2010/9/26 Thomas Backlund <tmb@iki.fi>
Giuseppe Ghibň skrev 26.9.2010 02:09:Well, here is a simple fact.
Not exactly. I'm not talking in just using -march=<something> but in
also pushing -mfpmath=sse -msse (and maybe -msse2) , which should be
much more than JUST 1-2% (1-2% is usually the benchmark
tolerance)...
Yes, going for i686 + SSE2 would/could give some benefits,
BUT
It would close support for all theese:
- Intel i586 (all)
- Intel Pentium Pro
- Intel Pentium II
- Intel Pentium III (Including some Pentium D)
- 32-bit AMD Athlon
- AMD Geode
- VIA C3
- Transmeta Crusoe
So, thats not an option.
Now what about only i686 + SSE then?
It would still close support for all theese:
- Intel i586 (all)
- Some Intel Pentium D
- 32-bit AMD Athlon < XP/MP
- AMD Geode < NX
- VIA C3 (Samuel* and Ezra)
- Transmeta Crusoe
So, still not an option.
Now what about only i686 then?
It would still close support for all theese:
- Intel i586 (all)
- Some Intel Pentium D
- AMD Geode < NX
- VIA C3 (Samuel* and Ezra)
- Transmeta Crusoe
So, still not an option.
And then to point out some other facts:
Some of the benefits are already accounted for like you noted earlier:And many applications capable of utilizing sse2 and other instruction sets already have runtime detection support, so no problem there.
"Of course we shouldn't forget that the MDV had already a system for providing optimized (look at /usr/lib/sse2 for instance) version of libraries according to instruction set supported."
So, where does that leave us?
Simple.
For 32bit installs, we will still support i586 as base.
For 64bit installs, it's simple as x86_64 is SSE2 by default.
Now, _if_ the Mageia community gets enough developers to think of _and_ support a "light netbook/legacy edition", then maybe we can think of doing a i586/i686 split, but for now we will use i586 as base.
And many of the users wanting i686/sse2 already have hardware capable of running x86_64, so that's what they should do, as there is where the future is.