Public SourceMage Discussion List

[sergey AT optimaltec.com]

OK, here's what is happening. GCC needs a working compiler to create
<bld>/gcc/cc1 binary. This working compiler in my case happens to be GCC 3.4,
so I have to use `-march=k8 -m64'. After the make process is done with cc1, it
apparently tries to use cc1 to build some other stuff, such as libiberty in
this case, I think. So make is passing the same switches to cc1. But the
problem is that GCC 3.3 doesn't support -march=k8 or -march=athlon64. It does
support `-march=athlon-xp m64' though. So the challenge is, how do I swap k8
with athlon-xp right before it starts doing make target named
"configure-target-libiberty." Ideas are very welcome!

Quoting sergey AT optimaltec.com:

> All,
>
> I need some help. I can't build gcc33 on amd64 to be able ton run perforce
> client. It goes through stages, than says something like "Configuring in
> x86_64-unknown-linux-gnu/libiberty..." and quits saying that
> gcc-3.3.5.bld/gcc/xgcc cannot build executables. In the
> gcc-3.3.5.bld/x86_64-unknown-linux-gnu/libiberty/config.log, the last error 
> I
> can see is this:
>
> cc1: error: bad value (athlon64) for -march= switch
>
> It was like this with -march=k8 (for GCC 3.4), so I replaced k8 with
> athlon64.
> But the same error (above) is produced anyway. Search for "cc1 bad value
> athlon64 for march switch" produces measly 7 hits none of which help. All I
> need is a simple thing called /usr/lib/libstdc++.so.5...