Re: Vanilla GCC 9 slower than Ubuntu GCC 9

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Fri, May 31, 2019 at 4:43 PM Tadeus Prastowo
<tadeus.prastowo@xxxxxxxx> wrote:
>
> On Wed, May 29, 2019 at 12:37 PM Tadeus Prastowo
> <tadeus.prastowo@xxxxxxxx> wrote:
> >
> > On Tue, May 28, 2019 at 10:00 PM Stefan Ring <stefanrin@xxxxxxxxx> wrote:
> > >
> > > On Mon, May 27, 2019 at 1:49 AM Tadeus Prastowo
> > > <tadeus.prastowo@xxxxxxxx> wrote:
> > > > Once debuild is done, I installed the aforementioned DEB files and
> > > > found out that the installed /usr/bin/g++-9 still compiles faster as
> > > > before.  So, I have failed to identify the factor in the PPA build
> > > > process that makes the resulting compiler faster than mine, which was
> > > > built from the release tarball using configure & make.  Any idea how
> > > > to find the factor that results in a faster compiler?  Is it really
> > > > the build process or the way the resulting compiler is installed that
> > > > gives the speed up?  For example, does symbol stripping have any
> > > > effect (I tried already to strip all symbols out of the executables
> > > > found in my GCC-9 installation with `strip -s' but to no avail)?
> > > > Perhaps some compiler libraries are installed in a particular way in
> > > > Ubuntu?
> > >
> > > I guess the Ubuntu package uses "make profiledbootstrap", which you
> > > don’t get (IIRC) when just issuing "make".
> >
> > Yes, that is true, but that did not ring my bell.  Thank you very much
> > for ringing the bell for me, quoting
> > https://gcc.gnu.org/install/build.html:
> > It is possible to use profile feedback to optimize the compiler
> > itself. This should result in a faster compiler binary. Experiments
> > done on x86 using gcc 3.3 showed approximately 7 percent speedup on
> > compiling C programs. To bootstrap the compiler with profile feedback,
> > use make profiledbootstrap.
> >
> > I will give it a try and report it back.
>
> Yes, it is very likely that the use of `make profiledbootstrap' is the
> main factor that boosts the compilation speed of vanilla GCC 9.
>
> I say ``very likely'' because compiling vanilla GCC 9 using `make
> profiledbootstrap' results in a missing Makefile error as detailed in
> this thread: https://gcc.gnu.org/ml/gcc-help/2019-05/msg00137.html
>
> However, forcing the PPA build not to use `profiledbootstrap' by
> editing the `debian/rules.defs' file results in DEB packages whose
> g++-9 runs slower by almost 9%.

To clarify: it runs slower by almost 9% than the vanilla GCC 9 whose
runtime was reported in the start of this thread.

The first trial gives:
real    6m9.464s
user    0m27.593s
sys    0m2.933s

The second trial gives:
real    6m29.064s
user    0m28.635s
sys    0m3.161s

> So, I hypothesize that if the vanilla
> GCC 9 were successful, it would also run faster by almost 15%.
>
> Thank you very much for your kind help, Stefan.
>
> --
> Best regards,
> Tadeus




[Index of Archives]     [Linux C Programming]     [Linux Kernel]     [eCos]     [Fedora Development]     [Fedora Announce]     [Autoconf]     [The DWARVES Debugging Tools]     [Yosemite Campsites]     [Yosemite News]     [Linux GCC]

  Powered by Linux