Re: RFA: xscale-elf -mstrict-prototypes

[James Lemke <jim at wasabisystems dot com>]

Sorry, schedules kept me from replying for a bit.

On Sun, 2004-02-22 at 16:50, Mark Mitchell wrote:
> James Lemke wrote:
> 
> >Given this source (gcc/testsuite/arm-g4-1.c):
> >short int addhi3 (short int a, short int b) __attribute__ ((naked));
> >short int addhi3 (short int a, short int b)
> >{
> >    return a + b + (short int)32;
> >}
> >
> >The assembler generated (xscale-elf-gcc -O2 -S) is rather poor:
> >
> I don't think we should add -mstrict-prototypes.
> 
> This problem is mostly a consequence of the old ARM ABI, and I think 
> will be fixed in the new ARM ABI.

I am trying to avoid the overhead of conditioning a shorter-than-int
parameter in the callee.  But there is a problem when a call is made to
a function like above without a prototype, and with an argument too
large for the receiving parameter.

For example:
int
main (int argc, char *argv[])
{
  return addhi3 (40000, -50000);
}
This call will pass out of range values to addhi3 which is incorrect but
may happen in existing code.

With -mstrict-prototypes addhi3 will not truncate and sign-extend the
values before use, so they will remain out of range.  This may introduce
additional undesired behaviour, such as an exception that didn't occur
before.

Therefore, I added this functionality under control of an option.

> And, if we are going to add it, it should be a -f option and applicable 
> to all architectures.
This may make sense, I haven't looked at the issue on other
architectures.

Jim.

-- 
James Lemke   jim@wasabisystems.com   Orillia, Ontario
http://www.wasabisystems.com