From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <gcc-bugzilla@gcc.gnu.org>
Received: by sourceware.org (Postfix, from userid 48)
	id 364F53857BAC; Wed, 14 Sep 2022 13:58:43 +0000 (GMT)
DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 364F53857BAC
DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gcc.gnu.org;
	s=default; t=1663163923;
	bh=3r39pR8PoLoP3Ifkk42A15XsY1PmpIMhBKZ90Gi7iF4=;
	h=From:To:Subject:Date:In-Reply-To:References:From;
	b=E77Sboo3CyW1ibo7SB/QAOQIQQjWFLyb23+vdpK18inW94iTzRcLsej1nEDD7SEXq
	 9e6bR7UM+giBpu/PRBnl07/Pfw8NIvZzcYMHuKyb3YM4+c/DCrblO5jMbN44UfVPLP
	 tgzoJyN9SeCcRpia3/I01k8OAKa9jxBgODmStkCk=
From: "marxin at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug ipa/106816] noreturn/pure attributes are not set correctly on
 multiversioned functions
Date: Wed, 14 Sep 2022 13:58:43 +0000
X-Bugzilla-Reason: CC
X-Bugzilla-Type: changed
X-Bugzilla-Watch-Reason: None
X-Bugzilla-Product: gcc
X-Bugzilla-Component: ipa
X-Bugzilla-Version: 12.2.0
X-Bugzilla-Keywords: 
X-Bugzilla-Severity: normal
X-Bugzilla-Who: marxin at gcc dot gnu.org
X-Bugzilla-Status: NEW
X-Bugzilla-Resolution: 
X-Bugzilla-Priority: P3
X-Bugzilla-Assigned-To: unassigned at gcc dot gnu.org
X-Bugzilla-Target-Milestone: ---
X-Bugzilla-Flags: 
X-Bugzilla-Changed-Fields: 
Message-ID: <bug-106816-4-kxwBDQoAea@http.gcc.gnu.org/bugzilla/>
In-Reply-To: <bug-106816-4@http.gcc.gnu.org/bugzilla/>
References: <bug-106816-4@http.gcc.gnu.org/bugzilla/>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
X-Bugzilla-URL: http://gcc.gnu.org/bugzilla/
Auto-Submitted: auto-generated
MIME-Version: 1.0
List-Id: <gcc-bugs.sourceware.org>

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=3D106816

--- Comment #6 from Martin Li=C5=A1ka <marxin at gcc dot gnu.org> ---
(In reply to Richard Biener from comment #5)
> The function should probably inherit all of the IPA pure/const/modref
> analysis result, that is all "IPA" state should be copied.  I think we wa=
nt
> some helper
> here - IPA clone creation must have something, no?

Well, in IPA clones utilize so-called function_summary that is a typical pl=
ace
where we store analysis results. And we don't typically create a new tree
declarations when we clone cgraph_nodes (clones share the same FE declarati=
on).

However, I noticed we want to do likely something similar to what cp/decl.c
does:

static void
merge_attribute_bits (tree newdecl, tree olddecl)
{
  TREE_THIS_VOLATILE (newdecl) |=3D TREE_THIS_VOLATILE (olddecl);
  TREE_THIS_VOLATILE (olddecl) |=3D TREE_THIS_VOLATILE (newdecl);
  TREE_NOTHROW (newdecl) |=3D TREE_NOTHROW (olddecl);
  TREE_NOTHROW (olddecl) |=3D TREE_NOTHROW (newdecl);
  TREE_READONLY (newdecl) |=3D TREE_READONLY (olddecl);
  TREE_READONLY (olddecl) |=3D TREE_READONLY (newdecl);
  DECL_IS_MALLOC (newdecl) |=3D DECL_IS_MALLOC (olddecl);
  DECL_IS_MALLOC (olddecl) |=3D DECL_IS_MALLOC (newdecl);
  DECL_PURE_P (newdecl) |=3D DECL_PURE_P (olddecl);
  DECL_PURE_P (olddecl) |=3D DECL_PURE_P (newdecl);
  DECL_UNINLINABLE (newdecl) |=3D DECL_UNINLINABLE (olddecl);
  DECL_UNINLINABLE (olddecl) |=3D DECL_UNINLINABLE (newdecl);
}

...

              /* Merge the noreturn bit.  */
              TREE_THIS_VOLATILE (olddecl) =3D TREE_THIS_VOLATILE (newdecl);
              TREE_READONLY (olddecl) =3D TREE_READONLY (newdecl);
              TREE_NOTHROW (olddecl) =3D TREE_NOTHROW (newdecl);
              DECL_IS_MALLOC (olddecl) =3D DECL_IS_MALLOC (newdecl);
              DECL_PURE_P (olddecl) =3D DECL_PURE_P (newdecl);

I can see IPA passes doing similar declaration clonning for VAR_DECLs
(gcc/ipa-param-manipulation.cc, gcc/tree-inline.c).

@Martin: Do we have a declaration cloning code for functions somewhere?=