soft-fp extensions and sNaNs

[Joseph S. Myers]

On Fri, 11 Oct 2013, David Miller wrote:

> The new test added to math/basic-test.c since glibc-2.18 fails on
> sparc for the quad float cases that use soft-fp:
> 
> Failure:  long double x = (long double) (float) sNaN, !issignaling
> Failure:  long double x = (long double) (double) sNaN, !issignaling
> 
> Looking specifically at 32-bit sparc's q_dtoq.c it seems to be
> implemented identically to soft-fp/extenddftf2.c, so this appears to
> be a generic soft-fp issue.
> 
> Joseph have you run into this already or know about it?

It's bug 16041.  It's on my list of soft-fp issues to address before 
attempting adding any features from the kernel version, but as by now I 
have six soft-fp patches outstanding pending review 
<https://sourceware.org/ml/libc-alpha/2013-10/msg00329.html> 
<https://sourceware.org/ml/libc-alpha/2013-10/msg00338.html> 
<https://sourceware.org/ml/libc-alpha/2013-10/msg00351.html> 
<https://sourceware.org/ml/libc-alpha/2013-10/msg00393.html> 
<https://sourceware.org/ml/libc-alpha/2013-10/msg00365.html> 
<https://sourceware.org/ml/libc-alpha/2013-10/msg00399.html>, I'm holding 
off on implementing more soft-fp patches until the backlog is reduced to 
avoid needing to track too many potential dependencies on previous 
patches.

-- 
Joseph S. Myers
joseph@codesourcery.com