[PATCH, committed] Fix PowerPC long double bootstrap

[Joseph S. Myers]

On Sat, 20 Jan 2007, David Edelsohn wrote:

> >>>>> Joseph S Myers writes:
> 
> Joseph> The whole of libgcc on PowerPC GNU/Linux is meant to be built with 
> Joseph> -mlong-double-128, whether or not glibc supports it, so that programs can 
> Joseph> be built with this option as long as they don't use libc/libm/libstdc++ 
> Joseph> long double functions without libc support.
> 
> 	This is an assertion.  Why?

I do not believe any of the principles have changed since the previous 
thread <http://gcc.gnu.org/ml/gcc-patches/2006-01/subjects.html#01563>.  
Building with -mlong-double-128 makes some additional functions to support 
TFmode available in libgcc.  It does not affect the functions for DFmode 
etc. that are present in any case.

The additional functions only cause problems if they contain undefined 
references to functions not in libgcc.  That was why Jakub restricted 
-mlong-double-128 to hard-float 
<http://gcc.gnu.org/ml/gcc-patches/2006-01/msg01849.html>.  Now soft-float 
long double support is present, the reason for this restriction no longer 
applies.

> 	Patches welcome.

I will remove those soft-float checks in t-ldbl128 in a resubmission of my 
E500 long double patch to apply to current mainline, which will remove the 
last case where long double support can lead to broken libgcc with 
undefined references.

-- 
Joseph S. Myers
joseph@codesourcery.com