[Haskell-cafe] benchmarking c/c++ and haskell

[Daniel Fischer]

On Wednesday 15 September 2010 02:50:15, David Terei wrote:
> On 13 September 2010 20:41, Vo Minh Thu <noteed at gmail.com> wrote:
> > ... the post is from 2008. No LLVM goodness. So I thought GHC 6.12.1
> > (not the latest and greatest HEAD) would be enough.
>
> I compiled the two programs myself out of curiosity and got the
> following times.
>
> Linux, 64bit, Ubuntu 10.10:
>
> 1e8
> clang: 0.180s
> gcc: 0.179s
> ghc 6.12.1 (viac): 0.187s
> ghc 6.12.1 (fasm): 0.218s
> ghc HEAD (viac): 0.186s
> ghc HEAD (fasm): 0.179s
> ghc HEAD (llvm): 0.174s
>
> 1e9
> clang: 1.657s
> gcc: 1.647s
> ghc 6.12.1 (viac): 1.653s
> ghc 6.12.1 (fasm): 1.975s
> ghc HEAD (viac): 1.648s
> ghc HEAD (fasm): 1.658s
> ghc HEAD (llvm): 1.646s
>
> So basically all have the same time except ghc 6.12.1 where fasm is a
> little slow.
>
> On windows xp 32bit I get quite different results which I trust less
> as the times are jumping around much more then they were on linux:
>
> 1e8
> gcc: 0.365s
> ghc 6.12.1 (viac): 5.287s
> ghc 6.12.1 (fasm): 1.332s
> ghc HEAD (viac): 5.292s
> ghc HEAD (fasm): 0.875s
> ghc HEAD (llvm): 0.359s
>
> Not sure why the results on windows are so different.

I have no idea why, but I remember that on several occasions timings for 
via-C compiled programmes on Windows have been abysmal.
Whether it's a general Windows/gcc mismatch or something GHC-specific, I've 
no idea (and can't find out, not having Windows).

For what it's worth, the approximate timings on my 32-bit linux (openSUSE 
11.1) box have been

1e8
gcc: ~0.4s
6.12.3 (viac): ~1.2s
6.12.3 (fasm): ~2.7s
HEAD the same as 6.12.3

(I have deleted the programmes and I don't remember the exact timings, if 
you're interested, I could reconstruct them).

> If anyone else
> wants to run the two programs on Windows and check that would be
> great.
>
> Cheers,
> David