Hmm... I think I made a little confusion so I put my finding here:
http://haisgwu.info/posts/2011-11-20-euler-problem-14.html

I do got stack overflow thus need several compile opts to fix it.
Not sure if it is what you mean by "You get overflow using 32-bit types here."

-Haisheng


On Sat, Nov 19, 2011 at 10:49 PM, Daniel Fischer <daniel.is.fischer@googlemail.com> wrote:
On Saturday 19 November 2011, 09:09:50, Haisheng Wu wrote:
> Hello,
>   I got great performance difference for the following code if I used
> type `Int` rather than `Data.Word.Word32`.
>   Anyone can help to explain why such difference?

Short answer: GHC optimises Int calculations far better than Word32
calculations, rather, it optimises more calculations for Int than for
Word32.  It also optimises more calculations for Word than for Word32, but
not as many as for Int.

The reason is that Int (and Word) are (at least expected to be) far more
often used, so the effort has gone to these types primarily. Work is being
done to get the fixed-width types on par, but it's not around the corner
yet.

You can try using Word instead of Word32, that's likely to be faster.

But

>   Thanks a lot.
>
> -Simon
>
> module Main where
> import Data.Word
>
> main :: IO ()
> main = print $ p14
>
> p14 = maximum [ (startChain n 0, n) | n <- [2..1000000] ]

You get overflow using 32-bit types here.

>
> startChain :: Word32 -> Int -> Int
> startChain 1 count    = count + 1
> startChain n count    = startChain (intTransform n) (count+1)
>
> intTransform :: Word32 -> Word32
> intTransform n
>
>   | even n         = n `div` 2
>   | otherwise      = 3 * n + 1