egcs-1.1.1 issues; Fortran

[David Edelsohn]

>>>>> Jeffrey A Law writes:

>> On the 7th of September Marc Dejardin <Marc.Dejardin@cern.ch>  
>> posted a report on an "internal error--insn does not satisfy its  
>> constraints" when compiling a routine from CERNLIB on a powerpc  
>> based machine.
>> 
>> Looks like a powerpc backend problem.

Jeff> No clue where we stand on this one.

	This looks like another PIC/GOT problem.  movdf is trying to
reference a symbol described by a LO_SUM.  I suspect that the "address"
needs to be split / legitimized somehow.  Probably movsf and movdf
define_expand patterns need some of the mechanism employed in movsi for
handling these compound addresses, but I am not sure.

David