From: Jason Merrill <jason@redhat.com>
To: Jakub Jelinek <jakub@redhat.com>,
Jonathan Wakely <jwakely@redhat.com>, Jan Hubicka <jh@suse.cz>
Cc: gcc-patches@gcc.gnu.org, Richard Biener <rguenther@suse.de>,
Patrick Palka <ppalka@redhat.com>
Subject: Re: [PATCH] c++, v3: Retry the aliasing of base/complete cdtor optimization at import_export_decl time [PR113208]
Date: Wed, 24 Apr 2024 18:39:33 -0400 [thread overview]
Message-ID: <cc3d167a-3bd8-4cc8-96bc-c9c313cf2e7e@redhat.com> (raw)
In-Reply-To: <Zikwc9BTXQErR++G@tucnak>
On 4/24/24 09:16, Jakub Jelinek wrote:
> On Wed, Apr 24, 2024 at 10:16:05AM +0100, Jonathan Wakely wrote:
>>> That fixes the testcases too, but seems to regress
>>> +FAIL: libstdc++-abi/abi_check
>
>> There are explicit instantiation definitions that should instantiate
>> those types:
>>
>> src/c++17/fs_dir.cc:template class std::__shared_ptr<fs::_Dir>;
>> src/c++17/fs_dir.cc:template class
>> std::__shared_ptr<fs::recursive_directory_iterator::_Dir_stack>;
>> src/c++17/fs_path.cc:template class std::__shared_ptr<const
>> fs::filesystem_error::_Impl>;
>>
>> So the missing symbols should be present in cow-fs_dir.o and cow-fs_path.o
>
> So this boils down to (cvise reduced):
> namespace __gnu_cxx { enum _Lock_policy { _S_single, _S_mutex, _S_atomic } const __default_lock_policy = _S_atomic; }
> namespace std {
> using __gnu_cxx::__default_lock_policy;
> using __gnu_cxx::_Lock_policy;
> template <typename, _Lock_policy = __default_lock_policy> struct __shared_ptr { constexpr __shared_ptr() {} };
> namespace filesystem {
> struct _Dir;
> struct directory_iterator { __shared_ptr<_Dir> _M_dir; };
> void end() { directory_iterator(); } }
> extern template class __shared_ptr<filesystem::_Dir>;
> }
> namespace fs = std::filesystem;
> template class std::__shared_ptr<fs::_Dir>;
> at -O2, previously it used to emit
> _ZNSt12__shared_ptrINSt10filesystem4_DirELN9__gnu_cxx12_Lock_policyE2EEC2Ev
> _ZNSt12__shared_ptrINSt10filesystem4_DirELN9__gnu_cxx12_Lock_policyE1EEC2Ev
> but now no longer does with the yesterday posted PR113208 patch.
>
> The following updated patch fixes that by calling note_vague_linkage_fn for
> the cdtor clones from maybe_clone_body if the flags suggest that the
> maybe-in-charge cdtor had tentative_decl_linkage -> note_vague_linkage_fn
> called too. And then I've noticed that in some cases the updated comdat
> group set by maybe_clone_body (*C5* or *D5*) was then overwritten again by
> maybe_make_one_only. So the patch tweaks cxx_comdat_group, so that when
> some comdat group has been chosen already it doesn't try to use some
> different one.
>
> Bootstrapped/regtested on x86_64-linux and i686-linux, this one doesn't
> regress anything unlike the earlier patch.
>
> 2024-04-24 Jakub Jelinek <jakub@redhat.com>
>
> PR lto/113208
> * decl2.cc (tentative_decl_linkage): Use comdat_linkage also
> for implicit instantiations of maybe in charge ctors/dtors
> if -fimplicit-templates or -fimplicit-inline-templates and
> -fweak and target supports aliases.
> * optimize.cc (maybe_clone_body): Call note_vague_linkage_fn
> on clone if fn has DECL_INTERFACE_KNOWN, DECL_NOT_REALLY_EXTERN
> and DECL_DEFER_OUTPUT flags set.
> * decl.cc (cxx_comdat_group): For DECL_CLONED_FUNCTION_P
> functions if SUPPORTS_ONE_ONLY return DECL_COMDAT_GROUP if already
> set.
>
> * g++.dg/abi/comdat2.C: New test.
> * g++.dg/abi/comdat3.C: New test.
> * g++.dg/lto/pr113208_0.C: New test.
> * g++.dg/lto/pr113208_1.C: New file.
> * g++.dg/lto/pr113208.h: New file.
>
> --- gcc/cp/decl2.cc.jj 2024-04-23 14:49:41.933186265 +0200
> +++ gcc/cp/decl2.cc 2024-04-24 15:17:09.043625729 +0200
> @@ -3314,7 +3314,16 @@ tentative_decl_linkage (tree decl)
> to mark the functions at this point. */
> if (DECL_DECLARED_INLINE_P (decl)
> && (!DECL_IMPLICIT_INSTANTIATION (decl)
> - || DECL_DEFAULTED_FN (decl)))
> + || DECL_DEFAULTED_FN (decl)
> + /* For implicit instantiations of cdtors,
> + if import_export_decl would use comdat linkage,
> + make sure to use it right away, so that maybe_clone_body
> + can use aliases. See PR113208. */
> + || (DECL_MAYBE_IN_CHARGE_CDTOR_P (decl)
> + && (flag_implicit_templates
> + || flag_implicit_inline_templates)
> + && flag_weak
> + && TARGET_SUPPORTS_ALIASES)))
It seems wrong to set DECL_INTERFACE_KNOWN for cdtors that might get an
explicit instantiation later, and likewise for comdat_linkage when
!flag_weak; instead of adding this condition to the if, how about adding
an else like
> else if (DECL_MAYBE_IN_CHARGE_CDTOR_P (decl))
> /* For implicit instantiations of cdtors,
> if import_export_decl would use comdat linkage,
> make sure to use it right away, so that maybe_clone_body
> can use aliases. See PR113208. */
> maybe_make_one_only (decl);
?
Jason
next prev parent reply other threads:[~2024-04-24 22:39 UTC|newest]
Thread overview: 25+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-04-17 7:42 [PATCH] c++: " Jakub Jelinek
2024-04-17 9:04 ` Jan Hubicka
2024-04-17 12:32 ` Jakub Jelinek
2024-04-17 13:26 ` Jan Hubicka
2024-04-17 14:07 ` Jakub Jelinek
2024-04-17 14:34 ` Jan Hubicka
2024-04-17 14:39 ` Jakub Jelinek
2024-04-22 15:42 ` [PATCH] c++, v2: " Jakub Jelinek
2024-04-23 3:14 ` Jason Merrill
2024-04-23 16:04 ` Jakub Jelinek
2024-04-24 9:16 ` Jonathan Wakely
2024-04-24 16:16 ` [PATCH] c++, v3: " Jakub Jelinek
2024-04-24 22:39 ` Jason Merrill [this message]
2024-04-24 22:47 ` Jakub Jelinek
2024-04-25 0:43 ` Jason Merrill
2024-04-25 12:02 ` [PATCH] c++, v4: " Jakub Jelinek
2024-04-25 14:22 ` Jakub Jelinek
2024-04-25 15:30 ` Jason Merrill
2024-04-25 18:42 ` [PATCH] c++, v5: " Jakub Jelinek
2024-05-09 18:20 ` [PATCH] c++: Optimize in maybe_clone_body aliases even when not at_eof [PR113208] Jakub Jelinek
2024-05-09 18:58 ` Marek Polacek
2024-05-09 19:05 ` Jakub Jelinek
2024-05-10 19:59 ` Jason Merrill
2024-05-13 10:19 ` Jakub Jelinek
2024-05-14 22:20 ` Jason Merrill
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=cc3d167a-3bd8-4cc8-96bc-c9c313cf2e7e@redhat.com \
--to=jason@redhat.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=jakub@redhat.com \
--cc=jh@suse.cz \
--cc=jwakely@redhat.com \
--cc=ppalka@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).