Re: Floating Point Support

[Jeff Johnston <jjohnstn at redhat dot com>]

Dietmar Ernst wrote:

> Hi,
>
> I am using newlib-1.10.0  for an AT91M55800A (ARM7 Core), my compiler is
> gcc-3.2.1 for Cygwin. I've some problems when performing floating point
> operations, for example "strtod". The conversion seems to work fine, but suddenly it
> jumps to label "undfl" and leaves the functions with 0x0. There's not much
> information available for this problem. Maybe it has something to do with
> option "-msoft-float"?
>
> Dietmar
>  
There has been no strtod problem like this that I am aware of.  Can you 
give details of how you configured?

Some possible problems could be:

 1. You did not build gcc and newlib in the same build directory.  When 
multilibs are involved,
      it is best to build gcc and newlib in the same place.  When you 
specify a compile with link,
      gcc will link in the correct multilib.  Otherwise, you have to 
manually specify which multilib
     to link in and if you get it wrong, undefined behavior results 
(e.g. mixing up little-endian and big-endian).

 2. Another possibility is that you have a compiler bug.  Is is 
possible for you to try a different compiler?
     If you still have problems with newlib, I suggest trying newlib-1.11.0

-- Jeff J.