type vs data performance in haskell

Question

I found answers explaining difference between newtype and data in Haskell. But if I have the following type synonym:

type Point = (Int,Int)

Would that be more efficient rather to use:

data Point = Pt (Int,Int) ?

Answer 1

Using type will be more efficient, as it incurs one less indirection than the data version.

Note that both are more inefficient than:

data Point = Point {-# UNPACK #-}!Int {-# UNPACK #-}!Int

as you can see from this earlier question on data representations.

Answer 2

Yes.

The Pt construction adds one word of overhead (in GHC) and the field (i.e. the pair) is stored as a pointer to a pair, adding one additional word, for a total of two words overhead (and an extra indirection to get to the values).

I recommend that you either use the type synonym or, better yet, define

data Point = Pt {-# UNPACK #-} !Int {-# UNPACK #-} !Int

This requires 4 words less than

type Point = (Int, Int)

and uses one less level of indirections (pointers).