This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: PR 56524: TREE_OPTIMIZATION_OPTABS vs. mips16


On Thu, Mar 07, 2013 at 10:51:14PM +0000, Richard Sandiford wrote:
>  void
> -save_optabs_if_changed (tree fndecl)
> +init_tree_optimization_optabs (tree optnode)
>  {
> -  /* ?? If this fails, we should temporarily restore the default
> -     target first (set_cfun (NULL) ??), do the rest of this function,
> -     and then restore it.  */
> -  gcc_assert (this_target_optabs == &default_target_optabs);
> +  /* Quick exit if we have already computed optabs for this target.  */
> +  if (TREE_OPTIMIZATION_BASE_OPTABS (optnode) == this_target_optabs)
> +    return;
>  
> +  /* Forget any previous information and set up for the current target.  */
> +  TREE_OPTIMIZATION_BASE_OPTABS (optnode) = this_target_optabs;
>    struct target_optabs *tmp_optabs = (struct target_optabs *)
> -    ggc_alloc_atomic (sizeof (struct target_optabs));
> -  tree optnode = DECL_FUNCTION_SPECIFIC_OPTIMIZATION (fndecl);
> +    TREE_OPTIMIZATION_OPTABS (optnode);
> +  if (tmp_optabs)
> +    memset (tmp_optabs, 0, sizeof (struct target_optabs));
> +  else
> +    tmp_optabs = (struct target_optabs *)
> +      ggc_alloc_atomic (sizeof (struct target_optabs));
>  
>    /* Generate a new set of optabs into tmp_optabs.  */
>    init_all_optabs (tmp_optabs);

If TARGET_OPTIMIZATION_OPTABS is non-NULL upon entering the function,
then we call memset above, and then

...

  /* If the optabs changed, record it.  */
  if (memcmp (tmp_optabs, this_target_optabs, sizeof (struct target_optabs)))
    {
      if (TREE_OPTIMIZATION_OPTABS (optnode))
        ggc_free (TREE_OPTIMIZATION_OPTABS (optnode));
                                            
      TREE_OPTIMIZATION_OPTABS (optnode) = (unsigned char *) tmp_optabs;

As tmp_optabs == TREE_OPTIMIZATION_OPTABS (optnode); in that case,
the ggc_free is wrong, TREE_OPTIMIZATION_OPTABS (optnode) will then point
to freed memory.

So I think the

      if (TREE_OPTIMIZATION_OPTABS (optnode))
        ggc_free (TREE_OPTIMIZATION_OPTABS (optnode));

has to be removed (of course the second ggc_free, if memcmp returned 0, is
desirable).  Otherwise looks good.

	Jakub


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]