my ($largest, $smallest) = (-9e9,9e9);                          #   in
+itialize to the wrong extremes
    ...
        for my $v (@array) {                                    #   [a
+nonymous monk]'s [id://1203660]: single pass through loop, without so
+rting; more efficient than brostad's single sort
            $largest = $v   if $v > $largest;
            $smallest = $v  if $v < $smallest;
        }
[download]

The only quarrel I have with this implementation is that it depends on assumptions about the smallest and largest representable numbers (the wrongest extremes) in the system. Even if the assumptions are true in a given system, all bets are off if you move, e.g., to a different platform: from a 32-bit float to a 64-, 80- or who-knows-how-many-bit float. And if they're not true:

c:\@Work\Perl\monks>perl -wMstrict -le
"my @array = (-9e9-123, -9e9-234);
 ;;
 my ($largest, $smallest) = (-9e9, 9e9);
 ;;
 for my $elem (@array) {
   $largest  = $elem if $elem > $largest;
   $smallest = $elem if $elem < $smallest;
   }
 print qq{smallest: $smallest;  largest: $largest};
"
smallest: -9000000234;  largest: -9000000000
[download]

Taking the initial smallest/largest value from the array itself is bulletproof: either the initializer is already the smallest/largest value, or some other value will be found in the array that is smaller/larger.

c:\@Work\Perl\monks>perl -wMstrict -le
"my @array = (-9e9-123, -9e9-234);
 ;;
 my ($largest, $smallest) = ($array[0], $array[0]);
 ;;
 for my $elem (@array) {
   $largest  = $elem if $elem > $largest;
   $smallest = $elem if $elem < $smallest;
   }
 print qq{smallest: $smallest;  largest: $largest};
"
smallest: -9000000234;  largest: -9000000123
[download]