CFLAGS Central (Part 2)

[kernelOfTruth]

Alright has anyone tested out march=naitve besides me? I've been running my system through hoops now to try to see if anything is broken and so far its good.

++

works fine here

use march=native only, and NOT march=native mtune=native or any other combination since that seems to do other not optimal optimizations which from time to time seems to break stuff ...

.. just my 2 Ct's

[Master Shake]

I am not currently on my linux box as I'm at work, but my cflags are something to the effect of -march=native -pipe -O2.

Works great. Actually noticed some improvements in World of Warcraft in wine. My old march was nocana.

[Veldrin]

++

I changed my CFLAGS to -march=native a month ago, when I switched to gcc-4.2.2. So far I have not encountered any problems.

Tough you have to be careful if you are cross-compiling... as native gets selected for the compiling processor.

just my $.02

[Master Shake]

That is a good point! Since I don't cross compile I don't need to worry about this, however does distcc take the cflags from the host machine?

[haarp]

Next question:
I have reasons to believe that if i compile wxGTK-2.8 with -fvisibility-inlines-hidden in my CXXFLAGS,
it breaks compilation of an app that depends on wxGTK. Is this possible? Up to now, I always assumed this flag was safe.

[MrCanis]

Hello all and happy new year ,

I receive today or in the next days a new Laptop

Code: Select all

Inspiron 6400 N-Series Pentium Dual Core T2130 (1.86Ghz, 1MB L2 Cache, 533Mhz)

And I don't know which parameter for

Code: Select all

-march=?

, in /etc/make.conf, I should use? I mean which is the optimal parameter? In gcc 4.2.2 is the possibility to set it to native but in Gentoo we are using gcc 4.1.2 and in this version there is no native option.

[desultory]

Merged the above post.

[Veldrin]

<cynic> if you don't know what to use, just leave it unset </cynic>

as it is yet another core cpu, use -march=prescott for 32bit, -march=nocona for 64bit

cheers
V.

for details, read here

[MrCanis]

Hello Veldrin,
thanks for your answer. You hit the nail exactly on the head.
That's the info I'm looking for.

[MrCanis]

Hello,
I'm wondering about the difference between the two GCC - flag combinations:

A)

Code: Select all

... -march=i686

B)

Code: Select all

... -march=i686 -mtune=i686

That are the CFLAG combinations A) for PENTIUM4 and B) for CORE2 in the kernel configuration under "Processor family...". The GNU documentation says that -march implies -mtune. If so, than would in combination B) the -mtune parameter unnecessary. I don't think so but what is the difference?

I have a Pentium Dual Core T2130. Processor family 6, according to the help text in the kernel configuration, is CORE2 the correct choice. And in /etc/make.conf (according to the WIKI SafeCFLAGS)

Code: Select all

-O2 -march=prescott -pipe

or

Code: Select all

-O2 -march=prescott -mfpmath=sse

or would it be better to use the same configuration as in the kernel configuration (In my case, configuration B)) and perhaps additionally -mfpmath=sse?

The background to this question is: My system, without Gnome (2.20.3), runs fine and fast but Gnome freezes unexpectedly, regardless whether a new home directory or not and if once froze, it happened ever again. The X server hasn't any problems, the same with blackbox and Firefox, Thunderbird and Sundbird. Therefore I don't think that it has something to do with one of the Mozilla programs. I'm a little bit at a loss.

Thanks in advance.

[Master Shake]

A T2130 is a Core Duo not a Core 2 Duo. Therefore you need to keep it prescott. I would recommend using native, however prescott is the best choice.

[MrCanis]

A T2130 is a Core Duo not a Core 2 Duo. Therefore you need to keep it prescott. I would recommend using native, however prescott is the best choice.

Hello,
thanks for your answer. I agree with you but in the kernel configuration is no prescott option. Only the two that I mentioned above ( A,B ). And again where is the difference between this both?

My GCC version is 4.1.2. How can I get 4.2.x?

Thanks in advance.

[Veldrin]

My GCC version is 4.1.2. How can I get 4.2.x?

keyword; upgrade; gcc-config to select new compiler - code is below.

Code: Select all

 # echo =sys-devel/gcc-4.2* ~x86 >> /etc/portage/package.keywords
# emerge -au gcc
# gcc-config -l
# gcc-config <2>

[MrCanis]

My GCC version is 4.1.2. How can I get 4.2.x?

keyword; upgrade; gcc-config to select new compiler - code is below.

Code: Select all

 # echo =sys-devel/gcc-4.2* ~x86 >> /etc/portage/package.keywords
# emerge -au gcc
# gcc-config -l
# gcc-config <2>

Hello,
thanks for your answer. That will simplify things, I hope.

[MrCanis]

Hello,
a little typo:

Code: Select all

 # echo =sys-devel/gcc-4.2* ~x86 >> /etc/portage/package.keywords

Should be:

Code: Select all

 # echo =sys-devel/gcc ~x86 >> /etc/portage/package.keywords

If not, emerge says: "Invalid atom..."

[Veldrin]

hmm weird. AFAIK you can unmask a package version with wild-chars. But at least it worked.

[MrCanis]

hmm weird. AFAIK you can unmask a package version with wild-chars. But at least it worked.

That's what I thought too, but emerge saws that different. Anyway, have I to run fix-libtool or something like that or was that only for the gcc 3.* relevant?

Now runs the build process for gcc 4.2.2. If that finished I will do a

Code: Select all

gcc-config 2 && emerge -e system && emerge -e world

with the new CFLAGS settings (-O2 -march=native -mtune=native -pipe) or have I before that something more to do?

Thanks in advance.

[Veldrin]

No, this should be alright. though there is a big discussion whether you really need to build the system twice (once in system, once in world), and the answer was pretty much no.

personally, I would rebuild the toolchain, and then the world.

Code: Select all

# emerge -1a binutils gcc-config gcc linux-headers glibc libstdc++v3 && emerge -e world

[MrCanis]

No, this should be alright. though there is a big discussion whether you really need to build the system twice (once in system, once in world), and the answer was pretty much no.

Yes I know, but I was unsure because of the gcc-update doc. There it would suggest that I have to do it in two steps. And actually I never did this two steps, but this gcc update is the first for me.

personally, I would rebuild the toolchain, and then the world.

Code: Select all

# emerge -1a binutils gcc-config gcc linux-headers glibc libstdc++v3 && emerge -e world

Your suggestion sounds good to me. I will do so.

Thanks.

[MrCanis]

hmm weird. AFAIK you can unmask a package version with wild-chars. But at least it worked.

Hello,
it isn't weird, I did something wrong. I recreated the entry and it works.

Sometimes are the finger faster than the brain, that shouldn't but...

[desultory]

What you are most likely want is to accept unstable versions of sys-devel/gcc in a specific slot, as opposed to matching glob pattern.

Code: Select all

sys-devel/gcc:4.2

[MrCanis]

What you are most likely want is to accept unstable versions of sys-devel/gcc in a specific slot, as opposed to matching glob pattern.

Code: Select all

sys-devel/gcc:4.2

Thanks for the info. For now I did it as Veldrin suggested. But my initial question isn't answered, perhaps you know the difference between the usage of "-march=i686" or "-march=i686 -mtune=i686". (Details some posts above)


Thanks in advance.

[desultory]

Considering that since -mtune was introduced -march is supposed to imply the equivalent -mtune, so far as I know they should be equivalent.

[MrCanis]

Considering that since -mtune was introduced -march is supposed to imply the equivalent -mtune, so far as I know they should be equivalent.

Hello,
thanks for your answer.

That's what I thought too, but if you have a look in the kernel configuration, you will see that both option will be used for menu entry "CORE2" and only -march for "PENTIUM4". Therefore I think there is a difference, but what? In the document (WIKI) Save CFLAGS for dual cores cites a email, which shall prove that for a dual core, prescott is the right option. This email stated that it should -march=prescott -mtune=generic but he wants look into it. (The wiki is, as I wrote this, not available)
That's not that important since gcc 4.2.2, because we have there the option native. But for those who do cross compiling or use distcc, it could be important and I think it is.

[Eckos]

what would be good flags for Athlon XP 2200+