Re: gcc-4.5/4.4: Bug in .subreg1 pass?

[Georg-Johann Lay <avr at gjlay dot de>]

Georg-Johann Lay schrieb:
> Denis Chertykov schrieb:
>
> > 2011/2/26 Georg Johann Lay <avr@gjlay.de>
> >
> > > Eric Botcazou schrieb:
> > >
> > > > > What does "word" mean here? Is it a 32-bit entity or is it 
> > > > > according to
> > > > > word_mode which is QImode for avr?
> > > >
> > > >
> > > > The latter, it is machine-dependent.
> > > >
> > > >
> > > >
> > > > > So the same should be true for QI-subregs of scalar modes if
> > > > > UNITS_PER_WORT = 1. Right?
> > > >
> > > >
> > > > Right.
> > >
> > >
> > > Thanks for that definite clarification.
> >
> >
> >
> > As I understand PR46779 and PR45291 no longer an AVR port bugs.
> > And the following comment is not a true anymore:
> > http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45291#c2
> > Am I right ?
>
>
> At least the analysis I gave there is incorrect.
>
> I remember a problem in avr backend concerning the frame pointer because 
> QImode was allowed in FP's hard regs... that sort of problem. Was that 
> fixed here or just in some distribution like WinAVR? So perhaps it's 
> still this FP problem?

Ok, here it is:

http://gcc.gnu.org/viewcvs?view=revision&revision=123263

But now there is the problem that we have

(insn 33 35 34 2 foo.c:5 (set (subreg:QI (reg:HI 50 [ D.1955 ]) 1)
        (reg:QI 41 [ D.1955 ])) ...

and reg 50 is allowed to get in R29:R28 (frame pointer regs) but the 
subreg is not, i.e. because QI is not allowed in FP regs, it is not 
possible to set low/high part seperately.

Johann