On Sun, Oct 03, 2004 at 04:03:55PM +0200, Carsten Schultz wrote:
Hi Tomasz!
Hi Carsten!
To my surprise, it was much slower and made many allocations: [... Then I noticed the cause: GHC.Prim.<# returns a boxed, heap allocated Bool, and so do other primitive comparison operators.
I should have asked one fundamental question first: am I right? ;)
That's not really the cause. A function returning a boxed value does not necessarily have to allocate it, it is just a vectored return afaik.
I haven't heard about 'vectored return' before. I will try to find more information about it. Maybe you can recommend something for me to read?
(Notice that $wfor again take three arguments, the last one being the state.)
Hmmm, I noticed that the number of arguments differed, there were even some quiet alarm bells in my head, but I ignored it.
for4 :: Int -> IO () -> IO () for4 n x = if n `gt` 0 == 0 then return () else x >> (for4 (n-1) x)
gt :: Int -> Int -> Int gt x y = if x > y then 1 else 0
If you test it, it should be fast.
It is even slightly faster than my fastest version :)
BTW, although counting upwards (and not solving the problem generally), the following is ok too:
for2 :: Int -> IO () -> IO () for2 n x = sequence_ [x | i <- [1..n]]
This one is amazing. It's 3 times faster than the previous one in spite of being written in high level style. I guess it's worth checking idiomatic Haskell style first, because there is a big chance that GHC was optimised for it :) However, it would be nice if all versions were as efficient...
Playing with the code generated by ghc is a great way to waste time for me.
Well, but you seem to be very good at it. Maybe it won't be such a waste of time in the long term :)
Wait until you have found the RULES-pragma :-)
I've already found it some time ago. I even tried to use them to optimise vector/matrix expressions (to eliminate intermediate vectors), but I remember that sometimes the rules didn't fire and I didn't understand why.
Have fun,
Carsten
Hope this will teach me to avoid premature conclusions :-/ Thanks, Tom -- .signature: Too many levels of symbolic links