Re: Followup for reg_equiv_invariant patch: Fix PR39871

[Joern Rennecke <amylaar at spamcop dot net>]

Quoting Jeff Law <law@redhat.com>:

> On 06/14/10 20:35, Joern Rennecke wrote:
> > Quoting Bernd Schmidt <bernds@codesourcery.com>:
> >
> > > Joern, any comments - do you recall any reason why this change would
> > > have been intentional?
> >
> > What happens if we have a (plus (REG:SI SP_REG) (symbol_ref foo)) ?
> > Or will this never happen for flag_pic?
> I can't see how this would ever be valid when flag_pic.

The question is if such invalid expressions might be in notes at these
points; if that might be the case, the code needs to reject them.

function_invariant_p will accept them.  An argument can be made that it
should reject them, but should it then be renamed to
function_invariant_and_suitable_for_pic_if_the_latter_matters_p or
rematerializable_invariant_p ?
The reason why I missed the issue with constants passed to  
LEGITIMATE_PIC_OPERAND_P back in 1998 was that I assumed the macro name
was descriptive and didn't check the documentation in tm.texi.