Re: Fix performance of generic atomics

Sokolov Yura <funny(dot)falcon(at)postgrespro(dot)ru> writes:
@@ -382,12 +358,8 @@ static inline uint64
pg_atomic_fetch_and_u64_impl(volatile pg_atomic_uint64 *ptr, uint64 and_)
{
uint64 old;
- while (true)
- {
- old = pg_atomic_read_u64_impl(ptr);
- if (pg_atomic_compare_exchange_u64_impl(ptr, &old, old & and_))
- break;
- }
+ old = pg_atomic_read_u64_impl(ptr);
+ while (!pg_atomic_compare_exchange_u64_impl(ptr, &old, old & and_));
return old;
}
#endif

FWIW, I do not think that writing the loops like that is good style.
It looks like a typo and will confuse readers. You could perhaps
write the same code with better formatting, eg

while (!pg_atomic_compare_exchange_u64_impl(ptr, &old, old & and_))
/* skip */ ;

but why not leave the formulation with while(true) and a break alone?

(I take no position on whether moving the read of "old" outside the
loop is a valid optimization.)

regards, tom lane