ia64 ltoff22x/ldxmov relaxation

[David Mosberger]

  Rich> Which brings me to my next point.  David, is there any *good*
  Rich> reason to have this 2**61 byte separation?  If we did not have
  Rich> it, then the 4MB block spanned by the gp would cover at least
  Rich> part of the .rodata section on many programs, which would
  Rich> allow us to optimize address loads of string constants and the
  Rich> like.

The main reason to keep it is to preserve the ability to change the
kernel's VM system in the future.  But if we can change the layout to
(almost) contiguous without breaking existing binaries, I suppose we
don't really lose any flexibility (since we could change it back
again, if necessary).

	--david