Re: [patch] Fix PR tree-optimization/46663

[Richard Guenther <richard dot guenther at gmail dot com>]

On Wed, Dec 1, 2010 at 11:13 AM, Ira Rosen <IRAR@il.ibm.com> wrote:
>
> Hi,
>
> This patch adds a check that FUNCTION_DECL exists for the call in
> vectorizer power pattern recognition.
> I am testing the patch for trunk on x86_64-suse-linux now.
>
> The problem goes back to 4.0, how far back should I backport this fix?

Till 4.3, but the fix is not complete, it also needs

> +  if (fn == NULL_TREE
          || DECL_BUILT_IN_CLASS (fn) !=  BUILT_IN_NORMAL)
> +   return NULL;

otherwise it'll match target and non-builtins as well.

Richard.

> Thanks,
> Ira
>
> ChangeLog:
>
> ? ? ? ?PR tree-optimization/46663
> ? ? ? ?* tree-vect-patterns.c (vect_recog_pow_pattern): Check that
> ? ? ? ?FUNCTION_DECL exists.
>
> testsuite/ChangeLog:
>
> ? ? ? ?PR tree-optimization/46663
> ? ? ? ?* gcc.dg/vect/pr46663.c: New test.
>
>
> Index: testsuite/gcc.dg/vect/pr46663.c
> ===================================================================
> --- testsuite/gcc.dg/vect/pr46663.c ? ? (revision 0)
> +++ testsuite/gcc.dg/vect/pr46663.c ? ? (revision 0)
> @@ -0,0 +1,14 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O -ftree-vectorize -fdump-tree-vect-details
> -fexceptions" } */
> +
> +typedef __attribute__ ((const)) int (*bart) (void);
> +
> +int foo (bart bar, int m)
> +{
> + ?int i, j = 0;
> + ?for (i = 0; i < m; i++)
> + ? ?j += bar();
> + ?return j;
> +}
> +
> +/* { dg-final { cleanup-tree-dump "vect" } } */
> Index: tree-vect-patterns.c
> ===================================================================
> --- tree-vect-patterns.c ? ? ? ?(revision 167324)
> +++ tree-vect-patterns.c ? ? ? ?(working copy)
> @@ -472,6 +472,9 @@ vect_recog_pow_pattern (gimple last_stmt
> ? ? return NULL;
>
> ? fn = gimple_call_fndecl (last_stmt);
> + ?if (fn == NULL_TREE)
> + ? return NULL;
> +
> ? switch (DECL_FUNCTION_CODE (fn))
> ? ? {
> ? ? case BUILT_IN_POWIF:
>
>