Natanael Copa wrote: > Note that there are 64 CPU cores. I have only tested with that many cores on > aarch64.
You could vary the number of CPU cores using the 'taskset' command. What I see in an Alpine Linux 3.20 / x86_64 VM: $ time taskset -c 0,1,2,3,4,5,6,7 ./test-pthread-rwlock real time 0.74 sec $ time taskset -c 0,1,2,3 ./test-pthread-rwlock $ time taskset -c 4,5,6,7 ./test-pthread-rwlock real time 1.06 sec $ time taskset -c 0,1 ./test-pthread-rwlock $ time taskset -c 2,3 ./test-pthread-rwlock $ time taskset -c 4,5 ./test-pthread-rwlock $ time taskset -c 6,7 ./test-pthread-rwlock real time 1.99 sec $ time taskset -c 0 ./test-pthread-rwlock $ time taskset -c 1 ./test-pthread-rwlock $ time taskset -c 2 ./test-pthread-rwlock real time 3.78 sec > Not sure the exact codepath it takes on GNU libc systems. Is it the > same as with musl libc? At this point, I would concentrate on the musl side. One of the interesting questions is why SCHED_FIFO and SCHED_RR are faster by 25% than SCHED_OTHER, even on x86_64. > > Note: Most Gnulib applications don't use pthread_rwlock directly, but > > the glthread_rwlock facility. On musl systems, it works around the > > possible writer starvation by reimplementing read-write locks based > > on condition variables. This may be slower for a single operation, > > but it is guaranteed to avoid writer starvation and therefore is > > preferrable globally. This is why you don't see a timeout in > > './test-lock', only in './test-pthread-rwlock'. > > Wait a second. The test does not run the gnulib locking? It just tests > the system (musl libc) pthread rwlock, while the app (gettext) would > use the gnulib implementation? You are right that it's confusing: The test contains some test fixture, that starts 20 threads, does atomic operations, and finally pthread_join(). And, within that test fixture, it does the pthread_rwlock_* operations. It could also be a problem with the test fixture, e.g. with the atomic operations, currently implemented via semaphores. You can try to replace in atomic-int-posix.h #define USE_VOLATILE 0 by #define USE_VOLATILE 1 or disable the use of semaphores, and see whether that has an effect. Bruno
