[Haskell-cafe] Haskell Speed Myth

[Thomas Davie]

Lo guys,
   I thought you'd like to know about this result.  I've been playing  
with the debian language shootout programs under OS X, looking at how  
fast Haskell code is compared to C on OS X, rather than linux.

Interestingly, Haskell comes out rather better on OS X than it did on  
Linux.  Here's my results (times in seconds):

		C	Haskell	Relative speed		Inverse
Binary Trees	6.842	1.228	0.179479684302835	5.57166123778502
Fannkuch	5.683	15.73	2.76790427591061	0.361284170375079
Mandelbrot	1.183	2.287	1.93322062552832	0.517271534761697
nbody		10.275	16.219	1.57849148418491	0.633516246377705
nsieve		0.167	0.253	1.51497005988024	0.660079051383399
nsieve-bits	0.471	0.713	1.51380042462845	0.660589060308555
partial sums	1.047	1.313	1.25405921680993	0.797410510281797
pidigits	1.238	1.4	1.13085621970921	0.884285714285714
recursive	1.554	3.594	2.31274131274131	0.432387312186978
spectral-norm	27.939	19.165	0.685958695729983	1.45781372293243
threadring	91.284	1.389	0.0152162481924543	65.719222462203
-------------------------------------------------------------------------
Averages			1.35333620432893	0.738914688605306

Some notes:
Hardware: 2Ghz Core2Duo, enough ram to not worry about paging

Some programs are not included, this is because the C code produced  
compile errors.  The Haskell code appears to be portable in *all* cases.

The average slowdown for running Haskell is only 1.3 times on OS X!   
That's pretty damn good.

I'm sure some people will say "yeh, but you have to optimise your code  
pretty heavily to get that kind of result".  Interestingly, the  
programs that have the biggest speed advantage over C here are also  
the most "naïvely" written ones.

The thing that seems to make C slower is the implementation of malloc  
in OS X.  This has a couple of implications, first, if apple replaced  
the malloc library, it would probably push the result back to the 1.7  
times slower we see under Linux.  Second, Fannkuch can probably be  
optimised differently for Haskell to look even better -- at the  
moment, the haskell code actually uses malloc!

Finally, that threading example... WOW! 65 times faster, and the code  
is *really* simple.  The C on the other hand is a massive mess.

Bob

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.haskell.org/pipermail/haskell-cafe/attachments/20080823/bcf3db95/attachment.htm