reload displacement optimization

[Jeffrey A Law]

  In message < 19980402024528.15289@dot.cygnus.com >you write:
  > On Thu, Mar 12, 1998 at 04:39:15PM -0800, Jim Wilson wrote:
  > > I am willing to accept LEGITIMIZE_RELOAD_ADDRESS as a short term solution
  > > with documentation cautioning that it must be used carefully.
  > 
  > I've cleaned up the documentation a bit and committed the generic
  > bits and my Alpha bits.  I did not attempt commentary on the x86
  > or sh bits;  I'll leave that for Joern to take care of at his leisure.
Note the documentation says nothing about the need to call push_reload
on the address you actually want reloaded.

I also vaguely remember a question about any structure sharing
assumptions made by LEGITIMIZE_RELOAD_ADDRESS.  Do you need to
call copy_rtx on the address?

Any chance you could expand the docs?  Particularly on the need to
call push_reload.

Think about how to describe the actions that LEGITIMIZE_RELOAD_ADDRESS
must take if it finds a better way to reload an address.  This is the
kind of information that an implementor will need if they want to
define this macro for a particular target.

jeff