Re: [patch] Update loops in cfg manipulation functions

[Diego Novillo <dnovillo at redhat dot com>]

Zdenek Dvorak wrote on 10/23/06 04:53:

> *************** execute_vrp (void)
> *** 4781,4787 ****
>   {
>     insert_range_assertions ();
>   
> !   current_loops = loop_optimizer_init (LOOPS_NORMAL);
>     if (current_loops)
>       scev_initialize (current_loops);
>   
> --- 4781,4787 ----
>   {
>     insert_range_assertions ();
>   
> !   loop_optimizer_init (LOOPS_NORMAL);
>     if (current_loops)
>       scev_initialize (current_loops);
>   
If we are going to be keeping loops up-to-date, wouldn't this
initialization be unnecessary?  Couldn't we just initialize loops
after the CFG is built?  If not, would it make sense to have that as a
long-term goal?

>   #define avail_temp_slots (cfun->x_avail_temp_slots)
>   #define temp_slot_level (cfun->x_temp_slot_level)
>   #define nonlocal_goto_handler_labels (cfun->x_nonlocal_goto_handler_labels)
> + #define current_loops (cfun->x_current_loops)
>   
For now I guess this is OK, but I would really prefer that we had CFUN
a parameter to this accessor.  Since Jan is working in this area, we
can do this afterwards.


Nice cleanup.  Thanks.  Any idea on compile-time impact?  We seem to
be doing more in the CFG low-level manipulators.

The patch crosses into the middle-end quite a bit, I'd like a
middle-end maintainer to look at it as well.