } // while
 
    // Mark the thread as active again (if it was previous marked as inactive)
    if (deactivated) {
      th->th.th_active = TRUE;
      if (TCR_4(th->th.th_in_pool)) {
        KMP_ATOMIC_INC(&__kmp_thread_pool_active_nth);
        th->th.th_active_in_pool = TRUE;
      }
    }
  }
  // We may have had the loop variable set before entering the loop body;
  // so we need to reset sleep_loc.
  TCW_PTR(th->th.th_sleep_loc, NULL);
  th->th.th_sleep_loc_type = flag_unset;
 
  KMP_DEBUG_ASSERT(!flag->is_sleeping());
  KMP_DEBUG_ASSERT(!th->th.th_sleep_loc);
#ifdef DEBUG_SUSPEND
  {
    char buffer[128];
    __kmp_print_cond(buffer, &th->th.th_suspend_cv);
    __kmp_printf("__kmp_suspend_template: T#%d has awakened: %s\n", th_gtid,
                 buffer);
  }
#endif
 
  __kmp_unlock_suspend_mx(th);
  KF_TRACE(30, ("__kmp_suspend_template: T#%d exit\n", th_gtid));
}
static inline void __kmp_suspend_template(int th_gtid, C *flag) {…}
 
template <bool C, bool S>
void __kmp_suspend_32(int th_gtid, kmp_flag_32<C, S> *flag) {
  __kmp_suspend_template(th_gtid, flag);
}
void __kmp_suspend_32(int th_gtid, kmp_flag_32<C, S> *flag) {…}
template <bool C, bool S>
void __kmp_suspend_64(int th_gtid, kmp_flag_64<C, S> *flag) {
  __kmp_suspend_template(th_gtid, flag);
}
void __kmp_suspend_64(int th_gtid, kmp_flag_64<C, S> *flag) {…}
template <bool C, bool S>
void __kmp_atomic_suspend_64(int th_gtid, kmp_atomic_flag_64<C, S> *flag) {
  __kmp_suspend_template(th_gtid, flag);
}
void __kmp_atomic_suspend_64(int th_gtid, kmp_atomic_flag_64<C, S> *flag) {…}
void __kmp_suspend_oncore(int th_gtid, kmp_flag_oncore *flag) {
  __kmp_suspend_template(th_gtid, flag);
}
void __kmp_suspend_oncore(int th_gtid, kmp_flag_oncore *flag) {…}
 
template void __kmp_suspend_32<false, false>(int, kmp_flag_32<false, false> *);
template void __kmp_suspend_64<false, true>(int, kmp_flag_64<false, true> *);
template void __kmp_suspend_64<true, false>(int, kmp_flag_64<true, false> *);
template void
__kmp_atomic_suspend_64<false, true>(int, kmp_atomic_flag_64<false, true> *);
template void
__kmp_atomic_suspend_64<true, false>(int, kmp_atomic_flag_64<true, false> *);
 
/* This routine signals the thread specified by target_gtid to wake up
   after setting the sleep bit indicated by the flag argument to FALSE.
   The target thread must already have called __kmp_suspend_template() */
template <class C>
static inline void __kmp_resume_template(int target_gtid, C *flag) {
  KMP_TIME_DEVELOPER_PARTITIONED_BLOCK(USER_resume);
  kmp_info_t *th = __kmp_threads[target_gtid];
  int status;
 
#ifdef KMP_DEBUG
  int gtid = TCR_4(__kmp_init_gtid) ? __kmp_get_gtid() : -1;
#endif
 
  KF_TRACE(30, ("__kmp_resume_template: T#%d wants to wakeup T#%d enter\n",
                gtid, target_gtid));
  KMP_DEBUG_ASSERT(gtid != target_gtid);
 
  __kmp_suspend_initialize_thread(th);
 
  __kmp_lock_suspend_mx(th);
 
  if (!flag || flag != th->th.th_sleep_loc) {
    // coming from __kmp_null_resume_wrapper, or thread is now sleeping on a
    // different location; wake up at new location
    flag = (C *)CCAST(void *, th->th.th_sleep_loc);
  }
 
  // First, check if the flag is null or its type has changed. If so, someone
  // else woke it up.
  if (!flag) { // Thread doesn't appear to be sleeping on anything
    KF_TRACE(5, ("__kmp_resume_template: T#%d exiting, thread T#%d already "
                 "awake: flag(%p)\n",
                 gtid, target_gtid, (void *)NULL));
    __kmp_unlock_suspend_mx(th);
    return;
  } else if (flag->get_type() != th->th.th_sleep_loc_type) {
    // Flag type does not appear to match this function template; possibly the
    // thread is sleeping on something else. Try null resume again.
    KF_TRACE(
        5,
        ("__kmp_resume_template: T#%d retrying, thread T#%d Mismatch flag(%p), "
         "spin(%p) type=%d ptr_type=%d\n",
         gtid, target_gtid, flag, flag->get(), flag->get_type(),
         th->th.th_sleep_loc_type));
    __kmp_unlock_suspend_mx(th);
    __kmp_null_resume_wrapper(th);
    return;
  } else { // if multiple threads are sleeping, flag should be internally
    // referring to a specific thread here
    if (!flag->is_sleeping()) {
      KF_TRACE(5, ("__kmp_resume_template: T#%d exiting, thread T#%d already "
                   "awake: flag(%p): %u\n",
                   gtid, target_gtid, flag->get(), (unsigned int)flag->load()));
      __kmp_unlock_suspend_mx(th);
      return;
    }
  }
  KMP_DEBUG_ASSERT(flag);
  flag->unset_sleeping();
  TCW_PTR(th->th.th_sleep_loc, NULL);
  th->th.th_sleep_loc_type = flag_unset;
 
  KF_TRACE(5, ("__kmp_resume_template: T#%d about to wakeup T#%d, reset "
               "sleep bit for flag's loc(%p): %u\n",
               gtid, target_gtid, flag->get(), (unsigned int)flag->load()));
 
#ifdef DEBUG_SUSPEND
  {
    char buffer[128];
    __kmp_print_cond(buffer, &th->th.th_suspend_cv);
    __kmp_printf("__kmp_resume_template: T#%d resuming T#%d: %s\n", gtid,
                 target_gtid, buffer);
  }
#endif
  status = pthread_cond_signal(&th->th.th_suspend_cv.c_cond);
  KMP_CHECK_SYSFAIL("pthread_cond_signal", status);
  __kmp_unlock_suspend_mx(th);
  KF_TRACE(30, ("__kmp_resume_template: T#%d exiting after signaling wake up"
                " for T#%d\n",
                gtid, target_gtid));
}
static inline void __kmp_resume_template(int target_gtid, C *flag) {…}
 
template <bool C, bool S>
void __kmp_resume_32(int target_gtid, kmp_flag_32<C, S> *flag) {
  __kmp_resume_template(target_gtid, flag);
}
void __kmp_resume_32(int target_gtid, kmp_flag_32<C, S> *flag) {…}
template <bool C, bool S>
void __kmp_resume_64(int target_gtid, kmp_flag_64<C, S> *flag) {
  __kmp_resume_template(target_gtid, flag);
}
void __kmp_resume_64(int target_gtid, kmp_flag_64<C, S> *flag) {…}
template <bool C, bool S>
void __kmp_atomic_resume_64(int target_gtid, kmp_atomic_flag_64<C, S> *flag) {
  __kmp_resume_template(target_gtid, flag);
}
void __kmp_atomic_resume_64(int target_gtid, kmp_atomic_flag_64<C, S> *flag) {…}
void __kmp_resume_oncore(int target_gtid, kmp_flag_oncore *flag) {
  __kmp_resume_template(target_gtid, flag);
}
void __kmp_resume_oncore(int target_gtid, kmp_flag_oncore *flag) {…}
 
template void __kmp_resume_32<false, true>(int, kmp_flag_32<false, true> *);
template void __kmp_resume_32<false, false>(int, kmp_flag_32<false, false> *);
template void __kmp_resume_64<false, true>(int, kmp_flag_64<false, true> *);
template void
__kmp_atomic_resume_64<false, true>(int, kmp_atomic_flag_64<false, true> *);
 
#if KMP_USE_MONITOR
void __kmp_resume_monitor() {
  KMP_TIME_DEVELOPER_PARTITIONED_BLOCK(USER_resume);
  int status;
#ifdef KMP_DEBUG
  int gtid = TCR_4(__kmp_init_gtid) ? __kmp_get_gtid() : -1;
  KF_TRACE(30, ("__kmp_resume_monitor: T#%d wants to wakeup T#%d enter\n", gtid,
                KMP_GTID_MONITOR));
  KMP_DEBUG_ASSERT(gtid != KMP_GTID_MONITOR);
#endif
  status = pthread_mutex_lock(&__kmp_wait_mx.m_mutex);
  KMP_CHECK_SYSFAIL("pthread_mutex_lock", status);
#ifdef DEBUG_SUSPEND
  {
    char buffer[128];
    __kmp_print_cond(buffer, &__kmp_wait_cv.c_cond);
    __kmp_printf("__kmp_resume_monitor: T#%d resuming T#%d: %s\n", gtid,
                 KMP_GTID_MONITOR, buffer);
  }
#endif
  status = pthread_cond_signal(&__kmp_wait_cv.c_cond);
  KMP_CHECK_SYSFAIL("pthread_cond_signal", status);
  status = pthread_mutex_unlock(&__kmp_wait_mx.m_mutex);
  KMP_CHECK_SYSFAIL("pthread_mutex_unlock", status);
  KF_TRACE(30, ("__kmp_resume_monitor: T#%d exiting after signaling wake up"
                " for T#%d\n",
                gtid, KMP_GTID_MONITOR));
}
#endif // KMP_USE_MONITOR
 
void __kmp_yield() { sched_yield(); }
 
void __kmp_gtid_set_specific(int gtid) {
  if (__kmp_init_gtid) {
    int status;
    status = pthread_setspecific(__kmp_gtid_threadprivate_key,
                                 (void *)(intptr_t)(gtid + 1));
    KMP_CHECK_SYSFAIL("pthread_setspecific", status);
  } else {
    KA_TRACE(50, ("__kmp_gtid_set_specific: runtime shutdown, returning\n"));
  }
}
void __kmp_gtid_set_specific(int gtid) {…}
 
int __kmp_gtid_get_specific() {
  int gtid;
  if (!__kmp_init_gtid) {
    KA_TRACE(50, ("__kmp_gtid_get_specific: runtime shutdown, returning "
                  "KMP_GTID_SHUTDOWN\n"));
    return KMP_GTID_SHUTDOWN;
  }
  gtid = (int)(size_t)pthread_getspecific(__kmp_gtid_threadprivate_key);
  if (gtid == 0) {
    gtid = KMP_GTID_DNE;
  } else {
    gtid--;
  }
  KA_TRACE(50, ("__kmp_gtid_get_specific: key:%d gtid:%d\n",
                __kmp_gtid_threadprivate_key, gtid));
  return gtid;
}
int __kmp_gtid_get_specific() {…}
 
double __kmp_read_cpu_time(void) {
  /*clock_t   t;*/
  struct tms buffer;
 
  /*t =*/times(&buffer);
 
  return (double)(buffer.tms_utime + buffer.tms_cutime) /
         (double)CLOCKS_PER_SEC;
}
double __kmp_read_cpu_time(void) {…}
 
int __kmp_read_system_info(struct kmp_sys_info *info) {
  int status;
  struct rusage r_usage;
 
  memset(info, 0, sizeof(*info));
 
  status = getrusage(RUSAGE_SELF, &r_usage);
  KMP_CHECK_SYSFAIL_ERRNO("getrusage", status);
 
#if !KMP_OS_WASI
  // The maximum resident set size utilized (in kilobytes)
  info->maxrss = r_usage.ru_maxrss;
  // The number of page faults serviced without any I/O
  info->minflt = r_usage.ru_minflt;
  // The number of page faults serviced that required I/O
  info->majflt = r_usage.ru_majflt;
  // The number of times a process was "swapped" out of memory
  info->nswap = r_usage.ru_nswap;
  // The number of times the file system had to perform input
  info->inblock = r_usage.ru_inblock;
  // The number of times the file system had to perform output
  info->oublock = r_usage.ru_oublock;
  // The number of times a context switch was voluntarily
  info->nvcsw = r_usage.ru_nvcsw;
  // The number of times a context switch was forced
  info->nivcsw = r_usage.ru_nivcsw;
#endif
 
  return (status != 0);
}
int __kmp_read_system_info(struct kmp_sys_info *info) {…}
 
void __kmp_read_system_time(double *delta) {
  double t_ns;
  struct timeval tval;
  struct timespec stop;
  int status;
 
  status = gettimeofday(&tval, NULL);
  KMP_CHECK_SYSFAIL_ERRNO("gettimeofday", status);
  TIMEVAL_TO_TIMESPEC(&tval, &stop);
  t_ns = (double)(TS2NS(stop) - TS2NS(__kmp_sys_timer_data.start));
  *delta = (t_ns * 1e-9);
}
void __kmp_read_system_time(double *delta) {…}
 
void __kmp_clear_system_time(void) {
  struct timeval tval;
  int status;
  status = gettimeofday(&tval, NULL);
  KMP_CHECK_SYSFAIL_ERRNO("gettimeofday", status);
  TIMEVAL_TO_TIMESPEC(&tval, &__kmp_sys_timer_data.start);
}
void __kmp_clear_system_time(void) {…}
 
static int __kmp_get_xproc(void) {
 
  int r = 0;
 
#if KMP_OS_LINUX
 
  __kmp_type_convert(sysconf(_SC_NPROCESSORS_CONF), &(r));
 
#elif KMP_OS_DRAGONFLY || KMP_OS_FREEBSD || KMP_OS_NETBSD || KMP_OS_OPENBSD || \
    KMP_OS_HAIKU || KMP_OS_HURD || KMP_OS_SOLARIS || KMP_OS_WASI || KMP_OS_AIX
 
  __kmp_type_convert(sysconf(_SC_NPROCESSORS_ONLN), &(r));
 
#elif KMP_OS_DARWIN
 
  size_t len = sizeof(r);
  sysctlbyname("hw.logicalcpu", &r, &len, NULL, 0);
 
#else
 
#error "Unknown or unsupported OS."
 
#endif
 
  return r > 0 ? r : 2; /* guess value of 2 if OS told us 0 */
 
} // __kmp_get_xproc
static int __kmp_get_xproc(void) {…}
 
int __kmp_read_from_file(char const *path, char const *format, ...) {
  int result;
  va_list args;
 
  va_start(args, format);
  FILE *f = fopen(path, "rb");
  if (f == NULL) {
    va_end(args);
    return 0;
  }
  result = vfscanf(f, format, args);
  fclose(f);
  va_end(args);
 
  return result;
}
int __kmp_read_from_file(char const *path, char const *format, ...) {…}
 
void __kmp_runtime_initialize(void) {
  int status;
  pthread_mutexattr_t mutex_attr;
  pthread_condattr_t cond_attr;
 
  if (__kmp_init_runtime) {
    return;
  }
 
#if (KMP_ARCH_X86 || KMP_ARCH_X86_64)
  if (!__kmp_cpuinfo.initialized) {
    __kmp_query_cpuid(&__kmp_cpuinfo);
  }
#endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
 
  __kmp_xproc = __kmp_get_xproc();
 
#if !KMP_32_BIT_ARCH
  struct rlimit rlim;
  // read stack size of calling thread, save it as default for worker threads;
  // this should be done before reading environment variables
  status = getrlimit(RLIMIT_STACK, &rlim);
  if (status == 0) { // success?
    __kmp_stksize = rlim.rlim_cur;
    __kmp_check_stksize(&__kmp_stksize); // check value and adjust if needed
  }
#endif /* KMP_32_BIT_ARCH */
 
  if (sysconf(_SC_THREADS)) {
 
    /* Query the maximum number of threads */
    __kmp_type_convert(sysconf(_SC_THREAD_THREADS_MAX), &(__kmp_sys_max_nth));
#ifdef __ve__
    if (__kmp_sys_max_nth == -1) {
      // VE's pthread supports only up to 64 threads per a VE process.
      // So we use that KMP_MAX_NTH (predefined as 64) here.
      __kmp_sys_max_nth = KMP_MAX_NTH;
    }
#else
    if (__kmp_sys_max_nth == -1) {
      /* Unlimited threads for NPTL */
      __kmp_sys_max_nth = INT_MAX;
    } else if (__kmp_sys_max_nth <= 1) {
      /* Can't tell, just use PTHREAD_THREADS_MAX */
      __kmp_sys_max_nth = KMP_MAX_NTH;
    }
#endif
 
    /* Query the minimum stack size */
    __kmp_sys_min_stksize = sysconf(_SC_THREAD_STACK_MIN);
    if (__kmp_sys_min_stksize <= 1) {
      __kmp_sys_min_stksize = KMP_MIN_STKSIZE;
    }
  }
 
  /* Set up minimum number of threads to switch to TLS gtid */
  __kmp_tls_gtid_min = KMP_TLS_GTID_MIN;
 
  status = pthread_key_create(&__kmp_gtid_threadprivate_key,
                              __kmp_internal_end_dest);
  KMP_CHECK_SYSFAIL("pthread_key_create", status);
  status = pthread_mutexattr_init(&mutex_attr);
  KMP_CHECK_SYSFAIL("pthread_mutexattr_init", status);
  status = pthread_mutex_init(&__kmp_wait_mx.m_mutex, &mutex_attr);
  KMP_CHECK_SYSFAIL("pthread_mutex_init", status);
  status = pthread_mutexattr_destroy(&mutex_attr);
  KMP_CHECK_SYSFAIL("pthread_mutexattr_destroy", status);
  status = pthread_condattr_init(&cond_attr);
  KMP_CHECK_SYSFAIL("pthread_condattr_init", status);
  status = pthread_cond_init(&__kmp_wait_cv.c_cond, &cond_attr);
  KMP_CHECK_SYSFAIL("pthread_cond_init", status);
  status = pthread_condattr_destroy(&cond_attr);
  KMP_CHECK_SYSFAIL("pthread_condattr_destroy", status);
#if USE_ITT_BUILD
  __kmp_itt_initialize();
#endif /* USE_ITT_BUILD */
 
  __kmp_init_runtime = TRUE;
}
void __kmp_runtime_initialize(void) {…}
 
void __kmp_runtime_destroy(void) {
  int status;
 
  if (!__kmp_init_runtime) {
    return; // Nothing to do.
  }
 
#if USE_ITT_BUILD
  __kmp_itt_destroy();
#endif /* USE_ITT_BUILD */
 
  status = pthread_key_delete(__kmp_gtid_threadprivate_key);
  KMP_CHECK_SYSFAIL("pthread_key_delete", status);
 
  status = pthread_mutex_destroy(&__kmp_wait_mx.m_mutex);
  if (status != 0 && status != EBUSY) {
    KMP_SYSFAIL("pthread_mutex_destroy", status);
  }
  status = pthread_cond_destroy(&__kmp_wait_cv.c_cond);
  if (status != 0 && status != EBUSY) {
    KMP_SYSFAIL("pthread_cond_destroy", status);
  }
#if KMP_AFFINITY_SUPPORTED
  __kmp_affinity_uninitialize();
#endif
 
  __kmp_init_runtime = FALSE;
}
void __kmp_runtime_destroy(void) {…}
 
/* Put the thread to sleep for a time period */
/* NOTE: not currently used anywhere */
void __kmp_thread_sleep(int millis) { sleep((millis + 500) / 1000); }
 
/* Calculate the elapsed wall clock time for the user */
void __kmp_elapsed(double *t) {
  int status;
#ifdef FIX_SGI_CLOCK
  struct timespec ts;
 
  status = clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &ts);
  KMP_CHECK_SYSFAIL_ERRNO("clock_gettime", status);
  *t =
      (double)ts.tv_nsec * (1.0 / (double)KMP_NSEC_PER_SEC) + (double)ts.tv_sec;
#else
  struct timeval tv;
 
  status = gettimeofday(&tv, NULL);
  KMP_CHECK_SYSFAIL_ERRNO("gettimeofday", status);
  *t =
      (double)tv.tv_usec * (1.0 / (double)KMP_USEC_PER_SEC) + (double)tv.tv_sec;
#endif
}
void __kmp_elapsed(double *t) {…}
 
/* Calculate the elapsed wall clock tick for the user */
void __kmp_elapsed_tick(double *t) { *t = 1 / (double)CLOCKS_PER_SEC; }
 
/* Return the current time stamp in nsec */
kmp_uint64 __kmp_now_nsec() {
  struct timeval t;
  gettimeofday(&t, NULL);
  kmp_uint64 nsec = (kmp_uint64)KMP_NSEC_PER_SEC * (kmp_uint64)t.tv_sec +
                    (kmp_uint64)1000 * (kmp_uint64)t.tv_usec;
  return nsec;
}
kmp_uint64 __kmp_now_nsec() {…}
 
#if KMP_ARCH_X86 || KMP_ARCH_X86_64
/* Measure clock ticks per millisecond */
void __kmp_initialize_system_tick() {
  kmp_uint64 now, nsec2, diff;
  kmp_uint64 delay = 1000000; // ~450 usec on most machines.
  kmp_uint64 nsec = __kmp_now_nsec();
  kmp_uint64 goal = __kmp_hardware_timestamp() + delay;
  while ((now = __kmp_hardware_timestamp()) < goal)
    ;
  nsec2 = __kmp_now_nsec();
  diff = nsec2 - nsec;
  if (diff > 0) {
    double tpus = 1000.0 * (double)(delay + (now - goal)) / (double)diff;
    if (tpus > 0.0) {
      __kmp_ticks_per_msec = (kmp_uint64)(tpus * 1000.0);
      __kmp_ticks_per_usec = (kmp_uint64)tpus;
    }
  }
}
#endif
 
/* Determine whether the given address is mapped into the current address
   space. */
 
int __kmp_is_address_mapped(void *addr) {
 
  int found = 0;
  int rc;
 
#if KMP_OS_LINUX || KMP_OS_HURD
 
  /* On GNUish OSes, read the /proc/<pid>/maps pseudo-file to get all the
     address ranges mapped into the address space. */
 
  char *name = __kmp_str_format("/proc/%d/maps", getpid());
  FILE *file = NULL;
 
  file = fopen(name, "r");
  KMP_ASSERT(file != NULL);
 
  for (;;) {
 
    void *beginning = NULL;
    void *ending = NULL;
    char perms[5];
 
    rc = fscanf(file, "%p-%p %4s %*[^\n]\n", &beginning, &ending, perms);
    if (rc == EOF) {
      break;
    }
    KMP_ASSERT(rc == 3 &&
               KMP_STRLEN(perms) == 4); // Make sure all fields are read.
 
    // Ending address is not included in the region, but beginning is.
    if ((addr >= beginning) && (addr < ending)) {
      perms[2] = 0; // 3th and 4th character does not matter.
      if (strcmp(perms, "rw") == 0) {
        // Memory we are looking for should be readable and writable.
        found = 1;
      }
      break;
    }
  }
 
  // Free resources.
  fclose(file);
  KMP_INTERNAL_FREE(name);
#elif KMP_OS_FREEBSD
  char *buf;
  size_t lstsz;
  int mib[] = {CTL_KERN, KERN_PROC, KERN_PROC_VMMAP, getpid()};
  rc = sysctl(mib, 4, NULL, &lstsz, NULL, 0);
  if (rc < 0)
    return 0;
  // We pass from number of vm entry's semantic
  // to size of whole entry map list.
  lstsz = lstsz * 4 / 3;
  buf = reinterpret_cast<char *>(KMP_INTERNAL_MALLOC(lstsz));
  rc = sysctl(mib, 4, buf, &lstsz, NULL, 0);
  if (rc < 0) {
    KMP_INTERNAL_FREE(buf);
    return 0;
  }
 
  char *lw = buf;
  char *up = buf + lstsz;
 
  while (lw < up) {
    struct kinfo_vmentry *cur = reinterpret_cast<struct kinfo_vmentry *>(lw);
    size_t cursz = cur->kve_structsize;
    if (cursz == 0)
      break;
    void *start = reinterpret_cast<void *>(cur->kve_start);
    void *end = reinterpret_cast<void *>(cur->kve_end);
    // Readable/Writable addresses within current map entry
    if ((addr >= start) && (addr < end)) {
      if ((cur->kve_protection & KVME_PROT_READ) != 0 &&
          (cur->kve_protection & KVME_PROT_WRITE) != 0) {
        found = 1;
        break;
      }
    }
    lw += cursz;
  }
  KMP_INTERNAL_FREE(buf);
#elif KMP_OS_DRAGONFLY
  char err[_POSIX2_LINE_MAX];
  kinfo_proc *proc;
  vmspace sp;
  vm_map *cur;
  vm_map_entry entry, *c;
  struct proc p;
  kvm_t *fd;
  uintptr_t uaddr;
  int num;
 
  fd = kvm_openfiles(nullptr, nullptr, nullptr, O_RDONLY, err);
  if (!fd) {
    return 0;
  }
 
  proc = kvm_getprocs(fd, KERN_PROC_PID, getpid(), &num);
 
  if (kvm_read(fd, static_cast<uintptr_t>(proc->kp_paddr), &p, sizeof(p)) !=
          sizeof(p) ||
      kvm_read(fd, reinterpret_cast<uintptr_t>(p.p_vmspace), &sp, sizeof(sp)) !=
          sizeof(sp)) {
    kvm_close(fd);
    return 0;
  }
 
  (void)rc;
  cur = &sp.vm_map;
  uaddr = reinterpret_cast<uintptr_t>(addr);
  for (c = kvm_vm_map_entry_first(fd, cur, &entry); c;
       c = kvm_vm_map_entry_next(fd, c, &entry)) {
    if ((uaddr >= entry.ba.start) && (uaddr <= entry.ba.end)) {
      if ((entry.protection & VM_PROT_READ) != 0 &&
          (entry.protection & VM_PROT_WRITE) != 0) {
        found = 1;
        break;
      }
    }
  }
 
  kvm_close(fd);
#elif KMP_OS_SOLARIS
  prxmap_t *cur, *map;
  void *buf;
  uintptr_t uaddr;
  ssize_t rd;
  int fd;
  pid_t pid = getpid();
  char *name = __kmp_str_format("/proc/%d/xmap", pid);
  fd = open(name, O_RDONLY);
  if (fd == -1) {
    KMP_INTERNAL_FREE(name);
    return 0;
  }
 
  size_t sz = (1 << 20);
  buf = KMP_INTERNAL_MALLOC(sz);
 
  while (sz > 0 && (rd = pread(fd, buf, sz, 0)) == sz) {
    void *newbuf;
    sz <<= 1;
    newbuf = KMP_INTERNAL_REALLOC(buf, sz);
    buf = newbuf;
  }
 
  map = reinterpret_cast<prxmap_t *>(buf);
  uaddr = reinterpret_cast<uintptr_t>(addr);
 
  for (cur = map; rd > 0; cur++, rd = -sizeof(*map)) {
    if (uaddr >= cur->pr_vaddr && uaddr < cur->pr_vaddr) {
      if ((cur->pr_mflags & MA_READ) != 0 && (cur->pr_mflags & MA_WRITE) != 0) {
        found = 1;
        break;
      }
    }
  }
 
  KMP_INTERNAL_FREE(map);
  close(fd);
  KMP_INTERNAL_FREE(name);
#elif KMP_OS_DARWIN
 
  /* On OS X*, /proc pseudo filesystem is not available. Try to read memory
     using vm interface. */
 
  int buffer;
  vm_size_t count;
  rc = vm_read_overwrite(
      mach_task_self(), // Task to read memory of.
      (vm_address_t)(addr), // Address to read from.
      1, // Number of bytes to be read.
      (vm_address_t)(&buffer), // Address of buffer to save read bytes in.
      &count // Address of var to save number of read bytes in.
  );
  if (rc == 0) {
    // Memory successfully read.
    found = 1;
  }
 
#elif KMP_OS_NETBSD
 
  int mib[5];
  mib[0] = CTL_VM;
  mib[1] = VM_PROC;
  mib[2] = VM_PROC_MAP;
  mib[3] = getpid();
  mib[4] = sizeof(struct kinfo_vmentry);
 
  size_t size;
  rc = sysctl(mib, __arraycount(mib), NULL, &size, NULL, 0);
  KMP_ASSERT(!rc);
  KMP_ASSERT(size);
 
  size = size * 4 / 3;
  struct kinfo_vmentry *kiv = (struct kinfo_vmentry *)KMP_INTERNAL_MALLOC(size);
  KMP_ASSERT(kiv);
 
  rc = sysctl(mib, __arraycount(mib), kiv, &size, NULL, 0);
  KMP_ASSERT(!rc);
  KMP_ASSERT(size);
 
  for (size_t i = 0; i < size; i++) {
    if (kiv[i].kve_start >= (uint64_t)addr &&
        kiv[i].kve_end <= (uint64_t)addr) {
      found = 1;
      break;
    }
  }
  KMP_INTERNAL_FREE(kiv);
#elif KMP_OS_OPENBSD
 
  int mib[3];
  mib[0] = CTL_KERN;
  mib[1] = KERN_PROC_VMMAP;
  mib[2] = getpid();
 
  size_t size;
  uint64_t end;
  rc = sysctl(mib, 3, NULL, &size, NULL, 0);
  KMP_ASSERT(!rc);
  KMP_ASSERT(size);
  end = size;
 
  struct kinfo_vmentry kiv = {.kve_start = 0};
 
  while ((rc = sysctl(mib, 3, &kiv, &size, NULL, 0)) == 0) {
    KMP_ASSERT(size);
    if (kiv.kve_end == end)
      break;
 
    if (kiv.kve_start >= (uint64_t)addr && kiv.kve_end <= (uint64_t)addr) {
      found = 1;
      break;
    }
    kiv.kve_start += 1;
  }
#elif KMP_OS_WASI
  found = (int)addr < (__builtin_wasm_memory_size(0) * PAGESIZE);
#elif KMP_OS_AIX
 
  uint32_t loadQueryBufSize = 4096u; // Default loadquery buffer size.
  char *loadQueryBuf;
 
  for (;;) {
    loadQueryBuf = (char *)KMP_INTERNAL_MALLOC(loadQueryBufSize);
    if (loadQueryBuf == NULL) {
      return 0;
    }
 
    rc = loadquery(L_GETXINFO | L_IGNOREUNLOAD, loadQueryBuf, loadQueryBufSize);
    if (rc < 0) {
      KMP_INTERNAL_FREE(loadQueryBuf);
      if (errno != ENOMEM) {
        return 0;
      }
      // errno == ENOMEM; double the size.
      loadQueryBufSize <<= 1;
      continue;
    }
    // Obtained the load info successfully.
    break;
  }
 
  struct ld_xinfo *curLdInfo = (struct ld_xinfo *)loadQueryBuf;
 
  // Loop through the load info to find if there is a match.
  for (;;) {
    uintptr_t curDataStart = (uintptr_t)curLdInfo->ldinfo_dataorg;
    uintptr_t curDataEnd = curDataStart + curLdInfo->ldinfo_datasize;
 
    // The data segment is readable and writable.
    if (curDataStart <= (uintptr_t)addr && (uintptr_t)addr < curDataEnd) {
      found = 1;
      break;
    }
    if (curLdInfo->ldinfo_next == 0u) {
      // Reached the end of load info.
      break;
    }
    curLdInfo = (struct ld_xinfo *)((char *)curLdInfo + curLdInfo->ldinfo_next);
  }
  KMP_INTERNAL_FREE(loadQueryBuf);
 
#elif KMP_OS_HAIKU
 
  found = 1;
#else
 
#error "Unknown or unsupported OS"
 
#endif
 
  return found;
 
} // __kmp_is_address_mapped
int __kmp_is_address_mapped(void *addr) {…}
 
#ifdef USE_LOAD_BALANCE
 
#if KMP_OS_DARWIN || KMP_OS_DRAGONFLY || KMP_OS_FREEBSD || KMP_OS_NETBSD ||    \
    KMP_OS_OPENBSD || KMP_OS_SOLARIS
 
// The function returns the rounded value of the system load average
// during given time interval which depends on the value of
// __kmp_load_balance_interval variable (default is 60 sec, other values
// may be 300 sec or 900 sec).
// It returns -1 in case of error.
int __kmp_get_load_balance(int max) {
  double averages[3];
  int ret_avg = 0;
 
  int res = getloadavg(averages, 3);
 
  // Check __kmp_load_balance_interval to determine which of averages to use.
  // getloadavg() may return the number of samples less than requested that is
  // less than 3.
  if (__kmp_load_balance_interval < 180 && (res >= 1)) {
    ret_avg = (int)averages[0]; // 1 min
  } else if ((__kmp_load_balance_interval >= 180 &&
              __kmp_load_balance_interval < 600) &&
             (res >= 2)) {
    ret_avg = (int)averages[1]; // 5 min
  } else if ((__kmp_load_balance_interval >= 600) && (res == 3)) {
    ret_avg = (int)averages[2]; // 15 min
  } else { // Error occurred
    return -1;
  }
 
  return ret_avg;
}
 
#elif KMP_OS_AIX
 
// The function returns number of running (not sleeping) threads, or -1 in case
// of error.
int __kmp_get_load_balance(int max) {
 
  static int glb_running_threads = 0; // Saved count of the running threads for
                                      // the thread balance algorithm.
  static double glb_call_time = 0; // Thread balance algorithm call time.
  int running_threads = 0; // Number of running threads in the system.
 
  double call_time = 0.0;
 
  __kmp_elapsed(&call_time);
 
  if (glb_call_time &&
      (call_time - glb_call_time < __kmp_load_balance_interval))
    return glb_running_threads;
 
  glb_call_time = call_time;
 
  if (max <= 0) {
    max = INT_MAX;
  }
 
  // Check how many perfstat_cpu_t structures are available.
  int logical_cpus = perfstat_cpu(NULL, NULL, sizeof(perfstat_cpu_t), 0);
  if (logical_cpus <= 0) {
    glb_call_time = -1;
    return -1;
  }
 
  perfstat_cpu_t *cpu_stat = (perfstat_cpu_t *)KMP_INTERNAL_MALLOC(
      logical_cpus * sizeof(perfstat_cpu_t));
  if (cpu_stat == NULL) {
    glb_call_time = -1;
    return -1;
  }
 
  // Set first CPU as the name of the first logical CPU for which the info is
  // desired.
  perfstat_id_t first_cpu_name;
  strcpy(first_cpu_name.name, FIRST_CPU);
 
  // Get the stat info of logical CPUs.
  int rc = perfstat_cpu(&first_cpu_name, cpu_stat, sizeof(perfstat_cpu_t),
                        logical_cpus);
  KMP_DEBUG_ASSERT(rc == logical_cpus);
  if (rc <= 0) {
    KMP_INTERNAL_FREE(cpu_stat);
    glb_call_time = -1;
    return -1;
  }
  for (int i = 0; i < logical_cpus; ++i) {
    running_threads += cpu_stat[i].runque;
    if (running_threads >= max)
      break;
  }
 
  // There _might_ be a timing hole where the thread executing this
  // code gets skipped in the load balance, and running_threads is 0.
  // Assert in the debug builds only!!!
  KMP_DEBUG_ASSERT(running_threads > 0);
  if (running_threads <= 0)
    running_threads = 1;
 
  KMP_INTERNAL_FREE(cpu_stat);
 
  glb_running_threads = running_threads;
 
  return running_threads;
}
 
#else // Linux* OS
 
// The function returns number of running (not sleeping) threads, or -1 in case
// of error. Error could be reported if Linux* OS kernel too old (without
// "/proc" support). Counting running threads stops if max running threads
// encountered.
int __kmp_get_load_balance(int max) {
  static int permanent_error = 0;
  static int glb_running_threads = 0; // Saved count of the running threads for
  // the thread balance algorithm
  static double glb_call_time = 0; /* Thread balance algorithm call time */
 
  int running_threads = 0; // Number of running threads in the system.
 
  DIR *proc_dir = NULL; // Handle of "/proc/" directory.
  struct dirent *proc_entry = NULL;
 
  kmp_str_buf_t task_path; // "/proc/<pid>/task/<tid>/" path.
  DIR *task_dir = NULL; // Handle of "/proc/<pid>/task/<tid>/" directory.
  struct dirent *task_entry = NULL;
  int task_path_fixed_len;
 
  kmp_str_buf_t stat_path; // "/proc/<pid>/task/<tid>/stat" path.
  int stat_file = -1;
  int stat_path_fixed_len;
 
#ifdef KMP_DEBUG
  int total_processes = 0; // Total number of processes in system.
#endif
 
  double call_time = 0.0;
 
  __kmp_str_buf_init(&task_path);
  __kmp_str_buf_init(&stat_path);
 
  __kmp_elapsed(&call_time);
 
  if (glb_call_time &&
      (call_time - glb_call_time < __kmp_load_balance_interval)) {
    running_threads = glb_running_threads;
    goto finish;
  }
 
  glb_call_time = call_time;
 
  // Do not spend time on scanning "/proc/" if we have a permanent error.
  if (permanent_error) {
    running_threads = -1;
    goto finish;
  }
 
  if (max <= 0) {
    max = INT_MAX;
  }
 
  // Open "/proc/" directory.
  proc_dir = opendir("/proc");
  if (proc_dir == NULL) {
    // Cannot open "/proc/". Probably the kernel does not support it. Return an
    // error now and in subsequent calls.
    running_threads = -1;
    permanent_error = 1;
    goto finish;
  }
 
  // Initialize fixed part of task_path. This part will not change.
  __kmp_str_buf_cat(&task_path, "/proc/", 6);
  task_path_fixed_len = task_path.used; // Remember number of used characters.
 
  proc_entry = readdir(proc_dir);
  while (proc_entry != NULL) {
    // Proc entry is a directory and name starts with a digit. Assume it is a
    // process' directory.
    if (proc_entry->d_type == DT_DIR && isdigit(proc_entry->d_name[0])) {
 
#ifdef KMP_DEBUG
      ++total_processes;
#endif
      // Make sure init process is the very first in "/proc", so we can replace
      // strcmp( proc_entry->d_name, "1" ) == 0 with simpler total_processes ==
      // 1. We are going to check that total_processes == 1 => d_name == "1" is
      // true (where "=>" is implication). Since C++ does not have => operator,
      // let us replace it with its equivalent: a => b == ! a || b.
      KMP_DEBUG_ASSERT(total_processes != 1 ||
                       strcmp(proc_entry->d_name, "1") == 0);
 
      // Construct task_path.
      task_path.used = task_path_fixed_len; // Reset task_path to "/proc/".
      __kmp_str_buf_cat(&task_path, proc_entry->d_name,
                        KMP_STRLEN(proc_entry->d_name));
      __kmp_str_buf_cat(&task_path, "/task", 5);
 
      task_dir = opendir(task_path.str);
      if (task_dir == NULL) {
        // Process can finish between reading "/proc/" directory entry and
        // opening process' "task/" directory. So, in general case we should not
        // complain, but have to skip this process and read the next one. But on
        // systems with no "task/" support we will spend lot of time to scan
        // "/proc/" tree again and again without any benefit. "init" process
        // (its pid is 1) should exist always, so, if we cannot open
        // "/proc/1/task/" directory, it means "task/" is not supported by
        // kernel. Report an error now and in the future.
        if (strcmp(proc_entry->d_name, "1") == 0) {
          running_threads = -1;
          permanent_error = 1;
          goto finish;
        }
      } else {
        // Construct fixed part of stat file path.
        __kmp_str_buf_clear(&stat_path);
        __kmp_str_buf_cat(&stat_path, task_path.str, task_path.used);
        __kmp_str_buf_cat(&stat_path, "/", 1);
        stat_path_fixed_len = stat_path.used;
 
        task_entry = readdir(task_dir);
        while (task_entry != NULL) {
          // It is a directory and name starts with a digit.
          if (proc_entry->d_type == DT_DIR && isdigit(task_entry->d_name[0])) {
 
            // Construct complete stat file path. Easiest way would be:
            //  __kmp_str_buf_print( & stat_path, "%s/%s/stat", task_path.str,
            //  task_entry->d_name );
            // but seriae of __kmp_str_buf_cat works a bit faster.
            stat_path.used =
                stat_path_fixed_len; // Reset stat path to its fixed part.
            __kmp_str_buf_cat(&stat_path, task_entry->d_name,
                              KMP_STRLEN(task_entry->d_name));
            __kmp_str_buf_cat(&stat_path, "/stat", 5);
 
            // Note: Low-level API (open/read/close) is used. High-level API
            // (fopen/fclose)  works ~ 30 % slower.
            stat_file = open(stat_path.str, O_RDONLY);
            if (stat_file == -1) {
              // We cannot report an error because task (thread) can terminate
              // just before reading this file.
            } else {
              /* Content of "stat" file looks like:
                 24285 (program) S ...
 
                 It is a single line (if program name does not include funny
                 symbols). First number is a thread id, then name of executable
                 file name in paretheses, then state of the thread. We need just
                 thread state.
 
                 Good news: Length of program name is 15 characters max. Longer
                 names are truncated.
 
                 Thus, we need rather short buffer: 15 chars for program name +
                 2 parenthesis, + 3 spaces + ~7 digits of pid = 37.
 
                 Bad news: Program name may contain special symbols like space,
                 closing parenthesis, or even new line. This makes parsing
                 "stat" file not 100 % reliable. In case of fanny program names
                 parsing may fail (report incorrect thread state).
 
                 Parsing "status" file looks more promissing (due to different
                 file structure and escaping special symbols) but reading and
                 parsing of "status" file works slower.
                  -- ln
              */
              char buffer[65];
              ssize_t len;
              len = read(stat_file, buffer, sizeof(buffer) - 1);
              if (len >= 0) {
                buffer[len] = 0;
                // Using scanf:
                //     sscanf( buffer, "%*d (%*s) %c ", & state );
                // looks very nice, but searching for a closing parenthesis
                // works a bit faster.
                char *close_parent = strstr(buffer, ") ");
                if (close_parent != NULL) {
                  char state = *(close_parent + 2);
                  if (state == 'R') {
                    ++running_threads;
                    if (running_threads >= max) {
                      goto finish;
                    }
                  }
                }
              }
              close(stat_file);
              stat_file = -1;
            }
          }
          task_entry = readdir(task_dir);
        }
        closedir(task_dir);
        task_dir = NULL;
      }
    }
    proc_entry = readdir(proc_dir);
  }
 
  // There _might_ be a timing hole where the thread executing this
  // code get skipped in the load balance, and running_threads is 0.
  // Assert in the debug builds only!!!
  KMP_DEBUG_ASSERT(running_threads > 0);
  if (running_threads <= 0) {
    running_threads = 1;
  }
 
finish: // Clean up and exit.
  if (proc_dir != NULL) {
    closedir(proc_dir);
  }
  __kmp_str_buf_free(&task_path);
  if (task_dir != NULL) {
    closedir(task_dir);
  }
  __kmp_str_buf_free(&stat_path);
  if (stat_file != -1) {
    close(stat_file);
  }
 
  glb_running_threads = running_threads;
 
  return running_threads;
 
} // __kmp_get_load_balance
 
#endif // KMP_OS_DARWIN
 
#endif // USE_LOAD_BALANCE
 
#if !(KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_MIC ||                            \
      ((KMP_OS_LINUX || KMP_OS_DARWIN) && KMP_ARCH_AARCH64) ||                 \
      KMP_ARCH_PPC64 || KMP_ARCH_RISCV64 || KMP_ARCH_LOONGARCH64 ||            \
      KMP_ARCH_ARM || KMP_ARCH_VE || KMP_ARCH_S390X || KMP_ARCH_PPC_XCOFF ||   \
      KMP_ARCH_AARCH64_32)
 
// Because WebAssembly will use `call_indirect` to invoke the microtask and
// WebAssembly indirect calls check that the called signature is a precise
// match, we need to cast each microtask function pointer back from `void *` to
// its original type.
typedef void (*microtask_t0)(int *, int *);
typedef void (*microtask_t1)(int *, int *, void *);
typedef void (*microtask_t2)(int *, int *, void *, void *);
typedef void (*microtask_t3)(int *, int *, void *, void *, void *);
typedef void (*microtask_t4)(int *, int *, void *, void *, void *, void *);
typedef void (*microtask_t5)(int *, int *, void *, void *, void *, void *,
                             void *);
typedef void (*microtask_t6)(int *, int *, void *, void *, void *, void *,
                             void *, void *);
typedef void (*microtask_t7)(int *, int *, void *, void *, void *, void *,
                             void *, void *, void *);
typedef void (*microtask_t8)(int *, int *, void *, void *, void *, void *,
                             void *, void *, void *, void *);
typedef void (*microtask_t9)(int *, int *, void *, void *, void *, void *,
                             void *, void *, void *, void *, void *);
typedef void (*microtask_t10)(int *, int *, void *, void *, void *, void *,
                              void *, void *, void *, void *, void *, void *);
typedef void (*microtask_t11)(int *, int *, void *, void *, void *, void *,
                              void *, void *, void *, void *, void *, void *,
                              void *);
typedef void (*microtask_t12)(int *, int *, void *, void *, void *, void *,
                              void *, void *, void *, void *, void *, void *,
                              void *, void *);
typedef void (*microtask_t13)(int *, int *, void *, void *, void *, void *,
                              void *, void *, void *, void *, void *, void *,
                              void *, void *, void *);
typedef void (*microtask_t14)(int *, int *, void *, void *, void *, void *,
                              void *, void *, void *, void *, void *, void *,
                              void *, void *, void *, void *);
typedef void (*microtask_t15)(int *, int *, void *, void *, void *, void *,
                              void *, void *, void *, void *, void *, void *,
                              void *, void *, void *, void *, void *);
 
// we really only need the case with 1 argument, because CLANG always build
// a struct of pointers to shared variables referenced in the outlined function
int __kmp_invoke_microtask(microtask_t pkfn, int gtid, int tid, int argc,
                           void *p_argv[]
#if OMPT_SUPPORT
                           ,
                           void **exit_frame_ptr
#endif
) {
#if OMPT_SUPPORT
  *exit_frame_ptr = OMPT_GET_FRAME_ADDRESS(0);
#endif
 
  switch (argc) {
  default:
    fprintf(stderr, "Too many args to microtask: %d!\n", argc);
    fflush(stderr);
    exit(-1);
  case 0:
    (*(microtask_t0)pkfn)(&gtid, &tid);
    break;
  case 1:
    (*(microtask_t1)pkfn)(&gtid, &tid, p_argv[0]);
    break;
  case 2:
    (*(microtask_t2)pkfn)(&gtid, &tid, p_argv[0], p_argv[1]);
    break;
  case 3:
    (*(microtask_t3)pkfn)(&gtid, &tid, p_argv[0], p_argv[1], p_argv[2]);
    break;
  case 4:
    (*(microtask_t4)pkfn)(&gtid, &tid, p_argv[0], p_argv[1], p_argv[2],
                          p_argv[3]);
    break;
  case 5:
    (*(microtask_t5)pkfn)(&gtid, &tid, p_argv[0], p_argv[1], p_argv[2],
                          p_argv[3], p_argv[4]);
    break;
  case 6:
    (*(microtask_t6)pkfn)(&gtid, &tid, p_argv[0], p_argv[1], p_argv[2],
                          p_argv[3], p_argv[4], p_argv[5]);
    break;
  case 7:
    (*(microtask_t7)pkfn)(&gtid, &tid, p_argv[0], p_argv[1], p_argv[2],
                          p_argv[3], p_argv[4], p_argv[5], p_argv[6]);
    break;
  case 8:
    (*(microtask_t8)pkfn)(&gtid, &tid, p_argv[0], p_argv[1], p_argv[2],
                          p_argv[3], p_argv[4], p_argv[5], p_argv[6],
                          p_argv[7]);
    break;
  case 9:
    (*(microtask_t9)pkfn)(&gtid, &tid, p_argv[0], p_argv[1], p_argv[2],
                          p_argv[3], p_argv[4], p_argv[5], p_argv[6], p_argv[7],
                          p_argv[8]);
    break;
  case 10:
    (*(microtask_t10)pkfn)(&gtid, &tid, p_argv[0], p_argv[1], p_argv[2],
                           p_argv[3], p_argv[4], p_argv[5], p_argv[6],
                           p_argv[7], p_argv[8], p_argv[9]);
    break;
  case 11:
    (*(microtask_t11)pkfn)(&gtid, &tid, p_argv[0], p_argv[1], p_argv[2],
                           p_argv[3], p_argv[4], p_argv[5], p_argv[6],
                           p_argv[7], p_argv[8], p_argv[9], p_argv[10]);
    break;
  case 12:
    (*(microtask_t12)pkfn)(&gtid, &tid, p_argv[0], p_argv[1], p_argv[2],
                           p_argv[3], p_argv[4], p_argv[5], p_argv[6],
                           p_argv[7], p_argv[8], p_argv[9], p_argv[10],
                           p_argv[11]);
    break;
  case 13:
    (*(microtask_t13)pkfn)(&gtid, &tid, p_argv[0], p_argv[1], p_argv[2],
                           p_argv[3], p_argv[4], p_argv[5], p_argv[6],
                           p_argv[7], p_argv[8], p_argv[9], p_argv[10],
                           p_argv[11], p_argv[12]);
    break;
  case 14:
    (*(microtask_t14)pkfn)(&gtid, &tid, p_argv[0], p_argv[1], p_argv[2],
                           p_argv[3], p_argv[4], p_argv[5], p_argv[6],
                           p_argv[7], p_argv[8], p_argv[9], p_argv[10],
                           p_argv[11], p_argv[12], p_argv[13]);
    break;
  case 15:
    (*(microtask_t15)pkfn)(&gtid, &tid, p_argv[0], p_argv[1], p_argv[2],
                           p_argv[3], p_argv[4], p_argv[5], p_argv[6],
                           p_argv[7], p_argv[8], p_argv[9], p_argv[10],
                           p_argv[11], p_argv[12], p_argv[13], p_argv[14]);
    break;
  }
 
  return 1;
}
int __kmp_invoke_microtask(microtask_t pkfn, int gtid, int tid, int argc, {…}
 
#endif
 
#if KMP_OS_LINUX
// Functions for hidden helper task
namespace {
// Condition variable for initializing hidden helper team
pthread_cond_t hidden_helper_threads_initz_cond_var;
pthread_mutex_t hidden_helper_threads_initz_lock;
volatile int hidden_helper_initz_signaled = FALSE;
 
// Condition variable for deinitializing hidden helper team
pthread_cond_t hidden_helper_threads_deinitz_cond_var;
pthread_mutex_t hidden_helper_threads_deinitz_lock;
volatile int hidden_helper_deinitz_signaled = FALSE;
 
// Condition variable for the wrapper function of main thread
pthread_cond_t hidden_helper_main_thread_cond_var;
pthread_mutex_t hidden_helper_main_thread_lock;
volatile int hidden_helper_main_thread_signaled = FALSE;
 
// Semaphore for worker threads. We don't use condition variable here in case
// that when multiple signals are sent at the same time, only one thread might
// be waken.
sem_t hidden_helper_task_sem;
} // namespace
 
void __kmp_hidden_helper_worker_thread_wait() {
  int status = sem_wait(&hidden_helper_task_sem);
  KMP_CHECK_SYSFAIL("sem_wait", status);
}
 
void __kmp_do_initialize_hidden_helper_threads() {
  // Initialize condition variable
  int status =
      pthread_cond_init(&hidden_helper_threads_initz_cond_var, nullptr);
  KMP_CHECK_SYSFAIL("pthread_cond_init", status);
 
  status = pthread_cond_init(&hidden_helper_threads_deinitz_cond_var, nullptr);
  KMP_CHECK_SYSFAIL("pthread_cond_init", status);
 
  status = pthread_cond_init(&hidden_helper_main_thread_cond_var, nullptr);
  KMP_CHECK_SYSFAIL("pthread_cond_init", status);
 
  status = pthread_mutex_init(&hidden_helper_threads_initz_lock, nullptr);
  KMP_CHECK_SYSFAIL("pthread_mutex_init", status);
 
  status = pthread_mutex_init(&hidden_helper_threads_deinitz_lock, nullptr);
  KMP_CHECK_SYSFAIL("pthread_mutex_init", status);
 
  status = pthread_mutex_init(&hidden_helper_main_thread_lock, nullptr);
  KMP_CHECK_SYSFAIL("pthread_mutex_init", status);
 
  // Initialize the semaphore
  status = sem_init(&hidden_helper_task_sem, 0, 0);
  KMP_CHECK_SYSFAIL("sem_init", status);
 
  // Create a new thread to finish initialization
  pthread_t handle;
  status = pthread_create(
      &handle, nullptr,
      [](void *) -> void * {
        __kmp_hidden_helper_threads_initz_routine();
        return nullptr;
      },
      nullptr);
  KMP_CHECK_SYSFAIL("pthread_create", status);
}
 
void __kmp_hidden_helper_threads_initz_wait() {
  // Initial thread waits here for the completion of the initialization. The
  // condition variable will be notified by main thread of hidden helper teams.
  int status = pthread_mutex_lock(&hidden_helper_threads_initz_lock);
  KMP_CHECK_SYSFAIL("pthread_mutex_lock", status);
 
  if (!TCR_4(hidden_helper_initz_signaled)) {
    status = pthread_cond_wait(&hidden_helper_threads_initz_cond_var,
                               &hidden_helper_threads_initz_lock);
    KMP_CHECK_SYSFAIL("pthread_cond_wait", status);
  }
 
  status = pthread_mutex_unlock(&hidden_helper_threads_initz_lock);
  KMP_CHECK_SYSFAIL("pthread_mutex_unlock", status);
}
 
void __kmp_hidden_helper_initz_release() {
  // After all initialization, reset __kmp_init_hidden_helper_threads to false.
  int status = pthread_mutex_lock(&hidden_helper_threads_initz_lock);
  KMP_CHECK_SYSFAIL("pthread_mutex_lock", status);
 
  status = pthread_cond_signal(&hidden_helper_threads_initz_cond_var);
  KMP_CHECK_SYSFAIL("pthread_cond_wait", status);
 
  TCW_SYNC_4(hidden_helper_initz_signaled, TRUE);
 
  status = pthread_mutex_unlock(&hidden_helper_threads_initz_lock);
  KMP_CHECK_SYSFAIL("pthread_mutex_unlock", status);
}
 
void __kmp_hidden_helper_main_thread_wait() {
  // The main thread of hidden helper team will be blocked here. The
  // condition variable can only be signal in the destructor of RTL.
  int status = pthread_mutex_lock(&hidden_helper_main_thread_lock);
  KMP_CHECK_SYSFAIL("pthread_mutex_lock", status);
 
  if (!TCR_4(hidden_helper_main_thread_signaled)) {
    status = pthread_cond_wait(&hidden_helper_main_thread_cond_var,
                               &hidden_helper_main_thread_lock);
    KMP_CHECK_SYSFAIL("pthread_cond_wait", status);
  }
 
  status = pthread_mutex_unlock(&hidden_helper_main_thread_lock);
  KMP_CHECK_SYSFAIL("pthread_mutex_unlock", status);
}
 
void __kmp_hidden_helper_main_thread_release() {
  // The initial thread of OpenMP RTL should call this function to wake up the
  // main thread of hidden helper team.
  int status = pthread_mutex_lock(&hidden_helper_main_thread_lock);
  KMP_CHECK_SYSFAIL("pthread_mutex_lock", status);
 
  status = pthread_cond_signal(&hidden_helper_main_thread_cond_var);
  KMP_CHECK_SYSFAIL("pthread_cond_signal", status);
 
  // The hidden helper team is done here
  TCW_SYNC_4(hidden_helper_main_thread_signaled, TRUE);
 
  status = pthread_mutex_unlock(&hidden_helper_main_thread_lock);
  KMP_CHECK_SYSFAIL("pthread_mutex_unlock", status);
}
 
void __kmp_hidden_helper_worker_thread_signal() {
  int status = sem_post(&hidden_helper_task_sem);
  KMP_CHECK_SYSFAIL("sem_post", status);
}
 
void __kmp_hidden_helper_threads_deinitz_wait() {
  // Initial thread waits here for the completion of the deinitialization. The
  // condition variable will be notified by main thread of hidden helper teams.
  int status = pthread_mutex_lock(&hidden_helper_threads_deinitz_lock);
  KMP_CHECK_SYSFAIL("pthread_mutex_lock", status);
 
  if (!TCR_4(hidden_helper_deinitz_signaled)) {
    status = pthread_cond_wait(&hidden_helper_threads_deinitz_cond_var,
                               &hidden_helper_threads_deinitz_lock);
    KMP_CHECK_SYSFAIL("pthread_cond_wait", status);
  }
 
  status = pthread_mutex_unlock(&hidden_helper_threads_deinitz_lock);
  KMP_CHECK_SYSFAIL("pthread_mutex_unlock", status);
}
 
void __kmp_hidden_helper_threads_deinitz_release() {
  int status = pthread_mutex_lock(&hidden_helper_threads_deinitz_lock);
  KMP_CHECK_SYSFAIL("pthread_mutex_lock", status);
 
  status = pthread_cond_signal(&hidden_helper_threads_deinitz_cond_var);
  KMP_CHECK_SYSFAIL("pthread_cond_wait", status);
 
  TCW_SYNC_4(hidden_helper_deinitz_signaled, TRUE);
 
  status = pthread_mutex_unlock(&hidden_helper_threads_deinitz_lock);
  KMP_CHECK_SYSFAIL("pthread_mutex_unlock", status);
}
#else // KMP_OS_LINUX
void __kmp_hidden_helper_worker_thread_wait() {
  KMP_ASSERT(0 && "Hidden helper task is not supported on this OS");
}
void __kmp_hidden_helper_worker_thread_wait() {…}
 
void __kmp_do_initialize_hidden_helper_threads() {
  KMP_ASSERT(0 && "Hidden helper task is not supported on this OS");
}
void __kmp_do_initialize_hidden_helper_threads() {…}
 
void __kmp_hidden_helper_threads_initz_wait() {
  KMP_ASSERT(0 && "Hidden helper task is not supported on this OS");
}
void __kmp_hidden_helper_threads_initz_wait() {…}
 
void __kmp_hidden_helper_initz_release() {
  KMP_ASSERT(0 && "Hidden helper task is not supported on this OS");
}
void __kmp_hidden_helper_initz_release() {…}
 
void __kmp_hidden_helper_main_thread_wait() {
  KMP_ASSERT(0 && "Hidden helper task is not supported on this OS");
}
void __kmp_hidden_helper_main_thread_wait() {…}
 
void __kmp_hidden_helper_main_thread_release() {
  KMP_ASSERT(0 && "Hidden helper task is not supported on this OS");
}
void __kmp_hidden_helper_main_thread_release() {…}
 
void __kmp_hidden_helper_worker_thread_signal() {
  KMP_ASSERT(0 && "Hidden helper task is not supported on this OS");
}
void __kmp_hidden_helper_worker_thread_signal() {…}
 
void __kmp_hidden_helper_threads_deinitz_wait() {
  KMP_ASSERT(0 && "Hidden helper task is not supported on this OS");
}
void __kmp_hidden_helper_threads_deinitz_wait() {…}
 
void __kmp_hidden_helper_threads_deinitz_release() {
  KMP_ASSERT(0 && "Hidden helper task is not supported on this OS");
}
void __kmp_hidden_helper_threads_deinitz_release() {…}
#endif // KMP_OS_LINUX
 
bool __kmp_detect_shm() {
  DIR *dir = opendir("/dev/shm");
  if (dir) { // /dev/shm exists
    closedir(dir);
    return true;
  } else if (ENOENT == errno) { // /dev/shm does not exist
    return false;
  } else { // opendir() failed
    return false;
  }
}
bool __kmp_detect_shm() {…}
 
bool __kmp_detect_tmp() {
  DIR *dir = opendir("/tmp");
  if (dir) { // /tmp exists
    closedir(dir);
    return true;
  } else if (ENOENT == errno) { // /tmp does not exist
    return false;
  } else { // opendir() failed
    return false;
  }
}
bool __kmp_detect_tmp() {…}
 
// end of file //