Transform Sequence Problem

[wbirkett]

Thanks to all those who offered help! Your suggestions helped me find a better solution.

use B::Deparse;

# make deparsing object
$deparse = B::Deparse->new("-p", "-sC");

# forward encoding transforms
@fwd = (
    sub {$_[0]/100, ($_[1] + 128)/255, ($_[2] + 128)/255},
    sub {$_[0] * 256/25700, ($_[1] + 128) * 256/65535, ($_[2] + 128) *
+ 256/65535},
    sub {$_[0]/100, ($_[1] + 128) * 256/65535, ($_[2] + 128) * 256/655
+35},
    sub {@_},
    sub {$_[0] + 116 * $_[1]/500, $_[0], $_[0] - 116 * $_[2]/200},
    sub {$_[0]/100 + 116 * $_[1]/50000, $_[0]/100, $_[0]/100 - 116 * $
+_[2]/20000},
    sub {@_},
    sub {$_[0] * 0.4821, $_[1] * 0.5, $_[2] * 0.41245},
    sub {$_[0] * 0.9642, $_[1], $_[2] * 0.8249},
    sub {@_},
    sub {$_[0] * 96.42, $_[1] * 100, $_[2] * 82.49},
);

# reverse encoding transforms
@rev = (
    sub {$_[0] * 100, $_[1] * 255 - 128, $_[2] * 255 - 128},
    sub {$_[0] * 25700/256, $_[1] * 65535/256 - 128, $_[2] * 65535/256
+ - 128},
    sub {$_[0] * 100, $_[1] * 65535/256 - 128, $_[2] * 65535/256 - 128
+},
    sub {@_},
    sub {$_[1], 500 * ($_[0] - $_[1])/116, 200 * ($_[1] - $_[2])/116},
    sub {$_[1] * 100, 50000 * ($_[0] - $_[1])/116, 20000 * ($_[1] - $_
+[2])/116},
    sub {@_},
    sub {$_[0]/0.4821, $_[1]/0.5, $_[2]/0.41245},
    sub {$_[0]/0.9642, $_[1], $_[2]/0.8249},
    sub {@_},
    sub {$_[0]/96.42, $_[1]/100, $_[2]/82.49},
);

# CIELAB transforms
sub _x2L {($_[0] > 216/24389) ? 116 * $_[0]**(1/3) - 16 : $_[0] * 2438
+9/27}
sub _xyz2Lab {my $L = _x2L($_[1]); $L, (_x2L($_[0]) - $L) * 500/116, (
+$L - _x2L($_[2])) * 200/116}
sub _L2x {($_[0] > 8) ? (($_[0] + 16)/116)**3 : $_[0] * 27/24389}
sub _Lab2xyz {_L2x($_[0] + $_[1] * 116/500), _L2x($_[0]), _L2x($_[0] -
+ $_[2] * 116/200)}

# make example transform sequence array
@seq = ($rev[1], \&_Lab2xyz, $fwd[10]);

# make combined transform
# parameter: (array_of_code_references)
# returns: (code_reference)
sub combine {
    
    # copy transform array
    my @t = @_;
    
    # initialize expression
    my $expr = '@_';
    
    # for each transform step
    for my $i (0 .. $#t) {
        
        # add to expression
        $expr = "\$t[$i]->($expr)";
        
    }
    
    # return combined transform
    eval("return(sub {$expr})");
    
}

# combined transform (fast and flexible, but requires initializaton (a
+pprox. 75 μs))
$trans3 = combine(@seq);

# deparse the code reference
$body = $deparse->coderef2text($trans3);

print "\$trans3: $body\n";

# call the transform
@out = &$trans3(@in);

print "@out\n";

The 'combine' function uses 'eval' to create a clean and efficient transform from the component parts. The overhead of this 'combine' step is equal to about 10 transform calls, which is insignificant in my application. ##### original posting ##### I need an efficient way to process a sequence of transforms. For example, here are some individual transforms as code references:

# some code references
$f1 = sub {map {$_ + 1} @_};
$f2 = sub {map {log($_)} @_};
$f3 = sub {map {$_ * 3} @_};

They may be combined into a sequence

 # combined transform
sub trans1 {&$f3(&$f2(&$f1))};

# some data
@data = (1, 2, 3);

# call the transform
@trans = trans1(@data);

print "@trans\n";

The sequence could have fewer or more steps, so I would like to make an array of code references:

# make a sequence array
@seq = ($f1, $f2, $f3);

# combined transform
sub trans2 {
    
    # for each code reference
    for (@seq) {
        
        # transform data
        @_ = &$_;
        
    }
    
    # return
    return(@_);
    
}

# call the transform
@trans = trans2(@data);

print "@trans\n";

This works okay, but I wonder if there is a better and faster way.