public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
From: Ilya Verbin <iverbin@gmail.com>
To: Jan Hubicka <hubicka@ucw.cz>
Cc: Jason Merrill <jason@redhat.com>,
	Jakub Jelinek <jakub@redhat.com>,
	gcc-patches@gcc.gnu.org, rguenther@suse.de
Subject: Re: Silence merge warnings on artificial types
Date: Thu, 02 Apr 2015 18:23:00 -0000	[thread overview]
Message-ID: <20150402182303.GA52553@msticlxl57.ims.intel.com> (raw)
In-Reply-To: <20150331075121.GB62830@kam.mff.cuni.cz>

On Tue, Mar 31, 2015 at 09:51:21 +0200, Jan Hubicka wrote:
> this patch adds the ARTIFICIAL flag check to avoid ODR merging to these.
> I oriignally tested DECL_ARTIFICIAL (decl) (that is TYPE_NAME) that randomly
> dropped type names on some classes but not all.
> 
> Jason, please do you know what is meaning of DECL_ARTIFICIAL on class type
> names? Perhaps we can drop them to 0 in free lang data?
> 
> With this bug I triggered wrong devirtualization because we no longer insert
> non-odr types into a type inheritance graph.  This is fixed by the lto_read_decls
> change and finally I triggered an ICE in ipa-devirt that due to the bug
> output a warning late and ICEd on streamer cache being NULL.  I guess it is
> better to guard it even though all wanrings should be output early.
> 
> Bootsrapped/regtested x86_64-linux, will commit it after chromium rebuild.
> 
> Honza
> 
> 	* tree.c (need_assembler_name_p): Artificial types have no ODR
> 	names.
> 	* ipa-devirt.c (warn_odr): Do not try to apply ODR cache when
> 	no caching is done.
> 
> 	* lto.c (lto_read_decls): Move code registering odr types out
> 	of TYPE_CANONICAL conditional and also register polymorphic types.
> Index: tree.c
> ===================================================================
> --- tree.c	(revision 221777)
> @@ -5139,6 +5145,7 @@ need_assembler_name_p (tree decl)
>        && decl == TYPE_NAME (TREE_TYPE (decl))
>        && !is_lang_specific (TREE_TYPE (decl))
>        && AGGREGATE_TYPE_P (TREE_TYPE (decl))
> +      && !TYPE_ARTIFICIAL (TREE_TYPE (decl))
>        && !variably_modified_type_p (TREE_TYPE (decl), NULL_TREE)
>        && !type_in_anonymous_namespace_p (TREE_TYPE (decl)))
>      return !DECL_ASSEMBLER_NAME_SET_P (decl);

Hmm, libgomp.c++/target-3.C still fails.
Here is what I see in need_assembler_name_p:

 <type_decl 0x2b5a8751aed8 .omp_data_s.3
    type <record_type 0x2b5a87536690 .omp_data_s.3 BLK
        size <integer_cst 0x2b5a873ce318 constant 192>
        unit size <integer_cst 0x2b5a873ce2e8 constant 24>
        align 64 symtab 0 alias set -1 canonical type 0x2b5a87536690
        fields <field_decl 0x2b5a87537000 b.0 type <pointer_type 0x2b5a87536150>
            unsigned DI file test.cpp line 8 col 13
            size <integer_cst 0x2b5a873ace58 constant 64>
            unit size <integer_cst 0x2b5a873ace70 constant 8>
            align 64 offset_align 128
            offset <integer_cst 0x2b5a873ace88 constant 0>
            bit offset <integer_cst 0x2b5a873aced0 constant 0> context <record_type 0x2b5a87536690 .omp_data_s.3> chain <field_decl 0x2b5a87537098 s>> reference_to_this <reference_type 0x2b5a87536738>>
    VOID file test.cpp line 11 col 13
    align 1>
(TYPE_ARTIFICIAL = 0)

 <type_decl 0x2b5a8751a850 .omp_data_s.3
    type <record_type 0x2b5a8752cdc8 .omp_data_s.3 BLK
        size <integer_cst 0x2b5a873ce318 constant 192>
        unit size <integer_cst 0x2b5a873ce2e8 constant 24>
        align 64 symtab 0 alias set -1 canonical type 0x2b5a8752cdc8
        fields <field_decl 0x2b5a8751a980 b.0 type <pointer_type 0x2b5a8752c498>
            unsigned DI file test.cpp line 8 col 13
            size <integer_cst 0x2b5a873ace58 constant 64>
            unit size <integer_cst 0x2b5a873ace70 constant 8>
            align 64 offset_align 128
            offset <integer_cst 0x2b5a873ace88 constant 0>
            bit offset <integer_cst 0x2b5a873aced0 constant 0> context <record_type 0x2b5a8752cdc8 .omp_data_s.3> chain <field_decl 0x2b5a8751a8e8 s>>
        pointer_to_this <pointer_type 0x2b5a87536888>>
    VOID file test.cpp line 11 col 13
    align 1>
(TYPE_ARTIFICIAL = 0)

  -- Ilya

  parent reply	other threads:[~2015-04-02 18:23 UTC|newest]

Thread overview: 16+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2015-03-30  3:03 Silence merge warnings on artiical types Jan Hubicka
2015-03-30  8:30 ` Richard Biener
2015-03-30  8:48   ` Jan Hubicka
2015-03-30 15:36 ` Ilya Verbin
2015-03-30 17:06   ` Jan Hubicka
2015-03-30 17:11     ` Ilya Verbin
2015-03-30 17:16       ` Jan Hubicka
2015-03-30 17:21     ` Jakub Jelinek
2015-03-30 17:23       ` Jan Hubicka
2015-03-30 17:53         ` Jason Merrill
2015-03-31  7:51           ` Silence merge warnings on artificial types Jan Hubicka
2015-03-31 13:26             ` Jason Merrill
2015-04-02 18:23             ` Ilya Verbin [this message]
2015-04-02 18:37               ` Jan Hubicka
2015-04-02 19:06               ` Jakub Jelinek
2015-04-03 13:36                 ` Jakub Jelinek

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20150402182303.GA52553@msticlxl57.ims.intel.com \
    --to=iverbin@gmail.com \
    --cc=gcc-patches@gcc.gnu.org \
    --cc=hubicka@ucw.cz \
    --cc=jakub@redhat.com \
    --cc=jason@redhat.com \
    --cc=rguenther@suse.de \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).