This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Problems crosscompiling 3.3.1 (CVS branch) from i386-linux toi686-pc-linux-gnu


Kelley Cook wrote:

In article <3EE28434.8000206@ntlworld.com>, dliviu@ntlworld.com says...


Hello,

After all the nice things that started to appear in 3.3 branch of gcc, I launched on a project to build
a GNU/Linux system tailored to my CPU. As a first step, I'm trying to build a crosscompiler from my
i386-linux environment (Debian testing) to a i686-pc-linux-gnu target using the 3.3 CVS branch.


However, building the bootstrap c compiler fails misserably with:



...


Actually, the first question I need to ask is why exactly are you trying to create a cross-compiler. Unless you are just experimenting then I think that what you actually want in just a native compile, at which point you seem to have gone WAY overboard on your configure options. I'm pretty sure that you would be happy with just:

../../src/gcc-3.3.1/configure --host=athlon_xp-pc-linux-gnu

(AND NOTHING ELSE)

since your current i386-linux will still happily run Athlon XP tuned Linux binaries.

First it's an experiment. Following a recent thread on gcc mailing list, I was curious about the
right set of commands to build a cross compiler in a portable way. The easiest cross-compiler
to test should be between two ix86 architectures, right?


Second, I have this crazy plan to learn how to build from scratch a linux distribution, where
you can fine tune your executables. Yes, I know that something like this already exists, it's
just for me to learn. I want to be able to build a chrooted environment with my own settings
where to try various things.



Going to the trouble of cross compiling is usually used for when your CPU/OS is incapable of running your newly compiled binaries. Like for example using your i386-linux machine to compile for a Motorola MC68328 based Palm-pilot (m68k-palmos-coff).




4) OT: even with all those options, the genattrtab always chooses the
'pentium' and 'pentium_fpu' Automaton. Any idea on how to change that?



You don't want to, no matter which default x86 binary you create it is still capable of creating tuned binaries for i386, i486, pentium, k6, pentiumpro, athlon, and pentium4 cores as well as all their deriatives.


Due to their architecture only the Pentium and Athlon pipes require the specific pipe modeling by that step that you are seeing during the build.

I am a bit lost. If each architecture requires a specific pipeline model, and I want to select the
Athlon one, can/how I do it?



Please followup to gcc-help@gcc.gnu.org


Kelley Cook



Regards,
Liviu


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]