OSDN Git Service

2014-01-25 Walter Lee <walt@tilera.com>
authorwalt <walt@138bc75d-0d04-0410-961f-82ee72b054a4>
Sat, 25 Jan 2014 22:49:41 +0000 (22:49 +0000)
committerwalt <walt@138bc75d-0d04-0410-961f-82ee72b054a4>
Sat, 25 Jan 2014 22:49:41 +0000 (22:49 +0000)
            Backport from mainline
            2014-01-25  Walter Lee  <walt@tilera.com>

            * config/tilepro/atomic.c (pre_atomic_barrier): Mark inline.
            (post_atomic_barrier): Ditto.
            (__fetch_and_do): New macro.
            (__atomic_fetch_and_do): Use __fetch_and_do.
            (__sync_fetch_and_do): New macro.
            (__sync_fetch_and_add_4): New function.
            (__sync_fetch_and_sub_4): New function.
            (__sync_fetch_and_or_4): New function.
            (__sync_fetch_and_and_4): New function.
            (__sync_fetch_and_xor_4): New function.
            (__sync_fetch_and_nand_4): New function.
            (__sync_fetch_and_add_8): New function.
            (__sync_fetch_and_sub_8): New function.
            (__sync_fetch_and_or_8): New function.
            (__sync_fetch_and_and_8): New function.
            (__sync_fetch_and_xor_8): New function.
            (__sync_fetch_and_nand_8): New function.
            (__do_and_fetch): New macro.
            (__atomic_do_and_fetch): Use __do_and_fetch.
            (__sync_do_and_fetch): New macro.
            (__sync_add_and_fetch_4): New function.
            (__sync_sub_and_fetch_4): New function.
            (__sync_or_and_fetch_4): New function.
            (__sync_and_and_fetch_4): New function.
            (__sync_xor_and_fetch_4): New function.
            (__sync_nand_and_fetch_4): New function.
            (__sync_add_and_fetch_8): New function.
            (__sync_sub_and_fetch_8): New function.
            (__sync_or_and_fetch_8): New function.
            (__sync_and_and_fetch_8): New function.
            (__sync_xor_and_fetch_8): New function.
            (__sync_nand_and_fetch_8): New function.
            (__sync_exchange_methods): New macro.
            (__sync_val_compare_and_swap_4): New function.
            (__sync_bool_compare_and_swap_4): New function.
            (__sync_lock_test_and_test_4): New function.
            (__sync_val_compare_and_swap_8): New function.
            (__sync_bool_compare_and_swap_8): New function.
            (__sync_lock_test_and_test_8): New function.
            (__subword_cmpxchg_body): New macro.
            (__atomic_compare_exchange_1): Use __subword_cmpxchg_body.
            (__atomic_compare_exchange_2): Ditto.
            (__sync_subword_cmpxchg): New macro.
            (__sync_val_compare_and_swap_1): New function.
            (__sync_bool_compare_and_swap_1): New function.
            (__sync_val_compare_and_swap_2): New function.
            (__sync_bool_compare_and_swap_2): New function.
            (__atomic_subword): Rename to ...
            (__subword): ... New name.
            (__atomic_subword_fetch): Use __subword.
            (__sync_subword_fetch): New macro.
            (__sync_fetch_and_add_1): New function.
            (__sync_fetch_and_sub_1): New function.
            (__sync_fetch_and_or_1): New function.
            (__sync_fetch_and_and_1): New function.
            (__sync_fetch_and_xor_1): New function.
            (__sync_fetch_and_nand_1): New function.
            (__sync_fetch_and_add_2): New function.
            (__sync_fetch_and_sub_2): New function.
            (__sync_fetch_and_or_2): New function.
            (__sync_fetch_and_and_2): New function.
            (__sync_fetch_and_xor_2): New function.
            (__sync_fetch_and_nand_2): New function.
            (__sync_add_and_fetch_1): New function.
            (__sync_sub_and_fetch_1): New function.
            (__sync_or_and_fetch_1): New function.
            (__sync_and_and_fetch_1): New function.
            (__sync_xor_and_fetch_1): New function.
            (__sync_nand_and_fetch_1): New function.
            (__sync_add_and_fetch_2): New function.
            (__sync_sub_and_fetch_2): New function.
            (__sync_or_and_fetch_2): New function.
            (__sync_and_and_fetch_2): New function.
            (__sync_xor_and_fetch_2): New function.
            (__sync_nand_and_fetch_2): New function.
            (__atomic_subword_lock): Use __subword.
            (__sync_subword_lock): New macro.
            (__sync_lock_test_and_set_1): New function.
            (__sync_lock_test_and_set_2): New function.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/gcc-4_7-branch@207107 138bc75d-0d04-0410-961f-82ee72b054a4

libgcc/ChangeLog
libgcc/config/tilepro/atomic.c

index f5db856..87f3eab 100644 (file)
@@ -3,6 +3,91 @@
        Backport from mainline
        2014-01-25  Walter Lee  <walt@tilera.com>
 
+       * config/tilepro/atomic.c (pre_atomic_barrier): Mark inline.
+       (post_atomic_barrier): Ditto.
+       (__fetch_and_do): New macro.
+       (__atomic_fetch_and_do): Use __fetch_and_do.
+       (__sync_fetch_and_do): New macro.
+       (__sync_fetch_and_add_4): New function.
+       (__sync_fetch_and_sub_4): New function.
+       (__sync_fetch_and_or_4): New function.
+       (__sync_fetch_and_and_4): New function.
+       (__sync_fetch_and_xor_4): New function.
+       (__sync_fetch_and_nand_4): New function.
+       (__sync_fetch_and_add_8): New function.
+       (__sync_fetch_and_sub_8): New function.
+       (__sync_fetch_and_or_8): New function.
+       (__sync_fetch_and_and_8): New function.
+       (__sync_fetch_and_xor_8): New function.
+       (__sync_fetch_and_nand_8): New function.
+       (__do_and_fetch): New macro.
+       (__atomic_do_and_fetch): Use __do_and_fetch.
+       (__sync_do_and_fetch): New macro.
+       (__sync_add_and_fetch_4): New function.
+       (__sync_sub_and_fetch_4): New function.
+       (__sync_or_and_fetch_4): New function.
+       (__sync_and_and_fetch_4): New function.
+       (__sync_xor_and_fetch_4): New function.
+       (__sync_nand_and_fetch_4): New function.
+       (__sync_add_and_fetch_8): New function.
+       (__sync_sub_and_fetch_8): New function.
+       (__sync_or_and_fetch_8): New function.
+       (__sync_and_and_fetch_8): New function.
+       (__sync_xor_and_fetch_8): New function.
+       (__sync_nand_and_fetch_8): New function.
+       (__sync_exchange_methods): New macro.
+       (__sync_val_compare_and_swap_4): New function.
+       (__sync_bool_compare_and_swap_4): New function.
+       (__sync_lock_test_and_test_4): New function.
+       (__sync_val_compare_and_swap_8): New function.
+       (__sync_bool_compare_and_swap_8): New function.
+       (__sync_lock_test_and_test_8): New function.
+       (__subword_cmpxchg_body): New macro.
+       (__atomic_compare_exchange_1): Use __subword_cmpxchg_body.
+       (__atomic_compare_exchange_2): Ditto.
+       (__sync_subword_cmpxchg): New macro.
+       (__sync_val_compare_and_swap_1): New function.
+       (__sync_bool_compare_and_swap_1): New function.
+       (__sync_val_compare_and_swap_2): New function.
+       (__sync_bool_compare_and_swap_2): New function.
+       (__atomic_subword): Rename to ...
+       (__subword): ... New name.
+       (__atomic_subword_fetch): Use __subword.
+       (__sync_subword_fetch): New macro.
+       (__sync_fetch_and_add_1): New function.
+       (__sync_fetch_and_sub_1): New function.
+       (__sync_fetch_and_or_1): New function.
+       (__sync_fetch_and_and_1): New function.
+       (__sync_fetch_and_xor_1): New function.
+       (__sync_fetch_and_nand_1): New function.
+       (__sync_fetch_and_add_2): New function.
+       (__sync_fetch_and_sub_2): New function.
+       (__sync_fetch_and_or_2): New function.
+       (__sync_fetch_and_and_2): New function.
+       (__sync_fetch_and_xor_2): New function.
+       (__sync_fetch_and_nand_2): New function.
+       (__sync_add_and_fetch_1): New function.
+       (__sync_sub_and_fetch_1): New function.
+       (__sync_or_and_fetch_1): New function.
+       (__sync_and_and_fetch_1): New function.
+       (__sync_xor_and_fetch_1): New function.
+       (__sync_nand_and_fetch_1): New function.
+       (__sync_add_and_fetch_2): New function.
+       (__sync_sub_and_fetch_2): New function.
+       (__sync_or_and_fetch_2): New function.
+       (__sync_and_and_fetch_2): New function.
+       (__sync_xor_and_fetch_2): New function.
+       (__sync_nand_and_fetch_2): New function.
+       (__atomic_subword_lock): Use __subword.
+       (__sync_subword_lock): New macro.
+       (__sync_lock_test_and_set_1): New function.
+       (__sync_lock_test_and_set_2): New function.
+
+2014-01-25  Walter Lee  <walt@tilera.com>
+
+       Backport from mainline
+       2014-01-25  Walter Lee  <walt@tilera.com>
+
        * config/tilepro/atomic.c (BIT_OFFSET): Define.
        (__atomic_subword_cmpxchg): Use BIT_OFFSET.
        (__atomic_subword): Ditto.
index 7188962..038091e 100644 (file)
@@ -29,7 +29,7 @@
 /* This code should be inlined by the compiler, but for now support
    it as out-of-line methods in libgcc.  */
 
-static void
+static inline void
 pre_atomic_barrier (int model)
 {
   switch ((enum memmodel) model)
@@ -45,7 +45,7 @@ pre_atomic_barrier (int model)
   return;
 }
 
-static void
+static inline void
 post_atomic_barrier (int model)
 {
   switch ((enum memmodel) model)
@@ -63,16 +63,21 @@ post_atomic_barrier (int model)
 
 #define __unused __attribute__((unused))
 
-#define __atomic_fetch_and_do(type, size, opname)              \
-type                                                           \
-__atomic_fetch_##opname##_##size(type* p, type i, int model)   \
+#define __fetch_and_do(proto, type, size, opname, top, bottom) \
+proto                                                          \
 {                                                              \
-  pre_atomic_barrier(model);                                   \
+  top;                                                         \
   type rv = arch_atomic_##opname(p, i);                                \
-  post_atomic_barrier(model);                                  \
+  bottom;                                                      \
   return rv;                                                   \
 }
 
+#define __atomic_fetch_and_do(type, size, opname)                      \
+  __fetch_and_do(type __atomic_fetch_##opname##_##size(type* p, type i, int model), \
+                type, size, opname,                                    \
+                pre_atomic_barrier(model),                             \
+                post_atomic_barrier(model))                            \
+
 __atomic_fetch_and_do (int, 4, add)
 __atomic_fetch_and_do (int, 4, sub)
 __atomic_fetch_and_do (int, 4, or)
@@ -86,15 +91,40 @@ __atomic_fetch_and_do (long long, 8, and)
 __atomic_fetch_and_do (long long, 8, xor)
 __atomic_fetch_and_do (long long, 8, nand)
 
-#define __atomic_do_and_fetch(type, size, opname, op, op2)     \
-type                                                           \
-__atomic_##opname##_fetch_##size(type* p, type i, int model)   \
-{                                                              \
-  pre_atomic_barrier(model);                                   \
-  type rv = op2 (arch_atomic_##opname(p, i) op i);             \
-  post_atomic_barrier(model);                                  \
-  return rv;                                                   \
+#define __sync_fetch_and_do(type, size, opname)                                \
+  __fetch_and_do(type __sync_fetch_and_##opname##_##size(type* p, type i), \
+                type, size, opname,                                    \
+                arch_atomic_write_barrier(),                           \
+                arch_atomic_read_barrier())
+
+__sync_fetch_and_do (int, 4, add)
+__sync_fetch_and_do (int, 4, sub)
+__sync_fetch_and_do (int, 4, or)
+__sync_fetch_and_do (int, 4, and)
+__sync_fetch_and_do (int, 4, xor)
+__sync_fetch_and_do (int, 4, nand)
+__sync_fetch_and_do (long long, 8, add)
+__sync_fetch_and_do (long long, 8, sub)
+__sync_fetch_and_do (long long, 8, or)
+__sync_fetch_and_do (long long, 8, and)
+__sync_fetch_and_do (long long, 8, xor)
+__sync_fetch_and_do (long long, 8, nand)
+
+#define __do_and_fetch(proto, type, size, opname, op, op2, top, bottom)        \
+proto                                                                  \
+{                                                                      \
+  top;                                                                 \
+  type rv = op2 (arch_atomic_##opname(p, i) op i);                     \
+  bottom;                                                              \
+  return rv;                                                           \
 }
+
+#define __atomic_do_and_fetch(type, size, opname, op, op2)             \
+  __do_and_fetch(type __atomic_##opname##_fetch_##size(type* p, type i, int model), \
+                type, size, opname, op, op2,                           \
+                pre_atomic_barrier(model),                             \
+                post_atomic_barrier(model))                            \
+
 __atomic_do_and_fetch (int, 4, add, +, )
 __atomic_do_and_fetch (int, 4, sub, -, )
 __atomic_do_and_fetch (int, 4, or, |, )
@@ -108,6 +138,25 @@ __atomic_do_and_fetch (long long, 8, and, &, )
 __atomic_do_and_fetch (long long, 8, xor, |, )
 __atomic_do_and_fetch (long long, 8, nand, &, ~)
 
+#define __sync_do_and_fetch(type, size, opname, op, op2)               \
+  __do_and_fetch(type __sync_##opname##_and_fetch_##size(type* p, type i), \
+                type, size, opname, op, op2,                           \
+                arch_atomic_write_barrier(),                           \
+                arch_atomic_read_barrier())                            \
+
+__sync_do_and_fetch (int, 4, add, +, )
+__sync_do_and_fetch (int, 4, sub, -, )
+__sync_do_and_fetch (int, 4, or, |, )
+__sync_do_and_fetch (int, 4, and, &, )
+__sync_do_and_fetch (int, 4, xor, |, )
+__sync_do_and_fetch (int, 4, nand, &, ~)
+__sync_do_and_fetch (long long, 8, add, +, )
+__sync_do_and_fetch (long long, 8, sub, -, )
+__sync_do_and_fetch (long long, 8, or, |, )
+__sync_do_and_fetch (long long, 8, and, &, )
+__sync_do_and_fetch (long long, 8, xor, |, )
+__sync_do_and_fetch (long long, 8, nand, &, ~)
+
 #define __atomic_exchange_methods(type, size)                          \
 bool                                                                   \
 __atomic_compare_exchange_##size(volatile type* ptr, type* oldvalp,    \
@@ -135,6 +184,36 @@ __atomic_exchange_##size(volatile type* ptr, type val, int model)  \
 __atomic_exchange_methods (int, 4)
 __atomic_exchange_methods (long long, 8)
 
+#define __sync_exchange_methods(type, size)                            \
+type                                                                   \
+__sync_val_compare_and_swap_##size(type* ptr, type oldval, type newval)        \
+{                                                                      \
+  arch_atomic_write_barrier();                                         \
+  type retval = arch_atomic_val_compare_and_exchange(ptr, oldval, newval); \
+  arch_atomic_read_barrier();                                          \
+  return retval;                                                       \
+}                                                                      \
+                                                                       \
+bool                                                                   \
+__sync_bool_compare_and_swap_##size(type* ptr, type oldval, type newval) \
+{                                                                      \
+  arch_atomic_write_barrier();                                         \
+  bool retval = arch_atomic_bool_compare_and_exchange(ptr, oldval, newval); \
+  arch_atomic_read_barrier();                                          \
+  return retval;                                                       \
+}                                                                      \
+                                                                       \
+type                                                                   \
+__sync_lock_test_and_set_##size(type* ptr, type val)                   \
+{                                                                      \
+  type retval = arch_atomic_exchange(ptr, val);                                \
+  arch_atomic_acquire_barrier_value(retval);                           \
+  return retval;                                                       \
+}
+
+__sync_exchange_methods (int, 4)
+__sync_exchange_methods (long long, 8)
+
 #ifdef __LITTLE_ENDIAN__
 #define BIT_OFFSET(n, type) ((n) * 8)
 #else
@@ -147,38 +226,66 @@ __atomic_exchange_methods (long long, 8)
 #define u8 unsigned char
 #define u16 unsigned short
 
+#define __subword_cmpxchg_body(type, size, ptr, guess, val)            \
+  ({                                                                   \
+    unsigned int *p = (unsigned int *)((unsigned long)ptr & ~3UL);     \
+    const int shift = BIT_OFFSET((unsigned long)ptr & 3UL, type);      \
+    const unsigned int valmask = (1 << (sizeof(type) * 8)) - 1;                \
+    const unsigned int bgmask = ~(valmask << shift);                   \
+    unsigned int oldword = *p;                                         \
+    type oldval = (oldword >> shift) & valmask;                                \
+    if (__builtin_expect((oldval == guess), 1)) {                      \
+      unsigned int word = (oldword & bgmask) | ((val & valmask) << shift); \
+      oldword = arch_atomic_val_compare_and_exchange(p, oldword, word);        \
+      oldval = (oldword >> shift) & valmask;                           \
+    }                                                                  \
+    oldval;                                                            \
+  })                                                                   \
+
 #define __atomic_subword_cmpxchg(type, size)                           \
                                                                        \
 bool                                                                   \
-__atomic_compare_exchange_##size(volatile type* ptr, type* guess,      \
+__atomic_compare_exchange_##size(volatile type* ptr, type* guess_ptr,  \
                                 type val, bool weak __unused, int models, \
                                 int modelf __unused)                   \
 {                                                                      \
   pre_atomic_barrier(models);                                          \
-  unsigned int *p = (unsigned int *)((unsigned long)ptr & ~3UL);       \
-  const int shift = BIT_OFFSET((unsigned long)ptr & 3UL, type);                \
-  const unsigned int valmask = (1 << (sizeof(type) * 8)) - 1;          \
-  const unsigned int bgmask = ~(valmask << shift);                     \
-  unsigned int oldword = *p;                                           \
-  type oldval = (oldword >> shift) & valmask;                          \
-  if (__builtin_expect((oldval == *guess), 1)) {                       \
-    unsigned int word = (oldword & bgmask) | ((val & valmask) << shift); \
-    oldword = arch_atomic_val_compare_and_exchange(p, oldword, word);  \
-    oldval = (oldword >> shift) & valmask;                             \
-  }                                                                    \
+  type guess = *guess_ptr;                                             \
+  type oldval = __subword_cmpxchg_body(type, size, ptr, guess, val);   \
   post_atomic_barrier(models);                                         \
-  bool success = (oldval == *guess);                                   \
-  *guess = oldval;                                                     \
+  bool success = (oldval == guess);                                    \
+  *guess_ptr = oldval;                                                 \
   return success;                                                      \
 }
 
 __atomic_subword_cmpxchg (u8, 1)
 __atomic_subword_cmpxchg (u16, 2)
 
+#define __sync_subword_cmpxchg(type, size)                             \
+                                                                       \
+type                                                                   \
+__sync_val_compare_and_swap_##size(type* ptr, type guess, type val)    \
+{                                                                      \
+  arch_atomic_write_barrier();                                         \
+  type oldval = __subword_cmpxchg_body(type, size, ptr, guess, val);   \
+  arch_atomic_read_barrier();                                          \
+  return oldval;                                                       \
+}                                                                      \
+                                                                       \
+bool                                                                   \
+__sync_bool_compare_and_swap_##size(type* ptr, type guess, type val)   \
+{                                                                      \
+  type oldval = __sync_val_compare_and_swap_##size(ptr, guess, val);   \
+  return oldval == guess;                                              \
+}
+
+__sync_subword_cmpxchg (u8, 1)
+__sync_subword_cmpxchg (u16, 2)
+
 /* For the atomic-update subword methods, we use the same approach as
    above, but we retry until we succeed if the compare-and-exchange
    fails.  */
-#define __atomic_subword(type, proto, top, expr, bottom)               \
+#define __subword(type, proto, top, expr, bottom)                      \
 proto                                                                  \
 {                                                                      \
   top                                                                  \
@@ -199,11 +306,11 @@ proto                                                                     \
 }
 
 #define __atomic_subword_fetch(type, funcname, expr, retval)           \
-  __atomic_subword(type,                                               \
-                  type __atomic_ ## funcname(volatile type *ptr, type i, int model), \
-                  pre_atomic_barrier(model);,                          \
-                  expr,                                                \
-                  post_atomic_barrier(model); return retval;)
+  __subword(type,                                                      \
+           type __atomic_ ## funcname(volatile type *ptr, type i, int model), \
+           pre_atomic_barrier(model);,                                 \
+           expr,                                                       \
+           post_atomic_barrier(model); return retval;)
 
 __atomic_subword_fetch (u8, fetch_add_1, oldval + i, oldval)
 __atomic_subword_fetch (u8, fetch_sub_1, oldval - i, oldval)
@@ -233,13 +340,57 @@ __atomic_subword_fetch (u16, and_fetch_2, oldval & i, val)
 __atomic_subword_fetch (u16, xor_fetch_2, oldval ^ i, val)
 __atomic_subword_fetch (u16, nand_fetch_2, ~(oldval & i), val)
 
+#define __sync_subword_fetch(type, funcname, expr, retval)     \
+  __subword(type,                                              \
+           type __sync_ ## funcname(type *ptr, type i),        \
+           arch_atomic_read_barrier();,                        \
+           expr,                                               \
+           arch_atomic_write_barrier(); return retval;)
+
+__sync_subword_fetch (u8, fetch_and_add_1, oldval + i, oldval)
+__sync_subword_fetch (u8, fetch_and_sub_1, oldval - i, oldval)
+__sync_subword_fetch (u8, fetch_and_or_1, oldval | i, oldval)
+__sync_subword_fetch (u8, fetch_and_and_1, oldval & i, oldval)
+__sync_subword_fetch (u8, fetch_and_xor_1, oldval ^ i, oldval)
+__sync_subword_fetch (u8, fetch_and_nand_1, ~(oldval & i), oldval)
+
+__sync_subword_fetch (u16, fetch_and_add_2, oldval + i, oldval)
+__sync_subword_fetch (u16, fetch_and_sub_2, oldval - i, oldval)
+__sync_subword_fetch (u16, fetch_and_or_2, oldval | i, oldval)
+__sync_subword_fetch (u16, fetch_and_and_2, oldval & i, oldval)
+__sync_subword_fetch (u16, fetch_and_xor_2, oldval ^ i, oldval)
+__sync_subword_fetch (u16, fetch_and_nand_2, ~(oldval & i), oldval)
+
+__sync_subword_fetch (u8, add_and_fetch_1, oldval + i, val)
+__sync_subword_fetch (u8, sub_and_fetch_1, oldval - i, val)
+__sync_subword_fetch (u8, or_and_fetch_1, oldval | i, val)
+__sync_subword_fetch (u8, and_and_fetch_1, oldval & i, val)
+__sync_subword_fetch (u8, xor_and_fetch_1, oldval ^ i, val)
+__sync_subword_fetch (u8, nand_and_fetch_1, ~(oldval & i), val)
+
+__sync_subword_fetch (u16, add_and_fetch_2, oldval + i, val)
+__sync_subword_fetch (u16, sub_and_fetch_2, oldval - i, val)
+__sync_subword_fetch (u16, or_and_fetch_2, oldval | i, val)
+__sync_subword_fetch (u16, and_and_fetch_2, oldval & i, val)
+__sync_subword_fetch (u16, xor_and_fetch_2, oldval ^ i, val)
+__sync_subword_fetch (u16, nand_and_fetch_2, ~(oldval & i), val)
+
 #define __atomic_subword_lock(type, size)                              \
-                                                                       \
-__atomic_subword(type,                                                 \
-                type __atomic_exchange_##size(volatile type* ptr, type nval, int model), \
-                pre_atomic_barrier(model);,                            \
-                nval,                                                  \
-                post_atomic_barrier(model); return oldval;)
+  __subword(type,                                                      \
+           type __atomic_exchange_##size(volatile type* ptr, type nval, int model), \
+           pre_atomic_barrier(model);,                                 \
+           nval,                                                       \
+           post_atomic_barrier(model); return oldval;)
 
 __atomic_subword_lock (u8, 1)
 __atomic_subword_lock (u16, 2)
+
+#define __sync_subword_lock(type, size)                                        \
+  __subword(type,                                                      \
+           type __sync_lock_test_and_set_##size(type* ptr, type nval), \
+           ,                                                           \
+           nval,                                                       \
+           arch_atomic_acquire_barrier_value(oldval); return oldval;)
+
+__sync_subword_lock (u8, 1)
+__sync_subword_lock (u16, 2)