first off: always use strictures (use strict; use warnings;).

my $out = "/home/users/user/tmp/$progname_$today.dat";
[download]

should be

my $out = "/home/users/user/tmp/${progname}_$today.dat";
[download]

or $progname_ will be seen as the variable name rather than $progname. Strictures should also force you to think a little more about lifetime of variables and how you pass information around to different parts of your program.

call_dir($p=1)
...
sub call_dir {
    if ($p == 1)
...
<p>would be better as:</p>
<c>
call_dir(1)
...
sub call_dir {
    my ($p) = @_;

    if ($p == 1)
[download]

although passing an array into the find call would be even better:

my $main = "/home/users/user";
my @rootDirs = ("$main/tmp/", "$main/exe");
...
find ({wanted => \&data_for_path, follow=>1, follow_skip=>2}, @rootDir
+s);
[download]

and avoids the interesting mismatch between the calls to call_dir and the implementation (3 calls, 2 valid return values).

However, I'd be inclined to read and parse today and yesterday's files in parallel to detect differences. That avoids having any more than a few lines of data in memory at any given time, but may make the parsing a little more interesting.

If instead you read the files in parallel as suggested above, but load a complete directory of files at a time into two arrays of lines (one for today's files and one for yesterday's) you could then use Algorithm::Diff to do the heavy lifting in differencing the two file sets. That limits the data in memory to one directory worth of files (x 2 - one image for each day), but probably simplifies the diff parsing substantially.