Re: GHC-6.4.1 much slower than GHC-6.4

2005-10-28 Thread Mirko Rahn 



Below is my profiling data (on Linux turing 2.6.11.4-21.9-default)
Any ideas?



   interesting-ghc-6.4 +RTS -p -RTS
total time  =  253.76 secs   (12688 ticks @ 20 ms)
total alloc = 15,991,765,312 bytes  (excludes profiling overheads)
transF NFA.Type  13.91.1



   interesting-ghc-6.4.1 +RTS -p -RTS
total time  = 1020.74 secs   (51037 ticks @ 20 ms)
total alloc = 70,086,586,572 bytes  (excludes profiling overheads)
transF NFA.Type  17.31.6


I have similar results, you can get it via

http://liinwww.ira.uka.de/~rahn/src/Interesting.6.4.prof.bin.prof
http://liinwww.ira.uka.de/~rahn/src/Interesting.6.4.1.prof.bin.prof

   Interesting.6.4.prof.bin +RTS -p -RTS
   total time  =  264.84 secs   (13242 ticks @ 20 ms)
   total alloc = 20,392,244,536 bytes  (excludes profiling overheads)
transF NFA.Type   6.92.4

   Interesting.6.4.1.prof.bin +RTS -p -RTS
   total time  = 1027.78 secs   (51389 ticks @ 20 ms)
   total alloc = 86,324,225,736 bytes  (excludes profiling overheads)
transF NFA.Type   8.93.4

As in your case, the 6.4.1-version allocs memory by about factor 4.3! I 
don't know why our profiles are pretty different but in my profiles the 
entry counts for some functions are much higher in the 6.4.1-version, 
e.G. 2559782 entries for Util.Words.isPartOf in 6.4-version and 10123468 
entries in 6.4.1-version, also about factor 4.


I guess the 6.4.1-version does more work, although I don't know why...

 Mirko

--
-- Mirko Rahn -- Tel +49-721 608 7504 --
--- http://liinwww.ira.uka.de/~rahn/ ---
___
Glasgow-haskell-bugs mailing list
Glasgow-haskell-bugs@haskell.org
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs

RE: GHC-6.4.1 much slower than GHC-6.4

2005-10-28 Thread Simon Marlow 
On 28 October 2005 16:09, Mirko Rahn wrote:

 I've found a way to speed up your code for ghc-6.4.1.
 Replace the where with a case in PCP.Suc.suc_rules (see below)
 
 Okay, thanks, that works fine.
 
 (Maybe that is a good idea in PCP.Rules.Suc, too)
 
 Seems to be unimportant whether the replacement is done in
 PCP.Rules.Suc too. In both cases the 6.4.1-version is slightly faster
 -- as is should be :-) -- than the 6.4.
 
 I don't know why this can be better optimized, but my profiling
 results 
 
 I even don't understand why it works. I think I have to reread the
 report once more...

It's a bug - we're looking into it.  Thanks for the report.

Cheers,
Simon
___
Glasgow-haskell-bugs mailing list
Glasgow-haskell-bugs@haskell.org
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs

Re: GHC-6.4.1 much slower than GHC-6.4

2005-10-28 Thread Christian Maeder 

Mirko Rahn wrote:



I've found a way to speed up your code for ghc-6.4.1.
Replace the where with a case in PCP.Suc.suc_rules (see below)


Okay, thanks, that works fine.


Yes, It's even quite fast without optimization (I did not check with 
ghc-6.4, though)



(Maybe that is a good idea in PCP.Rules.Suc, too)


Seems to be unimportant whether the replacement is done in PCP.Rules.Suc 


Yes, it is unimportant.

Christian
___
Glasgow-haskell-bugs mailing list
Glasgow-haskell-bugs@haskell.org
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs