Hi,
This is the 0.15.5 release of liburcu. The most relevant
change introduced by this release is the use of
"lock; addl" to replace the "mfence" instruction for
cmm_smp_mb() on x86-64 when users build liburcu without
"compiler builtins" atomics. Users wishing to synchronize
with I/O already need to use cmm_mb().
This is motivated by the fact that "lock; addl" is significantly
faster than "mfence".
Detailed changelog:
2025-11-10 Userspace RCU 0.15.5
* x86: Define cmm_smp_mb() as lock; addl rather than mfence
* Introduce barrier test
* Add test_uatomic to gitignore
* Cleanup: Remove stray space
* benchmark: Add uatomic benchmark
Project website: https://liburcu.org
Git repository: https://git.liburcu.org/userspace-rcu.git
Thanks,
Mathieu
--
Mathieu Desnoyers
EfficiOS Inc.
https://www.efficios.com
