From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-ej1-x635.google.com (mail-ej1-x635.google.com [IPv6:2a00:1450:4864:20::635]) by sourceware.org (Postfix) with ESMTPS id 3263538582B3 for ; Mon, 8 Aug 2022 08:57:02 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org 3263538582B3 Received: by mail-ej1-x635.google.com with SMTP id k26so15362853ejx.5 for ; Mon, 08 Aug 2022 01:57:02 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:mime-version:references:in-reply-to:from:date :message-id:subject:to:cc:content-transfer-encoding; bh=TbAA4XHLHzXPi6g6pGxY2U1oL50nR/pdyjJdu/X5P6A=; b=zcHDrn51BqO7OPwsNFZ+aLqvsxp9wB39t4YvYNA13n9xTb/3yTfPvhTNiu6oJpONrg hozL06HidCJ5c04GwMp7aXExtLqSoE7CvBbe5l9ynnG3K2P/IvDFhKY4EZqniuHBaYqm DgBMNiQCML4gjQFOPwnkJZ9N4ac8gYjT3rU+i/kc/s886AosinuoXAxA5BWu6vsYy9aO nvKy4Z/dCTQG2UGQO4cu/u4YwyIKoNHTz99+uJj5Jt32aFZjOg+Mr1rh0Cxxo03KNf4q vMID1Xh+lcBhRQ76z39VALjgCmN/dC6smdoOv20YY/GgNWzdJnSnj9OBB2K/SAmMvpBD UeDw== X-Gm-Message-State: ACgBeo33n6xg/4E6+CG5mReRZJBrrWHqSaV0UdrdsZsgYB3yI+YwDtui jkkOs9LMwXBWt32wKrcc4w7UtQzDJarC/zKiAvM= X-Google-Smtp-Source: AA6agR6mlybqBSGfj4GpKLv1hQffbDW3llOMd361A8nV/mbwuvjY7bKEdIu2gS/7gomXwu89wPuxdjlqQvG4A0B/d6Y= X-Received: by 2002:a17:907:948f:b0:731:3f2e:8916 with SMTP id dm15-20020a170907948f00b007313f2e8916mr5433106ejc.442.1659949020512; Mon, 08 Aug 2022 01:57:00 -0700 (PDT) MIME-Version: 1.0 References: In-Reply-To: From: Richard Biener Date: Mon, 8 Aug 2022 10:56:48 +0200 Message-ID: Subject: Re: ICE after folding svld1rq to vec_perm_expr duing forwprop To: Prathamesh Kulkarni Cc: gcc Patches , Richard Sandiford Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable X-Spam-Status: No, score=-2.1 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, FREEMAIL_FROM, RCVD_IN_DNSWL_NONE, SPF_HELO_NONE, SPF_PASS, TXREP, T_SCC_BODY_TEXT_LINE, WEIRD_PORT autolearn=ham autolearn_force=no version=3.4.6 X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on server2.sourceware.org X-BeenThere: gcc-patches@gcc.gnu.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Gcc-patches mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Mon, 08 Aug 2022 08:57:05 -0000 On Mon, Aug 1, 2022 at 5:17 AM Prathamesh Kulkarni wrote: > > On Thu, 21 Jul 2022 at 12:21, Richard Biener = wrote: > > > > On Wed, Jul 20, 2022 at 5:36 PM Prathamesh Kulkarni > > wrote: > > > > > > On Mon, 18 Jul 2022 at 11:57, Richard Biener wrote: > > > > > > > > On Fri, Jul 15, 2022 at 3:49 PM Prathamesh Kulkarni > > > > wrote: > > > > > > > > > > On Thu, 14 Jul 2022 at 17:22, Richard Sandiford > > > > > wrote: > > > > > > > > > > > > Richard Biener writes: > > > > > > > On Thu, Jul 14, 2022 at 9:55 AM Prathamesh Kulkarni > > > > > > > wrote: > > > > > > >> > > > > > > >> On Wed, 13 Jul 2022 at 12:22, Richard Biener wrote: > > > > > > >> > > > > > > > >> > On Tue, Jul 12, 2022 at 9:12 PM Prathamesh Kulkarni via Gc= c-patches > > > > > > >> > wrote: > > > > > > >> > > > > > > > > >> > > Hi Richard, > > > > > > >> > > For the following test: > > > > > > >> > > > > > > > > >> > > svint32_t f2(int a, int b, int c, int d) > > > > > > >> > > { > > > > > > >> > > int32x4_t v =3D (int32x4_t) {a, b, c, d}; > > > > > > >> > > return svld1rq_s32 (svptrue_b8 (), &v[0]); > > > > > > >> > > } > > > > > > >> > > > > > > > > >> > > The compiler emits following ICE with -O3 -mcpu=3Dgeneri= c+sve: > > > > > > >> > > foo.c: In function =E2=80=98f2=E2=80=99: > > > > > > >> > > foo.c:4:11: error: non-trivial conversion in =E2=80=98vi= ew_convert_expr=E2=80=99 > > > > > > >> > > 4 | svint32_t f2(int a, int b, int c, int d) > > > > > > >> > > | ^~ > > > > > > >> > > svint32_t > > > > > > >> > > __Int32x4_t > > > > > > >> > > _7 =3D VIEW_CONVERT_EXPR<__Int32x4_t>(_8); > > > > > > >> > > during GIMPLE pass: forwprop > > > > > > >> > > dump file: foo.c.109t.forwprop2 > > > > > > >> > > foo.c:4:11: internal compiler error: verify_gimple faile= d > > > > > > >> > > 0xfda04a verify_gimple_in_cfg(function*, bool) > > > > > > >> > > ../../gcc/gcc/tree-cfg.cc:5568 > > > > > > >> > > 0xe9371f execute_function_todo > > > > > > >> > > ../../gcc/gcc/passes.cc:2091 > > > > > > >> > > 0xe93ccb execute_todo > > > > > > >> > > ../../gcc/gcc/passes.cc:2145 > > > > > > >> > > > > > > > > >> > > This happens because, after folding svld1rq_s32 to vec_p= erm_expr, we have: > > > > > > >> > > int32x4_t v; > > > > > > >> > > __Int32x4_t _1; > > > > > > >> > > svint32_t _9; > > > > > > >> > > vector(4) int _11; > > > > > > >> > > > > > > > > >> > > : > > > > > > >> > > _1 =3D {a_3(D), b_4(D), c_5(D), d_6(D)}; > > > > > > >> > > v_12 =3D _1; > > > > > > >> > > _11 =3D v_12; > > > > > > >> > > _9 =3D VEC_PERM_EXPR <_11, _11, { 0, 1, 2, 3, ... }>; > > > > > > >> > > return _9; > > > > > > >> > > > > > > > > >> > > During forwprop, simplify_permutation simplifies vec_per= m_expr to > > > > > > >> > > view_convert_expr, > > > > > > >> > > and the end result becomes: > > > > > > >> > > svint32_t _7; > > > > > > >> > > __Int32x4_t _8; > > > > > > >> > > > > > > > > >> > > ;; basic block 2, loop depth 0 > > > > > > >> > > ;; pred: ENTRY > > > > > > >> > > _8 =3D {a_2(D), b_3(D), c_4(D), d_5(D)}; > > > > > > >> > > _7 =3D VIEW_CONVERT_EXPR<__Int32x4_t>(_8); > > > > > > >> > > return _7; > > > > > > >> > > ;; succ: EXIT > > > > > > >> > > > > > > > > >> > > which causes the error duing verify_gimple since VIEW_CO= NVERT_EXPR > > > > > > >> > > has incompatible types (svint32_t, int32x4_t). > > > > > > >> > > > > > > > > >> > > The attached patch disables simplification of VEC_PERM_E= XPR > > > > > > >> > > in simplify_permutation, if lhs and rhs have non compati= ble types, > > > > > > >> > > which resolves ICE, but am not sure if it's the correct = approach ? > > > > > > >> > > > > > > > >> > It for sure papers over the issue. I think the error happ= ens earlier, > > > > > > >> > the V_C_E should have been built with the type of the VEC_= PERM_EXPR > > > > > > >> > which is the type of the LHS. But then you probably run i= nto the > > > > > > >> > different sizes ICE (VLA vs constant size). I think for t= his case you > > > > > > >> > want a BIT_FIELD_REF instead of a VIEW_CONVERT_EXPR, > > > > > > >> > selecting the "low" part of the VLA vector. > > > > > > >> Hi Richard, > > > > > > >> Sorry I don't quite follow. In this case, we use VEC_PERM_EX= PR to > > > > > > >> represent dup operation > > > > > > >> from fixed width to VLA vector. I am not sure how folding it= to > > > > > > >> BIT_FIELD_REF will work. > > > > > > >> Could you please elaborate ? > > > > > > >> > > > > > > >> Also, the issue doesn't seem restricted to this case. > > > > > > >> The following test case also ICE's during forwprop: > > > > > > >> svint32_t foo() > > > > > > >> { > > > > > > >> int32x4_t v =3D (int32x4_t) {1, 2, 3, 4}; > > > > > > >> svint32_t v2 =3D svld1rq_s32 (svptrue_b8 (), &v[0]); > > > > > > >> return v2; > > > > > > >> } > > > > > > >> > > > > > > >> foo2.c: In function =E2=80=98foo=E2=80=99: > > > > > > >> foo2.c:9:1: error: non-trivial conversion in =E2=80=98vector= _cst=E2=80=99 > > > > > > >> 9 | } > > > > > > >> | ^ > > > > > > >> svint32_t > > > > > > >> int32x4_t > > > > > > >> v2_4 =3D { 1, 2, 3, 4 }; > > > > > > >> > > > > > > >> because simplify_permutation folds > > > > > > >> VEC_PERM_EXPR< {1, 2, 3, 4}, {1, 2, 3, 4}, {0, 1, 2, 3, ...}= > > > > > > > >> into: > > > > > > >> vector_cst {1, 2, 3, 4} > > > > > > >> > > > > > > >> and it complains during verify_gimple_assign_single because = we don't > > > > > > >> support assignment of vector_cst to VLA vector. > > > > > > >> > > > > > > >> I guess the issue really is that currently, only VEC_PERM_EX= PR > > > > > > >> supports lhs and rhs > > > > > > >> to have vector types with differing lengths, and simplifying= it to > > > > > > >> other tree codes, like above, > > > > > > >> will result in type errors ? > > > > > > > > > > > > > > That might be the case - Richard should know. > > > > > > > > > > > > I don't see anything particularly special about VEC_PERM_EXPR h= ere, > > > > > > or about the VLA vs. VLS thing. We would have the same issue t= rying > > > > > > to build a 128-bit vector from 2 64-bit vectors. And there are= other > > > > > > tree codes whose input types are/can be different from their ou= tput > > > > > > types. > > > > > > > > > > > > So it just seems like a normal type correctness issue: a VEC_PE= RM_EXPR > > > > > > of type T needs to be replaced by something of type T. Whether= T has a > > > > > > constant size or a variable size doesn't matter. > > > > > > > > > > > > > If so your type check > > > > > > > is still too late, you should instead recognize that we are p= ermuting > > > > > > > a VLA vector and then refuse to go any of the non-VEC_PERM ge= nerating > > > > > > > paths - that probably means just allowing the code =3D=3D VEC= _PERM_EXPR > > > > > > > case and not any of the CTOR/CST/VIEW_CONVERT_EXPR cases? > > > > > > > > > > > > Yeah. If we're talking about the match.pd code, I think only: > > > > > > > > > > > > (if (sel.series_p (0, 1, 0, 1)) > > > > > > { op0; } > > > > > > (if (sel.series_p (0, 1, nelts, 1)) > > > > > > { op1; } > > > > > > > > > > > > need a type compatibility check. For fold_vec_perm I think > > > > > > we should just rearrange: > > > > > > > > > > > > gcc_assert (known_eq (TYPE_VECTOR_SUBPARTS (type), nelts) > > > > > > && known_eq (TYPE_VECTOR_SUBPARTS (TREE_TYPE (arg= 0)), nelts) > > > > > > && known_eq (TYPE_VECTOR_SUBPARTS (TREE_TYPE (arg= 1)), nelts)); > > > > > > if (TREE_TYPE (TREE_TYPE (arg0)) !=3D TREE_TYPE (type) > > > > > > || TREE_TYPE (TREE_TYPE (arg1)) !=3D TREE_TYPE (type)) > > > > > > return NULL_TREE; > > > > > > > > > > > > so that the assert comes after the early-out. > > > > > > > > > > > > It would be good at some point to relax fold_vec_perm to cases = where the > > > > > > outputs are a different length from the inputs, since the all-c= onstant > > > > > > VEC_PERM_EXPR above could be folded to a VECTOR_CST. > > > > > Hi, > > > > > For the above case, I think the issue is that simplify_permutatio= n > > > > > uses TREE_TYPE (arg0) for res_type, > > > > > while it should now use type for lhs. > > > > > > > > > > /* Shuffle of a constructor. */ > > > > > bool ret =3D false; > > > > > tree res_type =3D TREE_TYPE (arg0); > > > > > tree opt =3D fold_ternary (VEC_PERM_EXPR, res_type, arg0, a= rg1, op2); > > > > > > > > > > Using, res_type =3D TREE_TYPE (gimple_get_lhs (stmt)), > > > > > resolves the ICE, and emits all constant VEC_PERM_EXPR: > > > > > > > > > > v2_4 =3D VEC_PERM_EXPR <{ 1, 2, 3, 4 }, { 1, 2, 3, 4 }, { 0, 1,= 2, 3, ... }>; > > > > > return v2_4; > > > > > > > > > > Does the patch look OK to commit after bootstrap+test ? > > > > > > > > Ok with using gimple_assign_lhs (stmt) instead of gimple_get_lhs (s= tmt). > > > Hi, > > > I committed the patch but unfortunately it caused PR106360. > > > The issue is that for slp-reduc-sad-2.c on ppc64le, > > > simplify_permutation sees the following during forwprop4: > > > > > > _78 =3D (void *) ivtmp.21_73; > > > _92 =3D MEM [(uint8_t *)_78]; > > > _91 =3D {_92, 0}; > > > vect__1.6_90 =3D VIEW_CONVERT_EXPR(_91); > > > _88 =3D MEM [(uint8_t *)_78 + 16B]; > > > _87 =3D {_88, 0}; > > > vect__1.7_86 =3D VIEW_CONVERT_EXPR(_87); > > > vect__1.8_85 =3D VEC_PERM_EXPR > > 3, 4, 5, 6, 7, 16, 17, 18, 19, 20, 21, 22, 23 }>; > > > > > > So for, > > > tree res_type =3D TREE_TYPE (gimple_assign_lhs (stmt)); > > > tree opt =3D fold_ternary (VEC_PERM_EXPR, res_type, arg0, arg1, op2); > > > > > > we have: > > > res_type =3D V16QI > > > arg0 =3D {_92, 0} > > > arg1 =3D {_88, 0} > > > op2 =3D {0, 2} > > > > > > and thus we hit the following assert in fold_vec_perm: > > > > > > gcc_assert (known_eq (TYPE_VECTOR_SUBPARTS (type), nelts) > > > && known_eq (TYPE_VECTOR_SUBPARTS (TREE_TYPE (arg0)), n= elts) > > > && known_eq (TYPE_VECTOR_SUBPARTS (TREE_TYPE (arg1)), n= elts)); > > > > > > since nelts =3D=3D 2, and TYPE_VECTOR_SUBPARTS (type) =3D=3D 16. > > > > > > If we revert the committed patch so we pass res_type =3D TREE_TYPE (a= rg0) instead, > > > it simplifies the above VEC_PERM_EXPR to VIEW_CONVERT_EXPR: > > > _92 =3D MEM [(uint8_t *)_78]; > > > _88 =3D MEM [(uint8_t *)_78 + 16B]; > > > _5 =3D {_92, _88}; > > > vect__1.8_85 =3D VIEW_CONVERT_EXPR(_5); > > > > > > I suppose it's legal to cast vector of one type to another as long as > > > sizes match ? > > > > > > IIUC, the above VIEW_CONVERT_EXPR will result in: > > > vect__1.8_85 =3D { (uint8_t) _92, 0, 0, 0, 0, 0, 0, 0, (uint8_t) _88,= 0, > > > 0, 0, 0, 0, 0, 0 } ? > > > > > > In the attached patch, it restores res_type to TREE_TYPE (arg0), and = checks > > > if lhs_type and res_type differ but have same size, and in that case = emit: > > > lhs =3D VIEW_CONVERT_EXPR (opt), > > > instead of: > > > lhs =3D VIEW_CONVERT_EXPR (opt) > > > where opt is result of folding VEC_PERM_EXPR > > > > > > Does it look OK ? > > > > Definitely the original change was bogus. > > > > + if (!operand_equal_p (TYPE_SIZE (lhs_type), TYPE_SIZE (res_ty= pe))) > > + return 0; > > > > just repeats your very original change though ... I'll note that > > fold_ternary will > > ICE on now valid VEC_PERM_EXPRs so we should fix it, possibly by > > returning NULL_TREE on cases it does not handle. > > > > I think what should be done is, in the > > > > /* If there are any VIEW_CONVERT_EXPRs found when finding permuta= tion > > operands source, check whether it's valid to transform and pre= pare > > the required new operands. */ > > if (code =3D=3D VIEW_CONVERT_EXPR || code2 =3D=3D VIEW_CONVERT_EX= PR) > > { > > ... > > > > path also transform the expected result type. It should remain V_C_E c= ompatible > > to TREE_TYPE (lhs) but get a new element type. > > > > But as said, > > > > tree > > fold_vec_perm (tree type, tree arg0, tree arg1, const vec_perm_indices = &sel) > > { > > unsigned int i; > > unsigned HOST_WIDE_INT nelts; > > bool need_ctor =3D false; > > > > if (!sel.length ().is_constant (&nelts)) > > return NULL_TREE; > > gcc_assert (known_eq (TYPE_VECTOR_SUBPARTS (type), nelts) > > && known_eq (TYPE_VECTOR_SUBPARTS (TREE_TYPE (arg0)), nel= ts) > > && known_eq (TYPE_VECTOR_SUBPARTS (TREE_TYPE (arg1)), nel= ts)); > > > > ^^^ this doesn't match what we allow for VEC_PERM_EXPRs now and fold_te= rnary > > doesn't guard according to those asserts (I think we should extend fold= _vec_perm > > to support the new constraints). > Hi Richard, > Thanks for the suggestions and sorry for late reply. I reverted the chang= e to > simplify_permutaton which resolved the ppc64le case ICE. > > The attached patch extends fold_vec_perm to handle vectors with > differing lengths. > For, > lhs =3D vec_perm_expr, > the patch: > (a) asserts lhs and mask have same vector length. > (b) asserts arg0 and arg1 have same vector length. > (c) returns NULL_TREE if element type differs for lhs, arg0 and arg1. > (d) if len(lhs) > len(arg0), then the patch allows permuting arg0, arg1 > if the mask has npatterns =3D=3D len(arg0) and nelts_per_pattern =3D=3D 1= . > The intent is to permute arg0 and arg1, and then to dup elements in resul= t > to target vector length. > So for eg: > vec_perm_expr< {1, 2, 3, 4}, {5, 6, 7, 8}, {0, 4, 1, 5, ...}> > will result in vla vector {1, 5, 2, 6, ....} with {1, 5, 2, 6} > replicated thru-out. > Does it look OK ? > > With the patch, we don't ICE for either of the aarch64 tests above. > For, > svint32_t f1() > { > int32x4_t v =3D {1, 2, 3, 4}; > return svld1rq_s32 (svptrue_b8 (), &v[0]); > } > > optimized dump shows: > svint32_t f1 () > { > int32x4_t v; > > : > return { 1, 2, 3, 4, ... }; > > } > > code-gen: > f1: > .LFB3900: > .cfi_startproc > ptrue p0.b, all > adrp x0, .LC0 > add x0, x0, :lo12:.LC0 > ld1rqw z0.s, p0/z, [x0] > ret > .LC0: > .word 1 > .word 2 > .word 3 > .word 4 > > I guess for this particular case, we could use index instead. > > For, > svint32_t f2(int a, int b, int c, int d) > { > int32x4_t v =3D {a, b, c, d}; > return svld1rq_s32 (svptrue_b8 (), &v[0]); > } > > optimized dump shows: > svint32_t f2 (int a, int b, int c, int d) > { > svint32_t _6; > > [local count: 1073741824]: > _6 =3D {a_1(D), b_2(D), c_3(D), d_4(D), ... }; > return _6; > > The code-gen seems pretty bad however: > f2: > .LFB3901: > .cfi_startproc > addvl sp, sp, #-4 > .cfi_escape 0xf,0x9,0x8f,0,0x92,0x2e,0,0x8,0x20,0x1e,0x22 > ptrue p0.b, all > addvl x4, sp, #3 > mov z0.b, #0 > st1w z0.s, p0, [sp, #3, mul vl] > str w0, [x4] > addvl x0, sp, #2 > ld1w z0.s, p0/z, [sp, #3, mul vl] > st1w z0.s, p0, [sp, #2, mul vl] > str w1, [x0, 4] > addvl x0, sp, #1 > ld1w z0.s, p0/z, [sp, #2, mul vl] > st1w z0.s, p0, [sp, #1, mul vl] > str w2, [x0, 8] > ld1w z0.s, p0/z, [sp, #1, mul vl] > st1w z0.s, p0, [sp] > str w3, [sp, 12] > ld1w z0.s, p0/z, [sp] > addvl sp, sp, #4 > .cfi_def_cfa_offset 0 > ret > > I will try to address code-gen issues in follow up patches. > Bootstrapped+tested on x64_64-linux-gnu and aarch64-linux-gnu. /* If result vector has greater length than input vector, + then allow permuting two vectors as long as: + a) sel.nelts_per_pattern =3D=3D 1 + b) sel.npatterns =3D=3D len of input vector. + The intent is to permute input vectors, and + dup the elements in resulting vector to target vector length. */ + + if (maybe_gt (TYPE_VECTOR_SUBPARTS (type), + TYPE_VECTOR_SUBPARTS (TREE_TYPE (arg0)))) + { + nelts =3D sel.encoding ().npatterns (); + if (sel.encoding ().nelts_per_pattern () !=3D 1 + || (!known_eq (nelts, TYPE_VECTOR_SUBPARTS (TREE_TYPE (arg0))))) + return NULL_TREE; + } so the only case you add is non-VLA to VLA and there explicitely only the case of a period that's same as the element count in the input vectors. @@ -2602,6 +2602,9 @@ dump_generic_node (pretty_printer *pp, tree node, int spc, dump_flags_t flags, pp_space (pp); } } + if (VECTOR_TYPE_P (TREE_TYPE (node)) + && !TYPE_VECTOR_SUBPARTS (TREE_TYPE (node)).is_constant ()) + pp_string (pp, ", ... "); pp_right_brace (pp); btw, I do wonder if VLA CONSTRUCTORs are a "thing"? Are they? I had hoped that you would make tree *in_elts =3D XALLOCAVEC (tree, nelts * 2); if (!vec_cst_ctor_to_array (arg0, nelts, in_elts) || !vec_cst_ctor_to_array (arg1, nelts, in_elts + nelts)) return NULL_TREE; VLA agnostic, thus support for example permuting { 0, 2, 4, 8, ... }, { 1, 3, 5, 7 ... }, { 0, 4, 1, 5 ... } as { 0, 1, 2, 3, ... }, etc. that should be entirely doable, to somebody familiar with VLA and the APIs even more so. Richard. > Thanks, > Prathamesh > > > > > Richard. > > > > > Thanks, > > > Prathamesh > > > > > > > > > > > > I will try to address the folding for above VEC_PERM_EXPR in foll= ow-up patch. > > > > > > > > > > Thanks, > > > > > Prathamesh > > > > > > > > > > > > Thanks, > > > > > > Richard