Doc for -fcall-used-REG and -fcall-saved-REG

Doc for -fcall-used-REG and -fcall-saved-REG

[Gavin Romig-Koch]

From invoke.texi, for -fcall-used-REG and -fcall-saved-REG:

   Use of this flag for a register that has a fixed pervasive role in the
   machine's execution model, such as the stack pointer or frame pointer,
   will produce disastrous results.

I'm writing a patch to have the compiler emit an error message 
if these options are used on the frame pointer or the stack pointer,
and was going to update the doc to reflect this, so some thing 
like:

   Use of this flag for the stack pointer or the frame pointer is
   an error.

My question is whether we need to keep the above comment for 
other fixed pervasive role registers (other than the sp and fp);
something like:

   Use of this flag for the stack pointer or the frame pointer is
   an error.  Use of this flag for a register that has a fixed 
   pervasive role in the machine's execution model, other than 
   the stack pointer and frame pointer, will produce disastrous
   results.

Or wether we know that there are no other such fixed registers,
and can drop that second sentance?  Does anyone know of other
fixed pervasive role (perhaps machine specific) registers  
exist?

I think the "will produce disastrous results" thing is a cop out,
and I'd like to replace it with an actual real life error message,
if we can hunt down all the bad cases.

                                         -gavin...

Re: Doc for -fcall-used-REG and -fcall-saved-REG

[Nick Ing-Simmons]

Gavin Romig-Koch <gavin@cygnus.com> writes:
>
>   Use of this flag for the stack pointer or the frame pointer is
>   an error.  Use of this flag for a register that has a fixed 
>   pervasive role in the machine's execution model, other than 
>   the stack pointer and frame pointer, will produce disastrous
>   results.
>
>Or wether we know that there are no other such fixed registers,
>and can drop that second sentance?  Does anyone know of other
>fixed pervasive role (perhaps machine specific) registers  
>exist?

On my (TI internal) port of GCC2 to TMS320C80's RISC, there is a "return 
address" register which is fixed-and pervasive in same manner. (Which register
is used does not matter but it has to be "pervasive" at least between caller and 
callee!, and it is easier to fix it as part of ABI.)

In fact in that port it is quite permissible to re-use frame pointer in any 
function which does not need alloca() style dynamic frame size.  

-- 
Nick Ing-Simmons <nik@tiuk.ti.com>
Via, but not speaking for: Texas Instruments Ltd.

Re: Doc for -fcall-used-REG and -fcall-saved-REG

[Jeffrey A Law]

  In message < m0zTED4-000s8JC@cetus.cygnus.com >you write:
  > My question is whether we need to keep the above comment for 
  > other fixed pervasive role registers (other than the sp and fp);
  > something like:
  > 
  >    Use of this flag for the stack pointer or the frame pointer is
  >    an error.  Use of this flag for a register that has a fixed 
  >    pervasive role in the machine's execution model, other than 
  >    the stack pointer and frame pointer, will produce disastrous
  >    results.
  > 
  > Or wether we know that there are no other such fixed registers,
  > and can drop that second sentance?  Does anyone know of other
  > fixed pervasive role (perhaps machine specific) registers  
  > exist?
argument & return value registers come to mind.  Registers which are in a class
by themselves (think HILO on mips, shift count registers and the like).  The
return pointer may be another worth worrying about.


jeff