sqlite> create virtual table n using wholenumber; sqlite> .timer on sqlite> select sum(sqrt(value)) from n where value between 1 and 1000; 21097.4558874807 Run Time: real 0.001 user 0.000000 sys 0.000000 sqlite> select sum(sqrt(value)) from n where value between 1 and 1000000; 666667166.458841 Run Time: real 0.160 user 0.156250 sys 0.000000 sqlite> select sum(sqrt(value)) from n where value between 1 and 1000000000; 21081851083598.4 Run Time: real 151.021 user 151.031250 sys 0.000000 sqlite> select sum(value) from n where value between 1 and 1000000000; 500000000500000000 Run Time: real 89.341 user 89.343750 sys 0.000000
A native sqrt takes about 60 ns per operation. >-----Original Message----- >From: sqlite-users-boun...@sqlite.org [mailto:sqlite-users- >boun...@sqlite.org] On Behalf Of big stone >Sent: Sunday, 9 March, 2014 03:35 >To: sqlite-users@sqlite.org >Subject: Re: [sqlite] Suggestion to add "locate" as a broader version of >"instr" > >Hello Max, > >Your link is pretty interesting. It looks that : >- method1 should be easily implemented with SQLite "floating point" >representation, >- and with a very very small code size. > >Here is the benchmarking of the two available methods : > >https://raw.github.com/stonebig/ztest_donotuse/master/square_rooting_benc >hmark.GIF > > >So : >- your method is only 3 times slower than the python sqrt(), >- if SQLite team accepts to sacrifice a few bytes to implement sqrt(), we >may benefit a 389% speed-up (300/27*.652/1.86) at least. > (300/27*.652/1.86) > >sqrt() is very interesting for statistics on-the-go over sql datas. > >Regards, >_______________________________________________ >sqlite-users mailing list >sqlite-users@sqlite.org >http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users _______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
