NSSet vs.unordered_set

Question

I performed a simple randomized test of the performance of Cocoa's NSSet vs. the STL's unordered_set . Both are implemented using hash tables. Being that C++ is known for its highly performant code, I was surprised to see that NSSet was ~1.5x faster than unordered_set . Could anyone shed some light on why this is?

PS The test was searching random strings (using NSString for NSSet and basic_string<unichar> for unordered_set ).

EDIT: Here's the code: https://github.com/fumoboy007/StringCollectionSpeedTest .

Answer 1

Enable optimization and enjoy how "fast" C++ containers are: zero nano seconds for the loop, that has been optimized away by the compiler! Didn't you here it mumbling: "Don't mess with me!" Compilers are smarter than you think. ;)

So, first trick the compiler: Restructure your test, so that a result will be generated by the loop, and that result MUST depend on the actions in each iteration (for example sum the count). That result must be used as well somewhere, for example print it to the console. Otherwise the compiler will again detect that you don't need that result and thus, don't need to perform the loop at all.

Edit:

Test Results (on my system):

It turned out that NSSet and std::unordered_set are about equal in the time for a lookup.