Alexander Barkov added a comment - 2020-06-16 06:34 - edited dmitriy.philimonov , thanks for you feedback. I've created a separate issue MDEV-22849 for the my_hash_sort_utf8xxx related improvement and pushed it to 10.5. The coming soon 10.5.4 release will including this change. For the my_strnncollsp_utf8 change, is there a chance to get your patch for unittest/gunit? We'd like to run various benchmarks with this change. As it is useful mostly for the non-default sql_mode=PAD_CHAR_TO_FULL_LENGTH, we need to make sure that this change does not introduce performance degradation for the default sql_mode, when PAD_CHAR_TO_FULL_LENGTH is not enabled. Thanks for your contribution!

Dmitriy Philimonov added a comment - 2020-06-16 09:55 Dear Alexander, I'm glad that I was allowed to share the benchmark code with you. It uses unittest/gunit mini-benchmark engine from gtest. If somehow you don't have this engine, it's quite simple to write something similar based on std::chrono: #define BENCHMARK(func) \ TEST(Microbenchmarks, func) { internal_do_microbenchmark(#func, func); }   void internal_do_microbenchmark( const char *name, void (*func)( size_t )) { //calibration phase func(calibration_iterations); //estimation phase func(num_iterations); //print time spent in the func } The benchmark emulates CHAR(120) field. The field is filled with English/Russian/Chinese letters (1/2/3 utf8 bytes per character). The test consists of the cases where the field is filled fully/partially + the first non-equal letter is placed in the beginning/middle/end, or the strings are fully equal. I hope this will help. Put the code in unittest/gunit directory, resolve dependencies, estimate the results. #include "benchmark.h" namespace NMiniBenchmark { static size_t bytesSize = 360; static size_t charsSize = 120; static std::string GenerateStringEnglish(){ std::string s(bytesSize, ' ' ); for ( size_t i=0;i<charsSize;++i) s[i]= 'A' ; // 1 byte return s; } static std::string GenerateStringEnglishQuater(){ std::string s(bytesSize, ' ' ); //360 bytes // only 30 out of 120 characters are filled for ( size_t i=0;i<charsSize/4;++i) s[i]= 'A' ; return s; } static std::string GenerateStringEnglish40(){ std::string s = GenerateStringEnglish(); s[40]= 'B' ; return s; } static std::string GenerateStringEnglishZ(){ std::string s(bytesSize, ' ' ); for ( size_t i=0;i<charsSize;++i) s[i]= 'Z' ; // 1 byte return s; } static std::string GenerateStringRussian(){ std::string s(bytesSize, ' ' ); std::string ya= "Б" ; // 2 bytes for ( size_t i=0,j=0;i<charsSize;++i,j+=2) memcpy (&s[j], ya.data(), ya.size()); return s; } static std::string GenerateStringRussian40(){ std::string s = GenerateStringRussian(); std::string ya= "Б" ; // 2 bytes memcpy (&s[80], ya.data(), ya.size()); return s; } static std::string GenerateStringRussianZ(){ std::string s(bytesSize, ' ' ); std::string ya= "Я" ; // 2 bytes for ( size_t i=0,j=0;i<charsSize;++i,j+=2) memcpy (&s[j], ya.data(), ya.size()); return s; } static std::string GenerateStringChinese(){ std::string s(bytesSize, ' ' ); std::string ya= "我" ; // 3 bytes for ( size_t i=0,j=0;i<charsSize;++i,j+=3) memcpy (&s[j], ya.data(), ya.size()); return s; } static std::string GenerateStringChinese40(){ std::string s = GenerateStringChinese(); std::string ya= "是" ; // 3 bytes memcpy (&s[120],ya.data(),ya.size()); return s; }   static std::string GenerateStringChineseZ(){ std::string s(bytesSize, ' ' ); std::string ya= "是" ; // 3 bytes for ( size_t i=0,j=0;i<charsSize;++i,j+=3) memcpy (&s[j], ya.data(), ya.size()); return s; }   static std::string _English = GenerateStringEnglish(); static std::string _English40 = GenerateStringEnglish40(); static std::string _EnglishQ = GenerateStringEnglishQuater(); static std::string _Russian = GenerateStringRussian(); static std::string _Russian40 = GenerateStringRussian40(); static std::string _Chinese = GenerateStringChinese(); static std::string _Chinese40 = GenerateStringChinese40(); static std::string _EnglishZ = GenerateStringEnglishZ(); static std::string _RussianZ = GenerateStringRussianZ(); static std::string _ChineseZ = GenerateStringChineseZ();   TEST(SkipTrailingSpaceImprovement,CHECK_SIZES){ std::string yarus= "Я" ; // 2 bytes std::string yachina= "我" ; // 3 bytes ASSERT_EQ(yarus.size(),2u); ASSERT_EQ(yachina.size(),3u); ASSERT_EQ(skip_trailing_space(( const uchar*)_Russian.data(),_Russian.size())-( const uchar*)_Russian.data(), 240u); ASSERT_EQ(skip_trailing_space(( const uchar*)_Chinese.data(),_Chinese.size())-( const uchar*)_Chinese.data(), 360u); }   #define STRNNCOLLSP(i1,i2) \ void STRNNCOLLSP_##i1 ##i2( size_t iterations){ \ for ( size_t i=0;i<iterations;++i) \ my_charset_utf8_general_ci.coll->strnncollsp(&my_charset_utf8_general_ci, \ ( const uchar*)i1.c_str(),i1.size(), \ ( const uchar*)i2.c_str(),i2.size(), \ 0); \ } \ BENCHMARK(STRNNCOLLSP_##i1 ##i2)     STRNNCOLLSP(_EnglishQ,_EnglishQ) STRNNCOLLSP(_English,_English) STRNNCOLLSP(_English,_English40) STRNNCOLLSP(_English,_EnglishZ) STRNNCOLLSP(_Russian,_Russian) STRNNCOLLSP(_Russian,_Russian40) STRNNCOLLSP(_Russian,_RussianZ) STRNNCOLLSP(_Chinese,_Chinese) STRNNCOLLSP(_Chinese,_Chinese40) STRNNCOLLSP(_Chinese,_ChineseZ) } Sincerely yours, Dmitriy Philimonov

Julien Fritsch added a comment - 2023-12-05 09:26 Automated message: ---------------------------- Since this issue has not been updated since 6 weeks, it's time to move it back to Stalled.

JiraAutomate added a comment - 2023-12-05 11:48 Automated message: ---------------------------- Since this issue has not been updated since 6 weeks, it's time to move it back to Stalled.

Alexander Barkov added a comment - 2024-06-19 08:55 After implementing MDEV-26572 in 10.7, these changes are not needed any more.