[atomic] On IBM, use light-weight sync for everything

lwsync() is a full read/write-barrier.  That's all we need, never
need sync().  I'm not sure why an isync() was used in fetch_and_add,
but since that's a read-modify-write, I just changed it to have
lwsync() on both sides.

diff --git a/src/hb-atomic-private.hh b/src/hb-atomic-private.hh
index 0d0badf..f60c46e 100644 (file)
--- a/src/hb-atomic-private.hh
+++ b/src/hb-atomic-private.hh
@@ -163,20 +163,20 @@ static inline void _hb_memory_barrier (void)      { OSMemoryBarrier (); }
 static inline int _hb_fetch_and_add(volatile int* AI, unsigned int V) {
   __lwsync();
   int result = __fetch_and_add(AI, V);
-  __isync();
+  __lwsync();
   return result;
 }
 static inline int _hb_compare_and_swaplp(volatile long* P, long O, long N) {
-  __sync();
+  __lwsync();
   int result = __compare_and_swaplp (P, &O, N);
-  __sync();
+  __lwsync();
   return result;
 }
 
 typedef int hb_atomic_int_impl_t;
 #define hb_atomic_int_impl_add(AI, V)           _hb_fetch_and_add ((AI), (V))
 
-static inline void _hb_memory_barrier (void)   { __sync(); }
+static inline void _hb_memory_barrier (void)   { __lwsync(); }
 
 #define hb_atomic_ptr_impl_cmpexch(P,O,N)       _hb_compare_and_swaplp ((long*)(P), (long)(O), (long)(N))