On 06/01/2011 06:06 PM, Georg-Johann Lay wrote: > From the internals description, HARD_FRAME_POINTER_REGNUM appears to > serve different purpose, and sources indicate that it is used similar, > i.e. per regno == HARD_FRAME_POINTER_REGNUM instead if having a rtx or > reg_class and test for overlaps.
Yes, but the difference is that if you define things this way, you can use HARD_FRAME_POINTER_REGNUM as a fairly normal register, as it appears only during IRA/reload. FRAME_POINTER_REGNUM however only occurs before (and during..) reload, and you can treat it specially since it isn't a real hard register. > Do you know of a target where the frame pointer spans several > word_mode registers? No, I think you're pretty unique there. So you may still run into some problems, but I think on the whole this setup would be worth a try. Bernd
