Sep 9 change in gen_lowpart_common causes aborts

[Jeffrey A Law]

  In message <Pine.SOL.3.90.980930155848.18576K-100000@matula.informatik.rwth-aachen.de>you write:
  > The following change
  > 
  > Wed Sep  9 23:31:36 1998  (Stephen L Moshier) <moshier@world.std.com>
  > 
  >         * emit-rtl.c (gen_lowpart_common): Disable optimization of
  >         initialized float-int union if the value is a NaN.
  > 
  > causes current egcs snapshots to abort when trying to compile the 
  > math tests in glibc-2.0.96 (on i586-linux).  Below is a small test
  > program that demonstrates the problem.  Basically, reload encounters
  > a SUBREG:SF of a SImode pseudo that is equivalent to a CONST_INT and
  > aborts in find_reloads_toplev because gen_lowpart doesn't handle it
  > anymore.
  > I can't really say what a correct fix would be, since I don't know
  > why the above change was necessary.
Bummer.  It was a bug & fix from the gcc lists.  Stephen, do you still have
the testcase which led to this change?

It's easy to forget that gen_lowpart_common must not fail for integers anymore
due to changes in how reload works.

methinks this problem needs to be fixed elsewhere.

jeff