[avr-gcc-list] Re: optimizer removes volatile pin access code.why?

[David Brown]

Weddington, Eric wrote:
> > -----Original Message----- From: 
> > address@hidden 
> > [mailto:address@hidden org]
> > On Behalf Of Joerg Wunsch Sent: Sunday, November 01, 2009 11:25 PM 
> > To: address@hidden Subject: Re: [avr-gcc-list] Re:
> > optimizer removes volatile pin access code.why?
> >
> > David Brown <address@hidden> wrote:
> >
> > > It might be worth filing a report for a better clarification in
> > > the documentation (does that need a bug report?) to make it
> > > clearer that "always_inline" has no effect without a normal
> > > "inline" declaration.
>
> This is what the GCC 4.3.2 manual says:
>
> always_inline Generally, functions are not inlined unless
> optimization is specified. For functions declared inline, this
> attribute inlines the function even if no optimization level was
> specified.
>
> So it does say "for functions declared inline". With this
> description, the "always_inline" attribute is really affecting how
> the optimization works, not how inlining works per se, i.e. one still
> has to mark the function as "inline" for inlining to happen. The
> always_inline attribute tells the optimizer to inline even at -O0,
> which apparently is not the default behaviour.

As far as I have seen, the compiler has been working exactly as it says 
in the manual (quoted above).  If anyone can post a code snippet and 
compiler flags that contradict that, then please do so!

Would it be reasonable to ask for a change in the documentation, since 
several people have been caught out by it?  Alternatively, should we ask 
for __attribute__((always_inline)) to imply inline, or give a warning if 
it is used for a non-inline function?  I certainly can't think of any 
situation where you would want always_inline but not normal inline.

mvh.,

David