Re: volatile function

[David Brown <david at westcontrol dot com>]

On 25/01/2011 10:37, Cedric Roux wrote:
> On 01/25/2011 09:04 AM, kevin diggs wrote:
> > Hi,
> >
> > The function in question is returning the value of the PLL config
> > register in a PowerPC 750GX. I wanted to make sure that the optimizer
> > does not move or eliminate the call. I thought that is one of the uses
> > of the volatile qualifier???
> >
> > Thanks!
> >
> > kevin
> >
> > > > static inline volatile unsigned int get_PLL(void)
> > > > {
> > > > unsigned int ret;
> > > >
> > > > __asm__ __volatile__ ("mfspr %0,%1":
> > > > "=r"(ret):
> > > > "i"(SPRN_HID1)
> > > > );
> > > >
> > > > return ret;
> > > > }
>
> I think that SPRN_HID1 has to be declared as volatile
> (if that has a meaning at all in your case) for gcc
> to force the load in all cases (my .02 euros, I don't
> know your target asm, I don't know what SPRN_HID1 is).

It's a PowerPC, and SPRN_HID1 is a constant (the "mfspr" instruction 
reads from a specified "special purpose register").

Drop the "volatile" on the function declaration, and then code is then 
correct.  It is the "__volatile__" in the assembly statement that 
ensures the read cannot be eliminated.