On Thu, May 12, 2016 at 12:19 PM, Ilya Enkovich <enkovich....@gmail.com> wrote:
> Hi,
>
> Currently we have a code in vect_determine_vectorization_factor to
> compute vectype for mask producers. It wasn't meant to be used for
> EXPR_COND assignments but it is used now in some cases causing wrong
> resulting vectype.
>
> Bootstrapped and regtested for x86_64-pc-linux-gnu. OK for trunk?
Ok.
Thanks,
Richard.
> Thanks,
> Ilya
> --
> gcc/
>
> 2016-05-12 Ilya Enkovich <ilya.enkov...@intel.com>
>
> PR tree-optimization/71006
> * tree-vect-loop.c (vect_determine_vectorization_factor): Don't
> consider COND_EXPR as a mask producer.
>
> gcc/testsuite/
>
> 2016-05-12 Ilya Enkovich <ilya.enkov...@intel.com>
>
> PR tree-optimization/71006
> * gcc.dg/pr71006.c: New test.
>
>
> diff --git a/gcc/testsuite/gcc.dg/pr71006.c b/gcc/testsuite/gcc.dg/pr71006.c
> new file mode 100644
> index 0000000..2b45aa0
> --- /dev/null
> +++ b/gcc/testsuite/gcc.dg/pr71006.c
> @@ -0,0 +1,16 @@
> +/* PR target/71006 */
> +/* { dg-do compile } */
> +/* { dg-options "-O1 -ftree-vectorize" } */
> +
> +unsigned char uu, gu, e2;
> +
> +void
> +fs (void)
> +{
> + char *nq = (char *)&gu, *k4 = (char *)&gu;
> + while (*k4 < 1)
> + {
> + uu += (*nq != 0 || e2 != 0);
> + ++*k4;
> + }
> +}
> diff --git a/gcc/tree-vect-loop.c b/gcc/tree-vect-loop.c
> index da98211..d673c67 100644
> --- a/gcc/tree-vect-loop.c
> +++ b/gcc/tree-vect-loop.c
> @@ -437,7 +437,9 @@ vect_determine_vectorization_factor (loop_vec_info
> loop_vinfo)
> /* Bool ops don't participate in vectorization factor
> computation. For comparison use compared types to
> compute a factor. */
> - if (TREE_CODE (scalar_type) == BOOLEAN_TYPE)
> + if (TREE_CODE (scalar_type) == BOOLEAN_TYPE
> + && is_gimple_assign (stmt)
> + && gimple_assign_rhs_code (stmt) != COND_EXPR)
> {
> if (STMT_VINFO_RELEVANT_P (stmt_info))
> mask_producers.safe_push (stmt_info);
