newlib "light!

Wolfgang Fritz
Wed Feb 27 10:58:00 GMT 2002

[CC: to you, my mail2news gateway seems to suppress messages

"J. Johnston" schrieb:
> Wolfgang Fritz wrote:
> >
> > Hi,
> >
> > I am developing communications applications on a 68LC302 platform using
> > the gcc toolchain and newlib. I've seen the binaries are much bigger as
> > the same application generated with a commercial toolchain. This is not
> > a compiler but a newlib issue. The commercial library does not have
> > floating point support, which is not needed in my application. (They
> > have a floating point version too).
> >
> > My question: Is it possible to strip down newlib to a  pure integer
> > version (printf etc.)? A short look into the sources showed some
> > #defines which may imply that it is possible, but I could not find
> > options to configure it. The newlib sources are a real jungle for me.
> >
> > Is there any documentation available how to tailor newlib?
> >
> There is no documentation in this respect other than what's in the code.
> I would suggest an experiment.  Try just #define'ing the various printf routines
> to point to their iprintf counterparts which are built without floating point
> support.

I tried that today. Here the results:

 printf: .text: 88f0 .data: 0778 .bss 21b0
iprintf: .text: 4ccc .data: 075c .bss 21b0
                3c24        001c            = 3c40 (15424 dec) bytes

That is comparable to the commercial toolchain. The floating point
routines are not loaded any more. That is what I wanted. Thanks.

> See how that lowers your application size.  If it solves your problem I can tie
> it
> to a configuration option.  If you are using scanf routines, there is a
> flag you can specify when building vfscanf.o.

Up to now I have no scanf in my application, but I will try your
suggestion if I have some spare time (but that won't be the next days,
because I expect my hardware prototypes tomorrow...)

Thanks again

> -- Jeff J.

More information about the Newlib mailing list