Safe Cflags AMD E350 / Thinkpad x120e

[gentian]

I'm currently using "-march=native" and I'd like to know which flags are needed for my e350 cpu. Here's what's on the gentoo wiki page for e350 cpus:

[Veldrin]

gcc translates -march=native into -march=amdfam10, which would be ok for the average case.
but e350 lacks the 3dnow instruction set, which is why it has been excluded in the cflags shown on the wiki.

In addition, e350 support some more instruction, which are not covered by -march=amdfam10, which is why -mcx16 -mpopcnt -msse3 -msse4a -mmmx have been added.

IIRC it does not matter, if you chose -march=native or -march=amdfam10 as long as you append -fno-3dnow, becuase otherwise the compiler will generate code, that cannot be interpreted by the cpu. the other options are just to make full use of the cpu.
and even 3dnow errors would only appear in multimedia applications (ffmpeg, mplayer, vlc, browsers to name a few).

HTH
V.
_________________
read the portage output!
If my answer is too concise, ask for an explanation.

[gentian]

[Veldrin]

That is correct, that in the past you could just use the arch of your cpu, even if gcc did not completely know it.

The problem with E350 is that gcc does not know it, so it uses the next best arch it knows: amdfam10.

BUT, E350 got stripped of the 3dnow commands, and amdfam10 still uses them, therefore code built with amdfam10 will have 3dnow instructions in it, which an E350 does not know how to interpret.


There are 2 possibilies to solve this issue: take the least common cpu that provides all instruction sets, that are understood by the cpu (which would -mtune=generic, there is not -march=generic), or take the next realted cpu and disable those parts it does not like (-march=amdfam10 -fno-3dnow).

that is for the explicit -march declaration.


if you want to use march=native, you have to make sure, that gcc knows your cpu (gcc-4.5 does not!) as it choses the best -march available.
The problem here again is that gcc chooses -march=amdfam10 which causes the above mentioned problem, and IIRC can be solved in exactly the same way (-march=native -fno-3dnow).


gcc-4.7 (currently in hardmasked testing) does provide -march=btver1, which coves all cpu features, and then -march=native works again without any fiddling, as gcc choses the correct -march.


I hope that clarified my previous post.
V.

PS. I ignored on purposes the part about the additional commands/instructions, to focus on the core issue.
_________________
read the portage output!
If my answer is too concise, ask for an explanation.

[gentian]

[Veldrin]

you can try.

my prediction: It will work fine till you hit the first 3dnow instruction, which will be used by some kind of multimedia application (that includes media players, web browsers (or its plugins), image manipulation and maybe some more).

I tend to be more conservative, as I built a -march=core2 system, and tried to run it on a phenom II cpu. works fine till i started the first browser (chromium iirc).
core2 includes -mssse3 which is not supported by amdfam10 cpu (phenom II), and the browser would just segfault.


V.
_________________
read the portage output!
If my answer is too concise, ask for an explanation.

[_SerEga_]

gentian
In case gcc-4.6 you can use -march=btver1 ....
But if you used -march=amdfam10 then you must add -fno-3dnow . Also you may add -mcx16 -mpopcnt -msse3 -msse4a -mmmx

[NeddySeagoon]

gcc-4.6 won't build a working grub, which is why its still masked.
Other than that, gcc-4.6 seems fine. Use grub-static in place of grub if you want to use gcc-4.6 systemwide.
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

[gentian]

[NeddySeagoon]

gentian,

My E350 media player uses

[gentian]

[NeddySeagoon]

gentian,

-march=native is shorthand for look at the CPU I'm running on and use the CFLAGS for the CPU I found.
Thats ok under the following circumstances.
a) you only want to run thre code on the CPU that built it
b) gcc knows your CPU exactly.

a) is generally not true when disctcc is used. The helper CPUs may net even be AMD. They could be PPC, SPARC, ARM ....
b) when gcc does not know your CPU exactly, it uses the CPU familiy. That generally OK but not for an E350.

In case b) you must euther get a newer gcc that does know your cpu (gcc-4.6.2) or help older gcc to get the CFLAGS right.
Most of the amdfam10 CPUs have 3dnow, 3dnowext and so on. The E350 does not.

For the code out of distcc to work, all of the contributing verions of gcc must be *identical*
I don't think distcc is worth the trouble any more. If you have 4G RAM, use gcc-4.6.2 and put /var/tmp/portage into /dev/shm

[gentian]

Would I need to use these USE flags as well? I got these settings from the x120e gentoo-wiki page

[Veldrin]

no - because these are cpu flags, and not useflags.
OTOH, for some of them useflags exist: mmx sse sse2 sse3 ssse3 sse4a
Those you could add. or just add them per package basis.

Can you link the page on gentoo-wiki? It seems down ATM, and maybe there is a cached version on google somewhere...

V.
_________________
read the portage output!
If my answer is too concise, ask for an explanation.

[gentian]

[Veldrin]

that is the usual safe cflags page. It does not mention anything about useflags. All those are required, are mentioned already.
_________________
read the portage output!
If my answer is too concise, ask for an explanation.

[gentian]

[Veldrin]

[emc]

I will try on my E450.