i try profile cpu-intensive part of program.
it stunned me, unordered_map's iscontain, insert, , remove really slow.
here real example of result profiling:-
physic_object* cave = ... function argument ... if(cave->id==3){ //< 0.1% simple int int asdfdssd=0; } //"concaves" myencapsulatorof_unordered_map<physic_object*> (it field) //there 500-1000 elements in unordered_map. if (phy_concaves.iscontain(cave)) { //0.8% unordered_map "iscontain" phy_concaves.remove_immediate(cave); //4.9% unordered_map "iscontain" + "erase" cave->finalize_inverse(); //6.0% lot of unordered_map function delete cave; //0.7% simple delete }else { } here cheap hash function.
//physic_object.h public:virtual long hqhashcode() const override final { return id; //id +1 every time when new physic_object create //i have static int "id_runner" track. } from whole profile, unordered_map's operations gobbling 70%-80% of game.
question: unordered_map slow that, or there wrong code?
more specifically, common insertion/deletion cost around 40+ simple arithmetic operation?
here how create unordered_map, don't think there wrong, provided in case.
//myencapsulatorof_unordered_map<k> , k can pointer or value template<typename t2> static const t2 * ptr(t2 const& obj) { return &obj; } //turn reference pointer! template<typename t2> static t2 * ptr(t2 * obj) { return obj; } //obj pointer, return it! template< class t2>class structtptr1 { public: std::size_t operator()(t2 const& t) const{ std::size_t h1 = ptr(t)->hqhashcode(); return h1; } }; template< class t2>class structtptr2 { public: bool operator() (const t2 & t1, const t2&t2) const { return t1 == t2; } }; public:std::unordered_set<k, structtptr1<k>, structtptr2<k>>main; note ran in debug mode visual c++, release mode yields similar result.
why unordered_map hashing slow simple array hash seems related question, doesn't indicate/profile performance of unordered_map.
i don't think gcc either. therefore link may not relate. is gcc std::unordered_map implementation slow? if - why?
