From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from foss.arm.com (foss.arm.com [217.140.110.172]) by sourceware.org (Postfix) with ESMTP id AE8B73858D35 for ; Thu, 22 Jun 2023 18:27:48 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.2 sourceware.org AE8B73858D35 Authentication-Results: sourceware.org; dmarc=pass (p=none dis=none) header.from=arm.com Authentication-Results: sourceware.org; spf=pass smtp.mailfrom=arm.com Received: from usa-sjc-imap-foss1.foss.arm.com (unknown [10.121.207.14]) by usa-sjc-mx-foss1.foss.arm.com (Postfix) with ESMTP id 3D965C14; Thu, 22 Jun 2023 11:28:32 -0700 (PDT) Received: from localhost (e121540-lin.manchester.arm.com [10.32.110.72]) by usa-sjc-imap-foss1.foss.arm.com (Postfix) with ESMTPSA id 9D03B3F64C; Thu, 22 Jun 2023 11:27:47 -0700 (PDT) From: Richard Sandiford To: juzhe.zhong@rivai.ai Mail-Followup-To: juzhe.zhong@rivai.ai,gcc-patches@gcc.gnu.org, rguenther@suse.de, richard.sandiford@arm.com Cc: gcc-patches@gcc.gnu.org, rguenther@suse.de Subject: Re: [PATCH V5] VECT: Apply LEN_MASK_{LOAD,STORE} into vectorizer References: <20230622135348.160496-1-juzhe.zhong@rivai.ai> Date: Thu, 22 Jun 2023 19:27:46 +0100 In-Reply-To: <20230622135348.160496-1-juzhe.zhong@rivai.ai> (juzhe zhong's message of "Thu, 22 Jun 2023 21:53:48 +0800") Message-ID: User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/26.3 (gnu/linux) MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: quoted-printable X-Spam-Status: No, score=-27.5 required=5.0 tests=BAYES_00,GIT_PATCH_0,KAM_DMARC_NONE,KAM_DMARC_STATUS,KAM_LAZY_DOMAIN_SECURITY,SPF_HELO_NONE,SPF_NONE,TXREP,T_SCC_BODY_TEXT_LINE autolearn=ham autolearn_force=no version=3.4.6 X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on server2.sourceware.org List-Id: juzhe.zhong@rivai.ai writes: > From: Ju-Zhe Zhong > > gcc/ChangeLog: > > * internal-fn.cc (expand_partial_store_optab_fn): Adapt for LEN_M= ASK_STORE. > (internal_load_fn_p): Add LEN_MASK_LOAD. > (internal_store_fn_p): Add LEN_MASK_STORE. > (internal_fn_mask_index): Add LEN_MASK_{LOAD,STORE}. > (internal_fn_stored_value_index): Add LEN_MASK_STORE. > (internal_len_load_store_bias): Add LEN_MASK_{LOAD,STORE}. > * optabs-tree.cc (can_vec_mask_load_store_p): Adapt for LEN_MASK_= {LOAD,STORE}. > (get_len_load_store_mode): Ditto. > * optabs-tree.h (can_vec_mask_load_store_p): Ditto. > (get_len_load_store_mode): Ditto. > * tree-vect-stmts.cc (check_load_store_for_partial_vectors): Ditt= o. > (get_all_ones_mask): New function. > (vectorizable_store): Apply LEN_MASK_{LOAD,STORE} into vectorizer. > (vectorizable_load): Ditto. > > --- > gcc/internal-fn.cc | 36 ++++++- > gcc/optabs-tree.cc | 85 +++++++++++++--- > gcc/optabs-tree.h | 6 +- > gcc/tree-vect-stmts.cc | 220 +++++++++++++++++++++++++++++------------ > 4 files changed, 265 insertions(+), 82 deletions(-) > > diff --git a/gcc/internal-fn.cc b/gcc/internal-fn.cc > index c911ae790cb..b90bd85df2c 100644 > --- a/gcc/internal-fn.cc > +++ b/gcc/internal-fn.cc > @@ -2949,7 +2949,7 @@ expand_partial_load_optab_fn (internal_fn, gcall *s= tmt, convert_optab optab) > * OPTAB. */ >=20=20 > static void > -expand_partial_store_optab_fn (internal_fn, gcall *stmt, convert_optab o= ptab) > +expand_partial_store_optab_fn (internal_fn ifn, gcall *stmt, convert_opt= ab optab) > { > class expand_operand ops[5]; > tree type, lhs, rhs, maskt, biast; > @@ -2957,7 +2957,7 @@ expand_partial_store_optab_fn (internal_fn, gcall *= stmt, convert_optab optab) > insn_code icode; >=20=20 > maskt =3D gimple_call_arg (stmt, 2); > - rhs =3D gimple_call_arg (stmt, 3); > + rhs =3D gimple_call_arg (stmt, internal_fn_stored_value_index (ifn)); > type =3D TREE_TYPE (rhs); > lhs =3D expand_call_mem_ref (type, stmt, 0); >=20=20 > @@ -4435,6 +4435,7 @@ internal_load_fn_p (internal_fn fn) > case IFN_GATHER_LOAD: > case IFN_MASK_GATHER_LOAD: > case IFN_LEN_LOAD: > + case IFN_LEN_MASK_LOAD: > return true; >=20=20 > default: > @@ -4455,6 +4456,7 @@ internal_store_fn_p (internal_fn fn) > case IFN_SCATTER_STORE: > case IFN_MASK_SCATTER_STORE: > case IFN_LEN_STORE: > + case IFN_LEN_MASK_STORE: > return true; >=20=20 > default: > @@ -4498,6 +4500,10 @@ internal_fn_mask_index (internal_fn fn) > case IFN_MASK_SCATTER_STORE: > return 4; >=20=20 > + case IFN_LEN_MASK_LOAD: > + case IFN_LEN_MASK_STORE: > + return 3; > + > default: > return (conditional_internal_fn_code (fn) !=3D ERROR_MARK > || get_unconditional_internal_fn (fn) !=3D IFN_LAST ? 0 : -1); > @@ -4519,6 +4525,9 @@ internal_fn_stored_value_index (internal_fn fn) > case IFN_LEN_STORE: > return 3; >=20=20 > + case IFN_LEN_MASK_STORE: > + return 4; > + > default: > return -1; > } > @@ -4583,13 +4592,32 @@ internal_len_load_store_bias (internal_fn ifn, ma= chine_mode mode) > { > optab optab =3D direct_internal_fn_optab (ifn); > insn_code icode =3D direct_optab_handler (optab, mode); > + int bias_opno =3D 3; > + > + if (icode =3D=3D CODE_FOR_nothing) > + { > + machine_mode mask_mode > + =3D targetm.vectorize.get_mask_mode (mode).require (); We can't require this to succeed, since the query is speculative. I think it should instead be: machine_mode mask_mode if (!targetm.vectorize.get_mask_mode (mode).exists (&mask_mode)) return VECT_PARTIAL_BIAS_UNSUPPORTED; > + if (ifn =3D=3D IFN_LEN_LOAD) > + { > + /* Try LEN_MASK_LOAD. */ > + optab =3D direct_internal_fn_optab (IFN_LEN_MASK_LOAD); > + } > + else > + { > + /* Try LEN_MASK_STORE. */ > + optab =3D direct_internal_fn_optab (IFN_LEN_MASK_STORE); > + } > + icode =3D convert_optab_handler (optab, mode, mask_mode); > + bias_opno =3D 4; > + } >=20=20 > if (icode !=3D CODE_FOR_nothing) > { > /* For now we only support biases of 0 or -1. Try both of them. = */ > - if (insn_operand_matches (icode, 3, GEN_INT (0))) > + if (insn_operand_matches (icode, bias_opno, GEN_INT (0))) > return 0; > - if (insn_operand_matches (icode, 3, GEN_INT (-1))) > + if (insn_operand_matches (icode, bias_opno, GEN_INT (-1))) > return -1; > } >=20=20 > diff --git a/gcc/optabs-tree.cc b/gcc/optabs-tree.cc > index 77bf745ae40..ab9514fc8e0 100644 > --- a/gcc/optabs-tree.cc > +++ b/gcc/optabs-tree.cc > @@ -543,19 +543,49 @@ target_supports_op_p (tree type, enum tree_code cod= e, > && optab_handler (ot, TYPE_MODE (type)) !=3D CODE_FOR_nothing); > } >=20=20 > -/* Return true if target supports vector masked load/store for mode. */ > +/* Return true if the target has support for masked load/store. > + We can support masked load/store by either mask{load,store} > + or len_mask{load,store}. > + This helper function checks whether target supports masked > + load/store and return corresponding IFN in the last argument > + (IFN_MASK_{LOAD,STORE} or IFN_LEN_MASK_{LOAD,STORE}). */ > + > +static bool > +target_supports_mask_load_store_p (machine_mode mode, machine_mode mask_= mode, > + bool is_load, internal_fn *ifn) > +{ > + optab op =3D is_load ? maskload_optab : maskstore_optab; > + optab len_op =3D is_load ? len_maskload_optab : len_maskstore_optab; > + if (convert_optab_handler (op, mode, mask_mode) !=3D CODE_FOR_nothing) > + { > + if (ifn) > + *ifn =3D is_load ? IFN_MASK_LOAD : IFN_MASK_STORE; > + return true; > + } > + else if (convert_optab_handler (len_op, mode, mask_mode) !=3D CODE_FOR= _nothing) > + { > + *ifn =3D is_load ? IFN_LEN_MASK_LOAD : IFN_LEN_MASK_STORE; I agree with Bernhard's comment here. > + return true; > + } > + return false; > +} > + > +/* Return true if target supports vector masked load/store for mode. > + An additional output in the last argumennt which is the IFN pointer. > + We set IFN as MASK_{LOAD,STORE} or LEN_MASK_{LOAD,STORE} according > + which optab is supported in the targe. */ >=20=20 > bool > can_vec_mask_load_store_p (machine_mode mode, > machine_mode mask_mode, > - bool is_load) > + bool is_load, > + internal_fn *ifn) > { > - optab op =3D is_load ? maskload_optab : maskstore_optab; > machine_mode vmode; >=20=20 > /* If mode is vector mode, check it directly. */ > if (VECTOR_MODE_P (mode)) > - return convert_optab_handler (op, mode, mask_mode) !=3D CODE_FOR_not= hing; > + return target_supports_mask_load_store_p (mode, mask_mode, is_load, = ifn); >=20=20 > /* Otherwise, return true if there is some vector mode with > the mask load/store supported. */ > @@ -569,7 +599,7 @@ can_vec_mask_load_store_p (machine_mode mode, > vmode =3D targetm.vectorize.preferred_simd_mode (smode); > if (VECTOR_MODE_P (vmode) > && targetm.vectorize.get_mask_mode (vmode).exists (&mask_mode) > - && convert_optab_handler (op, vmode, mask_mode) !=3D CODE_FOR_noth= ing) > + && target_supports_mask_load_store_p (vmode, mask_mode, is_load, i= fn)) > return true; >=20=20 > auto_vector_modes vector_modes; > @@ -577,33 +607,66 @@ can_vec_mask_load_store_p (machine_mode mode, > for (machine_mode base_mode : vector_modes) > if (related_vector_mode (base_mode, smode).exists (&vmode) > && targetm.vectorize.get_mask_mode (vmode).exists (&mask_mode) > - && convert_optab_handler (op, vmode, mask_mode) !=3D CODE_FOR_nothing) > + && target_supports_mask_load_store_p (vmode, mask_mode, is_load, ifn)) > return true; > return false; > } >=20=20 > +/* Return true if the target has support for len load/store. > + We can support len load/store by either len_{load,store} > + or len_mask{load,store}. > + This helper function checks whether target supports len > + load/store and return corresponding IFN in the last argument > + (IFN_LEN_{LOAD,STORE} or IFN_LEN_MASK_{LOAD,STORE}). */ > + > +static bool > +target_supports_len_load_store_p (machine_mode mode, bool is_load, > + internal_fn *ifn) > +{ > + optab op =3D is_load ? len_load_optab : len_store_optab; > + optab masked_op =3D is_load ? len_maskload_optab : len_maskstore_optab; > + > + if (direct_optab_handler (op, mode)) > + { > + if (ifn) > + *ifn =3D is_load ? IFN_LEN_LOAD : IFN_LEN_STORE; > + return true; > + } > + machine_mode mask_mode; > + if (targetm.vectorize.get_mask_mode (mode).exists (&mask_mode) > + && convert_optab_handler (masked_op, mode, mask_mode) !=3D CODE_FO= R_nothing) > + { > + if (ifn) > + *ifn =3D is_load ? IFN_LEN_MASK_LOAD : IFN_LEN_MASK_STORE; > + return true; > + } > + return false; > +} > + > /* If target supports vector load/store with length for vector mode MODE, > return the corresponding vector mode, otherwise return opt_machine_mo= de (). > There are two flavors for vector load/store with length, one is to me= asure > length with bytes, the other is to measure length with lanes. > As len_{load,store} optabs point out, for the flavor with bytes, we u= se > - VnQI to wrap the other supportable same size vector modes. */ > + VnQI to wrap the other supportable same size vector modes. > + An additional output in the last argumennt which is the IFN pointer. > + We set IFN as LEN_{LOAD,STORE} or LEN_MASK_{LOAD,STORE} according > + which optab is supported in the targe. */ >=20=20 > opt_machine_mode > -get_len_load_store_mode (machine_mode mode, bool is_load) > +get_len_load_store_mode (machine_mode mode, bool is_load, internal_fn *i= fn) > { > - optab op =3D is_load ? len_load_optab : len_store_optab; > gcc_assert (VECTOR_MODE_P (mode)); >=20=20 > /* Check if length in lanes supported for this mode directly. */ > - if (direct_optab_handler (op, mode)) > + if (target_supports_len_load_store_p (mode, is_load, ifn)) > return mode; >=20=20 > /* Check if length in bytes supported for same vector size VnQI. */ > machine_mode vmode; > poly_uint64 nunits =3D GET_MODE_SIZE (mode); > if (related_vector_mode (mode, QImode, nunits).exists (&vmode) > - && direct_optab_handler (op, vmode)) > + && target_supports_len_load_store_p (vmode, is_load, ifn)) > return vmode; >=20=20 > return opt_machine_mode (); > diff --git a/gcc/optabs-tree.h b/gcc/optabs-tree.h > index a3f79b6bd43..e421fc24289 100644 > --- a/gcc/optabs-tree.h > +++ b/gcc/optabs-tree.h > @@ -47,7 +47,9 @@ bool expand_vec_cond_expr_p (tree, tree, enum tree_code= ); > void init_tree_optimization_optabs (tree); > bool target_supports_op_p (tree, enum tree_code, > enum optab_subtype =3D optab_default); > -bool can_vec_mask_load_store_p (machine_mode, machine_mode, bool); > -opt_machine_mode get_len_load_store_mode (machine_mode, bool); > +bool can_vec_mask_load_store_p (machine_mode, machine_mode, bool, > + internal_fn * =3D nullptr); > +opt_machine_mode get_len_load_store_mode (machine_mode, bool, > + internal_fn * =3D nullptr); >=20=20 > #endif > diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc > index 056a0ecb2be..d53c4e4f2e5 100644 > --- a/gcc/tree-vect-stmts.cc > +++ b/gcc/tree-vect-stmts.cc > @@ -1819,16 +1819,8 @@ check_load_store_for_partial_vectors (loop_vec_inf= o loop_vinfo, tree vectype, > poly_uint64 nunits =3D TYPE_VECTOR_SUBPARTS (vectype); > poly_uint64 vf =3D LOOP_VINFO_VECT_FACTOR (loop_vinfo); > machine_mode mask_mode; > - bool using_partial_vectors_p =3D false; > - if (targetm.vectorize.get_mask_mode (vecmode).exists (&mask_mode) > - && can_vec_mask_load_store_p (vecmode, mask_mode, is_load)) > - { > - nvectors =3D group_memory_nvectors (group_size * vf, nunits); > - vect_record_loop_mask (loop_vinfo, masks, nvectors, vectype, scala= r_mask); > - using_partial_vectors_p =3D true; > - } > - > machine_mode vmode; > + bool using_partial_vectors_p =3D false; > if (get_len_load_store_mode (vecmode, is_load).exists (&vmode)) > { > nvectors =3D group_memory_nvectors (group_size * vf, nunits); > @@ -1837,6 +1829,13 @@ check_load_store_for_partial_vectors (loop_vec_inf= o loop_vinfo, tree vectype, > vect_record_loop_len (loop_vinfo, lens, nvectors, vectype, factor); > using_partial_vectors_p =3D true; > } > + else if (targetm.vectorize.get_mask_mode (vecmode).exists (&mask_mode) > + && can_vec_mask_load_store_p (vecmode, mask_mode, is_load)) > + { > + nvectors =3D group_memory_nvectors (group_size * vf, nunits); > + vect_record_loop_mask (loop_vinfo, masks, nvectors, vectype, scala= r_mask); > + using_partial_vectors_p =3D true; > + } >=20=20 > if (!using_partial_vectors_p) > { > @@ -3175,6 +3174,17 @@ vect_get_loop_variant_data_ptr_increment ( > return bump; > } >=20=20 > +/* Get all-ones vector mask for corresponding vectype. */ > + > +static tree > +get_all_ones_mask (machine_mode vmode) > +{ > + machine_mode maskmode =3D targetm.vectorize.get_mask_mode (vmode).requ= ire (); > + poly_uint64 nunits =3D GET_MODE_NUNITS (maskmode); > + tree masktype =3D build_truth_vector_type_for_mode (nunits, maskmode); > + return constant_boolean_node (true, masktype); > +} > + I think this is going against a previous comment, sorry, but FWIW: the WHILE_ULT code uses build_minus_one_cst (type) for this. Why do we pass the mode rather than the type to get_all_ones_mask? It looks like the mode comes directly from vectype. If we did pass the vectype, I think this could be: return build_minus_one_cst (build_truth_vector_type_for (vectype)); > /* Return the amount that should be added to a vector pointer to move > to the next or previous copy of AGGR_TYPE. DR_INFO is the data refer= ence > being vectorized and MEMORY_ACCESS_TYPE describes the type of > @@ -8944,30 +8954,58 @@ vectorizable_store (vec_info *vinfo, > vec_oprnd =3D new_temp; > } >=20=20 > - /* Arguments are ready. Create the new vector stmt. */ > - if (final_mask) > - { > - tree ptr =3D build_int_cst (ref_type, align * BITS_PER_UNIT); > - gcall *call > - =3D gimple_build_call_internal (IFN_MASK_STORE, 4, > - dataref_ptr, ptr, > - final_mask, vec_oprnd); > - gimple_call_set_nothrow (call, true); > - vect_finish_stmt_generation (vinfo, stmt_info, call, gsi); > - new_stmt =3D call; > - } > - else if (loop_lens) > + /* Compute IFN when LOOP_LENS or final_mask valid. */ > + machine_mode vmode =3D TYPE_MODE (vectype); > + machine_mode new_vmode =3D vmode; > + internal_fn partial_ifn =3D IFN_LAST; > + /* Produce 'len' and 'bias' argument. */ > + tree final_len =3D NULL_TREE; > + tree bias =3D NULL_TREE; > + if (loop_lens) > { > - machine_mode vmode =3D TYPE_MODE (vectype); > opt_machine_mode new_ovmode > =3D get_len_load_store_mode (vmode, false); > - machine_mode new_vmode =3D new_ovmode.require (); > + new_vmode =3D new_ovmode.require (); > unsigned factor > =3D (new_ovmode =3D=3D vmode) ? 1 : GET_MODE_UNIT_SIZE (vmode); > - tree final_len > - =3D vect_get_loop_len (loop_vinfo, gsi, loop_lens, > - vec_num * ncopies, vectype, > - vec_num * j + i, factor); > + final_len =3D vect_get_loop_len (loop_vinfo, gsi, loop_lens, > + vec_num * ncopies, vectype, > + vec_num * j + i, factor); > + } > + else if (final_mask) > + can_vec_mask_load_store_p (vmode, TYPE_MODE (mask_vectype), > + false, &partial_ifn); I think this should test the result and do gcc_unreachable () if it's false. > + > + if (partial_ifn =3D=3D IFN_LEN_MASK_STORE) > + { > + if (!final_len) > + { > + /* Pass VF value to 'len' argument of > + * LEN_MASK_STORE if LOOP_LENS is invalid. */ > + tree iv_type =3D LOOP_VINFO_RGROUP_IV_TYPE (loop_vinfo); > + final_len > + =3D build_int_cst (iv_type, > + TYPE_VECTOR_SUBPARTS (vectype)); > + } > + if (!final_mask) > + { > + /* Pass all ones value to 'mask' argument of > + * LEN_MASK_STORE if final_mask is invalid. */ Nit: should be no leading =E2=80=9C*=E2=80=9D on this line. Same for the l= ater copy. Thanks, Richard > + final_mask =3D get_all_ones_mask (vmode); > + } > + } > + if (final_len) > + { > + signed char biasval > + =3D LOOP_VINFO_PARTIAL_LOAD_STORE_BIAS (loop_vinfo); > + > + bias =3D build_int_cst (intQI_type_node, biasval); > + } > + > + /* Arguments are ready. Create the new vector stmt. */ > + if (final_len) > + { > + gcall *call; > tree ptr =3D build_int_cst (ref_type, align * BITS_PER_UNIT); > /* Need conversion if it's wrapped with VnQI. */ > if (vmode !=3D new_vmode) > @@ -8987,14 +9025,27 @@ vectorizable_store (vec_info *vinfo, > vec_oprnd =3D var; > } >=20=20 > - signed char biasval =3D > - LOOP_VINFO_PARTIAL_LOAD_STORE_BIAS (loop_vinfo); > - > - tree bias =3D build_int_cst (intQI_type_node, biasval); > + if (partial_ifn =3D=3D IFN_LEN_MASK_STORE) > + call =3D gimple_build_call_internal (IFN_LEN_MASK_STORE, 6, > + dataref_ptr, ptr, > + final_len, final_mask, > + vec_oprnd, bias); > + else > + call > + =3D gimple_build_call_internal (IFN_LEN_STORE, 5, > + dataref_ptr, ptr, final_len, > + vec_oprnd, bias); > + gimple_call_set_nothrow (call, true); > + vect_finish_stmt_generation (vinfo, stmt_info, call, gsi); > + new_stmt =3D call; > + } > + else if (final_mask) > + { > + tree ptr =3D build_int_cst (ref_type, align * BITS_PER_UNIT); > gcall *call > - =3D gimple_build_call_internal (IFN_LEN_STORE, 5, dataref_ptr, > - ptr, final_len, vec_oprnd, > - bias); > + =3D gimple_build_call_internal (IFN_MASK_STORE, 4, > + dataref_ptr, ptr, > + final_mask, vec_oprnd); > gimple_call_set_nothrow (call, true); > vect_finish_stmt_generation (vinfo, stmt_info, call, gsi); > new_stmt =3D call; > @@ -10304,45 +10355,72 @@ vectorizable_load (vec_info *vinfo, > align, misalign); > align =3D least_bit_hwi (misalign | align); >=20=20 > - if (final_mask) > - { > - tree ptr =3D build_int_cst (ref_type, > - align * BITS_PER_UNIT); > - gcall *call > - =3D gimple_build_call_internal (IFN_MASK_LOAD, 3, > - dataref_ptr, ptr, > - final_mask); > - gimple_call_set_nothrow (call, true); > - new_stmt =3D call; > - data_ref =3D NULL_TREE; > - } > - else if (loop_lens && memory_access_type !=3D VMAT_INVARIANT) > + /* Compute IFN when LOOP_LENS or final_mask valid. */ > + machine_mode vmode =3D TYPE_MODE (vectype); > + machine_mode new_vmode =3D vmode; > + internal_fn partial_ifn =3D IFN_LAST; > + /* Produce 'len' and 'bias' argument. */ > + tree final_len =3D NULL_TREE; > + tree bias =3D NULL_TREE; > + if (loop_lens) > { > - machine_mode vmode =3D TYPE_MODE (vectype); > opt_machine_mode new_ovmode > - =3D get_len_load_store_mode (vmode, true); > - machine_mode new_vmode =3D new_ovmode.require (); > + =3D get_len_load_store_mode (vmode, false); > + new_vmode =3D new_ovmode.require (); > unsigned factor =3D (new_ovmode =3D=3D vmode) > ? 1 > : GET_MODE_UNIT_SIZE (vmode); > - tree final_len > + final_len > =3D vect_get_loop_len (loop_vinfo, gsi, loop_lens, > vec_num * ncopies, vectype, > vec_num * j + i, factor); > - tree ptr > - =3D build_int_cst (ref_type, align * BITS_PER_UNIT); > - > - tree qi_type =3D unsigned_intQI_type_node; > + } > + else if (final_mask) > + can_vec_mask_load_store_p (vmode, > + TYPE_MODE (mask_vectype), > + false, &partial_ifn); >=20=20 > - signed char biasval =3D > - LOOP_VINFO_PARTIAL_LOAD_STORE_BIAS (loop_vinfo); > + if (partial_ifn =3D=3D IFN_LEN_MASK_LOAD) > + { > + if (!final_len) > + { > + /* Pass VF value to 'len' argument of > + * LEN_MASK_LOAD if LOOP_LENS is invalid. */ > + tree iv_type > + =3D LOOP_VINFO_RGROUP_IV_TYPE (loop_vinfo); > + final_len > + =3D build_int_cst (iv_type, > + TYPE_VECTOR_SUBPARTS (vectype)); > + } > + if (!final_mask) > + { > + /* Pass all ones value to 'mask' argument of > + * LEN_MASK_LOAD if final_mask is invalid. */ > + final_mask =3D get_all_ones_mask (vmode); > + } > + } > + if (final_len) > + { > + signed char biasval > + =3D LOOP_VINFO_PARTIAL_LOAD_STORE_BIAS (loop_vinfo); >=20=20 > - tree bias =3D build_int_cst (intQI_type_node, biasval); > + bias =3D build_int_cst (intQI_type_node, biasval); > + } >=20=20 > - gcall *call > - =3D gimple_build_call_internal (IFN_LEN_LOAD, 4, > - dataref_ptr, ptr, > - final_len, bias); > + if (final_len && memory_access_type !=3D VMAT_INVARIANT) > + { > + tree ptr > + =3D build_int_cst (ref_type, align * BITS_PER_UNIT); > + gcall *call; > + if (partial_ifn =3D=3D IFN_LEN_MASK_LOAD) > + call =3D gimple_build_call_internal (IFN_LEN_MASK_LOAD, > + 5, dataref_ptr, > + ptr, final_len, > + final_mask, bias); > + else > + call =3D gimple_build_call_internal (IFN_LEN_LOAD, 4, > + dataref_ptr, ptr, > + final_len, bias); > gimple_call_set_nothrow (call, true); > new_stmt =3D call; > data_ref =3D NULL_TREE; > @@ -10350,8 +10428,8 @@ vectorizable_load (vec_info *vinfo, > /* Need conversion if it's wrapped with VnQI. */ > if (vmode !=3D new_vmode) > { > - tree new_vtype > - =3D build_vector_type_for_mode (qi_type, new_vmode); > + tree new_vtype =3D build_vector_type_for_mode ( > + unsigned_intQI_type_node, new_vmode); > tree var =3D vect_get_new_ssa_name (new_vtype, > vect_simple_var); > gimple_set_lhs (call, var); > @@ -10363,6 +10441,18 @@ vectorizable_load (vec_info *vinfo, > VIEW_CONVERT_EXPR, op); > } > } > + else if (final_mask) > + { > + tree ptr =3D build_int_cst (ref_type, > + align * BITS_PER_UNIT); > + gcall *call > + =3D gimple_build_call_internal (IFN_MASK_LOAD, 3, > + dataref_ptr, ptr, > + final_mask); > + gimple_call_set_nothrow (call, true); > + new_stmt =3D call; > + data_ref =3D NULL_TREE; > + } > else > { > tree ltype =3D vectype;