Don Stewart wrote:
andrewcoppin:
Wait... you're telling me to give up? To not even try??
So all that bravado about Haskell enabling higher-level optimisations to
produce a result faster than C is actually complete nonesense?
What optimisations were operating in your md5 implementation that you
could expect it do match the C version?
I was hoping that things would be inlined, small loops unrolled, and I'd
end up with more or less the same imperative loops as the C version has.
I wasn't expecting to *beat* C in this particular case, but I expected
to get somewhere fairly near it.
If you want to write code that matches some heavily
optimised md5 in C then your
Haskell will need to compile down to similar data and control structures.
Agreed.
(did you look at the C implementation?)
I can't read C. (FWIW, I think I did briefly stare at the sources, but
eventually gave up because I simply had no clue what's going on.)
Does it?
I'm still figuring out how to answe that question.
[As in, "the profiler says 60% of my time is spent inside one function,
I've now tried writing that function 6 different ways, and currently it
compiles to about 3 pages of Core that I'm still attempting to
decypher". Given the size of the Core for this function, I'd say it's
probably inlining and unrolling just fine...]
So, as Bulat says, explain why your code should have similar performance
to the C version.
Because it executes the same algorithm? I mean, there's essentially only
one way round that you can perform the MD5 algorithm, so it just comes
down to how efficiently you execute the steps involved.
It would be much more useful than complaining about
the community who patiently help you day in and out.
Telling somebody "don't bother trying, it's impossible" doesn't strike
me as terribly helpful. It doesn't make my code go any faster, and it
certainly doesn't help me learn anything.
I guess my responce was a little harsh. But when you invest a lot of
time and effort in something, and somebody off-handedly tells you you're
wasting your time... it's quite upsetting.
_______________________________________________
Haskell-Cafe mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/haskell-cafe
