On Fri, 7 Oct 2011, Jakub Jelinek wrote:
> Hi!
>
> Calling vect_is_simple_cond from the pattern recognized was a bad idea,
> as it calls vect_is_simple_use_1 rather than vect_is_simple_use and the
> former looks at STMT_VINFO_VECTYPE which by that time is only set
> for the data ref stmts (which is why in my test it worked well, that
> one did (a[i] < b[i]) ? ... : .... and surprisingly nothing in the testsuite
> caught that either).
>
> Fixed thusly, bootstrapped/regtested on x86_64-linux and i686-linux.
> Ok for trunk?
Ok.
Thanks,
Richard.
> 2011-10-07 Jakub Jelinek <ja...@redhat.com>
>
> PR tree-optimization/50650
> * tree-vect-patterns.c (vect_recog_mixed_size_cond_pattern): Don't
> call vect_is_simple_cond here, instead fail if cond_expr isn't
> COMPARISON_CLASS_P or if get_vectype_for_scalar_type returns NULL
> for cond_expr's first operand.
> * tree-vect-stmts.c (vect_is_simple_cond): Static again.
> * tree-vectorizer.h (vect_is_simple_cond): Remove prototype.
>
> * gcc.c-torture/compile/pr50650.c: New test.
>
> --- gcc/tree-vect-patterns.c.jj 2011-10-07 10:03:28.000000000 +0200
> +++ gcc/tree-vect-patterns.c 2011-10-07 10:18:58.000000000 +0200
> @@ -1269,8 +1269,12 @@ vect_recog_mixed_size_cond_pattern (VEC
> || TREE_CODE (else_clause) != INTEGER_CST)
> return NULL;
>
> - if (!vect_is_simple_cond (cond_expr, loop_vinfo, &comp_vectype)
> - || !comp_vectype)
> + if (!COMPARISON_CLASS_P (cond_expr))
> + return NULL;
> +
> + comp_vectype
> + = get_vectype_for_scalar_type (TREE_TYPE (TREE_OPERAND (cond_expr, 0)));
> + if (comp_vectype == NULL_TREE)
> return NULL;
>
> type = gimple_expr_type (last_stmt);
> --- gcc/tree-vect-stmts.c.jj 2011-10-07 10:03:28.000000000 +0200
> +++ gcc/tree-vect-stmts.c 2011-10-07 10:04:45.000000000 +0200
> @@ -4698,7 +4698,7 @@ vectorizable_load (gimple stmt, gimple_s
> Returns whether a COND can be vectorized. Checks whether
> condition operands are supportable using vec_is_simple_use. */
>
> -bool
> +static bool
> vect_is_simple_cond (tree cond, loop_vec_info loop_vinfo, tree *comp_vectype)
> {
> tree lhs, rhs;
> --- gcc/tree-vectorizer.h.jj 2011-10-07 10:03:28.000000000 +0200
> +++ gcc/tree-vectorizer.h 2011-10-07 10:04:28.000000000 +0200
> @@ -818,7 +818,6 @@ extern bool vect_transform_stmt (gimple,
> bool *, slp_tree, slp_instance);
> extern void vect_remove_stores (gimple);
> extern bool vect_analyze_stmt (gimple, bool *, slp_tree);
> -extern bool vect_is_simple_cond (tree, loop_vec_info, tree *);
> extern bool vectorizable_condition (gimple, gimple_stmt_iterator *, gimple *,
> tree, int);
> extern void vect_get_load_cost (struct data_reference *, int, bool,
> --- gcc/testsuite/gcc.c-torture/compile/pr50650.c.jj 2011-10-07
> 11:12:27.000000000 +0200
> +++ gcc/testsuite/gcc.c-torture/compile/pr50650.c 2011-10-07
> 11:11:30.000000000 +0200
> @@ -0,0 +1,10 @@
> +/* PR tree-optimization/50650 */
> +
> +unsigned int
> +foo (unsigned int x, unsigned int y)
> +{
> + int i;
> + for (i = 8; i--; x <<= 1)
> + y ^= (x ^ y) & 0x80 ? 79U : 0U;
> + return y;
> +}
>
> Jakub
>
>
--
Richard Guenther <rguent...@suse.de>
SUSE / SUSE Labs
SUSE LINUX Products GmbH - Nuernberg - AG Nuernberg - HRB 16746
GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer
