Fwd: Re: [Shootout-list] Science-related benchmarks (speedoptimisation)
Jon Harrop
jon@ffconsultancy.com
Fri, 29 Apr 2005 18:18:33 +0100
On Thursday 28 April 2005 18:23, Brent Fulgham wrote:
> > this is a really interesting question. it does sound
> > sensible to compare
> > two compilers for the same language. but why stop
> > there? why not compare
> > the normal compiler/runtime switches (vanilla) with
> > the best possible
> > combination of compiler switches (all-flavors)?
> > because that is even
> > more interesting.
>
> Yes, this could be done.
I just tried playing with compiler switches for my ray tracer. On AMD64 it
makes little difference. On x86 it gives ~2x speed up! So I do think this is
worth mentioning.
Specifically, on 32-bit x86:
Using "g++-3.4 -O2 ray.cpp -o ray" takes 94s.
Using "g++-3.4 -march=athlon-tbird -O2 ray.cpp -o ray" takes 51s.
> > where do we stop comparing implementations? say that
> > a language (c) has two compilers (gcc and icc) and
> > that is ok to put both in the shootout.
> > what about another language (ocaml) that has two
> > compilers (ocamlopt and ocamlc). why deny this
> > language the same treatment as c gets?
>
> We don't. ocamlopt and ocamlc are both on the
> 'Great' version of the shootout:
>
> http://shootout.alioth.debian.org/great/benchmark.php?test=all&lang=ocaml&s
>ort=fullcpu
> http://shootout.alioth.debian.org/great/benchmark.php?test=all&lang=ocamlb&
>sort=fullcpu
Have you considered including ocamljit? I'm not sure about its reliability but
it is possibly worth a try.
--
Dr Jon D Harrop, Flying Frog Consultancy Ltd.
Objective CAML for Scientists
http://www.ffconsultancy.com/products/ocaml_for_scientists