The HLE example in the manual only commits when using bool
for the flag, because __atomic_clear only writes bool, and
HLE requires the acquire and release to match.
So when the example is copied with e.g. an int variable it
does not commit and causes slower than expected performance.
Some people are running into problems because of this.
Switch it over to use __atomic_store.
Also fix a minor typo nearby.
gcc/:
2013-06-21 Andi Kleen <ak@linux.intel.com>
* doc/extend.texi: Dont use __atomic_clear in HLE
example. Fix typo.
git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@200304
138bc75d-0d04-0410-961f-
82ee72b054a4
2013-06-21 Andi Kleen <ak@linux.intel.com>
+ * doc/extend.texi: Dont use __atomic_clear in HLE
+ example. Fix typo.
+
+2013-06-21 Andi Kleen <ak@linux.intel.com>
+
* doc/extend.texi: Document that __atomic_clear and
__atomic_test_and_set should only be used with bool.
Memory model must be @code{__ATOMIC_RELEASE} or stronger.
@end table
-When a lock acquire fails it's required for good performance to abort
+When a lock acquire fails it is required for good performance to abort
the transaction quickly. This can be done with a @code{_mm_pause}
@smallexample
#include <immintrin.h> // For _mm_pause
+int lockvar;
+
/* Acquire lock with lock elision */
while (__atomic_exchange_n(&lockvar, 1, __ATOMIC_ACQUIRE|__ATOMIC_HLE_ACQUIRE))
_mm_pause(); /* Abort failed transaction */
...
/* Free lock with lock elision */
-__atomic_clear(&lockvar, __ATOMIC_RELEASE|__ATOMIC_HLE_RELEASE);
+__atomic_store(&lockvar, 0, __ATOMIC_RELEASE|__ATOMIC_HLE_RELEASE);
@end smallexample
@node Object Size Checking