bench_lines_in_scalar.pl

#!/usr/bin/perl
use Benchmark qw(cmpthese);
use Config;
use Inline C => Config =>
#    BUILD_NOISY => 1,
#    FORCE_BUILD => 1,
    OPTIMIZE => " -O3 -march=native";

use Inline C => <<'END_OF_C_CODE';
SV* lines_in_string(SV* sv) {
    unsigned i, lines = 0;
    STRLEN len;
    char *ptr;
    ptr = SvPV(sv, len);
    for(i = 0; i + 8 <= len; i += 8) {
        uint64_t v = *(uint64_t*)(ptr + i) ^ 0x0a0a0a0a0a0a0a0aUL;
        lines += !(v & 0xFF00000000000000UL); 
        lines += !(v & 0xFF000000000000UL); 
        lines += !(v & 0xFF0000000000UL); 
        lines += !(v & 0xFF00000000UL); 
        lines += !(v & 0xFF000000U); 
        lines += !(v & 0xFF0000U); 
        lines += !(v & 0xFF00U); 
        lines += !(v & 0xffU); 
    }
    for(; i < len; ++i) {
        if(ptr[i] == '\n') ++lines;
    }
    if(len && ptr[len - 1] != '\n') ++lines;
    return newSVuv(lines);
}
 
END_OF_C_CODE

my $str = "ab\nc\ndefg\n" x 1048576;
my $lines;
cmpthese - 1, {
    m => sub {$lines = () = $str =~ m/\n/g},
    tr           => sub {$lines = ($str =~ tr/\n//)},
    while_dollar => sub {$lines = 0; $lines++ while $str =~ m/$/gsm},
    while_n      => sub {$lines = 0; $lines++ while $str =~ m/\n/gs},
    inlinec      => sub {$lines = lines_in_string($str)}
};

print lines_in_string($str) . "\n";
print scalar($str =~ tr/\n//) . "\n";

__END__
               Rate while_dollar           m     while_n          tr     inlinec
while_dollar 1.92/s           --        -47%        -61%        -99%       -100%
m            3.64/s          89%          --        -27%        -98%       -100%
while_n      4.95/s         157%         36%          --        -98%        -99%
tr            203/s       10442%       5475%       3995%          --        -76%
inlinec       852/s       44224%      23340%      17118%        320%          --