On Fri, 12 Dec 2014, Jakub Jelinek wrote:
> Hi!
>
> This testcase ICEs because I wasn't checking for overflow in the size
> computation. Only max (diff + len2, len1) <= 1024 cases are considered,
> so this patch gives up if either len2 or diff is > 1024.
>
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for
trunk/4.9/4.8?
Ok.
Thanks,
Richard.
> 2014-12-12 Jakub Jelinek <ja...@redhat.com>
>
> PR tree-optimization/64269
> * tree-ssa-forwprop.c (simplify_builtin_call): Bail out if
> len2 or diff are too large.
>
> * gcc.c-torture/compile/pr64269.c: New test.
>
> --- gcc/tree-ssa-forwprop.c.jj 2014-12-01 14:57:30.000000000 +0100
> +++ gcc/tree-ssa-forwprop.c 2014-12-12 09:46:05.790053928 +0100
> @@ -1288,7 +1288,8 @@ simplify_builtin_call (gimple_stmt_itera
> use_operand_p use_p;
>
> if (!tree_fits_shwi_p (val2)
> - || !tree_fits_uhwi_p (len2))
> + || !tree_fits_uhwi_p (len2)
> + || compare_tree_int (len2, 1024) == 1)
> break;
> if (is_gimple_call (stmt1))
> {
> @@ -1354,7 +1355,8 @@ simplify_builtin_call (gimple_stmt_itera
> is not constant, or is bigger than memcpy length, bail out. */
> if (diff == NULL
> || !tree_fits_uhwi_p (diff)
> - || tree_int_cst_lt (len1, diff))
> + || tree_int_cst_lt (len1, diff)
> + || compare_tree_int (diff, 1024) == 1)
> break;
>
> /* Use maximum of difference plus memset length and memcpy length
> --- gcc/testsuite/gcc.c-torture/compile/pr64269.c.jj 2014-12-12
> 09:47:04.795015479 +0100
> +++ gcc/testsuite/gcc.c-torture/compile/pr64269.c 2014-12-12
> 09:46:51.000000000 +0100
> @@ -0,0 +1,9 @@
> +/* PR tree-optimization/64269 */
> +
> +void
> +foo (char *p)
> +{
> + __SIZE_TYPE__ s = ~(__SIZE_TYPE__)0;
> + *p = 0;
> + __builtin_memset (p + 1, 0, s);
> +}
>
> Jakub
>
>
--
Richard Biener <rguent...@suse.de>
SUSE LINUX GmbH, GF: Felix Imendoerffer, Jane Smithard, Jennifer Guild,
Dilip Upmanyu, Graham Norton HRB 21284 (AG Nuernberg)
