[Patch, fortran] PR34861 and PR34854 - committed as obvious

[Paul Richard Thomas]

Talk about pride before a fall.....

PR34854 - I completely forgot, when reworking this, to include the
case of a renamed variable that had been previously loaded without
renaming and without an ONLY clause.

This caused a regression in libgomp/reduction_5.f90, which has become:

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34896

Thanks to HJ Lu for reporting this so quickly.  I'll fix it today.

Paul