[Haskell-cafe] Performance Help

[Dominic Steinitz]

On Friday 16 March 2007 21:29, David Brown wrote:
> Ian Lynagh wrote:
> > On Sun, Mar 11, 2007 at 08:18:44PM +0000, Dominic Steinitz wrote:
> >> I have re-written the sha1 code so that it is (hopefully) easy to see
>
> that it
>
> >> faithfully implements the algorithm (see
> >> http://www.itl.nist.gov/fipspubs/fip180-1.htm). Having got rid of the
>
> space
>
> >> leak, I have been trying to improve performance.
> >>
> >> Currently, the haskell code is 2 orders of magnitude slower than the
>
> sha1sum
>
> >> that ships with my linux.
> >
> > I don't know if this is useful to you, but darcs has some SHA1 code that
> > IIRC is much closer to C's performance. It currently uses darcs' own
> > FastPackedString library, but porting it to ByteString should be fairly
> > easy.
> >
> > See SHA1.lhs in http://www.abridgegame.org/repos/darcs-unstable
> >
> > It might even be able to be made faster still by calling lower-level
> > functions than {shift,rotate}{L,R} directly.
>
> I ended up deciding to call SHA1 routines out of openssl.  For
> applications where this is possible, it does very well, I got about
> 2.5 times the speed out of it, compared to ordinary C implementations.
>
> But, since harchive spends most of its CPU computing SHA1 hashes (and
> almost all of the rest in zlib), it is worth a complex binding there.
>
> Dave
Ian, Dave,

Thanks. My goal is to have natural haskell code that's reasonably efficient. I 
don't have a problem to solve that needs an efficient implementation of SHA1.

I'm going to try apfelmus' suggestions next and then (if I ever get yhc to 
build) start looking at what gets generated.

Dominic.