Richard Earnshaw wrote:
> 
> fnasser@redhat.com said:
> > Humm..., I am having second thoughts about this.  Isn't the problem
> > you  are seeing the same problem of not having the values peoperly
> > sign-extended?
> 
> No.  In this case we really need to copy the least significant 1 (or 2)
> bytes into the 1 or 2 bytes in the valbuf target.  That means doing a copy
> from the higher addresses.   So in that respect, the patch is correct.
> 
> But it breaks the case where the return value is more than one word.

Yes, I see that now.  Richard, how about a joint effort?
Would you be so kind as to fill in the empty else clause?