uClibc patches for sh-2

[Paul Mundt]

On Mon, Oct 23, 2006 at 03:05:38AM -0400, Mike Frysinger wrote:
> On Monday 23 October 2006 00:09, Paul Mundt wrote:
> > I don't remember the rationale at the moment, SH-3 wanted it for some
> > reason. Presumably it was the same sort of scenario as ARM, using special
> > tuning at some point and simply dropping it. Anyways, we have FPU
> > emulation in the kernel these days (where SH-3 is the primary user), so
> > we don't need it there anymore.
> 
> eh ?  the trend is to move away from userspace softfloat ?  all those illegal 
> instruction traps in kernel space seriously kills performance ...

No, userspace soft-float is still the way to go, just no one has written
up the patch for it yet. On the other hand, hardfloat emulation gives us
some better ABI consistency between SH-3 and SH-4. Since these binaries
would hit an illegal instruction otherwise, we're not really "losing"
anything by trapping them and fixing them up, even if the performance
is sub-optimal.