FastFormat - Fast, Safe C++ Formatting / Discussion / Help: fastformat 100 times slower of CString::Forma

Here is problem.
In Windows, VC 8 (VS 2005)
I built library as Release Multithreaded,

Either library was improperly built or fastformat is too s:low. 10M repeats of CString::Format takes ~6.5 sec in 32bit non-unicode build, while fastformat is some 10.5 minutes:

CString sink;
#define ATTEPMTS 10000000   
__int64 am = 3543000;
clock_t start1a = clock();
for( int t=0; t<ATTEPMTS; ++t)
    sink.Format ( "%I64d", am );
clock_t end1a = clock();
outut.Format( "CString::Format elapsed %.2f\n", (double(end1a) - double(start1a)) / CLOCKS_PER_SEC );
AfxMessageBox( outut );
start1a = clock();
for( int t=0; t<ATTEPMTS; ++t)
    fastformat::fmt( sink, "{0}", am );
end1a = clock();
outut.Format( "fastformat elapsed %.2f\n", (double(end1a) - double(start1a)) / CLOCKS_PER_SEC );
AfxMessageBox( outut );

That's because fmt() and write() append to the existing contents of the sink. So, in the statement

fastformat::fmt( sink, "{0}", am );

you're adding to a string each time, making a very large string: "354300035430003543000354300035430003543000354300035430003543000354300035430003543000354300035430003543000354300035430003543000354300035430003543000354300035430003543000 … etc."

Try placing the sink within the scope of the statement and you'll find the timings will be around the same. Here's a timing program I just knocked up:

#ifdef _DEBUG
    const unsigned  ITERATIONS  =   1000;
#else /* ? _DEBUG */
    const unsigned  ITERATIONS  =   1000000;
#endif /* _DEBUG */
    winstl::performance_counter counter;
    stlsoft::uint64_t   var;
    { for(int WARMUPS = 2; 0 != WARMUPS; --WARMUPS)
    {
        counter.start();
        { for(unsigned i = 0; i != ITERATIONS; ++i)
        {
            CString sink;
            sink.Format("%I64d", var);
        }}
        counter.stop();
        if(1 == WARMUPS)
        {
            ff::fmtln(std::cout, "CString::Format(): {0}ms", counter.get_milliseconds());
        }
        counter.start();
        { for(unsigned i = 0; i != ITERATIONS; ++i)
        {
            CString sink;
            ff::fmt(sink, "{0}", var);
        }}
        counter.stop();
        if(1 == WARMUPS)
        {
            ff::fmtln(std::cout, "ff::fmt(): {0}ms", counter.get_milliseconds());
        }
    }}

The output is

CString::Format(): 642ms
ff::fmt(): 598ms

So, in fact FastFormat is a little faster than CString::Format()

HTH

Matt

Trying again, to get the formatting correct:

{
#ifdef _DEBUG
    const unsigned  ITERATIONS  =   1000;
#else /* ? _DEBUG */
    const unsigned  ITERATIONS  =   1000000;
#endif /* _DEBUG */
    winstl::performance_counter counter;
    stlsoft::uint64_t   var;
    { for(int WARMUPS = 2; 0 != WARMUPS; --WARMUPS)
    {
        counter.start();
        { for(unsigned i = 0; i != ITERATIONS; ++i)
        {
            CString sink;
            sink.Format("%I64d", var);
        }}
        counter.stop();
        if(1 == WARMUPS)
        {
            ff::fmtln(std::cout, "CString::Format(): {0}ms", counter.get_milliseconds());
        }
        counter.start();
        { for(unsigned i = 0; i != ITERATIONS; ++i)
        {
            CString sink;
            ff::fmt(sink, "{0}", var);
        }}
        counter.stop();
        if(1 == WARMUPS)
        {
            ff::fmtln(std::cout, "ff::fmt(): {0}ms", counter.get_milliseconds());
        }
    }}
    /* . */
    return EXIT_SUCCESS;
}

Anonymous - 2010-05-12

Thank you,
I figured out it and was going to add comment about my mistake, but you already pointed that out.

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

fastformat 100 times slower of CString::Forma

Forums

Help

fastformat 100 times slower of CString::Forma document.SUBSCRIPTION_OPTIONS = { "thing": "topic", "subscribed": false, "url": "subscribe", "icon": { "css": "fa fa-envelope-o" } };

fastformat 100 times slower of CString::Forma