сравнение скорости регулярного выражения и поиска в хеше

[ocehb]

используем конструкцию (??{ code }); perldoc perlre

# for i in {1..10}; do print "Size: $[ 2**$i ]";
  perl -MBenchmark=timethese -e '$N = int ($ARGV[0]/2);
    @a = map { int rand(1024000) } 0..$ARGV[0];
    @hash{@a} = (1)x($#a+1);
    $re = join "|",keys %hash;
    timethese ($ARGV[1],
      { A => sub { return $a[$N] =~ m#^(\w+)(??{ exists $hash{$1} ? "" : "N/A" })$# },
        B => sub { return $a[$N] =~ m#^$re$# } } )' $[ 2**$i ] 1000000; done
Size: 2
Benchmark: timing 1000000 iterations of A, B...
         A:  3 wallclock secs ( 4.53 usr +  0.02 sys =  4.55 CPU) @ 219780.22/s (n=1000000)
         B:  0 wallclock secs ( 0.63 usr +  0.00 sys =  0.63 CPU) @ 1587301.59/s (n=1000000)
Size: 4
Benchmark: timing 1000000 iterations of A, B...
         A:  5 wallclock secs ( 4.54 usr +  0.00 sys =  4.54 CPU) @ 220264.32/s (n=1000000)
         B:  2 wallclock secs ( 1.10 usr +  0.00 sys =  1.10 CPU) @ 909090.91/s (n=1000000)
Size: 8
Benchmark: timing 1000000 iterations of A, B...
         A:  5 wallclock secs ( 4.49 usr +  0.02 sys =  4.51 CPU) @ 221729.49/s (n=1000000)
         B:  2 wallclock secs ( 1.16 usr +  0.02 sys =  1.18 CPU) @ 847457.63/s (n=1000000)
Size: 16
Benchmark: timing 1000000 iterations of A, B...
         A:  5 wallclock secs ( 4.86 usr +  0.01 sys =  4.87 CPU) @ 205338.81/s (n=1000000)
         B:  2 wallclock secs ( 1.24 usr +  0.00 sys =  1.24 CPU) @ 806451.61/s (n=1000000)
Size: 32
Benchmark: timing 1000000 iterations of A, B...
         A:  5 wallclock secs ( 4.53 usr +  0.01 sys =  4.54 CPU) @ 220264.32/s (n=1000000)
         B:  2 wallclock secs ( 1.44 usr +  0.00 sys =  1.44 CPU) @ 694444.44/s (n=1000000)
Size: 64
Benchmark: timing 1000000 iterations of A, B...
         A:  5 wallclock secs ( 4.53 usr + -0.01 sys =  4.52 CPU) @ 221238.94/s (n=1000000)
         B:  1 wallclock secs ( 1.76 usr +  0.01 sys =  1.77 CPU) @ 564971.75/s (n=1000000)
Size: 128
Benchmark: timing 1000000 iterations of A, B...
         A:  5 wallclock secs ( 4.45 usr +  0.03 sys =  4.48 CPU) @ 223214.29/s (n=1000000)
         B:  3 wallclock secs ( 2.45 usr +  0.00 sys =  2.45 CPU) @ 408163.27/s (n=1000000)
Size: 256
Benchmark: timing 1000000 iterations of A, B...
         A:  5 wallclock secs ( 4.54 usr +  0.01 sys =  4.55 CPU) @ 219780.22/s (n=1000000)
         B:  5 wallclock secs ( 3.88 usr +  0.02 sys =  3.90 CPU) @ 256410.26/s (n=1000000)
Size: 512
Benchmark: timing 1000000 iterations of A, B...
         A:  5 wallclock secs ( 4.50 usr +  0.02 sys =  4.52 CPU) @ 221238.94/s (n=1000000)
         B:  7 wallclock secs ( 6.64 usr +  0.01 sys =  6.65 CPU) @ 150375.94/s (n=1000000)
Size: 1024
Benchmark: timing 1000000 iterations of A, B...
         A:  5 wallclock secs ( 4.56 usr + -0.01 sys =  4.55 CPU) @ 219780.22/s (n=1000000)
         B: 13 wallclock secs (12.27 usr +  0.04 sys = 12.31 CPU) @ 81234.77/s (n=1000000)

резюме: на больших массивах имеет смысл использовать поиск через хеш.

update: при модификаторе /о регулярное выражение выигрывает у хеша, но до определенного размера.

Size: 16384
Benchmark: timing 1000000 iterations of A, B...
         A:  5 wallclock secs ( 4.58 usr + -0.01 sys =  4.57 CPU) @ 218818.38/s (n=1000000)
         B:  0 wallclock secs ( 0.90 usr +  0.02 sys =  0.92 CPU) @ 1086956.52/s (n=1000000)
Size: 32768
Benchmark: timing 1000000 iterations of A, B...
         A:  5 wallclock secs ( 4.55 usr +  0.00 sys =  4.55 CPU) @ 219780.22/s (n=1000000)
         B: 239 wallclock secs (237.26 usr +  0.37 sys = 237.63 CPU) @ 4208.22/s (n=1000000)