Re[2]: [Haskell-cafe] How to improve speed? (MersenneTwister is several times slower than C version)
Hello Donald, Thursday, November 2, 2006, 2:21:31 PM, you wrote: 10-20 times difference is typical for GHC programs. It's really more like 2-4x. Sometimes better than C. Where's this huge figure coming from Bulat? If you have code that behaves like this, you should report it. are you analyzed the cases where performance is close or even better? i does. it's either because C version is limited by memory performance or just use different, less efficient algorithm the cases which shows slowness of ghc-generated code is factorial algorithm and the program attached. despite that Haskell code is far uglier, C version outperforms it 20 times. run both programs with arrays of about 10k elements to see the difference: a.out 1 elements 10 iterations in February i've written detailed explanation (in ghc-users) of why this comes and made some suggestions on improving it. of course, main problem is ghc's own code generator which is far away from gcc or even ocaml ones -- Best regards, Bulatmailto:[EMAIL PROTECTED] full.c Description: Binary data newest.hs Description: Binary data ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
Re[2]: [Haskell-cafe] How to improve speed? (MersenneTwister is several times slower than C version)
Hello Lennart, Thursday, November 2, 2006, 4:34:04 AM, you wrote: A big problem with the Mersenne Twister is the shifts. As has been noted elsewhere, ghc doesn't do such a great job on those. #ifdef __GLASGOW_HASKELL__ (I# a) # (I# b) = (I# (a `iShiftL#` b)) (I# a) # (I# b) = (I# (a `uncheckedIShiftRL#` b)) #else /* ! __GLASGOW_HASKELL__ */ a # b = a `shiftL` b a # b = a `shiftR` b #endif /* ! __GLASGOW_HASKELL__ */ -- Best regards, Bulatmailto:[EMAIL PROTECTED] ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
Re[2]: [Haskell-cafe] How to improve speed? (MersenneTwister is several times slower than C version)
Hello Lennart, Thursday, November 2, 2006, 6:04:39 AM, you wrote: The whole point of writing the Mersenne Twister was that I wanted to show how a stateful computation could be encapsulated in the ST monad and none of it showing up outside. This aspect of the code is totally gone now when everything is in the IO monad. Is there some good reason to have it in the IO monad? i think no. ST computations internally are really the same as IO ones -- Best regards, Bulatmailto:[EMAIL PROTECTED] ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
