Problem with stack-adjustment code

[Jan Hubicka]

Oh, I should have noted, that after Richard's combined patch I believe thinks
works now correctly, so the rest of discussion is just about redundancy of code
in compute_argument_block_size and Mark's new function (I forgot its name
sorry). That is perhaps just aestetical. I am not aware of more bugs in
this area except of the problems Jeff is seeing on PA64.

Also I like Mark's code better than the one in compute_argument_block_size
so I would like to end up in this code moved to compute_argument_block_size
and with the alignment emitting code using similar way to combine adjustments
as it did so far (substract only multiples of PREFERRED_STACK_BOUNDARY
from the pending_stack_adjust).

Richard: Thanks for fixing the argument block allocation bug too. I was
aware of it and I believe I was sending patch to fix that, but perhaps
it is just waiting in the queue. Anyway it is nice to see it fixed.

Honza