gcc/ChangeLog

   1 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
   2
   3         * tree-vect-stmts.c (vect_get_store_rhs): New function.
   4         (vectorizable_mask_load_store): Delete.
   5         (vectorizable_call): Return false for masked loads and stores.
   6         (vectorizable_store): Handle IFN_MASK_STORE.  Use vect_get_store_rhs
   7         instead of gimple_assign_rhs1.
   8         (vectorizable_load): Handle IFN_MASK_LOAD.
   9         (vect_transform_stmt): Don't set is_store for call_vec_info_type.
  10
  11 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
  12
  13         * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
  14         split out from..,
  15         (vectorizable_mask_load_store): ...here.
  16         (vectorizable_load): ...and here.
  17
  18 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
  19
  20         * tree-vect-stmts.c (vect_build_all_ones_mask)
  21         (vect_build_zero_merge_argument): New functions, split out from...
  22         (vectorizable_load): ...here.
  23
  24 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
  25
  26         * tree-vect-stmts.c (vect_check_store_rhs): New function,
  27         split out from...
  28         (vectorizable_mask_load_store): ...here.
  29         (vectorizable_store): ...and here.
  30
  31 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
  32
  33         * tree-vect-stmts.c (vect_check_load_store_mask): New function,
  34         split out from...
  35         (vectorizable_mask_load_store): ...here.
  36
  37 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
  38
  39         * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
  40         (vect_model_store_cost): Take a vec_load_store_type instead of a
  41         vect_def_type.
  42         * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
  43         (vect_model_store_cost): Take a vec_load_store_type instead of a
  44         vect_def_type.
  45         (vectorizable_mask_load_store): Update accordingly.
  46         (vectorizable_store): Likewise.
  47         * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
  48
  49 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
  50
  51         * tree-vect-loop.c (vect_transform_loop): Stub out scalar
  52         IFN_MASK_LOAD calls here rather than...
  53         * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
  54
  55 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
  56             Alan Hayward  <alan.hayward@arm.com>
  57             David Sherwood  <david.sherwood@arm.com>
  58
  59         * expmed.c (extract_bit_field_1): For vector extracts,
  60         fall back to extract_bit_field_as_subreg if vec_extract
  61         isn't available.
  62
  63 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
  64             Alan Hayward  <alan.hayward@arm.com>
  65             David Sherwood  <david.sherwood@arm.com>
  66
  67         * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
  68         they are variable or constant sized.
  69         (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
  70         slots for constant-sized data.
  71
  72 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
  73             Alan Hayward  <alan.hayward@arm.com>
  74             David Sherwood  <david.sherwood@arm.com>
  75
  76         * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
  77         handling COND_EXPRs with boolean comparisons, try to find a better
  78         basis for the mask type than the boolean itself.
  79
  80 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
  81
  82         * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
  83         is calculated and how it can be overridden.
  84         * genmodes.c (max_bitsize_mode_any_mode): New variable.
  85         (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
  86         if defined.
  87         (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
  88         if nonzero.
  89
  90 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
  91             Alan Hayward  <alan.hayward@arm.com>
  92             David Sherwood  <david.sherwood@arm.com>
  93
  94         * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
  95         Remove the mode argument.
  96         (aarch64_simd_valid_immediate): Remove the mode and inverse
  97         arguments.
  98         * config/aarch64/iterators.md (bitsize): New iterator.
  99         * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
 100         (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
 101         * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
 102         aarch64_simd_valid_immediate.
 103         * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
 104         (aarch64_reg_or_bic_imm): Likewise.
 105         * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
 106         with an insn_type enum and msl with a modifier_type enum.
 107         Replace element_width with a scalar_mode.  Change the shift
 108         to unsigned int.  Add constructors for scalar_float_mode and
 109         scalar_int_mode elements.
 110         (aarch64_vect_float_const_representable_p): Delete.
 111         (aarch64_can_const_movi_rtx_p)
 112         (aarch64_simd_scalar_immediate_valid_for_move)
 113         (aarch64_simd_make_constant): Update call to
 114         aarch64_simd_valid_immediate.
 115         (aarch64_advsimd_valid_immediate_hs): New function.
 116         (aarch64_advsimd_valid_immediate): Likewise.
 117         (aarch64_simd_valid_immediate): Remove mode and inverse
 118         arguments.  Rewrite to use the above.  Use const_vec_duplicate_p
 119         to detect duplicated constants and use aarch64_float_const_zero_rtx_p
 120         and aarch64_float_const_representable_p on the result.
 121         (aarch64_output_simd_mov_immediate): Remove mode argument.
 122         Update call to aarch64_simd_valid_immediate and use of
 123         simd_immediate_info.
 124         (aarch64_output_scalar_simd_mov_immediate): Update call
 125         accordingly.
 126
 127 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 128             Alan Hayward  <alan.hayward@arm.com>
 129             David Sherwood  <david.sherwood@arm.com>
 130
 131         * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
 132         (mode_nunits): Likewise CONST_MODE_NUNITS.
 133         * machmode.def (ADJUST_NUNITS): Document.
 134         * genmodes.c (mode_data::need_nunits_adj): New field.
 135         (blank_mode): Update accordingly.
 136         (adj_nunits): New variable.
 137         (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
 138         parameter.
 139         (emit_mode_size_inline): Set need_bytesize_adj for all modes
 140         listed in adj_nunits.
 141         (emit_mode_nunits_inline): Set need_nunits_adj for all modes
 142         listed in adj_nunits.  Don't emit case statements for such modes.
 143         (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
 144         and CONST_MODE_PRECISION.  Make CONST_MODE_SIZE expand to
 145         nothing if adj_nunits is nonnull.
 146         (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
 147         (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
 148         (emit_mode_fbit): Update use of print_maybe_const_decl.
 149         (emit_move_size): Likewise.  Treat the array as non-const
 150         if adj_nunits.
 151         (emit_mode_adjustments): Handle adj_nunits.
 152
 153 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 154
 155         * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
 156         * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
 157         (VECTOR_MODES): Use it.
 158         (make_vector_modes): Take the prefix as an argument.
 159
 160 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 161             Alan Hayward  <alan.hayward@arm.com>
 162             David Sherwood  <david.sherwood@arm.com>
 163
 164         * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
 165         * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
 166         for MODE_VECTOR_BOOL.
 167         * machmode.def (VECTOR_BOOL_MODE): Document.
 168         * genmodes.c (VECTOR_BOOL_MODE): New macro.
 169         (make_vector_bool_mode): New function.
 170         (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
 171         MODE_VECTOR_BOOL.
 172         * lto-streamer-in.c (lto_input_mode_table): Likewise.
 173         * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
 174         Likewise.
 175         * stor-layout.c (int_mode_for_mode): Likewise.
 176         * tree.c (build_vector_type_for_mode): Likewise.
 177         * varasm.c (output_constant_pool_2): Likewise.
 178         * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
 179         CONSTM1_RTX (BImode) are the same thing.  Initialize const_tiny_rtx
 180         for MODE_VECTOR_BOOL.
 181         * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
 182         of mode class checks.
 183         * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
 184         instead of a list of mode class checks.
 185         (expand_vector_scalar_condition): Likewise.
 186         (type_for_widest_vector_mode): Handle BImode as an inner mode.
 187
 188 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 189             Alan Hayward  <alan.hayward@arm.com>
 190             David Sherwood  <david.sherwood@arm.com>
 191
 192         * machmode.h (mode_size): Change from unsigned short to
 193         poly_uint16_pod.
 194         (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
 195         (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
 196         or if measurement_type is not polynomial.
 197         (fixed_size_mode::includes_p): Check for constant-sized modes.
 198         * genmodes.c (emit_mode_size_inline): Make mode_size_inline
 199         return a poly_uint16 rather than an unsigned short.
 200         (emit_mode_size): Change the type of mode_size from unsigned short
 201         to poly_uint16_pod.  Use ZERO_COEFFS for the initializer.
 202         (emit_mode_adjustments): Cope with polynomial vector sizes.
 203         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
 204         for GET_MODE_SIZE.
 205         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
 206         for GET_MODE_SIZE.
 207         * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
 208         * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
 209         * caller-save.c (setup_save_areas): Likewise.
 210         (replace_reg_with_saved_mem): Likewise.
 211         * calls.c (emit_library_call_value_1): Likewise.
 212         * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
 213         * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
 214         (gen_lowpart_for_combine): Likewise.
 215         * convert.c (convert_to_integer_1): Likewise.
 216         * cse.c (equiv_constant, cse_insn): Likewise.
 217         * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
 218         (cselib_subst_to_values): Likewise.
 219         * dce.c (word_dce_process_block): Likewise.
 220         * df-problems.c (df_word_lr_mark_ref): Likewise.
 221         * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
 222         * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
 223         (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
 224         (rtl_for_decl_location): Likewise.
 225         * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
 226         * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
 227         * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
 228         (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
 229         (expand_expr_real_1): Likewise.
 230         * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
 231         (pad_below): Likewise.
 232         * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
 233         * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
 234         * ira.c (get_subreg_tracking_sizes): Likewise.
 235         * ira-build.c (ira_create_allocno_objects): Likewise.
 236         * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
 237         (ira_sort_regnos_for_alter_reg): Likewise.
 238         * ira-costs.c (record_operand_costs): Likewise.
 239         * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
 240         (resolve_simple_move): Likewise.
 241         * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
 242         (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
 243         (lra_constraints): Likewise.
 244         (CONST_POOL_OK_P): Reject variable-sized modes.
 245         * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
 246         (add_pseudo_to_slot, lra_spill): Likewise.
 247         * omp-low.c (omp_clause_aligned_alignment): Likewise.
 248         * optabs-query.c (get_best_extraction_insn): Likewise.
 249         * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
 250         * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
 251         (expand_mult_highpart, valid_multiword_target_p): Likewise.
 252         * recog.c (offsettable_address_addr_space_p): Likewise.
 253         * regcprop.c (maybe_mode_change): Likewise.
 254         * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
 255         * regrename.c (build_def_use): Likewise.
 256         * regstat.c (dump_reg_info): Likewise.
 257         * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
 258         (find_reloads, find_reloads_subreg_address): Likewise.
 259         * reload1.c (eliminate_regs_1): Likewise.
 260         * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
 261         * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
 262         (simplify_binary_operation_1, simplify_subreg): Likewise.
 263         * targhooks.c (default_function_arg_padding): Likewise.
 264         (default_hard_regno_nregs, default_class_max_nregs): Likewise.
 265         * tree-cfg.c (verify_gimple_assign_binary): Likewise.
 266         (verify_gimple_assign_ternary): Likewise.
 267         * tree-inline.c (estimate_move_cost): Likewise.
 268         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
 269         * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
 270         (get_address_cost_ainc): Likewise.
 271         * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
 272         (vect_supportable_dr_alignment): Likewise.
 273         * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
 274         (vectorizable_reduction): Likewise.
 275         * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
 276         (vectorizable_operation, vectorizable_load): Likewise.
 277         * tree.c (build_same_sized_truth_vector_type): Likewise.
 278         * valtrack.c (cleanup_auto_inc_dec): Likewise.
 279         * var-tracking.c (emit_note_insn_var_location): Likewise.
 280         * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
 281         (ADDR_VEC_ALIGN): Likewise.
 282
 283 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 284             Alan Hayward  <alan.hayward@arm.com>
 285             David Sherwood  <david.sherwood@arm.com>
 286
 287         * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
 288         unsigned short.
 289         (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
 290         or if measurement_type is polynomial.
 291         * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
 292         * combine.c (make_extraction): Likewise.
 293         * dse.c (find_shift_sequence): Likewise.
 294         * dwarf2out.c (mem_loc_descriptor): Likewise.
 295         * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
 296         (extract_bit_field, extract_low_bits): Likewise.
 297         * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
 298         (optimize_bitfield_assignment_op, expand_assignment): Likewise.
 299         (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
 300         * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
 301         * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
 302         * reload.c (find_reloads): Likewise.
 303         * reload1.c (alter_reg): Likewise.
 304         * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
 305         * targhooks.c (default_secondary_memory_needed_mode): Likewise.
 306         * tree-if-conv.c (predicate_mem_writes): Likewise.
 307         * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
 308         * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
 309         * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
 310         * valtrack.c (dead_debug_insert_temp): Likewise.
 311         * varasm.c (mergeable_constant_section): Likewise.
 312         * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
 313
 314 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 315             Alan Hayward  <alan.hayward@arm.com>
 316             David Sherwood  <david.sherwood@arm.com>
 317
 318         * expr.c (expand_assignment): Cope with polynomial mode sizes
 319         when assigning to a CONCAT.
 320
 321 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 322             Alan Hayward  <alan.hayward@arm.com>
 323             David Sherwood  <david.sherwood@arm.com>
 324
 325         * machmode.h (mode_precision): Change from unsigned short to
 326         poly_uint16_pod.
 327         (mode_to_precision): Return a poly_uint16 rather than an unsigned
 328         short.
 329         (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
 330         or if measurement_type is not polynomial.
 331         (HWI_COMPUTABLE_MODE_P): Turn into a function.  Optimize the case
 332         in which the mode is already known to be a scalar_int_mode.
 333         * genmodes.c (emit_mode_precision): Change the type of mode_precision
 334         from unsigned short to poly_uint16_pod.  Use ZERO_COEFFS for the
 335         initializer.
 336         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
 337         for GET_MODE_PRECISION.
 338         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
 339         for GET_MODE_PRECISION.
 340         * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
 341         as polynomial.
 342         (try_combine, find_split_point, combine_simplify_rtx): Likewise.
 343         (expand_field_assignment, make_extraction): Likewise.
 344         (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
 345         (get_last_value): Likewise.
 346         * convert.c (convert_to_integer_1): Likewise.
 347         * cse.c (cse_insn): Likewise.
 348         * expr.c (expand_expr_real_1): Likewise.
 349         * lra-constraints.c (simplify_operand_subreg): Likewise.
 350         * optabs-query.c (can_atomic_load_p): Likewise.
 351         * optabs.c (expand_atomic_load): Likewise.
 352         (expand_atomic_store): Likewise.
 353         * ree.c (combine_reaching_defs): Likewise.
 354         * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
 355         * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
 356         * tree.h (type_has_mode_precision_p): Likewise.
 357         * ubsan.c (instrument_si_overflow): Likewise.
 358
 359 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 360             Alan Hayward  <alan.hayward@arm.com>
 361             David Sherwood  <david.sherwood@arm.com>
 362
 363         * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
 364         polynomial numbers of units.
 365         (SET_TYPE_VECTOR_SUBPARTS): Likewise.
 366         (valid_vector_subparts_p): New function.
 367         (build_vector_type): Remove temporary shim and take the number
 368         of units as a poly_uint64 rather than an int.
 369         (build_opaque_vector_type): Take the number of units as a
 370         poly_uint64 rather than an int.
 371         * tree.c (build_vector_from_ctor): Handle polynomial
 372         TYPE_VECTOR_SUBPARTS.
 373         (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
 374         (uniform_vector_p, vector_type_mode, build_vector): Likewise.
 375         (build_vector_from_val): If the number of units is variable,
 376         use build_vec_duplicate_cst for constant operands and
 377         VEC_DUPLICATE_EXPR otherwise.
 378         (make_vector_type): Remove temporary is_constant ().
 379         (build_vector_type, build_opaque_vector_type): Take the number of
 380         units as a poly_uint64 rather than an int.
 381         (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
 382         VECTOR_CST_NELTS.
 383         * cfgexpand.c (expand_debug_expr): Likewise.
 384         * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
 385         (store_constructor, expand_expr_real_1): Likewise.
 386         (const_scalar_mask_from_tree): Likewise.
 387         * fold-const-call.c (fold_const_reduction): Likewise.
 388         * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
 389         (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
 390         (native_encode_vector, vec_cst_ctor_to_array): Likewise.
 391         (fold_relational_const): Likewise.
 392         (native_interpret_vector): Likewise.  Change the size from an
 393         int to an unsigned int.
 394         * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
 395         TYPE_VECTOR_SUBPARTS.
 396         (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
 397         (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
 398         duplicating a non-constant operand into a variable-length vector.
 399         * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
 400         TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
 401         * ipa-icf.c (sem_variable::equals): Likewise.
 402         * match.pd: Likewise.
 403         * omp-simd-clone.c (simd_clone_subparts): Likewise.
 404         * print-tree.c (print_node): Likewise.
 405         * stor-layout.c (layout_type): Likewise.
 406         * targhooks.c (default_builtin_vectorization_cost): Likewise.
 407         * tree-cfg.c (verify_gimple_comparison): Likewise.
 408         (verify_gimple_assign_binary): Likewise.
 409         (verify_gimple_assign_ternary): Likewise.
 410         (verify_gimple_assign_single): Likewise.
 411         * tree-pretty-print.c (dump_generic_node): Likewise.
 412         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
 413         (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
 414         * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
 415         (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
 416         (vect_shift_permute_load_chain): Likewise.
 417         * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
 418         (expand_vector_condition, optimize_vector_constructor): Likewise.
 419         (lower_vec_perm, get_compute_type): Likewise.
 420         * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
 421         (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
 422         * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
 423         (vect_recog_mask_conversion_pattern): Likewise.
 424         * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
 425         (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
 426         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
 427         (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
 428         (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
 429         (vectorizable_shift, vectorizable_operation, vectorizable_store)
 430         (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
 431         (supportable_widening_operation): Likewise.
 432         (supportable_narrowing_operation): Likewise.
 433         * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
 434         Likewise.
 435         * varasm.c (output_constant): Likewise.
 436
 437 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 438             Alan Hayward  <alan.hayward@arm.com>
 439             David Sherwood  <david.sherwood@arm.com>
 440
 441         * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
 442         so that both the length == 3 and length != 3 cases set up their
 443         own permute vectors.  Add comments explaining why we know the
 444         number of elements is constant.
 445         (vect_permute_load_chain): Likewise.
 446
 447 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 448             Alan Hayward  <alan.hayward@arm.com>
 449             David Sherwood  <david.sherwood@arm.com>
 450
 451         * machmode.h (mode_nunits): Change from unsigned char to
 452         poly_uint16_pod.
 453         (ONLY_FIXED_SIZE_MODES): New macro.
 454         (pod_mode::measurement_type, scalar_int_mode::measurement_type)
 455         (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
 456         (complex_mode::measurement_type, fixed_size_mode::measurement_type):
 457         New typedefs.
 458         (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
 459         (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
 460         or if measurement_type is not polynomial.
 461         * genmodes.c (ZERO_COEFFS): New macro.
 462         (emit_mode_nunits_inline): Make mode_nunits_inline return a
 463         poly_uint16.
 464         (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
 465         Use ZERO_COEFFS when emitting initializers.
 466         * data-streamer.h (bp_pack_poly_value): New function.
 467         (bp_unpack_poly_value): Likewise.
 468         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
 469         for GET_MODE_NUNITS.
 470         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
 471         for GET_MODE_NUNITS.
 472         * tree.c (make_vector_type): Remove temporary shim and make
 473         the real function take the number of units as a poly_uint64
 474         rather than an int.
 475         (build_vector_type_for_mode): Handle polynomial nunits.
 476         * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
 477         * emit-rtl.c (const_vec_series_p_1): Likewise.
 478         (gen_rtx_CONST_VECTOR): Likewise.
 479         * fold-const.c (test_vec_duplicate_folding): Likewise.
 480         * genrecog.c (validate_pattern): Likewise.
 481         * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
 482         * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
 483         * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
 484         (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
 485         (expand_vec_cond_expr, expand_mult_highpart): Likewise.
 486         * rtlanal.c (subreg_get_info): Likewise.
 487         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
 488         (vect_grouped_load_supported): Likewise.
 489         * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
 490         * tree-vect-loop.c (have_whole_vector_shift): Likewise.
 491         * simplify-rtx.c (simplify_unary_operation_1): Likewise.
 492         (simplify_const_unary_operation, simplify_binary_operation_1)
 493         (simplify_const_binary_operation, simplify_ternary_operation)
 494         (test_vector_ops_duplicate, test_vector_ops): Likewise.
 495         (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
 496         instead of CONST_VECTOR_NUNITS.
 497         * varasm.c (output_constant_pool_2): Likewise.
 498         * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
 499         explicit-encoded elements in the XVEC for variable-length vectors.
 500
 501 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 502
 503         * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
 504
 505 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 506             Alan Hayward  <alan.hayward@arm.com>
 507             David Sherwood  <david.sherwood@arm.com>
 508
 509         * coretypes.h (fixed_size_mode): Declare.
 510         (fixed_size_mode_pod): New typedef.
 511         * builtins.h (target_builtins::x_apply_args_mode)
 512         (target_builtins::x_apply_result_mode): Change type to
 513         fixed_size_mode_pod.
 514         * builtins.c (apply_args_size, apply_result_size, result_vector)
 515         (expand_builtin_apply_args_1, expand_builtin_apply)
 516         (expand_builtin_return): Update accordingly.
 517
 518 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 519
 520         * cse.c (hash_rtx_cb): Hash only the encoded elements.
 521         * cselib.c (cselib_hash_rtx): Likewise.
 522         * expmed.c (make_tree): Build VECTOR_CSTs directly from the
 523         CONST_VECTOR encoding.
 524
 525 2017-01-03  Jakub Jelinek  <jakub@redhat.com>
 526             Jeff Law  <law@redhat.com>
 527
 528         PR target/83641
 529         * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
 530         noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
 531         only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
 532         and add REG_CFA_ADJUST_CFA notes in that case to both insns.
 533
 534         PR target/83641
 535         * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
 536         explicitly probe *sp in a noreturn function if there were any callee
 537         register saves or frame pointer is needed.
 538
 539 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
 540
 541         PR debug/83621
 542         * cfgexpand.c (expand_debug_expr): Return NULL if mode is
 543         BLKmode for ternary, binary or unary expressions.
 544
 545         PR debug/83645
 546         * var-tracking.c (delete_vta_debug_insn): New inline function.
 547         (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
 548         insns from get_insns () to NULL instead of each bb separately.
 549         Use delete_vta_debug_insn.  No longer static.
 550         (vt_debug_insns_local, variable_tracking_main_1): Adjust
 551         delete_vta_debug_insns callers.
 552         * rtl.h (delete_vta_debug_insns): Declare.
 553         * final.c (rest_of_handle_final): Call delete_vta_debug_insns
 554         instead of variable_tracking_main.
 555
 556 2018-01-03  Martin Sebor  <msebor@redhat.com>
 557
 558         PR c/83559
 559         * doc/extend.texi (attribute const): Fix a typo.
 560         * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
 561         issuing -Wsuggest-attribute for void functions.
 562
 563 2018-01-03  Martin Sebor  <msebor@redhat.com>
 564
 565         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
 566         offset_int::from instead of wide_int::to_shwi.
 567         (maybe_diag_overlap): Remove assertion.
 568         Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
 569         * gimple-ssa-sprintf.c (format_directive): Same.
 570         (parse_directive): Same.
 571         (sprintf_dom_walker::compute_format_length): Same.
 572         (try_substitute_return_value): Same.
 573
 574 2017-01-03  Jeff Law  <law@redhat.com>
 575
 576         PR middle-end/83654
 577         * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
 578         non-constant residual for zero at runtime and avoid probing in
 579         that case.  Reorganize code for trailing problem to mirror handling
 580         of the residual.
 581
 582 2018-01-03  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
 583
 584         PR tree-optimization/83501
 585         * tree-ssa-strlen.c (get_string_cst): New.
 586         (handle_char_store): Call get_string_cst.
 587
 588 2018-01-03  Martin Liska  <mliska@suse.cz>
 589
 590         PR tree-optimization/83593
 591         * tree-ssa-strlen.c: Include tree-cfg.h.
 592         (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
 593         (strlen_dom_walker): Add new member variable m_cleanup_cfg.
 594         (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
 595         to false.
 596         (strlen_dom_walker::before_dom_children): Call
 597         gimple_purge_dead_eh_edges. Dump tranformation with details
 598         dump flags.
 599         (strlen_dom_walker::before_dom_children): Update call by adding
 600         new argument cleanup_eh.
 601         (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
 602
 603 2018-01-03  Martin Liska  <mliska@suse.cz>
 604
 605         PR ipa/83549
 606         * cif-code.def (VARIADIC_THUNK): New enum value.
 607         * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
 608         thunks.
 609
 610 2018-01-03  Jan Beulich  <jbeulich@suse.com>
 611
 612         * sse.md (mov<mode>_internal): Tighten condition for when to use
 613         vmovdqu<ssescalarsize> for TI and OI modes.
 614
 615 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
 616
 617         Update copyright years.
 618
 619 2018-01-03  Martin Liska  <mliska@suse.cz>
 620
 621         PR ipa/83594
 622         * ipa-visibility.c (function_and_variable_visibility): Skip
 623         functions with noipa attribure.
 624
 625 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
 626
 627         * gcc.c (process_command): Update copyright notice dates.
 628         * gcov-dump.c (print_version): Ditto.
 629         * gcov.c (print_version): Ditto.
 630         * gcov-tool.c (print_version): Ditto.
 631         * gengtype.c (create_file): Ditto.
 632         * doc/cpp.texi: Bump @copying's copyright year.
 633         * doc/cppinternals.texi: Ditto.
 634         * doc/gcc.texi: Ditto.
 635         * doc/gccint.texi: Ditto.
 636         * doc/gcov.texi: Ditto.
 637         * doc/install.texi: Ditto.
 638         * doc/invoke.texi: Ditto.
 639
 640 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 641
 642         * vector-builder.h (vector_builder::m_full_nelts): Change from
 643         unsigned int to poly_uint64.
 644         (vector_builder::full_nelts): Update prototype accordingly.
 645         (vector_builder::new_vector): Likewise.
 646         (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
 647         (vector_builder::operator ==): Likewise.
 648         (vector_builder::finalize): Likewise.
 649         * int-vector-builder.h (int_vector_builder::int_vector_builder):
 650         Take the number of elements as a poly_uint64 rather than an
 651         unsigned int.
 652         * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
 653         from unsigned int to poly_uint64.
 654         (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
 655         (vec_perm_indices::new_vector): Likewise.
 656         (vec_perm_indices::length): Likewise.
 657         (vec_perm_indices::nelts_per_input): Likewise.
 658         (vec_perm_indices::input_nelts): Likewise.
 659         * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
 660         number of elements per input as a poly_uint64 rather than an
 661         unsigned int.  Use the original encoding for variable-length
 662         vectors, rather than clamping each individual element.
 663         For the second and subsequent elements in each pattern,
 664         clamp the step and base before clamping their sum.
 665         (vec_perm_indices::series_p): Handle polynomial element counts.
 666         (vec_perm_indices::all_in_range_p): Likewise.
 667         (vec_perm_indices_to_tree): Likewise.
 668         (vec_perm_indices_to_rtx): Likewise.
 669         * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
 670         * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
 671         (tree_vector_builder::new_binary_operation): Handle polynomial
 672         element counts.  Return false if we need to know the number
 673         of elements at compile time.
 674         * fold-const.c (fold_vec_perm): Punt if the number of elements
 675         isn't known at compile time.
 676
 677 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 678
 679         * vec-perm-indices.h (vec_perm_builder): Change element type
 680         from HOST_WIDE_INT to poly_int64.
 681         (vec_perm_indices::element_type): Update accordingly.
 682         (vec_perm_indices::clamp): Handle polynomial element_types.
 683         * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
 684         (vec_perm_indices::all_in_range_p): Likewise.
 685         (tree_to_vec_perm_builder): Check for poly_int64 trees rather
 686         than shwi trees.
 687         * vector-builder.h (vector_builder::stepped_sequence_p): Handle
 688         polynomial vec_perm_indices element types.
 689         * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
 690         * fold-const.c (fold_vec_perm): Likewise.
 691         * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
 692         * tree-vect-generic.c (lower_vec_perm): Likewise.
 693         * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
 694         * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
 695         element type to HOST_WIDE_INT.
 696
 697 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 698             Alan Hayward  <alan.hayward@arm.com>
 699             David Sherwood  <david.sherwood@arm.com>
 700
 701         * alias.c (addr_side_effect_eval): Take the size as a poly_int64
 702         rather than an int.  Use plus_constant.
 703         (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
 704         Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
 705
 706 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 707             Alan Hayward  <alan.hayward@arm.com>
 708             David Sherwood  <david.sherwood@arm.com>
 709
 710         * calls.c (emit_call_1, expand_call): Change struct_value_size from
 711         a HOST_WIDE_INT to a poly_int64.
 712
 713 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 714             Alan Hayward  <alan.hayward@arm.com>
 715             David Sherwood  <david.sherwood@arm.com>
 716
 717         * calls.c (load_register_parameters): Cope with polynomial
 718         mode sizes.  Require a constant size for BLKmode parameters
 719         that aren't described by a PARALLEL.  If BLOCK_REG_PADDING
 720         forces a parameter to be padded at the lsb end in order to
 721         fill a complete number of words, require the parameter size
 722         to be ordered wrt UNITS_PER_WORD.
 723
 724 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 725             Alan Hayward  <alan.hayward@arm.com>
 726             David Sherwood  <david.sherwood@arm.com>
 727
 728         * reload1.c (spill_stack_slot_width): Change element type
 729         from unsigned int to poly_uint64_pod.
 730         (alter_reg): Treat mode sizes as polynomial.
 731
 732 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 733             Alan Hayward  <alan.hayward@arm.com>
 734             David Sherwood  <david.sherwood@arm.com>
 735
 736         * reload.c (complex_word_subreg_p): New function.
 737         (reload_inner_reg_of_subreg, push_reload): Use it.
 738
 739 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 740             Alan Hayward  <alan.hayward@arm.com>
 741             David Sherwood  <david.sherwood@arm.com>
 742
 743         * lra-constraints.c (process_alt_operands): Reject matched
 744         operands whose sizes aren't ordered.
 745         (match_reload): Refer to this check here.
 746
 747 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 748             Alan Hayward  <alan.hayward@arm.com>
 749             David Sherwood  <david.sherwood@arm.com>
 750
 751         * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
 752         that the mode size is in the set {1, 2, 4, 8, 16}.
 753
 754 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 755             Alan Hayward  <alan.hayward@arm.com>
 756             David Sherwood  <david.sherwood@arm.com>
 757
 758         * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
 759         Use plus_constant instead of gen_rtx_PLUS.
 760
 761 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 762             Alan Hayward  <alan.hayward@arm.com>
 763             David Sherwood  <david.sherwood@arm.com>
 764
 765         * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
 766         * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
 767         * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
 768         * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
 769         * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
 770         * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
 771         * config/i386/i386-protos.h (ix86_push_rounding): Declare.
 772         * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
 773         * config/i386/i386.c (ix86_push_rounding): ...this new function.
 774         * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
 775         a poly_int64.
 776         * config/m32c/m32c.c (m32c_push_rounding): Likewise.
 777         * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
 778         * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
 779         * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
 780         * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
 781         * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
 782         * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
 783         * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
 784         * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
 785         * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
 786         function.
 787         * expr.c (emit_move_resolve_push): Treat the input and result
 788         of PUSH_ROUNDING as a poly_int64.
 789         (emit_move_complex_push, emit_single_push_insn_1): Likewise.
 790         (emit_push_insn): Likewise.
 791         * lra-eliminations.c (mark_not_eliminable): Likewise.
 792         * recog.c (push_operand): Likewise.
 793         * reload1.c (elimination_effects): Likewise.
 794         * rtlanal.c (nonzero_bits1): Likewise.
 795         * calls.c (store_one_arg): Likewise.  Require the padding to be
 796         known at compile time.
 797
 798 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 799             Alan Hayward  <alan.hayward@arm.com>
 800             David Sherwood  <david.sherwood@arm.com>
 801
 802         * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
 803         Use plus_constant instead of gen_rtx_PLUS.
 804
 805 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 806             Alan Hayward  <alan.hayward@arm.com>
 807             David Sherwood  <david.sherwood@arm.com>
 808
 809         * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
 810         rather than an int.
 811
 812 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 813             Alan Hayward  <alan.hayward@arm.com>
 814             David Sherwood  <david.sherwood@arm.com>
 815
 816         * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
 817         instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
 818         via stack temporaries.  Treat the mode size as polynomial too.
 819
 820 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 821             Alan Hayward  <alan.hayward@arm.com>
 822             David Sherwood  <david.sherwood@arm.com>
 823
 824         * expr.c (expand_expr_real_2): When handling conversions involving
 825         unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
 826         multiplying int_size_in_bytes by BITS_PER_UNIT.  Treat GET_MODE_BISIZE
 827         as a poly_uint64 too.
 828
 829 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 830             Alan Hayward  <alan.hayward@arm.com>
 831             David Sherwood  <david.sherwood@arm.com>
 832
 833         * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
 834
 835 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 836             Alan Hayward  <alan.hayward@arm.com>
 837             David Sherwood  <david.sherwood@arm.com>
 838
 839         * combine.c (can_change_dest_mode): Handle polynomial
 840         REGMODE_NATURAL_SIZE.
 841         * expmed.c (store_bit_field_1): Likewise.
 842         * expr.c (store_constructor): Likewise.
 843         * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
 844         and polynomial REGMODE_NATURAL_SIZE.
 845         (gen_lowpart_common): Likewise.
 846         * reginfo.c (record_subregs_of_mode): Likewise.
 847         * rtlanal.c (read_modify_subreg_p): Likewise.
 848
 849 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 850             Alan Hayward  <alan.hayward@arm.com>
 851             David Sherwood  <david.sherwood@arm.com>
 852
 853         * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
 854         numbers of elements.
 855
 856 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 857             Alan Hayward  <alan.hayward@arm.com>
 858             David Sherwood  <david.sherwood@arm.com>
 859
 860         * match.pd: Cope with polynomial numbers of vector elements.
 861
 862 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 863             Alan Hayward  <alan.hayward@arm.com>
 864             David Sherwood  <david.sherwood@arm.com>
 865
 866         * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
 867         in a POINTER_PLUS_EXPR.
 868
 869 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 870             Alan Hayward  <alan.hayward@arm.com>
 871             David Sherwood  <david.sherwood@arm.com>
 872
 873         * omp-simd-clone.c (simd_clone_subparts): New function.
 874         (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
 875         (ipa_simd_modify_function_body): Likewise.
 876
 877 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 878             Alan Hayward  <alan.hayward@arm.com>
 879             David Sherwood  <david.sherwood@arm.com>
 880
 881         * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
 882         (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
 883         (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
 884         (expand_vector_condition, vector_element): Likewise.
 885         (subparts_gt): New function.
 886         (get_compute_type): Use subparts_gt.
 887         (count_type_subparts): Delete.
 888         (expand_vector_operations_1): Use subparts_gt instead of
 889         count_type_subparts.
 890
 891 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 892             Alan Hayward  <alan.hayward@arm.com>
 893             David Sherwood  <david.sherwood@arm.com>
 894
 895         * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
 896         (vect_compile_time_alias): ...this new function.  Do the calculation
 897         on poly_ints rather than trees.
 898         (vect_prune_runtime_alias_test_list): Update call accordingly.
 899
 900 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 901             Alan Hayward  <alan.hayward@arm.com>
 902             David Sherwood  <david.sherwood@arm.com>
 903
 904         * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
 905         numbers of units.
 906         (vect_schedule_slp_instance): Likewise.
 907
 908 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 909             Alan Hayward  <alan.hayward@arm.com>
 910             David Sherwood  <david.sherwood@arm.com>
 911
 912         * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
 913         constant and extern definitions for variable-length vectors.
 914         (vect_get_constant_vectors): Note that the number of units
 915         is known to be constant.
 916
 917 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 918             Alan Hayward  <alan.hayward@arm.com>
 919             David Sherwood  <david.sherwood@arm.com>
 920
 921         * tree-vect-stmts.c (vectorizable_conversion): Treat the number
 922         of units as polynomial.  Choose between WIDE and NARROW based
 923         on multiple_p.
 924
 925 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 926             Alan Hayward  <alan.hayward@arm.com>
 927             David Sherwood  <david.sherwood@arm.com>
 928
 929         * tree-vect-stmts.c (simd_clone_subparts): New function.
 930         (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
 931
 932 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 933             Alan Hayward  <alan.hayward@arm.com>
 934             David Sherwood  <david.sherwood@arm.com>
 935
 936         * tree-vect-stmts.c (vectorizable_call): Treat the number of
 937         vectors as polynomial.  Use build_index_vector for
 938         IFN_GOMP_SIMD_LANE.
 939
 940 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 941             Alan Hayward  <alan.hayward@arm.com>
 942             David Sherwood  <david.sherwood@arm.com>
 943
 944         * tree-vect-stmts.c (get_load_store_type): Treat the number of
 945         units as polynomial.  Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
 946         for variable-length vectors.
 947         (vectorizable_mask_load_store): Treat the number of units as
 948         polynomial, asserting that it is constant if the condition has
 949         already been enforced.
 950         (vectorizable_store, vectorizable_load): Likewise.
 951
 952 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 953             Alan Hayward  <alan.hayward@arm.com>
 954             David Sherwood  <david.sherwood@arm.com>
 955
 956         * tree-vect-loop.c (vectorizable_live_operation): Treat the number
 957         of units as polynomial.  Punt if we can't tell at compile time
 958         which vector contains the final result.
 959
 960 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 961             Alan Hayward  <alan.hayward@arm.com>
 962             David Sherwood  <david.sherwood@arm.com>
 963
 964         * tree-vect-loop.c (vectorizable_induction): Treat the number
 965         of units as polynomial.  Punt on SLP inductions.  Use an integer
 966         VEC_SERIES_EXPR for variable-length integer reductions.  Use a
 967         cast of such a series for variable-length floating-point
 968         reductions.
 969
 970 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 971             Alan Hayward  <alan.hayward@arm.com>
 972             David Sherwood  <david.sherwood@arm.com>
 973
 974         * tree.h (build_index_vector): Declare.
 975         * tree.c (build_index_vector): New function.
 976         * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
 977         of units as polynomial, forcibly converting it to a constant if
 978         vectorizable_reduction has already enforced the condition.
 979         (vect_create_epilog_for_reduction): Likewise.  Use build_index_vector
 980         to create a {1,2,3,...} vector.
 981         (vectorizable_reduction): Treat the number of units as polynomial.
 982         Choose vectype_in based on the largest scalar element size rather
 983         than the smallest number of units.  Enforce the restrictions
 984         relied on above.
 985
 986 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 987             Alan Hayward  <alan.hayward@arm.com>
 988             David Sherwood  <david.sherwood@arm.com>
 989
 990         * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
 991         number of units as polynomial.
 992
 993 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 994             Alan Hayward  <alan.hayward@arm.com>
 995             David Sherwood  <david.sherwood@arm.com>
 996
 997         * target.h (vector_sizes, auto_vector_sizes): New typedefs.
 998         * target.def (autovectorize_vector_sizes): Return the vector sizes
 999         by pointer, using vector_sizes rather than a bitmask.
1000         * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
1001         * targhooks.c (default_autovectorize_vector_sizes): Likewise.
1002         * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
1003         Likewise.
1004         * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
1005         * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
1006         * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
1007         * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
1008         * omp-general.c (omp_max_vf): Likewise.
1009         * omp-low.c (omp_clause_aligned_alignment): Likewise.
1010         * optabs-query.c (can_vec_mask_load_store_p): Likewise.
1011         * tree-vect-loop.c (vect_analyze_loop): Likewise.
1012         * tree-vect-slp.c (vect_slp_bb): Likewise.
1013         * doc/tm.texi: Regenerate.
1014         * tree-vectorizer.h (current_vector_size): Change from an unsigned int
1015         to a poly_uint64.
1016         * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
1017         the vector size as a poly_uint64 rather than an unsigned int.
1018         (current_vector_size): Change from an unsigned int to a poly_uint64.
1019         (get_vectype_for_scalar_type): Update accordingly.
1020         * tree.h (build_truth_vector_type): Take the size and number of
1021         units as a poly_uint64 rather than an unsigned int.
1022         (build_vector_type): Add a temporary overload that takes
1023         the number of units as a poly_uint64 rather than an unsigned int.
1024         * tree.c (make_vector_type): Likewise.
1025         (build_truth_vector_type): Take the number of units as a poly_uint64
1026         rather than an unsigned int.
1027
1028 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
1029             Alan Hayward  <alan.hayward@arm.com>
1030             David Sherwood  <david.sherwood@arm.com>
1031
1032         * target.def (get_mask_mode): Take the number of units and length
1033         as poly_uint64s rather than unsigned ints.
1034         * targhooks.h (default_get_mask_mode): Update accordingly.
1035         * targhooks.c (default_get_mask_mode): Likewise.
1036         * config/i386/i386.c (ix86_get_mask_mode): Likewise.
1037         * doc/tm.texi: Regenerate.
1038
1039 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
1040             Alan Hayward  <alan.hayward@arm.com>
1041             David Sherwood  <david.sherwood@arm.com>
1042
1043         * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
1044         * omp-general.c (omp_max_vf): Likewise.
1045         * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
1046         (expand_omp_simd): Handle polynomial safelen.
1047         * omp-low.c (omplow_simd_context): Add a default constructor.
1048         (omplow_simd_context::max_vf): Change from int to poly_uint64.
1049         (lower_rec_simd_input_clauses): Update accordingly.
1050         (lower_rec_input_clauses): Likewise.
1051
1052 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
1053             Alan Hayward  <alan.hayward@arm.com>
1054             David Sherwood  <david.sherwood@arm.com>
1055
1056         * tree-vectorizer.h (vect_nunits_for_cost): New function.
1057         * tree-vect-loop.c (vect_model_reduction_cost): Use it.
1058         * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
1059         (vect_analyze_slp_cost): Likewise.
1060         * tree-vect-stmts.c (vect_model_store_cost): Likewise.
1061         (vect_model_load_cost): Likewise.
1062
1063 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
1064             Alan Hayward  <alan.hayward@arm.com>
1065             David Sherwood  <david.sherwood@arm.com>
1066
1067         * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
1068         (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
1069         from an unsigned int * to a poly_uint64_pod *.
1070         (calculate_unrolling_factor): New function.
1071         (vect_analyze_slp_instance): Use it.  Track polynomial max_nunits.
1072
1073 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
1074             Alan Hayward  <alan.hayward@arm.com>
1075             David Sherwood  <david.sherwood@arm.com>
1076
1077         * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
1078         from an unsigned int to a poly_uint64.
1079         (_loop_vec_info::slp_unrolling_factor): Likewise.
1080         (_loop_vec_info::vectorization_factor): Change from an int
1081         to a poly_uint64.
1082         (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
1083         (vect_get_num_vectors): New function.
1084         (vect_update_max_nunits, vect_vf_for_cost): Likewise.
1085         (vect_get_num_copies): Use vect_get_num_vectors.
1086         (vect_analyze_data_ref_dependences): Change max_vf from an int *
1087         to an unsigned int *.
1088         (vect_analyze_data_refs): Change min_vf from an int * to a
1089         poly_uint64 *.
1090         (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
1091         than an unsigned HOST_WIDE_INT.
1092         * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
1093         (vect_analyze_data_ref_dependence): Change max_vf from an int *
1094         to an unsigned int *.
1095         (vect_analyze_data_ref_dependences): Likewise.
1096         (vect_compute_data_ref_alignment): Handle polynomial vf.
1097         (vect_enhance_data_refs_alignment): Likewise.
1098         (vect_prune_runtime_alias_test_list): Likewise.
1099         (vect_shift_permute_load_chain): Likewise.
1100         (vect_supportable_dr_alignment): Likewise.
1101         (dependence_distance_ge_vf): Take the vectorization factor as a
1102         poly_uint64 rather than an unsigned HOST_WIDE_INT.
1103         (vect_analyze_data_refs): Change min_vf from an int * to a
1104         poly_uint64 *.
1105         * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
1106         vfm1 as a poly_uint64 rather than an int.  Make the same change
1107         for the returned bound_scalar.
1108         (vect_gen_vector_loop_niters): Handle polynomial vf.
1109         (vect_do_peeling): Likewise.  Update call to
1110         vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
1111         (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
1112         be constant.
1113         * tree-vect-loop.c (vect_determine_vectorization_factor)
1114         (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
1115         (vect_get_known_peeling_cost): Likewise.
1116         (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
1117         (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
1118         (vect_transform_loop): Likewise.  Use the lowest possible VF when
1119         updating the upper bounds of the loop.
1120         (vect_min_worthwhile_factor): Make static.  Return an unsigned int
1121         rather than an int.
1122         * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
1123         polynomial unroll factors.
1124         (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
1125         (vect_make_slp_decision): Likewise.
1126         (vect_supported_load_permutation_p): Likewise, and polynomial
1127         vf too.
1128         (vect_analyze_slp_cost): Handle polynomial vf.
1129         (vect_slp_analyze_node_operations): Likewise.
1130         (vect_slp_analyze_bb_1): Likewise.
1131         (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
1132         than an unsigned HOST_WIDE_INT.
1133         * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
1134         (vectorizable_load): Handle polynomial vf.
1135         * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
1136         a poly_uint64.
1137         (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
1138
1139 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
1140             Alan Hayward  <alan.hayward@arm.com>
1141             David Sherwood  <david.sherwood@arm.com>
1142
1143         * match.pd: Handle bit operations involving three constants
1144         and try to fold one pair.
1145
1146 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
1147
1148         * tree-vect-loop-manip.c: Include gimple-fold.h.
1149         (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
1150         niters_maybe_zero parameters.  Handle other cases besides a step of 1.
1151         (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
1152         Add a path that uses a step of VF instead of 1, but disable it
1153         for now.
1154         (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
1155         and niters_no_overflow parameters.  Update calls to
1156         slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
1157         Create a new SSA name if the latter choses to use a ste other
1158         than zero, and return it via niters_vector_mult_vf_var.
1159         * tree-vect-loop.c (vect_transform_loop): Update calls to
1160         vect_do_peeling, vect_gen_vector_loop_niters and
1161         slpeel_make_loop_iterate_ntimes.
1162         * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
1163         (vect_gen_vector_loop_niters): Update declarations after above changes.
1164
1165 2018-01-02  Michael Meissner  <meissner@linux.vnet.ibm.com>
1166
1167         * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
1168         128-bit round to integer instructions.
1169         (ceil<mode>2): Likewise.
1170         (btrunc<mode>2): Likewise.
1171         (round<mode>2): Likewise.
1172
1173 2018-01-02  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
1174
1175         * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
1176         unaligned VSX load/store on P8/P9.
1177         (expand_block_clear): Allow the use of unaligned VSX
1178         load/store on P8/P9.
1179
1180 2018-01-02  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
1181
1182         * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
1183         New function.
1184         (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
1185         swap associated with both a load and a store.
1186
1187 2018-01-02  Andrew Waterman  <andrew@sifive.com>
1188
1189         * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
1190         * config/riscv/riscv.md (clear_cache): Use it.
1191
1192 2018-01-02  Artyom Skrobov  <tyomitch@gmail.com>
1193
1194         * web.c: Remove out-of-date comment.
1195
1196 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
1197
1198         * expr.c (fixup_args_size_notes): Check that any existing
1199         REG_ARGS_SIZE notes are correct, and don't try to re-add them.
1200         (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
1201         (emit_single_push_insn): ...here.
1202
1203 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
1204
1205         * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
1206         (const_vector_encoded_nelts): New function.
1207         (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
1208         (const_vector_int_elt, const_vector_elt): Declare.
1209         * emit-rtl.c (const_vector_int_elt_1): New function.
1210         (const_vector_elt): Likewise.
1211         * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
1212         of CONST_VECTOR_ELT.
1213
1214 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
1215
1216         * expr.c: Include rtx-vector-builder.h.
1217         (const_vector_mask_from_tree): Use rtx_vector_builder and operate
1218         directly on the tree encoding.
1219         (const_vector_from_tree): Likewise.
1220         * optabs.c: Include rtx-vector-builder.h.
1221         (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
1222         sequence of "u" values.
1223         * vec-perm-indices.c: Include rtx-vector-builder.h.
1224         (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
1225         directly on the vec_perm_indices encoding.
1226
1227 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
1228
1229         * doc/rtl.texi (const_vector): Describe new encoding scheme.
1230         * Makefile.in (OBJS): Add rtx-vector-builder.o.
1231         * rtx-vector-builder.h: New file.
1232         * rtx-vector-builder.c: Likewise.
1233         * rtl.h (rtx_def::u2): Add a const_vector field.
1234         (CONST_VECTOR_NPATTERNS): New macro.
1235         (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
1236         (CONST_VECTOR_DUPLICATE_P): Likewise.
1237         (CONST_VECTOR_STEPPED_P): Likewise.
1238         (CONST_VECTOR_ENCODED_ELT): Likewise.
1239         (const_vec_duplicate_p): Check for a duplicated vector encoding.
1240         (unwrap_const_vec_duplicate): Likewise.
1241         (const_vec_series_p): Check for a non-duplicated vector encoding.
1242         Say that the function only returns true for integer vectors.
1243         * emit-rtl.c: Include rtx-vector-builder.h.
1244         (gen_const_vec_duplicate_1): Delete.
1245         (gen_const_vector): Call gen_const_vec_duplicate instead of
1246         gen_const_vec_duplicate_1.
1247         (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
1248         (gen_const_vec_duplicate): Use rtx_vector_builder.
1249         (gen_const_vec_series): Likewise.
1250         (gen_rtx_CONST_VECTOR): Likewise.
1251         * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
1252         (swap_const_vector_halves): Take an rtx pointer rather than rtx.
1253         Build a new vector rather than modifying a CONST_VECTOR in-place.
1254         (handle_special_swappables): Update call accordingly.
1255         * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
1256         (swap_const_vector_halves): Take an rtx pointer rather than rtx.
1257         Build a new vector rather than modifying a CONST_VECTOR in-place.
1258         (handle_special_swappables): Update call accordingly.
1259
1260 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
1261
1262         * simplify-rtx.c (simplify_const_binary_operation): Use
1263         CONST_VECTOR_ELT instead of XVECEXP.
1264
1265 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
1266
1267         * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
1268         the selector elements to be different from the data elements
1269         if the selector is a VECTOR_CST.
1270         * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
1271         ssizetype for the selector.
1272
1273 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
1274
1275         * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
1276         before testing each element individually.
1277         * tree-vect-generic.c (lower_vec_perm): Likewise.
1278
1279 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
1280
1281         * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
1282         * selftest-run-tests.c (selftest::run_tests): Call it.
1283         * vector-builder.h (vector_builder::operator ==): New function.
1284         (vector_builder::operator !=): Likewise.
1285         * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
1286         (vec_perm_indices::all_from_input_p): New function.
1287         * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
1288         (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
1289         * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
1290         instead of reading the VECTOR_CST directly.  Detect whether both
1291         vector inputs are the same before constructing the vec_perm_indices,
1292         and update the number of inputs argument accordingly.  Use the
1293         utility functions added above.  Only construct sel2 if we need to.
1294
1295 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
1296
1297         * optabs.c (expand_vec_perm_var): Use an explicit encoding for
1298         the broadcast of the low byte.
1299         (expand_mult_highpart): Use an explicit encoding for the permutes.
1300         * optabs-query.c (can_mult_highpart_p): Likewise.
1301         * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
1302         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1303         (vectorizable_bswap): Likewise.
1304         * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
1305         explicit encoding for the power-of-2 permutes.
1306         (vect_permute_store_chain): Likewise.
1307         (vect_grouped_load_supported): Likewise.
1308         (vect_permute_load_chain): Likewise.
1309
1310 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
1311
1312         * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
1313         * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
1314         * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
1315         * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
1316         * tree-vect-stmts.c (vectorizable_bswap): Likewise.
1317         (vect_gen_perm_mask_any): Likewise.
1318
1319 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
1320
1321         * int-vector-builder.h: New file.
1322         * vec-perm-indices.h: Include int-vector-builder.h.
1323         (vec_perm_indices): Redefine as an int_vector_builder.
1324         (auto_vec_perm_indices): Delete.
1325         (vec_perm_builder): Redefine as a stand-alone class.
1326         (vec_perm_indices::vec_perm_indices): New function.
1327         (vec_perm_indices::clamp): Likewise.
1328         * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
1329         (vec_perm_indices::new_vector): New function.
1330         (vec_perm_indices::new_expanded_vector): Update for new
1331         vec_perm_indices class.
1332         (vec_perm_indices::rotate_inputs): New function.
1333         (vec_perm_indices::all_in_range_p): Operate directly on the
1334         encoded form, without computing elided elements.
1335         (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
1336         encoding.  Update for new vec_perm_indices class.
1337         * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
1338         the given vec_perm_builder.
1339         (expand_vec_perm_var): Update vec_perm_builder constructor.
1340         (expand_mult_highpart): Use vec_perm_builder instead of
1341         auto_vec_perm_indices.
1342         * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
1343         vec_perm_indices instead of auto_vec_perm_indices.  Use a single
1344         or double series encoding as appropriate.
1345         * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
1346         vec_perm_indices instead of auto_vec_perm_indices.
1347         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1348         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1349         (vect_permute_store_chain): Likewise.
1350         (vect_grouped_load_supported): Likewise.
1351         (vect_permute_load_chain): Likewise.
1352         (vect_shift_permute_load_chain): Likewise.
1353         * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
1354         (vect_transform_slp_perm_load): Likewise.
1355         (vect_schedule_slp_instance): Likewise.
1356         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1357         (vectorizable_mask_load_store): Likewise.
1358         (vectorizable_bswap): Likewise.
1359         (vectorizable_store): Likewise.
1360         (vectorizable_load): Likewise.
1361         * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
1362         vec_perm_indices instead of auto_vec_perm_indices.  Use
1363         tree_to_vec_perm_builder to read the vector from a tree.
1364         * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
1365         vec_perm_builder instead of a vec_perm_indices.
1366         (have_whole_vector_shift): Use vec_perm_builder and
1367         vec_perm_indices instead of auto_vec_perm_indices.  Leave the
1368         truncation to calc_vec_perm_mask_for_shift.
1369         (vect_create_epilog_for_reduction): Likewise.
1370         * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
1371         from auto_vec_perm_indices to vec_perm_indices.
1372         (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1373         instead of changing individual elements.
1374         (aarch64_vectorize_vec_perm_const): Use new_vector to install
1375         the vector in d.perm.
1376         * config/arm/arm.c (expand_vec_perm_d::perm): Change
1377         from auto_vec_perm_indices to vec_perm_indices.
1378         (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
1379         instead of changing individual elements.
1380         (arm_vectorize_vec_perm_const): Use new_vector to install
1381         the vector in d.perm.
1382         * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
1383         Update vec_perm_builder constructor.
1384         (rs6000_expand_interleave): Likewise.
1385         * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
1386         (rs6000_expand_interleave): Likewise.
1387
1388 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
1389
1390         * optabs-query.c (can_vec_perm_var_p): Check whether lowering
1391         to qimode could truncate the indices.
1392         * optabs.c (expand_vec_perm_var): Likewise.
1393
1394 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
1395
1396         * Makefile.in (OBJS): Add vec-perm-indices.o.
1397         * vec-perm-indices.h: New file.
1398         * vec-perm-indices.c: Likewise.
1399         * target.h (vec_perm_indices): Replace with a forward class
1400         declaration.
1401         (auto_vec_perm_indices): Move to vec-perm-indices.h.
1402         * optabs.h: Include vec-perm-indices.h.
1403         (expand_vec_perm): Delete.
1404         (selector_fits_mode_p, expand_vec_perm_var): Declare.
1405         (expand_vec_perm_const): Declare.
1406         * target.def (vec_perm_const_ok): Replace with...
1407         (vec_perm_const): ...this new hook.
1408         * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
1409         (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
1410         * doc/tm.texi: Regenerate.
1411         * optabs.def (vec_perm_const): Delete.
1412         * doc/md.texi (vec_perm_const): Likewise.
1413         (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
1414         * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
1415         expand_vec_perm for constant permutation vectors.  Assert that
1416         the mode of variable permutation vectors is the integer equivalent
1417         of the mode that is being permuted.
1418         * optabs-query.h (selector_fits_mode_p): Declare.
1419         * optabs-query.c: Include vec-perm-indices.h.
1420         (selector_fits_mode_p): New function.
1421         (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
1422         is defined, instead of checking whether the vec_perm_const_optab
1423         exists.  Use targetm.vectorize.vec_perm_const instead of
1424         targetm.vectorize.vec_perm_const_ok.  Check whether the indices
1425         fit in the vector mode before using a variable permute.
1426         * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
1427         vec_perm_indices instead of an rtx.
1428         (expand_vec_perm): Replace with...
1429         (expand_vec_perm_const): ...this new function.  Take the selector
1430         as a vec_perm_indices rather than an rtx.  Also take the mode of
1431         the selector.  Update call to shift_amt_for_vec_perm_mask.
1432         Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
1433         Use vec_perm_indices::new_expanded_vector to expand the original
1434         selector into bytes.  Check whether the indices fit in the vector
1435         mode before using a variable permute.
1436         (expand_vec_perm_var): Make global.
1437         (expand_mult_highpart): Use expand_vec_perm_const.
1438         * fold-const.c: Includes vec-perm-indices.h.
1439         * tree-ssa-forwprop.c: Likewise.
1440         * tree-vect-data-refs.c: Likewise.
1441         * tree-vect-generic.c: Likewise.
1442         * tree-vect-loop.c: Likewise.
1443         * tree-vect-slp.c: Likewise.
1444         * tree-vect-stmts.c: Likewise.
1445         * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
1446         Delete.
1447         * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
1448         * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
1449         (aarch64_vectorize_vec_perm_const_ok): Fuse into...
1450         (aarch64_vectorize_vec_perm_const): ...this new function.
1451         (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1452         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1453         * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
1454         * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
1455         * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1456         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1457         (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
1458         into...
1459         (arm_vectorize_vec_perm_const): ...this new function.  Explicitly
1460         check for NEON modes.
1461         * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
1462         * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
1463         * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
1464         (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
1465         into...
1466         (ix86_vectorize_vec_perm_const): ...this new function.  Incorporate
1467         the old VEC_PERM_CONST conditions.
1468         * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
1469         * config/ia64/vect.md (vec_perm_const<mode>): Delete.
1470         * config/ia64/ia64.c (ia64_expand_vec_perm_const)
1471         (ia64_vectorize_vec_perm_const_ok): Merge into...
1472         (ia64_vectorize_vec_perm_const): ...this new function.
1473         * config/mips/loongson.md (vec_perm_const<mode>): Delete.
1474         * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
1475         * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
1476         * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
1477         * config/mips/mips.c (mips_expand_vec_perm_const)
1478         (mips_vectorize_vec_perm_const_ok): Merge into...
1479         (mips_vectorize_vec_perm_const): ...this new function.
1480         * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
1481         * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
1482         * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
1483         * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
1484         * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
1485         (rs6000_expand_vec_perm_const): Delete.
1486         * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
1487         Delete.
1488         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1489         (altivec_expand_vec_perm_const_le): Take each operand individually.
1490         Operate on constant selectors rather than rtxes.
1491         (altivec_expand_vec_perm_const): Likewise.  Update call to
1492         altivec_expand_vec_perm_const_le.
1493         (rs6000_expand_vec_perm_const): Delete.
1494         (rs6000_vectorize_vec_perm_const_ok): Delete.
1495         (rs6000_vectorize_vec_perm_const): New function.
1496         (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
1497         an element count and rtx array.
1498         (rs6000_expand_extract_even): Update call accordingly.
1499         (rs6000_expand_interleave): Likewise.
1500         * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
1501         * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
1502         * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
1503         * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
1504         (rs6000_expand_vec_perm_const): Delete.
1505         * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
1506         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1507         (altivec_expand_vec_perm_const_le): Take each operand individually.
1508         Operate on constant selectors rather than rtxes.
1509         (altivec_expand_vec_perm_const): Likewise.  Update call to
1510         altivec_expand_vec_perm_const_le.
1511         (rs6000_expand_vec_perm_const): Delete.
1512         (rs6000_vectorize_vec_perm_const_ok): Delete.
1513         (rs6000_vectorize_vec_perm_const): New function.  Remove stray
1514         reference to the SPE evmerge intructions.
1515         (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
1516         an element count and rtx array.
1517         (rs6000_expand_extract_even): Update call accordingly.
1518         (rs6000_expand_interleave): Likewise.
1519         * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
1520         * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
1521         new function.
1522         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
1523
1524 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
1525
1526         * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
1527         vector mode and that that mode matches the mode of the data
1528         being permuted.
1529         (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
1530         out into expand_vec_perm_var.  Do all CONST_VECTOR handling here,
1531         directly using expand_vec_perm_1 when forcing selectors into
1532         registers.
1533         (expand_vec_perm_var): New function, split out from expand_vec_perm.
1534
1535 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
1536
1537         * optabs-query.h (can_vec_perm_p): Delete.
1538         (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
1539         * optabs-query.c (can_vec_perm_p): Split into...
1540         (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
1541         (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
1542         particular selector is valid.
1543         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
1544         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
1545         (vect_grouped_load_supported): Likewise.
1546         (vect_shift_permute_load_chain): Likewise.
1547         * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
1548         (vect_transform_slp_perm_load): Likewise.
1549         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
1550         (vectorizable_bswap): Likewise.
1551         (vect_gen_perm_mask_checked): Likewise.
1552         * fold-const.c (fold_ternary_loc): Likewise.  Don't take
1553         implementations of variable permutation vectors into account
1554         when deciding which selector to use.
1555         * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
1556         vec_perm_const_optab is supported; instead use can_vec_perm_const_p
1557         with a false third argument.
1558         * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
1559         to test whether the constant selector is valid and can_vec_perm_var_p
1560         to test whether a variable selector is valid.
1561
1562 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
1563
1564         * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
1565         * optabs-query.c (can_vec_perm_p): Likewise.
1566         * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
1567         instead of vec_perm_indices.
1568         * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
1569         (vect_gen_perm_mask_checked): Likewise,
1570         * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
1571         (vect_gen_perm_mask_checked): Likewise,
1572
1573 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
1574
1575         * optabs-query.h (qimode_for_vec_perm): Declare.
1576         * optabs-query.c (can_vec_perm_p): Split out qimode search to...
1577         (qimode_for_vec_perm): ...this new function.
1578         * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
1579
1580 2018-01-02  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
1581
1582         * rtlanal.c (canonicalize_condition): Return 0 if final rtx
1583         does not have a conditional at the top.
1584
1585 2018-01-02  Richard Biener  <rguenther@suse.de>
1586
1587         * ipa-inline.c (big_speedup_p): Fix expression.
1588
1589 2018-01-02  Jan Hubicka  <hubicka@ucw.cz>
1590
1591         PR target/81616
1592         * config/i386/x86-tune-costs.h: Increase cost of integer load costs
1593         for generic 4->6.
1594
1595 2018-01-02  Jan Hubicka  <hubicka@ucw.cz>
1596
1597         PR target/81616
1598         Generic tuning.
1599         * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
1600         cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
1601         and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
1602         cond_taken_branch_cost 3->4.
1603
1604 2018-01-01  Jakub Jelinek  <jakub@redhat.com>
1605
1606         PR tree-optimization/83581
1607         * tree-loop-distribution.c (pass_loop_distribution::execute): Return
1608         TODO_cleanup_cfg if any changes have been made.
1609
1610         PR middle-end/83608
1611         * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
1612         convert_modes if target mode has the right side, but different mode
1613         class.
1614
1615         PR middle-end/83609
1616         * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
1617         last argument when extracting from CONCAT.  If either from_real or
1618         from_imag is NULL, use expansion through memory.  If result is not
1619         a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
1620         the parts directly to inner mode, if even that fails, use expansion
1621         through memory.
1622
1623         PR middle-end/83623
1624         * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
1625         check for bswap in mode rather than HImode and use that in expand_unop
1626         too.
1627 \f
1628 Copyright (C) 2018 Free Software Foundation, Inc.
1629
1630 Copying and distribution of this file, with or without modification,
1631 are permitted in any medium without royalty provided the copyright
1632 notice and this notice are preserved.