a question about efficiency of STL

[palance]

#include <list>
 
int main()
{
        std::list<int> lst;
        for(int i=0; i<1000; i++)
        {
                for(int x=0; x<5000; x++)
                        lst.push_back(x);
                lst.clear();
        }
        return 0;
}

I compiled this code on VC8 (Release version ) and Cygwin G++ 3.4.4 and compare the result :

$time test_gcc
real 0m13.100s
user 0m13.077s
sys 0m0.031s

$time test_vc
real 0m3.552s
user 0m0.031s
sys 0m0.015s

why the difference is so far?

[gpraghuram]

Every compiler implements the STL in a different way.
But i am also surrised to see this much difference

Raghuram

[Banfa]

This doesn't even have to be an STL issue, it could simply be a compiler/optomiser issue.

For instance did you put switches on both compilers to optimise the output exe for time exectution or did you optimise one for time and the other for size?

For a long while gcc was lagging in producing efficent code to run on Windows (although I thought they had more or less caught up now) and in fact I believe the best compiler you can get for an Intel based PC is the Intel compiler (not too unsurprisingly they understand best how to optimise code to run on their chips).

The thing is that the way gcc and GNU compilers in general work is to cross compile to an intermediary code (p code?) and then compile that p code into a platform specific exe. This makes it very easy to add a new platform supported by many languages or add a new language to many platforms but the method is generic and because of that it is inherently hard for it to compete on efficiency with a targeted method. That is a compiler targeted at 1 platform and 1 language.

Like I said I thought the gcc developers had done something clever to get round this to some extent on the later releases so I would try fiddling with the compiler switches as a start.

[weaknessforcats]

Whose STL is being used?

Visual C++ uses Plaugher's templates at Dinkumware.com. These are optimized for speed.