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