Parrot vs. C cagematch (Just for fun)

[zentara]

I was looking at the latest parrot and saw in the examples, a mandelbrot generator, with the equivalent c source code. So I modified them slightly, and ran them thru 1000 generations. Parrot is fast, but it took 5 times the cpu, and roughly 3 times as long on my machine(Duron 800). It would be interesting to compare it to equivalent perl5 code, and if anyone can convert it, please post it. I'm almost there, but the C for statements are a little complicated for me to convert. :-)

Anyways, here is the code for you to try. I slightly modified the original code, to add timer variables.

#################################################
# mandel.pasm ...run with "parrot mandel.pasm"

        time N10
        set I0, 0

loop:
        set S1, " .:,;!/>)|&IH%*#"
        set I2, 30



YREDO:  #  C = y*0.1 - 1.5
        set N8, I2
        set N9, 0.1
        mul N7, N8, N9
        set N8, 1.5
        sub N7, N7, N8

        set I1, 0

XREDO:  # c = x*0.04 - 2
        set N8, I1
        set N9, 0.04
        mul N6, N8, N9
        set N8, 2.0
        sub N6, N6, N8
        set N3, 0
        set N4, 0

        set N1, N6
        set N2, N7

        set I3, 0

KREDO:  # t = z*z - Z*Z + r
        mul N8, N3, N3
        mul N9, N4, N4
        sub N5, N8, N9
        add N5, N5, N1

        # Z = 2*z*Z + i
        add N4, N4, N4
        mul N4, N4, N3
        add N4, N4, N2

        # z = t
        set N3, N5

        # if (z*z + Z*Z > 10) break;
        mul N8, N3, N3
        mul N9, N4, N4
        add N8, N8, N9
        gt N8, 10.0, PRINT

        inc I3
        lt I3, 112, KREDO

PRINT:  mod I4, I3, 16
        substr S2, S1, I4, 1
        print S2

        inc I1
        lt I1, 75, XREDO

        print "\n"
        dec I2
        gt I2, 0, YREDO


inc I0
lt I0, 1000, loop

print I0
print " count in "
time N11
sub N11,N10
print N11
print " seconds\n"

END:    end
############################################

And the c source

/* mandel.c  ... compile with gcc -o mandel mandel.c */
#include <stdio.h> 
#include <time.h>  

int main() {
  int x, y, k, counter;
  char *b = " .:,;!/>)|&IH%*#";
  float r, i, z, Z, t, c, C;
  time_t begin, end;
  counter = 0;

begin = time (NULL);

while(counter < 1000){
    for (y=30; puts(""), C = y*0.1 - 1.5, y--;) {
     for (x=0; c = x*0.04 - 2, z=0, Z=0, x++ < 75;) {
      for (r=c, i=C, k=0; t = z*z - Z*Z + r, Z = 2*z*Z + i, z=t, k<112
+; k++)
       if (z*z + Z*Z > 10) break;
        printf ("%c", b[k%16]);
      }
    }
counter++;
}

end = time (NULL);

printf ("%d count in %ld secs \n", counter, end - begin);

}

[tilly]

Here is the C translated into Perl in a straightforward way:

use strict;
my ($x, $y, $k, $counter);
my @b = split //, " .:,;!/>)|&IH%*#";
my ($r, $i, $z, $Z, $t, $c, $C);
my ($begin, $end);
my $counter = 0;

$begin = time;

while($counter < 1000){
    for ($y=30; print("\n"), $C = $y*0.1 - 1.5, $y--;) {
     for ($x=0; $c = $x*0.04 - 2, $z=0, $Z=0, $x++ < 75;) {
      for ($r=$c, $i=$C, $k=0; $t = $z*$z - $Z*$Z + $r, $Z = 2*$z*$Z +
+ $i, $z=$t
, $k<112; $k++) {
        last if ($z*$z + $Z*$Z > 10);
      }
      printf ($b[$k%16]);
    }
  }
  $counter++;
}

$end = time;

printf ("%d count in %ld secs \n", $counter, $end - $begin);

Note that this is going to be much slower than both Parrot and C. Without knowing much about Parrot, I'd be surprised if Perl 6 actually gets compiled to Parrot that is as efficient as the handrolled code above.

[zentara]

Note that this is going to be much slower than both Parrot and C. .. I'd be surprised if Perl 6 actually gets compiled to Parrot that is as efficient as the handrolled code above.

Yes, on my machine, the perl5 version used double the parrot cpu, and near 75% longer. Maybe Perl6 will do a bit better. It does look like Inline::Parrot could replace Inline::C in some circumstances. Parrot looks easy enough to understand.

[stvn]

Actually, there is a Perl6/Pugs version of mandel here. And Autrijus actually used this as his basis for the early Pugs->Parrot compiler. You can read more about it here.

[chromatic]

How did you compile Parrot and the C program? By default, Parrot builds without optimizations, and I believe the JIT is off by default too. Both of those will improve its performance.

[Joost]

#!/usr/local/bin/perl -w
use strict;

my ($x, $y, $k, $counter);
my @b = split //, " .:,;!/>)|&IH%*#";

my ($r, $i, $z, $Z, $t, $c, $C);
my ($begin, $end);
$counter = 0;

$begin = time;

while ($counter < 1000) {
    for ($y = 30 ; $C = $y * 0.1 - 1.5, $y--;) {
        print "\n";
        for ($x = 0 ; $c = $x * 0.04 - 2, $z = 0, $Z = 0, $x++ < 75 ;)
+ {
            for (
                    $r = $c, $i = $C, $k = 0 ;
                    $t = $z * $z - $Z * $Z + $r, $Z = 2 * $z * $Z + $i
+, $z = $t,
                    $k < 112 ;
                    $k++
                )
            {
                last if ($z * $z + $Z * $Z > 10);
            }
            print $b[ $k % 16 ];
        }
    }
    $counter++;
}

$end = time;

printf("%d count in %ld secs \n", $counter, $end - $begin);

updated: misplaced the counter++ statement.

[gellyfish]

This isn't actually equivalent but this is the mandelbrot generator I include as an example in Linux::Svgalib:

#!/usr/bin/perl -w
 
use strict;
use Linux::Svgalib;
 
use vars qw($P @Q $col $row @colours );
 
my (
    $max_iterations,
    $max_size )       = (512,4);
 
my $screen = Linux::Svgalib->new();
 
$screen->init();
$screen->setmode(8);
 
my ($XMax,$XMin,$YMax,$YMin) = (1.2,-2.0,1.2,-1.2);
 
my $maxcol = $screen->getxdim();
my $maxrow = $screen->getydim();
my $max_colours = $screen->getcolors();
 
my $deltaP = ($XMax - $XMin) / $maxcol;
my $deltaQ = ($YMax - $YMin) / $maxrow;
 
$Q[0] = $YMax;
 
 
$screen->clear();
 
for( 1 .. $maxrow )
{
    $Q[$_] = $Q[$_ -1] - $deltaQ;
}
 
$P = $XMin;
 
for $col ( 0 .. $maxcol -1 )
{
    for $row ( 0 .. $maxrow - 1)
    {
        my ($X,$Y,$XSquare,$YSquare) = (0,0,0,0);
 
        my $colour = 1;
 
        while (( $colour < $max_iterations ) &&
               (($XSquare + $YSquare ) < $max_size ))
        {
             $XSquare = $X * $X;
             $YSquare = $Y * $Y;
 
             $Y *= $X;
             $Y += $Y + $Q[$row];
             $X = $XSquare - $YSquare + $P;
             $colour++;
        }
           $screen->setcolor($colour % $max_colours);
           $screen->drawpixel($col,$row);
    }
    $P += $deltaP;
}
 
$screen->getch();
 
$screen->setmode(TEXT);

This is quite fast, especially when the exact same algorithm in C used to take nearly a day to render on an Amiga 500

/J\

[tall_man]

Perl isn't great at this sort of number-crunching application, but you can use C extension modules like PDL to improve performance. There is an example of a fast PDL Mandelbrot here.

[etj]

The latter link bitrotted, but is archived at https://web.archive.org/web/20080223091059/https://www.perl.org/tpc/1998/User_Applications/PDL/node11.html.

See also Re^2: [OT] Merry Christmas and gift of love; demo 3d in perldl also has Mandelbrot examples.

[spurperl]

Personally I find it a little sad that Parrot is so much slower than C.

Thinking on a conceptual level, Parrot shouldn't be much more difficult to translate to machine language efficiently than C, hence I expect it to be at least comparable.

Granted, C compilers lived through 30 years of optimizations and there probably isn't such a thing in existence for Parrot, yet.

[Limbic~Region]

spurperl,
I freely admit it has been more than half a year since I followed Parrot development. I can tell you that there is a constant struggle being waged between finishing specs, getting it to work, and speeding it up. More often then not, the optimizations take a back seat. I would like to think the finished product with optimizations and JIT are more closely comparable to C.