Re^3: Vowel search

Replies are listed 'Best First'.
Re^4: Vowel search by AnomalousMonk (Archbishop) on Jun 11, 2014 at 22:47 UTC
There doesn't seem to be a significant difference between the `[AaEeIiOoUu]{2}` and `[AaEeIiOoUu][AaEeIiOoUu]` variations under Strawberry 5.14.4, but I was a bit surprised that there's so little improvement over the `/i` version. c:\@Work\Perl\monks>perl -wMstrict -le "use Benchmark qw(cmpthese); ;; print 'Perl version: ', $]; ;; my $s = 'Aid bears out '; $s = $s x 10_000_000; print 'length: ', length $s; ;; cmpthese(-1, { '/i' => sub { $s =~ m{ (?i) [aeiou]{2} }xmsg }, '[Aa]{2}' => sub { $s =~ m{ [AaEeIiOoUu]{2} }xmsg }, '[Aa][Aa]' => sub { $s =~ m{ [AaEeIiOoUu][AaEeIiOoUu] }xmsg }, }); " Perl version: 5.014004 length: 140000000 Rate /i [Aa]{2} [Aa][Aa] /i 3276565/s -- -8% -9% [Aa]{2} 3558515/s 9% -- -1% [Aa][Aa] 3600879/s 10% 1% -- [download] The results are closer to what you suggest under ActiveState 5.8.9, but with `/i` still surprisingly high. `c:\@Work\Perl>perl -wMstrict -le "(source code as above) " Perl version: 5.008009 length: 140000000 Rate [Aa]{2} /i [Aa][Aa] [Aa]{2} 3276565/s -- -6% -16% /i 3480139/s 6% -- -11% [Aa][Aa] 3918166/s 20% 13% --` [download] Still, as you say, it's a bit of a micro-optimization.	[reply] [d/l] [select]
Re^5: Vowel search (benchmarks--) by tye (Sage) on Jun 12, 2014 at 01:43 UTC
It is too bad that people posting "benchmark" code so rarely take the care to validate that the numbers that they are posting actually have any meaning behind them. For example, when you get a "Rate" like "3918166/s", then you should probably just throw away your benchmark results as having no practical meaning. In this particular case, this was a hint that your `$s = $s x 10_000_000;` line was completely pointless because of other aspects of your code. Fixing that problem (and reducing the string size significantly so it will finish before I need to go to bed), gives: Read more... (1182 Bytes) `Perl version: v5.12.0 Length of string: 140000 Rate [Aa]{2} lc [Aa][Aa] /i [Aa]{2} 50.7/s -- -0% -1% -3% lc 50.8/s 0% -- -1% -3% [Aa][Aa] 51.5/s 1% 1% -- -2% /i 52.5/s 4% 3% 2% --` [download] which shows how the performance differences are minimal (and the actual difference you'll see in the performance of a real script will be even smaller than that). So the "which will become noticable" assertion was wrong. Update: Here, here! Learning++, (Good attitude)++. - tye	[reply] [d/l] [select]
Re^6: Vowel search (benchmarks++ after all) by AnomalousMonk (Archbishop) on Jun 12, 2014 at 04:03 UTC
I was a bit reluctant to post Re^4: Vowel search because I wasn't sure it was a real contribution to the discussion. Now I'm happy I did because I learned something, even if only not to slap myself in the forehead quite so hard when I realize I've done something stupid. So ++benchmarks!	[reply]
Re^5: Vowel search by Jim (Curate) on Jun 11, 2014 at 23:13 UTC
Of course, another approach would be to common-case all lines before matching... Adding this other approach to your benchmark test… C:\>type 634253.pl #!perl use strict; use warnings; use Benchmark qw( cmpthese ); use English qw( -no_match_vars ); local $OUTPUT_RECORD_SEPARATOR = "\n"; print 'Perl version: ', $PERL_VERSION; my $s = 'Aid bears out ' x 10_000_000; print 'Length of string: ', length $s; cmpthese(-1, { 'lc' => sub { lc($s) =~ m{ [aeiou][aeiou] }xmsg }, '[Aa][Aa]' => sub { $s =~ m{ [AaEeIiOoUu][AaEeIiOoUu] }xmsg }, '[Aa]{2}' => sub { $s =~ m{ [AaEeIiOoUu]{2} }xmsg }, '/i' => sub { $s =~ m{ (?i) [aeiou]{2} }xmsg }, }); exit 0; C:\>perl 634253.pl Perl version: v5.16.2 Length of string: 140000000 Rate lc [Aa][Aa] /i [Aa]{2} lc 5.20/s -- -100% -100% -100% [Aa][Aa] 2915419/s 56073120% -- -23% -25% /i 3764119/s 72396452% 29% -- -3% [Aa]{2} 3877869/s 74584244% 33% 3% -- C:\>perl 634253.pl Perl version: v5.16.2 Length of string: 140000000 Rate lc [Aa]{2} /i [Aa][Aa] lc 4.94/s -- -100% -100% -100% [Aa]{2} 2814401/s 57019656% -- -8% -25% /i 3044909/s 61689750% 8% -- -19% [Aa][Aa] 3762832/s 76234868% 34% 24% -- C:\> [download]	[reply] [d/l]