Building newlib without -mhard-float

Jeff Johnston jjohnstn@redhat.com
Wed Sep 5 22:45:00 GMT 2007


Rick Mann wrote:
> 
> On Sep 5, 2007, at 12:27 PM, Jeff Johnston wrote:
> 
>> My suggestion would be to back off this idea and simply build gcc and 
>> newlib without tinkering (disabling multilib and using TARGET_CFLAGS 
>> is tinkering).  This should give you all the possible arm-elf 
>> permutations and the various libraries will be in sync (i.e. newlib 
>> generated code won't reference any function that isn't found in 
>> libc/libm, libgloss, or libgcc).  Then, try and find one set of 
>> options that works for your target platform (e.g. arm-elf-gcc -g 
>> -mcpu=xscale -msoft-float test.c).  If none of the permutations work, 
>> then it means you need to add one to gcc's list so that you get a 
>> libgcc and newlib/libgloss that are in sync and use the options you 
>> want.  Adding a multilib is straightforward and has been done before.
> 
> Okay. That brings up the stage 1/2 approach someone else suggested I 
> use. A more recent post suggested I not do that, but I wanted to get 
> confirmation.
> 
> Since I need to rebuilt GCC (to enable multilib), I want to know: Can I 
> just build, *from scratch*, binutils, gcc, and then build newlib? Or 
> will I face some obstacles in doing so? That is, can I delete (or hide) 
> my existing /usr/local/arm directories and build everything in one pass?
> 
> When I build things, should I limit myself to --target=arm-elf and 
> --with-newlib?
>

That would be my starting point.

> I'll try this, but I'm not sure it will work. However, the thing that 
> originally motivated so much of this (getting VFP into all the built 
> pieces because of a single lib.a that I couldn't rebuild) may not be a 
> problem much longer.
> 
> 
> Thanks!
> 
> --Rick
> 
> 



More information about the Newlib mailing list