On Fri, Jul 9, 2010 at 11:15 AM, Patrick LeBoutillier
Now I find my code is much clearer (it almost writes itself...) and practically absent of low-level bit operations (except, for example, converting a Word into an actual Haskell Int32 to perform the actual arithmetic). However, I'm wondering about the cost of all the "packing/unpacking" via pattern-matching and/or accessors. Perhaps I should not worry about this and optimize later if required? Is this type of premature optimization (assuming the previous version was faster...) really the root of all evil?
Yes, you should test before worrying too much about it. There's only one bit you should do regardless: use strict and unpacked fields. For example: data Word = Word !Sign {-# UNPACK #-} !Byte {-# UNPACK #-} !Byte {-# UNPACK #-} !Byte {-# UNPACK #-} !Byte This should give you boost. I didn't give an UNPACK pragma to Sign because AFAIK it won't unpacked anyway. Of course there's a performance penalty, but if it matters or not you'll find only by testing. One possible solution is to never use pattern matching, only accessors. That way the code will work for any representation you choose. In fact, all bit-fiddling will be contained in your accessors. Cheers! =) -- Felipe.