From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: by sourceware.org (Postfix, from userid 48) id BBCB9385840B; Tue, 12 Mar 2024 07:37:19 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org BBCB9385840B DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gcc.gnu.org; s=default; t=1710229039; bh=pNhmow49NouxgB1FLcjNPVKRLDqKMfxgBkklj1oSEt8=; h=From:To:Subject:Date:In-Reply-To:References:From; b=BeR9/3LdNQhGL4B/vY73PDfdIBEn0P7Jwd5EPJFhsK6aXs1DktPFCiBmOIRAI0VX3 7kZgn2LHrv/FsfsfqIJQqoVnHlfrwMTQBntNfOesYvOICS2pvWPEvcuSIs534n6lOC smFKwF2p4EXlpEEwVnOQ8SDuBTpwEq850NPKGWw4= From: "rguenth at gcc dot gnu.org" To: gcc-bugs@gcc.gnu.org Subject: [Bug middle-end/96564] [11/12/13/14 Regression] New maybe use of uninitialized variable warning since r11-959 Date: Tue, 12 Mar 2024 07:37:18 +0000 X-Bugzilla-Reason: CC X-Bugzilla-Type: changed X-Bugzilla-Watch-Reason: None X-Bugzilla-Product: gcc X-Bugzilla-Component: middle-end X-Bugzilla-Version: 11.0 X-Bugzilla-Keywords: alias, diagnostic, missed-optimization X-Bugzilla-Severity: normal X-Bugzilla-Who: rguenth at gcc dot gnu.org X-Bugzilla-Status: NEW X-Bugzilla-Resolution: X-Bugzilla-Priority: P2 X-Bugzilla-Assigned-To: unassigned at gcc dot gnu.org X-Bugzilla-Target-Milestone: 11.5 X-Bugzilla-Flags: X-Bugzilla-Changed-Fields: Message-ID: In-Reply-To: References: 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: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=3D96564 --- Comment #16 from Richard Biener --- (In reply to Andrew Macleod from comment #15) > (In reply to Richard Biener from comment #13) > > (In reply to Jeffrey A. Law from comment #12) > > > So I think we could solve this with a bit of help from the alias orac= le. We > > > have the routine ptrs_compare_unequal, but points-to-null is going t= o get > > > in the way. > > >=20 > > > I think VRP and DOM have enough information to rule out NULL for both > > > objects in question. So if we could query the points-to information, > > > ignoring NULL then we could likely solve this particular bug. > > >=20 > > > Essentially VRP or DOM would prove NULL isn't in the set of possible = values > > > at the comparison point. Then we query the alias information ignorin= g NULL. > > > Voila we compute a static result for the comparison of the two pointe= rs and > > > the problematical block becomes unreachable and the bogus warning goe= s away. > > >=20 > > > Richi, any thoughts in viability of such an API? > >=20 > > We now treat pt.null conservatively and track non-null-ness derived from > > range-info in it. That means when VRP/DOM can prove a pointer is always > > not NULL they can do set_ptr_nonnull (p) on it. > >=20 > > This means the > >=20 > > /* ??? We'd like to handle ptr1 !=3D NULL and ptr1 !=3D ptr2 > > but those require pt.null to be conservatively correct. */ > >=20 > > is no longer true and we could finally implement it, like with > >=20 > > diff --git a/gcc/tree-ssa-alias.cc b/gcc/tree-ssa-alias.cc > > index e7c1c1aa624..5b6d9e0aa6a 100644 > > --- a/gcc/tree-ssa-alias.cc > > +++ b/gcc/tree-ssa-alias.cc > > @@ -479,9 +479,25 @@ ptrs_compare_unequal (tree ptr1, tree ptr2) > > } > > return !pt_solution_includes (&pi->pt, obj1); > > } > > - > > - /* ??? We'd like to handle ptr1 !=3D NULL and ptr1 !=3D ptr2 > > - but those require pt.null to be conservatively correct. */ > > + else if (TREE_CODE (ptr1) =3D=3D SSA_NAME) > > + { > > + struct ptr_info_def *pi1 =3D SSA_NAME_PTR_INFO (ptr1); > > + if (!pi1 > > + || pi1->pt.vars_contains_restrict > > + || pi1->pt.vars_contains_interposable) > > + return false; > > + if (integer_zerop (ptr2) && !pi1->pt.null) > > + return true; > > + if (TREE_CODE (ptr2) =3D=3D SSA_NAME) > > + { > > + struct ptr_info_def *pi2 =3D SSA_NAME_PTR_INFO (ptr2); > > + if (!pi2 > > + || pi2->pt.vars_contains_restrict > > + || pi2->pt.vars_contains_interposable) > > + if (!pi1->pt.null || !pi2->pt.null) > > + return !pt_solutions_intersect (&pi1->pt, &pi2->pt); > > + } > > + } > >=20=20 > > return false; > > } > >=20 > >=20 > > but the testcase shows the non-null-ness is only conditional which means > > we'd have to use a range query above which necessarily falls back to > > the global ranges given we don't have any context available here. The > > old EVRP adjusted global ranges during the walk but this is no longer d= one. > >=20 > You mean it lied? because x_1 is not NULL until after _8 =3D *x_1(D); > executes. It can still be NULL on that stmt can it not? Did it reset t= he > global value afterwards? Yes and yes, old EVRP turned global ranges into "ranges at the point of stmt evaluation/folding" (and restored them to the global values later). Note that EVRP didn't do any sort of iteration for loop handling and it folded stmts as it analyzed them. Using the global ranges as "lattice" had the advantage that all folding utilities picked up "local" ranges for free. IMO it was quite elegant and fast what EVRP did (with it's obvious limitations of course). > Contextually ranger knows both are non-null at EVRP time: > a.0_27 : [irange] int[0:D.xxxx] * [1, +INF] > 2->3 (T) x_1(D) : [irange] int * [1, +INF] > 2->3 (T) a.0_27 : [irange] int[0:D.xxxx] * [1, +INF] > 2->4 (F) x_1(D) : [irange] int * [1, +INF] > 2->4 (F) a.0_27 : [irange] int[0:D.xxxx] * [1, +INF] >=20 > So we know x_1 is non-NULL after the de-reference for the rest of the blo= ck > (and function). It also sets a.0_27 globally to be [1, +INF]. >=20 > > Note it's enough that one pointer is nonnull, so for your idea the > > API could be extended with a bool one_ptr_nonnull parameter. >=20 > ranger currently sets a.0 globally to be non-null in EVRP. After EVRP I see # PT =3D nonlocal null unsigned int * x_8(D) =3D x; ... : _1 =3D *x_8(D); # RANGE [irange] long unsigned int [0, 4294967295] MASK 0xffffffff VALUE = 0x0 _2 =3D (long unsigned int) _1; # PT =3D null { D.2781 } # ALIGN =3D 8, MISALIGN =3D 0 # USE =3D nonlocal escaped # CLB =3D nonlocal escaped a_10 =3D malloc (_2); if (a_10 =3D=3D 0B) ... : if (x_8(D) !=3D a_10) the last test is the one we want to eliminate. The proposed change (with a missing 'return false' fixed) isn't enough since it just looks at global ranges where both x_8 and a_10 can be null. In the ptrs_compare_unequal function I could at most use range_of_expr without a stmt context (as I don't have that) which wouldn't help. EVRP with the trick to adjust global ranges effectively had a "global context" it would use. I suppose that one could have something like that for ranger as well, add ranger::set_context (gimple *) which EVRP could set when folding a stmt (set it to right before 'stmt' execution) and which would be the context to fall back to when a folding dependent utility didn't specify one? Adding a global (not ranger specific) "folding context stack" might do the trick as well. Any utility that could take advantage of a context could look at the stack top (which might be NULL). That would be less churn than wrapping each and every folding function inside a "folder" class containing a context. Of course one has to be careful with such a thing, like with recursively invoking number of iteration or SCEV analysis which work on a more fuzzy context than a specific stmt contained in a loop.=