Ah that explains it! Thanks for the benchmarks. We only tested it with the old 4.8 implementation and decided to go with std::mutex instead. We might check out QMutex again! Does it support the behaviour of std::recursive_mutex?
On Wed, Jan 22, 2014 at 11:16 PM, Thiago Macieira <thiago.macie...@intel.com > wrote: > On quarta-feira, 22 de janeiro de 2014 22:01:39, Philipp Kursawe wrote: > > sorry got that mixed up. Qt uses Events under Windows (was that changed > in > > Qt5?), which is still not as efficient as CriticalSections. > > Yes, QMutex got a large rewrite in Qt 5. > > I'd like to see your benchmarks. This is what the one in > tests/benchmarks/corelib/thread/qmutex say: > > Non-recursive mutex: > - QMutex construction time and CriticalSection construction time are very > efficient. QMutex is 15.8% faster. > > - QMutex performance with uncontended locks is very good too. QMutex is > 29.8% > faster > > - for contended mutexes, QMutex also wins: 36.5% faster on the 1-mutex > testcase, 49.2% faster on the 2-mutex case > > Recursive mutex: > - QMutex construction is much worse, since it needs to allocate memory; > 5.4x > worse > > - uncontended locks: QMutex is 48% slower > > - contended locks: QMutex is between 100% to 500% slower > > The recursive QMutex and the critical section objects are very much alike. > The > only reason I can think of for why QMutex is that much slower is due to the > indirection and extra memory allocation. > > For Qt 6, it might be a good idea to make QMutex bigger so it can store the > fields for a recursive mutex without indirection and memory allocation. > > -- > Thiago Macieira - thiago.macieira (AT) intel.com > Software Architect - Intel Open Source Technology Center > > _______________________________________________ > Interest mailing list > Interest@qt-project.org > http://lists.qt-project.org/mailman/listinfo/interest > >
_______________________________________________ Interest mailing list Interest@qt-project.org http://lists.qt-project.org/mailman/listinfo/interest
