Probably a silly question, but has anyone performed benchmarks on using a std::for_each(begin, end, bind(&C::f, _1)); compared to the hand written loop? Under the debugger I see bind adds a 4 or 5 stack frames between the for_each frame and the C::f frame, so I'm wondering if there is any overhead in these? I could bench it for myself on my everyday compiler for my one task, but surely this was considered and is documented somewhere, much more extensively I'll ever can, no? Some boost libraries discuss performance, but I didn't such a section in the Boost.Bind doco. Thanks for pointers or insights, --DD
Dominique Devienne:
Probably a silly question, but has anyone performed benchmarks on using a std::for_each(begin, end, bind(&C::f, _1)); compared to the hand written loop? Under the debugger I see bind adds a 4 or 5 stack frames between the for_each frame and the C::f frame, so I'm wondering if there is any overhead in these?
It depends (heavily) on the amount of inlining the compiler does. In the simple case of #include <boost/bind.hpp> #include <algorithm> struct X { void f(); }; int main() { extern int N; extern X x[]; std::for_each( x, x + N, boost::bind( &X::f, _1 ) ); } the loop generated by MSVC 7.1 in a release build is $L9292: mov ecx, esi call ebx inc esi cmp esi, edi jne SHORT $L9292 where ebx points at X::f. More complicated examples would probably carry more overhead as the compiler inlining threshold is reached.
participants (2)
-
Dominique Devienne -
Peter Dimov