> In reading the gcc man page, and information from, say, AMD, I get
> the impression that gcc can generate executables on an x86-64 box
> that will run on a 32 bit machine.  But I can't make it work.

> I'm using SuSE 10.3 on a "Intel(R) Core(TM)2 CPU          4400  @ 2.00GHz
> stepping 02".

> The gcc man page says:

> -march=cpu-type
>   Generate instructions for the machine type cpu-type.  The choices for
> cpu-
>   type are the same as for -mtune.  Moreover, specifying -march=cpu-type
>   implies -mtune=cpu-type.

> Trying -march=i386 gives:

>  cc -g -I/home/mellman/src/ulib -o cross.o -march=i386 cross.c
>  cross.c:1: error: CPU you selected does not support x86-64 instruction
> set
>  cross.c:1: error: CPU you selected does not support x86-64 instruction
> set

> Right.  That's the point...

> And -march=i686 gives:

>  cc -g -I/home/mellman/src/ulib -o cross.o -march=i686 cross.c
>  cross.c:1: error: CPU you selected does not support x86-64 instruction
> set

> (just once this time)

> On a lark (because I don't find an explanation in the man page of what
> the difference with -march is), I tried -m32:

> Here I get:

>  cc -g -I/home/mellman/src/ulib -o cross.o -m32 cross.c
>  In file included from /usr/include/features.h:345,
>               from /usr/include/stdio.h:28,
>               from cross.c:1:
>  /usr/include/gnu/stubs.h:7:27: error: gnu/stubs-32.h: No such file or
> directory

> stubs-32.h is not on my system, and I can't find it on SuSE's site
> either.

> I find almost nothing about this when googling.  WRT stubs-32.h, the
> only thing I can find says to use different libraries (!).  The response
> is always that that didn't help.

> It would seem that this is a pretty common requirement - I have a 64 bit
> box at home and want to distribute my programs generally (like to my
> virtual server, which only has 32 bits).  The lack of general interest
> here must indicate that I'm completely on the wrong track somehow...

> Any ideas are appreciated...

>> Hi.

>> In reading the gcc man page, and information from, say, AMD, I get the
>> impression that gcc can generate executables on an x86-64 box that will
>> run on a 32 bit machine.  But I can't make it work.

>> I'm using SuSE 10.3 on a "Intel(R) Core(TM)2 CPU          4400  @
>> 2.00GHz stepping 02".

>> The gcc man page says:

>> -march=cpu-type
>>   Generate instructions for the machine type cpu-type.  The choices for
>> cpu-
>>   type are the same as for -mtune.  Moreover, specifying
>>   -march=cpu-type implies -mtune=cpu-type.

>> Trying -march=i386 gives:

>>  cc -g -I/home/mellman/src/ulib -o cross.o -march=i386 cross.c
>>  cross.c:1: error: CPU you selected does not support x86-64 instruction
>> set
>>  cross.c:1: error: CPU you selected does not support x86-64 instruction
>> set

>> Right.  That's the point...

>> And -march=i686 gives:

>>  cc -g -I/home/mellman/src/ulib -o cross.o -march=i686 cross.c
>>  cross.c:1: error: CPU you selected does not support x86-64 instruction
>> set

>> (just once this time)

>> On a lark (because I don't find an explanation in the man page of what
>> the difference with -march is), I tried -m32:

>> Here I get:

>>  cc -g -I/home/mellman/src/ulib -o cross.o -m32 cross.c In file
>>  included from /usr/include/features.h:345,
>>                   from /usr/include/stdio.h:28,
>>                   from cross.c:1:
>>  /usr/include/gnu/stubs.h:7:27: error: gnu/stubs-32.h: No such file or
>> directory

>> stubs-32.h is not on my system, and I can't find it on SuSE's site
>> either.

>> I find almost nothing about this when googling.  WRT stubs-32.h, the
>> only thing I can find says to use different libraries (!).  The
>> response is always that that didn't help.

>> It would seem that this is a pretty common requirement - I have a 64
>> bit box at home and want to distribute my programs generally (like to
>> my virtual server, which only has 32 bits).  The lack of general
>> interest here must indicate that I'm completely on the wrong track
>> somehow...

>> Any ideas are appreciated...

> install glibc-devel-32bit

> regards,
> lajos

>> On Wed, 26 Mar 2008, Charles T Smith wrote:

>>> Hi.

>>> In reading the gcc man page, and information from, say, AMD, I get the
>>> impression that gcc can generate executables on an x86-64 box that
>>> will run on a 32 bit machine.  But I can't make it work.

>>> I'm using SuSE 10.3 on a "Intel(R) Core(TM)2 CPU          4400  @
>>> 2.00GHz stepping 02".

>>> The gcc man page says:

>>> -march=cpu-type
>>>   Generate instructions for the machine type cpu-type.  The choices
>>>   for
>>> cpu-
>>>   type are the same as for -mtune.  Moreover, specifying
>>>   -march=cpu-type implies -mtune=cpu-type.

>>> Trying -march=i386 gives:

>>>  cc -g -I/home/mellman/src/ulib -o cross.o -march=i386 cross.c
>>>  cross.c:1: error: CPU you selected does not support x86-64
>>>  instruction
>>> set
>>>  cross.c:1: error: CPU you selected does not support x86-64
>>>  instruction
>>> set

>>> Right.  That's the point...

>>> And -march=i686 gives:

>>>  cc -g -I/home/mellman/src/ulib -o cross.o -march=i686 cross.c
>>>  cross.c:1: error: CPU you selected does not support x86-64
>>>  instruction
>>> set

>>> (just once this time)

>>> On a lark (because I don't find an explanation in the man page of what
>>> the difference with -march is), I tried -m32:

>>> Here I get:

>>>  cc -g -I/home/mellman/src/ulib -o cross.o -m32 cross.c In file
>>>  included from /usr/include/features.h:345,
>>>               from /usr/include/stdio.h:28,
>>>               from cross.c:1:
>>>  /usr/include/gnu/stubs.h:7:27: error: gnu/stubs-32.h: No such file or
>>> directory

>>> stubs-32.h is not on my system, and I can't find it on SuSE's site
>>> either.

>>> I find almost nothing about this when googling.  WRT stubs-32.h, the
>>> only thing I can find says to use different libraries (!).  The
>>> response is always that that didn't help.

>>> It would seem that this is a pretty common requirement - I have a 64
>>> bit box at home and want to distribute my programs generally (like to
>>> my virtual server, which only has 32 bits).  The lack of general
>>> interest here must indicate that I'm completely on the wrong track
>>> somehow...

>>> Any ideas are appreciated...

>> install glibc-devel-32bit

>> regards,
>> lajos

> Thank you.  That worked.  Actually, I got lost in the whirl of glibc-
> devel and glibc-32bit and missed that there was also a
> glibc-devel-32bit.

> Note for others: you will then have to install

> gcc42-32bit-4.2.1_20070724-17.x86_64.rpm libgomp (for 32 bits)
> libmudflap (for 32 bits)

> Now, is there an equally easy way to do the same thing for sparc? The
> last time I delved into that, it seemed like I'd have to build a special
> version of gcc ...

>> On Wed, 26 Mar 2008 15:07:45 +0100, FÖLDY Lajos wrote:

>>> On Wed, 26 Mar 2008, Charles T Smith wrote:
> make: *** [cross] Error 1

> > Now, is there an equally easy way to do the same thing for sparc?

> All recent versions of gcc, when configured and built on 64-bit
> Solaris/SPARC, support '-m32' (default) and '-m64' out of the box.

> Or is your question about Linux/SPARC?

> Cheers,
> --
> In order to understand recursion you must first understand recursion.
> Remove /-nsp/ for email.