Re: Kill pointers keeping gimple function body around after RTL is generated

[Richard Henderson <rth at redhat dot com>]

On Thu, Sep 02, 2004 at 11:59:06PM +0200, Jan Hubicka wrote:
> -   for (; !bsi_end_p (bsi); bsi_next (&bsi))
> +   for (; !bsi_end_p (bsi); bsi_remove (&bsi))

Why?  Aren't we coing to zap the entire statement list in a moment?

> -       if (TREE_CODE (t) == LABEL_EXPR)
> +       if (TREE_CODE (t) == LABEL_EXPR && label_to_block_map)

Which probably obviates this change, does it not?

>   void
>   fini_ssa_operands (void)
>   {
> +   ggc_free (build_defs);
> +   ggc_free (build_uses);
> +   ggc_free (build_v_may_defs);
> +   ggc_free (build_vuses);
> +   ggc_free (build_v_must_defs);
> +   build_defs = NULL;
> +   build_uses = NULL;
> +   build_v_may_defs = NULL;
> +   build_vuses = NULL;
> +   build_v_must_defs = NULL;
>   }
>   
>   
> Index: tree-ssanames.c
> ===================================================================
> RCS file: /cvs/gcc/gcc/gcc/tree-ssanames.c,v
> retrieving revision 2.8
> diff -c -3 -p -r2.8 tree-ssanames.c
> *** tree-ssanames.c	25 Aug 2004 21:21:19 -0000	2.8
> --- tree-ssanames.c	2 Sep 2004 21:54:51 -0000
> *************** init_ssanames (void)
> *** 155,160 ****
> --- 155,162 ----
>   void
>   fini_ssanames (void)
>   {
> +   ggc_free (ssa_names);
> +   ssa_names = NULL;
>     free_ssanames = NULL;
>   }

And these are completely unrelated changes?


r~