Re^2: Why this code is so slow if run in thread?

But then Perl's Core "Encode" is not thread safe? :(

Suppose I split my program in two. First one creates 2 files, with single-byte encoded data and double-byte encoded data (and says "Equal", of course -- this image has less than 255 "objects").

use strict;
use warnings;
use 5.020;

use PDL;
use PDL::IO::Image;
use PDL::Image2D;
use PDL::IO::FastRaw;
use Encode qw/ decode /;

my $img  = PDL::IO::Image-> new_from_file( 'test.png' );
my $pdl  = $img-> pixels_to_pdl-> short;

my $s1   = cc8compt( $pdl == 0 )-> byte;
my $s2   = cc8compt( $pdl == 0 );

say 'Equal' if decode( 'UTF16LE', ${ $s2-> get_dataref }) eq
                                  ${ $s1-> get_dataref };

writefraw( $s1, "fname-1" );
writefraw( $s2, "fname-2" );
[download]

The 2nd program doesn't use PDL modules, it reads data from disk (and uses some hard-coded numbers):

use strict;
use warnings;
use 5.020;

use threads;

use Encode qw/ decode /;

say "No thread (byte)\n---------";
say 'Count: ', scalar @{ test( 1 )};

say "\nThread (byte)\n---------";
say 'Count: ', scalar @{ threads-> create( \&test, 1 )-> join };

say "\nNo thread (short)\n---------";
say 'Count: ', scalar @{ test( 2 )};

say "\nThread (short)\n---------";
say 'Count: ', scalar @{ threads-> create( \&test, 2 )-> join };

sub test {
    my $arg = shift;
    
    open my $fh, '<', "fname-$arg";
    binmode $fh;
    my $str = do { local $/; <$fh> };
    close $fh;
    $str = decode( 'UTF16LE', $str ) if $arg == 2;

    my ( $w, $h ) = ( 7616, 1200 );

    my @b = map {
        [ [ $w, 0 ], [ $h, 0 ] ]
    } 0 .. 145;

    my $t = time;
    
    for my $y ( 0 .. $h - 1 ) {
        my $s = substr( $str, $y * $w, $w );
        while( $s =~ m[[^\0]+]g ) {
            my $c = ord( $& );
            $b[ $c ][ 0 ][ 0 ] = $-[0]     if $-[0]     < $b[ $c ][ 0 
+][ 0 ];
            $b[ $c ][ 0 ][ 1 ] = $+[0] - 1 if $+[0] - 1 > $b[ $c ][ 0 
+][ 1 ];
            $b[ $c ][ 1 ][ 0 ] = $y        if $y        < $b[ $c ][ 1 
+][ 0 ];
            $b[ $c ][ 1 ][ 1 ] = $y        if $y        > $b[ $c ][ 1 
+][ 1 ];
        }
    }
    say 'Time: ', time - $t;

    shift @b;
    return \@b;
}
[download]

The output:

No thread (byte)
---------
Time: 0
Count: 145

Thread (byte)
---------
Time: 1
Count: 145

No thread (short)
---------
Time: 1
Count: 145

Thread (short)
---------
Time: 51
Count: 145
[download]

Comment on Re^2: Why this code is so slow if run in thread? Select or Download Code

Replies are listed 'Best First'.
Re^3: Why this code is so slow if run in thread? (create) by tye (Sage) on Dec 11, 2016 at 19:24 UTC
With Perl threads, the thing that is usually the slowest is creating a thread. This is because creating a Perl thread requires cloning every data structure you have created. So I would first determine if that is where the time is being spent. You can report that by comparing the time at the start of the routine running against the time just before you ask for a thread to be created. If that is indeed where most of the time is spent, then you should do what just about anybody who has gotten good at using Perl threads does: Create your threads very early and just ship work to them, usually via a thread queue such as Thread::Queue. Though, looking at your code, I don't see where you would have a large data structure that should not have been destroyed before you create the thread. But, that data could be something cached (perhaps by accident) in some module you have pulled in, for example. Update: Or it could be time spent destroying the second interpreter instance. - tye	[reply]
Re^3: Why this code is so slow if run in thread? by ikegami (Patriarch) on Dec 11, 2016 at 19:14 UTC
Did you look at your results? They're mostly useless! We can tell that shorts are 50x slower with threads, but we have no idea how much slower bytes are with threads. Could be there not slower at all (0.9999s vs 1.0000s). Could be billions time slower (0.000000001 vs 1.000000000). Please re-run the tests with `use Time::HiRes qw( sleep );`. Or provide the data files.	[reply] [d/l]


good chemistry is complicated, and a little bit messy -LW
	PerlMonks