This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Various GCC compared (2.95.3 - 3.5-20040523)

On Fri, 2004-05-28 at 21:51, Rene Rebe wrote:
> Hi,
> On: Fri, 28 May 2004 20:52:51 +0200,
>     Laurent GUERBY <> wrote:
> > On Fri, 2004-05-28 at 18:28, Rene Rebe wrote:
> > > Now the gnupg results are not more reasonable and place recent gcc
> > > versions into a better light.
> > 
> > You need to configure GCC 3.5-exp with --disable-checking otherwise
> > build time are not "fair" vs releases because the exp version enable
> > some compiler internal checks by default. I believe you should
> > also report the date at which you took the sources from CVS
> > for 3.5-exp on your graph.
> I know. This is why I wrote in my initial mail that I have done
> this. Is there a runtime check so I can verify the resulting binary
> really does not include checking?

Sorry I missed the line in your post, I don't know how to check
the binary, "gcc -v" prints the configure line IIRC.

> Hm - I think (/ thought) relative number are better since I know my
> 800Mhz G3 is far slower than the 2.5Ghz Athlon-XP in the server room
> ... Not even to mention my 270Mhz UltraSPARC on the desk ... And
> taking the wide range of compile times (gzip vs. linux26) into account
> the diagrams would be harder to read.
> Why do you think absolute numbers are interesting?

To compare with other published benchmarks (eg: SPEC gzip and bzip2), to
enable back of the enveloppe extrapolations and to compare machines
with the GCC view. 

Even if you don't have homogeneous hardware,
other people might be able to use and complete the data, eg someone
with an Athlon 2000 will be able to evaluate a 2500
based on your results after checking the absolute
performance gain.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]