Re: tree profilng merge 3 - find_sub_basic_blocks fixes wrt computed gotos

[Jan Hubicka <jh at suse dot cz>]

> At the very moment I am simply removing the abormal edges and
> re-creating them from scratch using find_sub_basic_blocks since
> representation of some abnormal edges differ.  I have fixed few of these
> (link longjmps and some of EH) already so perhaps the situation is not
> too bad.  I can try to experiment with it after returning from the
> vacation at 14th.

Another problem with abnormal edges is that the expanders comonly put
some instructions after the instruction producing the edge (for case of
noncall EH it is for valid reasons, in the case of noreturn calls it is
bogus).  I would like to track these cases down, but it is bit involved.
So I guess for now we can simply collect the list of lables either in
find_basic_block or during the cfg expansion from tree level and use it
in the find_sub_basic_block assuming that none of those labels gets
removed.  Does this sound sane
I will also gradually fix the backends expanding code after functions
terminating the flow...

Honza
> 
> Honza
> >
> > 
> > 
> > r~