From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: by sourceware.org (Postfix, from userid 48) id E0BA83858423; Mon, 4 Oct 2021 17:15:10 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org E0BA83858423 From: "amacleod at redhat dot com" To: gcc-bugs@gcc.gnu.org Subject: [Bug tree-optimization/102540] [12 Regression] Dead Code Elimination Regression at -O3 since r12-476-gd846f225c25c5885 Date: Mon, 04 Oct 2021 17:15:10 +0000 X-Bugzilla-Reason: CC X-Bugzilla-Type: changed X-Bugzilla-Watch-Reason: None X-Bugzilla-Product: gcc X-Bugzilla-Component: tree-optimization X-Bugzilla-Version: 12.0 X-Bugzilla-Keywords: X-Bugzilla-Severity: normal X-Bugzilla-Who: amacleod at redhat dot com 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: 12.0 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 X-BeenThere: gcc-bugs@gcc.gnu.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Gcc-bugs mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Mon, 04 Oct 2021 17:15:11 -0000 https://gcc.gnu.org/bugzilla/show_bug.cgi?id=3D102540 --- Comment #6 from Andrew Macleod --- >=20 > > It removes a > > relationship between c_10 and _2. The reason ranger no longer can fold = _2 =3D=3D 0 > > is because the sequence is now: > >=20 > > a.0_1 =3D a; > > _2 =3D (unsigned int) a.0_1; > > b =3D _2; > > _6 =3D a.0_1 & 4294967295; > > c_10 =3D _6; > > if (c_10 !=3D 0) > > goto ; [INV] > >=20 > > We do not find _2 is non-zero on the outgoing edge because _2 is not re= lated to > > the calculation in the condition. (ie c_10 no longer has a dependency = on _2) > >=20 > > We do recalculate _2 based on the outgoing range of a.0_1, but with it = being a > > 64 bit value and _2 being 32 bits, we only know the outgoing range of a= .0_1 is > > non-zero.. we dont track any of the upper bits...=20 > > 2->3 (T) a.0_1 : long int [-INF, -1][1, +INF] > > And when we recalculate _2 using that value, we still get varying becau= se > > 0xFFFF0000 in not zero, but can still produce a zero in _2. > >=20 > > The problem is that the condition c_10 !=3D 0 no longer related to the = value of > > _2 in the IL... so ranger never sees it. and we cant represent the 2^16 > > subranges that end in [1,0xFFFF]. > >=20 > > Before that transformation,=20 > > _2 =3D (unsigned int) a.0_1; > > b =3D _2; > > c_10 =3D (long int) _2; > > The relationship is obvious, and ranger would relate the c_10 !=3D 0 to= _2 no > > problem. >=20 > I see - too bad. Note the transform made the dependence chain of _6 > one instruction shorter without increasing the number of instructions > so it's a profitable transform. >=20 > Btw, the relation is still there but only indirectly via a.0_1. The > old (E)VRP had this find_asserts(?) that produced assertions based > on the definitions - sth that now range-ops does(?), so it would > eventually have built assertions for a.0_1 for both conditions and > allow relations based on that? I can't seem to find my way around > the VRP code now - pieces moved all over the place and so my mind > fails me on the searching task :/ We do know that a.0_1 is non-zero on that edge: 2->3 (T) a.0_1 : long int [-INF, -1][1, +INF] the problem is that we can't currently represent that the bitmask operation causes all patterns ending in 0x00000000 to not occur.. we just leave it at ~[0,0]. which isn't sufficient for this use case.=20 we don't currently track any equivalences between values of different precision.. (even though ranger once did). Handling it as a general equivalence was fraught with issues.=20 We might be able to add a new equivalence class "slice" or something.. I had considered it but hadn't seen a great need case. This would make _6 a 32 = bit slice of a.0_1 with range [1, 0xffffffff]. Then when we are querying for the cast _2 =3D (unsigned int) a.0_1; we could also query the 32 bit equivalence slices of a.0_1, find _6, and get the outgoing range of [1,0xffffffff].. and apply that value. It would probably resolve an entire class of things where we don't recogniz= e an equivalence between a cast and a bitmask of equivalent precision. This would also mean the reverse would apply.. ie if we instead branched on= _2 !=3D 0 we would also understand that _6 will be non-zero.=