Hello,

When creating a package for own usage (so it doesn't need to be
portable), is it interessant to specify a particular architecture with
the -mcpu=cpu-type and -march=cpu-type of gcc instead of the default
386 ?

Is it really a gain in speed, or other advantage ?

Thanks in advance,
--
^^ Gauthier
(_____/°°-ç
| \_`-"
)/@mmm||
\nn \nn FOE-Belgium : http://www.amisdelaterre.be

On Thu, 12 Aug 2004 18:15:30 +0000, Gauthier wrote:

> Hello,
>
> When creating a package for own usage (so it doesn't need to be
> portable), is it interessant to specify a particular architecture with
> the -mcpu=cpu-type and -march=cpu-type of gcc instead of the default
> 386 ?

Probably: yes.

> Is it really a gain in speed,

Between i386 and i686 it /is/ for most things. However behond that: not
much. IME though, i586 as opposed to i486 doesn't generally gain much.

> or other advantage ?

Idunno.

> Thanks in advance,

Well, you might want to use "-O2". And "strip" the generated binaries.
Other then that, some specific programs doing math stuff gain some speed
from setting: MMX, 3D-Now, SSE, or whatever, options.

--
-Menno.

On Thu, 12 Aug 2004 19:49:31 +0000, Menno Duursma wrote:
> On Thu, 12 Aug 2004 18:15:30 +0000, Gauthier wrote:

[ GCC flags ]

>> or other advantage ?
>
> Idunno.

Thinking about this ... : if given program is to be run on a security
boundary (suid/sgid, daemon, kernel, etc) it might make writing an
exploit, for some vulnerability in it, somewhat harder. As such
"optimizing" may (in some cases) be considerd an obscurety measure.

Assuming here: the binary doesn't have the "readable" bit set for view
(from the atackers previlege level) they'd probably have to guess an ofset.

HTH.

--
-Menno.