Dominik Vogt <v...@linux.vnet.ibm.com> writes:
> With that problem fixed I still see one minor glitch. Maybe
> someone knows how to fix the following:
>
> * With a cross compiler that generates i686 binaries on s390x:
>
> $ i686-elf-gcc -c ~/foo.c -march=native
> /home/vogt/foo.c:1:0: error: bad value (native) for -march= switch
>
> This is all right because the x86 compiler just emits a brief
> error message because the argument to -march= is a string.
>
> * The other way round, generating s390x binaries on i686:
>
> $ s390x-linux-gcc -c ~/foo.c -march=native
> cc1: error: unrecognized argument in option '-march=native'
> cc1: note: valid arguments to '-march=' are: g5 g6 native z10 z13 z196
> z9-109 z9-ec z900 z990 zEC12
>
> So, the error message complains about "native" not being a valid
> argument to -march=, and then lists it as valid in the next line.
> This is because s390x uses an Enum option, and "native" is one
> of the allowed values.
Nice spot :-) One way would be to have:
%{mtune=native:%e-mtune=native is only supported for native configurations}
in the non-native MARCH_MTUNE_NATIVE_SPECS, and similarly for -march.
The problem is that that won't quote -mtune=native in the canonical
way and I don't know how well it would handle intl (probably not very well).
Maybe we should have a new .opt tag for native-only enum options.
Thanks,
Richard
