Re: [boost] [convert] Performance

Subject: Re: [boost] [convert] Performance
From: Vladimir Batov (vb.mail.247_at_[hidden])
Date: 2014-06-11 02:58:50

Joel,

Joel de Guzman wrote
> Benchmarks are a black art. See how we do our performance tests in Spirit:
>   
> https://github.com/boostorg/spirit/blob/master/workbench/qi/int_parser.cpp
> You can use our benchmark facility where all the black art is contained:
>    https://github.com/boostorg/spirit/blob/master/workbench/measure.hpp
> ...
> Joel de Guzman

Thank you for the pointers. I've 

1) #included "qi/int_parser.cpp";
2) made spirit_new_test from spirit_int_test;
3) added boost::convert()-based test;
4) compiled with gcc (Ubuntu 4.8.2-19ubuntu1) 4.8.2 with "g++ -O2 -std=c++11
..."
5) got the following results:

atoi_test:            2.2135431510 [s] {checksum: 730759d}
strtol_test:          2.1688206260 [s] {checksum: 730759d}
spirit_int_test:     0.5207926250 [s] {checksum: 730759d}
spirit_new_test:   0.5803735980 [s] {checksum: 730759d}
cnv_test:             0.6161884860 [s] {checksum: 730759d}

struct spirit_new_test : test::base
{
    static int parse(std::string const& str)
    {
        std::string::const_iterator beg = str.begin();
        std::string::const_iterator end = str.end();
        int n;

        if (boost::spirit::qi::parse(beg, end, boost::spirit::qi::int_, n))
            if (beg == end)
                return n;

        return (BOOST_ASSERT(0), 0);
    }
    void benchmark()
    {
        for (int i = 0; i < 9; ++i)
            this->val += parse(numbers[i]);
    }
};

struct cnv_test : test::base
{
    void benchmark()
    {
        for (int i = 0; i < 9; ++i)
            this->val += boost::convert<int>(numbers[i],
boost::cnv::spirit()).value();
    }
};

1. The changes in spirit_new_test compared to the original spirit_int_test
were additional conversion validity checks and the std::string... the same
way boost::cnv::spirit works. That slowed spirit_new_test some 11% compared
to spirit_int_test. Though raw conversion speed is important
(spirit_int_test) I do not expect it deployed like that (with no checks) in
a realistic setting.

2. boost::cnv::spirit implements the same process-flow as spirit_new_test
and boost::cnv::spirit was consistently 6% slower compared to raw spirit
conversion deployment with the same functionality (spirit_new_test).

Thank you, Joel, for sharing your performance testing framework. 

--
View this message in context: http://boost.2283326.n4.nabble.com/review-Convert-library-tp4662821p4663981.html
Sent from the Boost - Dev mailing list archive at Nabble.com.