Re: unwind info for epilogues

[Jan Hubicka <hubicka at ucw dot cz>]

> Ian Lance Taylor wrote:
> > This approach seems clearly superior.  I wonder if you can provide some
> >simple functions for the backends to call to add notes and such.
> 
> Possibly.  At the moment I can't really think of any, or at least
> anything more complicated than the existing add_reg_note.  I would
> expect these to become more apparent as ports are converted.
> 
> > One could also imagine a more complex function which simply walked the
> >prologue and did something like what the current dwarf2out.c code does,
> >except adding reg notes.  That might make it relatively easy to
> >transition the backends to the new scheme.
> 
> This same thought occurred to me.  Combine that with tracking the CFA
> state across basic block boundaries, and markers for the CFA state to
> be placed at hot/cold section boundaries and we'll fix two outstanding
> bugs with unwind code.

Needless to say that we need to add support for multiple entry points :)
> 
> Now if only someone would rewrite delayed-branch opt to not hork the
> CFG, and the aforementioned pass will be easy...

Sadly reorg.c was always black box for me. How many active targets use
it?

Honza
> 
> 
> r~