gcc/ChangeLog

   1 2018-01-03  Jan Beulich  <jbeulich@suse.com>
   2
   3         * sse.md (mov<mode>_internal): Tighten condition for when to use
   4         vmovdqu<ssescalarsize> for TI and OI modes.
   5
   6 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
   7
   8         Update copyright years.
   9
  10 2018-01-03  Martin Liska  <mliska@suse.cz>
  11
  12         PR ipa/83594
  13         * ipa-visibility.c (function_and_variable_visibility): Skip
  14         functions with noipa attribure.
  15
  16 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
  17
  18         * gcc.c (process_command): Update copyright notice dates.
  19         * gcov-dump.c (print_version): Ditto.
  20         * gcov.c (print_version): Ditto.
  21         * gcov-tool.c (print_version): Ditto.
  22         * gengtype.c (create_file): Ditto.
  23         * doc/cpp.texi: Bump @copying's copyright year.
  24         * doc/cppinternals.texi: Ditto.
  25         * doc/gcc.texi: Ditto.
  26         * doc/gccint.texi: Ditto.
  27         * doc/gcov.texi: Ditto.
  28         * doc/install.texi: Ditto.
  29         * doc/invoke.texi: Ditto.
  30
  31 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
  32
  33         * vector-builder.h (vector_builder::m_full_nelts): Change from
  34         unsigned int to poly_uint64.
  35         (vector_builder::full_nelts): Update prototype accordingly.
  36         (vector_builder::new_vector): Likewise.
  37         (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
  38         (vector_builder::operator ==): Likewise.
  39         (vector_builder::finalize): Likewise.
  40         * int-vector-builder.h (int_vector_builder::int_vector_builder):
  41         Take the number of elements as a poly_uint64 rather than an
  42         unsigned int.
  43         * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
  44         from unsigned int to poly_uint64.
  45         (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
  46         (vec_perm_indices::new_vector): Likewise.
  47         (vec_perm_indices::length): Likewise.
  48         (vec_perm_indices::nelts_per_input): Likewise.
  49         (vec_perm_indices::input_nelts): Likewise.
  50         * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
  51         number of elements per input as a poly_uint64 rather than an
  52         unsigned int.  Use the original encoding for variable-length
  53         vectors, rather than clamping each individual element.
  54         For the second and subsequent elements in each pattern,
  55         clamp the step and base before clamping their sum.
  56         (vec_perm_indices::series_p): Handle polynomial element counts.
  57         (vec_perm_indices::all_in_range_p): Likewise.
  58         (vec_perm_indices_to_tree): Likewise.
  59         (vec_perm_indices_to_rtx): Likewise.
  60         * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
  61         * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
  62         (tree_vector_builder::new_binary_operation): Handle polynomial
  63         element counts.  Return false if we need to know the number
  64         of elements at compile time.
  65         * fold-const.c (fold_vec_perm): Punt if the number of elements
  66         isn't known at compile time.
  67
  68 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
  69
  70         * vec-perm-indices.h (vec_perm_builder): Change element type
  71         from HOST_WIDE_INT to poly_int64.
  72         (vec_perm_indices::element_type): Update accordingly.
  73         (vec_perm_indices::clamp): Handle polynomial element_types.
  74         * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
  75         (vec_perm_indices::all_in_range_p): Likewise.
  76         (tree_to_vec_perm_builder): Check for poly_int64 trees rather
  77         than shwi trees.
  78         * vector-builder.h (vector_builder::stepped_sequence_p): Handle
  79         polynomial vec_perm_indices element types.
  80         * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
  81         * fold-const.c (fold_vec_perm): Likewise.
  82         * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
  83         * tree-vect-generic.c (lower_vec_perm): Likewise.
  84         * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
  85         * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
  86         element type to HOST_WIDE_INT.
  87
  88 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
  89             Alan Hayward  <alan.hayward@arm.com>
  90             David Sherwood  <david.sherwood@arm.com>
  91
  92         * alias.c (addr_side_effect_eval): Take the size as a poly_int64
  93         rather than an int.  Use plus_constant.
  94         (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
  95         Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
  96
  97 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
  98             Alan Hayward  <alan.hayward@arm.com>
  99             David Sherwood  <david.sherwood@arm.com>
 100
 101         * calls.c (emit_call_1, expand_call): Change struct_value_size from
 102         a HOST_WIDE_INT to a poly_int64.
 103
 104 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 105             Alan Hayward  <alan.hayward@arm.com>
 106             David Sherwood  <david.sherwood@arm.com>
 107
 108         * calls.c (load_register_parameters): Cope with polynomial
 109         mode sizes.  Require a constant size for BLKmode parameters
 110         that aren't described by a PARALLEL.  If BLOCK_REG_PADDING
 111         forces a parameter to be padded at the lsb end in order to
 112         fill a complete number of words, require the parameter size
 113         to be ordered wrt UNITS_PER_WORD.
 114
 115 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 116             Alan Hayward  <alan.hayward@arm.com>
 117             David Sherwood  <david.sherwood@arm.com>
 118
 119         * reload1.c (spill_stack_slot_width): Change element type
 120         from unsigned int to poly_uint64_pod.
 121         (alter_reg): Treat mode sizes as polynomial.
 122
 123 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 124             Alan Hayward  <alan.hayward@arm.com>
 125             David Sherwood  <david.sherwood@arm.com>
 126
 127         * reload.c (complex_word_subreg_p): New function.
 128         (reload_inner_reg_of_subreg, push_reload): Use it.
 129
 130 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 131             Alan Hayward  <alan.hayward@arm.com>
 132             David Sherwood  <david.sherwood@arm.com>
 133
 134         * lra-constraints.c (process_alt_operands): Reject matched
 135         operands whose sizes aren't ordered.
 136         (match_reload): Refer to this check here.
 137
 138 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 139             Alan Hayward  <alan.hayward@arm.com>
 140             David Sherwood  <david.sherwood@arm.com>
 141
 142         * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
 143         that the mode size is in the set {1, 2, 4, 8, 16}.
 144
 145 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 146             Alan Hayward  <alan.hayward@arm.com>
 147             David Sherwood  <david.sherwood@arm.com>
 148
 149         * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
 150         Use plus_constant instead of gen_rtx_PLUS.
 151
 152 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 153             Alan Hayward  <alan.hayward@arm.com>
 154             David Sherwood  <david.sherwood@arm.com>
 155
 156         * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
 157         * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
 158         * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
 159         * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
 160         * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
 161         * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
 162         * config/i386/i386-protos.h (ix86_push_rounding): Declare.
 163         * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
 164         * config/i386/i386.c (ix86_push_rounding): ...this new function.
 165         * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
 166         a poly_int64.
 167         * config/m32c/m32c.c (m32c_push_rounding): Likewise.
 168         * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
 169         * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
 170         * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
 171         * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
 172         * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
 173         * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
 174         * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
 175         * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
 176         * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
 177         function.
 178         * expr.c (emit_move_resolve_push): Treat the input and result
 179         of PUSH_ROUNDING as a poly_int64.
 180         (emit_move_complex_push, emit_single_push_insn_1): Likewise.
 181         (emit_push_insn): Likewise.
 182         * lra-eliminations.c (mark_not_eliminable): Likewise.
 183         * recog.c (push_operand): Likewise.
 184         * reload1.c (elimination_effects): Likewise.
 185         * rtlanal.c (nonzero_bits1): Likewise.
 186         * calls.c (store_one_arg): Likewise.  Require the padding to be
 187         known at compile time.
 188
 189 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 190             Alan Hayward  <alan.hayward@arm.com>
 191             David Sherwood  <david.sherwood@arm.com>
 192
 193         * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
 194         Use plus_constant instead of gen_rtx_PLUS.
 195
 196 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 197             Alan Hayward  <alan.hayward@arm.com>
 198             David Sherwood  <david.sherwood@arm.com>
 199
 200         * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
 201         rather than an int.
 202
 203 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 204             Alan Hayward  <alan.hayward@arm.com>
 205             David Sherwood  <david.sherwood@arm.com>
 206
 207         * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
 208         instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
 209         via stack temporaries.  Treat the mode size as polynomial too.
 210
 211 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 212             Alan Hayward  <alan.hayward@arm.com>
 213             David Sherwood  <david.sherwood@arm.com>
 214
 215         * expr.c (expand_expr_real_2): When handling conversions involving
 216         unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
 217         multiplying int_size_in_bytes by BITS_PER_UNIT.  Treat GET_MODE_BISIZE
 218         as a poly_uint64 too.
 219
 220 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 221             Alan Hayward  <alan.hayward@arm.com>
 222             David Sherwood  <david.sherwood@arm.com>
 223
 224         * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
 225
 226 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 227             Alan Hayward  <alan.hayward@arm.com>
 228             David Sherwood  <david.sherwood@arm.com>
 229
 230         * combine.c (can_change_dest_mode): Handle polynomial
 231         REGMODE_NATURAL_SIZE.
 232         * expmed.c (store_bit_field_1): Likewise.
 233         * expr.c (store_constructor): Likewise.
 234         * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
 235         and polynomial REGMODE_NATURAL_SIZE.
 236         (gen_lowpart_common): Likewise.
 237         * reginfo.c (record_subregs_of_mode): Likewise.
 238         * rtlanal.c (read_modify_subreg_p): Likewise.
 239
 240 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 241             Alan Hayward  <alan.hayward@arm.com>
 242             David Sherwood  <david.sherwood@arm.com>
 243
 244         * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
 245         numbers of elements.
 246
 247 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 248             Alan Hayward  <alan.hayward@arm.com>
 249             David Sherwood  <david.sherwood@arm.com>
 250
 251         * match.pd: Cope with polynomial numbers of vector elements.
 252
 253 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 254             Alan Hayward  <alan.hayward@arm.com>
 255             David Sherwood  <david.sherwood@arm.com>
 256
 257         * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
 258         in a POINTER_PLUS_EXPR.
 259
 260 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 261             Alan Hayward  <alan.hayward@arm.com>
 262             David Sherwood  <david.sherwood@arm.com>
 263
 264         * omp-simd-clone.c (simd_clone_subparts): New function.
 265         (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
 266         (ipa_simd_modify_function_body): Likewise.
 267
 268 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 269             Alan Hayward  <alan.hayward@arm.com>
 270             David Sherwood  <david.sherwood@arm.com>
 271
 272         * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
 273         (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
 274         (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
 275         (expand_vector_condition, vector_element): Likewise.
 276         (subparts_gt): New function.
 277         (get_compute_type): Use subparts_gt.
 278         (count_type_subparts): Delete.
 279         (expand_vector_operations_1): Use subparts_gt instead of
 280         count_type_subparts.
 281
 282 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 283             Alan Hayward  <alan.hayward@arm.com>
 284             David Sherwood  <david.sherwood@arm.com>
 285
 286         * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
 287         (vect_compile_time_alias): ...this new function.  Do the calculation
 288         on poly_ints rather than trees.
 289         (vect_prune_runtime_alias_test_list): Update call accordingly.
 290
 291 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 292             Alan Hayward  <alan.hayward@arm.com>
 293             David Sherwood  <david.sherwood@arm.com>
 294
 295         * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
 296         numbers of units.
 297         (vect_schedule_slp_instance): Likewise.
 298
 299 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 300             Alan Hayward  <alan.hayward@arm.com>
 301             David Sherwood  <david.sherwood@arm.com>
 302
 303         * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
 304         constant and extern definitions for variable-length vectors.
 305         (vect_get_constant_vectors): Note that the number of units
 306         is known to be constant.
 307
 308 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 309             Alan Hayward  <alan.hayward@arm.com>
 310             David Sherwood  <david.sherwood@arm.com>
 311
 312         * tree-vect-stmts.c (vectorizable_conversion): Treat the number
 313         of units as polynomial.  Choose between WIDE and NARROW based
 314         on multiple_p.
 315
 316 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 317             Alan Hayward  <alan.hayward@arm.com>
 318             David Sherwood  <david.sherwood@arm.com>
 319
 320         * tree-vect-stmts.c (simd_clone_subparts): New function.
 321         (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
 322
 323 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 324             Alan Hayward  <alan.hayward@arm.com>
 325             David Sherwood  <david.sherwood@arm.com>
 326
 327         * tree-vect-stmts.c (vectorizable_call): Treat the number of
 328         vectors as polynomial.  Use build_index_vector for
 329         IFN_GOMP_SIMD_LANE.
 330
 331 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 332             Alan Hayward  <alan.hayward@arm.com>
 333             David Sherwood  <david.sherwood@arm.com>
 334
 335         * tree-vect-stmts.c (get_load_store_type): Treat the number of
 336         units as polynomial.  Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
 337         for variable-length vectors.
 338         (vectorizable_mask_load_store): Treat the number of units as
 339         polynomial, asserting that it is constant if the condition has
 340         already been enforced.
 341         (vectorizable_store, vectorizable_load): Likewise.
 342
 343 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 344             Alan Hayward  <alan.hayward@arm.com>
 345             David Sherwood  <david.sherwood@arm.com>
 346
 347         * tree-vect-loop.c (vectorizable_live_operation): Treat the number
 348         of units as polynomial.  Punt if we can't tell at compile time
 349         which vector contains the final result.
 350
 351 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 352             Alan Hayward  <alan.hayward@arm.com>
 353             David Sherwood  <david.sherwood@arm.com>
 354
 355         * tree-vect-loop.c (vectorizable_induction): Treat the number
 356         of units as polynomial.  Punt on SLP inductions.  Use an integer
 357         VEC_SERIES_EXPR for variable-length integer reductions.  Use a
 358         cast of such a series for variable-length floating-point
 359         reductions.
 360
 361 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 362             Alan Hayward  <alan.hayward@arm.com>
 363             David Sherwood  <david.sherwood@arm.com>
 364
 365         * tree.h (build_index_vector): Declare.
 366         * tree.c (build_index_vector): New function.
 367         * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
 368         of units as polynomial, forcibly converting it to a constant if
 369         vectorizable_reduction has already enforced the condition.
 370         (vect_create_epilog_for_reduction): Likewise.  Use build_index_vector
 371         to create a {1,2,3,...} vector.
 372         (vectorizable_reduction): Treat the number of units as polynomial.
 373         Choose vectype_in based on the largest scalar element size rather
 374         than the smallest number of units.  Enforce the restrictions
 375         relied on above.
 376
 377 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 378             Alan Hayward  <alan.hayward@arm.com>
 379             David Sherwood  <david.sherwood@arm.com>
 380
 381         * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
 382         number of units as polynomial.
 383
 384 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 385             Alan Hayward  <alan.hayward@arm.com>
 386             David Sherwood  <david.sherwood@arm.com>
 387
 388         * target.h (vector_sizes, auto_vector_sizes): New typedefs.
 389         * target.def (autovectorize_vector_sizes): Return the vector sizes
 390         by pointer, using vector_sizes rather than a bitmask.
 391         * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
 392         * targhooks.c (default_autovectorize_vector_sizes): Likewise.
 393         * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
 394         Likewise.
 395         * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
 396         * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
 397         * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
 398         * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
 399         * omp-general.c (omp_max_vf): Likewise.
 400         * omp-low.c (omp_clause_aligned_alignment): Likewise.
 401         * optabs-query.c (can_vec_mask_load_store_p): Likewise.
 402         * tree-vect-loop.c (vect_analyze_loop): Likewise.
 403         * tree-vect-slp.c (vect_slp_bb): Likewise.
 404         * doc/tm.texi: Regenerate.
 405         * tree-vectorizer.h (current_vector_size): Change from an unsigned int
 406         to a poly_uint64.
 407         * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
 408         the vector size as a poly_uint64 rather than an unsigned int.
 409         (current_vector_size): Change from an unsigned int to a poly_uint64.
 410         (get_vectype_for_scalar_type): Update accordingly.
 411         * tree.h (build_truth_vector_type): Take the size and number of
 412         units as a poly_uint64 rather than an unsigned int.
 413         (build_vector_type): Add a temporary overload that takes
 414         the number of units as a poly_uint64 rather than an unsigned int.
 415         * tree.c (make_vector_type): Likewise.
 416         (build_truth_vector_type): Take the number of units as a poly_uint64
 417         rather than an unsigned int.
 418
 419 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 420             Alan Hayward  <alan.hayward@arm.com>
 421             David Sherwood  <david.sherwood@arm.com>
 422
 423         * target.def (get_mask_mode): Take the number of units and length
 424         as poly_uint64s rather than unsigned ints.
 425         * targhooks.h (default_get_mask_mode): Update accordingly.
 426         * targhooks.c (default_get_mask_mode): Likewise.
 427         * config/i386/i386.c (ix86_get_mask_mode): Likewise.
 428         * doc/tm.texi: Regenerate.
 429
 430 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 431             Alan Hayward  <alan.hayward@arm.com>
 432             David Sherwood  <david.sherwood@arm.com>
 433
 434         * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
 435         * omp-general.c (omp_max_vf): Likewise.
 436         * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
 437         (expand_omp_simd): Handle polynomial safelen.
 438         * omp-low.c (omplow_simd_context): Add a default constructor.
 439         (omplow_simd_context::max_vf): Change from int to poly_uint64.
 440         (lower_rec_simd_input_clauses): Update accordingly.
 441         (lower_rec_input_clauses): Likewise.
 442
 443 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 444             Alan Hayward  <alan.hayward@arm.com>
 445             David Sherwood  <david.sherwood@arm.com>
 446
 447         * tree-vectorizer.h (vect_nunits_for_cost): New function.
 448         * tree-vect-loop.c (vect_model_reduction_cost): Use it.
 449         * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
 450         (vect_analyze_slp_cost): Likewise.
 451         * tree-vect-stmts.c (vect_model_store_cost): Likewise.
 452         (vect_model_load_cost): Likewise.
 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         * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
 459         (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
 460         from an unsigned int * to a poly_uint64_pod *.
 461         (calculate_unrolling_factor): New function.
 462         (vect_analyze_slp_instance): Use it.  Track polynomial max_nunits.
 463
 464 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 465             Alan Hayward  <alan.hayward@arm.com>
 466             David Sherwood  <david.sherwood@arm.com>
 467
 468         * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
 469         from an unsigned int to a poly_uint64.
 470         (_loop_vec_info::slp_unrolling_factor): Likewise.
 471         (_loop_vec_info::vectorization_factor): Change from an int
 472         to a poly_uint64.
 473         (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
 474         (vect_get_num_vectors): New function.
 475         (vect_update_max_nunits, vect_vf_for_cost): Likewise.
 476         (vect_get_num_copies): Use vect_get_num_vectors.
 477         (vect_analyze_data_ref_dependences): Change max_vf from an int *
 478         to an unsigned int *.
 479         (vect_analyze_data_refs): Change min_vf from an int * to a
 480         poly_uint64 *.
 481         (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
 482         than an unsigned HOST_WIDE_INT.
 483         * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
 484         (vect_analyze_data_ref_dependence): Change max_vf from an int *
 485         to an unsigned int *.
 486         (vect_analyze_data_ref_dependences): Likewise.
 487         (vect_compute_data_ref_alignment): Handle polynomial vf.
 488         (vect_enhance_data_refs_alignment): Likewise.
 489         (vect_prune_runtime_alias_test_list): Likewise.
 490         (vect_shift_permute_load_chain): Likewise.
 491         (vect_supportable_dr_alignment): Likewise.
 492         (dependence_distance_ge_vf): Take the vectorization factor as a
 493         poly_uint64 rather than an unsigned HOST_WIDE_INT.
 494         (vect_analyze_data_refs): Change min_vf from an int * to a
 495         poly_uint64 *.
 496         * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
 497         vfm1 as a poly_uint64 rather than an int.  Make the same change
 498         for the returned bound_scalar.
 499         (vect_gen_vector_loop_niters): Handle polynomial vf.
 500         (vect_do_peeling): Likewise.  Update call to
 501         vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
 502         (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
 503         be constant.
 504         * tree-vect-loop.c (vect_determine_vectorization_factor)
 505         (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
 506         (vect_get_known_peeling_cost): Likewise.
 507         (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
 508         (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
 509         (vect_transform_loop): Likewise.  Use the lowest possible VF when
 510         updating the upper bounds of the loop.
 511         (vect_min_worthwhile_factor): Make static.  Return an unsigned int
 512         rather than an int.
 513         * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
 514         polynomial unroll factors.
 515         (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
 516         (vect_make_slp_decision): Likewise.
 517         (vect_supported_load_permutation_p): Likewise, and polynomial
 518         vf too.
 519         (vect_analyze_slp_cost): Handle polynomial vf.
 520         (vect_slp_analyze_node_operations): Likewise.
 521         (vect_slp_analyze_bb_1): Likewise.
 522         (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
 523         than an unsigned HOST_WIDE_INT.
 524         * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
 525         (vectorizable_load): Handle polynomial vf.
 526         * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
 527         a poly_uint64.
 528         (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
 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         * match.pd: Handle bit operations involving three constants
 535         and try to fold one pair.
 536
 537 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
 538
 539         * tree-vect-loop-manip.c: Include gimple-fold.h.
 540         (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
 541         niters_maybe_zero parameters.  Handle other cases besides a step of 1.
 542         (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
 543         Add a path that uses a step of VF instead of 1, but disable it
 544         for now.
 545         (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
 546         and niters_no_overflow parameters.  Update calls to
 547         slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
 548         Create a new SSA name if the latter choses to use a ste other
 549         than zero, and return it via niters_vector_mult_vf_var.
 550         * tree-vect-loop.c (vect_transform_loop): Update calls to
 551         vect_do_peeling, vect_gen_vector_loop_niters and
 552         slpeel_make_loop_iterate_ntimes.
 553         * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
 554         (vect_gen_vector_loop_niters): Update declarations after above changes.
 555
 556 2018-01-02  Michael Meissner  <meissner@linux.vnet.ibm.com>
 557
 558         * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
 559         128-bit round to integer instructions.
 560         (ceil<mode>2): Likewise.
 561         (btrunc<mode>2): Likewise.
 562         (round<mode>2): Likewise.
 563
 564 2018-01-02  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
 565
 566         * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
 567         unaligned VSX load/store on P8/P9.
 568         (expand_block_clear): Allow the use of unaligned VSX
 569         load/store on P8/P9.
 570
 571 2018-01-02  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
 572
 573         * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
 574         New function.
 575         (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
 576         swap associated with both a load and a store.
 577
 578 2018-01-02  Andrew Waterman  <andrew@sifive.com>
 579
 580         * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
 581         * config/riscv/riscv.md (clear_cache): Use it.
 582
 583 2018-01-02  Artyom Skrobov  <tyomitch@gmail.com>
 584
 585         * web.c: Remove out-of-date comment.
 586
 587 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
 588
 589         * expr.c (fixup_args_size_notes): Check that any existing
 590         REG_ARGS_SIZE notes are correct, and don't try to re-add them.
 591         (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
 592         (emit_single_push_insn): ...here.
 593
 594 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
 595
 596         * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
 597         (const_vector_encoded_nelts): New function.
 598         (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
 599         (const_vector_int_elt, const_vector_elt): Declare.
 600         * emit-rtl.c (const_vector_int_elt_1): New function.
 601         (const_vector_elt): Likewise.
 602         * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
 603         of CONST_VECTOR_ELT.
 604
 605 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
 606
 607         * expr.c: Include rtx-vector-builder.h.
 608         (const_vector_mask_from_tree): Use rtx_vector_builder and operate
 609         directly on the tree encoding.
 610         (const_vector_from_tree): Likewise.
 611         * optabs.c: Include rtx-vector-builder.h.
 612         (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
 613         sequence of "u" values.
 614         * vec-perm-indices.c: Include rtx-vector-builder.h.
 615         (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
 616         directly on the vec_perm_indices encoding.
 617
 618 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
 619
 620         * doc/rtl.texi (const_vector): Describe new encoding scheme.
 621         * Makefile.in (OBJS): Add rtx-vector-builder.o.
 622         * rtx-vector-builder.h: New file.
 623         * rtx-vector-builder.c: Likewise.
 624         * rtl.h (rtx_def::u2): Add a const_vector field.
 625         (CONST_VECTOR_NPATTERNS): New macro.
 626         (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
 627         (CONST_VECTOR_DUPLICATE_P): Likewise.
 628         (CONST_VECTOR_STEPPED_P): Likewise.
 629         (CONST_VECTOR_ENCODED_ELT): Likewise.
 630         (const_vec_duplicate_p): Check for a duplicated vector encoding.
 631         (unwrap_const_vec_duplicate): Likewise.
 632         (const_vec_series_p): Check for a non-duplicated vector encoding.
 633         Say that the function only returns true for integer vectors.
 634         * emit-rtl.c: Include rtx-vector-builder.h.
 635         (gen_const_vec_duplicate_1): Delete.
 636         (gen_const_vector): Call gen_const_vec_duplicate instead of
 637         gen_const_vec_duplicate_1.
 638         (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
 639         (gen_const_vec_duplicate): Use rtx_vector_builder.
 640         (gen_const_vec_series): Likewise.
 641         (gen_rtx_CONST_VECTOR): Likewise.
 642         * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
 643         (swap_const_vector_halves): Take an rtx pointer rather than rtx.
 644         Build a new vector rather than modifying a CONST_VECTOR in-place.
 645         (handle_special_swappables): Update call accordingly.
 646         * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
 647         (swap_const_vector_halves): Take an rtx pointer rather than rtx.
 648         Build a new vector rather than modifying a CONST_VECTOR in-place.
 649         (handle_special_swappables): Update call accordingly.
 650
 651 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
 652
 653         * simplify-rtx.c (simplify_const_binary_operation): Use
 654         CONST_VECTOR_ELT instead of XVECEXP.
 655
 656 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
 657
 658         * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
 659         the selector elements to be different from the data elements
 660         if the selector is a VECTOR_CST.
 661         * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
 662         ssizetype for the selector.
 663
 664 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
 665
 666         * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
 667         before testing each element individually.
 668         * tree-vect-generic.c (lower_vec_perm): Likewise.
 669
 670 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
 671
 672         * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
 673         * selftest-run-tests.c (selftest::run_tests): Call it.
 674         * vector-builder.h (vector_builder::operator ==): New function.
 675         (vector_builder::operator !=): Likewise.
 676         * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
 677         (vec_perm_indices::all_from_input_p): New function.
 678         * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
 679         (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
 680         * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
 681         instead of reading the VECTOR_CST directly.  Detect whether both
 682         vector inputs are the same before constructing the vec_perm_indices,
 683         and update the number of inputs argument accordingly.  Use the
 684         utility functions added above.  Only construct sel2 if we need to.
 685
 686 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
 687
 688         * optabs.c (expand_vec_perm_var): Use an explicit encoding for
 689         the broadcast of the low byte.
 690         (expand_mult_highpart): Use an explicit encoding for the permutes.
 691         * optabs-query.c (can_mult_highpart_p): Likewise.
 692         * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
 693         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
 694         (vectorizable_bswap): Likewise.
 695         * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
 696         explicit encoding for the power-of-2 permutes.
 697         (vect_permute_store_chain): Likewise.
 698         (vect_grouped_load_supported): Likewise.
 699         (vect_permute_load_chain): Likewise.
 700
 701 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
 702
 703         * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
 704         * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
 705         * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
 706         * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
 707         * tree-vect-stmts.c (vectorizable_bswap): Likewise.
 708         (vect_gen_perm_mask_any): Likewise.
 709
 710 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
 711
 712         * int-vector-builder.h: New file.
 713         * vec-perm-indices.h: Include int-vector-builder.h.
 714         (vec_perm_indices): Redefine as an int_vector_builder.
 715         (auto_vec_perm_indices): Delete.
 716         (vec_perm_builder): Redefine as a stand-alone class.
 717         (vec_perm_indices::vec_perm_indices): New function.
 718         (vec_perm_indices::clamp): Likewise.
 719         * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
 720         (vec_perm_indices::new_vector): New function.
 721         (vec_perm_indices::new_expanded_vector): Update for new
 722         vec_perm_indices class.
 723         (vec_perm_indices::rotate_inputs): New function.
 724         (vec_perm_indices::all_in_range_p): Operate directly on the
 725         encoded form, without computing elided elements.
 726         (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
 727         encoding.  Update for new vec_perm_indices class.
 728         * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
 729         the given vec_perm_builder.
 730         (expand_vec_perm_var): Update vec_perm_builder constructor.
 731         (expand_mult_highpart): Use vec_perm_builder instead of
 732         auto_vec_perm_indices.
 733         * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
 734         vec_perm_indices instead of auto_vec_perm_indices.  Use a single
 735         or double series encoding as appropriate.
 736         * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
 737         vec_perm_indices instead of auto_vec_perm_indices.
 738         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
 739         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
 740         (vect_permute_store_chain): Likewise.
 741         (vect_grouped_load_supported): Likewise.
 742         (vect_permute_load_chain): Likewise.
 743         (vect_shift_permute_load_chain): Likewise.
 744         * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
 745         (vect_transform_slp_perm_load): Likewise.
 746         (vect_schedule_slp_instance): Likewise.
 747         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
 748         (vectorizable_mask_load_store): Likewise.
 749         (vectorizable_bswap): Likewise.
 750         (vectorizable_store): Likewise.
 751         (vectorizable_load): Likewise.
 752         * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
 753         vec_perm_indices instead of auto_vec_perm_indices.  Use
 754         tree_to_vec_perm_builder to read the vector from a tree.
 755         * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
 756         vec_perm_builder instead of a vec_perm_indices.
 757         (have_whole_vector_shift): Use vec_perm_builder and
 758         vec_perm_indices instead of auto_vec_perm_indices.  Leave the
 759         truncation to calc_vec_perm_mask_for_shift.
 760         (vect_create_epilog_for_reduction): Likewise.
 761         * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
 762         from auto_vec_perm_indices to vec_perm_indices.
 763         (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
 764         instead of changing individual elements.
 765         (aarch64_vectorize_vec_perm_const): Use new_vector to install
 766         the vector in d.perm.
 767         * config/arm/arm.c (expand_vec_perm_d::perm): Change
 768         from auto_vec_perm_indices to vec_perm_indices.
 769         (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
 770         instead of changing individual elements.
 771         (arm_vectorize_vec_perm_const): Use new_vector to install
 772         the vector in d.perm.
 773         * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
 774         Update vec_perm_builder constructor.
 775         (rs6000_expand_interleave): Likewise.
 776         * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
 777         (rs6000_expand_interleave): Likewise.
 778
 779 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
 780
 781         * optabs-query.c (can_vec_perm_var_p): Check whether lowering
 782         to qimode could truncate the indices.
 783         * optabs.c (expand_vec_perm_var): Likewise.
 784
 785 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
 786
 787         * Makefile.in (OBJS): Add vec-perm-indices.o.
 788         * vec-perm-indices.h: New file.
 789         * vec-perm-indices.c: Likewise.
 790         * target.h (vec_perm_indices): Replace with a forward class
 791         declaration.
 792         (auto_vec_perm_indices): Move to vec-perm-indices.h.
 793         * optabs.h: Include vec-perm-indices.h.
 794         (expand_vec_perm): Delete.
 795         (selector_fits_mode_p, expand_vec_perm_var): Declare.
 796         (expand_vec_perm_const): Declare.
 797         * target.def (vec_perm_const_ok): Replace with...
 798         (vec_perm_const): ...this new hook.
 799         * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
 800         (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
 801         * doc/tm.texi: Regenerate.
 802         * optabs.def (vec_perm_const): Delete.
 803         * doc/md.texi (vec_perm_const): Likewise.
 804         (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
 805         * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
 806         expand_vec_perm for constant permutation vectors.  Assert that
 807         the mode of variable permutation vectors is the integer equivalent
 808         of the mode that is being permuted.
 809         * optabs-query.h (selector_fits_mode_p): Declare.
 810         * optabs-query.c: Include vec-perm-indices.h.
 811         (selector_fits_mode_p): New function.
 812         (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
 813         is defined, instead of checking whether the vec_perm_const_optab
 814         exists.  Use targetm.vectorize.vec_perm_const instead of
 815         targetm.vectorize.vec_perm_const_ok.  Check whether the indices
 816         fit in the vector mode before using a variable permute.
 817         * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
 818         vec_perm_indices instead of an rtx.
 819         (expand_vec_perm): Replace with...
 820         (expand_vec_perm_const): ...this new function.  Take the selector
 821         as a vec_perm_indices rather than an rtx.  Also take the mode of
 822         the selector.  Update call to shift_amt_for_vec_perm_mask.
 823         Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
 824         Use vec_perm_indices::new_expanded_vector to expand the original
 825         selector into bytes.  Check whether the indices fit in the vector
 826         mode before using a variable permute.
 827         (expand_vec_perm_var): Make global.
 828         (expand_mult_highpart): Use expand_vec_perm_const.
 829         * fold-const.c: Includes vec-perm-indices.h.
 830         * tree-ssa-forwprop.c: Likewise.
 831         * tree-vect-data-refs.c: Likewise.
 832         * tree-vect-generic.c: Likewise.
 833         * tree-vect-loop.c: Likewise.
 834         * tree-vect-slp.c: Likewise.
 835         * tree-vect-stmts.c: Likewise.
 836         * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
 837         Delete.
 838         * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
 839         * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
 840         (aarch64_vectorize_vec_perm_const_ok): Fuse into...
 841         (aarch64_vectorize_vec_perm_const): ...this new function.
 842         (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
 843         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
 844         * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
 845         * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
 846         * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
 847         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
 848         (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
 849         into...
 850         (arm_vectorize_vec_perm_const): ...this new function.  Explicitly
 851         check for NEON modes.
 852         * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
 853         * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
 854         * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
 855         (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
 856         into...
 857         (ix86_vectorize_vec_perm_const): ...this new function.  Incorporate
 858         the old VEC_PERM_CONST conditions.
 859         * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
 860         * config/ia64/vect.md (vec_perm_const<mode>): Delete.
 861         * config/ia64/ia64.c (ia64_expand_vec_perm_const)
 862         (ia64_vectorize_vec_perm_const_ok): Merge into...
 863         (ia64_vectorize_vec_perm_const): ...this new function.
 864         * config/mips/loongson.md (vec_perm_const<mode>): Delete.
 865         * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
 866         * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
 867         * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
 868         * config/mips/mips.c (mips_expand_vec_perm_const)
 869         (mips_vectorize_vec_perm_const_ok): Merge into...
 870         (mips_vectorize_vec_perm_const): ...this new function.
 871         * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
 872         * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
 873         * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
 874         * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
 875         * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
 876         (rs6000_expand_vec_perm_const): Delete.
 877         * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
 878         Delete.
 879         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
 880         (altivec_expand_vec_perm_const_le): Take each operand individually.
 881         Operate on constant selectors rather than rtxes.
 882         (altivec_expand_vec_perm_const): Likewise.  Update call to
 883         altivec_expand_vec_perm_const_le.
 884         (rs6000_expand_vec_perm_const): Delete.
 885         (rs6000_vectorize_vec_perm_const_ok): Delete.
 886         (rs6000_vectorize_vec_perm_const): New function.
 887         (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
 888         an element count and rtx array.
 889         (rs6000_expand_extract_even): Update call accordingly.
 890         (rs6000_expand_interleave): Likewise.
 891         * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
 892         * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
 893         * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
 894         * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
 895         (rs6000_expand_vec_perm_const): Delete.
 896         * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
 897         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
 898         (altivec_expand_vec_perm_const_le): Take each operand individually.
 899         Operate on constant selectors rather than rtxes.
 900         (altivec_expand_vec_perm_const): Likewise.  Update call to
 901         altivec_expand_vec_perm_const_le.
 902         (rs6000_expand_vec_perm_const): Delete.
 903         (rs6000_vectorize_vec_perm_const_ok): Delete.
 904         (rs6000_vectorize_vec_perm_const): New function.  Remove stray
 905         reference to the SPE evmerge intructions.
 906         (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
 907         an element count and rtx array.
 908         (rs6000_expand_extract_even): Update call accordingly.
 909         (rs6000_expand_interleave): Likewise.
 910         * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
 911         * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
 912         new function.
 913         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
 914
 915 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
 916
 917         * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
 918         vector mode and that that mode matches the mode of the data
 919         being permuted.
 920         (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
 921         out into expand_vec_perm_var.  Do all CONST_VECTOR handling here,
 922         directly using expand_vec_perm_1 when forcing selectors into
 923         registers.
 924         (expand_vec_perm_var): New function, split out from expand_vec_perm.
 925
 926 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
 927
 928         * optabs-query.h (can_vec_perm_p): Delete.
 929         (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
 930         * optabs-query.c (can_vec_perm_p): Split into...
 931         (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
 932         (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
 933         particular selector is valid.
 934         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
 935         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
 936         (vect_grouped_load_supported): Likewise.
 937         (vect_shift_permute_load_chain): Likewise.
 938         * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
 939         (vect_transform_slp_perm_load): Likewise.
 940         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
 941         (vectorizable_bswap): Likewise.
 942         (vect_gen_perm_mask_checked): Likewise.
 943         * fold-const.c (fold_ternary_loc): Likewise.  Don't take
 944         implementations of variable permutation vectors into account
 945         when deciding which selector to use.
 946         * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
 947         vec_perm_const_optab is supported; instead use can_vec_perm_const_p
 948         with a false third argument.
 949         * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
 950         to test whether the constant selector is valid and can_vec_perm_var_p
 951         to test whether a variable selector is valid.
 952
 953 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
 954
 955         * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
 956         * optabs-query.c (can_vec_perm_p): Likewise.
 957         * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
 958         instead of vec_perm_indices.
 959         * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
 960         (vect_gen_perm_mask_checked): Likewise,
 961         * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
 962         (vect_gen_perm_mask_checked): Likewise,
 963
 964 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
 965
 966         * optabs-query.h (qimode_for_vec_perm): Declare.
 967         * optabs-query.c (can_vec_perm_p): Split out qimode search to...
 968         (qimode_for_vec_perm): ...this new function.
 969         * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
 970
 971 2018-01-02  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
 972
 973         * rtlanal.c (canonicalize_condition): Return 0 if final rtx
 974         does not have a conditional at the top.
 975
 976 2018-01-02  Richard Biener  <rguenther@suse.de>
 977
 978         * ipa-inline.c (big_speedup_p): Fix expression.
 979
 980 2018-01-02  Jan Hubicka  <hubicka@ucw.cz>
 981
 982         PR target/81616
 983         * config/i386/x86-tune-costs.h: Increase cost of integer load costs
 984         for generic 4->6.
 985
 986 2018-01-02  Jan Hubicka  <hubicka@ucw.cz>
 987
 988         PR target/81616
 989         Generic tuning.
 990         * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
 991         cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
 992         and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
 993         cond_taken_branch_cost 3->4.
 994
 995 2018-01-01  Jakub Jelinek  <jakub@redhat.com>
 996
 997         PR tree-optimization/83581
 998         * tree-loop-distribution.c (pass_loop_distribution::execute): Return
 999         TODO_cleanup_cfg if any changes have been made.
1000
1001         PR middle-end/83608
1002         * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
1003         convert_modes if target mode has the right side, but different mode
1004         class.
1005
1006         PR middle-end/83609
1007         * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
1008         last argument when extracting from CONCAT.  If either from_real or
1009         from_imag is NULL, use expansion through memory.  If result is not
1010         a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
1011         the parts directly to inner mode, if even that fails, use expansion
1012         through memory.
1013
1014         PR middle-end/83623
1015         * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
1016         check for bswap in mode rather than HImode and use that in expand_unop
1017         too.
1018 \f
1019 Copyright (C) 2018 Free Software Foundation, Inc.
1020
1021 Copying and distribution of this file, with or without modification,
1022 are permitted in any medium without royalty provided the copyright
1023 notice and this notice are preserved.