gcc/ChangeLog

   1 2018-04-30  Jason Merrill  <jason@redhat.com>
   2
   3         PR c++/61982 - dead stores to destroyed objects.
   4         * gimplify.c (gimplify_modify_expr): Simplify complex lvalue on LHS
   5         of clobber.
   6
   7 2018-04-30  Jason Merrill  <jason@redhat.com>
   8
   9         * tree.c (build_clobber): New.
  10         * tree.h: Declare it.
  11         * gimplify.c (gimplify_bind_expr, gimplify_target_expr): Use it.
  12
  13 2018-04-30  David Malcolm  <dmalcolm@redhat.com>
  14
  15         * diagnostic-show-locus.c (layout::layout): Update for
  16         location_get_source_line returning a char_span.
  17         (struct char_span): Move to input.h.
  18         (struct correction): Update for fields in char_span becoming
  19         private.
  20         (struct source_line): Update for location_get_source_line
  21         returning a char_span.
  22         (layout::print_line): Likewise.
  23         * edit-context.c (edited_file::print_content): Likewise.
  24         (edited_file::print_diff_hunk): Likewise.
  25         (edited_file::print_run_of_changed_lines): Likewise.
  26         (edited_file::get_num_lines): Likewise.
  27         (edited_line::edited_line): Likewise.
  28         * final.c (asm_show_source): Likewise.
  29         * input.c (location_get_source_line): Convert return type
  30         from const char * to char_span, losing the final "line_len"
  31         param.
  32         (dump_location_info): Update for the above.
  33         (get_substring_ranges_for_loc): Likewise.  Use a char_span
  34         when handling the literal within the line.
  35         (test_reading_source_line): Update for location_get_source_line
  36         returning a char_span.
  37         * input.h (class char_span): Move here from
  38         diagnostic-show-locus.c, converting from a struct to a class.
  39         Make data members private.
  40         (char_span::operator bool): New.
  41         (char_span::length): New.
  42         (char_span::get_buffer): New.
  43         (char_span::operator[]): New.
  44         (char_span::subspan): Make const.
  45         (char_span::xstrdup): New.
  46         (location_get_source_line): Convert return type from const char *
  47         to char_span, losing the final "line_size" param.
  48
  49 2018-04-30  Jan Hubicka  <jh@suse.cz>
  50
  51         * lto-wrapper.c (ltrans_priorities): New static var.
  52         (cmp_priority): New.
  53         (run_gcc): Read priorities and if doing parallel build order
  54         the Makefile by them.
  55
  56 2018-04-30  David Malcolm  <dmalcolm@redhat.com>
  57
  58         * input.h (builtins_location_check): Convert to a STATIC_ASSERT.
  59
  60 2018-04-30  Richard Biener  <rguenther@suse.de>
  61
  62         * tree-cfg.c (verify_address): Remove base argument, add
  63         flag whether to check TREE_ADDRESSABLE and do that.
  64         (verify_expr): Remove.
  65         (verify_types_in_gimple_reference): Add pieces from verify_expr.
  66         (verify_gimple_assign_single): Likewise.
  67         (verify_gimple_switch): Likewise.
  68         (verify_expr_location_1): Dereference tp once.  Add (disabled)
  69         piece from verify_expr.
  70         (verify_gimple_in_cfg): Do not call verify_expr on all ops.
  71
  72 2018-04-30  Claudiu Zissulescu  <claziss@synopsys.com>
  73
  74         * config/arc/linux.h (CLEAR_INSN_CACHE): Define.
  75
  76 2018-04-30  Claudiu Zissulescu  <claziss@synopsys.com>
  77
  78         * config/arc/arc-protos.h (prepare_extend_operands): Remove.
  79         (small_data_pattern): Likewise.
  80         (arc_rewrite_small_data): Likewise.
  81         * config/arc/arc.c (LEGITIMATE_SMALL_DATA_OFFSET_P): Remove.
  82         (LEGITIMATE_SMALL_DATA_ADDRESS_P): Likewise.
  83         (get_symbol_alignment): New function.
  84         (legitimate_small_data_address_p): Likewise.
  85         (legitimate_scaled_address): Update, call
  86         legitimate_small_data_address_p.
  87         (output_sdata): New static variable.
  88         (arc_print_operand): Update how we handle small data operands.
  89         (arc_print_operand_address): Likewise.
  90         (arc_legitimate_address_p): Update, use
  91         legitimate_small_data_address_p.
  92         (arc_rewrite_small_data_p): Remove.
  93         (arc_rewrite_small_data_1): Likewise.
  94         (arc_rewrite_small_data): Likewise.
  95         (small_data_pattern): Likewise.
  96         (compact_sda_memory_operand): Update to use
  97         legitimate_small_data_address_p and get_symbol_alignment.
  98         (prepare_move_operands): Don't rewite sdata pattern.
  99         (prepare_extend_operands): Remove.
 100         * config/arc/arc.md (zero_extendqihi2): Don't rewrite sdata
 101         pattern.
 102         (zero_extendqisi2): Likewise.
 103         (zero_extendhisi2): Likewise.
 104         (extendqihi2): Likewise.
 105         (extendqisi2): Likewise.
 106         (extendhisi2): Likewise.
 107         (addsi3): Likewise.
 108         (subsi3): Likewise.
 109         (andsi3): Likewise.
 110         * config/arc/constraints.md (Usd): Change it to memory constraint.
 111
 112 2018-04-30  Claudiu Zissulescu  <claziss@synopsys.com>
 113
 114         * config/arc/arc.c (arc_split_move): Allow signed 6-bit constants
 115         as source of std instructions.
 116         * config/arc/arc.md (movsi_insn): Update pattern predicate to
 117         allow 6-bit constants as source for store instructions.
 118         (movdi_insn): Update instruction pattern to allow 6-bit constants
 119         as source for store instructions.
 120
 121 2018-04-30  Jonathan Wakely  <jwakely@redhat.com>
 122
 123         * doc/invoke.texi (-fdebug-types-section): Fix grammar.
 124
 125 2018-04-30  Nathan Sidwell  <nathan@acm.org>
 126             Sandra Loosemore <sandra@codesourcery.com>
 127
 128         * dumpfile.c (dump_open): Allow '-' for stdout.
 129         * doc/invoke.texi (Developer Options): Document dump filename
 130         determination early.  Document stdin/stdout selection.
 131
 132 2018-04-30 Andrew Sadek  <andrew.sadek.se@gmail.com>
 133
 134         Microblaze Target: PIC data text relative
 135
 136         * config/microblaze/microblaze.opt: add new option -mpic-data-text-rel.
 137         * config/microblaze/microblaze-protos.h (microblaze_constant_address_p):
 138         Add declaration.
 139         * gcc/config/microblaze/microblaze.h (microblaze_constant_address_p):
 140         CONSTANT_ADDRESS_P definition to microblaze_constant_address_p.
 141         * config/microblaze/microblaze.c (TARGET_PIC_DATA_TEXT_REL):
 142         New addressing mode for data-text relative position indepenedent code.
 143         (microblaze_classify_unspec): add 'UNSPEC_TEXT' case ->
 144         'ADDRESS_SYMBOLIC_TXT_REL'.
 145         (microblaze_classify_address): Add handling for UNSPEC + CONST_INT.
 146         (microblaze_legitimate_pic_operand): Exclude function calls from
 147         pic operands in case of TARGET_PIC_DATA_TEXT_REL option.
 148         (microblaze_legitimize_address): Generate 'UNSPEC_TEXT' for all possible
 149         addresses cases.
 150         (microblaze_address_insns): Add 'ADDRESS_SYMBOLIC_TXT_REL' case.
 151         (print_operand): Add 'ADDRESS_SYMBOLIC_TXT_REL' case.
 152         (print_operand_address): Add 'ADDRESS_SYMBOLIC_TXT_REL' case + handling
 153         for 'address + offset'.
 154         (microblaze_expand_prologue): Add new function prologue call for
 155         'r20' assignation.
 156         (microblaze_asm_generate_pic_addr_dif_vec): Override new target hook
 157         'TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC' to disable address diff vector
 158         table in case of TARGET_PIC_DATA_TEXT_REL.
 159         (expand_pic_symbol_ref): Add handling for 'UNSPEC_TEXT'.
 160         * gcc/config/microblaze/microblaze.md (TARGET_PIC_DATA_TEXT_REL):
 161         Add new macros 'UNSPEC_TEXT',
 162         'UNSPEC_SET_TEXT' + add rule for setting r20 in function prologue
 163         + exclude function calls from 'UNSPEC_PLT' in case of data text
 164         relative mode.
 165         * doc/tm.texi.in (TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC): Add
 166         new target hook for generating address diff vector tables in case of
 167         flag_pic.
 168         * doc/tm.texi : Regenerate.
 169         * stmt.c (TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC): Append new condition
 170         'targetm.asm_out.generate_pic_addr_diff_vec' to flag_pic in case
 171         of addr diff vector generation.
 172         * target.def (TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC): Add
 173         target hook definition.
 174         * targhooks.h, gcc/targhooks.c (TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC):
 175         Add default function for generate_pic_addr_diff_vec -> flag_pic.
 176         * doc/invoke.texi (Add new pic option): Add new microblaze pic
 177         option for data text relative.
 178
 179 2018-04-30  Richard Biener  <rguenther@suse.de>
 180
 181         * tree-chrec.h (evolution_function_is_constant_p): Remove
 182         redundant check.
 183         * tree-cfg.c (tree_node_can_be_shared): Re-order checks.
 184
 185 2018-04-30  Richard Biener  <rguenther@suse.de>
 186
 187         PR bootstrap/85571
 188         * dwarf2out.c (gen_producer_string): Ignore -fchecking[=].
 189
 190 2018-04-30  Richard Biener  <rguenther@suse.de>
 191
 192         PR tree-optimization/28364
 193         PR tree-optimization/85275
 194         * tree-ssa-loop-ch.c (ch_base::copy_headers): Stop after
 195         copying first exit test.
 196
 197 2018-04-28  Mark Wielaard  <mark@klomp.org>
 198
 199         * dwarf2out.c (dwarf2out_finish): Add .debug_addr table header for
 200         dwarf_version >= 5.
 201         (dwarf_AT): Handle DW_AT_addr_base.
 202         (add_top_level_skeleton_die_attrs): Use dwarf_AT for DW_AT_addr_base.
 203
 204 2018-04-28  Uros Bizjak  <ubizjak@gmail.com>
 205
 206         PR target/84431
 207         * config/i386/i386.md (*ashl<dwi>3_doubleword_mask): New pattern.
 208         (*ashl<dwi>3_doubleword_mask_1): Ditto.
 209         (*<shift_insn><dwi>3_doubleword_mask): Ditto.
 210         (*<shift_insn><dwi>3_doubleword_mask_1): Ditto.
 211
 212 2018-04-28  Richard Biener  <rguenther@suse.de>
 213
 214         * tree-cfg.c (verify_gimple_phi): Take a gphi * argument.
 215         (verify_gimple_in_cfg): Rename visited_stmts to visited_throwing_stmts
 216         to reflect use.  Only add interesting stmts.
 217
 218 2018-04-27  Martin Jambor  <mjambor@suse.cz>
 219
 220         PR ipa/85549
 221         * ipa-cp.c (find_aggregate_values_for_callers_subset): Make sure
 222         the jump function allows for passing through aggregate values.
 223
 224 2018-04-27  David Malcolm  <dmalcolm@redhat.com>
 225
 226         * input.h (in_system_header_at): Convert from macro to inline
 227         function.
 228         (from_macro_expansion_at): Likewise.
 229         (from_macro_definition_at): Likewise.
 230
 231 2018-04-27  Jeff Law  <law@redhat.com>
 232
 233         * config.gcc: Mark tile* targets as deprecated/obsolete.
 234
 235 2018-04-27  Richard Biener  <rguenther@suse.de>
 236
 237         * config/aarch64/aarch64.c: Simplify ap.__stack advance and
 238         fix for ILP32.
 239
 240 2018-04-27  Richard Biener  <rguenther@suse.de>
 241
 242         * tree-cfg.c (verify_expr): Make dead code hit gcc_unreachable.
 243
 244 2018-04-27  Uros Bizjak  <ubizjak@gmail.com>
 245
 246         * config/i386/i386.md (*movti_internal): Substitute Ye constraint
 247         with Yd constraint. Set "preferred_for_speed" attribute from
 248         TARGET_INTER_UNIT_MOVES_{FROM,TO}_VEC for alternatives
 249         with Yd constraint.
 250         (*movdi_internal): Ditto.
 251         (movti_interunit splitters): Remove
 252         TARGET_INTER_UNIT_MOVES_{FROM,TO}_VEC from insn condition.
 253         (movdi_interunit splitters): Ditto.
 254         * config/i386/constraints.md (Ye): Remove.
 255         (Yd): Do not depend on TARGET_INTER_UNIT_MOVES_{FROM,TO}_VEC.
 256
 257 2018-04-27  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
 258
 259         PR target/85512
 260         * config/aarch64/constraints.md (Usg): Limit to 31.
 261         (Usj): Limit to 63.
 262
 263 2018-04-27  Jakub Jelinek  <jakub@redhat.com>
 264
 265         PR tree-optimization/85529
 266         * tree-ssa-reassoc.c (optimize_range_tests_var_bound): Add FIRST_BB
 267         argument.  Don't call get_nonzero_bits if opcode is ERROR_MARK_NODE,
 268         rhs2 def stmt's bb is dominated by first_bb and it isn't an obvious
 269         zero extension or masking of the MSB bit.
 270         (optimize_range_tests): Add FIRST_BB argument, pass it through
 271         to optimize_range_tests_var_bound.
 272         (maybe_optimize_range_tests, reassociate_bb): Adjust
 273         optimize_range_tests callers.
 274
 275 2018-04-26  Richard Biener  <rguenther@suse.de>
 276             Jakub Jelinek  <jakub@redhat.com>
 277
 278         * cgraph.h (symbol_table): Just declare debug method here.
 279         * symtab.c (symbol_table::debug): Define.
 280
 281 2018-04-26  Eric Botcazou  <ebotcazou@adacore.com>
 282
 283         * loop-invariant.c (may_assign_reg_p): Return false for frame pointer.
 284
 285 2018-04-26  Uros Bizjak  <ubizjak@gmail.com>
 286
 287         * config/i386/i386.md ("isa" attribute): Add x64_sse2.
 288         ("enabled" attribute): Handle x64_sse2 "isa" attribute.
 289         (*movdi_internal): Substitute Yi and Yj constraint with x
 290         and Ym and Yn constraint with y constraint.  Update "isa"
 291         attribute and set "preferred_for_speed" attribute from
 292         TARGET_INTER_UNIT_MOVES_{FROM,TO}_VEC for updated alternatives.
 293         (*movsi_internal): Ditto.
 294         (*movdf_internal): Ditto.
 295         (*movsf_internal): Ditto.
 296         (*zero_extendsidi2): Ditto.
 297         * config/i386/sse.md (vec_set<mode>_0): Ditto.
 298         (sse2_loadld): Ditto.
 299         (*vec_extract<ssevecmodelower>_0): Ditto.
 300         (*vec_extractv4si_0_zext_sse4): Ditto.
 301         (vec_concatv2di): Ditto.
 302         (*vec_dup<mode>): Ditto.
 303         * config/i386/mmx.md (*mov<mode>_internal): Ditto.
 304         * config/i386/constraints.md (Yi): Remove.
 305         (Yj): Remove.
 306         (Ym): Remove.
 307         (Yn): Remove.
 308
 309 2018-04-26  Nathan Sidwell  <nathan@acm.org>
 310
 311         * dumpfile.c (dump_open): New.
 312         (dump_open_alternate_stream, dump_start, dump_begin): Call it.
 313         (dump_finish): Detect stdio/stderr by value not name.
 314
 315 2018-04-26  Jonathan Wakely  <jwakely@redhat.com>
 316
 317         * doc/invoke.texi (-Wreturn-type): Document default status for C++.
 318
 319 2018-04-26  Tom de Vries  <tom@codesourcery.com>
 320
 321         PR target/84952
 322         * config/nvptx/nvptx.c (verify_neutering_jumps)
 323         (verify_neutering_labels): New function
 324         (nvptx_single): Use verify_neutering_jumps and verify_neutering_labels.
 325
 326 2018-04-26  Tom de Vries  <tom@codesourcery.com>
 327
 328         PR target/84025
 329         * config/nvptx/nvptx.c (needs_neutering_p): New function.
 330         (nvptx_single): Use needs_neutering_p to skip over insns that do not
 331         need neutering.
 332
 333 2018-04-26  Richard Biener <rguenther@suse.de>
 334             Tom de Vries  <tom@codesourcery.com>
 335
 336         PR lto/85422
 337         * lto-streamer-out.c (output_function): Fixup loops if required to match
 338         discovery done in the reader.
 339
 340 2018-04-26  Richard Biener  <rguenther@suse.de>
 341
 342         PR tree-optimization/85116
 343         * tree-ssa-loop-ch.c (do_while_loop_p): A do-while loop should
 344         have a loop exit from the single latch predecessor.  Remove
 345         case of header with just condition.
 346         (ch_base::copy_headers): Exclude infinite loops from any
 347         processing.
 348         (pass_ch::execute): Record exits.
 349
 350 2018-04-26  Richard Biener  <rguenther@suse.de>
 351
 352         * tree-vect-data-refs.c (vect_get_data_access_cost): Get
 353         prologue cost vector and pass it to vect_get_load_cost.
 354         (vect_get_peeling_costs_all_drs): Likewise.
 355         (vect_peeling_hash_get_lowest_cost): Likewise.
 356         (vect_enhance_data_refs_alignment): Likewise.
 357
 358 2018-04-26  Richard Biener  <rguenther@suse.de>
 359
 360         PR middle-end/85450
 361         * tree-cfg.c (verify_gimple_assign_unary): Restore proper
 362         checking of integer<->pointer conversions.
 363         * omp-expand.c (expand_omp_for_static_nochunk): Avoid
 364         sign-/zero-extending pointer types.
 365         (expand_omp_for_static_chunk): Likewise.
 366
 367 2018-03-22  Hans-Peter Nilsson  <hp@axis.com>
 368             Jean Lee  <xiaoyur347@gmail.com>
 369
 370         * config/mips/mips.c (mips_asan_shadow_offset): New function.
 371         (TARGET_ASAN_SHADOW_OFFSET): Define.
 372         * config/mips/mips.h (FRAME_GROWS_DOWNWARD): Augment to also be
 373         true for -fsanitize=address.
 374
 375 2018-04-25  Mark Wielaard  <mark@klomp.org>
 376
 377         * dwarf2out.c (file_info_cmp): Sort longer dir prefixes before
 378         shorter ones.
 379
 380 2018-04-25  Jakub Jelinek  <jakub@redhat.com>
 381
 382         * config/i386/i386.md (*x86_mov<mode>cc_0_m1): Use type "alu1" rather
 383         than "alu", remove explicit "memory" and "imm_disp" attributes.
 384         (*x86_mov<mode>cc_0_m1_se, *x86_mov<mode>cc_0_m1_neg): Likewise.
 385
 386         PR middle-end/85414
 387         * simplify-rtx.c (simplify_unary_operation_1) <case SIGN_EXTEND,
 388         case ZERO_EXTEND>: Pass SUBREG_REG (op) rather than op to
 389         gen_lowpart_no_emit.
 390
 391 2018-04-25  Sebastian Peryt  <sebastian.peryt@intel.com>
 392
 393         PR target/85473
 394         * config/i386/i386.c (ix86_expand_builtin): Change memory
 395         operand to XI, extend p0 to Pmode.
 396         * config/i386/i386.md: Change unspec volatile and operand
 397         1 mode to XI, change operand 0 mode to P.
 398
 399 2018-04-25  Chung-Ju Wu  <jasonwucj@gmail.com>
 400
 401         * config/nds32/nds32-predicates.c (nds32_can_use_bclr_p): Mask with
 402         GET_MODE_MASK before any checking.
 403         (nds32_can_use_bset_p): Likewise.
 404         (nds32_can_use_btgl_p): Likewise.
 405
 406 2018-04-25  Chung-Ju Wu  <jasonwucj@gmail.com>
 407
 408         * config/nds32/nds32-doubleword.md: New define_split pattern for
 409         illegal register number.
 410
 411 2018-04-25  Chung-Ju Wu  <jasonwucj@gmail.com>
 412
 413         * config/nds32/nds32.c (nds32_print_operand): Set op_value ealier.
 414
 415 2018-04-25  Chung-Ju Wu  <jasonwucj@gmail.com>
 416
 417         * config/nds32/nds32.h (ASM_APP_ON): Add missing newline character.
 418
 419 2018-04-25  Richard Biener  <rguenther@suse.de>
 420
 421         * lto-streamer.h (LTO_major_version): Bump to 8.
 422
 423 2018-04-25  Jakub Jelinek  <jakub@redhat.com>
 424
 425         * BASE-VER: Set to 9.0.0.
 426
 427 2018-04-24  Segher Boessenkool  <segher@kernel.crashing.org>
 428
 429         * config/rs6000/rs6000.c (init_float128_ieee): Fix spelling mistakes
 430         in __abskf2 and __powikf2.
 431
 432 2018-04-24  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
 433
 434         PR target/85512
 435         * config/aarch64/constraints.md (Usg, Usj): New constraints.
 436         * config/aarch64/iterators.md (cmode_simd): New mode attribute.
 437         * config/aarch64/aarch64.md (*aarch64_ashr_sisd_or_int_<mode>3):
 438         Use the above on operand 2.  Reindent.
 439         (*aarch64_lshr_sisd_or_int_<mode>3): Likewise.
 440
 441 2018-04-24  H.J. Lu  <hongjiu.lu@intel.com>
 442
 443         PR target/85485
 444         * common/config/i386/i386-common.c (ix86_handle_option): Don't
 445         handle OPT_mcet.
 446         * config/i386/i386.opt (mcet): Removed.
 447         * doc/install.texi: Remove -mcet documentation.
 448         * doc/invoke.texi: Likewise.
 449
 450 2018-04-24  H.J. Lu  <hongjiu.lu@intel.com>
 451
 452         PR target/85485
 453         * doc/install.texi: Remove -mcet from bootstrap-cet.
 454
 455 2018-04-24  Jakub Jelinek  <jakub@redhat.com>
 456
 457         PR target/85511
 458         * config/i386/i386.c (ix86_init_mmx_sse_builtins): Don't define
 459         __builtin_ia32_readeflags_u32 and __builtin_ia32_writeeflags_u32
 460         if TARGET_64BIT.
 461
 462         PR target/85503
 463         * config/rs6000/rs6000-p8swap.c (const_load_sequence_p): Punt if
 464         const_vector is not CONST_VECTOR or SYMBOL_REF for a constant pool
 465         containing a CONST_VECTOR.
 466
 467 2018-04-24  Cesar Philippidis  <cesar@codesourcery.com>
 468
 469         * doc/install.texi: Update newlib dependency for nvptx.
 470
 471 2018-04-24  Jakub Jelinek  <jakub@redhat.com>
 472
 473         PR target/85508
 474         * config/i386/i386.c (ix86_expand_vector_init_one_var): Use UINTVAL
 475         instead of INTVAL when shifting x left.
 476
 477 2018-04-24  Andreas Krebbel  <krebbel@linux.ibm.com>
 478
 479         PR tree-optimization/85478
 480         * tree-vect-loop.c (vect_analyze_loop_2): Do not call
 481         vect_grouped_store_supported for single element vectors.
 482
 483 2018-04-24  Richard Biener  <rguenther@suse.de>
 484
 485         PR target/85491
 486         * config/i386/i386.c (ix86_add_stmt_cost): Restrict strided
 487         load cost increase to the case of non-constant step.
 488
 489 2018-04-24  Jakub Jelinek  <jakub@redhat.com>
 490
 491         PR target/84828
 492         * reg-stack.c (move_for_stack_reg): Don't fail assertion about dead
 493         destination if any_malformed_asm.
 494
 495 2018-04-23  Eric Botcazou  <ebotcazou@adacore.com>
 496
 497         PR middle-end/85496
 498         * expr.c (store_field): In the bitfield case, if the value comes from
 499         a function call and is returned in registers by means of a PARALLEL,
 500         do not change the mode of the temporary unless BLKmode and VOIDmode.
 501
 502 2018-04-23  Andrey Belevantsev  <abel@ispras.ru>
 503
 504         PR rtl-optimization/85423
 505         * sel-sched-ir.c (has_dependence_note_mem_dep): Only discard
 506         dependencies to debug insns when the previous insn is non-debug.
 507
 508 2018-04-23  Claudiu Zissulescu  <claziss@synopsys.com>
 509
 510         * config/arc/arc.md ("vunspec"): Delete it, unify all the unspec
 511         enums into a single definition.
 512         (fls): Fix predicates and printing.
 513         (seti): Likewise.
 514
 515 2018-04-23  Claudiu Zissulescu  <claziss@synopsys.com>
 516
 517         * config/arc/arc-protos.h (check_if_valid_sleep_operand): Remove.
 518         * config/arc/arc.c (arc_expand_builtin): Sleep accepts registers
 519         and short u6 immediate.
 520         (check_if_valid_sleep_operand): Remove.
 521         * config/arc/arc.md (Sleep): Accepts registers and u6 immediates.
 522
 523 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
 524
 525         * config/nds32/nds32.c (nds32_compute_stack_frame): Consider
 526         flag_always_save_lp condition.
 527         * config/nds32/nds32.opt (malways-save-lp): New option.
 528
 529 2018-04-22  Shiva Chen  <shiva0217@gmail.com>
 530
 531         * config/nds32/nds32-protos.h (nds32_use_load_post_increment): Declare.
 532         * config/nds32/nds32.c (nds32_use_load_post_increment): New.
 533         * config/nds32/nds32.h
 534         (USE_LOAD_POST_INCREMENT, USE_LOAD_POST_DECREMENT): Define.
 535         (USE_STORE_POST_INCREMENT, USE_STORE_POST_DECREMENT): Define.
 536
 537 2018-04-22  Shiva Chen  <shiva0217@gmail.com>
 538
 539         * config/nds32/nds32-protos.h (nds32_ls_333_p): Remove.
 540         * config/nds32/nds32.c (nds32_ls_333_p): Remove.
 541
 542 2018-04-22  Shiva Chen  <shiva0217@gmail.com>
 543             Chung-Ju Wu  <jasonwucj@gmail.com>
 544
 545         * config/nds32/nds32-protos.h (nds32_case_vector_shorten_mode):
 546         Declare.
 547         * config/nds32/nds32.c (nds32_case_vector_shorten_mode): New function.
 548         * config/nds32/nds32.h (CASE_VECTOR_SHORTEN_MODE): Modify.
 549
 550 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
 551
 552         * config/nds32/nds32.c (nds32_compute_stack_frame): Fix wrong value.
 553
 554 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
 555
 556         * config/nds32/nds32-protos.h (nds32_data_alignment,
 557         nds32_local_alignment): Declare.
 558         * config/nds32/nds32.c (nds32_data_alignment, nds32_constant_alignment,
 559         nds32_local_alignment): New functions.
 560         (TARGET_CONSTANT_ALIGNMENT): Define.
 561         * config/nds32/nds32.h (DATA_ALIGNMENT, LOCAL_ALIGNMENT): Define.
 562
 563 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
 564
 565         * config/nds32/nds32.c
 566         (TARGET_HARD_REGNO_MODE_OK): Move to the bottom of file.
 567         (TARGET_MODES_TIEABLE_P): Likewise.
 568
 569 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
 570
 571         * config/nds32/nds32.c (nds32_asm_file_start): Display optimization
 572         level Ofast and Og.
 573
 574 2018-04-22  Monk Chiang  <sh.chiang04@gmail.com>
 575             Chung-Ju Wu  <jasonwucj@gmail.com>
 576
 577         * config/nds32/constants.md (unspec_volatile_element): Add enum values
 578         for unaligned access.
 579         * config/nds32/nds32-intrinsic.c: Implementation of expanding
 580         unaligned access.
 581         * config/nds32/nds32-intrinsic.md: Likewise.
 582         * config/nds32/nds32_intrinsic.h: Likewise.
 583         * config/nds32/nds32.h (nds32_builtins): Likewise.
 584         * config/nds32/nds32.opt (munaligned-access): New option.
 585         * config/nds32/nds32.c (nds32_asm_file_start): Display
 586         flag_unaligned_access status.
 587
 588 2018-04-20  Kito Cheng  <kito.cheng@gmail.com>
 589
 590         * config/riscv/elf.h (LINK_SPEC): Pass --no-relax if
 591         -mno-relax is present.
 592         * config/riscv/linux.h (LINK_SPEC): Ditto.
 593
 594 2018-04-20  Martin Sebor  <msebor@redhat.com>
 595
 596         PR c/85365
 597         * gimple-fold.c (gimple_fold_builtin_strcpy): Suppress -Wrestrict
 598         for null pointers.
 599         (gimple_fold_builtin_stxcpy_chk): Same.
 600         * gimple-ssa-warn-restrict.c (check_bounds_or_overlap): Same.
 601
 602 2018-04-20  Michael Meissner  <meissner@linux.ibm.com>
 603
 604         PR target/85456
 605         * config/rs6000/rs6000.c (init_float128_ieee): Add support to call
 606         __powikf2 when long double is IEEE 128-bit.
 607
 608 2018-04-20  Kito Cheng  <kito.cheng@gmail.com>
 609
 610         * config/riscv/riscv.c (riscv_first_stack_step): Round up min
 611         step to make sure stack always aligned.
 612
 613 2018-04-20  Carl Love  <cel@us.ibm.com>
 614
 615         PR target/83402
 616         * config/rs6000/rs6000-c.c (rs6000_gimple_fold_builtin): Add
 617         size check for arg0.
 618
 619 2018-04-20  Nathan Sidwell  <nathan@codesourcery.com>
 620             Tom de Vries  <tom@codesourcery.com>
 621
 622         PR target/85445
 623         * config/nvptx/nvptx.c (nvptx_emit_forking, nvptx_emit_joining):
 624         Emit insns for calls too.
 625         (nvptx_find_par): Always look for worker-level predecessor insn.
 626         (nvptx_propagate): Add is_call parm, return bool.  Copy frame for
 627         calls.
 628         (nvptx_vpropagate, nvptx_wpropagate): Adjust.
 629         (nvptx_process_pars): Propagate frames for calls.
 630
 631 2018-04-20  H.J. Lu  <hongjiu.lu@intel.com>
 632
 633         PR target/85469
 634         * common/config/i386/i386-common.c (OPTION_MASK_ISA_IBT_SET):
 635         Removed.
 636         (OPTION_MASK_ISA_IBT_UNSET): Likewise.
 637         (ix86_handle_option): Don't handle OPT_mibt.
 638         * config/i386/cet.h: Check __CET__ instead of __IBT__ and
 639         __SHSTK__.
 640         * config/i386/driver-i386.c (host_detect_local_cpu): Remove
 641         has_ibt and ibt.
 642         * config/i386/i386-c.c (ix86_target_macros_internal): Don't
 643         check OPTION_MASK_ISA_IBT nor flag_cf_protection.
 644         (ix86_target_macros): Define __CET__ with flag_cf_protection
 645         for -fcf-protection.
 646         * config/i386/i386.c (isa2_opts): Remove -mibt.
 647         * config/i386/i386.h (TARGET_IBT): Removed.
 648         (TARGET_IBT_P): Likewise.
 649         (ix86_valid_target_attribute_inner_p): Don't check OPT_mibt.
 650         * config/i386/i386.md (nop_endbr): Don't check TARGET_IBT.
 651         * config/i386/i386.opt (mcet): Update help message.
 652         (mshstk): Likewise.
 653         (mibt): Removed.
 654         * doc/invoke.texi: Remove -mibt.  Document __CET__.  Document
 655         -mcet as an alias for -mshstk.
 656
 657 2018-04-20  Richard Biener <rguenther@suse.de>
 658
 659         PR middle-end/85475
 660         * match.pd ((X * CST) * Y -> (X * Y) * CST): Avoid exponential
 661         complexity by forcing a single use of the multiply operand.
 662
 663 2018-04-20  Martin Jambor  <mjambor@suse.cz>
 664
 665         ipa/85449
 666         * ipa-cp.c (cgraph_edge_brings_value_p): Move check for self-feeding
 667         recursion dependency to only apply to non-clones.
 668
 669 2018-04-20  Martin Jambor  <mjambor@suse.cz>
 670
 671         ipa/85447
 672         * ipa-cp.c (create_specialized_node): Check that clones of
 673         self-recursive edges exist during IPA-CP.
 674
 675 2018-04-19  Toon Moene  <toon@moene.org>
 676
 677         * doc/invoke.texi: Add -floop-unroll-and-jam to options enabled
 678         by -O3.
 679
 680 2018-04-19  Jakub Jelinek  <jakub@redhat.com>
 681
 682         PR tree-optimization/85467
 683         * fold-const.c (fold_ternary_loc) <case BIT_FIELD_REF>: Use
 684         VECTOR_TYPE_P macro.  If type is vector type, VIEW_CONVERT_EXPR the
 685         VECTOR_CST element to type.
 686
 687 2018-04-19  H.J. Lu  <hongjiu.lu@intel.com>
 688
 689         PR target/85397
 690         * config/i386/i386.h (STACK_SAVEAREA_MODE): New.
 691         * config/i386/i386.md (builtin_setjmp_setup): Removed.
 692         (builtin_longjmp): Likewise.
 693         (save_stack_nonlocal): New pattern.
 694         (restore_stack_nonlocal): Likewise.
 695
 696 2018-04-19  H.J. Lu  <hongjiu.lu@intel.com>
 697
 698         PR target/85404
 699         * config/i386/cet.c (file_end_indicate_exec_stack_and_cet):
 700         Replace ASM_OUTPUT_LABEL with fprintf.
 701
 702 2018-04-19  H.J. Lu  <hongjiu.lu@intel.com>
 703
 704         PR target/85417
 705         * config/i386/cet.c (file_end_indicate_exec_stack_and_cet):
 706         Check flag_cf_protection instead of TARGET_IBT and TARGET_SHSTK.
 707         * config/i386/i386-c.c (ix86_target_macros_internal): Also
 708         define __IBT__ and __SHSTK__ for -fcf-protection.
 709         * config/i386/i386.c (pass_insert_endbranch::gate): Don't check
 710         TARGET_IBT.
 711         (ix86_trampoline_init): Likewise.
 712         (x86_output_mi_thunk): Likewise.
 713         (ix86_notrack_prefixed_insn_p): Likewise.
 714         (ix86_option_override_internal): Don't disallow -fcf-protection.
 715         * config/i386/i386.md (rdssp<mode>): Also enable for
 716         -fcf-protection.
 717         (incssp<mode>): Likewise.
 718         (nop_endbr): Likewise.
 719         * config/i386/i386.opt (mcet): Change help message to built-in
 720         functions only.
 721         (mibt): Likewise.
 722         (mshstk): Likewise.
 723         * doc/invoke.texi: Remove -mcet, -mibt and -mshstk condition
 724         on -fcf-protection.  Change -mcet, -mibt and -mshstk to only
 725         enable CET built-in functions.
 726
 727 2018-04-19  Sebastian Peryt  <sebastian.peryt@intel.com>
 728
 729         * common/config/i386/i386-common.c
 730         (OPTION_MASK_ISA_MOVDIRI_SET, OPTION_MASK_ISA_MOVDIR64B_SET,
 731         OPTION_MASK_ISA_MOVDIRI_UNSET,
 732         OPTION_MASK_ISA_MOVDIR64B_UNSET): New defines.
 733         (ix86_handle_option): Handle -mmovdiri and -mmovdir64b.
 734         * config.gcc (movdirintrin.h): New header.
 735         * config/i386/cpuid.h (bit_MOVDIRI,
 736         bit_MOVDIR64B): New bits.
 737         * config/i386/driver-i386.c (host_detect_local_cpu): Detect -mmovdiri
 738         and -mmvodir64b.
 739         * config/i386/i386-builtin-types.def ((VOID, PUNSIGNED, UNSIGNED),
 740         (VOID, PVOID, PCVOID)): New function types.
 741         * config/i386/i386-builtin.def (__builtin_ia32_directstoreu_u32,
 742         __builtin_ia32_directstoreu_u64,
 743          __builtin_ia32_movdir64b): New builtins.
 744         * config/i386/i386-c.c (__MOVDIRI__, __MOVDIR64B__): New.
 745         * config/i386/i386.c (ix86_target_string): Added -mmovdir64b
 746         and -mmovdiri.
 747         (ix86_valid_target_attribute_inner_p): Ditto.
 748         (ix86_expand_special_args_builtin): Added VOID_FTYPE_PUNSIGNED_UNSIGNED
 749         and VOID_FTYPE_PUNSIGNED_UNSIGNED.
 750         (ix86_expand_builtin): Expand IX86_BUILTIN_MOVDIR64B.
 751         * config/i386/i386.h (TARGET_MOVDIRI, TARGET_MOVDIRI_P,
 752         TARGET_MOVDIR64B, TARGET_MOVDIR64B_P): New.
 753         * config/i386/i386.md (UNSPECV_MOVDIRI, UNSPECV_MOVDIR64B): New.
 754         (movdiri<mode>, movdir64b_<mode>): New.
 755         * config/i386/i386.opt: Add -mmovdiri and -mmovdir64b.
 756         * config/i386/immintrin.h: Include movdirintrin.h.
 757         * config/i386/movdirintrin.h: New file.
 758         * doc/invoke.texi: Added -mmovdiri and -mmovdir64b.
 759
 760 2018-04-19  Richard Biener  <rguenther@suse.de>
 761
 762         PR middle-end/85455
 763         * cfg.c (clear_bb_flags): When loop state says we have
 764         marked irreducible regions also preserve BB_IRREDUCIBLE_LOOP.
 765
 766 2018-04-19  Richard Biener  <rguenther@suse.de>
 767
 768         PR tree-optimization/84737
 769         * tree-vect-data-refs.c (vect_copy_ref_info): New function
 770         copying restrict info.
 771         (vect_setup_realignment): Use it.
 772         * tree-vectorizer.h (vect_copy_ref_info): Declare.
 773         * tree-vect-stmts.c (vectorizable_store): Copy ref info from
 774         the first DR to all generated stores.
 775         (vectorizable_load): Likewise for loads.
 776
 777 2018-04-19  Jakub Jelinek  <jakub@redhat.com>
 778
 779         PR tree-optimization/85446
 780         * match.pd ((intptr_t) x eq/ne CST to x eq/ne (typeof x) cst): Require
 781         the integral and pointer types to have the same precision.
 782
 783         * doc/install.texi: Document --disable-cet being the default and
 784         --enable-cet=auto.
 785
 786 2018-04-18  Martin Liska  <mliska@suse.cz>
 787
 788         * ipa-devirt.c (odr_subtypes_equivalent_p): Fix GNU coding
 789         style.
 790
 791 2018-04-18  Martin Liska  <mliska@suse.cz>
 792
 793         Revert
 794         2018-03-02  Eric Botcazou  <ebotcazou@adacore.com>
 795
 796         PR ipa/83983
 797         * ipa-devirt.c (odr_subtypes_equivalent_p): Get the ODR type of both
 798         arguments if they are comparable.
 799
 800 2018-04-18  Martin Liska  <mliska@suse.cz>
 801
 802         Revert
 803         2018-03-13  Eric Botcazou  <ebotcazou@adacore.com>
 804
 805         PR lto/84805
 806         * ipa-devirt.c (odr_subtypes_equivalent_p): Do not get the ODR type of
 807         incomplete types.
 808
 809 2018-04-18  H.J. Lu  <hongjiu.lu@intel.com>
 810
 811         PR target/85388
 812         * config/i386/i386.c (ix86_expand_split_stack_prologue): Insert
 813         ENDBR after calling __morestack.
 814
 815 2018-04-18  David Malcolm  <dmalcolm@redhat.com>
 816
 817         PR jit/85384
 818         * configure.ac (gcc-driver-name.h): Honor --with-gcc-major-version
 819         by using gcc_base_ver to generate a gcc_driver_version, and use
 820         it when generating GCC_DRIVER_NAME.
 821         * configure: Regenerate.
 822
 823 2018-04-18  Jakub Jelinek  <jakub@redhat.com>
 824
 825         PR target/81084
 826         * config.gcc: Obsolete powerpc*-*-*spe*.
 827
 828 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
 829
 830         PR debug/84637
 831         * dbxout.c (dbxout_int): Perform negation in unsigned int type.
 832         (stabstr_D): Change type of unum from unsigned int to
 833         unsigned HOST_WIDE_INT.  Perform negation in unsigned HOST_WIDE_INT
 834         type.
 835
 836 2018-04-17  Jim Wilson  <jimw@sifive.com>
 837
 838         PR 84856
 839         * config/riscv/riscv.c (riscv_compute_frame_info): Add calls to
 840         RISCV_STACK_ALIGN when using outgoing_args_size and pretend_args_size.
 841         Set arg_pointer_offset after using pretend_args_size.
 842
 843 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
 844
 845         PR rtl-optimization/85431
 846         * dse.c (record_store): Ignore zero width stores.
 847
 848         PR sanitizer/85230
 849         * asan.c (handle_builtin_stack_restore): Adjust comment.  Emit
 850         __asan_allocas_unpoison call and last_alloca_addr = new_sp before
 851         __builtin_stack_restore rather than after it.
 852         * builtins.c (expand_asan_emit_allocas_unpoison): Pass
 853         arg1 + (virtual_dynamic_stack_rtx - stack_pointer_rtx) as second
 854         argument instead of virtual_dynamic_stack_rtx.
 855
 856 2018-04-17  Kelvin Nilsen  <kelvin@gcc.gnu.org>
 857
 858         * config/rs6000/rs6000-protos.h (rs6000_builtin_is_supported_p):
 859         New prototype.
 860         * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin):
 861         Add note to error message to explain internal mapping of overloaded
 862         built-in function name to non-overloaded built-in function name.
 863         * config/rs6000/rs6000.c (rs6000_builtin_is_supported_p): New
 864         function.
 865
 866 2018-04-17  Michael Meissner  <meissner@linux.vnet.ibm.com>
 867
 868         PR target/85424
 869         * config/rs6000/rs6000.md (pack<mode>): Do not try handle a pack
 870         where the inputs overlap with the output.
 871
 872 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
 873
 874         * config/i386/sse.md (vec_extract_lo_<mode><mask_name>): Add
 875         (=v, v) alternative and explicit "memory" attribute.
 876         (vec_extract_lo_<mode><mask_name>): Likewise.  Also add
 877         "type", "prefix", "prefix_extra", "length_immediate" and "mode"
 878         attributes.
 879         (vec_extract_lo_<mode><mask_name>): Add (=v, v) alternative and use
 880         "sselog1" type instead of "sselog".
 881         (vec_extract_hi_<mode><mask_name>): Use "sselog1" type instead of
 882         "sselog".  Remove explicit "memory" attribute.
 883         (vec_extract_lo_v32hi): Add (=v, v) alternative and explicit "memory",
 884         "type", "prefix", "prefix_extra", "length_immediate" and "mode"
 885         attributes.
 886         (vec_extract_hi_v32hi): Merge all alternatives into one, use
 887         "sselog1" type instead of "sselog".  Remove explicit "memory"
 888         attribute.
 889         (vec_extract_hi_v16hi): Merge each pair of alternatives into one,
 890         use "sselog1" type instead of "sselog".  Remove explicit "memory"
 891         attribute.
 892         (vec_extract_lo_v64qi): Add (=v, v) alternative and explicit "memory",
 893         "type", "prefix", "prefix_extra", "length_immediate" and "mode"
 894         attributes.
 895         (vec_extract_hi_v64qi): Merge all alternatives into one, use
 896         "sselog1" type instead of "sselog".  Remove explicit "memory"
 897         attribute.
 898         (vec_extract_hi_v32qi): Merge each pair of alternatives into one,
 899         use "sselog1" type instead of "sselog".  Remove explicit "memory"
 900         attribute.
 901
 902         PR target/85430
 903         * config/i386/i386.md (*ashlqi3_1_slp): Use alu1 type instead of alu.
 904
 905         PR middle-end/85414
 906         * rtlhooks.c (gen_lowpart_if_possible): Don't call gen_lowpart_SUBREG
 907         on a SUBREG.
 908
 909 2018-04-17  Martin Jambor  <mjambor@suse.cz>
 910
 911         PR ipa/85421
 912         * ipa-cp.c (create_specialized_node): Call
 913         expand_all_artificial_thunks if necessary.
 914
 915 2018-04-17  Martin Liska  <mliska@suse.cz>
 916
 917         PR lto/85405
 918         * ipa-devirt.c (odr_types_equivalent_p): Remove trailing
 919         in message, remote space in between '_G' and '('.
 920
 921 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
 922
 923         PR target/85281
 924         * config/i386/sse.md (reduces<mode><mask_scalar_name>,
 925         avx512f_vmcmp<mode>3<round_saeonly_name>,
 926         avx512f_vmcmp<mode>3_mask<round_saeonly_name>,
 927         avx512f_sgetexp<mode><mask_scalar_name><round_saeonly_scalar_name>,
 928         avx512f_rndscale<mode><round_saeonly_name>,
 929         avx512dq_ranges<mode><mask_scalar_name><round_saeonly_scalar_name>,
 930         avx512f_vgetmant<mode><mask_scalar_name><round_saeonly_scalar_name>):
 931         Use %<iptr>2 instead of %2 for -masm=intel.
 932         (avx512f_vcvtss2usi<round_name>, avx512f_vcvtss2usiq<round_name>,
 933         avx512f_vcvttss2usi<round_saeonly_name>,
 934         avx512f_vcvttss2usiq<round_saeonly_name>): Use %k1 instead of %1 for
 935         -masm=intel.
 936         (avx512f_vcvtsd2usi<round_name>, avx512f_vcvtsd2usiq<round_name>,
 937         avx512f_vcvttsd2usi<round_saeonly_name>,
 938         avx512f_vcvttsd2usiq<round_saeonly_name>, ufloatv2siv2df2<mask_name>):
 939         Use %q1 instead of %1 for -masm=intel.
 940         (avx512f_sfixupimm<mode><sd_maskz_name><round_saeonly_name>,
 941         avx512f_sfixupimm<mode>_mask<round_saeonly_name>): Use %<iptr>3 instead
 942         of %3 for -masm=intel.
 943         (sse2_shufpd_v2df_mask): Fix a typo, change %{6%} to %{%6%} for
 944         -masm=intel.
 945         (*avx512vl_<code>v2div2qi2_store): Use %w0 instead of %0 for
 946         -masm=intel.
 947         (*avx512vl_<code><mode>v4qi2_store): Use %k0 instead of %0 for
 948         -masm=intel.
 949         (avx512vl_<code><mode>v4qi2_mask_store): Use a single pattern with
 950         %k0 and %1 for -masm=intel rather than two patterns, one with %0 and
 951         %g1.
 952         (*avx512vl_<code><mode>v8qi2_store): Use %q0 instead of %0 for
 953         -masm=intel.
 954         (avx512vl_<code><mode>v8qi2_mask_store): Use a single pattern with
 955         %q0 and %1 for -masm=intel rather than two patterns, one with %0 and
 956         %g1 and one with %0 and %1.
 957         (avx512er_vmrcp28<mode><round_saeonly_name>,
 958         avx512er_vmrsqrt28<mode><round_saeonly_name>): Use %<iptr>1 instead of
 959         %1 for -masm=intel.
 960         (avx5124fmaddps_4fmaddps_mask, avx5124fmaddps_4fmaddss_mask,
 961         avx5124fmaddps_4fnmaddps_mask, avx5124fmaddps_4fnmaddss_mask,
 962         avx5124vnniw_vp4dpwssd_mask, avx5124vnniw_vp4dpwssds_mask): Swap order
 963         of %0 and %{%4%} for -masm=intel.
 964         (avx5124fmaddps_4fmaddps_maskz, avx5124fmaddps_4fmaddss_maskz,
 965         avx5124fmaddps_4fnmaddps_maskz, avx5124fmaddps_4fnmaddss_maskz,
 966         avx5124vnniw_vp4dpwssd_maskz, avx5124vnniw_vp4dpwssds_maskz): Swap
 967         order of %0 and %{%5%}%{z%} for -masm=intel.
 968
 969 2018-04-17  Jan Hubicka  <jh@suse.cz>
 970
 971         PR lto/85405
 972         * ipa-devirt.c (odr_types_equivalent_p): Handle bit fields.
 973
 974 2018-04-17  Martin Liska  <mliska@suse.cz>
 975
 976         PR ipa/85329
 977         * multiple_target.c (create_dispatcher_calls): Set apostrophes
 978         for target_clone error message.  Make default implementation
 979         clone to be a local declaration.
 980         (separate_attrs): Add new argument and check for an empty
 981         string.
 982         (expand_target_clones): Handle it.
 983         (ipa_target_clone): Make redirection just for target_clones
 984         functions.
 985
 986 2018-04-16  Cesar Philippidis  <cesar@codesourcery.com>
 987             Tom de Vries  <tom@codesourcery.com>
 988
 989         PR middle-end/84955
 990         * omp-expand.c (expand_oacc_for): Add dummy false branch for
 991         tiled basic blocks without omp continue statements.
 992
 993 2018-04-16  Aaron Sawdey  <acsawdey@linux.ibm.com>
 994
 995         PR target/83660
 996         * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Mark
 997         vec_extract expression as having side effects to make sure it gets
 998         a cleanup point.
 999
1000 2018-04-16  H.J. Lu  <hongjiu.lu@intel.com>
1001
1002         PR target/85403
1003         * config/i386/i386.c (get_builtin_code_for_version): Check
1004         error_mark_node.
1005
1006 2018-04-16  Olga Makhotina  <olga.makhotina@intel.com>
1007
1008         PR target/84331
1009         * gcc/config.gcc: Support "skylake".
1010         * gcc/config/i386/i386-c.c (ix86_target_macros_internal): Handle
1011         PROCESSOR_SKYLAKE.
1012         * gcc/config/i386/i386.c (m_SKYLAKE): Define.
1013         (processor_target_table): Add "skylake".
1014         (ix86_option_override_internal): Add "skylake".
1015         (get_builtin_code_for_version): Handle PROCESSOR_SKYLAKE,
1016         PROCESSOR_CANNONLAKE.
1017         (get_builtin_code_for_version): Fix priority for
1018         PROCESSOR_ICELAKE_CLIENT, PROCESSOR_ICELAKE_SERVER,
1019         PROCESSOR_SKYLAKE-AVX512.
1020         * gcc/config/i386/i386.h (processor_costs): Define TARGET_SKYLAKE.
1021         (processor_type): Add PROCESSOR_SKYLAKE.
1022
1023 2018-04-16  Paolo Carlini  <paolo.carlini@oracle.com>
1024             Jason Merrill  <jason@redhat.com>
1025
1026         PR c++/85112
1027         * convert.c (convert_to_integer_1): Use direct recursion for
1028         enumeral types and types with a precision less than the number
1029         of bits in their mode.
1030
1031 2018-04-16  Julia Koval  <julia.koval@intel.com>
1032
1033         PR target/84413
1034         * config/i386/x86-tune.def (X86_TUNE_SSE_UNALIGNED_LOAD_OPTIMAL,
1035         X86_TUNE_SSE_UNALIGNED_STORE_OPTIMAL): Add m_SKYLAKE_AVX512
1036
1037 2018-04-14  Segher Boessenkool  <segher@kernel.crashing.org>
1038
1039         PR target/85293
1040         * config/rs6000/rs6000.opt (mdirect-move): Make deprecated.
1041         * doc/invoke.texi (RS/6000 and PowerPC Options): Remove -mdirect-move
1042         and -mno-direct-move.
1043
1044 2018-04-13  Paul A. Clarke  <pc@us.ibm.com>
1045
1046         PR target/83402
1047         * config/rs6000/emmintrin.h (_mm_slli_epi{16,32,64}):
1048         Ensure that vec_splat_s32 is only called with 0 <= shift < 16.
1049         Ensure negative shifts result in {0}.
1050
1051 2018-04-13  Vladimir Makarov  <vmakarov@redhat.com>
1052
1053         PR rtl-optimization/79916
1054         * config/rs6000/rs6000.c (rs6000_emit_move): Use assigned hard
1055         regs (if any) to define how to gnerate SD moves when LRA is in
1056         progress.
1057
1058 2018-04-13  Jakub Jelinek  <jakub@redhat.com>
1059
1060         PR rtl-optimization/85393
1061         * except.h (expand_dw2_landing_pad_for_region): Remove declaration.
1062         * except.c (expand_dw2_landing_pad_for_region): Make static.
1063         * bb-reorder.c (fix_up_crossing_landing_pad): In new_bb emit just
1064         a label and unconditional jump to old_bb, rather than
1065         expand_dw2_landing_pad_for_region insn(s) and jump to single_succ
1066         basic block.
1067
1068         PR rtl-optimization/85376
1069         * simplify-rtx.c (simplify_const_unary_operation): For CLZ and CTZ and
1070         zero op0, if C?Z_DEFINED_VALUE_AT_ZERO is false, return NULL_RTX
1071         instead of a specific value.
1072
1073 2018-04-13  Jan Hubicka  <hubicka@ucw.cz>
1074             Bin Cheng  <bin.cheng@arm.com>
1075
1076         PR tree-optimization/82965
1077         PR tree-optimization/83991
1078         * cfgloopanal.c (expected_loop_iterations_unbounded): Add
1079         by_profile_only parameter.
1080         * cfgloopmanip.c (scale_loop_profile): Further scale loop's profile
1081         information if the loop was predicted to iterate too many times.
1082         * cfgloop.h (expected_loop_iterations_unbounded): Update prototype
1083
1084 2018-04-13  Jan Hubicka  <hubicka@ucw.cz>
1085
1086         PR lto/71991
1087         * config/i386/i386.c (ix86_can_inline_p): Allow safe transitions for
1088         always inline.
1089
1090 2018-04-13  Martin Liska  <mliska@suse.cz>
1091             Jakub Jelinek  <jakub@redhat.com>
1092
1093         PR middle-end/81657
1094         * expr.h (enum block_op_methods): Add BLOCK_OP_NO_LIBCALL_RET.
1095         * expr.c (emit_block_move_hints): Handle BLOCK_OP_NO_LIBCALL_RET.
1096         * builtins.c (expand_builtin_memory_copy_args): Use
1097         BLOCK_OP_NO_LIBCALL_RET method for mempcpy with non-ignored target,
1098         handle dest_addr == pc_rtx.
1099
1100 2018-04-12  Segher Boessenkool  <segher@kernel.crashing.org>
1101
1102         PR target/85291
1103         * config/rs6000/rs6000.md (fix_trunc<mode>si2): Use legacy code if
1104         asked to not generate direct moves.
1105         (fix_trunc<mode>si2_stfiwx): Similar.
1106         (fix_trunc<mode>si2_internal): Similar.
1107
1108 2018-04-12  Jakub Jelinek  <jakub@redhat.com>
1109
1110         PR debug/83157
1111         * var-tracking.c (add_stores): Handle STRICT_LOW_PART SET_DEST.
1112         * cselib.c (cselib_record_sets): For STRICT_LOW_PART dest,
1113         lookup if dest in some wider mode is known to be const0_rtx and
1114         if so, record permanent equivalence for it to be ZERO_EXTEND of
1115         the narrower mode destination.
1116
1117 2018-04-12  Cesar Philippidis  <cesar@codesourcery.com>
1118
1119         * lto-streamer-out.c (output_function): Revert 259346.
1120         * omp-expand.c (expand_oacc_for): Likewise.
1121
1122 2018-04-12  Alexander Monakov  <amonakov@ispras.ru>
1123
1124         PR rtl-optimization/85354
1125         * sel-sched-ir.c (sel_init_pipelining): Move cfg_cleanup call...
1126         * sel-sched.c (sel_global_init): ... here.
1127
1128 2018-04-12  Eric Botcazou  <ebotcazou@adacore.com>
1129
1130         PR target/85238
1131         * lto-wrapper.c (debug_objcopy): Open the files in binary mode.
1132         * dwarf2out.c (dwarf2out_early_finish): Do not generate assembly in LTO
1133         mode for PE-COFF targets.
1134         * config/i386/i386-protos.h (i386_pe_asm_lto_start): Declare.
1135         (i386_pe_asm_lto_end): Likewise.
1136         * config/i386/cygming.h (TARGET_ASM_LTO_START): Define.
1137         (TARGET_ASM_LTO_END): Likewise.
1138         * config/i386/winnt.c (saved_debug_info_level): New static variable.
1139         (i386_pe_asm_lto_start): New function.
1140         (i386_pe_asm_lto_end): Likewise.
1141
1142 2018-04-12  Cesar Philippidis  <cesar@codesourcery.com>
1143             Richard Biener  <rguenther@suse.de>
1144
1145         PR middle-end/84955
1146         * lto-streamer-out.c (output_function): Fix CFG loop state before
1147         streaming out.
1148         * omp-expand.c (expand_oacc_for): Handle calls to internal
1149         functions like regular functions.
1150
1151 2018-04-12  Richard Biener  <rguenther@suse.de>
1152
1153         PR lto/85371
1154         * dwarf2out.c (init_sections_and_labels): Use debug_line_section[_label]
1155         for the early LTO debug to properly generate references to it
1156         during DIE emission.  Do not re-use that for the skeleton for
1157         split-dwarf.
1158         (dwarf2out_early_finish): Likewise.
1159
1160 2018-04-12  Jakub Jelinek  <jakub@redhat.com>
1161
1162         PR target/85328
1163         * config/i386/sse.md
1164         (<mask_codefor>avx512dq_vextract<shuffletype>64x2_1<mask_name> split,
1165         <mask_codefor>avx512f_vextract<shuffletype>32x4_1<mask_name> split,
1166         vec_extract_lo_<mode><mask_name> split, vec_extract_lo_v32hi,
1167         vec_extract_lo_v64qi): For non-AVX512VL if input is xmm16+ reg
1168         and output is a reg, avoid creating invalid lowpart subreg, but
1169         instead split into a 512-bit move.  Don't split if not AVX512VL,
1170         input is xmm16+ reg and output is a mem.
1171         (vec_extract_lo_<mode><mask_name>, vec_extract_lo_v32hi,
1172         vec_extract_lo_v64qi): Don't require split if not AVX512VL, input is
1173         xmm16+ reg and output is a mem.
1174
1175 2018-04-12  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
1176
1177         * config/s390/s390.c (s390_output_indirect_thunk_function): Check
1178         also for flag_dwarf2_cfi_asm.
1179
1180 2018-04-12  Jakub Jelinek  <jakub@redhat.com>
1181
1182         PR rtl-optimization/85342
1183         * regcprop.c (copyprop_hardreg_forward_1): Remove replaced array, use
1184         a bool scalar var inside of the loop instead.  Don't try to update
1185         recog_data.operand after failed apply_change_group.
1186
1187 2018-04-12  Tom de Vries  <tom@codesourcery.com>
1188
1189         PR target/85296
1190         * config/nvptx/nvptx.c (flexible_array_member_type_p): New function.
1191         (nvptx_assemble_decl_begin): Add undefined param.  Declare undefined
1192         array with flexible array member as array without given dimension.
1193         (nvptx_assemble_undefined_decl): Set nvptx_assemble_decl_begin call
1194         argument for undefined param to true.
1195
1196 2018-04-11  Aaron Sawdey  <acsawdey@linux.ibm.com>
1197
1198         PR target/85321
1199         * doc/invoke.texi (RS/6000 and PowerPC Options): Document options
1200         -mcall- and -mtraceback=. Remove options -mabi=spe and -mabi=no-spe
1201         from PowerPC section.
1202         * config/rs6000/sysv4.opt (mcall-): Improve help text.
1203         * config/rs6000/rs6000.opt (mblock-compare-inline-limit=): Trim
1204         help text that is too long.
1205         * config/rs6000/rs6000.opt (mblock-compare-inline-loop-limit=): Trim
1206         help text that is too long.
1207         * config/rs6000/rs6000.opt (mstring-compare-inline-limit=): Trim
1208         help text that is too long.
1209
1210 2018-04-11  Uros Bizjak  <ubizjak@gmail.com>
1211
1212         * config/alpha/alpha.md (stack_probe_internal): Rename
1213         from "probe_stack".  Update all callers.
1214
1215 2018-04-11  Alexander Monakov  <amonakov@ispras.ru>
1216
1217         PR rtl-optimization/84566
1218         * sched-deps.c (sched_analyze_insn): Check deps->readonly when invoking
1219         sched_macro_fuse_insns.
1220
1221 2018-04-11  Alexander Monakov  <amonakov@ispras.ru>
1222
1223         PR target/84301
1224         * sched-rgn.c (add_branch_dependences): Move sel_sched_p check here...
1225         (compute_block_dependences): ... from here.
1226
1227 2018-04-11  Jakub Jelinek  <jakub@redhat.com>
1228
1229         PR tree-optimization/85331
1230         * vec-perm-indices.h (vec_perm_indices::clamp): Change input type
1231         from int to HOST_WIDE_INT.
1232
1233 2018-04-11  Martin Jambor  <mjambor@suse.cz>
1234
1235         PR ipa/84149
1236         * ipa-cp.c (propagate_vals_across_pass_through): Expand comment.
1237         (cgraph_edge_brings_value_p): New parameter dest_val, check if it is
1238         not the same as the source val.
1239         (cgraph_edge_brings_value_p): New parameter.
1240         (gather_edges_for_value): Pass destination value to
1241         cgraph_edge_brings_value_p.
1242         (perhaps_add_new_callers): Likewise.
1243         (get_info_about_necessary_edges): Likewise and exclude values brought
1244         only by self-recursive edges.
1245         (create_specialized_node): Redirect only clones of self-calling edges.
1246         (+self_recursive_pass_through_p): New function.
1247         (find_more_scalar_values_for_callers_subset): Use it.
1248         (find_aggregate_values_for_callers_subset): Likewise.
1249         (known_aggs_to_agg_replacement_list): Removed.
1250         (decide_whether_version_node): Re-calculate known constants for all
1251         remaining context clones.
1252
1253 2018-04-11  Richard Biener  <rguenther@suse.de>
1254
1255         PR lto/85339
1256         * dwarf2out.c (dwarf2out_finish): Remove DW_AT_stmt_list attribute
1257         from early DWARF output.
1258         (dwarf2out_early_finish): Output line info unconditionally into
1259         early DWARF and add reference to it.
1260
1261 2018-04-11  Jakub Jelinek  <jakub@redhat.com>
1262
1263         PR target/85281
1264         * config/i386/sse.md (iptr): Add V16SFmode and V8DFmode cases.
1265         (<avx512>_vec_dup<mode><mask_name>): Use a single pattern for modes
1266         other than V2DFmode using iptr mode attribute.
1267         (<avx512>_vec_dup<mode><mask_name>): Use iptr mode attribute.
1268
1269 2018-04-11  Alexander Monakov  <amonakov@ispras.ru>
1270
1271         PR rtl-optimization/84659
1272         * sel-sched-ir.c (sel_init_pipelining): Invoke cleanup_cfg.
1273
1274 2018-04-11  Jakub Jelinek  <jakub@redhat.com>
1275
1276         PR debug/85302
1277         * dwarf2out.c (skip_loc_list_entry): Don't call size_of_locs if
1278         SIZEP is NULL.
1279         (output_loc_list): Pass address of a dummy size variable even in the
1280         locview handling loop.
1281         (index_location_lists): Add comment on why skip_loc_list_entry can't
1282         call size_of_locs.
1283
1284 2018-04-11  Thomas Preud'homme  <thomas.preudhomme@arm.com>
1285
1286         PR target/85261
1287         * config/arm/arm-builtins.c (arm_expand_builtin): Force input operand
1288         into register.
1289
1290 2018-04-10  Aaron Sawdey  <acsawdey@linux.ibm.com>
1291
1292         PR target/85321
1293         * doc/invoke.texi (RS/6000 and PowerPC Options): Document options
1294         -mblock-compare-inline-limit, -mblock-compare-inline-loop-limit,
1295         and -mstring-compare-inline-limit.
1296
1297 2018-04-10  Segher Boessenkool  <segher@kernel.crashing.org>
1298
1299         PR target/85287
1300         * gcc/config/rs6000/rs6000.md (allocate_stack): Put the residual size
1301         for stack clash protection in a register whenever we need it to be in
1302         a register.
1303
1304 2018-04-10  Segher Boessenkool  <segher@kernel.crashing.org>
1305
1306         * common/config/rs6000/rs6000-common.c (rs6000_option_init_struct):
1307         Enable -fasynchronous-unwind-tables by default if OBJECT_FORMAT_ELF.
1308
1309 2018-04-10  Segher Boessenkool  <segher@kernel.crashing.org>
1310
1311         PR target/85321
1312         * config/rs6000/rs6000.opt (mtraceback=): Show the allowed values in
1313         the help text.
1314         (mlong-double-): Ditto.
1315         * config/rs6000/sysv4.opt (msdata=): Ditto.
1316         (mtls-size=): Ditto.
1317
1318 2018-04-10  Kelvin Nilsen  <kelvin@gcc.gnu.org>
1319
1320         * config/rs6000/rs6000-c.c (altivec_overloaded_builtins): Remove
1321         erroneous entries for
1322         "vector int vec_ldl (int, long int *)", and
1323         "vector unsigned int vec_ldl (int, unsigned long int *)".
1324         Add comments and entries for
1325         "vector bool char vec_ldl (int, bool char *)",
1326         "vector bool short vec_ldl (int, bool short *)",
1327         "vector bool int vec_ldl (int, bool int *)",
1328         "vector bool long long vec_ldl (int, bool long long *)",
1329         "vector pixel vec_ldl (int, pixel *)",
1330         "vector long long vec_ldl (int, long long *)",
1331         "vector unsigned long long vec_ldl (int, unsigned long long *)".
1332         * config/rs6000/rs6000.c (rs6000_init_builtins): Initialize new
1333         type tree bool_long_long_type_node and correct definition of
1334         bool_V2DI_type_node to make reference to this new type tree.
1335         (rs6000_mangle_type): Replace erroneous reference to
1336         bool_long_type_node with bool_long_long_type_node.
1337         * config/rs6000/rs6000.h (enum rs6000_builtin_type_index): Add
1338         comments to emphasize sign distinctions for char and int types and
1339         replace RS6000_BTI_bool_long constant with
1340         RS6000_BTI_bool_long_long constant.  Also add comment to restrict
1341         use of RS6000_BTI_pixel.
1342         (bool_long_type_node): Remove this macro definition.
1343         (bool_long_long_type_node): New macro definition
1344
1345 2018-04-10  Jakub Jelinek  <jakub@redhat.com>
1346
1347         PR rtl-optimization/85300
1348         * combine.c (subst): Handle subst of CONST_SCALAR_INT_P new_rtx also
1349         into FLOAT and UNSIGNED_FLOAT like ZERO_EXTEND, return a CLOBBER if
1350         simplify_unary_operation fails.
1351
1352 2018-04-10  Martin Liska  <mliska@suse.cz>
1353
1354         * gdbhooks.py: Add pretty-printers for varpool_node, symtab_node,
1355         cgraph_edge and ipa_ref.
1356
1357 2018-04-10  Jakub Jelinek  <jakub@redhat.com>
1358
1359         PR target/85177
1360         PR target/85255
1361         * config/i386/sse.md
1362         (<extract_type>_vinsert<shuffletype><extract_suf>_mask): Fix
1363         computation of the VEC_MERGE selector from mask.
1364         (<extract_type>_vinsert<shuffletype><extract_suf>_1<mask_name>):
1365         Fix decoding of the VEC_MERGE selector into mask.
1366
1367 2018-04-10  Richard Sandiford  <richard.sandiford@linaro.org>
1368
1369         PR tree-optimization/85286
1370         * tree-vect-data-refs.c (vect_get_smallest_scalar_type):
1371
1372 2018-04-10  Richard Sandiford  <richard.sandiford@linaro.org>
1373
1374         * final.c (final_1): Set insn_last_address as well as
1375         insn_current_address.
1376
1377 2018-04-10  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
1378
1379         PR target/85173
1380         * explow.c (emit_stack_probe): Call validize_mem on memory location
1381         before passing it to gen_probe_stack.  Create address operand and
1382         legitimize it for the probe_stack_address case.
1383
1384 2018-04-09  Jan Hubicka  <jh@suse.cz>
1385
1386         PR lto/85078
1387         * ipa-devirt.c (rebuild_type_inheritance-hash): New.
1388         * ipa-utils.h (rebuild_type_inheritance-hash): Declare.
1389         * tree.c (free_lang_data_in_type): Fix handling of binfos;
1390         walk basetypes.
1391         (free_lang_data): Rebuild type inheritance graph.
1392
1393 2018-04-09  Martin Sebor  <msebor@redhat.com>
1394
1395         * invoke.texi (-finline-small-functions): Mention other optimization
1396         options.
1397         (-findirect-inlining, -fpartial-inlining): Same.
1398         (-finline-functions-called-once): Same.
1399         (-freorder-blocks-and-partition): Same.
1400
1401 2018-04-09  Jan Hubicka  <jh@suse.cz>
1402
1403         PR rtl/84058
1404         * cfgcleanup.c (try_forward_edges): Do not give up on crossing
1405         jumps; choose last target that matches the criteria (i.e.
1406         no partition changes for non-crossing jumps).
1407         * cfgrtl.c (cfg_layout_redirect_edge_and_branch): Add basic
1408         support for redirecting crossing jumps to non-crossing.
1409
1410 2018-04-09  Alexey Brodkin  <abrodkin@synopsys.com>
1411
1412         * config/arc/arc.c (arc_expand_prologue): Set stack usage info
1413         also for naked functions.
1414
1415 2018-04-09  Claudiu Zissulescu  <claziss@synopsys.com>
1416
1417         * config/arc/arc.md (add_shift): New pattern.
1418         (add_shift2): Likewise.
1419         (sub_shift): Likewise.
1420         (sub_shift_cmp0_noout): Likewise.
1421         (compare_si_ashiftsi): Likewise.
1422         (xbfu_cmp0_noout): New combine pattern.
1423         (xbfu_cmp0"): Likewise.
1424         (movsi_set_cc_insn): Place the predicable variant first.
1425         (commutative_binary_cmp0_noout): Remove clobber.
1426         (commutative_binary_cmp0): New pattern.
1427         (noncommutative_binary_cmp0): Likewise.
1428         (noncommutative_binary_cmp0_noout): Likewise.
1429         (noncommutative_binary_comparison_result_used): Removed.
1430         (rsub_cmp0): New pattern.
1431         (rsub_cmp0_noout): Likewise.
1432         (extzvsi): Changed, keep only meaningful variants.
1433         (SQH, SEZ): New iterators.
1434         (SQH_postfix): New mode attribute.
1435         (SEZ_prefix): New code attribute.
1436         (<SEZ_prefix>xt<SQH_postfix>_cmp0_noout): New instruction pattern.
1437         (<SEZ_prefix>xt<SQH_postfix>_cmp0): Likewise.
1438         * config/arc/predicates.md (cc_set_register): Use CC_REG instead
1439         of numerical value.
1440         (noncommutative_operator): Check the availability of barrel
1441         shifter option.
1442
1443 2018-04-09  Richard Biener  <rguenther@suse.de>
1444
1445         PR tree-optimization/85284
1446         * tree-ssa-loop-niter.c (number_of_iterations_exit_assumptions):
1447         Only use the niter constraining form of simple_iv when the exit
1448         is always executed.
1449
1450 2018-04-09  Tom de Vries  <tom@codesourcery.com>
1451
1452         PR target/84041
1453         * config/nvptx/nvptx.md (define_c_enum "unspecv"): Add UNSPECV_MEMBAR.
1454         (define_expand "*memory_barrier"): New define_expand.
1455         (define_insn "memory_barrier"): New insn.
1456
1457 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
1458
1459         PR rtl-optimization/80463
1460         PR rtl-optimization/83972
1461         PR rtl-optimization/83480
1462
1463         * sel-sched-ir.c (has_dependence_note_mem_dep): Take into account the
1464         correct producer for the insn.
1465         (tidy_control_flow): Fixup seqnos in case of debug insns.
1466
1467 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
1468
1469         PR rtl-optimization/83913
1470
1471         * sel-sched-ir.c (merge_expr_data): Choose the middle between two
1472         different sched-times when merging exprs.
1473
1474 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
1475
1476         PR rtl-optimization/83962
1477
1478         * sel-sched-ir.c (tidy_control_flow): Correct the order in which we call
1479         tidy_fallthru_edge and tidy_control_flow.
1480
1481 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
1482
1483         PR rtl-optimization/83530
1484
1485         * sel-sched.c (force_next_insn): New global variable.
1486         (remove_insn_for_debug): When force_next_insn is true, also leave only
1487         next insn in the ready list.
1488         (sel_sched_region): When the region wasn't scheduled, make another pass
1489         over it with force_next_insn set to 1.
1490
1491 2018-04-08  Monk Chiang  <sh.chiang04@gmail.com>
1492
1493         * config.gcc (nds32le-*-*, nds32be-*-*): Add nds32/nds32_intrinsic.h
1494         into tm_file.
1495         * config/nds32/constants.md (unspec_volatile_element): Add enum values
1496         for interrupt control.
1497         * config/nds32/nds32-intrinsic.c: Implementation of intrinsic
1498         functions for interrupt control.
1499         * config/nds32/nds32-intrinsic.md: Likewise.
1500         * config/nds32/nds32_intrinsic.h: Likewise.
1501         * config/nds32/nds32.h (nds32_builtins): Likewise.
1502
1503 2018-04-08  Chung-Ju Wu  <jasonwucj@gmail.com>
1504
1505         * config/nds32/nds32.c (nds32_init_machine_status,
1506         nds32_legitimate_index_p, nds32_legitimate_address_p): Consider
1507         strict_aligned_p field.
1508         (nds32_expand_to_rtl_hook): New function.
1509         (TARGET_EXPAND_TO_RTL_HOOK): Define.
1510         * config/nds32/nds32.h (machine_function): Add strict_aligned_p field.
1511
1512 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
1513             Chung-Ju Wu  <jasonwucj@gmail.com>
1514
1515         * config.gcc (nds32*-*-*): Check that n7 is valid to --with-cpu.
1516         * config/nds32/nds32-n7.md: New file.
1517         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_N7.
1518         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n7
1519         pipeline.
1520         * config/nds32/nds32-protos.h: More declarations for n7 pipeline.
1521         * config/nds32/nds32.md (pipeline_model): Add n7.
1522         * config/nds32/nds32.opt (mcpu): Support n7 pipeline cpus.
1523         * config/nds32/pipelines.md: Include n7 settings.
1524
1525 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
1526             Chung-Ju Wu  <jasonwucj@gmail.com>
1527
1528         * config.gcc (nds32*-*-*): Check that e8 is valid to --with-cpu.
1529         * config/nds32/nds32-e8.md: New file.
1530         * config/nds32/nds32-opts.h (nds32-cpu_type): Add CPU_E8.
1531         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for e8
1532         pipeline.
1533         * config/nds32/nds32-protos.h: More declarations for e8 pipeline.
1534         * config/nds32/nds32.md (pipeline_model): Add e8.
1535         * config/nds32/nds32.opt (mcpu): Support e8 pipeline cpus.
1536         * config/nds32/pipelines.md: Include e8 settings.
1537
1538 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
1539             Chung-Ju Wu  <jasonwucj@gmail.com>
1540
1541         * config.gcc (nds32*-*-*): Check that n6/n8/s8 are valid to --with-cpu.
1542         * config/nds32/nds32-n8.md: New file.
1543         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_N6 and CPU_N8.
1544         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n8
1545         pipeline.
1546         * config/nds32/nds32-protos.h: More declarations for n8 pipeline.
1547         * config/nds32/nds32-utils.c: More implementations for n8 pipeline.
1548         * config/nds32/nds32.md (pipeline_model): Add n8.
1549         * config/nds32/nds32.opt (mcpu): Support n8 pipeline cpus.
1550         * config/nds32/pipelines.md: Include n8 settings.
1551
1552 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
1553             Chung-Ju Wu  <jasonwucj@gmail.com>
1554
1555         * config.gcc (nds32*): Add nds32-utils.o into extra_objs.
1556         * config/nds32/nds32-n9-2r1w.md: New file.
1557         * config/nds32/nds32-n9-3r2w.md: New file.
1558         * config/nds32/nds32-opts.h (nds32_cpu_type, nds32_mul_type,
1559         nds32_register_ports): New or modify for cpu n9.
1560         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n9
1561         pipeline.
1562         * config/nds32/nds32-protos.h: More declarations for n9 pipeline.
1563         * config/nds32/nds32-utils.c: New file.
1564         * config/nds32/nds32.h (TARGET_PIPELINE_N9, TARGET_PIPELINE_SIMPLE,
1565         TARGET_MUL_SLOW): Define.
1566         * config/nds32/nds32.md (pipeline_model): New attribute.
1567         * config/nds32/nds32.opt (mcpu, mconfig-mul, mconfig-register-ports):
1568         New options that support cpu n9.
1569         * config/nds32/pipelines.md: Include n9 settings.
1570         * config/nds32/t-nds32 (nds32-utils.o): Add dependency.
1571
1572 2018-04-08  Chung-Ju Wu  <jasonwucj@gmail.com>
1573
1574         * config/nds32/nds32-md-auxiliary.c (output_cond_branch): Output align
1575         information if necessary.
1576         (output_cond_branch_compare_zero): Likewise.
1577         * config/nds32/nds32.c (nds32_adjust_insn_length): Consider align case.
1578         (nds32_target_alignment): Refine for alignment.
1579         * config/nds32/nds32.h (NDS32_ALIGN_P): Define.
1580         (FUNCTION_BOUNDARY): Modify.
1581         * config/nds32/nds32.md (call_internal, call_value_internal): Consider
1582         align case.
1583         * config/nds32/nds32.opt (malways-align, malign-functions): New.
1584
1585 2018-04-08  Monk Chiang  <sh.chiang04@gmail.com>
1586
1587         * config/nds32/constants.md (unspec_volatile_element): Add values for
1588         TLB operation and data prefetch.
1589         * config/nds32/nds32-intrinsic.c: Implementation of intrinsic
1590         functions for TLB operation and data prefetch.
1591         * config/nds32/nds32-intrinsic.md: Likewise.
1592         * config/nds32/nds32_intrinsic.h: Likewise.
1593         * config/nds32/nds32.c (nds32_dpref_names): Likewise.
1594         (nds32_print_operand): Likewise.
1595         * config/nds32/nds32.h (nds32_builtins): Likewise.
1596
1597 2018-04-07  Thomas Koenig  <tkoenig@gcc.gnu.org>
1598         Andrew Pinski <pinsika@gcc.gnu.org>
1599
1600         PR middle-end/82976
1601         * match.pd: Use constant_boolean_node of correct type instead of
1602         boolean_true_node or boolean_false_node for simplifying
1603         pointer comparisons to zero.
1604
1605 2018-04-07  Jakub Jelinek  <jakub@redhat.com>
1606
1607         PR tree-optimization/80021
1608         * tree.c (verify_type_variant): Make error call in verify_variant_match
1609         translatable and remove final full stop.
1610
1611 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
1612
1613         * config/nds32/constants.md (unspec_volatile_element): Add
1614         UNSPEC_VOLATILE_EH_RETURN.
1615         * config/nds32/nds32-md-auxiliary.c (nds32_output_stack_push,
1616         nds32_output_stack_pop): Support dwarf exception handling process.
1617         * config/nds32/nds32-protos.h (nds32_dynamic_chain_address): Declare.
1618         * config/nds32/nds32.c (nds32_init_machine_status): Support dwarf
1619         exception handling process.
1620         (nds32_compute_stack_frame): Likewise.
1621         (nds32_return_addr_rtx): Likewise.
1622         (nds32_initial_elimination_offset): Likewise.
1623         (nds32_expand_prologue): Likewise.
1624         (nds32_expand_epilogue): Likewise.
1625         (nds32_dynamic_chain_address): New function.
1626         * config/nds32/nds32.h (machine_function): Add fields for dwarf
1627         exception handling.
1628         (DYNAMIC_CHAIN_ADDRESS): Define.
1629         (EH_RETURN_DATA_REGNO): Define.
1630         (EH_RETURN_STACKADJ_RTX): Define.
1631         * config/nds32/nds32.md (eh_return, nds32_eh_return): Implement
1632         patterns for dwarf exception handling.
1633
1634 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
1635
1636         * config/nds32/nds32.h: Clean up obsolete macros.
1637
1638 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
1639
1640         * config/nds32/constants.md (unspec_element, unspec_volatile_element):
1641         Add enum values for particular instructions.
1642         * config/nds32/nds32-intrinsic.c: Implementation of expanding
1643         particular intrinsic functions.
1644         * config/nds32/nds32-intrinsic.md: Likewise.
1645         * config/nds32/nds32_intrinsic.h: Likewise.
1646         * config/nds32/nds32.h (nds32_builtins): Likewise.
1647         * config/nds32/nds32.md (type): Add pbsad and pbsada.
1648         (btst, ave): New patterns for particular instructions.
1649
1650 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
1651
1652         * config/nds32/constants.md (unspec_element, unspec_volatile_element):
1653         Add enum values for atomic load/store and memory sync.
1654         * config/nds32/nds32-intrinsic.c: Implementation for atomic load/store
1655         and memory sync.
1656         * config/nds32/nds32-intrinsic.md: Likewise.
1657         * config/nds32/nds32_intrinsic.h: Likewise.
1658         * config/nds32/nds32.h (nds32_builtins): Likewise.
1659
1660 2018-04-07  Jakub Jelinek  <jakub@redhat.com>
1661
1662         PR tree-optimization/85257
1663         * fold-const.c (native_encode_vector): If not all elts could fit
1664         and off is -1, return 0 rather than offset.
1665         * tree-ssa-sccvn.c (vn_reference_lookup_3): Pass
1666         (offseti - offset2) / BITS_PER_UNIT as 4th argument to
1667         native_encode_expr.  Verify len * BITS_PER_UNIT >= maxsizei.  Don't
1668         adjust buffer in native_interpret_expr call.
1669
1670 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
1671
1672         * config/nds32/constants.md (unspec_volatile_element): Add cache
1673         control enum values.
1674         * config/nds32/nds32-intrinsic.c: Add cache control expand functions.
1675         * config/nds32/nds32-intrinsic.md: Add cache control patterns.
1676         * config/nds32/nds32.c (nds32_cctl_names): New.
1677         (nds32_print_operand): Handle cache control register names.
1678         * config/nds32/nds32.h (nds32_builtins): New enum values.
1679         * config/nds32/nds32_intrinsic.h: Add cache control enum types and
1680         macros.
1681         * config/nds32/nds32.md (type): Add mmu.
1682         * config/nds32/pipelines.md (simple_insn): Add mmu.
1683
1684 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
1685
1686         * config/nds32/nds32.md (type): Remove call.
1687         * config/nds32/pipelines.md (simple_insn): Likewise.
1688
1689 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
1690
1691         * config/nds32/constants.md (unspec_volatile_element): Add
1692         UNSPEC_VOLATILE_FMFCSR, UNSPEC_VOLATILE_FMTCSR and
1693         UNSPEC_VOLATILE_FMFCFG.
1694         * config/nds32/nds32-intrinsic.c (bdesc_noarg): New builtin
1695         description for fmfcfg and fmfcsr.
1696         (bdesc_1arg): Add fmtcsr.
1697         (bdesc_2arg): Add fcpynss, fcpyss, fcpynsd and fcpysd.
1698         (nds32_expand_builtin_impl): Deal with FPU intrinsic functions.
1699         * config/nds32/nds32-intrinsic.md (unspec_fcpynsd, unspec_fcpysd,
1700         unspec_fcpynss, unspec_fcpysd, unspec_fcpyss, unspec_fmfcsr,
1701         unspec_fmfcfg): New patterns.
1702         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_FMFCFG,
1703         NDS32_BUILTIN_FMFCSR, NDS32_BUILTIN_FMTCSR, NDS32_BUILTIN_FCPYNSS,
1704         NDS32_BUILTIN_FCPYSS,NDS32_BUILTIN_FCPYNSD and NDS32_BUILTIN_FCPYSD.
1705         * config/nds32/nds32_intrinsic.h (__nds32__fcpynsd, __nds32__fcpynss,
1706         __nds32__fcpysd, __nds32__fcpyss, __nds32__fmfcsr, __nds32__fmtcsr,
1707         __nds32__fmfcfg): Define.
1708
1709 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
1710
1711         * config/nds32/nds32.c (nds32_intrinsic_register_names): Add more
1712         intrinsic register names.
1713         * config/nds32/nds32_intrinsic.h (nds32_intrinsic_registers): Add more
1714         intrinsic register enum values and macros.
1715
1716 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
1717
1718         * config/nds32/nds32.c (nds32_legitimate_index_p): Modify condition
1719         for load/store addressing form.
1720         (nds32_print_operand_address): Likewise.
1721
1722 2018-04-06  Eric Botcazou  <ebotcazou@adacore.com>
1723
1724         PR target/85196
1725         * config/sparc/sparc.c (sparc_expand_move): Deal with symbolic operands
1726         based on LABEL_REF.  Remove useless assertion.
1727         (pic_address_needs_scratch): Fix formatting.
1728         (sparc_legitimize_pic_address): Minor tweaks.
1729         (sparc_delegitimize_address): Adjust assertion accordingly.
1730         * config/sparc/sparc.md (movsi_pic_label_ref): Change label_ref_operand
1731         into symbolic_operand.
1732         (movsi_high_pic_label_ref): Likewise.
1733         (movsi_lo_sum_pic_label_ref): Likewise.
1734         (movdi_pic_label_ref): Likewise.
1735         (movdi_high_pic_label_ref): Likewise.
1736         (movdi_lo_sum_pic_label_ref): Likewise.
1737
1738 2018-04-06  Amaan Cheval  <amaan.cheval@gmail.com>
1739
1740         * config.gcc (x86_64-*-rtems*): Add rtems.h to tm_file for
1741         custom LIB_SPEC setup.
1742
1743 2018-04-06  Ruslan Bukin  <br@bsdpad.com>
1744             Kito Cheng  <kito.cheng@gmail.com>
1745
1746         * config.gcc (riscv*-*-freebsd*): Add RISC-V FreeBSD support.
1747         * config/riscv/freebsd.h: New.
1748
1749 2018-04-06  Chung-Ju Wu  <jasonwucj@gmail.com>
1750
1751         * config/nds32/nds32.c (nds32_adjust_insn_length): Refine.
1752         * config/nds32/nds32.h (ADJUST_INSN_LENGTH): Change the location in
1753         file.
1754
1755 2018-04-06  Chung-Ju Wu  <jasonwucj@gmail.com>
1756             Kito Cheng  <kito.cheng@gmail.com>
1757
1758         * config/nds32/nds32-md-auxiliary.c (nds32_output_return,
1759         nds32_output_call, nds32_symbol_binds_local_p): New functions.
1760         * config/nds32/nds32-protos.h (nds32_output_call,
1761         nds32_output_return): Declare.
1762         * config/nds32/nds32.md: Refine all the call and return patterns.
1763
1764 2018-04-06  Jakub Jelinek  <jakub@redhat.com>
1765
1766         PR debug/85252
1767         * dwarf2out.c (rtl_for_decl_init): For STRING_CST initializer only
1768         build CONST_STRING if TYPE_MAX_VALUE is non-NULL and is INTEGER_CST.
1769
1770         PR rtl-optimization/84872
1771         * cfgloopmanip.c (create_preheader): Use make_forwarder_block even if
1772         nentry == 1 when CP_FALLTHRU_PREHEADERS and single_entry is
1773         EDGE_CROSSING edge.
1774
1775 2018-04-06  Tamar Christina  <tamar.christina@arm.com>
1776
1777         * expr.c (copy_blkmode_to_reg): Revert 254862.
1778         * doc/sourcebuild.texi (word_mode_no_slow_unalign): Likewise.
1779
1780 2018-04-06  Richard Biener  <rguenther@suse.de>
1781
1782         PR middle-end/85244
1783         * tree-dfa.c (get_ref_base_and_extent): Reset seen_variable_array_ref
1784         after seeing a component reference with an adjacent field.  Treat
1785         refs to arrays at struct end of external decls similar to
1786         refs to unconstrained commons.
1787
1788 2018-04-06  Jakub Jelinek  <jakub@redhat.com>
1789
1790         PR sanitizer/85213
1791         * fold-const.c (twoval_comparison_p): Remove SAVE_P argument and don't
1792         look through SAVE_EXPRs with non-side-effects argument.  Adjust
1793         recursive calls.
1794         (fold_comparison): Adjust twoval_comparison_p caller, don't handle
1795         save_p here.
1796
1797 2018-04-06  Richard Biener  <rguenther@suse.de>
1798
1799         PR middle-end/85180
1800         * alias.c (find_base_term): New wrapper around find_base_term
1801         unwinding CSELIB_VAL_PTR changes.
1802         (find_base_term): Do not restore CSELIB_VAL_PTR during the
1803         recursion.
1804
1805 2018-04-06  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
1806
1807         * config/s390/s390.c (s390_z10_optimize_cmp): Expand dedicated NOP
1808         instructions.
1809         * config/s390/s390.md (UNSPECV_NOP_LR_0, UNSPECV_NOP_LR_1): New
1810         constant definitions.
1811         ("nop"): lr 0,0 -> nopr r0
1812         ("nop_lr0", "nop_lr1"): New insn definitions.
1813
1814 2018-04-06  Chung-Ju Wu  <jasonwucj@gmail.com>
1815
1816         * config/nds32/nds32.md (*stack_push, *stack_pop): Use
1817         NDS32_V3PUSH_AVAILABLE_P macro.
1818
1819 2018-04-06  Monk Chiang  <sh.chiang04@gmail.com>
1820             Chung-Ju Wu  <jasonwucj@gmail.com>
1821
1822         * config.gcc (nds32*-*-*): Add v2j v3f v3s checking.
1823         (nds32*-*-*): Add float and fpu_config into supported_defaults.
1824         * common/config/nds32/nds32-common.c (TARGET_DEFAULT_TARGET_FLAGS):
1825         Include TARGET_DEFAULT_FPU_ISA and TARGET_DEFAULT_FPU_FMA.
1826         * config/nds32/constants.md (unspec_element): Add UNSPEC_COPYSIGN,
1827         UNSPEC_FCPYNSD, UNSPEC_FCPYNSS, UNSPEC_FCPYSD and UNSPEC_FCPYSS.
1828         * config/nds32/constraints.md: New constraints and checking for hard
1829         float configuration.
1830         * config/nds32/iterators.md: New mode iterator and attribute for hard
1831         float configuration.
1832         * config/nds32/nds32-doubleword.md: Use hard float alternatives and
1833         patterns.
1834         * config/nds32/nds32-fpu.md: New file.
1835         * config/nds32/nds32-md-auxiliary.c: New functions and checkings to
1836         deal with hard float code generation.
1837         * config/nds32/nds32-opts.h (nds32_arch_type): Add ARCH_V3F and
1838         ARCH_V3S.
1839         (abi_type, float_reg_number): New enum type.
1840         * config/nds32/nds32-predicates.c: New predicates for hard float.
1841         * config/nds32/nds32-protos.h: Declare functions for hard float.
1842         * config/nds32/nds32.c: Implementation for hard float configuration.
1843         * config/nds32/nds32.h: Definitions for hard float configuration.
1844         * config/nds32/nds32.md: Include hard float machine description and
1845         modify patterns for hard float configuration.
1846         * config/nds32/nds32.opt: New options for hard float configuration.
1847         * config/nds32/predicates.md: New predicates for hard float
1848         configuration.
1849
1850 2018-04-06  Kuan-Lin Chen  <kuanlinchentw@gmail.com>
1851
1852         * common/config/nds32/nds32-common.c
1853         (nds32_option_optimization_table): Enable -mreleax-hint by default.
1854
1855 2018-04-05  Jakub Jelinek  <jakub@redhat.com>
1856
1857         PR middle-end/85195
1858         * match.pd (BIT_FIELD_REF CONSTRUCTOR@0 @1 @2): Use view_convert around
1859         CONSTRUCTOR_ELT (ctor, ...)->value.
1860
1861 2018-04-05  Uros Bizjak  <ubizjak@gmail.com>
1862
1863         PR target/85193
1864         * config/i386/i386.md (define_attr "memory"): Handle rotate1 type.
1865
1866 2018-04-05  Tom de Vries  <tom@codesourcery.com>
1867
1868         PR target/85204
1869         * config/nvptx/nvptx.c (nvptx_single): Fix neutering of bb with only
1870         cond jump.
1871
1872 2018-04-05  Shiva Chen  <shiva0217@gmail.com>
1873             Kito Cheng  <kito.cheng@gmail.com>
1874
1875         * config/nds32/constraints.md (U33): Fine-tune checking condition.
1876         * config/nds32/nds32-md-auxiliary.c (nds32_mem_format): Ditto.
1877         * config/nds32/nds32.h (nds32_16bit_address_type): Add
1878         ADDRESS_POST_MODIFY_LO_REG_IMM3U.
1879
1880 2018-04-05  Shiva Chen  <shiva0217@gmail.com>
1881             Kito Cheng  <kito.cheng@gmail.com>
1882
1883         * config/nds32/constraints.md (Ufe): New memory constraint.
1884         * config/nds32/nds32-md-auxiliary.c (nds32_mem_format,
1885         nds32_output_16bit_load): Consider r8 register for lwi45.fe format.
1886         * config/nds32/nds32.c (nds32_print_operand): Output lwi45.fe
1887         operands.
1888         * config/nds32/nds32.h (nds32_16bit_address_type): Add ADDRESS_R8_IMM7U.
1889         * config/nds32/nds32.md (*mov<mode>): Adjust pattern.
1890
1891 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1892
1893         * config/nds32/nds32.md: Use optimize_size in the condition for
1894         alu-shift instructions.
1895
1896 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1897
1898         * config/nds32/nds32.md (divsi4, udivsi4): New patterns.
1899
1900 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1901
1902         * config/nds32/nds32.md (negsi2): Refine pattern.
1903
1904 2018-04-05  Kito Cheng  <kito.cheng@gmail.com>
1905             Chung-Ju Wu  <jasonwucj@gmail.com>
1906
1907         * config/nds32/iterators.md (shift_rotate): New code iterator.
1908         (shift): New code attribute.
1909         * config/nds32/nds32-md-auxiliary.c (nds32_expand_constant): New.
1910         * config/nds32/nds32-protos.h (nds32_expand_constant): Declare.
1911         * config/nds32/nds32.c (nds32_print_operand): Deal with more cases.
1912         * config/nds32/nds32.md (addsi3, *add_srli): Refine implementation for
1913         bit-wise operations.
1914         (andsi3, *andsi3): Ditto.
1915         (iorsi3, *iorsi3, *or_slli, *or_srli): Ditto.
1916         (xorsi3, *xorsi3, *xor_slli, *xor_srli): Ditto.
1917         (<shift>si3, *ashlsi3, *ashrsi3, *lshrsi3, *rotrsi3): Ditto.
1918         * config/nds32/predicates.md (nds32_rimm5u_operand, nds32_and_operand,
1919         nds32_ior_operand, nds32_xor_operand): New predicates.
1920
1921 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1922
1923         * config/nds32/nds32.md (add<mode>3, sub<mode>3): Rename to ...
1924         (addsi3, subsi3): ... this.
1925
1926 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1927
1928         * config/nds32/nds32.md (*sub_srli, *and_slli): Fine-tune predicator.
1929
1930 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1931
1932         * config/nds32/nds32.md: Adjust indention.
1933
1934 2018-04-05  Kito Cheng  <kito.cheng@gmail.com>
1935
1936         * config/nds32/nds32.md (feature): New attribute.
1937
1938 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1939
1940         * config/nds32/nds32.md (subtype): New attribute.
1941
1942 2018-04-04  Thomas Preud'homme  <thomas.preudhomme@arm.com>
1943
1944         PR target/85203
1945         * config/arm/arm-builtins.c (arm_expand_builtin): Change
1946         expansion to perform a bitwise AND of the argument followed by a
1947         boolean negation of the result.
1948
1949 2018-04-04  Peter Bergner  <bergner@vnet.ibm.com>
1950
1951         PR rtl-optimization/84878
1952         * ddg.c (add_cross_iteration_register_deps): Use DF_REF_BB to determine
1953         the basic block.  Assert the use reference is not artificial and that
1954         it has an associated insn.
1955
1956 2018-04-04  Michael Matz  <matz@suse.de>
1957
1958         * builtins.c (compute_objsize): Pass correct operand
1959         to array_at_struct_end_p.
1960
1961 2018-04-04  Richard Biener  <rguenther@suse.de>
1962
1963         PR lto/85176
1964         * dwarf2out.c (dwarf2out_register_external_die): Peel namespaces
1965         from contexts for DINFO_LEVEL_TERSE and below.
1966
1967 2018-04-04  Kito Cheng  <kito.cheng@gmail.com>
1968
1969         * config/nds32/nds32-doubleword.md (move_<mode>): Require
1970         resiter_operand condition.
1971         * config/nds32/nds32.md (*move<mode>): Ditto.
1972
1973 2018-04-04  Kito Cheng  <kito.cheng@gmail.com>
1974             Monk Chiang  <sh.chiang04@gmail.com>
1975
1976         * config/nds32/nds32.md (movmisalign<mode>): New pattern.
1977
1978 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
1979
1980         * config/nds32/nds32.md (movqi, movhi): Merge into mov<mode>.
1981
1982 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
1983             Kito Cheng  <kito.cheng@gmail.com>
1984
1985         * config/nds32/nds32-md-auxiliary.c (nds32_inverse_cond_code,
1986         nds32_cond_code_str, output_cond_branch,
1987         output_cond_branch_compare_zero, nds32_expand_cbranch,
1988         nds32_expand_cstore, nds32_expand_movcc,
1989         nds32_output_cbranchsi4_equality_zero,
1990         nds32_output_cbranchsi4_equality_reg,
1991         nds32_output_cbranchsi4_equality_reg_or_const_int,
1992         nds32_output_cbranchsi4_greater_less_zero: New functions.
1993         * config/nds32/nds32-protos.h (nds32_expand_cbranch,
1994         nds32_expand_cstore, nds32_expand_movcc,
1995         nds32_output_cbranchsi4_equality_zero,
1996         nds32_output_cbranchsi4_equality_reg,
1997         nds32_output_cbranchsi4_equality_reg_or_const_int,
1998         nds32_output_cbranchsi4_greater_less_zero): Declare.
1999         * config/nds32/predicates.md (nds32_movecc_comparison_operator,
2000         nds32_rimm11s_operand): New predicates.
2001         * config/nds32/nds32.h (nds32_expand_result_type): New enum type.
2002         * config/nds32/nds32.md: Rewrite all the branch and conditional move
2003         patterns.
2004
2005 2018-04-04  Kito Cheng  <kito.cheng@gmail.com>
2006
2007         * config/nds32/nds32-doubleword.md: Refine all the instruction type.
2008         * config/nds32/nds32.md: Ditto.
2009         * config/nds32/pipelines.md: Ditto.
2010
2011 2018-04-04  Richard Biener  <rguenther@suse.de>
2012
2013         PR tree-optimization/85168
2014         * tree-ssa-sccvn.c (vn_reference_maybe_forwprop_address): Avoid
2015         propagating abnormals.
2016
2017 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
2018
2019         * config/nds32/nds32.md (enabled): Use yes/no for this attribute.
2020
2021 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
2022             Kito Cheng  <kito.cheng@gmail.com>
2023
2024         * config/nds32/nds32-md-auxiliary.c (nds32_long_call_p): New function.
2025         * config/nds32/nds32-protos.h (nds32_long_call_p): Declare.
2026         * config/nds32/nds32.c (nds32_function_ok_for_sibcall): New function.
2027         (TARGET_FUNCTION_OK_FOR_SIBCALL): Define.
2028         * config/nds32/nds32.md (sibcall_internal): New.
2029         (sibcall_register): Remove.
2030         (sibcall_immediate): Remove.
2031         (sibcall_value_internal): New.
2032         (sibcall_value_register): Remove.
2033         (sibcall_value_immediate): Remove.
2034         * config/nds32/predicates.md (nds32_general_register_operand): New.
2035         (nds32_call_address_operand): New.
2036
2037 2018-04-03  Jakub Jelinek  <jakub@redhat.com>
2038
2039         PR rtl-optimization/85167
2040         * shrink-wrap.c (move_insn_for_shrink_wrap): Don't set bb_uses and
2041         bb_defs if *split_p, instead preinitialize it to NULL.
2042
2043         PR tree-optimization/85156
2044         * builtins.c (fold_builtin_expect): Use save_expr on arg1 to avoid
2045         evaluating the argument multiple times.
2046
2047 2018-04-03  Bill Schmidt  <wschmidt@linux.ibm.com>
2048
2049         * config/rs6000/emmintrin.h (_mm_cvtpd_epi32): Use __vector rather
2050         than vector.
2051         (_mm_cvtpd_ps): Likewise.
2052         (_mm_cvttpd_epi32): Likewise.
2053         * config/rs6000/mmintrin.h (_mm_unpacklo_pi8): Likewise.
2054         * config/rs6000/xmmintrin.h: For strict-ANSI C++ or C11, undefine
2055         vector, pixel, and bool following altivec.h include.
2056
2057 2018-04-03  Martin Sebor  <msebor@redhat.com>
2058
2059         * doc/extend.texi (Common Function Attributes): Clarify.
2060         (const attribute): Likewise.
2061         (pure attribute): Likewise.
2062
2063 2018-04-03  Jakub Jelinek  <jakub@redhat.com>
2064
2065         PR target/85169
2066         * config/i386/i386.c (ix86_expand_vector_set): Use
2067         HOST_WIDE_INT_1U << elt instead of 1 << elt.  Formatting fix.
2068
2069 2018-04-03  Uros Bizjak  <ubizjak@gmail.com>
2070
2071         * config/i386/i386.c (emit_i387_cw_initialization): Always use logic
2072         instructions when changing rounding bits to preserve precision bits
2073         in the x87 control word.
2074
2075 2018-04-03  Martin Liska  <mliska@suse.cz>
2076
2077         PR tree-optimization/82491
2078         * rtl.h (strip_offset_and_add): Replace += suboffset with
2079         poly_uint64 () + suboffset.
2080
2081 2018-03-29  Martin Liska  <mliska@suse.cz>
2082             Martin Jambor  <mjambor@suse.cz>
2083
2084         PR ipa/84947
2085         * ipa-cp.c (propagate_bits_across_jump_function): Bail out if
2086         param_type is not an integral or pointer type.
2087
2088 2018-04-03  Richard Biener  <rguenther@suse.de>
2089
2090         * sese.h (recompute_all_dominators): Remove.
2091
2092 2018-04-02  Martin Sebor  <msebor@redhat.com>
2093
2094         * doc/invoke.texi (-Wrestrict): Fix typos.
2095
2096 2018-04-02  Jim Wilson  <jimw@sifive.com>
2097
2098         * config/riscv/riscv.h (SHIFT_COUNT_TRUNCATED): Set to zero.
2099         * config/riscv/riscv.md (<optab>si3): Use QImode shift count.
2100         (<optab>di3, <optab>si3_extend): Likewise.
2101         (<optab>si3_mask, <optab>si3_mask_1): New.
2102         (<optab>di3_mask, <optab>di3_mask_1): New.
2103         (<optab>si3_extend_mask, <optab>si3_extend_mask_1): New.
2104         (lshrsi3_zero_extend_1): Use VOIDmode shift count.
2105         * config/riscv/sync.md (atomic_test_and_set): Emit QImode shift count.
2106
2107 2018-04-02  Gerald Pfeifer  <gerald@pfeifer.com>
2108
2109         * doc/cpp.texi (Variadic Macros): Fix line continuation in an
2110         example.
2111
2112 2018-04-02  Chung-Ju Wu  <jasonwucj@gmail.com>
2113
2114         * config/nds32/nds32.c (TARGET_CANONICALIZE_COMPARISON): Define.
2115         (nds32_canonicalize_comparison): New function.
2116
2117 2018-04-01  Chung-Ju Wu  <jasonwucj@gmail.com>
2118             Kito Cheng  <kito.cheng@gmail.com>
2119             Kuan-Lin Chen  <kuanlinchentw@gmail.com>
2120
2121         * config.gcc (nds32): Add nds32-relax-opt.o into extra_objs.
2122         * config/nds32/constants.md (unspec_volatile_element): Add
2123         UNSPEC_VOLATILE_RELAX_GROUP.
2124         * config/nds32/nds32-relax-opt.c: New file.
2125         * config/nds32/nds32-predicates.c
2126         (nds32_symbol_load_store_p): New function.
2127         * config/nds32/nds32-protos.h
2128         (nds32_symbol_load_store_p): Declare function.
2129         (make_pass_nds32_relax_opt): Declare new rtl pass function.
2130         * config/nds32/nds32.c
2131         (nds32_register_pass): New function to register pass.
2132         (nds32_register_passes): New function to register passes.
2133         * config/nds32/nds32.md (relax_group): New pattern.
2134         * config/nds32/nds32.opt (mrelax-hint): New option.
2135         * config/nds32/t-nds32 (nds32-relax-opt.o): New dependency.
2136
2137 2018-04-01  Kito Cheng  <kito.cheng@gmail.com>
2138
2139         * config/nds32/t-nds32: Modify files dependency.
2140
2141 2018-04-01  Chung-Ju Wu  <jasonwucj@gmail.com>
2142
2143         * config/nds32/nds32.h (FUNCTION_PROFILER): Output newline character.
2144         (PROFILE_HOOK): Define its implementation.
2145
2146 2018-04-01  Chung-Ju Wu  <jasonwucj@gmail.com>
2147
2148         * config/nds32/nds32.h (WCHAR_TYPE, WCHAR_TYPE_SIZE): Use unsigned int
2149         type and 32-bit size.
2150
2151 2018-04-01  Jakub Jelinek  <jakub@redhat.com>
2152
2153         PR middle-end/85090
2154         * config/i386/sse.md (V): Add V64QI and V32HI for TARGET_AVX512F.
2155         (V_128_256): New mode iterator.
2156         (*avx512dq_vextract<shuffletype>64x2_1 splitter): New define_split.
2157         (*avx512f_vextract<shuffletype>32x4_1 splitter): Likewise.
2158         (xop_pcmov_<mode><avxsizesuffix>): Use V_128_256 mode iterator instead
2159         of V.
2160         * config/i386/i386.c (ix86_expand_vector_set): Improve V32HImode and
2161         V64QImode expansion for !TARGET_AVX512BW && TARGET_AVX512F.
2162
2163 2018-03-31  Segher Boessenkool  <segher@kernel.crashing.org>
2164
2165         PR target/83315
2166         * config/rs6000/xmmintrin.h (_mm_set_ps, _mm_max_ps): Handle (quiet)
2167         NaN inputs correctly.
2168
2169 2018-03-30  Peter Bergner  <bergner@vnet.ibm.com>
2170
2171         PR target/80546
2172         * config/rs6000/vsx.md (??r): New mode attribute.
2173         (*vsx_mov<mode>_64bit): Use it.
2174         (*vsx_mov<mode>_32bit): Likewise.
2175
2176 2018-03-30  Martin Sebor  <msebor@redhat.com>
2177
2178         PR tree-optimization/84818
2179         * builtins.c (check_access): Use warning_n.
2180
2181 2018-03-30  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
2182
2183         PR target/83822
2184         * config/rs6000/rs6000-string.c (expand_compare_loop): Fix redundant
2185         condition.
2186         * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): Fix redundant
2187         condition.
2188
2189 2018-03-30  Julia Koval  <julia.koval@intel.com>
2190
2191         PR target/84413
2192         * x86-tune.def (movx, partial_reg_dependency): Enable for
2193         m_SKYLAKE_AVX512.
2194
2195 2018-03-29  Vladimir Makarov  <vmakarov@redhat.com>
2196
2197         PR inline-asm/84985
2198         * lra-constraints.c (process_alt_operands): Move setting
2199         this_alternative_matches below.
2200
2201 2018-03-29  Martin Liska  <mliska@suse.cz>
2202
2203         PR lto/84995.
2204         * doc/invoke.texi: Document how LTO works with debug info.
2205         Describe auto-load support of binutils.  Mention 'x86-64'
2206         as valid option value of -march option.
2207
2208 2018-03-29  Jakub Jelinek  <jakub@redhat.com>
2209
2210         * config/i386/sse.md (<avx512>_blendm<mode>): Use <sseintprefix>.
2211
2212         PR c/85094
2213         * fold-const.c (operand_equal_p): Handle DEBUG_BEGIN_STMT.
2214         For STATEMENT_LIST, pass down OEP_LEXICOGRAPHIC and maybe
2215         OEP_NO_HASH_CHECK for recursive call, to avoid exponential
2216         checking.
2217
2218 2018-03-28  Peter Bergner  <bergner@vnet.ibm.com>
2219
2220         PR target/84912
2221         * config/rs6000/rs6000.h (RS6000_BTM_POWERPC64): New define.
2222         (RS6000_BTM_COMMON): Add RS6000_BTM_POWERPC64.
2223         * config/rs6000/rs6000.c (rs6000_builtin_mask_calculate): Add support
2224         for RS6000_BTM_POWERPC64.
2225         (rs6000_invalid_builtin): Add handling for RS6000_BTM_POWERPC64
2226         (rs6000_builtin_mask_names): Add RS6000_BTM_POWERPC64.
2227         * config/rs6000/rs6000-builtin.def (BU_P7_POWERPC64_MISC_2): New macro
2228         definition.
2229         (DIVDE): Use it.
2230         (DIVDEU): Likewise.
2231
2232 2018-03-28 Carl Love  <cel@us.ibm.com>
2233
2234         Revert
2235         2017-09-27  Carl Love  <cel@us.ibm.com>
2236
2237         * config/rs6000/rs6000-builtin.def (BU_FP_1MISC_1): Add define macro.
2238         (FCTID, FCTIW): Add BU_FP_MISC_1 macro expansion for builtins.
2239         * config/rs6000/rs6000.md (lrintsfsi2): Add define_insn for the
2240         fctiw instruction.
2241
2242 2018-03-28  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
2243
2244         * config/rs6000/xmmintrin.h (_mm_max_pi16): Use __vector __bool
2245         instead of __vector bool.
2246         (_mm_max_pu8): Likewise.
2247         (_mm_min_pi16): Likewise.
2248
2249 2018-03-28  Peter Bergner  <bergner@vnet.ibm.com>
2250
2251         PR target/84912
2252         * config/rs6000/rs6000-builtin.def (DIVWEO): Delete macro expansion.
2253         (DIVWEUO): Likewise.
2254         (DIVDEO): Likewise.
2255         (DIVDEUO): Likewise.
2256         * config/rs6000/rs6000.c (builtin_function_type): Remove support for
2257         DIVWEUO and DIVDEUO.
2258         * config/rs6000/rs6000.md (UNSPEC_DIVEO, UNSPEC_DIVEUO): Delete unspecs.
2259         (UNSPEC_DIV_EXTEND): Remove deleted unspecs.
2260         (div_extend): Likewise.
2261         * doc/extend.texi (__builtin_divweo): Remove documentation for deleted
2262         builtin function.
2263         (__builtin_divweuo): Likewise.
2264         (__builtin_divdeo): Likewise.
2265         (__builtin_divdeuo): Likewise.
2266
2267 2018-03-28  Jakub Jelinek  <jakub@redhat.com>
2268
2269         PR target/85095
2270         * config/i386/i386.md (*add<mode>3_carry_0, *addsi3_carry_zext_0,
2271         *sub<mode>3_carry_0, *subsi3_carry_zext_0): New patterns.
2272
2273         PR tree-optimization/82004
2274         * gimple-match-head.c (optimize_pow_to_exp): New function.
2275         * match.pd (pow(C,x) -> exp(log(C)*x)): Wrap with #if GIMPLE.
2276         Don't fold to exp if optimize_pow_to_exp is false.
2277
2278 2018-03-28  Martin Liska  <mliska@suse.cz>
2279
2280         PR other/84819
2281         * calls.c (initialize_argument_information): Fix trailing space.
2282         * common.opt: Fix typo and provide better explanation for
2283         -fsanitize-coverage option.
2284         * config/i386/i386.opt: Fix typo.
2285
2286 2018-03-28  Jakub Jelinek  <jakub@redhat.com>
2287             Martin Liska  <mliska@suse.cz>
2288
2289         PR sanitizer/85081
2290         * gimplify.c (asan_poison_variable): Don't do the check for
2291         gimplify_omp_ctxp here.
2292         (gimplify_decl_expr): Do it here.
2293         (gimplify_target_expr): Likewise.
2294
2295 2018-03-28  Martin Liska  <mliska@suse.cz>
2296
2297         PR target/84988
2298         * config/i386/i386.c (ix86_function_arg_advance): Do not call
2299         chkp_type_bounds_count if MPX is not enabled.
2300
2301 2018-03-27  Chung-Ju Wu  <jasonwucj@gmail.com>
2302
2303         * config/nds32/nds32.h (BRANCH_COST): Adjust cost.
2304
2305 2018-03-27  Michael Meissner  <meissner@linux.vnet.ibm.com>
2306
2307         PR target/84914
2308         * config/rs6000/rs6000.c (create_complex_muldiv): New helper
2309         function to create the function decl for complex long double
2310         multiply and divide for -mabi=ieeelongdouble.
2311         (init_float128_ieee): Call it.
2312
2313 2018-03-27  H.J. Lu  <hongjiu.lu@intel.com>
2314
2315         PR target/85044
2316         * config/i386/i386.c (ix86_trampoline_init): Insert ENDBR for
2317         -fcf-protection=branch -mibt.
2318         * config/i386/i386.h (TRAMPOLINE_SIZE): Increased by 4 bytes.
2319
2320 2018-03-27  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
2321
2322         PR target/81863
2323         * config/arm/arm.c (arm_valid_symbolic_address): Handle arm_word_relocations
2324
2325 2018-03-27  Cesar Philippidis  <cesar@codesourcery.com>
2326
2327         PR target/85056
2328         * config/nvptx/nvptx.c (nvptx_assemble_decl_begin): Add '[]' to
2329         extern array declarations.
2330
2331 2018-03-27  Richard Biener  <rguenther@suse.de>
2332
2333         PR middle-end/84067
2334         * match.pd ((A * C) +- (B * C) -> (A+-B) * C): Guard with
2335         explicit single_use checks.
2336
2337 2018-03-27  Richard Biener  <rguenther@suse.de>
2338
2339         PR tree-optimization/85082
2340         * tree-ssa-sccvn.c (vn_reference_lookup_or_insert_for_pieces):
2341         Valueize the VUSE.
2342
2343 2018-03-27  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
2344
2345         * config.gcc (aarch64*-*-linux*): New TARGET_DEFAULT_ASYNC_UNWIND_TABLES
2346         * common/config/aarch64/aarch64-common.c (aarch64_optimization_table[]):
2347         Turn on fasynchronous-unwind-tables and funwind-tables.
2348
2349 2018-03-26  Uros Bizjak  <ubizjak@gmail.com>
2350
2351         PR target/85073
2352         * config/i386/i386.md (*bmi_blsr_<mode>_cmp): New insn pattern.
2353         (*bmi_blsr_<mode>_ccz): Ditto.
2354
2355 2018-03-26  Tom de Vries  <tom@codesourcery.com>
2356
2357         PR tree-optimization/85063
2358         * omp-general.c (offloading_function_p): New function.  Factor out
2359         of ...
2360         * omp-offload.c (pass_omp_target_link::gate): ... here.
2361         * omp-general.h (offloading_function_p): Declare.
2362         * tree-switch-conversion.c (build_one_array): Mark CSWTCH.x variable
2363         with attribute omp declare target for offloading functions.
2364
2365 2018-03-24  Richard Sandiford  <richard.sandiford@linaro.org>
2366
2367         PR tree-optimization/84005
2368         * tree-data-ref.h (get_base_for_alignment): Declare.
2369         * tree-data-ref.c (get_base_for_alignment_1): New function.
2370         (get_base_for_alignment): Likewise.
2371         * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Use
2372         get_base_for_alignment to find a suitable base object, instead
2373         of always using drb->base_address.
2374
2375 2018-03-23  Jakub Jelinek  <jakub@redhat.com>
2376
2377         PR inline-asm/85022
2378         * emit-rtl.c (init_emit_regs): Indicate that VOIDmode MEMs don't have
2379         known size by default.
2380
2381 2018-03-23  Vladimir Makarov  <vmakarov@redhat.com>
2382
2383         PR inline-asm/85030
2384         * lra-constraints.c (process_alt_operands): Don't match BLKmode
2385         and non BLKmode operands.
2386
2387 2018-03-23  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
2388
2389         PR target/85026
2390         * config/arm/arm.md (unaligned_loadhis): Remove first alternative.
2391         Clean up attributes.
2392
2393 2018-03-23  Richard Biener  <rguenther@suse.de>
2394
2395         PR debug/85020
2396         * dwarf2out.c (rtl_for_decl_location): Do not generate RTL early when
2397         we are going to emit early debug for LTO.
2398
2399 2018-03-23  Jakub Jelinek  <jakub@redhat.com>
2400
2401         PR inline-asm/85034
2402         * function.c (match_asm_constraints_1): Don't optimize if input
2403         doesn't satisfy general_operand predicate for output's mode.
2404
2405         PR inline-asm/85022
2406         * alias.c (write_dependence_p): Don't require for x_canonicalized
2407         non-VOIDmode if x has VOIDmode.
2408
2409         PR sanitizer/85029
2410         * sanopt.c (maybe_optimize_ubsan_ptr_ifn): If DECL_REGISTER (base),
2411         just don't try to optimize it rather than assert it never happens.
2412
2413 2018-03-22  Kelvin Nilsen  <kelvin@gcc.gnu.org>
2414
2415         * config/rs6000/rs6000-builtin.def: Remove various BU_ALTIVEC_X
2416         macro expansions for definition of ST_INTERNAL_<mode> and
2417         LD_INTERNAL_<mode> builtins.
2418         * config/rs6000/rs6000-protos.h (rs6000_address_for_altivec):
2419         Remove prototype.
2420         * config/rs6000/rs6000.c (altivec_expand_ld_builtin): Delete this
2421         function.
2422         (altivec_expand_st_builtin): Likewise.
2423         (altivec_expand_builtin): Remove calls to deleted functions.
2424         (rs6000_address_for_altivec): Delete this function.
2425         * config/rs6000/vector.md: Remove expands for
2426         vector_altivec_load_<mode> and vector_altivec_store_<mode>.
2427
2428 2018-03-22  Sudakshina Das  <sudi.das@arm.com>
2429
2430         PR target/84826
2431         * config/arm/arm.h (machine_function): Add static_chain_stack_bytes.
2432         * config/arm/arm.c (arm_compute_static_chain_stack_bytes): Avoid
2433         re-computing once computed.
2434         (arm_expand_prologue): Compute machine->static_chain_stack_bytes.
2435         (arm_init_machine_status): Initialize
2436         machine->static_chain_stack_bytes.
2437
2438 2018-03-22  Kelvin Nilsen  <kelvin@gcc.gnu.org>
2439
2440         PR target/84760
2441         * doc/extend.texi: Add four new prototypes for vec_ld.
2442         * config/rs6000/rs6000-builtin.def (LVX_V1TI): Reorder symbol
2443         definitions for more logical presentation.
2444         * config/rs6000/rs6000-c.c: (altivec_overloaded_builtins): Add
2445         entries for V1TI variants of __builtin_altivec_ld builtin.
2446         * config/rs6000/rs6000.c: (altivec_expand_lv_builtin): Add test and
2447         handling of V1TI variant of LVX icode pattern.
2448         (altivec_expand_builtin): Add case for ALTIVEC_BUILTIN_LVX_V1TI.
2449         (rs6000_gimple_fold_builtin): Likewise.
2450         (altivec_init_builtins): Add code to define
2451         __builtin_altivec_lvx_v1ti function.
2452
2453 2018-03-22  Jakub Jelinek  <jakub@redhat.com>
2454
2455         PR inline-asm/84941
2456         * function.c (match_asm_constraints_1): Don't do the optimization
2457         if input isn't a REG, SUBREG, MEM or constant.
2458
2459 2018-03-22  Tom de Vries  <tom@codesourcery.com>
2460
2461         PR tree-optimization/84956
2462         * tree-ssa-tail-merge.c (find_clusters_1): Skip bbs with
2463         bb_has_abnormal_pred.
2464
2465 2018-03-22  Jakub Jelinek  <jakub@redhat.com>
2466
2467         PR sanitizer/85018
2468         * dwarf2asm.c (dw2_output_indirect_constant_1): Set
2469         DECL_INITIAL (decl) to decl at the end.
2470         * varasm.c (use_blocks_for_decl_p): Revert the 2018-03-20 change,
2471         adjust the comment.
2472
2473 2018-03-21  Joseph Myers  <joseph@codesourcery.com>
2474
2475         * doc/extend.texi (__builtin_tgmath): Document when complex
2476         integer types are treated as _Complex _Float64.
2477
2478 2018-03-21  Tom de Vries  <tom@codesourcery.com>
2479
2480         * doc/extend.texi (__builtin_extend_pointer): Remove pasto.
2481
2482 2018-03-21  Jakub Jelinek  <jakub@redhat.com>
2483
2484         PR tree-optimization/84960
2485         * tree-cfg.c (remove_bb): Don't move forced labels into bb->prev_bb
2486         if it is ENTRY block, move them into single succ of ENTRY in that case.
2487
2488 2018-03-21  Richard Sandiford  <richard.sandiford@linaro.org>
2489
2490         PR tree-optimization/84811
2491         * poly-int.h (poly_span_traits): Remove the T3 parameter and
2492         promote HOST_WIDE_INT T2 - T1 results to unsigned HOST_WIDE_INT.
2493         (maybe_in_range_p, known_in_range_p, ranges_known_overlap_p):
2494         (known_subrange_p): Update accordingly.  Cast each value involved
2495         in the size comparison, rather than casting the result of the
2496         subtraction.
2497
2498 2018-03-21  Jakub Jelinek  <jakub@redhat.com>
2499
2500         PR tree-optimization/84982
2501         * gimple-ssa-store-merging.c (invert_op): Handle boolean inversion
2502         by flipping the least significant bit rather than all bits from
2503         bitpos to bitpos + bitsize - 1.
2504
2505 2018-03-21  Nathan Sidwell  <nathan@acm.org>
2506
2507         * doc/extend.texi (Deprecated Features): Remove mention of
2508         long-deleted deprecations.
2509
2510 2018-03-21  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
2511
2512         PR jit/84288
2513         * configure.ac (gcc_cv_ld_soname) <*-*-solaris2*>: Set.
2514         * configure: Regenerate.
2515
2516 2018-03-21  Tom de Vries  <tom@codesourcery.com>
2517
2518         PR tree-optimization/83126
2519         * tree-parloops.c (num_phis): New function.
2520         (gen_parallel_loop): Detect and handle canonicalize_loop_ivs failure.
2521
2522 2018-03-21  Nathan Sidwell  <nathan@acm.org>
2523
2524         * doc/extend.texi (Deprecated Features): Update deprecated flags,
2525         mention anon-struct/union members and trailing attributes.
2526
2527 2018-03-21  Bin Cheng  <bin.cheng@arm.com>
2528
2529         PR tree-optimization/84969
2530         * tree-loop-distribution.c (fuse_memset_builtins): Don't reorder
2531         builtin memset partitions if they set different rhs values.
2532
2533 2018-03-21  Jakub Jelinek  <jakub@redhat.com>
2534
2535         PR rtl-optimization/84989
2536         * simplify-rtx.c (simplify_unary_operation_1): Don't try to simplify
2537         VEC_DUPLICATE with scalar result mode.
2538
2539 2018-03-21  Martin Liska  <mliska@suse.cz>
2540
2541         PR ipa/84963
2542         * ipa-icf.c (sem_item_optimizer::fixup_points_to_sets): Remove
2543         not intended return statement.
2544
2545 2018-03-21  Martin Liska  <mliska@suse.cz>
2546
2547         PR target/84988
2548         * tree-chkp.c (CHKP_ARRAY_MAX_CHECK_STEPS): Define a new macro.
2549         (chkp_find_bound_slots_1): Limit number of iterations.
2550
2551 2018-03-20  David H. Gutteridge  <dhgutteridge@sympatico.ca>
2552
2553         PR target/84838
2554         * Minor grammar fixes for x86 options.
2555
2556 2018-03-20  Jakub Jelinek  <jakub@redhat.com>
2557
2558         PR debug/84875
2559         * dce.c (delete_unmarked_insns): Don't remove frame related noop moves
2560         holding REG_CFA_RESTORE notes, instead turn them into a USE.
2561
2562 2018-03-20  Peter Bergner  <bergner@vnet.ibm.com>
2563
2564         PR target/83789
2565         * config/rs6000/altivec.md (altivec_lvx_<mode>_2op): Delete define_insn.
2566         (altivec_lvx_<mode>_1op): Likewise.
2567         (altivec_stvx_<mode>_2op): Likewise.
2568         (altivec_stvx_<mode>_1op): Likewise.
2569         (altivec_lvx_<VM2:mode>): New define_expand.
2570         (altivec_stvx_<VM2:mode>): Likewise.
2571         (altivec_lvx_<VM2:mode>_2op_<P:mptrsize>): New define_insn.
2572         (altivec_lvx_<VM2:mode>_1op_<P:mptrsize>): Likewise.
2573         (altivec_stvx_<VM2:mode>_2op_<P:mptrsize>): Likewise.
2574         (altivec_stvx_<VM2:mode>_1op_<P:mptrsize>): Likewise.
2575         * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Use new expanders.
2576         (rs6000_gen_lvx): Likewise.
2577         * config/rs6000/rs6000.c (altivec_expand_lv_builtin): Likewise.
2578         (altivec_expand_stv_builtin): Likewise.
2579         (altivec_expand_builtin): Likewise.
2580         * config/rs6000/vector.md: Likewise.
2581
2582 2018-03-20  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
2583
2584         PR target/82518
2585         * config/arm/arm.c (arm_array_mode_supported_p): Return false for
2586         BYTES_BIG_ENDIAN.
2587
2588 2018-03-20  Richard Biener  <rguenther@suse.de>
2589
2590         PR target/84986
2591         * config/i386/i386.c (ix86_add_stmt_cost): Only cost
2592         sign-conversions as zero, fall back to standard scalar_stmt
2593         cost for the rest.
2594
2595 2018-03-20  Martin Liska  <mliska@suse.cz>
2596
2597         PR ipa/84825
2598         * predict.c (rebuild_frequencies): Handle case when we have
2599         PROFILE_ABSENT, but flag_guess_branch_prob is false.
2600
2601 2018-03-20  Jakub Jelinek  <jakub@redhat.com>
2602
2603         PR target/84990
2604         * dwarf2asm.c (dw2_output_indirect_constant_1): Temporarily turn off
2605         flag_section_anchors.
2606         * varasm.c (use_blocks_for_decl_p): Remove hack for
2607         dw2_force_const_mem.
2608
2609         PR target/84845
2610         * config/aarch64/aarch64.md (*aarch64_reg_<mode>3_neg_mask2): Rename
2611         to ...
2612         (*aarch64_<optab>_reg_<mode>3_neg_mask2): ... this.  If pseudos can't
2613         be created, use lowpart_subreg of operands[0] rather than operands[0]
2614         itself.
2615         (*aarch64_reg_<mode>3_minus_mask): Rename to ...
2616         (*aarch64_ashl_reg_<mode>3_minus_mask): ... this.
2617         (*aarch64_<optab>_reg_di3_mask2): Use const_int_operand predicate
2618         and n constraint instead of aarch64_shift_imm_di and Usd.
2619         (*aarch64_reg_<optab>_minus<mode>3): Rename to ...
2620         (*aarch64_<optab>_reg_minus<mode>3): ... this.
2621
2622 2018-03-20  Sudakshina Das  <sudi.das@arm.com>
2623
2624         PR target/82989
2625         * config/arm/neon.md (ashldi3_neon): Update ?s for constraints
2626         to favor GPR over NEON registers.
2627         (<shift>di3_neon): Likewise.
2628
2629 2018-03-20  Tom de Vries  <tom@codesourcery.com>
2630
2631         PR target/84952
2632         * config/nvptx/nvptx.c (nvptx_single): Don't neuter bar.sync.
2633         (nvptx_process_pars): Emit bar.sync asap and alap.
2634
2635 2018-03-20  Tom de Vries  <tom@codesourcery.com>
2636
2637         PR target/84954
2638         * config/nvptx/nvptx.c (prevent_branch_around_nothing): Also update
2639         seen_label if seen_label is already set.
2640
2641 2018-03-20  Jakub Jelinek  <jakub@redhat.com>
2642
2643         PR target/84945
2644         * config/i386/i386.c (fold_builtin_cpu): For features above 31
2645         use __cpu_features2 variable instead of __cpu_model.__cpu_features[0].
2646         Use 1U instead of 1.  Formatting fixes.
2647
2648         PR c/84953
2649         * builtins.c (fold_builtin_strpbrk): For strpbrk(x, "") use type
2650         instead of TREE_TYPE (s1) for the return value.
2651
2652 2018-03-19  Jakub Jelinek  <jakub@redhat.com>
2653
2654         PR tree-optimization/84946
2655         * gimple-ssa-store-merging.c (mem_valid_for_store_merging): Compute
2656         bitsize + bitsize in poly_uint64 rather than poly_int64.
2657
2658         PR sanitizer/78651
2659         * dwarf2asm.c: Include fold-const.c.
2660         (dw2_output_indirect_constant_1): Set DECL_INITIAL (decl) to ADDR_EXPR
2661         of decl rather than decl itself.
2662
2663         PR rtl-optimization/84643
2664         * memmodel.h (enum memmodel): Add MEMMODEL_MAX enumerator.
2665
2666 2018-03-19  Maxim Ostapenko  <m.ostapenko@samsung.com>
2667
2668         PR sanitizer/78651
2669         * dwarf2asm.c (dw2_output_indirect_constant_1): Disable ASan before
2670         calling assemble_variable.
2671
2672 2018-03-19  Sudakshina Das  <sudi.das@arm.com>
2673
2674         PR target/81647
2675         * config/aarch64/aarch64-simd.md (vec_cmp<mode><v_int_equiv>): Modify
2676         instructions for UNLT, UNLE, UNGT, UNGE, UNEQ, UNORDERED and ORDERED.
2677
2678 2018-03-19  Jim Wilson  <jimw@sifive.com>
2679
2680         PR bootstrap/84856
2681         * config/riscv/riscv.c (riscv_function_arg_boundary): Use
2682         PREFERRED_STACK_BOUNDARY instead of STACK_BOUNDARY.
2683         (riscv_first_stack_step): Likewise.
2684         (riscv_option_override): Use STACK_BOUNDARY instead of
2685         MIN_STACK_BOUNDARY.
2686         * config/riscv/riscv.h (STACK_BOUNDARY): Renamed from
2687         MIN_STACK_BOUNDARY.
2688         (BIGGEST_ALIGNMENT): Set to 128.
2689         (PREFERRED_STACK_BOUNDARY): Renamed from STACK_BOUNDARY.
2690         (RISCV_STACK_ALIGN): Use PREFERRED_STACK_BOUNDARY instead of
2691         STACK_BOUNDARY.
2692
2693 2018-03-19  Richard Biener  <rguenther@suse.de>
2694
2695         PR tree-optimization/84933
2696         * tree-vrp.c (set_and_canonicalize_value_range): Treat out-of-bound
2697         values as -INF/INF when canonicalizing an ANTI_RANGE to a RANGE.
2698
2699 2018-03-19  Richard Biener  <rguenther@suse.de>
2700
2701         PR tree-optimization/84859
2702         * tree-ssa-phiopt.c (single_trailing_store_in_bb): New function.
2703         (cond_if_else_store_replacement): Perform sinking operation on
2704         single-store BBs regardless of MAX_STORES_TO_SINK setting.
2705         Generalize what a BB with a single eligible store is.
2706
2707 2018-03-19  Richard Biener  <rguenther@suse.de>
2708
2709         PR tree-optimization/84929
2710         * tree-data-ref.c (analyze_siv_subscript_cst_affine): Guard
2711         chrec_is_positive against non-chrec arg.
2712
2713 2018-03-19  Tamar Christina  <tamar.christina@arm.com>
2714
2715         PR target/84711
2716         * config/arm/arm.c (arm_can_change_mode_class): revert r258554.
2717
2718 2018-03-18  Martin Liska  <mliska@suse.cz>
2719
2720         PR rtl-optimization/84635
2721         * regrename.c (build_def_use): Use matches_mode only when
2722         matches >= 0.
2723
2724 2018-03-18  Richard Sandiford  <richard.sandiford@linaro.org>
2725
2726         PR tree-optimization/84913
2727         * tree-vect-loop.c (vectorizable_reduction): Don't try to
2728         vectorize chains of COND_EXPRs.
2729
2730 2018-03-18  Chung-Ju Wu  <jasonwucj@gmail.com>
2731
2732         * config/nds32/nds32.h (MAX_REGS_PER_ADDRESS): Fix the value.
2733
2734 2018-03-18  Chung-Ju Wu  <jasonwucj@gmail.com>
2735
2736         * config/nds32/nds32.h (LOGICAL_OP_NON_SHORT_CIRCUIT): Define.
2737
2738 2018-03-18  Chung-Ju Wu  <jasonwucj@gmail.com>
2739
2740         * config/nds32/nds32.h (CLZ_DEFINED_VALUE_AT_ZERO): Define.
2741
2742 2018-03-17  Chung-Ju Wu  <jasonwucj@gmail.com>
2743             Kito Cheng  <kito.cheng@gmail.com>
2744
2745         * config/nds32/nds32-protos.h (nds32_adjust_reg_alloc_order): Declare.
2746         * config/nds32/nds32.c (nds32_reg_alloc_order_for_speed): New array.
2747         (nds32_adjust_reg_alloc_order): New function.
2748         * config/nds32/nds32.h (ADJUST_REG_ALLOC_ORDER): Define.
2749
2750 2018-03-17  Kito Cheng  <kito.cheng@gmail.com>
2751
2752         * config/nds32/nds32.c (nds32_asm_output_mi_thunk,
2753         nds32_print_operand, nds32_print_operand_address): Use
2754         HOST_WIDE_INT_PRINT_DEC instead.
2755
2756 2018-03-17  Chung-Ju Wu  <jasonwucj@gmail.com>
2757
2758         * config/nds32/nds32.c (nds32_register_priority): Modify cost.
2759
2760 2018-03-17  Jakub Jelinek  <jakub@redhat.com>
2761
2762         PR target/84902
2763         * config/i386/i386.c (initial_ix86_tune_features,
2764         initial_ix86_arch_features): Use unsigned HOST_WIDE_INT rather than
2765         unsigned long long.
2766         (set_ix86_tune_features): Change ix86_tune_mask from unsigned int
2767         to unsigned HOST_WIDE_INT, initialize to HOST_WIDE_INT_1U << ix86_tune
2768         rather than 1u << ix86_tune.  Formatting fix.
2769         (ix86_option_override_internal): Change ix86_arch_mask from
2770         unsigned int to unsigned HOST_WIDE_INT, initialize to
2771         HOST_WIDE_INT_1U << ix86_arch rather than 1u << ix86_arch.
2772         (ix86_function_specific_restore): Likewise.
2773
2774 2018-03-16  Jakub Jelinek  <jakub@redhat.com>
2775
2776         PR target/84899
2777         * postreload.c (reload_combine_recognize_pattern): Perform
2778         INTVAL addition in unsigned HOST_WIDE_INT type to avoid UB and
2779         truncate_int_for_mode the result for the destination's mode.
2780
2781         PR c/84909
2782         * hsa-gen.c (mem_type_for_type): Fix comment typo.
2783         * tree-vect-loop-manip.c (vect_create_cond_for_niters_checks):
2784         Likewise.
2785         * gimple-ssa-warn-restrict.c (builtin_memref::set_base_and_offset):
2786         Likewise.
2787
2788 2018-03-16  Vladimir Makarov  <vmakarov@redhat.com>
2789
2790         PR target/84876
2791         * lra-assigns.c (lra_split_hard_reg_for): Don't use
2792         regno_allocno_class_array and sorted_pseudos.
2793         * lra-constraints.c (spill_hard_reg_in_range): Ignore hard regs in
2794         insns where regno is used.
2795
2796 2018-03-16  Martin Liska  <mliska@suse.cz>
2797
2798         PR ipa/84833
2799         * multiple_target.c (create_dispatcher_calls): Redirect
2800         reference in the symbol table.
2801
2802 2018-03-16  Martin Liska  <mliska@suse.cz>
2803
2804         PR ipa/84722
2805         * multiple_target.c (create_dispatcher_calls): Redirect also
2806         an alias.
2807
2808 2018-03-16  Jakub Jelinek  <jakub@redhat.com>
2809
2810         PR c++/79937
2811         PR c++/82410
2812         * tree.h (TARGET_EXPR_NO_ELIDE): Define.
2813         * gimplify.c (gimplify_modify_expr_rhs): Don't elide TARGET_EXPRs with
2814         TARGET_EXPR_NO_ELIDE flag set unless *expr_p is INIT_EXPR.
2815
2816 2018-03-16  Julia Koval  <julia.koval@intel.com>
2817
2818         * doc/invoke.texi (Skylake Server): Add CLWB.
2819         Cannonlake): Remove CLWB.
2820
2821 2018-03-16  Jakub Jelinek  <jakub@redhat.com>
2822
2823         PR tree-optimization/84841
2824         * tree-ssa-reassoc.c (INTEGER_CONST_TYPE): Change to 1 << 4 from
2825         1 << 3.
2826         (FLOAT_ONE_CONST_TYPE): Define.
2827         (constant_type): Return FLOAT_ONE_CONST_TYPE for -1.0 and 1.0.
2828         (sort_by_operand_rank): Put entries with higher constant_type last
2829         rather than first to match comments.
2830
2831 2018-03-15  Sandra Loosemore  <sandra@codesourcery.com>
2832
2833         * config/nios2/nios2.md (movsi_internal): Fix thinko in
2834         split predicate.
2835
2836 2018-03-15  Jakub Jelinek  <jakub@redhat.com>
2837
2838         PR c++/79085
2839         * calls.c (expand_call): For TREE_ADDRESSABLE rettype ignore alignment
2840         check and use address of target always.
2841
2842 2018-03-15  H.J. Lu  <hongjiu.lu@intel.com>
2843
2844         PR target/84574
2845         * config/i386/i386.c (indirect_thunk_needed): Update comments.
2846         (indirect_thunk_bnd_needed): Likewise.
2847         (indirect_thunks_used): Likewise.
2848         (indirect_thunks_bnd_used): Likewise.
2849         (indirect_return_needed): New.
2850         (indirect_return_bnd_needed): Likewise.
2851         (output_indirect_thunk_function): Add a bool argument for
2852         function return.
2853         (output_indirect_thunk_function): Don't generate alias for
2854         function return thunk.
2855         (ix86_code_end): Call output_indirect_thunk_function to generate
2856         function return thunks.
2857         (ix86_output_function_return): Set indirect_return_bnd_needed
2858         and indirect_return_needed instead of indirect_thunk_bnd_needed
2859         and indirect_thunk_needed.
2860
2861 2018-03-15  Olga Makhotina  <olga.makhotina@intel.com>
2862
2863         * config/i386/sgxintrin.h (_enclv_u32): New intrinsic.
2864         (__enclv_bc, __enclv_cd, __enclv_generic): New definitions.
2865         (ERDINFO, ETRACKC, ELDBC, ELDUC): New leaves.
2866
2867 2018-03-15  David Malcolm  <dmalcolm@redhat.com>
2868             Paul Hua <paul.hua.gm@gmail.com>
2869
2870         PR c/84852
2871         * gcc.dg/fixits-pr84852-1.c: Fix filename in dg-regexp.
2872
2873 2018-03-15  Segher Boessenkool  <segher@kernel.crashing.org>
2874
2875         * config/rs6000/rs6000.c (abi_v4_pass_in_fpr): Add back the
2876         TARGET_DOUBLE_FLOAT and TARGET_SINGLE_FLOAT conditions on the DFmode
2877         resp. SFmode cases.
2878
2879 2018-03-15  Tamar Christina  <tamar.christina@arm.com>
2880
2881         PR target/84711
2882         * config/arm/arm.c (arm_can_change_mode_class): Use GET_MODE_UNIT_SIZE
2883         instead of GET_MODE_SIZE when comparing Units.
2884
2885 2018-03-15  Vladimir Mezentsev  <vladimir.mezentsev@oracle.com>
2886
2887         PR target/68256
2888         * varasm.c (hash_section): Return an unchangeble hash value
2889         * config/aarch64/aarch64.c (aarch64_use_blocks_for_constant_p):
2890         Return !aarch64_can_use_per_function_literal_pools_p ().
2891
2892 2018-03-15  Jakub Jelinek  <jakub@redhat.com>
2893
2894         PR target/84860
2895         * optabs.c (emit_conditional_move): Pass address of cmode's copy
2896         rather than address of cmode as last argument to prepare_cmp_insn.
2897
2898 2018-03-15  Julia Koval  <julia.koval@intel.com>
2899
2900         * config/i386/i386.c (F_AVX512VBMI2, F_GFNI, F_VPCLMULQDQ,
2901         F_AVX512VNNI, F_AVX512BITALG): New.
2902
2903 2018-03-14  John David Anglin  <danglin@gcc.gnu.org>
2904
2905         PR target/83451
2906         * config/pa/pa.c (pa_emit_move_sequence):  Always emit secondary reload
2907         insn for floating-point loads and stores.
2908
2909 2018-03-14  Carl Love  <cel@us.ibm.com>
2910
2911         * config/rs6000/rs6000-c.c: Add macro definitions for
2912         ALTIVEC_BUILTIN_VEC_PERMXOR.
2913         * config/rs6000/rs6000.h: Add #define for vec_permxor builtin.
2914         * config/rs6000/rs6000-builtin.def: Add macro expansions for VPERMXOR.
2915         * config/rs6000/altivec.md (altivec_vpermxor): New define expand.
2916         * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Add case
2917         UNSPEC_VPERMXOR.
2918         * config/doc/extend.texi: Add prototypes for vec_permxor.
2919
2920 2018-03-14  David Malcolm  <dmalcolm@redhat.com>
2921
2922         PR c/84852
2923         * diagnostic-show-locus.c (class layout_point): Convert m_line
2924         from int to linenum_type.
2925         (line_span::comparator): Use linenum "compare" function when
2926         comparing line numbers.
2927         (test_line_span): New function.
2928         (layout_range::contains_point): Convert param "row" from int to
2929         linenum_type.
2930         (layout_range::intersects_line_p): Likewise.
2931         (layout::will_show_line_p): Likewise.
2932         (layout::print_source_line): Likewise.
2933         (layout::should_print_annotation_line_p): Likewise.
2934         (layout::print_annotation_line): Likewise.
2935         (layout::print_leading_fixits): Likewise.
2936         (layout::annotation_line_showed_range_p): Likewise.
2937         (struct line_corrections): Likewise for field m_row.
2938         (line_corrections::line_corrections): Likewise for param "row".
2939         (layout::print_trailing_fixits): Likewise.
2940         (layout::get_state_at_point): Likewise.
2941         (layout::get_x_bound_for_row): Likewise.
2942         (layout::print_line): Likewise.
2943         (diagnostic_show_locus): Likewise for locals "last_line" and
2944         "row".
2945         (selftest::diagnostic_show_locus_c_tests): Call test_line_span.
2946         * input.c (selftest::test_linenum_comparisons): New function.
2947         (selftest::input_c_tests): Call it.
2948         * selftest.c (selftest::test_assertions): Test ASSERT_GT,
2949         ASSERT_GT_AT, ASSERT_LT, and ASSERT_LT_AT.
2950         * selftest.h (ASSERT_GT): New macro.
2951         (ASSERT_GT_AT): New macro.
2952         (ASSERT_LT): New macro.
2953         (ASSERT_LT_AT): New macro.
2954
2955 2018-03-14  Segher Boessenkool  <segher@kernel.crashing.org>
2956
2957         PR rtl-optimization/84780
2958         * combine.c (distribute_links): Don't make a link based on pc_rtx.
2959
2960 2018-03-14  Martin Liska  <mliska@suse.cz>
2961
2962         * tree.c (record_node_allocation_statistics): Use
2963         get_stats_node_kind.
2964         (get_stats_node_kind): New function extracted from
2965         record_node_allocation_statistics.
2966         (free_node): Use get_stats_node_kind.
2967
2968 2018-03-14  Richard Biener  <rguenther@suse.de>
2969
2970         * tree-ssa-pre.c (compute_antic_aux): Remove code that asserts
2971         that the value-set of ANTIC_IN doesn't grow.
2972
2973         Revert
2974         * tree-ssa-pre.c (struct bb_bitmap_sets): Add visited_with_visited_succs
2975         member.
2976         (BB_VISITED_WITH_VISITED_SUCCS): New define.
2977         (compute_antic): Initialize BB_VISITED_WITH_VISITED_SUCCS.
2978
2979 2018-03-14  Julia Koval  <julia.koval@intel.com>
2980
2981         * config.gcc (icelake-client, icelake-server): New.
2982         (icelake): Remove.
2983         * config/i386/i386.c (initial_ix86_tune_features): Extend to 64 bit.
2984         (initial_ix86_arch_features): Ditto.
2985         (PTA_SKYLAKE): Add SGX.
2986         (PTA_ICELAKE): Remove.
2987         (PTA_ICELAKE_CLIENT): New.
2988         (PTA_ICELAKE_SERVER): New.
2989         (ix86_option_override_internal): Split up icelake on icelake client and
2990         icelake server.
2991         (get_builtin_code_for_version): Ditto.
2992         (fold_builtin_cpu): Ditto.
2993         * config/i386/driver-i386.c (config/i386/driver-i386.c): Ditto.
2994         * config/i386/i386-c.c (ix86_target_macros_internal): Ditto
2995         * config/i386/i386.h (processor_type): Ditto.
2996         * doc/invoke.texi: Ditto.
2997
2998 2018-03-14  Jakub Jelinek  <jakub@redhat.com>
2999
3000         PR sanitizer/83392
3001         * sanopt.c (maybe_optimize_ubsan_ptr_ifn): Handle also
3002         INTEGER_CST offset, add it together with bitpos / 8 and
3003         sign extend based on POINTER_SIZE.
3004
3005         PR target/84844
3006         Revert
3007         2017-04-20  Uros Bizjak  <ubizjak@gmail.com>
3008
3009         PR target/78090
3010         * config/i386/constraints.md (Yc): New register constraint.
3011         * config/i386/i386.md (*float<SWI48:mode><MODEF:mode>2_mixed):
3012         Use Yc constraint for alternative 2 of operand 0.  Remove
3013         preferred_for_speed attribute.
3014
3015 2018-03-14  Richard Biener  <rguenther@suse.de>
3016
3017         PR tree-optimization/84830
3018         * tree-ssa-pre.c (compute_antic_aux): Intersect the new ANTIC_IN
3019         with the old one to avoid oscillations.
3020
3021 2018-03-13  Vladimir Makarov  <vmakarov@redhat.com>
3022
3023         PR target/83712
3024         * lra-assigns.c (find_all_spills_for): Ignore uninteresting
3025         pseudos.
3026         (assign_by_spills): Return a flag of reload assignment failure.
3027         Do not process the reload assignment failures.  Do not spill other
3028         reload pseudos if they has the same reg class.  Update n if
3029         necessary.
3030         (lra_assign): Add a return arg.  Set up from the result of
3031         assign_by_spills call.
3032         (find_reload_regno_insns, lra_split_hard_reg_for): New functions.
3033         * lra-constraints.c (split_reg): Add a new arg.  Use it instead of
3034         usage_insns if it is not NULL.
3035         (spill_hard_reg_in_range): New function.
3036         (split_if_necessary, inherit_in_ebb): Pass a new arg to split_reg.
3037         * lra-int.h (spill_hard_reg_in_range, lra_split_hard_reg_for): New
3038         function prototypes.
3039         (lra_assign): Change prototype.
3040         * lra.c (lra): Add code to deal with fails by splitting hard reg
3041         live ranges.
3042
3043 2018-03-01  Palmer Dabbelt  <palmer@sifive.com>
3044
3045         * config/riscv/riscv.opt (mrelax): New option.
3046         * config/riscv/riscv.c (riscv_file_start): Emit ".option
3047         "norelax" when riscv_mrelax is disabled.
3048         * doc/invoke.texi (RISC-V): Document "-mrelax" and "-mno-relax".
3049
3050 2018-03-13  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
3051
3052         PR target/84743
3053         * config/rs6000/rs6000.c (rs6000_reassociation_width): Disable parallel
3054         reassociation for int modes.
3055
3056 2018-03-13  Richard Sandiford  <richard.sandiford@linaro.org>
3057
3058         * tree-vect-loop-manip.c (vect_maybe_permute_loop_masks):
3059         Reverse the choice between VEC_UNPACK_LO_EXPR and VEC_UNPACK_HI_EXPR
3060         for big-endian.
3061         * config/aarch64/iterators.md (hi_lanes_optab): New int attribute.
3062         * config/aarch64/aarch64-sve.md
3063         (*aarch64_sve_<perm_insn><perm_hilo><mode>): Rename to...
3064         (aarch64_sve_<perm_insn><perm_hilo><mode>): ...this.
3065         (*extend<mode><Vwide>2): Rename to...
3066         (aarch64_sve_extend<mode><Vwide>2): ...this.
3067         (vec_unpack<su>_<perm_hilo>_<mode>): Turn into a define_expand,
3068         renaming the old pattern to...
3069         (aarch64_sve_punpk<perm_hilo>_<mode>): ...this.  Only define
3070         unsigned packs.
3071         (vec_unpack<su>_<perm_hilo>_<SVE_BHSI:mode>): Turn into a
3072         define_expand, renaming the old pattern to...
3073         (aarch64_sve_<su>unpk<perm_hilo>_<SVE_BHSI:mode>): ...this.
3074         (*vec_unpacku_<perm_hilo>_<mode>_no_convert): Delete.
3075         (vec_unpacks_<perm_hilo>_<mode>): Take BYTES_BIG_ENDIAN into
3076         account when deciding which SVE instruction the optab should use.
3077         (vec_unpack<su_optab>_float_<perm_hilo>_vnx4si): Likewise.
3078
3079 2018-03-13  Richard Sandiford  <richard.sandiford@linaro.org>
3080
3081         * config/aarch64/aarch64.md (V4_REGNUM, V8_REGNUM, V12_REGNUM)
3082         (V20_REGNUM, V24_REGNUM, V28_REGNUM, P1_REGNUM, P2_REGNUM, P3_REGNUM)
3083         (P4_REGNUM, P5_REGNUM, P6_REGNUM, P8_REGNUM, P9_REGNUM, P10_REGNUM)
3084         (P11_REGNUM, P12_REGNUM, P13_REGNUM, P14_REGNUM): New define_constants.
3085         (tlsdesc_small_<mode>): Turn a define_expand and use
3086         tlsdesc_small_sve_<mode> for SVE.  Rename original define_insn to...
3087         (tlsdesc_small_advsimd_<mode>): ...this.
3088         (tlsdesc_small_sve_<mode>): New pattern.
3089
3090 2018-03-13  Richard Sandiford  <richard.sandiford@linaro.org>
3091
3092         * config/aarch64/iterators.md (UNSPEC_SMUL_HIGHPART)
3093         (UNSPEC_UMUL_HIGHPART): New constants.
3094         (MUL_HIGHPART): New int iteraor.
3095         (su): Handle UNSPEC_SMUL_HIGHPART and UNSPEC_UMUL_HIGHPART.
3096         * config/aarch64/aarch64-sve.md (<su>mul<mode>3_highpart): New
3097         define_expand.
3098         (*<su>mul<mode>3_highpart): New define_insn.
3099
3100 2018-03-13  Eric Botcazou  <ebotcazou@adacore.com>
3101
3102         PR lto/84805
3103         * ipa-devirt.c (odr_subtypes_equivalent_p): Do not get the ODR type of
3104         incomplete types.
3105
3106 2018-03-13  Martin Liska  <mliska@suse.cz>
3107
3108         PR ipa/84658.
3109         * (sem_item_optimizer::sem_item_optimizer): Initialize new
3110         vector.
3111         (sem_item_optimizer::~sem_item_optimizer): Release it.
3112         (sem_item_optimizer::merge_classes): Register variable aliases.
3113         (sem_item_optimizer::fixup_pt_set): New function.
3114         (sem_item_optimizer::fixup_points_to_sets): Likewise.
3115         * ipa-icf.h: Declare new variables and functions.
3116
3117 2018-03-13  Jakub Jelinek  <jakub@redhat.com>
3118
3119         PR middle-end/84834
3120         * match.pd ((A & C) != 0 ? D : 0): Use INTEGER_CST@2 instead of
3121         integer_pow2p@2 and test integer_pow2p in condition.
3122         (A < 0 ? C : 0): Similarly for @1.
3123
3124         PR middle-end/84831
3125         * stmt.c (parse_output_constraint): If the CONSTRAINT_LEN (*p, p)
3126         characters starting at p contain '\0' character, don't look beyond
3127         that.
3128
3129         PR target/84827
3130         * config/i386/i386.md (round<mode>2): For 387 fancy math, disable
3131         pattern if -ftrapping-math -fno-fp-int-builtin-inexact.
3132
3133         PR target/84828
3134         * reg-stack.c (change_stack): Change update_end var from int to
3135         rtx_insn *, if non-NULL don't update just BB_END (current_block), but
3136         also call set_block_for_insn on the newly added insns and rescan.
3137
3138         PR target/84786
3139         * config/i386/sse.md (sse2_loadhpd): Use Yv constraint rather than v
3140         on the last operand.
3141
3142         PR c++/84704
3143         * tree.c (stabilize_reference_1): Return save_expr (e) for
3144         STATEMENT_LIST even if it doesn't have side-effects.
3145
3146 2018-03-12  Jonathan Wakely  <jwakely@redhat.com>
3147
3148         * doc/invoke.texi (-mclflushopt): Fix spelling of option.
3149
3150 2018-03-12  Renlin Li  <renlin.li@arm.com>
3151
3152         * config/aarch64/aarch64.md (movhf_aarch64): Fix mode argument to
3153         aarch64_output_scalar_simd_mov_immediate.
3154
3155 2018-03-12  Martin Sebor  <msebor@redhat.com>
3156
3157         PR tree-optimization/83456
3158         * gimple-fold.c (gimple_fold_builtin_memory_op): Avoid warning
3159         for perfectly overlapping calls to memcpy.
3160         (gimple_fold_builtin_memory_chk): Same.
3161         (gimple_fold_builtin_strcpy): Handle no-warning.
3162         (gimple_fold_builtin_stxcpy_chk): Same.
3163         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Handle no-warning.
3164
3165 2018-03-12  Segher Boessenkool  <segher@kernel.crashing.org>
3166
3167         * config/rs6000/rs6000.c (abi_v4_pass_in_fpr): Add bool "named"
3168         parameter.  Use it for SFmode.
3169         (rs6000_function_arg_advance_1): Adjust.
3170         (rs6000_function_arg): Adjust.
3171         (rs6000_gimplify_va_arg): Pass false for that new parameter.
3172
3173 2018-03-12  Segher Boessenkool  <segher@kernel.crashing.org>
3174
3175         PR rtl-optimization/84169
3176         PR rtl-optimization/84780
3177         * combine.c (can_combine_p): Check for a 2-insn combination whether
3178         the destination register is used between the two insns, too.
3179
3180 2018-03-12  Richard Biener  <rguenther@suse.de>
3181
3182         PR tree-optimization/84803
3183         * tree-if-conv.c (ifcvt_memrefs_wont_trap): Don't do anything
3184         for refs DR analysis didn't process.
3185
3186 2018-03-12  Richard Biener  <rguenther@suse.de>
3187
3188         PR tree-optimization/84777
3189         * tree-ssa-loop-ch.c (should_duplicate_loop_header_p): For
3190         force-vectorize loops ignore whether we are optimizing for size.
3191
3192 2018-03-12  Chung-Ju Wu  <jasonwucj@gmail.com>
3193
3194         * config/nds32/nds32.c (nds32_md_asm_adjust): New function.
3195         (TARGET_MD_ASM_ADJUST): Define.
3196
3197 2018-03-12  Monk Chiang  <sh.chiang04@gmail.com>
3198             Kito Cheng  <kito.cheng@gmail.com>
3199             Chung-Ju Wu  <jasonwucj@gmail.com>
3200
3201         * config/nds32/nds32.c (nds32_compute_stack_frame,
3202         nds32_emit_stack_push_multiple, nds32_emit_stack_pop_multiple,
3203         nds32_emit_stack_v3push, nds32_emit_stack_v3pop,
3204         nds32_emit_adjust_frame, nds32_expand_prologue, nds32_expand_epilogue,
3205         nds32_expand_prologue_v3push, nds32_expand_epilogue_v3pop): Refine.
3206         * config/nds32/nds32.h (NDS32_FIRST_CALLEE_SAVE_GPR_REGNUM,
3207         NDS32_LAST_CALLEE_SAVE_GPR_REGNUM, NDS32_V3PUSH_AVAILABLE_P): New.
3208         * config/nds32/nds32.md (prologue, epilogue): Use macro
3209         NDS32_V3PUSH_AVAILABLE_P to do checking.
3210
3211 2018-03-11  Jakub Jelinek  <jakub@redhat.com>
3212
3213         PR debug/58150
3214         * dwarf2out.c (gen_enumeration_type_die): Don't guard adding
3215         DW_AT_declaration for ENUM_IS_OPAQUE on -gdwarf-4 or -gno-strict-dwarf,
3216         but on TYPE_SIZE.  Don't do anything for ENUM_IS_OPAQUE if not creating
3217         a new die.  Don't set TREE_ASM_WRITTEN if ENUM_IS_OPAQUE.  Guard
3218         addition of most attributes on !orig_type_die or the attribute not
3219         being present already.  Assert TYPE_VALUES is NULL for ENUM_IS_OPAQUE.
3220
3221 2018-03-11  Kito Cheng  <kito.cheng@gmail.com>
3222             Chung-Ju Wu  <jasonwucj@gmail.com>
3223
3224         * config/nds32/nds32.c (nds32_cpu_cpp_builtins): Modify to define
3225         __NDS32_VH__ macro.
3226         * config/nds32/nds32.opt (mvh): New option.
3227
3228 2018-03-11  Kito Cheng  <kito.cheng@gmail.com>
3229             Chung-Ju Wu  <jasonwucj@gmail.com>
3230
3231         * config/nds32/nds32-protos.h (nds32_cpu_cpp_builtins): Declare
3232         function.
3233         * config/nds32/nds32.c (nds32_cpu_cpp_builtins): New function.
3234         * config/nds32/nds32.h (TARGET_CPU_CPP_BUILTINS): Modify its
3235         definition.
3236
3237 2018-03-11  Kito Cheng  <kito.cheng@gmail.com>
3238             Chung-Ju Wu  <jasonwucj@gmail.com>
3239
3240         * config/nds32/nds32-memory-manipulation.c (nds32_expand_strlen): New
3241         function.
3242         * config/nds32/nds32-multiple.md (strlensi): New pattern.
3243         * config/nds32/nds32-protos.h (nds32_expand_strlen): Declare function.
3244
3245 2018-03-11  Monk Chiang  <sh.chiang04@gmail.com>
3246             Kito Cheng  <kito.cheng@gmail.com>
3247             Chung-Ju Wu  <jasonwucj@gmail.com>
3248
3249         * config/nds32/constants.md (unspec_element): Add UNSPEC_FFB,
3250         UNSPEC_FFMISM and UNSPEC_FLMISM.
3251         * config/nds32/nds32-intrinsic.c (bdesc_2arg): Add builtin description
3252         for ffb, ffmism and flmism.
3253         * config/nds32/nds32-intrinsic.md (unspec_ffb): Define new pattern.
3254         (unspec_ffmism): Ditto.
3255         (unspec_flmism): Ditto.
3256         (nds32_expand_builtin_impl): Check if string extension is available.
3257         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_FFB,
3258         NDS32_BUILTIN_FFMISM and NDS32_BUILTIN_FLMISM.
3259
3260 2018-03-10  Vladimir Makarov  <vmakarov@redhat.com>
3261
3262         Reverting patch:
3263         2018-03-09  Vladimir Makarov  <vmakarov@redhat.com>
3264
3265         PR target/83712
3266         * lra-assigns.c (assign_by_spills): Return a flag of reload
3267         assignment failure.  Do not process the reload assignment
3268         failures.  Do not spill other reload pseudos if they has the same
3269         reg class.
3270         (lra_assign): Add a return arg.  Set up from the result of
3271         assign_by_spills call.
3272         (find_reload_regno_insns, lra_split_hard_reg_for): New functions.
3273         * lra-constraints.c (split_reg): Add a new arg.  Use it instead of
3274         usage_insns if it is not NULL.
3275         (spill_hard_reg_in_range): New function.
3276         (split_if_necessary, inherit_in_ebb): Pass a new arg to split_reg.
3277         * lra-int.h (spill_hard_reg_in_range, lra_split_hard_reg_for): New
3278         function prototypes.
3279         (lra_assign): Change prototype.
3280         * lra.c (lra): Add code to deal with fails by splitting hard reg
3281         live ranges.
3282
3283 2018-03-10  H.J. Lu  <hongjiu.lu@intel.com>
3284
3285         PR target/84807
3286         * config/i386/i386.opt: Replace Enforcment with Enforcement.
3287
3288 2018-03-10  Alexandre Oliva  <aoliva@redhat.com>
3289
3290         PR debug/84620
3291         * dwarf2out.h (dw_val_class): Add dw_val_class_symview.
3292         (dw_val_node): Add val_symbolic_view.
3293         * dwarf2out.c (dw_line_info_table): Add symviews_since_reset.
3294         (symview_upper_bound): New.
3295         (new_line_info_table): Initialize symviews_since_reset.
3296         (dwarf2out_source_line): Count symviews_since_reset and set
3297         symview_upper_bound.
3298         (dw_val_equal_p): Handle symview.
3299         (add_AT_symview): New.
3300         (print_dw_val): Handle symview.
3301         (attr_checksum, attr_checksum_ordered): Likewise.
3302         (same_dw_val_p, size_of_die): Likewise.
3303         (value_format, output_die): Likewise.
3304         (add_high_low_attributes): Use add_AT_symview for entry_view.
3305         (dwarf2out_finish): Reset symview_upper_bound, clear
3306         zero_view_p.
3307
3308 2018-03-09  Peter Bergner  <bergner@vnet.ibm.com>
3309
3310         PR target/83969
3311         * config/rs6000/rs6000.c (rs6000_offsettable_memref_p): New prototype.
3312         Add strict argument and use it.
3313         (rs6000_split_multireg_move): Update for new strict argument.
3314         (mem_operand_gpr): Disallow all non-offsettable addresses.
3315         * config/rs6000/rs6000.md (*movdi_internal64): Use YZ constraint.
3316
3317 2018-03-09  Jakub Jelinek  <jakub@redhat.com>
3318
3319         PR target/84772
3320         * config/rs6000/rs6000.c (rs6000_gimplify_va_arg): Mark va_arg_tmp
3321         temporary TREE_ADDRESSABLE before gimplification of BUILT_IN_MEMCPY.
3322         * config/powerpcspe/powerpcspe.c (rs6000_gimplify_va_arg): Likewise.
3323
3324         PR c++/84767
3325         * tree-inline.c (copy_tree_body_r): For INDIRECT_REF of a remapped
3326         decl, use remap_type if we want to use the type.
3327
3328 2018-03-09  Martin Sebor  <msebor@redhat.com>
3329
3330         PR tree-optimization/84526
3331         * gimple-ssa-warn-restrict.c (builtin_memref::set_base_and_offset):
3332         Remove dead code.
3333         (builtin_access::generic_overlap): Be prepared to handle non-array
3334         base objects.
3335
3336 2018-03-09  Alexandre Oliva  <aoliva@redhat.com>
3337
3338         PR rtl-optimization/84682
3339         * lra-constraints.c (process_address_1): Check is_address flag
3340         for address constraints.
3341         (process_alt_operands): Likewise.
3342         * lra.c (lra_set_insn_recog_data): Pass asm operand locs to
3343         preprocess_constraints.
3344         * recog.h (preprocess_constraints): Add oploc parameter.
3345         Adjust callers.
3346         * recog.c (preprocess_constraints): Test address_operand for
3347         CT_ADDRESS constraints.
3348
3349 2018-03-09  Vladimir Makarov  <vmakarov@redhat.com>
3350
3351         PR target/83712
3352         * lra-assigns.c (assign_by_spills): Return a flag of reload
3353         assignment failure.  Do not process the reload assignment
3354         failures.  Do not spill other reload pseudos if they has the same
3355         reg class.
3356         (lra_assign): Add a return arg.  Set up from the result of
3357         assign_by_spills call.
3358         (find_reload_regno_insns, lra_split_hard_reg_for): New functions.
3359         * lra-constraints.c (split_reg): Add a new arg.  Use it instead of
3360         usage_insns if it is not NULL.
3361         (spill_hard_reg_in_range): New function.
3362         (split_if_necessary, inherit_in_ebb): Pass a new arg to split_reg.
3363         * lra-int.h (spill_hard_reg_in_range, lra_split_hard_reg_for): New
3364         function prototypes.
3365         (lra_assign): Change prototype.
3366         * lra.c (lra): Add code to deal with fails by splitting hard reg
3367         live ranges.
3368
3369 2018-03-09  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
3370
3371         PR target/83193
3372         * common/config/arm/arm-common.c (arm_parse_arch_option_name):
3373         Accept complain bool parameter.  Only emit errors if it is true.
3374         (arm_parse_cpu_option_name): Likewise.
3375         (arm_target_thumb_only): Adjust callers of the above.
3376         * config/arm/arm-protos.h (arm_parse_cpu_option_name): Adjust
3377         prototype to take a default true bool parameter.
3378         (arm_parse_arch_option_name): Likewise.
3379
3380 2018-03-09  David Malcolm  <dmalcolm@redhat.com>
3381             Francois-Xavier Coudert  <fxcoudert@gcc.gnu.org>
3382
3383         PR jit/64089
3384         PR jit/84288
3385         * Makefile.in (LD_VERSION_SCRIPT_OPTION, LD_SONAME_OPTION): New.
3386         * configure: Regenerate.
3387         * configure.ac ("linker --version-script option"): New.
3388         ("linker soname option"): New.
3389
3390 2018-03-09  Richard Biener  <rguenther@suse.de>
3391
3392         PR tree-optimization/84775
3393         * tree-if-conv.c (add_bb_predicate_gimplified_stmts): Delink
3394         immediate uses of predicate stmts and mark them modified.
3395
3396         Revert
3397         PR tree-optimization/84178
3398         * tree-if-conv.c (combine_blocks): Move insert_gimplified_predicates
3399         to caller.
3400         (version_loop_for_if_conversion): Delay update_ssa call.
3401         (tree_if_conversion): Delay update_ssa until after predicate
3402         insertion.
3403
3404 2018-03-09  Eric Botcazou  <ebotcazou@adacore.com>
3405
3406         PR target/84763
3407         * config/i386/winnt.c (i386_pe_seh_cold_init): Use small pre-allocation
3408         when the function accesses prior frames.
3409
3410 2018-03-08  Jakub Jelinek  <jakub@redhat.com>
3411
3412         PR debug/84456
3413         * dwarf2out.c (dw_loc_list): If list && loc_list->first->next, call
3414         gen_llsym, otherwise call maybe_gen_llsym.
3415
3416         PR inline-asm/84742
3417         * recog.c (asm_operand_ok): Return 0 if multi-character constraint
3418         has ',' character inside of it.
3419
3420 2018-03-08  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
3421
3422         PR target/84748
3423         * config/aarch64/aarch64.md (*compare_cstore<mode>_insn): Mark pattern
3424         as clobbering CC_REGNUM.
3425
3426 2018-03-08  Richard Biener  <rguenther@suse.de>
3427
3428         PR middle-end/84552
3429         * tree-scalar-evolution.c: Include tree-into-ssa.h.
3430         (follow_copies_to_constant): Do not follow SSA names registered
3431         for update.
3432
3433 2018-03-08  Richard Biener  <rguenther@suse.de>
3434
3435         PR tree-optimization/84178
3436         * tree-if-conv.c (combine_blocks): Move insert_gimplified_predicates
3437         to caller.
3438         (version_loop_for_if_conversion): Delay update_ssa call.
3439         (tree_if_conversion): Delay update_ssa until after predicate
3440         insertion.
3441
3442 2018-03-08  David Malcolm  <dmalcolm@redhat.com>
3443
3444         PR tree-optimization/84178
3445         * tree-if-conv.c (release_bb_predicate): Remove the
3446         the assertion that the stmts have NULL use_ops.
3447         Discard the statements, asserting that they haven't
3448         yet been added to a BB.
3449
3450 2018-03-08  Richard Biener  <rguenther@suse.de>
3451
3452         PR tree-optimization/84746
3453         * tree-ssa-pre.c (find_leader_in_sets): Deal with SET1 being NULL.
3454         (phi_translate): Pass in destination ANTIC_OUT set.
3455         (phi_translate_1): Likewise.  For a simplified result lookup
3456         a leader in ANTIC_OUT and AVAIL_OUT, not the ANTIC_IN sets.
3457         (phi_translate_set): Adjust.
3458         (do_pre_regular_insertion): Likewise.
3459         (do_pre_partial_partial_insertion): Likewise.
3460
3461 2018-03-08  Martin Liska  <mliska@suse.cz>
3462
3463         PR gcov-profile/84735
3464         * doc/gcov.texi: Document usage of profile files.
3465         * gcov-io.h: Document changes in the format.
3466
3467 2018-03-08  Alexandre Oliva  <aoliva@redhat.com>
3468
3469         PR debug/84404
3470         PR debug/84408
3471         * dwarf2out.c (struct dw_line_info_table): Update comments for
3472         view == -1.
3473         (FORCE_RESET_NEXT_VIEW): New.
3474         (FORCE_RESETTING_VIEW_P): New.
3475         (RESETTING_VIEW_P): Check for -1 too.
3476         (ZERO_VIEW_P): Likewise.
3477         (new_line_info_table): Force-reset next view.
3478         (dwarf2out_begin_function): Likewise.
3479         (dwarf2out_source_line): Simplify zero_view_p initialization.
3480         Test FORCE_RESETTING_VIEW_P and RESETTING_VIEW_P instead of
3481         view directly.  Omit view when omitting .loc at line 0.
3482
3483 2018-03-08  Jakub Jelinek  <jakub@redhat.com>
3484
3485         PR tree-optimization/84740
3486         * tree-switch-conversion.c (process_switch): Call build_constructors
3487         only if info.phi_count is non-zero.
3488
3489         PR tree-optimization/84739
3490         * tree-tailcall.c (find_tail_calls): Check call arguments against
3491         DECL_ARGUMENTS (current_function_decl) rather than
3492         DECL_ARGUMENTS (func) when checking for tail recursion.
3493
3494 2018-03-07  Jakub Jelinek  <jakub@redhat.com>
3495
3496         * doc/contrib.texi: Add entries for Martin Liska, David Malcolm,
3497         Marek Polacek, extend Vladimir Makarov's, Jonathan Wakely's and
3498         Volker Reichelt's entry and add entries for people that perform
3499         GCC fuzzy testing and report numerous bugs.
3500
3501 2018-03-07  Segher Boessenkool  <segher@kernel.crashing.org>
3502
3503         PR target/82411
3504         * config/rs6000/rs6000.c (rs6000_elf_in_small_data_p): Don't put
3505         readonly data in sdata, if that is disabled.
3506         * config/rs6000/sysv4.opt (mreadonly-in-sdata): New option.
3507         * doc/invoke.texi (RS/6000 and PowerPC Options): Document
3508         -mreadonly-in-sdata option.
3509
3510 2018-03-07  Martin Sebor  <msebor@redhat.com>
3511
3512         PR tree-optimization/84468
3513         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Consider successor
3514         basic block when looking for nul assignment.
3515
3516 2018-03-07  Eric Botcazou  <ebotcazou@adacore.com>
3517
3518         PR target/84277
3519         * except.h (output_function_exception_table): Adjust prototype.
3520         * except.c (output_function_exception_table): Remove FNNAME parameter
3521         and add SECTION parameter.  Ouput one part of the table at a time.
3522         * final.c (final_scan_insn_1) <NOTE_INSN_SWITCH_TEXT_SECTIONS>: Output
3523         the first part of the exception table and emit unwind directives.
3524         * config/i386/i386-protos.h (i386_pe_end_cold_function): Declare.
3525         (i386_pe_seh_cold_init): Likewise.
3526         * config/i386/cygming.h (ASM_DECLARE_COLD_FUNCTION_NAME): New macro.
3527         (ASM_DECLARE_COLD_FUNCTION_SIZE): Likewise.
3528         * config/i386/i386.c (x86_expand_epilogue): Fix wording in comment.
3529         (ix86_output_call_insn): Emit a nop in one more case for SEH.
3530         * config/i386/winnt.c: Include except.h.
3531         (struct seh_frame_state): Add reg_offset, after_prologue and
3532         in_cold_section fields.
3533         (i386_pe_seh_end_prologue): Set seh->after_prologue.
3534         (i386_pe_seh_cold_init): New function.
3535         (i386_pe_seh_fini): Add COLD parameter and bail out if it is not equal
3536         to seh->in_cold_section.
3537         (seh_emit_push): Record the offset of the push.
3538         (seh_emit_save): Record the offet of the save.
3539         (i386_pe_seh_unwind_emit): Deal with NOTE_INSN_SWITCH_TEXT_SECTIONS.
3540         Test seh->after_prologue to disregard the epilogue.
3541         (i386_pe_end_function): Pass FALSE to i386_pe_seh_fini.
3542         (i386_pe_end_cold_function): New function.
3543
3544 2018-03-07  Jakub Jelinek  <jakub@redhat.com>
3545
3546         PR fortran/84565
3547         * config/aarch64/predicates.md (aarch64_simd_reg_or_zero): Use
3548         aarch64_simd_or_scalar_imm_zero rather than aarch64_simd_imm_zero.
3549
3550         PR c++/84704
3551         * gimple-expr.c (create_tmp_var_raw): Set DECL_NAMELESS flag
3552         on tmp_var.
3553         * tree-pretty-print.c (dump_decl_name): For TDF_COMPARE_DEBUG,
3554         don't print names of DECL_NAMELESS DECL_IGNORED_P decls.
3555
3556         PR middle-end/84723
3557         * multiple_target.c: Include tree-inline.h and intl.h.
3558         (expand_target_clones): Diagnose and fail if node->definition and
3559         !tree_versionable_function_p (node->decl).
3560
3561 2018-03-06  John David Anglin  <danglin@gcc.gnu.org>
3562
3563         * config/pa/pa.h (ASM_GENERATE_INTERNAL_LABEL): Revise to use
3564         sprint_ul.
3565         (ASM_OUTPUT_ADDR_VEC_ELT): Revise for above change.
3566         (ASM_OUTPUT_ADDR_DIFF_ELT): Likewise.
3567         * config/pa/pa64-hpux.h (ASM_GENERATE_INTERNAL_LABEL): Revise as above.
3568
3569 2018-03-06  Jakub Jelinek  <jakub@redhat.com>
3570
3571         PR target/84710
3572         * combine.c (try_combine): Use reg_or_subregno instead of handling
3573         just paradoxical SUBREGs and REGs.
3574
3575 2018-03-06  Claudiu Zissulescu  <claziss@synopsys.com>
3576
3577          * config/arc/arc.c (arc_finalize_pic): Remove function.
3578          (arc_must_save_register): We use single base PIC register, remove
3579          checks to save/restore the PIC register.
3580          (arc_expand_prologue): Likewise.
3581          * config/arc/arc-protos.h (arc_set_default_type_attributes):
3582          Remove.
3583          (arc_verify_short): Likewise.
3584          (arc_attr_type): Likewise.
3585          * config/arc/arc.c (arc_set_default_type_attributes): Remove.
3586          (walk_stores): Likewise.
3587          (arc_address_cost): Make it static.
3588          (arc_verify_short): Likewise.
3589          (branch_dest): Likewise.
3590          (arc_attr_type): Likewise.
3591          * config/arc/arc.c (TARGET_ADJUST_INSN_LENGTH): Remove.
3592          (TARGET_INSN_LENGTH_PARAMETERS): Likewise.
3593          (arc_final_prescan_insn): Remove inserting the nops due to
3594          hardware hazards.  It is done in reorg step.
3595          (insn_length_variant_t): Remove.
3596          (insn_length_parameters_t): Likewise.
3597          (arc_insn_length_parameters): Likewise.
3598          (arc_get_insn_variants): Likewise.
3599          * config/arc/arc.h (TARGET_UPSIZE_DBR): Remove.
3600
3601 2018-03-06  Jakub Jelinek  <jakub@redhat.com>
3602
3603         PR inline-asm/84683
3604         * reg-stack.c (move_for_stack_reg): If any_malformed_asm, avoid
3605         assertion failure.
3606
3607         PR tree-optimization/84687
3608         * omp-simd-clone.c (simd_clone_create): Clear DECL_BUILT_IN_CLASS
3609         on new_node->decl.
3610         * match.pd (pow(C,x)*expN(y) -> expN(logN(C)*x+y)): New optimization.
3611
3612 2018-03-05  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
3613
3614         * config/rs6000/rs6000-builtin.def (rs6000_speculation_barrier):
3615         Rename to ppc_speculation_barrier.
3616         * config/rs6000/rs6000.c (rs6000_init_builtins): Rename builtin to
3617         __builtin_ppc_speculation_barrier.
3618
3619 2018-03-05  Jakub Jelinek  <jakub@redhat.com>
3620
3621         PR target/84700
3622         * combine.c (combine_simplify_rtx): Don't try to simplify if
3623         if_then_else_cond returned non-NULL, but either true_rtx or false_rtx
3624         are equal to x.
3625
3626 2018-03-05  Segher Boessenkool  <segher@kernel.crashing.org>
3627
3628         * config/rs6000/rs6000.c (rs6000_loop_align): Don't align tiny loops
3629         to 32 bytes when compiling for POWER9.
3630
3631 2018-03-05  Jakub Jelinek  <jakub@redhat.com>
3632
3633         PR target/84564
3634         * config/i386/i386.c (ix86_function_ok_for_sibcall): Check for
3635         regparm >= 3 with no arg reg available also for calls with
3636         flag_force_indirect_call.  Pass decl to ix86_function_regparm.
3637
3638         PR target/84524
3639         * config/i386/sse.md (*<code><mode>3): Replace <mask_prefix3> with
3640         orig,vex.
3641         (*<plusminus_insn><mode>3): Likewise.  Remove <mask_operand3> uses.
3642
3643 2018-03-05  Peter Bergner  <bergner@vnet.ibm.com>
3644
3645         PR target/84264
3646         * config/rs6000/vector.md (mov<mode>): Disallow altivec memory operands.
3647
3648 2018-03-05  Richard Biener  <rguenther@suse.de>
3649
3650         PR tree-optimization/84486
3651         * tree-ssa-pre.c (create_expression_by_pieces): Remove dead code.
3652         When inserting a __builtin_assume_aligned call set the LHS
3653         SSA name alignment info accordingly.
3654
3655 2018-03-05  Wilco Dijkstra  <wdijkstr@arm.com>
3656
3657         PR tree-optimization/84114
3658         * config/aarch64/aarch64.c (aarch64_reassociation_width)
3659         Avoid reassociation of FLOAT_MODE addition.
3660
3661 2018-03-05  Olga Makhotina  <olga.makhotina@intel.com>
3662
3663         * common/config/i386/i386-common.c (OPTION_MASK_ISA_PCONFIG_SET,
3664         OPTION_MASK_ISA_PCONFIG_UNSET, OPTION_MASK_ISA_WBNOINVD_SET,
3665         OPTION_MASK_ISA_WBNOINVD_UNSET): New definitions.
3666         (ix86_handle_option): Handle -mpconfig and -mwbnoinvd.
3667         * config.gcc (pconfigintrin.h, wbnoinvdintrin.h) : Add headers.
3668         * config/i386/cpuid.h (bit_PCONFIG, bit_WBNOINVD): New.
3669         * config/i386/driver-i386.c (host_detect_local_cpu): Detect -mpconfig
3670         and -mwbnoinvd.
3671         * config/i386/i386-builtin.def (__builtin_ia32_wbnoinvd,
3672         __builtin_ia32_wbinvd): New builtins.
3673         (SPECIAL_ARGS2): New.
3674         * config/i386/i386-c.c (__WBNOINVD__, __PCONFIG__): New.
3675         (SPECIAL_ARGS2): New.
3676         * config/i386/i386.c (ix86_target_string): Add -mpconfig and -mwbnoinvd.
3677         (ix86_valid_target_attribute_inner_p): Ditto.
3678         (ix86_init_mmx_sse_builtins): Add special_args2.
3679         * config/i386/i386.h (TARGET_PCONFIG, TARGET_PCONFIG_P, TARGET_WBNOINVD,
3680         TARGET_WBNOINVD_P): New.
3681         * config/i386/i386.md (UNSPECV_WBINVD, UNSPECV_WBNOINVD): New.
3682         (define_insn "wbinvd", define_insn "wbnoinvd"): New.
3683         * config/i386/i386.opt: Add -mpconfig and -mwbnoinvd.
3684         * config/i386/immintrin.h (_wbinvd): New intrinsic.
3685         * config/i386/pconfigintrin.h: New file.
3686         * config/i386/wbnoinvdintrin.h: Ditto.
3687         * config/i386/x86intrin.h: Add headers pconfigintrin.h and wbnoinvdintrin.h.
3688         * doc/invoke.texi (-mpconfig, -mwbnoinvd): New.
3689
3690 2018-03-05  Richard Biener  <rguenther@suse.de>
3691
3692         PR tree-optimization/84670
3693         * tree-ssa-pre.c (struct bb_bitmap_sets): Add visited_with_visited_succs
3694         member.
3695         (BB_VISITED_WITH_VISITED_SUCCS): New define.
3696         (compute_antic): Initialize BB_VISITED_WITH_VISITED_SUCCS.
3697         (compute_antic_aux): Only assert the number of values in ANTIC_IN
3698         doesn't grow if all successors (recursively) were visited at least
3699         once.
3700
3701 2018-03-05  Richard Biener  <rguenther@suse.de>
3702
3703         PR tree-optimization/84650
3704         * tree-ssa-loop-im.c (pass_lim::execute): Reset the SCEV cache
3705         if executed in the loop pipeline.
3706
3707 2018-03-05  Sandra Loosemore  <sandra@codesourcery.com>
3708
3709         * doc/configfiles.texi (Configuration Files): Move info about
3710         conditionalizing $target-protos.h to...
3711         * doc/sourcebuild.texi (Back End): Here.  Explain how $target.h
3712         differs from $target-protos.h.
3713
3714 2018-03-05  Kito Cheng  <kito.cheng@gmail.com>
3715             Chung-Ju Wu  <jasonwucj@gmail.com>
3716
3717         * config/nds32/nds32-protos.h (nds32_expand_setmem): Declare.
3718         * config/nds32/nds32-multiple.md (setmemsi): Define.
3719         * config/nds32/nds32-memory-manipulation.c
3720         (nds32_gen_dup_4_byte_to_word_value): New.
3721         (emit_setmem_word_loop): New.
3722         (emit_setmem_byte_loop): New.
3723         (nds32_expand_setmem_loop): New.
3724         (nds32_expand_setmem_loop_v3m): New.
3725         (nds32_expand_setmem_unroll): New.
3726         (nds32_expand_setmem): New.
3727
3728 2018-03-04  Kito Cheng  <kito.cheng@gmail.com>
3729             Chung-Ju Wu  <jasonwucj@gmail.com>
3730
3731         * config/nds32/nds32-memory-manipulation.c
3732         (nds32_emit_load_store): New.
3733         (nds32_emit_post_inc_load_store): New.
3734         (nds32_emit_mem_move): New.
3735         (nds32_emit_mem_move_block): New.
3736         (nds32_expand_movmemsi_loop_unknown_size): New.
3737         (nds32_expand_movmemsi_loop_known_size): New.
3738         (nds32_expand_movmemsi_loop): New.
3739         (nds32_expand_movmemsi_unroll): New.
3740         (nds32_expand_movmemqi): Rename ...
3741         (nds32_expand_movmemsi): ... to this.
3742         * config/nds32/nds32-multiple.md (movmemqi): Rename ...
3743         (movmemsi): ... to this.
3744         * config/nds32/nds32-protos.h (nds32_expand_movmemqi): Rename ...
3745         (nds32_expand_movmemsi): ... to this.
3746
3747 2018-03-04  Kito Cheng  <kito.cheng@gmail.com>
3748             Monk Chiang  <sh.chiang04@gmail.com>
3749             Chung-Ju Wu  <jasonwucj@gmail.com>
3750
3751         * config/nds32/nds32-protos.h
3752         (nds32_expand_load_multiple): New arguments.
3753         (nds32_expand_store_multiple): Ditto.
3754         (nds32_valid_multiple_load_store): Rename ...
3755         (nds32_valid_multiple_load_store_p): ... to this.
3756         * config/nds32/nds32-memory-manipulation.c
3757         (nds32_expand_load_multiple): Refine implementation.
3758         (nds32_expand_store_multiple): Ditto.
3759         * config/nds32/nds32-multiple.md
3760         (load_multiple): Update nds32_expand_load_multiple interface.
3761         (store_multiple): Update nds32_expand_store_multiple interface.
3762         * config/nds32/nds32-predicates.c
3763         (nds32_valid_multiple_load_store): Rename ...
3764         (nds32_valid_multiple_load_store_p): ... to this and refine
3765         implementation.
3766         * config/nds32/predicates.md
3767         (nds32_load_multiple_and_update_address_operation): New predicate.
3768         (nds32_store_multiple_and_update_address_operation): New predicate.
3769
3770 2018-03-04  Kito Cheng  <kito.cheng@gmail.com>
3771             Chung-Ju Wu  <jasonwucj@gmail.com>
3772
3773         * config/nds32/nds32.md (type): Add load_multiple and store_multiple.
3774         (combo): New attribute.
3775         * config/nds32/nds32-multiple.md: Refine patterns with new attributes.
3776
3777 2018-03-03  Chung-Ju Wu  <jasonwucj@gmail.com>
3778
3779         * config/nds32/nds32.opt: Change -mcmodel= default value.
3780
3781 2018-03-03  Kito Cheng  <kito.cheng@gmail.com>
3782             Monk Chiang  <sh.chiang04@gmail.com>
3783             Chung-Ju Wu  <jasonwucj@gmail.com>
3784
3785         * config/nds32/constants.md (unspec_element): New enum.
3786         * config/nds32/constraints.md (Umw): New constraint.
3787         * config/nds32/nds32-intrinsic.c: Add more builtin functions.
3788         * config/nds32/nds32-intrinsic.md: Likewise.
3789         * config/nds32/nds32-md-auxiliary.c (nds32_regno_to_enable4): New.
3790         (nds32_valid_smw_lwm_base_p): New.
3791         (nds32_output_smw_single_word): New.
3792         (nds32_output_lmw_single_word): New.
3793         (nds32_expand_unaligned_load): New.
3794         (nds32_expand_unaligned_store): New.
3795         * config/nds32/nds32-protos.h (nds32_valid_smw_lwm_base_p): Declare.
3796         (nds32_output_smw_single_word): Declare.
3797         (nds32_output_lmw_single_word): Declare.
3798         (nds32_expand_unaligned_load): Declare.
3799         (nds32_expand_unaligned_store): Declare.
3800         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_UALOAD_HW,
3801         NDS32_BUILTIN_UALOAD_W, NDS32_BUILTIN_UALOAD_DW,
3802         NDS32_BUILTIN_UASTORE_HW, NDS32_BUILTIN_UASTORE_W,
3803         NDS32_BUILTIN_UASTORE_DW.
3804         * config/nds32/predicates.md (nds32_lmw_smw_base_operand): New
3805         predicate.
3806
3807 2018-03-03  Monk Chiang  <sh.chiang04@gmail.com>
3808             Kito Cheng  <kito.cheng@gmail.com>
3809             Chung-Ju Wu  <jasonwucj@gmail.com>
3810
3811         * config/nds32/nds32-intrinsic.c
3812         (nds32_expand_builtin_null_ftype_reg): Delete.
3813         (nds32_expand_builtin_reg_ftype_imm): Ditto.
3814         (nds32_expand_builtin_null_ftype_reg_imm): Ditto.
3815         (nds32_read_argument): New.
3816         (nds32_legitimize_target): Ditto.
3817         (nds32_legitimize_argument): Ditto.
3818         (nds32_check_constant_argument): Ditto.
3819         (nds32_expand_unop_builtin): Ditto.
3820         (nds32_expand_unopimm_builtin): Ditto.
3821         (nds32_expand_binop_builtin): Ditto.
3822         (nds32_builtin_decl_impl): Ditto.
3823         (builtin_description): Ditto.
3824         (nds32_expand_builtin_impl): Rewrite with new infrastructure.
3825         (nds32_init_builtins_impl): Ditto.
3826         * config/nds32/nds32.c (TARGET_BUILTIN_DECL): Define.
3827         (nds32_builtin_decl): New.
3828         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_COUNT.
3829         * config/nds32/nds32-protos.h (nds32_builtin_decl_impl): Declare.
3830
3831 2018-03-02  Jeff Law  <law@redhat.com>
3832
3833         * reorg.c (stop_search_p): Handle DEBUG_INSN.
3834         (redundant_insn, fill_simple_delay_slots): Likewise.
3835         (fill_slots_from_thread): Likewise.
3836         * resource.c (mark_referenced_resources): Likewise.
3837         (mark_set_resources, find_dead_or_set_registers): Likewise.
3838
3839 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
3840
3841         * substring-locations.h (format_warning_va): Formatting fix for
3842         ATTRIBUTE_GCC_DIAG.
3843         (format_warning_at_substring): Fix up ATTRIBUTE_GCC_DIAG second
3844         argument.
3845         (format_warning_n_va, format_warning_at_substring_n): New prototypes.
3846         * substring-locations.c: Include intl.h.
3847         (format_warning_va): Turned into small wrapper around
3848         format_warning_n_va, renamed to ...
3849         (format_warning_n_va): ... this, add N and PLURAL_GMSGID arguments,
3850         rename GMSGID to SINGULAR_GMSGID, if SINGULAR_GMSGID != PLURAL_GMSGID,
3851         use ngettext.
3852         (format_warning_at_substring_n): New function.
3853         * gimple-ssa-sprintf.c: Remove GCC diagnostic ignored pragma.
3854         (fmtwarn): Add ATTRIBUTE_GCC_DIAG.  Turn into a copy of
3855         format_warning_at_substring with just a shorter name instead of
3856         const function pointer.
3857         (fmtwarn_n): New function.
3858         (maybe_warn, format_directive, parse_directive): Use fmtwarn_n where
3859         appropriate, get rid of all the fmtstr temporaries, move conditionals
3860         with G_() wrapped string literals directly into fmtwarn arguments,
3861         cast dir.len to (int), formatting fixes.
3862
3863 2018-03-02  Thomas Schwinge  <thomas@codesourcery.com>
3864
3865         * doc/invoke.texi: Remove "Cilk Plus" references.
3866
3867 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
3868             Richard Biener  <rguenther@suse.de>
3869
3870         PR ipa/84628
3871         * expr.c (expand_expr_real_1) <case CALL_EXPR>: Don't emit diagnostics
3872         for error or warning attributes if CALL_FROM_THUNK_P is set.
3873         Formatting fixes.
3874
3875 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
3876
3877         PR target/56540
3878         * config/pa/pa.h (TARGET_CPU_CPP_BUILTINS): Predefine
3879         __SIZEOF_128__ macro if HPUX_LONG_DOUBLE_LIBRARY.
3880
3881         PR target/56540
3882         * config/ia64/ia64.h (TARGET_CPU_CPP_BUILTINS): Predefine
3883         __SIZEOF_{FPREG,FLOAT{80,128}}__ macros.
3884
3885         * predict.c (test_prediction_value_range): Use PROB_UNINITIALIZED
3886         instead of -1U in last predictors element's probability member.
3887
3888 2018-03-02  Eric Botcazou  <ebotcazou@adacore.com>
3889
3890         PR ipa/83983
3891         * ipa-devirt.c (odr_subtypes_equivalent_p): Get the ODR type of both
3892         arguments if they are comparable.
3893
3894 2018-03-02  Richard Sandiford  <richard.sandiford@linaro.org>
3895
3896         PR tree-optimization/84634
3897         * tree-vect-stmts.c (vectorizable_store, vectorizable_load): Replace
3898         masks and masked_loop_p with a single loop_masks, making sure it's
3899         null for bb vectorization.
3900
3901 2018-03-02  Richard Sandiford  <richard.sandiford@linaro.org>
3902
3903         * tree-vect-data-refs.c (vect_analyze_data_ref_dependence)
3904         (vect_analyze_data_ref_access): Use loop->safe_len rather than
3905         loop->force_vectorize to check whether there is no alias.
3906
3907 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
3908
3909         PR target/84614
3910         * rtl.h (prev_real_nondebug_insn, next_real_nondebug_insn): New
3911         prototypes.
3912         * emit-rtl.c (next_real_insn, prev_real_insn): Fix up function
3913         comments.
3914         (next_real_nondebug_insn, prev_real_nondebug_insn): New functions.
3915         * cfgcleanup.c (try_head_merge_bb): Use prev_real_nondebug_insn
3916         instead of a loop around prev_real_insn.
3917         * combine.c (move_deaths): Use prev_real_nondebug_insn instead of
3918         prev_real_insn.
3919
3920         PR inline-asm/84625
3921         * config/i386/i386.c (ix86_print_operand): Use conditional
3922         output_operand_lossage instead of gcc_assert if CONST_VECTOR is not
3923         zero vector.
3924
3925 2018-03-02  Richard Biener  <rguenther@suse.de>
3926
3927         PR tree-optimization/84427
3928         * tree-ssa-pre.c (bitmap_remove_expr_from_set): Remove.
3929         (bitmap_set_subtract_values): Rewrite to handle multiple
3930         exprs per value.
3931         (clean): Likewise.
3932         (prune_clobbered_mems): Likewise.
3933         (phi_translate): Take edge instead of pred/phiblock.
3934         (phi_translate_1): Likewise.
3935         (phi_translate_set): Likewise.  Insert all translated
3936         exprs for a value into the set, keeping possibly multiple
3937         expressions per value.
3938         (compute_antic_aux): Adjust for phi_translate changes.
3939         When intersecting union the expressions and prune those
3940         not in the final value set, keeping possibly multiple
3941         expressions per value.  Do not use value-insertion
3942         for unioning ANTIC_OUT U EXP_GEN - TMP_GEN but merge
3943         all expressions.  Add verification that the value-sets
3944         only shrink during iteration.
3945         (compute_partial_antic_aux): Adjust for the phi_translate changes.
3946         (do_pre_regular_insertion): Likewise.
3947         (do_pre_partial_partial_insertion): Likewise.
3948
3949 2018-03-02  Richard Biener  <rguenther@suse.de>
3950
3951         PR target/82005
3952         * config/darwin.c (saved_debug_info_level): New static global.
3953         (darwin_asm_lto_start): Disable debug info generation for LTO out.
3954         (darwin_asm_lto_end): Restore debug info generation settings.
3955
3956 2018-03-01  Martin Liska  <mliska@suse.cz>
3957
3958         PR sanitizer/82484
3959         * sanopt.c (sanitize_rewrite_addressable_params): Do not handle
3960         volatile arguments.
3961
3962 2018-03-01  Richard Biener  <rguenther@suse.de>
3963
3964         PR debug/84645
3965         * dwarf2out.c (gen_variable_die): Properly handle late VLA
3966         type annotation with LTO when debug was disabled at compile-time.
3967
3968 2018-03-01  Matthew Fortune  <mfortune@gmail.com>
3969
3970         * config/mips/mips.c (mips_final_prescan_insn): Fix incorrect
3971         XINT with INTVAL.
3972         (mips_final_postscan_insn): Likewise.
3973
3974 2018-03-01  Richard Sandiford  <richard.sandiford@linaro.org>
3975
3976         PR rtl-optimization/84528
3977         * alias.c (init_alias_target): Add commentary.
3978         (init_alias_analysis): Only give HARD_FRAME_POINTER_REGNUM
3979         a unique base value if the frame pointer is not eliminated
3980         to the stack pointer.
3981
3982 2018-03-01  Tom de Vries  <tom@codesourcery.com>
3983
3984         PR rtl-optimization/83327
3985         * lra-int.h (hard_regs_spilled_into): Declare.
3986         * lra.c (hard_regs_spilled_into): Define.
3987         (init_reg_info): Init hard_regs_spilled_into.
3988         * lra-spills.c (assign_spill_hard_regs): Update hard_regs_spilled_into.
3989         * lra-lives.c (make_hard_regno_born, make_hard_regno_dead)
3990         (process_bb_lives): Handle hard_regs_spilled_into.
3991         (lra_create_live_ranges_1): Before doing liveness propagation, clear
3992         regs in all_hard_regs_bitmap if set in hard_regs_spilled_into.
3993
3994 2018-02-28  David Edelsohn  <dje.gcc@gmail.com>
3995
3996         * config.gcc (powerpc-ibm-aix7.1.*): New stanza.
3997         (powerpc-ibm-aix[789]*): Default to AIX 7.2.
3998         * config/rs6000/aix71.h (TARGET_DEFAULT): Revert to Power4 ISA.
3999         * config/rs6000/aix72.h: New file.
4000
4001 2018-02-28  Jakub Jelinek  <jakub@redhat.com>
4002
4003         * gimple-ssa-warn-restrict.c (maybe_diag_overlap): Use warning_n
4004         instead of warning_at with conditional singular and plural messages
4005         where possible.
4006
4007         PR target/52991
4008         * stor-layout.c (update_alignment_for_field): For
4009         targetm.ms_bitfield_layout_p (rli->t), if !is_bitfield
4010         && !DECL_PACKED (field), do the alignment update, just use
4011         only desired_align instead of MAX (type_align, desired_align)
4012         as the alignment.
4013         (place_field): Don't do known_align < desired_align handling
4014         early if targetm.ms_bitfield_layout_p (rli->t) and rli->prev_field
4015         is non-NULL, instead do it after rli->prev_field handling and
4016         only if not within a bitfield word.  For DECL_PACKED (field)
4017         use type_align of BITS_PER_UNIT.
4018
4019 2018-02-28  Eric Botcazou  <ebotcazou@adacore.com>
4020
4021         * config/aarch64/aarch64.c (aarch64_emit_probe_stack_range): Remove
4022         superfluous parentheses and trailing spaces.
4023
4024 2018-02-28  Richard Biener  <rguenther@suse.de>
4025
4026         PR tree-optimization/84584
4027         * graphite-scop-detection.c (scop_detection::add_scop): Discard
4028         SCoPs with fake exit edge.
4029
4030 2018-02-28  Martin Liska  <mliska@suse.cz>
4031
4032         PR testsuite/84597
4033         * timevar.c (timer::print): Fix format to properly print 100%
4034         values.
4035
4036 2018-02-28  Richard Biener  <rguenther@suse.de>
4037
4038         PR middle-end/84607
4039         * genmatch.c (capture_info::walk_match): Do not mark
4040         captured expressions without operands as expr_p given
4041         they act more like predicates and should be subject to
4042         "lost tail" side-effect preserving.
4043
4044 2018-02-28  Alexandre Oliva  <aoliva@redhat.com>
4045
4046         PR rtl-optimization/81611
4047         * auto-inc-dec.c (attempt_change): Move dead note from
4048         mem_insn if it's the next use of regno
4049         (find_address): Take address use of reg holding
4050         non-incremented value.  Add parm to limit search to the named
4051         reg only.
4052         (merge_in_block): Attempt to use a mem insn that is the next
4053         use of the original regno.
4054
4055 2018-02-27  Martin Sebor  <msebor@redhat.com>
4056
4057         PR c++/83871
4058         * gcc/doc/invoke.texi (-Wmissing-attributes): New option.
4059         * gcc/print-tree.c (print_node): Handle DECL_UNINLINABLE.
4060
4061 2018-02-27  Martin Sebor  <msebor@redhat.com>
4062
4063         PR translation/84207
4064         * diagnostic-core.h (warning_n, error_n, inform_n): Change
4065         n argument to unsigned HOST_WIDE_INT.
4066         * diagnostic.c (warning_n, error_n, inform_n): Ditto.
4067         (diagnostic_n_impl): Ditto.  Handle arguments in excess of LONG_MAX.
4068         * gimple-ssa-sprintf.c (format_directive): Simplify inform_n call.
4069         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Use warning_n.
4070
4071 2018-02-27  Richard Biener  <rguenther@suse.de>
4072
4073         PR tree-optimization/84512
4074         * tree-vect-loop.c (vect_compute_single_scalar_iteration_cost):
4075         Do not use the estimate returned from record_stmt_cost for
4076         the scalar iteration cost but sum properly using add_stmt_cost.
4077
4078 2018-02-27  Richard Biener  <rguenther@suse.de>
4079
4080         PR tree-optimization/84466
4081         * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
4082         Adjust last change to less strictly validate use operands.
4083
4084 2018-02-27  Martin Liska  <mliska@suse.cz>
4085
4086         PR gcov-profile/84548
4087         * gcov.c (process_file): Allow partial overlap and consider it
4088         also as group functions.
4089         (output_lines): Properly calculate range of lines for a group.
4090
4091 2018-02-27  Martin Liska  <mliska@suse.cz>
4092
4093         * timevar.c (timer::print_row): Remove 'usr', 'sys', 'wall' and
4094         'ggc' suffixes.  Change first column width.
4095         (timer::print): Fix formatting of the column.
4096
4097 2018-02-27  Alexandre Oliva  <aoliva@redhat.com>
4098
4099         * tree-ssa-live.c (remove_unused_scope_block_p): Do not
4100         preserve inline entry blocks for the sake of debug inline
4101         entry point markers alone.
4102         (remove_unused_locals): Suggest in comments a better place to
4103         force the preservation of inline entry blocks that are
4104         otherwise unused, but do not preserve them.
4105
4106 2018-02-26  H.J. Lu  <hongjiu.lu@intel.com>
4107
4108         * config/i386/i386.c (ix86_output_indirect_jmp): Update comments.
4109
4110 2018-02-26  H.J. Lu  <hongjiu.lu@intel.com>
4111
4112         PR target/84039
4113         * config/i386/constraints.md (Bs): Replace
4114         ix86_indirect_branch_register with
4115         TARGET_INDIRECT_BRANCH_REGISTER.
4116         (Bw): Likewise.
4117         * config/i386/i386.md (indirect_jump): Likewise.
4118         (tablejump): Likewise.
4119         (*sibcall_memory): Likewise.
4120         (*sibcall_value_memory): Likewise.
4121         Peepholes of indirect call and jump via memory: Likewise.
4122         (*sibcall_GOT_32): Disallowed for TARGET_INDIRECT_BRANCH_REGISTER.
4123         (*sibcall_value_GOT_32): Likewise.
4124         * config/i386/predicates.md (indirect_branch_operand): Likewise.
4125         (GOT_memory_operand): Likewise.
4126         (call_insn_operand): Likewise.
4127         (sibcall_insn_operand): Likewise.
4128         (GOT32_symbol_operand): Likewise.
4129         * config/i386/i386.h (TARGET_INDIRECT_BRANCH_REGISTER): New.
4130
4131 2018-02-26  Eric Botcazou  <ebotcazou@adacore.com>
4132
4133         PR rtl-optimization/83496
4134         * reorg.c (steal_delay_list_from_target): Change REDUNDANT array from
4135         booleans to RTXes.  Call fix_reg_dead_note on every non-null element.
4136         (steal_delay_list_from_fallthrough): Call fix_reg_dead_note on a
4137         redundant insn, if any.
4138         (relax_delay_slots): Likewise.
4139         (update_reg_unused_notes): Rename REDUNDANT_INSN to OTHER_INSN.
4140
4141 2018-02-26  Richard Sandiford  <richard.sandiford@linaro.org>
4142
4143         PR tree-optimization/83965
4144         * tree-vect-patterns.c (vect_reassociating_reduction_p): Assume
4145         that grouped statements are part of a reduction chain.  Return
4146         true if the statement is not marked as a reduction itself but
4147         is part of a group.
4148         (vect_recog_dot_prod_pattern): Don't check whether the statement
4149         is part of a group here.
4150         (vect_recog_sad_pattern): Likewise.
4151         (vect_recog_widen_sum_pattern): Likewise.
4152
4153 2018-02-26  Eric Botcazou  <ebotcazou@adacore.com>
4154
4155         PR debug/84545
4156         * final.c (rest_of_clean_state): Also look for calls inside sequences.
4157
4158 2018-02-26  H.J. Lu  <hongjiu.lu@intel.com>
4159
4160         PR target/84530
4161         * config/i386/i386-protos.h (ix86_output_indirect_jmp): Remove
4162         the bool argument.
4163         (ix86_output_indirect_function_return): New prototype.
4164         (ix86_split_simple_return_pop_internal): Likewise.
4165         * config/i386/i386.c (indirect_return_via_cx): New.
4166         (indirect_return_via_cx_bnd): Likewise.
4167         (indirect_thunk_name): Handle return va CX_REG.
4168         (output_indirect_thunk_function): Create alias for
4169         __x86_return_thunk_[re]cx and __x86_return_thunk_[re]cx_bnd.
4170         (ix86_output_indirect_jmp): Remove the bool argument.
4171         (ix86_output_indirect_function_return): New function.
4172         (ix86_split_simple_return_pop_internal): Likewise.
4173         * config/i386/i386.md (*indirect_jump): Don't pass false
4174         to ix86_output_indirect_jmp.
4175         (*tablejump_1): Likewise.
4176         (simple_return_pop_internal): Change it to define_insn_and_split.
4177         Call ix86_split_simple_return_pop_internal to split it for
4178         -mfunction-return=.
4179         (simple_return_indirect_internal): Call
4180         ix86_output_indirect_function_return instead of
4181         ix86_output_indirect_jmp.
4182
4183 2018-02-26  Jakub Jelinek  <jakub@redhat.com>
4184
4185         PR bootstrap/84405
4186         * vec.h (vec_default_construct): For BROKEN_VALUE_INITIALIZATION use
4187         memset and value initialization afterwards.
4188
4189 2018-02-26  Christophe Lyon  <christophe.lyon@linaro.org>
4190
4191         * Makefile.in (lto-wrapper): Use ALL_LINKERFLAGS.
4192
4193 2018-02-26  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
4194
4195         PR target/84521
4196         * common/config/aarch64/aarch64-common.c
4197         (aarch_option_optimization_table[]): Switch
4198         off fomit-frame-pointer
4199
4200 2018-02-26  Kito Cheng  <kito.cheng@gmail.com>
4201             Chung-Ju Wu  <jasonwucj@gmail.com>
4202
4203         * config/nds32/nds32-multiple.md (load_multiple): Disallow
4204         volatile memory.
4205         (store_multiple): Ditto.
4206
4207 2018-02-26  Kito Cheng  <kito.cheng@gmail.com>
4208
4209         * config.gcc: Add --with-cpu support for nds32 target.
4210         * config/nds32/nds32-opts.h (nds32_cpu_type): New.
4211         * config/nds32/nds32.opt: Add -mcpu= option.
4212
4213 2018-02-25  Segher Boessenkool  <segher@kernel.crashing.org>
4214
4215         * config/rs6000/rs6000.opt (mvrsave=no, mvrsave=yes, isel=no,
4216         isel=yes): Warn for these deprecated options.
4217
4218 2018-02-23  David Edelsohn  <dje.gcc@gmail.com>
4219
4220         * config/rs6000/aix71.h (TARGET_DEFAULT): Change to
4221         ISA_2_5_MASKS_EMBEDDED.
4222
4223 2018-02-23  Jakub Jelinek  <jakub@redhat.com>
4224
4225         * ipa-prop.c (ipa_vr_ggc_hash_traits::hash): Hash p->min and
4226         p->max as pointers rather than using iterative_hash_expr.
4227
4228 2018-02-23  Carl Love  <cel@us.ibm.com>
4229
4230         * config/rs6000/rs6000-builtin.def: Change VSIGNED2 and VUNSIGNED2
4231         macro expansions from BU_VSX_2 to BU_P8V_VSX_2 and BU_VSX_OVERLOAD_2 to
4232         BU_P8V_OVERLOAD_2.
4233         * config/rs6000/rs6000-c.c: Change VSX_BUILTIN_VEC_VSIGNED2 to
4234         P8V_BUILTIN_VEC_VSIGNED2.  Change VSX_BUILTIN_VEC_VUNSIGNED2 to
4235         P8V_BUILTIN_VEC_VUNSIGNED2.
4236
4237 2018-02-22  Vladimir Makarov  <vmakarov@redhat.com>
4238
4239         PR target/81572
4240         * lra-int.h (LRA_UNKNOWN_ALT, LRA_NON_CLOBBERED_ALT): New macros.
4241         * lra.c (lra_set_insn_recog_data, lra_update_insn_recog_data): Use
4242         LRA_UNKNOWN_ALT.
4243         * lra-constraints.c (curr_insn_transform): Set up
4244         LRA_NON_CLOBBERED_ALT for moves processed on the fast path.  Use
4245         LRA_UNKNOWN_ALT.
4246         (remove_inheritance_pseudos): Use LRA_UNKNOWN_ALT.
4247         * lra-eliminations.c (spill_pseudos): Ditto.
4248         (process_insn_for_elimination): Ditto.
4249         * lra-lives.c (reg_early_clobber_p): Use the new macros.
4250         * lra-spills.c (spill_pseudos): Use LRA_UNKNOWN_ALT and
4251         LRA_NON_CLOBBERED_ALT.
4252
4253 2018-02-22  Martin Sebor  <msebor@redhat.com>
4254
4255         PR tree-optimization/84480
4256         * gimple-fold.c (gimple_fold_builtin_strcpy): Move warnings
4257         to maybe_diag_stxncpy_trunc.  Call it.
4258         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Integrate warnings
4259         from gimple_fold_builtin_strcpy.  Print inlining stack.
4260         (handle_builtin_stxncpy): Print inlining stack.
4261         * tree-ssa-strlen.h (maybe_diag_stxncpy_trunc): Declare.
4262
4263 2018-02-22  H.J. Lu  <hongjiu.lu@intel.com>
4264
4265         PR target/84176
4266         * config/i386/i386.c (ix86_set_indirect_branch_type): Issue an
4267         error when -mindirect-branch=thunk-extern, -fcf-protection=branch
4268         and -fcheck-pointer-bounds are used together.
4269         (indirect_thunk_prefix): New enum.
4270         (indirect_thunk_need_prefix): New function.
4271         (indirect_thunk_name): Replace need_bnd_p with need_prefix.  Use
4272         "_nt" instead of "_bnd" for NOTRACK prefix.
4273         (output_indirect_thunk): Replace need_bnd_p with need_prefix.
4274         (output_indirect_thunk_function): Likewise.
4275         (): Likewise.
4276         (ix86_code_end): Update output_indirect_thunk_function calls.
4277         (ix86_output_indirect_branch_via_reg): Replace
4278         ix86_bnd_prefixed_insn_p with indirect_thunk_need_prefix.
4279         (ix86_output_indirect_branch_via_push): Likewise.
4280         (ix86_output_function_return): Likewise.
4281         * doc/invoke.texi: Document -mindirect-branch=thunk-extern is
4282         incompatible with -fcf-protection=branch and
4283         -fcheck-pointer-bounds.
4284
4285 2018-02-22  Steve Ellcey  <sellcey@cavium.com>
4286
4287         PR target/83335
4288         * config/aarch64/aarch64.c (aarch64_print_address_internal):
4289         Change gcc_assert call to output_operand_lossage.
4290
4291 2018-02-22  Steve Ellcey  <sellcey@cavium.com>
4292
4293         * doc/extend.texi (__builtin_extend_pointer): Document builtin.
4294
4295 2018-02-22  DJ Delorie  <dj@redhat.com>
4296             Sebastian Perta  <sebastian.perta@renesas.com>
4297             Oleg Endo  <olegendo@gcc.gnu.org>
4298
4299         * config/rx/rx.c (rx_rtx_costs): New function.
4300         (TARGET_RTX_COSTS): Override to use rx_rtx_costs.
4301
4302 2018-02-22  Thomas Preud'homme  <thomas.preudhomme@arm.com>
4303
4304         * config/arm/t-multilib: Map Armv8-R to Armv7 multilibs.
4305
4306 2018-02-22  Martin Liska  <mliska@suse.cz>
4307
4308         PR driver/83193
4309         * common/config/arm/arm-common.c (arm_print_hint_for_cpu_option):
4310         Add "native" as a possible value.
4311
4312 2018-02-22  Martin Liska  <mliska@suse.cz>
4313
4314         PR driver/83193
4315         * config/i386/i386.c (ix86_option_override_internal):
4316         Add "native" as a possible value for -march and -mtune.
4317
4318 2018-02-22  Jakub Jelinek  <jakub@redhat.com>
4319
4320         PR target/84502
4321         * stor-layout.c (finalize_type_size): Propagate TYPE_EMPTY_P flag
4322         to all type variants.
4323
4324         PR tree-optimization/84503
4325         * gimple-ssa-store-merging.c (merged_store_group::merge_into): Compute
4326         width as info->bitpos + info->bitsize - start.
4327         (merged_store_group::merge_overlapping): Simplify width computation.
4328         (check_no_overlap): New function.
4329         (imm_store_chain_info::try_coalesce_bswap): Compute expected
4330         start + width and last_order of the group, fail if check_no_overlap
4331         fails.
4332         (imm_store_chain_info::coalesce_immediate_stores): Don't merge info
4333         to group if check_no_overlap fails.
4334
4335 2018-02-21  Segher Boessenkool  <segher@kernel.crashing.org>
4336
4337         * config/rs6000/altivec.md: Delete contraint arguments to
4338         define_expand, define_split, and define_peephole2, and in
4339         define_insn_and_split if always unused.
4340         * config/rs6000/darwin.md: Ditto.
4341         * config/rs6000/dfp.md: Ditto.
4342         * config/rs6000/rs6000.md: Ditto.
4343         * config/rs6000/sync.md: Ditto.
4344         * config/rs6000/vector.md: Ditto.
4345         * config/rs6000/vsx.md: Ditto.
4346
4347 2018-02-21  Segher Boessenkool  <segher@kernel.crashing.org>
4348
4349         * config/rs6000/altivec.md: Write output control strings as braced
4350         blocks instead of double-quoted strings.
4351         * config/rs6000/darwin.md: Ditto.
4352         * config/rs6000/rs6000.md: Ditto.
4353         * config/rs6000/vector.md: Ditto.
4354         * config/rs6000/vsx.md: Ditto.
4355
4356 2018-02-21  Jason Merrill  <jason@redhat.com>
4357
4358         PR c++/84314 - ICE with templates and fastcall attribute.
4359         * attribs.c (build_type_attribute_qual_variant): Remove assert.
4360
4361 2018-02-21  Jan Hubicka  <hubicka@ucw.cz>
4362
4363         * ipa-cp.c (determine_versionability): Fix comment typos.
4364
4365 2018-02-21  Jan Hubicka  <hubicka@ucw.cz>
4366
4367         PR c/84229
4368         * ipa-cp.c (determine_versionability): Do not version functions caling
4369         va_arg_pack.
4370
4371 2018-02-21  Martin Liska  <mliska@suse.cz>
4372
4373         PR driver/83193
4374         * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch):
4375         Add "native" as a possible value.
4376         * config/aarch64/aarch64.h (HAVE_LOCAL_CPU_DETECT):  Define
4377         the macro when native cpu detection is available.
4378
4379 2018-02-21  Martin Liska  <mliska@suse.cz>
4380
4381         PR driver/83193
4382         * common/config/arm/arm-common.c (arm_print_hint_for_arch_option):
4383         Add "native" as a possible value.
4384         * config/arm/arm.h (HAVE_LOCAL_CPU_DETECT): Define the macro
4385         when native cpu detection is available.
4386
4387 2018-02-21  Jakub Jelinek  <jakub@redhat.com>
4388             Martin Sebor  <msebor@redhat.com>
4389
4390         PR tree-optimization/84478
4391         * gimple-fold.h (get_range_strlen): Add a bool argument defaulted to
4392         false.
4393         * gimple-fold.c (get_range_strlen): Make minlen const and assume it
4394         can't be NULL.  Change FUZZY from bool to int, for 1 add PHI/COND_EXPR
4395         support which is conservatively correct, for 2 only stay conservative
4396         for maxlen.  Formatting and comment capitalization fixes.  Add STRICT
4397         argument to the 2 argument get_range_strlen, adjust 6 arg
4398         get_range_strlen caller and clear minmaxlen[0] and [1] if it returned
4399         false.
4400         (get_maxval_strlen): Adjust 6 arg get_range_strlen caller.
4401         (gimple_fold_builtin_strlen): Pass true as last argument to
4402         get_range_strlen.
4403
4404 2018-02-20  Martin Sebor  <msebor@redhat.com>
4405
4406         PR middle-end/84095
4407         * gimple-ssa-warn-restrict.c (builtin_memref::extend_offset_range): New.
4408         (builtin_memref::set_base_and_offset): Same.  Handle inner references.
4409         (builtin_memref::builtin_memref): Factor out parts into
4410         set_base_and_offset and call it.
4411
4412 2018-02-20  Richard Sandiford  <richard.sandiford@linaro.org>
4413
4414         PR middle-end/84406
4415         * optabs-query.c (find_widening_optab_handler_and_mode): If from_mode
4416         is a scalar_int_mode, assert that to_mode is a scalar_int_mode with
4417         greater precision.  If to_mode is a MODE_PARTIAL_INT, stop the
4418         search at the associated MODE_INT.
4419
4420 2018-02-20  Jeff Law  <law@redhat.com>
4421
4422         PR middle-end/82123
4423         PR tree-optimization/81592
4424         PR middle-end/79257
4425         * gimple-ssa-sprintf.c (format_integer): Query EVRP range analyzer
4426         for range data rather than using global data.
4427         * gimple-ssa-sprintf.c (get_int_range): Query EVRP range analyzer for
4428         range data rather than using global data.
4429         * gimple-ssa-sprintf.c (get_int_range): Accept vr_values parameter
4430         pass it to children as needed.
4431         (struct directive::fmtresult): Similarly.
4432         (struct directive::set_width): Similarly.
4433         (struct directive::set_precision): Similarly.
4434         (format_integer, format_directive, parse_directive): Similarly.
4435         (format_none): Accept unnamed vr_values parameter.
4436         (format_percent, format_floating, format_character): Similarly.
4437         (format_string, format_plain): Similarly.
4438         * gimple-ssa-sprintf.c (sprintf_dom_walker::handle_gimple_call): Query
4439         the EVRP range analyzer for range data rather than using global data.
4440         * gimple-ssa-sprintf.c: Include alloc-pool.h, vr-values.h and
4441         gimple-ssa-evrp-analyze.h
4442         (class sprintf_dom_walker): Add after_dom_children member function.
4443         Add evrp_range_analyzer member.
4444         (sprintf_dom_walker::before_dom_children): Call into the EVRP
4445         range analyzer as needed.
4446         (sprintf_dom_walker::after_dom_children): New member function.
4447         * gimple-ssa-evrp-analyze.c (evrp_range_analyzer::enter): Do nothing
4448         if not optimizing.
4449         (evrp_range_analyzer::record_ranges_from_stmt): Likewise.
4450         (evrp_range_analyzer::pop_to_marker): Likewise.
4451
4452 2018-02-20  Richard Sandiford  <richard.sandiford@linaro.org>
4453
4454         PR tree-optimization/84419
4455         * internal-fn.c (expand_call_mem_ref): Create a TARGET_MEM_REF
4456         with the required type if its current type is compatible but
4457         different.
4458
4459 2018-02-20  Jakub Jelinek  <jakub@redhat.com>
4460
4461         PR middle-end/82004
4462         * match.pd (pow(C,x) -> exp(log(C)*x)): Delay all folding until
4463         after vectorization.
4464
4465 2018-02-20  Martin Liska  <mliska@suse.cz>
4466
4467         PR driver/83193
4468         * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch): Print
4469         possible values if we don't have a hint.
4470
4471 2018-02-20  Martin Liska  <mliska@suse.cz>
4472
4473         PR c/84310
4474         PR target/79747
4475         * final.c (shorten_branches): Build align_tab array with one
4476         more element.
4477         * opts.c (finish_options): Add alignment option limit check.
4478         (MAX_CODE_ALIGN): Likewise.
4479         (MAX_CODE_ALIGN_VALUE): Likewise.
4480         * doc/invoke.texi: Document maximum allowed option value for
4481         all -falign-* options.
4482
4483 2018-02-19  Jakub Jelinek  <jakub@redhat.com>
4484
4485         PR target/84146
4486         * reg-notes.def (REG_CALL_ARG_LOCATION): New reg note.
4487         * insn-notes.def (NOTE_INSN_CALL_ARG_LOCATION): Remove.
4488         * var-tracking.c (emit_note_insn_var_location): Remove all references
4489         to NOTE_INSN_CALL_ARG_LOCATION.
4490         (emit_notes_in_bb): Emit arguments as REG_CALL_ARG_LOCATION note on
4491         the CALL_INSN rather than separate NOTE_INSN_CALL_ARG_LOCATION note.
4492         Use copy_rtx_if_shared.
4493         * dwarf2out.c (gen_subprogram_die): Use XEXP with 0 instead of
4494         NOTE_VAR_LOCATION on ca_loc->call_arg_loc_note.
4495         (dwarf2out_var_location): Remove handling of
4496         NOTE_INSN_CALL_ARG_LOCATION, instead handle REG_CALL_ARG_LOCATION note
4497         on call_insn.
4498         * final.c (final_scan_insn): Remove all references to
4499         NOTE_INSN_CALL_ARG_LOCATION.
4500         (rest_of_clean_state): Likewise.  Remove REG_CALL_ARG_LOCATION notes
4501         before dumping final insns.
4502         * except.c (emit_note_eh_region_end): Remove all references to
4503         NOTE_INSN_CALL_ARG_LOCATION.
4504         * config/alpha/alpha.c (alpha_pad_function_end): Likewise.
4505         * config/c6x/c6x.c (c6x_gen_bundles): Likewise.
4506         * config/arc/arc.c (hwloop_optimize): Likewise.
4507         * config/arm/arm.c (create_fix_barrier): Likewise.
4508         * config/s390/s390.c (s390_chunkify_start): Likewise.
4509         * config/sh/sh.c (find_barrier): Likewise.
4510         * config/i386/i386.c (rest_of_insert_endbranch,
4511         ix86_seh_fixup_eh_fallthru): Likewise.
4512         * config/xtensa/xtensa.c (hwloop_optimize): Likewise.
4513         * config/iq2000/iq2000.c (final_prescan_insn): Likewise.
4514         * config/frv/frv.c (frv_function_prologue): Likewise.
4515         * emit-rtl.c (try_split): Likewise.  Copy over REG_CALL_ARG_LOCATION
4516         reg note.
4517         (note_outside_basic_block_p): Remove all references to
4518         NOTE_INSN_CALL_ARG_LOCATION.
4519         * gengtype.c (adjust_field_rtx_def): Likewise.
4520         * print-rtl.c (rtx_writer::print_rtx_operand_code_0, print_insn):
4521         Likewise.
4522         * jump.c (cleanup_barriers, delete_related_insns): Likewise.
4523         * cfgrtl.c (force_nonfallthru_and_redirect): Likewise.
4524
4525         PR c++/84444
4526         * builtins.c (builtin_mathfn_code): Don't check if CALL_EXPR_FN (t)
4527         is ADDR_EXPR.
4528
4529         PR tree-optimization/84452
4530         * tree-vect-patterns.c (vect_recog_pow_pattern): Don't call
4531         expand_simd_clones if targetm.simd_clone.compute_vecsize_and_simdlen
4532         is NULL.
4533
4534 2018-02-19  Martin Liska  <mliska@suse.cz>
4535
4536         PR sanitizer/82183
4537         * passes.def: Put pass_sancov_O0 before pass_lower_switch with -O0.
4538
4539 2018-02-19  Martin Liska  <mliska@suse.cz>
4540             Richard Sandiford  <richard.sandiford@linaro.org>
4541
4542         PR tree-optimization/82491
4543         * gimple-fold.c (get_base_constructor): Make earlier bail out
4544         to prevent ubsan.
4545
4546 2018-02-19  Carl Love  <cel@us.ibm.com>
4547
4548         * config/rs6000/rs6000-builtin.def: Change NEG macro expansions from
4549         BU_ALTIVEC_A to BU_P8V_AV_1 and BU_ALTIVEC_OVERLOAD_1 to
4550         BU_P8V_OVERLOAD_1.
4551         * config/rs6000/rs6000-c.c: Change ALTIVEC_BUILTIN_VEC_NEG to
4552         P8V_BUILTIN_VEC_NEG.
4553
4554 2018-02-19  Sebastian Perta  <sebastian.perta@renesas.com>
4555
4556         * config/rl78/rl78.md (movdf): New define expand.
4557
4558 2018-02-19  Martin Liska  <mliska@suse.cz>
4559
4560         PR other/80589
4561         * doc/invoke.texi: Fix typo.
4562         * params.def (PARAM_MAX_LOOP_HEADER_INSNS): Likewise.
4563
4564 2018-02-18  Segher Boessenkool  <segher@kernel.crashing.org>
4565
4566         * config/rs6000/rs6000.c (rs6000_option_override_internal): Don't
4567         handle rs6000_single_float and rs6000_double_float specially for
4568         e500 family CPUs.
4569
4570 2018-02-16  Jeff Law  <law@redhat.com>
4571
4572         * config/rx/rx.c (add_pop_cfi_notes): New function.;
4573         (pop_regs): Use it.
4574
4575 2018-02-16  Jakub Jelinek  <jakub@redhat.com>
4576
4577         PR ipa/84425
4578         * ipa-inline.c (inline_small_functions): Fix a typo.
4579
4580 2018-02-16  Nathan Sidwell  <nathan@acm.org>
4581
4582         * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
4583
4584 2018-02-16  Carl Love  <cel@us.ibm.com>
4585
4586         * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
4587         Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
4588         from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
4589         * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
4590         expansion to P8V_BUILTIN_VEC_FLOAT2.
4591
4592 2018-02-16  Vladimir Makarov  <vmakarov@redhat.com>
4593
4594         PR rtl-optimization/70023
4595         * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
4596         src_regno into account.
4597
4598 2018-02-16  Carl Love  <cel@us.ibm.com>
4599
4600         * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
4601         * config/rs6000/rs6000-builtin.def: Remove macro expansion for
4602         VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
4603         * config/rs6000/rs6000.c: Remove case statements for
4604         P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
4605         P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
4606         and P9V_BUILTIN_VEC_VINSERT4B.
4607         * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
4608         P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
4609         * config/rs6000/vsx.md:
4610         * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
4611         vec_insert4b.
4612
4613 2018-02-16  Carl Love  <cel@us.ibm.com>
4614
4615         * config/rs6000/altivec.h: Add builtin names vec_extract4b
4616         vec_insert4b.
4617         * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
4618         definitions.
4619         * config/rs6000/rs6000-c.c: Add the definitions for
4620         P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
4621         * config/rs6000/rs6000.c (altivec_expand_builtin): Add
4622         P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
4623         * config/rs6000/vsx.md: Add define_insn extract4b.  Add define_expand
4624         definition for insert4b and define insn *insert3b_internal.
4625         * doc/extend.texi: Add documentation for vec_extract4b.
4626
4627 2018-02-16  Nathan Sidwell  <nathan@acm.org>
4628
4629         * doc/extend.texi (Backwards Compatibility): Mention friend
4630         injection.  Note for-scope is deprecated.
4631         * doc/invoke.texi (-ffriend-injection): Deprecate.
4632
4633 2018-02-16  Segher Boessenkool  <segher@kernel.crashing.org>
4634
4635         * combine.c (try_combine): When adjusting LOG_LINKS for the destination
4636         that moved to I2, also allow destinations that are a paradoxical
4637         subreg (instead of a normal reg).
4638
4639 2018-02-16  Oleg Endo  <olegendo@gcc.gnu.org>
4640
4641         PR target/83831
4642         * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
4643         to QImode.
4644
4645 2018-02-16  Richard Biener  <rguenther@suse.de>
4646
4647         PR tree-optimization/84037
4648         PR tree-optimization/84016
4649         PR target/82862
4650         * config/i386/i386.c (ix86_builtin_vectorization_cost):
4651         Adjust vec_construct for the fact we need additional higher latency
4652         128bit inserts for AVX256 and AVX512 vector builds.
4653         (ix86_add_stmt_cost): Scale vector construction cost for
4654         elementwise loads.
4655
4656 2018-02-16  Richard Biener  <rguenther@suse.de>
4657
4658         PR tree-optimization/84417
4659         * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
4660         the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
4661         (non_rewritable_lvalue_p): Likewise, use poly-ints.
4662
4663 2018-02-16  Martin Liska  <mliska@suse.cz>
4664
4665         PR sanitizer/84307
4666         * internal-fn.def (ASAN_CHECK): Set proper flags.
4667         (ASAN_MARK): Likewise.
4668
4669 2018-02-16  Julia Koval  <julia.koval@intel.com>
4670
4671         * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
4672         from PTA_CANNONLAKE.
4673
4674 2018-02-16  Jakub Jelinek  <jakub@redhat.com>
4675
4676         PR target/84272
4677         * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
4678         Use ++iter rather than iter++ for std::list iterators.
4679         (func_fma_steering::dfs): Likewise.  Don't delete nodes right away,
4680         defer deleting them until all nodes in the forest are processed.  Do
4681         free even leaf nodes.  Change to_process into auto_vec.
4682
4683         PR bootstrap/84405
4684         * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
4685         * vec.h (vec_default_construct): Use memset instead of placement new
4686         if BROKEN_VALUE_INITIALIZATION is defined.
4687         * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
4688         memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
4689         is defined.
4690
4691         PR rtl-optimization/83723
4692         * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
4693         * lra.c (lra_substitute_pseudo): Likewise.  If true, use
4694         gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG.  Pass DEBUG_P to
4695         recursive calls.
4696         (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
4697         callers.
4698         * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
4699
4700 2018-02-16  Eric Botcazou  <ebotcazou@adacore.com>
4701
4702         PR rtl-optimization/81443
4703         * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
4704         from inner REGs to paradoxical SUBREGs.
4705
4706 2018-02-16  Richard Biener  <rguenther@suse.de>
4707
4708         PR tree-optimization/84399
4709         * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
4710         For operands we can analyze at their definition make sure we can
4711         analyze them at each use as well.
4712
4713 2018-02-16  Richard Biener  <rguenther@suse.de>
4714
4715         PR tree-optimization/84190
4716         * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
4717         volatile accesses if the decl isn't volatile.
4718
4719 2018-02-15  Jason Merrill  <jason@redhat.com>
4720
4721         PR c++/84314 - ICE with templates and fastcall attribute.
4722         * attribs.c (build_type_attribute_qual_variant): Don't clobber
4723         TYPE_CANONICAL on an existing type.
4724
4725 2018-02-15  Jakub Jelinek  <jakub@redhat.com>
4726
4727         PR tree-optimization/84383
4728         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
4729         dstoff nor call operand_equal_p if dstbase is NULL.
4730
4731         PR tree-optimization/84334
4732         * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
4733         also a CONSTANT_CLASS_P, punt.
4734
4735 2018-02-14  Jim Wilson  <jimw@sifive.com>
4736
4737         * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
4738         first SMALL_OPERAND check.  New local min_second_step.  Move assert
4739         to where locals are set.  Add TARGET_RVC support.
4740         * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
4741
4742 2018-02-14  Indu Bhagat  <indu.bhagat@oracle.com>
4743
4744         * doc/invoke.texi: Correct -Wformat-overflow code sample.
4745
4746 2018-02-14  Martin Sebor  <msebor@redhat.com>
4747
4748         PR tree-optimization/83698
4749         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
4750         arrays constrain the offset range to their bounds.
4751         (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
4752         (builtin_access::overlap): Avoid setting the size of overlap if it's
4753         already been set.
4754         (maybe_diag_overlap): Also consider arrays when deciding what values
4755         of offsets to include in diagnostics.
4756
4757 2018-02-14  Martin Sebor  <msebor@redhat.com>
4758
4759         PR c/84108
4760         * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
4761         that correspond to the kind of a declaration.
4762
4763 2018-02-14  John David Anglin  <danglin@gcc.gnu.org>
4764
4765         PR target/83984
4766         * config/pa/pa.md: Load address of PIC label using the linkage table
4767         if the label is nonlocal.
4768
4769 2018-02-14  Kelvin Nilsen  <kelvin@gcc.gnu.org>
4770
4771         * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
4772         warning message if user requests -maltivec=be.
4773         * doc/invoke.texi: Document deprecation of -maltivec=be.
4774
4775 2018-02-14  Will Schmidt  <will_schmidt@vnet.ibm.com>
4776
4777         PR target/84220
4778         * config/rs6000/rs6000-c.c: Update definitions for
4779         ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
4780         VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
4781
4782 2018-02-14  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
4783
4784         PR target/84239
4785         * config/i386/cetintrin.h: Remove _rdssp[d|q] and
4786         add _get_ssp intrinsics. Remove argument from
4787         __builtin_ia32_rdssp[d|q].
4788         * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
4789         * config/i386/i386-builtin.def: Remove argument from
4790         __builtin_ia32_rdssp[d|q].
4791         * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
4792         ix86_expand_special_args_builtin for _rdssp[d|q].
4793         * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
4794         Clear register before usage.
4795         * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
4796         Add documentation for new _get_ssp and _inc_ssp intrinsics.
4797
4798 2018-02-14  Richard Sandiford  <richard.sandiford@linaro.org>
4799
4800         PR tree-optimization/84357
4801         * tree-data-ref.c (object_address_invariant_in_loop_p): Check
4802         operand 1 of an ARRAY_REF too.
4803
4804 2018-02-14  Oleg Endo  <olegendo@gcc.gnu.org>
4805
4806         PR target/83831
4807         * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
4808         rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
4809         declarations.
4810         (set_of_reg): New struct.
4811         (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
4812         * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
4813         rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
4814         functions.
4815         * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
4816         Split into bitclr, bitset, bitinvert patterns if appropriate.
4817         (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
4818         use rx_fuse_in_memory_bitop.
4819         (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
4820         to named insn, correct maximum insn length.
4821
4822 2018-02-14  Jozef Lawrynowicz  <jozefl.gcc@gmail.com>
4823
4824         PR target/79242
4825         * machmode.def: Define a complex mode for PARTIAL_INT.
4826         * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
4827         MODE_PARTIAL_INT.
4828         * doc/rtl.texi: Document CSPImode.
4829         * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
4830         handling.
4831         (msp430_hard_regno_nregs_with_padding): Likewise.
4832
4833 2018-02-13  Peter Bergner  <bergner@vnet.ibm.com>
4834
4835         PR target/84279
4836         * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
4837
4838 2018-02-13  Segher Boessenkool  <segher@kernel.crashing.org>
4839
4840         PR rtl-optimization/84169
4841         * combine.c (try_combine): New variable split_i2i3.  Set it to true if
4842         we generated a parallel as new i3 and we split that to new i2 and i3
4843         instructions.  Handle split_i2i3 similar to swap_i2i3: scan the
4844         LOG_LINKs of i3 to see which of those need to link to i2 now.  Link
4845         those to i2, not i1.  Partially rewrite this scan code.
4846
4847 2018-02-13  Jakub Jelinek  <jakub@redhat.com>
4848
4849         PR c/82210
4850         * stor-layout.c (place_field): For variable length fields, adjust
4851         offset_align afterwards not just based on the field's alignment,
4852         but also on the size.
4853
4854         PR middle-end/84309
4855         * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
4856         of exps and logs in the use_exp2 case.
4857
4858 2018-02-13  Jeff Law  <law@redhat.com>
4859
4860         * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
4861         entry for "vector".
4862
4863         * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
4864         ARGS as unused.
4865
4866 2018-02-13  Alexandre Oliva  <aoliva@redhat.com>
4867
4868         PR debug/84342
4869         PR debug/84319
4870         * common.opt (gas-loc-support, gas-locview-support): New.
4871         (ginline-points, ginternal-reset-location-views): New.
4872         * doc/invoke.texi: Document them.  Use @itemx where intended.
4873         (gvariable-location-views): Adjust.
4874         * target.def (reset_location_view): New.
4875         * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
4876         (TARGET_RESET_LOCATION_VIEW): New.
4877         * doc/tm.texi: Rebuilt.
4878         * dwarf2out.c (dwarf2out_default_as_loc_support): New.
4879         (dwarf2out_default_as_locview_support): New.
4880         (output_asm_line_debug_info): Use option variables.
4881         (dwarf2out_maybe_output_loclist_view_pair): Likewise.
4882         (output_loc_list): Likewise.
4883         (add_high_low_attributes): Check option variables.
4884         Don't output entry view attribute in strict mode.
4885         (gen_inlined_subroutine_die): Check option variables.
4886         (dwarf2out_inline_entry): Likewise.
4887         (init_sections_and_labels): Likewise.
4888         (dwarf2out_early_finish): Likewise.
4889         (maybe_reset_location_view): New, from...
4890         (dwarf2out_var_location): ... here.  Call it.
4891         * debug.h (dwarf2out_default_as_loc_support): Declare.
4892         (dwarf2out_default_as_locview_support): Declare.
4893         * hooks.c (hook_int_rtx_insn_0): New.
4894         * hooks.h (hook_int_rtx_insn_0): Declare.
4895         * toplev.c (process_options): Take -gas-loc-support and
4896         -gas-locview-support from dwarf2out.  Enable
4897         -gvariable-location-views by default only with locview
4898         assembler support.  Enable -ginternal-reset-location-views by
4899         default only if the target defines the corresponding hook.
4900         Enable -ginline-points by default if location views are
4901         enabled; force it disabled if statement frontiers are
4902         disabled.
4903         * tree-inline.c (expand_call_inline): Check option variables.
4904         * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
4905
4906 2018-02-13  Richard Sandiford  <richard.sandiford@linaro.org>
4907
4908         PR tree-optimization/84321
4909         * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
4910         handling.  Also check whether the anti-range contains any values
4911         that satisfy the mask; switch to a VR_RANGE if not.
4912
4913 2018-02-13  Paolo Bonzini  <bonzini@gnu.org>
4914
4915         PR sanitizer/84340
4916         * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
4917
4918 2018-02-13  Martin Jambor  <mjambor@suse.cz>
4919
4920         PR c++/83990
4921         * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
4922         of call statements, also set location of a load to a temporary.
4923
4924 2018-02-13  Sebastian Perta  <sebastian.perta@renesas.com>
4925
4926         * config/rl78/rl78.c (add_vector_labels): New function.
4927         * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
4928         * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
4929         * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
4930         which checks that no arguments are passed.
4931         * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
4932         * doc/extend.texi: Documentation for the new attribute.
4933
4934 2018-02-13  Andreas Schwab  <schwab@suse.de>
4935
4936         * config/riscv/linux.h (CPP_SPEC): Define.
4937
4938 2018-02-13  Jakub Jelinek  <jakub@redhat.com>
4939
4940         PR target/84335
4941         * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
4942         OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
4943         OPTION_MASK_ISA_AES as first argument to def_builtin_const
4944         for AES builtins.  Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
4945         instead of OPTION_MASK_ISA_PCLMUL as first argument to
4946         def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
4947         * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
4948         temporarily for AES and PCLMUL builtins.
4949
4950         PR tree-optimization/84339
4951         * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
4952         ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
4953         Formatting fixes.
4954
4955         PR middle-end/84309
4956         * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
4957         exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
4958         * generic-match-head.c (canonicalize_math_after_vectorization_p): New
4959         inline function.
4960         * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
4961         inline function.
4962         * omp-simd-clone.h: New file.
4963         * omp-simd-clone.c: Include omp-simd-clone.h.
4964         (expand_simd_clones): No longer static.
4965         * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
4966         cgraph.h and omp-simd-clone.h.
4967         (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
4968         (vect_recog_widen_shift_pattern): Formatting fix.
4969         (vect_pattern_recog_1): Don't check optab for calls.
4970
4971         PR target/84336
4972         * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
4973         operands[2] into a REG before using gen_lowpart on it.
4974
4975 2018-02-12  Jeff Law  <law@redhat.com>
4976
4977         PR target/83760
4978         * config/sh/sh.c (find_barrier): Consider a sibling call
4979         a barrier as well.
4980
4981         * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
4982         successfully back substituting a reg.
4983
4984 2018-02-12  Richard Biener  <rguenther@suse.de>
4985
4986         PR tree-optimization/84037
4987         * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
4988         parameter, move visited init to caller.
4989         (vect_slp_analyze_operations): Separate cost from validity
4990         check, initialize visited once for all instances.
4991         (vect_schedule_slp): Analyze map to CSE vectorized nodes once
4992         for all instances.
4993         * tree-vect-stmts.c (vect_model_simple_cost): Make early
4994         out an assert.
4995         (vect_model_promotion_demotion_cost): Likewise.
4996         (vectorizable_bswap): Guard cost modeling with !slp_node
4997         instead of !PURE_SLP_STMT to avoid double-counting on hybrid
4998         SLP stmts.
4999         (vectorizable_call): Likewise.
5000         (vectorizable_conversion): Likewise.
5001         (vectorizable_assignment): Likewise.
5002         (vectorizable_shift): Likewise.
5003         (vectorizable_operation): Likewise.
5004         (vectorizable_store): Likewise.
5005         (vectorizable_load): Likewise.
5006         (vectorizable_condition): Likewise.
5007         (vectorizable_comparison): Likewise.
5008
5009 2018-02-12  Paolo Bonzini  <bonzini@gnu.org>
5010
5011         PR sanitizer/84307
5012         * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
5013         (ASAN_MARK): Fix fnspec to account for return value, change pointer
5014         argument from 'R' to 'W' so that the pointed-to datum is clobbered.
5015
5016 2018-02-08  Jan Hubicka  <hubicka@ucw.cz>
5017
5018         PR middle-end/83665
5019         * params.def (inline-min-speedup): Increase from 8 to 15.
5020         (max-inline-insns-auto): Decrease from 40 to 30.
5021         * ipa-split.c (consider_split): Add some buffer for function to
5022         be considered inlining candidate.
5023         * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
5024         default values.
5025
5026 2018-02-12  Richard Biener  <rguenther@suse.de>
5027
5028         PR tree-optimization/84037
5029         * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
5030         matched stmts if we cannot swap the non-matched ones.
5031
5032 2018-02-12  Olga Makhotina  <olga.makhotina@intel.com>
5033
5034         * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
5035         _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
5036         _mm_maskz_scalef_round_ss): New intrinsics.
5037         (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
5038         * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
5039         __builtin_ia32_scalefss_round): Remove.
5040         (__builtin_ia32_scalefsd_mask_round,
5041         __builtin_ia32_scalefss_mask_round): New intrinsics.
5042         * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
5043         (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
5044         ((match_operand:VF_128 2 "<round_nimm_predicate>"
5045         "<round_constraint>")): Changed to ...
5046         ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
5047         "<round_scalar_constraint>")): ... this.
5048         ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
5049         %0, %1, %2<round_op3>}"): Changed to ...
5050         ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
5051         %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
5052         %2<round_scalar_mask_op3>}"): ... this.
5053         * config/i386/subst.md (round_scalar_nimm_predicate): New.
5054
5055 2018-02-12  Olga Makhotina  <olga.makhotina@intel.com>
5056
5057         * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
5058         (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
5059         (_mm_maskz_sqrt_round_ss): New intrinsics.
5060         (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
5061         (__builtin_ia32_sqrtsd_mask_round)
5062         (__builtin_ia32_sqrtss_mask_round): New builtins.
5063         * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
5064         (__builtin_ia32_sqrtss_round): Remove.
5065         (__builtin_ia32_sqrtsd_mask_round)
5066         (__builtin_ia32_sqrtss_mask_round): New builtins.
5067         * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
5068         (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
5069         ((match_operand:VF_128 1 "vector_operand"
5070         "xBm,<round_constraint>")): Changed to ...
5071         ((match_operand:VF_128 1 "vector_operand"
5072         "xBm,<round_scalar_constraint>")): ... this.
5073         (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
5074         %0, %2, %<iptr>1<round_op3>}): Changed to ...
5075         (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
5076         %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
5077         %<iptr>1<round_scalar_mask_op3>}): ... this.
5078         ((set_attr "prefix" "<round_prefix>")): Changed to ...
5079         ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
5080
5081 2018-02-11  Steven Munroe  <munroesj@gcc.gnu.org>
5082
5083         PR target/84266
5084         * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
5085         Cast vec_cmpeq result to correct type.
5086         * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
5087         Cast vec_cmpgt result to correct type.
5088
5089 2018-02-11  Alexandre Oliva  <aoliva@redhat.com>
5090
5091         * final.c (final_scan_insn_1): Renamed from...
5092         (final_scan_insn): ... this.  New wrapper, to recover
5093         seen from the outermost call in recursive ones.
5094         * config/sparc/sparc.c (output_return): Drop seen from call.
5095         (output_sibcall): Likewise.
5096         * config/visium/visium.c (output_branch): Likewise.
5097
5098 2018-02-10  John David Anglin  <danglin@gcc.gnu.org>
5099
5100         * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
5101         function label.
5102
5103 2018-02-10  Alan Modra  <amodra@gmail.com>
5104
5105         PR target/84300
5106         * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
5107         Specify LR as an input.
5108
5109 2018-02-10  Jakub Jelinek  <jakub@redhat.com>
5110
5111         PR sanitizer/83987
5112         * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
5113         remove_member_access_dummy_vars): New functions.
5114         (lower_omp_for, lower_omp_taskreg, lower_omp_target,
5115         lower_omp_1, execute_lower_omp): Use them.
5116
5117         PR rtl-optimization/84308
5118         * shrink-wrap.c (spread_components): Release todo vector.
5119
5120 2018-02-09  Vladimir Makarov  <vmakarov@redhat.com>
5121
5122         PR rtl-optimization/57193
5123         * ira-color.c (struct allocno_color_data): Add member
5124         conflict_allocno_hard_prefs.
5125         (update_conflict_allocno_hard_prefs): New.
5126         (bucket_allocno_compare_func): Add a preference based on
5127         conflict_allocno_hard_prefs.
5128         (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
5129         (color_allocnos): Remove a dead code.  Initiate
5130         conflict_allocno_hard_prefs.  Call update_costs_from_prefs.
5131
5132 2018-02-09  Jakub Jelinek  <jakub@redhat.com>
5133
5134         PR target/84226
5135         * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
5136         constraint from =wa to wa.  Avoid a subreg on the output operand,
5137         instead use a pseudo and subreg it in a move.
5138         (p9_xxbrd_<mode>): Changed to ...
5139         (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
5140         (p9_xxbrd_v2df): New expander.
5141         (p9_xxbrw_<mode>): Changed to ...
5142         (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
5143         (p9_xxbrw_v4sf): New expander.
5144
5145 2018-02-09  Sebastian Perta  <sebastian.perta@renesas.com>
5146
5147         * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
5148
5149 2018-02-09  Peter Bergner  <bergner@vnet.ibm.com>
5150
5151         PR target/83926
5152         * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
5153         multiply in 32-bit mode.
5154         (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
5155         (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
5156         mode.
5157
5158 2018-02-09  Sebastian Perta  <sebastian.perta@renesas.com>
5159
5160         * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
5161         to allow or block "symbol_ref" depending on the value of TARGET_JSR.
5162         * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
5163         * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
5164
5165 2018-02-09  Pierre-Marie de Rodat  <derodat@adacore.com>
5166
5167         PR lto/84213
5168         * dwarf2out.c (is_trivial_indirect_ref): New function.
5169         (dwarf2out_late_global_decl): Do not generate a location
5170         attribute for variables that have a non-trivial DECL_VALUE_EXPR
5171         and that are not defined in the current unit.
5172
5173 2018-02-09  Eric Botcazou  <ebotcazou@adacore.com>
5174
5175         * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
5176         instead of a libcall for UNORDERED.
5177
5178 2018-02-09  Tamar Christina  <tamar.christina@arm.com>
5179
5180         PR target/82641
5181         * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
5182         __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
5183
5184 2018-02-09  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
5185
5186         PR target/PR84295
5187         * config/s390/s390.c (s390_set_current_function): Invoke
5188         s390_indirect_branch_settings also if fndecl didn't change.
5189
5190 2018-02-09  Alexandre Oliva  <aoliva@redhat.com>
5191
5192         * config/rs6000/rs6000.md (blockage): Set length to zero.
5193
5194 2018-02-09  Eric Botcazou  <ebotcazou@adacore.com>
5195
5196         * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
5197
5198 2018-02-09  Jakub Jelinek  <jakub@redhat.com>
5199
5200         PR sanitizer/84285
5201         * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
5202         STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
5203         -static-lib*san.
5204
5205         PR debug/84252
5206         * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
5207         PARALLEL incoming that failed vt_get_decl_and_offset check.
5208
5209         PR middle-end/84237
5210         * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
5211         * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
5212         TREE_READONLY bit.
5213         (get_variable_section): For decls in named .bss* sections pass true as
5214         second argument to bss_initializer_p.
5215
5216 2018-02-09  Marek Polacek  <polacek@redhat.com>
5217             Jakub Jelinek  <jakub@redhat.com>
5218
5219         PR c++/83659
5220         * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
5221         Formatting fixes.  Verify first that tree_fits_poly_int64_p (op01).
5222         Sync some changes from cxx_fold_indirect_ref.
5223
5224 2018-02-09  Alexandre Oliva  <aoliva@redhat.com>
5225
5226         * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
5227         markers.
5228         * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
5229         (BLOCK_INLINE_ENTRY_LABEL): New.
5230         (dwarf2out_var_location): Disregard inline entry markers.
5231         (inline_entry_data): New struct.
5232         (inline_entry_data_hasher): New hashtable type.
5233         (inline_entry_data_hasher::hash): New.
5234         (inline_entry_data_hasher::equal): New.
5235         (inline_entry_data_table): New variable.
5236         (add_high_low_attributes): Add DW_AT_entry_pc and
5237         DW_AT_GNU_entry_view attributes if a pending entry is found
5238         in inline_entry_data_table.  Add old entry_pc attribute only
5239         if debug nonbinding markers are disabled.
5240         (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
5241         markers are enabled.
5242         (block_within_block_p, dwarf2out_inline_entry): New.
5243         (dwarf2out_finish): Check that no entries remained in
5244         inline_entry_data_table.
5245         * final.c (reemit_insn_block_notes): Handle inline entry notes.
5246         (final_scan_insn, notice_source_line): Likewise.
5247         (rest_of_clean_state): Skip inline entry markers.
5248         * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
5249         markers.
5250         * gimple.c (gimple_build_debug_inline_entry): New.
5251         * gimple.h (enum gimple_debug_subcode): Add
5252         GIMPLE_DEBUG_INLINE_ENTRY.
5253         (gimple_build_debug_inline_entry): Declare.
5254         (gimple_debug_inline_entry_p): New.
5255         (gimple_debug_nonbind_marker_p): Adjust.
5256         * insn-notes.def (INLINE_ENTRY): New.
5257         * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
5258         inline entry marker notes.
5259         (print_insn): Likewise.
5260         * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
5261         (INSN_DEBUG_MARKER_KIND): Likewise.
5262         (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
5263         * tree-inline.c (expand_call_inline): Build and insert
5264         debug_inline_entry stmt.
5265         * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
5266         inline entry blocks early, if nonbind markers are enabled.
5267         (dump_scope_block): Dump fragment info.
5268         * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
5269         * doc/gimple.texi (gimple_debug_inline_entry_p): New.
5270         (gimple_build_debug_inline_entry): New.
5271         * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
5272         Enable/disable inline entry points too.
5273         * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
5274         (DEBUG_INSN): Describe inline entry markers.
5275
5276         * common.opt (gvariable-location-views): New.
5277         (gvariable-location-views=incompat5): New.
5278         * config.in: Rebuilt.
5279         * configure: Rebuilt.
5280         * configure.ac: Test assembler for view support.
5281         * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
5282         * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
5283         * dwarf2out.c (var_loc_view): New typedef.
5284         (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
5285         (dwarf2out_locviews_in_attribute): New.
5286         (dwarf2out_locviews_in_loclist): New.
5287         (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
5288         (enum dw_line_info_opcode): Add LI_adv_address.
5289         (struct dw_line_info_table): Add view.
5290         (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
5291         (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
5292         (zero_view_p): New variable.
5293         (ZERO_VIEW_P): New macro.
5294         (output_asm_line_debug_info): New.
5295         (struct var_loc_node): Add view.
5296         (add_AT_view_list, AT_loc_list): New.
5297         (add_var_loc_to_decl): Add view param.  Test it against last.
5298         (new_loc_list): Add view params.  Record them.
5299         (AT_loc_list_ptr): Handle loc and view lists.
5300         (view_list_to_loc_list_val_node): New.
5301         (print_dw_val): Handle dw_val_class_view_list.
5302         (size_of_die): Likewise.
5303         (value_format): Likewise.
5304         (loc_list_has_views): New.
5305         (gen_llsym): Set vl_symbol too.
5306         (maybe_gen_llsym, skip_loc_list_entry): New.
5307         (dwarf2out_maybe_output_loclist_view_pair): New.
5308         (output_loc_list): Output view list or entries too.
5309         (output_view_list_offset): New.
5310         (output_die): Handle dw_val_class_view_list.
5311         (output_dwarf_version): New.
5312         (output_compilation_unit_header): Use it.
5313         (output_skeleton_debug_sections): Likewise.
5314         (output_rnglists, output_line_info): Likewise.
5315         (output_pubnames, output_aranges): Update version comments.
5316         (output_one_line_info_table): Output view numbers in asm comments.
5317         (dw_loc_list): Determine current endview, pass it to new_loc_list.
5318         Call maybe_gen_llsym.
5319         (loc_list_from_tree_1): Adjust.
5320         (add_AT_location_description): Create view list attribute if
5321         needed, check it's absent otherwise.
5322         (convert_cfa_to_fb_loc_list): Adjust.
5323         (maybe_emit_file): Call output_asm_line_debug_info for test.
5324         (dwarf2out_var_location): Reset views as needed.  Precompute
5325         add_var_loc_to_decl args.  Call get_attr_min_length only if we have the
5326         attribute.  Set view.
5327         (new_line_info_table): Reset next view.
5328         (set_cur_line_info_table): Call output_asm_line_debug_info for test.
5329         (dwarf2out_source_line): Likewise.  Output view resets and labels to
5330         the assembler, or select appropriate line info opcodes.
5331         (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
5332         (optimize_string_length): Catch it.  Adjust.
5333         (resolve_addr): Copy vl_symbol along with ll_symbol.  Handle
5334         dw_val_class_view_list, and remove it if no longer needed.
5335         (hash_loc_list): Hash view numbers.
5336         (loc_list_hasher::equal): Compare them.
5337         (optimize_location_lists): Check whether a view list symbol is
5338         needed, and whether the locview attribute is present, and
5339         whether they match.  Remove the locview attribute if no longer
5340         needed.
5341         (index_location_lists): Call skip_loc_list_entry for test.
5342         (dwarf2out_finish): Call output_asm_line_debug_info for test.
5343         Use output_dwarf_version.
5344         * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
5345         (struct dw_val_node): Add val_view_list.
5346         * final.c (SEEN_NEXT_VIEW): New.
5347         (set_next_view_needed): New.
5348         (clear_next_view_needed): New.
5349         (maybe_output_next_view): New.
5350         (final_start_function): Rename to...
5351         (final_start_function_1): ... this.  Take pointer to FIRST,
5352         add SEEN parameter.  Emit param bindings in the initial view.
5353         (final_start_function): Reintroduce SEEN-less interface.
5354         (final): Rename to...
5355         (final_1): ... this.  Take SEEN parameter.  Output final pending
5356         next view at the end.
5357         (final): Reintroduce seen-less interface.
5358         (final_scan_insn): Output pending next view before switching
5359         sections or ending a block.  Mark the next view as needed when
5360         outputting variable locations.  Notify debug backend of section
5361         changes, and of location view changes.
5362         (rest_of_handle_final): Adjust.
5363         * toplev.c (process_options): Autodetect value for debug variable
5364         location views option.  Warn on incompat5 without -gdwarf-5.
5365         * doc/invoke.texi (gvariable-location-views): New.
5366         (gvariable-location-views=incompat5): New.
5367         (gno-variable-location-views): New.
5368
5369 2018-02-08  David Malcolm  <dmalcolm@redhat.com>
5370
5371         PR tree-optimization/84136
5372         * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
5373         that the result of find_edge is non-NULL.
5374
5375 2018-02-08  Sergey Shalnov  <sergey.shalnov@intel.com>
5376
5377         PR target/83008
5378         * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
5379         storing integer register in SImode.  Fix cost of 256 and 512
5380         byte aligned SSE register store.
5381
5382 2018-02-08  Sergey Shalnov  <sergey.shalnov@intel.com>
5383
5384         * config/i386/i386.c (ix86_multiplication_cost): Fix
5385         multiplication cost for TARGET_AVX512DQ.
5386
5387 2018-02-08  Marek Polacek  <polacek@redhat.com>
5388
5389         PR tree-optimization/84238
5390         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
5391         get_range_strlen.
5392
5393 2018-02-08  Richard Sandiford  <richard.sandiford@linaro.org>
5394
5395         PR tree-optimization/84265
5396         * tree-vect-stmts.c (vectorizable_store): Don't treat
5397         VMAT_CONTIGUOUS accesses as grouped.
5398         (vectorizable_load): Likewise.
5399
5400 2018-02-08  Richard Sandiford  <richard.sandiford@linaro.org>
5401
5402         PR tree-optimization/81635
5403         * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
5404         * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
5405         (test_round_for_mask): New functions.
5406         (wide_int_cc_tests): Call test_round_for_mask.
5407         * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
5408         * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
5409         * tree-data-ref.c (split_constant_offset_1): Use it to refine the
5410         range returned by get_range_info.
5411
5412 2018-02-08  Jan Hubicka  <hubicka@ucw.cz>
5413
5414         PR ipa/81360
5415         * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
5416         * symtab.c: Include builtins.h
5417         (symtab_node::output_to_lto_symbol_table_p): Move here
5418         from lto-streamer-out.c:output_symbol_p.
5419         * lto-streamer-out.c (write_symbol): Turn early exit to assert.
5420         (output_symbol_p): Move all logic to symtab.c
5421         (produce_symtab): Update.
5422
5423 2018-02-08  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
5424
5425         * config/s390/s390-opts.h (enum indirect_branch): Define.
5426         * config/s390/s390-protos.h (s390_return_addr_from_memory)
5427         (s390_indirect_branch_via_thunk)
5428         (s390_indirect_branch_via_inline_thunk): Add function prototypes.
5429         (enum s390_indirect_branch_type): Define.
5430         * config/s390/s390.c (struct s390_frame_layout, struct
5431         machine_function): Remove.
5432         (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
5433         (indirect_branch_table_label_no, indirect_branch_table_name):
5434         Define variables.
5435         (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
5436         (enum s390_indirect_branch_option): Define.
5437         (s390_return_addr_from_memory): New function.
5438         (s390_handle_string_attribute): New function.
5439         (s390_attribute_table): Add new attribute handler.
5440         (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
5441         (s390_indirect_branch_via_thunk): New function.
5442         (s390_indirect_branch_via_inline_thunk): New function.
5443         (s390_function_ok_for_sibcall): When jumping via thunk disallow
5444         sibling call optimization for non z10 compiles.
5445         (s390_emit_call): Force indirect branch target to be a single
5446         register.  Add r1 clobber for non-z10 compiles.
5447         (s390_emit_epilogue): Emit return jump via return_use expander.
5448         (s390_reorg): Handle JUMP_INSNs as execute targets.
5449         (s390_option_override_internal): Perform validity checks for the
5450         new command line options.
5451         (s390_indirect_branch_attrvalue): New function.
5452         (s390_indirect_branch_settings): New function.
5453         (s390_set_current_function): Invoke s390_indirect_branch_settings.
5454         (s390_output_indirect_thunk_function):  New function.
5455         (s390_code_end): Implement target hook.
5456         (s390_case_values_threshold): Implement target hook.
5457         (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
5458         macros.
5459         * config/s390/s390.h (struct s390_frame_layout)
5460         (struct machine_function): Move here from s390.c.
5461         (TARGET_INDIRECT_BRANCH_NOBP_RET)
5462         (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
5463         (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
5464         (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
5465         (TARGET_INDIRECT_BRANCH_NOBP_CALL)
5466         (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
5467         (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
5468         (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
5469         (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
5470         * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
5471         (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
5472         (mnemonic attribute): Add values which aren't recognized
5473         automatically.
5474         ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
5475         pattern for branch conversion.  Fix mnemonic attribute.
5476         ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
5477         indirect branch via thunk if requested.
5478         ("indirect_jump", "<code>"): Expand patterns for branch conversion.
5479         ("*indirect_jump"): Disable for branch conversion using out of
5480         line thunks.
5481         ("indirect_jump_via_thunk<mode>_z10")
5482         ("indirect_jump_via_thunk<mode>")
5483         ("indirect_jump_via_inlinethunk<mode>_z10")
5484         ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
5485         ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
5486         ("casesi_jump_via_inlinethunk<mode>_z10")
5487         ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
5488         ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
5489         ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
5490         ("*indirect2_jump"): Disable for branch conversion.
5491         ("casesi_jump"): Turn into expander and expand patterns for branch
5492         conversion.
5493         ("return_use"): New expander.
5494         ("*return"): Emit return via thunk and rename it to ...
5495         ("*return<mode>"): ... this one.
5496         * config/s390/s390.opt: Add new options and and enum for the
5497         option values.
5498
5499 2018-02-08  Richard Sandiford  <richard.sandiford@linaro.org>
5500
5501         * lra-constraints.c (match_reload): Unconditionally use
5502         gen_lowpart_SUBREG, rather than selecting between that
5503         and equivalent gen_rtx_SUBREG code.
5504
5505 2018-02-08  Richard Biener  <rguenther@suse.de>
5506
5507         PR tree-optimization/84233
5508         * tree-ssa-phiprop.c (propagate_with_phi): Use separate
5509         changed flag instead of boguously re-using phi_inserted.
5510
5511 2018-02-08  Martin Jambor  <mjambor@suse.cz>
5512
5513         * hsa-gen.c (get_symbol_for_decl): Set program allocation for
5514         static local variables.
5515
5516 2018-02-08  Richard Biener  <rguenther@suse.de>
5517
5518         PR tree-optimization/84278
5519         * tree-vect-stmts.c (vectorizable_store): When looking for
5520         smaller vector types to perform grouped strided loads/stores
5521         make sure the mode is supported by the target.
5522         (vectorizable_load): Likewise.
5523
5524 2018-02-08  Wilco Dijkstra  <wdijkstr@arm.com>
5525
5526         * config/aarch64/aarch64.c (aarch64_components_for_bb):
5527         Increase LDP/STP opportunities by adding adjacent callee-saves.
5528
5529 2018-02-08  Wilco Dijkstra  <wdijkstr@arm.com>
5530
5531         PR rtl-optimization/84068
5532         PR rtl-optimization/83459
5533         * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
5534
5535 2018-02-08  Aldy Hernandez  <aldyh@redhat.com>
5536
5537         PR tree-optimization/84224
5538         * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
5539         * calls.c (gimple_alloca_call_p): Only return TRUE when we have
5540         non-zero arguments.
5541
5542 2018-02-07  Iain Sandoe  <iain@codesourcery.com>
5543
5544         PR target/84113
5545         * config/rs6000/altivec.md (*restore_world): Remove LR use.
5546         * config/rs6000/predicates.md (restore_world_operation): Adjust op
5547         count, remove one USE.
5548
5549 2018-02-07  Michael Meissner  <meissner@linux.vnet.ibm.com>
5550
5551         * doc/install.texi (Configuration): Document the
5552         --with-long-double-format={ibm,ieee} PowerPC configuration
5553         options.
5554
5555         PR target/84154
5556         * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
5557         Convert from define_expand to be define_insn_and_split.  Rework
5558         float/double/_Float128 conversions to QI/HI/SImode to work with
5559         both ISA 2.07 (power8) or ISA 3.0 (power9).  Fix regression where
5560         conversions to QI/HImode types did a store and then a load to
5561         truncate the value.  For conversions to VSX registers, don't split
5562         the insn, instead emit the code directly.  Use the code iterator
5563         any_fix to combine signed and unsigned conversions.
5564         (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
5565         (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
5566         (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
5567         (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
5568         (fix_<mode>di2_hw): Likewise.
5569         (fixuns_<mode>di2_hw): Likewise.
5570         (fix_<mode>si2_hw): Likewise.
5571         (fixuns_<mode>si2_hw): Likewise.
5572         (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
5573         (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
5574         (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
5575         fix<uns>_trunc<SFDF:mode>si2_p8.
5576         (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
5577         used.
5578         (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
5579         (fix<uns>_<mode>_mem): Likewise.
5580         (fctiw<u>z_<mode>_mem): Likewise.
5581         (fix<uns>_<mode>_mem): Likewise.
5582         (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
5583         the register allocator from doing a direct move to the GPRs to do
5584         a store, and instead use the ISA 3.0 store byte/half-word from
5585         vector register instruction.  For IEEE 128-bit floating point,
5586         also optimize stores of 32-bit ints.
5587         (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
5588
5589 2018-02-07  Alan Hayward  <alan.hayward@arm.com>
5590
5591         * genextract.c (push_pathstr_operand): New function to support
5592         [a-zA-Z].
5593         (walk_rtx): Call push_pathstr_operand.
5594         (print_path): Support [a-zA-Z].
5595
5596 2018-02-07  Richard Biener  <rguenther@suse.de>
5597
5598         PR tree-optimization/84037
5599         * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
5600         (cse_and_gimplify_to_preheader): Declare.
5601         (vect_get_place_in_interleaving_chain): Likewise.
5602         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
5603         ivexpr_map.
5604         (_loop_vec_info::~_loop_vec_info): Delete it.
5605         (cse_and_gimplify_to_preheader): New function.
5606         * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
5607         * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
5608         (vectorizable_load): Likewise.  For grouped stores always base
5609         the IV on the first element.
5610         * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
5611         condition before gimplifying.
5612
5613 2018-02-07  Jakub Jelinek  <jakub@redhat.com>
5614
5615         * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
5616         *DIV_EXPR and *MOD_EXPR.
5617
5618 2018-02-07  H.J. Lu  <hongjiu.lu@intel.com>
5619
5620         PR target/84248
5621         * config/i386/i386.c (ix86_option_override_internal): Mask out
5622         the CF_SET bit when checking -fcf-protection.
5623
5624 2018-02-07  Tom de Vries  <tom@codesourcery.com>
5625
5626         PR libgomp/84217
5627         * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
5628         enough.
5629
5630 2018-02-07  Richard Biener  <rguenther@suse.de>
5631
5632         PR tree-optimization/84204
5633         * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
5634         this place.
5635
5636         PR tree-optimization/84205
5637         * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
5638         special-case isl_ast_op_zdiv_r.
5639
5640         PR tree-optimization/84223
5641         * graphite-scop-detection.c (gather_bbs::before_dom_children):
5642         Only add conditions from within the region.
5643         (gather_bbs::after_dom_children): Adjust.
5644
5645 2018-02-07  Georg-Johann Lay  <avr@gjlay.de>
5646
5647         PR target/84209
5648         * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
5649         * config/avr/avr.md: Only post-reload split REG-REG moves if
5650         either register is GENERAL_REG_P.
5651
5652 2018-02-07  Jakub Jelinek  <jakub@redhat.com>
5653
5654         PR tree-optimization/84235
5655         * tree-ssa-scopedtables.c
5656         (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
5657         if the subtraction is performed in floating point type where NaNs are
5658         honored.  For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
5659         build 1.  Formatting fix.
5660
5661 2018-02-06  Jakub Jelinek  <jakub@redhat.com>
5662
5663         PR target/84146
5664         * config/i386/i386.c (rest_of_insert_endbranch): Only skip
5665         NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
5666         and skip it regardless of bb boundaries.  Use CALL_P macro,
5667         don't test INSN_P (insn) together with CALL_P or JUMP_P check
5668         unnecessarily, formatting fix.
5669
5670 2018-02-06  Michael Collison  <michael.collison@arm.com>
5671
5672         * config/arm/thumb2.md:
5673         (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
5674         (*thumb_mov_notscc): Ditto.
5675
5676 2018-02-06  Michael Meissner  <meissner@linux.vnet.ibm.com>
5677
5678         PR target/84154
5679         * config/rs6000/rs6000.md (su code attribute): Use "u" for
5680         unsigned_fix, not "s".
5681
5682 2018-02-06  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
5683
5684         * configure.ac (gcc_fn_eh_frame_ro): New function.
5685         (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
5686         correct .eh_frame permissions.
5687         * configure: Regenerate.
5688
5689 2018-02-06  Andrew Jenner  <andrew@codeourcery.com>
5690
5691         * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
5692         irrelevant options.
5693
5694 2018-02-06  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
5695
5696         * config/rs6000/rs6000.c (rs6000_option_override_internal):
5697         Display warning message for -mno-speculate-indirect-jumps.
5698
5699 2018-02-06  Andrew Jenner  <andrew@codesourcery.com>
5700
5701         * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
5702         Undocumented.
5703         * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
5704
5705 2018-02-06  Aldy Hernandez  <aldyh@redhat.com>
5706
5707         PR tree-optimization/84225
5708         * tree-eh.c (find_trapping_overflow): Only call
5709         operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
5710
5711 2018-02-06  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
5712
5713         PR target/84145
5714         * config/i386/i386.c: Reimplement the check of possible options
5715         -mibt/-mshstk conbination. Change error messages.
5716         * doc/invoke.texi: Fix a typo: remove extra '='.
5717
5718 2018-02-06  Marek Polacek  <polacek@redhat.com>
5719
5720         PR tree-optimization/84228
5721         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
5722
5723 2018-02-06  Tamar Christina  <tamar.christina@arm.com>
5724
5725         PR target/82641
5726         * config/arm/arm.c (arm_print_asm_arch_directives): Record already
5727         emitted arch directives.
5728         * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
5729         __ARM_FEATURE_COPROC before changing architectures.
5730
5731 2018-02-06  Richard Biener  <rguenther@suse.de>
5732
5733         * config/i386/i386.c (print_reg): Fix typo.
5734         (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
5735
5736 2018-02-06  Eric Botcazou  <ebotcazou@adacore.com>
5737
5738         * configure: Regenerate.
5739
5740 2018-02-05  Martin Sebor  <msebor@redhat.com>
5741
5742         PR tree-optimization/83369
5743         * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
5744         inlining context.
5745
5746 2018-02-05  Martin Liska  <mliska@suse.cz>
5747
5748         * doc/invoke.texi: Cherry-pick upstream r323995.
5749
5750 2018-02-05  Richard Sandiford  <richard.sandiford@linaro.org>
5751
5752         * ira.c (ira_init_register_move_cost): Adjust comment.
5753
5754 2018-02-05  Martin Liska  <mliska@suse.cz>
5755
5756         PR gcov-profile/84137
5757         * doc/gcov.texi: Fix typo in documentation.
5758
5759 2018-02-05  Martin Liska  <mliska@suse.cz>
5760
5761         PR gcov-profile/83879
5762         * doc/gcov.texi: Document necessity of --dynamic-list-data when
5763         using dlopen functionality.
5764
5765 2018-02-05  Olga Makhotina  <olga.makhotina@intel.com>
5766
5767         * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
5768         _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
5769         _mm_maskz_range_ss, _mm_mask_range_round_ss,
5770         _mm_maskz_range_round_ss): New intrinsics.
5771         (__builtin_ia32_rangesd128_round)
5772         (__builtin_ia32_rangess128_round): Remove.
5773         (__builtin_ia32_rangesd128_mask_round,
5774         __builtin_ia32_rangess128_mask_round): New builtins.
5775         * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
5776         __builtin_ia32_rangess128_round): Remove.
5777         (__builtin_ia32_rangesd128_mask_round,
5778         __builtin_ia32_rangess128_mask_round): New builtins.
5779         * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
5780         (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
5781         ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
5782         "<round_saeonly_constraint>")): Changed to ...
5783         ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
5784         "<round_saeonly_scalar_constraint>")): ... this.
5785         ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
5786         %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
5787         ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
5788         %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
5789         %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
5790
5791 2018-02-02  Andrew Jenner  <andrew@codesourcery.com>
5792
5793         * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
5794         options.
5795         * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
5796         Remove all values except native, 8540 and 8548.
5797
5798 2018-02-02  H.J. Lu  <hongjiu.lu@intel.com>
5799
5800         * config/i386/i386.c (ix86_output_function_return): Pass
5801         INVALID_REGNUM, instead of -1, as invalid register number to
5802         indirect_thunk_name and output_indirect_thunk.
5803
5804 2018-02-02  Julia Koval  <julia.koval@intel.com>
5805
5806         * config.gcc: Add -march=icelake.
5807         * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
5808         * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
5809         * config/i386/i386.c (processor_costs): Add m_ICELAKE.
5810         (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
5811         PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
5812         (processor_target_table): Add icelake.
5813         (ix86_option_override_internal): Handle new PTAs.
5814         (get_builtin_code_for_version): Handle icelake.
5815         (M_INTEL_COREI7_ICELAKE): New.
5816         (fold_builtin_cpu): Handle icelake.
5817         * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
5818         * doc/invoke.texi: Add -march=icelake.
5819
5820 2018-02-02  Julia Koval  <julia.koval@intel.com>
5821
5822         * config/i386/i386.c (ix86_option_override_internal): Change flags type
5823         to wide_int_bitmask.
5824         * wide-int-bitmask.h: New.
5825
5826 2018-02-02  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
5827
5828         PR target/84066
5829         * config/i386/i386.md: Replace Pmode with word_mode in
5830         builtin_setjmp_setup and builtin_longjmp to support x32.
5831
5832 2018-02-01  Peter Bergner  <bergner@vnet.ibm.com>
5833
5834         PR target/56010
5835         PR target/83743
5836         * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
5837         #include "opts.h".
5838         (rs6000_supported_cpu_names): New static variable.
5839         (linux_cpu_translation_table): Likewise.
5840         (elf_platform) <cpu>: Define new static variable and use it.
5841         Translate kernel AT_PLATFORM name to canonical name if needed.
5842         Error if platform name is unknown.
5843
5844 2018-02-01  Aldy Hernandez  <aldyh@redhat.com>
5845
5846         PR target/84089
5847         * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
5848
5849 2018-02-01  Jeff Law  <law@redhat.com>
5850
5851         PR target/84128
5852         * config/i386/i386.c (release_scratch_register_on_entry): Add new
5853         OFFSET and RELEASE_VIA_POP arguments.  Use SP+OFFSET to restore
5854         the scratch if RELEASE_VIA_POP is false.
5855         (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
5856         If we have to save a temporary register, decrement SIZE appropriately.
5857         Pass new arguments to release_scratch_register_on_entry.
5858         (ix86_adjust_stack_and_probe): Likewise.
5859         (ix86_emit_probe_stack_range): Pass new arguments to
5860         release_scratch_register_on_entry.
5861
5862 2018-02-01  Uros Bizjak  <ubizjak@gmail.com>
5863
5864         PR rtl-optimization/84157
5865         * combine.c (change_zero_ext): Use REG_P predicate in
5866         front of HARD_REGISTER_P predicate.
5867
5868 2018-02-01  Georg-Johann Lay  <avr@gjlay.de>
5869
5870         * config/avr/avr.c (avr_option_override): Move disabling of
5871         -fdelete-null-pointer-checks to...
5872         * common/config/avr/avr-common.c (avr_option_optimization_table):
5873         ...here.
5874
5875 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
5876
5877         PR tree-optimization/81635
5878         * tree-data-ref.c (split_constant_offset_1): For types that
5879         wrap on overflow, try to use range info to prove that wrapping
5880         cannot occur.
5881
5882 2018-02-01  Renlin Li  <renlin.li@arm.com>
5883
5884         PR target/83370
5885         * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
5886         TAILCALL_ADDR_REGS.
5887         (aarch64_register_move_cost): Likewise.
5888         * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
5889         TAILCALL_ADDR_REGS.
5890         (REG_CLASS_NAMES): Likewise.
5891         (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
5892         TAILCALL_ADDR_REGS. Remove IP registers.
5893         * config/aarch64/aarch64.md (Ucs): Update register constraint.
5894
5895 2018-02-01  Richard Biener  <rguenther@suse.de>
5896
5897         * domwalk.h (dom_walker::dom_walker): Add additional constructor
5898         for specifying RPO order and allow NULL for that.
5899         * domwalk.c (dom_walker::dom_walker): Likewise.
5900         (dom_walker::walk): Handle NULL RPO order.
5901         * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
5902         in RPO order.
5903         (rewrite_update_dom_walker): Likewise.
5904         (mark_def_dom_walker): Likewise.
5905
5906 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
5907
5908         * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
5909         (aarch64_maybe_expand_sve_subreg_move): Declare.
5910         * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
5911         * config/aarch64/predicates.md (aarch64_any_register_operand): New
5912         predicate.
5913         * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
5914         that are semantically a reverse operation.
5915         (*aarch64_sve_mov<mode>_subreg_be): New pattern.
5916         * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
5917         (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
5918         functions.
5919         (aarch64_can_change_mode_class): For big-endian, forbid changes
5920         between two SVE modes if they have different element sizes.
5921
5922 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
5923
5924         * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
5925         the TImode handling for big-endian targets.
5926
5927 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
5928
5929         * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
5930         (*sve_ld1rq<Vesize>): ... this new pattern.  Handle all element sizes,
5931         not just bytes.
5932         * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
5933         Remove BSWAP handing for big-endian targets and use the form of
5934         LD1RQ appropariate for the mode.
5935
5936 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
5937
5938         * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
5939         all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
5940         duplicated element.
5941
5942 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
5943
5944         PR tearget/83845
5945         * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
5946         check for operands that need to go through aarch64_sve_reload_be.
5947
5948 2018-02-01  Jakub Jelinek  <jakub@redhat.com>
5949
5950         PR tree-optimization/81661
5951         PR tree-optimization/84117
5952         * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
5953         * tree-eh.c: Include gimplify.h.
5954         (find_trapping_overflow, replace_trapping_overflow,
5955         rewrite_to_non_trapping_overflow): New functions.
5956         * tree-vect-loop.c: Include tree-eh.h.
5957         (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
5958         * tree-data-ref.c: Include tree-eh.h.
5959         (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
5960
5961 2018-01-31  Uros Bizjak  <ubizjak@gmail.com>
5962
5963         PR rtl-optimization/84123
5964         * combine.c (change_zero_ext): Check if hard register satisfies
5965         can_change_dest_mode before calling gen_lowpart_SUBREG.
5966
5967 2018-01-31  Vladimir Makarov  <vmakarov@redhat.com>
5968
5969         PR target/82444
5970         * ira.c (ira_init_register_move_cost): Remove assert.
5971
5972 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
5973
5974         PR rtl-optimization/84071
5975         * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
5976         * doc/tm.texi: Regenerate.
5977
5978 2018-01-31  Richard Biener  <rguenther@suse.de>
5979
5980         PR tree-optimization/84132
5981         * tree-data-ref.c (analyze_miv_subscript): Properly
5982         check whether evolution_function_is_affine_multivariate_p
5983         before calling gcd_of_steps_may_divide_p.
5984
5985 2018-01-31  Julia Koval  <julia.koval@intel.com>
5986
5987         PR target/83618
5988         * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
5989         * config/i386/i386.md (rdpid_rex64) New.
5990         (rdpid): Make 32bit only.
5991
5992 2018-01-29  Aldy Hernandez  <aldyh@redhat.com>
5993
5994         PR lto/84105
5995         * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
5996         an IDENTIFIER_NODE for FUNCTION_TYPE's.
5997
5998 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
5999
6000         Revert
6001         2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
6002
6003         * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
6004
6005 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
6006
6007         PR rtl-optimization/84071
6008         * combine.c (record_dead_and_set_regs_1): Record the source unmodified
6009         for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
6010
6011 2018-01-31  Claudiu Zissulescu  <claziss@synopsys.com>
6012
6013         * config/arc/arc.c (arc_handle_aux_attribute): New function.
6014         (arc_attribute_table): Add 'aux' attribute.
6015         (arc_in_small_data_p): Consider aux like variables.
6016         (arc_is_aux_reg_p): New function.
6017         (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
6018         (arc_get_aux_arg): New function.
6019         (prepare_move_operands): Handle aux-register access.
6020         (arc_handle_aux_attribute): New function.
6021         * doc/extend.texi (ARC Variable attributes): Add subsection.
6022
6023 2018-01-31  Claudiu Zissulescu  <claziss@synopsys.com>
6024
6025         * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
6026         * config/arc/arc.c (arc_handle_uncached_attribute): New function.
6027         (arc_attribute_table): Add 'uncached' attribute.
6028         (arc_print_operand): Print '.di' flag for uncached memory
6029         accesses.
6030         (arc_in_small_data_p): Do not consider for small data the uncached
6031         types.
6032         (arc_is_uncached_mem_p): New function.
6033         * config/arc/predicates.md (compact_store_memory_operand): Check
6034         for uncached memory accesses.
6035         (nonvol_nonimm_operand): Likewise.
6036         * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
6037
6038 2018-01-31  Jakub Jelinek  <jakub@redhat.com>
6039
6040         PR c/84100
6041         * common.opt (falign-functions=, falign-jumps=, falign-labels=,
6042         falign-loops=): Add Optimization flag.
6043
6044 2018-01-30  Jeff Law  <law@redhat.com>
6045
6046         PR target/84064
6047         * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
6048         INT_REGISTERS_SAVED.  Check it prior to calling
6049         get_scratch_register_on_entry.
6050         (ix86_adjust_stack_and_probe): Similarly.
6051         (ix86_emit_probe_stack_range): Similarly.
6052         (ix86_expand_prologue): Corresponding changes.
6053
6054 2018-01-30  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
6055
6056         PR target/40411
6057         * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
6058         -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
6059
6060 2018-01-30  Vladimir Makarov  <vmakarov@redhat.com>
6061
6062         PR target/84112
6063         * lra-constraints.c (curr_insn_transform): Process AND in the
6064         address.
6065
6066 2018-01-30  Jakub Jelinek  <jakub@redhat.com>
6067
6068         PR rtl-optimization/83986
6069         * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
6070         dependence against last_pending_memory_flush in addition to
6071         pending_jump_insns.
6072
6073 2018-01-30  Alexandre Oliva  <aoliva@redhat.com>
6074
6075         PR tree-optimization/81611
6076         * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
6077         copies.
6078
6079 2018-01-30  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
6080
6081         PR target/83758
6082         * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
6083         a reg rtx.
6084
6085 2018-01-30  Richard Biener  <rguenther@suse.de>
6086             Jakub Jelinek  <jakub@redhat.com>
6087
6088         PR tree-optimization/84111
6089         * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
6090         inner loops added during recursion, as they don't have up-to-date
6091         SSA form.
6092
6093 2018-01-30  Jan Hubicka  <hubicka@ucw.cz>
6094
6095         PR ipa/81360
6096         * ipa-inline.c (can_inline_edge_p): Break out late tests to...
6097         (can_inline_edge_by_limits_p): ... here.
6098         (can_early_inline_edge_p, check_callers,
6099         update_caller_keys, update_callee_keys, recursive_inlining,
6100         add_new_edges_to_heap, speculation_useful_p,
6101         inline_small_functions,
6102         inline_small_functions, flatten_function,
6103         inline_to_all_callers_1): Update.
6104
6105 2018-01-30  Jan Hubicka  <hubicka@ucw.cz>
6106
6107         * profile-count.c (profile_count::combine_with_ipa_count): Handle
6108         zeros correctly.
6109
6110 2018-01-30  Richard Biener  <rguenther@suse.de>
6111
6112         PR tree-optimization/83008
6113         * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
6114         invariant and constant vector uses in stmts when they need
6115         more than one stmt.
6116
6117 2018-01-30  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
6118
6119         PR bootstrap/84017
6120         * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
6121         * configure: Regenerate.
6122
6123 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
6124
6125         * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
6126         pattern.
6127         (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
6128         Use gen_rtx_REG rather than gen_lowpart.
6129
6130 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
6131
6132         * lra-constraints.c (match_reload): Use subreg_lowpart_offset
6133         rather than 0 when creating partial subregs.
6134
6135 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
6136
6137         * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
6138         of usage.
6139
6140 2018-01-29  Michael Meissner  <meissner@linux.vnet.ibm.com>
6141
6142         PR target/81550
6143         * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
6144         and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
6145         -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
6146         flags.  This restores the settings used before the 2017-07-24.
6147         Turning off pre increment/decrement/modify allows IVOPTS to
6148         optimize DF/SF loops where the index is an int.
6149
6150 2018-01-29  Richard Biener  <rguenther@suse.de>
6151             Kelvin Nilsen  <kelvin@gcc.gnu.org>
6152
6153         PR bootstrap/80867
6154         * tree-vect-stmts.c (vectorizable_call): Don't call
6155         targetm.vectorize_builtin_md_vectorized_function if callee is
6156         NULL.
6157
6158 2018-01-22  Carl Love  <cel@us.ibm.com>
6159
6160         * doc/extend.tex: Fix typo in second arg in
6161         __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
6162
6163 2018-01-29  Richard Biener  <rguenther@suse.de>
6164
6165         PR tree-optimization/84086
6166         * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
6167         (flush_ssaname_freelist): When SSA names were released reset
6168         the SCEV hash table.
6169
6170 2018-01-29  Richard Biener  <rguenther@suse.de>
6171
6172         PR tree-optimization/84057
6173         * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
6174         removed paths when removing edges.
6175
6176 2018-01-27  H.J. Lu  <hongjiu.lu@intel.com>
6177
6178         * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
6179         -mfunction-return=@var{choice}.
6180
6181 2018-01-27  Bernd Edlinger  <bernd.edlinger@hotmail.de>
6182
6183         PR diagnostic/84034
6184         * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
6185         Handle CR like TAB.
6186         (layout::print_source_line): Likewise.
6187         (test_get_line_width_without_trailing_whitespace): Add test cases.
6188
6189 2018-01-27  Jakub Jelinek  <jakub@redhat.com>
6190
6191         PR middle-end/84040
6192         * sched-deps.c (sched_macro_fuse_insns): Return immediately for
6193         debug insns.
6194
6195 2018-01-26  Jim Wilson  <jimw@sifive.com>
6196
6197         * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
6198
6199         * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
6200         specified.
6201
6202 2018-01-26  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
6203
6204         * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
6205         and CMP + SUB-immediate -> SUBS.
6206
6207 2018-01-26  Martin Sebor  <msebor@redhat.com>
6208
6209         PR tree-optimization/83896
6210         * tree-ssa-strlen.c (get_string_len): Rename...
6211         (get_string_cst_length): ...to this.  Return HOST_WIDE_INT.
6212         Avoid assuming length is constant.
6213         (handle_char_store): Use HOST_WIDE_INT for string length.
6214
6215 2018-01-26  Uros Bizjak  <ubizjak@gmail.com>
6216
6217         PR target/81763
6218         * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
6219         to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
6220
6221 2018-01-26  Richard Biener  <rguenther@suse.de>
6222
6223         PR rtl-optimization/84003
6224         * dse.c (record_store): Only record redundant stores when
6225         the earlier store aliases at least all accesses the later one does.
6226
6227 2018-01-26  Jakub Jelinek  <jakub@redhat.com>
6228
6229         PR rtl-optimization/83985
6230         * dce.c (deletable_insn_p): Return false for separate shrink wrapping
6231         REG_CFA_RESTORE insns.
6232         (delete_unmarked_insns): Don't ignore separate shrink wrapping
6233         REG_CFA_RESTORE insns here.
6234
6235         PR c/83989
6236         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
6237         use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
6238
6239 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
6240
6241         * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
6242         * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
6243         (arc_init): Likewise.
6244         (arc_override_options): Likewise.
6245         (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
6246         value.
6247         (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
6248         support.
6249         * config/arc/arc.h (TARGET_DBNZ): Define.
6250         * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
6251         properly set the tune attribute.
6252         (dbnz): Use TARGET_DBNZ guard.
6253         * config/arc/arc.opt (mtune): Add core3 option.
6254
6255 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
6256
6257         * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
6258         recognize new pic like addresses.
6259         (arc_delegitimize_address): Clean up.
6260
6261 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
6262
6263         * config/arc/arc-arches.def: Option mrf16 valid for all
6264         architectures.
6265         * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
6266         * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
6267         * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
6268         * config/arc/arc-tables.opt: Regenerate.
6269         * config/arc/arc.c (arc_conditional_register_usage): Handle
6270         reduced register file case.
6271         (arc_file_start): Set must have build attributes.
6272         * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
6273         mrf16 option value.
6274         * config/arc/arc.opt (mrf16): Add new option.
6275         * config/arc/elf.h (ATTRIBUTE_PCS): Define.
6276         * config/arc/genmultilib.awk: Handle new mrf16 option.
6277         * config/arc/linux.h (ATTRIBUTE_PCS): Define.
6278         * config/arc/t-multilib: Regenerate.
6279         * doc/invoke.texi (ARC Options): Document mrf16 option.
6280
6281 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
6282
6283         * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
6284         * config/arc/arc.c (arc_handle_secure_attribute): New function.
6285         (arc_attribute_table): Add 'secure_call' attribute.
6286         (arc_print_operand): Print secure call operand.
6287         (arc_function_ok_for_sibcall): Don't optimize tail calls when
6288         secure.
6289         (arc_is_secure_call_p): New function.  * config/arc/arc.md
6290         (call_i): Add support for sjli instruction.
6291         (call_value_i): Likewise.
6292         * config/arc/constraints.md (Csc): New constraint.
6293
6294 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
6295             John Eric Martin  <John.Martin@emmicro-us.com>
6296
6297         * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
6298         * config/arc/arc.c (_arc_jli_section): New struct.
6299         (arc_jli_section): New type.
6300         (rc_jli_sections): New static variable.
6301         (arc_handle_jli_attribute): New function.
6302         (arc_attribute_table): Add jli_always and jli_fixed attribute.
6303         (arc_file_end): New function.
6304         (TARGET_ASM_FILE_END): Define.
6305         (arc_print_operand): Reuse 'S' letter for JLI output instruction.
6306         (arc_add_jli_section): New function.
6307         (jli_call_scan): Likewise.
6308         (arc_reorg): Call jli_call_scan.
6309         (arc_output_addsi): Remove 'S' from printing asm operand.
6310         (arc_is_jli_call_p): New function.
6311         * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
6312         operand.
6313         (movhi_insn): Likewise.
6314         (movsi_insn): Likewise.
6315         (movsi_set_cc_insn): Likewise.
6316         (loadqi_update): Likewise.
6317         (load_zeroextendqisi_update): Likewise.
6318         (load_signextendqisi_update): Likewise.
6319         (loadhi_update): Likewise.
6320         (load_zeroextendhisi_update): Likewise.
6321         (load_signextendhisi_update): Likewise.
6322         (loadsi_update): Likewise.
6323         (loadsf_update): Likewise.
6324         (movsicc_insn): Likewise.
6325         (bset_insn): Likewise.
6326         (bxor_insn): Likewise.
6327         (bclr_insn): Likewise.
6328         (bmsk_insn): Likewise.
6329         (bicsi3_insn): Likewise.
6330         (cmpsi_cc_c_insn): Likewise.
6331         (movsi_ne): Likewise.
6332         (movsi_cond_exec): Likewise.
6333         (clrsbsi2): Likewise.
6334         (norm_f): Likewise.
6335         (normw): Likewise.
6336         (swap): Likewise.
6337         (divaw): Likewise.
6338         (flag): Likewise.
6339         (sr): Likewise.
6340         (kflag): Likewise.
6341         (ffs): Likewise.
6342         (ffs_f): Likewise.
6343         (fls): Likewise.
6344         (call_i): Remove 'S' asm letter, add jli instruction.
6345         (call_value_i): Likewise.
6346         * config/arc/arc.op (mjli-always): New option.
6347         * config/arc/constraints.md (Cji): New constraint.
6348         * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
6349         operand.
6350         (subsf3_fpx): Likewise.
6351         (mulsf3_fpx): Likewise.
6352         * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
6353         asm operand.
6354         * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
6355         function attrbutes.
6356         * doc/invoke.texi (ARC): Document mjli-always option.
6357
6358 2018-01-26  Sebastian Perta  <sebastian.perta@renesas.com>
6359
6360         * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
6361         avoid addition with 0 and use incw and decw where possible.
6362
6363 2018-01-26  Richard Biener  <rguenther@suse.de>
6364
6365         PR tree-optimization/81082
6366         * fold-const.c (fold_plusminus_mult_expr): Do not perform the
6367         association if it requires casting to unsigned.
6368         * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
6369         from fold_plusminus_mult_expr to catch important cases late when
6370         range info is available.
6371
6372 2018-01-26  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
6373
6374         * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
6375         * configure.ac (hidden_linkonce): New test.
6376         * configure: Regenerate.
6377         * config.in: Regenerate.
6378
6379 2018-01-26  Julia Koval  <julia.koval@intel.com>
6380
6381         * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
6382         _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
6383         _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
6384         _mm_mask_bitshuffle_epi64_mask): Fix type.
6385         * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
6386         USI_FTYPE_V4DI_V4DI_USI): Remove.
6387         * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
6388         __builtin_ia32_vpshufbitqmb256_mask,
6389         __builtin_ia32_vpshufbitqmb128_mask): Fix types.
6390         * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
6391         * config/i386/sse.md (VI1_AVX512VLBW): Change types.
6392
6393 2018-01-26  Alan Modra  <amodra@gmail.com>
6394
6395         PR target/84033
6396         * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
6397         UNSPEC_VBPERMQ.  Sort other unspecs.
6398
6399 2018-01-25  David Edelsohn  <dje.gcc@gmail.com>
6400
6401         * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
6402
6403 2018-01-25  Jan Hubicka  <hubicka@ucw.cz>
6404
6405         PR middle-end/83055
6406         * predict.c (drop_profile): Do not push/pop cfun; update also
6407         node->count.
6408         (handle_missing_profiles): Fix logic looking for zero profiles.
6409
6410 2018-01-25  Jakub Jelinek  <jakub@redhat.com>
6411
6412         PR middle-end/83977
6413         * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
6414         on functions with #pragma omp declare simd or functions with simd
6415         attribute.
6416         * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
6417         * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
6418         Remove trailing \n from warning_at calls.
6419
6420 2018-01-25  Tom de Vries  <tom@codesourcery.com>
6421
6422         PR target/84028
6423         * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
6424         for neutered workers.
6425
6426 2018-01-24  Joseph Myers  <joseph@codesourcery.com>
6427
6428         PR target/68467
6429         * config/m68k/m68k.c (m68k_promote_function_mode): New function.
6430         (TARGET_PROMOTE_FUNCTION_MODE): New macro.
6431
6432 2018-01-24  Jeff Law  <law@redhat.com>
6433
6434         PR target/83994
6435         * i386.c (get_probe_interval): Move to earlier point.
6436         (ix86_compute_frame_layout): If -fstack-clash-protection and
6437         the frame is larger than the probe interval, then use pushes
6438         to save registers rather than reg->mem moves.
6439         (ix86_expand_prologue): Remove conditional for int_registers_saved
6440         assertion.
6441
6442 2018-01-24  Vladimir Makarov  <vmakarov@redhat.com>
6443
6444         PR target/84014
6445         * ira-build.c (setup_min_max_allocno_live_range_point): Set up
6446         min/max for never referenced object.
6447
6448 2018-01-24  Jakub Jelinek  <jakub@redhat.com>
6449
6450         PR middle-end/83977
6451         * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
6452         here.
6453         * omp-low.c (create_omp_child_function): Remove "omp declare simd"
6454         attributes from DECL_ATTRIBUTES (decl) without affecting
6455         DECL_ATTRIBUTES (current_function_decl).
6456         * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
6457         functions with non-NULL DECL_ABSTRACT_ORIGIN.
6458
6459 2018-01-24  Richard Sandiford  <richard.sandiford@linaro.org>
6460
6461         PR tree-optimization/83979
6462         * fold-const.c (fold_comparison): Use constant_boolean_node
6463         instead of boolean_{true,false}_node.
6464
6465 2018-01-24  Jan Hubicka  <hubicka@ucw.cz>
6466
6467         * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
6468         with zero counts.
6469
6470 2018-01-24  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
6471
6472         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
6473         Simplify the clause that sets the length attribute.
6474         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
6475         (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
6476         clause that sets the length attribute.
6477         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
6478
6479 2018-01-24  Tom de Vries  <tom@codesourcery.com>
6480
6481         PR target/83589
6482         * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
6483         (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
6484         Add strict parameter.
6485         (prevent_branch_around_nothing): Insert dummy insn between branch to
6486         label and label with no ptx insn inbetween.
6487         * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
6488
6489 2018-01-24  Tom de Vries  <tom@codesourcery.com>
6490
6491         PR target/81352
6492         * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
6493         for neutered threads in warp.
6494         * config/nvptx/nvptx.md (define_insn "exit"): New insn.
6495
6496 2018-01-24  Richard Biener  <rguenther@suse.de>
6497
6498         PR tree-optimization/83176
6499         * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
6500         operands.
6501
6502 2018-01-24  Richard Biener  <rguenther@suse.de>
6503
6504         PR tree-optimization/82819
6505         * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
6506         code generating pluses that are no-ops in the target precision.
6507
6508 2018-01-24  Richard Biener  <rguenther@suse.de>
6509
6510         PR middle-end/84000
6511         * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
6512
6513 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
6514
6515         * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
6516         to merge probabilities.
6517         * predict.c (probably_never_executed): Also mark as cold functions
6518         with global 0 profile and guessed local profile.
6519         * profile-count.c (profile_probability::combine_with_count): New
6520         member function.
6521         * profile-count.h (profile_probability::operator*,
6522         profile_probability::operator*=, profile_probability::operator/,
6523         profile_probability::operator/=): Reduce precision to adjusted
6524         and set value to guessed on contradictory divisions.
6525         (profile_probability::combine_with_freq): Remove.
6526         (profile_probability::combine_wiht_count): Declare.
6527         (profile_count::force_nonzero):: Set to adjusted.
6528         (profile_count::probability_in):: Set quality to adjusted.
6529         * tree-ssa-tail-merge.c (replace_block_by): Use
6530         combine_with_count.
6531
6532 2018-01-23  Andrew Waterman  <andrew@sifive.com>
6533             Jim Wilson  <jimw@sifive.com>
6534
6535         * config/riscv/riscv.c (riscv_stack_boundary): New.
6536         (riscv_option_override): Set riscv_stack_boundary.  Handle
6537         riscv_preferred_stack_boundary_arg.
6538         * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
6539         (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
6540         (STACK_BOUNDARY): Set to riscv_stack_boundary.
6541         (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
6542         * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
6543         * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
6544
6545 2018-01-23  H.J. Lu  <hongjiu.lu@intel.com>
6546
6547         PR target/83905
6548         * config/i386/i386.c (ix86_expand_prologue): Use cost reference
6549         of struct ix86_frame.
6550         (ix86_expand_epilogue): Likewise.  Add a local variable for
6551         the reg_save_offset field in struct ix86_frame.
6552
6553 2018-01-23  Bin Cheng  <bin.cheng@arm.com>
6554
6555         PR tree-optimization/82604
6556         * tree-loop-distribution.c (enum partition_kind): New enum item
6557         PKIND_PARTIAL_MEMSET.
6558         (partition_builtin_p): Support above new enum item.
6559         (generate_code_for_partition): Ditto.
6560         (compute_access_range): Differentiate cases that equality can be
6561         proven at all loops, the innermost loops or no loops.
6562         (classify_builtin_st, classify_builtin_ldst): Adjust call to above
6563         function.  Set PKIND_PARTIAL_MEMSET for partition appropriately.
6564         (finalize_partitions, distribute_loop): Don't fuse partition of
6565         PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
6566         (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
6567         parloop is enabled.
6568
6569 2018-01-23  Martin Liska  <mliska@suse.cz>
6570
6571         * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
6572         order to ignore the predictor.
6573         (PRED_POLYMORPHIC_CALL): Likewise.
6574         (PRED_RECURSIVE_CALL): Likewise.
6575
6576 2018-01-23  Martin Liska  <mliska@suse.cz>
6577
6578         * tree-profile.c (tree_profiling): Print function header to
6579         aware reader which function we are working on.
6580         * value-prof.c (gimple_find_values_to_profile): Do not print
6581         not interesting value histograms.
6582
6583 2018-01-23  Martin Liska  <mliska@suse.cz>
6584
6585         * profile-count.h (enum profile_quality): Add
6586         profile_uninitialized as the first value. Do not number values
6587         as they are zero based.
6588         (profile_count::verify): Update sanity check.
6589         (profile_probability::verify): Likewise.
6590
6591 2018-01-23  Nathan Sidwell  <nathan@acm.org>
6592
6593         * doc/invoke.texi (ffor-scope): Deprecate.
6594
6595 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
6596
6597         PR tree-optimization/83510
6598         * domwalk.c (set_all_edges_as_executable): New function.
6599         (dom_walker::dom_walker): Convert bool param
6600         "skip_unreachable_blocks" to enum reachability.  Move setup of
6601         edge flags to set_all_edges_as_executable and only do it when
6602         reachability is REACHABLE_BLOCKS.
6603         * domwalk.h (enum dom_walker::reachability): New enum.
6604         (dom_walker::dom_walker): Convert bool param
6605         "skip_unreachable_blocks" to enum reachability.
6606         (set_all_edges_as_executable): New decl.
6607         * graphite-scop-detection.c  (gather_bbs::gather_bbs): Convert
6608         from false for "skip_unreachable_blocks" to ALL_BLOCKS for
6609         "reachability".
6610         * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
6611         but converting true to REACHABLE_BLOCKS.
6612         * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
6613         * tree-vrp.c
6614         (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
6615         Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
6616         (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
6617         REACHABLE_BLOCKS.
6618         (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
6619         if check_all_array_refs will be called.
6620
6621 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
6622
6623         * tree.c (selftest::test_location_wrappers): Add more test
6624         coverage.
6625
6626 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
6627
6628         * sbitmap.c (selftest::test_set_range): Fix memory leaks.
6629         (selftest::test_bit_in_range): Likewise.
6630
6631 2018-01-23  Richard Sandiford  <richard.sandiford@linaro.org>
6632
6633         PR testsuite/83888
6634         * doc/sourcebuild.texi (vect_float): Say that the selector
6635         only describes the situation when -funsafe-math-optimizations is on.
6636         (vect_float_strict): Document.
6637
6638 2018-01-23  Richard Sandiford  <richard.sandiford@linaro.org>
6639
6640         PR tree-optimization/83965
6641         * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
6642         (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
6643         instead of checking only for a reduction.
6644         (vect_recog_widen_sum_pattern): Likewise.
6645
6646 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
6647
6648         * predict.c (probably_never_executed): Only use precise profile info.
6649         (compute_function_frequency): Skip after inlining hack since we now
6650         have quality checking.
6651
6652 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
6653
6654         * profile-count.h (profile_probability::very_unlikely,
6655         profile_probability::unlikely, profile_probability::even): Set
6656         precision to guessed.
6657
6658 2018-01-23  Richard Biener  <rguenther@suse.de>
6659
6660         PR tree-optimization/83963
6661         * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
6662         Properly terminate dominator walk when crossing the exit edge not
6663         when visiting its source block.
6664
6665 2018-01-23  Jakub Jelinek  <jakub@redhat.com>
6666
6667         PR c++/83918
6668         * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
6669         VIEW_CONVERT_EXPR to wrap CONST_DECLs.
6670
6671 2018-01-22  Jakub Jelinek  <jakub@redhat.com>
6672
6673         PR tree-optimization/83957
6674         * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs.  Remove
6675         semicolon after for body surrounded by braces.
6676
6677         PR tree-optimization/83081
6678         * profile-count.h (profile_probability::split): New method.
6679         * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
6680         Use profile_probability::split.
6681         (do_compare_rtx_and_jump): Fix adjustment of probabilities
6682         when splitting a single conditional jump into 2.
6683
6684 2018-01-22  David Malcolm  <dmalcolm@redhat.com>
6685
6686         PR tree-optimization/69452
6687         * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
6688         decl.
6689
6690 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6691
6692         * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
6693         * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
6694         * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
6695
6696 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6697
6698         * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
6699         * config/rl78/rl78.md (movdi): New define_expand.
6700         * config/rl78/rl78.c (rl78_split_movdi): New function.
6701
6702 2018-01-22  Michael Meissner  <meissner@linux.vnet.ibm.com>
6703
6704         PR target/83862
6705         * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
6706         no longer used.
6707         * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
6708         * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
6709         128-bit to produce an UNSPEC move to get the double word with the
6710         signbit and then a shift directly to do signbit.
6711         (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
6712         implementation with a new version that just does either a direct
6713         move or a regular move.  Move memory interface to separate insns.
6714         Move insns so they are next to the expander.
6715         (signbit<mode>2_dm_mem_be): New combiner insns to combine load
6716         with signbit move.  Split big and little endian case.
6717         (signbit<mode>2_dm_mem_le): Likewise.
6718         (signbit<mode>2_dm_<su>ext): Delete, no longer used.
6719         (signbit<mode>2_dm2): Likewise.
6720
6721 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6722
6723         * config/rl78/rl78.md (anddi3): New define_expand.
6724
6725 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6726
6727         * config/rl78/rl78.md (umindi3): New define_expand.
6728
6729 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6730
6731         * config/rl78/rl78.md (smindi3): New define_expand.
6732
6733 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6734
6735         * config/rl78/rl78.md (smaxdi3): New define_expand.
6736
6737 2018-01-22  Carl Love  <cel@us.ibm.com>
6738
6739         * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
6740         LVX_V1TI): Add macro expansion.
6741         * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
6742         definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
6743         VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
6744         * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
6745         Change check to determine if the instruction is a byte reversing
6746         entry.  Fix typo in comment.
6747         * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
6748         for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
6749         Add def_builtin calls for new builtins.
6750         * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
6751         Add define_insn expansion.
6752
6753 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6754
6755         * config/rl78/rl78.md (umaxdi3): New define_expand.
6756
6757 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6758
6759         * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
6760         for non-QImode registers.
6761
6762 2018-01-22  Richard Biener  <rguenther@suse.de>
6763
6764         PR tree-optimization/83963
6765         * graphite-scop-detection.c (scop_detection::get_sese): Delay
6766         including the loop exit block.
6767         (scop_detection::merge_sese): Likewise.
6768         (scop_detection::add_scop): Do it here instead.
6769
6770 2018-01-22  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
6771
6772         * doc/sourcebuild.texi (arm_softfloat): Document.
6773
6774 2018-01-21  John David Anglin  <danglin@gcc.gnu.org>
6775
6776         PR gcc/77734
6777         * config/pa/pa.c (pa_function_ok_for_sibcall): Use
6778         targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
6779         Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
6780
6781 2018-01-21  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
6782             David Edelsohn  <dje.gcc@gmail.com>
6783
6784         PR target/83946
6785         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
6786         Change "crset eq" to "crset 2".
6787         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
6788         (*call_indirect_aix<mode>_nospec): Likewise.
6789         (*call_value_indirect_aix<mode>_nospec): Likewise.
6790         (*call_indirect_elfv2<mode>_nospec): Likewise.
6791         (*call_value_indirect_elfv2<mode>_nospec): Likewise.
6792         (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
6793         change assembly output from . to $.
6794         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
6795         (indirect_jump<mode>_nospec): Change assembly output from . to $.
6796         (*tablejump<mode>_internal1_nospec): Likewise.
6797
6798 2018-01-21  Oleg Endo  <olegendo@gcc.gnu.org>
6799
6800         PR target/80870
6801         * config/sh/sh_optimize_sett_clrt.cc:
6802         Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
6803
6804 2018-01-20  Richard Sandiford  <richard.sandiford@linaro.org>
6805
6806         PR tree-optimization/83940
6807         * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
6808         offset_dt to vect_constant_def rather than vect_unknown_def_type.
6809         (vect_check_load_store_mask): Add a mask_dt_out parameter and
6810         use it to pass back the definition type.
6811         (vect_check_store_rhs): Likewise rhs_dt_out.
6812         (vect_build_gather_load_calls): Add a mask_dt argument and use
6813         it instead of a call to vect_is_simple_use.
6814         (vectorizable_store): Update calls to vect_check_load_store_mask
6815         and vect_check_store_rhs.  Use the dt returned by the latter instead
6816         of scatter_src_dt.  Use the cached mask_dt and gs_info.offset_dt
6817         instead of calls to vect_is_simple_use.  Pass the scalar rather
6818         than the vector operand to vect_is_simple_use when handling
6819         second and subsequent copies of an rhs value.
6820         (vectorizable_load): Update calls to vect_check_load_store_mask
6821         and vect_build_gather_load_calls.  Use the cached mask_dt and
6822         gs_info.offset_dt instead of calls to vect_is_simple_use.
6823
6824 2018-01-20  Jakub Jelinek  <jakub@redhat.com>
6825
6826         PR middle-end/83945
6827         * tree-emutls.c: Include gimplify.h.
6828         (lower_emutls_2): New function.
6829         (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
6830         with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
6831         it before further processing.
6832
6833         PR target/83930
6834         * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
6835         UINTVAL (trueop1) instead of INTVAL (op1).
6836
6837 2018-01-19  Jakub Jelinek  <jakub@redhat.com>
6838
6839         PR debug/81570
6840         PR debug/83728
6841         * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
6842         INCOMING_FRAME_SP_OFFSET if not defined.
6843         (scan_trace): Add ENTRY argument.  If true and
6844         DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
6845         emit a note to adjust the CFA offset.
6846         (create_cfi_notes): Adjust scan_trace callers.
6847         (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
6848         INCOMING_FRAME_SP_OFFSET in the CIE.
6849         * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
6850         * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
6851         Likewise.
6852         * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
6853         * doc/tm.texi: Regenerated.
6854
6855 2018-01-19  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
6856
6857         PR rtl-optimization/83147
6858         * lra-constraints.c (remove_inheritance_pseudos): Use
6859         lra_substitute_pseudo_within_insn.
6860
6861 2018-01-19  Tom de Vries  <tom@codesourcery.com>
6862             Cesar Philippidis  <cesar@codesourcery.com>
6863
6864         PR target/83920
6865         * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
6866
6867 2018-01-19  Cesar Philippidis  <cesar@codesourcery.com>
6868
6869         PR target/83790
6870         * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
6871         spaces for function labels.
6872
6873 2018-01-19  Martin Liska  <mliska@suse.cz>
6874
6875         * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
6876         (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
6877         (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
6878         (PRED_OPCODE_POSITIVE): Change from 64 to 59.
6879         (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
6880         (PRED_CONST_RETURN): Change from 69 to 65.
6881         (PRED_NULL_RETURN): Change from 91 to 71.
6882         (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
6883         (PRED_LOOP_GUARD): Change from 66 to 73.
6884
6885 2018-01-19  Martin Liska  <mliska@suse.cz>
6886
6887         * predict.c (predict_insn_def): Add new assert.
6888         (struct branch_predictor): Change type to signed integer.
6889         (test_prediction_value_range): Amend test to cover
6890         PROB_UNINITIALIZED.
6891         * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
6892         (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
6893         (PRED_LOOP_ITERATIONS_MAX): Likewise.
6894         (PRED_LOOP_IV_COMPARE): Likewise.
6895         * predict.h (PROB_UNINITIALIZED): Define new constant.
6896
6897 2018-01-19  Martin Liska  <mliska@suse.cz>
6898
6899         * predict.c (dump_prediction): Add new format for
6900         analyze_brprob.py script which is enabled with -details
6901         suboption.
6902         * profile-count.h (precise_p): New function.
6903
6904 2018-01-19  Richard Sandiford  <richard.sandiford@linaro.org>
6905
6906         PR tree-optimization/83922
6907         * tree-vect-loop.c (vect_verify_full_masking): Return false if
6908         there are no statements that need masking.
6909         (vect_active_double_reduction_p): New function.
6910         (vect_analyze_loop_operations): Use it when handling phis that
6911         are not in the loop header.
6912
6913 2018-01-19  Richard Sandiford  <richard.sandiford@linaro.org>
6914
6915         PR tree-optimization/83914
6916         * tree-vect-loop.c (vectorizable_induction): Don't convert
6917         init_expr or apply the peeling adjustment for inductions
6918         that are nested within the vectorized loop.
6919
6920 2018-01-19  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
6921
6922         * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
6923         instead of NEG.
6924
6925 2018-01-18  Jakub Jelinek  <jakub@redhat.com>
6926
6927         PR sanitizer/81715
6928         PR testsuite/83882
6929         * function.h (gimplify_parameters): Add gimple_seq * argument.
6930         * function.c: Include gimple.h and options.h.
6931         (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
6932         for the added local temporaries if needed.
6933         * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
6934         if there are any parameter cleanups, wrap whole body into a
6935         try/finally with the cleanups.
6936
6937 2018-01-18  Wilco Dijkstra  <wdijkstr@arm.com>
6938
6939         PR target/82964
6940         * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
6941         Use GET_MODE_CLASS for scalar floating point.
6942
6943 2018-01-18  Jan Hubicka  <hubicka@ucw.cz>
6944
6945         PR ipa/82256
6946         patch by PaX Team
6947         * cgraphclones.c (cgraph_node::create_version_clone_with_body):
6948         Fix call of call_cgraph_insertion_hooks.
6949
6950 2018-01-18  Martin Sebor  <msebor@redhat.com>
6951
6952         * doc/invoke.texi (-Wclass-memaccess): Tweak text.
6953
6954 2018-01-18  Jan Hubicka  <hubicka@ucw.cz>
6955
6956         PR ipa/83619
6957         * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
6958         frequencies.
6959
6960 2018-01-18  Boris Kolpackov  <boris@codesynthesis.com>
6961
6962         PR other/70268
6963         * common.opt: (-ffile-prefix-map): New option.
6964         * opts.c (common_handle_option): Defer it.
6965         * opts-global.c (handle_common_deferred_options): Handle it.
6966         * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
6967         * file-prefix-map.h: New file.
6968         (remap_debug_filename, add_debug_prefix_map): ...here.
6969         (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
6970         * final.c (debug_prefix_map, add_debug_prefix_map
6971         remap_debug_filename): Move to...
6972         * file-prefix-map.c: New file.
6973         (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
6974         generalize, get rid of alloca(), use strrchr() instead of strchr().
6975         (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
6976         Implement in terms of add_prefix_map().
6977         (remap_macro_filename, remap_debug_filename): Implement in term of
6978         remap_filename().
6979         * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
6980         * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
6981         * dbxout.c: Include file-prefix-map.h.
6982         * varasm.c: Likewise.
6983         * vmsdbgout.c: Likewise.
6984         * xcoffout.c: Likewise.
6985         * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
6986         * doc/cppopts.texi (-fmacro-prefix-map): Document.
6987         * doc/invoke.texi (-ffile-prefix-map): Document.
6988         (-fdebug-prefix-map): Update description.
6989
6990 2018-01-18  Martin Liska  <mliska@suse.cz>
6991
6992         * config/i386/i386.c (indirect_thunk_name): Document that also
6993         lfence is emitted.
6994         (output_indirect_thunk): Document why both instructions
6995         (pause and lfence) are generated.
6996
6997 2018-01-18  Richard Biener  <rguenther@suse.de>
6998
6999         PR tree-optimization/83887
7000         * graphite-scop-detection.c
7001         (scop_detection::get_nearest_dom_with_single_entry): Remove.
7002         (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
7003         (scop_detection::merge_sese): Re-implement with a flood-fill
7004         algorithm that properly finds a SESE region if it exists.
7005
7006 2018-01-18  Jakub Jelinek  <jakub@redhat.com>
7007
7008         PR c/61240
7009         * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
7010         pointer_diff optimizations use view_convert instead of convert.
7011
7012 2018-01-17  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
7013
7014         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
7015         Generate different code for -mno-speculate-indirect-jumps.
7016         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
7017         (*call_indirect_aix<mode>): Disable for
7018         -mno-speculate-indirect-jumps.
7019         (*call_indirect_aix<mode>_nospec): New define_insn.
7020         (*call_value_indirect_aix<mode>): Disable for
7021         -mno-speculate-indirect-jumps.
7022         (*call_value_indirect_aix<mode>_nospec): New define_insn.
7023         (*sibcall_nonlocal_sysv<mode>): Generate different code for
7024         -mno-speculate-indirect-jumps.
7025         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
7026
7027 2018-01-17  Michael Meissner  <meissner@linux.vnet.ibm.com>
7028
7029         * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
7030         long double type, set the flags for noting the default long double
7031         type, even if we don't pass or return a long double type.
7032
7033 2018-01-17  Jan Hubicka  <hubicka@ucw.cz>
7034
7035         PR ipa/83051
7036         * ipa-inline.c (flatten_function): Do not overwrite final inlining
7037         failure.
7038
7039 2018-01-17  Will Schmidt  <will_schmidt@vnet.ibm.com>
7040
7041         * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
7042         support for merge[hl].
7043         (fold_mergehl_helper): New helper function.
7044         (tree-vector-builder.h): New #include for tree_vector_builder usage.
7045         * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
7046         (altivec_vmrglw_direct): Add xxmrglw insn.
7047
7048 2018-01-17  Andrew Waterman  <andrew@sifive.com>
7049
7050         * config/riscv/riscv.c (riscv_conditional_register_usage): If
7051         UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
7052
7053 2018-01-17  David Malcolm  <dmalcolm@redhat.com>
7054
7055         PR lto/83121
7056         * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
7057         call the lto_location_cache before reading the
7058         DECL_SOURCE_LOCATION of the types.
7059
7060 2018-01-17  Wilco Dijkstra  <wdijkstr@arm.com>
7061             Richard Sandiford  <richard.sandiford@linaro.org>
7062
7063         * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
7064         * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
7065         (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
7066         SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
7067         * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
7068         Add declaration.
7069         * config/aarch64/constraints.md (aarch64_movti_operand):
7070         Limit immediates.
7071         * config/aarch64/predicates.md (Uti): Add new constraint.
7072
7073 2018-01-17 Carl Love  <cel@us.ibm.com>
7074
7075         * config/rs6000/vsx.md (define_expand xl_len_r,
7076         define_expand stxvl, define_expand *stxvl): Add match_dup argument.
7077         (define_insn): Add, match_dup 1 argument to define_insn stxvll and
7078         lxvll.
7079         (define_expand, define_insn): Move the shift left from  the
7080         define_insn to the define_expand for lxvl and stxvl instructions.
7081         * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
7082         and XL_LEN_R definitions to PURE.
7083
7084 2018-01-17  Uros Bizjak  <ubizjak@gmail.com>
7085
7086         * config/i386/i386.c (indirect_thunk_name): Declare regno
7087         as unsigned int.  Compare regno with INVALID_REGNUM.
7088         (output_indirect_thunk): Ditto.
7089         (output_indirect_thunk_function): Ditto.
7090         (ix86_code_end): Declare regno as unsigned int.  Use INVALID_REGNUM
7091         in the call to output_indirect_thunk_function.
7092
7093 2018-01-17  Richard Sandiford  <richard.sandiford@linaro.org>
7094
7095         PR middle-end/83884
7096         * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
7097         rather than the size of inner_type to determine the stack slot size
7098         when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
7099
7100 2018-01-16  Sebastian Peryt  <sebastian.peryt@intel.com>
7101
7102         PR target/83546
7103         * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
7104         to PTA_SILVERMONT.
7105
7106 2018-01-16  Michael Meissner  <meissner@linux.vnet.ibm.com>
7107
7108         * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
7109         endian Linux systems to optionally enable multilibs for selecting
7110         the long double type if the user configured an explicit type.
7111         * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
7112         have no long double multilibs if not defined.
7113         * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
7114         warn if the user used -mabi={ieee,ibm}longdouble and we built
7115         multilibs for long double.
7116         * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
7117         appropriate multilib option.
7118         (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
7119         multilib options.
7120         * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
7121         for building long double multilibs.
7122         * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
7123
7124 2018-01-16  John David Anglin  <danglin@gcc.gnu.org>
7125
7126         * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
7127         copies.
7128
7129         * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
7130         64 bits.
7131         * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
7132         128 bits.
7133
7134         * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
7135         variables.
7136
7137         * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
7138         return value.
7139
7140 2018-01-16  Eric Botcazou  <ebotcazou@adacore.com>
7141
7142         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
7143         ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
7144
7145 2018-01-16  Kelvin Nilsen  <kelvin@gcc.gnu.org>
7146
7147         * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
7148         different rtl trees depending on TARGET_64BIT.
7149         (rs6000_gen_lvx): Likewise.
7150
7151 2018-01-16  Eric Botcazou  <ebotcazou@adacore.com>
7152
7153         * config/visium/visium.md (nop): Tweak comment.
7154         (hazard_nop): Likewise.
7155
7156 2018-01-16  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
7157
7158         * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
7159         -mspeculate-indirect-jumps.
7160         * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
7161         for -mno-speculate-indirect-jumps.
7162         (*call_indirect_elfv2<mode>_nospec): New define_insn.
7163         (*call_value_indirect_elfv2<mode>): Disable for
7164         -mno-speculate-indirect-jumps.
7165         (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
7166         (indirect_jump): Emit different RTL for
7167         -mno-speculate-indirect-jumps.
7168         (*indirect_jump<mode>): Disable for
7169         -mno-speculate-indirect-jumps.
7170         (*indirect_jump<mode>_nospec): New define_insn.
7171         (tablejump): Emit different RTL for
7172         -mno-speculate-indirect-jumps.
7173         (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
7174         (tablejumpsi_nospec): New define_expand.
7175         (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
7176         (tablejumpdi_nospec): New define_expand.
7177         (*tablejump<mode>_internal1): Disable for
7178         -mno-speculate-indirect-jumps.
7179         (*tablejump<mode>_internal1_nospec): New define_insn.
7180         * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
7181         option.
7182
7183 2018-01-16  Artyom Skrobov tyomitch@gmail.com
7184
7185         * caller-save.c (insert_save): Drop unnecessary parameter.  All
7186         callers updated.
7187
7188 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
7189             Richard Biener  <rguenth@suse.de>
7190
7191         PR libgomp/83590
7192         * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
7193         return early, inline manually is_gimple_sizepos.  Make sure if we
7194         call gimplify_expr we don't end up with a gimple constant.
7195         * tree.c (variably_modified_type_p): Don't return true for
7196         is_gimple_constant (_t).  Inline manually is_gimple_sizepos.
7197         * gimplify.h (is_gimple_sizepos): Remove.
7198
7199 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
7200
7201         PR tree-optimization/83857
7202         * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
7203         vectorizable_live_operation for pure SLP statements.
7204         (vectorizable_live_operation): Handle PHIs.
7205
7206 2018-01-16  Richard Biener  <rguenther@suse.de>
7207
7208         PR tree-optimization/83867
7209         * tree-vect-stmts.c (vect_transform_stmt): Precompute
7210         nested_in_vect_loop_p since the scalar stmt may get invalidated.
7211
7212 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
7213
7214         PR c/83844
7215         * stor-layout.c (handle_warn_if_not_align): Use byte_position and
7216         multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
7217         If off is not INTEGER_CST, issue a may not be aligned warning
7218         rather than isn't aligned.  Use isn%'t rather than isn't.
7219         * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
7220         into MULT_EXPR.
7221         <case MULT_EXPR>: Improve the case when bottom and one of the
7222         MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
7223         operand, in that case check if the other operand is multiple of
7224         bottom divided by the INTEGER_CST operand.
7225
7226 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
7227
7228         PR target/83858
7229         * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
7230         * config/pa/pa-protos.h (pa_function_arg_size): Declare.
7231         * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
7232         pa_function_arg_size instead of FUNCTION_ARG_SIZE.
7233         * config/pa/pa.c (pa_function_arg_advance): Likewise.
7234         (pa_function_arg, pa_arg_partial_bytes): Likewise.
7235         (pa_function_arg_size): New function.
7236
7237 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
7238
7239         * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
7240         in a separate statement.
7241
7242 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
7243
7244         PR tree-optimization/83847
7245         * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
7246         group gathers and scatters.
7247
7248 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
7249
7250         PR rtl-optimization/86620
7251         * params.def (max-sched-ready-insns): Bump minimum value to 1.
7252
7253         PR rtl-optimization/83213
7254         * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
7255         to last if both are JUMP_INSNs.
7256
7257         PR tree-optimization/83843
7258         * gimple-ssa-store-merging.c
7259         (imm_store_chain_info::output_merged_store): Handle bit_not_p on
7260         store_immediate_info for bswap/nop orig_stores.
7261
7262 2018-01-15  Andrew Waterman  <andrew@sifive.com>
7263
7264         * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
7265         !TARGET_MUL.
7266         <UDIV>: Increase cost if !TARGET_DIV.
7267
7268 2018-01-15  Segher Boessenkool  <segher@kernel.crashing.org>
7269
7270         * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
7271         (define_attr "cr_logical_3op"): New.
7272         (cceq_ior_compare): Adjust.
7273         (cceq_ior_compare_complement): Adjust.
7274         (*cceq_rev_compare): Adjust.
7275         * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
7276         (is_cracked_insn): Adjust.
7277         (insn_must_be_first_in_group): Adjust.
7278         * config/rs6000/40x.md: Adjust.
7279         * config/rs6000/440.md: Adjust.
7280         * config/rs6000/476.md: Adjust.
7281         * config/rs6000/601.md: Adjust.
7282         * config/rs6000/603.md: Adjust.
7283         * config/rs6000/6xx.md: Adjust.
7284         * config/rs6000/7450.md: Adjust.
7285         * config/rs6000/7xx.md: Adjust.
7286         * config/rs6000/8540.md: Adjust.
7287         * config/rs6000/cell.md: Adjust.
7288         * config/rs6000/e300c2c3.md: Adjust.
7289         * config/rs6000/e500mc.md: Adjust.
7290         * config/rs6000/e500mc64.md: Adjust.
7291         * config/rs6000/e5500.md: Adjust.
7292         * config/rs6000/e6500.md: Adjust.
7293         * config/rs6000/mpc.md: Adjust.
7294         * config/rs6000/power4.md: Adjust.
7295         * config/rs6000/power5.md: Adjust.
7296         * config/rs6000/power6.md: Adjust.
7297         * config/rs6000/power7.md: Adjust.
7298         * config/rs6000/power8.md: Adjust.
7299         * config/rs6000/power9.md: Adjust.
7300         * config/rs6000/rs64.md: Adjust.
7301         * config/rs6000/titan.md: Adjust.
7302
7303 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
7304
7305         * config/i386/predicates.md (indirect_branch_operand): Rewrite
7306         ix86_indirect_branch_register logic.
7307
7308 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
7309
7310         * config/i386/constraints.md (Bs): Update
7311         ix86_indirect_branch_register check.  Don't check
7312         ix86_indirect_branch_register with GOT_memory_operand.
7313         (Bw): Likewise.
7314         * config/i386/predicates.md (GOT_memory_operand): Don't check
7315         ix86_indirect_branch_register here.
7316         (GOT32_symbol_operand): Likewise.
7317
7318 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
7319
7320         * config/i386/predicates.md (constant_call_address_operand):
7321         Rewrite ix86_indirect_branch_register logic.
7322         (sibcall_insn_operand): Likewise.
7323
7324 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
7325
7326         * config/i386/constraints.md (Bs): Replace
7327         ix86_indirect_branch_thunk_register with
7328         ix86_indirect_branch_register.
7329         (Bw): Likewise.
7330         * config/i386/i386.md (indirect_jump): Likewise.
7331         (tablejump): Likewise.
7332         (*sibcall_memory): Likewise.
7333         (*sibcall_value_memory): Likewise.
7334         Peepholes of indirect call and jump via memory: Likewise.
7335         * config/i386/i386.opt: Likewise.
7336         * config/i386/predicates.md (indirect_branch_operand): Likewise.
7337         (GOT_memory_operand): Likewise.
7338         (call_insn_operand): Likewise.
7339         (sibcall_insn_operand): Likewise.
7340         (GOT32_symbol_operand): Likewise.
7341
7342 2018-01-15  Jakub Jelinek  <jakub@redhat.com>
7343
7344         PR middle-end/83837
7345         * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
7346         type rather than type addr's type points to.
7347         (expand_omp_atomic_mutex): Likewise.
7348         (expand_omp_atomic): Likewise.
7349
7350 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
7351
7352         PR target/83839
7353         * config/i386/i386.c (output_indirect_thunk_function): Use
7354         ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
7355         for  __x86_return_thunk.
7356
7357 2018-01-15  Richard Biener  <rguenther@suse.de>
7358
7359         PR middle-end/83850
7360         * expmed.c (extract_bit_field_1): Fix typo.
7361
7362 2018-01-15  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
7363
7364         PR target/83687
7365         * config/arm/iterators.md (VF): New mode iterator.
7366         * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
7367         Remove integer-related logic from pattern.
7368         (neon_vabd<mode>_3): Likewise.
7369
7370 2018-01-15  Jakub Jelinek  <jakub@redhat.com>
7371
7372         PR middle-end/82694
7373         * common.opt (fstrict-overflow): No longer an alias.
7374         (fwrapv-pointer): New option.
7375         * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
7376         also for pointer types based on flag_wrapv_pointer.
7377         * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
7378         opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
7379         opts->x_flag_wrapv got set.
7380         * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
7381         changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
7382         POINTER_TYPE_OVERFLOW_UNDEFINED.
7383         * match.pd: Likewise in address comparison pattern.
7384         * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
7385
7386 2018-01-15  Richard Biener  <rguenther@suse.de>
7387
7388         PR lto/83804
7389         * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
7390         from TYPE_FIELDS.  Free TYPE_BINFO if not used by devirtualization.
7391         Reset type names to their identifier if their TYPE_DECL doesn't
7392         have linkage (and thus is used for ODR and devirt).
7393         (save_debug_info_for_decl): Remove.
7394         (save_debug_info_for_type): Likewise.
7395         (add_tree_to_fld_list): Adjust.
7396         * tree-pretty-print.c (dump_generic_node): Make dumping of
7397         type names more robust.
7398
7399 2018-01-15  Richard Biener  <rguenther@suse.de>
7400
7401         * BASE-VER: Bump to 8.0.1.
7402
7403 2018-01-14  Martin Sebor  <msebor@redhat.com>
7404
7405         PR other/83508
7406         * builtins.c (check_access): Avoid warning when the no-warning bit
7407         is set.
7408
7409 2018-01-14  Cory Fields  <cory-nospam-@coryfields.com>
7410
7411         * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
7412         * ira-color (allocno_hard_regs_compare): Likewise.
7413
7414 2018-01-14  Nathan Rossi  <nathan@nathanrossi.com>
7415
7416         PR target/83013
7417         * config/microblaze/microblaze.c (microblaze_asm_output_ident):
7418         Use .pushsection/.popsection.
7419
7420 2018-01-14  Martin Sebor  <msebor@redhat.com>
7421
7422         PR c++/81327
7423         * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
7424
7425 2018-01-14  Jakub Jelinek  <jakub@redhat.com>
7426
7427         * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
7428         entry from extra_headers.
7429         (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
7430         extra_headers, make the list bitwise identical to the i?86-*-* one.
7431
7432 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
7433
7434         * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
7435         -mcmodel=large with -mindirect-branch=thunk,
7436         -mindirect-branch=thunk-extern, -mfunction-return=thunk and
7437         -mfunction-return=thunk-extern.
7438         * doc/invoke.texi: Document -mcmodel=large is incompatible with
7439         -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
7440         -mfunction-return=thunk and -mfunction-return=thunk-extern.
7441
7442 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
7443
7444         * config/i386/i386.c (print_reg): Print the name of the full
7445         integer register without '%'.
7446         (ix86_print_operand): Handle 'V'.
7447          * doc/extend.texi: Document 'V' modifier.
7448
7449 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
7450
7451         * config/i386/constraints.md (Bs): Disallow memory operand for
7452         -mindirect-branch-register.
7453         (Bw): Likewise.
7454         * config/i386/predicates.md (indirect_branch_operand): Likewise.
7455         (GOT_memory_operand): Likewise.
7456         (call_insn_operand): Likewise.
7457         (sibcall_insn_operand): Likewise.
7458         (GOT32_symbol_operand): Likewise.
7459         * config/i386/i386.md (indirect_jump): Call convert_memory_address
7460         for -mindirect-branch-register.
7461         (tablejump): Likewise.
7462         (*sibcall_memory): Likewise.
7463         (*sibcall_value_memory): Likewise.
7464         Disallow peepholes of indirect call and jump via memory for
7465         -mindirect-branch-register.
7466         (*call_pop): Replace m with Bw.
7467         (*call_value_pop): Likewise.
7468         (*sibcall_pop_memory): Replace m with Bs.
7469         * config/i386/i386.opt (mindirect-branch-register): New option.
7470         * doc/invoke.texi: Document -mindirect-branch-register option.
7471
7472 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
7473
7474         * config/i386/i386-protos.h (ix86_output_function_return): New.
7475         * config/i386/i386.c (ix86_set_indirect_branch_type): Also
7476         set function_return_type.
7477         (indirect_thunk_name): Add ret_p to indicate thunk for function
7478         return.
7479         (output_indirect_thunk_function): Pass false to
7480         indirect_thunk_name.
7481         (ix86_output_indirect_branch_via_reg): Likewise.
7482         (ix86_output_indirect_branch_via_push): Likewise.
7483         (output_indirect_thunk_function): Create alias for function
7484         return thunk if regno < 0.
7485         (ix86_output_function_return): New function.
7486         (ix86_handle_fndecl_attribute): Handle function_return.
7487         (ix86_attribute_table): Add function_return.
7488         * config/i386/i386.h (machine_function): Add
7489         function_return_type.
7490         * config/i386/i386.md (simple_return_internal): Use
7491         ix86_output_function_return.
7492         (simple_return_internal_long): Likewise.
7493         * config/i386/i386.opt (mfunction-return=): New option.
7494         (indirect_branch): Mention -mfunction-return=.
7495         * doc/extend.texi: Document function_return function attribute.
7496         * doc/invoke.texi: Document -mfunction-return= option.
7497
7498 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
7499
7500         * config/i386/i386-opts.h (indirect_branch): New.
7501         * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
7502         * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
7503         with local indirect jump when converting indirect call and jump.
7504         (ix86_set_indirect_branch_type): New.
7505         (ix86_set_current_function): Call ix86_set_indirect_branch_type.
7506         (indirectlabelno): New.
7507         (indirect_thunk_needed): Likewise.
7508         (indirect_thunk_bnd_needed): Likewise.
7509         (indirect_thunks_used): Likewise.
7510         (indirect_thunks_bnd_used): Likewise.
7511         (INDIRECT_LABEL): Likewise.
7512         (indirect_thunk_name): Likewise.
7513         (output_indirect_thunk): Likewise.
7514         (output_indirect_thunk_function): Likewise.
7515         (ix86_output_indirect_branch_via_reg): Likewise.
7516         (ix86_output_indirect_branch_via_push): Likewise.
7517         (ix86_output_indirect_branch): Likewise.
7518         (ix86_output_indirect_jmp): Likewise.
7519         (ix86_code_end): Call output_indirect_thunk_function if needed.
7520         (ix86_output_call_insn): Call ix86_output_indirect_branch if
7521         needed.
7522         (ix86_handle_fndecl_attribute): Handle indirect_branch.
7523         (ix86_attribute_table): Add indirect_branch.
7524         * config/i386/i386.h (machine_function): Add indirect_branch_type
7525         and has_local_indirect_jump.
7526         * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
7527         to true.
7528         (tablejump): Likewise.
7529         (*indirect_jump): Use ix86_output_indirect_jmp.
7530         (*tablejump_1): Likewise.
7531         (simple_return_indirect_internal): Likewise.
7532         * config/i386/i386.opt (mindirect-branch=): New option.
7533         (indirect_branch): New.
7534         (keep): Likewise.
7535         (thunk): Likewise.
7536         (thunk-inline): Likewise.
7537         (thunk-extern): Likewise.
7538         * doc/extend.texi: Document indirect_branch function attribute.
7539         * doc/invoke.texi: Document -mindirect-branch= option.
7540
7541 2018-01-14  Jan Hubicka  <hubicka@ucw.cz>
7542
7543         PR ipa/83051
7544         * ipa-inline.c (edge_badness): Tolerate roundoff errors.
7545
7546 2018-01-14  Richard Sandiford  <richard.sandiford@linaro.org>
7547
7548         * ipa-inline.c (want_inline_small_function_p): Return false if
7549         inlining has already failed with CIF_FINAL_ERROR.
7550         (update_caller_keys): Call want_inline_small_function_p before
7551         can_inline_edge_p.
7552         (update_callee_keys): Likewise.
7553
7554 2018-01-10  Kelvin Nilsen  <kelvin@gcc.gnu.org>
7555
7556         * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
7557         New function.
7558         (rs6000_quadword_masked_address_p): Likewise.
7559         (quad_aligned_load_p): Likewise.
7560         (quad_aligned_store_p): Likewise.
7561         (const_load_sequence_p): Add comment to describe the outer-most loop.
7562         (mimic_memory_attributes_and_flags): New function.
7563         (rs6000_gen_stvx): Likewise.
7564         (replace_swapped_aligned_store): Likewise.
7565         (rs6000_gen_lvx): Likewise.
7566         (replace_swapped_aligned_load): Likewise.
7567         (replace_swapped_load_constant): Capitalize argument name in
7568         comment describing this function.
7569         (rs6000_analyze_swaps): Add a third pass to search for vector loads
7570         and stores that access quad-word aligned addresses and replace
7571         with stvx or lvx instructions when appropriate.
7572         * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
7573         New function prototype.
7574         (rs6000_quadword_masked_address_p): Likewise.
7575         (rs6000_gen_lvx): Likewise.
7576         (rs6000_gen_stvx): Likewise.
7577         * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
7578         VSX_D (V2DF, V2DI), modify this split to select lvx instruction
7579         when memory address is aligned.
7580         (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
7581         this split to select lvx instruction when memory address is aligned.
7582         (*vsx_le_perm_load_v8hi): Modify this split to select lvx
7583         instruction when memory address is aligned.
7584         (*vsx_le_perm_load_v16qi): Likewise.
7585         (four unnamed splitters): Modify to select the stvx instruction
7586         when memory is aligned.
7587
7588 2018-01-13  Jan Hubicka  <hubicka@ucw.cz>
7589
7590         * predict.c (determine_unlikely_bbs): Handle correctly BBs
7591         which appears in the queue multiple times.
7592
7593 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7594             Alan Hayward  <alan.hayward@arm.com>
7595             David Sherwood  <david.sherwood@arm.com>
7596
7597         * tree-vectorizer.h (vec_lower_bound): New structure.
7598         (_loop_vec_info): Add check_nonzero and lower_bounds.
7599         (LOOP_VINFO_CHECK_NONZERO): New macro.
7600         (LOOP_VINFO_LOWER_BOUNDS): Likewise.
7601         (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
7602         * tree-data-ref.h (dr_with_seg_len): Add access_size and align
7603         fields.  Make seg_len the distance travelled, not including the
7604         access size.
7605         (dr_direction_indicator): Declare.
7606         (dr_zero_step_indicator): Likewise.
7607         (dr_known_forward_stride_p): Likewise.
7608         * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
7609         tree-ssanames.h.
7610         (runtime_alias_check_p): Allow runtime alias checks with
7611         variable strides.
7612         (operator ==): Compare access_size and align.
7613         (prune_runtime_alias_test_list): Rework for new distinction between
7614         the access_size and seg_len.
7615         (create_intersect_range_checks_index): Likewise.  Cope with polynomial
7616         segment lengths.
7617         (get_segment_min_max): New function.
7618         (create_intersect_range_checks): Use it.
7619         (dr_step_indicator): New function.
7620         (dr_direction_indicator): Likewise.
7621         (dr_zero_step_indicator): Likewise.
7622         (dr_known_forward_stride_p): Likewise.
7623         * tree-loop-distribution.c (data_ref_segment_size): Return
7624         DR_STEP * (niters - 1).
7625         (compute_alias_check_pairs): Update call to the dr_with_seg_len
7626         constructor.
7627         * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
7628         (vect_preserves_scalar_order_p): New function, split out from...
7629         (vect_analyze_data_ref_dependence): ...here.  Check for zero steps.
7630         (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
7631         (vect_vfa_access_size): New function.
7632         (vect_vfa_align): Likewise.
7633         (vect_compile_time_alias): Take access_size_a and access_b arguments.
7634         (dump_lower_bound): New function.
7635         (vect_check_lower_bound): Likewise.
7636         (vect_small_gap_p): Likewise.
7637         (vectorizable_with_step_bound_p): Likewise.
7638         (vect_prune_runtime_alias_test_list): Ignore cross-iteration
7639         depencies if the vectorization factor is 1.  Convert the checks
7640         for nonzero steps into checks on the bounds of DR_STEP.  Try using
7641         a bunds check for variable steps if the minimum required step is
7642         relatively small. Update calls to the dr_with_seg_len
7643         constructor and to vect_compile_time_alias.
7644         * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
7645         function.
7646         (vect_loop_versioning): Call it.
7647         * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
7648         when retrying.
7649         (vect_estimate_min_profitable_iters): Account for any bounds checks.
7650
7651 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7652             Alan Hayward  <alan.hayward@arm.com>
7653             David Sherwood  <david.sherwood@arm.com>
7654
7655         * doc/sourcebuild.texi (vect_scatter_store): Document.
7656         * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
7657         optabs.
7658         * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
7659         Document.
7660         * genopinit.c (main): Add supports_vec_scatter_store and
7661         supports_vec_scatter_store_cached to target_optabs.
7662         * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
7663         IFN_MASK_SCATTER_STORE.
7664         * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
7665         functions.
7666         * internal-fn.h (internal_store_fn_p): Declare.
7667         (internal_fn_stored_value_index): Likewise.
7668         * internal-fn.c (scatter_store_direct): New macro.
7669         (expand_scatter_store_optab_fn): New function.
7670         (direct_scatter_store_optab_supported_p): New macro.
7671         (internal_store_fn_p): New function.
7672         (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
7673         IFN_MASK_SCATTER_STORE.
7674         (internal_fn_mask_index): Likewise.
7675         (internal_fn_stored_value_index): New function.
7676         (internal_gather_scatter_fn_supported_p): Adjust operand numbers
7677         for scatter stores.
7678         * optabs-query.h (supports_vec_scatter_store_p): Declare.
7679         * optabs-query.c (supports_vec_scatter_store_p): New function.
7680         * tree-vectorizer.h (vect_get_store_rhs): Declare.
7681         * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
7682         true for scatter stores.
7683         (vect_gather_scatter_fn_p): Handle scatter stores too.
7684         (vect_check_gather_scatter): Consider using scatter stores if
7685         supports_vec_scatter_store_p.
7686         * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
7687         scatter stores too.
7688         * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
7689         internal_fn_stored_value_index.
7690         (check_load_store_masking): Handle scatter stores too.
7691         (vect_get_store_rhs): Make public.
7692         (vectorizable_call): Use internal_store_fn_p.
7693         (vectorizable_store): Handle scatter store internal functions.
7694         (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
7695         when deciding whether the end of the group has been reached.
7696         * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
7697         * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
7698         (mask_scatter_store<mode>): New insns.
7699
7700 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7701             Alan Hayward  <alan.hayward@arm.com>
7702             David Sherwood  <david.sherwood@arm.com>
7703
7704         * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
7705         * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
7706         * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
7707         function.
7708         (vect_use_strided_gather_scatters_p): Take a masked_p argument.
7709         Use vect_truncate_gather_scatter_offset if we can't treat the
7710         operation as a normal gather load or scatter store.
7711         (get_group_load_store_type): Take the gather_scatter_info
7712         as argument.  Try using a gather load or scatter store for
7713         single-element groups.
7714         (get_load_store_type): Update calls to get_group_load_store_type
7715         and vect_use_strided_gather_scatters_p.
7716
7717 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7718             Alan Hayward  <alan.hayward@arm.com>
7719             David Sherwood  <david.sherwood@arm.com>
7720
7721         * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
7722         optional tree argument.
7723         * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
7724         null target hooks.
7725         (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
7726         but continue to use the current value as a fallback.
7727         (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
7728         to compare the updates.
7729         * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
7730         (get_load_store_type): Use it when handling a strided access.
7731         (vect_get_strided_load_store_ops): New function.
7732         (vect_get_data_ptr_increment): Likewise.
7733         (vectorizable_load): Handle strided gather loads.  Always pass
7734         a step to vect_create_data_ref_ptr and bump_vector_ptr.
7735
7736 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7737             Alan Hayward  <alan.hayward@arm.com>
7738             David Sherwood  <david.sherwood@arm.com>
7739
7740         * doc/md.texi (gather_load@var{m}): Document.
7741         (mask_gather_load@var{m}): Likewise.
7742         * genopinit.c (main): Add supports_vec_gather_load and
7743         supports_vec_gather_load_cached to target_optabs.
7744         * optabs-tree.c (init_tree_optimization_optabs): Use
7745         ggc_cleared_alloc to allocate target_optabs.
7746         * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
7747         * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
7748         functions.
7749         * internal-fn.h (internal_load_fn_p): Declare.
7750         (internal_gather_scatter_fn_p): Likewise.
7751         (internal_fn_mask_index): Likewise.
7752         (internal_gather_scatter_fn_supported_p): Likewise.
7753         * internal-fn.c (gather_load_direct): New macro.
7754         (expand_gather_load_optab_fn): New function.
7755         (direct_gather_load_optab_supported_p): New macro.
7756         (direct_internal_fn_optab): New function.
7757         (internal_load_fn_p): Likewise.
7758         (internal_gather_scatter_fn_p): Likewise.
7759         (internal_fn_mask_index): Likewise.
7760         (internal_gather_scatter_fn_supported_p): Likewise.
7761         * optabs-query.c (supports_at_least_one_mode_p): New function.
7762         (supports_vec_gather_load_p): Likewise.
7763         * optabs-query.h (supports_vec_gather_load_p): Declare.
7764         * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
7765         and memory_type field.
7766         (NUM_PATTERNS): Bump to 15.
7767         * tree-vect-data-refs.c: Include internal-fn.h.
7768         (vect_gather_scatter_fn_p): New function.
7769         (vect_describe_gather_scatter_call): Likewise.
7770         (vect_check_gather_scatter): Try using internal functions for
7771         gather loads.  Recognize existing calls to a gather load function.
7772         (vect_analyze_data_refs): Consider using gather loads if
7773         supports_vec_gather_load_p.
7774         * tree-vect-patterns.c (vect_get_load_store_mask): New function.
7775         (vect_get_gather_scatter_offset_type): Likewise.
7776         (vect_convert_mask_for_vectype): Likewise.
7777         (vect_add_conversion_to_patterm): Likewise.
7778         (vect_try_gather_scatter_pattern): Likewise.
7779         (vect_recog_gather_scatter_pattern): New pattern recognizer.
7780         (vect_vect_recog_func_ptrs): Add it.
7781         * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
7782         internal_fn_mask_index and internal_gather_scatter_fn_p.
7783         (check_load_store_masking): Take the gather_scatter_info as an
7784         argument and handle gather loads.
7785         (vect_get_gather_scatter_ops): New function.
7786         (vectorizable_call): Check internal_load_fn_p.
7787         (vectorizable_load): Likewise.  Handle gather load internal
7788         functions.
7789         (vectorizable_store): Update call to check_load_store_masking.
7790         * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
7791         * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
7792         * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
7793         (aarch64_gather_scale_operand_d): New predicates.
7794         * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
7795         (mask_gather_load<mode>): New insns.
7796
7797 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7798             Alan Hayward  <alan.hayward@arm.com>
7799             David Sherwood  <david.sherwood@arm.com>
7800
7801         * optabs.def (fold_left_plus_optab): New optab.
7802         * doc/md.texi (fold_left_plus_@var{m}): Document.
7803         * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
7804         * internal-fn.c (fold_left_direct): Define.
7805         (expand_fold_left_optab_fn): Likewise.
7806         (direct_fold_left_optab_supported_p): Likewise.
7807         * fold-const-call.c (fold_const_fold_left): New function.
7808         (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
7809         * tree-parloops.c (valid_reduction_p): New function.
7810         (gather_scalar_reductions): Use it.
7811         * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
7812         (vect_finish_replace_stmt): Declare.
7813         * tree-vect-loop.c (fold_left_reduction_fn): New function.
7814         (needs_fold_left_reduction_p): New function, split out from...
7815         (vect_is_simple_reduction): ...here.  Accept reductions that
7816         forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
7817         (vect_force_simple_reduction): Also store the reduction type in
7818         the assignment's STMT_VINFO_REDUC_TYPE.
7819         (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
7820         (merge_with_identity): New function.
7821         (vect_expand_fold_left): Likewise.
7822         (vectorize_fold_left_reduction): Likewise.
7823         (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION.  Leave the
7824         scalar phi in place for it.  Check for target support and reject
7825         cases that would reassociate the operation.  Defer the transform
7826         phase to vectorize_fold_left_reduction.
7827         * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
7828         * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
7829         (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
7830
7831 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7832
7833         * tree-if-conv.c (predicate_mem_writes): Remove redundant
7834         call to ifc_temp_var.
7835
7836 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7837             Alan Hayward  <alan.hayward@arm.com>
7838             David Sherwood  <david.sherwood@arm.com>
7839
7840         * target.def (legitimize_address_displacement): Take the original
7841         offset as a poly_int.
7842         * targhooks.h (default_legitimize_address_displacement): Update
7843         accordingly.
7844         * targhooks.c (default_legitimize_address_displacement): Likewise.
7845         * doc/tm.texi: Regenerate.
7846         * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
7847         as an argument, moving assert of ad->disp == ad->disp_term to...
7848         (process_address_1): ...here.  Update calls to base_plus_disp_to_reg.
7849         Try calling targetm.legitimize_address_displacement before expanding
7850         the address rather than afterwards, and adjust for the new interface.
7851         * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
7852         Match the new hook interface.  Handle SVE addresses.
7853         * config/sh/sh.c (sh_legitimize_address_displacement): Make the
7854         new hook interface.
7855
7856 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7857
7858         * Makefile.in (OBJS): Add early-remat.o.
7859         * target.def (select_early_remat_modes): New hook.
7860         * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
7861         * doc/tm.texi: Regenerate.
7862         * targhooks.h (default_select_early_remat_modes): Declare.
7863         * targhooks.c (default_select_early_remat_modes): New function.
7864         * timevar.def (TV_EARLY_REMAT): New timevar.
7865         * passes.def (pass_early_remat): New pass.
7866         * tree-pass.h (make_pass_early_remat): Declare.
7867         * early-remat.c: New file.
7868         * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
7869         function.
7870         (TARGET_SELECT_EARLY_REMAT_MODES): Define.
7871
7872 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7873             Alan Hayward  <alan.hayward@arm.com>
7874             David Sherwood  <david.sherwood@arm.com>
7875
7876         * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
7877         vfm1 with a bound_epilog parameter.
7878         (vect_do_peeling): Update calls accordingly, and move the prologue
7879         call earlier in the function.  Treat the base bound_epilog as 0 for
7880         fully-masked loops and retain vf - 1 for other loops.  Add 1 to
7881         this base when peeling for gaps.
7882         * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
7883         with fully-masked loops.
7884         (vect_estimate_min_profitable_iters): Handle the single peeled
7885         iteration in that case.
7886
7887 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7888             Alan Hayward  <alan.hayward@arm.com>
7889             David Sherwood  <david.sherwood@arm.com>
7890
7891         * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
7892         single-element interleaving even if the size is not a power of 2.
7893         * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
7894         accesses for single-element interleaving if the group size is
7895         not a power of 2.
7896
7897 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7898             Alan Hayward  <alan.hayward@arm.com>
7899             David Sherwood  <david.sherwood@arm.com>
7900
7901         * doc/md.texi (fold_extract_last_@var{m}): Document.
7902         * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
7903         * optabs.def (fold_extract_last_optab): New optab.
7904         * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
7905         * internal-fn.c (fold_extract_direct): New macro.
7906         (expand_fold_extract_optab_fn): Likewise.
7907         (direct_fold_extract_optab_supported_p): Likewise.
7908         * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
7909         * tree-vect-loop.c (vect_model_reduction_cost): Handle
7910         EXTRACT_LAST_REDUCTION.
7911         (get_initial_def_for_reduction): Do not create an initial vector
7912         for EXTRACT_LAST_REDUCTION reductions.
7913         (vectorizable_reduction): Leave the scalar phi in place for
7914         EXTRACT_LAST_REDUCTIONs.  Try using EXTRACT_LAST_REDUCTION
7915         ahead of INTEGER_INDUC_COND_REDUCTION.  Do not check for an
7916         epilogue code for EXTRACT_LAST_REDUCTION and defer the
7917         transform phase to vectorizable_condition.
7918         * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
7919         split out from...
7920         (vect_finish_stmt_generation): ...here.
7921         (vect_finish_replace_stmt): New function.
7922         (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
7923         * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
7924         pattern.
7925         * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
7926
7927 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7928             Alan Hayward  <alan.hayward@arm.com>
7929             David Sherwood  <david.sherwood@arm.com>
7930
7931         * doc/md.texi (extract_last_@var{m}): Document.
7932         * optabs.def (extract_last_optab): New optab.
7933         * internal-fn.def (EXTRACT_LAST): New internal function.
7934         * internal-fn.c (cond_unary_direct): New macro.
7935         (expand_cond_unary_optab_fn): Likewise.
7936         (direct_cond_unary_optab_supported_p): Likewise.
7937         * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
7938         loops using EXTRACT_LAST.
7939         * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
7940         (extract_last_<mode>): ...this optab.
7941         (vec_extract<mode><Vel>): Update accordingly.
7942
7943 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7944             Alan Hayward  <alan.hayward@arm.com>
7945             David Sherwood  <david.sherwood@arm.com>
7946
7947         * target.def (empty_mask_is_expensive): New hook.
7948         * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
7949         * doc/tm.texi: Regenerate.
7950         * targhooks.h (default_empty_mask_is_expensive): Declare.
7951         * targhooks.c (default_empty_mask_is_expensive): New function.
7952         * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
7953         if the target says that empty masks are expensive.
7954         * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
7955         New function.
7956         (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
7957
7958 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7959             Alan Hayward  <alan.hayward@arm.com>
7960             David Sherwood  <david.sherwood@arm.com>
7961
7962         * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
7963         (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
7964         (vect_use_loop_mask_for_alignment_p): New function.
7965         (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
7966         * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
7967         niters_skip argument.  Make sure that the first niters_skip elements
7968         of the first iteration are inactive.
7969         (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
7970         Update call to vect_set_loop_masks_directly.
7971         (get_misalign_in_elems): New function, split out from...
7972         (vect_gen_prolog_loop_niters): ...here.
7973         (vect_update_init_of_dr): Take a code argument that specifies whether
7974         the adjustment should be added or subtracted.
7975         (vect_update_init_of_drs): Likewise.
7976         (vect_prepare_for_masked_peels): New function.
7977         (vect_do_peeling): Skip prologue peeling if we're using a mask
7978         instead.  Update call to vect_update_inits_of_drs.
7979         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
7980         mask_skip_niters.
7981         (vect_analyze_loop_2): Allow fully-masked loops with peeling for
7982         alignment.  Do not include the number of peeled iterations in
7983         the minimum threshold in that case.
7984         (vectorizable_induction): Adjust the start value down by
7985         LOOP_VINFO_MASK_SKIP_NITERS iterations.
7986         (vect_transform_loop): Call vect_prepare_for_masked_peels.
7987         Take the number of skipped iterations into account when calculating
7988         the loop bounds.
7989         * tree-vect-stmts.c (vect_gen_while_not): New function.
7990
7991 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7992             Alan Hayward  <alan.hayward@arm.com>
7993             David Sherwood  <david.sherwood@arm.com>
7994
7995         * doc/sourcebuild.texi (vect_fully_masked): Document.
7996         * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
7997         default value to 0.
7998         * tree-vect-loop.c (vect_analyze_loop_costing): New function,
7999         split out from...
8000         (vect_analyze_loop_2): ...here. Don't check the vectorization
8001         factor against the number of loop iterations if the loop is
8002         fully-masked.
8003
8004 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8005             Alan Hayward  <alan.hayward@arm.com>
8006             David Sherwood  <david.sherwood@arm.com>
8007
8008         * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
8009         (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
8010         (dump_groups): Update accordingly.
8011         (iv_use::mem_type): New member variable.
8012         (address_p): New function.
8013         (record_use): Add a mem_type argument and initialize the new
8014         mem_type field.
8015         (record_group_use): Add a mem_type argument.  Use address_p.
8016         Remove obsolete null checks of base_object.  Update call to record_use.
8017         (find_interesting_uses_op): Update call to record_group_use.
8018         (find_interesting_uses_cond): Likewise.
8019         (find_interesting_uses_address): Likewise.
8020         (get_mem_type_for_internal_fn): New function.
8021         (find_address_like_use): Likewise.
8022         (find_interesting_uses_stmt): Try find_address_like_use before
8023         calling find_interesting_uses_op.
8024         (addr_offset_valid_p): Use the iv mem_type field as the type
8025         of the addressed memory.
8026         (add_autoinc_candidates): Likewise.
8027         (get_address_cost): Likewise.
8028         (split_small_address_groups_p): Use address_p.
8029         (split_address_groups): Likewise.
8030         (add_iv_candidate_for_use): Likewise.
8031         (autoinc_possible_for_pair): Likewise.
8032         (rewrite_groups): Likewise.
8033         (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
8034         (determine_group_iv_cost): Update after split of USE_ADDRESS.
8035         (get_alias_ptr_type_for_ptr_address): New function.
8036         (rewrite_use_address): Rewrite address uses in calls that were
8037         identified by find_address_like_use.
8038
8039 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8040             Alan Hayward  <alan.hayward@arm.com>
8041             David Sherwood  <david.sherwood@arm.com>
8042
8043         * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
8044         TARGET_MEM_REFs.
8045         * gimple-expr.h (is_gimple_addressable: Likewise.
8046         * gimple-expr.c (is_gimple_address): Likewise.
8047         * internal-fn.c (expand_call_mem_ref): New function.
8048         (expand_mask_load_optab_fn): Use it.
8049         (expand_mask_store_optab_fn): Likewise.
8050
8051 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8052             Alan Hayward  <alan.hayward@arm.com>
8053             David Sherwood  <david.sherwood@arm.com>
8054
8055         * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
8056         (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
8057         (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
8058         (cond_umax@var{mode}): Document.
8059         * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
8060         (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
8061         (cond_umin_optab, cond_umax_optab): New optabs.
8062         * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
8063         (COND_IOR, COND_XOR): New internal functions.
8064         * internal-fn.h (get_conditional_internal_fn): Declare.
8065         * internal-fn.c (cond_binary_direct): New macro.
8066         (expand_cond_binary_optab_fn): Likewise.
8067         (direct_cond_binary_optab_supported_p): Likewise.
8068         (get_conditional_internal_fn): New function.
8069         * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
8070         Cope with reduction statements that are vectorized as calls rather
8071         than assignments.
8072         * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
8073         * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
8074         (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
8075         (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
8076         (UNSPEC_COND_EOR): New unspecs.
8077         (optab): Add mappings for them.
8078         (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
8079         (sve_int_op, sve_fp_op): New int attributes.
8080
8081 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8082             Alan Hayward  <alan.hayward@arm.com>
8083             David Sherwood  <david.sherwood@arm.com>
8084
8085         * optabs.def (while_ult_optab): New optab.
8086         * doc/md.texi (while_ult@var{m}@var{n}): Document.
8087         * internal-fn.def (WHILE_ULT): New internal function.
8088         * internal-fn.h (direct_internal_fn_supported_p): New override
8089         that takes two types as argument.
8090         * internal-fn.c (while_direct): New macro.
8091         (expand_while_optab_fn): New function.
8092         (convert_optab_supported_p): Likewise.
8093         (direct_while_optab_supported_p): New macro.
8094         * wide-int.h (wi::udiv_ceil): New function.
8095         * tree-vectorizer.h (rgroup_masks): New structure.
8096         (vec_loop_masks): New typedef.
8097         (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
8098         and fully_masked_p.
8099         (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
8100         (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
8101         (vect_max_vf): New function.
8102         (slpeel_make_loop_iterate_ntimes): Delete.
8103         (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
8104         (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
8105         (vect_record_loop_mask, vect_get_loop_mask): Likewise.
8106         * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
8107         internal-fn.h, stor-layout.h and optabs-query.h.
8108         (vect_set_loop_mask): New function.
8109         (add_preheader_seq): Likewise.
8110         (add_header_seq): Likewise.
8111         (interleave_supported_p): Likewise.
8112         (vect_maybe_permute_loop_masks): Likewise.
8113         (vect_set_loop_masks_directly): Likewise.
8114         (vect_set_loop_condition_masked): Likewise.
8115         (vect_set_loop_condition_unmasked): New function, split out from
8116         slpeel_make_loop_iterate_ntimes.
8117         (slpeel_make_loop_iterate_ntimes): Rename to..
8118         (vect_set_loop_condition): ...this.  Use vect_set_loop_condition_masked
8119         for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
8120         (vect_do_peeling): Update call accordingly.
8121         (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
8122         loops.
8123         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
8124         mask_compare_type, can_fully_mask_p and fully_masked_p.
8125         (release_vec_loop_masks): New function.
8126         (_loop_vec_info): Use it to free the loop masks.
8127         (can_produce_all_loop_masks_p): New function.
8128         (vect_get_max_nscalars_per_iter): Likewise.
8129         (vect_verify_full_masking): Likewise.
8130         (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
8131         retries, and free the mask rgroups before retrying.  Check loop-wide
8132         reasons for disallowing fully-masked loops.  Make the final decision
8133         about whether use a fully-masked loop or not.
8134         (vect_estimate_min_profitable_iters): Do not assume that peeling
8135         for the number of iterations will be needed for fully-masked loops.
8136         (vectorizable_reduction): Disable fully-masked loops.
8137         (vectorizable_live_operation): Likewise.
8138         (vect_halve_mask_nunits): New function.
8139         (vect_double_mask_nunits): Likewise.
8140         (vect_record_loop_mask): Likewise.
8141         (vect_get_loop_mask): Likewise.
8142         (vect_transform_loop): Handle the case in which the final loop
8143         iteration might handle a partial vector.  Call vect_set_loop_condition
8144         instead of slpeel_make_loop_iterate_ntimes.
8145         * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
8146         (check_load_store_masking): New function.
8147         (prepare_load_store_mask): Likewise.
8148         (vectorizable_store): Handle fully-masked loops.
8149         (vectorizable_load): Likewise.
8150         (supportable_widening_operation): Use vect_halve_mask_nunits for
8151         booleans.
8152         (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
8153         (vect_gen_while): New function.
8154         * config/aarch64/aarch64.md (umax<mode>3): New expander.
8155         (aarch64_uqdec<mode>): New insn.
8156
8157 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8158             Alan Hayward  <alan.hayward@arm.com>
8159             David Sherwood  <david.sherwood@arm.com>
8160
8161         * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
8162         (reduc_xor_scal_optab): New optabs.
8163         * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
8164         (reduc_xor_scal_@var{m}): Document.
8165         * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
8166         * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
8167         internal functions.
8168         * fold-const-call.c (fold_const_call): Handle them.
8169         * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
8170         internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
8171         * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
8172         (*reduc_<bit_reduc>_scal_<mode>): New patterns.
8173         * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
8174         (UNSPEC_XORV): New unspecs.
8175         (optab): Add entries for them.
8176         (BITWISEV): New int iterator.
8177         (bit_reduc_op): New int attributes.
8178
8179 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8180             Alan Hayward  <alan.hayward@arm.com>
8181             David Sherwood  <david.sherwood@arm.com>
8182
8183         * doc/md.texi (vec_shl_insert_@var{m}): New optab.
8184         * internal-fn.def (VEC_SHL_INSERT): New internal function.
8185         * optabs.def (vec_shl_insert_optab): New optab.
8186         * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
8187         (duplicate_and_interleave): Likewise.
8188         * tree-vect-loop.c: Include internal-fn.h.
8189         (neutral_op_for_slp_reduction): New function, split out from
8190         get_initial_defs_for_reduction.
8191         (get_initial_def_for_reduction): Handle option 2 for variable-length
8192         vectors by loading the neutral value into a vector and then shifting
8193         the initial value into element 0.
8194         (get_initial_defs_for_reduction): Replace the code argument with
8195         the neutral value calculated by neutral_op_for_slp_reduction.
8196         Use gimple_build_vector for constant-length vectors.
8197         Use IFN_VEC_SHL_INSERT for variable-length vectors if all
8198         but the first group_size elements have a neutral value.
8199         Use duplicate_and_interleave otherwise.
8200         (vect_create_epilog_for_reduction): Take a neutral_op parameter.
8201         Update call to get_initial_defs_for_reduction.  Handle SLP
8202         reductions for variable-length vectors by creating one vector
8203         result for each scalar result, with the elements associated
8204         with other scalar results stubbed out with the neutral value.
8205         (vectorizable_reduction): Call neutral_op_for_slp_reduction.
8206         Require IFN_VEC_SHL_INSERT for double reductions on
8207         variable-length vectors, or SLP reductions that have
8208         a neutral value.  Require can_duplicate_and_interleave_p
8209         support for variable-length unchained SLP reductions if there
8210         is no neutral value, such as for MIN/MAX reductions.  Also require
8211         the number of vector elements to be a multiple of the number of
8212         SLP statements when doing variable-length unchained SLP reductions.
8213         Update call to vect_create_epilog_for_reduction.
8214         * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
8215         and remove initial values.
8216         (duplicate_and_interleave): Make public.
8217         * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
8218         * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
8219
8220 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8221             Alan Hayward  <alan.hayward@arm.com>
8222             David Sherwood  <david.sherwood@arm.com>
8223
8224         * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
8225         (can_duplicate_and_interleave_p): New function.
8226         (vect_get_and_check_slp_defs): Take the vector of statements
8227         rather than just the current one.  Remove excess parentheses.
8228         Restriction rejectinon of vect_constant_def and vect_external_def
8229         for variable-length vectors to boolean types, or types for which
8230         can_duplicate_and_interleave_p is false.
8231         (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
8232         (duplicate_and_interleave): New function.
8233         (vect_get_constant_vectors): Use gimple_build_vector for
8234         constant-length vectors and suitable variable-length constant
8235         vectors.  Use duplicate_and_interleave for other variable-length
8236         vectors.  Don't defer the update when inserting new statements.
8237
8238 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8239             Alan Hayward  <alan.hayward@arm.com>
8240             David Sherwood  <david.sherwood@arm.com>
8241
8242         * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
8243         min_profitable_iters doesn't go negative.
8244
8245 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8246             Alan Hayward  <alan.hayward@arm.com>
8247             David Sherwood  <david.sherwood@arm.com>
8248
8249         * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
8250         (vec_mask_store_lanes@var{m}@var{n}): Likewise.
8251         * optabs.def (vec_mask_load_lanes_optab): New optab.
8252         (vec_mask_store_lanes_optab): Likewise.
8253         * internal-fn.def (MASK_LOAD_LANES): New internal function.
8254         (MASK_STORE_LANES): Likewise.
8255         * internal-fn.c (mask_load_lanes_direct): New macro.
8256         (mask_store_lanes_direct): Likewise.
8257         (expand_mask_load_optab_fn): Handle masked operations.
8258         (expand_mask_load_lanes_optab_fn): New macro.
8259         (expand_mask_store_optab_fn): Handle masked operations.
8260         (expand_mask_store_lanes_optab_fn): New macro.
8261         (direct_mask_load_lanes_optab_supported_p): Likewise.
8262         (direct_mask_store_lanes_optab_supported_p): Likewise.
8263         * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
8264         parameter.
8265         (vect_load_lanes_supported): Likewise.
8266         * tree-vect-data-refs.c (strip_conversion): New function.
8267         (can_group_stmts_p): Likewise.
8268         (vect_analyze_data_ref_accesses): Use it instead of checking
8269         for a pair of assignments.
8270         (vect_store_lanes_supported): Take a masked_p parameter.
8271         (vect_load_lanes_supported): Likewise.
8272         * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
8273         vect_store_lanes_supported and vect_load_lanes_supported.
8274         * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
8275         * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
8276         parameter.  Don't allow gaps for masked accesses.
8277         Use vect_get_store_rhs.  Update calls to vect_store_lanes_supported
8278         and vect_load_lanes_supported.
8279         (get_load_store_type): Take a masked_p parameter and update
8280         call to get_group_load_store_type.
8281         (vectorizable_store): Update call to get_load_store_type.
8282         Handle IFN_MASK_STORE_LANES.
8283         (vectorizable_load): Update call to get_load_store_type.
8284         Handle IFN_MASK_LOAD_LANES.
8285
8286 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8287             Alan Hayward  <alan.hayward@arm.com>
8288             David Sherwood  <david.sherwood@arm.com>
8289
8290         * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
8291         modes for SVE.
8292         * config/aarch64/aarch64-protos.h
8293         (aarch64_sve_struct_memory_operand_p): Declare.
8294         * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
8295         (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
8296         (VPRED, vpred): Handle SVE structure modes.
8297         * config/aarch64/constraints.md (Utx): New constraint.
8298         * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
8299         (aarch64_sve_struct_nonimmediate_operand): New predicates.
8300         * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
8301         * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
8302         (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
8303         structure modes.  Split into pieces after RA.
8304         (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
8305         (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
8306         New patterns.
8307         * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
8308         SVE structure modes.
8309         (aarch64_classify_address): Likewise.
8310         (sizetochar): Move earlier in file.
8311         (aarch64_print_operand): Handle SVE register lists.
8312         (aarch64_array_mode): New function.
8313         (aarch64_sve_struct_memory_operand_p): Likewise.
8314         (TARGET_ARRAY_MODE): Redefine.
8315
8316 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8317             Alan Hayward  <alan.hayward@arm.com>
8318             David Sherwood  <david.sherwood@arm.com>
8319
8320         * target.def (array_mode): New target hook.
8321         * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
8322         * doc/tm.texi: Regenerate.
8323         * hooks.h (hook_optmode_mode_uhwi_none): Declare.
8324         * hooks.c (hook_optmode_mode_uhwi_none): New function.
8325         * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
8326         targetm.array_mode.
8327         * stor-layout.c (mode_for_array): Likewise.  Support polynomial
8328         type sizes.
8329
8330 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8331             Alan Hayward  <alan.hayward@arm.com>
8332             David Sherwood  <david.sherwood@arm.com>
8333
8334         * fold-const.c (fold_binary_loc): Check the argument types
8335         rather than the result type when testing for a vector operation.
8336
8337 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8338
8339         * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
8340         * doc/tm.texi: Regenerate.
8341
8342 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8343             Alan Hayward  <alan.hayward@arm.com>
8344             David Sherwood  <david.sherwood@arm.com>
8345
8346         * doc/invoke.texi (-msve-vector-bits=): Document new option.
8347         (sve): Document new AArch64 extension.
8348         * doc/md.texi (w): Extend the description of the AArch64
8349         constraint to include SVE vectors.
8350         (Upl, Upa): Document new AArch64 predicate constraints.
8351         * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
8352         enum.
8353         * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
8354         (msve-vector-bits=): New option.
8355         * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
8356         SVE when these are disabled.
8357         (sve): New extension.
8358         * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
8359         modes.  Adjust their number of units based on aarch64_sve_vg.
8360         (MAX_BITSIZE_MODE_ANY_MODE): Define.
8361         * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
8362         aarch64_addr_query_type.
8363         (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
8364         (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
8365         (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
8366         (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
8367         (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
8368         (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
8369         (aarch64_simd_imm_zero_p): Delete.
8370         (aarch64_check_zero_based_sve_index_immediate): Declare.
8371         (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
8372         (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
8373         (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
8374         (aarch64_sve_float_mul_immediate_p): Likewise.
8375         (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
8376         rather than an rtx.
8377         (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
8378         (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
8379         (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
8380         (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
8381         (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
8382         (aarch64_regmode_natural_size): Likewise.
8383         * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
8384         (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
8385         left one place.
8386         (AARCH64_ISA_SVE, TARGET_SVE): New macros.
8387         (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
8388         for VG and the SVE predicate registers.
8389         (V_ALIASES): Add a "z"-prefixed alias.
8390         (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
8391         (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
8392         (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
8393         (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
8394         (REG_CLASS_NAMES): Add entries for them.
8395         (REG_CLASS_CONTENTS): Likewise.  Update ALL_REGS to include VG
8396         and the predicate registers.
8397         (aarch64_sve_vg): Declare.
8398         (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
8399         (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
8400         (REGMODE_NATURAL_SIZE): Define.
8401         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
8402         SVE macros.
8403         * config/aarch64/aarch64.c: Include cfgrtl.h.
8404         (simd_immediate_info): Add a constructor for series vectors,
8405         and an associated step field.
8406         (aarch64_sve_vg): New variable.
8407         (aarch64_dbx_register_number): Handle VG and the predicate registers.
8408         (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
8409         (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
8410         (VEC_ANY_DATA, VEC_STRUCT): New constants.
8411         (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
8412         (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
8413         (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
8414         (aarch64_get_mask_mode): New functions.
8415         (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
8416         and FP_LO_REGS.  Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
8417         (aarch64_hard_regno_mode_ok): Handle VG.  Also handle the SVE
8418         predicate modes and predicate registers.  Explicitly restrict
8419         GPRs to modes of 16 bytes or smaller.  Only allow FP registers
8420         to store a vector mode if it is recognized by
8421         aarch64_classify_vector_mode.
8422         (aarch64_regmode_natural_size): New function.
8423         (aarch64_hard_regno_caller_save_mode): Return the original mode
8424         for predicates.
8425         (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
8426         (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
8427         (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
8428         (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
8429         functions.
8430         (aarch64_add_offset): Add a temp2 parameter.  Assert that temp1
8431         does not overlap dest if the function is frame-related.  Handle
8432         SVE constants.
8433         (aarch64_split_add_offset): New function.
8434         (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
8435         them aarch64_add_offset.
8436         (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
8437         and update call to aarch64_sub_sp.
8438         (aarch64_add_cfa_expression): New function.
8439         (aarch64_expand_prologue): Pass extra temporary registers to the
8440         functions above.  Handle the case in which we need to emit new
8441         DW_CFA_expressions for registers that were originally saved
8442         relative to the stack pointer, but now have to be expressed
8443         relative to the frame pointer.
8444         (aarch64_output_mi_thunk): Pass extra temporary registers to the
8445         functions above.
8446         (aarch64_expand_epilogue): Likewise.  Prevent inheritance of
8447         IP0 and IP1 values for SVE frames.
8448         (aarch64_expand_vec_series): New function.
8449         (aarch64_expand_sve_widened_duplicate): Likewise.
8450         (aarch64_expand_sve_const_vector): Likewise.
8451         (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
8452         Handle SVE constants.  Use emit_move_insn to move a force_const_mem
8453         into the register, rather than emitting a SET directly.
8454         (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
8455         (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
8456         (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
8457         (offset_9bit_signed_scaled_p): New functions.
8458         (aarch64_replicate_bitmask_imm): New function.
8459         (aarch64_bitmask_imm): Use it.
8460         (aarch64_cannot_force_const_mem): Reject expressions involving
8461         a CONST_POLY_INT.  Update call to aarch64_classify_symbol.
8462         (aarch64_classify_index): Handle SVE indices, by requiring
8463         a plain register index with a scale that matches the element size.
8464         (aarch64_classify_address): Handle SVE addresses.  Assert that
8465         the mode of the address is VOIDmode or an integer mode.
8466         Update call to aarch64_classify_symbol.
8467         (aarch64_classify_symbolic_expression): Update call to
8468         aarch64_classify_symbol.
8469         (aarch64_const_vec_all_in_range_p): New function.
8470         (aarch64_print_vector_float_operand): Likewise.
8471         (aarch64_print_operand): Handle 'N' and 'C'.  Use "zN" rather than
8472         "vN" for FP registers with SVE modes.  Handle (const ...) vectors
8473         and the FP immediates 1.0 and 0.5.
8474         (aarch64_print_address_internal): Handle SVE addresses.
8475         (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
8476         (aarch64_regno_regclass): Handle predicate registers.
8477         (aarch64_secondary_reload): Handle big-endian reloads of SVE
8478         data modes.
8479         (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
8480         (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
8481         (aarch64_convert_sve_vector_bits): New function.
8482         (aarch64_override_options): Use it to handle -msve-vector-bits=.
8483         (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
8484         rather than an rtx.
8485         (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
8486         Handle SVE vector and predicate modes.  Accept VL-based constants
8487         that need only one temporary register, and VL offsets that require
8488         no temporary registers.
8489         (aarch64_conditional_register_usage): Mark the predicate registers
8490         as fixed if SVE isn't available.
8491         (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
8492         Return true for SVE vector and predicate modes.
8493         (aarch64_simd_container_mode): Take the number of bits as a poly_int64
8494         rather than an unsigned int.  Handle SVE modes.
8495         (aarch64_preferred_simd_mode): Update call accordingly.  Handle
8496         SVE modes.
8497         (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
8498         if SVE is enabled.
8499         (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
8500         (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
8501         (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
8502         (aarch64_sve_float_mul_immediate_p): New functions.
8503         (aarch64_sve_valid_immediate): New function.
8504         (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
8505         Explicitly reject structure modes.  Check for INDEX constants.
8506         Handle PTRUE and PFALSE constants.
8507         (aarch64_check_zero_based_sve_index_immediate): New function.
8508         (aarch64_simd_imm_zero_p): Delete.
8509         (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
8510         vector modes.  Accept constants in the range of CNT[BHWD].
8511         (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
8512         ask for an Advanced SIMD mode.
8513         (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
8514         (aarch64_simd_vector_alignment): Handle SVE predicates.
8515         (aarch64_vectorize_preferred_vector_alignment): New function.
8516         (aarch64_simd_vector_alignment_reachable): Use it instead of
8517         the vector size.
8518         (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
8519         (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
8520         functions.
8521         (MAX_VECT_LEN): Delete.
8522         (expand_vec_perm_d): Add a vec_flags field.
8523         (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
8524         (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
8525         (aarch64_evpc_ext): Don't apply a big-endian lane correction
8526         for SVE modes.
8527         (aarch64_evpc_rev): Rename to...
8528         (aarch64_evpc_rev_local): ...this.  Use a predicated operation for SVE.
8529         (aarch64_evpc_rev_global): New function.
8530         (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
8531         (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
8532         MAX_VECT_LEN.
8533         (aarch64_evpc_sve_tbl): New function.
8534         (aarch64_expand_vec_perm_const_1): Update after rename of
8535         aarch64_evpc_rev.  Handle SVE permutes too, trying
8536         aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
8537         than aarch64_evpc_tbl.
8538         (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
8539         (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
8540         (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
8541         (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
8542         (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
8543         (aarch64_expand_sve_vcond): New functions.
8544         (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
8545         of aarch64_vector_mode_p.
8546         (aarch64_dwarf_poly_indeterminate_value): New function.
8547         (aarch64_compute_pressure_classes): Likewise.
8548         (aarch64_can_change_mode_class): Likewise.
8549         (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
8550         (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
8551         (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
8552         (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
8553         (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
8554         (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
8555         * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
8556         (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
8557         constraints.
8558         (Dn, Dl, Dr): Accept const as well as const_vector.
8559         (Dz): Likewise.  Compare against CONST0_RTX.
8560         * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
8561         of "vector" where appropriate.
8562         (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
8563         (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
8564         (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
8565         (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
8566         (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
8567         (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
8568         (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
8569         (v_int_equiv): Extend to SVE modes.
8570         (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
8571         mode attributes.
8572         (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
8573         (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
8574         (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
8575         (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
8576         (SVE_COND_FP_CMP): New int iterators.
8577         (perm_hilo): Handle the new unpack unspecs.
8578         (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
8579         attributes.
8580         * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
8581         (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
8582         (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
8583         (aarch64_equality_operator, aarch64_constant_vector_operand)
8584         (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
8585         (aarch64_sve_nonimmediate_operand): Likewise.
8586         (aarch64_sve_general_operand): Likewise.
8587         (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
8588         (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
8589         (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
8590         (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
8591         (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
8592         (aarch64_sve_float_arith_immediate): Likewise.
8593         (aarch64_sve_float_arith_with_sub_immediate): Likewise.
8594         (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
8595         (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
8596         (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
8597         (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
8598         (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
8599         (aarch64_sve_float_arith_operand): Likewise.
8600         (aarch64_sve_float_arith_with_sub_operand): Likewise.
8601         (aarch64_sve_float_mul_operand): Likewise.
8602         (aarch64_sve_vec_perm_operand): Likewise.
8603         (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
8604         (aarch64_mov_operand): Accept const_poly_int and const_vector.
8605         (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
8606         as well as const_vector.
8607         (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
8608         in file.  Use CONST0_RTX and CONSTM1_RTX.
8609         (aarch64_simd_or_scalar_imm_zero): Likewise.  Add match_codes.
8610         (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
8611         Use aarch64_simd_imm_zero.
8612         * config/aarch64/aarch64-sve.md: New file.
8613         * config/aarch64/aarch64.md: Include it.
8614         (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
8615         (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
8616         (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
8617         (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
8618         (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
8619         (sve): New attribute.
8620         (enabled): Disable instructions with the sve attribute unless
8621         TARGET_SVE.
8622         (movqi, movhi): Pass CONST_POLY_INT operaneds through
8623         aarch64_expand_mov_immediate.
8624         (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
8625         CNT[BHSD] immediates.
8626         (movti): Split CONST_POLY_INT moves into two halves.
8627         (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
8628         Split additions that need a temporary here if the destination
8629         is the stack pointer.
8630         (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
8631         (*add<mode>3_poly_1): New instruction.
8632         (set_clobber_cc): New expander.
8633
8634 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8635
8636         * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
8637         parameter and use it instead of GET_MODE_SIZE (innermode).  Use
8638         inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
8639         Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
8640         GET_MODE_NUNITS (innermode).  Also add a first_elem parameter.
8641         Change innermode from fixed_mode_size to machine_mode.
8642         (simplify_subreg): Update call accordingly.  Handle a constant-sized
8643         subreg of a variable-length CONST_VECTOR.
8644
8645 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8646             Alan Hayward  <alan.hayward@arm.com>
8647             David Sherwood  <david.sherwood@arm.com>
8648
8649         * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
8650         (add_offset_to_base): New function, split out from...
8651         (create_mem_ref): ...here.  When handling a scale other than 1,
8652         check first whether the address is valid without the offset.
8653         Add it into the base if so, leaving the index and scale as-is.
8654
8655 2018-01-12  Jakub Jelinek  <jakub@redhat.com>
8656
8657         PR c++/83778
8658         * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
8659         fold_for_warn before checking if arg2 is INTEGER_CST.
8660
8661 2018-01-12  Segher Boessenkool  <segher@kernel.crashing.org>
8662
8663         * config/rs6000/predicates.md (load_multiple_operation): Delete.
8664         (store_multiple_operation): Delete.
8665         * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
8666         * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
8667         * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
8668         guarded by TARGET_STRING.
8669         (rs6000_output_load_multiple): Delete.
8670         * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
8671         OPTION_MASK_STRING / TARGET_STRING handling.
8672         (print_operand) <'N', 'O'>: Add comment that these are unused now.
8673         (const rs6000_opt_masks) <"string">: Change mask to 0.
8674         * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
8675         (MASK_STRING): Delete.
8676         * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
8677         parts.  Simplify.
8678         (load_multiple): Delete.
8679         (*ldmsi8): Delete.
8680         (*ldmsi7): Delete.
8681         (*ldmsi6): Delete.
8682         (*ldmsi5): Delete.
8683         (*ldmsi4): Delete.
8684         (*ldmsi3): Delete.
8685         (store_multiple): Delete.
8686         (*stmsi8): Delete.
8687         (*stmsi7): Delete.
8688         (*stmsi6): Delete.
8689         (*stmsi5): Delete.
8690         (*stmsi4): Delete.
8691         (*stmsi3): Delete.
8692         (movmemsi_8reg): Delete.
8693         (corresponding unnamed define_insn): Delete.
8694         (movmemsi_6reg): Delete.
8695         (corresponding unnamed define_insn): Delete.
8696         (movmemsi_4reg): Delete.
8697         (corresponding unnamed define_insn): Delete.
8698         (movmemsi_2reg): Delete.
8699         (corresponding unnamed define_insn): Delete.
8700         (movmemsi_1reg): Delete.
8701         (corresponding unnamed define_insn): Delete.
8702         * config/rs6000/rs6000.opt (mno-string): New.
8703         (mstring): Replace by deprecation warning stub.
8704         * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
8705
8706 2018-01-12  Jakub Jelinek  <jakub@redhat.com>
8707
8708         * regrename.c (regrename_do_replace): If replacing the same
8709         reg multiple times, try to reuse last created gen_raw_REG.
8710
8711         PR debug/81155
8712         * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
8713         main to workaround a bug in GDB.
8714
8715 2018-01-12  Tom de Vries  <tom@codesourcery.com>
8716
8717         PR target/83737
8718         * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
8719
8720 2018-01-12  Vladimir Makarov  <vmakarov@redhat.com>
8721
8722         PR rtl-optimization/80481
8723         * ira-color.c (get_cap_member): New function.
8724         (allocnos_conflict_by_live_ranges_p): Use it.
8725         (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
8726         (setup_slot_coalesced_allocno_live_ranges): Ditto.
8727
8728 2018-01-12  Uros Bizjak  <ubizjak@gmail.com>
8729
8730         PR target/83628
8731         * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
8732         (*saddl_se_1): Ditto.
8733         (*ssubsi_1): Ditto.
8734         (*ssubl_se_1): Ditto.
8735
8736 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
8737
8738         * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
8739         rather than wi::to_widest for DR_INITs.
8740         * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
8741         wi::to_poly_offset rather than wi::to_offset for DR_INIT.
8742         (vect_analyze_data_ref_accesses): Require both DR_INITs to be
8743         INTEGER_CSTs.
8744         (vect_analyze_group_access_1): Note that here.
8745
8746 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
8747
8748         * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
8749         polynomial type sizes.
8750
8751 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
8752
8753         * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
8754         poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
8755         (gimple_add_tmp_var): Likewise.
8756
8757 2018-01-12  Martin Liska  <mliska@suse.cz>
8758
8759         * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
8760         (gimple_alloc_sizes): Likewise.
8761         (dump_gimple_statistics): Use PRIu64 in printf format.
8762         * gimple.h: Change uint64_t to int.
8763
8764 2018-01-12  Martin Liska  <mliska@suse.cz>
8765
8766         * tree-core.h: Use uint64_t instead of int.
8767         * tree.c (tree_node_counts): Likewise.
8768         (tree_node_sizes): Likewise.
8769         (dump_tree_statistics): Use PRIu64 in printf format.
8770
8771 2018-01-12  Martin Liska  <mliska@suse.cz>
8772
8773         * Makefile.in: As qsort_chk is implemented in vec.c, add
8774         vec.o to linkage of gencfn-macros.
8775         * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
8776         passing the info to record_node_allocation_statistics.
8777         (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
8778         and pass the info.
8779         * ggc-common.c (struct ggc_usage): Add operator== and use
8780         it in operator< and compare function.
8781         * mem-stats.h (struct mem_usage): Likewise.
8782         * vec.c (struct vec_usage): Remove operator< and compare
8783         function. Can be simply inherited.
8784
8785 2018-01-12  Martin Jambor  <mjambor@suse.cz>
8786
8787         PR target/81616
8788         * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
8789         * tree-ssa-math-opts.c: Include domwalk.h.
8790         (convert_mult_to_fma_1): New function.
8791         (fma_transformation_info): New type.
8792         (fma_deferring_state): Likewise.
8793         (cancel_fma_deferring): New function.
8794         (result_of_phi): Likewise.
8795         (last_fma_candidate_feeds_initial_phi): Likewise.
8796         (convert_mult_to_fma): Added deferring logic, split actual
8797         transformation to convert_mult_to_fma_1.
8798         (math_opts_dom_walker): New type.
8799         (math_opts_dom_walker::after_dom_children): New method, body moved
8800         here from pass_optimize_widening_mul::execute, added deferring logic
8801         bits.
8802         (pass_optimize_widening_mul::execute): Moved most of code to
8803         math_opts_dom_walker::after_dom_children.
8804         * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
8805         * config/i386/i386.c (ix86_option_override_internal): Added
8806         maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
8807
8808 2018-01-12  Richard Biener  <rguenther@suse.de>
8809
8810         PR debug/83157
8811         * dwarf2out.c (gen_variable_die): Do not reset old_die for
8812         inline instance vars.
8813
8814 2018-01-12  Oleg Endo  <olegendo@gcc.gnu.org>
8815
8816         PR target/81819
8817         * config/rx/rx.c (rx_is_restricted_memory_address):
8818         Handle SUBREG case.
8819
8820 2018-01-12  Richard Biener  <rguenther@suse.de>
8821
8822         PR tree-optimization/80846
8823         * target.def (split_reduction): New target hook.
8824         * targhooks.c (default_split_reduction): New function.
8825         * targhooks.h (default_split_reduction): Declare.
8826         * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
8827         target requests first reduce vectors by combining low and high
8828         parts.
8829         * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
8830         (get_vectype_for_scalar_type_and_size): Export.
8831         * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
8832         * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
8833         * doc/tm.texi: Regenerate.
8834         * config/i386/i386.c (ix86_split_reduction): Implement
8835         TARGET_VECTORIZE_SPLIT_REDUCTION.
8836
8837 2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
8838
8839         PR target/83368
8840         * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
8841         in PIC mode except for TARGET_VXWORKS_RTP.
8842         * config/sparc/sparc.c: Include cfgrtl.h.
8843         (TARGET_INIT_PIC_REG): Define.
8844         (TARGET_USE_PSEUDO_PIC_REG): Likewise.
8845         (sparc_pic_register_p): New predicate.
8846         (sparc_legitimate_address_p): Use it.
8847         (sparc_legitimize_pic_address): Likewise.
8848         (sparc_delegitimize_address): Likewise.
8849         (sparc_mode_dependent_address_p): Likewise.
8850         (gen_load_pcrel_sym): Remove 4th parameter.
8851         (load_got_register): Adjust call to above.  Remove obsolete stuff.
8852         (sparc_expand_prologue): Do not call load_got_register here.
8853         (sparc_flat_expand_prologue): Likewise.
8854         (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
8855         (sparc_use_pseudo_pic_reg): New function.
8856         (sparc_init_pic_reg): Likewise.
8857         * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
8858         (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
8859
8860 2018-01-12  Christophe Lyon  <christophe.lyon@linaro.org>
8861
8862         * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
8863         Add item for branch_cost.
8864
8865 2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
8866
8867         PR rtl-optimization/83565
8868         * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
8869         not extend the result to a larger mode for rotate operations.
8870         (num_sign_bit_copies1): Likewise.
8871
8872 2018-01-12  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
8873
8874         PR target/40411
8875         * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
8876         -symbolic.
8877         Use values-Xc.o for -pedantic.
8878         Link with values-xpg4.o for C90, values-xpg6.o otherwise.
8879
8880 2018-01-12  Martin Liska  <mliska@suse.cz>
8881
8882         PR ipa/83054
8883         * ipa-devirt.c (final_warning_record::grow_type_warnings):
8884         New function.
8885         (possible_polymorphic_call_targets): Use it.
8886         (ipa_devirt): Likewise.
8887
8888 2018-01-12  Martin Liska  <mliska@suse.cz>
8889
8890         * profile-count.h (enum profile_quality): Use 0 as invalid
8891         enum value of profile_quality.
8892
8893 2018-01-12  Chung-Ju Wu  <jasonwucj@gmail.com>
8894
8895         * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
8896         -mext-string options.
8897
8898 2018-01-12  Richard Biener  <rguenther@suse.de>
8899
8900         * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
8901         DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
8902         * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
8903         Likewise.
8904         * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
8905
8906 2018-01-11  Michael Meissner  <meissner@linux.vnet.ibm.com>
8907
8908         * configure.ac (--with-long-double-format): Add support for the
8909         configuration option to change the default long double format on
8910         PowerPC systems.
8911         * config.gcc (powerpc*-linux*-*): Likewise.
8912         * configure: Regenerate.
8913         * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
8914         double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
8915         used without modification.
8916
8917 2018-01-11  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
8918
8919         * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
8920         (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
8921         * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
8922         MISC_BUILTIN_SPEC_BARRIER.
8923         (rs6000_init_builtins): Likewise.
8924         * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
8925         enum value.
8926         (speculation_barrier): New define_insn.
8927         * doc/extend.texi: Document __builtin_speculation_barrier.
8928
8929 2018-01-11  Jakub Jelinek  <jakub@redhat.com>
8930
8931         PR target/83203
8932         * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
8933         is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
8934         * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
8935         iterators.
8936         (ssescalarmodesuffix): Add 512-bit vectors.  Use "d" or "q" for
8937         integral modes instead of "ss" and "sd".
8938         (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
8939         vectors with 32-bit and 64-bit elements.
8940         (vecdupssescalarmodesuffix): New mode attribute.
8941         (vec_dup<mode>): Use it.
8942
8943 2018-01-11  H.J. Lu  <hongjiu.lu@intel.com>
8944
8945         PR target/83330
8946         * config/i386/i386.c (ix86_compute_frame_layout): Align stack
8947         frame if argument is passed on stack.
8948
8949 2018-01-11  Jakub Jelinek  <jakub@redhat.com>
8950
8951         PR target/82682
8952         * ree.c (combine_reaching_defs): Optimize also
8953         reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
8954         reg2=any_extend(exp); reg1=reg2;, formatting fix.
8955
8956 2018-01-11  Jan Hubicka  <hubicka@ucw.cz>
8957
8958         PR middle-end/83189
8959         * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
8960
8961 2018-01-11  Jan Hubicka  <hubicka@ucw.cz>
8962
8963         PR middle-end/83718
8964         * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
8965         after they are computed.
8966
8967 2018-01-11  Bin Cheng  <bin.cheng@arm.com>
8968
8969         PR tree-optimization/83695
8970         * gimple-loop-linterchange.cc
8971         (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
8972         reset cached scev information after interchange.
8973         (pass_linterchange::execute): Remove call to scev_reset_htab.
8974
8975 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
8976
8977         * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
8978         vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
8979         vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
8980         vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
8981         vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
8982         vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
8983         * config/arm/arm_neon_builtins.def (vfmal_lane_low,
8984         vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
8985         vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
8986         vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
8987         vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
8988         * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
8989         (V_lane_reg): Likewise.
8990         * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
8991         New define_expand.
8992         (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
8993         (vfmal_lane_low<mode>_intrinsic,
8994         vfmal_lane_low<vfmlsel2><mode>_intrinsic,
8995         vfmal_lane_high<vfmlsel2><mode>_intrinsic,
8996         vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
8997         vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
8998         vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
8999         vfmsl_lane_high<mode>_intrinsic): New define_insns.
9000
9001 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
9002
9003         * config/arm/arm-cpus.in (fp16fml): New feature.
9004         (ALL_SIMD): Add fp16fml.
9005         (armv8.2-a): Add fp16fml as an option.
9006         (armv8.3-a): Likewise.
9007         (armv8.4-a): Add fp16fml as part of fp16.
9008         * config/arm/arm.h (TARGET_FP16FML): Define.
9009         * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
9010         when appropriate.
9011         * config/arm/arm-modes.def (V2HF): Define.
9012         * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
9013         vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
9014         vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
9015         * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
9016         vfmsl_low, vfmsl_high): New set of builtins.
9017         * config/arm/iterators.md (PLUSMINUS): New code iterator.
9018         (vfml_op): New code attribute.
9019         (VFMLHALVES): New int iterator.
9020         (VFML, VFMLSEL): New mode attributes.
9021         (V_reg): Define mapping for V2HF.
9022         (V_hi, V_lo): New mode attributes.
9023         (VF_constraint): Likewise.
9024         (vfml_half, vfml_half_selector): New int attributes.
9025         * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
9026         define_expand.
9027         (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
9028         vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
9029         New define_insn.
9030         * config/arm/t-arm-elf (v8_fps): Add fp16fml.
9031         * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
9032         * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
9033         * doc/invoke.texi (ARM Options): Document fp16fml.  Update armv8.4-a
9034         documentation.
9035         * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
9036         Document new effective target and option set.
9037
9038 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
9039
9040         * config/arm/arm-cpus.in (armv8_4): New feature.
9041         (ARMv8_4a): New fgroup.
9042         (armv8.4-a): New arch.
9043         * config/arm/arm-tables.opt: Regenerate.
9044         * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
9045         * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
9046         * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
9047         Add matching rules for -march=armv8.4-a and extensions.
9048         * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
9049
9050 2018-01-11  Oleg Endo  <olegendo@gcc.gnu.org>
9051
9052         PR target/81821
9053         * config/rx/rx.md (BW): New mode attribute.
9054         (sync_lock_test_and_setsi): Add mode suffix to insn output.
9055
9056 2018-01-11  Richard Biener  <rguenther@suse.de>
9057
9058         PR tree-optimization/83435
9059         * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
9060         * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
9061         * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
9062
9063 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
9064             Alan Hayward  <alan.hayward@arm.com>
9065             David Sherwood  <david.sherwood@arm.com>
9066
9067         * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
9068         field.
9069         (aarch64_classify_address): Initialize it.  Track polynomial offsets.
9070         (aarch64_print_address_internal): Use it to check for a zero offset.
9071
9072 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
9073             Alan Hayward  <alan.hayward@arm.com>
9074             David Sherwood  <david.sherwood@arm.com>
9075
9076         * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
9077         * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
9078         Return a poly_int64 rather than a HOST_WIDE_INT.
9079         (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
9080         rather than a HOST_WIDE_INT.
9081         * config/aarch64/aarch64.h (aarch64_frame): Protect with
9082         HAVE_POLY_INT_H rather than HOST_WIDE_INT.  Change locals_offset,
9083         hard_fp_offset, frame_size, initial_adjust, callee_offset and
9084         final_offset from HOST_WIDE_INT to poly_int64.
9085         * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
9086         to_constant when getting the number of units in an Advanced SIMD
9087         mode.
9088         (aarch64_builtin_vectorized_function): Check for a constant number
9089         of units.
9090         * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
9091         GET_MODE_SIZE.
9092         (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
9093         attribute instead of GET_MODE_NUNITS.
9094         * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
9095         (aarch64_class_max_nregs): Use the constant_lowest_bound of the
9096         GET_MODE_SIZE for fixed-size registers.
9097         (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
9098         (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
9099         (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
9100         (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
9101         (aarch64_print_operand, aarch64_print_address_internal)
9102         (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
9103         (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
9104         (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
9105         Handle polynomial GET_MODE_SIZE.
9106         (aarch64_hard_regno_caller_save_mode): Likewise.  Return modes
9107         wider than SImode without modification.
9108         (tls_symbolic_operand_type): Use strip_offset instead of split_const.
9109         (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
9110         (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
9111         passing and returning SVE modes.
9112         (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
9113         rather than GEN_INT.
9114         (aarch64_emit_probe_stack_range): Take the size as a poly_int64
9115         rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
9116         (aarch64_allocate_and_probe_stack_space): Likewise.
9117         (aarch64_layout_frame): Cope with polynomial offsets.
9118         (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
9119         start_offset as a poly_int64 rather than a HOST_WIDE_INT.  Track
9120         polynomial offsets.
9121         (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
9122         (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
9123         poly_int64 rather than a HOST_WIDE_INT.
9124         (aarch64_get_separate_components, aarch64_process_components)
9125         (aarch64_expand_prologue, aarch64_expand_epilogue)
9126         (aarch64_use_return_insn_p): Handle polynomial frame offsets.
9127         (aarch64_anchor_offset): New function, split out from...
9128         (aarch64_legitimize_address): ...here.
9129         (aarch64_builtin_vectorization_cost): Handle polynomial
9130         TYPE_VECTOR_SUBPARTS.
9131         (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
9132         GET_MODE_NUNITS.
9133         (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
9134         number of elements from the PARALLEL rather than the mode.
9135         (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
9136         rather than GET_MODE_BITSIZE.
9137         (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
9138         (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
9139         (aarch64_expand_vec_perm_const_1): Handle polynomial
9140         d->perm.length () and d->perm elements.
9141         (aarch64_evpc_tbl): Likewise.  Use nelt rather than GET_MODE_NUNITS.
9142         Apply to_constant to d->perm elements.
9143         (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
9144         polynomial CONST_VECTOR_NUNITS.
9145         (aarch64_move_pointer): Take amount as a poly_int64 rather
9146         than an int.
9147         (aarch64_progress_pointer): Avoid temporary variable.
9148         * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
9149         the mode attribute instead of GET_MODE.
9150
9151 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
9152             Alan Hayward  <alan.hayward@arm.com>
9153             David Sherwood  <david.sherwood@arm.com>
9154
9155         * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
9156         x exists before using it.
9157         (aarch64_add_constant_internal): Rename to...
9158         (aarch64_add_offset_1): ...this.  Replace regnum with separate
9159         src and dest rtxes.  Handle the case in which they're different,
9160         including when the offset is zero.  Replace scratchreg with an rtx.
9161         Use 2 additions if there is no spare register into which we can
9162         move a 16-bit constant.
9163         (aarch64_add_constant): Delete.
9164         (aarch64_add_offset): Replace reg with separate src and dest
9165         rtxes.  Take a poly_int64 offset instead of a HOST_WIDE_INT.
9166         Use aarch64_add_offset_1.
9167         (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
9168         an rtx rather than an int.  Take the delta as a poly_int64
9169         rather than a HOST_WIDE_INT.  Use aarch64_add_offset.
9170         (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
9171         (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
9172         aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
9173         (aarch64_expand_epilogue): Update calls to aarch64_add_offset
9174         and aarch64_add_sp.
9175         (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
9176         aarch64_add_constant.
9177
9178 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
9179
9180         * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
9181         Use scalar_float_mode.
9182
9183 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
9184
9185         * config/aarch64/aarch64-simd.md
9186         (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
9187         (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
9188         (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
9189         (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
9190         (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
9191         (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
9192         (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
9193         (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
9194         (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
9195         (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
9196
9197 2018-01-11  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
9198
9199         PR target/83514
9200         * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
9201         targ_options->x_arm_arch_string is non NULL.
9202
9203 2018-01-11  Tamar Christina  <tamar.christina@arm.com>
9204
9205         * config/aarch64/aarch64.h
9206         (AARCH64_FL_FOR_ARCH8_4): Add  AARCH64_FL_DOTPROD.
9207
9208 2018-01-11  Sudakshina Das  <sudi.das@arm.com>
9209
9210         PR target/82096
9211         * expmed.c (emit_store_flag_force): Swap if const op0
9212         and change VOIDmode to mode of op0.
9213
9214 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
9215
9216         PR rtl-optimization/83761
9217         * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
9218         than bytes to mode_for_size.
9219
9220 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
9221
9222         PR middle-end/83189
9223         * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
9224         * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
9225         profile.
9226
9227 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
9228
9229         PR middle-end/83575
9230         * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
9231         when in layout mode.
9232         (cfg_layout_finalize): Do not verify cfg before we are out of layout.
9233         * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
9234         partition fixup.
9235
9236 2018-01-10  Michael Collison  <michael.collison@arm.com>
9237
9238         * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
9239         * config/aarch64/aarch64-option-extension.def: Add
9240         AARCH64_OPT_EXTENSION of 'fp16fml'.
9241         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
9242         (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
9243         * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
9244         * config/aarch64/constraints.md (Ui7): New constraint.
9245         * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
9246         (VFMLA_SEL_W): Ditto.
9247         (f16quad): Ditto.
9248         (f16mac1): Ditto.
9249         (VFMLA16_LOW): New int iterator.
9250         (VFMLA16_HIGH): Ditto.
9251         (UNSPEC_FMLAL): New unspec.
9252         (UNSPEC_FMLSL): Ditto.
9253         (UNSPEC_FMLAL2): Ditto.
9254         (UNSPEC_FMLSL2): Ditto.
9255         (f16mac): New code attribute.
9256         * config/aarch64/aarch64-simd-builtins.def
9257         (aarch64_fmlal_lowv2sf): Ditto.
9258         (aarch64_fmlsl_lowv2sf): Ditto.
9259         (aarch64_fmlalq_lowv4sf): Ditto.
9260         (aarch64_fmlslq_lowv4sf): Ditto.
9261         (aarch64_fmlal_highv2sf): Ditto.
9262         (aarch64_fmlsl_highv2sf): Ditto.
9263         (aarch64_fmlalq_highv4sf): Ditto.
9264         (aarch64_fmlslq_highv4sf): Ditto.
9265         (aarch64_fmlal_lane_lowv2sf): Ditto.
9266         (aarch64_fmlsl_lane_lowv2sf): Ditto.
9267         (aarch64_fmlal_laneq_lowv2sf): Ditto.
9268         (aarch64_fmlsl_laneq_lowv2sf): Ditto.
9269         (aarch64_fmlalq_lane_lowv4sf): Ditto.
9270         (aarch64_fmlsl_lane_lowv4sf): Ditto.
9271         (aarch64_fmlalq_laneq_lowv4sf): Ditto.
9272         (aarch64_fmlsl_laneq_lowv4sf): Ditto.
9273         (aarch64_fmlal_lane_highv2sf): Ditto.
9274         (aarch64_fmlsl_lane_highv2sf): Ditto.
9275         (aarch64_fmlal_laneq_highv2sf): Ditto.
9276         (aarch64_fmlsl_laneq_highv2sf): Ditto.
9277         (aarch64_fmlalq_lane_highv4sf): Ditto.
9278         (aarch64_fmlsl_lane_highv4sf): Ditto.
9279         (aarch64_fmlalq_laneq_highv4sf): Ditto.
9280         (aarch64_fmlsl_laneq_highv4sf): Ditto.
9281         * config/aarch64/aarch64-simd.md:
9282         (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
9283         (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
9284         (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
9285         (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
9286         (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
9287         (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
9288         (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
9289         (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
9290         (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
9291         (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
9292         (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
9293         (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
9294         (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
9295         (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
9296         (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
9297         (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
9298         (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
9299         (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
9300         (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
9301         (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
9302         * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
9303         (vfmlsl_low_u32): Ditto.
9304         (vfmlalq_low_u32): Ditto.
9305         (vfmlslq_low_u32): Ditto.
9306         (vfmlal_high_u32): Ditto.
9307         (vfmlsl_high_u32): Ditto.
9308         (vfmlalq_high_u32): Ditto.
9309         (vfmlslq_high_u32): Ditto.
9310         (vfmlal_lane_low_u32): Ditto.
9311         (vfmlsl_lane_low_u32): Ditto.
9312         (vfmlal_laneq_low_u32): Ditto.
9313         (vfmlsl_laneq_low_u32): Ditto.
9314         (vfmlalq_lane_low_u32): Ditto.
9315         (vfmlslq_lane_low_u32): Ditto.
9316         (vfmlalq_laneq_low_u32): Ditto.
9317         (vfmlslq_laneq_low_u32): Ditto.
9318         (vfmlal_lane_high_u32): Ditto.
9319         (vfmlsl_lane_high_u32): Ditto.
9320         (vfmlal_laneq_high_u32): Ditto.
9321         (vfmlsl_laneq_high_u32): Ditto.
9322         (vfmlalq_lane_high_u32): Ditto.
9323         (vfmlslq_lane_high_u32): Ditto.
9324         (vfmlalq_laneq_high_u32): Ditto.
9325         (vfmlslq_laneq_high_u32): Ditto.
9326         * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
9327         (AARCH64_FL_FOR_ARCH8_4): New.
9328         (AARCH64_ISA_F16FML): New ISA flag.
9329         (TARGET_F16FML): New feature flag for fp16fml.
9330         (doc/invoke.texi): Document new fp16fml option.
9331
9332 2018-01-10  Michael Collison  <michael.collison@arm.com>
9333
9334         * config/aarch64/aarch64-builtins.c:
9335         (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
9336         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
9337         (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
9338         * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
9339         (AARCH64_ISA_SHA3): New ISA flag.
9340         (TARGET_SHA3): New feature flag for sha3.
9341         * config/aarch64/iterators.md (sha512_op): New int attribute.
9342         (CRYPTO_SHA512): New int iterator.
9343         (UNSPEC_SHA512H): New unspec.
9344         (UNSPEC_SHA512H2): Ditto.
9345         (UNSPEC_SHA512SU0): Ditto.
9346         (UNSPEC_SHA512SU1): Ditto.
9347         * config/aarch64/aarch64-simd-builtins.def
9348         (aarch64_crypto_sha512hqv2di): New builtin.
9349         (aarch64_crypto_sha512h2qv2di): Ditto.
9350         (aarch64_crypto_sha512su0qv2di): Ditto.
9351         (aarch64_crypto_sha512su1qv2di): Ditto.
9352         (aarch64_eor3qv8hi): Ditto.
9353         (aarch64_rax1qv2di): Ditto.
9354         (aarch64_xarqv2di): Ditto.
9355         (aarch64_bcaxqv8hi): Ditto.
9356         * config/aarch64/aarch64-simd.md:
9357         (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
9358         (aarch64_crypto_sha512su0qv2di): Ditto.
9359         (aarch64_crypto_sha512su1qv2di): Ditto.
9360         (aarch64_eor3qv8hi): Ditto.
9361         (aarch64_rax1qv2di): Ditto.
9362         (aarch64_xarqv2di): Ditto.
9363         (aarch64_bcaxqv8hi): Ditto.
9364         * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
9365         (vsha512h2q_u64): Ditto.
9366         (vsha512su0q_u64): Ditto.
9367         (vsha512su1q_u64): Ditto.
9368         (veor3q_u16): Ditto.
9369         (vrax1q_u64): Ditto.
9370         (vxarq_u64): Ditto.
9371         (vbcaxq_u16): Ditto.
9372         * config/arm/types.md (crypto_sha512): New type attribute.
9373         (crypto_sha3): Ditto.
9374         (doc/invoke.texi): Document new sha3 option.
9375
9376 2018-01-10  Michael Collison  <michael.collison@arm.com>
9377
9378         * config/aarch64/aarch64-builtins.c:
9379         (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
9380         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
9381         (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
9382         (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
9383         * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
9384         (AARCH64_ISA_SM4): New ISA flag.
9385         (TARGET_SM4): New feature flag for sm4.
9386         * config/aarch64/aarch64-simd-builtins.def
9387         (aarch64_sm3ss1qv4si): Ditto.
9388         (aarch64_sm3tt1aq4si): Ditto.
9389         (aarch64_sm3tt1bq4si): Ditto.
9390         (aarch64_sm3tt2aq4si): Ditto.
9391         (aarch64_sm3tt2bq4si): Ditto.
9392         (aarch64_sm3partw1qv4si): Ditto.
9393         (aarch64_sm3partw2qv4si): Ditto.
9394         (aarch64_sm4eqv4si): Ditto.
9395         (aarch64_sm4ekeyqv4si): Ditto.
9396         * config/aarch64/aarch64-simd.md:
9397         (aarch64_sm3ss1qv4si): Ditto.
9398         (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
9399         (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
9400         (aarch64_sm4eqv4si): Ditto.
9401         (aarch64_sm4ekeyqv4si): Ditto.
9402         * config/aarch64/iterators.md (sm3tt_op): New int iterator.
9403         (sm3part_op): Ditto.
9404         (CRYPTO_SM3TT): Ditto.
9405         (CRYPTO_SM3PART): Ditto.
9406         (UNSPEC_SM3SS1): New unspec.
9407         (UNSPEC_SM3TT1A): Ditto.
9408         (UNSPEC_SM3TT1B): Ditto.
9409         (UNSPEC_SM3TT2A): Ditto.
9410         (UNSPEC_SM3TT2B): Ditto.
9411         (UNSPEC_SM3PARTW1): Ditto.
9412         (UNSPEC_SM3PARTW2): Ditto.
9413         (UNSPEC_SM4E): Ditto.
9414         (UNSPEC_SM4EKEY): Ditto.
9415         * config/aarch64/constraints.md (Ui2): New constraint.
9416         * config/aarch64/predicates.md (aarch64_imm2): New predicate.
9417         * config/arm/types.md (crypto_sm3): New type attribute.
9418         (crypto_sm4): Ditto.
9419         * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
9420         (vsm3tt1aq_u32): Ditto.
9421         (vsm3tt1bq_u32): Ditto.
9422         (vsm3tt2aq_u32): Ditto.
9423         (vsm3tt2bq_u32): Ditto.
9424         (vsm3partw1q_u32): Ditto.
9425         (vsm3partw2q_u32): Ditto.
9426         (vsm4eq_u32): Ditto.
9427         (vsm4ekeyq_u32): Ditto.
9428         (doc/invoke.texi): Document new sm4 option.
9429
9430 2018-01-10  Michael Collison  <michael.collison@arm.com>
9431
9432         * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
9433         * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
9434         (AARCH64_FL_FOR_ARCH8_4): New.
9435         (AARCH64_FL_V8_4): New flag.
9436         (doc/invoke.texi): Document new armv8.4-a option.
9437
9438 2018-01-10  Michael Collison  <michael.collison@arm.com>
9439
9440         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
9441         (__ARM_FEATURE_AES): Define if TARGET_AES is true.
9442         (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
9443         * config/aarch64/aarch64-option-extension.def: Add
9444         AARCH64_OPT_EXTENSION of 'sha2'.
9445         (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
9446         (crypto): Disable sha2 and aes if crypto disabled.
9447         (crypto): Enable aes and sha2 if enabled.
9448         (simd): Disable sha2 and aes if simd disabled.
9449         * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
9450         New flags.
9451         (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
9452         (TARGET_SHA2): New feature flag for sha2.
9453         (TARGET_AES): New feature flag for aes.
9454         * config/aarch64/aarch64-simd.md:
9455         (aarch64_crypto_aes<aes_op>v16qi): Make pattern
9456         conditional on TARGET_AES.
9457         (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
9458         (aarch64_crypto_sha1hsi): Make pattern conditional
9459         on TARGET_SHA2.
9460         (aarch64_crypto_sha1hv4si): Ditto.
9461         (aarch64_be_crypto_sha1hv4si): Ditto.
9462         (aarch64_crypto_sha1su1v4si): Ditto.
9463         (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
9464         (aarch64_crypto_sha1su0v4si): Ditto.
9465         (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
9466         (aarch64_crypto_sha256su0v4si): Ditto.
9467         (aarch64_crypto_sha256su1v4si): Ditto.
9468         (doc/invoke.texi): Document new aes and sha2 options.
9469
9470 2018-01-10  Martin Sebor  <msebor@redhat.com>
9471
9472         PR tree-optimization/83781
9473         * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
9474         as string arrays.
9475
9476 2018-01-11  Martin Sebor  <msebor@gmail.com>
9477             Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
9478
9479         PR tree-optimization/83501
9480         PR tree-optimization/81703
9481
9482         * tree-ssa-strlen.c (get_string_cst): Rename...
9483         (get_string_len): ...to this.  Handle global constants.
9484         (handle_char_store): Adjust.
9485
9486 2018-01-10  Kito Cheng  <kito.cheng@gmail.com>
9487             Jim Wilson  <jimw@sifive.com>
9488
9489         * config/riscv/riscv-protos.h (riscv_output_return): New.
9490         * config/riscv/riscv.c (struct machine_function): New naked_p field.
9491         (riscv_attribute_table, riscv_output_return),
9492         (riscv_handle_fndecl_attribute, riscv_naked_function_p),
9493         (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
9494         (riscv_compute_frame_info): Only compute frame->mask if not a naked
9495         function.
9496         (riscv_expand_prologue): Add early return for naked function.
9497         (riscv_expand_epilogue): Likewise.
9498         (riscv_function_ok_for_sibcall): Return false for naked function.
9499         (riscv_set_current_function): New.
9500         (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
9501         (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
9502         * config/riscv/riscv.md (simple_return): Call riscv_output_return.
9503         * doc/extend.texi (RISC-V Function Attributes): New.
9504
9505 2018-01-10  Michael Meissner  <meissner@linux.vnet.ibm.com>
9506
9507         * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
9508         check for 128-bit long double before checking TCmode.
9509         * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
9510         128-bit long doubles before checking TFmode or TCmode.
9511         (FLOAT128_IBM_P): Likewise.
9512
9513 2018-01-10  Martin Sebor  <msebor@redhat.com>
9514
9515         PR tree-optimization/83671
9516         * builtins.c (c_strlen): Unconditionally return zero for the empty
9517         string.
9518         Use -Warray-bounds for warnings.
9519         * gimple-fold.c (get_range_strlen): Handle non-constant lengths
9520         for non-constant array indices with COMPONENT_REF, arrays of
9521         arrays, and pointers to arrays.
9522         (gimple_fold_builtin_strlen): Determine and set length range for
9523         non-constant character arrays.
9524
9525 2018-01-10  Aldy Hernandez  <aldyh@redhat.com>
9526
9527         PR middle-end/81897
9528         * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
9529         empty blocks.
9530
9531 2018-01-10  Eric Botcazou  <ebotcazou@adacore.com>
9532
9533         * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
9534
9535 2018-01-10  Peter Bergner  <bergner@vnet.ibm.com>
9536
9537         PR target/83399
9538         * config/rs6000/rs6000.c (print_operand) <'y'>: Use
9539         VECTOR_MEM_ALTIVEC_OR_VSX_P.
9540         * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
9541         indexed_or_indirect_operand predicate.
9542         (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
9543         (*vsx_le_perm_load_v8hi): Likewise.
9544         (*vsx_le_perm_load_v16qi): Likewise.
9545         (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
9546         (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
9547         (*vsx_le_perm_store_v8hi): Likewise.
9548         (*vsx_le_perm_store_v16qi): Likewise.
9549         (eight unnamed splitters): Likewise.
9550
9551 2018-01-10  Peter Bergner  <bergner@vnet.ibm.com>
9552
9553         * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
9554         * config/rs6000/emmintrin.h: Likewise.
9555         * config/rs6000/mmintrin.h: Likewise.
9556         * config/rs6000/xmmintrin.h: Likewise.
9557
9558 2018-01-10  David Malcolm  <dmalcolm@redhat.com>
9559
9560         PR c++/43486
9561         * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
9562         "public_flag".
9563         * tree.c (tree_nop_conversion): Return true for location wrapper
9564         nodes.
9565         (maybe_wrap_with_location): New function.
9566         (selftest::check_strip_nops): New function.
9567         (selftest::test_location_wrappers): New function.
9568         (selftest::tree_c_tests): Call it.
9569         * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
9570         (maybe_wrap_with_location): New decl.
9571         (EXPR_LOCATION_WRAPPER_P): New macro.
9572         (location_wrapper_p): New inline function.
9573         (tree_strip_any_location_wrapper): New inline function.
9574
9575 2018-01-10  H.J. Lu  <hongjiu.lu@intel.com>
9576
9577         PR target/83735
9578         * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
9579         stack_realign_offset for the largest alignment of stack slot
9580         actually used.
9581         (ix86_find_max_used_stack_alignment): New function.
9582         (ix86_finalize_stack_frame_flags): Use it.  Set
9583         max_used_stack_alignment if we don't realign stack.
9584         * config/i386/i386.h (machine_function): Add
9585         max_used_stack_alignment.
9586
9587 2018-01-10  Christophe Lyon  <christophe.lyon@linaro.org>
9588
9589         * config/arm/arm.opt (-mbranch-cost): New option.
9590         * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
9591         account.
9592
9593 2018-01-10  Segher Boessenkool  <segher@kernel.crashing.org>
9594
9595         PR target/83629
9596         * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
9597         load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
9598
9599 2018-01-10  Richard Biener  <rguenther@suse.de>
9600
9601         PR debug/83765
9602         * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
9603         early out so it also covers the case where we have a non-NULL
9604         origin.
9605
9606 2018-01-10  Richard Sandiford  <richard.sandiford@linaro.org>
9607
9608         PR tree-optimization/83753
9609         * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
9610         for non-strided grouped accesses if the number of elements is 1.
9611
9612 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
9613
9614         PR target/81616
9615         * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
9616         * i386.h (TARGET_USE_GATHER): Define.
9617         * x86-tune.def (X86_TUNE_USE_GATHER): New.
9618
9619 2018-01-10  Martin Liska  <mliska@suse.cz>
9620
9621         PR bootstrap/82831
9622         * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
9623         * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
9624         partitioning.
9625         * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
9626         CLEANUP_NO_PARTITIONING is not set.
9627
9628 2018-01-10  Richard Sandiford  <richard.sandiford@linaro.org>
9629
9630         * doc/rtl.texi: Remove documentation of (const ...) wrappers
9631         for vectors, as a partial revert of r254296.
9632         * rtl.h (const_vec_p): Delete.
9633         (const_vec_duplicate_p): Don't test for vector CONSTs.
9634         (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
9635         * expmed.c (make_tree): Likewise.
9636
9637         Revert:
9638         * common.md (E, F): Use CONSTANT_P instead of checking for
9639         CONST_VECTOR.
9640         * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
9641         checking for CONST_VECTOR.
9642
9643 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
9644
9645         PR middle-end/83575
9646         * predict.c (force_edge_cold): Handle in more sane way edges
9647         with no prediction.
9648
9649 2018-01-09  Carl Love  <cel@us.ibm.com>
9650
9651         * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
9652         V4SI, V4SF types.
9653         (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
9654         * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
9655         VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
9656         VMRGOW_V2DI, VMRGOW_V2DF.  Remove definition for VMRGOW.
9657         * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
9658         P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW):  Add definitions.
9659         * config/rs6000/rs6000-protos.h: Add extern defition for
9660         rs6000_generate_float2_double_code.
9661         * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
9662         function.
9663         * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
9664         (float2_v2df): Add define_expand.
9665
9666 2018-01-09  Uros Bizjak  <ubizjak@gmail.com>
9667
9668         PR target/83628
9669         * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
9670         op_mode in the force_to_mode call.
9671
9672 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
9673
9674         * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
9675         instead of checking each element individually.
9676         (aarch64_evpc_uzp): Likewise.
9677         (aarch64_evpc_zip): Likewise.
9678         (aarch64_evpc_ext): Likewise.
9679         (aarch64_evpc_rev): Likewise.
9680         (aarch64_evpc_dup): Test the encoding for a single duplicated element,
9681         instead of checking each element individually.  Return true without
9682         generating rtl if
9683         (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
9684         whether all selected elements come from the same input, instead of
9685         checking each element individually.  Remove calls to gen_rtx_REG,
9686         start_sequence and end_sequence and instead assert that no rtl is
9687         generated.
9688
9689 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
9690
9691         * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
9692         order of HIGH and CONST checks.
9693
9694 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
9695
9696         * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
9697         if the destination isn't an SSA_NAME.
9698
9699 2018-01-09  Richard Biener  <rguenther@suse.de>
9700
9701         PR tree-optimization/83668
9702         * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
9703         move prologue...
9704         (canonicalize_loop_form): ... here, renamed from ...
9705         (canonicalize_loop_closed_ssa_form): ... this and amended to
9706         swap successor edges for loop exit blocks to make us use
9707         the RPO order we need for initial schedule generation.
9708
9709 2018-01-09  Joseph Myers  <joseph@codesourcery.com>
9710
9711         PR tree-optimization/64811
9712         * match.pd: When optimizing comparisons with Inf, avoid
9713         introducing or losing exceptions from comparisons with NaN.
9714
9715 2018-01-09  Martin Liska  <mliska@suse.cz>
9716
9717         PR sanitizer/82517
9718         * asan.c (shadow_mem_size): Add gcc_assert.
9719
9720 2018-01-09  Georg-Johann Lay  <avr@gjlay.de>
9721
9722         Don't save registers in main().
9723
9724         PR target/83738
9725         * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
9726         * config/avr/avr.opt (-mmain-is-OS_task): New target option.
9727         * config/avr/avr.c (avr_set_current_function): Don't error if
9728         naked, OS_task or OS_main are specified at the same time.
9729         (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
9730         OS_main.
9731         (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
9732         attribute.
9733         * common/config/avr/avr-common.c (avr_option_optimization_table):
9734         Switch on -mmain-is-OS_task for optimizing compilations.
9735
9736 2018-01-09  Richard Biener  <rguenther@suse.de>
9737
9738         PR tree-optimization/83572
9739         * graphite.c: Include cfganal.h.
9740         (graphite_transform_loops): Connect infinite loops to exit
9741         and remove fake edges at the end.
9742
9743 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
9744
9745         * ipa-inline.c (edge_badness): Revert accidental checkin.
9746
9747 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
9748
9749         PR ipa/80763
9750         * ipa-comdats.c (set_comdat_group): Only set comdat group of real
9751         symbols; not inline clones.
9752
9753 2018-01-09  Jakub Jelinek  <jakub@redhat.com>
9754
9755         PR target/83507
9756         * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
9757         hard registers.  Formatting fixes.
9758
9759         PR preprocessor/83722
9760         * gcc.c (try_generate_repro): Pass
9761         &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
9762         &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
9763         do_report_bug.
9764
9765 2018-01-08  Monk Chiang  <sh.chiang04@gmail.com>
9766             Kito Cheng  <kito.cheng@gmail.com>
9767
9768         * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
9769         (riscv_leaf_function_p): Delete.
9770         (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
9771
9772 2018-01-08  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
9773
9774         * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
9775         function.
9776         (do_ifelse): New function.
9777         (do_isel): New function.
9778         (do_sub3): New function.
9779         (do_add3): New function.
9780         (do_load_mask_compare): New function.
9781         (do_overlap_load_compare): New function.
9782         (expand_compare_loop): New function.
9783         (expand_block_compare): Call expand_compare_loop() when appropriate.
9784         * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
9785         option description.
9786         (-mblock-compare-inline-loop-limit): New option.
9787
9788 2018-01-08  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
9789
9790         PR target/83677
9791         * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
9792         Reverse order of second and third operands in first alternative.
9793         * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
9794         of first and second elements in UNSPEC_VPERMR vector.
9795         (altivec_expand_vec_perm_le): Likewise.
9796
9797 2018-01-08  Jeff Law  <law@redhat.com>
9798
9799         PR rtl-optimizatin/81308
9800         * tree-switch-conversion.c (cfg_altered): New file scoped static.
9801         (process_switch): If group_case_labels makes a change, then set
9802         cfg_altered.
9803         (pass_convert_switch::execute): If a switch is converted, then
9804         set cfg_altered.  Return TODO_cfg_cleanup if cfg_altered is true.
9805
9806         PR rtl-optimization/81308
9807         * recog.c (split_all_insns): Conditionally cleanup the CFG after
9808         splitting insns.
9809
9810 2018-01-08  Vidya Praveen  <vidyapraveen@arm.com>
9811
9812         PR target/83663 - Revert r255946
9813         * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
9814         generation for cases where splatting a value is not useful.
9815         * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
9816         across a vec_duplicate and a paradoxical subreg forming a vector
9817         mode to a vec_concat.
9818
9819 2018-01-08  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
9820
9821         * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
9822         -march=armv8.3-a variants.
9823         * config/arm/t-multilib: Likewise.
9824         * config/arm/t-arm-elf: Likewise.  Handle dotprod extension.
9825
9826 2018-01-08  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
9827
9828         * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
9829         to generate rtl.
9830         (cceq_ior_compare_complement): Give it a name so I can use it, and
9831         change boolean_or_operator predicate to boolean_operator so it can
9832         be used to generate a crand.
9833         (eqne): New code iterator.
9834         (bd/bd_neg): New code_attrs.
9835         (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
9836         a single define_insn.
9837         (<bd>tf_<mode>): A new insn pattern for the conditional form branch
9838         decrement (bdnzt/bdnzf/bdzt/bdzf).
9839         * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
9840         with the new names of the branch decrement patterns, and added the
9841         names of the branch decrement conditional patterns.
9842
9843 2018-01-08  Richard Biener  <rguenther@suse.de>
9844
9845         PR tree-optimization/83563
9846         * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
9847         cache.
9848
9849 2018-01-08  Richard Biener  <rguenther@suse.de>
9850
9851         PR middle-end/83713
9852         * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
9853
9854 2018-01-08  Richard Biener  <rguenther@suse.de>
9855
9856         PR tree-optimization/83685
9857         * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
9858         references to abnormals.
9859
9860 2018-01-08  Richard Biener  <rguenther@suse.de>
9861
9862         PR lto/83719
9863         * dwarf2out.c (output_indirect_strings): Handle empty
9864         skeleton_debug_str_hash.
9865         (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
9866
9867 2018-01-08  Claudiu Zissulescu  <claziss@synopsys.com>
9868
9869         * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
9870         (emit_store_direct): Likewise.
9871         (arc_trampoline_adjust_address): Likewise.
9872         (arc_asm_trampoline_template): New function.
9873         (arc_initialize_trampoline): Use asm_trampoline_template.
9874         (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
9875         * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
9876         * config/arc/arc.md (flush_icache): Delete pattern.
9877
9878 2018-01-08  Claudiu Zissulescu  <claziss@synopsys.com>
9879
9880         * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
9881         * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
9882         munaligned-access.
9883
9884 2018-01-08  Sebastian Huber  <sebastian.huber@embedded-brains.de>
9885
9886         PR target/83681
9887         * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
9888         by not USED_FOR_TARGET.
9889         (make_pass_resolve_sw_modes): Likewise.
9890
9891 2018-01-08  Sebastian Huber  <sebastian.huber@embedded-brains.de>
9892
9893         * config/nios2/nios2.h (nios2_section_threshold): Guard by not
9894         USED_FOR_TARGET.
9895
9896 2018-01-08  Richard Biener  <rguenther@suse.de>
9897
9898         PR middle-end/83580
9899         * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
9900
9901 2018-01-08  Richard Biener  <rguenther@suse.de>
9902
9903         PR middle-end/83517
9904         * match.pd ((t * 2) / 2) -> t): Add missing :c.
9905
9906 2018-01-06  Aldy Hernandez  <aldyh@redhat.com>
9907
9908         PR middle-end/81897
9909         * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
9910         basic blocks with a small number of successors.
9911         (convert_control_dep_chain_into_preds): Improve handling of
9912         forwarder blocks.
9913         (dump_predicates): Split apart into...
9914         (dump_pred_chain): ...here...
9915         (dump_pred_info): ...and here.
9916         (can_one_predicate_be_invalidated_p): Add debugging printfs.
9917         (can_chain_union_be_invalidated_p): Improve check for invalidation
9918         of paths.
9919         (uninit_uses_cannot_happen): Avoid unnecessary if
9920         convert_control_dep_chain_into_preds yielded nothing.
9921
9922 2018-01-06  Martin Sebor  <msebor@redhat.com>
9923
9924         PR tree-optimization/83640
9925         * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
9926         subtracting negative offset from size.
9927         (builtin_access::overlap): Adjust offset bounds of the access to fall
9928         within the size of the object if possible.
9929
9930 2018-01-06  Richard Sandiford  <richard.sandiford@linaro.org>
9931
9932         PR rtl-optimization/83699
9933         * expmed.c (extract_bit_field_1): Restrict the vector usage of
9934         extract_bit_field_as_subreg to cases in which the extracted
9935         value is also a vector.
9936
9937         * lra-constraints.c (process_alt_operands): Test for the equivalence
9938         substitutions when detecting a possible reload cycle.
9939
9940 2018-01-06  Jakub Jelinek  <jakub@redhat.com>
9941
9942         PR debug/83480
9943         * toplev.c (process_options): Don't enable debug_nonbind_markers_p
9944         by default if flag_selective_schedling{,2}.  Formatting fixes.
9945
9946         PR rtl-optimization/83682
9947         * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
9948         if it has non-VECTOR_MODE element mode.
9949         (vec_duplicate_p): Likewise.
9950
9951         PR middle-end/83694
9952         * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
9953         and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
9954
9955 2018-01-05  Jakub Jelinek  <jakub@redhat.com>
9956
9957         PR target/83604
9958         * config/i386/i386-builtin.def
9959         (__builtin_ia32_vgf2p8affineinvqb_v64qi,
9960         __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
9961         Require also OPTION_MASK_ISA_AVX512F in addition to
9962         OPTION_MASK_ISA_GFNI.
9963         (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
9964         __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
9965         OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
9966         to OPTION_MASK_ISA_GFNI.
9967         (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
9968         OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
9969         OPTION_MASK_ISA_AVX512BW.
9970         (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
9971         OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
9972         addition to OPTION_MASK_ISA_GFNI.
9973         (__builtin_ia32_vgf2p8affineinvqb_v16qi,
9974         __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
9975         Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
9976         to OPTION_MASK_ISA_GFNI.
9977         * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
9978         a requirement for all ISAs rather than any of them with a few
9979         exceptions.
9980         (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
9981         processing.
9982         (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
9983         bitmasks to be enabled with 3 exceptions, instead of requiring any
9984         enabled ISA with lots of exceptions.
9985         * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
9986         vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
9987         Change avx512bw in isa attribute to avx512f.
9988         * config/i386/sgxintrin.h: Add license boilerplate.
9989         * config/i386/vaesintrin.h: Likewise.  Fix macro spelling __AVX512F
9990         to __AVX512F__ and __AVX512VL to __AVX512VL__.
9991         (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
9992         _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
9993         defined.
9994         * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
9995         _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
9996         temporarily sse2 rather than sse if not enabled already.
9997
9998         PR target/83604
9999         * config/i386/sse.md (VI248_VLBW): Rename to ...
10000         (VI248_AVX512VL): ... this.  Don't guard V32HI with TARGET_AVX512BW.
10001         (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
10002         vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
10003         vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
10004         vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
10005         mode iterator instead of VI248_VLBW.
10006
10007 2018-01-05  Jan Hubicka  <hubicka@ucw.cz>
10008
10009         * ipa-fnsummary.c (record_modified_bb_info): Add OP.
10010         (record_modified): Skip clobbers; add debug output.
10011         (param_change_prob): Use sreal frequencies.
10012
10013 2018-01-05  Richard Sandiford  <richard.sandiford@linaro.org>
10014
10015         * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
10016         punt for user-aligned variables.
10017
10018 2018-01-05  Richard Sandiford  <richard.sandiford@linaro.org>
10019
10020         * tree-chrec.c (chrec_contains_symbols): Return true for
10021         POLY_INT_CST.
10022
10023 2018-01-05  Sudakshina Das  <sudi.das@arm.com>
10024
10025         PR target/82439
10026         * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
10027         of (x|y) == x for BICS pattern.
10028
10029 2018-01-05  Jakub Jelinek  <jakub@redhat.com>
10030
10031         PR tree-optimization/83605
10032         * gimple-ssa-strength-reduction.c: Include tree-eh.h.
10033         (find_candidates_dom_walker::before_dom_children): Ignore stmts that
10034         can throw.
10035
10036 2018-01-05  Sebastian Huber  <sebastian.huber@embedded-brains.de>
10037
10038         * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
10039         * config/epiphany/rtems.h: New file.
10040
10041 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
10042             Uros Bizjak  <ubizjak@gmail.com>
10043
10044         PR target/83554
10045         * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
10046         QIreg_operand instead of register_operand predicate.
10047         * config/i386/i386.c (ix86_rop_should_change_byte_p,
10048         set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
10049         comments instead of -fmitigate[-_]rop.
10050
10051 2018-01-04  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
10052
10053         PR bootstrap/81926
10054         * cgraphunit.c (symbol_table::compile): Switch to text_section
10055         before calling assembly_start debug hook.
10056         * run-rtl-passes.c (run_rtl_passes): Likewise.
10057         Include output.h.
10058
10059 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
10060
10061         * tree-vrp.c (extract_range_from_binary_expr_1): Check
10062         range_int_cst_p rather than !symbolic_range_p before calling
10063         extract_range_from_multiplicative_op_1.
10064
10065 2018-01-04  Jeff Law  <law@redhat.com>
10066
10067         * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
10068         redundant test in assertion.
10069
10070 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
10071
10072         * doc/rtl.texi: Document machine_mode wrapper classes.
10073
10074 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
10075
10076         * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
10077         using tree_to_uhwi.
10078
10079 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
10080
10081         * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
10082         the VEC_PERM_EXPR fold to fail.
10083
10084 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
10085
10086         PR debug/83585
10087         * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
10088         to switched_sections.
10089
10090 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
10091
10092         PR target/83680
10093         * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
10094         test for d.testing.
10095
10096 2018-01-04  Peter Bergner  <bergner@vnet.ibm.com>
10097
10098         PR target/83387
10099         * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
10100         allow arguments in FP registers if TARGET_HARD_FLOAT is false.
10101
10102 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
10103
10104         PR debug/83666
10105         * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
10106         is BLKmode and bitpos not zero or mode change is needed.
10107
10108 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
10109
10110         PR target/83675
10111         * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
10112         TARGET_VIS2.
10113
10114 2018-01-04  Uros Bizjak  <ubizjak@gmail.com>
10115
10116         PR target/83628
10117         * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
10118         instead of MULT rtx.  Update all corresponding splitters.
10119         (*saddl_se): Ditto.
10120         (*ssub<modesuffix>): Ditto.
10121         (*ssubl_se): Ditto.
10122         (*cmp_sadd_di): Update split patterns.
10123         (*cmp_sadd_si): Ditto.
10124         (*cmp_sadd_sidi): Ditto.
10125         (*cmp_ssub_di): Ditto.
10126         (*cmp_ssub_si): Ditto.
10127         (*cmp_ssub_sidi): Ditto.
10128         * config/alpha/predicates.md (const23_operand): New predicate.
10129         * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
10130         Look for ASHIFT, not MULT inner operand.
10131         (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
10132
10133 2018-01-04  Martin Liska  <mliska@suse.cz>
10134
10135         PR gcov-profile/83669
10136         * gcov.c (output_intermediate_file): Add version to intermediate
10137         gcov file.
10138         * doc/gcov.texi: Document new field 'version' in intermediate
10139         file format. Fix location of '-k' option of gcov command.
10140
10141 2018-01-04  Martin Liska  <mliska@suse.cz>
10142
10143         PR ipa/82352
10144         * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
10145
10146 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
10147
10148         * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
10149
10150 2018-01-03  Martin Sebor  <msebor@redhat.com>
10151
10152         PR tree-optimization/83655
10153         * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
10154         checking calls with invalid arguments.
10155
10156 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10157
10158         * tree-vect-stmts.c (vect_get_store_rhs): New function.
10159         (vectorizable_mask_load_store): Delete.
10160         (vectorizable_call): Return false for masked loads and stores.
10161         (vectorizable_store): Handle IFN_MASK_STORE.  Use vect_get_store_rhs
10162         instead of gimple_assign_rhs1.
10163         (vectorizable_load): Handle IFN_MASK_LOAD.
10164         (vect_transform_stmt): Don't set is_store for call_vec_info_type.
10165
10166 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10167
10168         * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
10169         split out from..,
10170         (vectorizable_mask_load_store): ...here.
10171         (vectorizable_load): ...and here.
10172
10173 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10174
10175         * tree-vect-stmts.c (vect_build_all_ones_mask)
10176         (vect_build_zero_merge_argument): New functions, split out from...
10177         (vectorizable_load): ...here.
10178
10179 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10180
10181         * tree-vect-stmts.c (vect_check_store_rhs): New function,
10182         split out from...
10183         (vectorizable_mask_load_store): ...here.
10184         (vectorizable_store): ...and here.
10185
10186 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10187
10188         * tree-vect-stmts.c (vect_check_load_store_mask): New function,
10189         split out from...
10190         (vectorizable_mask_load_store): ...here.
10191
10192 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10193
10194         * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
10195         (vect_model_store_cost): Take a vec_load_store_type instead of a
10196         vect_def_type.
10197         * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
10198         (vect_model_store_cost): Take a vec_load_store_type instead of a
10199         vect_def_type.
10200         (vectorizable_mask_load_store): Update accordingly.
10201         (vectorizable_store): Likewise.
10202         * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
10203
10204 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10205
10206         * tree-vect-loop.c (vect_transform_loop): Stub out scalar
10207         IFN_MASK_LOAD calls here rather than...
10208         * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
10209
10210 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10211             Alan Hayward  <alan.hayward@arm.com>
10212             David Sherwood  <david.sherwood@arm.com>
10213
10214         * expmed.c (extract_bit_field_1): For vector extracts,
10215         fall back to extract_bit_field_as_subreg if vec_extract
10216         isn't available.
10217
10218 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10219             Alan Hayward  <alan.hayward@arm.com>
10220             David Sherwood  <david.sherwood@arm.com>
10221
10222         * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
10223         they are variable or constant sized.
10224         (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
10225         slots for constant-sized data.
10226
10227 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10228             Alan Hayward  <alan.hayward@arm.com>
10229             David Sherwood  <david.sherwood@arm.com>
10230
10231         * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
10232         handling COND_EXPRs with boolean comparisons, try to find a better
10233         basis for the mask type than the boolean itself.
10234
10235 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10236
10237         * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
10238         is calculated and how it can be overridden.
10239         * genmodes.c (max_bitsize_mode_any_mode): New variable.
10240         (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
10241         if defined.
10242         (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
10243         if nonzero.
10244
10245 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10246             Alan Hayward  <alan.hayward@arm.com>
10247             David Sherwood  <david.sherwood@arm.com>
10248
10249         * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
10250         Remove the mode argument.
10251         (aarch64_simd_valid_immediate): Remove the mode and inverse
10252         arguments.
10253         * config/aarch64/iterators.md (bitsize): New iterator.
10254         * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
10255         (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
10256         * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
10257         aarch64_simd_valid_immediate.
10258         * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
10259         (aarch64_reg_or_bic_imm): Likewise.
10260         * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
10261         with an insn_type enum and msl with a modifier_type enum.
10262         Replace element_width with a scalar_mode.  Change the shift
10263         to unsigned int.  Add constructors for scalar_float_mode and
10264         scalar_int_mode elements.
10265         (aarch64_vect_float_const_representable_p): Delete.
10266         (aarch64_can_const_movi_rtx_p)
10267         (aarch64_simd_scalar_immediate_valid_for_move)
10268         (aarch64_simd_make_constant): Update call to
10269         aarch64_simd_valid_immediate.
10270         (aarch64_advsimd_valid_immediate_hs): New function.
10271         (aarch64_advsimd_valid_immediate): Likewise.
10272         (aarch64_simd_valid_immediate): Remove mode and inverse
10273         arguments.  Rewrite to use the above.  Use const_vec_duplicate_p
10274         to detect duplicated constants and use aarch64_float_const_zero_rtx_p
10275         and aarch64_float_const_representable_p on the result.
10276         (aarch64_output_simd_mov_immediate): Remove mode argument.
10277         Update call to aarch64_simd_valid_immediate and use of
10278         simd_immediate_info.
10279         (aarch64_output_scalar_simd_mov_immediate): Update call
10280         accordingly.
10281
10282 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10283             Alan Hayward  <alan.hayward@arm.com>
10284             David Sherwood  <david.sherwood@arm.com>
10285
10286         * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
10287         (mode_nunits): Likewise CONST_MODE_NUNITS.
10288         * machmode.def (ADJUST_NUNITS): Document.
10289         * genmodes.c (mode_data::need_nunits_adj): New field.
10290         (blank_mode): Update accordingly.
10291         (adj_nunits): New variable.
10292         (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
10293         parameter.
10294         (emit_mode_size_inline): Set need_bytesize_adj for all modes
10295         listed in adj_nunits.
10296         (emit_mode_nunits_inline): Set need_nunits_adj for all modes
10297         listed in adj_nunits.  Don't emit case statements for such modes.
10298         (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
10299         and CONST_MODE_PRECISION.  Make CONST_MODE_SIZE expand to
10300         nothing if adj_nunits is nonnull.
10301         (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
10302         (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
10303         (emit_mode_fbit): Update use of print_maybe_const_decl.
10304         (emit_move_size): Likewise.  Treat the array as non-const
10305         if adj_nunits.
10306         (emit_mode_adjustments): Handle adj_nunits.
10307
10308 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10309
10310         * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
10311         * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
10312         (VECTOR_MODES): Use it.
10313         (make_vector_modes): Take the prefix as an argument.
10314
10315 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10316             Alan Hayward  <alan.hayward@arm.com>
10317             David Sherwood  <david.sherwood@arm.com>
10318
10319         * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
10320         * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
10321         for MODE_VECTOR_BOOL.
10322         * machmode.def (VECTOR_BOOL_MODE): Document.
10323         * genmodes.c (VECTOR_BOOL_MODE): New macro.
10324         (make_vector_bool_mode): New function.
10325         (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
10326         MODE_VECTOR_BOOL.
10327         * lto-streamer-in.c (lto_input_mode_table): Likewise.
10328         * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
10329         Likewise.
10330         * stor-layout.c (int_mode_for_mode): Likewise.
10331         * tree.c (build_vector_type_for_mode): Likewise.
10332         * varasm.c (output_constant_pool_2): Likewise.
10333         * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
10334         CONSTM1_RTX (BImode) are the same thing.  Initialize const_tiny_rtx
10335         for MODE_VECTOR_BOOL.
10336         * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
10337         of mode class checks.
10338         * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
10339         instead of a list of mode class checks.
10340         (expand_vector_scalar_condition): Likewise.
10341         (type_for_widest_vector_mode): Handle BImode as an inner mode.
10342
10343 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10344             Alan Hayward  <alan.hayward@arm.com>
10345             David Sherwood  <david.sherwood@arm.com>
10346
10347         * machmode.h (mode_size): Change from unsigned short to
10348         poly_uint16_pod.
10349         (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
10350         (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
10351         or if measurement_type is not polynomial.
10352         (fixed_size_mode::includes_p): Check for constant-sized modes.
10353         * genmodes.c (emit_mode_size_inline): Make mode_size_inline
10354         return a poly_uint16 rather than an unsigned short.
10355         (emit_mode_size): Change the type of mode_size from unsigned short
10356         to poly_uint16_pod.  Use ZERO_COEFFS for the initializer.
10357         (emit_mode_adjustments): Cope with polynomial vector sizes.
10358         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
10359         for GET_MODE_SIZE.
10360         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
10361         for GET_MODE_SIZE.
10362         * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
10363         * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
10364         * caller-save.c (setup_save_areas): Likewise.
10365         (replace_reg_with_saved_mem): Likewise.
10366         * calls.c (emit_library_call_value_1): Likewise.
10367         * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
10368         * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
10369         (gen_lowpart_for_combine): Likewise.
10370         * convert.c (convert_to_integer_1): Likewise.
10371         * cse.c (equiv_constant, cse_insn): Likewise.
10372         * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
10373         (cselib_subst_to_values): Likewise.
10374         * dce.c (word_dce_process_block): Likewise.
10375         * df-problems.c (df_word_lr_mark_ref): Likewise.
10376         * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
10377         * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
10378         (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
10379         (rtl_for_decl_location): Likewise.
10380         * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
10381         * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
10382         * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
10383         (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
10384         (expand_expr_real_1): Likewise.
10385         * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
10386         (pad_below): Likewise.
10387         * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
10388         * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
10389         * ira.c (get_subreg_tracking_sizes): Likewise.
10390         * ira-build.c (ira_create_allocno_objects): Likewise.
10391         * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
10392         (ira_sort_regnos_for_alter_reg): Likewise.
10393         * ira-costs.c (record_operand_costs): Likewise.
10394         * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
10395         (resolve_simple_move): Likewise.
10396         * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
10397         (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
10398         (lra_constraints): Likewise.
10399         (CONST_POOL_OK_P): Reject variable-sized modes.
10400         * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
10401         (add_pseudo_to_slot, lra_spill): Likewise.
10402         * omp-low.c (omp_clause_aligned_alignment): Likewise.
10403         * optabs-query.c (get_best_extraction_insn): Likewise.
10404         * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
10405         * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
10406         (expand_mult_highpart, valid_multiword_target_p): Likewise.
10407         * recog.c (offsettable_address_addr_space_p): Likewise.
10408         * regcprop.c (maybe_mode_change): Likewise.
10409         * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
10410         * regrename.c (build_def_use): Likewise.
10411         * regstat.c (dump_reg_info): Likewise.
10412         * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
10413         (find_reloads, find_reloads_subreg_address): Likewise.
10414         * reload1.c (eliminate_regs_1): Likewise.
10415         * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
10416         * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
10417         (simplify_binary_operation_1, simplify_subreg): Likewise.
10418         * targhooks.c (default_function_arg_padding): Likewise.
10419         (default_hard_regno_nregs, default_class_max_nregs): Likewise.
10420         * tree-cfg.c (verify_gimple_assign_binary): Likewise.
10421         (verify_gimple_assign_ternary): Likewise.
10422         * tree-inline.c (estimate_move_cost): Likewise.
10423         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
10424         * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
10425         (get_address_cost_ainc): Likewise.
10426         * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
10427         (vect_supportable_dr_alignment): Likewise.
10428         * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
10429         (vectorizable_reduction): Likewise.
10430         * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
10431         (vectorizable_operation, vectorizable_load): Likewise.
10432         * tree.c (build_same_sized_truth_vector_type): Likewise.
10433         * valtrack.c (cleanup_auto_inc_dec): Likewise.
10434         * var-tracking.c (emit_note_insn_var_location): Likewise.
10435         * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
10436         (ADDR_VEC_ALIGN): Likewise.
10437
10438 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10439             Alan Hayward  <alan.hayward@arm.com>
10440             David Sherwood  <david.sherwood@arm.com>
10441
10442         * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
10443         unsigned short.
10444         (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
10445         or if measurement_type is polynomial.
10446         * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
10447         * combine.c (make_extraction): Likewise.
10448         * dse.c (find_shift_sequence): Likewise.
10449         * dwarf2out.c (mem_loc_descriptor): Likewise.
10450         * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
10451         (extract_bit_field, extract_low_bits): Likewise.
10452         * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
10453         (optimize_bitfield_assignment_op, expand_assignment): Likewise.
10454         (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
10455         * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
10456         * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
10457         * reload.c (find_reloads): Likewise.
10458         * reload1.c (alter_reg): Likewise.
10459         * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
10460         * targhooks.c (default_secondary_memory_needed_mode): Likewise.
10461         * tree-if-conv.c (predicate_mem_writes): Likewise.
10462         * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
10463         * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
10464         * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
10465         * valtrack.c (dead_debug_insert_temp): Likewise.
10466         * varasm.c (mergeable_constant_section): Likewise.
10467         * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
10468
10469 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10470             Alan Hayward  <alan.hayward@arm.com>
10471             David Sherwood  <david.sherwood@arm.com>
10472
10473         * expr.c (expand_assignment): Cope with polynomial mode sizes
10474         when assigning to a CONCAT.
10475
10476 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10477             Alan Hayward  <alan.hayward@arm.com>
10478             David Sherwood  <david.sherwood@arm.com>
10479
10480         * machmode.h (mode_precision): Change from unsigned short to
10481         poly_uint16_pod.
10482         (mode_to_precision): Return a poly_uint16 rather than an unsigned
10483         short.
10484         (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
10485         or if measurement_type is not polynomial.
10486         (HWI_COMPUTABLE_MODE_P): Turn into a function.  Optimize the case
10487         in which the mode is already known to be a scalar_int_mode.
10488         * genmodes.c (emit_mode_precision): Change the type of mode_precision
10489         from unsigned short to poly_uint16_pod.  Use ZERO_COEFFS for the
10490         initializer.
10491         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
10492         for GET_MODE_PRECISION.
10493         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
10494         for GET_MODE_PRECISION.
10495         * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
10496         as polynomial.
10497         (try_combine, find_split_point, combine_simplify_rtx): Likewise.
10498         (expand_field_assignment, make_extraction): Likewise.
10499         (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
10500         (get_last_value): Likewise.
10501         * convert.c (convert_to_integer_1): Likewise.
10502         * cse.c (cse_insn): Likewise.
10503         * expr.c (expand_expr_real_1): Likewise.
10504         * lra-constraints.c (simplify_operand_subreg): Likewise.
10505         * optabs-query.c (can_atomic_load_p): Likewise.
10506         * optabs.c (expand_atomic_load): Likewise.
10507         (expand_atomic_store): Likewise.
10508         * ree.c (combine_reaching_defs): Likewise.
10509         * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
10510         * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
10511         * tree.h (type_has_mode_precision_p): Likewise.
10512         * ubsan.c (instrument_si_overflow): Likewise.
10513
10514 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10515             Alan Hayward  <alan.hayward@arm.com>
10516             David Sherwood  <david.sherwood@arm.com>
10517
10518         * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
10519         polynomial numbers of units.
10520         (SET_TYPE_VECTOR_SUBPARTS): Likewise.
10521         (valid_vector_subparts_p): New function.
10522         (build_vector_type): Remove temporary shim and take the number
10523         of units as a poly_uint64 rather than an int.
10524         (build_opaque_vector_type): Take the number of units as a
10525         poly_uint64 rather than an int.
10526         * tree.c (build_vector_from_ctor): Handle polynomial
10527         TYPE_VECTOR_SUBPARTS.
10528         (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
10529         (uniform_vector_p, vector_type_mode, build_vector): Likewise.
10530         (build_vector_from_val): If the number of units is variable,
10531         use build_vec_duplicate_cst for constant operands and
10532         VEC_DUPLICATE_EXPR otherwise.
10533         (make_vector_type): Remove temporary is_constant ().
10534         (build_vector_type, build_opaque_vector_type): Take the number of
10535         units as a poly_uint64 rather than an int.
10536         (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
10537         VECTOR_CST_NELTS.
10538         * cfgexpand.c (expand_debug_expr): Likewise.
10539         * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
10540         (store_constructor, expand_expr_real_1): Likewise.
10541         (const_scalar_mask_from_tree): Likewise.
10542         * fold-const-call.c (fold_const_reduction): Likewise.
10543         * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
10544         (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
10545         (native_encode_vector, vec_cst_ctor_to_array): Likewise.
10546         (fold_relational_const): Likewise.
10547         (native_interpret_vector): Likewise.  Change the size from an
10548         int to an unsigned int.
10549         * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
10550         TYPE_VECTOR_SUBPARTS.
10551         (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
10552         (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
10553         duplicating a non-constant operand into a variable-length vector.
10554         * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
10555         TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
10556         * ipa-icf.c (sem_variable::equals): Likewise.
10557         * match.pd: Likewise.
10558         * omp-simd-clone.c (simd_clone_subparts): Likewise.
10559         * print-tree.c (print_node): Likewise.
10560         * stor-layout.c (layout_type): Likewise.
10561         * targhooks.c (default_builtin_vectorization_cost): Likewise.
10562         * tree-cfg.c (verify_gimple_comparison): Likewise.
10563         (verify_gimple_assign_binary): Likewise.
10564         (verify_gimple_assign_ternary): Likewise.
10565         (verify_gimple_assign_single): Likewise.
10566         * tree-pretty-print.c (dump_generic_node): Likewise.
10567         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
10568         (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
10569         * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
10570         (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
10571         (vect_shift_permute_load_chain): Likewise.
10572         * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
10573         (expand_vector_condition, optimize_vector_constructor): Likewise.
10574         (lower_vec_perm, get_compute_type): Likewise.
10575         * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
10576         (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
10577         * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
10578         (vect_recog_mask_conversion_pattern): Likewise.
10579         * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
10580         (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
10581         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
10582         (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
10583         (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
10584         (vectorizable_shift, vectorizable_operation, vectorizable_store)
10585         (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
10586         (supportable_widening_operation): Likewise.
10587         (supportable_narrowing_operation): Likewise.
10588         * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
10589         Likewise.
10590         * varasm.c (output_constant): Likewise.
10591
10592 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10593             Alan Hayward  <alan.hayward@arm.com>
10594             David Sherwood  <david.sherwood@arm.com>
10595
10596         * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
10597         so that both the length == 3 and length != 3 cases set up their
10598         own permute vectors.  Add comments explaining why we know the
10599         number of elements is constant.
10600         (vect_permute_load_chain): Likewise.
10601
10602 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10603             Alan Hayward  <alan.hayward@arm.com>
10604             David Sherwood  <david.sherwood@arm.com>
10605
10606         * machmode.h (mode_nunits): Change from unsigned char to
10607         poly_uint16_pod.
10608         (ONLY_FIXED_SIZE_MODES): New macro.
10609         (pod_mode::measurement_type, scalar_int_mode::measurement_type)
10610         (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
10611         (complex_mode::measurement_type, fixed_size_mode::measurement_type):
10612         New typedefs.
10613         (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
10614         (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
10615         or if measurement_type is not polynomial.
10616         * genmodes.c (ZERO_COEFFS): New macro.
10617         (emit_mode_nunits_inline): Make mode_nunits_inline return a
10618         poly_uint16.
10619         (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
10620         Use ZERO_COEFFS when emitting initializers.
10621         * data-streamer.h (bp_pack_poly_value): New function.
10622         (bp_unpack_poly_value): Likewise.
10623         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
10624         for GET_MODE_NUNITS.
10625         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
10626         for GET_MODE_NUNITS.
10627         * tree.c (make_vector_type): Remove temporary shim and make
10628         the real function take the number of units as a poly_uint64
10629         rather than an int.
10630         (build_vector_type_for_mode): Handle polynomial nunits.
10631         * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
10632         * emit-rtl.c (const_vec_series_p_1): Likewise.
10633         (gen_rtx_CONST_VECTOR): Likewise.
10634         * fold-const.c (test_vec_duplicate_folding): Likewise.
10635         * genrecog.c (validate_pattern): Likewise.
10636         * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
10637         * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
10638         * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
10639         (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
10640         (expand_vec_cond_expr, expand_mult_highpart): Likewise.
10641         * rtlanal.c (subreg_get_info): Likewise.
10642         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
10643         (vect_grouped_load_supported): Likewise.
10644         * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
10645         * tree-vect-loop.c (have_whole_vector_shift): Likewise.
10646         * simplify-rtx.c (simplify_unary_operation_1): Likewise.
10647         (simplify_const_unary_operation, simplify_binary_operation_1)
10648         (simplify_const_binary_operation, simplify_ternary_operation)
10649         (test_vector_ops_duplicate, test_vector_ops): Likewise.
10650         (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
10651         instead of CONST_VECTOR_NUNITS.
10652         * varasm.c (output_constant_pool_2): Likewise.
10653         * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
10654         explicit-encoded elements in the XVEC for variable-length vectors.
10655
10656 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10657
10658         * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
10659
10660 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10661             Alan Hayward  <alan.hayward@arm.com>
10662             David Sherwood  <david.sherwood@arm.com>
10663
10664         * coretypes.h (fixed_size_mode): Declare.
10665         (fixed_size_mode_pod): New typedef.
10666         * builtins.h (target_builtins::x_apply_args_mode)
10667         (target_builtins::x_apply_result_mode): Change type to
10668         fixed_size_mode_pod.
10669         * builtins.c (apply_args_size, apply_result_size, result_vector)
10670         (expand_builtin_apply_args_1, expand_builtin_apply)
10671         (expand_builtin_return): Update accordingly.
10672
10673 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10674
10675         * cse.c (hash_rtx_cb): Hash only the encoded elements.
10676         * cselib.c (cselib_hash_rtx): Likewise.
10677         * expmed.c (make_tree): Build VECTOR_CSTs directly from the
10678         CONST_VECTOR encoding.
10679
10680 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
10681             Jeff Law  <law@redhat.com>
10682
10683         PR target/83641
10684         * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
10685         noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
10686         only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
10687         and add REG_CFA_ADJUST_CFA notes in that case to both insns.
10688
10689         PR target/83641
10690         * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
10691         explicitly probe *sp in a noreturn function if there were any callee
10692         register saves or frame pointer is needed.
10693
10694 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
10695
10696         PR debug/83621
10697         * cfgexpand.c (expand_debug_expr): Return NULL if mode is
10698         BLKmode for ternary, binary or unary expressions.
10699
10700         PR debug/83645
10701         * var-tracking.c (delete_vta_debug_insn): New inline function.
10702         (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
10703         insns from get_insns () to NULL instead of each bb separately.
10704         Use delete_vta_debug_insn.  No longer static.
10705         (vt_debug_insns_local, variable_tracking_main_1): Adjust
10706         delete_vta_debug_insns callers.
10707         * rtl.h (delete_vta_debug_insns): Declare.
10708         * final.c (rest_of_handle_final): Call delete_vta_debug_insns
10709         instead of variable_tracking_main.
10710
10711 2018-01-03  Martin Sebor  <msebor@redhat.com>
10712
10713         PR tree-optimization/83603
10714         * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
10715         arguments past the endof the argument list in functions declared
10716         without a prototype.
10717         * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
10718         Avoid checking when arguments are null.
10719
10720 2018-01-03  Martin Sebor  <msebor@redhat.com>
10721
10722         PR c/83559
10723         * doc/extend.texi (attribute const): Fix a typo.
10724         * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
10725         issuing -Wsuggest-attribute for void functions.
10726
10727 2018-01-03  Martin Sebor  <msebor@redhat.com>
10728
10729         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
10730         offset_int::from instead of wide_int::to_shwi.
10731         (maybe_diag_overlap): Remove assertion.
10732         Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
10733         * gimple-ssa-sprintf.c (format_directive): Same.
10734         (parse_directive): Same.
10735         (sprintf_dom_walker::compute_format_length): Same.
10736         (try_substitute_return_value): Same.
10737
10738 2018-01-03  Jeff Law  <law@redhat.com>
10739
10740         PR middle-end/83654
10741         * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
10742         non-constant residual for zero at runtime and avoid probing in
10743         that case.  Reorganize code for trailing problem to mirror handling
10744         of the residual.
10745
10746 2018-01-03  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
10747
10748         PR tree-optimization/83501
10749         * tree-ssa-strlen.c (get_string_cst): New.
10750         (handle_char_store): Call get_string_cst.
10751
10752 2018-01-03  Martin Liska  <mliska@suse.cz>
10753
10754         PR tree-optimization/83593
10755         * tree-ssa-strlen.c: Include tree-cfg.h.
10756         (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
10757         (strlen_dom_walker): Add new member variable m_cleanup_cfg.
10758         (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
10759         to false.
10760         (strlen_dom_walker::before_dom_children): Call
10761         gimple_purge_dead_eh_edges. Dump tranformation with details
10762         dump flags.
10763         (strlen_dom_walker::before_dom_children): Update call by adding
10764         new argument cleanup_eh.
10765         (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
10766
10767 2018-01-03  Martin Liska  <mliska@suse.cz>
10768
10769         PR ipa/83549
10770         * cif-code.def (VARIADIC_THUNK): New enum value.
10771         * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
10772         thunks.
10773
10774 2018-01-03  Jan Beulich  <jbeulich@suse.com>
10775
10776         * sse.md (mov<mode>_internal): Tighten condition for when to use
10777         vmovdqu<ssescalarsize> for TI and OI modes.
10778
10779 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
10780
10781         Update copyright years.
10782
10783 2018-01-03  Martin Liska  <mliska@suse.cz>
10784
10785         PR ipa/83594
10786         * ipa-visibility.c (function_and_variable_visibility): Skip
10787         functions with noipa attribure.
10788
10789 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
10790
10791         * gcc.c (process_command): Update copyright notice dates.
10792         * gcov-dump.c (print_version): Ditto.
10793         * gcov.c (print_version): Ditto.
10794         * gcov-tool.c (print_version): Ditto.
10795         * gengtype.c (create_file): Ditto.
10796         * doc/cpp.texi: Bump @copying's copyright year.
10797         * doc/cppinternals.texi: Ditto.
10798         * doc/gcc.texi: Ditto.
10799         * doc/gccint.texi: Ditto.
10800         * doc/gcov.texi: Ditto.
10801         * doc/install.texi: Ditto.
10802         * doc/invoke.texi: Ditto.
10803
10804 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10805
10806         * vector-builder.h (vector_builder::m_full_nelts): Change from
10807         unsigned int to poly_uint64.
10808         (vector_builder::full_nelts): Update prototype accordingly.
10809         (vector_builder::new_vector): Likewise.
10810         (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
10811         (vector_builder::operator ==): Likewise.
10812         (vector_builder::finalize): Likewise.
10813         * int-vector-builder.h (int_vector_builder::int_vector_builder):
10814         Take the number of elements as a poly_uint64 rather than an
10815         unsigned int.
10816         * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
10817         from unsigned int to poly_uint64.
10818         (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
10819         (vec_perm_indices::new_vector): Likewise.
10820         (vec_perm_indices::length): Likewise.
10821         (vec_perm_indices::nelts_per_input): Likewise.
10822         (vec_perm_indices::input_nelts): Likewise.
10823         * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
10824         number of elements per input as a poly_uint64 rather than an
10825         unsigned int.  Use the original encoding for variable-length
10826         vectors, rather than clamping each individual element.
10827         For the second and subsequent elements in each pattern,
10828         clamp the step and base before clamping their sum.
10829         (vec_perm_indices::series_p): Handle polynomial element counts.
10830         (vec_perm_indices::all_in_range_p): Likewise.
10831         (vec_perm_indices_to_tree): Likewise.
10832         (vec_perm_indices_to_rtx): Likewise.
10833         * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
10834         * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
10835         (tree_vector_builder::new_binary_operation): Handle polynomial
10836         element counts.  Return false if we need to know the number
10837         of elements at compile time.
10838         * fold-const.c (fold_vec_perm): Punt if the number of elements
10839         isn't known at compile time.
10840
10841 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10842
10843         * vec-perm-indices.h (vec_perm_builder): Change element type
10844         from HOST_WIDE_INT to poly_int64.
10845         (vec_perm_indices::element_type): Update accordingly.
10846         (vec_perm_indices::clamp): Handle polynomial element_types.
10847         * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
10848         (vec_perm_indices::all_in_range_p): Likewise.
10849         (tree_to_vec_perm_builder): Check for poly_int64 trees rather
10850         than shwi trees.
10851         * vector-builder.h (vector_builder::stepped_sequence_p): Handle
10852         polynomial vec_perm_indices element types.
10853         * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
10854         * fold-const.c (fold_vec_perm): Likewise.
10855         * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
10856         * tree-vect-generic.c (lower_vec_perm): Likewise.
10857         * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
10858         * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
10859         element type to HOST_WIDE_INT.
10860
10861 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10862             Alan Hayward  <alan.hayward@arm.com>
10863             David Sherwood  <david.sherwood@arm.com>
10864
10865         * alias.c (addr_side_effect_eval): Take the size as a poly_int64
10866         rather than an int.  Use plus_constant.
10867         (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
10868         Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
10869
10870 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10871             Alan Hayward  <alan.hayward@arm.com>
10872             David Sherwood  <david.sherwood@arm.com>
10873
10874         * calls.c (emit_call_1, expand_call): Change struct_value_size from
10875         a HOST_WIDE_INT to a poly_int64.
10876
10877 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10878             Alan Hayward  <alan.hayward@arm.com>
10879             David Sherwood  <david.sherwood@arm.com>
10880
10881         * calls.c (load_register_parameters): Cope with polynomial
10882         mode sizes.  Require a constant size for BLKmode parameters
10883         that aren't described by a PARALLEL.  If BLOCK_REG_PADDING
10884         forces a parameter to be padded at the lsb end in order to
10885         fill a complete number of words, require the parameter size
10886         to be ordered wrt UNITS_PER_WORD.
10887
10888 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10889             Alan Hayward  <alan.hayward@arm.com>
10890             David Sherwood  <david.sherwood@arm.com>
10891
10892         * reload1.c (spill_stack_slot_width): Change element type
10893         from unsigned int to poly_uint64_pod.
10894         (alter_reg): Treat mode sizes as polynomial.
10895
10896 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10897             Alan Hayward  <alan.hayward@arm.com>
10898             David Sherwood  <david.sherwood@arm.com>
10899
10900         * reload.c (complex_word_subreg_p): New function.
10901         (reload_inner_reg_of_subreg, push_reload): Use it.
10902
10903 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10904             Alan Hayward  <alan.hayward@arm.com>
10905             David Sherwood  <david.sherwood@arm.com>
10906
10907         * lra-constraints.c (process_alt_operands): Reject matched
10908         operands whose sizes aren't ordered.
10909         (match_reload): Refer to this check here.
10910
10911 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10912             Alan Hayward  <alan.hayward@arm.com>
10913             David Sherwood  <david.sherwood@arm.com>
10914
10915         * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
10916         that the mode size is in the set {1, 2, 4, 8, 16}.
10917
10918 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10919             Alan Hayward  <alan.hayward@arm.com>
10920             David Sherwood  <david.sherwood@arm.com>
10921
10922         * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
10923         Use plus_constant instead of gen_rtx_PLUS.
10924
10925 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10926             Alan Hayward  <alan.hayward@arm.com>
10927             David Sherwood  <david.sherwood@arm.com>
10928
10929         * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
10930         * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
10931         * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
10932         * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
10933         * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
10934         * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
10935         * config/i386/i386-protos.h (ix86_push_rounding): Declare.
10936         * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
10937         * config/i386/i386.c (ix86_push_rounding): ...this new function.
10938         * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
10939         a poly_int64.
10940         * config/m32c/m32c.c (m32c_push_rounding): Likewise.
10941         * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
10942         * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
10943         * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
10944         * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
10945         * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
10946         * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
10947         * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
10948         * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
10949         * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
10950         function.
10951         * expr.c (emit_move_resolve_push): Treat the input and result
10952         of PUSH_ROUNDING as a poly_int64.
10953         (emit_move_complex_push, emit_single_push_insn_1): Likewise.
10954         (emit_push_insn): Likewise.
10955         * lra-eliminations.c (mark_not_eliminable): Likewise.
10956         * recog.c (push_operand): Likewise.
10957         * reload1.c (elimination_effects): Likewise.
10958         * rtlanal.c (nonzero_bits1): Likewise.
10959         * calls.c (store_one_arg): Likewise.  Require the padding to be
10960         known at compile time.
10961
10962 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10963             Alan Hayward  <alan.hayward@arm.com>
10964             David Sherwood  <david.sherwood@arm.com>
10965
10966         * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
10967         Use plus_constant instead of gen_rtx_PLUS.
10968
10969 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10970             Alan Hayward  <alan.hayward@arm.com>
10971             David Sherwood  <david.sherwood@arm.com>
10972
10973         * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
10974         rather than an int.
10975
10976 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10977             Alan Hayward  <alan.hayward@arm.com>
10978             David Sherwood  <david.sherwood@arm.com>
10979
10980         * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
10981         instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
10982         via stack temporaries.  Treat the mode size as polynomial too.
10983
10984 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10985             Alan Hayward  <alan.hayward@arm.com>
10986             David Sherwood  <david.sherwood@arm.com>
10987
10988         * expr.c (expand_expr_real_2): When handling conversions involving
10989         unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
10990         multiplying int_size_in_bytes by BITS_PER_UNIT.  Treat GET_MODE_BISIZE
10991         as a poly_uint64 too.
10992
10993 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10994             Alan Hayward  <alan.hayward@arm.com>
10995             David Sherwood  <david.sherwood@arm.com>
10996
10997         * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
10998
10999 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11000             Alan Hayward  <alan.hayward@arm.com>
11001             David Sherwood  <david.sherwood@arm.com>
11002
11003         * combine.c (can_change_dest_mode): Handle polynomial
11004         REGMODE_NATURAL_SIZE.
11005         * expmed.c (store_bit_field_1): Likewise.
11006         * expr.c (store_constructor): Likewise.
11007         * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
11008         and polynomial REGMODE_NATURAL_SIZE.
11009         (gen_lowpart_common): Likewise.
11010         * reginfo.c (record_subregs_of_mode): Likewise.
11011         * rtlanal.c (read_modify_subreg_p): Likewise.
11012
11013 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11014             Alan Hayward  <alan.hayward@arm.com>
11015             David Sherwood  <david.sherwood@arm.com>
11016
11017         * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
11018         numbers of elements.
11019
11020 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11021             Alan Hayward  <alan.hayward@arm.com>
11022             David Sherwood  <david.sherwood@arm.com>
11023
11024         * match.pd: Cope with polynomial numbers of vector elements.
11025
11026 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11027             Alan Hayward  <alan.hayward@arm.com>
11028             David Sherwood  <david.sherwood@arm.com>
11029
11030         * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
11031         in a POINTER_PLUS_EXPR.
11032
11033 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11034             Alan Hayward  <alan.hayward@arm.com>
11035             David Sherwood  <david.sherwood@arm.com>
11036
11037         * omp-simd-clone.c (simd_clone_subparts): New function.
11038         (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
11039         (ipa_simd_modify_function_body): Likewise.
11040
11041 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11042             Alan Hayward  <alan.hayward@arm.com>
11043             David Sherwood  <david.sherwood@arm.com>
11044
11045         * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
11046         (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
11047         (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
11048         (expand_vector_condition, vector_element): Likewise.
11049         (subparts_gt): New function.
11050         (get_compute_type): Use subparts_gt.
11051         (count_type_subparts): Delete.
11052         (expand_vector_operations_1): Use subparts_gt instead of
11053         count_type_subparts.
11054
11055 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11056             Alan Hayward  <alan.hayward@arm.com>
11057             David Sherwood  <david.sherwood@arm.com>
11058
11059         * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
11060         (vect_compile_time_alias): ...this new function.  Do the calculation
11061         on poly_ints rather than trees.
11062         (vect_prune_runtime_alias_test_list): Update call accordingly.
11063
11064 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11065             Alan Hayward  <alan.hayward@arm.com>
11066             David Sherwood  <david.sherwood@arm.com>
11067
11068         * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
11069         numbers of units.
11070         (vect_schedule_slp_instance): Likewise.
11071
11072 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11073             Alan Hayward  <alan.hayward@arm.com>
11074             David Sherwood  <david.sherwood@arm.com>
11075
11076         * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
11077         constant and extern definitions for variable-length vectors.
11078         (vect_get_constant_vectors): Note that the number of units
11079         is known to be constant.
11080
11081 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11082             Alan Hayward  <alan.hayward@arm.com>
11083             David Sherwood  <david.sherwood@arm.com>
11084
11085         * tree-vect-stmts.c (vectorizable_conversion): Treat the number
11086         of units as polynomial.  Choose between WIDE and NARROW based
11087         on multiple_p.
11088
11089 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11090             Alan Hayward  <alan.hayward@arm.com>
11091             David Sherwood  <david.sherwood@arm.com>
11092
11093         * tree-vect-stmts.c (simd_clone_subparts): New function.
11094         (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
11095
11096 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11097             Alan Hayward  <alan.hayward@arm.com>
11098             David Sherwood  <david.sherwood@arm.com>
11099
11100         * tree-vect-stmts.c (vectorizable_call): Treat the number of
11101         vectors as polynomial.  Use build_index_vector for
11102         IFN_GOMP_SIMD_LANE.
11103
11104 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11105             Alan Hayward  <alan.hayward@arm.com>
11106             David Sherwood  <david.sherwood@arm.com>
11107
11108         * tree-vect-stmts.c (get_load_store_type): Treat the number of
11109         units as polynomial.  Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
11110         for variable-length vectors.
11111         (vectorizable_mask_load_store): Treat the number of units as
11112         polynomial, asserting that it is constant if the condition has
11113         already been enforced.
11114         (vectorizable_store, vectorizable_load): Likewise.
11115
11116 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11117             Alan Hayward  <alan.hayward@arm.com>
11118             David Sherwood  <david.sherwood@arm.com>
11119
11120         * tree-vect-loop.c (vectorizable_live_operation): Treat the number
11121         of units as polynomial.  Punt if we can't tell at compile time
11122         which vector contains the final result.
11123
11124 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11125             Alan Hayward  <alan.hayward@arm.com>
11126             David Sherwood  <david.sherwood@arm.com>
11127
11128         * tree-vect-loop.c (vectorizable_induction): Treat the number
11129         of units as polynomial.  Punt on SLP inductions.  Use an integer
11130         VEC_SERIES_EXPR for variable-length integer reductions.  Use a
11131         cast of such a series for variable-length floating-point
11132         reductions.
11133
11134 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11135             Alan Hayward  <alan.hayward@arm.com>
11136             David Sherwood  <david.sherwood@arm.com>
11137
11138         * tree.h (build_index_vector): Declare.
11139         * tree.c (build_index_vector): New function.
11140         * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
11141         of units as polynomial, forcibly converting it to a constant if
11142         vectorizable_reduction has already enforced the condition.
11143         (vect_create_epilog_for_reduction): Likewise.  Use build_index_vector
11144         to create a {1,2,3,...} vector.
11145         (vectorizable_reduction): Treat the number of units as polynomial.
11146         Choose vectype_in based on the largest scalar element size rather
11147         than the smallest number of units.  Enforce the restrictions
11148         relied on above.
11149
11150 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11151             Alan Hayward  <alan.hayward@arm.com>
11152             David Sherwood  <david.sherwood@arm.com>
11153
11154         * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
11155         number of units as polynomial.
11156
11157 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11158             Alan Hayward  <alan.hayward@arm.com>
11159             David Sherwood  <david.sherwood@arm.com>
11160
11161         * target.h (vector_sizes, auto_vector_sizes): New typedefs.
11162         * target.def (autovectorize_vector_sizes): Return the vector sizes
11163         by pointer, using vector_sizes rather than a bitmask.
11164         * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
11165         * targhooks.c (default_autovectorize_vector_sizes): Likewise.
11166         * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
11167         Likewise.
11168         * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
11169         * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
11170         * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
11171         * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
11172         * omp-general.c (omp_max_vf): Likewise.
11173         * omp-low.c (omp_clause_aligned_alignment): Likewise.
11174         * optabs-query.c (can_vec_mask_load_store_p): Likewise.
11175         * tree-vect-loop.c (vect_analyze_loop): Likewise.
11176         * tree-vect-slp.c (vect_slp_bb): Likewise.
11177         * doc/tm.texi: Regenerate.
11178         * tree-vectorizer.h (current_vector_size): Change from an unsigned int
11179         to a poly_uint64.
11180         * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
11181         the vector size as a poly_uint64 rather than an unsigned int.
11182         (current_vector_size): Change from an unsigned int to a poly_uint64.
11183         (get_vectype_for_scalar_type): Update accordingly.
11184         * tree.h (build_truth_vector_type): Take the size and number of
11185         units as a poly_uint64 rather than an unsigned int.
11186         (build_vector_type): Add a temporary overload that takes
11187         the number of units as a poly_uint64 rather than an unsigned int.
11188         * tree.c (make_vector_type): Likewise.
11189         (build_truth_vector_type): Take the number of units as a poly_uint64
11190         rather than an unsigned int.
11191
11192 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11193             Alan Hayward  <alan.hayward@arm.com>
11194             David Sherwood  <david.sherwood@arm.com>
11195
11196         * target.def (get_mask_mode): Take the number of units and length
11197         as poly_uint64s rather than unsigned ints.
11198         * targhooks.h (default_get_mask_mode): Update accordingly.
11199         * targhooks.c (default_get_mask_mode): Likewise.
11200         * config/i386/i386.c (ix86_get_mask_mode): Likewise.
11201         * doc/tm.texi: Regenerate.
11202
11203 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11204             Alan Hayward  <alan.hayward@arm.com>
11205             David Sherwood  <david.sherwood@arm.com>
11206
11207         * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
11208         * omp-general.c (omp_max_vf): Likewise.
11209         * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
11210         (expand_omp_simd): Handle polynomial safelen.
11211         * omp-low.c (omplow_simd_context): Add a default constructor.
11212         (omplow_simd_context::max_vf): Change from int to poly_uint64.
11213         (lower_rec_simd_input_clauses): Update accordingly.
11214         (lower_rec_input_clauses): Likewise.
11215
11216 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11217             Alan Hayward  <alan.hayward@arm.com>
11218             David Sherwood  <david.sherwood@arm.com>
11219
11220         * tree-vectorizer.h (vect_nunits_for_cost): New function.
11221         * tree-vect-loop.c (vect_model_reduction_cost): Use it.
11222         * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
11223         (vect_analyze_slp_cost): Likewise.
11224         * tree-vect-stmts.c (vect_model_store_cost): Likewise.
11225         (vect_model_load_cost): Likewise.
11226
11227 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11228             Alan Hayward  <alan.hayward@arm.com>
11229             David Sherwood  <david.sherwood@arm.com>
11230
11231         * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
11232         (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
11233         from an unsigned int * to a poly_uint64_pod *.
11234         (calculate_unrolling_factor): New function.
11235         (vect_analyze_slp_instance): Use it.  Track polynomial max_nunits.
11236
11237 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11238             Alan Hayward  <alan.hayward@arm.com>
11239             David Sherwood  <david.sherwood@arm.com>
11240
11241         * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
11242         from an unsigned int to a poly_uint64.
11243         (_loop_vec_info::slp_unrolling_factor): Likewise.
11244         (_loop_vec_info::vectorization_factor): Change from an int
11245         to a poly_uint64.
11246         (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
11247         (vect_get_num_vectors): New function.
11248         (vect_update_max_nunits, vect_vf_for_cost): Likewise.
11249         (vect_get_num_copies): Use vect_get_num_vectors.
11250         (vect_analyze_data_ref_dependences): Change max_vf from an int *
11251         to an unsigned int *.
11252         (vect_analyze_data_refs): Change min_vf from an int * to a
11253         poly_uint64 *.
11254         (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
11255         than an unsigned HOST_WIDE_INT.
11256         * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
11257         (vect_analyze_data_ref_dependence): Change max_vf from an int *
11258         to an unsigned int *.
11259         (vect_analyze_data_ref_dependences): Likewise.
11260         (vect_compute_data_ref_alignment): Handle polynomial vf.
11261         (vect_enhance_data_refs_alignment): Likewise.
11262         (vect_prune_runtime_alias_test_list): Likewise.
11263         (vect_shift_permute_load_chain): Likewise.
11264         (vect_supportable_dr_alignment): Likewise.
11265         (dependence_distance_ge_vf): Take the vectorization factor as a
11266         poly_uint64 rather than an unsigned HOST_WIDE_INT.
11267         (vect_analyze_data_refs): Change min_vf from an int * to a
11268         poly_uint64 *.
11269         * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
11270         vfm1 as a poly_uint64 rather than an int.  Make the same change
11271         for the returned bound_scalar.
11272         (vect_gen_vector_loop_niters): Handle polynomial vf.
11273         (vect_do_peeling): Likewise.  Update call to
11274         vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
11275         (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
11276         be constant.
11277         * tree-vect-loop.c (vect_determine_vectorization_factor)
11278         (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
11279         (vect_get_known_peeling_cost): Likewise.
11280         (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
11281         (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
11282         (vect_transform_loop): Likewise.  Use the lowest possible VF when
11283         updating the upper bounds of the loop.
11284         (vect_min_worthwhile_factor): Make static.  Return an unsigned int
11285         rather than an int.
11286         * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
11287         polynomial unroll factors.
11288         (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
11289         (vect_make_slp_decision): Likewise.
11290         (vect_supported_load_permutation_p): Likewise, and polynomial
11291         vf too.
11292         (vect_analyze_slp_cost): Handle polynomial vf.
11293         (vect_slp_analyze_node_operations): Likewise.
11294         (vect_slp_analyze_bb_1): Likewise.
11295         (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
11296         than an unsigned HOST_WIDE_INT.
11297         * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
11298         (vectorizable_load): Handle polynomial vf.
11299         * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
11300         a poly_uint64.
11301         (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
11302
11303 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11304             Alan Hayward  <alan.hayward@arm.com>
11305             David Sherwood  <david.sherwood@arm.com>
11306
11307         * match.pd: Handle bit operations involving three constants
11308         and try to fold one pair.
11309
11310 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11311
11312         * tree-vect-loop-manip.c: Include gimple-fold.h.
11313         (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
11314         niters_maybe_zero parameters.  Handle other cases besides a step of 1.
11315         (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
11316         Add a path that uses a step of VF instead of 1, but disable it
11317         for now.
11318         (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
11319         and niters_no_overflow parameters.  Update calls to
11320         slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
11321         Create a new SSA name if the latter choses to use a ste other
11322         than zero, and return it via niters_vector_mult_vf_var.
11323         * tree-vect-loop.c (vect_transform_loop): Update calls to
11324         vect_do_peeling, vect_gen_vector_loop_niters and
11325         slpeel_make_loop_iterate_ntimes.
11326         * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
11327         (vect_gen_vector_loop_niters): Update declarations after above changes.
11328
11329 2018-01-02  Michael Meissner  <meissner@linux.vnet.ibm.com>
11330
11331         * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
11332         128-bit round to integer instructions.
11333         (ceil<mode>2): Likewise.
11334         (btrunc<mode>2): Likewise.
11335         (round<mode>2): Likewise.
11336
11337 2018-01-02  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
11338
11339         * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
11340         unaligned VSX load/store on P8/P9.
11341         (expand_block_clear): Allow the use of unaligned VSX
11342         load/store on P8/P9.
11343
11344 2018-01-02  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
11345
11346         * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
11347         New function.
11348         (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
11349         swap associated with both a load and a store.
11350
11351 2018-01-02  Andrew Waterman  <andrew@sifive.com>
11352
11353         * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
11354         * config/riscv/riscv.md (clear_cache): Use it.
11355
11356 2018-01-02  Artyom Skrobov  <tyomitch@gmail.com>
11357
11358         * web.c: Remove out-of-date comment.
11359
11360 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11361
11362         * expr.c (fixup_args_size_notes): Check that any existing
11363         REG_ARGS_SIZE notes are correct, and don't try to re-add them.
11364         (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
11365         (emit_single_push_insn): ...here.
11366
11367 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11368
11369         * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
11370         (const_vector_encoded_nelts): New function.
11371         (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
11372         (const_vector_int_elt, const_vector_elt): Declare.
11373         * emit-rtl.c (const_vector_int_elt_1): New function.
11374         (const_vector_elt): Likewise.
11375         * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
11376         of CONST_VECTOR_ELT.
11377
11378 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11379
11380         * expr.c: Include rtx-vector-builder.h.
11381         (const_vector_mask_from_tree): Use rtx_vector_builder and operate
11382         directly on the tree encoding.
11383         (const_vector_from_tree): Likewise.
11384         * optabs.c: Include rtx-vector-builder.h.
11385         (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
11386         sequence of "u" values.
11387         * vec-perm-indices.c: Include rtx-vector-builder.h.
11388         (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
11389         directly on the vec_perm_indices encoding.
11390
11391 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11392
11393         * doc/rtl.texi (const_vector): Describe new encoding scheme.
11394         * Makefile.in (OBJS): Add rtx-vector-builder.o.
11395         * rtx-vector-builder.h: New file.
11396         * rtx-vector-builder.c: Likewise.
11397         * rtl.h (rtx_def::u2): Add a const_vector field.
11398         (CONST_VECTOR_NPATTERNS): New macro.
11399         (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
11400         (CONST_VECTOR_DUPLICATE_P): Likewise.
11401         (CONST_VECTOR_STEPPED_P): Likewise.
11402         (CONST_VECTOR_ENCODED_ELT): Likewise.
11403         (const_vec_duplicate_p): Check for a duplicated vector encoding.
11404         (unwrap_const_vec_duplicate): Likewise.
11405         (const_vec_series_p): Check for a non-duplicated vector encoding.
11406         Say that the function only returns true for integer vectors.
11407         * emit-rtl.c: Include rtx-vector-builder.h.
11408         (gen_const_vec_duplicate_1): Delete.
11409         (gen_const_vector): Call gen_const_vec_duplicate instead of
11410         gen_const_vec_duplicate_1.
11411         (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
11412         (gen_const_vec_duplicate): Use rtx_vector_builder.
11413         (gen_const_vec_series): Likewise.
11414         (gen_rtx_CONST_VECTOR): Likewise.
11415         * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
11416         (swap_const_vector_halves): Take an rtx pointer rather than rtx.
11417         Build a new vector rather than modifying a CONST_VECTOR in-place.
11418         (handle_special_swappables): Update call accordingly.
11419         * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
11420         (swap_const_vector_halves): Take an rtx pointer rather than rtx.
11421         Build a new vector rather than modifying a CONST_VECTOR in-place.
11422         (handle_special_swappables): Update call accordingly.
11423
11424 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11425
11426         * simplify-rtx.c (simplify_const_binary_operation): Use
11427         CONST_VECTOR_ELT instead of XVECEXP.
11428
11429 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11430
11431         * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
11432         the selector elements to be different from the data elements
11433         if the selector is a VECTOR_CST.
11434         * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
11435         ssizetype for the selector.
11436
11437 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11438
11439         * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
11440         before testing each element individually.
11441         * tree-vect-generic.c (lower_vec_perm): Likewise.
11442
11443 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11444
11445         * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
11446         * selftest-run-tests.c (selftest::run_tests): Call it.
11447         * vector-builder.h (vector_builder::operator ==): New function.
11448         (vector_builder::operator !=): Likewise.
11449         * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
11450         (vec_perm_indices::all_from_input_p): New function.
11451         * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
11452         (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
11453         * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
11454         instead of reading the VECTOR_CST directly.  Detect whether both
11455         vector inputs are the same before constructing the vec_perm_indices,
11456         and update the number of inputs argument accordingly.  Use the
11457         utility functions added above.  Only construct sel2 if we need to.
11458
11459 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11460
11461         * optabs.c (expand_vec_perm_var): Use an explicit encoding for
11462         the broadcast of the low byte.
11463         (expand_mult_highpart): Use an explicit encoding for the permutes.
11464         * optabs-query.c (can_mult_highpart_p): Likewise.
11465         * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
11466         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
11467         (vectorizable_bswap): Likewise.
11468         * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
11469         explicit encoding for the power-of-2 permutes.
11470         (vect_permute_store_chain): Likewise.
11471         (vect_grouped_load_supported): Likewise.
11472         (vect_permute_load_chain): Likewise.
11473
11474 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11475
11476         * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
11477         * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
11478         * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
11479         * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
11480         * tree-vect-stmts.c (vectorizable_bswap): Likewise.
11481         (vect_gen_perm_mask_any): Likewise.
11482
11483 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11484
11485         * int-vector-builder.h: New file.
11486         * vec-perm-indices.h: Include int-vector-builder.h.
11487         (vec_perm_indices): Redefine as an int_vector_builder.
11488         (auto_vec_perm_indices): Delete.
11489         (vec_perm_builder): Redefine as a stand-alone class.
11490         (vec_perm_indices::vec_perm_indices): New function.
11491         (vec_perm_indices::clamp): Likewise.
11492         * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
11493         (vec_perm_indices::new_vector): New function.
11494         (vec_perm_indices::new_expanded_vector): Update for new
11495         vec_perm_indices class.
11496         (vec_perm_indices::rotate_inputs): New function.
11497         (vec_perm_indices::all_in_range_p): Operate directly on the
11498         encoded form, without computing elided elements.
11499         (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
11500         encoding.  Update for new vec_perm_indices class.
11501         * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
11502         the given vec_perm_builder.
11503         (expand_vec_perm_var): Update vec_perm_builder constructor.
11504         (expand_mult_highpart): Use vec_perm_builder instead of
11505         auto_vec_perm_indices.
11506         * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
11507         vec_perm_indices instead of auto_vec_perm_indices.  Use a single
11508         or double series encoding as appropriate.
11509         * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
11510         vec_perm_indices instead of auto_vec_perm_indices.
11511         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
11512         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
11513         (vect_permute_store_chain): Likewise.
11514         (vect_grouped_load_supported): Likewise.
11515         (vect_permute_load_chain): Likewise.
11516         (vect_shift_permute_load_chain): Likewise.
11517         * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
11518         (vect_transform_slp_perm_load): Likewise.
11519         (vect_schedule_slp_instance): Likewise.
11520         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
11521         (vectorizable_mask_load_store): Likewise.
11522         (vectorizable_bswap): Likewise.
11523         (vectorizable_store): Likewise.
11524         (vectorizable_load): Likewise.
11525         * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
11526         vec_perm_indices instead of auto_vec_perm_indices.  Use
11527         tree_to_vec_perm_builder to read the vector from a tree.
11528         * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
11529         vec_perm_builder instead of a vec_perm_indices.
11530         (have_whole_vector_shift): Use vec_perm_builder and
11531         vec_perm_indices instead of auto_vec_perm_indices.  Leave the
11532         truncation to calc_vec_perm_mask_for_shift.
11533         (vect_create_epilog_for_reduction): Likewise.
11534         * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
11535         from auto_vec_perm_indices to vec_perm_indices.
11536         (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
11537         instead of changing individual elements.
11538         (aarch64_vectorize_vec_perm_const): Use new_vector to install
11539         the vector in d.perm.
11540         * config/arm/arm.c (expand_vec_perm_d::perm): Change
11541         from auto_vec_perm_indices to vec_perm_indices.
11542         (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
11543         instead of changing individual elements.
11544         (arm_vectorize_vec_perm_const): Use new_vector to install
11545         the vector in d.perm.
11546         * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
11547         Update vec_perm_builder constructor.
11548         (rs6000_expand_interleave): Likewise.
11549         * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
11550         (rs6000_expand_interleave): Likewise.
11551
11552 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11553
11554         * optabs-query.c (can_vec_perm_var_p): Check whether lowering
11555         to qimode could truncate the indices.
11556         * optabs.c (expand_vec_perm_var): Likewise.
11557
11558 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11559
11560         * Makefile.in (OBJS): Add vec-perm-indices.o.
11561         * vec-perm-indices.h: New file.
11562         * vec-perm-indices.c: Likewise.
11563         * target.h (vec_perm_indices): Replace with a forward class
11564         declaration.
11565         (auto_vec_perm_indices): Move to vec-perm-indices.h.
11566         * optabs.h: Include vec-perm-indices.h.
11567         (expand_vec_perm): Delete.
11568         (selector_fits_mode_p, expand_vec_perm_var): Declare.
11569         (expand_vec_perm_const): Declare.
11570         * target.def (vec_perm_const_ok): Replace with...
11571         (vec_perm_const): ...this new hook.
11572         * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
11573         (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
11574         * doc/tm.texi: Regenerate.
11575         * optabs.def (vec_perm_const): Delete.
11576         * doc/md.texi (vec_perm_const): Likewise.
11577         (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
11578         * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
11579         expand_vec_perm for constant permutation vectors.  Assert that
11580         the mode of variable permutation vectors is the integer equivalent
11581         of the mode that is being permuted.
11582         * optabs-query.h (selector_fits_mode_p): Declare.
11583         * optabs-query.c: Include vec-perm-indices.h.
11584         (selector_fits_mode_p): New function.
11585         (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
11586         is defined, instead of checking whether the vec_perm_const_optab
11587         exists.  Use targetm.vectorize.vec_perm_const instead of
11588         targetm.vectorize.vec_perm_const_ok.  Check whether the indices
11589         fit in the vector mode before using a variable permute.
11590         * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
11591         vec_perm_indices instead of an rtx.
11592         (expand_vec_perm): Replace with...
11593         (expand_vec_perm_const): ...this new function.  Take the selector
11594         as a vec_perm_indices rather than an rtx.  Also take the mode of
11595         the selector.  Update call to shift_amt_for_vec_perm_mask.
11596         Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
11597         Use vec_perm_indices::new_expanded_vector to expand the original
11598         selector into bytes.  Check whether the indices fit in the vector
11599         mode before using a variable permute.
11600         (expand_vec_perm_var): Make global.
11601         (expand_mult_highpart): Use expand_vec_perm_const.
11602         * fold-const.c: Includes vec-perm-indices.h.
11603         * tree-ssa-forwprop.c: Likewise.
11604         * tree-vect-data-refs.c: Likewise.
11605         * tree-vect-generic.c: Likewise.
11606         * tree-vect-loop.c: Likewise.
11607         * tree-vect-slp.c: Likewise.
11608         * tree-vect-stmts.c: Likewise.
11609         * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
11610         Delete.
11611         * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
11612         * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
11613         (aarch64_vectorize_vec_perm_const_ok): Fuse into...
11614         (aarch64_vectorize_vec_perm_const): ...this new function.
11615         (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
11616         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
11617         * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
11618         * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
11619         * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
11620         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
11621         (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
11622         into...
11623         (arm_vectorize_vec_perm_const): ...this new function.  Explicitly
11624         check for NEON modes.
11625         * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
11626         * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
11627         * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
11628         (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
11629         into...
11630         (ix86_vectorize_vec_perm_const): ...this new function.  Incorporate
11631         the old VEC_PERM_CONST conditions.
11632         * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
11633         * config/ia64/vect.md (vec_perm_const<mode>): Delete.
11634         * config/ia64/ia64.c (ia64_expand_vec_perm_const)
11635         (ia64_vectorize_vec_perm_const_ok): Merge into...
11636         (ia64_vectorize_vec_perm_const): ...this new function.
11637         * config/mips/loongson.md (vec_perm_const<mode>): Delete.
11638         * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
11639         * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
11640         * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
11641         * config/mips/mips.c (mips_expand_vec_perm_const)
11642         (mips_vectorize_vec_perm_const_ok): Merge into...
11643         (mips_vectorize_vec_perm_const): ...this new function.
11644         * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
11645         * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
11646         * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
11647         * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
11648         * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
11649         (rs6000_expand_vec_perm_const): Delete.
11650         * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
11651         Delete.
11652         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
11653         (altivec_expand_vec_perm_const_le): Take each operand individually.
11654         Operate on constant selectors rather than rtxes.
11655         (altivec_expand_vec_perm_const): Likewise.  Update call to
11656         altivec_expand_vec_perm_const_le.
11657         (rs6000_expand_vec_perm_const): Delete.
11658         (rs6000_vectorize_vec_perm_const_ok): Delete.
11659         (rs6000_vectorize_vec_perm_const): New function.
11660         (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
11661         an element count and rtx array.
11662         (rs6000_expand_extract_even): Update call accordingly.
11663         (rs6000_expand_interleave): Likewise.
11664         * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
11665         * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
11666         * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
11667         * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
11668         (rs6000_expand_vec_perm_const): Delete.
11669         * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
11670         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
11671         (altivec_expand_vec_perm_const_le): Take each operand individually.
11672         Operate on constant selectors rather than rtxes.
11673         (altivec_expand_vec_perm_const): Likewise.  Update call to
11674         altivec_expand_vec_perm_const_le.
11675         (rs6000_expand_vec_perm_const): Delete.
11676         (rs6000_vectorize_vec_perm_const_ok): Delete.
11677         (rs6000_vectorize_vec_perm_const): New function.  Remove stray
11678         reference to the SPE evmerge intructions.
11679         (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
11680         an element count and rtx array.
11681         (rs6000_expand_extract_even): Update call accordingly.
11682         (rs6000_expand_interleave): Likewise.
11683         * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
11684         * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
11685         new function.
11686         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
11687
11688 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11689
11690         * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
11691         vector mode and that that mode matches the mode of the data
11692         being permuted.
11693         (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
11694         out into expand_vec_perm_var.  Do all CONST_VECTOR handling here,
11695         directly using expand_vec_perm_1 when forcing selectors into
11696         registers.
11697         (expand_vec_perm_var): New function, split out from expand_vec_perm.
11698
11699 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11700
11701         * optabs-query.h (can_vec_perm_p): Delete.
11702         (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
11703         * optabs-query.c (can_vec_perm_p): Split into...
11704         (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
11705         (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
11706         particular selector is valid.
11707         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
11708         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
11709         (vect_grouped_load_supported): Likewise.
11710         (vect_shift_permute_load_chain): Likewise.
11711         * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
11712         (vect_transform_slp_perm_load): Likewise.
11713         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
11714         (vectorizable_bswap): Likewise.
11715         (vect_gen_perm_mask_checked): Likewise.
11716         * fold-const.c (fold_ternary_loc): Likewise.  Don't take
11717         implementations of variable permutation vectors into account
11718         when deciding which selector to use.
11719         * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
11720         vec_perm_const_optab is supported; instead use can_vec_perm_const_p
11721         with a false third argument.
11722         * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
11723         to test whether the constant selector is valid and can_vec_perm_var_p
11724         to test whether a variable selector is valid.
11725
11726 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11727
11728         * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
11729         * optabs-query.c (can_vec_perm_p): Likewise.
11730         * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
11731         instead of vec_perm_indices.
11732         * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
11733         (vect_gen_perm_mask_checked): Likewise,
11734         * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
11735         (vect_gen_perm_mask_checked): Likewise,
11736
11737 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11738
11739         * optabs-query.h (qimode_for_vec_perm): Declare.
11740         * optabs-query.c (can_vec_perm_p): Split out qimode search to...
11741         (qimode_for_vec_perm): ...this new function.
11742         * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
11743
11744 2018-01-02  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
11745
11746         * rtlanal.c (canonicalize_condition): Return 0 if final rtx
11747         does not have a conditional at the top.
11748
11749 2018-01-02  Richard Biener  <rguenther@suse.de>
11750
11751         * ipa-inline.c (big_speedup_p): Fix expression.
11752
11753 2018-01-02  Jan Hubicka  <hubicka@ucw.cz>
11754
11755         PR target/81616
11756         * config/i386/x86-tune-costs.h: Increase cost of integer load costs
11757         for generic 4->6.
11758
11759 2018-01-02  Jan Hubicka  <hubicka@ucw.cz>
11760
11761         PR target/81616
11762         Generic tuning.
11763         * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
11764         cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
11765         and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
11766         cond_taken_branch_cost 3->4.
11767
11768 2018-01-01  Jakub Jelinek  <jakub@redhat.com>
11769
11770         PR tree-optimization/83581
11771         * tree-loop-distribution.c (pass_loop_distribution::execute): Return
11772         TODO_cleanup_cfg if any changes have been made.
11773
11774         PR middle-end/83608
11775         * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
11776         convert_modes if target mode has the right side, but different mode
11777         class.
11778
11779         PR middle-end/83609
11780         * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
11781         last argument when extracting from CONCAT.  If either from_real or
11782         from_imag is NULL, use expansion through memory.  If result is not
11783         a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
11784         the parts directly to inner mode, if even that fails, use expansion
11785         through memory.
11786
11787         PR middle-end/83623
11788         * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
11789         check for bswap in mode rather than HImode and use that in expand_unop
11790         too.
11791 \f
11792 Copyright (C) 2018 Free Software Foundation, Inc.
11793
11794 Copying and distribution of this file, with or without modification,
11795 are permitted in any medium without royalty provided the copyright
11796 notice and this notice are preserved.