PATCH: PA PIC symbolic memory fix PART 1

[Jeffrey A Law]

  In message < 200102200421.XAA01885@hiauly1.hia.nrc.ca >you write:
  > < http://gcc.gnu.org/ml/gcc-bugs/2000-12/msg00115.html >.
  > 
  > > You also don't explain why IS_RELOADING_PSEUDO_P isn't needed.  It was
  > > put in there for a reason, you should explain why it isn't needed if
  > > you're planning to take it out.
  > 
  > My understanding is that reload does re-recognize insns (see comment
  > at line 1168 in reload1.c).
Only in special circumstances.  In fact, if it had consistently 
rerecognized insns, then we wouldn't have had the bug which ultimately
caused me to merge the PIC LO_SUM pattern with the general movsi/movid
patterns.

  > indicate that this is in fact the case.  IS_RELOADING_PSEUDO_P was
  > apparently added to handle the fact that reload didn't re-recognize
  > the insns that it generates causing constrain_operands to fail.
And I believe it is still very necessary.

  > removed but I don't have a 2.0 machine available for testing.  Basically,
  > I am skeptical of the code because it is clear that PIC code generation
  > on pa1.1 machines was never well tested.
Actually it was, better than quite a few systems.  It's built some pretty
big codebases.

The problems we've been having with PIC on the PA are directly related to
update_equiv_regs getting smarter about finding equivalences than it used
to be.

jeff