Re: [Haskell-cafe] estimating the speed of operation
Is this benchmark ment to measure something in particular? Did you not copy and paste it? It doesn't even compile: > parse error on input `module' > not in scope: `pack' > Couldn't match expected type `(Integer, Integer)' against inferred type `IO > Integer' > Couldn't match expected type `(Integer, Integer)' against inferred type `IO > (Integer, Integer)' etc etc. Also, you don't have a 'main' with any print statements which makes me think you are interpreting it - a poor way to benchmark. It should be compiled with -O2. After fixing up these issue, and adding a main: > ghc md5.hs -O2 --make ; ./md5 [[(0,16),(0,5)],[(0,10),(0,11)],[(0,78),(0,77)],[(0,770),(0,746)]] And the code is: > import qualified Data.ByteString.Lazy.Char8 as L8 > import Data.Digest.Pure.MD5 > import System.IO.Unsafe > import System.Time > import System.Random > import Data.Char > > clockTime2Tuple (TOD sec pic) = (sec, pic) > > getClockTimeTuple :: IO (Integer, Integer) > getClockTimeTuple = getClockTime >>= return . clockTime2Tuple > > subTimeTuples (s2,ps2) (s1,ps1) = > let dps = ps2-ps1 > (s,ps) = (s2 -s1 - (if dps < 0 then 1 else 0), if dps<0 then > dps+(10^12) else dps) > in (s, ps `div` (10^6)) > > test_md5 :: [L8.ByteString] -> Int -> IO [(Integer,Integer)] > test_md5 test_input n = flip mapM test_input >( \ input_row -> do >t1 <- getClockTimeTuple >sequence_ (replicate n (return $ md5 input_row)) >t2 <- getClockTimeTuple >dt <- return $ subTimeTuples t2 t1 >return dt >) > > test_list = [ L8.pack "Hello world!"] ++ [L8.replicate 100 ':'] > > main = do > x <- mapM (test_md5 test_list) [1000, 1, 10, 100] > print x Could someone with lots of spare time please write an e-mail filter that attempts to compile alledged code segments and cans the e-mail if it fails? Tom On Wed, Feb 18, 2009 at 5:21 AM, Belka wrote: > > Thanks alot, Bulat! > > New results are much better: > --- > test_list = > [L8.pack "Hello world!"] ++ > [L8.replicate 100 ':'] > --- > The results are (iterations_count, microseconds): > (1000, [300 +/- 200 , 18400 +/- 100]) > (1, [1030 +/- 10 , 19950 +/- 50 ]) > (10, [9100 +/- 100 , 55000 +/- 15000]) > (100, [89850 +/- 500 ,185000 +/- 500 ]) > > I also played a bit with bang patterns, to make input stricter, but this > gave the same result. > --- > For the first test string time grow almost linearly, what can't be said > about the second string. I still wonder if (and how) GHC optimizes the > process. > > Belka > -- > View this message in context: > http://www.nabble.com/estimating-the-speed-of-operation-tp22075843p22078560.html > Sent from the Haskell - Haskell-Cafe mailing list archive at Nabble.com. > > ___ > Haskell-Cafe mailing list > Haskell-Cafe@haskell.org > http://www.haskell.org/mailman/listinfo/haskell-cafe > ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
Re: [Haskell-cafe] estimating the speed of operation
Thanks alot, Bulat! New results are much better: --- test_list = [L8.pack "Hello world!"] ++ [L8.replicate 100 ':'] --- The results are (iterations_count, microseconds): (1000, [300 +/- 200 , 18400 +/- 100]) (1, [1030 +/- 10 , 19950 +/- 50 ]) (10, [9100 +/- 100 , 55000 +/- 15000]) (100, [89850 +/- 500 ,185000 +/- 500 ]) I also played a bit with bang patterns, to make input stricter, but this gave the same result. --- For the first test string time grow almost linearly, what can't be said about the second string. I still wonder if (and how) GHC optimizes the process. Belka -- View this message in context: http://www.nabble.com/estimating-the-speed-of-operation-tp22075843p22078560.html Sent from the Haskell - Haskell-Cafe mailing list archive at Nabble.com. ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
Re: [Haskell-cafe] estimating the speed of operation
Hello Belka, Wednesday, February 18, 2009, 1:15:32 PM, you wrote: > sequence_ (replicate n (return $ md5 input_row)) $! -- Best regards, Bulatmailto:bulat.zigans...@gmail.com ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
[Haskell-cafe] estimating the speed of operation
Hello, communion people! I have a little problem and ask for an advice. I'm trying to estimate the performance speed of one pure function, but get some strange results. - import qualified module Data.ByteString.Lazy.Char8 as L8 import module Data.Digest.Pure.MD5 import module System.IO.Unsafe import module System.Time import module System.Random import module Data.Char clockTime2Tuple (TOD sec pic) = (sec, pic) getClockTimeTuple = getClockTime >>= (\ x -> return $ clockTime2Tuple x) subTimeTuples (s2,ps2) (s1,ps1) = let dps = ps2-ps1 in let (s,ps) = (s2-s1-(if dps<0 then 1 else 0), if dps<0 then dps+(10^12) else dps) in (s, (div) ps (10^6)) test_md5 test_input n = flip mapM test_input ( \ input_row -> do t1 <- getClockTimeTuple sequence_ (replicate n (return $ md5 input_row)) t2 <- getClockTimeTuple dt <- subTimeTuples t2 t1 return (dt) ) test_list = [ pack "Hello world!"] ++ [L8.replicate 100 ':'] - The results are (iterations_count, microseconds): (1000, [105,105]) (+/- 10) (1, [1000,950]) (+/- 50) (10, [9050,9000]) (+/- 50) (100, [89200,89150]) (+/- 100) - I suspect following problems, which make the results non-objective: 1. I cant get out of laziness 2. I don't turn off GHC internal optimizers So, perhaps, GHC evaluates the MD5 once, but cycles on something else?.. For now I can only guess. Could anybody, please clarify and maybe suggest configuration, which would allow objective speed estimation? Thanks in advance, Best regards, Belka -- View this message in context: http://www.nabble.com/estimating-the-speed-of-operation-tp22075843p22075843.html Sent from the Haskell - Haskell-Cafe mailing list archive at Nabble.com. ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
