another ia64 (and maybe generic) vector mode issue

[Jan Beulich]

>Yes of course it's dangerous.  Which is why it has been brought
>up on this list before that the only safe thing to do is to
>*always* ignore the mode of the argument, and rely only on the
>type node.

So it is just on oversight/ommission that this isn't done so in
ia64.c?

>You havn't said, however, what Intel does with their vector
>parameters.

They pass them in integer registers, and also return them there.
They're not supporting anything but (in gcc notation) V4SFmode at
present, though. I opened an issue with them to clarify whether they
consider their compiler behavior ABI compliant.

Jan