Re: bind2nd (and bind1st) wasting a copy ?
On 23 Jun, 02:08, Vincent Poinot <clcppm-pos...@this.is.invalid>
wrote:
Maxim Yegorushkin wrote:
Could you provide results of profiling your hypothesis please?
Well, I wrote a small test with an object whose copy constructor is not too
cheap (two strings). Here is the code:
#include <functional>
#include <ctime>
#include <iostream>
using namespace std ;
class Person
{
public:
Person(const string& first_name, const string& last_name) :
first_name_(first_name), last_name_(last_name)
{
}
string first_name() const { return first_name_ ; }
string last_name() const { return last_name_ ; }
private:
string first_name_ ;
string last_name_ ;
} ;
inline bool operator==(const Person& left, const Person& right)
{
return left.first_name() == right.first_name() &&
left.last_name() == right.last_name() ;
}
int main()
{
const Person jean("Sergei", "Eisenstein") ;
unsigned long max = 200000000 ;
time_t start = time(0) ;
for (int i = 0 ; i <= max ; ++ i)
bind2nd(equal_to<Person>(), jean) ;
cout << "Test took: " << time(0) - start << "\n" ;
}
The test, when using the STL version of bind2nd (GNU), takes 68s on my
machine. If I run the same test with the version I suggest, it drops to 37s
which is not far the half actually.
(I did the same test on Solaris and got similar results).
It is so indeed, please accept my apology for being sceptical.
Here are results I got with gcc 4.1 and Sun CC 5.9 (Sun Studio 12) on
Linux Fedora Core 5. The code has been slightly simplified, it
benchmarks std::bind2nd (GNU and stlport versions) against your
proposed version.
$ cat test.cc
#include <functional>
#include <ctime>
#include <vector>
#include <iostream>
#include <boost/type_traits/is_same.hpp>
struct data
{
std::vector<long> payload;
data() : payload(0x100) {}
};
template <class _Operation, class _Tp>
inline std::binder2nd<_Operation>
new_bind2nd(const _Operation& __fn, const _Tp& __x)
{
typedef typename _Operation::second_argument_type _Arg2_type;
// the dead branch is eliminated at compile time
return boost::is_same<_Arg2_type, _Tp>::value
? std::binder2nd<_Operation>(__fn, __x)
: std::binder2nd<_Operation>(__fn, _Arg2_type(__x))
;
}
int main()
{
data d;
unsigned const max = 10000000;
for(unsigned N = 3; N--;)
{
{
time_t start = time(0) ;
for (unsigned i = 0 ; i < max ; ++i)
std::bind2nd(std::equal_to<data>(), d);
time_t total = time(0) - start;
std::cout << "std::bind took: " << total << "\n";
}
{
time_t start = time(0) ;
for (unsigned i = 0 ; i < max ; ++i)
new_bind2nd(std::equal_to<data>(), d);
time_t total = time(0) - start;
std::cout << "new_bind took: " << total << "\n";
}
}
}
$ g++ --version
g++ (GCC) 4.1.1 20070105 (Red Hat 4.1.1-51)
Copyright (C) 2006 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There
is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR
PURPOSE.
$ g++ -Wall -Winline -O3 -o test.g++ test.cc
$ CC -V
CC: Sun C++ 5.9 Linux_i386 2007/05/03
$ CC +w2 -fast -library=stlport4 -o test.CC test.cc
"test.cc", line 34: Warning: "std::bind2nd<std::equal_to<data>,
data>(const std::equal_to<data>&, const data&)" is too large and will
not be expanded inline.
"test.cc", line 42: Warning: "new_bind2nd<std::equal_to<data>,
data>(const std::equal_to<data>&, const data&)" is too large and will
not be expanded inline.
2 Warning(s) detected.
$ ./test.g++
std::bind took: 14
new_bind took: 7
std::bind took: 14
new_bind took: 7
std::bind took: 14
new_bind took: 7
$ ./test.CC
std::bind took: 15
new_bind took: 7
std::bind took: 15
new_bind took: 7
std::bind took: 15
new_bind took: 8
--
[ See http://www.gotw.ca/resources/clcm.htm for info about ]
[ comp.lang.c++.moderated. First time posters: Do this! ]