Re^2: Speed/Efficiency tweaks for a fannkuch benchmark script?

in reply to Re: Speed/Efficiency tweaks for a fannkuch benchmark script?
in thread Speed/Efficiency tweaks for a fannkuch benchmark script?

I noticed that your fast permutation algorithm passes an arrayref, but then copies the values out. There's no particular savings there compared to passing the array. That and a few other micro-optimizations squeezed 210+% more speed out (if you don't print — the savings are probably less significant with printing turned on.) Code (update: fixed, reducing savings, as expected) follows.

{
  my $level = 0;
  sub permute_new {
    local @_ = @_;
    my $index = $level;
    my $printing = $level == $#_;
    if ($printing) {
      do {
        #print "@_\n";
        @_[$index-1, $index] = @_[$index, $index-1]
            if $index != 0;
      } while $index-- > 0;
    }
    else {
      do {
        ++$level;
        &permute_new;
        --$level;
        @_[$index-1, $index] = @_[$index, $index-1]
            if $index != 0;
      } while $index-- > 0;
    }
  }
}

sub permute_orig {
    my ($aref, $level) = (@_, 0);
    my ($index, $copy, $printing) = ($level, [@$aref], $level+1 == @$a
+ref);
    do {
        if ($printing) {
            #print "@$copy\n";
        } else {
            permute_orig($copy, 1+$level);
        }
        @$copy[$index-1, $index] = @$copy[$index, $index-1]
            if $index != 0;

    } while $index-- > 0;
}

my $to = shift;

use Time::HiRes qw( gettimeofday tv_interval );
my @start_time = gettimeofday();

permute_orig([1..$to]);
print "\n";

my @end_time = gettimeofday();
print tv_interval ( \@start_time, \@end_time )," elapsed seconds for o
+riginal.\n\n";

@start_time = gettimeofday();

permute_new(1..$to);

@end_time = gettimeofday();
print tv_interval ( \@start_time, \@end_time )," elapsed seconds for n
+ew.\n\n";
[download]

Caution: Contents may have been coded under pressure.

In Section Seekers of Perl Wisdom