[gentoo-user] About this graphite stuff

[gentoo-user] About this graphite stuff

[Alex Schuster]

Hi there!

Due to a MAJOR hardware problem I just got a new PC. It has an AMD
FX(tm)-4100 Quad-Core Processor which is a lot faster than my previous
AMD 4850e dual core machine. I'd like to emerge -e @world with
-march=native now, but I think about starting to use that graphite stuff.
You know, enabling the graphite USE flag for gcc, and adding
-floop-interchange -floop-strip-mine -floop-block to the CFLAGS in order
to gain some speed by parallelizing stuff or something like that. 

What are your impressions on this? Is it fun? Will there be a noticeable
speed difference? Do packages fail to build? We just had a 'Graphite
causing trouble' thread here, the problem was that dev-libs/cloog-ppl
has to be rebuilt when dev-libs/ppl has been updated. Can there be other
problems, which would make me waste much more time than I could
possibly gain by using these optimizations?

	Wonko

Re: [gentoo-user] About this graphite stuff

[victor romanchuk]

Alex Schuster wrote, at 01/29/2012 11:23 PM:
> What are your impressions on this? Is it fun? Will there be a noticeable
> speed difference? Do packages fail to build? We just had a 'Graphite
> causing trouble' thread here, the problem was that dev-libs/cloog-ppl
> has to be rebuilt when dev-libs/ppl has been updated. Can there be other
> problems, which would make me waste much more time than I could
> possibly gain by using these optimizations?
>

i'm using graphite on core-i7 (950), x86_64 since the release of gcc-4.4.5 and
consider it as 'just fun' - i did not observe significant speed difference but
it should depend on a software you're going to 'graphitize'. things installed on
my desktop are mostly for development (emacs, gdb, *sql, php, perl) with trivial
multimedia (mplayer with gnome frontend), a set of web browsers and ordinary
office framework: thunderbird, pidgin and libreoffice

just for reference these are CFLAGS from my /etc/make.conf:

CFLAGS="-O2 -g0 -march=core2 -msse4 -mcx16 -mpopcnt -msahf \
        -ftree-loop-distribution -ftree-loop-linear -mmmx \
        -floop-interchange -floop-strip-mine -floop-block -pipe"

i do not use 'native' flag because that machine acts as distcc server for
several smaller core-i<x> computers

the only graphite incompatibility i've detected is x11-wm/compiz; however this
is easily worked around using portage environment quirks in
/etc/portage/env/x11-wm/compiz:

CFLAGS="-O2 -g0 -march=core2 -msse4 -mmmx -pipe"
CXXFLAGS=${CFLAGS}

Re: [gentoo-user] About this graphite stuff

[Alex Schuster]

victor romanchuk writes:

> Alex Schuster wrote, at 01/29/2012 11:23 PM:
> > What are your impressions on this? Is it fun? Will there be a
> > noticeable speed difference? Do packages fail to build? We just had a
> > 'Graphite causing trouble' thread here, the problem was that
> > dev-libs/cloog-ppl has to be rebuilt when dev-libs/ppl has been
> > updated. Can there be other problems, which would make me waste much
> > more time than I could possibly gain by using these optimizations?
> 
> i'm using graphite on core-i7 (950), x86_64 since the release of
> gcc-4.4.5 and consider it as 'just fun' - i did not observe significant
> speed difference but it should depend on a software you're going to
> 'graphitize'. things installed on my desktop are mostly for development
> (emacs, gdb, *sql, php, perl) with trivial multimedia (mplayer with
> gnome frontend), a set of web browsers and ordinary office framework:
> thunderbird, pidgin and libreoffice

I'm a KDE user, and it always felt too slow, so I thought about giving
graphite a try. Now, with the new machine, it is faster, and with 16 GB
of RAM swapping no longer occurs. So there is no longer a need for more
optimization, but I'm emerging -e @world anyway, so I think I'll just try
it.

> just for reference these are CFLAGS from my /etc/make.conf:
> 
> CFLAGS="-O2 -g0 -march=core2 -msse4 -mcx16 -mpopcnt -msahf \
>         -ftree-loop-distribution -ftree-loop-linear -mmmx \
>         -floop-interchange -floop-strip-mine -floop-block -pipe"
> 
> i do not use 'native' flag because that machine acts as distcc server
> for several smaller core-i<x> computers

That won't be a problem, unless the machine will be a distcc _client_. In
this case, your CFLAGS will be sent over to the distcc servers, and they
will apply _their_ native settings.
Just make sure the smaller computers don't have -march=native in their
CFLAGS, as they would get code delivered compiled for the i7.

> the only graphite incompatibility i've detected is x11-wm/compiz;
> however this is easily worked around using portage environment quirks in
> /etc/portage/env/x11-wm/compiz:
> 
> CFLAGS="-O2 -g0 -march=core2 -msse4 -mmmx -pipe"
> CXXFLAGS=${CFLAGS}

I don't use that, but I guess in case I experience strange compile
problems I can always try to avoid those special CFLAGS, and it might
work.

Thanks for your input on this,

	Wonko

Re: [gentoo-user] About this graphite stuff

[Volker Armin Hemmann]

last time I played with that I had to pretty much rebuild the system because 
nothing worked anymore.

-- 
#163933

[gentoo-user] Re: About this graphite stuff

[walt]

On 01/29/2012 11:23 AM, Alex Schuster wrote:
> Do packages fail to build? We just had a 'Graphite
> causing trouble' thread here,

I've read in this group that an occasional package fails when using
-j2 or higher (which you will certainly be doing) but that has nothing
to do with graphite.

I've been using graphite for a long time with no problem whatever.
Except for a rare rebuild of cloog-ppl, of course, but I already knew
how to bail myself out of that situation, and now you know too :)

Re: [gentoo-user] Re: About this graphite stuff

[Pandu Poluan]

[-- Attachment #1: Type: text/plain, Size: 817 bytes --]

On Jan 30, 2012 4:52 AM, "walt" <w41ter@gmail.com> wrote:
>
> On 01/29/2012 11:23 AM, Alex Schuster wrote:
> > Do packages fail to build? We just had a 'Graphite
> > causing trouble' thread here,
>
> I've read in this group that an occasional package fails when using
> -j2 or higher (which you will certainly be doing) but that has nothing
> to do with graphite.
>

In my case, as long as I stay -j3 or lower, everything compile well.

> I've been using graphite for a long time with no problem whatever.

I took the plunge when graphite went stable and production-ready in the
4.5.x series.

> Except for a rare rebuild of cloog-ppl, of course, but I already knew
> how to bail myself out of that situation, and now you know too :)
>

Let me add that my servers *feel* faster with graphite. Of course, YMMV.

Rgds,

[-- Attachment #2: Type: text/html, Size: 1029 bytes --]

Re: OT: [gentoo-user] About this graphite stuff

[v_2e]

  Hello!
  I was reading this thread and felt that the "graphite" USE flag seems
familiar to me, but I just couldn't remember where I had seen it. So I
checked and discovered that there are 2 packages with such USE flag on
my system:

$ equery hasuse graphite
 * Searching for USE flag graphite ... 
[IP-] [  ] app-office/libreoffice-3.5.0.1:0
[IP-] [  ] sys-devel/gcc-4.5.3-r2:4.5

But the description of this USE flag is different for these two
packages:

 - - graphite  : Add support for the framework for loop optimizations
based on a polyhedral intermediate representation

 + + graphite  : Enable support for non-Roman fonts via
media-gfx/graphite2

  So, is it normal to have the same USE flag for two different meanings?

  Thanks.
    Vladimir

----- 
 <v_2e@ukr.net>

Re: OT: [gentoo-user] About this graphite stuff

[Michael Mol]

On Mon, Jan 30, 2012 at 8:38 AM,  <v_2e@ukr.net> wrote:
>  Hello!
>  I was reading this thread and felt that the "graphite" USE flag seems
> familiar to me, but I just couldn't remember where I had seen it. So I
> checked and discovered that there are 2 packages with such USE flag on
> my system:
>
> $ equery hasuse graphite
>  * Searching for USE flag graphite ...
> [IP-] [  ] app-office/libreoffice-3.5.0.1:0
> [IP-] [  ] sys-devel/gcc-4.5.3-r2:4.5
>
> But the description of this USE flag is different for these two
> packages:
>
>  - - graphite  : Add support for the framework for loop optimizations
> based on a polyhedral intermediate representation
>
>  + + graphite  : Enable support for non-Roman fonts via
> media-gfx/graphite2
>
>  So, is it normal to have the same USE flag for two different meanings?

Generally undesireale, but it happens. This is part of why USE flags
are lumped into two categories, 'global' and 'local'. Some USE flags
(such as 'debug') might be found as a global flag, but their meaning
can differ significantly package to package.

It's probably worth noting that you can set USE flags per-package,
though I've found most USE flags I touch are safe to set in make.conf.


-- 
:wq