outstanding regressions

[David Edelsohn]

> powerpc-eabisim g77.sum g77.f-torture/execute/19981119-0.f
> 
> is due to rth's changes of 2002-07-21;

	This is some sort of dependency problem.  It was on my list to
help investigate.  It only fails with inlining.

The Fortran is

      N=10
      call sub(N,a)
      if (a(-N,N) .ne. 0d0) call abort

where subroutine "sub" sets element a(-10,10) to 0d0.  The relevant part
of the PowerPC assembly output is:

        lfd 13,LC..3(2)		; load constant 0d0 into $f13
        lfd 0,3360(9)		; load a(-10,10) into $f0
        stfd 13,3360(9)		; store 0d0 into a(-10,10)
        fcmpu 0,0,13		; compare 0d0 with old value of a(-10,10)

I know I left my causality around here somewhere...

David