Re: Return value address

Nicola.Musatti@ObjectWay.it (Nicola Musatti) wrote in message news:04-09-100...
> "Christoph Neubauer" <christoph.neubauer@siemens.com> wrote
> > What about a stack model, that allocates:
> >
> > r allways at 0
> > arguments starting at 4
> > local variables afterwards
> > temp variables afterwards
>
> In my ignorance I thought that just about every compiler worked in
> this way, except maybe when return value and arguments are directly
> allocated into registers. Are there reasons to do otherwise?


If you can use registers instead of a stack, that speeds up things.
MIPS (and PowerPC) has 4 registers for arguments (and the rest go on
stack). The argument registers are not used for any other purpose
[i.e. save them before calling another procedure]. When a procedure is
called, it saves the registers it wants to use onto stack and then
uses it -and that way, you have the speed of registers and don't end
up creating copies of unused registers.


> > [It also seemed to me that the obvious solution is for the caller to
> > allocate the return value cell. -John]


Yes -the caller creates space for the return value and writes a return
address onto stack, and the callee writes the return value and does a
jmp to the specified return address.


> Wouldn't this also give the opportunity to optimize it away, e.g. by
> directly using the variable to which the function result is assigned?


Not sure I understand. Could you also clarify whether the compiler is
going to generate binary machine code or something else?


thsnks
-kamal

Return to the comp.compilers page.
Search the comp.compilers archives again.