An alpha patch for egcs 1.1

[Richard Henderson]

On Sat, Aug 15, 1998 at 10:01:13PM -0700, H.J. Lu wrote:
> While we are on it, I have an impression there may be a prolem in all
> ASM_OUTPUT_MI_THUNKs. Should we check all ASM_OUTPUT_MI_THUNKs to make
> sure thay are ok? We know x86 and alpha are fine. How about others?

Probably a good idea.  I don't know enough about anything but Sparc
to comment.  Sparc32 is fine, as the aggregate return pointer is
hidden in a special stack slot. 

Sparc64 is not, as aggregates larger than 32 bytes are returned via
an in-band argument register pointer just as on Alpha.   This should
not hold up egcs 1.1, as Sparc64 is not really supported well at the
moment anyway.  

Incidentally, we should have a variant of one of the thunk tests that
has a larger class size.  The existing tests have size 16 on sparc64,
which would be returned in registers and so not show the bug.


r~