FIRST_PARM_OFFSET() depending on "frame_pointer_needed" and "regs_ever_live[]". Can it?

["Mike Laman" <mlaman at cox dot net>]

Hi,

I've been working on a port for almost three months now (3.1 at this time)
of evenings when I can squeeze in the time. I've read the documentation
(a bit like drinking from a fire hose :-)), and I have managed to get
through
a lot of issues. As challenging and difficult as this has been, I am having
fun...
I have had my compiler generating code for a month, but now I'm working on
getting it
to generate the correct code :-).

I'm working on the function call logic (so it will work and be efficient in
space(!)).
I'm trying to let the compiler use only the stack pointer if it needs to
access the
arguments and local variables. Therefore I'm trying to let the compiler
generate
code to access the arguments and local data (well, everything) from the
stack pointer.

The problem I'm having (which could very well be self induced), is that
FIRST_PARM_OFFSET()
is sometimes getting values of  "frame_pointer_needed" and
"regs_ever_live[]"
that are different than when my prologue and epilogue generating functions
are called.
Consequently, the offsets from the stack pointer to the arguments and local
variables
are wrong. (I'm working on code generated with and without the frame pointer
being omitted).

Is there something else I'm supposed to define? Am I even allowed to use the
"frame_pointer_needed" and "regs_ever_live[]" in FIRST_PARM_OFFSET()?

I will be very disappointed if I am required to define/use an argument
register, but I'm not
seeing any way out of it right now, so I thought I'd check with the group to
see if
I am developing the code in an accepted direction.

I'm sorry to bother the group. I've tried very hard to answer all my own
questions,
and I certainly have answered many to get to this point. But I think I'm at
to point
on this one that I should get some guidance.

Thanks in advance,
Mike