reload vs adds with clobbers

[Richard Kenner]

    > The way I've fixed it is to prevent gen_reload from ever taking that
    > path by marking all PLUS expressions that would otherwise go there
    > as needing a secondary reload (via SECONDARY_INPUT_RELOAD_CLASS) and
    > doing the right thing in reload_insi ...
    
    There's got to be a better way to handle this situation.  Why can't
    reload just use the right expander?

What's "right"?  If an addition needs a scratch register (it was said
that it had a CLOBBER), the only way to produce that is via the secondary
reloads.  The approach above sounds exactly correct.