rs6000 3.1 Ada bootstrap problem

[David Edelsohn]

>>>>> Richard Kenner writes:

Richard> Well, the fundamental problem here is that we have a configuration
Richard> where the sizes are a function of the -m values.  So gnatpsta (which
Richard> is just a utility to print sizes and such) doesn't really have a good
Richard> way of outputting the "right" value.  Moreover, you also don't want to
Richard> try to pull in everything involved in computing that value.

Richard> This needs to be kludged somehow to return a "reasonable" value (i.e.,
Richard> the one with no -m switches), but it's not at all clear what the best
Richard> way to do that is given the structure of the rs6000 config files.

	The other values work because they rely on target_flags.  The
problem for rs6000 is that we ran out of bits in target_flags.

	Long double support currently is experimental, so I can set the
Ada macro to 64 which is correct for now.  But that fixes gnatpsta while
hurting other uses of targtyps.c which correctly link it with $(BACKEND).

David