sm-sorcery AT lists.ibiblio.org
Subject: Discussion of Sorcery related topics
List archive
Re[2]: [SM-Sorcery]Architecture Selection Code - Will this work?
- From: Mark Andrews <msasgl AT msa-enterprises.com>
- To: Wolfgang Scheicher <worf AT sbox.tu-graz.ac.at>
- Cc: sm-sorcery AT lists.ibiblio.org
- Subject: Re[2]: [SM-Sorcery]Architecture Selection Code - Will this work?
- Date: Tue, 1 Oct 2002 10:46:37 +0100
Hello Wolfgang,
Monday, September 30, 2002, 8:10:46 PM, you wrote:
MA> I've just been looking at the best place to put my auto-processor
MA> selection code.
WS> is this for general hardware detection or for compiler flag stuff?
Both.
WS> because for the compiler optimisation settings i would be careful.
With this implementation the auto option is an add on to the current
optimisation menus, that way novice users (and probably nearly
everyone else if this works as well as I think it will) can use auto
which will be specific to their machine. Advanced users can continue
to use the current system and hand select exactly what they want if
they prefer i586 code to Athlon for example.
The whole point of gathering the cpuinfo is so I can be careful and
automagically pick the exact optimisation settings that work best on
that processor. This is early doors on this project, but believe me I
am looking for the simplest, most reliable implementation I can.
The code:-
CPUNAME=`egrep "model name" /proc/cpuinfo | cut -d: -f2 | uniq`
was just an example of what I might be doing in the routine to
illustrate my point. The final code will be different to this (need to
allow for SMP) and the optimisation flags will be generated from a
series of checks and possibly user input if it can't be avoided. At
the moment it looks like a different "auto" script will be needed for
each architecture.
Hence why I want to know where the archspecs files are called from. I
really need to know if I can do things like - prompt the user for input
in an archspecs routine / provide non standard optimisation options -
or whether that will break things. Yes I will look at the code, but I
figure someone already knows where archspecs are hooked and can tell
me where to start. Saves time and it lets me focus on the new stuff
I'm doing.
WS> maybe default to i386/486/586/686, but not that athlon stuff - or at
least
WS> ask... and we should find out what code runs on what machine - and display
WS> that info... rebuilding the whole thing is quite a task - and i would not
WS> like to do it too often.
Me neither. That's what started this. If the computer can detect the
best options itself then we can use them from the install onwards and
lack of user knowledge is less of a problem. For example, I didn't
know if my processor had an APIC or not when I compiled my kernel so I
said yes. If I'd known then about cpuinfo then I could have just
looked at 'flags' in /proc/cpuinfo, seen that 'apic' wasn't there
and known I didn't need APIC. Better still if sorcery had done it for
me ;-)
That's why I said both general hardware detection and compiler flag
stuff. When I started looking at the processor detection possibilities
I was just thinking optimisation settings. Then I saw that it can be
used for other things too. Why ask the user questions when the
computer can figure out the answer itself?
--
Best regards,
Mark mailto:msasgl AT msa-enterprises.com
-
Re[2]: [SM-Sorcery]Architecture Selection Code - Will this work?,
Mark Andrews, 10/01/2002
- Re: Re[2]: [SM-Sorcery]Architecture Selection Code - Will this work?, Dufflebunk, 10/01/2002
Archive powered by MHonArc 2.6.24.