gcc/ChangeLog

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