On Wed, 18 Nov 2020, Jan Hubicka wrote:
> Hi,
> main purpose of obj_type_ref is to hold the type that was used in
> virutal call. We do not hash this info in hash_operand that causes a
> lot of miscompares at ICF time. With LTO this is quite important for
> icf performance and in that case we do have manged type names (for
> non-anonymous types)
>
> Building libxul without patch we get 1477890 miscompares at:
> libxul.so.wpa.076i.icf: false returned: 'operand_equal_p failed' in
> compare_operand at ../../gcc/ipa-icf-gimple.c:356
> With patch this turns into 242454.
>
> Bootstrapped/regtested x86_64-linux, OK?
OK.
Richard.
> Honza
> PR ipa/92535
> * fold-const.c (operand_compare::hash_operand): Hash ODR name of
> obj_type_ref_class.
> diff --git a/gcc/fold-const.c b/gcc/fold-const.c
> index ddf18f27cb7..e759ddb1e60 100644
> --- a/gcc/fold-const.c
> +++ b/gcc/fold-const.c
> @@ -3866,6 +3866,16 @@ operand_compare::hash_operand (const_tree t,
> inchash::hash &hstate,
> inchash::add_expr (OBJ_TYPE_REF_EXPR (t), hstate, flags);
> inchash::add_expr (OBJ_TYPE_REF_TOKEN (t), hstate, flags);
> inchash::add_expr (OBJ_TYPE_REF_OBJECT (t), hstate, flags);
> + if (tree c = obj_type_ref_class (t))
> + {
> + c = TYPE_NAME (TYPE_MAIN_VARIANT (c));
> + /* We compute mangled names only when free_lang_data is run.
> + In that case we can hash precisely. */
> + if (DECL_ASSEMBLER_NAME_SET_P (c))
> + hstate.add_object
> + (IDENTIFIER_HASH_VALUE
> + (DECL_ASSEMBLER_NAME (c)));
> + }
> return;
> default:
> break;
>
--
Richard Biener <[email protected]>
SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 Nuernberg,
Germany; GF: Felix Imend
