Gold timings

[Michal Nowak]

----- "Ian Lance Taylor" <iant at google.com> wrote:

> Hi, Tom Tromey pointed me at your message
> http://lists.fedoraproject.org/pipermail/devel/2010-March/133039.html

Hi Ian.

> I was curious what you are timing when you compare ld and gold.  Is
> that the total time that it takes to build the package, or just the
> time that the linker runs?

You may be familiar with Mock which builds source RPMs in chroot. Mock 
prepares build root (e.g. fedora-12-x86_64), installs BuildRequires,
and builds given SRPM. Such a build should be repeatable. What is done
is that run of this `mock' process is timed -- bad thing obviously here
is that the build time includes part which is not related to build itself
(i.e. build root preparation, package installation), however good is that
the unrelated time is pretty much constant among runs.

If anyone knows hot to measure just the build time, I'd be glad to know.

> As you mention, xulrunner and thunderbird are much slower, which is
> rather odd and completely contrary to my own testing.  Do you know if
> those builds use linker scripts?  Do you have the build logs?

Unfortunately logs are gone with my yesterday's reinstall. However I am
building xulrunner and thunderbird right now on different box. Will let
you know till Monday.

No idea about the linking script since logs are gone.

> Also, what sort of machine were you running this on, and how much RAM
> did it have?

Dualcore, Intel(R) Xeon(R) CPU 5110 @ 1.60GHz, 2GB RAM, 4GB swap.

> I'd certainly like to get to the bottom of any cases where gold is
> significantly slower than GNU ld.
> 
> I'm not on the devel at lists.fedoraproject.org mailing list.
> 
> Thanks.
> 
> Ian

Michal