Use language TLS specifier if it is possible.

To access TLS, it is faster to use language TLS specifier instead
of using pthread_get/setspecific functions.

Original proposal is: Use native thread locals. #3665

6 files changed, 78 insertions, 2 deletions

diff --git a/ractor.h b/ractor.h
index 4cd89522a7..d3de06b559 100644
--- a/ractor.h
+++ b/ractor.h
@@ -205,7 +205,15 @@ rb_ractor_thread_switch(rb_ractor_t *cr, rb_thread_t *th)
 static inline void
 rb_ractor_set_current_ec(rb_ractor_t *cr, rb_execution_context_t *ec)
 {
+#ifdef RB_THREAD_LOCAL_SPECIFIER
+  #if __APPLE__
+    rb_current_ec_set(ec);
+  #else
+    ruby_current_ec = ec;
+  #endif
+#else
     native_tls_set(ruby_current_ec_key, ec);
+#endif
 
     if (cr->threads.running_ec != ec) {
         if (0) fprintf(stderr, "rb_ractor_set_current_ec ec:%p->%p\n",
diff --git a/thread_pthread.c b/thread_pthread.c
index 427897cfd8..71667aec69 100644
--- a/thread_pthread.c
+++ b/thread_pthread.c
@@ -550,7 +550,11 @@ native_cond_timeout(rb_nativethread_cond_t *cond, const rb_hrtime_t rel)
 #define native_cleanup_push pthread_cleanup_push
 #define native_cleanup_pop  pthread_cleanup_pop
 
+#ifdef RB_THREAD_LOCAL_SPECIFIER
+static RB_THREAD_LOCAL_SPECIFIER rb_thread_t *ruby_native_thread;
+#else
 static pthread_key_t ruby_native_thread_key;
+#endif
 
 static void
 null_func(int i)
@@ -561,7 +565,11 @@ null_func(int i)
 static rb_thread_t *
 ruby_thread_from_native(void)
 {
+#ifdef RB_THREAD_LOCAL_SPECIFIER
+    return ruby_native_thread;
+#else
     return pthread_getspecific(ruby_native_thread_key);
+#endif
 }
 
 static int
@@ -570,7 +578,12 @@ ruby_thread_set_native(rb_thread_t *th)
     if (th && th->ec) {
         rb_ractor_set_current_ec(th->ractor, th->ec);
     }
+#ifdef RB_THREAD_LOCAL_SPECIFIER
+    ruby_native_thread = th;
+    return 1;
+#else
     return pthread_setspecific(ruby_native_thread_key, th) == 0;
+#endif
 }
 
 static void native_thread_init(rb_thread_t *th);
@@ -587,12 +600,15 @@ Init_native_thread(rb_thread_t *th)
         if (r) condattr_monotonic = NULL;
     }
 #endif
+
+#ifndef RB_THREAD_LOCAL_SPECIFIER
     if (pthread_key_create(&ruby_native_thread_key, 0) == EAGAIN) {
         rb_bug("pthread_key_create failed (ruby_native_thread_key)");
     }
     if (pthread_key_create(&ruby_current_ec_key, 0) == EAGAIN) {
         rb_bug("pthread_key_create failed (ruby_current_ec_key)");
     }
+#endif
     th->thread_id = pthread_self();
     ruby_thread_set_native(th);
     fill_thread_id_str(th);
diff --git a/thread_pthread.h b/thread_pthread.h
index d14857b05a..fa375b3e55 100644
--- a/thread_pthread.h
+++ b/thread_pthread.h
@@ -83,6 +83,14 @@ typedef struct rb_global_vm_lock_struct {
     int wait_yield;
 } rb_global_vm_lock_t;
 
+
+#if __STDC_VERSION__ >= 201112
+  #define RB_THREAD_LOCAL_SPECIFIER _Thread_local
+#elif defined(__GNUC__)
+  /* note that ICC (linux) and Clang are covered by __GNUC__ */
+  #define RB_THREAD_LOCAL_SPECIFIER __thread
+#else
+
 typedef pthread_key_t native_tls_key_t;
 
 static inline void *
@@ -102,5 +110,20 @@ native_tls_set(native_tls_key_t key, void *ptr)
         rb_bug("pthread_setspecific error");
     }
 }
+#endif
+
+RUBY_SYMBOL_EXPORT_BEGIN
+#ifdef RB_THREAD_LOCAL_SPECIFIER
+  #if __APPLE__
+    // on Darwin, TLS can not be accessed across .so
+    struct rb_execution_context_struct *rb_current_ec();
+    void rb_current_ec_set(struct rb_execution_context_struct *);
+  #else
+    RUBY_EXTERN RB_THREAD_LOCAL_SPECIFIER struct rb_execution_context_struct *ruby_current_ec;
+  #endif
+#else
+  RUBY_EXTERN native_tls_key_t ruby_current_ec_key;
+#endif
+RUBY_SYMBOL_EXPORT_END
 
 #endif /* RUBY_THREAD_PTHREAD_H */
diff --git a/thread_win32.h b/thread_win32.h
index 0d95731587..cdcc159b2d 100644
--- a/thread_win32.h
+++ b/thread_win32.h
@@ -63,4 +63,8 @@ void rb_native_cond_timedwait(rb_nativethread_cond_t *cond, rb_nativethread_lock
 void rb_native_cond_initialize(rb_nativethread_cond_t *cond);
 void rb_native_cond_destroy(rb_nativethread_cond_t *cond);
 
+RUBY_SYMBOL_EXPORT_BEGIN
+RUBY_EXTERN native_tls_key_t ruby_current_ec_key;
+RUBY_SYMBOL_EXPORT_END
+
 #endif /* RUBY_THREAD_WIN32_H */
diff --git a/vm.c b/vm.c
index 77a0659dd1..879814a14b 100644
--- a/vm.c
+++ b/vm.c
@@ -379,7 +379,26 @@ VALUE rb_block_param_proxy;
 #define ruby_vm_redefined_flag GET_VM()->redefined_flag
 VALUE ruby_vm_const_missing_count = 0;
 rb_vm_t *ruby_current_vm_ptr = NULL;
+
+#ifdef RB_THREAD_LOCAL_SPECIFIER
+RB_THREAD_LOCAL_SPECIFIER rb_execution_context_t *ruby_current_ec;
+
+#ifdef __APPLE__
+  rb_execution_context_t *
+  rb_current_ec(void)
+  {
+      return ruby_current_ec;
+  }
+  void
+  rb_current_ec_set(rb_execution_context_t *ec)
+  {
+      ruby_current_ec = ec;
+  }
+#endif
+
+#else
 native_tls_key_t ruby_current_ec_key;
+#endif
 
 rb_event_flag_t ruby_vm_event_flags;
 rb_event_flag_t ruby_vm_event_enabled_global_flags;
diff --git a/vm_core.h b/vm_core.h
index 73b6be52f6..f644e8a6bc 100644
--- a/vm_core.h
+++ b/vm_core.h
@@ -1721,8 +1721,6 @@ RUBY_EXTERN rb_event_flag_t ruby_vm_event_flags;
 RUBY_EXTERN rb_event_flag_t ruby_vm_event_enabled_global_flags;
 RUBY_EXTERN unsigned int    ruby_vm_event_local_num;
 
-RUBY_EXTERN native_tls_key_t ruby_current_ec_key;
-
 RUBY_SYMBOL_EXPORT_END
 
 #define GET_VM()     rb_current_vm()
@@ -1764,7 +1762,15 @@ rb_ec_vm_ptr(const rb_execution_context_t *ec)
 static inline rb_execution_context_t *
 rb_current_execution_context(void)
 {
+#ifdef RB_THREAD_LOCAL_SPECIFIER
+  #if __APPLE__
+    rb_execution_context_t *ec = rb_current_ec();
+  #else
+    rb_execution_context_t *ec = ruby_current_ec;
+  #endif
+#else
     rb_execution_context_t *ec = native_tls_get(ruby_current_ec_key);
+#endif
     VM_ASSERT(ec != NULL);
     return ec;
 }