stack pointer adjustment problem

[Joern Rennecke]

> the reason i want to do such a thing is this: i've been working on
> implementing proper tail calls.  to this end i had to implement a new
> calling convention in which the callee pops the arguments.  on most
> risc machines, like the alpha, space for outgoing arguments is
> allocated in the prologue of the caller.  now, if a function with the
> new calling convention is called (without the call being a proper tail
> call), the stack pointer is modified when the function returns.  this
> modification (the popping of the arguments by the callee) has to be
> reverted.

How about explicitly setting the stack pointer to ints new value in
the call insn?
E.g. for call_value:

(call_insn (parallel [(set (reg foo) (call ...))
                      (set (sp) (plus (sp) (const_int bar)))]))