gcc/ChangeLog

   1 2018-05-01  Jakub Jelinek  <jakub@redhat.com>
   2
   3         * configure.ac (LD_AS_NEEDED_OPTION, LD_NO_AS_NEEDED_OPTION): Use
   4         --push-state --as-needed and --pop-state instead of --as-needed and
   5         --no-as-needed if ld supports it.
   6         * configure: Regenerated.
   7
   8         PR web/85578
   9         * doc/install.texi2html: Replace _002d with - and _002a with * in
  10         generated html files using sed.
  11
  12 2018-04-30  David Malcolm  <dmalcolm@redhat.com>
  13
  14         PR c++/85523
  15         * gcc-rich-location.c (blank_line_before_p): New function.
  16         (use_new_line): New function.
  17         (gcc_rich_location::add_fixit_insert_formatted): New function.
  18         * gcc-rich-location.h
  19         (gcc_rich_location::add_fixit_insert_formatted): New function.
  20
  21 2018-04-30  David Malcolm  <dmalcolm@redhat.com>
  22
  23         * selftest.c (assert_streq): Rename "expected" and "actual" to
  24         "val1" and "val2".  Extend NULL-handling to cover both inputs
  25         symmetrically, while still requiring both to be non-NULL for a pass.
  26         * selftest.h (assert_streq): Rename "expected" and "actual" to
  27         "val1" and "val2".
  28         (ASSERT_EQ): Likewise.
  29         (ASSERT_EQ_AT): Likewise.
  30         (ASSERT_KNOWN_EQ): Likewise.
  31         (ASSERT_KNOWN_EQ_AT): Likewise.
  32         (ASSERT_NE): Likewise.
  33         (ASSERT_MAYBE_NE): Likewise.
  34         (ASSERT_MAYBE_NE_AT): Likewise.
  35         (ASSERT_STREQ): Likewise.  Clarify that both must be non-NULL for
  36         the assertion to pass.
  37         (ASSERT_STREQ_AT): Likewise.
  38
  39 2018-04-30  Jonathan Wakely  <jwakely@redhat.com>
  40
  41         * doc/invoke.texi (-fpie, -fPIE): Fix grammar and clarify
  42         interaction with -pie.
  43
  44 2018-04-30  David Malcolm  <dmalcolm@redhat.com>
  45
  46         * selftest.h: Fix alphabetization of per-source-file selftest
  47         declarations.
  48
  49 2018-04-30  Jason Merrill  <jason@redhat.com>
  50
  51         PR c++/61982 - dead stores to destroyed objects.
  52         * gimplify.c (gimplify_modify_expr): Simplify complex lvalue on LHS
  53         of clobber.
  54
  55 2018-04-30  Jason Merrill  <jason@redhat.com>
  56
  57         * tree.c (build_clobber): New.
  58         * tree.h: Declare it.
  59         * gimplify.c (gimplify_bind_expr, gimplify_target_expr): Use it.
  60
  61 2018-04-30  David Malcolm  <dmalcolm@redhat.com>
  62
  63         * diagnostic-show-locus.c (layout::layout): Update for
  64         location_get_source_line returning a char_span.
  65         (struct char_span): Move to input.h.
  66         (struct correction): Update for fields in char_span becoming
  67         private.
  68         (struct source_line): Update for location_get_source_line
  69         returning a char_span.
  70         (layout::print_line): Likewise.
  71         * edit-context.c (edited_file::print_content): Likewise.
  72         (edited_file::print_diff_hunk): Likewise.
  73         (edited_file::print_run_of_changed_lines): Likewise.
  74         (edited_file::get_num_lines): Likewise.
  75         (edited_line::edited_line): Likewise.
  76         * final.c (asm_show_source): Likewise.
  77         * input.c (location_get_source_line): Convert return type
  78         from const char * to char_span, losing the final "line_len"
  79         param.
  80         (dump_location_info): Update for the above.
  81         (get_substring_ranges_for_loc): Likewise.  Use a char_span
  82         when handling the literal within the line.
  83         (test_reading_source_line): Update for location_get_source_line
  84         returning a char_span.
  85         * input.h (class char_span): Move here from
  86         diagnostic-show-locus.c, converting from a struct to a class.
  87         Make data members private.
  88         (char_span::operator bool): New.
  89         (char_span::length): New.
  90         (char_span::get_buffer): New.
  91         (char_span::operator[]): New.
  92         (char_span::subspan): Make const.
  93         (char_span::xstrdup): New.
  94         (location_get_source_line): Convert return type from const char *
  95         to char_span, losing the final "line_size" param.
  96
  97 2018-04-30  Jan Hubicka  <jh@suse.cz>
  98
  99         * lto-wrapper.c (ltrans_priorities): New static var.
 100         (cmp_priority): New.
 101         (run_gcc): Read priorities and if doing parallel build order
 102         the Makefile by them.
 103
 104 2018-04-30  David Malcolm  <dmalcolm@redhat.com>
 105
 106         * input.h (builtins_location_check): Convert to a STATIC_ASSERT.
 107
 108 2018-04-30  Richard Biener  <rguenther@suse.de>
 109
 110         * tree-cfg.c (verify_address): Remove base argument, add
 111         flag whether to check TREE_ADDRESSABLE and do that.
 112         (verify_expr): Remove.
 113         (verify_types_in_gimple_reference): Add pieces from verify_expr.
 114         (verify_gimple_assign_single): Likewise.
 115         (verify_gimple_switch): Likewise.
 116         (verify_expr_location_1): Dereference tp once.  Add (disabled)
 117         piece from verify_expr.
 118         (verify_gimple_in_cfg): Do not call verify_expr on all ops.
 119
 120 2018-04-30  Claudiu Zissulescu  <claziss@synopsys.com>
 121
 122         * config/arc/linux.h (CLEAR_INSN_CACHE): Define.
 123
 124 2018-04-30  Claudiu Zissulescu  <claziss@synopsys.com>
 125
 126         * config/arc/arc-protos.h (prepare_extend_operands): Remove.
 127         (small_data_pattern): Likewise.
 128         (arc_rewrite_small_data): Likewise.
 129         * config/arc/arc.c (LEGITIMATE_SMALL_DATA_OFFSET_P): Remove.
 130         (LEGITIMATE_SMALL_DATA_ADDRESS_P): Likewise.
 131         (get_symbol_alignment): New function.
 132         (legitimate_small_data_address_p): Likewise.
 133         (legitimate_scaled_address): Update, call
 134         legitimate_small_data_address_p.
 135         (output_sdata): New static variable.
 136         (arc_print_operand): Update how we handle small data operands.
 137         (arc_print_operand_address): Likewise.
 138         (arc_legitimate_address_p): Update, use
 139         legitimate_small_data_address_p.
 140         (arc_rewrite_small_data_p): Remove.
 141         (arc_rewrite_small_data_1): Likewise.
 142         (arc_rewrite_small_data): Likewise.
 143         (small_data_pattern): Likewise.
 144         (compact_sda_memory_operand): Update to use
 145         legitimate_small_data_address_p and get_symbol_alignment.
 146         (prepare_move_operands): Don't rewite sdata pattern.
 147         (prepare_extend_operands): Remove.
 148         * config/arc/arc.md (zero_extendqihi2): Don't rewrite sdata
 149         pattern.
 150         (zero_extendqisi2): Likewise.
 151         (zero_extendhisi2): Likewise.
 152         (extendqihi2): Likewise.
 153         (extendqisi2): Likewise.
 154         (extendhisi2): Likewise.
 155         (addsi3): Likewise.
 156         (subsi3): Likewise.
 157         (andsi3): Likewise.
 158         * config/arc/constraints.md (Usd): Change it to memory constraint.
 159
 160 2018-04-30  Claudiu Zissulescu  <claziss@synopsys.com>
 161
 162         * config/arc/arc.c (arc_split_move): Allow signed 6-bit constants
 163         as source of std instructions.
 164         * config/arc/arc.md (movsi_insn): Update pattern predicate to
 165         allow 6-bit constants as source for store instructions.
 166         (movdi_insn): Update instruction pattern to allow 6-bit constants
 167         as source for store instructions.
 168
 169 2018-04-30  Jonathan Wakely  <jwakely@redhat.com>
 170
 171         * doc/invoke.texi (-fdebug-types-section): Fix grammar.
 172
 173 2018-04-30  Nathan Sidwell  <nathan@acm.org>
 174             Sandra Loosemore <sandra@codesourcery.com>
 175
 176         * dumpfile.c (dump_open): Allow '-' for stdout.
 177         * doc/invoke.texi (Developer Options): Document dump filename
 178         determination early.  Document stdin/stdout selection.
 179
 180 2018-04-30 Andrew Sadek  <andrew.sadek.se@gmail.com>
 181
 182         Microblaze Target: PIC data text relative
 183
 184         * config/microblaze/microblaze.opt: add new option -mpic-data-text-rel.
 185         * config/microblaze/microblaze-protos.h (microblaze_constant_address_p):
 186         Add declaration.
 187         * gcc/config/microblaze/microblaze.h (microblaze_constant_address_p):
 188         CONSTANT_ADDRESS_P definition to microblaze_constant_address_p.
 189         * config/microblaze/microblaze.c (TARGET_PIC_DATA_TEXT_REL):
 190         New addressing mode for data-text relative position indepenedent code.
 191         (microblaze_classify_unspec): add 'UNSPEC_TEXT' case ->
 192         'ADDRESS_SYMBOLIC_TXT_REL'.
 193         (microblaze_classify_address): Add handling for UNSPEC + CONST_INT.
 194         (microblaze_legitimate_pic_operand): Exclude function calls from
 195         pic operands in case of TARGET_PIC_DATA_TEXT_REL option.
 196         (microblaze_legitimize_address): Generate 'UNSPEC_TEXT' for all possible
 197         addresses cases.
 198         (microblaze_address_insns): Add 'ADDRESS_SYMBOLIC_TXT_REL' case.
 199         (print_operand): Add 'ADDRESS_SYMBOLIC_TXT_REL' case.
 200         (print_operand_address): Add 'ADDRESS_SYMBOLIC_TXT_REL' case + handling
 201         for 'address + offset'.
 202         (microblaze_expand_prologue): Add new function prologue call for
 203         'r20' assignation.
 204         (microblaze_asm_generate_pic_addr_dif_vec): Override new target hook
 205         'TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC' to disable address diff vector
 206         table in case of TARGET_PIC_DATA_TEXT_REL.
 207         (expand_pic_symbol_ref): Add handling for 'UNSPEC_TEXT'.
 208         * gcc/config/microblaze/microblaze.md (TARGET_PIC_DATA_TEXT_REL):
 209         Add new macros 'UNSPEC_TEXT',
 210         'UNSPEC_SET_TEXT' + add rule for setting r20 in function prologue
 211         + exclude function calls from 'UNSPEC_PLT' in case of data text
 212         relative mode.
 213         * doc/tm.texi.in (TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC): Add
 214         new target hook for generating address diff vector tables in case of
 215         flag_pic.
 216         * doc/tm.texi : Regenerate.
 217         * stmt.c (TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC): Append new condition
 218         'targetm.asm_out.generate_pic_addr_diff_vec' to flag_pic in case
 219         of addr diff vector generation.
 220         * target.def (TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC): Add
 221         target hook definition.
 222         * targhooks.h, gcc/targhooks.c (TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC):
 223         Add default function for generate_pic_addr_diff_vec -> flag_pic.
 224         * doc/invoke.texi (Add new pic option): Add new microblaze pic
 225         option for data text relative.
 226
 227 2018-04-30  Richard Biener  <rguenther@suse.de>
 228
 229         * tree-chrec.h (evolution_function_is_constant_p): Remove
 230         redundant check.
 231         * tree-cfg.c (tree_node_can_be_shared): Re-order checks.
 232
 233 2018-04-30  Richard Biener  <rguenther@suse.de>
 234
 235         PR bootstrap/85571
 236         * dwarf2out.c (gen_producer_string): Ignore -fchecking[=].
 237
 238 2018-04-30  Richard Biener  <rguenther@suse.de>
 239
 240         PR tree-optimization/28364
 241         PR tree-optimization/85275
 242         * tree-ssa-loop-ch.c (ch_base::copy_headers): Stop after
 243         copying first exit test.
 244
 245 2018-04-28  Mark Wielaard  <mark@klomp.org>
 246
 247         * dwarf2out.c (dwarf2out_finish): Add .debug_addr table header for
 248         dwarf_version >= 5.
 249         (dwarf_AT): Handle DW_AT_addr_base.
 250         (add_top_level_skeleton_die_attrs): Use dwarf_AT for DW_AT_addr_base.
 251
 252 2018-04-28  Uros Bizjak  <ubizjak@gmail.com>
 253
 254         PR target/84431
 255         * config/i386/i386.md (*ashl<dwi>3_doubleword_mask): New pattern.
 256         (*ashl<dwi>3_doubleword_mask_1): Ditto.
 257         (*<shift_insn><dwi>3_doubleword_mask): Ditto.
 258         (*<shift_insn><dwi>3_doubleword_mask_1): Ditto.
 259
 260 2018-04-28  Richard Biener  <rguenther@suse.de>
 261
 262         * tree-cfg.c (verify_gimple_phi): Take a gphi * argument.
 263         (verify_gimple_in_cfg): Rename visited_stmts to visited_throwing_stmts
 264         to reflect use.  Only add interesting stmts.
 265
 266 2018-04-27  Martin Jambor  <mjambor@suse.cz>
 267
 268         PR ipa/85549
 269         * ipa-cp.c (find_aggregate_values_for_callers_subset): Make sure
 270         the jump function allows for passing through aggregate values.
 271
 272 2018-04-27  David Malcolm  <dmalcolm@redhat.com>
 273
 274         * input.h (in_system_header_at): Convert from macro to inline
 275         function.
 276         (from_macro_expansion_at): Likewise.
 277         (from_macro_definition_at): Likewise.
 278
 279 2018-04-27  Jeff Law  <law@redhat.com>
 280
 281         * config.gcc: Mark tile* targets as deprecated/obsolete.
 282
 283 2018-04-27  Richard Biener  <rguenther@suse.de>
 284
 285         * config/aarch64/aarch64.c: Simplify ap.__stack advance and
 286         fix for ILP32.
 287
 288 2018-04-27  Richard Biener  <rguenther@suse.de>
 289
 290         * tree-cfg.c (verify_expr): Make dead code hit gcc_unreachable.
 291
 292 2018-04-27  Uros Bizjak  <ubizjak@gmail.com>
 293
 294         * config/i386/i386.md (*movti_internal): Substitute Ye constraint
 295         with Yd constraint. Set "preferred_for_speed" attribute from
 296         TARGET_INTER_UNIT_MOVES_{FROM,TO}_VEC for alternatives
 297         with Yd constraint.
 298         (*movdi_internal): Ditto.
 299         (movti_interunit splitters): Remove
 300         TARGET_INTER_UNIT_MOVES_{FROM,TO}_VEC from insn condition.
 301         (movdi_interunit splitters): Ditto.
 302         * config/i386/constraints.md (Ye): Remove.
 303         (Yd): Do not depend on TARGET_INTER_UNIT_MOVES_{FROM,TO}_VEC.
 304
 305 2018-04-27  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
 306
 307         PR target/85512
 308         * config/aarch64/constraints.md (Usg): Limit to 31.
 309         (Usj): Limit to 63.
 310
 311 2018-04-27  Jakub Jelinek  <jakub@redhat.com>
 312
 313         PR tree-optimization/85529
 314         * tree-ssa-reassoc.c (optimize_range_tests_var_bound): Add FIRST_BB
 315         argument.  Don't call get_nonzero_bits if opcode is ERROR_MARK_NODE,
 316         rhs2 def stmt's bb is dominated by first_bb and it isn't an obvious
 317         zero extension or masking of the MSB bit.
 318         (optimize_range_tests): Add FIRST_BB argument, pass it through
 319         to optimize_range_tests_var_bound.
 320         (maybe_optimize_range_tests, reassociate_bb): Adjust
 321         optimize_range_tests callers.
 322
 323 2018-04-26  Richard Biener  <rguenther@suse.de>
 324             Jakub Jelinek  <jakub@redhat.com>
 325
 326         * cgraph.h (symbol_table): Just declare debug method here.
 327         * symtab.c (symbol_table::debug): Define.
 328
 329 2018-04-26  Eric Botcazou  <ebotcazou@adacore.com>
 330
 331         * loop-invariant.c (may_assign_reg_p): Return false for frame pointer.
 332
 333 2018-04-26  Uros Bizjak  <ubizjak@gmail.com>
 334
 335         * config/i386/i386.md ("isa" attribute): Add x64_sse2.
 336         ("enabled" attribute): Handle x64_sse2 "isa" attribute.
 337         (*movdi_internal): Substitute Yi and Yj constraint with x
 338         and Ym and Yn constraint with y constraint.  Update "isa"
 339         attribute and set "preferred_for_speed" attribute from
 340         TARGET_INTER_UNIT_MOVES_{FROM,TO}_VEC for updated alternatives.
 341         (*movsi_internal): Ditto.
 342         (*movdf_internal): Ditto.
 343         (*movsf_internal): Ditto.
 344         (*zero_extendsidi2): Ditto.
 345         * config/i386/sse.md (vec_set<mode>_0): Ditto.
 346         (sse2_loadld): Ditto.
 347         (*vec_extract<ssevecmodelower>_0): Ditto.
 348         (*vec_extractv4si_0_zext_sse4): Ditto.
 349         (vec_concatv2di): Ditto.
 350         (*vec_dup<mode>): Ditto.
 351         * config/i386/mmx.md (*mov<mode>_internal): Ditto.
 352         * config/i386/constraints.md (Yi): Remove.
 353         (Yj): Remove.
 354         (Ym): Remove.
 355         (Yn): Remove.
 356
 357 2018-04-26  Nathan Sidwell  <nathan@acm.org>
 358
 359         * dumpfile.c (dump_open): New.
 360         (dump_open_alternate_stream, dump_start, dump_begin): Call it.
 361         (dump_finish): Detect stdio/stderr by value not name.
 362
 363 2018-04-26  Jonathan Wakely  <jwakely@redhat.com>
 364
 365         * doc/invoke.texi (-Wreturn-type): Document default status for C++.
 366
 367 2018-04-26  Tom de Vries  <tom@codesourcery.com>
 368
 369         PR target/84952
 370         * config/nvptx/nvptx.c (verify_neutering_jumps)
 371         (verify_neutering_labels): New function
 372         (nvptx_single): Use verify_neutering_jumps and verify_neutering_labels.
 373
 374 2018-04-26  Tom de Vries  <tom@codesourcery.com>
 375
 376         PR target/84025
 377         * config/nvptx/nvptx.c (needs_neutering_p): New function.
 378         (nvptx_single): Use needs_neutering_p to skip over insns that do not
 379         need neutering.
 380
 381 2018-04-26  Richard Biener <rguenther@suse.de>
 382             Tom de Vries  <tom@codesourcery.com>
 383
 384         PR lto/85422
 385         * lto-streamer-out.c (output_function): Fixup loops if required to match
 386         discovery done in the reader.
 387
 388 2018-04-26  Richard Biener  <rguenther@suse.de>
 389
 390         PR tree-optimization/85116
 391         * tree-ssa-loop-ch.c (do_while_loop_p): A do-while loop should
 392         have a loop exit from the single latch predecessor.  Remove
 393         case of header with just condition.
 394         (ch_base::copy_headers): Exclude infinite loops from any
 395         processing.
 396         (pass_ch::execute): Record exits.
 397
 398 2018-04-26  Richard Biener  <rguenther@suse.de>
 399
 400         * tree-vect-data-refs.c (vect_get_data_access_cost): Get
 401         prologue cost vector and pass it to vect_get_load_cost.
 402         (vect_get_peeling_costs_all_drs): Likewise.
 403         (vect_peeling_hash_get_lowest_cost): Likewise.
 404         (vect_enhance_data_refs_alignment): Likewise.
 405
 406 2018-04-26  Richard Biener  <rguenther@suse.de>
 407
 408         PR middle-end/85450
 409         * tree-cfg.c (verify_gimple_assign_unary): Restore proper
 410         checking of integer<->pointer conversions.
 411         * omp-expand.c (expand_omp_for_static_nochunk): Avoid
 412         sign-/zero-extending pointer types.
 413         (expand_omp_for_static_chunk): Likewise.
 414
 415 2018-03-22  Hans-Peter Nilsson  <hp@axis.com>
 416             Jean Lee  <xiaoyur347@gmail.com>
 417
 418         * config/mips/mips.c (mips_asan_shadow_offset): New function.
 419         (TARGET_ASAN_SHADOW_OFFSET): Define.
 420         * config/mips/mips.h (FRAME_GROWS_DOWNWARD): Augment to also be
 421         true for -fsanitize=address.
 422
 423 2018-04-25  Mark Wielaard  <mark@klomp.org>
 424
 425         * dwarf2out.c (file_info_cmp): Sort longer dir prefixes before
 426         shorter ones.
 427
 428 2018-04-25  Jakub Jelinek  <jakub@redhat.com>
 429
 430         * config/i386/i386.md (*x86_mov<mode>cc_0_m1): Use type "alu1" rather
 431         than "alu", remove explicit "memory" and "imm_disp" attributes.
 432         (*x86_mov<mode>cc_0_m1_se, *x86_mov<mode>cc_0_m1_neg): Likewise.
 433
 434         PR middle-end/85414
 435         * simplify-rtx.c (simplify_unary_operation_1) <case SIGN_EXTEND,
 436         case ZERO_EXTEND>: Pass SUBREG_REG (op) rather than op to
 437         gen_lowpart_no_emit.
 438
 439 2018-04-25  Sebastian Peryt  <sebastian.peryt@intel.com>
 440
 441         PR target/85473
 442         * config/i386/i386.c (ix86_expand_builtin): Change memory
 443         operand to XI, extend p0 to Pmode.
 444         * config/i386/i386.md: Change unspec volatile and operand
 445         1 mode to XI, change operand 0 mode to P.
 446
 447 2018-04-25  Chung-Ju Wu  <jasonwucj@gmail.com>
 448
 449         * config/nds32/nds32-predicates.c (nds32_can_use_bclr_p): Mask with
 450         GET_MODE_MASK before any checking.
 451         (nds32_can_use_bset_p): Likewise.
 452         (nds32_can_use_btgl_p): Likewise.
 453
 454 2018-04-25  Chung-Ju Wu  <jasonwucj@gmail.com>
 455
 456         * config/nds32/nds32-doubleword.md: New define_split pattern for
 457         illegal register number.
 458
 459 2018-04-25  Chung-Ju Wu  <jasonwucj@gmail.com>
 460
 461         * config/nds32/nds32.c (nds32_print_operand): Set op_value ealier.
 462
 463 2018-04-25  Chung-Ju Wu  <jasonwucj@gmail.com>
 464
 465         * config/nds32/nds32.h (ASM_APP_ON): Add missing newline character.
 466
 467 2018-04-25  Richard Biener  <rguenther@suse.de>
 468
 469         * lto-streamer.h (LTO_major_version): Bump to 8.
 470
 471 2018-04-25  Jakub Jelinek  <jakub@redhat.com>
 472
 473         * BASE-VER: Set to 9.0.0.
 474
 475 2018-04-24  Segher Boessenkool  <segher@kernel.crashing.org>
 476
 477         * config/rs6000/rs6000.c (init_float128_ieee): Fix spelling mistakes
 478         in __abskf2 and __powikf2.
 479
 480 2018-04-24  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
 481
 482         PR target/85512
 483         * config/aarch64/constraints.md (Usg, Usj): New constraints.
 484         * config/aarch64/iterators.md (cmode_simd): New mode attribute.
 485         * config/aarch64/aarch64.md (*aarch64_ashr_sisd_or_int_<mode>3):
 486         Use the above on operand 2.  Reindent.
 487         (*aarch64_lshr_sisd_or_int_<mode>3): Likewise.
 488
 489 2018-04-24  H.J. Lu  <hongjiu.lu@intel.com>
 490
 491         PR target/85485
 492         * common/config/i386/i386-common.c (ix86_handle_option): Don't
 493         handle OPT_mcet.
 494         * config/i386/i386.opt (mcet): Removed.
 495         * doc/install.texi: Remove -mcet documentation.
 496         * doc/invoke.texi: Likewise.
 497
 498 2018-04-24  H.J. Lu  <hongjiu.lu@intel.com>
 499
 500         PR target/85485
 501         * doc/install.texi: Remove -mcet from bootstrap-cet.
 502
 503 2018-04-24  Jakub Jelinek  <jakub@redhat.com>
 504
 505         PR target/85511
 506         * config/i386/i386.c (ix86_init_mmx_sse_builtins): Don't define
 507         __builtin_ia32_readeflags_u32 and __builtin_ia32_writeeflags_u32
 508         if TARGET_64BIT.
 509
 510         PR target/85503
 511         * config/rs6000/rs6000-p8swap.c (const_load_sequence_p): Punt if
 512         const_vector is not CONST_VECTOR or SYMBOL_REF for a constant pool
 513         containing a CONST_VECTOR.
 514
 515 2018-04-24  Cesar Philippidis  <cesar@codesourcery.com>
 516
 517         * doc/install.texi: Update newlib dependency for nvptx.
 518
 519 2018-04-24  Jakub Jelinek  <jakub@redhat.com>
 520
 521         PR target/85508
 522         * config/i386/i386.c (ix86_expand_vector_init_one_var): Use UINTVAL
 523         instead of INTVAL when shifting x left.
 524
 525 2018-04-24  Andreas Krebbel  <krebbel@linux.ibm.com>
 526
 527         PR tree-optimization/85478
 528         * tree-vect-loop.c (vect_analyze_loop_2): Do not call
 529         vect_grouped_store_supported for single element vectors.
 530
 531 2018-04-24  Richard Biener  <rguenther@suse.de>
 532
 533         PR target/85491
 534         * config/i386/i386.c (ix86_add_stmt_cost): Restrict strided
 535         load cost increase to the case of non-constant step.
 536
 537 2018-04-24  Jakub Jelinek  <jakub@redhat.com>
 538
 539         PR target/84828
 540         * reg-stack.c (move_for_stack_reg): Don't fail assertion about dead
 541         destination if any_malformed_asm.
 542
 543 2018-04-23  Eric Botcazou  <ebotcazou@adacore.com>
 544
 545         PR middle-end/85496
 546         * expr.c (store_field): In the bitfield case, if the value comes from
 547         a function call and is returned in registers by means of a PARALLEL,
 548         do not change the mode of the temporary unless BLKmode and VOIDmode.
 549
 550 2018-04-23  Andrey Belevantsev  <abel@ispras.ru>
 551
 552         PR rtl-optimization/85423
 553         * sel-sched-ir.c (has_dependence_note_mem_dep): Only discard
 554         dependencies to debug insns when the previous insn is non-debug.
 555
 556 2018-04-23  Claudiu Zissulescu  <claziss@synopsys.com>
 557
 558         * config/arc/arc.md ("vunspec"): Delete it, unify all the unspec
 559         enums into a single definition.
 560         (fls): Fix predicates and printing.
 561         (seti): Likewise.
 562
 563 2018-04-23  Claudiu Zissulescu  <claziss@synopsys.com>
 564
 565         * config/arc/arc-protos.h (check_if_valid_sleep_operand): Remove.
 566         * config/arc/arc.c (arc_expand_builtin): Sleep accepts registers
 567         and short u6 immediate.
 568         (check_if_valid_sleep_operand): Remove.
 569         * config/arc/arc.md (Sleep): Accepts registers and u6 immediates.
 570
 571 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
 572
 573         * config/nds32/nds32.c (nds32_compute_stack_frame): Consider
 574         flag_always_save_lp condition.
 575         * config/nds32/nds32.opt (malways-save-lp): New option.
 576
 577 2018-04-22  Shiva Chen  <shiva0217@gmail.com>
 578
 579         * config/nds32/nds32-protos.h (nds32_use_load_post_increment): Declare.
 580         * config/nds32/nds32.c (nds32_use_load_post_increment): New.
 581         * config/nds32/nds32.h
 582         (USE_LOAD_POST_INCREMENT, USE_LOAD_POST_DECREMENT): Define.
 583         (USE_STORE_POST_INCREMENT, USE_STORE_POST_DECREMENT): Define.
 584
 585 2018-04-22  Shiva Chen  <shiva0217@gmail.com>
 586
 587         * config/nds32/nds32-protos.h (nds32_ls_333_p): Remove.
 588         * config/nds32/nds32.c (nds32_ls_333_p): Remove.
 589
 590 2018-04-22  Shiva Chen  <shiva0217@gmail.com>
 591             Chung-Ju Wu  <jasonwucj@gmail.com>
 592
 593         * config/nds32/nds32-protos.h (nds32_case_vector_shorten_mode):
 594         Declare.
 595         * config/nds32/nds32.c (nds32_case_vector_shorten_mode): New function.
 596         * config/nds32/nds32.h (CASE_VECTOR_SHORTEN_MODE): Modify.
 597
 598 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
 599
 600         * config/nds32/nds32.c (nds32_compute_stack_frame): Fix wrong value.
 601
 602 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
 603
 604         * config/nds32/nds32-protos.h (nds32_data_alignment,
 605         nds32_local_alignment): Declare.
 606         * config/nds32/nds32.c (nds32_data_alignment, nds32_constant_alignment,
 607         nds32_local_alignment): New functions.
 608         (TARGET_CONSTANT_ALIGNMENT): Define.
 609         * config/nds32/nds32.h (DATA_ALIGNMENT, LOCAL_ALIGNMENT): Define.
 610
 611 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
 612
 613         * config/nds32/nds32.c
 614         (TARGET_HARD_REGNO_MODE_OK): Move to the bottom of file.
 615         (TARGET_MODES_TIEABLE_P): Likewise.
 616
 617 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
 618
 619         * config/nds32/nds32.c (nds32_asm_file_start): Display optimization
 620         level Ofast and Og.
 621
 622 2018-04-22  Monk Chiang  <sh.chiang04@gmail.com>
 623             Chung-Ju Wu  <jasonwucj@gmail.com>
 624
 625         * config/nds32/constants.md (unspec_volatile_element): Add enum values
 626         for unaligned access.
 627         * config/nds32/nds32-intrinsic.c: Implementation of expanding
 628         unaligned access.
 629         * config/nds32/nds32-intrinsic.md: Likewise.
 630         * config/nds32/nds32_intrinsic.h: Likewise.
 631         * config/nds32/nds32.h (nds32_builtins): Likewise.
 632         * config/nds32/nds32.opt (munaligned-access): New option.
 633         * config/nds32/nds32.c (nds32_asm_file_start): Display
 634         flag_unaligned_access status.
 635
 636 2018-04-20  Kito Cheng  <kito.cheng@gmail.com>
 637
 638         * config/riscv/elf.h (LINK_SPEC): Pass --no-relax if
 639         -mno-relax is present.
 640         * config/riscv/linux.h (LINK_SPEC): Ditto.
 641
 642 2018-04-20  Martin Sebor  <msebor@redhat.com>
 643
 644         PR c/85365
 645         * gimple-fold.c (gimple_fold_builtin_strcpy): Suppress -Wrestrict
 646         for null pointers.
 647         (gimple_fold_builtin_stxcpy_chk): Same.
 648         * gimple-ssa-warn-restrict.c (check_bounds_or_overlap): Same.
 649
 650 2018-04-20  Michael Meissner  <meissner@linux.ibm.com>
 651
 652         PR target/85456
 653         * config/rs6000/rs6000.c (init_float128_ieee): Add support to call
 654         __powikf2 when long double is IEEE 128-bit.
 655
 656 2018-04-20  Kito Cheng  <kito.cheng@gmail.com>
 657
 658         * config/riscv/riscv.c (riscv_first_stack_step): Round up min
 659         step to make sure stack always aligned.
 660
 661 2018-04-20  Carl Love  <cel@us.ibm.com>
 662
 663         PR target/83402
 664         * config/rs6000/rs6000-c.c (rs6000_gimple_fold_builtin): Add
 665         size check for arg0.
 666
 667 2018-04-20  Nathan Sidwell  <nathan@codesourcery.com>
 668             Tom de Vries  <tom@codesourcery.com>
 669
 670         PR target/85445
 671         * config/nvptx/nvptx.c (nvptx_emit_forking, nvptx_emit_joining):
 672         Emit insns for calls too.
 673         (nvptx_find_par): Always look for worker-level predecessor insn.
 674         (nvptx_propagate): Add is_call parm, return bool.  Copy frame for
 675         calls.
 676         (nvptx_vpropagate, nvptx_wpropagate): Adjust.
 677         (nvptx_process_pars): Propagate frames for calls.
 678
 679 2018-04-20  H.J. Lu  <hongjiu.lu@intel.com>
 680
 681         PR target/85469
 682         * common/config/i386/i386-common.c (OPTION_MASK_ISA_IBT_SET):
 683         Removed.
 684         (OPTION_MASK_ISA_IBT_UNSET): Likewise.
 685         (ix86_handle_option): Don't handle OPT_mibt.
 686         * config/i386/cet.h: Check __CET__ instead of __IBT__ and
 687         __SHSTK__.
 688         * config/i386/driver-i386.c (host_detect_local_cpu): Remove
 689         has_ibt and ibt.
 690         * config/i386/i386-c.c (ix86_target_macros_internal): Don't
 691         check OPTION_MASK_ISA_IBT nor flag_cf_protection.
 692         (ix86_target_macros): Define __CET__ with flag_cf_protection
 693         for -fcf-protection.
 694         * config/i386/i386.c (isa2_opts): Remove -mibt.
 695         * config/i386/i386.h (TARGET_IBT): Removed.
 696         (TARGET_IBT_P): Likewise.
 697         (ix86_valid_target_attribute_inner_p): Don't check OPT_mibt.
 698         * config/i386/i386.md (nop_endbr): Don't check TARGET_IBT.
 699         * config/i386/i386.opt (mcet): Update help message.
 700         (mshstk): Likewise.
 701         (mibt): Removed.
 702         * doc/invoke.texi: Remove -mibt.  Document __CET__.  Document
 703         -mcet as an alias for -mshstk.
 704
 705 2018-04-20  Richard Biener <rguenther@suse.de>
 706
 707         PR middle-end/85475
 708         * match.pd ((X * CST) * Y -> (X * Y) * CST): Avoid exponential
 709         complexity by forcing a single use of the multiply operand.
 710
 711 2018-04-20  Martin Jambor  <mjambor@suse.cz>
 712
 713         ipa/85449
 714         * ipa-cp.c (cgraph_edge_brings_value_p): Move check for self-feeding
 715         recursion dependency to only apply to non-clones.
 716
 717 2018-04-20  Martin Jambor  <mjambor@suse.cz>
 718
 719         ipa/85447
 720         * ipa-cp.c (create_specialized_node): Check that clones of
 721         self-recursive edges exist during IPA-CP.
 722
 723 2018-04-19  Toon Moene  <toon@moene.org>
 724
 725         * doc/invoke.texi: Add -floop-unroll-and-jam to options enabled
 726         by -O3.
 727
 728 2018-04-19  Jakub Jelinek  <jakub@redhat.com>
 729
 730         PR tree-optimization/85467
 731         * fold-const.c (fold_ternary_loc) <case BIT_FIELD_REF>: Use
 732         VECTOR_TYPE_P macro.  If type is vector type, VIEW_CONVERT_EXPR the
 733         VECTOR_CST element to type.
 734
 735 2018-04-19  H.J. Lu  <hongjiu.lu@intel.com>
 736
 737         PR target/85397
 738         * config/i386/i386.h (STACK_SAVEAREA_MODE): New.
 739         * config/i386/i386.md (builtin_setjmp_setup): Removed.
 740         (builtin_longjmp): Likewise.
 741         (save_stack_nonlocal): New pattern.
 742         (restore_stack_nonlocal): Likewise.
 743
 744 2018-04-19  H.J. Lu  <hongjiu.lu@intel.com>
 745
 746         PR target/85404
 747         * config/i386/cet.c (file_end_indicate_exec_stack_and_cet):
 748         Replace ASM_OUTPUT_LABEL with fprintf.
 749
 750 2018-04-19  H.J. Lu  <hongjiu.lu@intel.com>
 751
 752         PR target/85417
 753         * config/i386/cet.c (file_end_indicate_exec_stack_and_cet):
 754         Check flag_cf_protection instead of TARGET_IBT and TARGET_SHSTK.
 755         * config/i386/i386-c.c (ix86_target_macros_internal): Also
 756         define __IBT__ and __SHSTK__ for -fcf-protection.
 757         * config/i386/i386.c (pass_insert_endbranch::gate): Don't check
 758         TARGET_IBT.
 759         (ix86_trampoline_init): Likewise.
 760         (x86_output_mi_thunk): Likewise.
 761         (ix86_notrack_prefixed_insn_p): Likewise.
 762         (ix86_option_override_internal): Don't disallow -fcf-protection.
 763         * config/i386/i386.md (rdssp<mode>): Also enable for
 764         -fcf-protection.
 765         (incssp<mode>): Likewise.
 766         (nop_endbr): Likewise.
 767         * config/i386/i386.opt (mcet): Change help message to built-in
 768         functions only.
 769         (mibt): Likewise.
 770         (mshstk): Likewise.
 771         * doc/invoke.texi: Remove -mcet, -mibt and -mshstk condition
 772         on -fcf-protection.  Change -mcet, -mibt and -mshstk to only
 773         enable CET built-in functions.
 774
 775 2018-04-19  Sebastian Peryt  <sebastian.peryt@intel.com>
 776
 777         * common/config/i386/i386-common.c
 778         (OPTION_MASK_ISA_MOVDIRI_SET, OPTION_MASK_ISA_MOVDIR64B_SET,
 779         OPTION_MASK_ISA_MOVDIRI_UNSET,
 780         OPTION_MASK_ISA_MOVDIR64B_UNSET): New defines.
 781         (ix86_handle_option): Handle -mmovdiri and -mmovdir64b.
 782         * config.gcc (movdirintrin.h): New header.
 783         * config/i386/cpuid.h (bit_MOVDIRI,
 784         bit_MOVDIR64B): New bits.
 785         * config/i386/driver-i386.c (host_detect_local_cpu): Detect -mmovdiri
 786         and -mmvodir64b.
 787         * config/i386/i386-builtin-types.def ((VOID, PUNSIGNED, UNSIGNED),
 788         (VOID, PVOID, PCVOID)): New function types.
 789         * config/i386/i386-builtin.def (__builtin_ia32_directstoreu_u32,
 790         __builtin_ia32_directstoreu_u64,
 791          __builtin_ia32_movdir64b): New builtins.
 792         * config/i386/i386-c.c (__MOVDIRI__, __MOVDIR64B__): New.
 793         * config/i386/i386.c (ix86_target_string): Added -mmovdir64b
 794         and -mmovdiri.
 795         (ix86_valid_target_attribute_inner_p): Ditto.
 796         (ix86_expand_special_args_builtin): Added VOID_FTYPE_PUNSIGNED_UNSIGNED
 797         and VOID_FTYPE_PUNSIGNED_UNSIGNED.
 798         (ix86_expand_builtin): Expand IX86_BUILTIN_MOVDIR64B.
 799         * config/i386/i386.h (TARGET_MOVDIRI, TARGET_MOVDIRI_P,
 800         TARGET_MOVDIR64B, TARGET_MOVDIR64B_P): New.
 801         * config/i386/i386.md (UNSPECV_MOVDIRI, UNSPECV_MOVDIR64B): New.
 802         (movdiri<mode>, movdir64b_<mode>): New.
 803         * config/i386/i386.opt: Add -mmovdiri and -mmovdir64b.
 804         * config/i386/immintrin.h: Include movdirintrin.h.
 805         * config/i386/movdirintrin.h: New file.
 806         * doc/invoke.texi: Added -mmovdiri and -mmovdir64b.
 807
 808 2018-04-19  Richard Biener  <rguenther@suse.de>
 809
 810         PR middle-end/85455
 811         * cfg.c (clear_bb_flags): When loop state says we have
 812         marked irreducible regions also preserve BB_IRREDUCIBLE_LOOP.
 813
 814 2018-04-19  Richard Biener  <rguenther@suse.de>
 815
 816         PR tree-optimization/84737
 817         * tree-vect-data-refs.c (vect_copy_ref_info): New function
 818         copying restrict info.
 819         (vect_setup_realignment): Use it.
 820         * tree-vectorizer.h (vect_copy_ref_info): Declare.
 821         * tree-vect-stmts.c (vectorizable_store): Copy ref info from
 822         the first DR to all generated stores.
 823         (vectorizable_load): Likewise for loads.
 824
 825 2018-04-19  Jakub Jelinek  <jakub@redhat.com>
 826
 827         PR tree-optimization/85446
 828         * match.pd ((intptr_t) x eq/ne CST to x eq/ne (typeof x) cst): Require
 829         the integral and pointer types to have the same precision.
 830
 831         * doc/install.texi: Document --disable-cet being the default and
 832         --enable-cet=auto.
 833
 834 2018-04-18  Martin Liska  <mliska@suse.cz>
 835
 836         * ipa-devirt.c (odr_subtypes_equivalent_p): Fix GNU coding
 837         style.
 838
 839 2018-04-18  Martin Liska  <mliska@suse.cz>
 840
 841         Revert
 842         2018-03-02  Eric Botcazou  <ebotcazou@adacore.com>
 843
 844         PR ipa/83983
 845         * ipa-devirt.c (odr_subtypes_equivalent_p): Get the ODR type of both
 846         arguments if they are comparable.
 847
 848 2018-04-18  Martin Liska  <mliska@suse.cz>
 849
 850         Revert
 851         2018-03-13  Eric Botcazou  <ebotcazou@adacore.com>
 852
 853         PR lto/84805
 854         * ipa-devirt.c (odr_subtypes_equivalent_p): Do not get the ODR type of
 855         incomplete types.
 856
 857 2018-04-18  H.J. Lu  <hongjiu.lu@intel.com>
 858
 859         PR target/85388
 860         * config/i386/i386.c (ix86_expand_split_stack_prologue): Insert
 861         ENDBR after calling __morestack.
 862
 863 2018-04-18  David Malcolm  <dmalcolm@redhat.com>
 864
 865         PR jit/85384
 866         * configure.ac (gcc-driver-name.h): Honor --with-gcc-major-version
 867         by using gcc_base_ver to generate a gcc_driver_version, and use
 868         it when generating GCC_DRIVER_NAME.
 869         * configure: Regenerate.
 870
 871 2018-04-18  Jakub Jelinek  <jakub@redhat.com>
 872
 873         PR target/81084
 874         * config.gcc: Obsolete powerpc*-*-*spe*.
 875
 876 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
 877
 878         PR debug/84637
 879         * dbxout.c (dbxout_int): Perform negation in unsigned int type.
 880         (stabstr_D): Change type of unum from unsigned int to
 881         unsigned HOST_WIDE_INT.  Perform negation in unsigned HOST_WIDE_INT
 882         type.
 883
 884 2018-04-17  Jim Wilson  <jimw@sifive.com>
 885
 886         PR 84856
 887         * config/riscv/riscv.c (riscv_compute_frame_info): Add calls to
 888         RISCV_STACK_ALIGN when using outgoing_args_size and pretend_args_size.
 889         Set arg_pointer_offset after using pretend_args_size.
 890
 891 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
 892
 893         PR rtl-optimization/85431
 894         * dse.c (record_store): Ignore zero width stores.
 895
 896         PR sanitizer/85230
 897         * asan.c (handle_builtin_stack_restore): Adjust comment.  Emit
 898         __asan_allocas_unpoison call and last_alloca_addr = new_sp before
 899         __builtin_stack_restore rather than after it.
 900         * builtins.c (expand_asan_emit_allocas_unpoison): Pass
 901         arg1 + (virtual_dynamic_stack_rtx - stack_pointer_rtx) as second
 902         argument instead of virtual_dynamic_stack_rtx.
 903
 904 2018-04-17  Kelvin Nilsen  <kelvin@gcc.gnu.org>
 905
 906         * config/rs6000/rs6000-protos.h (rs6000_builtin_is_supported_p):
 907         New prototype.
 908         * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin):
 909         Add note to error message to explain internal mapping of overloaded
 910         built-in function name to non-overloaded built-in function name.
 911         * config/rs6000/rs6000.c (rs6000_builtin_is_supported_p): New
 912         function.
 913
 914 2018-04-17  Michael Meissner  <meissner@linux.vnet.ibm.com>
 915
 916         PR target/85424
 917         * config/rs6000/rs6000.md (pack<mode>): Do not try handle a pack
 918         where the inputs overlap with the output.
 919
 920 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
 921
 922         * config/i386/sse.md (vec_extract_lo_<mode><mask_name>): Add
 923         (=v, v) alternative and explicit "memory" attribute.
 924         (vec_extract_lo_<mode><mask_name>): Likewise.  Also add
 925         "type", "prefix", "prefix_extra", "length_immediate" and "mode"
 926         attributes.
 927         (vec_extract_lo_<mode><mask_name>): Add (=v, v) alternative and use
 928         "sselog1" type instead of "sselog".
 929         (vec_extract_hi_<mode><mask_name>): Use "sselog1" type instead of
 930         "sselog".  Remove explicit "memory" attribute.
 931         (vec_extract_lo_v32hi): Add (=v, v) alternative and explicit "memory",
 932         "type", "prefix", "prefix_extra", "length_immediate" and "mode"
 933         attributes.
 934         (vec_extract_hi_v32hi): Merge all alternatives into one, use
 935         "sselog1" type instead of "sselog".  Remove explicit "memory"
 936         attribute.
 937         (vec_extract_hi_v16hi): Merge each pair of alternatives into one,
 938         use "sselog1" type instead of "sselog".  Remove explicit "memory"
 939         attribute.
 940         (vec_extract_lo_v64qi): Add (=v, v) alternative and explicit "memory",
 941         "type", "prefix", "prefix_extra", "length_immediate" and "mode"
 942         attributes.
 943         (vec_extract_hi_v64qi): Merge all alternatives into one, use
 944         "sselog1" type instead of "sselog".  Remove explicit "memory"
 945         attribute.
 946         (vec_extract_hi_v32qi): Merge each pair of alternatives into one,
 947         use "sselog1" type instead of "sselog".  Remove explicit "memory"
 948         attribute.
 949
 950         PR target/85430
 951         * config/i386/i386.md (*ashlqi3_1_slp): Use alu1 type instead of alu.
 952
 953         PR middle-end/85414
 954         * rtlhooks.c (gen_lowpart_if_possible): Don't call gen_lowpart_SUBREG
 955         on a SUBREG.
 956
 957 2018-04-17  Martin Jambor  <mjambor@suse.cz>
 958
 959         PR ipa/85421
 960         * ipa-cp.c (create_specialized_node): Call
 961         expand_all_artificial_thunks if necessary.
 962
 963 2018-04-17  Martin Liska  <mliska@suse.cz>
 964
 965         PR lto/85405
 966         * ipa-devirt.c (odr_types_equivalent_p): Remove trailing
 967         in message, remote space in between '_G' and '('.
 968
 969 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
 970
 971         PR target/85281
 972         * config/i386/sse.md (reduces<mode><mask_scalar_name>,
 973         avx512f_vmcmp<mode>3<round_saeonly_name>,
 974         avx512f_vmcmp<mode>3_mask<round_saeonly_name>,
 975         avx512f_sgetexp<mode><mask_scalar_name><round_saeonly_scalar_name>,
 976         avx512f_rndscale<mode><round_saeonly_name>,
 977         avx512dq_ranges<mode><mask_scalar_name><round_saeonly_scalar_name>,
 978         avx512f_vgetmant<mode><mask_scalar_name><round_saeonly_scalar_name>):
 979         Use %<iptr>2 instead of %2 for -masm=intel.
 980         (avx512f_vcvtss2usi<round_name>, avx512f_vcvtss2usiq<round_name>,
 981         avx512f_vcvttss2usi<round_saeonly_name>,
 982         avx512f_vcvttss2usiq<round_saeonly_name>): Use %k1 instead of %1 for
 983         -masm=intel.
 984         (avx512f_vcvtsd2usi<round_name>, avx512f_vcvtsd2usiq<round_name>,
 985         avx512f_vcvttsd2usi<round_saeonly_name>,
 986         avx512f_vcvttsd2usiq<round_saeonly_name>, ufloatv2siv2df2<mask_name>):
 987         Use %q1 instead of %1 for -masm=intel.
 988         (avx512f_sfixupimm<mode><sd_maskz_name><round_saeonly_name>,
 989         avx512f_sfixupimm<mode>_mask<round_saeonly_name>): Use %<iptr>3 instead
 990         of %3 for -masm=intel.
 991         (sse2_shufpd_v2df_mask): Fix a typo, change %{6%} to %{%6%} for
 992         -masm=intel.
 993         (*avx512vl_<code>v2div2qi2_store): Use %w0 instead of %0 for
 994         -masm=intel.
 995         (*avx512vl_<code><mode>v4qi2_store): Use %k0 instead of %0 for
 996         -masm=intel.
 997         (avx512vl_<code><mode>v4qi2_mask_store): Use a single pattern with
 998         %k0 and %1 for -masm=intel rather than two patterns, one with %0 and
 999         %g1.
1000         (*avx512vl_<code><mode>v8qi2_store): Use %q0 instead of %0 for
1001         -masm=intel.
1002         (avx512vl_<code><mode>v8qi2_mask_store): Use a single pattern with
1003         %q0 and %1 for -masm=intel rather than two patterns, one with %0 and
1004         %g1 and one with %0 and %1.
1005         (avx512er_vmrcp28<mode><round_saeonly_name>,
1006         avx512er_vmrsqrt28<mode><round_saeonly_name>): Use %<iptr>1 instead of
1007         %1 for -masm=intel.
1008         (avx5124fmaddps_4fmaddps_mask, avx5124fmaddps_4fmaddss_mask,
1009         avx5124fmaddps_4fnmaddps_mask, avx5124fmaddps_4fnmaddss_mask,
1010         avx5124vnniw_vp4dpwssd_mask, avx5124vnniw_vp4dpwssds_mask): Swap order
1011         of %0 and %{%4%} for -masm=intel.
1012         (avx5124fmaddps_4fmaddps_maskz, avx5124fmaddps_4fmaddss_maskz,
1013         avx5124fmaddps_4fnmaddps_maskz, avx5124fmaddps_4fnmaddss_maskz,
1014         avx5124vnniw_vp4dpwssd_maskz, avx5124vnniw_vp4dpwssds_maskz): Swap
1015         order of %0 and %{%5%}%{z%} for -masm=intel.
1016
1017 2018-04-17  Jan Hubicka  <jh@suse.cz>
1018
1019         PR lto/85405
1020         * ipa-devirt.c (odr_types_equivalent_p): Handle bit fields.
1021
1022 2018-04-17  Martin Liska  <mliska@suse.cz>
1023
1024         PR ipa/85329
1025         * multiple_target.c (create_dispatcher_calls): Set apostrophes
1026         for target_clone error message.  Make default implementation
1027         clone to be a local declaration.
1028         (separate_attrs): Add new argument and check for an empty
1029         string.
1030         (expand_target_clones): Handle it.
1031         (ipa_target_clone): Make redirection just for target_clones
1032         functions.
1033
1034 2018-04-16  Cesar Philippidis  <cesar@codesourcery.com>
1035             Tom de Vries  <tom@codesourcery.com>
1036
1037         PR middle-end/84955
1038         * omp-expand.c (expand_oacc_for): Add dummy false branch for
1039         tiled basic blocks without omp continue statements.
1040
1041 2018-04-16  Aaron Sawdey  <acsawdey@linux.ibm.com>
1042
1043         PR target/83660
1044         * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Mark
1045         vec_extract expression as having side effects to make sure it gets
1046         a cleanup point.
1047
1048 2018-04-16  H.J. Lu  <hongjiu.lu@intel.com>
1049
1050         PR target/85403
1051         * config/i386/i386.c (get_builtin_code_for_version): Check
1052         error_mark_node.
1053
1054 2018-04-16  Olga Makhotina  <olga.makhotina@intel.com>
1055
1056         PR target/84331
1057         * gcc/config.gcc: Support "skylake".
1058         * gcc/config/i386/i386-c.c (ix86_target_macros_internal): Handle
1059         PROCESSOR_SKYLAKE.
1060         * gcc/config/i386/i386.c (m_SKYLAKE): Define.
1061         (processor_target_table): Add "skylake".
1062         (ix86_option_override_internal): Add "skylake".
1063         (get_builtin_code_for_version): Handle PROCESSOR_SKYLAKE,
1064         PROCESSOR_CANNONLAKE.
1065         (get_builtin_code_for_version): Fix priority for
1066         PROCESSOR_ICELAKE_CLIENT, PROCESSOR_ICELAKE_SERVER,
1067         PROCESSOR_SKYLAKE-AVX512.
1068         * gcc/config/i386/i386.h (processor_costs): Define TARGET_SKYLAKE.
1069         (processor_type): Add PROCESSOR_SKYLAKE.
1070
1071 2018-04-16  Paolo Carlini  <paolo.carlini@oracle.com>
1072             Jason Merrill  <jason@redhat.com>
1073
1074         PR c++/85112
1075         * convert.c (convert_to_integer_1): Use direct recursion for
1076         enumeral types and types with a precision less than the number
1077         of bits in their mode.
1078
1079 2018-04-16  Julia Koval  <julia.koval@intel.com>
1080
1081         PR target/84413
1082         * config/i386/x86-tune.def (X86_TUNE_SSE_UNALIGNED_LOAD_OPTIMAL,
1083         X86_TUNE_SSE_UNALIGNED_STORE_OPTIMAL): Add m_SKYLAKE_AVX512
1084
1085 2018-04-14  Segher Boessenkool  <segher@kernel.crashing.org>
1086
1087         PR target/85293
1088         * config/rs6000/rs6000.opt (mdirect-move): Make deprecated.
1089         * doc/invoke.texi (RS/6000 and PowerPC Options): Remove -mdirect-move
1090         and -mno-direct-move.
1091
1092 2018-04-13  Paul A. Clarke  <pc@us.ibm.com>
1093
1094         PR target/83402
1095         * config/rs6000/emmintrin.h (_mm_slli_epi{16,32,64}):
1096         Ensure that vec_splat_s32 is only called with 0 <= shift < 16.
1097         Ensure negative shifts result in {0}.
1098
1099 2018-04-13  Vladimir Makarov  <vmakarov@redhat.com>
1100
1101         PR rtl-optimization/79916
1102         * config/rs6000/rs6000.c (rs6000_emit_move): Use assigned hard
1103         regs (if any) to define how to gnerate SD moves when LRA is in
1104         progress.
1105
1106 2018-04-13  Jakub Jelinek  <jakub@redhat.com>
1107
1108         PR rtl-optimization/85393
1109         * except.h (expand_dw2_landing_pad_for_region): Remove declaration.
1110         * except.c (expand_dw2_landing_pad_for_region): Make static.
1111         * bb-reorder.c (fix_up_crossing_landing_pad): In new_bb emit just
1112         a label and unconditional jump to old_bb, rather than
1113         expand_dw2_landing_pad_for_region insn(s) and jump to single_succ
1114         basic block.
1115
1116         PR rtl-optimization/85376
1117         * simplify-rtx.c (simplify_const_unary_operation): For CLZ and CTZ and
1118         zero op0, if C?Z_DEFINED_VALUE_AT_ZERO is false, return NULL_RTX
1119         instead of a specific value.
1120
1121 2018-04-13  Jan Hubicka  <hubicka@ucw.cz>
1122             Bin Cheng  <bin.cheng@arm.com>
1123
1124         PR tree-optimization/82965
1125         PR tree-optimization/83991
1126         * cfgloopanal.c (expected_loop_iterations_unbounded): Add
1127         by_profile_only parameter.
1128         * cfgloopmanip.c (scale_loop_profile): Further scale loop's profile
1129         information if the loop was predicted to iterate too many times.
1130         * cfgloop.h (expected_loop_iterations_unbounded): Update prototype
1131
1132 2018-04-13  Jan Hubicka  <hubicka@ucw.cz>
1133
1134         PR lto/71991
1135         * config/i386/i386.c (ix86_can_inline_p): Allow safe transitions for
1136         always inline.
1137
1138 2018-04-13  Martin Liska  <mliska@suse.cz>
1139             Jakub Jelinek  <jakub@redhat.com>
1140
1141         PR middle-end/81657
1142         * expr.h (enum block_op_methods): Add BLOCK_OP_NO_LIBCALL_RET.
1143         * expr.c (emit_block_move_hints): Handle BLOCK_OP_NO_LIBCALL_RET.
1144         * builtins.c (expand_builtin_memory_copy_args): Use
1145         BLOCK_OP_NO_LIBCALL_RET method for mempcpy with non-ignored target,
1146         handle dest_addr == pc_rtx.
1147
1148 2018-04-12  Segher Boessenkool  <segher@kernel.crashing.org>
1149
1150         PR target/85291
1151         * config/rs6000/rs6000.md (fix_trunc<mode>si2): Use legacy code if
1152         asked to not generate direct moves.
1153         (fix_trunc<mode>si2_stfiwx): Similar.
1154         (fix_trunc<mode>si2_internal): Similar.
1155
1156 2018-04-12  Jakub Jelinek  <jakub@redhat.com>
1157
1158         PR debug/83157
1159         * var-tracking.c (add_stores): Handle STRICT_LOW_PART SET_DEST.
1160         * cselib.c (cselib_record_sets): For STRICT_LOW_PART dest,
1161         lookup if dest in some wider mode is known to be const0_rtx and
1162         if so, record permanent equivalence for it to be ZERO_EXTEND of
1163         the narrower mode destination.
1164
1165 2018-04-12  Cesar Philippidis  <cesar@codesourcery.com>
1166
1167         * lto-streamer-out.c (output_function): Revert 259346.
1168         * omp-expand.c (expand_oacc_for): Likewise.
1169
1170 2018-04-12  Alexander Monakov  <amonakov@ispras.ru>
1171
1172         PR rtl-optimization/85354
1173         * sel-sched-ir.c (sel_init_pipelining): Move cfg_cleanup call...
1174         * sel-sched.c (sel_global_init): ... here.
1175
1176 2018-04-12  Eric Botcazou  <ebotcazou@adacore.com>
1177
1178         PR target/85238
1179         * lto-wrapper.c (debug_objcopy): Open the files in binary mode.
1180         * dwarf2out.c (dwarf2out_early_finish): Do not generate assembly in LTO
1181         mode for PE-COFF targets.
1182         * config/i386/i386-protos.h (i386_pe_asm_lto_start): Declare.
1183         (i386_pe_asm_lto_end): Likewise.
1184         * config/i386/cygming.h (TARGET_ASM_LTO_START): Define.
1185         (TARGET_ASM_LTO_END): Likewise.
1186         * config/i386/winnt.c (saved_debug_info_level): New static variable.
1187         (i386_pe_asm_lto_start): New function.
1188         (i386_pe_asm_lto_end): Likewise.
1189
1190 2018-04-12  Cesar Philippidis  <cesar@codesourcery.com>
1191             Richard Biener  <rguenther@suse.de>
1192
1193         PR middle-end/84955
1194         * lto-streamer-out.c (output_function): Fix CFG loop state before
1195         streaming out.
1196         * omp-expand.c (expand_oacc_for): Handle calls to internal
1197         functions like regular functions.
1198
1199 2018-04-12  Richard Biener  <rguenther@suse.de>
1200
1201         PR lto/85371
1202         * dwarf2out.c (init_sections_and_labels): Use debug_line_section[_label]
1203         for the early LTO debug to properly generate references to it
1204         during DIE emission.  Do not re-use that for the skeleton for
1205         split-dwarf.
1206         (dwarf2out_early_finish): Likewise.
1207
1208 2018-04-12  Jakub Jelinek  <jakub@redhat.com>
1209
1210         PR target/85328
1211         * config/i386/sse.md
1212         (<mask_codefor>avx512dq_vextract<shuffletype>64x2_1<mask_name> split,
1213         <mask_codefor>avx512f_vextract<shuffletype>32x4_1<mask_name> split,
1214         vec_extract_lo_<mode><mask_name> split, vec_extract_lo_v32hi,
1215         vec_extract_lo_v64qi): For non-AVX512VL if input is xmm16+ reg
1216         and output is a reg, avoid creating invalid lowpart subreg, but
1217         instead split into a 512-bit move.  Don't split if not AVX512VL,
1218         input is xmm16+ reg and output is a mem.
1219         (vec_extract_lo_<mode><mask_name>, vec_extract_lo_v32hi,
1220         vec_extract_lo_v64qi): Don't require split if not AVX512VL, input is
1221         xmm16+ reg and output is a mem.
1222
1223 2018-04-12  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
1224
1225         * config/s390/s390.c (s390_output_indirect_thunk_function): Check
1226         also for flag_dwarf2_cfi_asm.
1227
1228 2018-04-12  Jakub Jelinek  <jakub@redhat.com>
1229
1230         PR rtl-optimization/85342
1231         * regcprop.c (copyprop_hardreg_forward_1): Remove replaced array, use
1232         a bool scalar var inside of the loop instead.  Don't try to update
1233         recog_data.operand after failed apply_change_group.
1234
1235 2018-04-12  Tom de Vries  <tom@codesourcery.com>
1236
1237         PR target/85296
1238         * config/nvptx/nvptx.c (flexible_array_member_type_p): New function.
1239         (nvptx_assemble_decl_begin): Add undefined param.  Declare undefined
1240         array with flexible array member as array without given dimension.
1241         (nvptx_assemble_undefined_decl): Set nvptx_assemble_decl_begin call
1242         argument for undefined param to true.
1243
1244 2018-04-11  Aaron Sawdey  <acsawdey@linux.ibm.com>
1245
1246         PR target/85321
1247         * doc/invoke.texi (RS/6000 and PowerPC Options): Document options
1248         -mcall- and -mtraceback=. Remove options -mabi=spe and -mabi=no-spe
1249         from PowerPC section.
1250         * config/rs6000/sysv4.opt (mcall-): Improve help text.
1251         * config/rs6000/rs6000.opt (mblock-compare-inline-limit=): Trim
1252         help text that is too long.
1253         * config/rs6000/rs6000.opt (mblock-compare-inline-loop-limit=): Trim
1254         help text that is too long.
1255         * config/rs6000/rs6000.opt (mstring-compare-inline-limit=): Trim
1256         help text that is too long.
1257
1258 2018-04-11  Uros Bizjak  <ubizjak@gmail.com>
1259
1260         * config/alpha/alpha.md (stack_probe_internal): Rename
1261         from "probe_stack".  Update all callers.
1262
1263 2018-04-11  Alexander Monakov  <amonakov@ispras.ru>
1264
1265         PR rtl-optimization/84566
1266         * sched-deps.c (sched_analyze_insn): Check deps->readonly when invoking
1267         sched_macro_fuse_insns.
1268
1269 2018-04-11  Alexander Monakov  <amonakov@ispras.ru>
1270
1271         PR target/84301
1272         * sched-rgn.c (add_branch_dependences): Move sel_sched_p check here...
1273         (compute_block_dependences): ... from here.
1274
1275 2018-04-11  Jakub Jelinek  <jakub@redhat.com>
1276
1277         PR tree-optimization/85331
1278         * vec-perm-indices.h (vec_perm_indices::clamp): Change input type
1279         from int to HOST_WIDE_INT.
1280
1281 2018-04-11  Martin Jambor  <mjambor@suse.cz>
1282
1283         PR ipa/84149
1284         * ipa-cp.c (propagate_vals_across_pass_through): Expand comment.
1285         (cgraph_edge_brings_value_p): New parameter dest_val, check if it is
1286         not the same as the source val.
1287         (cgraph_edge_brings_value_p): New parameter.
1288         (gather_edges_for_value): Pass destination value to
1289         cgraph_edge_brings_value_p.
1290         (perhaps_add_new_callers): Likewise.
1291         (get_info_about_necessary_edges): Likewise and exclude values brought
1292         only by self-recursive edges.
1293         (create_specialized_node): Redirect only clones of self-calling edges.
1294         (+self_recursive_pass_through_p): New function.
1295         (find_more_scalar_values_for_callers_subset): Use it.
1296         (find_aggregate_values_for_callers_subset): Likewise.
1297         (known_aggs_to_agg_replacement_list): Removed.
1298         (decide_whether_version_node): Re-calculate known constants for all
1299         remaining context clones.
1300
1301 2018-04-11  Richard Biener  <rguenther@suse.de>
1302
1303         PR lto/85339
1304         * dwarf2out.c (dwarf2out_finish): Remove DW_AT_stmt_list attribute
1305         from early DWARF output.
1306         (dwarf2out_early_finish): Output line info unconditionally into
1307         early DWARF and add reference to it.
1308
1309 2018-04-11  Jakub Jelinek  <jakub@redhat.com>
1310
1311         PR target/85281
1312         * config/i386/sse.md (iptr): Add V16SFmode and V8DFmode cases.
1313         (<avx512>_vec_dup<mode><mask_name>): Use a single pattern for modes
1314         other than V2DFmode using iptr mode attribute.
1315         (<avx512>_vec_dup<mode><mask_name>): Use iptr mode attribute.
1316
1317 2018-04-11  Alexander Monakov  <amonakov@ispras.ru>
1318
1319         PR rtl-optimization/84659
1320         * sel-sched-ir.c (sel_init_pipelining): Invoke cleanup_cfg.
1321
1322 2018-04-11  Jakub Jelinek  <jakub@redhat.com>
1323
1324         PR debug/85302
1325         * dwarf2out.c (skip_loc_list_entry): Don't call size_of_locs if
1326         SIZEP is NULL.
1327         (output_loc_list): Pass address of a dummy size variable even in the
1328         locview handling loop.
1329         (index_location_lists): Add comment on why skip_loc_list_entry can't
1330         call size_of_locs.
1331
1332 2018-04-11  Thomas Preud'homme  <thomas.preudhomme@arm.com>
1333
1334         PR target/85261
1335         * config/arm/arm-builtins.c (arm_expand_builtin): Force input operand
1336         into register.
1337
1338 2018-04-10  Aaron Sawdey  <acsawdey@linux.ibm.com>
1339
1340         PR target/85321
1341         * doc/invoke.texi (RS/6000 and PowerPC Options): Document options
1342         -mblock-compare-inline-limit, -mblock-compare-inline-loop-limit,
1343         and -mstring-compare-inline-limit.
1344
1345 2018-04-10  Segher Boessenkool  <segher@kernel.crashing.org>
1346
1347         PR target/85287
1348         * gcc/config/rs6000/rs6000.md (allocate_stack): Put the residual size
1349         for stack clash protection in a register whenever we need it to be in
1350         a register.
1351
1352 2018-04-10  Segher Boessenkool  <segher@kernel.crashing.org>
1353
1354         * common/config/rs6000/rs6000-common.c (rs6000_option_init_struct):
1355         Enable -fasynchronous-unwind-tables by default if OBJECT_FORMAT_ELF.
1356
1357 2018-04-10  Segher Boessenkool  <segher@kernel.crashing.org>
1358
1359         PR target/85321
1360         * config/rs6000/rs6000.opt (mtraceback=): Show the allowed values in
1361         the help text.
1362         (mlong-double-): Ditto.
1363         * config/rs6000/sysv4.opt (msdata=): Ditto.
1364         (mtls-size=): Ditto.
1365
1366 2018-04-10  Kelvin Nilsen  <kelvin@gcc.gnu.org>
1367
1368         * config/rs6000/rs6000-c.c (altivec_overloaded_builtins): Remove
1369         erroneous entries for
1370         "vector int vec_ldl (int, long int *)", and
1371         "vector unsigned int vec_ldl (int, unsigned long int *)".
1372         Add comments and entries for
1373         "vector bool char vec_ldl (int, bool char *)",
1374         "vector bool short vec_ldl (int, bool short *)",
1375         "vector bool int vec_ldl (int, bool int *)",
1376         "vector bool long long vec_ldl (int, bool long long *)",
1377         "vector pixel vec_ldl (int, pixel *)",
1378         "vector long long vec_ldl (int, long long *)",
1379         "vector unsigned long long vec_ldl (int, unsigned long long *)".
1380         * config/rs6000/rs6000.c (rs6000_init_builtins): Initialize new
1381         type tree bool_long_long_type_node and correct definition of
1382         bool_V2DI_type_node to make reference to this new type tree.
1383         (rs6000_mangle_type): Replace erroneous reference to
1384         bool_long_type_node with bool_long_long_type_node.
1385         * config/rs6000/rs6000.h (enum rs6000_builtin_type_index): Add
1386         comments to emphasize sign distinctions for char and int types and
1387         replace RS6000_BTI_bool_long constant with
1388         RS6000_BTI_bool_long_long constant.  Also add comment to restrict
1389         use of RS6000_BTI_pixel.
1390         (bool_long_type_node): Remove this macro definition.
1391         (bool_long_long_type_node): New macro definition
1392
1393 2018-04-10  Jakub Jelinek  <jakub@redhat.com>
1394
1395         PR rtl-optimization/85300
1396         * combine.c (subst): Handle subst of CONST_SCALAR_INT_P new_rtx also
1397         into FLOAT and UNSIGNED_FLOAT like ZERO_EXTEND, return a CLOBBER if
1398         simplify_unary_operation fails.
1399
1400 2018-04-10  Martin Liska  <mliska@suse.cz>
1401
1402         * gdbhooks.py: Add pretty-printers for varpool_node, symtab_node,
1403         cgraph_edge and ipa_ref.
1404
1405 2018-04-10  Jakub Jelinek  <jakub@redhat.com>
1406
1407         PR target/85177
1408         PR target/85255
1409         * config/i386/sse.md
1410         (<extract_type>_vinsert<shuffletype><extract_suf>_mask): Fix
1411         computation of the VEC_MERGE selector from mask.
1412         (<extract_type>_vinsert<shuffletype><extract_suf>_1<mask_name>):
1413         Fix decoding of the VEC_MERGE selector into mask.
1414
1415 2018-04-10  Richard Sandiford  <richard.sandiford@linaro.org>
1416
1417         PR tree-optimization/85286
1418         * tree-vect-data-refs.c (vect_get_smallest_scalar_type):
1419
1420 2018-04-10  Richard Sandiford  <richard.sandiford@linaro.org>
1421
1422         * final.c (final_1): Set insn_last_address as well as
1423         insn_current_address.
1424
1425 2018-04-10  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
1426
1427         PR target/85173
1428         * explow.c (emit_stack_probe): Call validize_mem on memory location
1429         before passing it to gen_probe_stack.  Create address operand and
1430         legitimize it for the probe_stack_address case.
1431
1432 2018-04-09  Jan Hubicka  <jh@suse.cz>
1433
1434         PR lto/85078
1435         * ipa-devirt.c (rebuild_type_inheritance-hash): New.
1436         * ipa-utils.h (rebuild_type_inheritance-hash): Declare.
1437         * tree.c (free_lang_data_in_type): Fix handling of binfos;
1438         walk basetypes.
1439         (free_lang_data): Rebuild type inheritance graph.
1440
1441 2018-04-09  Martin Sebor  <msebor@redhat.com>
1442
1443         * invoke.texi (-finline-small-functions): Mention other optimization
1444         options.
1445         (-findirect-inlining, -fpartial-inlining): Same.
1446         (-finline-functions-called-once): Same.
1447         (-freorder-blocks-and-partition): Same.
1448
1449 2018-04-09  Jan Hubicka  <jh@suse.cz>
1450
1451         PR rtl/84058
1452         * cfgcleanup.c (try_forward_edges): Do not give up on crossing
1453         jumps; choose last target that matches the criteria (i.e.
1454         no partition changes for non-crossing jumps).
1455         * cfgrtl.c (cfg_layout_redirect_edge_and_branch): Add basic
1456         support for redirecting crossing jumps to non-crossing.
1457
1458 2018-04-09  Alexey Brodkin  <abrodkin@synopsys.com>
1459
1460         * config/arc/arc.c (arc_expand_prologue): Set stack usage info
1461         also for naked functions.
1462
1463 2018-04-09  Claudiu Zissulescu  <claziss@synopsys.com>
1464
1465         * config/arc/arc.md (add_shift): New pattern.
1466         (add_shift2): Likewise.
1467         (sub_shift): Likewise.
1468         (sub_shift_cmp0_noout): Likewise.
1469         (compare_si_ashiftsi): Likewise.
1470         (xbfu_cmp0_noout): New combine pattern.
1471         (xbfu_cmp0"): Likewise.
1472         (movsi_set_cc_insn): Place the predicable variant first.
1473         (commutative_binary_cmp0_noout): Remove clobber.
1474         (commutative_binary_cmp0): New pattern.
1475         (noncommutative_binary_cmp0): Likewise.
1476         (noncommutative_binary_cmp0_noout): Likewise.
1477         (noncommutative_binary_comparison_result_used): Removed.
1478         (rsub_cmp0): New pattern.
1479         (rsub_cmp0_noout): Likewise.
1480         (extzvsi): Changed, keep only meaningful variants.
1481         (SQH, SEZ): New iterators.
1482         (SQH_postfix): New mode attribute.
1483         (SEZ_prefix): New code attribute.
1484         (<SEZ_prefix>xt<SQH_postfix>_cmp0_noout): New instruction pattern.
1485         (<SEZ_prefix>xt<SQH_postfix>_cmp0): Likewise.
1486         * config/arc/predicates.md (cc_set_register): Use CC_REG instead
1487         of numerical value.
1488         (noncommutative_operator): Check the availability of barrel
1489         shifter option.
1490
1491 2018-04-09  Richard Biener  <rguenther@suse.de>
1492
1493         PR tree-optimization/85284
1494         * tree-ssa-loop-niter.c (number_of_iterations_exit_assumptions):
1495         Only use the niter constraining form of simple_iv when the exit
1496         is always executed.
1497
1498 2018-04-09  Tom de Vries  <tom@codesourcery.com>
1499
1500         PR target/84041
1501         * config/nvptx/nvptx.md (define_c_enum "unspecv"): Add UNSPECV_MEMBAR.
1502         (define_expand "*memory_barrier"): New define_expand.
1503         (define_insn "memory_barrier"): New insn.
1504
1505 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
1506
1507         PR rtl-optimization/80463
1508         PR rtl-optimization/83972
1509         PR rtl-optimization/83480
1510
1511         * sel-sched-ir.c (has_dependence_note_mem_dep): Take into account the
1512         correct producer for the insn.
1513         (tidy_control_flow): Fixup seqnos in case of debug insns.
1514
1515 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
1516
1517         PR rtl-optimization/83913
1518
1519         * sel-sched-ir.c (merge_expr_data): Choose the middle between two
1520         different sched-times when merging exprs.
1521
1522 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
1523
1524         PR rtl-optimization/83962
1525
1526         * sel-sched-ir.c (tidy_control_flow): Correct the order in which we call
1527         tidy_fallthru_edge and tidy_control_flow.
1528
1529 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
1530
1531         PR rtl-optimization/83530
1532
1533         * sel-sched.c (force_next_insn): New global variable.
1534         (remove_insn_for_debug): When force_next_insn is true, also leave only
1535         next insn in the ready list.
1536         (sel_sched_region): When the region wasn't scheduled, make another pass
1537         over it with force_next_insn set to 1.
1538
1539 2018-04-08  Monk Chiang  <sh.chiang04@gmail.com>
1540
1541         * config.gcc (nds32le-*-*, nds32be-*-*): Add nds32/nds32_intrinsic.h
1542         into tm_file.
1543         * config/nds32/constants.md (unspec_volatile_element): Add enum values
1544         for interrupt control.
1545         * config/nds32/nds32-intrinsic.c: Implementation of intrinsic
1546         functions for interrupt control.
1547         * config/nds32/nds32-intrinsic.md: Likewise.
1548         * config/nds32/nds32_intrinsic.h: Likewise.
1549         * config/nds32/nds32.h (nds32_builtins): Likewise.
1550
1551 2018-04-08  Chung-Ju Wu  <jasonwucj@gmail.com>
1552
1553         * config/nds32/nds32.c (nds32_init_machine_status,
1554         nds32_legitimate_index_p, nds32_legitimate_address_p): Consider
1555         strict_aligned_p field.
1556         (nds32_expand_to_rtl_hook): New function.
1557         (TARGET_EXPAND_TO_RTL_HOOK): Define.
1558         * config/nds32/nds32.h (machine_function): Add strict_aligned_p field.
1559
1560 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
1561             Chung-Ju Wu  <jasonwucj@gmail.com>
1562
1563         * config.gcc (nds32*-*-*): Check that n7 is valid to --with-cpu.
1564         * config/nds32/nds32-n7.md: New file.
1565         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_N7.
1566         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n7
1567         pipeline.
1568         * config/nds32/nds32-protos.h: More declarations for n7 pipeline.
1569         * config/nds32/nds32.md (pipeline_model): Add n7.
1570         * config/nds32/nds32.opt (mcpu): Support n7 pipeline cpus.
1571         * config/nds32/pipelines.md: Include n7 settings.
1572
1573 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
1574             Chung-Ju Wu  <jasonwucj@gmail.com>
1575
1576         * config.gcc (nds32*-*-*): Check that e8 is valid to --with-cpu.
1577         * config/nds32/nds32-e8.md: New file.
1578         * config/nds32/nds32-opts.h (nds32-cpu_type): Add CPU_E8.
1579         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for e8
1580         pipeline.
1581         * config/nds32/nds32-protos.h: More declarations for e8 pipeline.
1582         * config/nds32/nds32.md (pipeline_model): Add e8.
1583         * config/nds32/nds32.opt (mcpu): Support e8 pipeline cpus.
1584         * config/nds32/pipelines.md: Include e8 settings.
1585
1586 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
1587             Chung-Ju Wu  <jasonwucj@gmail.com>
1588
1589         * config.gcc (nds32*-*-*): Check that n6/n8/s8 are valid to --with-cpu.
1590         * config/nds32/nds32-n8.md: New file.
1591         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_N6 and CPU_N8.
1592         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n8
1593         pipeline.
1594         * config/nds32/nds32-protos.h: More declarations for n8 pipeline.
1595         * config/nds32/nds32-utils.c: More implementations for n8 pipeline.
1596         * config/nds32/nds32.md (pipeline_model): Add n8.
1597         * config/nds32/nds32.opt (mcpu): Support n8 pipeline cpus.
1598         * config/nds32/pipelines.md: Include n8 settings.
1599
1600 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
1601             Chung-Ju Wu  <jasonwucj@gmail.com>
1602
1603         * config.gcc (nds32*): Add nds32-utils.o into extra_objs.
1604         * config/nds32/nds32-n9-2r1w.md: New file.
1605         * config/nds32/nds32-n9-3r2w.md: New file.
1606         * config/nds32/nds32-opts.h (nds32_cpu_type, nds32_mul_type,
1607         nds32_register_ports): New or modify for cpu n9.
1608         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n9
1609         pipeline.
1610         * config/nds32/nds32-protos.h: More declarations for n9 pipeline.
1611         * config/nds32/nds32-utils.c: New file.
1612         * config/nds32/nds32.h (TARGET_PIPELINE_N9, TARGET_PIPELINE_SIMPLE,
1613         TARGET_MUL_SLOW): Define.
1614         * config/nds32/nds32.md (pipeline_model): New attribute.
1615         * config/nds32/nds32.opt (mcpu, mconfig-mul, mconfig-register-ports):
1616         New options that support cpu n9.
1617         * config/nds32/pipelines.md: Include n9 settings.
1618         * config/nds32/t-nds32 (nds32-utils.o): Add dependency.
1619
1620 2018-04-08  Chung-Ju Wu  <jasonwucj@gmail.com>
1621
1622         * config/nds32/nds32-md-auxiliary.c (output_cond_branch): Output align
1623         information if necessary.
1624         (output_cond_branch_compare_zero): Likewise.
1625         * config/nds32/nds32.c (nds32_adjust_insn_length): Consider align case.
1626         (nds32_target_alignment): Refine for alignment.
1627         * config/nds32/nds32.h (NDS32_ALIGN_P): Define.
1628         (FUNCTION_BOUNDARY): Modify.
1629         * config/nds32/nds32.md (call_internal, call_value_internal): Consider
1630         align case.
1631         * config/nds32/nds32.opt (malways-align, malign-functions): New.
1632
1633 2018-04-08  Monk Chiang  <sh.chiang04@gmail.com>
1634
1635         * config/nds32/constants.md (unspec_volatile_element): Add values for
1636         TLB operation and data prefetch.
1637         * config/nds32/nds32-intrinsic.c: Implementation of intrinsic
1638         functions for TLB operation and data prefetch.
1639         * config/nds32/nds32-intrinsic.md: Likewise.
1640         * config/nds32/nds32_intrinsic.h: Likewise.
1641         * config/nds32/nds32.c (nds32_dpref_names): Likewise.
1642         (nds32_print_operand): Likewise.
1643         * config/nds32/nds32.h (nds32_builtins): Likewise.
1644
1645 2018-04-07  Thomas Koenig  <tkoenig@gcc.gnu.org>
1646         Andrew Pinski <pinsika@gcc.gnu.org>
1647
1648         PR middle-end/82976
1649         * match.pd: Use constant_boolean_node of correct type instead of
1650         boolean_true_node or boolean_false_node for simplifying
1651         pointer comparisons to zero.
1652
1653 2018-04-07  Jakub Jelinek  <jakub@redhat.com>
1654
1655         PR tree-optimization/80021
1656         * tree.c (verify_type_variant): Make error call in verify_variant_match
1657         translatable and remove final full stop.
1658
1659 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
1660
1661         * config/nds32/constants.md (unspec_volatile_element): Add
1662         UNSPEC_VOLATILE_EH_RETURN.
1663         * config/nds32/nds32-md-auxiliary.c (nds32_output_stack_push,
1664         nds32_output_stack_pop): Support dwarf exception handling process.
1665         * config/nds32/nds32-protos.h (nds32_dynamic_chain_address): Declare.
1666         * config/nds32/nds32.c (nds32_init_machine_status): Support dwarf
1667         exception handling process.
1668         (nds32_compute_stack_frame): Likewise.
1669         (nds32_return_addr_rtx): Likewise.
1670         (nds32_initial_elimination_offset): Likewise.
1671         (nds32_expand_prologue): Likewise.
1672         (nds32_expand_epilogue): Likewise.
1673         (nds32_dynamic_chain_address): New function.
1674         * config/nds32/nds32.h (machine_function): Add fields for dwarf
1675         exception handling.
1676         (DYNAMIC_CHAIN_ADDRESS): Define.
1677         (EH_RETURN_DATA_REGNO): Define.
1678         (EH_RETURN_STACKADJ_RTX): Define.
1679         * config/nds32/nds32.md (eh_return, nds32_eh_return): Implement
1680         patterns for dwarf exception handling.
1681
1682 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
1683
1684         * config/nds32/nds32.h: Clean up obsolete macros.
1685
1686 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
1687
1688         * config/nds32/constants.md (unspec_element, unspec_volatile_element):
1689         Add enum values for particular instructions.
1690         * config/nds32/nds32-intrinsic.c: Implementation of expanding
1691         particular intrinsic functions.
1692         * config/nds32/nds32-intrinsic.md: Likewise.
1693         * config/nds32/nds32_intrinsic.h: Likewise.
1694         * config/nds32/nds32.h (nds32_builtins): Likewise.
1695         * config/nds32/nds32.md (type): Add pbsad and pbsada.
1696         (btst, ave): New patterns for particular instructions.
1697
1698 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
1699
1700         * config/nds32/constants.md (unspec_element, unspec_volatile_element):
1701         Add enum values for atomic load/store and memory sync.
1702         * config/nds32/nds32-intrinsic.c: Implementation for atomic load/store
1703         and memory sync.
1704         * config/nds32/nds32-intrinsic.md: Likewise.
1705         * config/nds32/nds32_intrinsic.h: Likewise.
1706         * config/nds32/nds32.h (nds32_builtins): Likewise.
1707
1708 2018-04-07  Jakub Jelinek  <jakub@redhat.com>
1709
1710         PR tree-optimization/85257
1711         * fold-const.c (native_encode_vector): If not all elts could fit
1712         and off is -1, return 0 rather than offset.
1713         * tree-ssa-sccvn.c (vn_reference_lookup_3): Pass
1714         (offseti - offset2) / BITS_PER_UNIT as 4th argument to
1715         native_encode_expr.  Verify len * BITS_PER_UNIT >= maxsizei.  Don't
1716         adjust buffer in native_interpret_expr call.
1717
1718 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
1719
1720         * config/nds32/constants.md (unspec_volatile_element): Add cache
1721         control enum values.
1722         * config/nds32/nds32-intrinsic.c: Add cache control expand functions.
1723         * config/nds32/nds32-intrinsic.md: Add cache control patterns.
1724         * config/nds32/nds32.c (nds32_cctl_names): New.
1725         (nds32_print_operand): Handle cache control register names.
1726         * config/nds32/nds32.h (nds32_builtins): New enum values.
1727         * config/nds32/nds32_intrinsic.h: Add cache control enum types and
1728         macros.
1729         * config/nds32/nds32.md (type): Add mmu.
1730         * config/nds32/pipelines.md (simple_insn): Add mmu.
1731
1732 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
1733
1734         * config/nds32/nds32.md (type): Remove call.
1735         * config/nds32/pipelines.md (simple_insn): Likewise.
1736
1737 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
1738
1739         * config/nds32/constants.md (unspec_volatile_element): Add
1740         UNSPEC_VOLATILE_FMFCSR, UNSPEC_VOLATILE_FMTCSR and
1741         UNSPEC_VOLATILE_FMFCFG.
1742         * config/nds32/nds32-intrinsic.c (bdesc_noarg): New builtin
1743         description for fmfcfg and fmfcsr.
1744         (bdesc_1arg): Add fmtcsr.
1745         (bdesc_2arg): Add fcpynss, fcpyss, fcpynsd and fcpysd.
1746         (nds32_expand_builtin_impl): Deal with FPU intrinsic functions.
1747         * config/nds32/nds32-intrinsic.md (unspec_fcpynsd, unspec_fcpysd,
1748         unspec_fcpynss, unspec_fcpysd, unspec_fcpyss, unspec_fmfcsr,
1749         unspec_fmfcfg): New patterns.
1750         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_FMFCFG,
1751         NDS32_BUILTIN_FMFCSR, NDS32_BUILTIN_FMTCSR, NDS32_BUILTIN_FCPYNSS,
1752         NDS32_BUILTIN_FCPYSS,NDS32_BUILTIN_FCPYNSD and NDS32_BUILTIN_FCPYSD.
1753         * config/nds32/nds32_intrinsic.h (__nds32__fcpynsd, __nds32__fcpynss,
1754         __nds32__fcpysd, __nds32__fcpyss, __nds32__fmfcsr, __nds32__fmtcsr,
1755         __nds32__fmfcfg): Define.
1756
1757 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
1758
1759         * config/nds32/nds32.c (nds32_intrinsic_register_names): Add more
1760         intrinsic register names.
1761         * config/nds32/nds32_intrinsic.h (nds32_intrinsic_registers): Add more
1762         intrinsic register enum values and macros.
1763
1764 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
1765
1766         * config/nds32/nds32.c (nds32_legitimate_index_p): Modify condition
1767         for load/store addressing form.
1768         (nds32_print_operand_address): Likewise.
1769
1770 2018-04-06  Eric Botcazou  <ebotcazou@adacore.com>
1771
1772         PR target/85196
1773         * config/sparc/sparc.c (sparc_expand_move): Deal with symbolic operands
1774         based on LABEL_REF.  Remove useless assertion.
1775         (pic_address_needs_scratch): Fix formatting.
1776         (sparc_legitimize_pic_address): Minor tweaks.
1777         (sparc_delegitimize_address): Adjust assertion accordingly.
1778         * config/sparc/sparc.md (movsi_pic_label_ref): Change label_ref_operand
1779         into symbolic_operand.
1780         (movsi_high_pic_label_ref): Likewise.
1781         (movsi_lo_sum_pic_label_ref): Likewise.
1782         (movdi_pic_label_ref): Likewise.
1783         (movdi_high_pic_label_ref): Likewise.
1784         (movdi_lo_sum_pic_label_ref): Likewise.
1785
1786 2018-04-06  Amaan Cheval  <amaan.cheval@gmail.com>
1787
1788         * config.gcc (x86_64-*-rtems*): Add rtems.h to tm_file for
1789         custom LIB_SPEC setup.
1790
1791 2018-04-06  Ruslan Bukin  <br@bsdpad.com>
1792             Kito Cheng  <kito.cheng@gmail.com>
1793
1794         * config.gcc (riscv*-*-freebsd*): Add RISC-V FreeBSD support.
1795         * config/riscv/freebsd.h: New.
1796
1797 2018-04-06  Chung-Ju Wu  <jasonwucj@gmail.com>
1798
1799         * config/nds32/nds32.c (nds32_adjust_insn_length): Refine.
1800         * config/nds32/nds32.h (ADJUST_INSN_LENGTH): Change the location in
1801         file.
1802
1803 2018-04-06  Chung-Ju Wu  <jasonwucj@gmail.com>
1804             Kito Cheng  <kito.cheng@gmail.com>
1805
1806         * config/nds32/nds32-md-auxiliary.c (nds32_output_return,
1807         nds32_output_call, nds32_symbol_binds_local_p): New functions.
1808         * config/nds32/nds32-protos.h (nds32_output_call,
1809         nds32_output_return): Declare.
1810         * config/nds32/nds32.md: Refine all the call and return patterns.
1811
1812 2018-04-06  Jakub Jelinek  <jakub@redhat.com>
1813
1814         PR debug/85252
1815         * dwarf2out.c (rtl_for_decl_init): For STRING_CST initializer only
1816         build CONST_STRING if TYPE_MAX_VALUE is non-NULL and is INTEGER_CST.
1817
1818         PR rtl-optimization/84872
1819         * cfgloopmanip.c (create_preheader): Use make_forwarder_block even if
1820         nentry == 1 when CP_FALLTHRU_PREHEADERS and single_entry is
1821         EDGE_CROSSING edge.
1822
1823 2018-04-06  Tamar Christina  <tamar.christina@arm.com>
1824
1825         * expr.c (copy_blkmode_to_reg): Revert 254862.
1826         * doc/sourcebuild.texi (word_mode_no_slow_unalign): Likewise.
1827
1828 2018-04-06  Richard Biener  <rguenther@suse.de>
1829
1830         PR middle-end/85244
1831         * tree-dfa.c (get_ref_base_and_extent): Reset seen_variable_array_ref
1832         after seeing a component reference with an adjacent field.  Treat
1833         refs to arrays at struct end of external decls similar to
1834         refs to unconstrained commons.
1835
1836 2018-04-06  Jakub Jelinek  <jakub@redhat.com>
1837
1838         PR sanitizer/85213
1839         * fold-const.c (twoval_comparison_p): Remove SAVE_P argument and don't
1840         look through SAVE_EXPRs with non-side-effects argument.  Adjust
1841         recursive calls.
1842         (fold_comparison): Adjust twoval_comparison_p caller, don't handle
1843         save_p here.
1844
1845 2018-04-06  Richard Biener  <rguenther@suse.de>
1846
1847         PR middle-end/85180
1848         * alias.c (find_base_term): New wrapper around find_base_term
1849         unwinding CSELIB_VAL_PTR changes.
1850         (find_base_term): Do not restore CSELIB_VAL_PTR during the
1851         recursion.
1852
1853 2018-04-06  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
1854
1855         * config/s390/s390.c (s390_z10_optimize_cmp): Expand dedicated NOP
1856         instructions.
1857         * config/s390/s390.md (UNSPECV_NOP_LR_0, UNSPECV_NOP_LR_1): New
1858         constant definitions.
1859         ("nop"): lr 0,0 -> nopr r0
1860         ("nop_lr0", "nop_lr1"): New insn definitions.
1861
1862 2018-04-06  Chung-Ju Wu  <jasonwucj@gmail.com>
1863
1864         * config/nds32/nds32.md (*stack_push, *stack_pop): Use
1865         NDS32_V3PUSH_AVAILABLE_P macro.
1866
1867 2018-04-06  Monk Chiang  <sh.chiang04@gmail.com>
1868             Chung-Ju Wu  <jasonwucj@gmail.com>
1869
1870         * config.gcc (nds32*-*-*): Add v2j v3f v3s checking.
1871         (nds32*-*-*): Add float and fpu_config into supported_defaults.
1872         * common/config/nds32/nds32-common.c (TARGET_DEFAULT_TARGET_FLAGS):
1873         Include TARGET_DEFAULT_FPU_ISA and TARGET_DEFAULT_FPU_FMA.
1874         * config/nds32/constants.md (unspec_element): Add UNSPEC_COPYSIGN,
1875         UNSPEC_FCPYNSD, UNSPEC_FCPYNSS, UNSPEC_FCPYSD and UNSPEC_FCPYSS.
1876         * config/nds32/constraints.md: New constraints and checking for hard
1877         float configuration.
1878         * config/nds32/iterators.md: New mode iterator and attribute for hard
1879         float configuration.
1880         * config/nds32/nds32-doubleword.md: Use hard float alternatives and
1881         patterns.
1882         * config/nds32/nds32-fpu.md: New file.
1883         * config/nds32/nds32-md-auxiliary.c: New functions and checkings to
1884         deal with hard float code generation.
1885         * config/nds32/nds32-opts.h (nds32_arch_type): Add ARCH_V3F and
1886         ARCH_V3S.
1887         (abi_type, float_reg_number): New enum type.
1888         * config/nds32/nds32-predicates.c: New predicates for hard float.
1889         * config/nds32/nds32-protos.h: Declare functions for hard float.
1890         * config/nds32/nds32.c: Implementation for hard float configuration.
1891         * config/nds32/nds32.h: Definitions for hard float configuration.
1892         * config/nds32/nds32.md: Include hard float machine description and
1893         modify patterns for hard float configuration.
1894         * config/nds32/nds32.opt: New options for hard float configuration.
1895         * config/nds32/predicates.md: New predicates for hard float
1896         configuration.
1897
1898 2018-04-06  Kuan-Lin Chen  <kuanlinchentw@gmail.com>
1899
1900         * common/config/nds32/nds32-common.c
1901         (nds32_option_optimization_table): Enable -mreleax-hint by default.
1902
1903 2018-04-05  Jakub Jelinek  <jakub@redhat.com>
1904
1905         PR middle-end/85195
1906         * match.pd (BIT_FIELD_REF CONSTRUCTOR@0 @1 @2): Use view_convert around
1907         CONSTRUCTOR_ELT (ctor, ...)->value.
1908
1909 2018-04-05  Uros Bizjak  <ubizjak@gmail.com>
1910
1911         PR target/85193
1912         * config/i386/i386.md (define_attr "memory"): Handle rotate1 type.
1913
1914 2018-04-05  Tom de Vries  <tom@codesourcery.com>
1915
1916         PR target/85204
1917         * config/nvptx/nvptx.c (nvptx_single): Fix neutering of bb with only
1918         cond jump.
1919
1920 2018-04-05  Shiva Chen  <shiva0217@gmail.com>
1921             Kito Cheng  <kito.cheng@gmail.com>
1922
1923         * config/nds32/constraints.md (U33): Fine-tune checking condition.
1924         * config/nds32/nds32-md-auxiliary.c (nds32_mem_format): Ditto.
1925         * config/nds32/nds32.h (nds32_16bit_address_type): Add
1926         ADDRESS_POST_MODIFY_LO_REG_IMM3U.
1927
1928 2018-04-05  Shiva Chen  <shiva0217@gmail.com>
1929             Kito Cheng  <kito.cheng@gmail.com>
1930
1931         * config/nds32/constraints.md (Ufe): New memory constraint.
1932         * config/nds32/nds32-md-auxiliary.c (nds32_mem_format,
1933         nds32_output_16bit_load): Consider r8 register for lwi45.fe format.
1934         * config/nds32/nds32.c (nds32_print_operand): Output lwi45.fe
1935         operands.
1936         * config/nds32/nds32.h (nds32_16bit_address_type): Add ADDRESS_R8_IMM7U.
1937         * config/nds32/nds32.md (*mov<mode>): Adjust pattern.
1938
1939 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1940
1941         * config/nds32/nds32.md: Use optimize_size in the condition for
1942         alu-shift instructions.
1943
1944 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1945
1946         * config/nds32/nds32.md (divsi4, udivsi4): New patterns.
1947
1948 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1949
1950         * config/nds32/nds32.md (negsi2): Refine pattern.
1951
1952 2018-04-05  Kito Cheng  <kito.cheng@gmail.com>
1953             Chung-Ju Wu  <jasonwucj@gmail.com>
1954
1955         * config/nds32/iterators.md (shift_rotate): New code iterator.
1956         (shift): New code attribute.
1957         * config/nds32/nds32-md-auxiliary.c (nds32_expand_constant): New.
1958         * config/nds32/nds32-protos.h (nds32_expand_constant): Declare.
1959         * config/nds32/nds32.c (nds32_print_operand): Deal with more cases.
1960         * config/nds32/nds32.md (addsi3, *add_srli): Refine implementation for
1961         bit-wise operations.
1962         (andsi3, *andsi3): Ditto.
1963         (iorsi3, *iorsi3, *or_slli, *or_srli): Ditto.
1964         (xorsi3, *xorsi3, *xor_slli, *xor_srli): Ditto.
1965         (<shift>si3, *ashlsi3, *ashrsi3, *lshrsi3, *rotrsi3): Ditto.
1966         * config/nds32/predicates.md (nds32_rimm5u_operand, nds32_and_operand,
1967         nds32_ior_operand, nds32_xor_operand): New predicates.
1968
1969 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1970
1971         * config/nds32/nds32.md (add<mode>3, sub<mode>3): Rename to ...
1972         (addsi3, subsi3): ... this.
1973
1974 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1975
1976         * config/nds32/nds32.md (*sub_srli, *and_slli): Fine-tune predicator.
1977
1978 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1979
1980         * config/nds32/nds32.md: Adjust indention.
1981
1982 2018-04-05  Kito Cheng  <kito.cheng@gmail.com>
1983
1984         * config/nds32/nds32.md (feature): New attribute.
1985
1986 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1987
1988         * config/nds32/nds32.md (subtype): New attribute.
1989
1990 2018-04-04  Thomas Preud'homme  <thomas.preudhomme@arm.com>
1991
1992         PR target/85203
1993         * config/arm/arm-builtins.c (arm_expand_builtin): Change
1994         expansion to perform a bitwise AND of the argument followed by a
1995         boolean negation of the result.
1996
1997 2018-04-04  Peter Bergner  <bergner@vnet.ibm.com>
1998
1999         PR rtl-optimization/84878
2000         * ddg.c (add_cross_iteration_register_deps): Use DF_REF_BB to determine
2001         the basic block.  Assert the use reference is not artificial and that
2002         it has an associated insn.
2003
2004 2018-04-04  Michael Matz  <matz@suse.de>
2005
2006         * builtins.c (compute_objsize): Pass correct operand
2007         to array_at_struct_end_p.
2008
2009 2018-04-04  Richard Biener  <rguenther@suse.de>
2010
2011         PR lto/85176
2012         * dwarf2out.c (dwarf2out_register_external_die): Peel namespaces
2013         from contexts for DINFO_LEVEL_TERSE and below.
2014
2015 2018-04-04  Kito Cheng  <kito.cheng@gmail.com>
2016
2017         * config/nds32/nds32-doubleword.md (move_<mode>): Require
2018         resiter_operand condition.
2019         * config/nds32/nds32.md (*move<mode>): Ditto.
2020
2021 2018-04-04  Kito Cheng  <kito.cheng@gmail.com>
2022             Monk Chiang  <sh.chiang04@gmail.com>
2023
2024         * config/nds32/nds32.md (movmisalign<mode>): New pattern.
2025
2026 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
2027
2028         * config/nds32/nds32.md (movqi, movhi): Merge into mov<mode>.
2029
2030 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
2031             Kito Cheng  <kito.cheng@gmail.com>
2032
2033         * config/nds32/nds32-md-auxiliary.c (nds32_inverse_cond_code,
2034         nds32_cond_code_str, output_cond_branch,
2035         output_cond_branch_compare_zero, nds32_expand_cbranch,
2036         nds32_expand_cstore, nds32_expand_movcc,
2037         nds32_output_cbranchsi4_equality_zero,
2038         nds32_output_cbranchsi4_equality_reg,
2039         nds32_output_cbranchsi4_equality_reg_or_const_int,
2040         nds32_output_cbranchsi4_greater_less_zero: New functions.
2041         * config/nds32/nds32-protos.h (nds32_expand_cbranch,
2042         nds32_expand_cstore, nds32_expand_movcc,
2043         nds32_output_cbranchsi4_equality_zero,
2044         nds32_output_cbranchsi4_equality_reg,
2045         nds32_output_cbranchsi4_equality_reg_or_const_int,
2046         nds32_output_cbranchsi4_greater_less_zero): Declare.
2047         * config/nds32/predicates.md (nds32_movecc_comparison_operator,
2048         nds32_rimm11s_operand): New predicates.
2049         * config/nds32/nds32.h (nds32_expand_result_type): New enum type.
2050         * config/nds32/nds32.md: Rewrite all the branch and conditional move
2051         patterns.
2052
2053 2018-04-04  Kito Cheng  <kito.cheng@gmail.com>
2054
2055         * config/nds32/nds32-doubleword.md: Refine all the instruction type.
2056         * config/nds32/nds32.md: Ditto.
2057         * config/nds32/pipelines.md: Ditto.
2058
2059 2018-04-04  Richard Biener  <rguenther@suse.de>
2060
2061         PR tree-optimization/85168
2062         * tree-ssa-sccvn.c (vn_reference_maybe_forwprop_address): Avoid
2063         propagating abnormals.
2064
2065 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
2066
2067         * config/nds32/nds32.md (enabled): Use yes/no for this attribute.
2068
2069 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
2070             Kito Cheng  <kito.cheng@gmail.com>
2071
2072         * config/nds32/nds32-md-auxiliary.c (nds32_long_call_p): New function.
2073         * config/nds32/nds32-protos.h (nds32_long_call_p): Declare.
2074         * config/nds32/nds32.c (nds32_function_ok_for_sibcall): New function.
2075         (TARGET_FUNCTION_OK_FOR_SIBCALL): Define.
2076         * config/nds32/nds32.md (sibcall_internal): New.
2077         (sibcall_register): Remove.
2078         (sibcall_immediate): Remove.
2079         (sibcall_value_internal): New.
2080         (sibcall_value_register): Remove.
2081         (sibcall_value_immediate): Remove.
2082         * config/nds32/predicates.md (nds32_general_register_operand): New.
2083         (nds32_call_address_operand): New.
2084
2085 2018-04-03  Jakub Jelinek  <jakub@redhat.com>
2086
2087         PR rtl-optimization/85167
2088         * shrink-wrap.c (move_insn_for_shrink_wrap): Don't set bb_uses and
2089         bb_defs if *split_p, instead preinitialize it to NULL.
2090
2091         PR tree-optimization/85156
2092         * builtins.c (fold_builtin_expect): Use save_expr on arg1 to avoid
2093         evaluating the argument multiple times.
2094
2095 2018-04-03  Bill Schmidt  <wschmidt@linux.ibm.com>
2096
2097         * config/rs6000/emmintrin.h (_mm_cvtpd_epi32): Use __vector rather
2098         than vector.
2099         (_mm_cvtpd_ps): Likewise.
2100         (_mm_cvttpd_epi32): Likewise.
2101         * config/rs6000/mmintrin.h (_mm_unpacklo_pi8): Likewise.
2102         * config/rs6000/xmmintrin.h: For strict-ANSI C++ or C11, undefine
2103         vector, pixel, and bool following altivec.h include.
2104
2105 2018-04-03  Martin Sebor  <msebor@redhat.com>
2106
2107         * doc/extend.texi (Common Function Attributes): Clarify.
2108         (const attribute): Likewise.
2109         (pure attribute): Likewise.
2110
2111 2018-04-03  Jakub Jelinek  <jakub@redhat.com>
2112
2113         PR target/85169
2114         * config/i386/i386.c (ix86_expand_vector_set): Use
2115         HOST_WIDE_INT_1U << elt instead of 1 << elt.  Formatting fix.
2116
2117 2018-04-03  Uros Bizjak  <ubizjak@gmail.com>
2118
2119         * config/i386/i386.c (emit_i387_cw_initialization): Always use logic
2120         instructions when changing rounding bits to preserve precision bits
2121         in the x87 control word.
2122
2123 2018-04-03  Martin Liska  <mliska@suse.cz>
2124
2125         PR tree-optimization/82491
2126         * rtl.h (strip_offset_and_add): Replace += suboffset with
2127         poly_uint64 () + suboffset.
2128
2129 2018-03-29  Martin Liska  <mliska@suse.cz>
2130             Martin Jambor  <mjambor@suse.cz>
2131
2132         PR ipa/84947
2133         * ipa-cp.c (propagate_bits_across_jump_function): Bail out if
2134         param_type is not an integral or pointer type.
2135
2136 2018-04-03  Richard Biener  <rguenther@suse.de>
2137
2138         * sese.h (recompute_all_dominators): Remove.
2139
2140 2018-04-02  Martin Sebor  <msebor@redhat.com>
2141
2142         * doc/invoke.texi (-Wrestrict): Fix typos.
2143
2144 2018-04-02  Jim Wilson  <jimw@sifive.com>
2145
2146         * config/riscv/riscv.h (SHIFT_COUNT_TRUNCATED): Set to zero.
2147         * config/riscv/riscv.md (<optab>si3): Use QImode shift count.
2148         (<optab>di3, <optab>si3_extend): Likewise.
2149         (<optab>si3_mask, <optab>si3_mask_1): New.
2150         (<optab>di3_mask, <optab>di3_mask_1): New.
2151         (<optab>si3_extend_mask, <optab>si3_extend_mask_1): New.
2152         (lshrsi3_zero_extend_1): Use VOIDmode shift count.
2153         * config/riscv/sync.md (atomic_test_and_set): Emit QImode shift count.
2154
2155 2018-04-02  Gerald Pfeifer  <gerald@pfeifer.com>
2156
2157         * doc/cpp.texi (Variadic Macros): Fix line continuation in an
2158         example.
2159
2160 2018-04-02  Chung-Ju Wu  <jasonwucj@gmail.com>
2161
2162         * config/nds32/nds32.c (TARGET_CANONICALIZE_COMPARISON): Define.
2163         (nds32_canonicalize_comparison): New function.
2164
2165 2018-04-01  Chung-Ju Wu  <jasonwucj@gmail.com>
2166             Kito Cheng  <kito.cheng@gmail.com>
2167             Kuan-Lin Chen  <kuanlinchentw@gmail.com>
2168
2169         * config.gcc (nds32): Add nds32-relax-opt.o into extra_objs.
2170         * config/nds32/constants.md (unspec_volatile_element): Add
2171         UNSPEC_VOLATILE_RELAX_GROUP.
2172         * config/nds32/nds32-relax-opt.c: New file.
2173         * config/nds32/nds32-predicates.c
2174         (nds32_symbol_load_store_p): New function.
2175         * config/nds32/nds32-protos.h
2176         (nds32_symbol_load_store_p): Declare function.
2177         (make_pass_nds32_relax_opt): Declare new rtl pass function.
2178         * config/nds32/nds32.c
2179         (nds32_register_pass): New function to register pass.
2180         (nds32_register_passes): New function to register passes.
2181         * config/nds32/nds32.md (relax_group): New pattern.
2182         * config/nds32/nds32.opt (mrelax-hint): New option.
2183         * config/nds32/t-nds32 (nds32-relax-opt.o): New dependency.
2184
2185 2018-04-01  Kito Cheng  <kito.cheng@gmail.com>
2186
2187         * config/nds32/t-nds32: Modify files dependency.
2188
2189 2018-04-01  Chung-Ju Wu  <jasonwucj@gmail.com>
2190
2191         * config/nds32/nds32.h (FUNCTION_PROFILER): Output newline character.
2192         (PROFILE_HOOK): Define its implementation.
2193
2194 2018-04-01  Chung-Ju Wu  <jasonwucj@gmail.com>
2195
2196         * config/nds32/nds32.h (WCHAR_TYPE, WCHAR_TYPE_SIZE): Use unsigned int
2197         type and 32-bit size.
2198
2199 2018-04-01  Jakub Jelinek  <jakub@redhat.com>
2200
2201         PR middle-end/85090
2202         * config/i386/sse.md (V): Add V64QI and V32HI for TARGET_AVX512F.
2203         (V_128_256): New mode iterator.
2204         (*avx512dq_vextract<shuffletype>64x2_1 splitter): New define_split.
2205         (*avx512f_vextract<shuffletype>32x4_1 splitter): Likewise.
2206         (xop_pcmov_<mode><avxsizesuffix>): Use V_128_256 mode iterator instead
2207         of V.
2208         * config/i386/i386.c (ix86_expand_vector_set): Improve V32HImode and
2209         V64QImode expansion for !TARGET_AVX512BW && TARGET_AVX512F.
2210
2211 2018-03-31  Segher Boessenkool  <segher@kernel.crashing.org>
2212
2213         PR target/83315
2214         * config/rs6000/xmmintrin.h (_mm_set_ps, _mm_max_ps): Handle (quiet)
2215         NaN inputs correctly.
2216
2217 2018-03-30  Peter Bergner  <bergner@vnet.ibm.com>
2218
2219         PR target/80546
2220         * config/rs6000/vsx.md (??r): New mode attribute.
2221         (*vsx_mov<mode>_64bit): Use it.
2222         (*vsx_mov<mode>_32bit): Likewise.
2223
2224 2018-03-30  Martin Sebor  <msebor@redhat.com>
2225
2226         PR tree-optimization/84818
2227         * builtins.c (check_access): Use warning_n.
2228
2229 2018-03-30  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
2230
2231         PR target/83822
2232         * config/rs6000/rs6000-string.c (expand_compare_loop): Fix redundant
2233         condition.
2234         * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): Fix redundant
2235         condition.
2236
2237 2018-03-30  Julia Koval  <julia.koval@intel.com>
2238
2239         PR target/84413
2240         * x86-tune.def (movx, partial_reg_dependency): Enable for
2241         m_SKYLAKE_AVX512.
2242
2243 2018-03-29  Vladimir Makarov  <vmakarov@redhat.com>
2244
2245         PR inline-asm/84985
2246         * lra-constraints.c (process_alt_operands): Move setting
2247         this_alternative_matches below.
2248
2249 2018-03-29  Martin Liska  <mliska@suse.cz>
2250
2251         PR lto/84995.
2252         * doc/invoke.texi: Document how LTO works with debug info.
2253         Describe auto-load support of binutils.  Mention 'x86-64'
2254         as valid option value of -march option.
2255
2256 2018-03-29  Jakub Jelinek  <jakub@redhat.com>
2257
2258         * config/i386/sse.md (<avx512>_blendm<mode>): Use <sseintprefix>.
2259
2260         PR c/85094
2261         * fold-const.c (operand_equal_p): Handle DEBUG_BEGIN_STMT.
2262         For STATEMENT_LIST, pass down OEP_LEXICOGRAPHIC and maybe
2263         OEP_NO_HASH_CHECK for recursive call, to avoid exponential
2264         checking.
2265
2266 2018-03-28  Peter Bergner  <bergner@vnet.ibm.com>
2267
2268         PR target/84912
2269         * config/rs6000/rs6000.h (RS6000_BTM_POWERPC64): New define.
2270         (RS6000_BTM_COMMON): Add RS6000_BTM_POWERPC64.
2271         * config/rs6000/rs6000.c (rs6000_builtin_mask_calculate): Add support
2272         for RS6000_BTM_POWERPC64.
2273         (rs6000_invalid_builtin): Add handling for RS6000_BTM_POWERPC64
2274         (rs6000_builtin_mask_names): Add RS6000_BTM_POWERPC64.
2275         * config/rs6000/rs6000-builtin.def (BU_P7_POWERPC64_MISC_2): New macro
2276         definition.
2277         (DIVDE): Use it.
2278         (DIVDEU): Likewise.
2279
2280 2018-03-28 Carl Love  <cel@us.ibm.com>
2281
2282         Revert
2283         2017-09-27  Carl Love  <cel@us.ibm.com>
2284
2285         * config/rs6000/rs6000-builtin.def (BU_FP_1MISC_1): Add define macro.
2286         (FCTID, FCTIW): Add BU_FP_MISC_1 macro expansion for builtins.
2287         * config/rs6000/rs6000.md (lrintsfsi2): Add define_insn for the
2288         fctiw instruction.
2289
2290 2018-03-28  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
2291
2292         * config/rs6000/xmmintrin.h (_mm_max_pi16): Use __vector __bool
2293         instead of __vector bool.
2294         (_mm_max_pu8): Likewise.
2295         (_mm_min_pi16): Likewise.
2296
2297 2018-03-28  Peter Bergner  <bergner@vnet.ibm.com>
2298
2299         PR target/84912
2300         * config/rs6000/rs6000-builtin.def (DIVWEO): Delete macro expansion.
2301         (DIVWEUO): Likewise.
2302         (DIVDEO): Likewise.
2303         (DIVDEUO): Likewise.
2304         * config/rs6000/rs6000.c (builtin_function_type): Remove support for
2305         DIVWEUO and DIVDEUO.
2306         * config/rs6000/rs6000.md (UNSPEC_DIVEO, UNSPEC_DIVEUO): Delete unspecs.
2307         (UNSPEC_DIV_EXTEND): Remove deleted unspecs.
2308         (div_extend): Likewise.
2309         * doc/extend.texi (__builtin_divweo): Remove documentation for deleted
2310         builtin function.
2311         (__builtin_divweuo): Likewise.
2312         (__builtin_divdeo): Likewise.
2313         (__builtin_divdeuo): Likewise.
2314
2315 2018-03-28  Jakub Jelinek  <jakub@redhat.com>
2316
2317         PR target/85095
2318         * config/i386/i386.md (*add<mode>3_carry_0, *addsi3_carry_zext_0,
2319         *sub<mode>3_carry_0, *subsi3_carry_zext_0): New patterns.
2320
2321         PR tree-optimization/82004
2322         * gimple-match-head.c (optimize_pow_to_exp): New function.
2323         * match.pd (pow(C,x) -> exp(log(C)*x)): Wrap with #if GIMPLE.
2324         Don't fold to exp if optimize_pow_to_exp is false.
2325
2326 2018-03-28  Martin Liska  <mliska@suse.cz>
2327
2328         PR other/84819
2329         * calls.c (initialize_argument_information): Fix trailing space.
2330         * common.opt: Fix typo and provide better explanation for
2331         -fsanitize-coverage option.
2332         * config/i386/i386.opt: Fix typo.
2333
2334 2018-03-28  Jakub Jelinek  <jakub@redhat.com>
2335             Martin Liska  <mliska@suse.cz>
2336
2337         PR sanitizer/85081
2338         * gimplify.c (asan_poison_variable): Don't do the check for
2339         gimplify_omp_ctxp here.
2340         (gimplify_decl_expr): Do it here.
2341         (gimplify_target_expr): Likewise.
2342
2343 2018-03-28  Martin Liska  <mliska@suse.cz>
2344
2345         PR target/84988
2346         * config/i386/i386.c (ix86_function_arg_advance): Do not call
2347         chkp_type_bounds_count if MPX is not enabled.
2348
2349 2018-03-27  Chung-Ju Wu  <jasonwucj@gmail.com>
2350
2351         * config/nds32/nds32.h (BRANCH_COST): Adjust cost.
2352
2353 2018-03-27  Michael Meissner  <meissner@linux.vnet.ibm.com>
2354
2355         PR target/84914
2356         * config/rs6000/rs6000.c (create_complex_muldiv): New helper
2357         function to create the function decl for complex long double
2358         multiply and divide for -mabi=ieeelongdouble.
2359         (init_float128_ieee): Call it.
2360
2361 2018-03-27  H.J. Lu  <hongjiu.lu@intel.com>
2362
2363         PR target/85044
2364         * config/i386/i386.c (ix86_trampoline_init): Insert ENDBR for
2365         -fcf-protection=branch -mibt.
2366         * config/i386/i386.h (TRAMPOLINE_SIZE): Increased by 4 bytes.
2367
2368 2018-03-27  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
2369
2370         PR target/81863
2371         * config/arm/arm.c (arm_valid_symbolic_address): Handle arm_word_relocations
2372
2373 2018-03-27  Cesar Philippidis  <cesar@codesourcery.com>
2374
2375         PR target/85056
2376         * config/nvptx/nvptx.c (nvptx_assemble_decl_begin): Add '[]' to
2377         extern array declarations.
2378
2379 2018-03-27  Richard Biener  <rguenther@suse.de>
2380
2381         PR middle-end/84067
2382         * match.pd ((A * C) +- (B * C) -> (A+-B) * C): Guard with
2383         explicit single_use checks.
2384
2385 2018-03-27  Richard Biener  <rguenther@suse.de>
2386
2387         PR tree-optimization/85082
2388         * tree-ssa-sccvn.c (vn_reference_lookup_or_insert_for_pieces):
2389         Valueize the VUSE.
2390
2391 2018-03-27  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
2392
2393         * config.gcc (aarch64*-*-linux*): New TARGET_DEFAULT_ASYNC_UNWIND_TABLES
2394         * common/config/aarch64/aarch64-common.c (aarch64_optimization_table[]):
2395         Turn on fasynchronous-unwind-tables and funwind-tables.
2396
2397 2018-03-26  Uros Bizjak  <ubizjak@gmail.com>
2398
2399         PR target/85073
2400         * config/i386/i386.md (*bmi_blsr_<mode>_cmp): New insn pattern.
2401         (*bmi_blsr_<mode>_ccz): Ditto.
2402
2403 2018-03-26  Tom de Vries  <tom@codesourcery.com>
2404
2405         PR tree-optimization/85063
2406         * omp-general.c (offloading_function_p): New function.  Factor out
2407         of ...
2408         * omp-offload.c (pass_omp_target_link::gate): ... here.
2409         * omp-general.h (offloading_function_p): Declare.
2410         * tree-switch-conversion.c (build_one_array): Mark CSWTCH.x variable
2411         with attribute omp declare target for offloading functions.
2412
2413 2018-03-24  Richard Sandiford  <richard.sandiford@linaro.org>
2414
2415         PR tree-optimization/84005
2416         * tree-data-ref.h (get_base_for_alignment): Declare.
2417         * tree-data-ref.c (get_base_for_alignment_1): New function.
2418         (get_base_for_alignment): Likewise.
2419         * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Use
2420         get_base_for_alignment to find a suitable base object, instead
2421         of always using drb->base_address.
2422
2423 2018-03-23  Jakub Jelinek  <jakub@redhat.com>
2424
2425         PR inline-asm/85022
2426         * emit-rtl.c (init_emit_regs): Indicate that VOIDmode MEMs don't have
2427         known size by default.
2428
2429 2018-03-23  Vladimir Makarov  <vmakarov@redhat.com>
2430
2431         PR inline-asm/85030
2432         * lra-constraints.c (process_alt_operands): Don't match BLKmode
2433         and non BLKmode operands.
2434
2435 2018-03-23  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
2436
2437         PR target/85026
2438         * config/arm/arm.md (unaligned_loadhis): Remove first alternative.
2439         Clean up attributes.
2440
2441 2018-03-23  Richard Biener  <rguenther@suse.de>
2442
2443         PR debug/85020
2444         * dwarf2out.c (rtl_for_decl_location): Do not generate RTL early when
2445         we are going to emit early debug for LTO.
2446
2447 2018-03-23  Jakub Jelinek  <jakub@redhat.com>
2448
2449         PR inline-asm/85034
2450         * function.c (match_asm_constraints_1): Don't optimize if input
2451         doesn't satisfy general_operand predicate for output's mode.
2452
2453         PR inline-asm/85022
2454         * alias.c (write_dependence_p): Don't require for x_canonicalized
2455         non-VOIDmode if x has VOIDmode.
2456
2457         PR sanitizer/85029
2458         * sanopt.c (maybe_optimize_ubsan_ptr_ifn): If DECL_REGISTER (base),
2459         just don't try to optimize it rather than assert it never happens.
2460
2461 2018-03-22  Kelvin Nilsen  <kelvin@gcc.gnu.org>
2462
2463         * config/rs6000/rs6000-builtin.def: Remove various BU_ALTIVEC_X
2464         macro expansions for definition of ST_INTERNAL_<mode> and
2465         LD_INTERNAL_<mode> builtins.
2466         * config/rs6000/rs6000-protos.h (rs6000_address_for_altivec):
2467         Remove prototype.
2468         * config/rs6000/rs6000.c (altivec_expand_ld_builtin): Delete this
2469         function.
2470         (altivec_expand_st_builtin): Likewise.
2471         (altivec_expand_builtin): Remove calls to deleted functions.
2472         (rs6000_address_for_altivec): Delete this function.
2473         * config/rs6000/vector.md: Remove expands for
2474         vector_altivec_load_<mode> and vector_altivec_store_<mode>.
2475
2476 2018-03-22  Sudakshina Das  <sudi.das@arm.com>
2477
2478         PR target/84826
2479         * config/arm/arm.h (machine_function): Add static_chain_stack_bytes.
2480         * config/arm/arm.c (arm_compute_static_chain_stack_bytes): Avoid
2481         re-computing once computed.
2482         (arm_expand_prologue): Compute machine->static_chain_stack_bytes.
2483         (arm_init_machine_status): Initialize
2484         machine->static_chain_stack_bytes.
2485
2486 2018-03-22  Kelvin Nilsen  <kelvin@gcc.gnu.org>
2487
2488         PR target/84760
2489         * doc/extend.texi: Add four new prototypes for vec_ld.
2490         * config/rs6000/rs6000-builtin.def (LVX_V1TI): Reorder symbol
2491         definitions for more logical presentation.
2492         * config/rs6000/rs6000-c.c: (altivec_overloaded_builtins): Add
2493         entries for V1TI variants of __builtin_altivec_ld builtin.
2494         * config/rs6000/rs6000.c: (altivec_expand_lv_builtin): Add test and
2495         handling of V1TI variant of LVX icode pattern.
2496         (altivec_expand_builtin): Add case for ALTIVEC_BUILTIN_LVX_V1TI.
2497         (rs6000_gimple_fold_builtin): Likewise.
2498         (altivec_init_builtins): Add code to define
2499         __builtin_altivec_lvx_v1ti function.
2500
2501 2018-03-22  Jakub Jelinek  <jakub@redhat.com>
2502
2503         PR inline-asm/84941
2504         * function.c (match_asm_constraints_1): Don't do the optimization
2505         if input isn't a REG, SUBREG, MEM or constant.
2506
2507 2018-03-22  Tom de Vries  <tom@codesourcery.com>
2508
2509         PR tree-optimization/84956
2510         * tree-ssa-tail-merge.c (find_clusters_1): Skip bbs with
2511         bb_has_abnormal_pred.
2512
2513 2018-03-22  Jakub Jelinek  <jakub@redhat.com>
2514
2515         PR sanitizer/85018
2516         * dwarf2asm.c (dw2_output_indirect_constant_1): Set
2517         DECL_INITIAL (decl) to decl at the end.
2518         * varasm.c (use_blocks_for_decl_p): Revert the 2018-03-20 change,
2519         adjust the comment.
2520
2521 2018-03-21  Joseph Myers  <joseph@codesourcery.com>
2522
2523         * doc/extend.texi (__builtin_tgmath): Document when complex
2524         integer types are treated as _Complex _Float64.
2525
2526 2018-03-21  Tom de Vries  <tom@codesourcery.com>
2527
2528         * doc/extend.texi (__builtin_extend_pointer): Remove pasto.
2529
2530 2018-03-21  Jakub Jelinek  <jakub@redhat.com>
2531
2532         PR tree-optimization/84960
2533         * tree-cfg.c (remove_bb): Don't move forced labels into bb->prev_bb
2534         if it is ENTRY block, move them into single succ of ENTRY in that case.
2535
2536 2018-03-21  Richard Sandiford  <richard.sandiford@linaro.org>
2537
2538         PR tree-optimization/84811
2539         * poly-int.h (poly_span_traits): Remove the T3 parameter and
2540         promote HOST_WIDE_INT T2 - T1 results to unsigned HOST_WIDE_INT.
2541         (maybe_in_range_p, known_in_range_p, ranges_known_overlap_p):
2542         (known_subrange_p): Update accordingly.  Cast each value involved
2543         in the size comparison, rather than casting the result of the
2544         subtraction.
2545
2546 2018-03-21  Jakub Jelinek  <jakub@redhat.com>
2547
2548         PR tree-optimization/84982
2549         * gimple-ssa-store-merging.c (invert_op): Handle boolean inversion
2550         by flipping the least significant bit rather than all bits from
2551         bitpos to bitpos + bitsize - 1.
2552
2553 2018-03-21  Nathan Sidwell  <nathan@acm.org>
2554
2555         * doc/extend.texi (Deprecated Features): Remove mention of
2556         long-deleted deprecations.
2557
2558 2018-03-21  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
2559
2560         PR jit/84288
2561         * configure.ac (gcc_cv_ld_soname) <*-*-solaris2*>: Set.
2562         * configure: Regenerate.
2563
2564 2018-03-21  Tom de Vries  <tom@codesourcery.com>
2565
2566         PR tree-optimization/83126
2567         * tree-parloops.c (num_phis): New function.
2568         (gen_parallel_loop): Detect and handle canonicalize_loop_ivs failure.
2569
2570 2018-03-21  Nathan Sidwell  <nathan@acm.org>
2571
2572         * doc/extend.texi (Deprecated Features): Update deprecated flags,
2573         mention anon-struct/union members and trailing attributes.
2574
2575 2018-03-21  Bin Cheng  <bin.cheng@arm.com>
2576
2577         PR tree-optimization/84969
2578         * tree-loop-distribution.c (fuse_memset_builtins): Don't reorder
2579         builtin memset partitions if they set different rhs values.
2580
2581 2018-03-21  Jakub Jelinek  <jakub@redhat.com>
2582
2583         PR rtl-optimization/84989
2584         * simplify-rtx.c (simplify_unary_operation_1): Don't try to simplify
2585         VEC_DUPLICATE with scalar result mode.
2586
2587 2018-03-21  Martin Liska  <mliska@suse.cz>
2588
2589         PR ipa/84963
2590         * ipa-icf.c (sem_item_optimizer::fixup_points_to_sets): Remove
2591         not intended return statement.
2592
2593 2018-03-21  Martin Liska  <mliska@suse.cz>
2594
2595         PR target/84988
2596         * tree-chkp.c (CHKP_ARRAY_MAX_CHECK_STEPS): Define a new macro.
2597         (chkp_find_bound_slots_1): Limit number of iterations.
2598
2599 2018-03-20  David H. Gutteridge  <dhgutteridge@sympatico.ca>
2600
2601         PR target/84838
2602         * Minor grammar fixes for x86 options.
2603
2604 2018-03-20  Jakub Jelinek  <jakub@redhat.com>
2605
2606         PR debug/84875
2607         * dce.c (delete_unmarked_insns): Don't remove frame related noop moves
2608         holding REG_CFA_RESTORE notes, instead turn them into a USE.
2609
2610 2018-03-20  Peter Bergner  <bergner@vnet.ibm.com>
2611
2612         PR target/83789
2613         * config/rs6000/altivec.md (altivec_lvx_<mode>_2op): Delete define_insn.
2614         (altivec_lvx_<mode>_1op): Likewise.
2615         (altivec_stvx_<mode>_2op): Likewise.
2616         (altivec_stvx_<mode>_1op): Likewise.
2617         (altivec_lvx_<VM2:mode>): New define_expand.
2618         (altivec_stvx_<VM2:mode>): Likewise.
2619         (altivec_lvx_<VM2:mode>_2op_<P:mptrsize>): New define_insn.
2620         (altivec_lvx_<VM2:mode>_1op_<P:mptrsize>): Likewise.
2621         (altivec_stvx_<VM2:mode>_2op_<P:mptrsize>): Likewise.
2622         (altivec_stvx_<VM2:mode>_1op_<P:mptrsize>): Likewise.
2623         * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Use new expanders.
2624         (rs6000_gen_lvx): Likewise.
2625         * config/rs6000/rs6000.c (altivec_expand_lv_builtin): Likewise.
2626         (altivec_expand_stv_builtin): Likewise.
2627         (altivec_expand_builtin): Likewise.
2628         * config/rs6000/vector.md: Likewise.
2629
2630 2018-03-20  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
2631
2632         PR target/82518
2633         * config/arm/arm.c (arm_array_mode_supported_p): Return false for
2634         BYTES_BIG_ENDIAN.
2635
2636 2018-03-20  Richard Biener  <rguenther@suse.de>
2637
2638         PR target/84986
2639         * config/i386/i386.c (ix86_add_stmt_cost): Only cost
2640         sign-conversions as zero, fall back to standard scalar_stmt
2641         cost for the rest.
2642
2643 2018-03-20  Martin Liska  <mliska@suse.cz>
2644
2645         PR ipa/84825
2646         * predict.c (rebuild_frequencies): Handle case when we have
2647         PROFILE_ABSENT, but flag_guess_branch_prob is false.
2648
2649 2018-03-20  Jakub Jelinek  <jakub@redhat.com>
2650
2651         PR target/84990
2652         * dwarf2asm.c (dw2_output_indirect_constant_1): Temporarily turn off
2653         flag_section_anchors.
2654         * varasm.c (use_blocks_for_decl_p): Remove hack for
2655         dw2_force_const_mem.
2656
2657         PR target/84845
2658         * config/aarch64/aarch64.md (*aarch64_reg_<mode>3_neg_mask2): Rename
2659         to ...
2660         (*aarch64_<optab>_reg_<mode>3_neg_mask2): ... this.  If pseudos can't
2661         be created, use lowpart_subreg of operands[0] rather than operands[0]
2662         itself.
2663         (*aarch64_reg_<mode>3_minus_mask): Rename to ...
2664         (*aarch64_ashl_reg_<mode>3_minus_mask): ... this.
2665         (*aarch64_<optab>_reg_di3_mask2): Use const_int_operand predicate
2666         and n constraint instead of aarch64_shift_imm_di and Usd.
2667         (*aarch64_reg_<optab>_minus<mode>3): Rename to ...
2668         (*aarch64_<optab>_reg_minus<mode>3): ... this.
2669
2670 2018-03-20  Sudakshina Das  <sudi.das@arm.com>
2671
2672         PR target/82989
2673         * config/arm/neon.md (ashldi3_neon): Update ?s for constraints
2674         to favor GPR over NEON registers.
2675         (<shift>di3_neon): Likewise.
2676
2677 2018-03-20  Tom de Vries  <tom@codesourcery.com>
2678
2679         PR target/84952
2680         * config/nvptx/nvptx.c (nvptx_single): Don't neuter bar.sync.
2681         (nvptx_process_pars): Emit bar.sync asap and alap.
2682
2683 2018-03-20  Tom de Vries  <tom@codesourcery.com>
2684
2685         PR target/84954
2686         * config/nvptx/nvptx.c (prevent_branch_around_nothing): Also update
2687         seen_label if seen_label is already set.
2688
2689 2018-03-20  Jakub Jelinek  <jakub@redhat.com>
2690
2691         PR target/84945
2692         * config/i386/i386.c (fold_builtin_cpu): For features above 31
2693         use __cpu_features2 variable instead of __cpu_model.__cpu_features[0].
2694         Use 1U instead of 1.  Formatting fixes.
2695
2696         PR c/84953
2697         * builtins.c (fold_builtin_strpbrk): For strpbrk(x, "") use type
2698         instead of TREE_TYPE (s1) for the return value.
2699
2700 2018-03-19  Jakub Jelinek  <jakub@redhat.com>
2701
2702         PR tree-optimization/84946
2703         * gimple-ssa-store-merging.c (mem_valid_for_store_merging): Compute
2704         bitsize + bitsize in poly_uint64 rather than poly_int64.
2705
2706         PR sanitizer/78651
2707         * dwarf2asm.c: Include fold-const.c.
2708         (dw2_output_indirect_constant_1): Set DECL_INITIAL (decl) to ADDR_EXPR
2709         of decl rather than decl itself.
2710
2711         PR rtl-optimization/84643
2712         * memmodel.h (enum memmodel): Add MEMMODEL_MAX enumerator.
2713
2714 2018-03-19  Maxim Ostapenko  <m.ostapenko@samsung.com>
2715
2716         PR sanitizer/78651
2717         * dwarf2asm.c (dw2_output_indirect_constant_1): Disable ASan before
2718         calling assemble_variable.
2719
2720 2018-03-19  Sudakshina Das  <sudi.das@arm.com>
2721
2722         PR target/81647
2723         * config/aarch64/aarch64-simd.md (vec_cmp<mode><v_int_equiv>): Modify
2724         instructions for UNLT, UNLE, UNGT, UNGE, UNEQ, UNORDERED and ORDERED.
2725
2726 2018-03-19  Jim Wilson  <jimw@sifive.com>
2727
2728         PR bootstrap/84856
2729         * config/riscv/riscv.c (riscv_function_arg_boundary): Use
2730         PREFERRED_STACK_BOUNDARY instead of STACK_BOUNDARY.
2731         (riscv_first_stack_step): Likewise.
2732         (riscv_option_override): Use STACK_BOUNDARY instead of
2733         MIN_STACK_BOUNDARY.
2734         * config/riscv/riscv.h (STACK_BOUNDARY): Renamed from
2735         MIN_STACK_BOUNDARY.
2736         (BIGGEST_ALIGNMENT): Set to 128.
2737         (PREFERRED_STACK_BOUNDARY): Renamed from STACK_BOUNDARY.
2738         (RISCV_STACK_ALIGN): Use PREFERRED_STACK_BOUNDARY instead of
2739         STACK_BOUNDARY.
2740
2741 2018-03-19  Richard Biener  <rguenther@suse.de>
2742
2743         PR tree-optimization/84933
2744         * tree-vrp.c (set_and_canonicalize_value_range): Treat out-of-bound
2745         values as -INF/INF when canonicalizing an ANTI_RANGE to a RANGE.
2746
2747 2018-03-19  Richard Biener  <rguenther@suse.de>
2748
2749         PR tree-optimization/84859
2750         * tree-ssa-phiopt.c (single_trailing_store_in_bb): New function.
2751         (cond_if_else_store_replacement): Perform sinking operation on
2752         single-store BBs regardless of MAX_STORES_TO_SINK setting.
2753         Generalize what a BB with a single eligible store is.
2754
2755 2018-03-19  Richard Biener  <rguenther@suse.de>
2756
2757         PR tree-optimization/84929
2758         * tree-data-ref.c (analyze_siv_subscript_cst_affine): Guard
2759         chrec_is_positive against non-chrec arg.
2760
2761 2018-03-19  Tamar Christina  <tamar.christina@arm.com>
2762
2763         PR target/84711
2764         * config/arm/arm.c (arm_can_change_mode_class): revert r258554.
2765
2766 2018-03-18  Martin Liska  <mliska@suse.cz>
2767
2768         PR rtl-optimization/84635
2769         * regrename.c (build_def_use): Use matches_mode only when
2770         matches >= 0.
2771
2772 2018-03-18  Richard Sandiford  <richard.sandiford@linaro.org>
2773
2774         PR tree-optimization/84913
2775         * tree-vect-loop.c (vectorizable_reduction): Don't try to
2776         vectorize chains of COND_EXPRs.
2777
2778 2018-03-18  Chung-Ju Wu  <jasonwucj@gmail.com>
2779
2780         * config/nds32/nds32.h (MAX_REGS_PER_ADDRESS): Fix the value.
2781
2782 2018-03-18  Chung-Ju Wu  <jasonwucj@gmail.com>
2783
2784         * config/nds32/nds32.h (LOGICAL_OP_NON_SHORT_CIRCUIT): Define.
2785
2786 2018-03-18  Chung-Ju Wu  <jasonwucj@gmail.com>
2787
2788         * config/nds32/nds32.h (CLZ_DEFINED_VALUE_AT_ZERO): Define.
2789
2790 2018-03-17  Chung-Ju Wu  <jasonwucj@gmail.com>
2791             Kito Cheng  <kito.cheng@gmail.com>
2792
2793         * config/nds32/nds32-protos.h (nds32_adjust_reg_alloc_order): Declare.
2794         * config/nds32/nds32.c (nds32_reg_alloc_order_for_speed): New array.
2795         (nds32_adjust_reg_alloc_order): New function.
2796         * config/nds32/nds32.h (ADJUST_REG_ALLOC_ORDER): Define.
2797
2798 2018-03-17  Kito Cheng  <kito.cheng@gmail.com>
2799
2800         * config/nds32/nds32.c (nds32_asm_output_mi_thunk,
2801         nds32_print_operand, nds32_print_operand_address): Use
2802         HOST_WIDE_INT_PRINT_DEC instead.
2803
2804 2018-03-17  Chung-Ju Wu  <jasonwucj@gmail.com>
2805
2806         * config/nds32/nds32.c (nds32_register_priority): Modify cost.
2807
2808 2018-03-17  Jakub Jelinek  <jakub@redhat.com>
2809
2810         PR target/84902
2811         * config/i386/i386.c (initial_ix86_tune_features,
2812         initial_ix86_arch_features): Use unsigned HOST_WIDE_INT rather than
2813         unsigned long long.
2814         (set_ix86_tune_features): Change ix86_tune_mask from unsigned int
2815         to unsigned HOST_WIDE_INT, initialize to HOST_WIDE_INT_1U << ix86_tune
2816         rather than 1u << ix86_tune.  Formatting fix.
2817         (ix86_option_override_internal): Change ix86_arch_mask from
2818         unsigned int to unsigned HOST_WIDE_INT, initialize to
2819         HOST_WIDE_INT_1U << ix86_arch rather than 1u << ix86_arch.
2820         (ix86_function_specific_restore): Likewise.
2821
2822 2018-03-16  Jakub Jelinek  <jakub@redhat.com>
2823
2824         PR target/84899
2825         * postreload.c (reload_combine_recognize_pattern): Perform
2826         INTVAL addition in unsigned HOST_WIDE_INT type to avoid UB and
2827         truncate_int_for_mode the result for the destination's mode.
2828
2829         PR c/84909
2830         * hsa-gen.c (mem_type_for_type): Fix comment typo.
2831         * tree-vect-loop-manip.c (vect_create_cond_for_niters_checks):
2832         Likewise.
2833         * gimple-ssa-warn-restrict.c (builtin_memref::set_base_and_offset):
2834         Likewise.
2835
2836 2018-03-16  Vladimir Makarov  <vmakarov@redhat.com>
2837
2838         PR target/84876
2839         * lra-assigns.c (lra_split_hard_reg_for): Don't use
2840         regno_allocno_class_array and sorted_pseudos.
2841         * lra-constraints.c (spill_hard_reg_in_range): Ignore hard regs in
2842         insns where regno is used.
2843
2844 2018-03-16  Martin Liska  <mliska@suse.cz>
2845
2846         PR ipa/84833
2847         * multiple_target.c (create_dispatcher_calls): Redirect
2848         reference in the symbol table.
2849
2850 2018-03-16  Martin Liska  <mliska@suse.cz>
2851
2852         PR ipa/84722
2853         * multiple_target.c (create_dispatcher_calls): Redirect also
2854         an alias.
2855
2856 2018-03-16  Jakub Jelinek  <jakub@redhat.com>
2857
2858         PR c++/79937
2859         PR c++/82410
2860         * tree.h (TARGET_EXPR_NO_ELIDE): Define.
2861         * gimplify.c (gimplify_modify_expr_rhs): Don't elide TARGET_EXPRs with
2862         TARGET_EXPR_NO_ELIDE flag set unless *expr_p is INIT_EXPR.
2863
2864 2018-03-16  Julia Koval  <julia.koval@intel.com>
2865
2866         * doc/invoke.texi (Skylake Server): Add CLWB.
2867         Cannonlake): Remove CLWB.
2868
2869 2018-03-16  Jakub Jelinek  <jakub@redhat.com>
2870
2871         PR tree-optimization/84841
2872         * tree-ssa-reassoc.c (INTEGER_CONST_TYPE): Change to 1 << 4 from
2873         1 << 3.
2874         (FLOAT_ONE_CONST_TYPE): Define.
2875         (constant_type): Return FLOAT_ONE_CONST_TYPE for -1.0 and 1.0.
2876         (sort_by_operand_rank): Put entries with higher constant_type last
2877         rather than first to match comments.
2878
2879 2018-03-15  Sandra Loosemore  <sandra@codesourcery.com>
2880
2881         * config/nios2/nios2.md (movsi_internal): Fix thinko in
2882         split predicate.
2883
2884 2018-03-15  Jakub Jelinek  <jakub@redhat.com>
2885
2886         PR c++/79085
2887         * calls.c (expand_call): For TREE_ADDRESSABLE rettype ignore alignment
2888         check and use address of target always.
2889
2890 2018-03-15  H.J. Lu  <hongjiu.lu@intel.com>
2891
2892         PR target/84574
2893         * config/i386/i386.c (indirect_thunk_needed): Update comments.
2894         (indirect_thunk_bnd_needed): Likewise.
2895         (indirect_thunks_used): Likewise.
2896         (indirect_thunks_bnd_used): Likewise.
2897         (indirect_return_needed): New.
2898         (indirect_return_bnd_needed): Likewise.
2899         (output_indirect_thunk_function): Add a bool argument for
2900         function return.
2901         (output_indirect_thunk_function): Don't generate alias for
2902         function return thunk.
2903         (ix86_code_end): Call output_indirect_thunk_function to generate
2904         function return thunks.
2905         (ix86_output_function_return): Set indirect_return_bnd_needed
2906         and indirect_return_needed instead of indirect_thunk_bnd_needed
2907         and indirect_thunk_needed.
2908
2909 2018-03-15  Olga Makhotina  <olga.makhotina@intel.com>
2910
2911         * config/i386/sgxintrin.h (_enclv_u32): New intrinsic.
2912         (__enclv_bc, __enclv_cd, __enclv_generic): New definitions.
2913         (ERDINFO, ETRACKC, ELDBC, ELDUC): New leaves.
2914
2915 2018-03-15  David Malcolm  <dmalcolm@redhat.com>
2916             Paul Hua <paul.hua.gm@gmail.com>
2917
2918         PR c/84852
2919         * gcc.dg/fixits-pr84852-1.c: Fix filename in dg-regexp.
2920
2921 2018-03-15  Segher Boessenkool  <segher@kernel.crashing.org>
2922
2923         * config/rs6000/rs6000.c (abi_v4_pass_in_fpr): Add back the
2924         TARGET_DOUBLE_FLOAT and TARGET_SINGLE_FLOAT conditions on the DFmode
2925         resp. SFmode cases.
2926
2927 2018-03-15  Tamar Christina  <tamar.christina@arm.com>
2928
2929         PR target/84711
2930         * config/arm/arm.c (arm_can_change_mode_class): Use GET_MODE_UNIT_SIZE
2931         instead of GET_MODE_SIZE when comparing Units.
2932
2933 2018-03-15  Vladimir Mezentsev  <vladimir.mezentsev@oracle.com>
2934
2935         PR target/68256
2936         * varasm.c (hash_section): Return an unchangeble hash value
2937         * config/aarch64/aarch64.c (aarch64_use_blocks_for_constant_p):
2938         Return !aarch64_can_use_per_function_literal_pools_p ().
2939
2940 2018-03-15  Jakub Jelinek  <jakub@redhat.com>
2941
2942         PR target/84860
2943         * optabs.c (emit_conditional_move): Pass address of cmode's copy
2944         rather than address of cmode as last argument to prepare_cmp_insn.
2945
2946 2018-03-15  Julia Koval  <julia.koval@intel.com>
2947
2948         * config/i386/i386.c (F_AVX512VBMI2, F_GFNI, F_VPCLMULQDQ,
2949         F_AVX512VNNI, F_AVX512BITALG): New.
2950
2951 2018-03-14  John David Anglin  <danglin@gcc.gnu.org>
2952
2953         PR target/83451
2954         * config/pa/pa.c (pa_emit_move_sequence):  Always emit secondary reload
2955         insn for floating-point loads and stores.
2956
2957 2018-03-14  Carl Love  <cel@us.ibm.com>
2958
2959         * config/rs6000/rs6000-c.c: Add macro definitions for
2960         ALTIVEC_BUILTIN_VEC_PERMXOR.
2961         * config/rs6000/rs6000.h: Add #define for vec_permxor builtin.
2962         * config/rs6000/rs6000-builtin.def: Add macro expansions for VPERMXOR.
2963         * config/rs6000/altivec.md (altivec_vpermxor): New define expand.
2964         * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Add case
2965         UNSPEC_VPERMXOR.
2966         * config/doc/extend.texi: Add prototypes for vec_permxor.
2967
2968 2018-03-14  David Malcolm  <dmalcolm@redhat.com>
2969
2970         PR c/84852
2971         * diagnostic-show-locus.c (class layout_point): Convert m_line
2972         from int to linenum_type.
2973         (line_span::comparator): Use linenum "compare" function when
2974         comparing line numbers.
2975         (test_line_span): New function.
2976         (layout_range::contains_point): Convert param "row" from int to
2977         linenum_type.
2978         (layout_range::intersects_line_p): Likewise.
2979         (layout::will_show_line_p): Likewise.
2980         (layout::print_source_line): Likewise.
2981         (layout::should_print_annotation_line_p): Likewise.
2982         (layout::print_annotation_line): Likewise.
2983         (layout::print_leading_fixits): Likewise.
2984         (layout::annotation_line_showed_range_p): Likewise.
2985         (struct line_corrections): Likewise for field m_row.
2986         (line_corrections::line_corrections): Likewise for param "row".
2987         (layout::print_trailing_fixits): Likewise.
2988         (layout::get_state_at_point): Likewise.
2989         (layout::get_x_bound_for_row): Likewise.
2990         (layout::print_line): Likewise.
2991         (diagnostic_show_locus): Likewise for locals "last_line" and
2992         "row".
2993         (selftest::diagnostic_show_locus_c_tests): Call test_line_span.
2994         * input.c (selftest::test_linenum_comparisons): New function.
2995         (selftest::input_c_tests): Call it.
2996         * selftest.c (selftest::test_assertions): Test ASSERT_GT,
2997         ASSERT_GT_AT, ASSERT_LT, and ASSERT_LT_AT.
2998         * selftest.h (ASSERT_GT): New macro.
2999         (ASSERT_GT_AT): New macro.
3000         (ASSERT_LT): New macro.
3001         (ASSERT_LT_AT): New macro.
3002
3003 2018-03-14  Segher Boessenkool  <segher@kernel.crashing.org>
3004
3005         PR rtl-optimization/84780
3006         * combine.c (distribute_links): Don't make a link based on pc_rtx.
3007
3008 2018-03-14  Martin Liska  <mliska@suse.cz>
3009
3010         * tree.c (record_node_allocation_statistics): Use
3011         get_stats_node_kind.
3012         (get_stats_node_kind): New function extracted from
3013         record_node_allocation_statistics.
3014         (free_node): Use get_stats_node_kind.
3015
3016 2018-03-14  Richard Biener  <rguenther@suse.de>
3017
3018         * tree-ssa-pre.c (compute_antic_aux): Remove code that asserts
3019         that the value-set of ANTIC_IN doesn't grow.
3020
3021         Revert
3022         * tree-ssa-pre.c (struct bb_bitmap_sets): Add visited_with_visited_succs
3023         member.
3024         (BB_VISITED_WITH_VISITED_SUCCS): New define.
3025         (compute_antic): Initialize BB_VISITED_WITH_VISITED_SUCCS.
3026
3027 2018-03-14  Julia Koval  <julia.koval@intel.com>
3028
3029         * config.gcc (icelake-client, icelake-server): New.
3030         (icelake): Remove.
3031         * config/i386/i386.c (initial_ix86_tune_features): Extend to 64 bit.
3032         (initial_ix86_arch_features): Ditto.
3033         (PTA_SKYLAKE): Add SGX.
3034         (PTA_ICELAKE): Remove.
3035         (PTA_ICELAKE_CLIENT): New.
3036         (PTA_ICELAKE_SERVER): New.
3037         (ix86_option_override_internal): Split up icelake on icelake client and
3038         icelake server.
3039         (get_builtin_code_for_version): Ditto.
3040         (fold_builtin_cpu): Ditto.
3041         * config/i386/driver-i386.c (config/i386/driver-i386.c): Ditto.
3042         * config/i386/i386-c.c (ix86_target_macros_internal): Ditto
3043         * config/i386/i386.h (processor_type): Ditto.
3044         * doc/invoke.texi: Ditto.
3045
3046 2018-03-14  Jakub Jelinek  <jakub@redhat.com>
3047
3048         PR sanitizer/83392
3049         * sanopt.c (maybe_optimize_ubsan_ptr_ifn): Handle also
3050         INTEGER_CST offset, add it together with bitpos / 8 and
3051         sign extend based on POINTER_SIZE.
3052
3053         PR target/84844
3054         Revert
3055         2017-04-20  Uros Bizjak  <ubizjak@gmail.com>
3056
3057         PR target/78090
3058         * config/i386/constraints.md (Yc): New register constraint.
3059         * config/i386/i386.md (*float<SWI48:mode><MODEF:mode>2_mixed):
3060         Use Yc constraint for alternative 2 of operand 0.  Remove
3061         preferred_for_speed attribute.
3062
3063 2018-03-14  Richard Biener  <rguenther@suse.de>
3064
3065         PR tree-optimization/84830
3066         * tree-ssa-pre.c (compute_antic_aux): Intersect the new ANTIC_IN
3067         with the old one to avoid oscillations.
3068
3069 2018-03-13  Vladimir Makarov  <vmakarov@redhat.com>
3070
3071         PR target/83712
3072         * lra-assigns.c (find_all_spills_for): Ignore uninteresting
3073         pseudos.
3074         (assign_by_spills): Return a flag of reload assignment failure.
3075         Do not process the reload assignment failures.  Do not spill other
3076         reload pseudos if they has the same reg class.  Update n if
3077         necessary.
3078         (lra_assign): Add a return arg.  Set up from the result of
3079         assign_by_spills call.
3080         (find_reload_regno_insns, lra_split_hard_reg_for): New functions.
3081         * lra-constraints.c (split_reg): Add a new arg.  Use it instead of
3082         usage_insns if it is not NULL.
3083         (spill_hard_reg_in_range): New function.
3084         (split_if_necessary, inherit_in_ebb): Pass a new arg to split_reg.
3085         * lra-int.h (spill_hard_reg_in_range, lra_split_hard_reg_for): New
3086         function prototypes.
3087         (lra_assign): Change prototype.
3088         * lra.c (lra): Add code to deal with fails by splitting hard reg
3089         live ranges.
3090
3091 2018-03-01  Palmer Dabbelt  <palmer@sifive.com>
3092
3093         * config/riscv/riscv.opt (mrelax): New option.
3094         * config/riscv/riscv.c (riscv_file_start): Emit ".option
3095         "norelax" when riscv_mrelax is disabled.
3096         * doc/invoke.texi (RISC-V): Document "-mrelax" and "-mno-relax".
3097
3098 2018-03-13  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
3099
3100         PR target/84743
3101         * config/rs6000/rs6000.c (rs6000_reassociation_width): Disable parallel
3102         reassociation for int modes.
3103
3104 2018-03-13  Richard Sandiford  <richard.sandiford@linaro.org>
3105
3106         * tree-vect-loop-manip.c (vect_maybe_permute_loop_masks):
3107         Reverse the choice between VEC_UNPACK_LO_EXPR and VEC_UNPACK_HI_EXPR
3108         for big-endian.
3109         * config/aarch64/iterators.md (hi_lanes_optab): New int attribute.
3110         * config/aarch64/aarch64-sve.md
3111         (*aarch64_sve_<perm_insn><perm_hilo><mode>): Rename to...
3112         (aarch64_sve_<perm_insn><perm_hilo><mode>): ...this.
3113         (*extend<mode><Vwide>2): Rename to...
3114         (aarch64_sve_extend<mode><Vwide>2): ...this.
3115         (vec_unpack<su>_<perm_hilo>_<mode>): Turn into a define_expand,
3116         renaming the old pattern to...
3117         (aarch64_sve_punpk<perm_hilo>_<mode>): ...this.  Only define
3118         unsigned packs.
3119         (vec_unpack<su>_<perm_hilo>_<SVE_BHSI:mode>): Turn into a
3120         define_expand, renaming the old pattern to...
3121         (aarch64_sve_<su>unpk<perm_hilo>_<SVE_BHSI:mode>): ...this.
3122         (*vec_unpacku_<perm_hilo>_<mode>_no_convert): Delete.
3123         (vec_unpacks_<perm_hilo>_<mode>): Take BYTES_BIG_ENDIAN into
3124         account when deciding which SVE instruction the optab should use.
3125         (vec_unpack<su_optab>_float_<perm_hilo>_vnx4si): Likewise.
3126
3127 2018-03-13  Richard Sandiford  <richard.sandiford@linaro.org>
3128
3129         * config/aarch64/aarch64.md (V4_REGNUM, V8_REGNUM, V12_REGNUM)
3130         (V20_REGNUM, V24_REGNUM, V28_REGNUM, P1_REGNUM, P2_REGNUM, P3_REGNUM)
3131         (P4_REGNUM, P5_REGNUM, P6_REGNUM, P8_REGNUM, P9_REGNUM, P10_REGNUM)
3132         (P11_REGNUM, P12_REGNUM, P13_REGNUM, P14_REGNUM): New define_constants.
3133         (tlsdesc_small_<mode>): Turn a define_expand and use
3134         tlsdesc_small_sve_<mode> for SVE.  Rename original define_insn to...
3135         (tlsdesc_small_advsimd_<mode>): ...this.
3136         (tlsdesc_small_sve_<mode>): New pattern.
3137
3138 2018-03-13  Richard Sandiford  <richard.sandiford@linaro.org>
3139
3140         * config/aarch64/iterators.md (UNSPEC_SMUL_HIGHPART)
3141         (UNSPEC_UMUL_HIGHPART): New constants.
3142         (MUL_HIGHPART): New int iteraor.
3143         (su): Handle UNSPEC_SMUL_HIGHPART and UNSPEC_UMUL_HIGHPART.
3144         * config/aarch64/aarch64-sve.md (<su>mul<mode>3_highpart): New
3145         define_expand.
3146         (*<su>mul<mode>3_highpart): New define_insn.
3147
3148 2018-03-13  Eric Botcazou  <ebotcazou@adacore.com>
3149
3150         PR lto/84805
3151         * ipa-devirt.c (odr_subtypes_equivalent_p): Do not get the ODR type of
3152         incomplete types.
3153
3154 2018-03-13  Martin Liska  <mliska@suse.cz>
3155
3156         PR ipa/84658.
3157         * (sem_item_optimizer::sem_item_optimizer): Initialize new
3158         vector.
3159         (sem_item_optimizer::~sem_item_optimizer): Release it.
3160         (sem_item_optimizer::merge_classes): Register variable aliases.
3161         (sem_item_optimizer::fixup_pt_set): New function.
3162         (sem_item_optimizer::fixup_points_to_sets): Likewise.
3163         * ipa-icf.h: Declare new variables and functions.
3164
3165 2018-03-13  Jakub Jelinek  <jakub@redhat.com>
3166
3167         PR middle-end/84834
3168         * match.pd ((A & C) != 0 ? D : 0): Use INTEGER_CST@2 instead of
3169         integer_pow2p@2 and test integer_pow2p in condition.
3170         (A < 0 ? C : 0): Similarly for @1.
3171
3172         PR middle-end/84831
3173         * stmt.c (parse_output_constraint): If the CONSTRAINT_LEN (*p, p)
3174         characters starting at p contain '\0' character, don't look beyond
3175         that.
3176
3177         PR target/84827
3178         * config/i386/i386.md (round<mode>2): For 387 fancy math, disable
3179         pattern if -ftrapping-math -fno-fp-int-builtin-inexact.
3180
3181         PR target/84828
3182         * reg-stack.c (change_stack): Change update_end var from int to
3183         rtx_insn *, if non-NULL don't update just BB_END (current_block), but
3184         also call set_block_for_insn on the newly added insns and rescan.
3185
3186         PR target/84786
3187         * config/i386/sse.md (sse2_loadhpd): Use Yv constraint rather than v
3188         on the last operand.
3189
3190         PR c++/84704
3191         * tree.c (stabilize_reference_1): Return save_expr (e) for
3192         STATEMENT_LIST even if it doesn't have side-effects.
3193
3194 2018-03-12  Jonathan Wakely  <jwakely@redhat.com>
3195
3196         * doc/invoke.texi (-mclflushopt): Fix spelling of option.
3197
3198 2018-03-12  Renlin Li  <renlin.li@arm.com>
3199
3200         * config/aarch64/aarch64.md (movhf_aarch64): Fix mode argument to
3201         aarch64_output_scalar_simd_mov_immediate.
3202
3203 2018-03-12  Martin Sebor  <msebor@redhat.com>
3204
3205         PR tree-optimization/83456
3206         * gimple-fold.c (gimple_fold_builtin_memory_op): Avoid warning
3207         for perfectly overlapping calls to memcpy.
3208         (gimple_fold_builtin_memory_chk): Same.
3209         (gimple_fold_builtin_strcpy): Handle no-warning.
3210         (gimple_fold_builtin_stxcpy_chk): Same.
3211         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Handle no-warning.
3212
3213 2018-03-12  Segher Boessenkool  <segher@kernel.crashing.org>
3214
3215         * config/rs6000/rs6000.c (abi_v4_pass_in_fpr): Add bool "named"
3216         parameter.  Use it for SFmode.
3217         (rs6000_function_arg_advance_1): Adjust.
3218         (rs6000_function_arg): Adjust.
3219         (rs6000_gimplify_va_arg): Pass false for that new parameter.
3220
3221 2018-03-12  Segher Boessenkool  <segher@kernel.crashing.org>
3222
3223         PR rtl-optimization/84169
3224         PR rtl-optimization/84780
3225         * combine.c (can_combine_p): Check for a 2-insn combination whether
3226         the destination register is used between the two insns, too.
3227
3228 2018-03-12  Richard Biener  <rguenther@suse.de>
3229
3230         PR tree-optimization/84803
3231         * tree-if-conv.c (ifcvt_memrefs_wont_trap): Don't do anything
3232         for refs DR analysis didn't process.
3233
3234 2018-03-12  Richard Biener  <rguenther@suse.de>
3235
3236         PR tree-optimization/84777
3237         * tree-ssa-loop-ch.c (should_duplicate_loop_header_p): For
3238         force-vectorize loops ignore whether we are optimizing for size.
3239
3240 2018-03-12  Chung-Ju Wu  <jasonwucj@gmail.com>
3241
3242         * config/nds32/nds32.c (nds32_md_asm_adjust): New function.
3243         (TARGET_MD_ASM_ADJUST): Define.
3244
3245 2018-03-12  Monk Chiang  <sh.chiang04@gmail.com>
3246             Kito Cheng  <kito.cheng@gmail.com>
3247             Chung-Ju Wu  <jasonwucj@gmail.com>
3248
3249         * config/nds32/nds32.c (nds32_compute_stack_frame,
3250         nds32_emit_stack_push_multiple, nds32_emit_stack_pop_multiple,
3251         nds32_emit_stack_v3push, nds32_emit_stack_v3pop,
3252         nds32_emit_adjust_frame, nds32_expand_prologue, nds32_expand_epilogue,
3253         nds32_expand_prologue_v3push, nds32_expand_epilogue_v3pop): Refine.
3254         * config/nds32/nds32.h (NDS32_FIRST_CALLEE_SAVE_GPR_REGNUM,
3255         NDS32_LAST_CALLEE_SAVE_GPR_REGNUM, NDS32_V3PUSH_AVAILABLE_P): New.
3256         * config/nds32/nds32.md (prologue, epilogue): Use macro
3257         NDS32_V3PUSH_AVAILABLE_P to do checking.
3258
3259 2018-03-11  Jakub Jelinek  <jakub@redhat.com>
3260
3261         PR debug/58150
3262         * dwarf2out.c (gen_enumeration_type_die): Don't guard adding
3263         DW_AT_declaration for ENUM_IS_OPAQUE on -gdwarf-4 or -gno-strict-dwarf,
3264         but on TYPE_SIZE.  Don't do anything for ENUM_IS_OPAQUE if not creating
3265         a new die.  Don't set TREE_ASM_WRITTEN if ENUM_IS_OPAQUE.  Guard
3266         addition of most attributes on !orig_type_die or the attribute not
3267         being present already.  Assert TYPE_VALUES is NULL for ENUM_IS_OPAQUE.
3268
3269 2018-03-11  Kito Cheng  <kito.cheng@gmail.com>
3270             Chung-Ju Wu  <jasonwucj@gmail.com>
3271
3272         * config/nds32/nds32.c (nds32_cpu_cpp_builtins): Modify to define
3273         __NDS32_VH__ macro.
3274         * config/nds32/nds32.opt (mvh): New option.
3275
3276 2018-03-11  Kito Cheng  <kito.cheng@gmail.com>
3277             Chung-Ju Wu  <jasonwucj@gmail.com>
3278
3279         * config/nds32/nds32-protos.h (nds32_cpu_cpp_builtins): Declare
3280         function.
3281         * config/nds32/nds32.c (nds32_cpu_cpp_builtins): New function.
3282         * config/nds32/nds32.h (TARGET_CPU_CPP_BUILTINS): Modify its
3283         definition.
3284
3285 2018-03-11  Kito Cheng  <kito.cheng@gmail.com>
3286             Chung-Ju Wu  <jasonwucj@gmail.com>
3287
3288         * config/nds32/nds32-memory-manipulation.c (nds32_expand_strlen): New
3289         function.
3290         * config/nds32/nds32-multiple.md (strlensi): New pattern.
3291         * config/nds32/nds32-protos.h (nds32_expand_strlen): Declare function.
3292
3293 2018-03-11  Monk Chiang  <sh.chiang04@gmail.com>
3294             Kito Cheng  <kito.cheng@gmail.com>
3295             Chung-Ju Wu  <jasonwucj@gmail.com>
3296
3297         * config/nds32/constants.md (unspec_element): Add UNSPEC_FFB,
3298         UNSPEC_FFMISM and UNSPEC_FLMISM.
3299         * config/nds32/nds32-intrinsic.c (bdesc_2arg): Add builtin description
3300         for ffb, ffmism and flmism.
3301         * config/nds32/nds32-intrinsic.md (unspec_ffb): Define new pattern.
3302         (unspec_ffmism): Ditto.
3303         (unspec_flmism): Ditto.
3304         (nds32_expand_builtin_impl): Check if string extension is available.
3305         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_FFB,
3306         NDS32_BUILTIN_FFMISM and NDS32_BUILTIN_FLMISM.
3307
3308 2018-03-10  Vladimir Makarov  <vmakarov@redhat.com>
3309
3310         Reverting patch:
3311         2018-03-09  Vladimir Makarov  <vmakarov@redhat.com>
3312
3313         PR target/83712
3314         * lra-assigns.c (assign_by_spills): Return a flag of reload
3315         assignment failure.  Do not process the reload assignment
3316         failures.  Do not spill other reload pseudos if they has the same
3317         reg class.
3318         (lra_assign): Add a return arg.  Set up from the result of
3319         assign_by_spills call.
3320         (find_reload_regno_insns, lra_split_hard_reg_for): New functions.
3321         * lra-constraints.c (split_reg): Add a new arg.  Use it instead of
3322         usage_insns if it is not NULL.
3323         (spill_hard_reg_in_range): New function.
3324         (split_if_necessary, inherit_in_ebb): Pass a new arg to split_reg.
3325         * lra-int.h (spill_hard_reg_in_range, lra_split_hard_reg_for): New
3326         function prototypes.
3327         (lra_assign): Change prototype.
3328         * lra.c (lra): Add code to deal with fails by splitting hard reg
3329         live ranges.
3330
3331 2018-03-10  H.J. Lu  <hongjiu.lu@intel.com>
3332
3333         PR target/84807
3334         * config/i386/i386.opt: Replace Enforcment with Enforcement.
3335
3336 2018-03-10  Alexandre Oliva  <aoliva@redhat.com>
3337
3338         PR debug/84620
3339         * dwarf2out.h (dw_val_class): Add dw_val_class_symview.
3340         (dw_val_node): Add val_symbolic_view.
3341         * dwarf2out.c (dw_line_info_table): Add symviews_since_reset.
3342         (symview_upper_bound): New.
3343         (new_line_info_table): Initialize symviews_since_reset.
3344         (dwarf2out_source_line): Count symviews_since_reset and set
3345         symview_upper_bound.
3346         (dw_val_equal_p): Handle symview.
3347         (add_AT_symview): New.
3348         (print_dw_val): Handle symview.
3349         (attr_checksum, attr_checksum_ordered): Likewise.
3350         (same_dw_val_p, size_of_die): Likewise.
3351         (value_format, output_die): Likewise.
3352         (add_high_low_attributes): Use add_AT_symview for entry_view.
3353         (dwarf2out_finish): Reset symview_upper_bound, clear
3354         zero_view_p.
3355
3356 2018-03-09  Peter Bergner  <bergner@vnet.ibm.com>
3357
3358         PR target/83969
3359         * config/rs6000/rs6000.c (rs6000_offsettable_memref_p): New prototype.
3360         Add strict argument and use it.
3361         (rs6000_split_multireg_move): Update for new strict argument.
3362         (mem_operand_gpr): Disallow all non-offsettable addresses.
3363         * config/rs6000/rs6000.md (*movdi_internal64): Use YZ constraint.
3364
3365 2018-03-09  Jakub Jelinek  <jakub@redhat.com>
3366
3367         PR target/84772
3368         * config/rs6000/rs6000.c (rs6000_gimplify_va_arg): Mark va_arg_tmp
3369         temporary TREE_ADDRESSABLE before gimplification of BUILT_IN_MEMCPY.
3370         * config/powerpcspe/powerpcspe.c (rs6000_gimplify_va_arg): Likewise.
3371
3372         PR c++/84767
3373         * tree-inline.c (copy_tree_body_r): For INDIRECT_REF of a remapped
3374         decl, use remap_type if we want to use the type.
3375
3376 2018-03-09  Martin Sebor  <msebor@redhat.com>
3377
3378         PR tree-optimization/84526
3379         * gimple-ssa-warn-restrict.c (builtin_memref::set_base_and_offset):
3380         Remove dead code.
3381         (builtin_access::generic_overlap): Be prepared to handle non-array
3382         base objects.
3383
3384 2018-03-09  Alexandre Oliva  <aoliva@redhat.com>
3385
3386         PR rtl-optimization/84682
3387         * lra-constraints.c (process_address_1): Check is_address flag
3388         for address constraints.
3389         (process_alt_operands): Likewise.
3390         * lra.c (lra_set_insn_recog_data): Pass asm operand locs to
3391         preprocess_constraints.
3392         * recog.h (preprocess_constraints): Add oploc parameter.
3393         Adjust callers.
3394         * recog.c (preprocess_constraints): Test address_operand for
3395         CT_ADDRESS constraints.
3396
3397 2018-03-09  Vladimir Makarov  <vmakarov@redhat.com>
3398
3399         PR target/83712
3400         * lra-assigns.c (assign_by_spills): Return a flag of reload
3401         assignment failure.  Do not process the reload assignment
3402         failures.  Do not spill other reload pseudos if they has the same
3403         reg class.
3404         (lra_assign): Add a return arg.  Set up from the result of
3405         assign_by_spills call.
3406         (find_reload_regno_insns, lra_split_hard_reg_for): New functions.
3407         * lra-constraints.c (split_reg): Add a new arg.  Use it instead of
3408         usage_insns if it is not NULL.
3409         (spill_hard_reg_in_range): New function.
3410         (split_if_necessary, inherit_in_ebb): Pass a new arg to split_reg.
3411         * lra-int.h (spill_hard_reg_in_range, lra_split_hard_reg_for): New
3412         function prototypes.
3413         (lra_assign): Change prototype.
3414         * lra.c (lra): Add code to deal with fails by splitting hard reg
3415         live ranges.
3416
3417 2018-03-09  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
3418
3419         PR target/83193
3420         * common/config/arm/arm-common.c (arm_parse_arch_option_name):
3421         Accept complain bool parameter.  Only emit errors if it is true.
3422         (arm_parse_cpu_option_name): Likewise.
3423         (arm_target_thumb_only): Adjust callers of the above.
3424         * config/arm/arm-protos.h (arm_parse_cpu_option_name): Adjust
3425         prototype to take a default true bool parameter.
3426         (arm_parse_arch_option_name): Likewise.
3427
3428 2018-03-09  David Malcolm  <dmalcolm@redhat.com>
3429             Francois-Xavier Coudert  <fxcoudert@gcc.gnu.org>
3430
3431         PR jit/64089
3432         PR jit/84288
3433         * Makefile.in (LD_VERSION_SCRIPT_OPTION, LD_SONAME_OPTION): New.
3434         * configure: Regenerate.
3435         * configure.ac ("linker --version-script option"): New.
3436         ("linker soname option"): New.
3437
3438 2018-03-09  Richard Biener  <rguenther@suse.de>
3439
3440         PR tree-optimization/84775
3441         * tree-if-conv.c (add_bb_predicate_gimplified_stmts): Delink
3442         immediate uses of predicate stmts and mark them modified.
3443
3444         Revert
3445         PR tree-optimization/84178
3446         * tree-if-conv.c (combine_blocks): Move insert_gimplified_predicates
3447         to caller.
3448         (version_loop_for_if_conversion): Delay update_ssa call.
3449         (tree_if_conversion): Delay update_ssa until after predicate
3450         insertion.
3451
3452 2018-03-09  Eric Botcazou  <ebotcazou@adacore.com>
3453
3454         PR target/84763
3455         * config/i386/winnt.c (i386_pe_seh_cold_init): Use small pre-allocation
3456         when the function accesses prior frames.
3457
3458 2018-03-08  Jakub Jelinek  <jakub@redhat.com>
3459
3460         PR debug/84456
3461         * dwarf2out.c (dw_loc_list): If list && loc_list->first->next, call
3462         gen_llsym, otherwise call maybe_gen_llsym.
3463
3464         PR inline-asm/84742
3465         * recog.c (asm_operand_ok): Return 0 if multi-character constraint
3466         has ',' character inside of it.
3467
3468 2018-03-08  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
3469
3470         PR target/84748
3471         * config/aarch64/aarch64.md (*compare_cstore<mode>_insn): Mark pattern
3472         as clobbering CC_REGNUM.
3473
3474 2018-03-08  Richard Biener  <rguenther@suse.de>
3475
3476         PR middle-end/84552
3477         * tree-scalar-evolution.c: Include tree-into-ssa.h.
3478         (follow_copies_to_constant): Do not follow SSA names registered
3479         for update.
3480
3481 2018-03-08  Richard Biener  <rguenther@suse.de>
3482
3483         PR tree-optimization/84178
3484         * tree-if-conv.c (combine_blocks): Move insert_gimplified_predicates
3485         to caller.
3486         (version_loop_for_if_conversion): Delay update_ssa call.
3487         (tree_if_conversion): Delay update_ssa until after predicate
3488         insertion.
3489
3490 2018-03-08  David Malcolm  <dmalcolm@redhat.com>
3491
3492         PR tree-optimization/84178
3493         * tree-if-conv.c (release_bb_predicate): Remove the
3494         the assertion that the stmts have NULL use_ops.
3495         Discard the statements, asserting that they haven't
3496         yet been added to a BB.
3497
3498 2018-03-08  Richard Biener  <rguenther@suse.de>
3499
3500         PR tree-optimization/84746
3501         * tree-ssa-pre.c (find_leader_in_sets): Deal with SET1 being NULL.
3502         (phi_translate): Pass in destination ANTIC_OUT set.
3503         (phi_translate_1): Likewise.  For a simplified result lookup
3504         a leader in ANTIC_OUT and AVAIL_OUT, not the ANTIC_IN sets.
3505         (phi_translate_set): Adjust.
3506         (do_pre_regular_insertion): Likewise.
3507         (do_pre_partial_partial_insertion): Likewise.
3508
3509 2018-03-08  Martin Liska  <mliska@suse.cz>
3510
3511         PR gcov-profile/84735
3512         * doc/gcov.texi: Document usage of profile files.
3513         * gcov-io.h: Document changes in the format.
3514
3515 2018-03-08  Alexandre Oliva  <aoliva@redhat.com>
3516
3517         PR debug/84404
3518         PR debug/84408
3519         * dwarf2out.c (struct dw_line_info_table): Update comments for
3520         view == -1.
3521         (FORCE_RESET_NEXT_VIEW): New.
3522         (FORCE_RESETTING_VIEW_P): New.
3523         (RESETTING_VIEW_P): Check for -1 too.
3524         (ZERO_VIEW_P): Likewise.
3525         (new_line_info_table): Force-reset next view.
3526         (dwarf2out_begin_function): Likewise.
3527         (dwarf2out_source_line): Simplify zero_view_p initialization.
3528         Test FORCE_RESETTING_VIEW_P and RESETTING_VIEW_P instead of
3529         view directly.  Omit view when omitting .loc at line 0.
3530
3531 2018-03-08  Jakub Jelinek  <jakub@redhat.com>
3532
3533         PR tree-optimization/84740
3534         * tree-switch-conversion.c (process_switch): Call build_constructors
3535         only if info.phi_count is non-zero.
3536
3537         PR tree-optimization/84739
3538         * tree-tailcall.c (find_tail_calls): Check call arguments against
3539         DECL_ARGUMENTS (current_function_decl) rather than
3540         DECL_ARGUMENTS (func) when checking for tail recursion.
3541
3542 2018-03-07  Jakub Jelinek  <jakub@redhat.com>
3543
3544         * doc/contrib.texi: Add entries for Martin Liska, David Malcolm,
3545         Marek Polacek, extend Vladimir Makarov's, Jonathan Wakely's and
3546         Volker Reichelt's entry and add entries for people that perform
3547         GCC fuzzy testing and report numerous bugs.
3548
3549 2018-03-07  Segher Boessenkool  <segher@kernel.crashing.org>
3550
3551         PR target/82411
3552         * config/rs6000/rs6000.c (rs6000_elf_in_small_data_p): Don't put
3553         readonly data in sdata, if that is disabled.
3554         * config/rs6000/sysv4.opt (mreadonly-in-sdata): New option.
3555         * doc/invoke.texi (RS/6000 and PowerPC Options): Document
3556         -mreadonly-in-sdata option.
3557
3558 2018-03-07  Martin Sebor  <msebor@redhat.com>
3559
3560         PR tree-optimization/84468
3561         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Consider successor
3562         basic block when looking for nul assignment.
3563
3564 2018-03-07  Eric Botcazou  <ebotcazou@adacore.com>
3565
3566         PR target/84277
3567         * except.h (output_function_exception_table): Adjust prototype.
3568         * except.c (output_function_exception_table): Remove FNNAME parameter
3569         and add SECTION parameter.  Ouput one part of the table at a time.
3570         * final.c (final_scan_insn_1) <NOTE_INSN_SWITCH_TEXT_SECTIONS>: Output
3571         the first part of the exception table and emit unwind directives.
3572         * config/i386/i386-protos.h (i386_pe_end_cold_function): Declare.
3573         (i386_pe_seh_cold_init): Likewise.
3574         * config/i386/cygming.h (ASM_DECLARE_COLD_FUNCTION_NAME): New macro.
3575         (ASM_DECLARE_COLD_FUNCTION_SIZE): Likewise.
3576         * config/i386/i386.c (x86_expand_epilogue): Fix wording in comment.
3577         (ix86_output_call_insn): Emit a nop in one more case for SEH.
3578         * config/i386/winnt.c: Include except.h.
3579         (struct seh_frame_state): Add reg_offset, after_prologue and
3580         in_cold_section fields.
3581         (i386_pe_seh_end_prologue): Set seh->after_prologue.
3582         (i386_pe_seh_cold_init): New function.
3583         (i386_pe_seh_fini): Add COLD parameter and bail out if it is not equal
3584         to seh->in_cold_section.
3585         (seh_emit_push): Record the offset of the push.
3586         (seh_emit_save): Record the offet of the save.
3587         (i386_pe_seh_unwind_emit): Deal with NOTE_INSN_SWITCH_TEXT_SECTIONS.
3588         Test seh->after_prologue to disregard the epilogue.
3589         (i386_pe_end_function): Pass FALSE to i386_pe_seh_fini.
3590         (i386_pe_end_cold_function): New function.
3591
3592 2018-03-07  Jakub Jelinek  <jakub@redhat.com>
3593
3594         PR fortran/84565
3595         * config/aarch64/predicates.md (aarch64_simd_reg_or_zero): Use
3596         aarch64_simd_or_scalar_imm_zero rather than aarch64_simd_imm_zero.
3597
3598         PR c++/84704
3599         * gimple-expr.c (create_tmp_var_raw): Set DECL_NAMELESS flag
3600         on tmp_var.
3601         * tree-pretty-print.c (dump_decl_name): For TDF_COMPARE_DEBUG,
3602         don't print names of DECL_NAMELESS DECL_IGNORED_P decls.
3603
3604         PR middle-end/84723
3605         * multiple_target.c: Include tree-inline.h and intl.h.
3606         (expand_target_clones): Diagnose and fail if node->definition and
3607         !tree_versionable_function_p (node->decl).
3608
3609 2018-03-06  John David Anglin  <danglin@gcc.gnu.org>
3610
3611         * config/pa/pa.h (ASM_GENERATE_INTERNAL_LABEL): Revise to use
3612         sprint_ul.
3613         (ASM_OUTPUT_ADDR_VEC_ELT): Revise for above change.
3614         (ASM_OUTPUT_ADDR_DIFF_ELT): Likewise.
3615         * config/pa/pa64-hpux.h (ASM_GENERATE_INTERNAL_LABEL): Revise as above.
3616
3617 2018-03-06  Jakub Jelinek  <jakub@redhat.com>
3618
3619         PR target/84710
3620         * combine.c (try_combine): Use reg_or_subregno instead of handling
3621         just paradoxical SUBREGs and REGs.
3622
3623 2018-03-06  Claudiu Zissulescu  <claziss@synopsys.com>
3624
3625          * config/arc/arc.c (arc_finalize_pic): Remove function.
3626          (arc_must_save_register): We use single base PIC register, remove
3627          checks to save/restore the PIC register.
3628          (arc_expand_prologue): Likewise.
3629          * config/arc/arc-protos.h (arc_set_default_type_attributes):
3630          Remove.
3631          (arc_verify_short): Likewise.
3632          (arc_attr_type): Likewise.
3633          * config/arc/arc.c (arc_set_default_type_attributes): Remove.
3634          (walk_stores): Likewise.
3635          (arc_address_cost): Make it static.
3636          (arc_verify_short): Likewise.
3637          (branch_dest): Likewise.
3638          (arc_attr_type): Likewise.
3639          * config/arc/arc.c (TARGET_ADJUST_INSN_LENGTH): Remove.
3640          (TARGET_INSN_LENGTH_PARAMETERS): Likewise.
3641          (arc_final_prescan_insn): Remove inserting the nops due to
3642          hardware hazards.  It is done in reorg step.
3643          (insn_length_variant_t): Remove.
3644          (insn_length_parameters_t): Likewise.
3645          (arc_insn_length_parameters): Likewise.
3646          (arc_get_insn_variants): Likewise.
3647          * config/arc/arc.h (TARGET_UPSIZE_DBR): Remove.
3648
3649 2018-03-06  Jakub Jelinek  <jakub@redhat.com>
3650
3651         PR inline-asm/84683
3652         * reg-stack.c (move_for_stack_reg): If any_malformed_asm, avoid
3653         assertion failure.
3654
3655         PR tree-optimization/84687
3656         * omp-simd-clone.c (simd_clone_create): Clear DECL_BUILT_IN_CLASS
3657         on new_node->decl.
3658         * match.pd (pow(C,x)*expN(y) -> expN(logN(C)*x+y)): New optimization.
3659
3660 2018-03-05  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
3661
3662         * config/rs6000/rs6000-builtin.def (rs6000_speculation_barrier):
3663         Rename to ppc_speculation_barrier.
3664         * config/rs6000/rs6000.c (rs6000_init_builtins): Rename builtin to
3665         __builtin_ppc_speculation_barrier.
3666
3667 2018-03-05  Jakub Jelinek  <jakub@redhat.com>
3668
3669         PR target/84700
3670         * combine.c (combine_simplify_rtx): Don't try to simplify if
3671         if_then_else_cond returned non-NULL, but either true_rtx or false_rtx
3672         are equal to x.
3673
3674 2018-03-05  Segher Boessenkool  <segher@kernel.crashing.org>
3675
3676         * config/rs6000/rs6000.c (rs6000_loop_align): Don't align tiny loops
3677         to 32 bytes when compiling for POWER9.
3678
3679 2018-03-05  Jakub Jelinek  <jakub@redhat.com>
3680
3681         PR target/84564
3682         * config/i386/i386.c (ix86_function_ok_for_sibcall): Check for
3683         regparm >= 3 with no arg reg available also for calls with
3684         flag_force_indirect_call.  Pass decl to ix86_function_regparm.
3685
3686         PR target/84524
3687         * config/i386/sse.md (*<code><mode>3): Replace <mask_prefix3> with
3688         orig,vex.
3689         (*<plusminus_insn><mode>3): Likewise.  Remove <mask_operand3> uses.
3690
3691 2018-03-05  Peter Bergner  <bergner@vnet.ibm.com>
3692
3693         PR target/84264
3694         * config/rs6000/vector.md (mov<mode>): Disallow altivec memory operands.
3695
3696 2018-03-05  Richard Biener  <rguenther@suse.de>
3697
3698         PR tree-optimization/84486
3699         * tree-ssa-pre.c (create_expression_by_pieces): Remove dead code.
3700         When inserting a __builtin_assume_aligned call set the LHS
3701         SSA name alignment info accordingly.
3702
3703 2018-03-05  Wilco Dijkstra  <wdijkstr@arm.com>
3704
3705         PR tree-optimization/84114
3706         * config/aarch64/aarch64.c (aarch64_reassociation_width)
3707         Avoid reassociation of FLOAT_MODE addition.
3708
3709 2018-03-05  Olga Makhotina  <olga.makhotina@intel.com>
3710
3711         * common/config/i386/i386-common.c (OPTION_MASK_ISA_PCONFIG_SET,
3712         OPTION_MASK_ISA_PCONFIG_UNSET, OPTION_MASK_ISA_WBNOINVD_SET,
3713         OPTION_MASK_ISA_WBNOINVD_UNSET): New definitions.
3714         (ix86_handle_option): Handle -mpconfig and -mwbnoinvd.
3715         * config.gcc (pconfigintrin.h, wbnoinvdintrin.h) : Add headers.
3716         * config/i386/cpuid.h (bit_PCONFIG, bit_WBNOINVD): New.
3717         * config/i386/driver-i386.c (host_detect_local_cpu): Detect -mpconfig
3718         and -mwbnoinvd.
3719         * config/i386/i386-builtin.def (__builtin_ia32_wbnoinvd,
3720         __builtin_ia32_wbinvd): New builtins.
3721         (SPECIAL_ARGS2): New.
3722         * config/i386/i386-c.c (__WBNOINVD__, __PCONFIG__): New.
3723         (SPECIAL_ARGS2): New.
3724         * config/i386/i386.c (ix86_target_string): Add -mpconfig and -mwbnoinvd.
3725         (ix86_valid_target_attribute_inner_p): Ditto.
3726         (ix86_init_mmx_sse_builtins): Add special_args2.
3727         * config/i386/i386.h (TARGET_PCONFIG, TARGET_PCONFIG_P, TARGET_WBNOINVD,
3728         TARGET_WBNOINVD_P): New.
3729         * config/i386/i386.md (UNSPECV_WBINVD, UNSPECV_WBNOINVD): New.
3730         (define_insn "wbinvd", define_insn "wbnoinvd"): New.
3731         * config/i386/i386.opt: Add -mpconfig and -mwbnoinvd.
3732         * config/i386/immintrin.h (_wbinvd): New intrinsic.
3733         * config/i386/pconfigintrin.h: New file.
3734         * config/i386/wbnoinvdintrin.h: Ditto.
3735         * config/i386/x86intrin.h: Add headers pconfigintrin.h and wbnoinvdintrin.h.
3736         * doc/invoke.texi (-mpconfig, -mwbnoinvd): New.
3737
3738 2018-03-05  Richard Biener  <rguenther@suse.de>
3739
3740         PR tree-optimization/84670
3741         * tree-ssa-pre.c (struct bb_bitmap_sets): Add visited_with_visited_succs
3742         member.
3743         (BB_VISITED_WITH_VISITED_SUCCS): New define.
3744         (compute_antic): Initialize BB_VISITED_WITH_VISITED_SUCCS.
3745         (compute_antic_aux): Only assert the number of values in ANTIC_IN
3746         doesn't grow if all successors (recursively) were visited at least
3747         once.
3748
3749 2018-03-05  Richard Biener  <rguenther@suse.de>
3750
3751         PR tree-optimization/84650
3752         * tree-ssa-loop-im.c (pass_lim::execute): Reset the SCEV cache
3753         if executed in the loop pipeline.
3754
3755 2018-03-05  Sandra Loosemore  <sandra@codesourcery.com>
3756
3757         * doc/configfiles.texi (Configuration Files): Move info about
3758         conditionalizing $target-protos.h to...
3759         * doc/sourcebuild.texi (Back End): Here.  Explain how $target.h
3760         differs from $target-protos.h.
3761
3762 2018-03-05  Kito Cheng  <kito.cheng@gmail.com>
3763             Chung-Ju Wu  <jasonwucj@gmail.com>
3764
3765         * config/nds32/nds32-protos.h (nds32_expand_setmem): Declare.
3766         * config/nds32/nds32-multiple.md (setmemsi): Define.
3767         * config/nds32/nds32-memory-manipulation.c
3768         (nds32_gen_dup_4_byte_to_word_value): New.
3769         (emit_setmem_word_loop): New.
3770         (emit_setmem_byte_loop): New.
3771         (nds32_expand_setmem_loop): New.
3772         (nds32_expand_setmem_loop_v3m): New.
3773         (nds32_expand_setmem_unroll): New.
3774         (nds32_expand_setmem): New.
3775
3776 2018-03-04  Kito Cheng  <kito.cheng@gmail.com>
3777             Chung-Ju Wu  <jasonwucj@gmail.com>
3778
3779         * config/nds32/nds32-memory-manipulation.c
3780         (nds32_emit_load_store): New.
3781         (nds32_emit_post_inc_load_store): New.
3782         (nds32_emit_mem_move): New.
3783         (nds32_emit_mem_move_block): New.
3784         (nds32_expand_movmemsi_loop_unknown_size): New.
3785         (nds32_expand_movmemsi_loop_known_size): New.
3786         (nds32_expand_movmemsi_loop): New.
3787         (nds32_expand_movmemsi_unroll): New.
3788         (nds32_expand_movmemqi): Rename ...
3789         (nds32_expand_movmemsi): ... to this.
3790         * config/nds32/nds32-multiple.md (movmemqi): Rename ...
3791         (movmemsi): ... to this.
3792         * config/nds32/nds32-protos.h (nds32_expand_movmemqi): Rename ...
3793         (nds32_expand_movmemsi): ... to this.
3794
3795 2018-03-04  Kito Cheng  <kito.cheng@gmail.com>
3796             Monk Chiang  <sh.chiang04@gmail.com>
3797             Chung-Ju Wu  <jasonwucj@gmail.com>
3798
3799         * config/nds32/nds32-protos.h
3800         (nds32_expand_load_multiple): New arguments.
3801         (nds32_expand_store_multiple): Ditto.
3802         (nds32_valid_multiple_load_store): Rename ...
3803         (nds32_valid_multiple_load_store_p): ... to this.
3804         * config/nds32/nds32-memory-manipulation.c
3805         (nds32_expand_load_multiple): Refine implementation.
3806         (nds32_expand_store_multiple): Ditto.
3807         * config/nds32/nds32-multiple.md
3808         (load_multiple): Update nds32_expand_load_multiple interface.
3809         (store_multiple): Update nds32_expand_store_multiple interface.
3810         * config/nds32/nds32-predicates.c
3811         (nds32_valid_multiple_load_store): Rename ...
3812         (nds32_valid_multiple_load_store_p): ... to this and refine
3813         implementation.
3814         * config/nds32/predicates.md
3815         (nds32_load_multiple_and_update_address_operation): New predicate.
3816         (nds32_store_multiple_and_update_address_operation): New predicate.
3817
3818 2018-03-04  Kito Cheng  <kito.cheng@gmail.com>
3819             Chung-Ju Wu  <jasonwucj@gmail.com>
3820
3821         * config/nds32/nds32.md (type): Add load_multiple and store_multiple.
3822         (combo): New attribute.
3823         * config/nds32/nds32-multiple.md: Refine patterns with new attributes.
3824
3825 2018-03-03  Chung-Ju Wu  <jasonwucj@gmail.com>
3826
3827         * config/nds32/nds32.opt: Change -mcmodel= default value.
3828
3829 2018-03-03  Kito Cheng  <kito.cheng@gmail.com>
3830             Monk Chiang  <sh.chiang04@gmail.com>
3831             Chung-Ju Wu  <jasonwucj@gmail.com>
3832
3833         * config/nds32/constants.md (unspec_element): New enum.
3834         * config/nds32/constraints.md (Umw): New constraint.
3835         * config/nds32/nds32-intrinsic.c: Add more builtin functions.
3836         * config/nds32/nds32-intrinsic.md: Likewise.
3837         * config/nds32/nds32-md-auxiliary.c (nds32_regno_to_enable4): New.
3838         (nds32_valid_smw_lwm_base_p): New.
3839         (nds32_output_smw_single_word): New.
3840         (nds32_output_lmw_single_word): New.
3841         (nds32_expand_unaligned_load): New.
3842         (nds32_expand_unaligned_store): New.
3843         * config/nds32/nds32-protos.h (nds32_valid_smw_lwm_base_p): Declare.
3844         (nds32_output_smw_single_word): Declare.
3845         (nds32_output_lmw_single_word): Declare.
3846         (nds32_expand_unaligned_load): Declare.
3847         (nds32_expand_unaligned_store): Declare.
3848         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_UALOAD_HW,
3849         NDS32_BUILTIN_UALOAD_W, NDS32_BUILTIN_UALOAD_DW,
3850         NDS32_BUILTIN_UASTORE_HW, NDS32_BUILTIN_UASTORE_W,
3851         NDS32_BUILTIN_UASTORE_DW.
3852         * config/nds32/predicates.md (nds32_lmw_smw_base_operand): New
3853         predicate.
3854
3855 2018-03-03  Monk Chiang  <sh.chiang04@gmail.com>
3856             Kito Cheng  <kito.cheng@gmail.com>
3857             Chung-Ju Wu  <jasonwucj@gmail.com>
3858
3859         * config/nds32/nds32-intrinsic.c
3860         (nds32_expand_builtin_null_ftype_reg): Delete.
3861         (nds32_expand_builtin_reg_ftype_imm): Ditto.
3862         (nds32_expand_builtin_null_ftype_reg_imm): Ditto.
3863         (nds32_read_argument): New.
3864         (nds32_legitimize_target): Ditto.
3865         (nds32_legitimize_argument): Ditto.
3866         (nds32_check_constant_argument): Ditto.
3867         (nds32_expand_unop_builtin): Ditto.
3868         (nds32_expand_unopimm_builtin): Ditto.
3869         (nds32_expand_binop_builtin): Ditto.
3870         (nds32_builtin_decl_impl): Ditto.
3871         (builtin_description): Ditto.
3872         (nds32_expand_builtin_impl): Rewrite with new infrastructure.
3873         (nds32_init_builtins_impl): Ditto.
3874         * config/nds32/nds32.c (TARGET_BUILTIN_DECL): Define.
3875         (nds32_builtin_decl): New.
3876         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_COUNT.
3877         * config/nds32/nds32-protos.h (nds32_builtin_decl_impl): Declare.
3878
3879 2018-03-02  Jeff Law  <law@redhat.com>
3880
3881         * reorg.c (stop_search_p): Handle DEBUG_INSN.
3882         (redundant_insn, fill_simple_delay_slots): Likewise.
3883         (fill_slots_from_thread): Likewise.
3884         * resource.c (mark_referenced_resources): Likewise.
3885         (mark_set_resources, find_dead_or_set_registers): Likewise.
3886
3887 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
3888
3889         * substring-locations.h (format_warning_va): Formatting fix for
3890         ATTRIBUTE_GCC_DIAG.
3891         (format_warning_at_substring): Fix up ATTRIBUTE_GCC_DIAG second
3892         argument.
3893         (format_warning_n_va, format_warning_at_substring_n): New prototypes.
3894         * substring-locations.c: Include intl.h.
3895         (format_warning_va): Turned into small wrapper around
3896         format_warning_n_va, renamed to ...
3897         (format_warning_n_va): ... this, add N and PLURAL_GMSGID arguments,
3898         rename GMSGID to SINGULAR_GMSGID, if SINGULAR_GMSGID != PLURAL_GMSGID,
3899         use ngettext.
3900         (format_warning_at_substring_n): New function.
3901         * gimple-ssa-sprintf.c: Remove GCC diagnostic ignored pragma.
3902         (fmtwarn): Add ATTRIBUTE_GCC_DIAG.  Turn into a copy of
3903         format_warning_at_substring with just a shorter name instead of
3904         const function pointer.
3905         (fmtwarn_n): New function.
3906         (maybe_warn, format_directive, parse_directive): Use fmtwarn_n where
3907         appropriate, get rid of all the fmtstr temporaries, move conditionals
3908         with G_() wrapped string literals directly into fmtwarn arguments,
3909         cast dir.len to (int), formatting fixes.
3910
3911 2018-03-02  Thomas Schwinge  <thomas@codesourcery.com>
3912
3913         * doc/invoke.texi: Remove "Cilk Plus" references.
3914
3915 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
3916             Richard Biener  <rguenther@suse.de>
3917
3918         PR ipa/84628
3919         * expr.c (expand_expr_real_1) <case CALL_EXPR>: Don't emit diagnostics
3920         for error or warning attributes if CALL_FROM_THUNK_P is set.
3921         Formatting fixes.
3922
3923 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
3924
3925         PR target/56540
3926         * config/pa/pa.h (TARGET_CPU_CPP_BUILTINS): Predefine
3927         __SIZEOF_128__ macro if HPUX_LONG_DOUBLE_LIBRARY.
3928
3929         PR target/56540
3930         * config/ia64/ia64.h (TARGET_CPU_CPP_BUILTINS): Predefine
3931         __SIZEOF_{FPREG,FLOAT{80,128}}__ macros.
3932
3933         * predict.c (test_prediction_value_range): Use PROB_UNINITIALIZED
3934         instead of -1U in last predictors element's probability member.
3935
3936 2018-03-02  Eric Botcazou  <ebotcazou@adacore.com>
3937
3938         PR ipa/83983
3939         * ipa-devirt.c (odr_subtypes_equivalent_p): Get the ODR type of both
3940         arguments if they are comparable.
3941
3942 2018-03-02  Richard Sandiford  <richard.sandiford@linaro.org>
3943
3944         PR tree-optimization/84634
3945         * tree-vect-stmts.c (vectorizable_store, vectorizable_load): Replace
3946         masks and masked_loop_p with a single loop_masks, making sure it's
3947         null for bb vectorization.
3948
3949 2018-03-02  Richard Sandiford  <richard.sandiford@linaro.org>
3950
3951         * tree-vect-data-refs.c (vect_analyze_data_ref_dependence)
3952         (vect_analyze_data_ref_access): Use loop->safe_len rather than
3953         loop->force_vectorize to check whether there is no alias.
3954
3955 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
3956
3957         PR target/84614
3958         * rtl.h (prev_real_nondebug_insn, next_real_nondebug_insn): New
3959         prototypes.
3960         * emit-rtl.c (next_real_insn, prev_real_insn): Fix up function
3961         comments.
3962         (next_real_nondebug_insn, prev_real_nondebug_insn): New functions.
3963         * cfgcleanup.c (try_head_merge_bb): Use prev_real_nondebug_insn
3964         instead of a loop around prev_real_insn.
3965         * combine.c (move_deaths): Use prev_real_nondebug_insn instead of
3966         prev_real_insn.
3967
3968         PR inline-asm/84625
3969         * config/i386/i386.c (ix86_print_operand): Use conditional
3970         output_operand_lossage instead of gcc_assert if CONST_VECTOR is not
3971         zero vector.
3972
3973 2018-03-02  Richard Biener  <rguenther@suse.de>
3974
3975         PR tree-optimization/84427
3976         * tree-ssa-pre.c (bitmap_remove_expr_from_set): Remove.
3977         (bitmap_set_subtract_values): Rewrite to handle multiple
3978         exprs per value.
3979         (clean): Likewise.
3980         (prune_clobbered_mems): Likewise.
3981         (phi_translate): Take edge instead of pred/phiblock.
3982         (phi_translate_1): Likewise.
3983         (phi_translate_set): Likewise.  Insert all translated
3984         exprs for a value into the set, keeping possibly multiple
3985         expressions per value.
3986         (compute_antic_aux): Adjust for phi_translate changes.
3987         When intersecting union the expressions and prune those
3988         not in the final value set, keeping possibly multiple
3989         expressions per value.  Do not use value-insertion
3990         for unioning ANTIC_OUT U EXP_GEN - TMP_GEN but merge
3991         all expressions.  Add verification that the value-sets
3992         only shrink during iteration.
3993         (compute_partial_antic_aux): Adjust for the phi_translate changes.
3994         (do_pre_regular_insertion): Likewise.
3995         (do_pre_partial_partial_insertion): Likewise.
3996
3997 2018-03-02  Richard Biener  <rguenther@suse.de>
3998
3999         PR target/82005
4000         * config/darwin.c (saved_debug_info_level): New static global.
4001         (darwin_asm_lto_start): Disable debug info generation for LTO out.
4002         (darwin_asm_lto_end): Restore debug info generation settings.
4003
4004 2018-03-01  Martin Liska  <mliska@suse.cz>
4005
4006         PR sanitizer/82484
4007         * sanopt.c (sanitize_rewrite_addressable_params): Do not handle
4008         volatile arguments.
4009
4010 2018-03-01  Richard Biener  <rguenther@suse.de>
4011
4012         PR debug/84645
4013         * dwarf2out.c (gen_variable_die): Properly handle late VLA
4014         type annotation with LTO when debug was disabled at compile-time.
4015
4016 2018-03-01  Matthew Fortune  <mfortune@gmail.com>
4017
4018         * config/mips/mips.c (mips_final_prescan_insn): Fix incorrect
4019         XINT with INTVAL.
4020         (mips_final_postscan_insn): Likewise.
4021
4022 2018-03-01  Richard Sandiford  <richard.sandiford@linaro.org>
4023
4024         PR rtl-optimization/84528
4025         * alias.c (init_alias_target): Add commentary.
4026         (init_alias_analysis): Only give HARD_FRAME_POINTER_REGNUM
4027         a unique base value if the frame pointer is not eliminated
4028         to the stack pointer.
4029
4030 2018-03-01  Tom de Vries  <tom@codesourcery.com>
4031
4032         PR rtl-optimization/83327
4033         * lra-int.h (hard_regs_spilled_into): Declare.
4034         * lra.c (hard_regs_spilled_into): Define.
4035         (init_reg_info): Init hard_regs_spilled_into.
4036         * lra-spills.c (assign_spill_hard_regs): Update hard_regs_spilled_into.
4037         * lra-lives.c (make_hard_regno_born, make_hard_regno_dead)
4038         (process_bb_lives): Handle hard_regs_spilled_into.
4039         (lra_create_live_ranges_1): Before doing liveness propagation, clear
4040         regs in all_hard_regs_bitmap if set in hard_regs_spilled_into.
4041
4042 2018-02-28  David Edelsohn  <dje.gcc@gmail.com>
4043
4044         * config.gcc (powerpc-ibm-aix7.1.*): New stanza.
4045         (powerpc-ibm-aix[789]*): Default to AIX 7.2.
4046         * config/rs6000/aix71.h (TARGET_DEFAULT): Revert to Power4 ISA.
4047         * config/rs6000/aix72.h: New file.
4048
4049 2018-02-28  Jakub Jelinek  <jakub@redhat.com>
4050
4051         * gimple-ssa-warn-restrict.c (maybe_diag_overlap): Use warning_n
4052         instead of warning_at with conditional singular and plural messages
4053         where possible.
4054
4055         PR target/52991
4056         * stor-layout.c (update_alignment_for_field): For
4057         targetm.ms_bitfield_layout_p (rli->t), if !is_bitfield
4058         && !DECL_PACKED (field), do the alignment update, just use
4059         only desired_align instead of MAX (type_align, desired_align)
4060         as the alignment.
4061         (place_field): Don't do known_align < desired_align handling
4062         early if targetm.ms_bitfield_layout_p (rli->t) and rli->prev_field
4063         is non-NULL, instead do it after rli->prev_field handling and
4064         only if not within a bitfield word.  For DECL_PACKED (field)
4065         use type_align of BITS_PER_UNIT.
4066
4067 2018-02-28  Eric Botcazou  <ebotcazou@adacore.com>
4068
4069         * config/aarch64/aarch64.c (aarch64_emit_probe_stack_range): Remove
4070         superfluous parentheses and trailing spaces.
4071
4072 2018-02-28  Richard Biener  <rguenther@suse.de>
4073
4074         PR tree-optimization/84584
4075         * graphite-scop-detection.c (scop_detection::add_scop): Discard
4076         SCoPs with fake exit edge.
4077
4078 2018-02-28  Martin Liska  <mliska@suse.cz>
4079
4080         PR testsuite/84597
4081         * timevar.c (timer::print): Fix format to properly print 100%
4082         values.
4083
4084 2018-02-28  Richard Biener  <rguenther@suse.de>
4085
4086         PR middle-end/84607
4087         * genmatch.c (capture_info::walk_match): Do not mark
4088         captured expressions without operands as expr_p given
4089         they act more like predicates and should be subject to
4090         "lost tail" side-effect preserving.
4091
4092 2018-02-28  Alexandre Oliva  <aoliva@redhat.com>
4093
4094         PR rtl-optimization/81611
4095         * auto-inc-dec.c (attempt_change): Move dead note from
4096         mem_insn if it's the next use of regno
4097         (find_address): Take address use of reg holding
4098         non-incremented value.  Add parm to limit search to the named
4099         reg only.
4100         (merge_in_block): Attempt to use a mem insn that is the next
4101         use of the original regno.
4102
4103 2018-02-27  Martin Sebor  <msebor@redhat.com>
4104
4105         PR c++/83871
4106         * gcc/doc/invoke.texi (-Wmissing-attributes): New option.
4107         * gcc/print-tree.c (print_node): Handle DECL_UNINLINABLE.
4108
4109 2018-02-27  Martin Sebor  <msebor@redhat.com>
4110
4111         PR translation/84207
4112         * diagnostic-core.h (warning_n, error_n, inform_n): Change
4113         n argument to unsigned HOST_WIDE_INT.
4114         * diagnostic.c (warning_n, error_n, inform_n): Ditto.
4115         (diagnostic_n_impl): Ditto.  Handle arguments in excess of LONG_MAX.
4116         * gimple-ssa-sprintf.c (format_directive): Simplify inform_n call.
4117         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Use warning_n.
4118
4119 2018-02-27  Richard Biener  <rguenther@suse.de>
4120
4121         PR tree-optimization/84512
4122         * tree-vect-loop.c (vect_compute_single_scalar_iteration_cost):
4123         Do not use the estimate returned from record_stmt_cost for
4124         the scalar iteration cost but sum properly using add_stmt_cost.
4125
4126 2018-02-27  Richard Biener  <rguenther@suse.de>
4127
4128         PR tree-optimization/84466
4129         * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
4130         Adjust last change to less strictly validate use operands.
4131
4132 2018-02-27  Martin Liska  <mliska@suse.cz>
4133
4134         PR gcov-profile/84548
4135         * gcov.c (process_file): Allow partial overlap and consider it
4136         also as group functions.
4137         (output_lines): Properly calculate range of lines for a group.
4138
4139 2018-02-27  Martin Liska  <mliska@suse.cz>
4140
4141         * timevar.c (timer::print_row): Remove 'usr', 'sys', 'wall' and
4142         'ggc' suffixes.  Change first column width.
4143         (timer::print): Fix formatting of the column.
4144
4145 2018-02-27  Alexandre Oliva  <aoliva@redhat.com>
4146
4147         * tree-ssa-live.c (remove_unused_scope_block_p): Do not
4148         preserve inline entry blocks for the sake of debug inline
4149         entry point markers alone.
4150         (remove_unused_locals): Suggest in comments a better place to
4151         force the preservation of inline entry blocks that are
4152         otherwise unused, but do not preserve them.
4153
4154 2018-02-26  H.J. Lu  <hongjiu.lu@intel.com>
4155
4156         * config/i386/i386.c (ix86_output_indirect_jmp): Update comments.
4157
4158 2018-02-26  H.J. Lu  <hongjiu.lu@intel.com>
4159
4160         PR target/84039
4161         * config/i386/constraints.md (Bs): Replace
4162         ix86_indirect_branch_register with
4163         TARGET_INDIRECT_BRANCH_REGISTER.
4164         (Bw): Likewise.
4165         * config/i386/i386.md (indirect_jump): Likewise.
4166         (tablejump): Likewise.
4167         (*sibcall_memory): Likewise.
4168         (*sibcall_value_memory): Likewise.
4169         Peepholes of indirect call and jump via memory: Likewise.
4170         (*sibcall_GOT_32): Disallowed for TARGET_INDIRECT_BRANCH_REGISTER.
4171         (*sibcall_value_GOT_32): Likewise.
4172         * config/i386/predicates.md (indirect_branch_operand): Likewise.
4173         (GOT_memory_operand): Likewise.
4174         (call_insn_operand): Likewise.
4175         (sibcall_insn_operand): Likewise.
4176         (GOT32_symbol_operand): Likewise.
4177         * config/i386/i386.h (TARGET_INDIRECT_BRANCH_REGISTER): New.
4178
4179 2018-02-26  Eric Botcazou  <ebotcazou@adacore.com>
4180
4181         PR rtl-optimization/83496
4182         * reorg.c (steal_delay_list_from_target): Change REDUNDANT array from
4183         booleans to RTXes.  Call fix_reg_dead_note on every non-null element.
4184         (steal_delay_list_from_fallthrough): Call fix_reg_dead_note on a
4185         redundant insn, if any.
4186         (relax_delay_slots): Likewise.
4187         (update_reg_unused_notes): Rename REDUNDANT_INSN to OTHER_INSN.
4188
4189 2018-02-26  Richard Sandiford  <richard.sandiford@linaro.org>
4190
4191         PR tree-optimization/83965
4192         * tree-vect-patterns.c (vect_reassociating_reduction_p): Assume
4193         that grouped statements are part of a reduction chain.  Return
4194         true if the statement is not marked as a reduction itself but
4195         is part of a group.
4196         (vect_recog_dot_prod_pattern): Don't check whether the statement
4197         is part of a group here.
4198         (vect_recog_sad_pattern): Likewise.
4199         (vect_recog_widen_sum_pattern): Likewise.
4200
4201 2018-02-26  Eric Botcazou  <ebotcazou@adacore.com>
4202
4203         PR debug/84545
4204         * final.c (rest_of_clean_state): Also look for calls inside sequences.
4205
4206 2018-02-26  H.J. Lu  <hongjiu.lu@intel.com>
4207
4208         PR target/84530
4209         * config/i386/i386-protos.h (ix86_output_indirect_jmp): Remove
4210         the bool argument.
4211         (ix86_output_indirect_function_return): New prototype.
4212         (ix86_split_simple_return_pop_internal): Likewise.
4213         * config/i386/i386.c (indirect_return_via_cx): New.
4214         (indirect_return_via_cx_bnd): Likewise.
4215         (indirect_thunk_name): Handle return va CX_REG.
4216         (output_indirect_thunk_function): Create alias for
4217         __x86_return_thunk_[re]cx and __x86_return_thunk_[re]cx_bnd.
4218         (ix86_output_indirect_jmp): Remove the bool argument.
4219         (ix86_output_indirect_function_return): New function.
4220         (ix86_split_simple_return_pop_internal): Likewise.
4221         * config/i386/i386.md (*indirect_jump): Don't pass false
4222         to ix86_output_indirect_jmp.
4223         (*tablejump_1): Likewise.
4224         (simple_return_pop_internal): Change it to define_insn_and_split.
4225         Call ix86_split_simple_return_pop_internal to split it for
4226         -mfunction-return=.
4227         (simple_return_indirect_internal): Call
4228         ix86_output_indirect_function_return instead of
4229         ix86_output_indirect_jmp.
4230
4231 2018-02-26  Jakub Jelinek  <jakub@redhat.com>
4232
4233         PR bootstrap/84405
4234         * vec.h (vec_default_construct): For BROKEN_VALUE_INITIALIZATION use
4235         memset and value initialization afterwards.
4236
4237 2018-02-26  Christophe Lyon  <christophe.lyon@linaro.org>
4238
4239         * Makefile.in (lto-wrapper): Use ALL_LINKERFLAGS.
4240
4241 2018-02-26  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
4242
4243         PR target/84521
4244         * common/config/aarch64/aarch64-common.c
4245         (aarch_option_optimization_table[]): Switch
4246         off fomit-frame-pointer
4247
4248 2018-02-26  Kito Cheng  <kito.cheng@gmail.com>
4249             Chung-Ju Wu  <jasonwucj@gmail.com>
4250
4251         * config/nds32/nds32-multiple.md (load_multiple): Disallow
4252         volatile memory.
4253         (store_multiple): Ditto.
4254
4255 2018-02-26  Kito Cheng  <kito.cheng@gmail.com>
4256
4257         * config.gcc: Add --with-cpu support for nds32 target.
4258         * config/nds32/nds32-opts.h (nds32_cpu_type): New.
4259         * config/nds32/nds32.opt: Add -mcpu= option.
4260
4261 2018-02-25  Segher Boessenkool  <segher@kernel.crashing.org>
4262
4263         * config/rs6000/rs6000.opt (mvrsave=no, mvrsave=yes, isel=no,
4264         isel=yes): Warn for these deprecated options.
4265
4266 2018-02-23  David Edelsohn  <dje.gcc@gmail.com>
4267
4268         * config/rs6000/aix71.h (TARGET_DEFAULT): Change to
4269         ISA_2_5_MASKS_EMBEDDED.
4270
4271 2018-02-23  Jakub Jelinek  <jakub@redhat.com>
4272
4273         * ipa-prop.c (ipa_vr_ggc_hash_traits::hash): Hash p->min and
4274         p->max as pointers rather than using iterative_hash_expr.
4275
4276 2018-02-23  Carl Love  <cel@us.ibm.com>
4277
4278         * config/rs6000/rs6000-builtin.def: Change VSIGNED2 and VUNSIGNED2
4279         macro expansions from BU_VSX_2 to BU_P8V_VSX_2 and BU_VSX_OVERLOAD_2 to
4280         BU_P8V_OVERLOAD_2.
4281         * config/rs6000/rs6000-c.c: Change VSX_BUILTIN_VEC_VSIGNED2 to
4282         P8V_BUILTIN_VEC_VSIGNED2.  Change VSX_BUILTIN_VEC_VUNSIGNED2 to
4283         P8V_BUILTIN_VEC_VUNSIGNED2.
4284
4285 2018-02-22  Vladimir Makarov  <vmakarov@redhat.com>
4286
4287         PR target/81572
4288         * lra-int.h (LRA_UNKNOWN_ALT, LRA_NON_CLOBBERED_ALT): New macros.
4289         * lra.c (lra_set_insn_recog_data, lra_update_insn_recog_data): Use
4290         LRA_UNKNOWN_ALT.
4291         * lra-constraints.c (curr_insn_transform): Set up
4292         LRA_NON_CLOBBERED_ALT for moves processed on the fast path.  Use
4293         LRA_UNKNOWN_ALT.
4294         (remove_inheritance_pseudos): Use LRA_UNKNOWN_ALT.
4295         * lra-eliminations.c (spill_pseudos): Ditto.
4296         (process_insn_for_elimination): Ditto.
4297         * lra-lives.c (reg_early_clobber_p): Use the new macros.
4298         * lra-spills.c (spill_pseudos): Use LRA_UNKNOWN_ALT and
4299         LRA_NON_CLOBBERED_ALT.
4300
4301 2018-02-22  Martin Sebor  <msebor@redhat.com>
4302
4303         PR tree-optimization/84480
4304         * gimple-fold.c (gimple_fold_builtin_strcpy): Move warnings
4305         to maybe_diag_stxncpy_trunc.  Call it.
4306         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Integrate warnings
4307         from gimple_fold_builtin_strcpy.  Print inlining stack.
4308         (handle_builtin_stxncpy): Print inlining stack.
4309         * tree-ssa-strlen.h (maybe_diag_stxncpy_trunc): Declare.
4310
4311 2018-02-22  H.J. Lu  <hongjiu.lu@intel.com>
4312
4313         PR target/84176
4314         * config/i386/i386.c (ix86_set_indirect_branch_type): Issue an
4315         error when -mindirect-branch=thunk-extern, -fcf-protection=branch
4316         and -fcheck-pointer-bounds are used together.
4317         (indirect_thunk_prefix): New enum.
4318         (indirect_thunk_need_prefix): New function.
4319         (indirect_thunk_name): Replace need_bnd_p with need_prefix.  Use
4320         "_nt" instead of "_bnd" for NOTRACK prefix.
4321         (output_indirect_thunk): Replace need_bnd_p with need_prefix.
4322         (output_indirect_thunk_function): Likewise.
4323         (): Likewise.
4324         (ix86_code_end): Update output_indirect_thunk_function calls.
4325         (ix86_output_indirect_branch_via_reg): Replace
4326         ix86_bnd_prefixed_insn_p with indirect_thunk_need_prefix.
4327         (ix86_output_indirect_branch_via_push): Likewise.
4328         (ix86_output_function_return): Likewise.
4329         * doc/invoke.texi: Document -mindirect-branch=thunk-extern is
4330         incompatible with -fcf-protection=branch and
4331         -fcheck-pointer-bounds.
4332
4333 2018-02-22  Steve Ellcey  <sellcey@cavium.com>
4334
4335         PR target/83335
4336         * config/aarch64/aarch64.c (aarch64_print_address_internal):
4337         Change gcc_assert call to output_operand_lossage.
4338
4339 2018-02-22  Steve Ellcey  <sellcey@cavium.com>
4340
4341         * doc/extend.texi (__builtin_extend_pointer): Document builtin.
4342
4343 2018-02-22  DJ Delorie  <dj@redhat.com>
4344             Sebastian Perta  <sebastian.perta@renesas.com>
4345             Oleg Endo  <olegendo@gcc.gnu.org>
4346
4347         * config/rx/rx.c (rx_rtx_costs): New function.
4348         (TARGET_RTX_COSTS): Override to use rx_rtx_costs.
4349
4350 2018-02-22  Thomas Preud'homme  <thomas.preudhomme@arm.com>
4351
4352         * config/arm/t-multilib: Map Armv8-R to Armv7 multilibs.
4353
4354 2018-02-22  Martin Liska  <mliska@suse.cz>
4355
4356         PR driver/83193
4357         * common/config/arm/arm-common.c (arm_print_hint_for_cpu_option):
4358         Add "native" as a possible value.
4359
4360 2018-02-22  Martin Liska  <mliska@suse.cz>
4361
4362         PR driver/83193
4363         * config/i386/i386.c (ix86_option_override_internal):
4364         Add "native" as a possible value for -march and -mtune.
4365
4366 2018-02-22  Jakub Jelinek  <jakub@redhat.com>
4367
4368         PR target/84502
4369         * stor-layout.c (finalize_type_size): Propagate TYPE_EMPTY_P flag
4370         to all type variants.
4371
4372         PR tree-optimization/84503
4373         * gimple-ssa-store-merging.c (merged_store_group::merge_into): Compute
4374         width as info->bitpos + info->bitsize - start.
4375         (merged_store_group::merge_overlapping): Simplify width computation.
4376         (check_no_overlap): New function.
4377         (imm_store_chain_info::try_coalesce_bswap): Compute expected
4378         start + width and last_order of the group, fail if check_no_overlap
4379         fails.
4380         (imm_store_chain_info::coalesce_immediate_stores): Don't merge info
4381         to group if check_no_overlap fails.
4382
4383 2018-02-21  Segher Boessenkool  <segher@kernel.crashing.org>
4384
4385         * config/rs6000/altivec.md: Delete contraint arguments to
4386         define_expand, define_split, and define_peephole2, and in
4387         define_insn_and_split if always unused.
4388         * config/rs6000/darwin.md: Ditto.
4389         * config/rs6000/dfp.md: Ditto.
4390         * config/rs6000/rs6000.md: Ditto.
4391         * config/rs6000/sync.md: Ditto.
4392         * config/rs6000/vector.md: Ditto.
4393         * config/rs6000/vsx.md: Ditto.
4394
4395 2018-02-21  Segher Boessenkool  <segher@kernel.crashing.org>
4396
4397         * config/rs6000/altivec.md: Write output control strings as braced
4398         blocks instead of double-quoted strings.
4399         * config/rs6000/darwin.md: Ditto.
4400         * config/rs6000/rs6000.md: Ditto.
4401         * config/rs6000/vector.md: Ditto.
4402         * config/rs6000/vsx.md: Ditto.
4403
4404 2018-02-21  Jason Merrill  <jason@redhat.com>
4405
4406         PR c++/84314 - ICE with templates and fastcall attribute.
4407         * attribs.c (build_type_attribute_qual_variant): Remove assert.
4408
4409 2018-02-21  Jan Hubicka  <hubicka@ucw.cz>
4410
4411         * ipa-cp.c (determine_versionability): Fix comment typos.
4412
4413 2018-02-21  Jan Hubicka  <hubicka@ucw.cz>
4414
4415         PR c/84229
4416         * ipa-cp.c (determine_versionability): Do not version functions caling
4417         va_arg_pack.
4418
4419 2018-02-21  Martin Liska  <mliska@suse.cz>
4420
4421         PR driver/83193
4422         * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch):
4423         Add "native" as a possible value.
4424         * config/aarch64/aarch64.h (HAVE_LOCAL_CPU_DETECT):  Define
4425         the macro when native cpu detection is available.
4426
4427 2018-02-21  Martin Liska  <mliska@suse.cz>
4428
4429         PR driver/83193
4430         * common/config/arm/arm-common.c (arm_print_hint_for_arch_option):
4431         Add "native" as a possible value.
4432         * config/arm/arm.h (HAVE_LOCAL_CPU_DETECT): Define the macro
4433         when native cpu detection is available.
4434
4435 2018-02-21  Jakub Jelinek  <jakub@redhat.com>
4436             Martin Sebor  <msebor@redhat.com>
4437
4438         PR tree-optimization/84478
4439         * gimple-fold.h (get_range_strlen): Add a bool argument defaulted to
4440         false.
4441         * gimple-fold.c (get_range_strlen): Make minlen const and assume it
4442         can't be NULL.  Change FUZZY from bool to int, for 1 add PHI/COND_EXPR
4443         support which is conservatively correct, for 2 only stay conservative
4444         for maxlen.  Formatting and comment capitalization fixes.  Add STRICT
4445         argument to the 2 argument get_range_strlen, adjust 6 arg
4446         get_range_strlen caller and clear minmaxlen[0] and [1] if it returned
4447         false.
4448         (get_maxval_strlen): Adjust 6 arg get_range_strlen caller.
4449         (gimple_fold_builtin_strlen): Pass true as last argument to
4450         get_range_strlen.
4451
4452 2018-02-20  Martin Sebor  <msebor@redhat.com>
4453
4454         PR middle-end/84095
4455         * gimple-ssa-warn-restrict.c (builtin_memref::extend_offset_range): New.
4456         (builtin_memref::set_base_and_offset): Same.  Handle inner references.
4457         (builtin_memref::builtin_memref): Factor out parts into
4458         set_base_and_offset and call it.
4459
4460 2018-02-20  Richard Sandiford  <richard.sandiford@linaro.org>
4461
4462         PR middle-end/84406
4463         * optabs-query.c (find_widening_optab_handler_and_mode): If from_mode
4464         is a scalar_int_mode, assert that to_mode is a scalar_int_mode with
4465         greater precision.  If to_mode is a MODE_PARTIAL_INT, stop the
4466         search at the associated MODE_INT.
4467
4468 2018-02-20  Jeff Law  <law@redhat.com>
4469
4470         PR middle-end/82123
4471         PR tree-optimization/81592
4472         PR middle-end/79257
4473         * gimple-ssa-sprintf.c (format_integer): Query EVRP range analyzer
4474         for range data rather than using global data.
4475         * gimple-ssa-sprintf.c (get_int_range): Query EVRP range analyzer for
4476         range data rather than using global data.
4477         * gimple-ssa-sprintf.c (get_int_range): Accept vr_values parameter
4478         pass it to children as needed.
4479         (struct directive::fmtresult): Similarly.
4480         (struct directive::set_width): Similarly.
4481         (struct directive::set_precision): Similarly.
4482         (format_integer, format_directive, parse_directive): Similarly.
4483         (format_none): Accept unnamed vr_values parameter.
4484         (format_percent, format_floating, format_character): Similarly.
4485         (format_string, format_plain): Similarly.
4486         * gimple-ssa-sprintf.c (sprintf_dom_walker::handle_gimple_call): Query
4487         the EVRP range analyzer for range data rather than using global data.
4488         * gimple-ssa-sprintf.c: Include alloc-pool.h, vr-values.h and
4489         gimple-ssa-evrp-analyze.h
4490         (class sprintf_dom_walker): Add after_dom_children member function.
4491         Add evrp_range_analyzer member.
4492         (sprintf_dom_walker::before_dom_children): Call into the EVRP
4493         range analyzer as needed.
4494         (sprintf_dom_walker::after_dom_children): New member function.
4495         * gimple-ssa-evrp-analyze.c (evrp_range_analyzer::enter): Do nothing
4496         if not optimizing.
4497         (evrp_range_analyzer::record_ranges_from_stmt): Likewise.
4498         (evrp_range_analyzer::pop_to_marker): Likewise.
4499
4500 2018-02-20  Richard Sandiford  <richard.sandiford@linaro.org>
4501
4502         PR tree-optimization/84419
4503         * internal-fn.c (expand_call_mem_ref): Create a TARGET_MEM_REF
4504         with the required type if its current type is compatible but
4505         different.
4506
4507 2018-02-20  Jakub Jelinek  <jakub@redhat.com>
4508
4509         PR middle-end/82004
4510         * match.pd (pow(C,x) -> exp(log(C)*x)): Delay all folding until
4511         after vectorization.
4512
4513 2018-02-20  Martin Liska  <mliska@suse.cz>
4514
4515         PR driver/83193
4516         * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch): Print
4517         possible values if we don't have a hint.
4518
4519 2018-02-20  Martin Liska  <mliska@suse.cz>
4520
4521         PR c/84310
4522         PR target/79747
4523         * final.c (shorten_branches): Build align_tab array with one
4524         more element.
4525         * opts.c (finish_options): Add alignment option limit check.
4526         (MAX_CODE_ALIGN): Likewise.
4527         (MAX_CODE_ALIGN_VALUE): Likewise.
4528         * doc/invoke.texi: Document maximum allowed option value for
4529         all -falign-* options.
4530
4531 2018-02-19  Jakub Jelinek  <jakub@redhat.com>
4532
4533         PR target/84146
4534         * reg-notes.def (REG_CALL_ARG_LOCATION): New reg note.
4535         * insn-notes.def (NOTE_INSN_CALL_ARG_LOCATION): Remove.
4536         * var-tracking.c (emit_note_insn_var_location): Remove all references
4537         to NOTE_INSN_CALL_ARG_LOCATION.
4538         (emit_notes_in_bb): Emit arguments as REG_CALL_ARG_LOCATION note on
4539         the CALL_INSN rather than separate NOTE_INSN_CALL_ARG_LOCATION note.
4540         Use copy_rtx_if_shared.
4541         * dwarf2out.c (gen_subprogram_die): Use XEXP with 0 instead of
4542         NOTE_VAR_LOCATION on ca_loc->call_arg_loc_note.
4543         (dwarf2out_var_location): Remove handling of
4544         NOTE_INSN_CALL_ARG_LOCATION, instead handle REG_CALL_ARG_LOCATION note
4545         on call_insn.
4546         * final.c (final_scan_insn): Remove all references to
4547         NOTE_INSN_CALL_ARG_LOCATION.
4548         (rest_of_clean_state): Likewise.  Remove REG_CALL_ARG_LOCATION notes
4549         before dumping final insns.
4550         * except.c (emit_note_eh_region_end): Remove all references to
4551         NOTE_INSN_CALL_ARG_LOCATION.
4552         * config/alpha/alpha.c (alpha_pad_function_end): Likewise.
4553         * config/c6x/c6x.c (c6x_gen_bundles): Likewise.
4554         * config/arc/arc.c (hwloop_optimize): Likewise.
4555         * config/arm/arm.c (create_fix_barrier): Likewise.
4556         * config/s390/s390.c (s390_chunkify_start): Likewise.
4557         * config/sh/sh.c (find_barrier): Likewise.
4558         * config/i386/i386.c (rest_of_insert_endbranch,
4559         ix86_seh_fixup_eh_fallthru): Likewise.
4560         * config/xtensa/xtensa.c (hwloop_optimize): Likewise.
4561         * config/iq2000/iq2000.c (final_prescan_insn): Likewise.
4562         * config/frv/frv.c (frv_function_prologue): Likewise.
4563         * emit-rtl.c (try_split): Likewise.  Copy over REG_CALL_ARG_LOCATION
4564         reg note.
4565         (note_outside_basic_block_p): Remove all references to
4566         NOTE_INSN_CALL_ARG_LOCATION.
4567         * gengtype.c (adjust_field_rtx_def): Likewise.
4568         * print-rtl.c (rtx_writer::print_rtx_operand_code_0, print_insn):
4569         Likewise.
4570         * jump.c (cleanup_barriers, delete_related_insns): Likewise.
4571         * cfgrtl.c (force_nonfallthru_and_redirect): Likewise.
4572
4573         PR c++/84444
4574         * builtins.c (builtin_mathfn_code): Don't check if CALL_EXPR_FN (t)
4575         is ADDR_EXPR.
4576
4577         PR tree-optimization/84452
4578         * tree-vect-patterns.c (vect_recog_pow_pattern): Don't call
4579         expand_simd_clones if targetm.simd_clone.compute_vecsize_and_simdlen
4580         is NULL.
4581
4582 2018-02-19  Martin Liska  <mliska@suse.cz>
4583
4584         PR sanitizer/82183
4585         * passes.def: Put pass_sancov_O0 before pass_lower_switch with -O0.
4586
4587 2018-02-19  Martin Liska  <mliska@suse.cz>
4588             Richard Sandiford  <richard.sandiford@linaro.org>
4589
4590         PR tree-optimization/82491
4591         * gimple-fold.c (get_base_constructor): Make earlier bail out
4592         to prevent ubsan.
4593
4594 2018-02-19  Carl Love  <cel@us.ibm.com>
4595
4596         * config/rs6000/rs6000-builtin.def: Change NEG macro expansions from
4597         BU_ALTIVEC_A to BU_P8V_AV_1 and BU_ALTIVEC_OVERLOAD_1 to
4598         BU_P8V_OVERLOAD_1.
4599         * config/rs6000/rs6000-c.c: Change ALTIVEC_BUILTIN_VEC_NEG to
4600         P8V_BUILTIN_VEC_NEG.
4601
4602 2018-02-19  Sebastian Perta  <sebastian.perta@renesas.com>
4603
4604         * config/rl78/rl78.md (movdf): New define expand.
4605
4606 2018-02-19  Martin Liska  <mliska@suse.cz>
4607
4608         PR other/80589
4609         * doc/invoke.texi: Fix typo.
4610         * params.def (PARAM_MAX_LOOP_HEADER_INSNS): Likewise.
4611
4612 2018-02-18  Segher Boessenkool  <segher@kernel.crashing.org>
4613
4614         * config/rs6000/rs6000.c (rs6000_option_override_internal): Don't
4615         handle rs6000_single_float and rs6000_double_float specially for
4616         e500 family CPUs.
4617
4618 2018-02-16  Jeff Law  <law@redhat.com>
4619
4620         * config/rx/rx.c (add_pop_cfi_notes): New function.;
4621         (pop_regs): Use it.
4622
4623 2018-02-16  Jakub Jelinek  <jakub@redhat.com>
4624
4625         PR ipa/84425
4626         * ipa-inline.c (inline_small_functions): Fix a typo.
4627
4628 2018-02-16  Nathan Sidwell  <nathan@acm.org>
4629
4630         * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
4631
4632 2018-02-16  Carl Love  <cel@us.ibm.com>
4633
4634         * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
4635         Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
4636         from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
4637         * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
4638         expansion to P8V_BUILTIN_VEC_FLOAT2.
4639
4640 2018-02-16  Vladimir Makarov  <vmakarov@redhat.com>
4641
4642         PR rtl-optimization/70023
4643         * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
4644         src_regno into account.
4645
4646 2018-02-16  Carl Love  <cel@us.ibm.com>
4647
4648         * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
4649         * config/rs6000/rs6000-builtin.def: Remove macro expansion for
4650         VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
4651         * config/rs6000/rs6000.c: Remove case statements for
4652         P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
4653         P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
4654         and P9V_BUILTIN_VEC_VINSERT4B.
4655         * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
4656         P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
4657         * config/rs6000/vsx.md:
4658         * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
4659         vec_insert4b.
4660
4661 2018-02-16  Carl Love  <cel@us.ibm.com>
4662
4663         * config/rs6000/altivec.h: Add builtin names vec_extract4b
4664         vec_insert4b.
4665         * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
4666         definitions.
4667         * config/rs6000/rs6000-c.c: Add the definitions for
4668         P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
4669         * config/rs6000/rs6000.c (altivec_expand_builtin): Add
4670         P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
4671         * config/rs6000/vsx.md: Add define_insn extract4b.  Add define_expand
4672         definition for insert4b and define insn *insert3b_internal.
4673         * doc/extend.texi: Add documentation for vec_extract4b.
4674
4675 2018-02-16  Nathan Sidwell  <nathan@acm.org>
4676
4677         * doc/extend.texi (Backwards Compatibility): Mention friend
4678         injection.  Note for-scope is deprecated.
4679         * doc/invoke.texi (-ffriend-injection): Deprecate.
4680
4681 2018-02-16  Segher Boessenkool  <segher@kernel.crashing.org>
4682
4683         * combine.c (try_combine): When adjusting LOG_LINKS for the destination
4684         that moved to I2, also allow destinations that are a paradoxical
4685         subreg (instead of a normal reg).
4686
4687 2018-02-16  Oleg Endo  <olegendo@gcc.gnu.org>
4688
4689         PR target/83831
4690         * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
4691         to QImode.
4692
4693 2018-02-16  Richard Biener  <rguenther@suse.de>
4694
4695         PR tree-optimization/84037
4696         PR tree-optimization/84016
4697         PR target/82862
4698         * config/i386/i386.c (ix86_builtin_vectorization_cost):
4699         Adjust vec_construct for the fact we need additional higher latency
4700         128bit inserts for AVX256 and AVX512 vector builds.
4701         (ix86_add_stmt_cost): Scale vector construction cost for
4702         elementwise loads.
4703
4704 2018-02-16  Richard Biener  <rguenther@suse.de>
4705
4706         PR tree-optimization/84417
4707         * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
4708         the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
4709         (non_rewritable_lvalue_p): Likewise, use poly-ints.
4710
4711 2018-02-16  Martin Liska  <mliska@suse.cz>
4712
4713         PR sanitizer/84307
4714         * internal-fn.def (ASAN_CHECK): Set proper flags.
4715         (ASAN_MARK): Likewise.
4716
4717 2018-02-16  Julia Koval  <julia.koval@intel.com>
4718
4719         * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
4720         from PTA_CANNONLAKE.
4721
4722 2018-02-16  Jakub Jelinek  <jakub@redhat.com>
4723
4724         PR target/84272
4725         * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
4726         Use ++iter rather than iter++ for std::list iterators.
4727         (func_fma_steering::dfs): Likewise.  Don't delete nodes right away,
4728         defer deleting them until all nodes in the forest are processed.  Do
4729         free even leaf nodes.  Change to_process into auto_vec.
4730
4731         PR bootstrap/84405
4732         * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
4733         * vec.h (vec_default_construct): Use memset instead of placement new
4734         if BROKEN_VALUE_INITIALIZATION is defined.
4735         * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
4736         memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
4737         is defined.
4738
4739         PR rtl-optimization/83723
4740         * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
4741         * lra.c (lra_substitute_pseudo): Likewise.  If true, use
4742         gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG.  Pass DEBUG_P to
4743         recursive calls.
4744         (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
4745         callers.
4746         * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
4747
4748 2018-02-16  Eric Botcazou  <ebotcazou@adacore.com>
4749
4750         PR rtl-optimization/81443
4751         * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
4752         from inner REGs to paradoxical SUBREGs.
4753
4754 2018-02-16  Richard Biener  <rguenther@suse.de>
4755
4756         PR tree-optimization/84399
4757         * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
4758         For operands we can analyze at their definition make sure we can
4759         analyze them at each use as well.
4760
4761 2018-02-16  Richard Biener  <rguenther@suse.de>
4762
4763         PR tree-optimization/84190
4764         * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
4765         volatile accesses if the decl isn't volatile.
4766
4767 2018-02-15  Jason Merrill  <jason@redhat.com>
4768
4769         PR c++/84314 - ICE with templates and fastcall attribute.
4770         * attribs.c (build_type_attribute_qual_variant): Don't clobber
4771         TYPE_CANONICAL on an existing type.
4772
4773 2018-02-15  Jakub Jelinek  <jakub@redhat.com>
4774
4775         PR tree-optimization/84383
4776         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
4777         dstoff nor call operand_equal_p if dstbase is NULL.
4778
4779         PR tree-optimization/84334
4780         * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
4781         also a CONSTANT_CLASS_P, punt.
4782
4783 2018-02-14  Jim Wilson  <jimw@sifive.com>
4784
4785         * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
4786         first SMALL_OPERAND check.  New local min_second_step.  Move assert
4787         to where locals are set.  Add TARGET_RVC support.
4788         * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
4789
4790 2018-02-14  Indu Bhagat  <indu.bhagat@oracle.com>
4791
4792         * doc/invoke.texi: Correct -Wformat-overflow code sample.
4793
4794 2018-02-14  Martin Sebor  <msebor@redhat.com>
4795
4796         PR tree-optimization/83698
4797         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
4798         arrays constrain the offset range to their bounds.
4799         (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
4800         (builtin_access::overlap): Avoid setting the size of overlap if it's
4801         already been set.
4802         (maybe_diag_overlap): Also consider arrays when deciding what values
4803         of offsets to include in diagnostics.
4804
4805 2018-02-14  Martin Sebor  <msebor@redhat.com>
4806
4807         PR c/84108
4808         * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
4809         that correspond to the kind of a declaration.
4810
4811 2018-02-14  John David Anglin  <danglin@gcc.gnu.org>
4812
4813         PR target/83984
4814         * config/pa/pa.md: Load address of PIC label using the linkage table
4815         if the label is nonlocal.
4816
4817 2018-02-14  Kelvin Nilsen  <kelvin@gcc.gnu.org>
4818
4819         * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
4820         warning message if user requests -maltivec=be.
4821         * doc/invoke.texi: Document deprecation of -maltivec=be.
4822
4823 2018-02-14  Will Schmidt  <will_schmidt@vnet.ibm.com>
4824
4825         PR target/84220
4826         * config/rs6000/rs6000-c.c: Update definitions for
4827         ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
4828         VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
4829
4830 2018-02-14  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
4831
4832         PR target/84239
4833         * config/i386/cetintrin.h: Remove _rdssp[d|q] and
4834         add _get_ssp intrinsics. Remove argument from
4835         __builtin_ia32_rdssp[d|q].
4836         * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
4837         * config/i386/i386-builtin.def: Remove argument from
4838         __builtin_ia32_rdssp[d|q].
4839         * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
4840         ix86_expand_special_args_builtin for _rdssp[d|q].
4841         * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
4842         Clear register before usage.
4843         * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
4844         Add documentation for new _get_ssp and _inc_ssp intrinsics.
4845
4846 2018-02-14  Richard Sandiford  <richard.sandiford@linaro.org>
4847
4848         PR tree-optimization/84357
4849         * tree-data-ref.c (object_address_invariant_in_loop_p): Check
4850         operand 1 of an ARRAY_REF too.
4851
4852 2018-02-14  Oleg Endo  <olegendo@gcc.gnu.org>
4853
4854         PR target/83831
4855         * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
4856         rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
4857         declarations.
4858         (set_of_reg): New struct.
4859         (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
4860         * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
4861         rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
4862         functions.
4863         * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
4864         Split into bitclr, bitset, bitinvert patterns if appropriate.
4865         (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
4866         use rx_fuse_in_memory_bitop.
4867         (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
4868         to named insn, correct maximum insn length.
4869
4870 2018-02-14  Jozef Lawrynowicz  <jozefl.gcc@gmail.com>
4871
4872         PR target/79242
4873         * machmode.def: Define a complex mode for PARTIAL_INT.
4874         * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
4875         MODE_PARTIAL_INT.
4876         * doc/rtl.texi: Document CSPImode.
4877         * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
4878         handling.
4879         (msp430_hard_regno_nregs_with_padding): Likewise.
4880
4881 2018-02-13  Peter Bergner  <bergner@vnet.ibm.com>
4882
4883         PR target/84279
4884         * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
4885
4886 2018-02-13  Segher Boessenkool  <segher@kernel.crashing.org>
4887
4888         PR rtl-optimization/84169
4889         * combine.c (try_combine): New variable split_i2i3.  Set it to true if
4890         we generated a parallel as new i3 and we split that to new i2 and i3
4891         instructions.  Handle split_i2i3 similar to swap_i2i3: scan the
4892         LOG_LINKs of i3 to see which of those need to link to i2 now.  Link
4893         those to i2, not i1.  Partially rewrite this scan code.
4894
4895 2018-02-13  Jakub Jelinek  <jakub@redhat.com>
4896
4897         PR c/82210
4898         * stor-layout.c (place_field): For variable length fields, adjust
4899         offset_align afterwards not just based on the field's alignment,
4900         but also on the size.
4901
4902         PR middle-end/84309
4903         * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
4904         of exps and logs in the use_exp2 case.
4905
4906 2018-02-13  Jeff Law  <law@redhat.com>
4907
4908         * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
4909         entry for "vector".
4910
4911         * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
4912         ARGS as unused.
4913
4914 2018-02-13  Alexandre Oliva  <aoliva@redhat.com>
4915
4916         PR debug/84342
4917         PR debug/84319
4918         * common.opt (gas-loc-support, gas-locview-support): New.
4919         (ginline-points, ginternal-reset-location-views): New.
4920         * doc/invoke.texi: Document them.  Use @itemx where intended.
4921         (gvariable-location-views): Adjust.
4922         * target.def (reset_location_view): New.
4923         * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
4924         (TARGET_RESET_LOCATION_VIEW): New.
4925         * doc/tm.texi: Rebuilt.
4926         * dwarf2out.c (dwarf2out_default_as_loc_support): New.
4927         (dwarf2out_default_as_locview_support): New.
4928         (output_asm_line_debug_info): Use option variables.
4929         (dwarf2out_maybe_output_loclist_view_pair): Likewise.
4930         (output_loc_list): Likewise.
4931         (add_high_low_attributes): Check option variables.
4932         Don't output entry view attribute in strict mode.
4933         (gen_inlined_subroutine_die): Check option variables.
4934         (dwarf2out_inline_entry): Likewise.
4935         (init_sections_and_labels): Likewise.
4936         (dwarf2out_early_finish): Likewise.
4937         (maybe_reset_location_view): New, from...
4938         (dwarf2out_var_location): ... here.  Call it.
4939         * debug.h (dwarf2out_default_as_loc_support): Declare.
4940         (dwarf2out_default_as_locview_support): Declare.
4941         * hooks.c (hook_int_rtx_insn_0): New.
4942         * hooks.h (hook_int_rtx_insn_0): Declare.
4943         * toplev.c (process_options): Take -gas-loc-support and
4944         -gas-locview-support from dwarf2out.  Enable
4945         -gvariable-location-views by default only with locview
4946         assembler support.  Enable -ginternal-reset-location-views by
4947         default only if the target defines the corresponding hook.
4948         Enable -ginline-points by default if location views are
4949         enabled; force it disabled if statement frontiers are
4950         disabled.
4951         * tree-inline.c (expand_call_inline): Check option variables.
4952         * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
4953
4954 2018-02-13  Richard Sandiford  <richard.sandiford@linaro.org>
4955
4956         PR tree-optimization/84321
4957         * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
4958         handling.  Also check whether the anti-range contains any values
4959         that satisfy the mask; switch to a VR_RANGE if not.
4960
4961 2018-02-13  Paolo Bonzini  <bonzini@gnu.org>
4962
4963         PR sanitizer/84340
4964         * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
4965
4966 2018-02-13  Martin Jambor  <mjambor@suse.cz>
4967
4968         PR c++/83990
4969         * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
4970         of call statements, also set location of a load to a temporary.
4971
4972 2018-02-13  Sebastian Perta  <sebastian.perta@renesas.com>
4973
4974         * config/rl78/rl78.c (add_vector_labels): New function.
4975         * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
4976         * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
4977         * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
4978         which checks that no arguments are passed.
4979         * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
4980         * doc/extend.texi: Documentation for the new attribute.
4981
4982 2018-02-13  Andreas Schwab  <schwab@suse.de>
4983
4984         * config/riscv/linux.h (CPP_SPEC): Define.
4985
4986 2018-02-13  Jakub Jelinek  <jakub@redhat.com>
4987
4988         PR target/84335
4989         * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
4990         OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
4991         OPTION_MASK_ISA_AES as first argument to def_builtin_const
4992         for AES builtins.  Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
4993         instead of OPTION_MASK_ISA_PCLMUL as first argument to
4994         def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
4995         * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
4996         temporarily for AES and PCLMUL builtins.
4997
4998         PR tree-optimization/84339
4999         * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
5000         ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
5001         Formatting fixes.
5002
5003         PR middle-end/84309
5004         * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
5005         exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
5006         * generic-match-head.c (canonicalize_math_after_vectorization_p): New
5007         inline function.
5008         * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
5009         inline function.
5010         * omp-simd-clone.h: New file.
5011         * omp-simd-clone.c: Include omp-simd-clone.h.
5012         (expand_simd_clones): No longer static.
5013         * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
5014         cgraph.h and omp-simd-clone.h.
5015         (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
5016         (vect_recog_widen_shift_pattern): Formatting fix.
5017         (vect_pattern_recog_1): Don't check optab for calls.
5018
5019         PR target/84336
5020         * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
5021         operands[2] into a REG before using gen_lowpart on it.
5022
5023 2018-02-12  Jeff Law  <law@redhat.com>
5024
5025         PR target/83760
5026         * config/sh/sh.c (find_barrier): Consider a sibling call
5027         a barrier as well.
5028
5029         * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
5030         successfully back substituting a reg.
5031
5032 2018-02-12  Richard Biener  <rguenther@suse.de>
5033
5034         PR tree-optimization/84037
5035         * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
5036         parameter, move visited init to caller.
5037         (vect_slp_analyze_operations): Separate cost from validity
5038         check, initialize visited once for all instances.
5039         (vect_schedule_slp): Analyze map to CSE vectorized nodes once
5040         for all instances.
5041         * tree-vect-stmts.c (vect_model_simple_cost): Make early
5042         out an assert.
5043         (vect_model_promotion_demotion_cost): Likewise.
5044         (vectorizable_bswap): Guard cost modeling with !slp_node
5045         instead of !PURE_SLP_STMT to avoid double-counting on hybrid
5046         SLP stmts.
5047         (vectorizable_call): Likewise.
5048         (vectorizable_conversion): Likewise.
5049         (vectorizable_assignment): Likewise.
5050         (vectorizable_shift): Likewise.
5051         (vectorizable_operation): Likewise.
5052         (vectorizable_store): Likewise.
5053         (vectorizable_load): Likewise.
5054         (vectorizable_condition): Likewise.
5055         (vectorizable_comparison): Likewise.
5056
5057 2018-02-12  Paolo Bonzini  <bonzini@gnu.org>
5058
5059         PR sanitizer/84307
5060         * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
5061         (ASAN_MARK): Fix fnspec to account for return value, change pointer
5062         argument from 'R' to 'W' so that the pointed-to datum is clobbered.
5063
5064 2018-02-08  Jan Hubicka  <hubicka@ucw.cz>
5065
5066         PR middle-end/83665
5067         * params.def (inline-min-speedup): Increase from 8 to 15.
5068         (max-inline-insns-auto): Decrease from 40 to 30.
5069         * ipa-split.c (consider_split): Add some buffer for function to
5070         be considered inlining candidate.
5071         * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
5072         default values.
5073
5074 2018-02-12  Richard Biener  <rguenther@suse.de>
5075
5076         PR tree-optimization/84037
5077         * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
5078         matched stmts if we cannot swap the non-matched ones.
5079
5080 2018-02-12  Olga Makhotina  <olga.makhotina@intel.com>
5081
5082         * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
5083         _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
5084         _mm_maskz_scalef_round_ss): New intrinsics.
5085         (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
5086         * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
5087         __builtin_ia32_scalefss_round): Remove.
5088         (__builtin_ia32_scalefsd_mask_round,
5089         __builtin_ia32_scalefss_mask_round): New intrinsics.
5090         * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
5091         (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
5092         ((match_operand:VF_128 2 "<round_nimm_predicate>"
5093         "<round_constraint>")): Changed to ...
5094         ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
5095         "<round_scalar_constraint>")): ... this.
5096         ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
5097         %0, %1, %2<round_op3>}"): Changed to ...
5098         ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
5099         %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
5100         %2<round_scalar_mask_op3>}"): ... this.
5101         * config/i386/subst.md (round_scalar_nimm_predicate): New.
5102
5103 2018-02-12  Olga Makhotina  <olga.makhotina@intel.com>
5104
5105         * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
5106         (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
5107         (_mm_maskz_sqrt_round_ss): New intrinsics.
5108         (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
5109         (__builtin_ia32_sqrtsd_mask_round)
5110         (__builtin_ia32_sqrtss_mask_round): New builtins.
5111         * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
5112         (__builtin_ia32_sqrtss_round): Remove.
5113         (__builtin_ia32_sqrtsd_mask_round)
5114         (__builtin_ia32_sqrtss_mask_round): New builtins.
5115         * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
5116         (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
5117         ((match_operand:VF_128 1 "vector_operand"
5118         "xBm,<round_constraint>")): Changed to ...
5119         ((match_operand:VF_128 1 "vector_operand"
5120         "xBm,<round_scalar_constraint>")): ... this.
5121         (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
5122         %0, %2, %<iptr>1<round_op3>}): Changed to ...
5123         (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
5124         %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
5125         %<iptr>1<round_scalar_mask_op3>}): ... this.
5126         ((set_attr "prefix" "<round_prefix>")): Changed to ...
5127         ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
5128
5129 2018-02-11  Steven Munroe  <munroesj@gcc.gnu.org>
5130
5131         PR target/84266
5132         * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
5133         Cast vec_cmpeq result to correct type.
5134         * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
5135         Cast vec_cmpgt result to correct type.
5136
5137 2018-02-11  Alexandre Oliva  <aoliva@redhat.com>
5138
5139         * final.c (final_scan_insn_1): Renamed from...
5140         (final_scan_insn): ... this.  New wrapper, to recover
5141         seen from the outermost call in recursive ones.
5142         * config/sparc/sparc.c (output_return): Drop seen from call.
5143         (output_sibcall): Likewise.
5144         * config/visium/visium.c (output_branch): Likewise.
5145
5146 2018-02-10  John David Anglin  <danglin@gcc.gnu.org>
5147
5148         * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
5149         function label.
5150
5151 2018-02-10  Alan Modra  <amodra@gmail.com>
5152
5153         PR target/84300
5154         * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
5155         Specify LR as an input.
5156
5157 2018-02-10  Jakub Jelinek  <jakub@redhat.com>
5158
5159         PR sanitizer/83987
5160         * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
5161         remove_member_access_dummy_vars): New functions.
5162         (lower_omp_for, lower_omp_taskreg, lower_omp_target,
5163         lower_omp_1, execute_lower_omp): Use them.
5164
5165         PR rtl-optimization/84308
5166         * shrink-wrap.c (spread_components): Release todo vector.
5167
5168 2018-02-09  Vladimir Makarov  <vmakarov@redhat.com>
5169
5170         PR rtl-optimization/57193
5171         * ira-color.c (struct allocno_color_data): Add member
5172         conflict_allocno_hard_prefs.
5173         (update_conflict_allocno_hard_prefs): New.
5174         (bucket_allocno_compare_func): Add a preference based on
5175         conflict_allocno_hard_prefs.
5176         (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
5177         (color_allocnos): Remove a dead code.  Initiate
5178         conflict_allocno_hard_prefs.  Call update_costs_from_prefs.
5179
5180 2018-02-09  Jakub Jelinek  <jakub@redhat.com>
5181
5182         PR target/84226
5183         * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
5184         constraint from =wa to wa.  Avoid a subreg on the output operand,
5185         instead use a pseudo and subreg it in a move.
5186         (p9_xxbrd_<mode>): Changed to ...
5187         (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
5188         (p9_xxbrd_v2df): New expander.
5189         (p9_xxbrw_<mode>): Changed to ...
5190         (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
5191         (p9_xxbrw_v4sf): New expander.
5192
5193 2018-02-09  Sebastian Perta  <sebastian.perta@renesas.com>
5194
5195         * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
5196
5197 2018-02-09  Peter Bergner  <bergner@vnet.ibm.com>
5198
5199         PR target/83926
5200         * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
5201         multiply in 32-bit mode.
5202         (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
5203         (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
5204         mode.
5205
5206 2018-02-09  Sebastian Perta  <sebastian.perta@renesas.com>
5207
5208         * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
5209         to allow or block "symbol_ref" depending on the value of TARGET_JSR.
5210         * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
5211         * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
5212
5213 2018-02-09  Pierre-Marie de Rodat  <derodat@adacore.com>
5214
5215         PR lto/84213
5216         * dwarf2out.c (is_trivial_indirect_ref): New function.
5217         (dwarf2out_late_global_decl): Do not generate a location
5218         attribute for variables that have a non-trivial DECL_VALUE_EXPR
5219         and that are not defined in the current unit.
5220
5221 2018-02-09  Eric Botcazou  <ebotcazou@adacore.com>
5222
5223         * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
5224         instead of a libcall for UNORDERED.
5225
5226 2018-02-09  Tamar Christina  <tamar.christina@arm.com>
5227
5228         PR target/82641
5229         * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
5230         __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
5231
5232 2018-02-09  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
5233
5234         PR target/PR84295
5235         * config/s390/s390.c (s390_set_current_function): Invoke
5236         s390_indirect_branch_settings also if fndecl didn't change.
5237
5238 2018-02-09  Alexandre Oliva  <aoliva@redhat.com>
5239
5240         * config/rs6000/rs6000.md (blockage): Set length to zero.
5241
5242 2018-02-09  Eric Botcazou  <ebotcazou@adacore.com>
5243
5244         * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
5245
5246 2018-02-09  Jakub Jelinek  <jakub@redhat.com>
5247
5248         PR sanitizer/84285
5249         * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
5250         STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
5251         -static-lib*san.
5252
5253         PR debug/84252
5254         * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
5255         PARALLEL incoming that failed vt_get_decl_and_offset check.
5256
5257         PR middle-end/84237
5258         * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
5259         * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
5260         TREE_READONLY bit.
5261         (get_variable_section): For decls in named .bss* sections pass true as
5262         second argument to bss_initializer_p.
5263
5264 2018-02-09  Marek Polacek  <polacek@redhat.com>
5265             Jakub Jelinek  <jakub@redhat.com>
5266
5267         PR c++/83659
5268         * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
5269         Formatting fixes.  Verify first that tree_fits_poly_int64_p (op01).
5270         Sync some changes from cxx_fold_indirect_ref.
5271
5272 2018-02-09  Alexandre Oliva  <aoliva@redhat.com>
5273
5274         * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
5275         markers.
5276         * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
5277         (BLOCK_INLINE_ENTRY_LABEL): New.
5278         (dwarf2out_var_location): Disregard inline entry markers.
5279         (inline_entry_data): New struct.
5280         (inline_entry_data_hasher): New hashtable type.
5281         (inline_entry_data_hasher::hash): New.
5282         (inline_entry_data_hasher::equal): New.
5283         (inline_entry_data_table): New variable.
5284         (add_high_low_attributes): Add DW_AT_entry_pc and
5285         DW_AT_GNU_entry_view attributes if a pending entry is found
5286         in inline_entry_data_table.  Add old entry_pc attribute only
5287         if debug nonbinding markers are disabled.
5288         (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
5289         markers are enabled.
5290         (block_within_block_p, dwarf2out_inline_entry): New.
5291         (dwarf2out_finish): Check that no entries remained in
5292         inline_entry_data_table.
5293         * final.c (reemit_insn_block_notes): Handle inline entry notes.
5294         (final_scan_insn, notice_source_line): Likewise.
5295         (rest_of_clean_state): Skip inline entry markers.
5296         * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
5297         markers.
5298         * gimple.c (gimple_build_debug_inline_entry): New.
5299         * gimple.h (enum gimple_debug_subcode): Add
5300         GIMPLE_DEBUG_INLINE_ENTRY.
5301         (gimple_build_debug_inline_entry): Declare.
5302         (gimple_debug_inline_entry_p): New.
5303         (gimple_debug_nonbind_marker_p): Adjust.
5304         * insn-notes.def (INLINE_ENTRY): New.
5305         * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
5306         inline entry marker notes.
5307         (print_insn): Likewise.
5308         * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
5309         (INSN_DEBUG_MARKER_KIND): Likewise.
5310         (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
5311         * tree-inline.c (expand_call_inline): Build and insert
5312         debug_inline_entry stmt.
5313         * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
5314         inline entry blocks early, if nonbind markers are enabled.
5315         (dump_scope_block): Dump fragment info.
5316         * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
5317         * doc/gimple.texi (gimple_debug_inline_entry_p): New.
5318         (gimple_build_debug_inline_entry): New.
5319         * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
5320         Enable/disable inline entry points too.
5321         * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
5322         (DEBUG_INSN): Describe inline entry markers.
5323
5324         * common.opt (gvariable-location-views): New.
5325         (gvariable-location-views=incompat5): New.
5326         * config.in: Rebuilt.
5327         * configure: Rebuilt.
5328         * configure.ac: Test assembler for view support.
5329         * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
5330         * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
5331         * dwarf2out.c (var_loc_view): New typedef.
5332         (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
5333         (dwarf2out_locviews_in_attribute): New.
5334         (dwarf2out_locviews_in_loclist): New.
5335         (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
5336         (enum dw_line_info_opcode): Add LI_adv_address.
5337         (struct dw_line_info_table): Add view.
5338         (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
5339         (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
5340         (zero_view_p): New variable.
5341         (ZERO_VIEW_P): New macro.
5342         (output_asm_line_debug_info): New.
5343         (struct var_loc_node): Add view.
5344         (add_AT_view_list, AT_loc_list): New.
5345         (add_var_loc_to_decl): Add view param.  Test it against last.
5346         (new_loc_list): Add view params.  Record them.
5347         (AT_loc_list_ptr): Handle loc and view lists.
5348         (view_list_to_loc_list_val_node): New.
5349         (print_dw_val): Handle dw_val_class_view_list.
5350         (size_of_die): Likewise.
5351         (value_format): Likewise.
5352         (loc_list_has_views): New.
5353         (gen_llsym): Set vl_symbol too.
5354         (maybe_gen_llsym, skip_loc_list_entry): New.
5355         (dwarf2out_maybe_output_loclist_view_pair): New.
5356         (output_loc_list): Output view list or entries too.
5357         (output_view_list_offset): New.
5358         (output_die): Handle dw_val_class_view_list.
5359         (output_dwarf_version): New.
5360         (output_compilation_unit_header): Use it.
5361         (output_skeleton_debug_sections): Likewise.
5362         (output_rnglists, output_line_info): Likewise.
5363         (output_pubnames, output_aranges): Update version comments.
5364         (output_one_line_info_table): Output view numbers in asm comments.
5365         (dw_loc_list): Determine current endview, pass it to new_loc_list.
5366         Call maybe_gen_llsym.
5367         (loc_list_from_tree_1): Adjust.
5368         (add_AT_location_description): Create view list attribute if
5369         needed, check it's absent otherwise.
5370         (convert_cfa_to_fb_loc_list): Adjust.
5371         (maybe_emit_file): Call output_asm_line_debug_info for test.
5372         (dwarf2out_var_location): Reset views as needed.  Precompute
5373         add_var_loc_to_decl args.  Call get_attr_min_length only if we have the
5374         attribute.  Set view.
5375         (new_line_info_table): Reset next view.
5376         (set_cur_line_info_table): Call output_asm_line_debug_info for test.
5377         (dwarf2out_source_line): Likewise.  Output view resets and labels to
5378         the assembler, or select appropriate line info opcodes.
5379         (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
5380         (optimize_string_length): Catch it.  Adjust.
5381         (resolve_addr): Copy vl_symbol along with ll_symbol.  Handle
5382         dw_val_class_view_list, and remove it if no longer needed.
5383         (hash_loc_list): Hash view numbers.
5384         (loc_list_hasher::equal): Compare them.
5385         (optimize_location_lists): Check whether a view list symbol is
5386         needed, and whether the locview attribute is present, and
5387         whether they match.  Remove the locview attribute if no longer
5388         needed.
5389         (index_location_lists): Call skip_loc_list_entry for test.
5390         (dwarf2out_finish): Call output_asm_line_debug_info for test.
5391         Use output_dwarf_version.
5392         * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
5393         (struct dw_val_node): Add val_view_list.
5394         * final.c (SEEN_NEXT_VIEW): New.
5395         (set_next_view_needed): New.
5396         (clear_next_view_needed): New.
5397         (maybe_output_next_view): New.
5398         (final_start_function): Rename to...
5399         (final_start_function_1): ... this.  Take pointer to FIRST,
5400         add SEEN parameter.  Emit param bindings in the initial view.
5401         (final_start_function): Reintroduce SEEN-less interface.
5402         (final): Rename to...
5403         (final_1): ... this.  Take SEEN parameter.  Output final pending
5404         next view at the end.
5405         (final): Reintroduce seen-less interface.
5406         (final_scan_insn): Output pending next view before switching
5407         sections or ending a block.  Mark the next view as needed when
5408         outputting variable locations.  Notify debug backend of section
5409         changes, and of location view changes.
5410         (rest_of_handle_final): Adjust.
5411         * toplev.c (process_options): Autodetect value for debug variable
5412         location views option.  Warn on incompat5 without -gdwarf-5.
5413         * doc/invoke.texi (gvariable-location-views): New.
5414         (gvariable-location-views=incompat5): New.
5415         (gno-variable-location-views): New.
5416
5417 2018-02-08  David Malcolm  <dmalcolm@redhat.com>
5418
5419         PR tree-optimization/84136
5420         * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
5421         that the result of find_edge is non-NULL.
5422
5423 2018-02-08  Sergey Shalnov  <sergey.shalnov@intel.com>
5424
5425         PR target/83008
5426         * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
5427         storing integer register in SImode.  Fix cost of 256 and 512
5428         byte aligned SSE register store.
5429
5430 2018-02-08  Sergey Shalnov  <sergey.shalnov@intel.com>
5431
5432         * config/i386/i386.c (ix86_multiplication_cost): Fix
5433         multiplication cost for TARGET_AVX512DQ.
5434
5435 2018-02-08  Marek Polacek  <polacek@redhat.com>
5436
5437         PR tree-optimization/84238
5438         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
5439         get_range_strlen.
5440
5441 2018-02-08  Richard Sandiford  <richard.sandiford@linaro.org>
5442
5443         PR tree-optimization/84265
5444         * tree-vect-stmts.c (vectorizable_store): Don't treat
5445         VMAT_CONTIGUOUS accesses as grouped.
5446         (vectorizable_load): Likewise.
5447
5448 2018-02-08  Richard Sandiford  <richard.sandiford@linaro.org>
5449
5450         PR tree-optimization/81635
5451         * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
5452         * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
5453         (test_round_for_mask): New functions.
5454         (wide_int_cc_tests): Call test_round_for_mask.
5455         * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
5456         * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
5457         * tree-data-ref.c (split_constant_offset_1): Use it to refine the
5458         range returned by get_range_info.
5459
5460 2018-02-08  Jan Hubicka  <hubicka@ucw.cz>
5461
5462         PR ipa/81360
5463         * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
5464         * symtab.c: Include builtins.h
5465         (symtab_node::output_to_lto_symbol_table_p): Move here
5466         from lto-streamer-out.c:output_symbol_p.
5467         * lto-streamer-out.c (write_symbol): Turn early exit to assert.
5468         (output_symbol_p): Move all logic to symtab.c
5469         (produce_symtab): Update.
5470
5471 2018-02-08  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
5472
5473         * config/s390/s390-opts.h (enum indirect_branch): Define.
5474         * config/s390/s390-protos.h (s390_return_addr_from_memory)
5475         (s390_indirect_branch_via_thunk)
5476         (s390_indirect_branch_via_inline_thunk): Add function prototypes.
5477         (enum s390_indirect_branch_type): Define.
5478         * config/s390/s390.c (struct s390_frame_layout, struct
5479         machine_function): Remove.
5480         (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
5481         (indirect_branch_table_label_no, indirect_branch_table_name):
5482         Define variables.
5483         (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
5484         (enum s390_indirect_branch_option): Define.
5485         (s390_return_addr_from_memory): New function.
5486         (s390_handle_string_attribute): New function.
5487         (s390_attribute_table): Add new attribute handler.
5488         (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
5489         (s390_indirect_branch_via_thunk): New function.
5490         (s390_indirect_branch_via_inline_thunk): New function.
5491         (s390_function_ok_for_sibcall): When jumping via thunk disallow
5492         sibling call optimization for non z10 compiles.
5493         (s390_emit_call): Force indirect branch target to be a single
5494         register.  Add r1 clobber for non-z10 compiles.
5495         (s390_emit_epilogue): Emit return jump via return_use expander.
5496         (s390_reorg): Handle JUMP_INSNs as execute targets.
5497         (s390_option_override_internal): Perform validity checks for the
5498         new command line options.
5499         (s390_indirect_branch_attrvalue): New function.
5500         (s390_indirect_branch_settings): New function.
5501         (s390_set_current_function): Invoke s390_indirect_branch_settings.
5502         (s390_output_indirect_thunk_function):  New function.
5503         (s390_code_end): Implement target hook.
5504         (s390_case_values_threshold): Implement target hook.
5505         (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
5506         macros.
5507         * config/s390/s390.h (struct s390_frame_layout)
5508         (struct machine_function): Move here from s390.c.
5509         (TARGET_INDIRECT_BRANCH_NOBP_RET)
5510         (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
5511         (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
5512         (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
5513         (TARGET_INDIRECT_BRANCH_NOBP_CALL)
5514         (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
5515         (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
5516         (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
5517         (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
5518         * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
5519         (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
5520         (mnemonic attribute): Add values which aren't recognized
5521         automatically.
5522         ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
5523         pattern for branch conversion.  Fix mnemonic attribute.
5524         ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
5525         indirect branch via thunk if requested.
5526         ("indirect_jump", "<code>"): Expand patterns for branch conversion.
5527         ("*indirect_jump"): Disable for branch conversion using out of
5528         line thunks.
5529         ("indirect_jump_via_thunk<mode>_z10")
5530         ("indirect_jump_via_thunk<mode>")
5531         ("indirect_jump_via_inlinethunk<mode>_z10")
5532         ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
5533         ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
5534         ("casesi_jump_via_inlinethunk<mode>_z10")
5535         ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
5536         ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
5537         ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
5538         ("*indirect2_jump"): Disable for branch conversion.
5539         ("casesi_jump"): Turn into expander and expand patterns for branch
5540         conversion.
5541         ("return_use"): New expander.
5542         ("*return"): Emit return via thunk and rename it to ...
5543         ("*return<mode>"): ... this one.
5544         * config/s390/s390.opt: Add new options and and enum for the
5545         option values.
5546
5547 2018-02-08  Richard Sandiford  <richard.sandiford@linaro.org>
5548
5549         * lra-constraints.c (match_reload): Unconditionally use
5550         gen_lowpart_SUBREG, rather than selecting between that
5551         and equivalent gen_rtx_SUBREG code.
5552
5553 2018-02-08  Richard Biener  <rguenther@suse.de>
5554
5555         PR tree-optimization/84233
5556         * tree-ssa-phiprop.c (propagate_with_phi): Use separate
5557         changed flag instead of boguously re-using phi_inserted.
5558
5559 2018-02-08  Martin Jambor  <mjambor@suse.cz>
5560
5561         * hsa-gen.c (get_symbol_for_decl): Set program allocation for
5562         static local variables.
5563
5564 2018-02-08  Richard Biener  <rguenther@suse.de>
5565
5566         PR tree-optimization/84278
5567         * tree-vect-stmts.c (vectorizable_store): When looking for
5568         smaller vector types to perform grouped strided loads/stores
5569         make sure the mode is supported by the target.
5570         (vectorizable_load): Likewise.
5571
5572 2018-02-08  Wilco Dijkstra  <wdijkstr@arm.com>
5573
5574         * config/aarch64/aarch64.c (aarch64_components_for_bb):
5575         Increase LDP/STP opportunities by adding adjacent callee-saves.
5576
5577 2018-02-08  Wilco Dijkstra  <wdijkstr@arm.com>
5578
5579         PR rtl-optimization/84068
5580         PR rtl-optimization/83459
5581         * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
5582
5583 2018-02-08  Aldy Hernandez  <aldyh@redhat.com>
5584
5585         PR tree-optimization/84224
5586         * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
5587         * calls.c (gimple_alloca_call_p): Only return TRUE when we have
5588         non-zero arguments.
5589
5590 2018-02-07  Iain Sandoe  <iain@codesourcery.com>
5591
5592         PR target/84113
5593         * config/rs6000/altivec.md (*restore_world): Remove LR use.
5594         * config/rs6000/predicates.md (restore_world_operation): Adjust op
5595         count, remove one USE.
5596
5597 2018-02-07  Michael Meissner  <meissner@linux.vnet.ibm.com>
5598
5599         * doc/install.texi (Configuration): Document the
5600         --with-long-double-format={ibm,ieee} PowerPC configuration
5601         options.
5602
5603         PR target/84154
5604         * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
5605         Convert from define_expand to be define_insn_and_split.  Rework
5606         float/double/_Float128 conversions to QI/HI/SImode to work with
5607         both ISA 2.07 (power8) or ISA 3.0 (power9).  Fix regression where
5608         conversions to QI/HImode types did a store and then a load to
5609         truncate the value.  For conversions to VSX registers, don't split
5610         the insn, instead emit the code directly.  Use the code iterator
5611         any_fix to combine signed and unsigned conversions.
5612         (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
5613         (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
5614         (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
5615         (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
5616         (fix_<mode>di2_hw): Likewise.
5617         (fixuns_<mode>di2_hw): Likewise.
5618         (fix_<mode>si2_hw): Likewise.
5619         (fixuns_<mode>si2_hw): Likewise.
5620         (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
5621         (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
5622         (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
5623         fix<uns>_trunc<SFDF:mode>si2_p8.
5624         (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
5625         used.
5626         (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
5627         (fix<uns>_<mode>_mem): Likewise.
5628         (fctiw<u>z_<mode>_mem): Likewise.
5629         (fix<uns>_<mode>_mem): Likewise.
5630         (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
5631         the register allocator from doing a direct move to the GPRs to do
5632         a store, and instead use the ISA 3.0 store byte/half-word from
5633         vector register instruction.  For IEEE 128-bit floating point,
5634         also optimize stores of 32-bit ints.
5635         (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
5636
5637 2018-02-07  Alan Hayward  <alan.hayward@arm.com>
5638
5639         * genextract.c (push_pathstr_operand): New function to support
5640         [a-zA-Z].
5641         (walk_rtx): Call push_pathstr_operand.
5642         (print_path): Support [a-zA-Z].
5643
5644 2018-02-07  Richard Biener  <rguenther@suse.de>
5645
5646         PR tree-optimization/84037
5647         * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
5648         (cse_and_gimplify_to_preheader): Declare.
5649         (vect_get_place_in_interleaving_chain): Likewise.
5650         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
5651         ivexpr_map.
5652         (_loop_vec_info::~_loop_vec_info): Delete it.
5653         (cse_and_gimplify_to_preheader): New function.
5654         * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
5655         * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
5656         (vectorizable_load): Likewise.  For grouped stores always base
5657         the IV on the first element.
5658         * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
5659         condition before gimplifying.
5660
5661 2018-02-07  Jakub Jelinek  <jakub@redhat.com>
5662
5663         * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
5664         *DIV_EXPR and *MOD_EXPR.
5665
5666 2018-02-07  H.J. Lu  <hongjiu.lu@intel.com>
5667
5668         PR target/84248
5669         * config/i386/i386.c (ix86_option_override_internal): Mask out
5670         the CF_SET bit when checking -fcf-protection.
5671
5672 2018-02-07  Tom de Vries  <tom@codesourcery.com>
5673
5674         PR libgomp/84217
5675         * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
5676         enough.
5677
5678 2018-02-07  Richard Biener  <rguenther@suse.de>
5679
5680         PR tree-optimization/84204
5681         * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
5682         this place.
5683
5684         PR tree-optimization/84205
5685         * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
5686         special-case isl_ast_op_zdiv_r.
5687
5688         PR tree-optimization/84223
5689         * graphite-scop-detection.c (gather_bbs::before_dom_children):
5690         Only add conditions from within the region.
5691         (gather_bbs::after_dom_children): Adjust.
5692
5693 2018-02-07  Georg-Johann Lay  <avr@gjlay.de>
5694
5695         PR target/84209
5696         * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
5697         * config/avr/avr.md: Only post-reload split REG-REG moves if
5698         either register is GENERAL_REG_P.
5699
5700 2018-02-07  Jakub Jelinek  <jakub@redhat.com>
5701
5702         PR tree-optimization/84235
5703         * tree-ssa-scopedtables.c
5704         (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
5705         if the subtraction is performed in floating point type where NaNs are
5706         honored.  For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
5707         build 1.  Formatting fix.
5708
5709 2018-02-06  Jakub Jelinek  <jakub@redhat.com>
5710
5711         PR target/84146
5712         * config/i386/i386.c (rest_of_insert_endbranch): Only skip
5713         NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
5714         and skip it regardless of bb boundaries.  Use CALL_P macro,
5715         don't test INSN_P (insn) together with CALL_P or JUMP_P check
5716         unnecessarily, formatting fix.
5717
5718 2018-02-06  Michael Collison  <michael.collison@arm.com>
5719
5720         * config/arm/thumb2.md:
5721         (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
5722         (*thumb_mov_notscc): Ditto.
5723
5724 2018-02-06  Michael Meissner  <meissner@linux.vnet.ibm.com>
5725
5726         PR target/84154
5727         * config/rs6000/rs6000.md (su code attribute): Use "u" for
5728         unsigned_fix, not "s".
5729
5730 2018-02-06  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
5731
5732         * configure.ac (gcc_fn_eh_frame_ro): New function.
5733         (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
5734         correct .eh_frame permissions.
5735         * configure: Regenerate.
5736
5737 2018-02-06  Andrew Jenner  <andrew@codeourcery.com>
5738
5739         * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
5740         irrelevant options.
5741
5742 2018-02-06  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
5743
5744         * config/rs6000/rs6000.c (rs6000_option_override_internal):
5745         Display warning message for -mno-speculate-indirect-jumps.
5746
5747 2018-02-06  Andrew Jenner  <andrew@codesourcery.com>
5748
5749         * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
5750         Undocumented.
5751         * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
5752
5753 2018-02-06  Aldy Hernandez  <aldyh@redhat.com>
5754
5755         PR tree-optimization/84225
5756         * tree-eh.c (find_trapping_overflow): Only call
5757         operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
5758
5759 2018-02-06  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
5760
5761         PR target/84145
5762         * config/i386/i386.c: Reimplement the check of possible options
5763         -mibt/-mshstk conbination. Change error messages.
5764         * doc/invoke.texi: Fix a typo: remove extra '='.
5765
5766 2018-02-06  Marek Polacek  <polacek@redhat.com>
5767
5768         PR tree-optimization/84228
5769         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
5770
5771 2018-02-06  Tamar Christina  <tamar.christina@arm.com>
5772
5773         PR target/82641
5774         * config/arm/arm.c (arm_print_asm_arch_directives): Record already
5775         emitted arch directives.
5776         * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
5777         __ARM_FEATURE_COPROC before changing architectures.
5778
5779 2018-02-06  Richard Biener  <rguenther@suse.de>
5780
5781         * config/i386/i386.c (print_reg): Fix typo.
5782         (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
5783
5784 2018-02-06  Eric Botcazou  <ebotcazou@adacore.com>
5785
5786         * configure: Regenerate.
5787
5788 2018-02-05  Martin Sebor  <msebor@redhat.com>
5789
5790         PR tree-optimization/83369
5791         * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
5792         inlining context.
5793
5794 2018-02-05  Martin Liska  <mliska@suse.cz>
5795
5796         * doc/invoke.texi: Cherry-pick upstream r323995.
5797
5798 2018-02-05  Richard Sandiford  <richard.sandiford@linaro.org>
5799
5800         * ira.c (ira_init_register_move_cost): Adjust comment.
5801
5802 2018-02-05  Martin Liska  <mliska@suse.cz>
5803
5804         PR gcov-profile/84137
5805         * doc/gcov.texi: Fix typo in documentation.
5806
5807 2018-02-05  Martin Liska  <mliska@suse.cz>
5808
5809         PR gcov-profile/83879
5810         * doc/gcov.texi: Document necessity of --dynamic-list-data when
5811         using dlopen functionality.
5812
5813 2018-02-05  Olga Makhotina  <olga.makhotina@intel.com>
5814
5815         * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
5816         _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
5817         _mm_maskz_range_ss, _mm_mask_range_round_ss,
5818         _mm_maskz_range_round_ss): New intrinsics.
5819         (__builtin_ia32_rangesd128_round)
5820         (__builtin_ia32_rangess128_round): Remove.
5821         (__builtin_ia32_rangesd128_mask_round,
5822         __builtin_ia32_rangess128_mask_round): New builtins.
5823         * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
5824         __builtin_ia32_rangess128_round): Remove.
5825         (__builtin_ia32_rangesd128_mask_round,
5826         __builtin_ia32_rangess128_mask_round): New builtins.
5827         * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
5828         (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
5829         ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
5830         "<round_saeonly_constraint>")): Changed to ...
5831         ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
5832         "<round_saeonly_scalar_constraint>")): ... this.
5833         ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
5834         %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
5835         ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
5836         %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
5837         %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
5838
5839 2018-02-02  Andrew Jenner  <andrew@codesourcery.com>
5840
5841         * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
5842         options.
5843         * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
5844         Remove all values except native, 8540 and 8548.
5845
5846 2018-02-02  H.J. Lu  <hongjiu.lu@intel.com>
5847
5848         * config/i386/i386.c (ix86_output_function_return): Pass
5849         INVALID_REGNUM, instead of -1, as invalid register number to
5850         indirect_thunk_name and output_indirect_thunk.
5851
5852 2018-02-02  Julia Koval  <julia.koval@intel.com>
5853
5854         * config.gcc: Add -march=icelake.
5855         * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
5856         * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
5857         * config/i386/i386.c (processor_costs): Add m_ICELAKE.
5858         (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
5859         PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
5860         (processor_target_table): Add icelake.
5861         (ix86_option_override_internal): Handle new PTAs.
5862         (get_builtin_code_for_version): Handle icelake.
5863         (M_INTEL_COREI7_ICELAKE): New.
5864         (fold_builtin_cpu): Handle icelake.
5865         * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
5866         * doc/invoke.texi: Add -march=icelake.
5867
5868 2018-02-02  Julia Koval  <julia.koval@intel.com>
5869
5870         * config/i386/i386.c (ix86_option_override_internal): Change flags type
5871         to wide_int_bitmask.
5872         * wide-int-bitmask.h: New.
5873
5874 2018-02-02  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
5875
5876         PR target/84066
5877         * config/i386/i386.md: Replace Pmode with word_mode in
5878         builtin_setjmp_setup and builtin_longjmp to support x32.
5879
5880 2018-02-01  Peter Bergner  <bergner@vnet.ibm.com>
5881
5882         PR target/56010
5883         PR target/83743
5884         * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
5885         #include "opts.h".
5886         (rs6000_supported_cpu_names): New static variable.
5887         (linux_cpu_translation_table): Likewise.
5888         (elf_platform) <cpu>: Define new static variable and use it.
5889         Translate kernel AT_PLATFORM name to canonical name if needed.
5890         Error if platform name is unknown.
5891
5892 2018-02-01  Aldy Hernandez  <aldyh@redhat.com>
5893
5894         PR target/84089
5895         * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
5896
5897 2018-02-01  Jeff Law  <law@redhat.com>
5898
5899         PR target/84128
5900         * config/i386/i386.c (release_scratch_register_on_entry): Add new
5901         OFFSET and RELEASE_VIA_POP arguments.  Use SP+OFFSET to restore
5902         the scratch if RELEASE_VIA_POP is false.
5903         (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
5904         If we have to save a temporary register, decrement SIZE appropriately.
5905         Pass new arguments to release_scratch_register_on_entry.
5906         (ix86_adjust_stack_and_probe): Likewise.
5907         (ix86_emit_probe_stack_range): Pass new arguments to
5908         release_scratch_register_on_entry.
5909
5910 2018-02-01  Uros Bizjak  <ubizjak@gmail.com>
5911
5912         PR rtl-optimization/84157
5913         * combine.c (change_zero_ext): Use REG_P predicate in
5914         front of HARD_REGISTER_P predicate.
5915
5916 2018-02-01  Georg-Johann Lay  <avr@gjlay.de>
5917
5918         * config/avr/avr.c (avr_option_override): Move disabling of
5919         -fdelete-null-pointer-checks to...
5920         * common/config/avr/avr-common.c (avr_option_optimization_table):
5921         ...here.
5922
5923 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
5924
5925         PR tree-optimization/81635
5926         * tree-data-ref.c (split_constant_offset_1): For types that
5927         wrap on overflow, try to use range info to prove that wrapping
5928         cannot occur.
5929
5930 2018-02-01  Renlin Li  <renlin.li@arm.com>
5931
5932         PR target/83370
5933         * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
5934         TAILCALL_ADDR_REGS.
5935         (aarch64_register_move_cost): Likewise.
5936         * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
5937         TAILCALL_ADDR_REGS.
5938         (REG_CLASS_NAMES): Likewise.
5939         (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
5940         TAILCALL_ADDR_REGS. Remove IP registers.
5941         * config/aarch64/aarch64.md (Ucs): Update register constraint.
5942
5943 2018-02-01  Richard Biener  <rguenther@suse.de>
5944
5945         * domwalk.h (dom_walker::dom_walker): Add additional constructor
5946         for specifying RPO order and allow NULL for that.
5947         * domwalk.c (dom_walker::dom_walker): Likewise.
5948         (dom_walker::walk): Handle NULL RPO order.
5949         * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
5950         in RPO order.
5951         (rewrite_update_dom_walker): Likewise.
5952         (mark_def_dom_walker): Likewise.
5953
5954 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
5955
5956         * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
5957         (aarch64_maybe_expand_sve_subreg_move): Declare.
5958         * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
5959         * config/aarch64/predicates.md (aarch64_any_register_operand): New
5960         predicate.
5961         * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
5962         that are semantically a reverse operation.
5963         (*aarch64_sve_mov<mode>_subreg_be): New pattern.
5964         * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
5965         (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
5966         functions.
5967         (aarch64_can_change_mode_class): For big-endian, forbid changes
5968         between two SVE modes if they have different element sizes.
5969
5970 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
5971
5972         * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
5973         the TImode handling for big-endian targets.
5974
5975 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
5976
5977         * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
5978         (*sve_ld1rq<Vesize>): ... this new pattern.  Handle all element sizes,
5979         not just bytes.
5980         * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
5981         Remove BSWAP handing for big-endian targets and use the form of
5982         LD1RQ appropariate for the mode.
5983
5984 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
5985
5986         * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
5987         all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
5988         duplicated element.
5989
5990 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
5991
5992         PR tearget/83845
5993         * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
5994         check for operands that need to go through aarch64_sve_reload_be.
5995
5996 2018-02-01  Jakub Jelinek  <jakub@redhat.com>
5997
5998         PR tree-optimization/81661
5999         PR tree-optimization/84117
6000         * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
6001         * tree-eh.c: Include gimplify.h.
6002         (find_trapping_overflow, replace_trapping_overflow,
6003         rewrite_to_non_trapping_overflow): New functions.
6004         * tree-vect-loop.c: Include tree-eh.h.
6005         (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
6006         * tree-data-ref.c: Include tree-eh.h.
6007         (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
6008
6009 2018-01-31  Uros Bizjak  <ubizjak@gmail.com>
6010
6011         PR rtl-optimization/84123
6012         * combine.c (change_zero_ext): Check if hard register satisfies
6013         can_change_dest_mode before calling gen_lowpart_SUBREG.
6014
6015 2018-01-31  Vladimir Makarov  <vmakarov@redhat.com>
6016
6017         PR target/82444
6018         * ira.c (ira_init_register_move_cost): Remove assert.
6019
6020 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
6021
6022         PR rtl-optimization/84071
6023         * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
6024         * doc/tm.texi: Regenerate.
6025
6026 2018-01-31  Richard Biener  <rguenther@suse.de>
6027
6028         PR tree-optimization/84132
6029         * tree-data-ref.c (analyze_miv_subscript): Properly
6030         check whether evolution_function_is_affine_multivariate_p
6031         before calling gcd_of_steps_may_divide_p.
6032
6033 2018-01-31  Julia Koval  <julia.koval@intel.com>
6034
6035         PR target/83618
6036         * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
6037         * config/i386/i386.md (rdpid_rex64) New.
6038         (rdpid): Make 32bit only.
6039
6040 2018-01-29  Aldy Hernandez  <aldyh@redhat.com>
6041
6042         PR lto/84105
6043         * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
6044         an IDENTIFIER_NODE for FUNCTION_TYPE's.
6045
6046 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
6047
6048         Revert
6049         2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
6050
6051         * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
6052
6053 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
6054
6055         PR rtl-optimization/84071
6056         * combine.c (record_dead_and_set_regs_1): Record the source unmodified
6057         for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
6058
6059 2018-01-31  Claudiu Zissulescu  <claziss@synopsys.com>
6060
6061         * config/arc/arc.c (arc_handle_aux_attribute): New function.
6062         (arc_attribute_table): Add 'aux' attribute.
6063         (arc_in_small_data_p): Consider aux like variables.
6064         (arc_is_aux_reg_p): New function.
6065         (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
6066         (arc_get_aux_arg): New function.
6067         (prepare_move_operands): Handle aux-register access.
6068         (arc_handle_aux_attribute): New function.
6069         * doc/extend.texi (ARC Variable attributes): Add subsection.
6070
6071 2018-01-31  Claudiu Zissulescu  <claziss@synopsys.com>
6072
6073         * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
6074         * config/arc/arc.c (arc_handle_uncached_attribute): New function.
6075         (arc_attribute_table): Add 'uncached' attribute.
6076         (arc_print_operand): Print '.di' flag for uncached memory
6077         accesses.
6078         (arc_in_small_data_p): Do not consider for small data the uncached
6079         types.
6080         (arc_is_uncached_mem_p): New function.
6081         * config/arc/predicates.md (compact_store_memory_operand): Check
6082         for uncached memory accesses.
6083         (nonvol_nonimm_operand): Likewise.
6084         * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
6085
6086 2018-01-31  Jakub Jelinek  <jakub@redhat.com>
6087
6088         PR c/84100
6089         * common.opt (falign-functions=, falign-jumps=, falign-labels=,
6090         falign-loops=): Add Optimization flag.
6091
6092 2018-01-30  Jeff Law  <law@redhat.com>
6093
6094         PR target/84064
6095         * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
6096         INT_REGISTERS_SAVED.  Check it prior to calling
6097         get_scratch_register_on_entry.
6098         (ix86_adjust_stack_and_probe): Similarly.
6099         (ix86_emit_probe_stack_range): Similarly.
6100         (ix86_expand_prologue): Corresponding changes.
6101
6102 2018-01-30  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
6103
6104         PR target/40411
6105         * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
6106         -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
6107
6108 2018-01-30  Vladimir Makarov  <vmakarov@redhat.com>
6109
6110         PR target/84112
6111         * lra-constraints.c (curr_insn_transform): Process AND in the
6112         address.
6113
6114 2018-01-30  Jakub Jelinek  <jakub@redhat.com>
6115
6116         PR rtl-optimization/83986
6117         * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
6118         dependence against last_pending_memory_flush in addition to
6119         pending_jump_insns.
6120
6121 2018-01-30  Alexandre Oliva  <aoliva@redhat.com>
6122
6123         PR tree-optimization/81611
6124         * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
6125         copies.
6126
6127 2018-01-30  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
6128
6129         PR target/83758
6130         * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
6131         a reg rtx.
6132
6133 2018-01-30  Richard Biener  <rguenther@suse.de>
6134             Jakub Jelinek  <jakub@redhat.com>
6135
6136         PR tree-optimization/84111
6137         * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
6138         inner loops added during recursion, as they don't have up-to-date
6139         SSA form.
6140
6141 2018-01-30  Jan Hubicka  <hubicka@ucw.cz>
6142
6143         PR ipa/81360
6144         * ipa-inline.c (can_inline_edge_p): Break out late tests to...
6145         (can_inline_edge_by_limits_p): ... here.
6146         (can_early_inline_edge_p, check_callers,
6147         update_caller_keys, update_callee_keys, recursive_inlining,
6148         add_new_edges_to_heap, speculation_useful_p,
6149         inline_small_functions,
6150         inline_small_functions, flatten_function,
6151         inline_to_all_callers_1): Update.
6152
6153 2018-01-30  Jan Hubicka  <hubicka@ucw.cz>
6154
6155         * profile-count.c (profile_count::combine_with_ipa_count): Handle
6156         zeros correctly.
6157
6158 2018-01-30  Richard Biener  <rguenther@suse.de>
6159
6160         PR tree-optimization/83008
6161         * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
6162         invariant and constant vector uses in stmts when they need
6163         more than one stmt.
6164
6165 2018-01-30  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
6166
6167         PR bootstrap/84017
6168         * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
6169         * configure: Regenerate.
6170
6171 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
6172
6173         * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
6174         pattern.
6175         (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
6176         Use gen_rtx_REG rather than gen_lowpart.
6177
6178 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
6179
6180         * lra-constraints.c (match_reload): Use subreg_lowpart_offset
6181         rather than 0 when creating partial subregs.
6182
6183 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
6184
6185         * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
6186         of usage.
6187
6188 2018-01-29  Michael Meissner  <meissner@linux.vnet.ibm.com>
6189
6190         PR target/81550
6191         * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
6192         and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
6193         -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
6194         flags.  This restores the settings used before the 2017-07-24.
6195         Turning off pre increment/decrement/modify allows IVOPTS to
6196         optimize DF/SF loops where the index is an int.
6197
6198 2018-01-29  Richard Biener  <rguenther@suse.de>
6199             Kelvin Nilsen  <kelvin@gcc.gnu.org>
6200
6201         PR bootstrap/80867
6202         * tree-vect-stmts.c (vectorizable_call): Don't call
6203         targetm.vectorize_builtin_md_vectorized_function if callee is
6204         NULL.
6205
6206 2018-01-22  Carl Love  <cel@us.ibm.com>
6207
6208         * doc/extend.tex: Fix typo in second arg in
6209         __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
6210
6211 2018-01-29  Richard Biener  <rguenther@suse.de>
6212
6213         PR tree-optimization/84086
6214         * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
6215         (flush_ssaname_freelist): When SSA names were released reset
6216         the SCEV hash table.
6217
6218 2018-01-29  Richard Biener  <rguenther@suse.de>
6219
6220         PR tree-optimization/84057
6221         * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
6222         removed paths when removing edges.
6223
6224 2018-01-27  H.J. Lu  <hongjiu.lu@intel.com>
6225
6226         * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
6227         -mfunction-return=@var{choice}.
6228
6229 2018-01-27  Bernd Edlinger  <bernd.edlinger@hotmail.de>
6230
6231         PR diagnostic/84034
6232         * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
6233         Handle CR like TAB.
6234         (layout::print_source_line): Likewise.
6235         (test_get_line_width_without_trailing_whitespace): Add test cases.
6236
6237 2018-01-27  Jakub Jelinek  <jakub@redhat.com>
6238
6239         PR middle-end/84040
6240         * sched-deps.c (sched_macro_fuse_insns): Return immediately for
6241         debug insns.
6242
6243 2018-01-26  Jim Wilson  <jimw@sifive.com>
6244
6245         * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
6246
6247         * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
6248         specified.
6249
6250 2018-01-26  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
6251
6252         * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
6253         and CMP + SUB-immediate -> SUBS.
6254
6255 2018-01-26  Martin Sebor  <msebor@redhat.com>
6256
6257         PR tree-optimization/83896
6258         * tree-ssa-strlen.c (get_string_len): Rename...
6259         (get_string_cst_length): ...to this.  Return HOST_WIDE_INT.
6260         Avoid assuming length is constant.
6261         (handle_char_store): Use HOST_WIDE_INT for string length.
6262
6263 2018-01-26  Uros Bizjak  <ubizjak@gmail.com>
6264
6265         PR target/81763
6266         * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
6267         to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
6268
6269 2018-01-26  Richard Biener  <rguenther@suse.de>
6270
6271         PR rtl-optimization/84003
6272         * dse.c (record_store): Only record redundant stores when
6273         the earlier store aliases at least all accesses the later one does.
6274
6275 2018-01-26  Jakub Jelinek  <jakub@redhat.com>
6276
6277         PR rtl-optimization/83985
6278         * dce.c (deletable_insn_p): Return false for separate shrink wrapping
6279         REG_CFA_RESTORE insns.
6280         (delete_unmarked_insns): Don't ignore separate shrink wrapping
6281         REG_CFA_RESTORE insns here.
6282
6283         PR c/83989
6284         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
6285         use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
6286
6287 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
6288
6289         * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
6290         * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
6291         (arc_init): Likewise.
6292         (arc_override_options): Likewise.
6293         (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
6294         value.
6295         (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
6296         support.
6297         * config/arc/arc.h (TARGET_DBNZ): Define.
6298         * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
6299         properly set the tune attribute.
6300         (dbnz): Use TARGET_DBNZ guard.
6301         * config/arc/arc.opt (mtune): Add core3 option.
6302
6303 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
6304
6305         * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
6306         recognize new pic like addresses.
6307         (arc_delegitimize_address): Clean up.
6308
6309 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
6310
6311         * config/arc/arc-arches.def: Option mrf16 valid for all
6312         architectures.
6313         * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
6314         * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
6315         * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
6316         * config/arc/arc-tables.opt: Regenerate.
6317         * config/arc/arc.c (arc_conditional_register_usage): Handle
6318         reduced register file case.
6319         (arc_file_start): Set must have build attributes.
6320         * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
6321         mrf16 option value.
6322         * config/arc/arc.opt (mrf16): Add new option.
6323         * config/arc/elf.h (ATTRIBUTE_PCS): Define.
6324         * config/arc/genmultilib.awk: Handle new mrf16 option.
6325         * config/arc/linux.h (ATTRIBUTE_PCS): Define.
6326         * config/arc/t-multilib: Regenerate.
6327         * doc/invoke.texi (ARC Options): Document mrf16 option.
6328
6329 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
6330
6331         * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
6332         * config/arc/arc.c (arc_handle_secure_attribute): New function.
6333         (arc_attribute_table): Add 'secure_call' attribute.
6334         (arc_print_operand): Print secure call operand.
6335         (arc_function_ok_for_sibcall): Don't optimize tail calls when
6336         secure.
6337         (arc_is_secure_call_p): New function.  * config/arc/arc.md
6338         (call_i): Add support for sjli instruction.
6339         (call_value_i): Likewise.
6340         * config/arc/constraints.md (Csc): New constraint.
6341
6342 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
6343             John Eric Martin  <John.Martin@emmicro-us.com>
6344
6345         * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
6346         * config/arc/arc.c (_arc_jli_section): New struct.
6347         (arc_jli_section): New type.
6348         (rc_jli_sections): New static variable.
6349         (arc_handle_jli_attribute): New function.
6350         (arc_attribute_table): Add jli_always and jli_fixed attribute.
6351         (arc_file_end): New function.
6352         (TARGET_ASM_FILE_END): Define.
6353         (arc_print_operand): Reuse 'S' letter for JLI output instruction.
6354         (arc_add_jli_section): New function.
6355         (jli_call_scan): Likewise.
6356         (arc_reorg): Call jli_call_scan.
6357         (arc_output_addsi): Remove 'S' from printing asm operand.
6358         (arc_is_jli_call_p): New function.
6359         * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
6360         operand.
6361         (movhi_insn): Likewise.
6362         (movsi_insn): Likewise.
6363         (movsi_set_cc_insn): Likewise.
6364         (loadqi_update): Likewise.
6365         (load_zeroextendqisi_update): Likewise.
6366         (load_signextendqisi_update): Likewise.
6367         (loadhi_update): Likewise.
6368         (load_zeroextendhisi_update): Likewise.
6369         (load_signextendhisi_update): Likewise.
6370         (loadsi_update): Likewise.
6371         (loadsf_update): Likewise.
6372         (movsicc_insn): Likewise.
6373         (bset_insn): Likewise.
6374         (bxor_insn): Likewise.
6375         (bclr_insn): Likewise.
6376         (bmsk_insn): Likewise.
6377         (bicsi3_insn): Likewise.
6378         (cmpsi_cc_c_insn): Likewise.
6379         (movsi_ne): Likewise.
6380         (movsi_cond_exec): Likewise.
6381         (clrsbsi2): Likewise.
6382         (norm_f): Likewise.
6383         (normw): Likewise.
6384         (swap): Likewise.
6385         (divaw): Likewise.
6386         (flag): Likewise.
6387         (sr): Likewise.
6388         (kflag): Likewise.
6389         (ffs): Likewise.
6390         (ffs_f): Likewise.
6391         (fls): Likewise.
6392         (call_i): Remove 'S' asm letter, add jli instruction.
6393         (call_value_i): Likewise.
6394         * config/arc/arc.op (mjli-always): New option.
6395         * config/arc/constraints.md (Cji): New constraint.
6396         * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
6397         operand.
6398         (subsf3_fpx): Likewise.
6399         (mulsf3_fpx): Likewise.
6400         * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
6401         asm operand.
6402         * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
6403         function attrbutes.
6404         * doc/invoke.texi (ARC): Document mjli-always option.
6405
6406 2018-01-26  Sebastian Perta  <sebastian.perta@renesas.com>
6407
6408         * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
6409         avoid addition with 0 and use incw and decw where possible.
6410
6411 2018-01-26  Richard Biener  <rguenther@suse.de>
6412
6413         PR tree-optimization/81082
6414         * fold-const.c (fold_plusminus_mult_expr): Do not perform the
6415         association if it requires casting to unsigned.
6416         * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
6417         from fold_plusminus_mult_expr to catch important cases late when
6418         range info is available.
6419
6420 2018-01-26  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
6421
6422         * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
6423         * configure.ac (hidden_linkonce): New test.
6424         * configure: Regenerate.
6425         * config.in: Regenerate.
6426
6427 2018-01-26  Julia Koval  <julia.koval@intel.com>
6428
6429         * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
6430         _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
6431         _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
6432         _mm_mask_bitshuffle_epi64_mask): Fix type.
6433         * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
6434         USI_FTYPE_V4DI_V4DI_USI): Remove.
6435         * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
6436         __builtin_ia32_vpshufbitqmb256_mask,
6437         __builtin_ia32_vpshufbitqmb128_mask): Fix types.
6438         * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
6439         * config/i386/sse.md (VI1_AVX512VLBW): Change types.
6440
6441 2018-01-26  Alan Modra  <amodra@gmail.com>
6442
6443         PR target/84033
6444         * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
6445         UNSPEC_VBPERMQ.  Sort other unspecs.
6446
6447 2018-01-25  David Edelsohn  <dje.gcc@gmail.com>
6448
6449         * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
6450
6451 2018-01-25  Jan Hubicka  <hubicka@ucw.cz>
6452
6453         PR middle-end/83055
6454         * predict.c (drop_profile): Do not push/pop cfun; update also
6455         node->count.
6456         (handle_missing_profiles): Fix logic looking for zero profiles.
6457
6458 2018-01-25  Jakub Jelinek  <jakub@redhat.com>
6459
6460         PR middle-end/83977
6461         * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
6462         on functions with #pragma omp declare simd or functions with simd
6463         attribute.
6464         * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
6465         * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
6466         Remove trailing \n from warning_at calls.
6467
6468 2018-01-25  Tom de Vries  <tom@codesourcery.com>
6469
6470         PR target/84028
6471         * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
6472         for neutered workers.
6473
6474 2018-01-24  Joseph Myers  <joseph@codesourcery.com>
6475
6476         PR target/68467
6477         * config/m68k/m68k.c (m68k_promote_function_mode): New function.
6478         (TARGET_PROMOTE_FUNCTION_MODE): New macro.
6479
6480 2018-01-24  Jeff Law  <law@redhat.com>
6481
6482         PR target/83994
6483         * i386.c (get_probe_interval): Move to earlier point.
6484         (ix86_compute_frame_layout): If -fstack-clash-protection and
6485         the frame is larger than the probe interval, then use pushes
6486         to save registers rather than reg->mem moves.
6487         (ix86_expand_prologue): Remove conditional for int_registers_saved
6488         assertion.
6489
6490 2018-01-24  Vladimir Makarov  <vmakarov@redhat.com>
6491
6492         PR target/84014
6493         * ira-build.c (setup_min_max_allocno_live_range_point): Set up
6494         min/max for never referenced object.
6495
6496 2018-01-24  Jakub Jelinek  <jakub@redhat.com>
6497
6498         PR middle-end/83977
6499         * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
6500         here.
6501         * omp-low.c (create_omp_child_function): Remove "omp declare simd"
6502         attributes from DECL_ATTRIBUTES (decl) without affecting
6503         DECL_ATTRIBUTES (current_function_decl).
6504         * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
6505         functions with non-NULL DECL_ABSTRACT_ORIGIN.
6506
6507 2018-01-24  Richard Sandiford  <richard.sandiford@linaro.org>
6508
6509         PR tree-optimization/83979
6510         * fold-const.c (fold_comparison): Use constant_boolean_node
6511         instead of boolean_{true,false}_node.
6512
6513 2018-01-24  Jan Hubicka  <hubicka@ucw.cz>
6514
6515         * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
6516         with zero counts.
6517
6518 2018-01-24  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
6519
6520         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
6521         Simplify the clause that sets the length attribute.
6522         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
6523         (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
6524         clause that sets the length attribute.
6525         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
6526
6527 2018-01-24  Tom de Vries  <tom@codesourcery.com>
6528
6529         PR target/83589
6530         * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
6531         (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
6532         Add strict parameter.
6533         (prevent_branch_around_nothing): Insert dummy insn between branch to
6534         label and label with no ptx insn inbetween.
6535         * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
6536
6537 2018-01-24  Tom de Vries  <tom@codesourcery.com>
6538
6539         PR target/81352
6540         * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
6541         for neutered threads in warp.
6542         * config/nvptx/nvptx.md (define_insn "exit"): New insn.
6543
6544 2018-01-24  Richard Biener  <rguenther@suse.de>
6545
6546         PR tree-optimization/83176
6547         * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
6548         operands.
6549
6550 2018-01-24  Richard Biener  <rguenther@suse.de>
6551
6552         PR tree-optimization/82819
6553         * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
6554         code generating pluses that are no-ops in the target precision.
6555
6556 2018-01-24  Richard Biener  <rguenther@suse.de>
6557
6558         PR middle-end/84000
6559         * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
6560
6561 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
6562
6563         * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
6564         to merge probabilities.
6565         * predict.c (probably_never_executed): Also mark as cold functions
6566         with global 0 profile and guessed local profile.
6567         * profile-count.c (profile_probability::combine_with_count): New
6568         member function.
6569         * profile-count.h (profile_probability::operator*,
6570         profile_probability::operator*=, profile_probability::operator/,
6571         profile_probability::operator/=): Reduce precision to adjusted
6572         and set value to guessed on contradictory divisions.
6573         (profile_probability::combine_with_freq): Remove.
6574         (profile_probability::combine_wiht_count): Declare.
6575         (profile_count::force_nonzero):: Set to adjusted.
6576         (profile_count::probability_in):: Set quality to adjusted.
6577         * tree-ssa-tail-merge.c (replace_block_by): Use
6578         combine_with_count.
6579
6580 2018-01-23  Andrew Waterman  <andrew@sifive.com>
6581             Jim Wilson  <jimw@sifive.com>
6582
6583         * config/riscv/riscv.c (riscv_stack_boundary): New.
6584         (riscv_option_override): Set riscv_stack_boundary.  Handle
6585         riscv_preferred_stack_boundary_arg.
6586         * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
6587         (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
6588         (STACK_BOUNDARY): Set to riscv_stack_boundary.
6589         (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
6590         * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
6591         * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
6592
6593 2018-01-23  H.J. Lu  <hongjiu.lu@intel.com>
6594
6595         PR target/83905
6596         * config/i386/i386.c (ix86_expand_prologue): Use cost reference
6597         of struct ix86_frame.
6598         (ix86_expand_epilogue): Likewise.  Add a local variable for
6599         the reg_save_offset field in struct ix86_frame.
6600
6601 2018-01-23  Bin Cheng  <bin.cheng@arm.com>
6602
6603         PR tree-optimization/82604
6604         * tree-loop-distribution.c (enum partition_kind): New enum item
6605         PKIND_PARTIAL_MEMSET.
6606         (partition_builtin_p): Support above new enum item.
6607         (generate_code_for_partition): Ditto.
6608         (compute_access_range): Differentiate cases that equality can be
6609         proven at all loops, the innermost loops or no loops.
6610         (classify_builtin_st, classify_builtin_ldst): Adjust call to above
6611         function.  Set PKIND_PARTIAL_MEMSET for partition appropriately.
6612         (finalize_partitions, distribute_loop): Don't fuse partition of
6613         PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
6614         (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
6615         parloop is enabled.
6616
6617 2018-01-23  Martin Liska  <mliska@suse.cz>
6618
6619         * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
6620         order to ignore the predictor.
6621         (PRED_POLYMORPHIC_CALL): Likewise.
6622         (PRED_RECURSIVE_CALL): Likewise.
6623
6624 2018-01-23  Martin Liska  <mliska@suse.cz>
6625
6626         * tree-profile.c (tree_profiling): Print function header to
6627         aware reader which function we are working on.
6628         * value-prof.c (gimple_find_values_to_profile): Do not print
6629         not interesting value histograms.
6630
6631 2018-01-23  Martin Liska  <mliska@suse.cz>
6632
6633         * profile-count.h (enum profile_quality): Add
6634         profile_uninitialized as the first value. Do not number values
6635         as they are zero based.
6636         (profile_count::verify): Update sanity check.
6637         (profile_probability::verify): Likewise.
6638
6639 2018-01-23  Nathan Sidwell  <nathan@acm.org>
6640
6641         * doc/invoke.texi (ffor-scope): Deprecate.
6642
6643 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
6644
6645         PR tree-optimization/83510
6646         * domwalk.c (set_all_edges_as_executable): New function.
6647         (dom_walker::dom_walker): Convert bool param
6648         "skip_unreachable_blocks" to enum reachability.  Move setup of
6649         edge flags to set_all_edges_as_executable and only do it when
6650         reachability is REACHABLE_BLOCKS.
6651         * domwalk.h (enum dom_walker::reachability): New enum.
6652         (dom_walker::dom_walker): Convert bool param
6653         "skip_unreachable_blocks" to enum reachability.
6654         (set_all_edges_as_executable): New decl.
6655         * graphite-scop-detection.c  (gather_bbs::gather_bbs): Convert
6656         from false for "skip_unreachable_blocks" to ALL_BLOCKS for
6657         "reachability".
6658         * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
6659         but converting true to REACHABLE_BLOCKS.
6660         * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
6661         * tree-vrp.c
6662         (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
6663         Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
6664         (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
6665         REACHABLE_BLOCKS.
6666         (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
6667         if check_all_array_refs will be called.
6668
6669 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
6670
6671         * tree.c (selftest::test_location_wrappers): Add more test
6672         coverage.
6673
6674 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
6675
6676         * sbitmap.c (selftest::test_set_range): Fix memory leaks.
6677         (selftest::test_bit_in_range): Likewise.
6678
6679 2018-01-23  Richard Sandiford  <richard.sandiford@linaro.org>
6680
6681         PR testsuite/83888
6682         * doc/sourcebuild.texi (vect_float): Say that the selector
6683         only describes the situation when -funsafe-math-optimizations is on.
6684         (vect_float_strict): Document.
6685
6686 2018-01-23  Richard Sandiford  <richard.sandiford@linaro.org>
6687
6688         PR tree-optimization/83965
6689         * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
6690         (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
6691         instead of checking only for a reduction.
6692         (vect_recog_widen_sum_pattern): Likewise.
6693
6694 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
6695
6696         * predict.c (probably_never_executed): Only use precise profile info.
6697         (compute_function_frequency): Skip after inlining hack since we now
6698         have quality checking.
6699
6700 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
6701
6702         * profile-count.h (profile_probability::very_unlikely,
6703         profile_probability::unlikely, profile_probability::even): Set
6704         precision to guessed.
6705
6706 2018-01-23  Richard Biener  <rguenther@suse.de>
6707
6708         PR tree-optimization/83963
6709         * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
6710         Properly terminate dominator walk when crossing the exit edge not
6711         when visiting its source block.
6712
6713 2018-01-23  Jakub Jelinek  <jakub@redhat.com>
6714
6715         PR c++/83918
6716         * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
6717         VIEW_CONVERT_EXPR to wrap CONST_DECLs.
6718
6719 2018-01-22  Jakub Jelinek  <jakub@redhat.com>
6720
6721         PR tree-optimization/83957
6722         * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs.  Remove
6723         semicolon after for body surrounded by braces.
6724
6725         PR tree-optimization/83081
6726         * profile-count.h (profile_probability::split): New method.
6727         * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
6728         Use profile_probability::split.
6729         (do_compare_rtx_and_jump): Fix adjustment of probabilities
6730         when splitting a single conditional jump into 2.
6731
6732 2018-01-22  David Malcolm  <dmalcolm@redhat.com>
6733
6734         PR tree-optimization/69452
6735         * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
6736         decl.
6737
6738 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6739
6740         * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
6741         * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
6742         * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
6743
6744 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6745
6746         * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
6747         * config/rl78/rl78.md (movdi): New define_expand.
6748         * config/rl78/rl78.c (rl78_split_movdi): New function.
6749
6750 2018-01-22  Michael Meissner  <meissner@linux.vnet.ibm.com>
6751
6752         PR target/83862
6753         * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
6754         no longer used.
6755         * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
6756         * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
6757         128-bit to produce an UNSPEC move to get the double word with the
6758         signbit and then a shift directly to do signbit.
6759         (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
6760         implementation with a new version that just does either a direct
6761         move or a regular move.  Move memory interface to separate insns.
6762         Move insns so they are next to the expander.
6763         (signbit<mode>2_dm_mem_be): New combiner insns to combine load
6764         with signbit move.  Split big and little endian case.
6765         (signbit<mode>2_dm_mem_le): Likewise.
6766         (signbit<mode>2_dm_<su>ext): Delete, no longer used.
6767         (signbit<mode>2_dm2): Likewise.
6768
6769 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6770
6771         * config/rl78/rl78.md (anddi3): New define_expand.
6772
6773 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6774
6775         * config/rl78/rl78.md (umindi3): New define_expand.
6776
6777 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6778
6779         * config/rl78/rl78.md (smindi3): New define_expand.
6780
6781 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6782
6783         * config/rl78/rl78.md (smaxdi3): New define_expand.
6784
6785 2018-01-22  Carl Love  <cel@us.ibm.com>
6786
6787         * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
6788         LVX_V1TI): Add macro expansion.
6789         * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
6790         definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
6791         VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
6792         * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
6793         Change check to determine if the instruction is a byte reversing
6794         entry.  Fix typo in comment.
6795         * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
6796         for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
6797         Add def_builtin calls for new builtins.
6798         * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
6799         Add define_insn expansion.
6800
6801 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6802
6803         * config/rl78/rl78.md (umaxdi3): New define_expand.
6804
6805 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6806
6807         * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
6808         for non-QImode registers.
6809
6810 2018-01-22  Richard Biener  <rguenther@suse.de>
6811
6812         PR tree-optimization/83963
6813         * graphite-scop-detection.c (scop_detection::get_sese): Delay
6814         including the loop exit block.
6815         (scop_detection::merge_sese): Likewise.
6816         (scop_detection::add_scop): Do it here instead.
6817
6818 2018-01-22  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
6819
6820         * doc/sourcebuild.texi (arm_softfloat): Document.
6821
6822 2018-01-21  John David Anglin  <danglin@gcc.gnu.org>
6823
6824         PR gcc/77734
6825         * config/pa/pa.c (pa_function_ok_for_sibcall): Use
6826         targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
6827         Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
6828
6829 2018-01-21  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
6830             David Edelsohn  <dje.gcc@gmail.com>
6831
6832         PR target/83946
6833         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
6834         Change "crset eq" to "crset 2".
6835         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
6836         (*call_indirect_aix<mode>_nospec): Likewise.
6837         (*call_value_indirect_aix<mode>_nospec): Likewise.
6838         (*call_indirect_elfv2<mode>_nospec): Likewise.
6839         (*call_value_indirect_elfv2<mode>_nospec): Likewise.
6840         (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
6841         change assembly output from . to $.
6842         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
6843         (indirect_jump<mode>_nospec): Change assembly output from . to $.
6844         (*tablejump<mode>_internal1_nospec): Likewise.
6845
6846 2018-01-21  Oleg Endo  <olegendo@gcc.gnu.org>
6847
6848         PR target/80870
6849         * config/sh/sh_optimize_sett_clrt.cc:
6850         Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
6851
6852 2018-01-20  Richard Sandiford  <richard.sandiford@linaro.org>
6853
6854         PR tree-optimization/83940
6855         * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
6856         offset_dt to vect_constant_def rather than vect_unknown_def_type.
6857         (vect_check_load_store_mask): Add a mask_dt_out parameter and
6858         use it to pass back the definition type.
6859         (vect_check_store_rhs): Likewise rhs_dt_out.
6860         (vect_build_gather_load_calls): Add a mask_dt argument and use
6861         it instead of a call to vect_is_simple_use.
6862         (vectorizable_store): Update calls to vect_check_load_store_mask
6863         and vect_check_store_rhs.  Use the dt returned by the latter instead
6864         of scatter_src_dt.  Use the cached mask_dt and gs_info.offset_dt
6865         instead of calls to vect_is_simple_use.  Pass the scalar rather
6866         than the vector operand to vect_is_simple_use when handling
6867         second and subsequent copies of an rhs value.
6868         (vectorizable_load): Update calls to vect_check_load_store_mask
6869         and vect_build_gather_load_calls.  Use the cached mask_dt and
6870         gs_info.offset_dt instead of calls to vect_is_simple_use.
6871
6872 2018-01-20  Jakub Jelinek  <jakub@redhat.com>
6873
6874         PR middle-end/83945
6875         * tree-emutls.c: Include gimplify.h.
6876         (lower_emutls_2): New function.
6877         (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
6878         with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
6879         it before further processing.
6880
6881         PR target/83930
6882         * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
6883         UINTVAL (trueop1) instead of INTVAL (op1).
6884
6885 2018-01-19  Jakub Jelinek  <jakub@redhat.com>
6886
6887         PR debug/81570
6888         PR debug/83728
6889         * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
6890         INCOMING_FRAME_SP_OFFSET if not defined.
6891         (scan_trace): Add ENTRY argument.  If true and
6892         DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
6893         emit a note to adjust the CFA offset.
6894         (create_cfi_notes): Adjust scan_trace callers.
6895         (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
6896         INCOMING_FRAME_SP_OFFSET in the CIE.
6897         * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
6898         * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
6899         Likewise.
6900         * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
6901         * doc/tm.texi: Regenerated.
6902
6903 2018-01-19  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
6904
6905         PR rtl-optimization/83147
6906         * lra-constraints.c (remove_inheritance_pseudos): Use
6907         lra_substitute_pseudo_within_insn.
6908
6909 2018-01-19  Tom de Vries  <tom@codesourcery.com>
6910             Cesar Philippidis  <cesar@codesourcery.com>
6911
6912         PR target/83920
6913         * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
6914
6915 2018-01-19  Cesar Philippidis  <cesar@codesourcery.com>
6916
6917         PR target/83790
6918         * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
6919         spaces for function labels.
6920
6921 2018-01-19  Martin Liska  <mliska@suse.cz>
6922
6923         * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
6924         (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
6925         (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
6926         (PRED_OPCODE_POSITIVE): Change from 64 to 59.
6927         (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
6928         (PRED_CONST_RETURN): Change from 69 to 65.
6929         (PRED_NULL_RETURN): Change from 91 to 71.
6930         (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
6931         (PRED_LOOP_GUARD): Change from 66 to 73.
6932
6933 2018-01-19  Martin Liska  <mliska@suse.cz>
6934
6935         * predict.c (predict_insn_def): Add new assert.
6936         (struct branch_predictor): Change type to signed integer.
6937         (test_prediction_value_range): Amend test to cover
6938         PROB_UNINITIALIZED.
6939         * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
6940         (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
6941         (PRED_LOOP_ITERATIONS_MAX): Likewise.
6942         (PRED_LOOP_IV_COMPARE): Likewise.
6943         * predict.h (PROB_UNINITIALIZED): Define new constant.
6944
6945 2018-01-19  Martin Liska  <mliska@suse.cz>
6946
6947         * predict.c (dump_prediction): Add new format for
6948         analyze_brprob.py script which is enabled with -details
6949         suboption.
6950         * profile-count.h (precise_p): New function.
6951
6952 2018-01-19  Richard Sandiford  <richard.sandiford@linaro.org>
6953
6954         PR tree-optimization/83922
6955         * tree-vect-loop.c (vect_verify_full_masking): Return false if
6956         there are no statements that need masking.
6957         (vect_active_double_reduction_p): New function.
6958         (vect_analyze_loop_operations): Use it when handling phis that
6959         are not in the loop header.
6960
6961 2018-01-19  Richard Sandiford  <richard.sandiford@linaro.org>
6962
6963         PR tree-optimization/83914
6964         * tree-vect-loop.c (vectorizable_induction): Don't convert
6965         init_expr or apply the peeling adjustment for inductions
6966         that are nested within the vectorized loop.
6967
6968 2018-01-19  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
6969
6970         * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
6971         instead of NEG.
6972
6973 2018-01-18  Jakub Jelinek  <jakub@redhat.com>
6974
6975         PR sanitizer/81715
6976         PR testsuite/83882
6977         * function.h (gimplify_parameters): Add gimple_seq * argument.
6978         * function.c: Include gimple.h and options.h.
6979         (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
6980         for the added local temporaries if needed.
6981         * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
6982         if there are any parameter cleanups, wrap whole body into a
6983         try/finally with the cleanups.
6984
6985 2018-01-18  Wilco Dijkstra  <wdijkstr@arm.com>
6986
6987         PR target/82964
6988         * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
6989         Use GET_MODE_CLASS for scalar floating point.
6990
6991 2018-01-18  Jan Hubicka  <hubicka@ucw.cz>
6992
6993         PR ipa/82256
6994         patch by PaX Team
6995         * cgraphclones.c (cgraph_node::create_version_clone_with_body):
6996         Fix call of call_cgraph_insertion_hooks.
6997
6998 2018-01-18  Martin Sebor  <msebor@redhat.com>
6999
7000         * doc/invoke.texi (-Wclass-memaccess): Tweak text.
7001
7002 2018-01-18  Jan Hubicka  <hubicka@ucw.cz>
7003
7004         PR ipa/83619
7005         * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
7006         frequencies.
7007
7008 2018-01-18  Boris Kolpackov  <boris@codesynthesis.com>
7009
7010         PR other/70268
7011         * common.opt: (-ffile-prefix-map): New option.
7012         * opts.c (common_handle_option): Defer it.
7013         * opts-global.c (handle_common_deferred_options): Handle it.
7014         * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
7015         * file-prefix-map.h: New file.
7016         (remap_debug_filename, add_debug_prefix_map): ...here.
7017         (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
7018         * final.c (debug_prefix_map, add_debug_prefix_map
7019         remap_debug_filename): Move to...
7020         * file-prefix-map.c: New file.
7021         (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
7022         generalize, get rid of alloca(), use strrchr() instead of strchr().
7023         (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
7024         Implement in terms of add_prefix_map().
7025         (remap_macro_filename, remap_debug_filename): Implement in term of
7026         remap_filename().
7027         * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
7028         * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
7029         * dbxout.c: Include file-prefix-map.h.
7030         * varasm.c: Likewise.
7031         * vmsdbgout.c: Likewise.
7032         * xcoffout.c: Likewise.
7033         * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
7034         * doc/cppopts.texi (-fmacro-prefix-map): Document.
7035         * doc/invoke.texi (-ffile-prefix-map): Document.
7036         (-fdebug-prefix-map): Update description.
7037
7038 2018-01-18  Martin Liska  <mliska@suse.cz>
7039
7040         * config/i386/i386.c (indirect_thunk_name): Document that also
7041         lfence is emitted.
7042         (output_indirect_thunk): Document why both instructions
7043         (pause and lfence) are generated.
7044
7045 2018-01-18  Richard Biener  <rguenther@suse.de>
7046
7047         PR tree-optimization/83887
7048         * graphite-scop-detection.c
7049         (scop_detection::get_nearest_dom_with_single_entry): Remove.
7050         (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
7051         (scop_detection::merge_sese): Re-implement with a flood-fill
7052         algorithm that properly finds a SESE region if it exists.
7053
7054 2018-01-18  Jakub Jelinek  <jakub@redhat.com>
7055
7056         PR c/61240
7057         * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
7058         pointer_diff optimizations use view_convert instead of convert.
7059
7060 2018-01-17  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
7061
7062         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
7063         Generate different code for -mno-speculate-indirect-jumps.
7064         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
7065         (*call_indirect_aix<mode>): Disable for
7066         -mno-speculate-indirect-jumps.
7067         (*call_indirect_aix<mode>_nospec): New define_insn.
7068         (*call_value_indirect_aix<mode>): Disable for
7069         -mno-speculate-indirect-jumps.
7070         (*call_value_indirect_aix<mode>_nospec): New define_insn.
7071         (*sibcall_nonlocal_sysv<mode>): Generate different code for
7072         -mno-speculate-indirect-jumps.
7073         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
7074
7075 2018-01-17  Michael Meissner  <meissner@linux.vnet.ibm.com>
7076
7077         * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
7078         long double type, set the flags for noting the default long double
7079         type, even if we don't pass or return a long double type.
7080
7081 2018-01-17  Jan Hubicka  <hubicka@ucw.cz>
7082
7083         PR ipa/83051
7084         * ipa-inline.c (flatten_function): Do not overwrite final inlining
7085         failure.
7086
7087 2018-01-17  Will Schmidt  <will_schmidt@vnet.ibm.com>
7088
7089         * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
7090         support for merge[hl].
7091         (fold_mergehl_helper): New helper function.
7092         (tree-vector-builder.h): New #include for tree_vector_builder usage.
7093         * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
7094         (altivec_vmrglw_direct): Add xxmrglw insn.
7095
7096 2018-01-17  Andrew Waterman  <andrew@sifive.com>
7097
7098         * config/riscv/riscv.c (riscv_conditional_register_usage): If
7099         UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
7100
7101 2018-01-17  David Malcolm  <dmalcolm@redhat.com>
7102
7103         PR lto/83121
7104         * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
7105         call the lto_location_cache before reading the
7106         DECL_SOURCE_LOCATION of the types.
7107
7108 2018-01-17  Wilco Dijkstra  <wdijkstr@arm.com>
7109             Richard Sandiford  <richard.sandiford@linaro.org>
7110
7111         * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
7112         * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
7113         (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
7114         SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
7115         * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
7116         Add declaration.
7117         * config/aarch64/constraints.md (aarch64_movti_operand):
7118         Limit immediates.
7119         * config/aarch64/predicates.md (Uti): Add new constraint.
7120
7121 2018-01-17 Carl Love  <cel@us.ibm.com>
7122
7123         * config/rs6000/vsx.md (define_expand xl_len_r,
7124         define_expand stxvl, define_expand *stxvl): Add match_dup argument.
7125         (define_insn): Add, match_dup 1 argument to define_insn stxvll and
7126         lxvll.
7127         (define_expand, define_insn): Move the shift left from  the
7128         define_insn to the define_expand for lxvl and stxvl instructions.
7129         * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
7130         and XL_LEN_R definitions to PURE.
7131
7132 2018-01-17  Uros Bizjak  <ubizjak@gmail.com>
7133
7134         * config/i386/i386.c (indirect_thunk_name): Declare regno
7135         as unsigned int.  Compare regno with INVALID_REGNUM.
7136         (output_indirect_thunk): Ditto.
7137         (output_indirect_thunk_function): Ditto.
7138         (ix86_code_end): Declare regno as unsigned int.  Use INVALID_REGNUM
7139         in the call to output_indirect_thunk_function.
7140
7141 2018-01-17  Richard Sandiford  <richard.sandiford@linaro.org>
7142
7143         PR middle-end/83884
7144         * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
7145         rather than the size of inner_type to determine the stack slot size
7146         when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
7147
7148 2018-01-16  Sebastian Peryt  <sebastian.peryt@intel.com>
7149
7150         PR target/83546
7151         * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
7152         to PTA_SILVERMONT.
7153
7154 2018-01-16  Michael Meissner  <meissner@linux.vnet.ibm.com>
7155
7156         * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
7157         endian Linux systems to optionally enable multilibs for selecting
7158         the long double type if the user configured an explicit type.
7159         * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
7160         have no long double multilibs if not defined.
7161         * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
7162         warn if the user used -mabi={ieee,ibm}longdouble and we built
7163         multilibs for long double.
7164         * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
7165         appropriate multilib option.
7166         (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
7167         multilib options.
7168         * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
7169         for building long double multilibs.
7170         * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
7171
7172 2018-01-16  John David Anglin  <danglin@gcc.gnu.org>
7173
7174         * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
7175         copies.
7176
7177         * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
7178         64 bits.
7179         * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
7180         128 bits.
7181
7182         * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
7183         variables.
7184
7185         * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
7186         return value.
7187
7188 2018-01-16  Eric Botcazou  <ebotcazou@adacore.com>
7189
7190         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
7191         ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
7192
7193 2018-01-16  Kelvin Nilsen  <kelvin@gcc.gnu.org>
7194
7195         * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
7196         different rtl trees depending on TARGET_64BIT.
7197         (rs6000_gen_lvx): Likewise.
7198
7199 2018-01-16  Eric Botcazou  <ebotcazou@adacore.com>
7200
7201         * config/visium/visium.md (nop): Tweak comment.
7202         (hazard_nop): Likewise.
7203
7204 2018-01-16  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
7205
7206         * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
7207         -mspeculate-indirect-jumps.
7208         * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
7209         for -mno-speculate-indirect-jumps.
7210         (*call_indirect_elfv2<mode>_nospec): New define_insn.
7211         (*call_value_indirect_elfv2<mode>): Disable for
7212         -mno-speculate-indirect-jumps.
7213         (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
7214         (indirect_jump): Emit different RTL for
7215         -mno-speculate-indirect-jumps.
7216         (*indirect_jump<mode>): Disable for
7217         -mno-speculate-indirect-jumps.
7218         (*indirect_jump<mode>_nospec): New define_insn.
7219         (tablejump): Emit different RTL for
7220         -mno-speculate-indirect-jumps.
7221         (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
7222         (tablejumpsi_nospec): New define_expand.
7223         (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
7224         (tablejumpdi_nospec): New define_expand.
7225         (*tablejump<mode>_internal1): Disable for
7226         -mno-speculate-indirect-jumps.
7227         (*tablejump<mode>_internal1_nospec): New define_insn.
7228         * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
7229         option.
7230
7231 2018-01-16  Artyom Skrobov tyomitch@gmail.com
7232
7233         * caller-save.c (insert_save): Drop unnecessary parameter.  All
7234         callers updated.
7235
7236 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
7237             Richard Biener  <rguenth@suse.de>
7238
7239         PR libgomp/83590
7240         * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
7241         return early, inline manually is_gimple_sizepos.  Make sure if we
7242         call gimplify_expr we don't end up with a gimple constant.
7243         * tree.c (variably_modified_type_p): Don't return true for
7244         is_gimple_constant (_t).  Inline manually is_gimple_sizepos.
7245         * gimplify.h (is_gimple_sizepos): Remove.
7246
7247 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
7248
7249         PR tree-optimization/83857
7250         * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
7251         vectorizable_live_operation for pure SLP statements.
7252         (vectorizable_live_operation): Handle PHIs.
7253
7254 2018-01-16  Richard Biener  <rguenther@suse.de>
7255
7256         PR tree-optimization/83867
7257         * tree-vect-stmts.c (vect_transform_stmt): Precompute
7258         nested_in_vect_loop_p since the scalar stmt may get invalidated.
7259
7260 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
7261
7262         PR c/83844
7263         * stor-layout.c (handle_warn_if_not_align): Use byte_position and
7264         multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
7265         If off is not INTEGER_CST, issue a may not be aligned warning
7266         rather than isn't aligned.  Use isn%'t rather than isn't.
7267         * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
7268         into MULT_EXPR.
7269         <case MULT_EXPR>: Improve the case when bottom and one of the
7270         MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
7271         operand, in that case check if the other operand is multiple of
7272         bottom divided by the INTEGER_CST operand.
7273
7274 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
7275
7276         PR target/83858
7277         * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
7278         * config/pa/pa-protos.h (pa_function_arg_size): Declare.
7279         * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
7280         pa_function_arg_size instead of FUNCTION_ARG_SIZE.
7281         * config/pa/pa.c (pa_function_arg_advance): Likewise.
7282         (pa_function_arg, pa_arg_partial_bytes): Likewise.
7283         (pa_function_arg_size): New function.
7284
7285 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
7286
7287         * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
7288         in a separate statement.
7289
7290 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
7291
7292         PR tree-optimization/83847
7293         * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
7294         group gathers and scatters.
7295
7296 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
7297
7298         PR rtl-optimization/86620
7299         * params.def (max-sched-ready-insns): Bump minimum value to 1.
7300
7301         PR rtl-optimization/83213
7302         * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
7303         to last if both are JUMP_INSNs.
7304
7305         PR tree-optimization/83843
7306         * gimple-ssa-store-merging.c
7307         (imm_store_chain_info::output_merged_store): Handle bit_not_p on
7308         store_immediate_info for bswap/nop orig_stores.
7309
7310 2018-01-15  Andrew Waterman  <andrew@sifive.com>
7311
7312         * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
7313         !TARGET_MUL.
7314         <UDIV>: Increase cost if !TARGET_DIV.
7315
7316 2018-01-15  Segher Boessenkool  <segher@kernel.crashing.org>
7317
7318         * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
7319         (define_attr "cr_logical_3op"): New.
7320         (cceq_ior_compare): Adjust.
7321         (cceq_ior_compare_complement): Adjust.
7322         (*cceq_rev_compare): Adjust.
7323         * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
7324         (is_cracked_insn): Adjust.
7325         (insn_must_be_first_in_group): Adjust.
7326         * config/rs6000/40x.md: Adjust.
7327         * config/rs6000/440.md: Adjust.
7328         * config/rs6000/476.md: Adjust.
7329         * config/rs6000/601.md: Adjust.
7330         * config/rs6000/603.md: Adjust.
7331         * config/rs6000/6xx.md: Adjust.
7332         * config/rs6000/7450.md: Adjust.
7333         * config/rs6000/7xx.md: Adjust.
7334         * config/rs6000/8540.md: Adjust.
7335         * config/rs6000/cell.md: Adjust.
7336         * config/rs6000/e300c2c3.md: Adjust.
7337         * config/rs6000/e500mc.md: Adjust.
7338         * config/rs6000/e500mc64.md: Adjust.
7339         * config/rs6000/e5500.md: Adjust.
7340         * config/rs6000/e6500.md: Adjust.
7341         * config/rs6000/mpc.md: Adjust.
7342         * config/rs6000/power4.md: Adjust.
7343         * config/rs6000/power5.md: Adjust.
7344         * config/rs6000/power6.md: Adjust.
7345         * config/rs6000/power7.md: Adjust.
7346         * config/rs6000/power8.md: Adjust.
7347         * config/rs6000/power9.md: Adjust.
7348         * config/rs6000/rs64.md: Adjust.
7349         * config/rs6000/titan.md: Adjust.
7350
7351 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
7352
7353         * config/i386/predicates.md (indirect_branch_operand): Rewrite
7354         ix86_indirect_branch_register logic.
7355
7356 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
7357
7358         * config/i386/constraints.md (Bs): Update
7359         ix86_indirect_branch_register check.  Don't check
7360         ix86_indirect_branch_register with GOT_memory_operand.
7361         (Bw): Likewise.
7362         * config/i386/predicates.md (GOT_memory_operand): Don't check
7363         ix86_indirect_branch_register here.
7364         (GOT32_symbol_operand): Likewise.
7365
7366 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
7367
7368         * config/i386/predicates.md (constant_call_address_operand):
7369         Rewrite ix86_indirect_branch_register logic.
7370         (sibcall_insn_operand): Likewise.
7371
7372 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
7373
7374         * config/i386/constraints.md (Bs): Replace
7375         ix86_indirect_branch_thunk_register with
7376         ix86_indirect_branch_register.
7377         (Bw): Likewise.
7378         * config/i386/i386.md (indirect_jump): Likewise.
7379         (tablejump): Likewise.
7380         (*sibcall_memory): Likewise.
7381         (*sibcall_value_memory): Likewise.
7382         Peepholes of indirect call and jump via memory: Likewise.
7383         * config/i386/i386.opt: Likewise.
7384         * config/i386/predicates.md (indirect_branch_operand): Likewise.
7385         (GOT_memory_operand): Likewise.
7386         (call_insn_operand): Likewise.
7387         (sibcall_insn_operand): Likewise.
7388         (GOT32_symbol_operand): Likewise.
7389
7390 2018-01-15  Jakub Jelinek  <jakub@redhat.com>
7391
7392         PR middle-end/83837
7393         * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
7394         type rather than type addr's type points to.
7395         (expand_omp_atomic_mutex): Likewise.
7396         (expand_omp_atomic): Likewise.
7397
7398 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
7399
7400         PR target/83839
7401         * config/i386/i386.c (output_indirect_thunk_function): Use
7402         ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
7403         for  __x86_return_thunk.
7404
7405 2018-01-15  Richard Biener  <rguenther@suse.de>
7406
7407         PR middle-end/83850
7408         * expmed.c (extract_bit_field_1): Fix typo.
7409
7410 2018-01-15  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
7411
7412         PR target/83687
7413         * config/arm/iterators.md (VF): New mode iterator.
7414         * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
7415         Remove integer-related logic from pattern.
7416         (neon_vabd<mode>_3): Likewise.
7417
7418 2018-01-15  Jakub Jelinek  <jakub@redhat.com>
7419
7420         PR middle-end/82694
7421         * common.opt (fstrict-overflow): No longer an alias.
7422         (fwrapv-pointer): New option.
7423         * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
7424         also for pointer types based on flag_wrapv_pointer.
7425         * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
7426         opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
7427         opts->x_flag_wrapv got set.
7428         * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
7429         changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
7430         POINTER_TYPE_OVERFLOW_UNDEFINED.
7431         * match.pd: Likewise in address comparison pattern.
7432         * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
7433
7434 2018-01-15  Richard Biener  <rguenther@suse.de>
7435
7436         PR lto/83804
7437         * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
7438         from TYPE_FIELDS.  Free TYPE_BINFO if not used by devirtualization.
7439         Reset type names to their identifier if their TYPE_DECL doesn't
7440         have linkage (and thus is used for ODR and devirt).
7441         (save_debug_info_for_decl): Remove.
7442         (save_debug_info_for_type): Likewise.
7443         (add_tree_to_fld_list): Adjust.
7444         * tree-pretty-print.c (dump_generic_node): Make dumping of
7445         type names more robust.
7446
7447 2018-01-15  Richard Biener  <rguenther@suse.de>
7448
7449         * BASE-VER: Bump to 8.0.1.
7450
7451 2018-01-14  Martin Sebor  <msebor@redhat.com>
7452
7453         PR other/83508
7454         * builtins.c (check_access): Avoid warning when the no-warning bit
7455         is set.
7456
7457 2018-01-14  Cory Fields  <cory-nospam-@coryfields.com>
7458
7459         * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
7460         * ira-color (allocno_hard_regs_compare): Likewise.
7461
7462 2018-01-14  Nathan Rossi  <nathan@nathanrossi.com>
7463
7464         PR target/83013
7465         * config/microblaze/microblaze.c (microblaze_asm_output_ident):
7466         Use .pushsection/.popsection.
7467
7468 2018-01-14  Martin Sebor  <msebor@redhat.com>
7469
7470         PR c++/81327
7471         * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
7472
7473 2018-01-14  Jakub Jelinek  <jakub@redhat.com>
7474
7475         * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
7476         entry from extra_headers.
7477         (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
7478         extra_headers, make the list bitwise identical to the i?86-*-* one.
7479
7480 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
7481
7482         * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
7483         -mcmodel=large with -mindirect-branch=thunk,
7484         -mindirect-branch=thunk-extern, -mfunction-return=thunk and
7485         -mfunction-return=thunk-extern.
7486         * doc/invoke.texi: Document -mcmodel=large is incompatible with
7487         -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
7488         -mfunction-return=thunk and -mfunction-return=thunk-extern.
7489
7490 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
7491
7492         * config/i386/i386.c (print_reg): Print the name of the full
7493         integer register without '%'.
7494         (ix86_print_operand): Handle 'V'.
7495          * doc/extend.texi: Document 'V' modifier.
7496
7497 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
7498
7499         * config/i386/constraints.md (Bs): Disallow memory operand for
7500         -mindirect-branch-register.
7501         (Bw): Likewise.
7502         * config/i386/predicates.md (indirect_branch_operand): Likewise.
7503         (GOT_memory_operand): Likewise.
7504         (call_insn_operand): Likewise.
7505         (sibcall_insn_operand): Likewise.
7506         (GOT32_symbol_operand): Likewise.
7507         * config/i386/i386.md (indirect_jump): Call convert_memory_address
7508         for -mindirect-branch-register.
7509         (tablejump): Likewise.
7510         (*sibcall_memory): Likewise.
7511         (*sibcall_value_memory): Likewise.
7512         Disallow peepholes of indirect call and jump via memory for
7513         -mindirect-branch-register.
7514         (*call_pop): Replace m with Bw.
7515         (*call_value_pop): Likewise.
7516         (*sibcall_pop_memory): Replace m with Bs.
7517         * config/i386/i386.opt (mindirect-branch-register): New option.
7518         * doc/invoke.texi: Document -mindirect-branch-register option.
7519
7520 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
7521
7522         * config/i386/i386-protos.h (ix86_output_function_return): New.
7523         * config/i386/i386.c (ix86_set_indirect_branch_type): Also
7524         set function_return_type.
7525         (indirect_thunk_name): Add ret_p to indicate thunk for function
7526         return.
7527         (output_indirect_thunk_function): Pass false to
7528         indirect_thunk_name.
7529         (ix86_output_indirect_branch_via_reg): Likewise.
7530         (ix86_output_indirect_branch_via_push): Likewise.
7531         (output_indirect_thunk_function): Create alias for function
7532         return thunk if regno < 0.
7533         (ix86_output_function_return): New function.
7534         (ix86_handle_fndecl_attribute): Handle function_return.
7535         (ix86_attribute_table): Add function_return.
7536         * config/i386/i386.h (machine_function): Add
7537         function_return_type.
7538         * config/i386/i386.md (simple_return_internal): Use
7539         ix86_output_function_return.
7540         (simple_return_internal_long): Likewise.
7541         * config/i386/i386.opt (mfunction-return=): New option.
7542         (indirect_branch): Mention -mfunction-return=.
7543         * doc/extend.texi: Document function_return function attribute.
7544         * doc/invoke.texi: Document -mfunction-return= option.
7545
7546 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
7547
7548         * config/i386/i386-opts.h (indirect_branch): New.
7549         * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
7550         * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
7551         with local indirect jump when converting indirect call and jump.
7552         (ix86_set_indirect_branch_type): New.
7553         (ix86_set_current_function): Call ix86_set_indirect_branch_type.
7554         (indirectlabelno): New.
7555         (indirect_thunk_needed): Likewise.
7556         (indirect_thunk_bnd_needed): Likewise.
7557         (indirect_thunks_used): Likewise.
7558         (indirect_thunks_bnd_used): Likewise.
7559         (INDIRECT_LABEL): Likewise.
7560         (indirect_thunk_name): Likewise.
7561         (output_indirect_thunk): Likewise.
7562         (output_indirect_thunk_function): Likewise.
7563         (ix86_output_indirect_branch_via_reg): Likewise.
7564         (ix86_output_indirect_branch_via_push): Likewise.
7565         (ix86_output_indirect_branch): Likewise.
7566         (ix86_output_indirect_jmp): Likewise.
7567         (ix86_code_end): Call output_indirect_thunk_function if needed.
7568         (ix86_output_call_insn): Call ix86_output_indirect_branch if
7569         needed.
7570         (ix86_handle_fndecl_attribute): Handle indirect_branch.
7571         (ix86_attribute_table): Add indirect_branch.
7572         * config/i386/i386.h (machine_function): Add indirect_branch_type
7573         and has_local_indirect_jump.
7574         * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
7575         to true.
7576         (tablejump): Likewise.
7577         (*indirect_jump): Use ix86_output_indirect_jmp.
7578         (*tablejump_1): Likewise.
7579         (simple_return_indirect_internal): Likewise.
7580         * config/i386/i386.opt (mindirect-branch=): New option.
7581         (indirect_branch): New.
7582         (keep): Likewise.
7583         (thunk): Likewise.
7584         (thunk-inline): Likewise.
7585         (thunk-extern): Likewise.
7586         * doc/extend.texi: Document indirect_branch function attribute.
7587         * doc/invoke.texi: Document -mindirect-branch= option.
7588
7589 2018-01-14  Jan Hubicka  <hubicka@ucw.cz>
7590
7591         PR ipa/83051
7592         * ipa-inline.c (edge_badness): Tolerate roundoff errors.
7593
7594 2018-01-14  Richard Sandiford  <richard.sandiford@linaro.org>
7595
7596         * ipa-inline.c (want_inline_small_function_p): Return false if
7597         inlining has already failed with CIF_FINAL_ERROR.
7598         (update_caller_keys): Call want_inline_small_function_p before
7599         can_inline_edge_p.
7600         (update_callee_keys): Likewise.
7601
7602 2018-01-10  Kelvin Nilsen  <kelvin@gcc.gnu.org>
7603
7604         * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
7605         New function.
7606         (rs6000_quadword_masked_address_p): Likewise.
7607         (quad_aligned_load_p): Likewise.
7608         (quad_aligned_store_p): Likewise.
7609         (const_load_sequence_p): Add comment to describe the outer-most loop.
7610         (mimic_memory_attributes_and_flags): New function.
7611         (rs6000_gen_stvx): Likewise.
7612         (replace_swapped_aligned_store): Likewise.
7613         (rs6000_gen_lvx): Likewise.
7614         (replace_swapped_aligned_load): Likewise.
7615         (replace_swapped_load_constant): Capitalize argument name in
7616         comment describing this function.
7617         (rs6000_analyze_swaps): Add a third pass to search for vector loads
7618         and stores that access quad-word aligned addresses and replace
7619         with stvx or lvx instructions when appropriate.
7620         * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
7621         New function prototype.
7622         (rs6000_quadword_masked_address_p): Likewise.
7623         (rs6000_gen_lvx): Likewise.
7624         (rs6000_gen_stvx): Likewise.
7625         * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
7626         VSX_D (V2DF, V2DI), modify this split to select lvx instruction
7627         when memory address is aligned.
7628         (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
7629         this split to select lvx instruction when memory address is aligned.
7630         (*vsx_le_perm_load_v8hi): Modify this split to select lvx
7631         instruction when memory address is aligned.
7632         (*vsx_le_perm_load_v16qi): Likewise.
7633         (four unnamed splitters): Modify to select the stvx instruction
7634         when memory is aligned.
7635
7636 2018-01-13  Jan Hubicka  <hubicka@ucw.cz>
7637
7638         * predict.c (determine_unlikely_bbs): Handle correctly BBs
7639         which appears in the queue multiple times.
7640
7641 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7642             Alan Hayward  <alan.hayward@arm.com>
7643             David Sherwood  <david.sherwood@arm.com>
7644
7645         * tree-vectorizer.h (vec_lower_bound): New structure.
7646         (_loop_vec_info): Add check_nonzero and lower_bounds.
7647         (LOOP_VINFO_CHECK_NONZERO): New macro.
7648         (LOOP_VINFO_LOWER_BOUNDS): Likewise.
7649         (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
7650         * tree-data-ref.h (dr_with_seg_len): Add access_size and align
7651         fields.  Make seg_len the distance travelled, not including the
7652         access size.
7653         (dr_direction_indicator): Declare.
7654         (dr_zero_step_indicator): Likewise.
7655         (dr_known_forward_stride_p): Likewise.
7656         * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
7657         tree-ssanames.h.
7658         (runtime_alias_check_p): Allow runtime alias checks with
7659         variable strides.
7660         (operator ==): Compare access_size and align.
7661         (prune_runtime_alias_test_list): Rework for new distinction between
7662         the access_size and seg_len.
7663         (create_intersect_range_checks_index): Likewise.  Cope with polynomial
7664         segment lengths.
7665         (get_segment_min_max): New function.
7666         (create_intersect_range_checks): Use it.
7667         (dr_step_indicator): New function.
7668         (dr_direction_indicator): Likewise.
7669         (dr_zero_step_indicator): Likewise.
7670         (dr_known_forward_stride_p): Likewise.
7671         * tree-loop-distribution.c (data_ref_segment_size): Return
7672         DR_STEP * (niters - 1).
7673         (compute_alias_check_pairs): Update call to the dr_with_seg_len
7674         constructor.
7675         * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
7676         (vect_preserves_scalar_order_p): New function, split out from...
7677         (vect_analyze_data_ref_dependence): ...here.  Check for zero steps.
7678         (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
7679         (vect_vfa_access_size): New function.
7680         (vect_vfa_align): Likewise.
7681         (vect_compile_time_alias): Take access_size_a and access_b arguments.
7682         (dump_lower_bound): New function.
7683         (vect_check_lower_bound): Likewise.
7684         (vect_small_gap_p): Likewise.
7685         (vectorizable_with_step_bound_p): Likewise.
7686         (vect_prune_runtime_alias_test_list): Ignore cross-iteration
7687         depencies if the vectorization factor is 1.  Convert the checks
7688         for nonzero steps into checks on the bounds of DR_STEP.  Try using
7689         a bunds check for variable steps if the minimum required step is
7690         relatively small. Update calls to the dr_with_seg_len
7691         constructor and to vect_compile_time_alias.
7692         * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
7693         function.
7694         (vect_loop_versioning): Call it.
7695         * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
7696         when retrying.
7697         (vect_estimate_min_profitable_iters): Account for any bounds checks.
7698
7699 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7700             Alan Hayward  <alan.hayward@arm.com>
7701             David Sherwood  <david.sherwood@arm.com>
7702
7703         * doc/sourcebuild.texi (vect_scatter_store): Document.
7704         * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
7705         optabs.
7706         * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
7707         Document.
7708         * genopinit.c (main): Add supports_vec_scatter_store and
7709         supports_vec_scatter_store_cached to target_optabs.
7710         * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
7711         IFN_MASK_SCATTER_STORE.
7712         * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
7713         functions.
7714         * internal-fn.h (internal_store_fn_p): Declare.
7715         (internal_fn_stored_value_index): Likewise.
7716         * internal-fn.c (scatter_store_direct): New macro.
7717         (expand_scatter_store_optab_fn): New function.
7718         (direct_scatter_store_optab_supported_p): New macro.
7719         (internal_store_fn_p): New function.
7720         (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
7721         IFN_MASK_SCATTER_STORE.
7722         (internal_fn_mask_index): Likewise.
7723         (internal_fn_stored_value_index): New function.
7724         (internal_gather_scatter_fn_supported_p): Adjust operand numbers
7725         for scatter stores.
7726         * optabs-query.h (supports_vec_scatter_store_p): Declare.
7727         * optabs-query.c (supports_vec_scatter_store_p): New function.
7728         * tree-vectorizer.h (vect_get_store_rhs): Declare.
7729         * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
7730         true for scatter stores.
7731         (vect_gather_scatter_fn_p): Handle scatter stores too.
7732         (vect_check_gather_scatter): Consider using scatter stores if
7733         supports_vec_scatter_store_p.
7734         * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
7735         scatter stores too.
7736         * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
7737         internal_fn_stored_value_index.
7738         (check_load_store_masking): Handle scatter stores too.
7739         (vect_get_store_rhs): Make public.
7740         (vectorizable_call): Use internal_store_fn_p.
7741         (vectorizable_store): Handle scatter store internal functions.
7742         (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
7743         when deciding whether the end of the group has been reached.
7744         * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
7745         * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
7746         (mask_scatter_store<mode>): New insns.
7747
7748 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7749             Alan Hayward  <alan.hayward@arm.com>
7750             David Sherwood  <david.sherwood@arm.com>
7751
7752         * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
7753         * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
7754         * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
7755         function.
7756         (vect_use_strided_gather_scatters_p): Take a masked_p argument.
7757         Use vect_truncate_gather_scatter_offset if we can't treat the
7758         operation as a normal gather load or scatter store.
7759         (get_group_load_store_type): Take the gather_scatter_info
7760         as argument.  Try using a gather load or scatter store for
7761         single-element groups.
7762         (get_load_store_type): Update calls to get_group_load_store_type
7763         and vect_use_strided_gather_scatters_p.
7764
7765 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7766             Alan Hayward  <alan.hayward@arm.com>
7767             David Sherwood  <david.sherwood@arm.com>
7768
7769         * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
7770         optional tree argument.
7771         * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
7772         null target hooks.
7773         (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
7774         but continue to use the current value as a fallback.
7775         (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
7776         to compare the updates.
7777         * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
7778         (get_load_store_type): Use it when handling a strided access.
7779         (vect_get_strided_load_store_ops): New function.
7780         (vect_get_data_ptr_increment): Likewise.
7781         (vectorizable_load): Handle strided gather loads.  Always pass
7782         a step to vect_create_data_ref_ptr and bump_vector_ptr.
7783
7784 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7785             Alan Hayward  <alan.hayward@arm.com>
7786             David Sherwood  <david.sherwood@arm.com>
7787
7788         * doc/md.texi (gather_load@var{m}): Document.
7789         (mask_gather_load@var{m}): Likewise.
7790         * genopinit.c (main): Add supports_vec_gather_load and
7791         supports_vec_gather_load_cached to target_optabs.
7792         * optabs-tree.c (init_tree_optimization_optabs): Use
7793         ggc_cleared_alloc to allocate target_optabs.
7794         * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
7795         * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
7796         functions.
7797         * internal-fn.h (internal_load_fn_p): Declare.
7798         (internal_gather_scatter_fn_p): Likewise.
7799         (internal_fn_mask_index): Likewise.
7800         (internal_gather_scatter_fn_supported_p): Likewise.
7801         * internal-fn.c (gather_load_direct): New macro.
7802         (expand_gather_load_optab_fn): New function.
7803         (direct_gather_load_optab_supported_p): New macro.
7804         (direct_internal_fn_optab): New function.
7805         (internal_load_fn_p): Likewise.
7806         (internal_gather_scatter_fn_p): Likewise.
7807         (internal_fn_mask_index): Likewise.
7808         (internal_gather_scatter_fn_supported_p): Likewise.
7809         * optabs-query.c (supports_at_least_one_mode_p): New function.
7810         (supports_vec_gather_load_p): Likewise.
7811         * optabs-query.h (supports_vec_gather_load_p): Declare.
7812         * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
7813         and memory_type field.
7814         (NUM_PATTERNS): Bump to 15.
7815         * tree-vect-data-refs.c: Include internal-fn.h.
7816         (vect_gather_scatter_fn_p): New function.
7817         (vect_describe_gather_scatter_call): Likewise.
7818         (vect_check_gather_scatter): Try using internal functions for
7819         gather loads.  Recognize existing calls to a gather load function.
7820         (vect_analyze_data_refs): Consider using gather loads if
7821         supports_vec_gather_load_p.
7822         * tree-vect-patterns.c (vect_get_load_store_mask): New function.
7823         (vect_get_gather_scatter_offset_type): Likewise.
7824         (vect_convert_mask_for_vectype): Likewise.
7825         (vect_add_conversion_to_patterm): Likewise.
7826         (vect_try_gather_scatter_pattern): Likewise.
7827         (vect_recog_gather_scatter_pattern): New pattern recognizer.
7828         (vect_vect_recog_func_ptrs): Add it.
7829         * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
7830         internal_fn_mask_index and internal_gather_scatter_fn_p.
7831         (check_load_store_masking): Take the gather_scatter_info as an
7832         argument and handle gather loads.
7833         (vect_get_gather_scatter_ops): New function.
7834         (vectorizable_call): Check internal_load_fn_p.
7835         (vectorizable_load): Likewise.  Handle gather load internal
7836         functions.
7837         (vectorizable_store): Update call to check_load_store_masking.
7838         * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
7839         * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
7840         * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
7841         (aarch64_gather_scale_operand_d): New predicates.
7842         * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
7843         (mask_gather_load<mode>): New insns.
7844
7845 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7846             Alan Hayward  <alan.hayward@arm.com>
7847             David Sherwood  <david.sherwood@arm.com>
7848
7849         * optabs.def (fold_left_plus_optab): New optab.
7850         * doc/md.texi (fold_left_plus_@var{m}): Document.
7851         * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
7852         * internal-fn.c (fold_left_direct): Define.
7853         (expand_fold_left_optab_fn): Likewise.
7854         (direct_fold_left_optab_supported_p): Likewise.
7855         * fold-const-call.c (fold_const_fold_left): New function.
7856         (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
7857         * tree-parloops.c (valid_reduction_p): New function.
7858         (gather_scalar_reductions): Use it.
7859         * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
7860         (vect_finish_replace_stmt): Declare.
7861         * tree-vect-loop.c (fold_left_reduction_fn): New function.
7862         (needs_fold_left_reduction_p): New function, split out from...
7863         (vect_is_simple_reduction): ...here.  Accept reductions that
7864         forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
7865         (vect_force_simple_reduction): Also store the reduction type in
7866         the assignment's STMT_VINFO_REDUC_TYPE.
7867         (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
7868         (merge_with_identity): New function.
7869         (vect_expand_fold_left): Likewise.
7870         (vectorize_fold_left_reduction): Likewise.
7871         (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION.  Leave the
7872         scalar phi in place for it.  Check for target support and reject
7873         cases that would reassociate the operation.  Defer the transform
7874         phase to vectorize_fold_left_reduction.
7875         * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
7876         * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
7877         (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
7878
7879 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7880
7881         * tree-if-conv.c (predicate_mem_writes): Remove redundant
7882         call to ifc_temp_var.
7883
7884 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7885             Alan Hayward  <alan.hayward@arm.com>
7886             David Sherwood  <david.sherwood@arm.com>
7887
7888         * target.def (legitimize_address_displacement): Take the original
7889         offset as a poly_int.
7890         * targhooks.h (default_legitimize_address_displacement): Update
7891         accordingly.
7892         * targhooks.c (default_legitimize_address_displacement): Likewise.
7893         * doc/tm.texi: Regenerate.
7894         * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
7895         as an argument, moving assert of ad->disp == ad->disp_term to...
7896         (process_address_1): ...here.  Update calls to base_plus_disp_to_reg.
7897         Try calling targetm.legitimize_address_displacement before expanding
7898         the address rather than afterwards, and adjust for the new interface.
7899         * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
7900         Match the new hook interface.  Handle SVE addresses.
7901         * config/sh/sh.c (sh_legitimize_address_displacement): Make the
7902         new hook interface.
7903
7904 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7905
7906         * Makefile.in (OBJS): Add early-remat.o.
7907         * target.def (select_early_remat_modes): New hook.
7908         * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
7909         * doc/tm.texi: Regenerate.
7910         * targhooks.h (default_select_early_remat_modes): Declare.
7911         * targhooks.c (default_select_early_remat_modes): New function.
7912         * timevar.def (TV_EARLY_REMAT): New timevar.
7913         * passes.def (pass_early_remat): New pass.
7914         * tree-pass.h (make_pass_early_remat): Declare.
7915         * early-remat.c: New file.
7916         * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
7917         function.
7918         (TARGET_SELECT_EARLY_REMAT_MODES): Define.
7919
7920 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7921             Alan Hayward  <alan.hayward@arm.com>
7922             David Sherwood  <david.sherwood@arm.com>
7923
7924         * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
7925         vfm1 with a bound_epilog parameter.
7926         (vect_do_peeling): Update calls accordingly, and move the prologue
7927         call earlier in the function.  Treat the base bound_epilog as 0 for
7928         fully-masked loops and retain vf - 1 for other loops.  Add 1 to
7929         this base when peeling for gaps.
7930         * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
7931         with fully-masked loops.
7932         (vect_estimate_min_profitable_iters): Handle the single peeled
7933         iteration in that case.
7934
7935 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7936             Alan Hayward  <alan.hayward@arm.com>
7937             David Sherwood  <david.sherwood@arm.com>
7938
7939         * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
7940         single-element interleaving even if the size is not a power of 2.
7941         * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
7942         accesses for single-element interleaving if the group size is
7943         not a power of 2.
7944
7945 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7946             Alan Hayward  <alan.hayward@arm.com>
7947             David Sherwood  <david.sherwood@arm.com>
7948
7949         * doc/md.texi (fold_extract_last_@var{m}): Document.
7950         * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
7951         * optabs.def (fold_extract_last_optab): New optab.
7952         * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
7953         * internal-fn.c (fold_extract_direct): New macro.
7954         (expand_fold_extract_optab_fn): Likewise.
7955         (direct_fold_extract_optab_supported_p): Likewise.
7956         * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
7957         * tree-vect-loop.c (vect_model_reduction_cost): Handle
7958         EXTRACT_LAST_REDUCTION.
7959         (get_initial_def_for_reduction): Do not create an initial vector
7960         for EXTRACT_LAST_REDUCTION reductions.
7961         (vectorizable_reduction): Leave the scalar phi in place for
7962         EXTRACT_LAST_REDUCTIONs.  Try using EXTRACT_LAST_REDUCTION
7963         ahead of INTEGER_INDUC_COND_REDUCTION.  Do not check for an
7964         epilogue code for EXTRACT_LAST_REDUCTION and defer the
7965         transform phase to vectorizable_condition.
7966         * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
7967         split out from...
7968         (vect_finish_stmt_generation): ...here.
7969         (vect_finish_replace_stmt): New function.
7970         (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
7971         * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
7972         pattern.
7973         * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
7974
7975 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7976             Alan Hayward  <alan.hayward@arm.com>
7977             David Sherwood  <david.sherwood@arm.com>
7978
7979         * doc/md.texi (extract_last_@var{m}): Document.
7980         * optabs.def (extract_last_optab): New optab.
7981         * internal-fn.def (EXTRACT_LAST): New internal function.
7982         * internal-fn.c (cond_unary_direct): New macro.
7983         (expand_cond_unary_optab_fn): Likewise.
7984         (direct_cond_unary_optab_supported_p): Likewise.
7985         * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
7986         loops using EXTRACT_LAST.
7987         * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
7988         (extract_last_<mode>): ...this optab.
7989         (vec_extract<mode><Vel>): Update accordingly.
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         * target.def (empty_mask_is_expensive): New hook.
7996         * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
7997         * doc/tm.texi: Regenerate.
7998         * targhooks.h (default_empty_mask_is_expensive): Declare.
7999         * targhooks.c (default_empty_mask_is_expensive): New function.
8000         * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
8001         if the target says that empty masks are expensive.
8002         * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
8003         New function.
8004         (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
8005
8006 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8007             Alan Hayward  <alan.hayward@arm.com>
8008             David Sherwood  <david.sherwood@arm.com>
8009
8010         * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
8011         (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
8012         (vect_use_loop_mask_for_alignment_p): New function.
8013         (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
8014         * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
8015         niters_skip argument.  Make sure that the first niters_skip elements
8016         of the first iteration are inactive.
8017         (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
8018         Update call to vect_set_loop_masks_directly.
8019         (get_misalign_in_elems): New function, split out from...
8020         (vect_gen_prolog_loop_niters): ...here.
8021         (vect_update_init_of_dr): Take a code argument that specifies whether
8022         the adjustment should be added or subtracted.
8023         (vect_update_init_of_drs): Likewise.
8024         (vect_prepare_for_masked_peels): New function.
8025         (vect_do_peeling): Skip prologue peeling if we're using a mask
8026         instead.  Update call to vect_update_inits_of_drs.
8027         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
8028         mask_skip_niters.
8029         (vect_analyze_loop_2): Allow fully-masked loops with peeling for
8030         alignment.  Do not include the number of peeled iterations in
8031         the minimum threshold in that case.
8032         (vectorizable_induction): Adjust the start value down by
8033         LOOP_VINFO_MASK_SKIP_NITERS iterations.
8034         (vect_transform_loop): Call vect_prepare_for_masked_peels.
8035         Take the number of skipped iterations into account when calculating
8036         the loop bounds.
8037         * tree-vect-stmts.c (vect_gen_while_not): New function.
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         * doc/sourcebuild.texi (vect_fully_masked): Document.
8044         * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
8045         default value to 0.
8046         * tree-vect-loop.c (vect_analyze_loop_costing): New function,
8047         split out from...
8048         (vect_analyze_loop_2): ...here. Don't check the vectorization
8049         factor against the number of loop iterations if the loop is
8050         fully-masked.
8051
8052 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8053             Alan Hayward  <alan.hayward@arm.com>
8054             David Sherwood  <david.sherwood@arm.com>
8055
8056         * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
8057         (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
8058         (dump_groups): Update accordingly.
8059         (iv_use::mem_type): New member variable.
8060         (address_p): New function.
8061         (record_use): Add a mem_type argument and initialize the new
8062         mem_type field.
8063         (record_group_use): Add a mem_type argument.  Use address_p.
8064         Remove obsolete null checks of base_object.  Update call to record_use.
8065         (find_interesting_uses_op): Update call to record_group_use.
8066         (find_interesting_uses_cond): Likewise.
8067         (find_interesting_uses_address): Likewise.
8068         (get_mem_type_for_internal_fn): New function.
8069         (find_address_like_use): Likewise.
8070         (find_interesting_uses_stmt): Try find_address_like_use before
8071         calling find_interesting_uses_op.
8072         (addr_offset_valid_p): Use the iv mem_type field as the type
8073         of the addressed memory.
8074         (add_autoinc_candidates): Likewise.
8075         (get_address_cost): Likewise.
8076         (split_small_address_groups_p): Use address_p.
8077         (split_address_groups): Likewise.
8078         (add_iv_candidate_for_use): Likewise.
8079         (autoinc_possible_for_pair): Likewise.
8080         (rewrite_groups): Likewise.
8081         (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
8082         (determine_group_iv_cost): Update after split of USE_ADDRESS.
8083         (get_alias_ptr_type_for_ptr_address): New function.
8084         (rewrite_use_address): Rewrite address uses in calls that were
8085         identified by find_address_like_use.
8086
8087 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8088             Alan Hayward  <alan.hayward@arm.com>
8089             David Sherwood  <david.sherwood@arm.com>
8090
8091         * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
8092         TARGET_MEM_REFs.
8093         * gimple-expr.h (is_gimple_addressable: Likewise.
8094         * gimple-expr.c (is_gimple_address): Likewise.
8095         * internal-fn.c (expand_call_mem_ref): New function.
8096         (expand_mask_load_optab_fn): Use it.
8097         (expand_mask_store_optab_fn): Likewise.
8098
8099 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8100             Alan Hayward  <alan.hayward@arm.com>
8101             David Sherwood  <david.sherwood@arm.com>
8102
8103         * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
8104         (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
8105         (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
8106         (cond_umax@var{mode}): Document.
8107         * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
8108         (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
8109         (cond_umin_optab, cond_umax_optab): New optabs.
8110         * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
8111         (COND_IOR, COND_XOR): New internal functions.
8112         * internal-fn.h (get_conditional_internal_fn): Declare.
8113         * internal-fn.c (cond_binary_direct): New macro.
8114         (expand_cond_binary_optab_fn): Likewise.
8115         (direct_cond_binary_optab_supported_p): Likewise.
8116         (get_conditional_internal_fn): New function.
8117         * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
8118         Cope with reduction statements that are vectorized as calls rather
8119         than assignments.
8120         * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
8121         * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
8122         (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
8123         (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
8124         (UNSPEC_COND_EOR): New unspecs.
8125         (optab): Add mappings for them.
8126         (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
8127         (sve_int_op, sve_fp_op): New int attributes.
8128
8129 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8130             Alan Hayward  <alan.hayward@arm.com>
8131             David Sherwood  <david.sherwood@arm.com>
8132
8133         * optabs.def (while_ult_optab): New optab.
8134         * doc/md.texi (while_ult@var{m}@var{n}): Document.
8135         * internal-fn.def (WHILE_ULT): New internal function.
8136         * internal-fn.h (direct_internal_fn_supported_p): New override
8137         that takes two types as argument.
8138         * internal-fn.c (while_direct): New macro.
8139         (expand_while_optab_fn): New function.
8140         (convert_optab_supported_p): Likewise.
8141         (direct_while_optab_supported_p): New macro.
8142         * wide-int.h (wi::udiv_ceil): New function.
8143         * tree-vectorizer.h (rgroup_masks): New structure.
8144         (vec_loop_masks): New typedef.
8145         (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
8146         and fully_masked_p.
8147         (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
8148         (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
8149         (vect_max_vf): New function.
8150         (slpeel_make_loop_iterate_ntimes): Delete.
8151         (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
8152         (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
8153         (vect_record_loop_mask, vect_get_loop_mask): Likewise.
8154         * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
8155         internal-fn.h, stor-layout.h and optabs-query.h.
8156         (vect_set_loop_mask): New function.
8157         (add_preheader_seq): Likewise.
8158         (add_header_seq): Likewise.
8159         (interleave_supported_p): Likewise.
8160         (vect_maybe_permute_loop_masks): Likewise.
8161         (vect_set_loop_masks_directly): Likewise.
8162         (vect_set_loop_condition_masked): Likewise.
8163         (vect_set_loop_condition_unmasked): New function, split out from
8164         slpeel_make_loop_iterate_ntimes.
8165         (slpeel_make_loop_iterate_ntimes): Rename to..
8166         (vect_set_loop_condition): ...this.  Use vect_set_loop_condition_masked
8167         for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
8168         (vect_do_peeling): Update call accordingly.
8169         (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
8170         loops.
8171         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
8172         mask_compare_type, can_fully_mask_p and fully_masked_p.
8173         (release_vec_loop_masks): New function.
8174         (_loop_vec_info): Use it to free the loop masks.
8175         (can_produce_all_loop_masks_p): New function.
8176         (vect_get_max_nscalars_per_iter): Likewise.
8177         (vect_verify_full_masking): Likewise.
8178         (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
8179         retries, and free the mask rgroups before retrying.  Check loop-wide
8180         reasons for disallowing fully-masked loops.  Make the final decision
8181         about whether use a fully-masked loop or not.
8182         (vect_estimate_min_profitable_iters): Do not assume that peeling
8183         for the number of iterations will be needed for fully-masked loops.
8184         (vectorizable_reduction): Disable fully-masked loops.
8185         (vectorizable_live_operation): Likewise.
8186         (vect_halve_mask_nunits): New function.
8187         (vect_double_mask_nunits): Likewise.
8188         (vect_record_loop_mask): Likewise.
8189         (vect_get_loop_mask): Likewise.
8190         (vect_transform_loop): Handle the case in which the final loop
8191         iteration might handle a partial vector.  Call vect_set_loop_condition
8192         instead of slpeel_make_loop_iterate_ntimes.
8193         * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
8194         (check_load_store_masking): New function.
8195         (prepare_load_store_mask): Likewise.
8196         (vectorizable_store): Handle fully-masked loops.
8197         (vectorizable_load): Likewise.
8198         (supportable_widening_operation): Use vect_halve_mask_nunits for
8199         booleans.
8200         (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
8201         (vect_gen_while): New function.
8202         * config/aarch64/aarch64.md (umax<mode>3): New expander.
8203         (aarch64_uqdec<mode>): New insn.
8204
8205 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8206             Alan Hayward  <alan.hayward@arm.com>
8207             David Sherwood  <david.sherwood@arm.com>
8208
8209         * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
8210         (reduc_xor_scal_optab): New optabs.
8211         * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
8212         (reduc_xor_scal_@var{m}): Document.
8213         * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
8214         * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
8215         internal functions.
8216         * fold-const-call.c (fold_const_call): Handle them.
8217         * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
8218         internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
8219         * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
8220         (*reduc_<bit_reduc>_scal_<mode>): New patterns.
8221         * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
8222         (UNSPEC_XORV): New unspecs.
8223         (optab): Add entries for them.
8224         (BITWISEV): New int iterator.
8225         (bit_reduc_op): New int attributes.
8226
8227 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8228             Alan Hayward  <alan.hayward@arm.com>
8229             David Sherwood  <david.sherwood@arm.com>
8230
8231         * doc/md.texi (vec_shl_insert_@var{m}): New optab.
8232         * internal-fn.def (VEC_SHL_INSERT): New internal function.
8233         * optabs.def (vec_shl_insert_optab): New optab.
8234         * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
8235         (duplicate_and_interleave): Likewise.
8236         * tree-vect-loop.c: Include internal-fn.h.
8237         (neutral_op_for_slp_reduction): New function, split out from
8238         get_initial_defs_for_reduction.
8239         (get_initial_def_for_reduction): Handle option 2 for variable-length
8240         vectors by loading the neutral value into a vector and then shifting
8241         the initial value into element 0.
8242         (get_initial_defs_for_reduction): Replace the code argument with
8243         the neutral value calculated by neutral_op_for_slp_reduction.
8244         Use gimple_build_vector for constant-length vectors.
8245         Use IFN_VEC_SHL_INSERT for variable-length vectors if all
8246         but the first group_size elements have a neutral value.
8247         Use duplicate_and_interleave otherwise.
8248         (vect_create_epilog_for_reduction): Take a neutral_op parameter.
8249         Update call to get_initial_defs_for_reduction.  Handle SLP
8250         reductions for variable-length vectors by creating one vector
8251         result for each scalar result, with the elements associated
8252         with other scalar results stubbed out with the neutral value.
8253         (vectorizable_reduction): Call neutral_op_for_slp_reduction.
8254         Require IFN_VEC_SHL_INSERT for double reductions on
8255         variable-length vectors, or SLP reductions that have
8256         a neutral value.  Require can_duplicate_and_interleave_p
8257         support for variable-length unchained SLP reductions if there
8258         is no neutral value, such as for MIN/MAX reductions.  Also require
8259         the number of vector elements to be a multiple of the number of
8260         SLP statements when doing variable-length unchained SLP reductions.
8261         Update call to vect_create_epilog_for_reduction.
8262         * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
8263         and remove initial values.
8264         (duplicate_and_interleave): Make public.
8265         * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
8266         * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
8267
8268 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8269             Alan Hayward  <alan.hayward@arm.com>
8270             David Sherwood  <david.sherwood@arm.com>
8271
8272         * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
8273         (can_duplicate_and_interleave_p): New function.
8274         (vect_get_and_check_slp_defs): Take the vector of statements
8275         rather than just the current one.  Remove excess parentheses.
8276         Restriction rejectinon of vect_constant_def and vect_external_def
8277         for variable-length vectors to boolean types, or types for which
8278         can_duplicate_and_interleave_p is false.
8279         (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
8280         (duplicate_and_interleave): New function.
8281         (vect_get_constant_vectors): Use gimple_build_vector for
8282         constant-length vectors and suitable variable-length constant
8283         vectors.  Use duplicate_and_interleave for other variable-length
8284         vectors.  Don't defer the update when inserting new statements.
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         * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
8291         min_profitable_iters doesn't go negative.
8292
8293 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8294             Alan Hayward  <alan.hayward@arm.com>
8295             David Sherwood  <david.sherwood@arm.com>
8296
8297         * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
8298         (vec_mask_store_lanes@var{m}@var{n}): Likewise.
8299         * optabs.def (vec_mask_load_lanes_optab): New optab.
8300         (vec_mask_store_lanes_optab): Likewise.
8301         * internal-fn.def (MASK_LOAD_LANES): New internal function.
8302         (MASK_STORE_LANES): Likewise.
8303         * internal-fn.c (mask_load_lanes_direct): New macro.
8304         (mask_store_lanes_direct): Likewise.
8305         (expand_mask_load_optab_fn): Handle masked operations.
8306         (expand_mask_load_lanes_optab_fn): New macro.
8307         (expand_mask_store_optab_fn): Handle masked operations.
8308         (expand_mask_store_lanes_optab_fn): New macro.
8309         (direct_mask_load_lanes_optab_supported_p): Likewise.
8310         (direct_mask_store_lanes_optab_supported_p): Likewise.
8311         * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
8312         parameter.
8313         (vect_load_lanes_supported): Likewise.
8314         * tree-vect-data-refs.c (strip_conversion): New function.
8315         (can_group_stmts_p): Likewise.
8316         (vect_analyze_data_ref_accesses): Use it instead of checking
8317         for a pair of assignments.
8318         (vect_store_lanes_supported): Take a masked_p parameter.
8319         (vect_load_lanes_supported): Likewise.
8320         * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
8321         vect_store_lanes_supported and vect_load_lanes_supported.
8322         * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
8323         * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
8324         parameter.  Don't allow gaps for masked accesses.
8325         Use vect_get_store_rhs.  Update calls to vect_store_lanes_supported
8326         and vect_load_lanes_supported.
8327         (get_load_store_type): Take a masked_p parameter and update
8328         call to get_group_load_store_type.
8329         (vectorizable_store): Update call to get_load_store_type.
8330         Handle IFN_MASK_STORE_LANES.
8331         (vectorizable_load): Update call to get_load_store_type.
8332         Handle IFN_MASK_LOAD_LANES.
8333
8334 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8335             Alan Hayward  <alan.hayward@arm.com>
8336             David Sherwood  <david.sherwood@arm.com>
8337
8338         * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
8339         modes for SVE.
8340         * config/aarch64/aarch64-protos.h
8341         (aarch64_sve_struct_memory_operand_p): Declare.
8342         * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
8343         (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
8344         (VPRED, vpred): Handle SVE structure modes.
8345         * config/aarch64/constraints.md (Utx): New constraint.
8346         * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
8347         (aarch64_sve_struct_nonimmediate_operand): New predicates.
8348         * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
8349         * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
8350         (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
8351         structure modes.  Split into pieces after RA.
8352         (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
8353         (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
8354         New patterns.
8355         * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
8356         SVE structure modes.
8357         (aarch64_classify_address): Likewise.
8358         (sizetochar): Move earlier in file.
8359         (aarch64_print_operand): Handle SVE register lists.
8360         (aarch64_array_mode): New function.
8361         (aarch64_sve_struct_memory_operand_p): Likewise.
8362         (TARGET_ARRAY_MODE): Redefine.
8363
8364 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8365             Alan Hayward  <alan.hayward@arm.com>
8366             David Sherwood  <david.sherwood@arm.com>
8367
8368         * target.def (array_mode): New target hook.
8369         * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
8370         * doc/tm.texi: Regenerate.
8371         * hooks.h (hook_optmode_mode_uhwi_none): Declare.
8372         * hooks.c (hook_optmode_mode_uhwi_none): New function.
8373         * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
8374         targetm.array_mode.
8375         * stor-layout.c (mode_for_array): Likewise.  Support polynomial
8376         type sizes.
8377
8378 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8379             Alan Hayward  <alan.hayward@arm.com>
8380             David Sherwood  <david.sherwood@arm.com>
8381
8382         * fold-const.c (fold_binary_loc): Check the argument types
8383         rather than the result type when testing for a vector operation.
8384
8385 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8386
8387         * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
8388         * doc/tm.texi: Regenerate.
8389
8390 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8391             Alan Hayward  <alan.hayward@arm.com>
8392             David Sherwood  <david.sherwood@arm.com>
8393
8394         * doc/invoke.texi (-msve-vector-bits=): Document new option.
8395         (sve): Document new AArch64 extension.
8396         * doc/md.texi (w): Extend the description of the AArch64
8397         constraint to include SVE vectors.
8398         (Upl, Upa): Document new AArch64 predicate constraints.
8399         * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
8400         enum.
8401         * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
8402         (msve-vector-bits=): New option.
8403         * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
8404         SVE when these are disabled.
8405         (sve): New extension.
8406         * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
8407         modes.  Adjust their number of units based on aarch64_sve_vg.
8408         (MAX_BITSIZE_MODE_ANY_MODE): Define.
8409         * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
8410         aarch64_addr_query_type.
8411         (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
8412         (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
8413         (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
8414         (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
8415         (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
8416         (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
8417         (aarch64_simd_imm_zero_p): Delete.
8418         (aarch64_check_zero_based_sve_index_immediate): Declare.
8419         (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
8420         (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
8421         (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
8422         (aarch64_sve_float_mul_immediate_p): Likewise.
8423         (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
8424         rather than an rtx.
8425         (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
8426         (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
8427         (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
8428         (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
8429         (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
8430         (aarch64_regmode_natural_size): Likewise.
8431         * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
8432         (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
8433         left one place.
8434         (AARCH64_ISA_SVE, TARGET_SVE): New macros.
8435         (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
8436         for VG and the SVE predicate registers.
8437         (V_ALIASES): Add a "z"-prefixed alias.
8438         (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
8439         (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
8440         (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
8441         (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
8442         (REG_CLASS_NAMES): Add entries for them.
8443         (REG_CLASS_CONTENTS): Likewise.  Update ALL_REGS to include VG
8444         and the predicate registers.
8445         (aarch64_sve_vg): Declare.
8446         (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
8447         (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
8448         (REGMODE_NATURAL_SIZE): Define.
8449         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
8450         SVE macros.
8451         * config/aarch64/aarch64.c: Include cfgrtl.h.
8452         (simd_immediate_info): Add a constructor for series vectors,
8453         and an associated step field.
8454         (aarch64_sve_vg): New variable.
8455         (aarch64_dbx_register_number): Handle VG and the predicate registers.
8456         (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
8457         (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
8458         (VEC_ANY_DATA, VEC_STRUCT): New constants.
8459         (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
8460         (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
8461         (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
8462         (aarch64_get_mask_mode): New functions.
8463         (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
8464         and FP_LO_REGS.  Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
8465         (aarch64_hard_regno_mode_ok): Handle VG.  Also handle the SVE
8466         predicate modes and predicate registers.  Explicitly restrict
8467         GPRs to modes of 16 bytes or smaller.  Only allow FP registers
8468         to store a vector mode if it is recognized by
8469         aarch64_classify_vector_mode.
8470         (aarch64_regmode_natural_size): New function.
8471         (aarch64_hard_regno_caller_save_mode): Return the original mode
8472         for predicates.
8473         (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
8474         (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
8475         (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
8476         (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
8477         functions.
8478         (aarch64_add_offset): Add a temp2 parameter.  Assert that temp1
8479         does not overlap dest if the function is frame-related.  Handle
8480         SVE constants.
8481         (aarch64_split_add_offset): New function.
8482         (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
8483         them aarch64_add_offset.
8484         (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
8485         and update call to aarch64_sub_sp.
8486         (aarch64_add_cfa_expression): New function.
8487         (aarch64_expand_prologue): Pass extra temporary registers to the
8488         functions above.  Handle the case in which we need to emit new
8489         DW_CFA_expressions for registers that were originally saved
8490         relative to the stack pointer, but now have to be expressed
8491         relative to the frame pointer.
8492         (aarch64_output_mi_thunk): Pass extra temporary registers to the
8493         functions above.
8494         (aarch64_expand_epilogue): Likewise.  Prevent inheritance of
8495         IP0 and IP1 values for SVE frames.
8496         (aarch64_expand_vec_series): New function.
8497         (aarch64_expand_sve_widened_duplicate): Likewise.
8498         (aarch64_expand_sve_const_vector): Likewise.
8499         (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
8500         Handle SVE constants.  Use emit_move_insn to move a force_const_mem
8501         into the register, rather than emitting a SET directly.
8502         (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
8503         (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
8504         (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
8505         (offset_9bit_signed_scaled_p): New functions.
8506         (aarch64_replicate_bitmask_imm): New function.
8507         (aarch64_bitmask_imm): Use it.
8508         (aarch64_cannot_force_const_mem): Reject expressions involving
8509         a CONST_POLY_INT.  Update call to aarch64_classify_symbol.
8510         (aarch64_classify_index): Handle SVE indices, by requiring
8511         a plain register index with a scale that matches the element size.
8512         (aarch64_classify_address): Handle SVE addresses.  Assert that
8513         the mode of the address is VOIDmode or an integer mode.
8514         Update call to aarch64_classify_symbol.
8515         (aarch64_classify_symbolic_expression): Update call to
8516         aarch64_classify_symbol.
8517         (aarch64_const_vec_all_in_range_p): New function.
8518         (aarch64_print_vector_float_operand): Likewise.
8519         (aarch64_print_operand): Handle 'N' and 'C'.  Use "zN" rather than
8520         "vN" for FP registers with SVE modes.  Handle (const ...) vectors
8521         and the FP immediates 1.0 and 0.5.
8522         (aarch64_print_address_internal): Handle SVE addresses.
8523         (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
8524         (aarch64_regno_regclass): Handle predicate registers.
8525         (aarch64_secondary_reload): Handle big-endian reloads of SVE
8526         data modes.
8527         (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
8528         (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
8529         (aarch64_convert_sve_vector_bits): New function.
8530         (aarch64_override_options): Use it to handle -msve-vector-bits=.
8531         (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
8532         rather than an rtx.
8533         (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
8534         Handle SVE vector and predicate modes.  Accept VL-based constants
8535         that need only one temporary register, and VL offsets that require
8536         no temporary registers.
8537         (aarch64_conditional_register_usage): Mark the predicate registers
8538         as fixed if SVE isn't available.
8539         (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
8540         Return true for SVE vector and predicate modes.
8541         (aarch64_simd_container_mode): Take the number of bits as a poly_int64
8542         rather than an unsigned int.  Handle SVE modes.
8543         (aarch64_preferred_simd_mode): Update call accordingly.  Handle
8544         SVE modes.
8545         (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
8546         if SVE is enabled.
8547         (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
8548         (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
8549         (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
8550         (aarch64_sve_float_mul_immediate_p): New functions.
8551         (aarch64_sve_valid_immediate): New function.
8552         (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
8553         Explicitly reject structure modes.  Check for INDEX constants.
8554         Handle PTRUE and PFALSE constants.
8555         (aarch64_check_zero_based_sve_index_immediate): New function.
8556         (aarch64_simd_imm_zero_p): Delete.
8557         (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
8558         vector modes.  Accept constants in the range of CNT[BHWD].
8559         (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
8560         ask for an Advanced SIMD mode.
8561         (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
8562         (aarch64_simd_vector_alignment): Handle SVE predicates.
8563         (aarch64_vectorize_preferred_vector_alignment): New function.
8564         (aarch64_simd_vector_alignment_reachable): Use it instead of
8565         the vector size.
8566         (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
8567         (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
8568         functions.
8569         (MAX_VECT_LEN): Delete.
8570         (expand_vec_perm_d): Add a vec_flags field.
8571         (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
8572         (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
8573         (aarch64_evpc_ext): Don't apply a big-endian lane correction
8574         for SVE modes.
8575         (aarch64_evpc_rev): Rename to...
8576         (aarch64_evpc_rev_local): ...this.  Use a predicated operation for SVE.
8577         (aarch64_evpc_rev_global): New function.
8578         (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
8579         (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
8580         MAX_VECT_LEN.
8581         (aarch64_evpc_sve_tbl): New function.
8582         (aarch64_expand_vec_perm_const_1): Update after rename of
8583         aarch64_evpc_rev.  Handle SVE permutes too, trying
8584         aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
8585         than aarch64_evpc_tbl.
8586         (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
8587         (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
8588         (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
8589         (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
8590         (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
8591         (aarch64_expand_sve_vcond): New functions.
8592         (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
8593         of aarch64_vector_mode_p.
8594         (aarch64_dwarf_poly_indeterminate_value): New function.
8595         (aarch64_compute_pressure_classes): Likewise.
8596         (aarch64_can_change_mode_class): Likewise.
8597         (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
8598         (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
8599         (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
8600         (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
8601         (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
8602         (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
8603         * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
8604         (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
8605         constraints.
8606         (Dn, Dl, Dr): Accept const as well as const_vector.
8607         (Dz): Likewise.  Compare against CONST0_RTX.
8608         * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
8609         of "vector" where appropriate.
8610         (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
8611         (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
8612         (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
8613         (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
8614         (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
8615         (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
8616         (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
8617         (v_int_equiv): Extend to SVE modes.
8618         (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
8619         mode attributes.
8620         (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
8621         (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
8622         (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
8623         (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
8624         (SVE_COND_FP_CMP): New int iterators.
8625         (perm_hilo): Handle the new unpack unspecs.
8626         (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
8627         attributes.
8628         * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
8629         (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
8630         (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
8631         (aarch64_equality_operator, aarch64_constant_vector_operand)
8632         (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
8633         (aarch64_sve_nonimmediate_operand): Likewise.
8634         (aarch64_sve_general_operand): Likewise.
8635         (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
8636         (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
8637         (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
8638         (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
8639         (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
8640         (aarch64_sve_float_arith_immediate): Likewise.
8641         (aarch64_sve_float_arith_with_sub_immediate): Likewise.
8642         (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
8643         (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
8644         (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
8645         (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
8646         (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
8647         (aarch64_sve_float_arith_operand): Likewise.
8648         (aarch64_sve_float_arith_with_sub_operand): Likewise.
8649         (aarch64_sve_float_mul_operand): Likewise.
8650         (aarch64_sve_vec_perm_operand): Likewise.
8651         (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
8652         (aarch64_mov_operand): Accept const_poly_int and const_vector.
8653         (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
8654         as well as const_vector.
8655         (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
8656         in file.  Use CONST0_RTX and CONSTM1_RTX.
8657         (aarch64_simd_or_scalar_imm_zero): Likewise.  Add match_codes.
8658         (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
8659         Use aarch64_simd_imm_zero.
8660         * config/aarch64/aarch64-sve.md: New file.
8661         * config/aarch64/aarch64.md: Include it.
8662         (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
8663         (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
8664         (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
8665         (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
8666         (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
8667         (sve): New attribute.
8668         (enabled): Disable instructions with the sve attribute unless
8669         TARGET_SVE.
8670         (movqi, movhi): Pass CONST_POLY_INT operaneds through
8671         aarch64_expand_mov_immediate.
8672         (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
8673         CNT[BHSD] immediates.
8674         (movti): Split CONST_POLY_INT moves into two halves.
8675         (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
8676         Split additions that need a temporary here if the destination
8677         is the stack pointer.
8678         (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
8679         (*add<mode>3_poly_1): New instruction.
8680         (set_clobber_cc): New expander.
8681
8682 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8683
8684         * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
8685         parameter and use it instead of GET_MODE_SIZE (innermode).  Use
8686         inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
8687         Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
8688         GET_MODE_NUNITS (innermode).  Also add a first_elem parameter.
8689         Change innermode from fixed_mode_size to machine_mode.
8690         (simplify_subreg): Update call accordingly.  Handle a constant-sized
8691         subreg of a variable-length CONST_VECTOR.
8692
8693 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8694             Alan Hayward  <alan.hayward@arm.com>
8695             David Sherwood  <david.sherwood@arm.com>
8696
8697         * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
8698         (add_offset_to_base): New function, split out from...
8699         (create_mem_ref): ...here.  When handling a scale other than 1,
8700         check first whether the address is valid without the offset.
8701         Add it into the base if so, leaving the index and scale as-is.
8702
8703 2018-01-12  Jakub Jelinek  <jakub@redhat.com>
8704
8705         PR c++/83778
8706         * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
8707         fold_for_warn before checking if arg2 is INTEGER_CST.
8708
8709 2018-01-12  Segher Boessenkool  <segher@kernel.crashing.org>
8710
8711         * config/rs6000/predicates.md (load_multiple_operation): Delete.
8712         (store_multiple_operation): Delete.
8713         * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
8714         * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
8715         * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
8716         guarded by TARGET_STRING.
8717         (rs6000_output_load_multiple): Delete.
8718         * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
8719         OPTION_MASK_STRING / TARGET_STRING handling.
8720         (print_operand) <'N', 'O'>: Add comment that these are unused now.
8721         (const rs6000_opt_masks) <"string">: Change mask to 0.
8722         * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
8723         (MASK_STRING): Delete.
8724         * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
8725         parts.  Simplify.
8726         (load_multiple): Delete.
8727         (*ldmsi8): Delete.
8728         (*ldmsi7): Delete.
8729         (*ldmsi6): Delete.
8730         (*ldmsi5): Delete.
8731         (*ldmsi4): Delete.
8732         (*ldmsi3): Delete.
8733         (store_multiple): Delete.
8734         (*stmsi8): Delete.
8735         (*stmsi7): Delete.
8736         (*stmsi6): Delete.
8737         (*stmsi5): Delete.
8738         (*stmsi4): Delete.
8739         (*stmsi3): Delete.
8740         (movmemsi_8reg): Delete.
8741         (corresponding unnamed define_insn): Delete.
8742         (movmemsi_6reg): Delete.
8743         (corresponding unnamed define_insn): Delete.
8744         (movmemsi_4reg): Delete.
8745         (corresponding unnamed define_insn): Delete.
8746         (movmemsi_2reg): Delete.
8747         (corresponding unnamed define_insn): Delete.
8748         (movmemsi_1reg): Delete.
8749         (corresponding unnamed define_insn): Delete.
8750         * config/rs6000/rs6000.opt (mno-string): New.
8751         (mstring): Replace by deprecation warning stub.
8752         * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
8753
8754 2018-01-12  Jakub Jelinek  <jakub@redhat.com>
8755
8756         * regrename.c (regrename_do_replace): If replacing the same
8757         reg multiple times, try to reuse last created gen_raw_REG.
8758
8759         PR debug/81155
8760         * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
8761         main to workaround a bug in GDB.
8762
8763 2018-01-12  Tom de Vries  <tom@codesourcery.com>
8764
8765         PR target/83737
8766         * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
8767
8768 2018-01-12  Vladimir Makarov  <vmakarov@redhat.com>
8769
8770         PR rtl-optimization/80481
8771         * ira-color.c (get_cap_member): New function.
8772         (allocnos_conflict_by_live_ranges_p): Use it.
8773         (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
8774         (setup_slot_coalesced_allocno_live_ranges): Ditto.
8775
8776 2018-01-12  Uros Bizjak  <ubizjak@gmail.com>
8777
8778         PR target/83628
8779         * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
8780         (*saddl_se_1): Ditto.
8781         (*ssubsi_1): Ditto.
8782         (*ssubl_se_1): Ditto.
8783
8784 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
8785
8786         * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
8787         rather than wi::to_widest for DR_INITs.
8788         * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
8789         wi::to_poly_offset rather than wi::to_offset for DR_INIT.
8790         (vect_analyze_data_ref_accesses): Require both DR_INITs to be
8791         INTEGER_CSTs.
8792         (vect_analyze_group_access_1): Note that here.
8793
8794 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
8795
8796         * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
8797         polynomial type sizes.
8798
8799 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
8800
8801         * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
8802         poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
8803         (gimple_add_tmp_var): Likewise.
8804
8805 2018-01-12  Martin Liska  <mliska@suse.cz>
8806
8807         * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
8808         (gimple_alloc_sizes): Likewise.
8809         (dump_gimple_statistics): Use PRIu64 in printf format.
8810         * gimple.h: Change uint64_t to int.
8811
8812 2018-01-12  Martin Liska  <mliska@suse.cz>
8813
8814         * tree-core.h: Use uint64_t instead of int.
8815         * tree.c (tree_node_counts): Likewise.
8816         (tree_node_sizes): Likewise.
8817         (dump_tree_statistics): Use PRIu64 in printf format.
8818
8819 2018-01-12  Martin Liska  <mliska@suse.cz>
8820
8821         * Makefile.in: As qsort_chk is implemented in vec.c, add
8822         vec.o to linkage of gencfn-macros.
8823         * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
8824         passing the info to record_node_allocation_statistics.
8825         (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
8826         and pass the info.
8827         * ggc-common.c (struct ggc_usage): Add operator== and use
8828         it in operator< and compare function.
8829         * mem-stats.h (struct mem_usage): Likewise.
8830         * vec.c (struct vec_usage): Remove operator< and compare
8831         function. Can be simply inherited.
8832
8833 2018-01-12  Martin Jambor  <mjambor@suse.cz>
8834
8835         PR target/81616
8836         * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
8837         * tree-ssa-math-opts.c: Include domwalk.h.
8838         (convert_mult_to_fma_1): New function.
8839         (fma_transformation_info): New type.
8840         (fma_deferring_state): Likewise.
8841         (cancel_fma_deferring): New function.
8842         (result_of_phi): Likewise.
8843         (last_fma_candidate_feeds_initial_phi): Likewise.
8844         (convert_mult_to_fma): Added deferring logic, split actual
8845         transformation to convert_mult_to_fma_1.
8846         (math_opts_dom_walker): New type.
8847         (math_opts_dom_walker::after_dom_children): New method, body moved
8848         here from pass_optimize_widening_mul::execute, added deferring logic
8849         bits.
8850         (pass_optimize_widening_mul::execute): Moved most of code to
8851         math_opts_dom_walker::after_dom_children.
8852         * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
8853         * config/i386/i386.c (ix86_option_override_internal): Added
8854         maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
8855
8856 2018-01-12  Richard Biener  <rguenther@suse.de>
8857
8858         PR debug/83157
8859         * dwarf2out.c (gen_variable_die): Do not reset old_die for
8860         inline instance vars.
8861
8862 2018-01-12  Oleg Endo  <olegendo@gcc.gnu.org>
8863
8864         PR target/81819
8865         * config/rx/rx.c (rx_is_restricted_memory_address):
8866         Handle SUBREG case.
8867
8868 2018-01-12  Richard Biener  <rguenther@suse.de>
8869
8870         PR tree-optimization/80846
8871         * target.def (split_reduction): New target hook.
8872         * targhooks.c (default_split_reduction): New function.
8873         * targhooks.h (default_split_reduction): Declare.
8874         * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
8875         target requests first reduce vectors by combining low and high
8876         parts.
8877         * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
8878         (get_vectype_for_scalar_type_and_size): Export.
8879         * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
8880         * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
8881         * doc/tm.texi: Regenerate.
8882         * config/i386/i386.c (ix86_split_reduction): Implement
8883         TARGET_VECTORIZE_SPLIT_REDUCTION.
8884
8885 2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
8886
8887         PR target/83368
8888         * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
8889         in PIC mode except for TARGET_VXWORKS_RTP.
8890         * config/sparc/sparc.c: Include cfgrtl.h.
8891         (TARGET_INIT_PIC_REG): Define.
8892         (TARGET_USE_PSEUDO_PIC_REG): Likewise.
8893         (sparc_pic_register_p): New predicate.
8894         (sparc_legitimate_address_p): Use it.
8895         (sparc_legitimize_pic_address): Likewise.
8896         (sparc_delegitimize_address): Likewise.
8897         (sparc_mode_dependent_address_p): Likewise.
8898         (gen_load_pcrel_sym): Remove 4th parameter.
8899         (load_got_register): Adjust call to above.  Remove obsolete stuff.
8900         (sparc_expand_prologue): Do not call load_got_register here.
8901         (sparc_flat_expand_prologue): Likewise.
8902         (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
8903         (sparc_use_pseudo_pic_reg): New function.
8904         (sparc_init_pic_reg): Likewise.
8905         * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
8906         (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
8907
8908 2018-01-12  Christophe Lyon  <christophe.lyon@linaro.org>
8909
8910         * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
8911         Add item for branch_cost.
8912
8913 2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
8914
8915         PR rtl-optimization/83565
8916         * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
8917         not extend the result to a larger mode for rotate operations.
8918         (num_sign_bit_copies1): Likewise.
8919
8920 2018-01-12  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
8921
8922         PR target/40411
8923         * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
8924         -symbolic.
8925         Use values-Xc.o for -pedantic.
8926         Link with values-xpg4.o for C90, values-xpg6.o otherwise.
8927
8928 2018-01-12  Martin Liska  <mliska@suse.cz>
8929
8930         PR ipa/83054
8931         * ipa-devirt.c (final_warning_record::grow_type_warnings):
8932         New function.
8933         (possible_polymorphic_call_targets): Use it.
8934         (ipa_devirt): Likewise.
8935
8936 2018-01-12  Martin Liska  <mliska@suse.cz>
8937
8938         * profile-count.h (enum profile_quality): Use 0 as invalid
8939         enum value of profile_quality.
8940
8941 2018-01-12  Chung-Ju Wu  <jasonwucj@gmail.com>
8942
8943         * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
8944         -mext-string options.
8945
8946 2018-01-12  Richard Biener  <rguenther@suse.de>
8947
8948         * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
8949         DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
8950         * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
8951         Likewise.
8952         * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
8953
8954 2018-01-11  Michael Meissner  <meissner@linux.vnet.ibm.com>
8955
8956         * configure.ac (--with-long-double-format): Add support for the
8957         configuration option to change the default long double format on
8958         PowerPC systems.
8959         * config.gcc (powerpc*-linux*-*): Likewise.
8960         * configure: Regenerate.
8961         * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
8962         double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
8963         used without modification.
8964
8965 2018-01-11  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
8966
8967         * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
8968         (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
8969         * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
8970         MISC_BUILTIN_SPEC_BARRIER.
8971         (rs6000_init_builtins): Likewise.
8972         * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
8973         enum value.
8974         (speculation_barrier): New define_insn.
8975         * doc/extend.texi: Document __builtin_speculation_barrier.
8976
8977 2018-01-11  Jakub Jelinek  <jakub@redhat.com>
8978
8979         PR target/83203
8980         * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
8981         is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
8982         * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
8983         iterators.
8984         (ssescalarmodesuffix): Add 512-bit vectors.  Use "d" or "q" for
8985         integral modes instead of "ss" and "sd".
8986         (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
8987         vectors with 32-bit and 64-bit elements.
8988         (vecdupssescalarmodesuffix): New mode attribute.
8989         (vec_dup<mode>): Use it.
8990
8991 2018-01-11  H.J. Lu  <hongjiu.lu@intel.com>
8992
8993         PR target/83330
8994         * config/i386/i386.c (ix86_compute_frame_layout): Align stack
8995         frame if argument is passed on stack.
8996
8997 2018-01-11  Jakub Jelinek  <jakub@redhat.com>
8998
8999         PR target/82682
9000         * ree.c (combine_reaching_defs): Optimize also
9001         reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
9002         reg2=any_extend(exp); reg1=reg2;, formatting fix.
9003
9004 2018-01-11  Jan Hubicka  <hubicka@ucw.cz>
9005
9006         PR middle-end/83189
9007         * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
9008
9009 2018-01-11  Jan Hubicka  <hubicka@ucw.cz>
9010
9011         PR middle-end/83718
9012         * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
9013         after they are computed.
9014
9015 2018-01-11  Bin Cheng  <bin.cheng@arm.com>
9016
9017         PR tree-optimization/83695
9018         * gimple-loop-linterchange.cc
9019         (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
9020         reset cached scev information after interchange.
9021         (pass_linterchange::execute): Remove call to scev_reset_htab.
9022
9023 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
9024
9025         * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
9026         vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
9027         vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
9028         vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
9029         vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
9030         vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
9031         * config/arm/arm_neon_builtins.def (vfmal_lane_low,
9032         vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
9033         vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
9034         vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
9035         vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
9036         * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
9037         (V_lane_reg): Likewise.
9038         * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
9039         New define_expand.
9040         (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
9041         (vfmal_lane_low<mode>_intrinsic,
9042         vfmal_lane_low<vfmlsel2><mode>_intrinsic,
9043         vfmal_lane_high<vfmlsel2><mode>_intrinsic,
9044         vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
9045         vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
9046         vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
9047         vfmsl_lane_high<mode>_intrinsic): New define_insns.
9048
9049 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
9050
9051         * config/arm/arm-cpus.in (fp16fml): New feature.
9052         (ALL_SIMD): Add fp16fml.
9053         (armv8.2-a): Add fp16fml as an option.
9054         (armv8.3-a): Likewise.
9055         (armv8.4-a): Add fp16fml as part of fp16.
9056         * config/arm/arm.h (TARGET_FP16FML): Define.
9057         * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
9058         when appropriate.
9059         * config/arm/arm-modes.def (V2HF): Define.
9060         * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
9061         vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
9062         vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
9063         * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
9064         vfmsl_low, vfmsl_high): New set of builtins.
9065         * config/arm/iterators.md (PLUSMINUS): New code iterator.
9066         (vfml_op): New code attribute.
9067         (VFMLHALVES): New int iterator.
9068         (VFML, VFMLSEL): New mode attributes.
9069         (V_reg): Define mapping for V2HF.
9070         (V_hi, V_lo): New mode attributes.
9071         (VF_constraint): Likewise.
9072         (vfml_half, vfml_half_selector): New int attributes.
9073         * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
9074         define_expand.
9075         (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
9076         vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
9077         New define_insn.
9078         * config/arm/t-arm-elf (v8_fps): Add fp16fml.
9079         * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
9080         * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
9081         * doc/invoke.texi (ARM Options): Document fp16fml.  Update armv8.4-a
9082         documentation.
9083         * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
9084         Document new effective target and option set.
9085
9086 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
9087
9088         * config/arm/arm-cpus.in (armv8_4): New feature.
9089         (ARMv8_4a): New fgroup.
9090         (armv8.4-a): New arch.
9091         * config/arm/arm-tables.opt: Regenerate.
9092         * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
9093         * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
9094         * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
9095         Add matching rules for -march=armv8.4-a and extensions.
9096         * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
9097
9098 2018-01-11  Oleg Endo  <olegendo@gcc.gnu.org>
9099
9100         PR target/81821
9101         * config/rx/rx.md (BW): New mode attribute.
9102         (sync_lock_test_and_setsi): Add mode suffix to insn output.
9103
9104 2018-01-11  Richard Biener  <rguenther@suse.de>
9105
9106         PR tree-optimization/83435
9107         * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
9108         * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
9109         * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
9110
9111 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
9112             Alan Hayward  <alan.hayward@arm.com>
9113             David Sherwood  <david.sherwood@arm.com>
9114
9115         * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
9116         field.
9117         (aarch64_classify_address): Initialize it.  Track polynomial offsets.
9118         (aarch64_print_address_internal): Use it to check for a zero offset.
9119
9120 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
9121             Alan Hayward  <alan.hayward@arm.com>
9122             David Sherwood  <david.sherwood@arm.com>
9123
9124         * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
9125         * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
9126         Return a poly_int64 rather than a HOST_WIDE_INT.
9127         (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
9128         rather than a HOST_WIDE_INT.
9129         * config/aarch64/aarch64.h (aarch64_frame): Protect with
9130         HAVE_POLY_INT_H rather than HOST_WIDE_INT.  Change locals_offset,
9131         hard_fp_offset, frame_size, initial_adjust, callee_offset and
9132         final_offset from HOST_WIDE_INT to poly_int64.
9133         * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
9134         to_constant when getting the number of units in an Advanced SIMD
9135         mode.
9136         (aarch64_builtin_vectorized_function): Check for a constant number
9137         of units.
9138         * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
9139         GET_MODE_SIZE.
9140         (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
9141         attribute instead of GET_MODE_NUNITS.
9142         * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
9143         (aarch64_class_max_nregs): Use the constant_lowest_bound of the
9144         GET_MODE_SIZE for fixed-size registers.
9145         (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
9146         (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
9147         (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
9148         (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
9149         (aarch64_print_operand, aarch64_print_address_internal)
9150         (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
9151         (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
9152         (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
9153         Handle polynomial GET_MODE_SIZE.
9154         (aarch64_hard_regno_caller_save_mode): Likewise.  Return modes
9155         wider than SImode without modification.
9156         (tls_symbolic_operand_type): Use strip_offset instead of split_const.
9157         (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
9158         (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
9159         passing and returning SVE modes.
9160         (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
9161         rather than GEN_INT.
9162         (aarch64_emit_probe_stack_range): Take the size as a poly_int64
9163         rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
9164         (aarch64_allocate_and_probe_stack_space): Likewise.
9165         (aarch64_layout_frame): Cope with polynomial offsets.
9166         (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
9167         start_offset as a poly_int64 rather than a HOST_WIDE_INT.  Track
9168         polynomial offsets.
9169         (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
9170         (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
9171         poly_int64 rather than a HOST_WIDE_INT.
9172         (aarch64_get_separate_components, aarch64_process_components)
9173         (aarch64_expand_prologue, aarch64_expand_epilogue)
9174         (aarch64_use_return_insn_p): Handle polynomial frame offsets.
9175         (aarch64_anchor_offset): New function, split out from...
9176         (aarch64_legitimize_address): ...here.
9177         (aarch64_builtin_vectorization_cost): Handle polynomial
9178         TYPE_VECTOR_SUBPARTS.
9179         (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
9180         GET_MODE_NUNITS.
9181         (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
9182         number of elements from the PARALLEL rather than the mode.
9183         (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
9184         rather than GET_MODE_BITSIZE.
9185         (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
9186         (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
9187         (aarch64_expand_vec_perm_const_1): Handle polynomial
9188         d->perm.length () and d->perm elements.
9189         (aarch64_evpc_tbl): Likewise.  Use nelt rather than GET_MODE_NUNITS.
9190         Apply to_constant to d->perm elements.
9191         (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
9192         polynomial CONST_VECTOR_NUNITS.
9193         (aarch64_move_pointer): Take amount as a poly_int64 rather
9194         than an int.
9195         (aarch64_progress_pointer): Avoid temporary variable.
9196         * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
9197         the mode attribute instead of GET_MODE.
9198
9199 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
9200             Alan Hayward  <alan.hayward@arm.com>
9201             David Sherwood  <david.sherwood@arm.com>
9202
9203         * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
9204         x exists before using it.
9205         (aarch64_add_constant_internal): Rename to...
9206         (aarch64_add_offset_1): ...this.  Replace regnum with separate
9207         src and dest rtxes.  Handle the case in which they're different,
9208         including when the offset is zero.  Replace scratchreg with an rtx.
9209         Use 2 additions if there is no spare register into which we can
9210         move a 16-bit constant.
9211         (aarch64_add_constant): Delete.
9212         (aarch64_add_offset): Replace reg with separate src and dest
9213         rtxes.  Take a poly_int64 offset instead of a HOST_WIDE_INT.
9214         Use aarch64_add_offset_1.
9215         (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
9216         an rtx rather than an int.  Take the delta as a poly_int64
9217         rather than a HOST_WIDE_INT.  Use aarch64_add_offset.
9218         (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
9219         (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
9220         aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
9221         (aarch64_expand_epilogue): Update calls to aarch64_add_offset
9222         and aarch64_add_sp.
9223         (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
9224         aarch64_add_constant.
9225
9226 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
9227
9228         * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
9229         Use scalar_float_mode.
9230
9231 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
9232
9233         * config/aarch64/aarch64-simd.md
9234         (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
9235         (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
9236         (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
9237         (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
9238         (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
9239         (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
9240         (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
9241         (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
9242         (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
9243         (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
9244
9245 2018-01-11  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
9246
9247         PR target/83514
9248         * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
9249         targ_options->x_arm_arch_string is non NULL.
9250
9251 2018-01-11  Tamar Christina  <tamar.christina@arm.com>
9252
9253         * config/aarch64/aarch64.h
9254         (AARCH64_FL_FOR_ARCH8_4): Add  AARCH64_FL_DOTPROD.
9255
9256 2018-01-11  Sudakshina Das  <sudi.das@arm.com>
9257
9258         PR target/82096
9259         * expmed.c (emit_store_flag_force): Swap if const op0
9260         and change VOIDmode to mode of op0.
9261
9262 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
9263
9264         PR rtl-optimization/83761
9265         * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
9266         than bytes to mode_for_size.
9267
9268 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
9269
9270         PR middle-end/83189
9271         * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
9272         * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
9273         profile.
9274
9275 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
9276
9277         PR middle-end/83575
9278         * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
9279         when in layout mode.
9280         (cfg_layout_finalize): Do not verify cfg before we are out of layout.
9281         * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
9282         partition fixup.
9283
9284 2018-01-10  Michael Collison  <michael.collison@arm.com>
9285
9286         * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
9287         * config/aarch64/aarch64-option-extension.def: Add
9288         AARCH64_OPT_EXTENSION of 'fp16fml'.
9289         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
9290         (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
9291         * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
9292         * config/aarch64/constraints.md (Ui7): New constraint.
9293         * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
9294         (VFMLA_SEL_W): Ditto.
9295         (f16quad): Ditto.
9296         (f16mac1): Ditto.
9297         (VFMLA16_LOW): New int iterator.
9298         (VFMLA16_HIGH): Ditto.
9299         (UNSPEC_FMLAL): New unspec.
9300         (UNSPEC_FMLSL): Ditto.
9301         (UNSPEC_FMLAL2): Ditto.
9302         (UNSPEC_FMLSL2): Ditto.
9303         (f16mac): New code attribute.
9304         * config/aarch64/aarch64-simd-builtins.def
9305         (aarch64_fmlal_lowv2sf): Ditto.
9306         (aarch64_fmlsl_lowv2sf): Ditto.
9307         (aarch64_fmlalq_lowv4sf): Ditto.
9308         (aarch64_fmlslq_lowv4sf): Ditto.
9309         (aarch64_fmlal_highv2sf): Ditto.
9310         (aarch64_fmlsl_highv2sf): Ditto.
9311         (aarch64_fmlalq_highv4sf): Ditto.
9312         (aarch64_fmlslq_highv4sf): Ditto.
9313         (aarch64_fmlal_lane_lowv2sf): Ditto.
9314         (aarch64_fmlsl_lane_lowv2sf): Ditto.
9315         (aarch64_fmlal_laneq_lowv2sf): Ditto.
9316         (aarch64_fmlsl_laneq_lowv2sf): Ditto.
9317         (aarch64_fmlalq_lane_lowv4sf): Ditto.
9318         (aarch64_fmlsl_lane_lowv4sf): Ditto.
9319         (aarch64_fmlalq_laneq_lowv4sf): Ditto.
9320         (aarch64_fmlsl_laneq_lowv4sf): Ditto.
9321         (aarch64_fmlal_lane_highv2sf): Ditto.
9322         (aarch64_fmlsl_lane_highv2sf): Ditto.
9323         (aarch64_fmlal_laneq_highv2sf): Ditto.
9324         (aarch64_fmlsl_laneq_highv2sf): Ditto.
9325         (aarch64_fmlalq_lane_highv4sf): Ditto.
9326         (aarch64_fmlsl_lane_highv4sf): Ditto.
9327         (aarch64_fmlalq_laneq_highv4sf): Ditto.
9328         (aarch64_fmlsl_laneq_highv4sf): Ditto.
9329         * config/aarch64/aarch64-simd.md:
9330         (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
9331         (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
9332         (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
9333         (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
9334         (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
9335         (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
9336         (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
9337         (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
9338         (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
9339         (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
9340         (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
9341         (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
9342         (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
9343         (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
9344         (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
9345         (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
9346         (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
9347         (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
9348         (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
9349         (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
9350         * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
9351         (vfmlsl_low_u32): Ditto.
9352         (vfmlalq_low_u32): Ditto.
9353         (vfmlslq_low_u32): Ditto.
9354         (vfmlal_high_u32): Ditto.
9355         (vfmlsl_high_u32): Ditto.
9356         (vfmlalq_high_u32): Ditto.
9357         (vfmlslq_high_u32): Ditto.
9358         (vfmlal_lane_low_u32): Ditto.
9359         (vfmlsl_lane_low_u32): Ditto.
9360         (vfmlal_laneq_low_u32): Ditto.
9361         (vfmlsl_laneq_low_u32): Ditto.
9362         (vfmlalq_lane_low_u32): Ditto.
9363         (vfmlslq_lane_low_u32): Ditto.
9364         (vfmlalq_laneq_low_u32): Ditto.
9365         (vfmlslq_laneq_low_u32): Ditto.
9366         (vfmlal_lane_high_u32): Ditto.
9367         (vfmlsl_lane_high_u32): Ditto.
9368         (vfmlal_laneq_high_u32): Ditto.
9369         (vfmlsl_laneq_high_u32): Ditto.
9370         (vfmlalq_lane_high_u32): Ditto.
9371         (vfmlslq_lane_high_u32): Ditto.
9372         (vfmlalq_laneq_high_u32): Ditto.
9373         (vfmlslq_laneq_high_u32): Ditto.
9374         * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
9375         (AARCH64_FL_FOR_ARCH8_4): New.
9376         (AARCH64_ISA_F16FML): New ISA flag.
9377         (TARGET_F16FML): New feature flag for fp16fml.
9378         (doc/invoke.texi): Document new fp16fml option.
9379
9380 2018-01-10  Michael Collison  <michael.collison@arm.com>
9381
9382         * config/aarch64/aarch64-builtins.c:
9383         (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
9384         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
9385         (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
9386         * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
9387         (AARCH64_ISA_SHA3): New ISA flag.
9388         (TARGET_SHA3): New feature flag for sha3.
9389         * config/aarch64/iterators.md (sha512_op): New int attribute.
9390         (CRYPTO_SHA512): New int iterator.
9391         (UNSPEC_SHA512H): New unspec.
9392         (UNSPEC_SHA512H2): Ditto.
9393         (UNSPEC_SHA512SU0): Ditto.
9394         (UNSPEC_SHA512SU1): Ditto.
9395         * config/aarch64/aarch64-simd-builtins.def
9396         (aarch64_crypto_sha512hqv2di): New builtin.
9397         (aarch64_crypto_sha512h2qv2di): Ditto.
9398         (aarch64_crypto_sha512su0qv2di): Ditto.
9399         (aarch64_crypto_sha512su1qv2di): Ditto.
9400         (aarch64_eor3qv8hi): Ditto.
9401         (aarch64_rax1qv2di): Ditto.
9402         (aarch64_xarqv2di): Ditto.
9403         (aarch64_bcaxqv8hi): Ditto.
9404         * config/aarch64/aarch64-simd.md:
9405         (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
9406         (aarch64_crypto_sha512su0qv2di): Ditto.
9407         (aarch64_crypto_sha512su1qv2di): Ditto.
9408         (aarch64_eor3qv8hi): Ditto.
9409         (aarch64_rax1qv2di): Ditto.
9410         (aarch64_xarqv2di): Ditto.
9411         (aarch64_bcaxqv8hi): Ditto.
9412         * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
9413         (vsha512h2q_u64): Ditto.
9414         (vsha512su0q_u64): Ditto.
9415         (vsha512su1q_u64): Ditto.
9416         (veor3q_u16): Ditto.
9417         (vrax1q_u64): Ditto.
9418         (vxarq_u64): Ditto.
9419         (vbcaxq_u16): Ditto.
9420         * config/arm/types.md (crypto_sha512): New type attribute.
9421         (crypto_sha3): Ditto.
9422         (doc/invoke.texi): Document new sha3 option.
9423
9424 2018-01-10  Michael Collison  <michael.collison@arm.com>
9425
9426         * config/aarch64/aarch64-builtins.c:
9427         (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
9428         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
9429         (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
9430         (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
9431         * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
9432         (AARCH64_ISA_SM4): New ISA flag.
9433         (TARGET_SM4): New feature flag for sm4.
9434         * config/aarch64/aarch64-simd-builtins.def
9435         (aarch64_sm3ss1qv4si): Ditto.
9436         (aarch64_sm3tt1aq4si): Ditto.
9437         (aarch64_sm3tt1bq4si): Ditto.
9438         (aarch64_sm3tt2aq4si): Ditto.
9439         (aarch64_sm3tt2bq4si): Ditto.
9440         (aarch64_sm3partw1qv4si): Ditto.
9441         (aarch64_sm3partw2qv4si): Ditto.
9442         (aarch64_sm4eqv4si): Ditto.
9443         (aarch64_sm4ekeyqv4si): Ditto.
9444         * config/aarch64/aarch64-simd.md:
9445         (aarch64_sm3ss1qv4si): Ditto.
9446         (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
9447         (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
9448         (aarch64_sm4eqv4si): Ditto.
9449         (aarch64_sm4ekeyqv4si): Ditto.
9450         * config/aarch64/iterators.md (sm3tt_op): New int iterator.
9451         (sm3part_op): Ditto.
9452         (CRYPTO_SM3TT): Ditto.
9453         (CRYPTO_SM3PART): Ditto.
9454         (UNSPEC_SM3SS1): New unspec.
9455         (UNSPEC_SM3TT1A): Ditto.
9456         (UNSPEC_SM3TT1B): Ditto.
9457         (UNSPEC_SM3TT2A): Ditto.
9458         (UNSPEC_SM3TT2B): Ditto.
9459         (UNSPEC_SM3PARTW1): Ditto.
9460         (UNSPEC_SM3PARTW2): Ditto.
9461         (UNSPEC_SM4E): Ditto.
9462         (UNSPEC_SM4EKEY): Ditto.
9463         * config/aarch64/constraints.md (Ui2): New constraint.
9464         * config/aarch64/predicates.md (aarch64_imm2): New predicate.
9465         * config/arm/types.md (crypto_sm3): New type attribute.
9466         (crypto_sm4): Ditto.
9467         * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
9468         (vsm3tt1aq_u32): Ditto.
9469         (vsm3tt1bq_u32): Ditto.
9470         (vsm3tt2aq_u32): Ditto.
9471         (vsm3tt2bq_u32): Ditto.
9472         (vsm3partw1q_u32): Ditto.
9473         (vsm3partw2q_u32): Ditto.
9474         (vsm4eq_u32): Ditto.
9475         (vsm4ekeyq_u32): Ditto.
9476         (doc/invoke.texi): Document new sm4 option.
9477
9478 2018-01-10  Michael Collison  <michael.collison@arm.com>
9479
9480         * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
9481         * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
9482         (AARCH64_FL_FOR_ARCH8_4): New.
9483         (AARCH64_FL_V8_4): New flag.
9484         (doc/invoke.texi): Document new armv8.4-a option.
9485
9486 2018-01-10  Michael Collison  <michael.collison@arm.com>
9487
9488         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
9489         (__ARM_FEATURE_AES): Define if TARGET_AES is true.
9490         (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
9491         * config/aarch64/aarch64-option-extension.def: Add
9492         AARCH64_OPT_EXTENSION of 'sha2'.
9493         (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
9494         (crypto): Disable sha2 and aes if crypto disabled.
9495         (crypto): Enable aes and sha2 if enabled.
9496         (simd): Disable sha2 and aes if simd disabled.
9497         * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
9498         New flags.
9499         (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
9500         (TARGET_SHA2): New feature flag for sha2.
9501         (TARGET_AES): New feature flag for aes.
9502         * config/aarch64/aarch64-simd.md:
9503         (aarch64_crypto_aes<aes_op>v16qi): Make pattern
9504         conditional on TARGET_AES.
9505         (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
9506         (aarch64_crypto_sha1hsi): Make pattern conditional
9507         on TARGET_SHA2.
9508         (aarch64_crypto_sha1hv4si): Ditto.
9509         (aarch64_be_crypto_sha1hv4si): Ditto.
9510         (aarch64_crypto_sha1su1v4si): Ditto.
9511         (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
9512         (aarch64_crypto_sha1su0v4si): Ditto.
9513         (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
9514         (aarch64_crypto_sha256su0v4si): Ditto.
9515         (aarch64_crypto_sha256su1v4si): Ditto.
9516         (doc/invoke.texi): Document new aes and sha2 options.
9517
9518 2018-01-10  Martin Sebor  <msebor@redhat.com>
9519
9520         PR tree-optimization/83781
9521         * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
9522         as string arrays.
9523
9524 2018-01-11  Martin Sebor  <msebor@gmail.com>
9525             Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
9526
9527         PR tree-optimization/83501
9528         PR tree-optimization/81703
9529
9530         * tree-ssa-strlen.c (get_string_cst): Rename...
9531         (get_string_len): ...to this.  Handle global constants.
9532         (handle_char_store): Adjust.
9533
9534 2018-01-10  Kito Cheng  <kito.cheng@gmail.com>
9535             Jim Wilson  <jimw@sifive.com>
9536
9537         * config/riscv/riscv-protos.h (riscv_output_return): New.
9538         * config/riscv/riscv.c (struct machine_function): New naked_p field.
9539         (riscv_attribute_table, riscv_output_return),
9540         (riscv_handle_fndecl_attribute, riscv_naked_function_p),
9541         (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
9542         (riscv_compute_frame_info): Only compute frame->mask if not a naked
9543         function.
9544         (riscv_expand_prologue): Add early return for naked function.
9545         (riscv_expand_epilogue): Likewise.
9546         (riscv_function_ok_for_sibcall): Return false for naked function.
9547         (riscv_set_current_function): New.
9548         (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
9549         (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
9550         * config/riscv/riscv.md (simple_return): Call riscv_output_return.
9551         * doc/extend.texi (RISC-V Function Attributes): New.
9552
9553 2018-01-10  Michael Meissner  <meissner@linux.vnet.ibm.com>
9554
9555         * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
9556         check for 128-bit long double before checking TCmode.
9557         * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
9558         128-bit long doubles before checking TFmode or TCmode.
9559         (FLOAT128_IBM_P): Likewise.
9560
9561 2018-01-10  Martin Sebor  <msebor@redhat.com>
9562
9563         PR tree-optimization/83671
9564         * builtins.c (c_strlen): Unconditionally return zero for the empty
9565         string.
9566         Use -Warray-bounds for warnings.
9567         * gimple-fold.c (get_range_strlen): Handle non-constant lengths
9568         for non-constant array indices with COMPONENT_REF, arrays of
9569         arrays, and pointers to arrays.
9570         (gimple_fold_builtin_strlen): Determine and set length range for
9571         non-constant character arrays.
9572
9573 2018-01-10  Aldy Hernandez  <aldyh@redhat.com>
9574
9575         PR middle-end/81897
9576         * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
9577         empty blocks.
9578
9579 2018-01-10  Eric Botcazou  <ebotcazou@adacore.com>
9580
9581         * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
9582
9583 2018-01-10  Peter Bergner  <bergner@vnet.ibm.com>
9584
9585         PR target/83399
9586         * config/rs6000/rs6000.c (print_operand) <'y'>: Use
9587         VECTOR_MEM_ALTIVEC_OR_VSX_P.
9588         * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
9589         indexed_or_indirect_operand predicate.
9590         (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
9591         (*vsx_le_perm_load_v8hi): Likewise.
9592         (*vsx_le_perm_load_v16qi): Likewise.
9593         (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
9594         (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
9595         (*vsx_le_perm_store_v8hi): Likewise.
9596         (*vsx_le_perm_store_v16qi): Likewise.
9597         (eight unnamed splitters): Likewise.
9598
9599 2018-01-10  Peter Bergner  <bergner@vnet.ibm.com>
9600
9601         * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
9602         * config/rs6000/emmintrin.h: Likewise.
9603         * config/rs6000/mmintrin.h: Likewise.
9604         * config/rs6000/xmmintrin.h: Likewise.
9605
9606 2018-01-10  David Malcolm  <dmalcolm@redhat.com>
9607
9608         PR c++/43486
9609         * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
9610         "public_flag".
9611         * tree.c (tree_nop_conversion): Return true for location wrapper
9612         nodes.
9613         (maybe_wrap_with_location): New function.
9614         (selftest::check_strip_nops): New function.
9615         (selftest::test_location_wrappers): New function.
9616         (selftest::tree_c_tests): Call it.
9617         * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
9618         (maybe_wrap_with_location): New decl.
9619         (EXPR_LOCATION_WRAPPER_P): New macro.
9620         (location_wrapper_p): New inline function.
9621         (tree_strip_any_location_wrapper): New inline function.
9622
9623 2018-01-10  H.J. Lu  <hongjiu.lu@intel.com>
9624
9625         PR target/83735
9626         * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
9627         stack_realign_offset for the largest alignment of stack slot
9628         actually used.
9629         (ix86_find_max_used_stack_alignment): New function.
9630         (ix86_finalize_stack_frame_flags): Use it.  Set
9631         max_used_stack_alignment if we don't realign stack.
9632         * config/i386/i386.h (machine_function): Add
9633         max_used_stack_alignment.
9634
9635 2018-01-10  Christophe Lyon  <christophe.lyon@linaro.org>
9636
9637         * config/arm/arm.opt (-mbranch-cost): New option.
9638         * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
9639         account.
9640
9641 2018-01-10  Segher Boessenkool  <segher@kernel.crashing.org>
9642
9643         PR target/83629
9644         * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
9645         load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
9646
9647 2018-01-10  Richard Biener  <rguenther@suse.de>
9648
9649         PR debug/83765
9650         * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
9651         early out so it also covers the case where we have a non-NULL
9652         origin.
9653
9654 2018-01-10  Richard Sandiford  <richard.sandiford@linaro.org>
9655
9656         PR tree-optimization/83753
9657         * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
9658         for non-strided grouped accesses if the number of elements is 1.
9659
9660 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
9661
9662         PR target/81616
9663         * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
9664         * i386.h (TARGET_USE_GATHER): Define.
9665         * x86-tune.def (X86_TUNE_USE_GATHER): New.
9666
9667 2018-01-10  Martin Liska  <mliska@suse.cz>
9668
9669         PR bootstrap/82831
9670         * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
9671         * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
9672         partitioning.
9673         * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
9674         CLEANUP_NO_PARTITIONING is not set.
9675
9676 2018-01-10  Richard Sandiford  <richard.sandiford@linaro.org>
9677
9678         * doc/rtl.texi: Remove documentation of (const ...) wrappers
9679         for vectors, as a partial revert of r254296.
9680         * rtl.h (const_vec_p): Delete.
9681         (const_vec_duplicate_p): Don't test for vector CONSTs.
9682         (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
9683         * expmed.c (make_tree): Likewise.
9684
9685         Revert:
9686         * common.md (E, F): Use CONSTANT_P instead of checking for
9687         CONST_VECTOR.
9688         * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
9689         checking for CONST_VECTOR.
9690
9691 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
9692
9693         PR middle-end/83575
9694         * predict.c (force_edge_cold): Handle in more sane way edges
9695         with no prediction.
9696
9697 2018-01-09  Carl Love  <cel@us.ibm.com>
9698
9699         * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
9700         V4SI, V4SF types.
9701         (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
9702         * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
9703         VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
9704         VMRGOW_V2DI, VMRGOW_V2DF.  Remove definition for VMRGOW.
9705         * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
9706         P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW):  Add definitions.
9707         * config/rs6000/rs6000-protos.h: Add extern defition for
9708         rs6000_generate_float2_double_code.
9709         * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
9710         function.
9711         * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
9712         (float2_v2df): Add define_expand.
9713
9714 2018-01-09  Uros Bizjak  <ubizjak@gmail.com>
9715
9716         PR target/83628
9717         * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
9718         op_mode in the force_to_mode call.
9719
9720 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
9721
9722         * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
9723         instead of checking each element individually.
9724         (aarch64_evpc_uzp): Likewise.
9725         (aarch64_evpc_zip): Likewise.
9726         (aarch64_evpc_ext): Likewise.
9727         (aarch64_evpc_rev): Likewise.
9728         (aarch64_evpc_dup): Test the encoding for a single duplicated element,
9729         instead of checking each element individually.  Return true without
9730         generating rtl if
9731         (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
9732         whether all selected elements come from the same input, instead of
9733         checking each element individually.  Remove calls to gen_rtx_REG,
9734         start_sequence and end_sequence and instead assert that no rtl is
9735         generated.
9736
9737 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
9738
9739         * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
9740         order of HIGH and CONST checks.
9741
9742 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
9743
9744         * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
9745         if the destination isn't an SSA_NAME.
9746
9747 2018-01-09  Richard Biener  <rguenther@suse.de>
9748
9749         PR tree-optimization/83668
9750         * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
9751         move prologue...
9752         (canonicalize_loop_form): ... here, renamed from ...
9753         (canonicalize_loop_closed_ssa_form): ... this and amended to
9754         swap successor edges for loop exit blocks to make us use
9755         the RPO order we need for initial schedule generation.
9756
9757 2018-01-09  Joseph Myers  <joseph@codesourcery.com>
9758
9759         PR tree-optimization/64811
9760         * match.pd: When optimizing comparisons with Inf, avoid
9761         introducing or losing exceptions from comparisons with NaN.
9762
9763 2018-01-09  Martin Liska  <mliska@suse.cz>
9764
9765         PR sanitizer/82517
9766         * asan.c (shadow_mem_size): Add gcc_assert.
9767
9768 2018-01-09  Georg-Johann Lay  <avr@gjlay.de>
9769
9770         Don't save registers in main().
9771
9772         PR target/83738
9773         * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
9774         * config/avr/avr.opt (-mmain-is-OS_task): New target option.
9775         * config/avr/avr.c (avr_set_current_function): Don't error if
9776         naked, OS_task or OS_main are specified at the same time.
9777         (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
9778         OS_main.
9779         (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
9780         attribute.
9781         * common/config/avr/avr-common.c (avr_option_optimization_table):
9782         Switch on -mmain-is-OS_task for optimizing compilations.
9783
9784 2018-01-09  Richard Biener  <rguenther@suse.de>
9785
9786         PR tree-optimization/83572
9787         * graphite.c: Include cfganal.h.
9788         (graphite_transform_loops): Connect infinite loops to exit
9789         and remove fake edges at the end.
9790
9791 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
9792
9793         * ipa-inline.c (edge_badness): Revert accidental checkin.
9794
9795 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
9796
9797         PR ipa/80763
9798         * ipa-comdats.c (set_comdat_group): Only set comdat group of real
9799         symbols; not inline clones.
9800
9801 2018-01-09  Jakub Jelinek  <jakub@redhat.com>
9802
9803         PR target/83507
9804         * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
9805         hard registers.  Formatting fixes.
9806
9807         PR preprocessor/83722
9808         * gcc.c (try_generate_repro): Pass
9809         &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
9810         &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
9811         do_report_bug.
9812
9813 2018-01-08  Monk Chiang  <sh.chiang04@gmail.com>
9814             Kito Cheng  <kito.cheng@gmail.com>
9815
9816         * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
9817         (riscv_leaf_function_p): Delete.
9818         (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
9819
9820 2018-01-08  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
9821
9822         * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
9823         function.
9824         (do_ifelse): New function.
9825         (do_isel): New function.
9826         (do_sub3): New function.
9827         (do_add3): New function.
9828         (do_load_mask_compare): New function.
9829         (do_overlap_load_compare): New function.
9830         (expand_compare_loop): New function.
9831         (expand_block_compare): Call expand_compare_loop() when appropriate.
9832         * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
9833         option description.
9834         (-mblock-compare-inline-loop-limit): New option.
9835
9836 2018-01-08  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
9837
9838         PR target/83677
9839         * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
9840         Reverse order of second and third operands in first alternative.
9841         * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
9842         of first and second elements in UNSPEC_VPERMR vector.
9843         (altivec_expand_vec_perm_le): Likewise.
9844
9845 2018-01-08  Jeff Law  <law@redhat.com>
9846
9847         PR rtl-optimizatin/81308
9848         * tree-switch-conversion.c (cfg_altered): New file scoped static.
9849         (process_switch): If group_case_labels makes a change, then set
9850         cfg_altered.
9851         (pass_convert_switch::execute): If a switch is converted, then
9852         set cfg_altered.  Return TODO_cfg_cleanup if cfg_altered is true.
9853
9854         PR rtl-optimization/81308
9855         * recog.c (split_all_insns): Conditionally cleanup the CFG after
9856         splitting insns.
9857
9858 2018-01-08  Vidya Praveen  <vidyapraveen@arm.com>
9859
9860         PR target/83663 - Revert r255946
9861         * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
9862         generation for cases where splatting a value is not useful.
9863         * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
9864         across a vec_duplicate and a paradoxical subreg forming a vector
9865         mode to a vec_concat.
9866
9867 2018-01-08  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
9868
9869         * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
9870         -march=armv8.3-a variants.
9871         * config/arm/t-multilib: Likewise.
9872         * config/arm/t-arm-elf: Likewise.  Handle dotprod extension.
9873
9874 2018-01-08  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
9875
9876         * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
9877         to generate rtl.
9878         (cceq_ior_compare_complement): Give it a name so I can use it, and
9879         change boolean_or_operator predicate to boolean_operator so it can
9880         be used to generate a crand.
9881         (eqne): New code iterator.
9882         (bd/bd_neg): New code_attrs.
9883         (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
9884         a single define_insn.
9885         (<bd>tf_<mode>): A new insn pattern for the conditional form branch
9886         decrement (bdnzt/bdnzf/bdzt/bdzf).
9887         * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
9888         with the new names of the branch decrement patterns, and added the
9889         names of the branch decrement conditional patterns.
9890
9891 2018-01-08  Richard Biener  <rguenther@suse.de>
9892
9893         PR tree-optimization/83563
9894         * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
9895         cache.
9896
9897 2018-01-08  Richard Biener  <rguenther@suse.de>
9898
9899         PR middle-end/83713
9900         * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
9901
9902 2018-01-08  Richard Biener  <rguenther@suse.de>
9903
9904         PR tree-optimization/83685
9905         * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
9906         references to abnormals.
9907
9908 2018-01-08  Richard Biener  <rguenther@suse.de>
9909
9910         PR lto/83719
9911         * dwarf2out.c (output_indirect_strings): Handle empty
9912         skeleton_debug_str_hash.
9913         (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
9914
9915 2018-01-08  Claudiu Zissulescu  <claziss@synopsys.com>
9916
9917         * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
9918         (emit_store_direct): Likewise.
9919         (arc_trampoline_adjust_address): Likewise.
9920         (arc_asm_trampoline_template): New function.
9921         (arc_initialize_trampoline): Use asm_trampoline_template.
9922         (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
9923         * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
9924         * config/arc/arc.md (flush_icache): Delete pattern.
9925
9926 2018-01-08  Claudiu Zissulescu  <claziss@synopsys.com>
9927
9928         * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
9929         * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
9930         munaligned-access.
9931
9932 2018-01-08  Sebastian Huber  <sebastian.huber@embedded-brains.de>
9933
9934         PR target/83681
9935         * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
9936         by not USED_FOR_TARGET.
9937         (make_pass_resolve_sw_modes): Likewise.
9938
9939 2018-01-08  Sebastian Huber  <sebastian.huber@embedded-brains.de>
9940
9941         * config/nios2/nios2.h (nios2_section_threshold): Guard by not
9942         USED_FOR_TARGET.
9943
9944 2018-01-08  Richard Biener  <rguenther@suse.de>
9945
9946         PR middle-end/83580
9947         * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
9948
9949 2018-01-08  Richard Biener  <rguenther@suse.de>
9950
9951         PR middle-end/83517
9952         * match.pd ((t * 2) / 2) -> t): Add missing :c.
9953
9954 2018-01-06  Aldy Hernandez  <aldyh@redhat.com>
9955
9956         PR middle-end/81897
9957         * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
9958         basic blocks with a small number of successors.
9959         (convert_control_dep_chain_into_preds): Improve handling of
9960         forwarder blocks.
9961         (dump_predicates): Split apart into...
9962         (dump_pred_chain): ...here...
9963         (dump_pred_info): ...and here.
9964         (can_one_predicate_be_invalidated_p): Add debugging printfs.
9965         (can_chain_union_be_invalidated_p): Improve check for invalidation
9966         of paths.
9967         (uninit_uses_cannot_happen): Avoid unnecessary if
9968         convert_control_dep_chain_into_preds yielded nothing.
9969
9970 2018-01-06  Martin Sebor  <msebor@redhat.com>
9971
9972         PR tree-optimization/83640
9973         * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
9974         subtracting negative offset from size.
9975         (builtin_access::overlap): Adjust offset bounds of the access to fall
9976         within the size of the object if possible.
9977
9978 2018-01-06  Richard Sandiford  <richard.sandiford@linaro.org>
9979
9980         PR rtl-optimization/83699
9981         * expmed.c (extract_bit_field_1): Restrict the vector usage of
9982         extract_bit_field_as_subreg to cases in which the extracted
9983         value is also a vector.
9984
9985         * lra-constraints.c (process_alt_operands): Test for the equivalence
9986         substitutions when detecting a possible reload cycle.
9987
9988 2018-01-06  Jakub Jelinek  <jakub@redhat.com>
9989
9990         PR debug/83480
9991         * toplev.c (process_options): Don't enable debug_nonbind_markers_p
9992         by default if flag_selective_schedling{,2}.  Formatting fixes.
9993
9994         PR rtl-optimization/83682
9995         * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
9996         if it has non-VECTOR_MODE element mode.
9997         (vec_duplicate_p): Likewise.
9998
9999         PR middle-end/83694
10000         * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
10001         and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
10002
10003 2018-01-05  Jakub Jelinek  <jakub@redhat.com>
10004
10005         PR target/83604
10006         * config/i386/i386-builtin.def
10007         (__builtin_ia32_vgf2p8affineinvqb_v64qi,
10008         __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
10009         Require also OPTION_MASK_ISA_AVX512F in addition to
10010         OPTION_MASK_ISA_GFNI.
10011         (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
10012         __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
10013         OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
10014         to OPTION_MASK_ISA_GFNI.
10015         (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
10016         OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
10017         OPTION_MASK_ISA_AVX512BW.
10018         (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
10019         OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
10020         addition to OPTION_MASK_ISA_GFNI.
10021         (__builtin_ia32_vgf2p8affineinvqb_v16qi,
10022         __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
10023         Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
10024         to OPTION_MASK_ISA_GFNI.
10025         * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
10026         a requirement for all ISAs rather than any of them with a few
10027         exceptions.
10028         (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
10029         processing.
10030         (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
10031         bitmasks to be enabled with 3 exceptions, instead of requiring any
10032         enabled ISA with lots of exceptions.
10033         * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
10034         vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
10035         Change avx512bw in isa attribute to avx512f.
10036         * config/i386/sgxintrin.h: Add license boilerplate.
10037         * config/i386/vaesintrin.h: Likewise.  Fix macro spelling __AVX512F
10038         to __AVX512F__ and __AVX512VL to __AVX512VL__.
10039         (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
10040         _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
10041         defined.
10042         * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
10043         _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
10044         temporarily sse2 rather than sse if not enabled already.
10045
10046         PR target/83604
10047         * config/i386/sse.md (VI248_VLBW): Rename to ...
10048         (VI248_AVX512VL): ... this.  Don't guard V32HI with TARGET_AVX512BW.
10049         (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
10050         vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
10051         vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
10052         vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
10053         mode iterator instead of VI248_VLBW.
10054
10055 2018-01-05  Jan Hubicka  <hubicka@ucw.cz>
10056
10057         * ipa-fnsummary.c (record_modified_bb_info): Add OP.
10058         (record_modified): Skip clobbers; add debug output.
10059         (param_change_prob): Use sreal frequencies.
10060
10061 2018-01-05  Richard Sandiford  <richard.sandiford@linaro.org>
10062
10063         * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
10064         punt for user-aligned variables.
10065
10066 2018-01-05  Richard Sandiford  <richard.sandiford@linaro.org>
10067
10068         * tree-chrec.c (chrec_contains_symbols): Return true for
10069         POLY_INT_CST.
10070
10071 2018-01-05  Sudakshina Das  <sudi.das@arm.com>
10072
10073         PR target/82439
10074         * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
10075         of (x|y) == x for BICS pattern.
10076
10077 2018-01-05  Jakub Jelinek  <jakub@redhat.com>
10078
10079         PR tree-optimization/83605
10080         * gimple-ssa-strength-reduction.c: Include tree-eh.h.
10081         (find_candidates_dom_walker::before_dom_children): Ignore stmts that
10082         can throw.
10083
10084 2018-01-05  Sebastian Huber  <sebastian.huber@embedded-brains.de>
10085
10086         * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
10087         * config/epiphany/rtems.h: New file.
10088
10089 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
10090             Uros Bizjak  <ubizjak@gmail.com>
10091
10092         PR target/83554
10093         * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
10094         QIreg_operand instead of register_operand predicate.
10095         * config/i386/i386.c (ix86_rop_should_change_byte_p,
10096         set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
10097         comments instead of -fmitigate[-_]rop.
10098
10099 2018-01-04  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
10100
10101         PR bootstrap/81926
10102         * cgraphunit.c (symbol_table::compile): Switch to text_section
10103         before calling assembly_start debug hook.
10104         * run-rtl-passes.c (run_rtl_passes): Likewise.
10105         Include output.h.
10106
10107 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
10108
10109         * tree-vrp.c (extract_range_from_binary_expr_1): Check
10110         range_int_cst_p rather than !symbolic_range_p before calling
10111         extract_range_from_multiplicative_op_1.
10112
10113 2018-01-04  Jeff Law  <law@redhat.com>
10114
10115         * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
10116         redundant test in assertion.
10117
10118 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
10119
10120         * doc/rtl.texi: Document machine_mode wrapper classes.
10121
10122 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
10123
10124         * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
10125         using tree_to_uhwi.
10126
10127 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
10128
10129         * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
10130         the VEC_PERM_EXPR fold to fail.
10131
10132 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
10133
10134         PR debug/83585
10135         * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
10136         to switched_sections.
10137
10138 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
10139
10140         PR target/83680
10141         * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
10142         test for d.testing.
10143
10144 2018-01-04  Peter Bergner  <bergner@vnet.ibm.com>
10145
10146         PR target/83387
10147         * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
10148         allow arguments in FP registers if TARGET_HARD_FLOAT is false.
10149
10150 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
10151
10152         PR debug/83666
10153         * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
10154         is BLKmode and bitpos not zero or mode change is needed.
10155
10156 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
10157
10158         PR target/83675
10159         * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
10160         TARGET_VIS2.
10161
10162 2018-01-04  Uros Bizjak  <ubizjak@gmail.com>
10163
10164         PR target/83628
10165         * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
10166         instead of MULT rtx.  Update all corresponding splitters.
10167         (*saddl_se): Ditto.
10168         (*ssub<modesuffix>): Ditto.
10169         (*ssubl_se): Ditto.
10170         (*cmp_sadd_di): Update split patterns.
10171         (*cmp_sadd_si): Ditto.
10172         (*cmp_sadd_sidi): Ditto.
10173         (*cmp_ssub_di): Ditto.
10174         (*cmp_ssub_si): Ditto.
10175         (*cmp_ssub_sidi): Ditto.
10176         * config/alpha/predicates.md (const23_operand): New predicate.
10177         * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
10178         Look for ASHIFT, not MULT inner operand.
10179         (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
10180
10181 2018-01-04  Martin Liska  <mliska@suse.cz>
10182
10183         PR gcov-profile/83669
10184         * gcov.c (output_intermediate_file): Add version to intermediate
10185         gcov file.
10186         * doc/gcov.texi: Document new field 'version' in intermediate
10187         file format. Fix location of '-k' option of gcov command.
10188
10189 2018-01-04  Martin Liska  <mliska@suse.cz>
10190
10191         PR ipa/82352
10192         * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
10193
10194 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
10195
10196         * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
10197
10198 2018-01-03  Martin Sebor  <msebor@redhat.com>
10199
10200         PR tree-optimization/83655
10201         * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
10202         checking calls with invalid arguments.
10203
10204 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10205
10206         * tree-vect-stmts.c (vect_get_store_rhs): New function.
10207         (vectorizable_mask_load_store): Delete.
10208         (vectorizable_call): Return false for masked loads and stores.
10209         (vectorizable_store): Handle IFN_MASK_STORE.  Use vect_get_store_rhs
10210         instead of gimple_assign_rhs1.
10211         (vectorizable_load): Handle IFN_MASK_LOAD.
10212         (vect_transform_stmt): Don't set is_store for call_vec_info_type.
10213
10214 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10215
10216         * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
10217         split out from..,
10218         (vectorizable_mask_load_store): ...here.
10219         (vectorizable_load): ...and here.
10220
10221 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10222
10223         * tree-vect-stmts.c (vect_build_all_ones_mask)
10224         (vect_build_zero_merge_argument): New functions, split out from...
10225         (vectorizable_load): ...here.
10226
10227 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10228
10229         * tree-vect-stmts.c (vect_check_store_rhs): New function,
10230         split out from...
10231         (vectorizable_mask_load_store): ...here.
10232         (vectorizable_store): ...and here.
10233
10234 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10235
10236         * tree-vect-stmts.c (vect_check_load_store_mask): New function,
10237         split out from...
10238         (vectorizable_mask_load_store): ...here.
10239
10240 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10241
10242         * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
10243         (vect_model_store_cost): Take a vec_load_store_type instead of a
10244         vect_def_type.
10245         * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
10246         (vect_model_store_cost): Take a vec_load_store_type instead of a
10247         vect_def_type.
10248         (vectorizable_mask_load_store): Update accordingly.
10249         (vectorizable_store): Likewise.
10250         * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
10251
10252 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10253
10254         * tree-vect-loop.c (vect_transform_loop): Stub out scalar
10255         IFN_MASK_LOAD calls here rather than...
10256         * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
10257
10258 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10259             Alan Hayward  <alan.hayward@arm.com>
10260             David Sherwood  <david.sherwood@arm.com>
10261
10262         * expmed.c (extract_bit_field_1): For vector extracts,
10263         fall back to extract_bit_field_as_subreg if vec_extract
10264         isn't available.
10265
10266 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10267             Alan Hayward  <alan.hayward@arm.com>
10268             David Sherwood  <david.sherwood@arm.com>
10269
10270         * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
10271         they are variable or constant sized.
10272         (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
10273         slots for constant-sized data.
10274
10275 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10276             Alan Hayward  <alan.hayward@arm.com>
10277             David Sherwood  <david.sherwood@arm.com>
10278
10279         * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
10280         handling COND_EXPRs with boolean comparisons, try to find a better
10281         basis for the mask type than the boolean itself.
10282
10283 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10284
10285         * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
10286         is calculated and how it can be overridden.
10287         * genmodes.c (max_bitsize_mode_any_mode): New variable.
10288         (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
10289         if defined.
10290         (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
10291         if nonzero.
10292
10293 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10294             Alan Hayward  <alan.hayward@arm.com>
10295             David Sherwood  <david.sherwood@arm.com>
10296
10297         * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
10298         Remove the mode argument.
10299         (aarch64_simd_valid_immediate): Remove the mode and inverse
10300         arguments.
10301         * config/aarch64/iterators.md (bitsize): New iterator.
10302         * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
10303         (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
10304         * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
10305         aarch64_simd_valid_immediate.
10306         * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
10307         (aarch64_reg_or_bic_imm): Likewise.
10308         * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
10309         with an insn_type enum and msl with a modifier_type enum.
10310         Replace element_width with a scalar_mode.  Change the shift
10311         to unsigned int.  Add constructors for scalar_float_mode and
10312         scalar_int_mode elements.
10313         (aarch64_vect_float_const_representable_p): Delete.
10314         (aarch64_can_const_movi_rtx_p)
10315         (aarch64_simd_scalar_immediate_valid_for_move)
10316         (aarch64_simd_make_constant): Update call to
10317         aarch64_simd_valid_immediate.
10318         (aarch64_advsimd_valid_immediate_hs): New function.
10319         (aarch64_advsimd_valid_immediate): Likewise.
10320         (aarch64_simd_valid_immediate): Remove mode and inverse
10321         arguments.  Rewrite to use the above.  Use const_vec_duplicate_p
10322         to detect duplicated constants and use aarch64_float_const_zero_rtx_p
10323         and aarch64_float_const_representable_p on the result.
10324         (aarch64_output_simd_mov_immediate): Remove mode argument.
10325         Update call to aarch64_simd_valid_immediate and use of
10326         simd_immediate_info.
10327         (aarch64_output_scalar_simd_mov_immediate): Update call
10328         accordingly.
10329
10330 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10331             Alan Hayward  <alan.hayward@arm.com>
10332             David Sherwood  <david.sherwood@arm.com>
10333
10334         * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
10335         (mode_nunits): Likewise CONST_MODE_NUNITS.
10336         * machmode.def (ADJUST_NUNITS): Document.
10337         * genmodes.c (mode_data::need_nunits_adj): New field.
10338         (blank_mode): Update accordingly.
10339         (adj_nunits): New variable.
10340         (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
10341         parameter.
10342         (emit_mode_size_inline): Set need_bytesize_adj for all modes
10343         listed in adj_nunits.
10344         (emit_mode_nunits_inline): Set need_nunits_adj for all modes
10345         listed in adj_nunits.  Don't emit case statements for such modes.
10346         (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
10347         and CONST_MODE_PRECISION.  Make CONST_MODE_SIZE expand to
10348         nothing if adj_nunits is nonnull.
10349         (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
10350         (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
10351         (emit_mode_fbit): Update use of print_maybe_const_decl.
10352         (emit_move_size): Likewise.  Treat the array as non-const
10353         if adj_nunits.
10354         (emit_mode_adjustments): Handle adj_nunits.
10355
10356 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10357
10358         * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
10359         * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
10360         (VECTOR_MODES): Use it.
10361         (make_vector_modes): Take the prefix as an argument.
10362
10363 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10364             Alan Hayward  <alan.hayward@arm.com>
10365             David Sherwood  <david.sherwood@arm.com>
10366
10367         * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
10368         * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
10369         for MODE_VECTOR_BOOL.
10370         * machmode.def (VECTOR_BOOL_MODE): Document.
10371         * genmodes.c (VECTOR_BOOL_MODE): New macro.
10372         (make_vector_bool_mode): New function.
10373         (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
10374         MODE_VECTOR_BOOL.
10375         * lto-streamer-in.c (lto_input_mode_table): Likewise.
10376         * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
10377         Likewise.
10378         * stor-layout.c (int_mode_for_mode): Likewise.
10379         * tree.c (build_vector_type_for_mode): Likewise.
10380         * varasm.c (output_constant_pool_2): Likewise.
10381         * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
10382         CONSTM1_RTX (BImode) are the same thing.  Initialize const_tiny_rtx
10383         for MODE_VECTOR_BOOL.
10384         * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
10385         of mode class checks.
10386         * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
10387         instead of a list of mode class checks.
10388         (expand_vector_scalar_condition): Likewise.
10389         (type_for_widest_vector_mode): Handle BImode as an inner mode.
10390
10391 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10392             Alan Hayward  <alan.hayward@arm.com>
10393             David Sherwood  <david.sherwood@arm.com>
10394
10395         * machmode.h (mode_size): Change from unsigned short to
10396         poly_uint16_pod.
10397         (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
10398         (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
10399         or if measurement_type is not polynomial.
10400         (fixed_size_mode::includes_p): Check for constant-sized modes.
10401         * genmodes.c (emit_mode_size_inline): Make mode_size_inline
10402         return a poly_uint16 rather than an unsigned short.
10403         (emit_mode_size): Change the type of mode_size from unsigned short
10404         to poly_uint16_pod.  Use ZERO_COEFFS for the initializer.
10405         (emit_mode_adjustments): Cope with polynomial vector sizes.
10406         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
10407         for GET_MODE_SIZE.
10408         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
10409         for GET_MODE_SIZE.
10410         * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
10411         * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
10412         * caller-save.c (setup_save_areas): Likewise.
10413         (replace_reg_with_saved_mem): Likewise.
10414         * calls.c (emit_library_call_value_1): Likewise.
10415         * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
10416         * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
10417         (gen_lowpart_for_combine): Likewise.
10418         * convert.c (convert_to_integer_1): Likewise.
10419         * cse.c (equiv_constant, cse_insn): Likewise.
10420         * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
10421         (cselib_subst_to_values): Likewise.
10422         * dce.c (word_dce_process_block): Likewise.
10423         * df-problems.c (df_word_lr_mark_ref): Likewise.
10424         * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
10425         * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
10426         (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
10427         (rtl_for_decl_location): Likewise.
10428         * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
10429         * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
10430         * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
10431         (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
10432         (expand_expr_real_1): Likewise.
10433         * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
10434         (pad_below): Likewise.
10435         * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
10436         * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
10437         * ira.c (get_subreg_tracking_sizes): Likewise.
10438         * ira-build.c (ira_create_allocno_objects): Likewise.
10439         * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
10440         (ira_sort_regnos_for_alter_reg): Likewise.
10441         * ira-costs.c (record_operand_costs): Likewise.
10442         * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
10443         (resolve_simple_move): Likewise.
10444         * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
10445         (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
10446         (lra_constraints): Likewise.
10447         (CONST_POOL_OK_P): Reject variable-sized modes.
10448         * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
10449         (add_pseudo_to_slot, lra_spill): Likewise.
10450         * omp-low.c (omp_clause_aligned_alignment): Likewise.
10451         * optabs-query.c (get_best_extraction_insn): Likewise.
10452         * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
10453         * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
10454         (expand_mult_highpart, valid_multiword_target_p): Likewise.
10455         * recog.c (offsettable_address_addr_space_p): Likewise.
10456         * regcprop.c (maybe_mode_change): Likewise.
10457         * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
10458         * regrename.c (build_def_use): Likewise.
10459         * regstat.c (dump_reg_info): Likewise.
10460         * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
10461         (find_reloads, find_reloads_subreg_address): Likewise.
10462         * reload1.c (eliminate_regs_1): Likewise.
10463         * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
10464         * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
10465         (simplify_binary_operation_1, simplify_subreg): Likewise.
10466         * targhooks.c (default_function_arg_padding): Likewise.
10467         (default_hard_regno_nregs, default_class_max_nregs): Likewise.
10468         * tree-cfg.c (verify_gimple_assign_binary): Likewise.
10469         (verify_gimple_assign_ternary): Likewise.
10470         * tree-inline.c (estimate_move_cost): Likewise.
10471         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
10472         * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
10473         (get_address_cost_ainc): Likewise.
10474         * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
10475         (vect_supportable_dr_alignment): Likewise.
10476         * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
10477         (vectorizable_reduction): Likewise.
10478         * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
10479         (vectorizable_operation, vectorizable_load): Likewise.
10480         * tree.c (build_same_sized_truth_vector_type): Likewise.
10481         * valtrack.c (cleanup_auto_inc_dec): Likewise.
10482         * var-tracking.c (emit_note_insn_var_location): Likewise.
10483         * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
10484         (ADDR_VEC_ALIGN): Likewise.
10485
10486 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10487             Alan Hayward  <alan.hayward@arm.com>
10488             David Sherwood  <david.sherwood@arm.com>
10489
10490         * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
10491         unsigned short.
10492         (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
10493         or if measurement_type is polynomial.
10494         * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
10495         * combine.c (make_extraction): Likewise.
10496         * dse.c (find_shift_sequence): Likewise.
10497         * dwarf2out.c (mem_loc_descriptor): Likewise.
10498         * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
10499         (extract_bit_field, extract_low_bits): Likewise.
10500         * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
10501         (optimize_bitfield_assignment_op, expand_assignment): Likewise.
10502         (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
10503         * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
10504         * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
10505         * reload.c (find_reloads): Likewise.
10506         * reload1.c (alter_reg): Likewise.
10507         * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
10508         * targhooks.c (default_secondary_memory_needed_mode): Likewise.
10509         * tree-if-conv.c (predicate_mem_writes): Likewise.
10510         * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
10511         * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
10512         * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
10513         * valtrack.c (dead_debug_insert_temp): Likewise.
10514         * varasm.c (mergeable_constant_section): Likewise.
10515         * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
10516
10517 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10518             Alan Hayward  <alan.hayward@arm.com>
10519             David Sherwood  <david.sherwood@arm.com>
10520
10521         * expr.c (expand_assignment): Cope with polynomial mode sizes
10522         when assigning to a CONCAT.
10523
10524 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10525             Alan Hayward  <alan.hayward@arm.com>
10526             David Sherwood  <david.sherwood@arm.com>
10527
10528         * machmode.h (mode_precision): Change from unsigned short to
10529         poly_uint16_pod.
10530         (mode_to_precision): Return a poly_uint16 rather than an unsigned
10531         short.
10532         (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
10533         or if measurement_type is not polynomial.
10534         (HWI_COMPUTABLE_MODE_P): Turn into a function.  Optimize the case
10535         in which the mode is already known to be a scalar_int_mode.
10536         * genmodes.c (emit_mode_precision): Change the type of mode_precision
10537         from unsigned short to poly_uint16_pod.  Use ZERO_COEFFS for the
10538         initializer.
10539         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
10540         for GET_MODE_PRECISION.
10541         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
10542         for GET_MODE_PRECISION.
10543         * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
10544         as polynomial.
10545         (try_combine, find_split_point, combine_simplify_rtx): Likewise.
10546         (expand_field_assignment, make_extraction): Likewise.
10547         (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
10548         (get_last_value): Likewise.
10549         * convert.c (convert_to_integer_1): Likewise.
10550         * cse.c (cse_insn): Likewise.
10551         * expr.c (expand_expr_real_1): Likewise.
10552         * lra-constraints.c (simplify_operand_subreg): Likewise.
10553         * optabs-query.c (can_atomic_load_p): Likewise.
10554         * optabs.c (expand_atomic_load): Likewise.
10555         (expand_atomic_store): Likewise.
10556         * ree.c (combine_reaching_defs): Likewise.
10557         * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
10558         * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
10559         * tree.h (type_has_mode_precision_p): Likewise.
10560         * ubsan.c (instrument_si_overflow): Likewise.
10561
10562 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10563             Alan Hayward  <alan.hayward@arm.com>
10564             David Sherwood  <david.sherwood@arm.com>
10565
10566         * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
10567         polynomial numbers of units.
10568         (SET_TYPE_VECTOR_SUBPARTS): Likewise.
10569         (valid_vector_subparts_p): New function.
10570         (build_vector_type): Remove temporary shim and take the number
10571         of units as a poly_uint64 rather than an int.
10572         (build_opaque_vector_type): Take the number of units as a
10573         poly_uint64 rather than an int.
10574         * tree.c (build_vector_from_ctor): Handle polynomial
10575         TYPE_VECTOR_SUBPARTS.
10576         (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
10577         (uniform_vector_p, vector_type_mode, build_vector): Likewise.
10578         (build_vector_from_val): If the number of units is variable,
10579         use build_vec_duplicate_cst for constant operands and
10580         VEC_DUPLICATE_EXPR otherwise.
10581         (make_vector_type): Remove temporary is_constant ().
10582         (build_vector_type, build_opaque_vector_type): Take the number of
10583         units as a poly_uint64 rather than an int.
10584         (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
10585         VECTOR_CST_NELTS.
10586         * cfgexpand.c (expand_debug_expr): Likewise.
10587         * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
10588         (store_constructor, expand_expr_real_1): Likewise.
10589         (const_scalar_mask_from_tree): Likewise.
10590         * fold-const-call.c (fold_const_reduction): Likewise.
10591         * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
10592         (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
10593         (native_encode_vector, vec_cst_ctor_to_array): Likewise.
10594         (fold_relational_const): Likewise.
10595         (native_interpret_vector): Likewise.  Change the size from an
10596         int to an unsigned int.
10597         * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
10598         TYPE_VECTOR_SUBPARTS.
10599         (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
10600         (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
10601         duplicating a non-constant operand into a variable-length vector.
10602         * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
10603         TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
10604         * ipa-icf.c (sem_variable::equals): Likewise.
10605         * match.pd: Likewise.
10606         * omp-simd-clone.c (simd_clone_subparts): Likewise.
10607         * print-tree.c (print_node): Likewise.
10608         * stor-layout.c (layout_type): Likewise.
10609         * targhooks.c (default_builtin_vectorization_cost): Likewise.
10610         * tree-cfg.c (verify_gimple_comparison): Likewise.
10611         (verify_gimple_assign_binary): Likewise.
10612         (verify_gimple_assign_ternary): Likewise.
10613         (verify_gimple_assign_single): Likewise.
10614         * tree-pretty-print.c (dump_generic_node): Likewise.
10615         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
10616         (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
10617         * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
10618         (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
10619         (vect_shift_permute_load_chain): Likewise.
10620         * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
10621         (expand_vector_condition, optimize_vector_constructor): Likewise.
10622         (lower_vec_perm, get_compute_type): Likewise.
10623         * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
10624         (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
10625         * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
10626         (vect_recog_mask_conversion_pattern): Likewise.
10627         * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
10628         (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
10629         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
10630         (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
10631         (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
10632         (vectorizable_shift, vectorizable_operation, vectorizable_store)
10633         (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
10634         (supportable_widening_operation): Likewise.
10635         (supportable_narrowing_operation): Likewise.
10636         * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
10637         Likewise.
10638         * varasm.c (output_constant): Likewise.
10639
10640 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10641             Alan Hayward  <alan.hayward@arm.com>
10642             David Sherwood  <david.sherwood@arm.com>
10643
10644         * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
10645         so that both the length == 3 and length != 3 cases set up their
10646         own permute vectors.  Add comments explaining why we know the
10647         number of elements is constant.
10648         (vect_permute_load_chain): Likewise.
10649
10650 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10651             Alan Hayward  <alan.hayward@arm.com>
10652             David Sherwood  <david.sherwood@arm.com>
10653
10654         * machmode.h (mode_nunits): Change from unsigned char to
10655         poly_uint16_pod.
10656         (ONLY_FIXED_SIZE_MODES): New macro.
10657         (pod_mode::measurement_type, scalar_int_mode::measurement_type)
10658         (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
10659         (complex_mode::measurement_type, fixed_size_mode::measurement_type):
10660         New typedefs.
10661         (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
10662         (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
10663         or if measurement_type is not polynomial.
10664         * genmodes.c (ZERO_COEFFS): New macro.
10665         (emit_mode_nunits_inline): Make mode_nunits_inline return a
10666         poly_uint16.
10667         (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
10668         Use ZERO_COEFFS when emitting initializers.
10669         * data-streamer.h (bp_pack_poly_value): New function.
10670         (bp_unpack_poly_value): Likewise.
10671         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
10672         for GET_MODE_NUNITS.
10673         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
10674         for GET_MODE_NUNITS.
10675         * tree.c (make_vector_type): Remove temporary shim and make
10676         the real function take the number of units as a poly_uint64
10677         rather than an int.
10678         (build_vector_type_for_mode): Handle polynomial nunits.
10679         * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
10680         * emit-rtl.c (const_vec_series_p_1): Likewise.
10681         (gen_rtx_CONST_VECTOR): Likewise.
10682         * fold-const.c (test_vec_duplicate_folding): Likewise.
10683         * genrecog.c (validate_pattern): Likewise.
10684         * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
10685         * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
10686         * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
10687         (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
10688         (expand_vec_cond_expr, expand_mult_highpart): Likewise.
10689         * rtlanal.c (subreg_get_info): Likewise.
10690         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
10691         (vect_grouped_load_supported): Likewise.
10692         * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
10693         * tree-vect-loop.c (have_whole_vector_shift): Likewise.
10694         * simplify-rtx.c (simplify_unary_operation_1): Likewise.
10695         (simplify_const_unary_operation, simplify_binary_operation_1)
10696         (simplify_const_binary_operation, simplify_ternary_operation)
10697         (test_vector_ops_duplicate, test_vector_ops): Likewise.
10698         (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
10699         instead of CONST_VECTOR_NUNITS.
10700         * varasm.c (output_constant_pool_2): Likewise.
10701         * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
10702         explicit-encoded elements in the XVEC for variable-length vectors.
10703
10704 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10705
10706         * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
10707
10708 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10709             Alan Hayward  <alan.hayward@arm.com>
10710             David Sherwood  <david.sherwood@arm.com>
10711
10712         * coretypes.h (fixed_size_mode): Declare.
10713         (fixed_size_mode_pod): New typedef.
10714         * builtins.h (target_builtins::x_apply_args_mode)
10715         (target_builtins::x_apply_result_mode): Change type to
10716         fixed_size_mode_pod.
10717         * builtins.c (apply_args_size, apply_result_size, result_vector)
10718         (expand_builtin_apply_args_1, expand_builtin_apply)
10719         (expand_builtin_return): Update accordingly.
10720
10721 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10722
10723         * cse.c (hash_rtx_cb): Hash only the encoded elements.
10724         * cselib.c (cselib_hash_rtx): Likewise.
10725         * expmed.c (make_tree): Build VECTOR_CSTs directly from the
10726         CONST_VECTOR encoding.
10727
10728 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
10729             Jeff Law  <law@redhat.com>
10730
10731         PR target/83641
10732         * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
10733         noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
10734         only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
10735         and add REG_CFA_ADJUST_CFA notes in that case to both insns.
10736
10737         PR target/83641
10738         * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
10739         explicitly probe *sp in a noreturn function if there were any callee
10740         register saves or frame pointer is needed.
10741
10742 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
10743
10744         PR debug/83621
10745         * cfgexpand.c (expand_debug_expr): Return NULL if mode is
10746         BLKmode for ternary, binary or unary expressions.
10747
10748         PR debug/83645
10749         * var-tracking.c (delete_vta_debug_insn): New inline function.
10750         (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
10751         insns from get_insns () to NULL instead of each bb separately.
10752         Use delete_vta_debug_insn.  No longer static.
10753         (vt_debug_insns_local, variable_tracking_main_1): Adjust
10754         delete_vta_debug_insns callers.
10755         * rtl.h (delete_vta_debug_insns): Declare.
10756         * final.c (rest_of_handle_final): Call delete_vta_debug_insns
10757         instead of variable_tracking_main.
10758
10759 2018-01-03  Martin Sebor  <msebor@redhat.com>
10760
10761         PR tree-optimization/83603
10762         * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
10763         arguments past the endof the argument list in functions declared
10764         without a prototype.
10765         * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
10766         Avoid checking when arguments are null.
10767
10768 2018-01-03  Martin Sebor  <msebor@redhat.com>
10769
10770         PR c/83559
10771         * doc/extend.texi (attribute const): Fix a typo.
10772         * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
10773         issuing -Wsuggest-attribute for void functions.
10774
10775 2018-01-03  Martin Sebor  <msebor@redhat.com>
10776
10777         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
10778         offset_int::from instead of wide_int::to_shwi.
10779         (maybe_diag_overlap): Remove assertion.
10780         Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
10781         * gimple-ssa-sprintf.c (format_directive): Same.
10782         (parse_directive): Same.
10783         (sprintf_dom_walker::compute_format_length): Same.
10784         (try_substitute_return_value): Same.
10785
10786 2018-01-03  Jeff Law  <law@redhat.com>
10787
10788         PR middle-end/83654
10789         * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
10790         non-constant residual for zero at runtime and avoid probing in
10791         that case.  Reorganize code for trailing problem to mirror handling
10792         of the residual.
10793
10794 2018-01-03  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
10795
10796         PR tree-optimization/83501
10797         * tree-ssa-strlen.c (get_string_cst): New.
10798         (handle_char_store): Call get_string_cst.
10799
10800 2018-01-03  Martin Liska  <mliska@suse.cz>
10801
10802         PR tree-optimization/83593
10803         * tree-ssa-strlen.c: Include tree-cfg.h.
10804         (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
10805         (strlen_dom_walker): Add new member variable m_cleanup_cfg.
10806         (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
10807         to false.
10808         (strlen_dom_walker::before_dom_children): Call
10809         gimple_purge_dead_eh_edges. Dump tranformation with details
10810         dump flags.
10811         (strlen_dom_walker::before_dom_children): Update call by adding
10812         new argument cleanup_eh.
10813         (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
10814
10815 2018-01-03  Martin Liska  <mliska@suse.cz>
10816
10817         PR ipa/83549
10818         * cif-code.def (VARIADIC_THUNK): New enum value.
10819         * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
10820         thunks.
10821
10822 2018-01-03  Jan Beulich  <jbeulich@suse.com>
10823
10824         * sse.md (mov<mode>_internal): Tighten condition for when to use
10825         vmovdqu<ssescalarsize> for TI and OI modes.
10826
10827 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
10828
10829         Update copyright years.
10830
10831 2018-01-03  Martin Liska  <mliska@suse.cz>
10832
10833         PR ipa/83594
10834         * ipa-visibility.c (function_and_variable_visibility): Skip
10835         functions with noipa attribure.
10836
10837 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
10838
10839         * gcc.c (process_command): Update copyright notice dates.
10840         * gcov-dump.c (print_version): Ditto.
10841         * gcov.c (print_version): Ditto.
10842         * gcov-tool.c (print_version): Ditto.
10843         * gengtype.c (create_file): Ditto.
10844         * doc/cpp.texi: Bump @copying's copyright year.
10845         * doc/cppinternals.texi: Ditto.
10846         * doc/gcc.texi: Ditto.
10847         * doc/gccint.texi: Ditto.
10848         * doc/gcov.texi: Ditto.
10849         * doc/install.texi: Ditto.
10850         * doc/invoke.texi: Ditto.
10851
10852 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10853
10854         * vector-builder.h (vector_builder::m_full_nelts): Change from
10855         unsigned int to poly_uint64.
10856         (vector_builder::full_nelts): Update prototype accordingly.
10857         (vector_builder::new_vector): Likewise.
10858         (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
10859         (vector_builder::operator ==): Likewise.
10860         (vector_builder::finalize): Likewise.
10861         * int-vector-builder.h (int_vector_builder::int_vector_builder):
10862         Take the number of elements as a poly_uint64 rather than an
10863         unsigned int.
10864         * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
10865         from unsigned int to poly_uint64.
10866         (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
10867         (vec_perm_indices::new_vector): Likewise.
10868         (vec_perm_indices::length): Likewise.
10869         (vec_perm_indices::nelts_per_input): Likewise.
10870         (vec_perm_indices::input_nelts): Likewise.
10871         * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
10872         number of elements per input as a poly_uint64 rather than an
10873         unsigned int.  Use the original encoding for variable-length
10874         vectors, rather than clamping each individual element.
10875         For the second and subsequent elements in each pattern,
10876         clamp the step and base before clamping their sum.
10877         (vec_perm_indices::series_p): Handle polynomial element counts.
10878         (vec_perm_indices::all_in_range_p): Likewise.
10879         (vec_perm_indices_to_tree): Likewise.
10880         (vec_perm_indices_to_rtx): Likewise.
10881         * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
10882         * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
10883         (tree_vector_builder::new_binary_operation): Handle polynomial
10884         element counts.  Return false if we need to know the number
10885         of elements at compile time.
10886         * fold-const.c (fold_vec_perm): Punt if the number of elements
10887         isn't known at compile time.
10888
10889 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10890
10891         * vec-perm-indices.h (vec_perm_builder): Change element type
10892         from HOST_WIDE_INT to poly_int64.
10893         (vec_perm_indices::element_type): Update accordingly.
10894         (vec_perm_indices::clamp): Handle polynomial element_types.
10895         * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
10896         (vec_perm_indices::all_in_range_p): Likewise.
10897         (tree_to_vec_perm_builder): Check for poly_int64 trees rather
10898         than shwi trees.
10899         * vector-builder.h (vector_builder::stepped_sequence_p): Handle
10900         polynomial vec_perm_indices element types.
10901         * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
10902         * fold-const.c (fold_vec_perm): Likewise.
10903         * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
10904         * tree-vect-generic.c (lower_vec_perm): Likewise.
10905         * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
10906         * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
10907         element type to HOST_WIDE_INT.
10908
10909 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10910             Alan Hayward  <alan.hayward@arm.com>
10911             David Sherwood  <david.sherwood@arm.com>
10912
10913         * alias.c (addr_side_effect_eval): Take the size as a poly_int64
10914         rather than an int.  Use plus_constant.
10915         (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
10916         Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
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         * calls.c (emit_call_1, expand_call): Change struct_value_size from
10923         a HOST_WIDE_INT to a poly_int64.
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         * calls.c (load_register_parameters): Cope with polynomial
10930         mode sizes.  Require a constant size for BLKmode parameters
10931         that aren't described by a PARALLEL.  If BLOCK_REG_PADDING
10932         forces a parameter to be padded at the lsb end in order to
10933         fill a complete number of words, require the parameter size
10934         to be ordered wrt UNITS_PER_WORD.
10935
10936 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10937             Alan Hayward  <alan.hayward@arm.com>
10938             David Sherwood  <david.sherwood@arm.com>
10939
10940         * reload1.c (spill_stack_slot_width): Change element type
10941         from unsigned int to poly_uint64_pod.
10942         (alter_reg): Treat mode sizes as polynomial.
10943
10944 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10945             Alan Hayward  <alan.hayward@arm.com>
10946             David Sherwood  <david.sherwood@arm.com>
10947
10948         * reload.c (complex_word_subreg_p): New function.
10949         (reload_inner_reg_of_subreg, push_reload): Use it.
10950
10951 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10952             Alan Hayward  <alan.hayward@arm.com>
10953             David Sherwood  <david.sherwood@arm.com>
10954
10955         * lra-constraints.c (process_alt_operands): Reject matched
10956         operands whose sizes aren't ordered.
10957         (match_reload): Refer to this check here.
10958
10959 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10960             Alan Hayward  <alan.hayward@arm.com>
10961             David Sherwood  <david.sherwood@arm.com>
10962
10963         * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
10964         that the mode size is in the set {1, 2, 4, 8, 16}.
10965
10966 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10967             Alan Hayward  <alan.hayward@arm.com>
10968             David Sherwood  <david.sherwood@arm.com>
10969
10970         * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
10971         Use plus_constant instead of gen_rtx_PLUS.
10972
10973 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10974             Alan Hayward  <alan.hayward@arm.com>
10975             David Sherwood  <david.sherwood@arm.com>
10976
10977         * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
10978         * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
10979         * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
10980         * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
10981         * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
10982         * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
10983         * config/i386/i386-protos.h (ix86_push_rounding): Declare.
10984         * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
10985         * config/i386/i386.c (ix86_push_rounding): ...this new function.
10986         * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
10987         a poly_int64.
10988         * config/m32c/m32c.c (m32c_push_rounding): Likewise.
10989         * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
10990         * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
10991         * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
10992         * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
10993         * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
10994         * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
10995         * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
10996         * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
10997         * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
10998         function.
10999         * expr.c (emit_move_resolve_push): Treat the input and result
11000         of PUSH_ROUNDING as a poly_int64.
11001         (emit_move_complex_push, emit_single_push_insn_1): Likewise.
11002         (emit_push_insn): Likewise.
11003         * lra-eliminations.c (mark_not_eliminable): Likewise.
11004         * recog.c (push_operand): Likewise.
11005         * reload1.c (elimination_effects): Likewise.
11006         * rtlanal.c (nonzero_bits1): Likewise.
11007         * calls.c (store_one_arg): Likewise.  Require the padding to be
11008         known at compile time.
11009
11010 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11011             Alan Hayward  <alan.hayward@arm.com>
11012             David Sherwood  <david.sherwood@arm.com>
11013
11014         * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
11015         Use plus_constant instead of gen_rtx_PLUS.
11016
11017 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11018             Alan Hayward  <alan.hayward@arm.com>
11019             David Sherwood  <david.sherwood@arm.com>
11020
11021         * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
11022         rather than an int.
11023
11024 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11025             Alan Hayward  <alan.hayward@arm.com>
11026             David Sherwood  <david.sherwood@arm.com>
11027
11028         * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
11029         instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
11030         via stack temporaries.  Treat the mode size as polynomial too.
11031
11032 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11033             Alan Hayward  <alan.hayward@arm.com>
11034             David Sherwood  <david.sherwood@arm.com>
11035
11036         * expr.c (expand_expr_real_2): When handling conversions involving
11037         unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
11038         multiplying int_size_in_bytes by BITS_PER_UNIT.  Treat GET_MODE_BISIZE
11039         as a poly_uint64 too.
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         * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
11046
11047 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11048             Alan Hayward  <alan.hayward@arm.com>
11049             David Sherwood  <david.sherwood@arm.com>
11050
11051         * combine.c (can_change_dest_mode): Handle polynomial
11052         REGMODE_NATURAL_SIZE.
11053         * expmed.c (store_bit_field_1): Likewise.
11054         * expr.c (store_constructor): Likewise.
11055         * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
11056         and polynomial REGMODE_NATURAL_SIZE.
11057         (gen_lowpart_common): Likewise.
11058         * reginfo.c (record_subregs_of_mode): Likewise.
11059         * rtlanal.c (read_modify_subreg_p): Likewise.
11060
11061 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11062             Alan Hayward  <alan.hayward@arm.com>
11063             David Sherwood  <david.sherwood@arm.com>
11064
11065         * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
11066         numbers of elements.
11067
11068 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11069             Alan Hayward  <alan.hayward@arm.com>
11070             David Sherwood  <david.sherwood@arm.com>
11071
11072         * match.pd: Cope with polynomial numbers of vector elements.
11073
11074 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11075             Alan Hayward  <alan.hayward@arm.com>
11076             David Sherwood  <david.sherwood@arm.com>
11077
11078         * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
11079         in a POINTER_PLUS_EXPR.
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         * omp-simd-clone.c (simd_clone_subparts): New function.
11086         (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
11087         (ipa_simd_modify_function_body): Likewise.
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-generic.c (nunits_for_known_piecewise_op): New function.
11094         (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
11095         (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
11096         (expand_vector_condition, vector_element): Likewise.
11097         (subparts_gt): New function.
11098         (get_compute_type): Use subparts_gt.
11099         (count_type_subparts): Delete.
11100         (expand_vector_operations_1): Use subparts_gt instead of
11101         count_type_subparts.
11102
11103 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11104             Alan Hayward  <alan.hayward@arm.com>
11105             David Sherwood  <david.sherwood@arm.com>
11106
11107         * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
11108         (vect_compile_time_alias): ...this new function.  Do the calculation
11109         on poly_ints rather than trees.
11110         (vect_prune_runtime_alias_test_list): Update call accordingly.
11111
11112 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11113             Alan Hayward  <alan.hayward@arm.com>
11114             David Sherwood  <david.sherwood@arm.com>
11115
11116         * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
11117         numbers of units.
11118         (vect_schedule_slp_instance): Likewise.
11119
11120 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11121             Alan Hayward  <alan.hayward@arm.com>
11122             David Sherwood  <david.sherwood@arm.com>
11123
11124         * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
11125         constant and extern definitions for variable-length vectors.
11126         (vect_get_constant_vectors): Note that the number of units
11127         is known to be constant.
11128
11129 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11130             Alan Hayward  <alan.hayward@arm.com>
11131             David Sherwood  <david.sherwood@arm.com>
11132
11133         * tree-vect-stmts.c (vectorizable_conversion): Treat the number
11134         of units as polynomial.  Choose between WIDE and NARROW based
11135         on multiple_p.
11136
11137 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11138             Alan Hayward  <alan.hayward@arm.com>
11139             David Sherwood  <david.sherwood@arm.com>
11140
11141         * tree-vect-stmts.c (simd_clone_subparts): New function.
11142         (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
11143
11144 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11145             Alan Hayward  <alan.hayward@arm.com>
11146             David Sherwood  <david.sherwood@arm.com>
11147
11148         * tree-vect-stmts.c (vectorizable_call): Treat the number of
11149         vectors as polynomial.  Use build_index_vector for
11150         IFN_GOMP_SIMD_LANE.
11151
11152 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11153             Alan Hayward  <alan.hayward@arm.com>
11154             David Sherwood  <david.sherwood@arm.com>
11155
11156         * tree-vect-stmts.c (get_load_store_type): Treat the number of
11157         units as polynomial.  Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
11158         for variable-length vectors.
11159         (vectorizable_mask_load_store): Treat the number of units as
11160         polynomial, asserting that it is constant if the condition has
11161         already been enforced.
11162         (vectorizable_store, vectorizable_load): Likewise.
11163
11164 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11165             Alan Hayward  <alan.hayward@arm.com>
11166             David Sherwood  <david.sherwood@arm.com>
11167
11168         * tree-vect-loop.c (vectorizable_live_operation): Treat the number
11169         of units as polynomial.  Punt if we can't tell at compile time
11170         which vector contains the final result.
11171
11172 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11173             Alan Hayward  <alan.hayward@arm.com>
11174             David Sherwood  <david.sherwood@arm.com>
11175
11176         * tree-vect-loop.c (vectorizable_induction): Treat the number
11177         of units as polynomial.  Punt on SLP inductions.  Use an integer
11178         VEC_SERIES_EXPR for variable-length integer reductions.  Use a
11179         cast of such a series for variable-length floating-point
11180         reductions.
11181
11182 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11183             Alan Hayward  <alan.hayward@arm.com>
11184             David Sherwood  <david.sherwood@arm.com>
11185
11186         * tree.h (build_index_vector): Declare.
11187         * tree.c (build_index_vector): New function.
11188         * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
11189         of units as polynomial, forcibly converting it to a constant if
11190         vectorizable_reduction has already enforced the condition.
11191         (vect_create_epilog_for_reduction): Likewise.  Use build_index_vector
11192         to create a {1,2,3,...} vector.
11193         (vectorizable_reduction): Treat the number of units as polynomial.
11194         Choose vectype_in based on the largest scalar element size rather
11195         than the smallest number of units.  Enforce the restrictions
11196         relied on above.
11197
11198 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11199             Alan Hayward  <alan.hayward@arm.com>
11200             David Sherwood  <david.sherwood@arm.com>
11201
11202         * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
11203         number of units as polynomial.
11204
11205 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11206             Alan Hayward  <alan.hayward@arm.com>
11207             David Sherwood  <david.sherwood@arm.com>
11208
11209         * target.h (vector_sizes, auto_vector_sizes): New typedefs.
11210         * target.def (autovectorize_vector_sizes): Return the vector sizes
11211         by pointer, using vector_sizes rather than a bitmask.
11212         * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
11213         * targhooks.c (default_autovectorize_vector_sizes): Likewise.
11214         * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
11215         Likewise.
11216         * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
11217         * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
11218         * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
11219         * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
11220         * omp-general.c (omp_max_vf): Likewise.
11221         * omp-low.c (omp_clause_aligned_alignment): Likewise.
11222         * optabs-query.c (can_vec_mask_load_store_p): Likewise.
11223         * tree-vect-loop.c (vect_analyze_loop): Likewise.
11224         * tree-vect-slp.c (vect_slp_bb): Likewise.
11225         * doc/tm.texi: Regenerate.
11226         * tree-vectorizer.h (current_vector_size): Change from an unsigned int
11227         to a poly_uint64.
11228         * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
11229         the vector size as a poly_uint64 rather than an unsigned int.
11230         (current_vector_size): Change from an unsigned int to a poly_uint64.
11231         (get_vectype_for_scalar_type): Update accordingly.
11232         * tree.h (build_truth_vector_type): Take the size and number of
11233         units as a poly_uint64 rather than an unsigned int.
11234         (build_vector_type): Add a temporary overload that takes
11235         the number of units as a poly_uint64 rather than an unsigned int.
11236         * tree.c (make_vector_type): Likewise.
11237         (build_truth_vector_type): Take the number of units as a poly_uint64
11238         rather than an unsigned int.
11239
11240 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11241             Alan Hayward  <alan.hayward@arm.com>
11242             David Sherwood  <david.sherwood@arm.com>
11243
11244         * target.def (get_mask_mode): Take the number of units and length
11245         as poly_uint64s rather than unsigned ints.
11246         * targhooks.h (default_get_mask_mode): Update accordingly.
11247         * targhooks.c (default_get_mask_mode): Likewise.
11248         * config/i386/i386.c (ix86_get_mask_mode): Likewise.
11249         * doc/tm.texi: Regenerate.
11250
11251 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11252             Alan Hayward  <alan.hayward@arm.com>
11253             David Sherwood  <david.sherwood@arm.com>
11254
11255         * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
11256         * omp-general.c (omp_max_vf): Likewise.
11257         * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
11258         (expand_omp_simd): Handle polynomial safelen.
11259         * omp-low.c (omplow_simd_context): Add a default constructor.
11260         (omplow_simd_context::max_vf): Change from int to poly_uint64.
11261         (lower_rec_simd_input_clauses): Update accordingly.
11262         (lower_rec_input_clauses): Likewise.
11263
11264 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11265             Alan Hayward  <alan.hayward@arm.com>
11266             David Sherwood  <david.sherwood@arm.com>
11267
11268         * tree-vectorizer.h (vect_nunits_for_cost): New function.
11269         * tree-vect-loop.c (vect_model_reduction_cost): Use it.
11270         * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
11271         (vect_analyze_slp_cost): Likewise.
11272         * tree-vect-stmts.c (vect_model_store_cost): Likewise.
11273         (vect_model_load_cost): Likewise.
11274
11275 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11276             Alan Hayward  <alan.hayward@arm.com>
11277             David Sherwood  <david.sherwood@arm.com>
11278
11279         * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
11280         (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
11281         from an unsigned int * to a poly_uint64_pod *.
11282         (calculate_unrolling_factor): New function.
11283         (vect_analyze_slp_instance): Use it.  Track polynomial max_nunits.
11284
11285 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11286             Alan Hayward  <alan.hayward@arm.com>
11287             David Sherwood  <david.sherwood@arm.com>
11288
11289         * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
11290         from an unsigned int to a poly_uint64.
11291         (_loop_vec_info::slp_unrolling_factor): Likewise.
11292         (_loop_vec_info::vectorization_factor): Change from an int
11293         to a poly_uint64.
11294         (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
11295         (vect_get_num_vectors): New function.
11296         (vect_update_max_nunits, vect_vf_for_cost): Likewise.
11297         (vect_get_num_copies): Use vect_get_num_vectors.
11298         (vect_analyze_data_ref_dependences): Change max_vf from an int *
11299         to an unsigned int *.
11300         (vect_analyze_data_refs): Change min_vf from an int * to a
11301         poly_uint64 *.
11302         (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
11303         than an unsigned HOST_WIDE_INT.
11304         * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
11305         (vect_analyze_data_ref_dependence): Change max_vf from an int *
11306         to an unsigned int *.
11307         (vect_analyze_data_ref_dependences): Likewise.
11308         (vect_compute_data_ref_alignment): Handle polynomial vf.
11309         (vect_enhance_data_refs_alignment): Likewise.
11310         (vect_prune_runtime_alias_test_list): Likewise.
11311         (vect_shift_permute_load_chain): Likewise.
11312         (vect_supportable_dr_alignment): Likewise.
11313         (dependence_distance_ge_vf): Take the vectorization factor as a
11314         poly_uint64 rather than an unsigned HOST_WIDE_INT.
11315         (vect_analyze_data_refs): Change min_vf from an int * to a
11316         poly_uint64 *.
11317         * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
11318         vfm1 as a poly_uint64 rather than an int.  Make the same change
11319         for the returned bound_scalar.
11320         (vect_gen_vector_loop_niters): Handle polynomial vf.
11321         (vect_do_peeling): Likewise.  Update call to
11322         vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
11323         (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
11324         be constant.
11325         * tree-vect-loop.c (vect_determine_vectorization_factor)
11326         (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
11327         (vect_get_known_peeling_cost): Likewise.
11328         (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
11329         (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
11330         (vect_transform_loop): Likewise.  Use the lowest possible VF when
11331         updating the upper bounds of the loop.
11332         (vect_min_worthwhile_factor): Make static.  Return an unsigned int
11333         rather than an int.
11334         * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
11335         polynomial unroll factors.
11336         (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
11337         (vect_make_slp_decision): Likewise.
11338         (vect_supported_load_permutation_p): Likewise, and polynomial
11339         vf too.
11340         (vect_analyze_slp_cost): Handle polynomial vf.
11341         (vect_slp_analyze_node_operations): Likewise.
11342         (vect_slp_analyze_bb_1): Likewise.
11343         (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
11344         than an unsigned HOST_WIDE_INT.
11345         * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
11346         (vectorizable_load): Handle polynomial vf.
11347         * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
11348         a poly_uint64.
11349         (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
11350
11351 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11352             Alan Hayward  <alan.hayward@arm.com>
11353             David Sherwood  <david.sherwood@arm.com>
11354
11355         * match.pd: Handle bit operations involving three constants
11356         and try to fold one pair.
11357
11358 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11359
11360         * tree-vect-loop-manip.c: Include gimple-fold.h.
11361         (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
11362         niters_maybe_zero parameters.  Handle other cases besides a step of 1.
11363         (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
11364         Add a path that uses a step of VF instead of 1, but disable it
11365         for now.
11366         (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
11367         and niters_no_overflow parameters.  Update calls to
11368         slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
11369         Create a new SSA name if the latter choses to use a ste other
11370         than zero, and return it via niters_vector_mult_vf_var.
11371         * tree-vect-loop.c (vect_transform_loop): Update calls to
11372         vect_do_peeling, vect_gen_vector_loop_niters and
11373         slpeel_make_loop_iterate_ntimes.
11374         * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
11375         (vect_gen_vector_loop_niters): Update declarations after above changes.
11376
11377 2018-01-02  Michael Meissner  <meissner@linux.vnet.ibm.com>
11378
11379         * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
11380         128-bit round to integer instructions.
11381         (ceil<mode>2): Likewise.
11382         (btrunc<mode>2): Likewise.
11383         (round<mode>2): Likewise.
11384
11385 2018-01-02  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
11386
11387         * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
11388         unaligned VSX load/store on P8/P9.
11389         (expand_block_clear): Allow the use of unaligned VSX
11390         load/store on P8/P9.
11391
11392 2018-01-02  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
11393
11394         * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
11395         New function.
11396         (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
11397         swap associated with both a load and a store.
11398
11399 2018-01-02  Andrew Waterman  <andrew@sifive.com>
11400
11401         * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
11402         * config/riscv/riscv.md (clear_cache): Use it.
11403
11404 2018-01-02  Artyom Skrobov  <tyomitch@gmail.com>
11405
11406         * web.c: Remove out-of-date comment.
11407
11408 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11409
11410         * expr.c (fixup_args_size_notes): Check that any existing
11411         REG_ARGS_SIZE notes are correct, and don't try to re-add them.
11412         (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
11413         (emit_single_push_insn): ...here.
11414
11415 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11416
11417         * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
11418         (const_vector_encoded_nelts): New function.
11419         (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
11420         (const_vector_int_elt, const_vector_elt): Declare.
11421         * emit-rtl.c (const_vector_int_elt_1): New function.
11422         (const_vector_elt): Likewise.
11423         * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
11424         of CONST_VECTOR_ELT.
11425
11426 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11427
11428         * expr.c: Include rtx-vector-builder.h.
11429         (const_vector_mask_from_tree): Use rtx_vector_builder and operate
11430         directly on the tree encoding.
11431         (const_vector_from_tree): Likewise.
11432         * optabs.c: Include rtx-vector-builder.h.
11433         (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
11434         sequence of "u" values.
11435         * vec-perm-indices.c: Include rtx-vector-builder.h.
11436         (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
11437         directly on the vec_perm_indices encoding.
11438
11439 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11440
11441         * doc/rtl.texi (const_vector): Describe new encoding scheme.
11442         * Makefile.in (OBJS): Add rtx-vector-builder.o.
11443         * rtx-vector-builder.h: New file.
11444         * rtx-vector-builder.c: Likewise.
11445         * rtl.h (rtx_def::u2): Add a const_vector field.
11446         (CONST_VECTOR_NPATTERNS): New macro.
11447         (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
11448         (CONST_VECTOR_DUPLICATE_P): Likewise.
11449         (CONST_VECTOR_STEPPED_P): Likewise.
11450         (CONST_VECTOR_ENCODED_ELT): Likewise.
11451         (const_vec_duplicate_p): Check for a duplicated vector encoding.
11452         (unwrap_const_vec_duplicate): Likewise.
11453         (const_vec_series_p): Check for a non-duplicated vector encoding.
11454         Say that the function only returns true for integer vectors.
11455         * emit-rtl.c: Include rtx-vector-builder.h.
11456         (gen_const_vec_duplicate_1): Delete.
11457         (gen_const_vector): Call gen_const_vec_duplicate instead of
11458         gen_const_vec_duplicate_1.
11459         (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
11460         (gen_const_vec_duplicate): Use rtx_vector_builder.
11461         (gen_const_vec_series): Likewise.
11462         (gen_rtx_CONST_VECTOR): Likewise.
11463         * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
11464         (swap_const_vector_halves): Take an rtx pointer rather than rtx.
11465         Build a new vector rather than modifying a CONST_VECTOR in-place.
11466         (handle_special_swappables): Update call accordingly.
11467         * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
11468         (swap_const_vector_halves): Take an rtx pointer rather than rtx.
11469         Build a new vector rather than modifying a CONST_VECTOR in-place.
11470         (handle_special_swappables): Update call accordingly.
11471
11472 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11473
11474         * simplify-rtx.c (simplify_const_binary_operation): Use
11475         CONST_VECTOR_ELT instead of XVECEXP.
11476
11477 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11478
11479         * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
11480         the selector elements to be different from the data elements
11481         if the selector is a VECTOR_CST.
11482         * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
11483         ssizetype for the selector.
11484
11485 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11486
11487         * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
11488         before testing each element individually.
11489         * tree-vect-generic.c (lower_vec_perm): Likewise.
11490
11491 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11492
11493         * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
11494         * selftest-run-tests.c (selftest::run_tests): Call it.
11495         * vector-builder.h (vector_builder::operator ==): New function.
11496         (vector_builder::operator !=): Likewise.
11497         * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
11498         (vec_perm_indices::all_from_input_p): New function.
11499         * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
11500         (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
11501         * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
11502         instead of reading the VECTOR_CST directly.  Detect whether both
11503         vector inputs are the same before constructing the vec_perm_indices,
11504         and update the number of inputs argument accordingly.  Use the
11505         utility functions added above.  Only construct sel2 if we need to.
11506
11507 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11508
11509         * optabs.c (expand_vec_perm_var): Use an explicit encoding for
11510         the broadcast of the low byte.
11511         (expand_mult_highpart): Use an explicit encoding for the permutes.
11512         * optabs-query.c (can_mult_highpart_p): Likewise.
11513         * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
11514         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
11515         (vectorizable_bswap): Likewise.
11516         * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
11517         explicit encoding for the power-of-2 permutes.
11518         (vect_permute_store_chain): Likewise.
11519         (vect_grouped_load_supported): Likewise.
11520         (vect_permute_load_chain): Likewise.
11521
11522 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11523
11524         * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
11525         * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
11526         * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
11527         * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
11528         * tree-vect-stmts.c (vectorizable_bswap): Likewise.
11529         (vect_gen_perm_mask_any): Likewise.
11530
11531 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11532
11533         * int-vector-builder.h: New file.
11534         * vec-perm-indices.h: Include int-vector-builder.h.
11535         (vec_perm_indices): Redefine as an int_vector_builder.
11536         (auto_vec_perm_indices): Delete.
11537         (vec_perm_builder): Redefine as a stand-alone class.
11538         (vec_perm_indices::vec_perm_indices): New function.
11539         (vec_perm_indices::clamp): Likewise.
11540         * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
11541         (vec_perm_indices::new_vector): New function.
11542         (vec_perm_indices::new_expanded_vector): Update for new
11543         vec_perm_indices class.
11544         (vec_perm_indices::rotate_inputs): New function.
11545         (vec_perm_indices::all_in_range_p): Operate directly on the
11546         encoded form, without computing elided elements.
11547         (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
11548         encoding.  Update for new vec_perm_indices class.
11549         * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
11550         the given vec_perm_builder.
11551         (expand_vec_perm_var): Update vec_perm_builder constructor.
11552         (expand_mult_highpart): Use vec_perm_builder instead of
11553         auto_vec_perm_indices.
11554         * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
11555         vec_perm_indices instead of auto_vec_perm_indices.  Use a single
11556         or double series encoding as appropriate.
11557         * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
11558         vec_perm_indices instead of auto_vec_perm_indices.
11559         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
11560         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
11561         (vect_permute_store_chain): Likewise.
11562         (vect_grouped_load_supported): Likewise.
11563         (vect_permute_load_chain): Likewise.
11564         (vect_shift_permute_load_chain): Likewise.
11565         * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
11566         (vect_transform_slp_perm_load): Likewise.
11567         (vect_schedule_slp_instance): Likewise.
11568         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
11569         (vectorizable_mask_load_store): Likewise.
11570         (vectorizable_bswap): Likewise.
11571         (vectorizable_store): Likewise.
11572         (vectorizable_load): Likewise.
11573         * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
11574         vec_perm_indices instead of auto_vec_perm_indices.  Use
11575         tree_to_vec_perm_builder to read the vector from a tree.
11576         * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
11577         vec_perm_builder instead of a vec_perm_indices.
11578         (have_whole_vector_shift): Use vec_perm_builder and
11579         vec_perm_indices instead of auto_vec_perm_indices.  Leave the
11580         truncation to calc_vec_perm_mask_for_shift.
11581         (vect_create_epilog_for_reduction): Likewise.
11582         * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
11583         from auto_vec_perm_indices to vec_perm_indices.
11584         (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
11585         instead of changing individual elements.
11586         (aarch64_vectorize_vec_perm_const): Use new_vector to install
11587         the vector in d.perm.
11588         * config/arm/arm.c (expand_vec_perm_d::perm): Change
11589         from auto_vec_perm_indices to vec_perm_indices.
11590         (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
11591         instead of changing individual elements.
11592         (arm_vectorize_vec_perm_const): Use new_vector to install
11593         the vector in d.perm.
11594         * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
11595         Update vec_perm_builder constructor.
11596         (rs6000_expand_interleave): Likewise.
11597         * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
11598         (rs6000_expand_interleave): Likewise.
11599
11600 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11601
11602         * optabs-query.c (can_vec_perm_var_p): Check whether lowering
11603         to qimode could truncate the indices.
11604         * optabs.c (expand_vec_perm_var): Likewise.
11605
11606 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11607
11608         * Makefile.in (OBJS): Add vec-perm-indices.o.
11609         * vec-perm-indices.h: New file.
11610         * vec-perm-indices.c: Likewise.
11611         * target.h (vec_perm_indices): Replace with a forward class
11612         declaration.
11613         (auto_vec_perm_indices): Move to vec-perm-indices.h.
11614         * optabs.h: Include vec-perm-indices.h.
11615         (expand_vec_perm): Delete.
11616         (selector_fits_mode_p, expand_vec_perm_var): Declare.
11617         (expand_vec_perm_const): Declare.
11618         * target.def (vec_perm_const_ok): Replace with...
11619         (vec_perm_const): ...this new hook.
11620         * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
11621         (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
11622         * doc/tm.texi: Regenerate.
11623         * optabs.def (vec_perm_const): Delete.
11624         * doc/md.texi (vec_perm_const): Likewise.
11625         (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
11626         * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
11627         expand_vec_perm for constant permutation vectors.  Assert that
11628         the mode of variable permutation vectors is the integer equivalent
11629         of the mode that is being permuted.
11630         * optabs-query.h (selector_fits_mode_p): Declare.
11631         * optabs-query.c: Include vec-perm-indices.h.
11632         (selector_fits_mode_p): New function.
11633         (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
11634         is defined, instead of checking whether the vec_perm_const_optab
11635         exists.  Use targetm.vectorize.vec_perm_const instead of
11636         targetm.vectorize.vec_perm_const_ok.  Check whether the indices
11637         fit in the vector mode before using a variable permute.
11638         * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
11639         vec_perm_indices instead of an rtx.
11640         (expand_vec_perm): Replace with...
11641         (expand_vec_perm_const): ...this new function.  Take the selector
11642         as a vec_perm_indices rather than an rtx.  Also take the mode of
11643         the selector.  Update call to shift_amt_for_vec_perm_mask.
11644         Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
11645         Use vec_perm_indices::new_expanded_vector to expand the original
11646         selector into bytes.  Check whether the indices fit in the vector
11647         mode before using a variable permute.
11648         (expand_vec_perm_var): Make global.
11649         (expand_mult_highpart): Use expand_vec_perm_const.
11650         * fold-const.c: Includes vec-perm-indices.h.
11651         * tree-ssa-forwprop.c: Likewise.
11652         * tree-vect-data-refs.c: Likewise.
11653         * tree-vect-generic.c: Likewise.
11654         * tree-vect-loop.c: Likewise.
11655         * tree-vect-slp.c: Likewise.
11656         * tree-vect-stmts.c: Likewise.
11657         * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
11658         Delete.
11659         * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
11660         * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
11661         (aarch64_vectorize_vec_perm_const_ok): Fuse into...
11662         (aarch64_vectorize_vec_perm_const): ...this new function.
11663         (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
11664         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
11665         * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
11666         * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
11667         * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
11668         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
11669         (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
11670         into...
11671         (arm_vectorize_vec_perm_const): ...this new function.  Explicitly
11672         check for NEON modes.
11673         * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
11674         * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
11675         * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
11676         (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
11677         into...
11678         (ix86_vectorize_vec_perm_const): ...this new function.  Incorporate
11679         the old VEC_PERM_CONST conditions.
11680         * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
11681         * config/ia64/vect.md (vec_perm_const<mode>): Delete.
11682         * config/ia64/ia64.c (ia64_expand_vec_perm_const)
11683         (ia64_vectorize_vec_perm_const_ok): Merge into...
11684         (ia64_vectorize_vec_perm_const): ...this new function.
11685         * config/mips/loongson.md (vec_perm_const<mode>): Delete.
11686         * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
11687         * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
11688         * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
11689         * config/mips/mips.c (mips_expand_vec_perm_const)
11690         (mips_vectorize_vec_perm_const_ok): Merge into...
11691         (mips_vectorize_vec_perm_const): ...this new function.
11692         * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
11693         * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
11694         * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
11695         * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
11696         * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
11697         (rs6000_expand_vec_perm_const): Delete.
11698         * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
11699         Delete.
11700         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
11701         (altivec_expand_vec_perm_const_le): Take each operand individually.
11702         Operate on constant selectors rather than rtxes.
11703         (altivec_expand_vec_perm_const): Likewise.  Update call to
11704         altivec_expand_vec_perm_const_le.
11705         (rs6000_expand_vec_perm_const): Delete.
11706         (rs6000_vectorize_vec_perm_const_ok): Delete.
11707         (rs6000_vectorize_vec_perm_const): New function.
11708         (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
11709         an element count and rtx array.
11710         (rs6000_expand_extract_even): Update call accordingly.
11711         (rs6000_expand_interleave): Likewise.
11712         * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
11713         * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
11714         * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
11715         * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
11716         (rs6000_expand_vec_perm_const): Delete.
11717         * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
11718         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
11719         (altivec_expand_vec_perm_const_le): Take each operand individually.
11720         Operate on constant selectors rather than rtxes.
11721         (altivec_expand_vec_perm_const): Likewise.  Update call to
11722         altivec_expand_vec_perm_const_le.
11723         (rs6000_expand_vec_perm_const): Delete.
11724         (rs6000_vectorize_vec_perm_const_ok): Delete.
11725         (rs6000_vectorize_vec_perm_const): New function.  Remove stray
11726         reference to the SPE evmerge intructions.
11727         (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
11728         an element count and rtx array.
11729         (rs6000_expand_extract_even): Update call accordingly.
11730         (rs6000_expand_interleave): Likewise.
11731         * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
11732         * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
11733         new function.
11734         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
11735
11736 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11737
11738         * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
11739         vector mode and that that mode matches the mode of the data
11740         being permuted.
11741         (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
11742         out into expand_vec_perm_var.  Do all CONST_VECTOR handling here,
11743         directly using expand_vec_perm_1 when forcing selectors into
11744         registers.
11745         (expand_vec_perm_var): New function, split out from expand_vec_perm.
11746
11747 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11748
11749         * optabs-query.h (can_vec_perm_p): Delete.
11750         (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
11751         * optabs-query.c (can_vec_perm_p): Split into...
11752         (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
11753         (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
11754         particular selector is valid.
11755         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
11756         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
11757         (vect_grouped_load_supported): Likewise.
11758         (vect_shift_permute_load_chain): Likewise.
11759         * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
11760         (vect_transform_slp_perm_load): Likewise.
11761         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
11762         (vectorizable_bswap): Likewise.
11763         (vect_gen_perm_mask_checked): Likewise.
11764         * fold-const.c (fold_ternary_loc): Likewise.  Don't take
11765         implementations of variable permutation vectors into account
11766         when deciding which selector to use.
11767         * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
11768         vec_perm_const_optab is supported; instead use can_vec_perm_const_p
11769         with a false third argument.
11770         * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
11771         to test whether the constant selector is valid and can_vec_perm_var_p
11772         to test whether a variable selector is valid.
11773
11774 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11775
11776         * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
11777         * optabs-query.c (can_vec_perm_p): Likewise.
11778         * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
11779         instead of vec_perm_indices.
11780         * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
11781         (vect_gen_perm_mask_checked): Likewise,
11782         * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
11783         (vect_gen_perm_mask_checked): Likewise,
11784
11785 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11786
11787         * optabs-query.h (qimode_for_vec_perm): Declare.
11788         * optabs-query.c (can_vec_perm_p): Split out qimode search to...
11789         (qimode_for_vec_perm): ...this new function.
11790         * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
11791
11792 2018-01-02  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
11793
11794         * rtlanal.c (canonicalize_condition): Return 0 if final rtx
11795         does not have a conditional at the top.
11796
11797 2018-01-02  Richard Biener  <rguenther@suse.de>
11798
11799         * ipa-inline.c (big_speedup_p): Fix expression.
11800
11801 2018-01-02  Jan Hubicka  <hubicka@ucw.cz>
11802
11803         PR target/81616
11804         * config/i386/x86-tune-costs.h: Increase cost of integer load costs
11805         for generic 4->6.
11806
11807 2018-01-02  Jan Hubicka  <hubicka@ucw.cz>
11808
11809         PR target/81616
11810         Generic tuning.
11811         * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
11812         cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
11813         and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
11814         cond_taken_branch_cost 3->4.
11815
11816 2018-01-01  Jakub Jelinek  <jakub@redhat.com>
11817
11818         PR tree-optimization/83581
11819         * tree-loop-distribution.c (pass_loop_distribution::execute): Return
11820         TODO_cleanup_cfg if any changes have been made.
11821
11822         PR middle-end/83608
11823         * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
11824         convert_modes if target mode has the right side, but different mode
11825         class.
11826
11827         PR middle-end/83609
11828         * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
11829         last argument when extracting from CONCAT.  If either from_real or
11830         from_imag is NULL, use expansion through memory.  If result is not
11831         a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
11832         the parts directly to inner mode, if even that fails, use expansion
11833         through memory.
11834
11835         PR middle-end/83623
11836         * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
11837         check for bswap in mode rather than HImode and use that in expand_unop
11838         too.
11839 \f
11840 Copyright (C) 2018 Free Software Foundation, Inc.
11841
11842 Copying and distribution of this file, with or without modification,
11843 are permitted in any medium without royalty provided the copyright
11844 notice and this notice are preserved.