So we are not forced to use the generic be_Spill, be_Reload functions
which don't work correctly with double-width register requirements.
This should really fix sparc float spill/reload this time.