[jira] Commented: (STDCXX-518) std::string copy constructor slow

["Mark Brown (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/STDCXX-518?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12519279 ] 

Mark Brown commented on STDCXX-518:
-----------------------------------

I thought timings for multipled threads might be of interest. Unfortunately, I have only 2 CPUs to test on.

$ time ./string-copy.gcc 100000000

real    0m9.739s
user    0m19.257s
sys     0m0.060s
$ time ./string-copy.stlport 100000000

real    0m20.906s
user    0m40.211s
sys     0m0.056s
$ time LD_LIBRARY_PATH=../lib ./string-copy 100000000

real    0m17.585s
user    0m34.698s
sys     0m0.076s



#include <cassert>
#include <cstdlib>
#include <string>
#include <pthread.h>


std::string strings [256];
const std::size_t num_strings = sizeof strings / sizeof *strings;
unsigned long n;

extern "C" void* test_copy_ctor (void*)
{
    for (unsigned long i = 0; i != n; ++i) {
        const std::size_t index = i % num_strings;
        const std::string copy (strings [index]);
        assert (copy.length () == index);
    }

    return 0;
}

int main (int argc, char *argv[])
{
    n = argc < 2 ? 0 : std::strtoul (argv [1], 0, 10);

    for (unsigned long i = 0; i != num_strings; ++i)
        strings [i].assign (i, 'x');

    pthread_t thread [2];
    const std::size_t num_threads = sizeof thread / sizeof *thread;

    for (std::size_t i = 0; i < num_threads; ++i)
        pthread_create (thread + i, 0, test_copy_ctor, 0);

    for (std::size_t i = 0; i < num_threads; ++i)
        pthread_join (thread [i], 0);
}


> std::string copy constructor slow
> ---------------------------------
>
>                 Key: STDCXX-518
>                 URL: https://issues.apache.org/jira/browse/STDCXX-518
>             Project: C++ Standard Library
>          Issue Type: Improvement
>          Components: 21. Strings
>    Affects Versions: 4.1.3
>         Environment: gcc 4.1.2, x86_64
>            Reporter: Mark Brown
>
> When thread safety is enabled in stdcxx the string copy constructor is more than ten times slower than when it's not, and twice as slow as the same copy constructor in gcc 4.1.2 on Linux (x86_64), but slightly faster than with STLport 5.1.3. The timings were done on Intel x86_64 at 2.4GHz:
> BUILDMODE=shared,optimized:
> $ time LD_LIBRARY_PATH=../lib ./string-copy 100000000
> real    0m0.482s
> user    0m0.480s
> sys     0m0.000s
> BUILDMODE=shared,optimized,pthread:
> $ time LD_LIBRARY_PATH=../lib ./string-copy 100000000
> real    0m10.157s
> user    0m10.041s
> sys     0m0.032s
> gcc 4.1.2 with -O2 -m64:
> $ time ./string-copy.gcc 100000000
> real    0m4.280s
> user    0m4.260s
> sys     0m0.020s
> gcc 4.1.2 with STLport 5.1.3 (-D_REENTRANT -O2 -m64 -lpthread):
> $ time ./string-copy.stlport 100000000
> real    0m12.479s
> user    0m12.473s
> sys     0m0.004s
> #include <cassert>
> #include <cstdlib>
> #include <string>
> int main (int argc, char *argv[])
> {
>     const unsigned long n = argc < 2 ? 0 : std::strtoul (argv [1], 0, 10);
>     std::string strings [256];
>     const std::size_t num_strings = sizeof strings / sizeof *strings;
>     for (unsigned long i = 0; i != num_strings; ++i)
>         strings [i].assign (i, 'x');
>     for (unsigned long i = 0; i < n; ++i) {
>         const std::size_t length = i % num_strings;
>         const std::string str (strings [length]);
>         assert (str.size () == length);
>     }
> }

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.