Re: Compilation with clang

[David Brown]

Hi,

Glad I could help - it's always nice to get one's "good deed for the 
day" over by morning coffee, and relax and do evil the rest of the day :-)

More importantly, I think clang is going to have to be taught about 
__tmp_reg__ anyway, because it is used in many places by avrlibc and 
inline assembly.  However, while the gcc avr backend makes heavy use of 
__tmp_reg__, perhaps clang avr backend does not (I don't know any of the 
details around it).  It's possible to have a local uint8_t variable here 
and pass that to the inline assembly for use as a temporary register, 
letting the compiler pick whichever one it likes.

mvh.,

David




On 15/10/2021 10:31, Marian Buschsieweke wrote:
> Hi David,
>
> that is an excellent idea. The attached patch implements that idea (slightly
> different) and almost makes clang happy. It know fails with:
>
> > /usr/avr/include/avr/wdt.h:436:5: error: invalid operand for instruction
> >                                  "in __tmp_reg__,__SREG__" "\n\t"
> >                                  ^
> > <inline asm>:1:5: note: instantiated into assembly here
> >          in __tmp_reg__,__SREG__
> >             ^
>
> So I guess for my application I know only have to teach clang the value of
> __tmp_reg__ (and maybe also a few other constants) to get it compiling :-)
>
> Kind regards,
> Marian
>
> On Fri, 15 Oct 2021 09:49:30 +0200
> David Brown <david@westcontrol.com> wrote:
>
> > On 13/10/2021 18:05, Marian Buschsieweke wrote:
> > > Hi together,
> > >
> > > parts of the avrlibc headers are not compatible with clang. (The use case is
> > > not to compile avrlibc itself with clang, but rather an application using
> > > a vanilla GCC compiled avrlibc.) However, the issues are not always trivial to
> > > fix.
> > >
> > > E.g. in avr/wdt.h there are several instances of code like this:
> > >    
> > > > static __inline__
> > > > __attribute__ ((__always_inline__))
> > > > void wdt_enable (const uint8_t value)
> > > > {
> > > >         if (_SFR_IO_REG_P (_WD_CONTROL_REG))
> > > >         {
> > > >                 __asm__ __volatile__ (
> > > >                                 [...]
> > > >                                 : /* no outputs */
> > > >                                 : "I" (_SFR_IO_ADDR(_WD_CONTROL_REG)),
> > > >                                 [...]
> > > >                                 : "r0"
> > > >                 );
> > > >         }
> > > >         else {
> > > >                 /* variant not using _WD_CONTROL_REG as immediate */
> > > >                 [...]
> > > >        
> > > > }
> >   [...]
> >
> > It's been a /long/ time since I have done anything with the AVR, and I
> > don't have a modern avr gcc compiler (or any clang avr compiler at all)
> > handy for testing.  However, I have an idea you can try:
> >
> >         "I" (_SFR_IO_ADDR(_WD_CONTROL_REG) || 1)
> >
> > If "_SFR_IO_ADDR(_WD_CONTROL_REG)" evaluates to false, then this
> > expression would give "1", which clang might be happier with.