This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] Re: [3.3/3.4/3.5 Regression] Performace regression: poor optimization of const memory
On Tue, Mar 02, 2004 at 10:02:46AM +0100, Jakub Jelinek wrote:
> On Tue, Mar 02, 2004 at 09:46:22AM +0100, Jakub Jelinek wrote:
> > But certainly maybe_set_unchanging should at least replace:
> > if ((TREE_READONLY (t) && DECL_P (t)
> > - && (TREE_STATIC (t) || ! AGGREGATE_TYPE_P (TREE_TYPE (t)))
> > + && (TREE_STATIC (t) || DECL_EXTERNAL (t) || ! AGGREGATE_TYPE_P (TREE_TYPE (t)))
> > && (TREE_CODE (t) == PARM_DECL
> > || (DECL_INITIAL (t) && TREE_CONSTANT (DECL_INITIAL (t)))))
> > || TREE_CODE_CLASS (TREE_CODE (t)) == 'c')
> > RTX_UNCHANGING_P (ref) = 1;
>
> Actually that is not enough, Jason in Dec, 2002 removed DECL_INITIAL (t) ==
> NULL case.
These patches to maybe_set_unchanging() broke "my" port as well. The only
solution I had, was to revert them in my tree. Maybe your patch restores
the old behaviour of RTX_UNCHANGING_P.
Gunther
> But this certainly makes Andrew's testcase read from a only once.
>
> 2004-03-02 Jakub Jelinek <jakub@redhat.com>
>
> * explow.c (maybe_set_unchanging): Set RTX_UNCHANGING_P even for
> read-only DECL_EXTERNAL decls.
>
> --- gcc/explow.c.jj 2003-12-15 09:37:32.000000000 +0100
> +++ gcc/explow.c 2004-03-02 12:07:55.034840235 +0100
> @@ -608,9 +608,11 @@ maybe_set_unchanging (rtx ref, tree t)
> solution, perhaps based on alias sets. */
>
> if ((TREE_READONLY (t) && DECL_P (t)
> - && (TREE_STATIC (t) || ! AGGREGATE_TYPE_P (TREE_TYPE (t)))
> - && (TREE_CODE (t) == PARM_DECL
> - || (DECL_INITIAL (t) && TREE_CONSTANT (DECL_INITIAL (t)))))
> + && (DECL_EXTERNAL (t)
> + || ((TREE_STATIC (t) || ! AGGREGATE_TYPE_P (TREE_TYPE (t)))
> + && (TREE_CODE (t) == PARM_DECL
> + || (DECL_INITIAL (t)
> + && TREE_CONSTANT (DECL_INITIAL (t)))))))
> || TREE_CODE_CLASS (TREE_CODE (t)) == 'c')
> RTX_UNCHANGING_P (ref) = 1;
> }