new g77 x86 ICE in 19990328

[Jeffrey A Law]

  In message < 14088.38333.660804.262809@ongaonga.elec.canterbury.ac.nz >you writ
e:
  > Jeffrey A Law writes:
  > 
  >  > Michael Hayes (privately) noted one of the bugs potential bugs in this c
  > ode
  >  > REGNO (op) + i might not be a hard register due to the same problem.
  > 
  > However, I can't see that this should be a problem unless the MD is
  > screwed up.  It would be a simple matter to check HARD_REGNO_NREGS to
  > be sure.
It can certainly be a problem.

Consider if we wanted subword 2 of the last FP register on the x86.  We'd
peek a HARD_REGNO_MODE_OK for something that wasn't a hard register, which
is bad.  Depending on the value we found, we might then return the first pseudo
register, which is also wrong.


jeff