[PowerPC] PR26459 again

[Alan Modra]

On Thu, Mar 30, 2006 at 04:50:32PM -0500, David Edelsohn wrote:
> 	What bothers me about this change is that it converts a test that
> does not need to call reg_classes_intersect_p() into one that does.  This
> seems to be slowing down the common case for no good reason.  The first
> test does not just cover floating point modes but integer modes as well.
> That shortcut seems to have disappeared in your change.

Well, we do need an extra call for TARGET_E500_DOUBLE and TARGET_SPE
of course.  After all, removing the "shortcut" in that case was the
whole point of the change!  I assume you're not talking about that.

However, I can't see any other case where reg_classes_intersect_p is
called now where it previously wasn't.

-- 
Alan Modra
IBM OzLabs - Linux Technology Centre