From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from nikam.ms.mff.cuni.cz (nikam.ms.mff.cuni.cz [195.113.20.16]) by sourceware.org (Postfix) with ESMTPS id 9DA9D3858D20 for ; Wed, 17 Apr 2024 09:04:28 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.2 sourceware.org 9DA9D3858D20 Authentication-Results: sourceware.org; dmarc=fail (p=none dis=none) header.from=ucw.cz Authentication-Results: sourceware.org; spf=pass smtp.mailfrom=kam.mff.cuni.cz ARC-Filter: OpenARC Filter v1.0.0 sourceware.org 9DA9D3858D20 Authentication-Results: server2.sourceware.org; arc=none smtp.remote-ip=195.113.20.16 ARC-Seal: i=1; a=rsa-sha256; d=sourceware.org; s=key; t=1713344679; cv=none; b=mt6suIk/Yfq5okepeT+FWvRZMUXbIp8o0SNsHnzM4JT4oLFW3M/gHh4QgFsx7vgXyu15P2iRIoR7ZmnrurXYmkKxiYAOkln9QIzZ5ywLljqEL8rzPsVna0M/8/7C/qc0hDOTaQaEMhrfZAkrqclOXaS5Dnym/ntSsQNN6mInH/o= ARC-Message-Signature: i=1; a=rsa-sha256; d=sourceware.org; s=key; t=1713344679; c=relaxed/simple; bh=fNWsEzAmAOI2yA4QfZwYlPM2yxe+Th+JMFcxXQ/uhLk=; h=DKIM-Signature:Date:From:To:Subject:Message-ID:MIME-Version; b=pMoF6IwmrhwfRNnc9OOOcHTW1jZGKT+zat2Z/RkpSarnlZzAYY+PybTVIlrxhsrhs5obRjNqPd4TkNZh+cVC9IzCQZrk+plr310iqJNC4VPZIrZHzcehISGZjNPniFvihPSJK6f0z6qbVjVit46RuqR0i+xydow3dSTXQcpVSXQ= ARC-Authentication-Results: i=1; server2.sourceware.org Received: by nikam.ms.mff.cuni.cz (Postfix, from userid 16202) id B375A283811; Wed, 17 Apr 2024 11:04:26 +0200 (CEST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=ucw.cz; s=gen1; t=1713344666; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version:content-type:content-type: in-reply-to:in-reply-to:references:references; bh=bOxKLNLB/oXEn3xeXz4fOdOJgmAWtT5hQEs5oXYEkyo=; b=CJFSXfqtxCh4EYrXoLrv1TqQp/niS9+2LI4dEwUw73eqOyZG6x2/2GN+xve0oVkGZuXNmx d1+ggnR3Z8t19YulkKIc8FqbqdS+c+oUUAqWsiwaOjpCe4WR6OUssMvaadh0lLJQLljlaK FBpF5tHd/upkQWMsBm0x0LfT+bj+a5Q= Date: Wed, 17 Apr 2024 11:04:26 +0200 From: Jan Hubicka To: Jakub Jelinek Cc: Jason Merrill , Richard Biener , Patrick Palka , gcc-patches@gcc.gnu.org Subject: Re: [PATCH] c++: Retry the aliasing of base/complete cdtor optimization at import_export_decl time [PR113208] Message-ID: References: MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: X-Spam-Status: No, score=-4.6 required=5.0 tests=BAYES_00,DKIM_SIGNED,DKIM_VALID,DKIM_VALID_AU,HEADER_FROM_DIFFERENT_DOMAINS,JMQ_SPF_NEUTRAL,RCVD_IN_MSPIKE_H3,RCVD_IN_MSPIKE_WL,SPF_HELO_NONE,SPF_PASS,TXREP autolearn=no autolearn_force=no version=3.4.6 X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on server2.sourceware.org List-Id: > Hi! Hello, > The reason the optimization doesn't trigger when the constructor is > constexpr is that expand_or_defer_fn is called in that case much earlier > than when it is not constexpr; in the former case it is called when we > try to constant evaluate that constructor. But DECL_INTERFACE_KNOWN > is false in that case and comdat_linkage hasn't been called either > (I think it is desirable, because comdat group is stored in the cgraph > node and am not sure it is a good idea to create cgraph nodes for > something that might not be needed later on at all), so maybe_clone_body > clones the bodies, but doesn't make them as aliases. Thanks for working this out! Creating cgraph node with no body is harmless as it will be removed later. > > The following patch is an attempt to redo this optimization when > import_export_decl is called at_eof time on the base/complete cdtor > (or deleting dtor). It will not do anything if maybe_clone_body > hasn't been called uyet (the TREE_ASM_WRITTEN check on the > DECL_MAYBE_IN_CHARGE_CDTOR_P), or when one or both of the base/complete > cdtors have been lowered already, or when maybe_clone_body called > maybe_thunk_body and it was successful. Otherwise retries the > can_alias_cdtor check and makes the complete cdtor alias to the > base cdtor with adjustments to the comdat group. > > Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk? > > 2024-04-17 Jakub Jelinek > > PR lto/113208 > * cp-tree.h (maybe_optimize_cdtor): Declare. > * decl2.cc (import_export_decl): Call it for cloned cdtors. > * optimize.cc (maybe_optimize_cdtor): New function. > > * 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/cp-tree.h.jj 2024-04-16 17:18:37.286279533 +0200 > +++ gcc/cp/cp-tree.h 2024-04-16 17:45:01.685635709 +0200 > @@ -7447,6 +7447,7 @@ extern bool handle_module_option (unsign > /* In optimize.cc */ > extern tree clone_attrs (tree); > extern bool maybe_clone_body (tree); > +extern void maybe_optimize_cdtor (tree); > > /* In parser.cc */ > extern tree cp_convert_range_for (tree, tree, tree, cp_decomp *, bool, > --- gcc/cp/decl2.cc.jj 2024-04-16 17:18:37.287279519 +0200 > +++ gcc/cp/decl2.cc 2024-04-16 17:45:01.686635695 +0200 > @@ -3568,6 +3568,9 @@ import_export_decl (tree decl) > } > > DECL_INTERFACE_KNOWN (decl) = 1; > + > + if (DECL_CLONED_FUNCTION_P (decl)) > + maybe_optimize_cdtor (decl); > } > > /* Return an expression that performs the destruction of DECL, which > --- gcc/cp/optimize.cc.jj 2024-04-16 17:18:37.374278327 +0200 > +++ gcc/cp/optimize.cc 2024-04-16 21:35:53.597188774 +0200 > @@ -723,3 +723,58 @@ maybe_clone_body (tree fn) > /* We don't need to process the original function any further. */ > return 1; > } > + > +/* If maybe_clone_body is called while the cdtor is still tentative, > + DECL_ONE_ONLY will be false and so will be can_alias_cdtor (fn). > + In that case we wouldn't try to optimize using an alias and instead > + would emit separate base and complete cdtor. The following function > + attempts to still optimize that case when we import_export_decl > + is called first time on one of the clones. */ > + > +void > +maybe_optimize_cdtor (tree orig_decl) > +{ > + tree fns[3]; > + tree fn = DECL_CLONED_FUNCTION (orig_decl); > + gcc_checking_assert (DECL_MAYBE_IN_CHARGE_CDTOR_P (fn)); > + if (DECL_INTERFACE_KNOWN (fn) > + || !TREE_ASM_WRITTEN (fn) > + || !DECL_ONE_ONLY (orig_decl) > + || symtab->global_info_ready) > + return; > + > + populate_clone_array (fn, fns); > + > + if (!fns[0] || !fns[1]) > + return; > + for (int i = 2 - !fns[2]; i >= 0; --i) > + if (fns[i] != orig_decl && DECL_INTERFACE_KNOWN (fns[i])) > + return; > + DECL_INTERFACE_KNOWN (fn) = 1; > + comdat_linkage (fn); > + if (!can_alias_cdtor (fn)) > + return; > + /* For comdat base and complete cdtors put them into the same, > + *[CD]5* comdat group instead of *[CD][12]*. */ > + auto n0 = cgraph_node::get_create (fns[0]); > + auto n1 = cgraph_node::get_create (fns[1]); > + auto n2 = fns[2] ? cgraph_node::get_create (fns[1]) : NULL; > + if (n0->lowered || n1->lowered || (n2 && n2->lowered)) > + return; > + import_export_decl (fns[0]); > + n1->definition = false; > + if (!n0->create_same_body_alias (fns[1], fns[0])) > + return; > + > + tree comdat_group = cdtor_comdat_group (fns[1], fns[0]); > + n1 = cgraph_node::get (fns[1]); > + n0->set_comdat_group (comdat_group); > + if (n1->same_comdat_group) > + n1->remove_from_same_comdat_group (); > + n1->add_to_same_comdat_group (n0); > + if (fns[2]) > + n2->add_to_same_comdat_group (n0); > + import_export_decl (fns[1]); So this is handling the case where symbol was already inserted into one comdat group and later we need to move it into the C5 group? As long as we move everythingf rom old comdat group, this should be fine. > + /* Remove the body now that it is an alias. */ > + DECL_SAVED_TREE (fns[1]) = NULL_TREE; Maybe using release_function_body since it also knows how to remove DECL_STRUCT_FUNCTION that exists at this stage? I was thinking how to solve the problem on cgrpah side. We definitely have long lasting bug where aliases are handled incorrectly for which I made WIP patch (but probably would like to hold it after release branch is created). When foo has alias bar and foo is praviled then the alias target is prevailed too. This is what causes the ICE about cross comdat section alias. However fixing this is not enough as I do not think we can handle incremental linking correctly (as discussed briefly on IRC technically we should keep both sections but that would require two symbols of same name in single .o file). With the aliasing fixed we turn the other symbol into static but keep alias, so we end up with one comdat group having the non-aliased constructor and second comdat group (C5) exporting only the alias, which is not quite reight either. Honza