Fix some declarations/definitions in __threading_support which have

inconsistent qualifiers.

libcxx/include/__threading_support

@@ -287,8 +287,10 @@ int __libcpp_tls_set(__libcpp_tls_key __key, void *__p);
 
 #if defined(_LIBCUDACXX_HAS_THREAD_API_CUDA)
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 void __libcpp_thread_yield() {}
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 void __libcpp_thread_sleep_for(chrono::nanoseconds __ns)
 {
 #if __CUDA_ARCH__ >= 700
@@ -302,6 +304,7 @@ void __libcpp_thread_sleep_for(chrono::nanoseconds __ns)
 
 #elif defined(_LIBCUDACXX_HAS_THREAD_API_PTHREAD)
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 __libcpp_timespec_t __libcpp_to_timespec(const chrono::nanoseconds& __ns)
 {
      using namespace chrono;
@@ -323,6 +326,7 @@ __libcpp_timespec_t __libcpp_to_timespec(const chrono::nanoseconds& __ns)
      return __ts;
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 int __libcpp_recursive_mutex_init(__libcpp_recursive_mutex_t *__m)
 {
     pthread_mutexattr_t attr;
@@ -347,68 +351,81 @@ int __libcpp_recursive_mutex_init(__libcpp_recursive_mutex_t *__m)
     return 0;
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 int __libcpp_recursive_mutex_lock(__libcpp_recursive_mutex_t *__m)
 {
     return pthread_mutex_lock(__m);
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 bool __libcpp_recursive_mutex_trylock(__libcpp_recursive_mutex_t *__m)
 {
     return pthread_mutex_trylock(__m) == 0;
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 int __libcpp_recursive_mutex_unlock(__libcpp_mutex_t *__m)
 {
     return pthread_mutex_unlock(__m);
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 int __libcpp_recursive_mutex_destroy(__libcpp_recursive_mutex_t *__m)
 {
     return pthread_mutex_destroy(__m);
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 int __libcpp_mutex_lock(__libcpp_mutex_t *__m)
 {
     return pthread_mutex_lock(__m);
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 bool __libcpp_mutex_trylock(__libcpp_mutex_t *__m)
 {
     return pthread_mutex_trylock(__m) == 0;
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 int __libcpp_mutex_unlock(__libcpp_mutex_t *__m)
 {
     return pthread_mutex_unlock(__m);
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 int __libcpp_mutex_destroy(__libcpp_mutex_t *__m)
 {
   return pthread_mutex_destroy(__m);
 }
 
 // Condition Variable
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 int __libcpp_condvar_signal(__libcpp_condvar_t *__cv)
 {
     return pthread_cond_signal(__cv);
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 int __libcpp_condvar_broadcast(__libcpp_condvar_t *__cv)
 {
     return pthread_cond_broadcast(__cv);
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 int __libcpp_condvar_wait(__libcpp_condvar_t *__cv, __libcpp_mutex_t *__m)
 {
     return pthread_cond_wait(__cv, __m);
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 int __libcpp_condvar_timedwait(__libcpp_condvar_t *__cv, __libcpp_mutex_t *__m,
                                __libcpp_timespec_t *__ts)
 {
     return pthread_cond_timedwait(__cv, __m, __ts);
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 int __libcpp_condvar_destroy(__libcpp_condvar_t *__cv)
 {
     return pthread_cond_destroy(__cv);
@@ -446,26 +463,31 @@ bool __libcpp_semaphore_wait_timed(__libcpp_semaphore_t* __sem, chrono::nanoseco
 
 #else
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 bool __libcpp_semaphore_init(__libcpp_semaphore_t* __sem, int __init)
 {
     return sem_init(__sem, 0, __init) == 0;
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 bool __libcpp_semaphore_destroy(__libcpp_semaphore_t* __sem)
 {
     return sem_destroy(__sem) == 0;
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 bool __libcpp_semaphore_post(__libcpp_semaphore_t* __sem)
 {
     return sem_post(__sem) == 0;
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 bool __libcpp_semaphore_wait(__libcpp_semaphore_t* __sem)
 {
     return sem_wait(__sem) == 0;
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 bool __libcpp_semaphore_wait_timed(__libcpp_semaphore_t* __sem, chrono::nanoseconds const& __ns)
 {
     __libcpp_timespec_t __ts = __libcpp_to_timespec(__ns);
@@ -475,77 +497,91 @@ bool __libcpp_semaphore_wait_timed(__libcpp_semaphore_t* __sem, chrono::nanoseco
 #endif //__APPLE__
 
 // Execute once
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 int __libcpp_execute_once(__libcpp_exec_once_flag *flag, void (*init_routine)())
 {
     return pthread_once(flag, init_routine);
 }
 
 // Thread id
 // Returns non-zero if the thread ids are equal, otherwise 0
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 bool __libcpp_thread_id_equal(__libcpp_thread_id t1, __libcpp_thread_id t2)
 {
     return pthread_equal(t1, t2) != 0;
 }
 
 // Returns non-zero if t1 < t2, otherwise 0
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 bool __libcpp_thread_id_less(__libcpp_thread_id t1, __libcpp_thread_id t2)
 {
     return t1 < t2;
 }
 
 // Thread
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 bool __libcpp_thread_isnull(const __libcpp_thread_t *__t)
 {
     return *__t == 0;
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 int __libcpp_thread_create(__libcpp_thread_t *__t, void *(*__func)(void *),
                            void *__arg)
 {
     return pthread_create(__t, 0, __func, __arg);
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 __libcpp_thread_id __libcpp_thread_get_current_id()
 {
     return pthread_self();
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 __libcpp_thread_id __libcpp_thread_get_id(const __libcpp_thread_t *__t)
 {
     return *__t;
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 int __libcpp_thread_join(__libcpp_thread_t *__t)
 {
     return pthread_join(*__t, 0);
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 int __libcpp_thread_detach(__libcpp_thread_t *__t)
 {
     return pthread_detach(*__t);
 }
 
 // Thread local storage
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 int __libcpp_tls_create(__libcpp_tls_key *__key, void (*__at_exit)(void *))
 {
     return pthread_key_create(__key, __at_exit);
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 void *__libcpp_tls_get(__libcpp_tls_key __key)
 {
     return pthread_getspecific(__key);
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 int __libcpp_tls_set(__libcpp_tls_key __key, void *__p)
 {
     return pthread_setspecific(__key, __p);
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 void __libcpp_thread_yield()
 {
     sched_yield();
 }
 
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 void __libcpp_thread_sleep_for(chrono::nanoseconds __ns)
 {
     __libcpp_timespec_t __ts = __libcpp_to_timespec(__ns);
@@ -596,6 +632,7 @@ void __libcpp_thread_sleep_for(chrono::nanoseconds __ns)
 #endif // !defined(_LIBCUDACXX_HAS_THREAD_LIBRARY_EXTERNAL) || defined(_LIBCUDACXX_BUILDING_THREAD_LIBRARY_EXTERNAL)
 
 template<class _Fn>
+_LIBCUDACXX_THREAD_ABI_VISIBILITY
 bool __libcpp_thread_poll_with_backoff(_Fn && __f, chrono::nanoseconds __max)
 {
     chrono::high_resolution_clock::time_point const __start = chrono::high_resolution_clock::now();