[patch] Fix PR43464: update loop closed SSA form once copy prop is done

[Richard Guenther]

On Mon, 29 Mar 2010, Sebastian Pop wrote:

> On Fri, Mar 26, 2010 at 04:51, Richard Guenther <rguenther@suse.de> wrote:
> > But we already analyzed that copyprop is _not_ the problem. Â The
> > problem is the pass that introduces the two-arg loop-closed PHI
> > node.
> 
> Note that pass_tree_unswitch is not the only pass that introduces
> the two-arg loop-closed PHI.  Copyprop is another pass that does that,
> as it has a TODO_cleanup_cfg that calls cleanup_tree_cfg.
> 
> cleanup_tree_cfg_noloop called from cleanup_tree_cfg is destroying the
> CFG structure needed for the LCSSA form: it removes all the single entry
> basic blocks that were placed after the loops for the LCSSA form.
> 
> Interestingly, cleanup_tree_cfg is calling repair_loop_structures
> 
>   if (current_loops != NULL
>       && loops_state_satisfies_p (LOOPS_NEED_FIXUP))
>     repair_loop_structures ();
> 
> that in turn calls rewrite_into_loop_closed_ssa like this:
> 
>   /* This usually does nothing.  But sometimes parts of cfg that originally
>      were inside a loop get out of it due to edge removal (since they
>      become unreachable by back edges from latch).  */
>   if (loops_state_satisfies_p (LOOP_CLOSED_SSA))
>     rewrite_into_loop_closed_ssa (changed_bbs, TODO_update_ssa);
> 
> and this call to rewrite_into_loop_closed_ssa does not add the single
> entry basic block needed to have a proper LCSSA form.
> 
> Should rewrite_into_loop_closed_ssa replace the missing basic blocks,
> or should cleanup_tree_cfg_noloop be fixed to not destroy the CFG
> form needed for the LCSSA?

Well, I'd say avoid breaking LCSSA will be cheaper in the end.

But Zdenek should know more here.

Richard.