Lennart Augustsson <[EMAIL PROTECTED]> wrote,
> So I did a test run too, on a 450MHz Pentuim II running NetBSD.
> I used Manuel's Haskell code, but I added a type signature for
> the expand function, and the original C++ code.
>
> dogbert% cc c.cc -o c.out -lstdc++
> dogbert% time c.out
> 1942530768
> 1.043u 0.010s 0:01.05 100.0% 0+0k 0+1io 0pf+0w
> dogbert% hbc h3.hs
> dogbert% time a.out
> [1,9,4,2,5,3,0,7,6,8]
> 0.761u 0.040s 0:00.80 100.0% 0+0k 0+0io 0pf+0w
> dogbert%
>
> So if you compile naively with no optimization the Haskell code
> is actually FASTER than the C++ code.
That's interesting. Actually, I didn't add the type
signature, because when I tried using -O optimisation, it
didn't make a difference and I wanted to minimise the delta
to the original code.
> no opt opt
> hbc 0.761u 0.663u
> cc 1.055u 0.167u
Overall, I have (on a 300MHz Celeron running Linux and
including the type signature to expand),
no opt opt
ghc 4.04pl1: 8.61u 2.63u
egcs 1.1.2 : 1.70u 0.31u
So, either Lennart has a worse C++ compiler or a better
Haskell compiler - or, I got something completely wrong.
Without wanting to draw an early conclusion, wasn't there
this request in GHC's documentation to report any program
performing better when compiled with HBC as a bug :-)
Cheers,
Manuel
PS: Note that my machine is much slower machine than
Lennart's, so an absolute comparison of our times is
meaningless.
