Am 21.09.2010 11:08, schrieb John Lato:
Which compiler/version are you using for your tests? Is the behavior the same for other compilers/versions? In particular, is it similar for ghc-6.10, ghc-6.12, and ghc-HEAD?
I've used ghc-6.12.3. My top-level file is Bench.hs: -- import Criterion.Main import Interspersing testCase f = print . length . f ',' main = do let s = replicate 10000000 'A' print $ length s defaultMain [ bench "isgo" $ testCase isgo s , bench "isrec" $ testCase isrec s ] -- (using Daniel's Interspersing module) I compiled by: ghc -O2 -fforce-recomp -fext-core --make Bench.hs and ran: ./Bench -s 10
When benchmarking variants like this, I'm always suspicious that my optimizations may be ghc-version-specific, since I've had occasional experience to the contrary.
Yes, and after years, you'll sometimes find out that your pragmas, bangs and other quirks actually made things worse. Cheers Christian