[gentoo-dev] GCC 4.6.0

[gentoo-dev] GCC 4.6.0

[Ryan Hill]

[-- Attachment #1: Type: text/plain, Size: 1287 bytes --]

I just added 4.6.0 to the tree.  We probably won't be unmasking it any time
soon (after 4.6.1 for sure), but please start testing your packages now so we
can get things rolling.

http://gcc.gnu.org/gcc-4.6/changes.html

For general testing you may want to add the gcc-porting overlay to avoid
known failures that haven't been fixed in the tree yet.

Outstanding issues I know about:

  - hardened is broken (stage comparison mismatch)
  - sys-boot/grub-0.97 is miscompiled (bug #360513)

  https://overlays.gentoo.org/proj/gcc-porting/browser/README-4.6

Common errors:

  https://overlays.gentoo.org/proj/gcc-porting/browser/ERRORS-4.6
  http://lists.fedoraproject.org/pipermail/devel/2011-February/148523.html

  You may also want to test your packages with the new -Ofast option to be
  sure it doesn't have any hardcoded assumptions about -O flags.

Please assign any bugs you find in GCC itself to toolchain, and any bugs in
packages to their respective maintainers with a block on "gcc-4.6" (ie. bug
#346809).


-- 
fonts, gcc-porting,                  it makes no sense how it makes no sense
toolchain, wxwidgets                           but i'll take it free anytime
@ gentoo.org                EFFD 380E 047A 4B51 D2BD C64F 8AA8 8346 F9A4 0662

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 198 bytes --]

[gentoo-dev] Re: GCC 4.6.0

[Duncan]

Ryan Hill posted on Sat, 02 Apr 2011 22:11:12 -0600 as excerpted:

>   You may also want to test your packages with the new -Ofast option to
>   be sure it doesn't have any hardcoded assumptions about -O flags.

The release description I've read for -Ofast says it includes -fast-math, 
among other things, a flag Gentoo has always strongly discouraged (you 
break with it, you keep the pieces) and which can get bugs resolved/
invalid as a result.

Now that gcc 4.6 itself is more strongly supporting it as enabled with one 
of the -O options, is that policy going to change, or is Gentoo going to 
officially not support -Ofast, as well?

Or is that yet to be established, thru testing?

FWIW, I've always stayed away from that flag, but if Gentoo's going to 
support it now, that may well change, tho I'd certainly disable it for 
specific packages using /etc/portage/env/*, as I already do for 
-combine, in my default CFLAGS but not CXXFLAGS, for instance.

-- 
Duncan - List replies preferred.   No HTML msgs.
"Every nonfree program has a lord, a master --
and if you use the program, he is your master."  Richard Stallman

[gentoo-dev] Re: GCC 4.6.0

[Ryan Hill]

[-- Attachment #1: Type: text/plain, Size: 1231 bytes --]

On Sun, 3 Apr 2011 05:50:32 +0000 (UTC)
Duncan <1i5t5.duncan@cox.net> wrote:

> Ryan Hill posted on Sat, 02 Apr 2011 22:11:12 -0600 as excerpted:
> 
> >   You may also want to test your packages with the new -Ofast option to
> >   be sure it doesn't have any hardcoded assumptions about -O flags.
> 
> The release description I've read for -Ofast says it includes -fast-math, 
> among other things, a flag Gentoo has always strongly discouraged (you 
> break with it, you keep the pieces) and which can get bugs resolved/
> invalid as a result.
> 
> Now that gcc 4.6 itself is more strongly supporting it as enabled with one 
> of the -O options, is that policy going to change, or is Gentoo going to 
> officially not support -Ofast, as well?

I doubt we will.  If a package breaks because of -Ofast there's really
nothing we can do about it.  It's not a bug in the compiler or the package,
it's that you explicitly told it to generate non-standard-conformant code.


-- 
fonts, gcc-porting,                  it makes no sense how it makes no sense
toolchain, wxwidgets                           but i'll take it free anytime
@ gentoo.org                EFFD 380E 047A 4B51 D2BD C64F 8AA8 8346 F9A4 0662

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 198 bytes --]

Re: [gentoo-dev] Re: GCC 4.6.0

[Mike Frysinger]

On Sun, Apr 3, 2011 at 6:19 AM, Ryan Hill wrote:
> On Sun, 3 Apr 2011 05:50:32 Duncan wrote:
>> Ryan Hill posted on Sat, 02 Apr 2011 22:11:12 -0600 as excerpted:
>>> You may also want to test your packages with the new -Ofast option to
>>> be sure it doesn't have any hardcoded assumptions about -O flags.
>>
>> The release description I've read for -Ofast says it includes -fast-math,
>> among other things, a flag Gentoo has always strongly discouraged (you
>> break with it, you keep the pieces) and which can get bugs resolved/
>> invalid as a result.
>>
>> Now that gcc 4.6 itself is more strongly supporting it as enabled with one
>> of the -O options, is that policy going to change, or is Gentoo going to
>> officially not support -Ofast, as well?
>
> I doubt we will.  If a package breaks because of -Ofast there's really
> nothing we can do about it.  It's not a bug in the compiler or the package,
> it's that you explicitly told it to generate non-standard-conformant code.

obviously we will look at ICEs and such, but in terms of apps
misbehaving at runtime, most likely we'll write it up as not a bug
like Ryan says
-mike

Re: [gentoo-dev] GCC 4.6.0

[Andreas K. Huettel]

[-- Attachment #1: Type: Text/Plain, Size: 440 bytes --]

On Sunday 03 April 2011 06:11:12 Ryan Hill wrote:
> 
> For general testing you may want to add the gcc-porting overlay to avoid
> known failures that haven't been fixed in the tree yet.
> 

Just for clarification... I thought the toolchain overlay was for the always-
gentoo-most-experimental compilers. What's the difference?!

-- 

Andreas K. Huettel
Gentoo Linux developer 
dilfridge@gentoo.org
http://www.akhuettel.de/


[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 198 bytes --]

Re: [gentoo-dev] Re: GCC 4.6.0

[Branko Badrljica]

On 03. 04. 2011 16:04, Mike Frysinger wrote:
> On Sun, Apr 3, 2011 at 6:19 AM, Ryan Hill wrote:
>> On Sun, 3 Apr 2011 05:50:32 Duncan wrote:
>>> Ryan Hill posted on Sat, 02 Apr 2011 22:11:12 -0600 as excerpted:
>>>>  You may also want to test your packages with the new -Ofast option to
>>>>  be sure it doesn't have any hardcoded assumptions about -O flags.
>>>
>>> The release description I've read for -Ofast says it includes -fast-math,
>>> among other things, a flag Gentoo has always strongly discouraged (you
>>> break with it, you keep the pieces) and which can get bugs resolved/
>>> invalid as a result.
>>>
>>> Now that gcc 4.6 itself is more strongly supporting it as enabled with one
>>> of the -O options, is that policy going to change, or is Gentoo going to
>>> officially not support -Ofast, as well?
>>
>> I doubt we will.  If a package breaks because of -Ofast there's really
>> nothing we can do about it.  It's not a bug in the compiler or the package,
>> it's that you explicitly told it to generate non-standard-conformant code.
> 
> obviously we will look at ICEs and such, but in terms of apps
> misbehaving at runtime, most likely we'll write it up as not a bug
> like Ryan says
> -mike
> 
> 

Maybe slightly off topic, but still..

1. I've noticed that -Ofast and couple other bits on gcc which I have
seen on Open64 before. Are these new optimisations "imported" from
Open64 or is this simply the result of good old competition of both teams ?


2. Is there any info on gcc version that will support -march=Bulldozer ?
I have googled a couple of gcc-related posts about optimizing for this
CPU architecture intricacies and I have hoped to see support for it in
4.6... Is this stuff still in early development or is it just waiting
for AMD to ship the chips due to some kind of NDA ?

Re: [gentoo-dev] Re: GCC 4.6.0

[Chí-Thanh Christopher Nguyễn]

Branko Badrljica schrieb:
> 2. Is there any info on gcc version that will support -march=Bulldozer ?
> I have googled a couple of gcc-related posts about optimizing for this
> CPU architecture intricacies and I have hoped to see support for it in
> 4.6... Is this stuff still in early development or is it just waiting
> for AMD to ship the chips due to some kind of NDA ?

gcc-4.6.0 adds targets for two new AMD CPU families, bdver1 and btver1.
It is all public, both the mailing list posts leading to the inclusion
and gcc/config/i386/i386.c

The gcc manpage was not updated, I think maybe by mistake.


Best regards,
Chi-Thanh Christopher Nguyen

Re: [gentoo-dev] GCC 4.6.0

[Mike Frysinger]

On Sun, Apr 3, 2011 at 10:56 AM, Andreas K. Huettel wrote:
> On Sunday 03 April 2011 06:11:12 Ryan Hill wrote:
>> For general testing you may want to add the gcc-porting overlay to avoid
>> known failures that haven't been fixed in the tree yet.
>
> Just for clarification... I thought the toolchain overlay was for the always-
> gentoo-most-experimental compilers. What's the difference?!

one is for toolchains and one is for porting.  the former only has
toolchain related packages while the latter has a lot of random
patched packages.
-mike

[gentoo-dev] Re: GCC 4.6.0

[Diego Elio Pettenò]

Il giorno sab, 02/04/2011 alle 22.11 -0600, Ryan Hill ha scritto:
> Common errors:

I've been running my tinderbox with GCC 4.6 now, so I hope to help out
discovering the issue asap, but in the mean time I'd like to point out
that GCC 4.6 (a little more than others, afaict) could cause ./configure
scripts to fail (or misdetect availability of functions).

If something does not seem to build right, but doesn't appear directly
related to GCC 4.6, make sure to attach the config.log of the configure
execution.

In particular, since with GCC 4.5 (and modern glibc) doing things such
as

write(fd, buf, bufsize);

would have caused "return value ignored" warnings, which would have
thrown off detections using -Werror, they were rewritten as

int n = write(fd, buf, bufsize);

... too bad that this *now* causes the "unused but set" warning.

-- 
Diego Elio Pettenò — Flameeyes
http://blog.flameeyes.eu/