Hi Constantin. Thank you for looking at my benchmark. > Am 24.07.2015 um 08:57 schrieb Constantin Makshin <cmaks...@gmail.com>: > > Well, after looking at the code I can say that the was you wrote this > benchmark "abuses" the QVector's copy-on-write semantic, making it > somewhat biased towards std::vector — you use only non-const versions of > QVector::begin(), QVector::end() and indexing methods. That leads to a > lot of [obviously non-free] checks for sharing and possible detachment. >
Actually imo this is the most important case. Why should i iterate over a vector and not change the values? If i search in a vector it should be sorted. The sorted case was tested in the other tests together with map, but not the const case, so i added that now, but there is not much of a difference here. > A few modifications (see the attachment) made the difference in > read-only access negligible, in some tests QVector gave slightly better > results than std::vector. QVector_fwd_it became more than 2x faster. > You also change the order when inserting, so instead of jumping from container to container, you insert in a row, but i did that intentionally to minimize L1 cache effect. The other iteration tests i added additionally. But the results show still a 50% better performance for std::vector when inserting and 2 times better performance when iteration non const. In the other cases QVector is on par with std::vector. So i still say choose QVector only when you really need the implicit sharing. Regards, Gunnar Roth _______________________________________________ Interest mailing list Interest@qt-project.org http://lists.qt-project.org/mailman/listinfo/interest
