#include #include #include #include // I noticed std::regex_search cropping up in profiling results. // Switching to boost::regex_search yielded a huge benefit: sometimes a 15x // increase in regex matching (on VS builds in particular). // Should be easily reproducible using these micro-benchmarks. namespace { class NumberExamples : public benchmark::Fixture { protected: std::vector m_numbers{ "0", "123", "0.123", ".123", "1e9", "1.87E-18", "012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789.012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789", }; }; class WhitespaceExamples : public benchmark::Fixture { protected: std::vector m_whitespace{ "", " 1", " 123", }; }; } BENCHMARK_F(NumberExamples, StdParseNumber)(benchmark::State& state) { using namespace math::server::lexer::details; for (auto _ : state) { for (const auto& src : m_numbers) { impl::std_parse_number(src); } } } BENCHMARK_F(NumberExamples, BoostParseNumber)(benchmark::State& state) { using namespace math::server::lexer::details; for (auto _ : state) { for (const auto& src : m_numbers) { impl::boost_parse_number(src); } } } BENCHMARK_F(WhitespaceExamples, StdParseWhitespace)(benchmark::State& state) { using namespace math::server::lexer::details; for (auto _ : state) { for (const auto& src : m_whitespace) { impl::std_parse_whitespace(src); } } } BENCHMARK_F(WhitespaceExamples, BoostParseWhitespace)(benchmark::State& state) { using namespace math::server::lexer::details; for (auto _ : state) { for (const auto& src : m_whitespace) { impl::boost_parse_whitespace(src); } } }