Athlon64-sse3

[Saar]

Hey all, just thought I'd share something sort of oddball that I came across whilst installing/compiling mplayer this evening.

I was experimenting with CFLAGS (the system I was working with is an Athlon64 X2 3800+ socket AM2) and I decided to make what I thought was a small change, which ultimately hosed compiling of anything.

The particular family of processors I'm dealing with (Windsor) support sse3. There are two ways of specifying this that I'm aware of: 1) -msse3 and 2) -march=athlon64-sse3. From what I read here my impression was that these two methods were synonymous. This is not the case, functionally. Using the latter method will prevent sse3 from being stripped off for certain packages (when it might be required for the compiler to work). One such package is apparently libdca.

I'm still somewhat newb-ish to Gentoo, but reverting back to -msse3 instead of -march=athlon64-sse3 fixed the problem for me. Hopefully my understanding of "why" is correct and/or helpful. Figured I'd share just to be on the safe side.

[96140]

--

[Eckos]

i use -march=native -msse3 and I have a AMD Athlon 64 X2 3800+.

[F-0_ICE]

so to clarify is there any difference between:

Code: Select all

CFLAGS="-march=athlon64 -O2 -pipe -msse3"

(my current setup)
and

Code: Select all

CFLAGS="-march=athlon64-sse3 -O2 -pipe"

from what i understand there shouldn't be any problems compiling if i switch to the new setting?

[Enlight]

so to clarify is there any difference between:

Code: Select all

CFLAGS="-march=athlon64 -O2 -pipe -msse3"

(my current setup)
and

Code: Select all

CFLAGS="-march=athlon64-sse3 -O2 -pipe"

from what i understand there shouldn't be any problems compiling if i switch to the new setting?

my understanding is that both CFLAGS do the same, portage knows how to strip one, not the other one, but some package can't be built withh sse3 support activated.

[yngwin]

No problems here with libdca.