3

I'm coding a classical cipher by substitution in C++, using all the printable characters in ASCII, and I'm wondering which is faster? A search in an array (edit: a non associative one, just something like letters[] = {'a', 'b', ...); (linear or binary) or a switch statement? The compiler can optimize the switch, doesn't it?. Maybe the difference is the memory usage? My choice is the switch, although the code is bigger, but maybe I'm missing something.

(Perhaps this question seems subjective, but I think would be objective reasons to choose one or another way. And sorry for my English).

Tae
  • 1,665
  • 5
  • 24
  • 45
  • 1
    Why do you need to search an array? There aren't many ASCII characters. Store the substitutions in one table and a lookup consists of a single array lookup. – sigfpe Jun 04 '11 at 14:34
  • @user207442 I'm not using associative arrays, because we don't study them in class yet. – Tae Jun 04 '11 at 20:43

3 Answers3

6

Why do you need a search? Just have a 128-entry array indexed by your ASCII character. That's basically what the compiler does with your switch. (You could subtract 32 and use a 96 entry array, thus saving the space used by the non-printables.)

andrewdski
  • 5,255
  • 2
  • 20
  • 20
  • 2
    Have you ever looked at assembly generated for switch statement? Compilers don't use arrays to implement switch. They use offset jumps. –  Jun 04 '11 at 14:37
  • 1
    @Vlad: GCC also sometimes uses a binary tree of branches. It depends on the switch values. – Peter Alexander Jun 04 '11 at 14:43
  • 2
    @Vlad: They use jump tables, which are arrays of jump targets, effectively. – Puppy Jun 04 '11 at 15:31
  • I'm thinking in a search because we don't study associative arrays in class yet. – Tae Jun 04 '11 at 20:36
2

Which one is faster depends on a lot of factors: your compiler's optimization (you'd have to look at the assembly to know what it does) and how you'd set your array up.

If performance is really important, and as both solutions are likely to be fairly simple to implement, I'd implement them both and benchmark them in an as-real-as-possible setup to find out which one is faster.

rlc
  • 2,808
  • 18
  • 23
1

There's certainly a chance that a sufficiently smart compiler could optimize the switch to be a lookup, which would be faster than a binary search. But you could do that optimization yourself and get short code:

char alphabet[] = {
    'Z', 'E', 'B', 'R', 'A', 'S', 'C', 'D', 'F', 'G', 'H', 'I', 'J',
    'K', 'L', 'M', 'N', 'O', 'P', 'Q', 'T', 'U', 'V', 'W', 'X', 'Y'
};
// now you can get the ciphertext for a single uppercase character with:
alphabet[ch - 'A'];
John Calsbeek
  • 35,947
  • 7
  • 94
  • 101
  • I get lost in the alphabet[ch - 'A'] part; Could you help me with a little bit of context? – Tae Jun 04 '11 at 20:03
  • Yeah. It's the same basic idea as having an array indexed by the ASCII code of your character. We know that the ASCII codes for A through Z are sequential, and obviously `'A' - 'A'` is 0. So `'B' - 'A'` is 1, all the way through `'Z' - 'A'` being 25. So it's a way to index an array that just contains capital letters. – John Calsbeek Jun 04 '11 at 20:05
  • Oh, right. I forgot mention that we don't study associative arrays in class yet. – Tae Jun 04 '11 at 20:34
  • 1
    This isn't technically an associative array. It's just an array, indexed 0 through 25, which correspond to A through Z. – John Calsbeek Jun 04 '11 at 20:40