#!/usr/bin/perl -w

use strict;


# Check if "pack" gives the required IEEE 64bit float
#
my $little_endian;

# Some test data
my $teststr = pack "d", 1.2345;
my @hexdata =(0x8D, 0x97, 0x6E, 0x12, 0x83, 0xC0, 0xF3, 0x3F);
my $number  = pack "C8", @hexdata;


if ($number eq $teststr) {
    # Little Endian
    $little_endian = 1;
}
elsif ($number eq reverse($teststr)) {
    # Big Endian
    $little_endian = 0;
}
else {
    # Give up, gracefully or otherwise.
    die "Required floating point format not supported on this platform
+.";
}


# Now use this information to pack a double
#
my $double = 3.1415;
my $packed_double;

if ($little_endian) {
    $packed_double = pack "d", $double;
}
else {
    $packed_double = reverse pack "d", $double;
}
[download]

#!/usr/bin/perl -w

use strict;
use Config;

my $double = 3.1415;
my $packed_double;

if ($Config{byteorder} == 1234) {
    $packed_double = pack "d", $double;
}
else {
    $packed_double = reverse pack "d", $double;
}
[download]

For portability, you can just skip pack and store your floating-point numbers as strings. With no extra work, you risk losing only about 3 bits of accuracy in the process (and you can lose that much with just a few arithmatic calculations so this might not be a problem at all).

If those three bits are important, then you can write your floating-point numbers via sprintf "%.20e",$num and read them just the normal Perl way and you won't lose any bits of accuracy between platforms with the same accuracy of floating-point numbers (the best you can do). ["%.20e" should be exact for up-to-64-bit mantissas, which should be overkill for most platforms. "%.16e" is probably all you really need.]