On Thu, 7 Dec 2023, Jakub Jelinek wrote:

> Hi!
> 
> The following testcase ICEs because maybe_optimize_arith_overflow
> uses build_nonstandard_integer_type, which is inappropriate if
> type is large BITINT_TYPE.
> 
> Fixed thusly, bootstrapped/regtested on x86_64-linux and i686-linux,
> ok for trunk?

OK.

> 2023-12-07  Jakub Jelinek  <ja...@redhat.com>
> 
>       PR tree-optimization/112880
>       * tree-ssa-dce.cc (maybe_optimize_arith_overflow): Use
>       unsigned_type_for instead of conditionally calling
>       build_nonstandard_integer_type.
> 
>       * gcc.dg/bitint-49.c: New test.
> 
> --- gcc/tree-ssa-dce.cc.jj    2023-12-05 16:14:36.833248092 +0100
> +++ gcc/tree-ssa-dce.cc       2023-12-06 17:16:42.740873447 +0100
> @@ -1241,9 +1241,7 @@ maybe_optimize_arith_overflow (gimple_st
>    tree arg1 = gimple_call_arg (stmt, 1);
>    location_t loc = gimple_location (stmt);
>    tree type = TREE_TYPE (TREE_TYPE (lhs));
> -  tree utype = type;
> -  if (!TYPE_UNSIGNED (type))
> -    utype = build_nonstandard_integer_type (TYPE_PRECISION (type), 1);
> +  tree utype = unsigned_type_for (type);
>    tree result = fold_build2_loc (loc, subcode, utype,
>                                fold_convert_loc (loc, utype, arg0),
>                                fold_convert_loc (loc, utype, arg1));
> --- gcc/testsuite/gcc.dg/bitint-49.c.jj       2023-12-06 17:15:30.461888327 
> +0100
> +++ gcc/testsuite/gcc.dg/bitint-49.c  2023-12-06 17:15:01.669292609 +0100
> @@ -0,0 +1,37 @@
> +/* PR tree-optimization/112880 */
> +/* { dg-do compile { target bitint } } */
> +/* { dg-options "-std=c23 -O2" } */
> +
> +#if __BITINT_MAXWIDTH__ >= 1024
> +_BitInt(1024) a, b, c, d, e, f;
> +
> +void
> +foo (void)
> +{
> +  __builtin_add_overflow (a, b, &a);
> +  __builtin_sub_overflow (c, d, &c);
> +  __builtin_mul_overflow (e, f, &e);
> +}
> +#endif
> +
> +#if __BITINT_MAXWIDTH__ >= 512
> +_BitInt(512) g, h, i, j, k, l;
> +
> +void
> +bar (void)
> +{
> +  __builtin_add_overflow (g, h, &g);
> +  __builtin_sub_overflow (i, j, &i);
> +  __builtin_mul_overflow (k, l, &k);
> +}
> +#endif
> +
> +_BitInt(32) m, n, o, p, q, r;
> +
> +void
> +baz (void)
> +{
> +  __builtin_add_overflow (m, n, &m);
> +  __builtin_sub_overflow (o, p, &o);
> +  __builtin_mul_overflow (q, r, &q);
> +}
> 
>       Jakub
> 
> 

-- 
Richard Biener <rguent...@suse.de>
SUSE Software Solutions Germany GmbH,
Frankenstrasse 146, 90461 Nuernberg, Germany;
GF: Ivo Totev, Andrew McDonald, Werner Knoblich; (HRB 36809, AG Nuernberg)

Reply via email to