processor compatibility when building packages

Danny Smith danny_r_smith_2001@yahoo.co.nz
Sat Oct 13 17:33:00 GMT 2001


 --- Christopher Faylor <cgf@redhat.com> wrote: > On Sun, Oct 14, 2001 at
08:22:58AM +1000, Danny Smith wrote:
> >This topic came up about a year ago on this list: 
> > http://sources.redhat.com/ml/cygwin-apps/2000-07/msg00039.html
> >
> >and I've recently raised it again on mingw lists.
> >
> >Looking at cygwin's spec file it says:
> >
> >
> >*cpp_cpu_default:
> >%(cpp_686)
> >
> >
> >To me that means that, unless user overrides with command line -march or
> >-mcpu option, code is tuned for i686.  To get that into the specs file,
> one
> >has to either configure gcc so that TARGET_CPU_DEFAULT=3 (=pentium) or
> pass
> >that define to make when building gcc.
> 
> I do neither of those things when configuring gcc for cygwin.
> 
> I do configure everything as i686-pc-cygwin, though.
> 
> >That interpretion is different from the conclusion that was posted in
> the
> >message pointed to above, which says that cygwin-produced code will run
> on
> >any i[35467]86 arch.  
> >
> >I'm not talking about the platform triplet moniker -- that I realise is
> >just a name.  I'm asking: Has anyone complained that cygwin-built apps
> fail
> >on i386 but not on i686? 
> 
> Nope, never.
> 
> cgf 

Thanks.  I must be misreading what "cpp_cpu_default" does then.  If
configure without the cpu part of the triplet, that spec is left undefined,
unless I do the defines I mentioned above.  Ahh, maybe it just affects
*scheduling* of code rather than code generation, ie. (-mcpu=i686)!=
(-march=i686).  

Danny

 

http://briefcase.yahoo.com.au - Yahoo! Briefcase
- Manage your files online.



More information about the Cygwin-apps mailing list