This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: TREE_ADDRESSABLE problem
> On Fri, Nov 21, 2003 at 12:24:02AM +0100, Jan Hubicka wrote:
> > However we need to ensure that VAR_DECLs won't be modified in some other
> > ways by the optimizers?
>
> No idea.
>
> > Perhaps we want do both - duplicate var_decls and compute addressable
> > flag as part of remove_useless_* or rewrite_into_ssa?
>
> I don't think so.
>
> It doesn't seem like it ought to be that difficult to just delay
> compilation of any inline candidate until we've decided that we're
> done with all inlining.
>
> Which is *much* more robust than copying things around.
Actually copying things around would help me to implement the recursive
inlining (ie inlinine recursive calls up to given depth so function
growth to specified size).
But I will definitly update the sorting routines to try first compile
all functions we inlined into and then the inline candidate.
However if we want to maintain the constraint that for non-recursive
graphs we compile callees before caller, we end up needing copies from
time to time.
OK, for start I will implement the tracking of ADDR_EXPRs so we won't
get tons of checking errors.
find_referenced_vars looks like most plausible place to hook into.
Honza
>
>
> r~