[PATCH v4 3/3] Optimize CAS in __pthread_mutex_lock_full [BZ #28537]
H.J. Lu
hjl.tools@gmail.com
Wed Nov 10 00:16:14 GMT 2021
1. Do an atomic load and skip CAS if compare may fail to reduce cache
line bouncing on contended locks.
2. Replace atomic_compare_and_exchange_bool_acq with
atomic_compare_and_exchange_val_acq to avoid the extra load.
---
nptl/pthread_mutex_lock.c | 15 ++++++++++-----
1 file changed, 10 insertions(+), 5 deletions(-)
diff --git a/nptl/pthread_mutex_lock.c b/nptl/pthread_mutex_lock.c
index 1126ecba95..9a18073f29 100644
--- a/nptl/pthread_mutex_lock.c
+++ b/nptl/pthread_mutex_lock.c
@@ -309,12 +309,17 @@ __pthread_mutex_lock_full (pthread_mutex_t *mutex)
meantime. */
if ((oldval & FUTEX_WAITERS) == 0)
{
- if (atomic_compare_and_exchange_bool_acq (&mutex->__data.__lock,
- oldval | FUTEX_WAITERS,
- oldval)
- != 0)
+ int currval = atomic_load_relaxed (&mutex->__data.__lock);
+ if (currval != oldval)
+ {
+ oldval = currval;
+ continue;
+ }
+ int newval = atomic_compare_and_exchange_val_acq
+ (&mutex->__data.__lock, oldval | FUTEX_WAITERS, oldval);
+ if (newval != oldval)
{
- oldval = mutex->__data.__lock;
+ oldval = newval;
continue;
}
oldval |= FUTEX_WAITERS;
--
2.33.1
More information about the Libc-alpha
mailing list