gcc/ChangeLog

   1 2018-02-06  Eric Botcazou  <ebotcazou@adacore.com>
   2
   3         * configure: Regenerate.
   4
   5 2018-02-05  Martin Sebor  <msebor@redhat.com>
   6
   7         PR tree-optimization/83369
   8         * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
   9         inlining context.
  10
  11 2018-02-05  Martin Liska  <mliska@suse.cz>
  12
  13         * doc/invoke.texi: Cherry-pick upstream r323995.
  14
  15 2018-02-05  Richard Sandiford  <richard.sandiford@linaro.org>
  16
  17         * ira.c (ira_init_register_move_cost): Adjust comment.
  18
  19 2018-02-05  Martin Liska  <mliska@suse.cz>
  20
  21         PR gcov-profile/84137
  22         * doc/gcov.texi: Fix typo in documentation.
  23
  24 2018-02-05  Martin Liska  <mliska@suse.cz>
  25
  26         PR gcov-profile/83879
  27         * doc/gcov.texi: Document necessity of --dynamic-list-data when
  28         using dlopen functionality.
  29
  30 2018-02-05  Olga Makhotina  <olga.makhotina@intel.com>
  31
  32         * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
  33         _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
  34         _mm_maskz_range_ss, _mm_mask_range_round_ss,
  35         _mm_maskz_range_round_ss): New intrinsics.
  36         (__builtin_ia32_rangesd128_round)
  37         (__builtin_ia32_rangess128_round): Remove.
  38         (__builtin_ia32_rangesd128_mask_round,
  39         __builtin_ia32_rangess128_mask_round): New builtins.
  40         * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
  41         __builtin_ia32_rangess128_round): Remove.
  42         (__builtin_ia32_rangesd128_mask_round,
  43         __builtin_ia32_rangess128_mask_round): New builtins.
  44         * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
  45         (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
  46         ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
  47         "<round_saeonly_constraint>")): Changed to ...
  48         ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
  49         "<round_saeonly_scalar_constraint>")): ... this.
  50         ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
  51         %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
  52         ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
  53         %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
  54         %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
  55
  56 2018-02-02  Andrew Jenner  <andrew@codesourcery.com>
  57
  58         * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
  59         options.
  60         * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
  61         Remove all values except native, 8540 and 8548.
  62
  63 2018-02-02  H.J. Lu  <hongjiu.lu@intel.com>
  64
  65         * config/i386/i386.c (ix86_output_function_return): Pass
  66         INVALID_REGNUM, instead of -1, as invalid register number to
  67         indirect_thunk_name and output_indirect_thunk.
  68
  69 2018-02-02  Julia Koval  <julia.koval@intel.com>
  70
  71         * config.gcc: Add -march=icelake.
  72         * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
  73         * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
  74         * config/i386/i386.c (processor_costs): Add m_ICELAKE.
  75         (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
  76         PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
  77         (processor_target_table): Add icelake.
  78         (ix86_option_override_internal): Handle new PTAs.
  79         (get_builtin_code_for_version): Handle icelake.
  80         (M_INTEL_COREI7_ICELAKE): New.
  81         (fold_builtin_cpu): Handle icelake.
  82         * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
  83         * doc/invoke.texi: Add -march=icelake.
  84
  85 2018-02-02  Julia Koval  <julia.koval@intel.com>
  86
  87         * config/i386/i386.c (ix86_option_override_internal): Change flags type
  88         to wide_int_bitmask.
  89         * wide-int-bitmask.h: New.
  90
  91 2018-02-02  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
  92
  93         PR target/84066
  94         * config/i386/i386.md: Replace Pmode with word_mode in
  95         builtin_setjmp_setup and builtin_longjmp to support x32.
  96
  97 2018-02-01  Peter Bergner  <bergner@vnet.ibm.com>
  98
  99         PR target/56010
 100         PR target/83743
 101         * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
 102         #include "opts.h".
 103         (rs6000_supported_cpu_names): New static variable.
 104         (linux_cpu_translation_table): Likewise.
 105         (elf_platform) <cpu>: Define new static variable and use it.
 106         Translate kernel AT_PLATFORM name to canonical name if needed.
 107         Error if platform name is unknown.
 108
 109 2018-02-01  Aldy Hernandez  <aldyh@redhat.com>
 110
 111         PR target/84089
 112         * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
 113
 114 2018-02-01  Jeff Law  <law@redhat.com>
 115
 116         PR target/84128
 117         * config/i386/i386.c (release_scratch_register_on_entry): Add new
 118         OFFSET and RELEASE_VIA_POP arguments.  Use SP+OFFSET to restore
 119         the scratch if RELEASE_VIA_POP is false.
 120         (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
 121         If we have to save a temporary register, decrement SIZE appropriately.
 122         Pass new arguments to release_scratch_register_on_entry.
 123         (ix86_adjust_stack_and_probe): Likewise.
 124         (ix86_emit_probe_stack_range): Pass new arguments to
 125         release_scratch_register_on_entry.
 126
 127 2018-02-01  Uros Bizjak  <ubizjak@gmail.com>
 128
 129         PR rtl-optimization/84157
 130         * combine.c (change_zero_ext): Use REG_P predicate in
 131         front of HARD_REGISTER_P predicate.
 132
 133 2018-02-01  Georg-Johann Lay  <avr@gjlay.de>
 134
 135         * config/avr/avr.c (avr_option_override): Move disabling of
 136         -fdelete-null-pointer-checks to...
 137         * common/config/avr/avr-common.c (avr_option_optimization_table):
 138         ...here.
 139
 140 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
 141
 142         PR tree-optimization/81635
 143         * tree-data-ref.c (split_constant_offset_1): For types that
 144         wrap on overflow, try to use range info to prove that wrapping
 145         cannot occur.
 146
 147 2018-02-01  Renlin Li  <renlin.li@arm.com>
 148
 149         PR target/83370
 150         * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
 151         TAILCALL_ADDR_REGS.
 152         (aarch64_register_move_cost): Likewise.
 153         * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
 154         TAILCALL_ADDR_REGS.
 155         (REG_CLASS_NAMES): Likewise.
 156         (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
 157         TAILCALL_ADDR_REGS. Remove IP registers.
 158         * config/aarch64/aarch64.md (Ucs): Update register constraint.
 159
 160 2018-02-01  Richard Biener  <rguenther@suse.de>
 161
 162         * domwalk.h (dom_walker::dom_walker): Add additional constructor
 163         for specifying RPO order and allow NULL for that.
 164         * domwalk.c (dom_walker::dom_walker): Likewise.
 165         (dom_walker::walk): Handle NULL RPO order.
 166         * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
 167         in RPO order.
 168         (rewrite_update_dom_walker): Likewise.
 169         (mark_def_dom_walker): Likewise.
 170
 171 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
 172
 173         * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
 174         (aarch64_maybe_expand_sve_subreg_move): Declare.
 175         * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
 176         * config/aarch64/predicates.md (aarch64_any_register_operand): New
 177         predicate.
 178         * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
 179         that are semantically a reverse operation.
 180         (*aarch64_sve_mov<mode>_subreg_be): New pattern.
 181         * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
 182         (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
 183         functions.
 184         (aarch64_can_change_mode_class): For big-endian, forbid changes
 185         between two SVE modes if they have different element sizes.
 186
 187 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
 188
 189         * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
 190         the TImode handling for big-endian targets.
 191
 192 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
 193
 194         * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
 195         (*sve_ld1rq<Vesize>): ... this new pattern.  Handle all element sizes,
 196         not just bytes.
 197         * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
 198         Remove BSWAP handing for big-endian targets and use the form of
 199         LD1RQ appropariate for the mode.
 200
 201 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
 202
 203         * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
 204         all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
 205         duplicated element.
 206
 207 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
 208
 209         PR tearget/83845
 210         * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
 211         check for operands that need to go through aarch64_sve_reload_be.
 212
 213 2018-02-01  Jakub Jelinek  <jakub@redhat.com>
 214
 215         PR tree-optimization/81661
 216         PR tree-optimization/84117
 217         * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
 218         * tree-eh.c: Include gimplify.h.
 219         (find_trapping_overflow, replace_trapping_overflow,
 220         rewrite_to_non_trapping_overflow): New functions.
 221         * tree-vect-loop.c: Include tree-eh.h.
 222         (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
 223         * tree-data-ref.c: Include tree-eh.h.
 224         (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
 225
 226 2018-01-31  Uros Bizjak  <ubizjak@gmail.com>
 227
 228         PR rtl-optimization/84123
 229         * combine.c (change_zero_ext): Check if hard register satisfies
 230         can_change_dest_mode before calling gen_lowpart_SUBREG.
 231
 232 2018-01-31  Vladimir Makarov  <vmakarov@redhat.com>
 233
 234         PR target/82444
 235         * ira.c (ira_init_register_move_cost): Remove assert.
 236
 237 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
 238
 239         PR rtl-optimization/84071
 240         * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
 241         * doc/tm.texi: Regenerate.
 242
 243 2018-01-31  Richard Biener  <rguenther@suse.de>
 244
 245         PR tree-optimization/84132
 246         * tree-data-ref.c (analyze_miv_subscript): Properly
 247         check whether evolution_function_is_affine_multivariate_p
 248         before calling gcd_of_steps_may_divide_p.
 249
 250 2018-01-31  Julia Koval  <julia.koval@intel.com>
 251
 252         PR target/83618
 253         * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
 254         * config/i386/i386.md (rdpid_rex64) New.
 255         (rdpid): Make 32bit only.
 256
 257 2018-01-29  Aldy Hernandez  <aldyh@redhat.com>
 258
 259         PR lto/84105
 260         * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
 261         an IDENTIFIER_NODE for FUNCTION_TYPE's.
 262
 263 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
 264
 265         Revert
 266         2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
 267
 268         * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
 269
 270 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
 271
 272         PR rtl-optimization/84071
 273         * combine.c (record_dead_and_set_regs_1): Record the source unmodified
 274         for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
 275
 276 2018-01-31  Claudiu Zissulescu  <claziss@synopsys.com>
 277
 278         * config/arc/arc.c (arc_handle_aux_attribute): New function.
 279         (arc_attribute_table): Add 'aux' attribute.
 280         (arc_in_small_data_p): Consider aux like variables.
 281         (arc_is_aux_reg_p): New function.
 282         (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
 283         (arc_get_aux_arg): New function.
 284         (prepare_move_operands): Handle aux-register access.
 285         (arc_handle_aux_attribute): New function.
 286         * doc/extend.texi (ARC Variable attributes): Add subsection.
 287
 288 2018-01-31  Claudiu Zissulescu  <claziss@synopsys.com>
 289
 290         * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
 291         * config/arc/arc.c (arc_handle_uncached_attribute): New function.
 292         (arc_attribute_table): Add 'uncached' attribute.
 293         (arc_print_operand): Print '.di' flag for uncached memory
 294         accesses.
 295         (arc_in_small_data_p): Do not consider for small data the uncached
 296         types.
 297         (arc_is_uncached_mem_p): New function.
 298         * config/arc/predicates.md (compact_store_memory_operand): Check
 299         for uncached memory accesses.
 300         (nonvol_nonimm_operand): Likewise.
 301         * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
 302
 303 2018-01-31  Jakub Jelinek  <jakub@redhat.com>
 304
 305         PR c/84100
 306         * common.opt (falign-functions=, falign-jumps=, falign-labels=,
 307         falign-loops=): Add Optimization flag.
 308
 309 2018-01-30  Jeff Law  <law@redhat.com>
 310
 311         PR target/84064
 312         * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
 313         INT_REGISTERS_SAVED.  Check it prior to calling
 314         get_scratch_register_on_entry.
 315         (ix86_adjust_stack_and_probe): Similarly.
 316         (ix86_emit_probe_stack_range): Similarly.
 317         (ix86_expand_prologue): Corresponding changes.
 318
 319 2018-01-30  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
 320
 321         PR target/40411
 322         * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
 323         -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
 324
 325 2018-01-30  Vladimir Makarov  <vmakarov@redhat.com>
 326
 327         PR target/84112
 328         * lra-constraints.c (curr_insn_transform): Process AND in the
 329         address.
 330
 331 2018-01-30  Jakub Jelinek  <jakub@redhat.com>
 332
 333         PR rtl-optimization/83986
 334         * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
 335         dependence against last_pending_memory_flush in addition to
 336         pending_jump_insns.
 337
 338 2018-01-30  Alexandre Oliva <aoliva@redhat.com>
 339
 340         PR tree-optimization/81611
 341         * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
 342         copies.
 343
 344 2018-01-30  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
 345
 346         * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
 347         a reg rtx.
 348
 349 2018-01-30  Richard Biener  <rguenther@suse.de>
 350             Jakub Jelinek  <jakub@redhat.com>
 351
 352         PR tree-optimization/84111
 353         * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
 354         inner loops added during recursion, as they don't have up-to-date
 355         SSA form.
 356
 357 2018-01-30  Jan Hubicka  <hubicka@ucw.cz>
 358
 359         PR ipa/81360
 360         * ipa-inline.c (can_inline_edge_p): Break out late tests to...
 361         (can_inline_edge_by_limits_p): ... here.
 362         (can_early_inline_edge_p, check_callers,
 363         update_caller_keys, update_callee_keys, recursive_inlining,
 364         add_new_edges_to_heap, speculation_useful_p,
 365         inline_small_functions,
 366         inline_small_functions, flatten_function,
 367         inline_to_all_callers_1): Update.
 368
 369 2018-01-30  Jan Hubicka  <hubicka@ucw.cz>
 370
 371         * profile-count.c (profile_count::combine_with_ipa_count): Handle
 372         zeros correctly.
 373
 374 2018-01-30  Richard Biener  <rguenther@suse.de>
 375
 376         PR tree-optimization/83008
 377         * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
 378         invariant and constant vector uses in stmts when they need
 379         more than one stmt.
 380
 381 2018-01-30  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
 382
 383         PR bootstrap/84017
 384         * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
 385         * configure: Regenerate.
 386
 387 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
 388
 389         * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
 390         pattern.
 391         (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
 392         Use gen_rtx_REG rather than gen_lowpart.
 393
 394 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
 395
 396         * lra-constraints.c (match_reload): Use subreg_lowpart_offset
 397         rather than 0 when creating partial subregs.
 398
 399 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
 400
 401         * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
 402         of usage.
 403
 404 2018-01-29  Michael Meissner  <meissner@linux.vnet.ibm.com>
 405
 406         PR target/81550
 407         * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
 408         and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
 409         -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
 410         flags.  This restores the settings used before the 2017-07-24.
 411         Turning off pre increment/decrement/modify allows IVOPTS to
 412         optimize DF/SF loops where the index is an int.
 413
 414 2018-01-29  Richard Biener <rguenther@suse.de>
 415             Kelvin Nilsen  <kelvin@gcc.gnu.org>
 416
 417         PR bootstrap/80867
 418         * tree-vect-stmts.c (vectorizable_call): Don't call
 419         targetm.vectorize_builtin_md_vectorized_function if callee is
 420         NULL.
 421
 422 2018-01-22 Carl Love <cel@us.ibm.com>
 423
 424         * doc/extend.tex: Fix typo in second arg in
 425         __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
 426
 427 2018-01-29  Richard Biener  <rguenther@suse.de>
 428
 429         PR tree-optimization/84086
 430         * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
 431         (flush_ssaname_freelist): When SSA names were released reset
 432         the SCEV hash table.
 433
 434 2018-01-29  Richard Biener  <rguenther@suse.de>
 435
 436         PR tree-optimization/84057
 437         * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
 438         removed paths when removing edges.
 439
 440 2018-01-27  H.J. Lu  <hongjiu.lu@intel.com>
 441
 442         * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
 443         -mfunction-return=@var{choice}.
 444
 445 2018-01-27  Bernd Edlinger  <bernd.edlinger@hotmail.de>
 446
 447         PR diagnostic/84034
 448         * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
 449         Handle CR like TAB.
 450         (layout::print_source_line): Likewise.
 451         (test_get_line_width_without_trailing_whitespace): Add test cases.
 452
 453 2018-01-27  Jakub Jelinek  <jakub@redhat.com>
 454
 455         PR middle-end/84040
 456         * sched-deps.c (sched_macro_fuse_insns): Return immediately for
 457         debug insns.
 458
 459 2018-01-26  Jim Wilson  <jimw@sifive.com>
 460
 461         * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
 462
 463         * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
 464         specified.
 465
 466 2018-01-26  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
 467
 468         * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
 469         and CMP + SUB-immediate -> SUBS.
 470
 471 2018-01-26  Martin Sebor  <msebor@redhat.com>
 472
 473         PR tree-optimization/83896
 474         * tree-ssa-strlen.c (get_string_len): Rename...
 475         (get_string_cst_length): ...to this.  Return HOST_WIDE_INT.
 476         Avoid assuming length is constant.
 477         (handle_char_store): Use HOST_WIDE_INT for string length.
 478
 479 2018-01-26  Uros Bizjak  <ubizjak@gmail.com>
 480
 481         PR target/81763
 482         * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
 483         to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
 484
 485 2018-01-26  Richard Biener  <rguenther@suse.de>
 486
 487         PR rtl-optimization/84003
 488         * dse.c (record_store): Only record redundant stores when
 489         the earlier store aliases at least all accesses the later one does.
 490
 491 2018-01-26  Jakub Jelinek  <jakub@redhat.com>
 492
 493         PR rtl-optimization/83985
 494         * dce.c (deletable_insn_p): Return false for separate shrink wrapping
 495         REG_CFA_RESTORE insns.
 496         (delete_unmarked_insns): Don't ignore separate shrink wrapping
 497         REG_CFA_RESTORE insns here.
 498
 499         PR c/83989
 500         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
 501         use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
 502
 503 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
 504
 505         * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
 506         * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
 507         (arc_init): Likewise.
 508         (arc_override_options): Likewise.
 509         (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
 510         value.
 511         (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
 512         support.
 513         * config/arc/arc.h (TARGET_DBNZ): Define.
 514         * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
 515         properly set the tune attribute.
 516         (dbnz): Use TARGET_DBNZ guard.
 517         * config/arc/arc.opt (mtune): Add core3 option.
 518
 519 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
 520
 521         * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
 522         recognize new pic like addresses.
 523         (arc_delegitimize_address): Clean up.
 524
 525 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
 526
 527         * config/arc/arc-arches.def: Option mrf16 valid for all
 528         architectures.
 529         * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
 530         * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
 531         * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
 532         * config/arc/arc-tables.opt: Regenerate.
 533         * config/arc/arc.c (arc_conditional_register_usage): Handle
 534         reduced register file case.
 535         (arc_file_start): Set must have build attributes.
 536         * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
 537         mrf16 option value.
 538         * config/arc/arc.opt (mrf16): Add new option.
 539         * config/arc/elf.h (ATTRIBUTE_PCS): Define.
 540         * config/arc/genmultilib.awk: Handle new mrf16 option.
 541         * config/arc/linux.h (ATTRIBUTE_PCS): Define.
 542         * config/arc/t-multilib: Regenerate.
 543         * doc/invoke.texi (ARC Options): Document mrf16 option.
 544
 545 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
 546
 547         * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
 548         * config/arc/arc.c (arc_handle_secure_attribute): New function.
 549         (arc_attribute_table): Add 'secure_call' attribute.
 550         (arc_print_operand): Print secure call operand.
 551         (arc_function_ok_for_sibcall): Don't optimize tail calls when
 552         secure.
 553         (arc_is_secure_call_p): New function.  * config/arc/arc.md
 554         (call_i): Add support for sjli instruction.
 555         (call_value_i): Likewise.
 556         * config/arc/constraints.md (Csc): New constraint.
 557
 558 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
 559             John Eric Martin <John.Martin@emmicro-us.com>
 560
 561         * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
 562         * config/arc/arc.c (_arc_jli_section): New struct.
 563         (arc_jli_section): New type.
 564         (rc_jli_sections): New static variable.
 565         (arc_handle_jli_attribute): New function.
 566         (arc_attribute_table): Add jli_always and jli_fixed attribute.
 567         (arc_file_end): New function.
 568         (TARGET_ASM_FILE_END): Define.
 569         (arc_print_operand): Reuse 'S' letter for JLI output instruction.
 570         (arc_add_jli_section): New function.
 571         (jli_call_scan): Likewise.
 572         (arc_reorg): Call jli_call_scan.
 573         (arc_output_addsi): Remove 'S' from printing asm operand.
 574         (arc_is_jli_call_p): New function.
 575         * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
 576         operand.
 577         (movhi_insn): Likewise.
 578         (movsi_insn): Likewise.
 579         (movsi_set_cc_insn): Likewise.
 580         (loadqi_update): Likewise.
 581         (load_zeroextendqisi_update): Likewise.
 582         (load_signextendqisi_update): Likewise.
 583         (loadhi_update): Likewise.
 584         (load_zeroextendhisi_update): Likewise.
 585         (load_signextendhisi_update): Likewise.
 586         (loadsi_update): Likewise.
 587         (loadsf_update): Likewise.
 588         (movsicc_insn): Likewise.
 589         (bset_insn): Likewise.
 590         (bxor_insn): Likewise.
 591         (bclr_insn): Likewise.
 592         (bmsk_insn): Likewise.
 593         (bicsi3_insn): Likewise.
 594         (cmpsi_cc_c_insn): Likewise.
 595         (movsi_ne): Likewise.
 596         (movsi_cond_exec): Likewise.
 597         (clrsbsi2): Likewise.
 598         (norm_f): Likewise.
 599         (normw): Likewise.
 600         (swap): Likewise.
 601         (divaw): Likewise.
 602         (flag): Likewise.
 603         (sr): Likewise.
 604         (kflag): Likewise.
 605         (ffs): Likewise.
 606         (ffs_f): Likewise.
 607         (fls): Likewise.
 608         (call_i): Remove 'S' asm letter, add jli instruction.
 609         (call_value_i): Likewise.
 610         * config/arc/arc.op (mjli-always): New option.
 611         * config/arc/constraints.md (Cji): New constraint.
 612         * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
 613         operand.
 614         (subsf3_fpx): Likewise.
 615         (mulsf3_fpx): Likewise.
 616         * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
 617         asm operand.
 618         * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
 619         function attrbutes.
 620         * doc/invoke.texi (ARC): Document mjli-always option.
 621
 622 2018-01-26  Sebastian Perta  <sebastian.perta@renesas.com>
 623
 624         * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
 625         and use incw and decw where possible
 626         * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
 627
 628 2018-01-26  Richard Biener  <rguenther@suse.de>
 629
 630         PR tree-optimization/81082
 631         * fold-const.c (fold_plusminus_mult_expr): Do not perform the
 632         association if it requires casting to unsigned.
 633         * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
 634         from fold_plusminus_mult_expr to catch important cases late when
 635         range info is available.
 636
 637 2018-01-26  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
 638
 639         * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
 640         * configure.ac (hidden_linkonce): New test.
 641         * configure: Regenerate.
 642         * config.in: Regenerate.
 643
 644 2018-01-26  Julia Koval  <julia.koval@intel.com>
 645
 646         * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
 647         _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
 648         _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
 649         _mm_mask_bitshuffle_epi64_mask): Fix type.
 650         * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
 651         USI_FTYPE_V4DI_V4DI_USI): Remove.
 652         * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
 653         __builtin_ia32_vpshufbitqmb256_mask,
 654         __builtin_ia32_vpshufbitqmb128_mask): Fix types.
 655         * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
 656         * config/i386/sse.md (VI1_AVX512VLBW): Change types.
 657
 658 2018-01-26  Alan Modra  <amodra@gmail.com>
 659
 660         PR target/84033
 661         * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
 662         UNSPEC_VBPERMQ.  Sort other unspecs.
 663
 664 2018-01-25  David Edelsohn  <dje.gcc@gmail.com>
 665
 666         * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
 667
 668 2018-01-25  Jan Hubicka  <hubicka@ucw.cz>
 669
 670         PR middle-end/83055
 671         * predict.c (drop_profile): Do not push/pop cfun; update also
 672         node->count.
 673         (handle_missing_profiles): Fix logic looking for zero profiles.
 674
 675 2018-01-25  Jakub Jelinek  <jakub@redhat.com>
 676
 677         PR middle-end/83977
 678         * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
 679         on functions with #pragma omp declare simd or functions with simd
 680         attribute.
 681         * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
 682         * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
 683         Remove trailing \n from warning_at calls.
 684
 685 2018-01-25  Tom de Vries  <tom@codesourcery.com>
 686
 687         PR target/84028
 688         * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
 689         for neutered workers.
 690
 691 2018-01-24  Joseph Myers  <joseph@codesourcery.com>
 692
 693         PR target/68467
 694         * config/m68k/m68k.c (m68k_promote_function_mode): New function.
 695         (TARGET_PROMOTE_FUNCTION_MODE): New macro.
 696
 697 2018-01-24  Jeff Law  <law@redhat.com>
 698
 699         PR target/83994
 700         * i386.c (get_probe_interval): Move to earlier point.
 701         (ix86_compute_frame_layout): If -fstack-clash-protection and
 702         the frame is larger than the probe interval, then use pushes
 703         to save registers rather than reg->mem moves.
 704         (ix86_expand_prologue): Remove conditional for int_registers_saved
 705         assertion.
 706
 707 2018-01-24  Vladimir Makarov  <vmakarov@redhat.com>
 708
 709         PR target/84014
 710         * ira-build.c (setup_min_max_allocno_live_range_point): Set up
 711         min/max for never referenced object.
 712
 713 2018-01-24  Jakub Jelinek  <jakub@redhat.com>
 714
 715         PR middle-end/83977
 716         * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
 717         here.
 718         * omp-low.c (create_omp_child_function): Remove "omp declare simd"
 719         attributes from DECL_ATTRIBUTES (decl) without affecting
 720         DECL_ATTRIBUTES (current_function_decl).
 721         * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
 722         functions with non-NULL DECL_ABSTRACT_ORIGIN.
 723
 724 2018-01-24  Richard Sandiford  <richard.sandiford@linaro.org>
 725
 726         PR tree-optimization/83979
 727         * fold-const.c (fold_comparison): Use constant_boolean_node
 728         instead of boolean_{true,false}_node.
 729
 730 2018-01-24  Jan Hubicka  <hubicka@ucw.cz>
 731
 732         * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
 733         with zero counts.
 734
 735 2018-01-24  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
 736
 737         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
 738         Simplify the clause that sets the length attribute.
 739         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
 740         (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
 741         clause that sets the length attribute.
 742         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
 743
 744 2018-01-24  Tom de Vries  <tom@codesourcery.com>
 745
 746         PR target/83589
 747         * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
 748         (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
 749         Add strict parameter.
 750         (prevent_branch_around_nothing): Insert dummy insn between branch to
 751         label and label with no ptx insn inbetween.
 752         * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
 753
 754 2018-01-24  Tom de Vries  <tom@codesourcery.com>
 755
 756         PR target/81352
 757         * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
 758         for neutered threads in warp.
 759         * config/nvptx/nvptx.md (define_insn "exit"): New insn.
 760
 761 2018-01-24  Richard Biener  <rguenther@suse.de>
 762
 763         PR tree-optimization/83176
 764         * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
 765         operands.
 766
 767 2018-01-24  Richard Biener  <rguenther@suse.de>
 768
 769         PR tree-optimization/82819
 770         * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
 771         code generating pluses that are no-ops in the target precision.
 772
 773 2018-01-24  Richard Biener  <rguenther@suse.de>
 774
 775         PR middle-end/84000
 776         * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
 777
 778 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
 779
 780         * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
 781         to merge probabilities.
 782         * predict.c (probably_never_executed): Also mark as cold functions
 783         with global 0 profile and guessed local profile.
 784         * profile-count.c (profile_probability::combine_with_count): New
 785         member function.
 786         * profile-count.h (profile_probability::operator*,
 787         profile_probability::operator*=, profile_probability::operator/,
 788         profile_probability::operator/=): Reduce precision to adjusted
 789         and set value to guessed on contradictory divisions.
 790         (profile_probability::combine_with_freq): Remove.
 791         (profile_probability::combine_wiht_count): Declare.
 792         (profile_count::force_nonzero):: Set to adjusted.
 793         (profile_count::probability_in):: Set quality to adjusted.
 794         * tree-ssa-tail-merge.c (replace_block_by): Use
 795         combine_with_count.
 796
 797 2018-01-23  Andrew Waterman  <andrew@sifive.com>
 798             Jim Wilson  <jimw@sifive.com>
 799
 800         * config/riscv/riscv.c (riscv_stack_boundary): New.
 801         (riscv_option_override): Set riscv_stack_boundary.  Handle
 802         riscv_preferred_stack_boundary_arg.
 803         * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
 804         (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
 805         (STACK_BOUNDARY): Set to riscv_stack_boundary.
 806         (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
 807         * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
 808         * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
 809
 810 2018-01-23  H.J. Lu  <hongjiu.lu@intel.com>
 811
 812         PR target/83905
 813         * config/i386/i386.c (ix86_expand_prologue): Use cost reference
 814         of struct ix86_frame.
 815         (ix86_expand_epilogue): Likewise.  Add a local variable for
 816         the reg_save_offset field in struct ix86_frame.
 817
 818 2018-01-23  Bin Cheng  <bin.cheng@arm.com>
 819
 820         PR tree-optimization/82604
 821         * tree-loop-distribution.c (enum partition_kind): New enum item
 822         PKIND_PARTIAL_MEMSET.
 823         (partition_builtin_p): Support above new enum item.
 824         (generate_code_for_partition): Ditto.
 825         (compute_access_range): Differentiate cases that equality can be
 826         proven at all loops, the innermost loops or no loops.
 827         (classify_builtin_st, classify_builtin_ldst): Adjust call to above
 828         function.  Set PKIND_PARTIAL_MEMSET for partition appropriately.
 829         (finalize_partitions, distribute_loop): Don't fuse partition of
 830         PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
 831         (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
 832         parloop is enabled.
 833
 834 2018-01-23  Martin Liska  <mliska@suse.cz>
 835
 836         * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
 837         order to ignore the predictor.
 838         (PRED_POLYMORPHIC_CALL): Likewise.
 839         (PRED_RECURSIVE_CALL): Likewise.
 840
 841 2018-01-23  Martin Liska  <mliska@suse.cz>
 842
 843         * tree-profile.c (tree_profiling): Print function header to
 844         aware reader which function we are working on.
 845         * value-prof.c (gimple_find_values_to_profile): Do not print
 846         not interesting value histograms.
 847
 848 2018-01-23  Martin Liska  <mliska@suse.cz>
 849
 850         * profile-count.h (enum profile_quality): Add
 851         profile_uninitialized as the first value. Do not number values
 852         as they are zero based.
 853         (profile_count::verify): Update sanity check.
 854         (profile_probability::verify): Likewise.
 855
 856 2018-01-23  Nathan Sidwell  <nathan@acm.org>
 857
 858         * doc/invoke.texi (ffor-scope): Deprecate.
 859
 860 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
 861
 862         PR tree-optimization/83510
 863         * domwalk.c (set_all_edges_as_executable): New function.
 864         (dom_walker::dom_walker): Convert bool param
 865         "skip_unreachable_blocks" to enum reachability.  Move setup of
 866         edge flags to set_all_edges_as_executable and only do it when
 867         reachability is REACHABLE_BLOCKS.
 868         * domwalk.h (enum dom_walker::reachability): New enum.
 869         (dom_walker::dom_walker): Convert bool param
 870         "skip_unreachable_blocks" to enum reachability.
 871         (set_all_edges_as_executable): New decl.
 872         * graphite-scop-detection.c  (gather_bbs::gather_bbs): Convert
 873         from false for "skip_unreachable_blocks" to ALL_BLOCKS for
 874         "reachability".
 875         * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
 876         but converting true to REACHABLE_BLOCKS.
 877         * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
 878         * tree-vrp.c
 879         (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
 880         Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
 881         (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
 882         REACHABLE_BLOCKS.
 883         (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
 884         if check_all_array_refs will be called.
 885
 886 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
 887
 888         * tree.c (selftest::test_location_wrappers): Add more test
 889         coverage.
 890
 891 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
 892
 893         * sbitmap.c (selftest::test_set_range): Fix memory leaks.
 894         (selftest::test_bit_in_range): Likewise.
 895
 896 2018-01-23  Richard Sandiford  <richard.sandiford@linaro.org>
 897
 898         PR testsuite/83888
 899         * doc/sourcebuild.texi (vect_float): Say that the selector
 900         only describes the situation when -funsafe-math-optimizations is on.
 901         (vect_float_strict): Document.
 902
 903 2018-01-23  Richard Sandiford  <richard.sandiford@linaro.org>
 904
 905         PR tree-optimization/83965
 906         * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
 907         (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
 908         instead of checking only for a reduction.
 909         (vect_recog_widen_sum_pattern): Likewise.
 910
 911 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
 912
 913         * predict.c (probably_never_executed): Only use precise profile info.
 914         (compute_function_frequency): Skip after inlining hack since we now
 915         have quality checking.
 916
 917 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
 918
 919         * profile-count.h (profile_probability::very_unlikely,
 920         profile_probability::unlikely, profile_probability::even): Set
 921         precision to guessed.
 922
 923 2018-01-23  Richard Biener  <rguenther@suse.de>
 924
 925         PR tree-optimization/83963
 926         * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
 927         Properly terminate dominator walk when crossing the exit edge not
 928         when visiting its source block.
 929
 930 2018-01-23  Jakub Jelinek  <jakub@redhat.com>
 931
 932         PR c++/83918
 933         * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
 934         VIEW_CONVERT_EXPR to wrap CONST_DECLs.
 935
 936 2018-01-22  Jakub Jelinek  <jakub@redhat.com>
 937
 938         PR tree-optimization/83957
 939         * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs.  Remove
 940         semicolon after for body surrounded by braces.
 941
 942         PR tree-optimization/83081
 943         * profile-count.h (profile_probability::split): New method.
 944         * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
 945         Use profile_probability::split.
 946         (do_compare_rtx_and_jump): Fix adjustment of probabilities
 947         when splitting a single conditional jump into 2.
 948
 949 2018-01-22  David Malcolm  <dmalcolm@redhat.com>
 950
 951         PR tree-optimization/69452
 952         * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
 953         decl.
 954
 955 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
 956
 957         * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
 958         * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
 959         * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
 960
 961 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
 962
 963         * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
 964         * config/rl78/rl78.md: New define_expand "movdi"
 965         * config/rl78/rl78.c: New function definition rl78_split_movdi
 966
 967 2018-01-22  Michael Meissner  <meissner@linux.vnet.ibm.com>
 968
 969         PR target/83862
 970         * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
 971         no longer used.
 972         * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
 973         * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
 974         128-bit to produce an UNSPEC move to get the double word with the
 975         signbit and then a shift directly to do signbit.
 976         (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
 977         implementation with a new version that just does either a direct
 978         move or a regular move.  Move memory interface to separate insns.
 979         Move insns so they are next to the expander.
 980         (signbit<mode>2_dm_mem_be): New combiner insns to combine load
 981         with signbit move.  Split big and little endian case.
 982         (signbit<mode>2_dm_mem_le): Likewise.
 983         (signbit<mode>2_dm_<su>ext): Delete, no longer used.
 984         (signbit<mode>2_dm2): Likewise.
 985
 986 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
 987
 988         * config/rl78/rl78.md: New define_expand "anddi3".
 989
 990 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
 991
 992         * config/rl78/rl78.md: New define_expand "umindi3".
 993
 994 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
 995
 996         * config/rl78/rl78.md: New define_expand "smindi3".
 997
 998 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
 999
1000         * config/rl78/rl78.md: New define_expand "smaxdi3".
1001
1002 2018-01-22 Carl Love <cel@us.ibm.com>
1003
1004         * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1005         LVX_V1TI): Add macro expansion.
1006         * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1007         definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1008         VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1009         * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1010         Change check to determine if the instruction is a byte reversing
1011         entry.  Fix typo in comment.
1012         * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1013         for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1014         Add def_builtin calls for new builtins.
1015         * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1016         Add define_insn expansion.
1017
1018 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
1019
1020         * config/rl78/rl78.md: New define_expand "umaxdi3".
1021
1022 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
1023
1024         * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1025         for non-QImode registers
1026
1027 2018-01-22  Richard Biener  <rguenther@suse.de>
1028
1029         PR tree-optimization/83963
1030         * graphite-scop-detection.c (scop_detection::get_sese): Delay
1031         including the loop exit block.
1032         (scop_detection::merge_sese): Likewise.
1033         (scop_detection::add_scop): Do it here instead.
1034
1035 2018-01-22  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
1036
1037         * doc/sourcebuild.texi (arm_softfloat): Document.
1038
1039 2018-01-21  John David Anglin  <danglin@gcc.gnu.org>
1040
1041         PR gcc/77734
1042         * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1043         targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1044         Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1045
1046 2018-01-21  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
1047             David Edelsohn <dje.gcc@gmail.com>
1048
1049         PR target/83946
1050         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1051         Change "crset eq" to "crset 2".
1052         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1053         (*call_indirect_aix<mode>_nospec): Likewise.
1054         (*call_value_indirect_aix<mode>_nospec): Likewise.
1055         (*call_indirect_elfv2<mode>_nospec): Likewise.
1056         (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1057         (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1058         change assembly output from . to $.
1059         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1060         (indirect_jump<mode>_nospec): Change assembly output from . to $.
1061         (*tablejump<mode>_internal1_nospec): Likewise.
1062
1063 2018-01-21  Oleg Endo  <olegendo@gcc.gnu.org>
1064
1065         PR target/80870
1066         * config/sh/sh_optimize_sett_clrt.cc:
1067         Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1068
1069 2018-01-20  Richard Sandiford  <richard.sandiford@linaro.org>
1070
1071         PR tree-optimization/83940
1072         * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1073         offset_dt to vect_constant_def rather than vect_unknown_def_type.
1074         (vect_check_load_store_mask): Add a mask_dt_out parameter and
1075         use it to pass back the definition type.
1076         (vect_check_store_rhs): Likewise rhs_dt_out.
1077         (vect_build_gather_load_calls): Add a mask_dt argument and use
1078         it instead of a call to vect_is_simple_use.
1079         (vectorizable_store): Update calls to vect_check_load_store_mask
1080         and vect_check_store_rhs.  Use the dt returned by the latter instead
1081         of scatter_src_dt.  Use the cached mask_dt and gs_info.offset_dt
1082         instead of calls to vect_is_simple_use.  Pass the scalar rather
1083         than the vector operand to vect_is_simple_use when handling
1084         second and subsequent copies of an rhs value.
1085         (vectorizable_load): Update calls to vect_check_load_store_mask
1086         and vect_build_gather_load_calls.  Use the cached mask_dt and
1087         gs_info.offset_dt instead of calls to vect_is_simple_use.
1088
1089 2018-01-20  Jakub Jelinek  <jakub@redhat.com>
1090
1091         PR middle-end/83945
1092         * tree-emutls.c: Include gimplify.h.
1093         (lower_emutls_2): New function.
1094         (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1095         with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1096         it before further processing.
1097
1098         PR target/83930
1099         * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1100         UINTVAL (trueop1) instead of INTVAL (op1).
1101
1102 2018-01-19  Jakub Jelinek  <jakub@redhat.com>
1103
1104         PR debug/81570
1105         PR debug/83728
1106         * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1107         INCOMING_FRAME_SP_OFFSET if not defined.
1108         (scan_trace): Add ENTRY argument.  If true and
1109         DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1110         emit a note to adjust the CFA offset.
1111         (create_cfi_notes): Adjust scan_trace callers.
1112         (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1113         INCOMING_FRAME_SP_OFFSET in the CIE.
1114         * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1115         * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1116         Likewise.
1117         * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1118         * doc/tm.texi: Regenerated.
1119
1120 2018-01-19  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
1121
1122         PR rtl-optimization/83147
1123         * lra-constraints.c (remove_inheritance_pseudos): Use
1124         lra_substitute_pseudo_within_insn.
1125
1126 2018-01-19  Tom de Vries  <tom@codesourcery.com>
1127             Cesar Philippidis  <cesar@codesourcery.com>
1128
1129         PR target/83920
1130         * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1131
1132 2018-01-19  Cesar Philippidis  <cesar@codesourcery.com>
1133
1134         PR target/83790
1135         * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1136         spaces for function labels.
1137
1138 2018-01-19  Martin Liska  <mliska@suse.cz>
1139
1140         * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1141         (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1142         (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1143         (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1144         (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1145         (PRED_CONST_RETURN): Change from 69 to 65.
1146         (PRED_NULL_RETURN): Change from 91 to 71.
1147         (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1148         (PRED_LOOP_GUARD): Change from 66 to 73.
1149
1150 2018-01-19  Martin Liska  <mliska@suse.cz>
1151
1152         * predict.c (predict_insn_def): Add new assert.
1153         (struct branch_predictor): Change type to signed integer.
1154         (test_prediction_value_range): Amend test to cover
1155         PROB_UNINITIALIZED.
1156         * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1157         (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1158         (PRED_LOOP_ITERATIONS_MAX): Likewise.
1159         (PRED_LOOP_IV_COMPARE): Likewise.
1160         * predict.h (PROB_UNINITIALIZED): Define new constant.
1161
1162 2018-01-19  Martin Liska  <mliska@suse.cz>
1163
1164         * predict.c (dump_prediction): Add new format for
1165         analyze_brprob.py script which is enabled with -details
1166         suboption.
1167         * profile-count.h (precise_p): New function.
1168
1169 2018-01-19  Richard Sandiford  <richard.sandiford@linaro.org>
1170
1171         PR tree-optimization/83922
1172         * tree-vect-loop.c (vect_verify_full_masking): Return false if
1173         there are no statements that need masking.
1174         (vect_active_double_reduction_p): New function.
1175         (vect_analyze_loop_operations): Use it when handling phis that
1176         are not in the loop header.
1177
1178 2018-01-19  Richard Sandiford  <richard.sandiford@linaro.org>
1179
1180         PR tree-optimization/83914
1181         * tree-vect-loop.c (vectorizable_induction): Don't convert
1182         init_expr or apply the peeling adjustment for inductions
1183         that are nested within the vectorized loop.
1184
1185 2018-01-19  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
1186
1187         * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1188         instead of NEG.
1189
1190 2018-01-18  Jakub Jelinek  <jakub@redhat.com>
1191
1192         PR sanitizer/81715
1193         PR testsuite/83882
1194         * function.h (gimplify_parameters): Add gimple_seq * argument.
1195         * function.c: Include gimple.h and options.h.
1196         (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1197         for the added local temporaries if needed.
1198         * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1199         if there are any parameter cleanups, wrap whole body into a
1200         try/finally with the cleanups.
1201
1202 2018-01-18  Wilco Dijkstra  <wdijkstr@arm.com>
1203
1204         PR target/82964
1205         * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1206         Use GET_MODE_CLASS for scalar floating point.
1207
1208 2018-01-18  Jan Hubicka  <hubicka@ucw.cz>
1209
1210         PR ipa/82256
1211         patch by PaX Team
1212         * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1213         Fix call of call_cgraph_insertion_hooks.
1214
1215 2018-01-18  Martin Sebor  <msebor@redhat.com>
1216
1217         * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1218
1219 2018-01-18  Jan Hubicka  <hubicka@ucw.cz>
1220
1221         PR ipa/83619
1222         * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1223         frequencies.
1224
1225 2018-01-18  Boris Kolpackov  <boris@codesynthesis.com>
1226
1227         PR other/70268
1228         * common.opt: (-ffile-prefix-map): New option.
1229         * opts.c (common_handle_option): Defer it.
1230         * opts-global.c (handle_common_deferred_options): Handle it.
1231         * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1232         * file-prefix-map.h: New file.
1233         (remap_debug_filename, add_debug_prefix_map): ...here.
1234         (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1235         * final.c (debug_prefix_map, add_debug_prefix_map
1236         remap_debug_filename): Move to...
1237         * file-prefix-map.c: New file.
1238         (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1239         generalize, get rid of alloca(), use strrchr() instead of strchr().
1240         (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1241         Implement in terms of add_prefix_map().
1242         (remap_macro_filename, remap_debug_filename): Implement in term of
1243         remap_filename().
1244         * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1245         * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1246         * dbxout.c: Include file-prefix-map.h.
1247         * varasm.c: Likewise.
1248         * vmsdbgout.c: Likewise.
1249         * xcoffout.c: Likewise.
1250         * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1251         * doc/cppopts.texi (-fmacro-prefix-map): Document.
1252         * doc/invoke.texi (-ffile-prefix-map): Document.
1253         (-fdebug-prefix-map): Update description.
1254
1255 2018-01-18  Martin Liska  <mliska@suse.cz>
1256
1257         * config/i386/i386.c (indirect_thunk_name): Document that also
1258         lfence is emitted.
1259         (output_indirect_thunk): Document why both instructions
1260         (pause and lfence) are generated.
1261
1262 2018-01-18  Richard Biener  <rguenther@suse.de>
1263
1264         PR tree-optimization/83887
1265         * graphite-scop-detection.c
1266         (scop_detection::get_nearest_dom_with_single_entry): Remove.
1267         (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1268         (scop_detection::merge_sese): Re-implement with a flood-fill
1269         algorithm that properly finds a SESE region if it exists.
1270
1271 2018-01-18  Jakub Jelinek  <jakub@redhat.com>
1272
1273         PR c/61240
1274         * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1275         pointer_diff optimizations use view_convert instead of convert.
1276
1277 2018-01-17  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
1278
1279         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1280         Generate different code for -mno-speculate-indirect-jumps.
1281         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1282         (*call_indirect_aix<mode>): Disable for
1283         -mno-speculate-indirect-jumps.
1284         (*call_indirect_aix<mode>_nospec): New define_insn.
1285         (*call_value_indirect_aix<mode>): Disable for
1286         -mno-speculate-indirect-jumps.
1287         (*call_value_indirect_aix<mode>_nospec): New define_insn.
1288         (*sibcall_nonlocal_sysv<mode>): Generate different code for
1289         -mno-speculate-indirect-jumps.
1290         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1291
1292 2018-01-17  Michael Meissner  <meissner@linux.vnet.ibm.com>
1293
1294         * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1295         long double type, set the flags for noting the default long double
1296         type, even if we don't pass or return a long double type.
1297
1298 2018-01-17  Jan Hubicka  <hubicka@ucw.cz>
1299
1300         PR ipa/83051
1301         * ipa-inline.c (flatten_function): Do not overwrite final inlining
1302         failure.
1303
1304 2018-01-17  Will Schmidt  <will_schmidt@vnet.ibm.com>
1305
1306         * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1307         support for merge[hl].
1308         (fold_mergehl_helper): New helper function.
1309         (tree-vector-builder.h): New #include for tree_vector_builder usage.
1310         * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1311         (altivec_vmrglw_direct): Add xxmrglw insn.
1312
1313 2018-01-17  Andrew Waterman  <andrew@sifive.com>
1314
1315         * config/riscv/riscv.c (riscv_conditional_register_usage): If
1316         UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1317
1318 2018-01-17  David Malcolm  <dmalcolm@redhat.com>
1319
1320         PR lto/83121
1321         * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1322         call the lto_location_cache before reading the
1323         DECL_SOURCE_LOCATION of the types.
1324
1325 2018-01-17  Wilco Dijkstra  <wdijkstr@arm.com>
1326             Richard Sandiford  <richard.sandiford@linaro.org>
1327
1328         * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1329         * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1330         (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1331         SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1332         * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1333         Add declaration.
1334         * config/aarch64/constraints.md (aarch64_movti_operand):
1335         Limit immediates.
1336         * config/aarch64/predicates.md (Uti): Add new constraint.
1337
1338 2018-01-17 Carl Love  <cel@us.ibm.com>
1339         * config/rs6000/vsx.md (define_expand xl_len_r,
1340         define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1341         (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1342         lxvll.
1343         (define_expand, define_insn): Move the shift left from  the
1344         define_insn to the define_expand for lxvl and stxvl instructions.
1345         * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1346         and XL_LEN_R definitions to PURE.
1347
1348 2018-01-17  Uros Bizjak  <ubizjak@gmail.com>
1349
1350         * config/i386/i386.c (indirect_thunk_name): Declare regno
1351         as unsigned int.  Compare regno with INVALID_REGNUM.
1352         (output_indirect_thunk): Ditto.
1353         (output_indirect_thunk_function): Ditto.
1354         (ix86_code_end): Declare regno as unsigned int.  Use INVALID_REGNUM
1355         in the call to output_indirect_thunk_function.
1356
1357 2018-01-17  Richard Sandiford  <richard.sandiford@linaro.org>
1358
1359         PR middle-end/83884
1360         * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1361         rather than the size of inner_type to determine the stack slot size
1362         when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1363
1364 2018-01-16  Sebastian Peryt  <sebastian.peryt@intel.com>
1365
1366         PR target/83546
1367         * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1368         to PTA_SILVERMONT.
1369
1370 2018-01-16  Michael Meissner  <meissner@linux.vnet.ibm.com>
1371
1372         * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1373         endian Linux systems to optionally enable multilibs for selecting
1374         the long double type if the user configured an explicit type.
1375         * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1376         have no long double multilibs if not defined.
1377         * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1378         warn if the user used -mabi={ieee,ibm}longdouble and we built
1379         multilibs for long double.
1380         * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1381         appropriate multilib option.
1382         (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1383         multilib options.
1384         * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1385         for building long double multilibs.
1386         * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1387
1388 2018-01-16  John David Anglin  <danglin@gcc.gnu.org>
1389
1390         * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1391         copies.
1392
1393         * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1394         64 bits.
1395         * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1396         128 bits.
1397
1398         * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1399         variables.
1400
1401         * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1402         return value.
1403
1404 2018-01-16  Eric Botcazou  <ebotcazou@adacore.com>
1405
1406         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1407         ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1408
1409 2018-01-16  Kelvin Nilsen  <kelvin@gcc.gnu.org>
1410
1411         * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1412         different rtl trees depending on TARGET_64BIT.
1413         (rs6000_gen_lvx): Likewise.
1414
1415 2018-01-16  Eric Botcazou  <ebotcazou@adacore.com>
1416
1417         * config/visium/visium.md (nop): Tweak comment.
1418         (hazard_nop): Likewise.
1419
1420 2018-01-16  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
1421
1422         * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1423         -mspeculate-indirect-jumps.
1424         * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1425         for -mno-speculate-indirect-jumps.
1426         (*call_indirect_elfv2<mode>_nospec): New define_insn.
1427         (*call_value_indirect_elfv2<mode>): Disable for
1428         -mno-speculate-indirect-jumps.
1429         (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1430         (indirect_jump): Emit different RTL for
1431         -mno-speculate-indirect-jumps.
1432         (*indirect_jump<mode>): Disable for
1433         -mno-speculate-indirect-jumps.
1434         (*indirect_jump<mode>_nospec): New define_insn.
1435         (tablejump): Emit different RTL for
1436         -mno-speculate-indirect-jumps.
1437         (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1438         (tablejumpsi_nospec): New define_expand.
1439         (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1440         (tablejumpdi_nospec): New define_expand.
1441         (*tablejump<mode>_internal1): Disable for
1442         -mno-speculate-indirect-jumps.
1443         (*tablejump<mode>_internal1_nospec): New define_insn.
1444         * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1445         option.
1446
1447 2018-01-16  Artyom Skrobov tyomitch@gmail.com
1448
1449         * caller-save.c (insert_save): Drop unnecessary parameter.  All
1450         callers updated.
1451
1452 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
1453             Richard Biener  <rguenth@suse.de>
1454
1455         PR libgomp/83590
1456         * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1457         return early, inline manually is_gimple_sizepos.  Make sure if we
1458         call gimplify_expr we don't end up with a gimple constant.
1459         * tree.c (variably_modified_type_p): Don't return true for
1460         is_gimple_constant (_t).  Inline manually is_gimple_sizepos.
1461         * gimplify.h (is_gimple_sizepos): Remove.
1462
1463 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
1464
1465         PR tree-optimization/83857
1466         * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1467         vectorizable_live_operation for pure SLP statements.
1468         (vectorizable_live_operation): Handle PHIs.
1469
1470 2018-01-16  Richard Biener  <rguenther@suse.de>
1471
1472         PR tree-optimization/83867
1473         * tree-vect-stmts.c (vect_transform_stmt): Precompute
1474         nested_in_vect_loop_p since the scalar stmt may get invalidated.
1475
1476 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
1477
1478         PR c/83844
1479         * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1480         multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1481         If off is not INTEGER_CST, issue a may not be aligned warning
1482         rather than isn't aligned.  Use isn%'t rather than isn't.
1483         * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1484         into MULT_EXPR.
1485         <case MULT_EXPR>: Improve the case when bottom and one of the
1486         MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1487         operand, in that case check if the other operand is multiple of
1488         bottom divided by the INTEGER_CST operand.
1489
1490 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
1491
1492         PR target/83858
1493         * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1494         * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1495         * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1496         pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1497         * config/pa/pa.c (pa_function_arg_advance): Likewise.
1498         (pa_function_arg, pa_arg_partial_bytes): Likewise.
1499         (pa_function_arg_size): New function.
1500
1501 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
1502
1503         * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1504         in a separate statement.
1505
1506 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
1507
1508         PR tree-optimization/83847
1509         * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1510         group gathers and scatters.
1511
1512 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
1513
1514         PR rtl-optimization/86620
1515         * params.def (max-sched-ready-insns): Bump minimum value to 1.
1516
1517         PR rtl-optimization/83213
1518         * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1519         to last if both are JUMP_INSNs.
1520
1521         PR tree-optimization/83843
1522         * gimple-ssa-store-merging.c
1523         (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1524         store_immediate_info for bswap/nop orig_stores.
1525
1526 2018-01-15  Andrew Waterman  <andrew@sifive.com>
1527
1528         * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1529         !TARGET_MUL.
1530         <UDIV>: Increase cost if !TARGET_DIV.
1531
1532 2018-01-15  Segher Boessenkool  <segher@kernel.crashing.org>
1533
1534         * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1535         (define_attr "cr_logical_3op"): New.
1536         (cceq_ior_compare): Adjust.
1537         (cceq_ior_compare_complement): Adjust.
1538         (*cceq_rev_compare): Adjust.
1539         * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1540         (is_cracked_insn): Adjust.
1541         (insn_must_be_first_in_group): Adjust.
1542         * config/rs6000/40x.md: Adjust.
1543         * config/rs6000/440.md: Adjust.
1544         * config/rs6000/476.md: Adjust.
1545         * config/rs6000/601.md: Adjust.
1546         * config/rs6000/603.md: Adjust.
1547         * config/rs6000/6xx.md: Adjust.
1548         * config/rs6000/7450.md: Adjust.
1549         * config/rs6000/7xx.md: Adjust.
1550         * config/rs6000/8540.md: Adjust.
1551         * config/rs6000/cell.md: Adjust.
1552         * config/rs6000/e300c2c3.md: Adjust.
1553         * config/rs6000/e500mc.md: Adjust.
1554         * config/rs6000/e500mc64.md: Adjust.
1555         * config/rs6000/e5500.md: Adjust.
1556         * config/rs6000/e6500.md: Adjust.
1557         * config/rs6000/mpc.md: Adjust.
1558         * config/rs6000/power4.md: Adjust.
1559         * config/rs6000/power5.md: Adjust.
1560         * config/rs6000/power6.md: Adjust.
1561         * config/rs6000/power7.md: Adjust.
1562         * config/rs6000/power8.md: Adjust.
1563         * config/rs6000/power9.md: Adjust.
1564         * config/rs6000/rs64.md: Adjust.
1565         * config/rs6000/titan.md: Adjust.
1566
1567 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
1568
1569         * config/i386/predicates.md (indirect_branch_operand): Rewrite
1570         ix86_indirect_branch_register logic.
1571
1572 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
1573
1574         * config/i386/constraints.md (Bs): Update
1575         ix86_indirect_branch_register check.  Don't check
1576         ix86_indirect_branch_register with GOT_memory_operand.
1577         (Bw): Likewise.
1578         * config/i386/predicates.md (GOT_memory_operand): Don't check
1579         ix86_indirect_branch_register here.
1580         (GOT32_symbol_operand): Likewise.
1581
1582 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
1583
1584         * config/i386/predicates.md (constant_call_address_operand):
1585         Rewrite ix86_indirect_branch_register logic.
1586         (sibcall_insn_operand): Likewise.
1587
1588 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
1589
1590         * config/i386/constraints.md (Bs): Replace
1591         ix86_indirect_branch_thunk_register with
1592         ix86_indirect_branch_register.
1593         (Bw): Likewise.
1594         * config/i386/i386.md (indirect_jump): Likewise.
1595         (tablejump): Likewise.
1596         (*sibcall_memory): Likewise.
1597         (*sibcall_value_memory): Likewise.
1598         Peepholes of indirect call and jump via memory: Likewise.
1599         * config/i386/i386.opt: Likewise.
1600         * config/i386/predicates.md (indirect_branch_operand): Likewise.
1601         (GOT_memory_operand): Likewise.
1602         (call_insn_operand): Likewise.
1603         (sibcall_insn_operand): Likewise.
1604         (GOT32_symbol_operand): Likewise.
1605
1606 2018-01-15  Jakub Jelinek  <jakub@redhat.com>
1607
1608         PR middle-end/83837
1609         * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1610         type rather than type addr's type points to.
1611         (expand_omp_atomic_mutex): Likewise.
1612         (expand_omp_atomic): Likewise.
1613
1614 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
1615
1616         PR target/83839
1617         * config/i386/i386.c (output_indirect_thunk_function): Use
1618         ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1619         for  __x86_return_thunk.
1620
1621 2018-01-15  Richard Biener  <rguenther@suse.de>
1622
1623         PR middle-end/83850
1624         * expmed.c (extract_bit_field_1): Fix typo.
1625
1626 2018-01-15  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
1627
1628         PR target/83687
1629         * config/arm/iterators.md (VF): New mode iterator.
1630         * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1631         Remove integer-related logic from pattern.
1632         (neon_vabd<mode>_3): Likewise.
1633
1634 2018-01-15  Jakub Jelinek  <jakub@redhat.com>
1635
1636         PR middle-end/82694
1637         * common.opt (fstrict-overflow): No longer an alias.
1638         (fwrapv-pointer): New option.
1639         * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1640         also for pointer types based on flag_wrapv_pointer.
1641         * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1642         opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1643         opts->x_flag_wrapv got set.
1644         * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1645         changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1646         POINTER_TYPE_OVERFLOW_UNDEFINED.
1647         * match.pd: Likewise in address comparison pattern.
1648         * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1649
1650 2018-01-15  Richard Biener  <rguenther@suse.de>
1651
1652         PR lto/83804
1653         * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1654         from TYPE_FIELDS.  Free TYPE_BINFO if not used by devirtualization.
1655         Reset type names to their identifier if their TYPE_DECL doesn't
1656         have linkage (and thus is used for ODR and devirt).
1657         (save_debug_info_for_decl): Remove.
1658         (save_debug_info_for_type): Likewise.
1659         (add_tree_to_fld_list): Adjust.
1660         * tree-pretty-print.c (dump_generic_node): Make dumping of
1661         type names more robust.
1662
1663 2018-01-15  Richard Biener  <rguenther@suse.de>
1664
1665         * BASE-VER: Bump to 8.0.1.
1666
1667 2018-01-14  Martin Sebor  <msebor@redhat.com>
1668
1669         PR other/83508
1670         * builtins.c (check_access): Avoid warning when the no-warning bit
1671         is set.
1672
1673 2018-01-14  Cory Fields  <cory-nospam-@coryfields.com>
1674
1675         * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1676         * ira-color (allocno_hard_regs_compare): Likewise.
1677
1678 2018-01-14  Nathan Rossi  <nathan@nathanrossi.com>
1679
1680         PR target/83013
1681         * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1682         Use .pushsection/.popsection.
1683
1684 2018-01-14  Martin Sebor  <msebor@redhat.com>
1685
1686         PR c++/81327
1687         * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1688
1689 2018-01-14  Jakub Jelinek  <jakub@redhat.com>
1690
1691         * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1692         entry from extra_headers.
1693         (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1694         extra_headers, make the list bitwise identical to the i?86-*-* one.
1695
1696 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
1697
1698         * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1699         -mcmodel=large with -mindirect-branch=thunk,
1700         -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1701         -mfunction-return=thunk-extern.
1702         * doc/invoke.texi: Document -mcmodel=large is incompatible with
1703         -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1704         -mfunction-return=thunk and -mfunction-return=thunk-extern.
1705
1706 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
1707
1708         * config/i386/i386.c (print_reg): Print the name of the full
1709         integer register without '%'.
1710         (ix86_print_operand): Handle 'V'.
1711          * doc/extend.texi: Document 'V' modifier.
1712
1713 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
1714
1715         * config/i386/constraints.md (Bs): Disallow memory operand for
1716         -mindirect-branch-register.
1717         (Bw): Likewise.
1718         * config/i386/predicates.md (indirect_branch_operand): Likewise.
1719         (GOT_memory_operand): Likewise.
1720         (call_insn_operand): Likewise.
1721         (sibcall_insn_operand): Likewise.
1722         (GOT32_symbol_operand): Likewise.
1723         * config/i386/i386.md (indirect_jump): Call convert_memory_address
1724         for -mindirect-branch-register.
1725         (tablejump): Likewise.
1726         (*sibcall_memory): Likewise.
1727         (*sibcall_value_memory): Likewise.
1728         Disallow peepholes of indirect call and jump via memory for
1729         -mindirect-branch-register.
1730         (*call_pop): Replace m with Bw.
1731         (*call_value_pop): Likewise.
1732         (*sibcall_pop_memory): Replace m with Bs.
1733         * config/i386/i386.opt (mindirect-branch-register): New option.
1734         * doc/invoke.texi: Document -mindirect-branch-register option.
1735
1736 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
1737
1738         * config/i386/i386-protos.h (ix86_output_function_return): New.
1739         * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1740         set function_return_type.
1741         (indirect_thunk_name): Add ret_p to indicate thunk for function
1742         return.
1743         (output_indirect_thunk_function): Pass false to
1744         indirect_thunk_name.
1745         (ix86_output_indirect_branch_via_reg): Likewise.
1746         (ix86_output_indirect_branch_via_push): Likewise.
1747         (output_indirect_thunk_function): Create alias for function
1748         return thunk if regno < 0.
1749         (ix86_output_function_return): New function.
1750         (ix86_handle_fndecl_attribute): Handle function_return.
1751         (ix86_attribute_table): Add function_return.
1752         * config/i386/i386.h (machine_function): Add
1753         function_return_type.
1754         * config/i386/i386.md (simple_return_internal): Use
1755         ix86_output_function_return.
1756         (simple_return_internal_long): Likewise.
1757         * config/i386/i386.opt (mfunction-return=): New option.
1758         (indirect_branch): Mention -mfunction-return=.
1759         * doc/extend.texi: Document function_return function attribute.
1760         * doc/invoke.texi: Document -mfunction-return= option.
1761
1762 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
1763
1764         * config/i386/i386-opts.h (indirect_branch): New.
1765         * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1766         * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1767         with local indirect jump when converting indirect call and jump.
1768         (ix86_set_indirect_branch_type): New.
1769         (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1770         (indirectlabelno): New.
1771         (indirect_thunk_needed): Likewise.
1772         (indirect_thunk_bnd_needed): Likewise.
1773         (indirect_thunks_used): Likewise.
1774         (indirect_thunks_bnd_used): Likewise.
1775         (INDIRECT_LABEL): Likewise.
1776         (indirect_thunk_name): Likewise.
1777         (output_indirect_thunk): Likewise.
1778         (output_indirect_thunk_function): Likewise.
1779         (ix86_output_indirect_branch_via_reg): Likewise.
1780         (ix86_output_indirect_branch_via_push): Likewise.
1781         (ix86_output_indirect_branch): Likewise.
1782         (ix86_output_indirect_jmp): Likewise.
1783         (ix86_code_end): Call output_indirect_thunk_function if needed.
1784         (ix86_output_call_insn): Call ix86_output_indirect_branch if
1785         needed.
1786         (ix86_handle_fndecl_attribute): Handle indirect_branch.
1787         (ix86_attribute_table): Add indirect_branch.
1788         * config/i386/i386.h (machine_function): Add indirect_branch_type
1789         and has_local_indirect_jump.
1790         * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1791         to true.
1792         (tablejump): Likewise.
1793         (*indirect_jump): Use ix86_output_indirect_jmp.
1794         (*tablejump_1): Likewise.
1795         (simple_return_indirect_internal): Likewise.
1796         * config/i386/i386.opt (mindirect-branch=): New option.
1797         (indirect_branch): New.
1798         (keep): Likewise.
1799         (thunk): Likewise.
1800         (thunk-inline): Likewise.
1801         (thunk-extern): Likewise.
1802         * doc/extend.texi: Document indirect_branch function attribute.
1803         * doc/invoke.texi: Document -mindirect-branch= option.
1804
1805 2018-01-14  Jan Hubicka  <hubicka@ucw.cz>
1806
1807         PR ipa/83051
1808         * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1809
1810 2018-01-14  Richard Sandiford  <richard.sandiford@linaro.org>
1811
1812         * ipa-inline.c (want_inline_small_function_p): Return false if
1813         inlining has already failed with CIF_FINAL_ERROR.
1814         (update_caller_keys): Call want_inline_small_function_p before
1815         can_inline_edge_p.
1816         (update_callee_keys): Likewise.
1817
1818 2018-01-10  Kelvin Nilsen  <kelvin@gcc.gnu.org>
1819
1820         * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1821         New function.
1822         (rs6000_quadword_masked_address_p): Likewise.
1823         (quad_aligned_load_p): Likewise.
1824         (quad_aligned_store_p): Likewise.
1825         (const_load_sequence_p): Add comment to describe the outer-most loop.
1826         (mimic_memory_attributes_and_flags): New function.
1827         (rs6000_gen_stvx): Likewise.
1828         (replace_swapped_aligned_store): Likewise.
1829         (rs6000_gen_lvx): Likewise.
1830         (replace_swapped_aligned_load): Likewise.
1831         (replace_swapped_load_constant): Capitalize argument name in
1832         comment describing this function.
1833         (rs6000_analyze_swaps): Add a third pass to search for vector loads
1834         and stores that access quad-word aligned addresses and replace
1835         with stvx or lvx instructions when appropriate.
1836         * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1837         New function prototype.
1838         (rs6000_quadword_masked_address_p): Likewise.
1839         (rs6000_gen_lvx): Likewise.
1840         (rs6000_gen_stvx): Likewise.
1841         * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1842         VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1843         when memory address is aligned.
1844         (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1845         this split to select lvx instruction when memory address is aligned.
1846         (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1847         instruction when memory address is aligned.
1848         (*vsx_le_perm_load_v16qi): Likewise.
1849         (four unnamed splitters): Modify to select the stvx instruction
1850         when memory is aligned.
1851
1852 2018-01-13  Jan Hubicka  <hubicka@ucw.cz>
1853
1854         * predict.c (determine_unlikely_bbs): Handle correctly BBs
1855         which appears in the queue multiple times.
1856
1857 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
1858             Alan Hayward  <alan.hayward@arm.com>
1859             David Sherwood  <david.sherwood@arm.com>
1860
1861         * tree-vectorizer.h (vec_lower_bound): New structure.
1862         (_loop_vec_info): Add check_nonzero and lower_bounds.
1863         (LOOP_VINFO_CHECK_NONZERO): New macro.
1864         (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1865         (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1866         * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1867         fields.  Make seg_len the distance travelled, not including the
1868         access size.
1869         (dr_direction_indicator): Declare.
1870         (dr_zero_step_indicator): Likewise.
1871         (dr_known_forward_stride_p): Likewise.
1872         * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1873         tree-ssanames.h.
1874         (runtime_alias_check_p): Allow runtime alias checks with
1875         variable strides.
1876         (operator ==): Compare access_size and align.
1877         (prune_runtime_alias_test_list): Rework for new distinction between
1878         the access_size and seg_len.
1879         (create_intersect_range_checks_index): Likewise.  Cope with polynomial
1880         segment lengths.
1881         (get_segment_min_max): New function.
1882         (create_intersect_range_checks): Use it.
1883         (dr_step_indicator): New function.
1884         (dr_direction_indicator): Likewise.
1885         (dr_zero_step_indicator): Likewise.
1886         (dr_known_forward_stride_p): Likewise.
1887         * tree-loop-distribution.c (data_ref_segment_size): Return
1888         DR_STEP * (niters - 1).
1889         (compute_alias_check_pairs): Update call to the dr_with_seg_len
1890         constructor.
1891         * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1892         (vect_preserves_scalar_order_p): New function, split out from...
1893         (vect_analyze_data_ref_dependence): ...here.  Check for zero steps.
1894         (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1895         (vect_vfa_access_size): New function.
1896         (vect_vfa_align): Likewise.
1897         (vect_compile_time_alias): Take access_size_a and access_b arguments.
1898         (dump_lower_bound): New function.
1899         (vect_check_lower_bound): Likewise.
1900         (vect_small_gap_p): Likewise.
1901         (vectorizable_with_step_bound_p): Likewise.
1902         (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1903         depencies if the vectorization factor is 1.  Convert the checks
1904         for nonzero steps into checks on the bounds of DR_STEP.  Try using
1905         a bunds check for variable steps if the minimum required step is
1906         relatively small. Update calls to the dr_with_seg_len
1907         constructor and to vect_compile_time_alias.
1908         * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1909         function.
1910         (vect_loop_versioning): Call it.
1911         * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1912         when retrying.
1913         (vect_estimate_min_profitable_iters): Account for any bounds checks.
1914
1915 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
1916             Alan Hayward  <alan.hayward@arm.com>
1917             David Sherwood  <david.sherwood@arm.com>
1918
1919         * doc/sourcebuild.texi (vect_scatter_store): Document.
1920         * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1921         optabs.
1922         * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1923         Document.
1924         * genopinit.c (main): Add supports_vec_scatter_store and
1925         supports_vec_scatter_store_cached to target_optabs.
1926         * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1927         IFN_MASK_SCATTER_STORE.
1928         * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1929         functions.
1930         * internal-fn.h (internal_store_fn_p): Declare.
1931         (internal_fn_stored_value_index): Likewise.
1932         * internal-fn.c (scatter_store_direct): New macro.
1933         (expand_scatter_store_optab_fn): New function.
1934         (direct_scatter_store_optab_supported_p): New macro.
1935         (internal_store_fn_p): New function.
1936         (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1937         IFN_MASK_SCATTER_STORE.
1938         (internal_fn_mask_index): Likewise.
1939         (internal_fn_stored_value_index): New function.
1940         (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1941         for scatter stores.
1942         * optabs-query.h (supports_vec_scatter_store_p): Declare.
1943         * optabs-query.c (supports_vec_scatter_store_p): New function.
1944         * tree-vectorizer.h (vect_get_store_rhs): Declare.
1945         * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1946         true for scatter stores.
1947         (vect_gather_scatter_fn_p): Handle scatter stores too.
1948         (vect_check_gather_scatter): Consider using scatter stores if
1949         supports_vec_scatter_store_p.
1950         * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1951         scatter stores too.
1952         * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1953         internal_fn_stored_value_index.
1954         (check_load_store_masking): Handle scatter stores too.
1955         (vect_get_store_rhs): Make public.
1956         (vectorizable_call): Use internal_store_fn_p.
1957         (vectorizable_store): Handle scatter store internal functions.
1958         (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1959         when deciding whether the end of the group has been reached.
1960         * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1961         * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1962         (mask_scatter_store<mode>): New insns.
1963
1964 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
1965             Alan Hayward  <alan.hayward@arm.com>
1966             David Sherwood  <david.sherwood@arm.com>
1967
1968         * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1969         * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1970         * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1971         function.
1972         (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1973         Use vect_truncate_gather_scatter_offset if we can't treat the
1974         operation as a normal gather load or scatter store.
1975         (get_group_load_store_type): Take the gather_scatter_info
1976         as argument.  Try using a gather load or scatter store for
1977         single-element groups.
1978         (get_load_store_type): Update calls to get_group_load_store_type
1979         and vect_use_strided_gather_scatters_p.
1980
1981 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
1982             Alan Hayward  <alan.hayward@arm.com>
1983             David Sherwood  <david.sherwood@arm.com>
1984
1985         * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1986         optional tree argument.
1987         * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1988         null target hooks.
1989         (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1990         but continue to use the current value as a fallback.
1991         (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1992         to compare the updates.
1993         * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1994         (get_load_store_type): Use it when handling a strided access.
1995         (vect_get_strided_load_store_ops): New function.
1996         (vect_get_data_ptr_increment): Likewise.
1997         (vectorizable_load): Handle strided gather loads.  Always pass
1998         a step to vect_create_data_ref_ptr and bump_vector_ptr.
1999
2000 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2001             Alan Hayward  <alan.hayward@arm.com>
2002             David Sherwood  <david.sherwood@arm.com>
2003
2004         * doc/md.texi (gather_load@var{m}): Document.
2005         (mask_gather_load@var{m}): Likewise.
2006         * genopinit.c (main): Add supports_vec_gather_load and
2007         supports_vec_gather_load_cached to target_optabs.
2008         * optabs-tree.c (init_tree_optimization_optabs): Use
2009         ggc_cleared_alloc to allocate target_optabs.
2010         * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2011         * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2012         functions.
2013         * internal-fn.h (internal_load_fn_p): Declare.
2014         (internal_gather_scatter_fn_p): Likewise.
2015         (internal_fn_mask_index): Likewise.
2016         (internal_gather_scatter_fn_supported_p): Likewise.
2017         * internal-fn.c (gather_load_direct): New macro.
2018         (expand_gather_load_optab_fn): New function.
2019         (direct_gather_load_optab_supported_p): New macro.
2020         (direct_internal_fn_optab): New function.
2021         (internal_load_fn_p): Likewise.
2022         (internal_gather_scatter_fn_p): Likewise.
2023         (internal_fn_mask_index): Likewise.
2024         (internal_gather_scatter_fn_supported_p): Likewise.
2025         * optabs-query.c (supports_at_least_one_mode_p): New function.
2026         (supports_vec_gather_load_p): Likewise.
2027         * optabs-query.h (supports_vec_gather_load_p): Declare.
2028         * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2029         and memory_type field.
2030         (NUM_PATTERNS): Bump to 15.
2031         * tree-vect-data-refs.c: Include internal-fn.h.
2032         (vect_gather_scatter_fn_p): New function.
2033         (vect_describe_gather_scatter_call): Likewise.
2034         (vect_check_gather_scatter): Try using internal functions for
2035         gather loads.  Recognize existing calls to a gather load function.
2036         (vect_analyze_data_refs): Consider using gather loads if
2037         supports_vec_gather_load_p.
2038         * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2039         (vect_get_gather_scatter_offset_type): Likewise.
2040         (vect_convert_mask_for_vectype): Likewise.
2041         (vect_add_conversion_to_patterm): Likewise.
2042         (vect_try_gather_scatter_pattern): Likewise.
2043         (vect_recog_gather_scatter_pattern): New pattern recognizer.
2044         (vect_vect_recog_func_ptrs): Add it.
2045         * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2046         internal_fn_mask_index and internal_gather_scatter_fn_p.
2047         (check_load_store_masking): Take the gather_scatter_info as an
2048         argument and handle gather loads.
2049         (vect_get_gather_scatter_ops): New function.
2050         (vectorizable_call): Check internal_load_fn_p.
2051         (vectorizable_load): Likewise.  Handle gather load internal
2052         functions.
2053         (vectorizable_store): Update call to check_load_store_masking.
2054         * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2055         * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2056         * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2057         (aarch64_gather_scale_operand_d): New predicates.
2058         * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2059         (mask_gather_load<mode>): New insns.
2060
2061 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2062             Alan Hayward  <alan.hayward@arm.com>
2063             David Sherwood  <david.sherwood@arm.com>
2064
2065         * optabs.def (fold_left_plus_optab): New optab.
2066         * doc/md.texi (fold_left_plus_@var{m}): Document.
2067         * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2068         * internal-fn.c (fold_left_direct): Define.
2069         (expand_fold_left_optab_fn): Likewise.
2070         (direct_fold_left_optab_supported_p): Likewise.
2071         * fold-const-call.c (fold_const_fold_left): New function.
2072         (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2073         * tree-parloops.c (valid_reduction_p): New function.
2074         (gather_scalar_reductions): Use it.
2075         * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2076         (vect_finish_replace_stmt): Declare.
2077         * tree-vect-loop.c (fold_left_reduction_fn): New function.
2078         (needs_fold_left_reduction_p): New function, split out from...
2079         (vect_is_simple_reduction): ...here.  Accept reductions that
2080         forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2081         (vect_force_simple_reduction): Also store the reduction type in
2082         the assignment's STMT_VINFO_REDUC_TYPE.
2083         (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2084         (merge_with_identity): New function.
2085         (vect_expand_fold_left): Likewise.
2086         (vectorize_fold_left_reduction): Likewise.
2087         (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION.  Leave the
2088         scalar phi in place for it.  Check for target support and reject
2089         cases that would reassociate the operation.  Defer the transform
2090         phase to vectorize_fold_left_reduction.
2091         * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2092         * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2093         (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2094
2095 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2096
2097         * tree-if-conv.c (predicate_mem_writes): Remove redundant
2098         call to ifc_temp_var.
2099
2100 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2101             Alan Hayward  <alan.hayward@arm.com>
2102             David Sherwood  <david.sherwood@arm.com>
2103
2104         * target.def (legitimize_address_displacement): Take the original
2105         offset as a poly_int.
2106         * targhooks.h (default_legitimize_address_displacement): Update
2107         accordingly.
2108         * targhooks.c (default_legitimize_address_displacement): Likewise.
2109         * doc/tm.texi: Regenerate.
2110         * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2111         as an argument, moving assert of ad->disp == ad->disp_term to...
2112         (process_address_1): ...here.  Update calls to base_plus_disp_to_reg.
2113         Try calling targetm.legitimize_address_displacement before expanding
2114         the address rather than afterwards, and adjust for the new interface.
2115         * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2116         Match the new hook interface.  Handle SVE addresses.
2117         * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2118         new hook interface.
2119
2120 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2121
2122         * Makefile.in (OBJS): Add early-remat.o.
2123         * target.def (select_early_remat_modes): New hook.
2124         * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2125         * doc/tm.texi: Regenerate.
2126         * targhooks.h (default_select_early_remat_modes): Declare.
2127         * targhooks.c (default_select_early_remat_modes): New function.
2128         * timevar.def (TV_EARLY_REMAT): New timevar.
2129         * passes.def (pass_early_remat): New pass.
2130         * tree-pass.h (make_pass_early_remat): Declare.
2131         * early-remat.c: New file.
2132         * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2133         function.
2134         (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2135
2136 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2137             Alan Hayward  <alan.hayward@arm.com>
2138             David Sherwood  <david.sherwood@arm.com>
2139
2140         * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2141         vfm1 with a bound_epilog parameter.
2142         (vect_do_peeling): Update calls accordingly, and move the prologue
2143         call earlier in the function.  Treat the base bound_epilog as 0 for
2144         fully-masked loops and retain vf - 1 for other loops.  Add 1 to
2145         this base when peeling for gaps.
2146         * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2147         with fully-masked loops.
2148         (vect_estimate_min_profitable_iters): Handle the single peeled
2149         iteration in that case.
2150
2151 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2152             Alan Hayward  <alan.hayward@arm.com>
2153             David Sherwood  <david.sherwood@arm.com>
2154
2155         * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2156         single-element interleaving even if the size is not a power of 2.
2157         * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2158         accesses for single-element interleaving if the group size is
2159         not a power of 2.
2160
2161 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2162             Alan Hayward  <alan.hayward@arm.com>
2163             David Sherwood  <david.sherwood@arm.com>
2164
2165         * doc/md.texi (fold_extract_last_@var{m}): Document.
2166         * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2167         * optabs.def (fold_extract_last_optab): New optab.
2168         * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2169         * internal-fn.c (fold_extract_direct): New macro.
2170         (expand_fold_extract_optab_fn): Likewise.
2171         (direct_fold_extract_optab_supported_p): Likewise.
2172         * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2173         * tree-vect-loop.c (vect_model_reduction_cost): Handle
2174         EXTRACT_LAST_REDUCTION.
2175         (get_initial_def_for_reduction): Do not create an initial vector
2176         for EXTRACT_LAST_REDUCTION reductions.
2177         (vectorizable_reduction): Leave the scalar phi in place for
2178         EXTRACT_LAST_REDUCTIONs.  Try using EXTRACT_LAST_REDUCTION
2179         ahead of INTEGER_INDUC_COND_REDUCTION.  Do not check for an
2180         epilogue code for EXTRACT_LAST_REDUCTION and defer the
2181         transform phase to vectorizable_condition.
2182         * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2183         split out from...
2184         (vect_finish_stmt_generation): ...here.
2185         (vect_finish_replace_stmt): New function.
2186         (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2187         * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2188         pattern.
2189         * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2190
2191 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2192             Alan Hayward  <alan.hayward@arm.com>
2193             David Sherwood  <david.sherwood@arm.com>
2194
2195         * doc/md.texi (extract_last_@var{m}): Document.
2196         * optabs.def (extract_last_optab): New optab.
2197         * internal-fn.def (EXTRACT_LAST): New internal function.
2198         * internal-fn.c (cond_unary_direct): New macro.
2199         (expand_cond_unary_optab_fn): Likewise.
2200         (direct_cond_unary_optab_supported_p): Likewise.
2201         * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2202         loops using EXTRACT_LAST.
2203         * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2204         (extract_last_<mode>): ...this optab.
2205         (vec_extract<mode><Vel>): Update accordingly.
2206
2207 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2208             Alan Hayward  <alan.hayward@arm.com>
2209             David Sherwood  <david.sherwood@arm.com>
2210
2211         * target.def (empty_mask_is_expensive): New hook.
2212         * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2213         * doc/tm.texi: Regenerate.
2214         * targhooks.h (default_empty_mask_is_expensive): Declare.
2215         * targhooks.c (default_empty_mask_is_expensive): New function.
2216         * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2217         if the target says that empty masks are expensive.
2218         * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2219         New function.
2220         (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2221
2222 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2223             Alan Hayward  <alan.hayward@arm.com>
2224             David Sherwood  <david.sherwood@arm.com>
2225
2226         * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2227         (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2228         (vect_use_loop_mask_for_alignment_p): New function.
2229         (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2230         * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2231         niters_skip argument.  Make sure that the first niters_skip elements
2232         of the first iteration are inactive.
2233         (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2234         Update call to vect_set_loop_masks_directly.
2235         (get_misalign_in_elems): New function, split out from...
2236         (vect_gen_prolog_loop_niters): ...here.
2237         (vect_update_init_of_dr): Take a code argument that specifies whether
2238         the adjustment should be added or subtracted.
2239         (vect_update_init_of_drs): Likewise.
2240         (vect_prepare_for_masked_peels): New function.
2241         (vect_do_peeling): Skip prologue peeling if we're using a mask
2242         instead.  Update call to vect_update_inits_of_drs.
2243         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2244         mask_skip_niters.
2245         (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2246         alignment.  Do not include the number of peeled iterations in
2247         the minimum threshold in that case.
2248         (vectorizable_induction): Adjust the start value down by
2249         LOOP_VINFO_MASK_SKIP_NITERS iterations.
2250         (vect_transform_loop): Call vect_prepare_for_masked_peels.
2251         Take the number of skipped iterations into account when calculating
2252         the loop bounds.
2253         * tree-vect-stmts.c (vect_gen_while_not): New function.
2254
2255 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2256             Alan Hayward  <alan.hayward@arm.com>
2257             David Sherwood  <david.sherwood@arm.com>
2258
2259         * doc/sourcebuild.texi (vect_fully_masked): Document.
2260         * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2261         default value to 0.
2262         * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2263         split out from...
2264         (vect_analyze_loop_2): ...here. Don't check the vectorization
2265         factor against the number of loop iterations if the loop is
2266         fully-masked.
2267
2268 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2269             Alan Hayward  <alan.hayward@arm.com>
2270             David Sherwood  <david.sherwood@arm.com>
2271
2272         * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2273         (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2274         (dump_groups): Update accordingly.
2275         (iv_use::mem_type): New member variable.
2276         (address_p): New function.
2277         (record_use): Add a mem_type argument and initialize the new
2278         mem_type field.
2279         (record_group_use): Add a mem_type argument.  Use address_p.
2280         Remove obsolete null checks of base_object.  Update call to record_use.
2281         (find_interesting_uses_op): Update call to record_group_use.
2282         (find_interesting_uses_cond): Likewise.
2283         (find_interesting_uses_address): Likewise.
2284         (get_mem_type_for_internal_fn): New function.
2285         (find_address_like_use): Likewise.
2286         (find_interesting_uses_stmt): Try find_address_like_use before
2287         calling find_interesting_uses_op.
2288         (addr_offset_valid_p): Use the iv mem_type field as the type
2289         of the addressed memory.
2290         (add_autoinc_candidates): Likewise.
2291         (get_address_cost): Likewise.
2292         (split_small_address_groups_p): Use address_p.
2293         (split_address_groups): Likewise.
2294         (add_iv_candidate_for_use): Likewise.
2295         (autoinc_possible_for_pair): Likewise.
2296         (rewrite_groups): Likewise.
2297         (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2298         (determine_group_iv_cost): Update after split of USE_ADDRESS.
2299         (get_alias_ptr_type_for_ptr_address): New function.
2300         (rewrite_use_address): Rewrite address uses in calls that were
2301         identified by find_address_like_use.
2302
2303 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2304             Alan Hayward  <alan.hayward@arm.com>
2305             David Sherwood  <david.sherwood@arm.com>
2306
2307         * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2308         TARGET_MEM_REFs.
2309         * gimple-expr.h (is_gimple_addressable: Likewise.
2310         * gimple-expr.c (is_gimple_address): Likewise.
2311         * internal-fn.c (expand_call_mem_ref): New function.
2312         (expand_mask_load_optab_fn): Use it.
2313         (expand_mask_store_optab_fn): Likewise.
2314
2315 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2316             Alan Hayward  <alan.hayward@arm.com>
2317             David Sherwood  <david.sherwood@arm.com>
2318
2319         * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2320         (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2321         (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2322         (cond_umax@var{mode}): Document.
2323         * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2324         (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2325         (cond_umin_optab, cond_umax_optab): New optabs.
2326         * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2327         (COND_IOR, COND_XOR): New internal functions.
2328         * internal-fn.h (get_conditional_internal_fn): Declare.
2329         * internal-fn.c (cond_binary_direct): New macro.
2330         (expand_cond_binary_optab_fn): Likewise.
2331         (direct_cond_binary_optab_supported_p): Likewise.
2332         (get_conditional_internal_fn): New function.
2333         * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2334         Cope with reduction statements that are vectorized as calls rather
2335         than assignments.
2336         * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2337         * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2338         (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2339         (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2340         (UNSPEC_COND_EOR): New unspecs.
2341         (optab): Add mappings for them.
2342         (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2343         (sve_int_op, sve_fp_op): New int attributes.
2344
2345 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2346             Alan Hayward  <alan.hayward@arm.com>
2347             David Sherwood  <david.sherwood@arm.com>
2348
2349         * optabs.def (while_ult_optab): New optab.
2350         * doc/md.texi (while_ult@var{m}@var{n}): Document.
2351         * internal-fn.def (WHILE_ULT): New internal function.
2352         * internal-fn.h (direct_internal_fn_supported_p): New override
2353         that takes two types as argument.
2354         * internal-fn.c (while_direct): New macro.
2355         (expand_while_optab_fn): New function.
2356         (convert_optab_supported_p): Likewise.
2357         (direct_while_optab_supported_p): New macro.
2358         * wide-int.h (wi::udiv_ceil): New function.
2359         * tree-vectorizer.h (rgroup_masks): New structure.
2360         (vec_loop_masks): New typedef.
2361         (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2362         and fully_masked_p.
2363         (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2364         (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2365         (vect_max_vf): New function.
2366         (slpeel_make_loop_iterate_ntimes): Delete.
2367         (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2368         (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2369         (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2370         * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2371         internal-fn.h, stor-layout.h and optabs-query.h.
2372         (vect_set_loop_mask): New function.
2373         (add_preheader_seq): Likewise.
2374         (add_header_seq): Likewise.
2375         (interleave_supported_p): Likewise.
2376         (vect_maybe_permute_loop_masks): Likewise.
2377         (vect_set_loop_masks_directly): Likewise.
2378         (vect_set_loop_condition_masked): Likewise.
2379         (vect_set_loop_condition_unmasked): New function, split out from
2380         slpeel_make_loop_iterate_ntimes.
2381         (slpeel_make_loop_iterate_ntimes): Rename to..
2382         (vect_set_loop_condition): ...this.  Use vect_set_loop_condition_masked
2383         for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2384         (vect_do_peeling): Update call accordingly.
2385         (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2386         loops.
2387         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2388         mask_compare_type, can_fully_mask_p and fully_masked_p.
2389         (release_vec_loop_masks): New function.
2390         (_loop_vec_info): Use it to free the loop masks.
2391         (can_produce_all_loop_masks_p): New function.
2392         (vect_get_max_nscalars_per_iter): Likewise.
2393         (vect_verify_full_masking): Likewise.
2394         (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2395         retries, and free the mask rgroups before retrying.  Check loop-wide
2396         reasons for disallowing fully-masked loops.  Make the final decision
2397         about whether use a fully-masked loop or not.
2398         (vect_estimate_min_profitable_iters): Do not assume that peeling
2399         for the number of iterations will be needed for fully-masked loops.
2400         (vectorizable_reduction): Disable fully-masked loops.
2401         (vectorizable_live_operation): Likewise.
2402         (vect_halve_mask_nunits): New function.
2403         (vect_double_mask_nunits): Likewise.
2404         (vect_record_loop_mask): Likewise.
2405         (vect_get_loop_mask): Likewise.
2406         (vect_transform_loop): Handle the case in which the final loop
2407         iteration might handle a partial vector.  Call vect_set_loop_condition
2408         instead of slpeel_make_loop_iterate_ntimes.
2409         * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2410         (check_load_store_masking): New function.
2411         (prepare_load_store_mask): Likewise.
2412         (vectorizable_store): Handle fully-masked loops.
2413         (vectorizable_load): Likewise.
2414         (supportable_widening_operation): Use vect_halve_mask_nunits for
2415         booleans.
2416         (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2417         (vect_gen_while): New function.
2418         * config/aarch64/aarch64.md (umax<mode>3): New expander.
2419         (aarch64_uqdec<mode>): New insn.
2420
2421 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2422             Alan Hayward  <alan.hayward@arm.com>
2423             David Sherwood  <david.sherwood@arm.com>
2424
2425         * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2426         (reduc_xor_scal_optab): New optabs.
2427         * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2428         (reduc_xor_scal_@var{m}): Document.
2429         * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2430         * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2431         internal functions.
2432         * fold-const-call.c (fold_const_call): Handle them.
2433         * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2434         internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2435         * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2436         (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2437         * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2438         (UNSPEC_XORV): New unspecs.
2439         (optab): Add entries for them.
2440         (BITWISEV): New int iterator.
2441         (bit_reduc_op): New int attributes.
2442
2443 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2444             Alan Hayward  <alan.hayward@arm.com>
2445             David Sherwood  <david.sherwood@arm.com>
2446
2447         * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2448         * internal-fn.def (VEC_SHL_INSERT): New internal function.
2449         * optabs.def (vec_shl_insert_optab): New optab.
2450         * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2451         (duplicate_and_interleave): Likewise.
2452         * tree-vect-loop.c: Include internal-fn.h.
2453         (neutral_op_for_slp_reduction): New function, split out from
2454         get_initial_defs_for_reduction.
2455         (get_initial_def_for_reduction): Handle option 2 for variable-length
2456         vectors by loading the neutral value into a vector and then shifting
2457         the initial value into element 0.
2458         (get_initial_defs_for_reduction): Replace the code argument with
2459         the neutral value calculated by neutral_op_for_slp_reduction.
2460         Use gimple_build_vector for constant-length vectors.
2461         Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2462         but the first group_size elements have a neutral value.
2463         Use duplicate_and_interleave otherwise.
2464         (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2465         Update call to get_initial_defs_for_reduction.  Handle SLP
2466         reductions for variable-length vectors by creating one vector
2467         result for each scalar result, with the elements associated
2468         with other scalar results stubbed out with the neutral value.
2469         (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2470         Require IFN_VEC_SHL_INSERT for double reductions on
2471         variable-length vectors, or SLP reductions that have
2472         a neutral value.  Require can_duplicate_and_interleave_p
2473         support for variable-length unchained SLP reductions if there
2474         is no neutral value, such as for MIN/MAX reductions.  Also require
2475         the number of vector elements to be a multiple of the number of
2476         SLP statements when doing variable-length unchained SLP reductions.
2477         Update call to vect_create_epilog_for_reduction.
2478         * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2479         and remove initial values.
2480         (duplicate_and_interleave): Make public.
2481         * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2482         * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2483
2484 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2485             Alan Hayward  <alan.hayward@arm.com>
2486             David Sherwood  <david.sherwood@arm.com>
2487
2488         * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2489         (can_duplicate_and_interleave_p): New function.
2490         (vect_get_and_check_slp_defs): Take the vector of statements
2491         rather than just the current one.  Remove excess parentheses.
2492         Restriction rejectinon of vect_constant_def and vect_external_def
2493         for variable-length vectors to boolean types, or types for which
2494         can_duplicate_and_interleave_p is false.
2495         (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2496         (duplicate_and_interleave): New function.
2497         (vect_get_constant_vectors): Use gimple_build_vector for
2498         constant-length vectors and suitable variable-length constant
2499         vectors.  Use duplicate_and_interleave for other variable-length
2500         vectors.  Don't defer the update when inserting new statements.
2501
2502 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2503             Alan Hayward  <alan.hayward@arm.com>
2504             David Sherwood  <david.sherwood@arm.com>
2505
2506         * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2507         min_profitable_iters doesn't go negative.
2508
2509 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2510             Alan Hayward  <alan.hayward@arm.com>
2511             David Sherwood  <david.sherwood@arm.com>
2512
2513         * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2514         (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2515         * optabs.def (vec_mask_load_lanes_optab): New optab.
2516         (vec_mask_store_lanes_optab): Likewise.
2517         * internal-fn.def (MASK_LOAD_LANES): New internal function.
2518         (MASK_STORE_LANES): Likewise.
2519         * internal-fn.c (mask_load_lanes_direct): New macro.
2520         (mask_store_lanes_direct): Likewise.
2521         (expand_mask_load_optab_fn): Handle masked operations.
2522         (expand_mask_load_lanes_optab_fn): New macro.
2523         (expand_mask_store_optab_fn): Handle masked operations.
2524         (expand_mask_store_lanes_optab_fn): New macro.
2525         (direct_mask_load_lanes_optab_supported_p): Likewise.
2526         (direct_mask_store_lanes_optab_supported_p): Likewise.
2527         * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2528         parameter.
2529         (vect_load_lanes_supported): Likewise.
2530         * tree-vect-data-refs.c (strip_conversion): New function.
2531         (can_group_stmts_p): Likewise.
2532         (vect_analyze_data_ref_accesses): Use it instead of checking
2533         for a pair of assignments.
2534         (vect_store_lanes_supported): Take a masked_p parameter.
2535         (vect_load_lanes_supported): Likewise.
2536         * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2537         vect_store_lanes_supported and vect_load_lanes_supported.
2538         * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2539         * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2540         parameter.  Don't allow gaps for masked accesses.
2541         Use vect_get_store_rhs.  Update calls to vect_store_lanes_supported
2542         and vect_load_lanes_supported.
2543         (get_load_store_type): Take a masked_p parameter and update
2544         call to get_group_load_store_type.
2545         (vectorizable_store): Update call to get_load_store_type.
2546         Handle IFN_MASK_STORE_LANES.
2547         (vectorizable_load): Update call to get_load_store_type.
2548         Handle IFN_MASK_LOAD_LANES.
2549
2550 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2551             Alan Hayward  <alan.hayward@arm.com>
2552             David Sherwood  <david.sherwood@arm.com>
2553
2554         * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2555         modes for SVE.
2556         * config/aarch64/aarch64-protos.h
2557         (aarch64_sve_struct_memory_operand_p): Declare.
2558         * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2559         (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2560         (VPRED, vpred): Handle SVE structure modes.
2561         * config/aarch64/constraints.md (Utx): New constraint.
2562         * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2563         (aarch64_sve_struct_nonimmediate_operand): New predicates.
2564         * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2565         * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2566         (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2567         structure modes.  Split into pieces after RA.
2568         (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2569         (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2570         New patterns.
2571         * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2572         SVE structure modes.
2573         (aarch64_classify_address): Likewise.
2574         (sizetochar): Move earlier in file.
2575         (aarch64_print_operand): Handle SVE register lists.
2576         (aarch64_array_mode): New function.
2577         (aarch64_sve_struct_memory_operand_p): Likewise.
2578         (TARGET_ARRAY_MODE): Redefine.
2579
2580 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2581             Alan Hayward  <alan.hayward@arm.com>
2582             David Sherwood  <david.sherwood@arm.com>
2583
2584         * target.def (array_mode): New target hook.
2585         * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2586         * doc/tm.texi: Regenerate.
2587         * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2588         * hooks.c (hook_optmode_mode_uhwi_none): New function.
2589         * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2590         targetm.array_mode.
2591         * stor-layout.c (mode_for_array): Likewise.  Support polynomial
2592         type sizes.
2593
2594 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2595             Alan Hayward  <alan.hayward@arm.com>
2596             David Sherwood  <david.sherwood@arm.com>
2597
2598         * fold-const.c (fold_binary_loc): Check the argument types
2599         rather than the result type when testing for a vector operation.
2600
2601 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2602
2603         * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2604         * doc/tm.texi: Regenerate.
2605
2606 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2607             Alan Hayward  <alan.hayward@arm.com>
2608             David Sherwood  <david.sherwood@arm.com>
2609
2610         * doc/invoke.texi (-msve-vector-bits=): Document new option.
2611         (sve): Document new AArch64 extension.
2612         * doc/md.texi (w): Extend the description of the AArch64
2613         constraint to include SVE vectors.
2614         (Upl, Upa): Document new AArch64 predicate constraints.
2615         * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2616         enum.
2617         * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2618         (msve-vector-bits=): New option.
2619         * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2620         SVE when these are disabled.
2621         (sve): New extension.
2622         * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2623         modes.  Adjust their number of units based on aarch64_sve_vg.
2624         (MAX_BITSIZE_MODE_ANY_MODE): Define.
2625         * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2626         aarch64_addr_query_type.
2627         (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2628         (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2629         (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2630         (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2631         (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2632         (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2633         (aarch64_simd_imm_zero_p): Delete.
2634         (aarch64_check_zero_based_sve_index_immediate): Declare.
2635         (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2636         (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2637         (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2638         (aarch64_sve_float_mul_immediate_p): Likewise.
2639         (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2640         rather than an rtx.
2641         (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2642         (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2643         (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2644         (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2645         (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2646         (aarch64_regmode_natural_size): Likewise.
2647         * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2648         (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2649         left one place.
2650         (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2651         (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2652         for VG and the SVE predicate registers.
2653         (V_ALIASES): Add a "z"-prefixed alias.
2654         (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2655         (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2656         (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2657         (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2658         (REG_CLASS_NAMES): Add entries for them.
2659         (REG_CLASS_CONTENTS): Likewise.  Update ALL_REGS to include VG
2660         and the predicate registers.
2661         (aarch64_sve_vg): Declare.
2662         (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2663         (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2664         (REGMODE_NATURAL_SIZE): Define.
2665         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2666         SVE macros.
2667         * config/aarch64/aarch64.c: Include cfgrtl.h.
2668         (simd_immediate_info): Add a constructor for series vectors,
2669         and an associated step field.
2670         (aarch64_sve_vg): New variable.
2671         (aarch64_dbx_register_number): Handle VG and the predicate registers.
2672         (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2673         (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2674         (VEC_ANY_DATA, VEC_STRUCT): New constants.
2675         (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2676         (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2677         (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2678         (aarch64_get_mask_mode): New functions.
2679         (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2680         and FP_LO_REGS.  Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2681         (aarch64_hard_regno_mode_ok): Handle VG.  Also handle the SVE
2682         predicate modes and predicate registers.  Explicitly restrict
2683         GPRs to modes of 16 bytes or smaller.  Only allow FP registers
2684         to store a vector mode if it is recognized by
2685         aarch64_classify_vector_mode.
2686         (aarch64_regmode_natural_size): New function.
2687         (aarch64_hard_regno_caller_save_mode): Return the original mode
2688         for predicates.
2689         (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2690         (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2691         (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2692         (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2693         functions.
2694         (aarch64_add_offset): Add a temp2 parameter.  Assert that temp1
2695         does not overlap dest if the function is frame-related.  Handle
2696         SVE constants.
2697         (aarch64_split_add_offset): New function.
2698         (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2699         them aarch64_add_offset.
2700         (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2701         and update call to aarch64_sub_sp.
2702         (aarch64_add_cfa_expression): New function.
2703         (aarch64_expand_prologue): Pass extra temporary registers to the
2704         functions above.  Handle the case in which we need to emit new
2705         DW_CFA_expressions for registers that were originally saved
2706         relative to the stack pointer, but now have to be expressed
2707         relative to the frame pointer.
2708         (aarch64_output_mi_thunk): Pass extra temporary registers to the
2709         functions above.
2710         (aarch64_expand_epilogue): Likewise.  Prevent inheritance of
2711         IP0 and IP1 values for SVE frames.
2712         (aarch64_expand_vec_series): New function.
2713         (aarch64_expand_sve_widened_duplicate): Likewise.
2714         (aarch64_expand_sve_const_vector): Likewise.
2715         (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2716         Handle SVE constants.  Use emit_move_insn to move a force_const_mem
2717         into the register, rather than emitting a SET directly.
2718         (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2719         (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2720         (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2721         (offset_9bit_signed_scaled_p): New functions.
2722         (aarch64_replicate_bitmask_imm): New function.
2723         (aarch64_bitmask_imm): Use it.
2724         (aarch64_cannot_force_const_mem): Reject expressions involving
2725         a CONST_POLY_INT.  Update call to aarch64_classify_symbol.
2726         (aarch64_classify_index): Handle SVE indices, by requiring
2727         a plain register index with a scale that matches the element size.
2728         (aarch64_classify_address): Handle SVE addresses.  Assert that
2729         the mode of the address is VOIDmode or an integer mode.
2730         Update call to aarch64_classify_symbol.
2731         (aarch64_classify_symbolic_expression): Update call to
2732         aarch64_classify_symbol.
2733         (aarch64_const_vec_all_in_range_p): New function.
2734         (aarch64_print_vector_float_operand): Likewise.
2735         (aarch64_print_operand): Handle 'N' and 'C'.  Use "zN" rather than
2736         "vN" for FP registers with SVE modes.  Handle (const ...) vectors
2737         and the FP immediates 1.0 and 0.5.
2738         (aarch64_print_address_internal): Handle SVE addresses.
2739         (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2740         (aarch64_regno_regclass): Handle predicate registers.
2741         (aarch64_secondary_reload): Handle big-endian reloads of SVE
2742         data modes.
2743         (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2744         (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2745         (aarch64_convert_sve_vector_bits): New function.
2746         (aarch64_override_options): Use it to handle -msve-vector-bits=.
2747         (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2748         rather than an rtx.
2749         (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2750         Handle SVE vector and predicate modes.  Accept VL-based constants
2751         that need only one temporary register, and VL offsets that require
2752         no temporary registers.
2753         (aarch64_conditional_register_usage): Mark the predicate registers
2754         as fixed if SVE isn't available.
2755         (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2756         Return true for SVE vector and predicate modes.
2757         (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2758         rather than an unsigned int.  Handle SVE modes.
2759         (aarch64_preferred_simd_mode): Update call accordingly.  Handle
2760         SVE modes.
2761         (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2762         if SVE is enabled.
2763         (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2764         (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2765         (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2766         (aarch64_sve_float_mul_immediate_p): New functions.
2767         (aarch64_sve_valid_immediate): New function.
2768         (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2769         Explicitly reject structure modes.  Check for INDEX constants.
2770         Handle PTRUE and PFALSE constants.
2771         (aarch64_check_zero_based_sve_index_immediate): New function.
2772         (aarch64_simd_imm_zero_p): Delete.
2773         (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2774         vector modes.  Accept constants in the range of CNT[BHWD].
2775         (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2776         ask for an Advanced SIMD mode.
2777         (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2778         (aarch64_simd_vector_alignment): Handle SVE predicates.
2779         (aarch64_vectorize_preferred_vector_alignment): New function.
2780         (aarch64_simd_vector_alignment_reachable): Use it instead of
2781         the vector size.
2782         (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2783         (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2784         functions.
2785         (MAX_VECT_LEN): Delete.
2786         (expand_vec_perm_d): Add a vec_flags field.
2787         (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2788         (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2789         (aarch64_evpc_ext): Don't apply a big-endian lane correction
2790         for SVE modes.
2791         (aarch64_evpc_rev): Rename to...
2792         (aarch64_evpc_rev_local): ...this.  Use a predicated operation for SVE.
2793         (aarch64_evpc_rev_global): New function.
2794         (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2795         (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2796         MAX_VECT_LEN.
2797         (aarch64_evpc_sve_tbl): New function.
2798         (aarch64_expand_vec_perm_const_1): Update after rename of
2799         aarch64_evpc_rev.  Handle SVE permutes too, trying
2800         aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2801         than aarch64_evpc_tbl.
2802         (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2803         (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2804         (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2805         (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2806         (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2807         (aarch64_expand_sve_vcond): New functions.
2808         (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2809         of aarch64_vector_mode_p.
2810         (aarch64_dwarf_poly_indeterminate_value): New function.
2811         (aarch64_compute_pressure_classes): Likewise.
2812         (aarch64_can_change_mode_class): Likewise.
2813         (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2814         (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2815         (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2816         (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2817         (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2818         (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2819         * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2820         (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2821         constraints.
2822         (Dn, Dl, Dr): Accept const as well as const_vector.
2823         (Dz): Likewise.  Compare against CONST0_RTX.
2824         * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2825         of "vector" where appropriate.
2826         (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2827         (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2828         (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2829         (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2830         (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2831         (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2832         (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2833         (v_int_equiv): Extend to SVE modes.
2834         (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2835         mode attributes.
2836         (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2837         (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2838         (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2839         (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2840         (SVE_COND_FP_CMP): New int iterators.
2841         (perm_hilo): Handle the new unpack unspecs.
2842         (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2843         attributes.
2844         * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2845         (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2846         (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2847         (aarch64_equality_operator, aarch64_constant_vector_operand)
2848         (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2849         (aarch64_sve_nonimmediate_operand): Likewise.
2850         (aarch64_sve_general_operand): Likewise.
2851         (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2852         (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2853         (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2854         (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2855         (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2856         (aarch64_sve_float_arith_immediate): Likewise.
2857         (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2858         (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2859         (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2860         (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2861         (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2862         (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2863         (aarch64_sve_float_arith_operand): Likewise.
2864         (aarch64_sve_float_arith_with_sub_operand): Likewise.
2865         (aarch64_sve_float_mul_operand): Likewise.
2866         (aarch64_sve_vec_perm_operand): Likewise.
2867         (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2868         (aarch64_mov_operand): Accept const_poly_int and const_vector.
2869         (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2870         as well as const_vector.
2871         (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2872         in file.  Use CONST0_RTX and CONSTM1_RTX.
2873         (aarch64_simd_or_scalar_imm_zero): Likewise.  Add match_codes.
2874         (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2875         Use aarch64_simd_imm_zero.
2876         * config/aarch64/aarch64-sve.md: New file.
2877         * config/aarch64/aarch64.md: Include it.
2878         (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2879         (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2880         (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2881         (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2882         (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2883         (sve): New attribute.
2884         (enabled): Disable instructions with the sve attribute unless
2885         TARGET_SVE.
2886         (movqi, movhi): Pass CONST_POLY_INT operaneds through
2887         aarch64_expand_mov_immediate.
2888         (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2889         CNT[BHSD] immediates.
2890         (movti): Split CONST_POLY_INT moves into two halves.
2891         (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2892         Split additions that need a temporary here if the destination
2893         is the stack pointer.
2894         (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2895         (*add<mode>3_poly_1): New instruction.
2896         (set_clobber_cc): New expander.
2897
2898 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2899
2900         * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2901         parameter and use it instead of GET_MODE_SIZE (innermode).  Use
2902         inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2903         Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2904         GET_MODE_NUNITS (innermode).  Also add a first_elem parameter.
2905         Change innermode from fixed_mode_size to machine_mode.
2906         (simplify_subreg): Update call accordingly.  Handle a constant-sized
2907         subreg of a variable-length CONST_VECTOR.
2908
2909 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
2910             Alan Hayward  <alan.hayward@arm.com>
2911             David Sherwood  <david.sherwood@arm.com>
2912
2913         * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2914         (add_offset_to_base): New function, split out from...
2915         (create_mem_ref): ...here.  When handling a scale other than 1,
2916         check first whether the address is valid without the offset.
2917         Add it into the base if so, leaving the index and scale as-is.
2918
2919 2018-01-12  Jakub Jelinek  <jakub@redhat.com>
2920
2921         PR c++/83778
2922         * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2923         fold_for_warn before checking if arg2 is INTEGER_CST.
2924
2925 2018-01-12  Segher Boessenkool  <segher@kernel.crashing.org>
2926
2927         * config/rs6000/predicates.md (load_multiple_operation): Delete.
2928         (store_multiple_operation): Delete.
2929         * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2930         * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2931         * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2932         guarded by TARGET_STRING.
2933         (rs6000_output_load_multiple): Delete.
2934         * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2935         OPTION_MASK_STRING / TARGET_STRING handling.
2936         (print_operand) <'N', 'O'>: Add comment that these are unused now.
2937         (const rs6000_opt_masks) <"string">: Change mask to 0.
2938         * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2939         (MASK_STRING): Delete.
2940         * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2941         parts.  Simplify.
2942         (load_multiple): Delete.
2943         (*ldmsi8): Delete.
2944         (*ldmsi7): Delete.
2945         (*ldmsi6): Delete.
2946         (*ldmsi5): Delete.
2947         (*ldmsi4): Delete.
2948         (*ldmsi3): Delete.
2949         (store_multiple): Delete.
2950         (*stmsi8): Delete.
2951         (*stmsi7): Delete.
2952         (*stmsi6): Delete.
2953         (*stmsi5): Delete.
2954         (*stmsi4): Delete.
2955         (*stmsi3): Delete.
2956         (movmemsi_8reg): Delete.
2957         (corresponding unnamed define_insn): Delete.
2958         (movmemsi_6reg): Delete.
2959         (corresponding unnamed define_insn): Delete.
2960         (movmemsi_4reg): Delete.
2961         (corresponding unnamed define_insn): Delete.
2962         (movmemsi_2reg): Delete.
2963         (corresponding unnamed define_insn): Delete.
2964         (movmemsi_1reg): Delete.
2965         (corresponding unnamed define_insn): Delete.
2966         * config/rs6000/rs6000.opt (mno-string): New.
2967         (mstring): Replace by deprecation warning stub.
2968         * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2969
2970 2018-01-12  Jakub Jelinek  <jakub@redhat.com>
2971
2972         * regrename.c (regrename_do_replace): If replacing the same
2973         reg multiple times, try to reuse last created gen_raw_REG.
2974
2975         PR debug/81155
2976         * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2977         main to workaround a bug in GDB.
2978
2979 2018-01-12  Tom de Vries  <tom@codesourcery.com>
2980
2981         PR target/83737
2982         * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2983
2984 2018-01-12  Vladimir Makarov  <vmakarov@redhat.com>
2985
2986         PR rtl-optimization/80481
2987         * ira-color.c (get_cap_member): New function.
2988         (allocnos_conflict_by_live_ranges_p): Use it.
2989         (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2990         (setup_slot_coalesced_allocno_live_ranges): Ditto.
2991
2992 2018-01-12  Uros Bizjak  <ubizjak@gmail.com>
2993
2994         PR target/83628
2995         * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2996         (*saddl_se_1): Ditto.
2997         (*ssubsi_1): Ditto.
2998         (*ssubl_se_1): Ditto.
2999
3000 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
3001
3002         * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3003         rather than wi::to_widest for DR_INITs.
3004         * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3005         wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3006         (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3007         INTEGER_CSTs.
3008         (vect_analyze_group_access_1): Note that here.
3009
3010 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
3011
3012         * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3013         polynomial type sizes.
3014
3015 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
3016
3017         * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3018         poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3019         (gimple_add_tmp_var): Likewise.
3020
3021 2018-01-12  Martin Liska  <mliska@suse.cz>
3022
3023         * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3024         (gimple_alloc_sizes): Likewise.
3025         (dump_gimple_statistics): Use PRIu64 in printf format.
3026         * gimple.h: Change uint64_t to int.
3027
3028 2018-01-12  Martin Liska  <mliska@suse.cz>
3029
3030         * tree-core.h: Use uint64_t instead of int.
3031         * tree.c (tree_node_counts): Likewise.
3032         (tree_node_sizes): Likewise.
3033         (dump_tree_statistics): Use PRIu64 in printf format.
3034
3035 2018-01-12  Martin Liska  <mliska@suse.cz>
3036
3037         * Makefile.in: As qsort_chk is implemented in vec.c, add
3038         vec.o to linkage of gencfn-macros.
3039         * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3040         passing the info to record_node_allocation_statistics.
3041         (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3042         and pass the info.
3043         * ggc-common.c (struct ggc_usage): Add operator== and use
3044         it in operator< and compare function.
3045         * mem-stats.h (struct mem_usage): Likewise.
3046         * vec.c (struct vec_usage): Remove operator< and compare
3047         function. Can be simply inherited.
3048
3049 2018-01-12  Martin Jambor  <mjambor@suse.cz>
3050
3051         PR target/81616
3052         * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3053         * tree-ssa-math-opts.c: Include domwalk.h.
3054         (convert_mult_to_fma_1): New function.
3055         (fma_transformation_info): New type.
3056         (fma_deferring_state): Likewise.
3057         (cancel_fma_deferring): New function.
3058         (result_of_phi): Likewise.
3059         (last_fma_candidate_feeds_initial_phi): Likewise.
3060         (convert_mult_to_fma): Added deferring logic, split actual
3061         transformation to convert_mult_to_fma_1.
3062         (math_opts_dom_walker): New type.
3063         (math_opts_dom_walker::after_dom_children): New method, body moved
3064         here from pass_optimize_widening_mul::execute, added deferring logic
3065         bits.
3066         (pass_optimize_widening_mul::execute): Moved most of code to
3067         math_opts_dom_walker::after_dom_children.
3068         * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3069         * config/i386/i386.c (ix86_option_override_internal): Added
3070         maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3071
3072 2018-01-12  Richard Biener  <rguenther@suse.de>
3073
3074         PR debug/83157
3075         * dwarf2out.c (gen_variable_die): Do not reset old_die for
3076         inline instance vars.
3077
3078 2018-01-12  Oleg Endo  <olegendo@gcc.gnu.org>
3079
3080         PR target/81819
3081         * config/rx/rx.c (rx_is_restricted_memory_address):
3082         Handle SUBREG case.
3083
3084 2018-01-12  Richard Biener  <rguenther@suse.de>
3085
3086         PR tree-optimization/80846
3087         * target.def (split_reduction): New target hook.
3088         * targhooks.c (default_split_reduction): New function.
3089         * targhooks.h (default_split_reduction): Declare.
3090         * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3091         target requests first reduce vectors by combining low and high
3092         parts.
3093         * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3094         (get_vectype_for_scalar_type_and_size): Export.
3095         * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3096         * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3097         * doc/tm.texi: Regenerate.
3098         * config/i386/i386.c (ix86_split_reduction): Implement
3099         TARGET_VECTORIZE_SPLIT_REDUCTION.
3100
3101 2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
3102
3103         PR target/83368
3104         * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3105         in PIC mode except for TARGET_VXWORKS_RTP.
3106         * config/sparc/sparc.c: Include cfgrtl.h.
3107         (TARGET_INIT_PIC_REG): Define.
3108         (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3109         (sparc_pic_register_p): New predicate.
3110         (sparc_legitimate_address_p): Use it.
3111         (sparc_legitimize_pic_address): Likewise.
3112         (sparc_delegitimize_address): Likewise.
3113         (sparc_mode_dependent_address_p): Likewise.
3114         (gen_load_pcrel_sym): Remove 4th parameter.
3115         (load_got_register): Adjust call to above.  Remove obsolete stuff.
3116         (sparc_expand_prologue): Do not call load_got_register here.
3117         (sparc_flat_expand_prologue): Likewise.
3118         (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3119         (sparc_use_pseudo_pic_reg): New function.
3120         (sparc_init_pic_reg): Likewise.
3121         * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3122         (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3123
3124 2018-01-12  Christophe Lyon  <christophe.lyon@linaro.org>
3125
3126         * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3127         Add item for branch_cost.
3128
3129 2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
3130
3131         PR rtl-optimization/83565
3132         * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3133         not extend the result to a larger mode for rotate operations.
3134         (num_sign_bit_copies1): Likewise.
3135
3136 2018-01-12  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
3137
3138         PR target/40411
3139         * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3140         -symbolic.
3141         Use values-Xc.o for -pedantic.
3142         Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3143
3144 2018-01-12  Martin Liska  <mliska@suse.cz>
3145
3146         PR ipa/83054
3147         * ipa-devirt.c (final_warning_record::grow_type_warnings):
3148         New function.
3149         (possible_polymorphic_call_targets): Use it.
3150         (ipa_devirt): Likewise.
3151
3152 2018-01-12  Martin Liska  <mliska@suse.cz>
3153
3154         * profile-count.h (enum profile_quality): Use 0 as invalid
3155         enum value of profile_quality.
3156
3157 2018-01-12  Chung-Ju Wu  <jasonwucj@gmail.com>
3158
3159         * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3160         -mext-string options.
3161
3162 2018-01-12  Richard Biener  <rguenther@suse.de>
3163
3164         * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3165         DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3166         * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3167         Likewise.
3168         * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3169
3170 2018-01-11  Michael Meissner  <meissner@linux.vnet.ibm.com>
3171
3172         * configure.ac (--with-long-double-format): Add support for the
3173         configuration option to change the default long double format on
3174         PowerPC systems.
3175         * config.gcc (powerpc*-linux*-*): Likewise.
3176         * configure: Regenerate.
3177         * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3178         double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3179         used without modification.
3180
3181 2018-01-11  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
3182
3183         * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3184         (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3185         * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3186         MISC_BUILTIN_SPEC_BARRIER.
3187         (rs6000_init_builtins): Likewise.
3188         * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3189         enum value.
3190         (speculation_barrier): New define_insn.
3191         * doc/extend.texi: Document __builtin_speculation_barrier.
3192
3193 2018-01-11  Jakub Jelinek  <jakub@redhat.com>
3194
3195         PR target/83203
3196         * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3197         is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3198         * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3199         iterators.
3200         (ssescalarmodesuffix): Add 512-bit vectors.  Use "d" or "q" for
3201         integral modes instead of "ss" and "sd".
3202         (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3203         vectors with 32-bit and 64-bit elements.
3204         (vecdupssescalarmodesuffix): New mode attribute.
3205         (vec_dup<mode>): Use it.
3206
3207 2018-01-11  H.J. Lu  <hongjiu.lu@intel.com>
3208
3209         PR target/83330
3210         * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3211         frame if argument is passed on stack.
3212
3213 2018-01-11  Jakub Jelinek  <jakub@redhat.com>
3214
3215         PR target/82682
3216         * ree.c (combine_reaching_defs): Optimize also
3217         reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3218         reg2=any_extend(exp); reg1=reg2;, formatting fix.
3219
3220 2018-01-11  Jan Hubicka  <hubicka@ucw.cz>
3221
3222         PR middle-end/83189
3223         * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3224
3225 2018-01-11  Jan Hubicka  <hubicka@ucw.cz>
3226
3227         PR middle-end/83718
3228         * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3229         after they are computed.
3230
3231 2018-01-11  Bin Cheng  <bin.cheng@arm.com>
3232
3233         PR tree-optimization/83695
3234         * gimple-loop-linterchange.cc
3235         (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3236         reset cached scev information after interchange.
3237         (pass_linterchange::execute): Remove call to scev_reset_htab.
3238
3239 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
3240
3241         * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3242         vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3243         vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3244         vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3245         vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3246         vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3247         * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3248         vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3249         vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3250         vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3251         vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3252         * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3253         (V_lane_reg): Likewise.
3254         * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3255         New define_expand.
3256         (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3257         (vfmal_lane_low<mode>_intrinsic,
3258         vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3259         vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3260         vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3261         vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3262         vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3263         vfmsl_lane_high<mode>_intrinsic): New define_insns.
3264
3265 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
3266
3267         * config/arm/arm-cpus.in (fp16fml): New feature.
3268         (ALL_SIMD): Add fp16fml.
3269         (armv8.2-a): Add fp16fml as an option.
3270         (armv8.3-a): Likewise.
3271         (armv8.4-a): Add fp16fml as part of fp16.
3272         * config/arm/arm.h (TARGET_FP16FML): Define.
3273         * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3274         when appropriate.
3275         * config/arm/arm-modes.def (V2HF): Define.
3276         * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3277         vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3278         vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3279         * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3280         vfmsl_low, vfmsl_high): New set of builtins.
3281         * config/arm/iterators.md (PLUSMINUS): New code iterator.
3282         (vfml_op): New code attribute.
3283         (VFMLHALVES): New int iterator.
3284         (VFML, VFMLSEL): New mode attributes.
3285         (V_reg): Define mapping for V2HF.
3286         (V_hi, V_lo): New mode attributes.
3287         (VF_constraint): Likewise.
3288         (vfml_half, vfml_half_selector): New int attributes.
3289         * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3290         define_expand.
3291         (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3292         vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3293         New define_insn.
3294         * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3295         * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3296         * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3297         * doc/invoke.texi (ARM Options): Document fp16fml.  Update armv8.4-a
3298         documentation.
3299         * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3300         Document new effective target and option set.
3301
3302 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
3303
3304         * config/arm/arm-cpus.in (armv8_4): New feature.
3305         (ARMv8_4a): New fgroup.
3306         (armv8.4-a): New arch.
3307         * config/arm/arm-tables.opt: Regenerate.
3308         * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3309         * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3310         * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3311         Add matching rules for -march=armv8.4-a and extensions.
3312         * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3313
3314 2018-01-11  Oleg Endo  <olegendo@gcc.gnu.org>
3315
3316         PR target/81821
3317         * config/rx/rx.md (BW): New mode attribute.
3318         (sync_lock_test_and_setsi): Add mode suffix to insn output.
3319
3320 2018-01-11  Richard Biener  <rguenther@suse.de>
3321
3322         PR tree-optimization/83435
3323         * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3324         * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3325         * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3326
3327 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
3328             Alan Hayward  <alan.hayward@arm.com>
3329             David Sherwood  <david.sherwood@arm.com>
3330
3331         * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3332         field.
3333         (aarch64_classify_address): Initialize it.  Track polynomial offsets.
3334         (aarch64_print_address_internal): Use it to check for a zero offset.
3335
3336 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
3337             Alan Hayward  <alan.hayward@arm.com>
3338             David Sherwood  <david.sherwood@arm.com>
3339
3340         * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3341         * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3342         Return a poly_int64 rather than a HOST_WIDE_INT.
3343         (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3344         rather than a HOST_WIDE_INT.
3345         * config/aarch64/aarch64.h (aarch64_frame): Protect with
3346         HAVE_POLY_INT_H rather than HOST_WIDE_INT.  Change locals_offset,
3347         hard_fp_offset, frame_size, initial_adjust, callee_offset and
3348         final_offset from HOST_WIDE_INT to poly_int64.
3349         * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3350         to_constant when getting the number of units in an Advanced SIMD
3351         mode.
3352         (aarch64_builtin_vectorized_function): Check for a constant number
3353         of units.
3354         * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3355         GET_MODE_SIZE.
3356         (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3357         attribute instead of GET_MODE_NUNITS.
3358         * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3359         (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3360         GET_MODE_SIZE for fixed-size registers.
3361         (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3362         (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3363         (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3364         (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3365         (aarch64_print_operand, aarch64_print_address_internal)
3366         (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3367         (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3368         (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3369         Handle polynomial GET_MODE_SIZE.
3370         (aarch64_hard_regno_caller_save_mode): Likewise.  Return modes
3371         wider than SImode without modification.
3372         (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3373         (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3374         (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3375         passing and returning SVE modes.
3376         (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3377         rather than GEN_INT.
3378         (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3379         rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3380         (aarch64_allocate_and_probe_stack_space): Likewise.
3381         (aarch64_layout_frame): Cope with polynomial offsets.
3382         (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3383         start_offset as a poly_int64 rather than a HOST_WIDE_INT.  Track
3384         polynomial offsets.
3385         (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3386         (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3387         poly_int64 rather than a HOST_WIDE_INT.
3388         (aarch64_get_separate_components, aarch64_process_components)
3389         (aarch64_expand_prologue, aarch64_expand_epilogue)
3390         (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3391         (aarch64_anchor_offset): New function, split out from...
3392         (aarch64_legitimize_address): ...here.
3393         (aarch64_builtin_vectorization_cost): Handle polynomial
3394         TYPE_VECTOR_SUBPARTS.
3395         (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3396         GET_MODE_NUNITS.
3397         (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3398         number of elements from the PARALLEL rather than the mode.
3399         (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3400         rather than GET_MODE_BITSIZE.
3401         (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3402         (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3403         (aarch64_expand_vec_perm_const_1): Handle polynomial
3404         d->perm.length () and d->perm elements.
3405         (aarch64_evpc_tbl): Likewise.  Use nelt rather than GET_MODE_NUNITS.
3406         Apply to_constant to d->perm elements.
3407         (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3408         polynomial CONST_VECTOR_NUNITS.
3409         (aarch64_move_pointer): Take amount as a poly_int64 rather
3410         than an int.
3411         (aarch64_progress_pointer): Avoid temporary variable.
3412         * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3413         the mode attribute instead of GET_MODE.
3414
3415 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
3416             Alan Hayward  <alan.hayward@arm.com>
3417             David Sherwood  <david.sherwood@arm.com>
3418
3419         * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3420         x exists before using it.
3421         (aarch64_add_constant_internal): Rename to...
3422         (aarch64_add_offset_1): ...this.  Replace regnum with separate
3423         src and dest rtxes.  Handle the case in which they're different,
3424         including when the offset is zero.  Replace scratchreg with an rtx.
3425         Use 2 additions if there is no spare register into which we can
3426         move a 16-bit constant.
3427         (aarch64_add_constant): Delete.
3428         (aarch64_add_offset): Replace reg with separate src and dest
3429         rtxes.  Take a poly_int64 offset instead of a HOST_WIDE_INT.
3430         Use aarch64_add_offset_1.
3431         (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3432         an rtx rather than an int.  Take the delta as a poly_int64
3433         rather than a HOST_WIDE_INT.  Use aarch64_add_offset.
3434         (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3435         (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3436         aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3437         (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3438         and aarch64_add_sp.
3439         (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3440         aarch64_add_constant.
3441
3442 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
3443
3444         * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3445         Use scalar_float_mode.
3446
3447 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
3448
3449         * config/aarch64/aarch64-simd.md
3450         (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3451         (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3452         (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3453         (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3454         (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3455         (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3456         (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3457         (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3458         (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3459         (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3460
3461 2018-01-11  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
3462
3463         PR target/83514
3464         * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3465         targ_options->x_arm_arch_string is non NULL.
3466
3467 2018-01-11  Tamar Christina  <tamar.christina@arm.com>
3468
3469         * config/aarch64/aarch64.h
3470         (AARCH64_FL_FOR_ARCH8_4): Add  AARCH64_FL_DOTPROD.
3471
3472 2018-01-11  Sudakshina Das  <sudi.das@arm.com>
3473
3474         PR target/82096
3475         * expmed.c (emit_store_flag_force): Swap if const op0
3476         and change VOIDmode to mode of op0.
3477
3478 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
3479
3480         PR rtl-optimization/83761
3481         * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3482         than bytes to mode_for_size.
3483
3484 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
3485
3486         PR middle-end/83189
3487         * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3488         * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3489         profile.
3490
3491 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
3492
3493         PR middle-end/83575
3494         * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3495         when in layout mode.
3496         (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3497         * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3498         partition fixup.
3499
3500 2018-01-10  Michael Collison  <michael.collison@arm.com>
3501
3502         * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3503         * config/aarch64/aarch64-option-extension.def: Add
3504         AARCH64_OPT_EXTENSION of 'fp16fml'.
3505         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3506         (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3507         * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3508         * config/aarch64/constraints.md (Ui7): New constraint.
3509         * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3510         (VFMLA_SEL_W): Ditto.
3511         (f16quad): Ditto.
3512         (f16mac1): Ditto.
3513         (VFMLA16_LOW): New int iterator.
3514         (VFMLA16_HIGH): Ditto.
3515         (UNSPEC_FMLAL): New unspec.
3516         (UNSPEC_FMLSL): Ditto.
3517         (UNSPEC_FMLAL2): Ditto.
3518         (UNSPEC_FMLSL2): Ditto.
3519         (f16mac): New code attribute.
3520         * config/aarch64/aarch64-simd-builtins.def
3521         (aarch64_fmlal_lowv2sf): Ditto.
3522         (aarch64_fmlsl_lowv2sf): Ditto.
3523         (aarch64_fmlalq_lowv4sf): Ditto.
3524         (aarch64_fmlslq_lowv4sf): Ditto.
3525         (aarch64_fmlal_highv2sf): Ditto.
3526         (aarch64_fmlsl_highv2sf): Ditto.
3527         (aarch64_fmlalq_highv4sf): Ditto.
3528         (aarch64_fmlslq_highv4sf): Ditto.
3529         (aarch64_fmlal_lane_lowv2sf): Ditto.
3530         (aarch64_fmlsl_lane_lowv2sf): Ditto.
3531         (aarch64_fmlal_laneq_lowv2sf): Ditto.
3532         (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3533         (aarch64_fmlalq_lane_lowv4sf): Ditto.
3534         (aarch64_fmlsl_lane_lowv4sf): Ditto.
3535         (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3536         (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3537         (aarch64_fmlal_lane_highv2sf): Ditto.
3538         (aarch64_fmlsl_lane_highv2sf): Ditto.
3539         (aarch64_fmlal_laneq_highv2sf): Ditto.
3540         (aarch64_fmlsl_laneq_highv2sf): Ditto.
3541         (aarch64_fmlalq_lane_highv4sf): Ditto.
3542         (aarch64_fmlsl_lane_highv4sf): Ditto.
3543         (aarch64_fmlalq_laneq_highv4sf): Ditto.
3544         (aarch64_fmlsl_laneq_highv4sf): Ditto.
3545         * config/aarch64/aarch64-simd.md:
3546         (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3547         (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3548         (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3549         (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3550         (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3551         (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3552         (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3553         (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3554         (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3555         (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3556         (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3557         (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3558         (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3559         (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3560         (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3561         (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3562         (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3563         (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3564         (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3565         (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3566         * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3567         (vfmlsl_low_u32): Ditto.
3568         (vfmlalq_low_u32): Ditto.
3569         (vfmlslq_low_u32): Ditto.
3570         (vfmlal_high_u32): Ditto.
3571         (vfmlsl_high_u32): Ditto.
3572         (vfmlalq_high_u32): Ditto.
3573         (vfmlslq_high_u32): Ditto.
3574         (vfmlal_lane_low_u32): Ditto.
3575         (vfmlsl_lane_low_u32): Ditto.
3576         (vfmlal_laneq_low_u32): Ditto.
3577         (vfmlsl_laneq_low_u32): Ditto.
3578         (vfmlalq_lane_low_u32): Ditto.
3579         (vfmlslq_lane_low_u32): Ditto.
3580         (vfmlalq_laneq_low_u32): Ditto.
3581         (vfmlslq_laneq_low_u32): Ditto.
3582         (vfmlal_lane_high_u32): Ditto.
3583         (vfmlsl_lane_high_u32): Ditto.
3584         (vfmlal_laneq_high_u32): Ditto.
3585         (vfmlsl_laneq_high_u32): Ditto.
3586         (vfmlalq_lane_high_u32): Ditto.
3587         (vfmlslq_lane_high_u32): Ditto.
3588         (vfmlalq_laneq_high_u32): Ditto.
3589         (vfmlslq_laneq_high_u32): Ditto.
3590         * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3591         (AARCH64_FL_FOR_ARCH8_4): New.
3592         (AARCH64_ISA_F16FML): New ISA flag.
3593         (TARGET_F16FML): New feature flag for fp16fml.
3594         (doc/invoke.texi): Document new fp16fml option.
3595
3596 2018-01-10  Michael Collison  <michael.collison@arm.com>
3597
3598         * config/aarch64/aarch64-builtins.c:
3599         (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3600         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3601         (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3602         * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3603         (AARCH64_ISA_SHA3): New ISA flag.
3604         (TARGET_SHA3): New feature flag for sha3.
3605         * config/aarch64/iterators.md (sha512_op): New int attribute.
3606         (CRYPTO_SHA512): New int iterator.
3607         (UNSPEC_SHA512H): New unspec.
3608         (UNSPEC_SHA512H2): Ditto.
3609         (UNSPEC_SHA512SU0): Ditto.
3610         (UNSPEC_SHA512SU1): Ditto.
3611         * config/aarch64/aarch64-simd-builtins.def
3612         (aarch64_crypto_sha512hqv2di): New builtin.
3613         (aarch64_crypto_sha512h2qv2di): Ditto.
3614         (aarch64_crypto_sha512su0qv2di): Ditto.
3615         (aarch64_crypto_sha512su1qv2di): Ditto.
3616         (aarch64_eor3qv8hi): Ditto.
3617         (aarch64_rax1qv2di): Ditto.
3618         (aarch64_xarqv2di): Ditto.
3619         (aarch64_bcaxqv8hi): Ditto.
3620         * config/aarch64/aarch64-simd.md:
3621         (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3622         (aarch64_crypto_sha512su0qv2di): Ditto.
3623         (aarch64_crypto_sha512su1qv2di): Ditto.
3624         (aarch64_eor3qv8hi): Ditto.
3625         (aarch64_rax1qv2di): Ditto.
3626         (aarch64_xarqv2di): Ditto.
3627         (aarch64_bcaxqv8hi): Ditto.
3628         * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3629         (vsha512h2q_u64): Ditto.
3630         (vsha512su0q_u64): Ditto.
3631         (vsha512su1q_u64): Ditto.
3632         (veor3q_u16): Ditto.
3633         (vrax1q_u64): Ditto.
3634         (vxarq_u64): Ditto.
3635         (vbcaxq_u16): Ditto.
3636         * config/arm/types.md (crypto_sha512): New type attribute.
3637         (crypto_sha3): Ditto.
3638         (doc/invoke.texi): Document new sha3 option.
3639
3640 2018-01-10  Michael Collison  <michael.collison@arm.com>
3641
3642         * config/aarch64/aarch64-builtins.c:
3643         (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3644         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3645         (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3646         (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3647         * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3648         (AARCH64_ISA_SM4): New ISA flag.
3649         (TARGET_SM4): New feature flag for sm4.
3650         * config/aarch64/aarch64-simd-builtins.def
3651         (aarch64_sm3ss1qv4si): Ditto.
3652         (aarch64_sm3tt1aq4si): Ditto.
3653         (aarch64_sm3tt1bq4si): Ditto.
3654         (aarch64_sm3tt2aq4si): Ditto.
3655         (aarch64_sm3tt2bq4si): Ditto.
3656         (aarch64_sm3partw1qv4si): Ditto.
3657         (aarch64_sm3partw2qv4si): Ditto.
3658         (aarch64_sm4eqv4si): Ditto.
3659         (aarch64_sm4ekeyqv4si): Ditto.
3660         * config/aarch64/aarch64-simd.md:
3661         (aarch64_sm3ss1qv4si): Ditto.
3662         (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3663         (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3664         (aarch64_sm4eqv4si): Ditto.
3665         (aarch64_sm4ekeyqv4si): Ditto.
3666         * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3667         (sm3part_op): Ditto.
3668         (CRYPTO_SM3TT): Ditto.
3669         (CRYPTO_SM3PART): Ditto.
3670         (UNSPEC_SM3SS1): New unspec.
3671         (UNSPEC_SM3TT1A): Ditto.
3672         (UNSPEC_SM3TT1B): Ditto.
3673         (UNSPEC_SM3TT2A): Ditto.
3674         (UNSPEC_SM3TT2B): Ditto.
3675         (UNSPEC_SM3PARTW1): Ditto.
3676         (UNSPEC_SM3PARTW2): Ditto.
3677         (UNSPEC_SM4E): Ditto.
3678         (UNSPEC_SM4EKEY): Ditto.
3679         * config/aarch64/constraints.md (Ui2): New constraint.
3680         * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3681         * config/arm/types.md (crypto_sm3): New type attribute.
3682         (crypto_sm4): Ditto.
3683         * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3684         (vsm3tt1aq_u32): Ditto.
3685         (vsm3tt1bq_u32): Ditto.
3686         (vsm3tt2aq_u32): Ditto.
3687         (vsm3tt2bq_u32): Ditto.
3688         (vsm3partw1q_u32): Ditto.
3689         (vsm3partw2q_u32): Ditto.
3690         (vsm4eq_u32): Ditto.
3691         (vsm4ekeyq_u32): Ditto.
3692         (doc/invoke.texi): Document new sm4 option.
3693
3694 2018-01-10  Michael Collison  <michael.collison@arm.com>
3695
3696         * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3697         * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3698         (AARCH64_FL_FOR_ARCH8_4): New.
3699         (AARCH64_FL_V8_4): New flag.
3700         (doc/invoke.texi): Document new armv8.4-a option.
3701
3702 2018-01-10  Michael Collison  <michael.collison@arm.com>
3703
3704         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3705         (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3706         (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3707         * config/aarch64/aarch64-option-extension.def: Add
3708         AARCH64_OPT_EXTENSION of 'sha2'.
3709         (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3710         (crypto): Disable sha2 and aes if crypto disabled.
3711         (crypto): Enable aes and sha2 if enabled.
3712         (simd): Disable sha2 and aes if simd disabled.
3713         * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3714         New flags.
3715         (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3716         (TARGET_SHA2): New feature flag for sha2.
3717         (TARGET_AES): New feature flag for aes.
3718         * config/aarch64/aarch64-simd.md:
3719         (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3720         conditional on TARGET_AES.
3721         (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3722         (aarch64_crypto_sha1hsi): Make pattern conditional
3723         on TARGET_SHA2.
3724         (aarch64_crypto_sha1hv4si): Ditto.
3725         (aarch64_be_crypto_sha1hv4si): Ditto.
3726         (aarch64_crypto_sha1su1v4si): Ditto.
3727         (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3728         (aarch64_crypto_sha1su0v4si): Ditto.
3729         (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3730         (aarch64_crypto_sha256su0v4si): Ditto.
3731         (aarch64_crypto_sha256su1v4si): Ditto.
3732         (doc/invoke.texi): Document new aes and sha2 options.
3733
3734 2018-01-10  Martin Sebor  <msebor@redhat.com>
3735
3736         PR tree-optimization/83781
3737         * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3738         as string arrays.
3739
3740 2018-01-11  Martin Sebor  <msebor@gmail.com>
3741             Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
3742
3743         PR tree-optimization/83501
3744         PR tree-optimization/81703
3745
3746         * tree-ssa-strlen.c (get_string_cst): Rename...
3747         (get_string_len): ...to this.  Handle global constants.
3748         (handle_char_store): Adjust.
3749
3750 2018-01-10  Kito Cheng  <kito.cheng@gmail.com>
3751             Jim Wilson  <jimw@sifive.com>
3752
3753         * config/riscv/riscv-protos.h (riscv_output_return): New.
3754         * config/riscv/riscv.c (struct machine_function): New naked_p field.
3755         (riscv_attribute_table, riscv_output_return),
3756         (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3757         (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3758         (riscv_compute_frame_info): Only compute frame->mask if not a naked
3759         function.
3760         (riscv_expand_prologue): Add early return for naked function.
3761         (riscv_expand_epilogue): Likewise.
3762         (riscv_function_ok_for_sibcall): Return false for naked function.
3763         (riscv_set_current_function): New.
3764         (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3765         (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3766         * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3767         * doc/extend.texi (RISC-V Function Attributes): New.
3768
3769 2018-01-10  Michael Meissner  <meissner@linux.vnet.ibm.com>
3770
3771         * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3772         check for 128-bit long double before checking TCmode.
3773         * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3774         128-bit long doubles before checking TFmode or TCmode.
3775         (FLOAT128_IBM_P): Likewise.
3776
3777 2018-01-10  Martin Sebor  <msebor@redhat.com>
3778
3779         PR tree-optimization/83671
3780         * builtins.c (c_strlen): Unconditionally return zero for the empty
3781         string.
3782         Use -Warray-bounds for warnings.
3783         * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3784         for non-constant array indices with COMPONENT_REF, arrays of
3785         arrays, and pointers to arrays.
3786         (gimple_fold_builtin_strlen): Determine and set length range for
3787         non-constant character arrays.
3788
3789 2018-01-10  Aldy Hernandez  <aldyh@redhat.com>
3790
3791         PR middle-end/81897
3792         * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3793         empty blocks.
3794
3795 2018-01-10  Eric Botcazou  <ebotcazou@adacore.com>
3796
3797         * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3798
3799 2018-01-10  Peter Bergner  <bergner@vnet.ibm.com>
3800
3801         PR target/83399
3802         * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3803         VECTOR_MEM_ALTIVEC_OR_VSX_P.
3804         * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3805         indexed_or_indirect_operand predicate.
3806         (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3807         (*vsx_le_perm_load_v8hi): Likewise.
3808         (*vsx_le_perm_load_v16qi): Likewise.
3809         (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3810         (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3811         (*vsx_le_perm_store_v8hi): Likewise.
3812         (*vsx_le_perm_store_v16qi): Likewise.
3813         (eight unnamed splitters): Likewise.
3814
3815 2018-01-10  Peter Bergner  <bergner@vnet.ibm.com>
3816
3817         * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3818         * config/rs6000/emmintrin.h: Likewise.
3819         * config/rs6000/mmintrin.h: Likewise.
3820         * config/rs6000/xmmintrin.h: Likewise.
3821
3822 2018-01-10  David Malcolm  <dmalcolm@redhat.com>
3823
3824         PR c++/43486
3825         * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3826         "public_flag".
3827         * tree.c (tree_nop_conversion): Return true for location wrapper
3828         nodes.
3829         (maybe_wrap_with_location): New function.
3830         (selftest::check_strip_nops): New function.
3831         (selftest::test_location_wrappers): New function.
3832         (selftest::tree_c_tests): Call it.
3833         * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3834         (maybe_wrap_with_location): New decl.
3835         (EXPR_LOCATION_WRAPPER_P): New macro.
3836         (location_wrapper_p): New inline function.
3837         (tree_strip_any_location_wrapper): New inline function.
3838
3839 2018-01-10  H.J. Lu  <hongjiu.lu@intel.com>
3840
3841         PR target/83735
3842         * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3843         stack_realign_offset for the largest alignment of stack slot
3844         actually used.
3845         (ix86_find_max_used_stack_alignment): New function.
3846         (ix86_finalize_stack_frame_flags): Use it.  Set
3847         max_used_stack_alignment if we don't realign stack.
3848         * config/i386/i386.h (machine_function): Add
3849         max_used_stack_alignment.
3850
3851 2018-01-10  Christophe Lyon  <christophe.lyon@linaro.org>
3852
3853         * config/arm/arm.opt (-mbranch-cost): New option.
3854         * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3855         account.
3856
3857 2018-01-10  Segher Boessenkool  <segher@kernel.crashing.org>
3858
3859         PR target/83629
3860         * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3861         load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3862
3863 2018-01-10  Richard Biener  <rguenther@suse.de>
3864
3865         PR debug/83765
3866         * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3867         early out so it also covers the case where we have a non-NULL
3868         origin.
3869
3870 2018-01-10  Richard Sandiford  <richard.sandiford@linaro.org>
3871
3872         PR tree-optimization/83753
3873         * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3874         for non-strided grouped accesses if the number of elements is 1.
3875
3876 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
3877
3878         PR target/81616
3879         * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3880         * i386.h (TARGET_USE_GATHER): Define.
3881         * x86-tune.def (X86_TUNE_USE_GATHER): New.
3882
3883 2018-01-10  Martin Liska  <mliska@suse.cz>
3884
3885         PR bootstrap/82831
3886         * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3887         * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3888         partitioning.
3889         * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3890         CLEANUP_NO_PARTITIONING is not set.
3891
3892 2018-01-10  Richard Sandiford  <richard.sandiford@linaro.org>
3893
3894         * doc/rtl.texi: Remove documentation of (const ...) wrappers
3895         for vectors, as a partial revert of r254296.
3896         * rtl.h (const_vec_p): Delete.
3897         (const_vec_duplicate_p): Don't test for vector CONSTs.
3898         (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3899         * expmed.c (make_tree): Likewise.
3900
3901         Revert:
3902         * common.md (E, F): Use CONSTANT_P instead of checking for
3903         CONST_VECTOR.
3904         * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3905         checking for CONST_VECTOR.
3906
3907 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
3908
3909         PR middle-end/83575
3910         * predict.c (force_edge_cold): Handle in more sane way edges
3911         with no prediction.
3912
3913 2018-01-09  Carl Love  <cel@us.ibm.com>
3914
3915         * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3916         V4SI, V4SF types.
3917         (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3918         * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3919         VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3920         VMRGOW_V2DI, VMRGOW_V2DF.  Remove definition for VMRGOW.
3921         * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3922         P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW):  Add definitions.
3923         * config/rs6000/rs6000-protos.h: Add extern defition for
3924         rs6000_generate_float2_double_code.
3925         * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3926         function.
3927         * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3928         (float2_v2df): Add define_expand.
3929
3930 2018-01-09  Uros Bizjak  <ubizjak@gmail.com>
3931
3932         PR target/83628
3933         * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3934         op_mode in the force_to_mode call.
3935
3936 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
3937
3938         * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3939         instead of checking each element individually.
3940         (aarch64_evpc_uzp): Likewise.
3941         (aarch64_evpc_zip): Likewise.
3942         (aarch64_evpc_ext): Likewise.
3943         (aarch64_evpc_rev): Likewise.
3944         (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3945         instead of checking each element individually.  Return true without
3946         generating rtl if
3947         (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3948         whether all selected elements come from the same input, instead of
3949         checking each element individually.  Remove calls to gen_rtx_REG,
3950         start_sequence and end_sequence and instead assert that no rtl is
3951         generated.
3952
3953 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
3954
3955         * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3956         order of HIGH and CONST checks.
3957
3958 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
3959
3960         * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3961         if the destination isn't an SSA_NAME.
3962
3963 2018-01-09  Richard Biener  <rguenther@suse.de>
3964
3965         PR tree-optimization/83668
3966         * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3967         move prologue...
3968         (canonicalize_loop_form): ... here, renamed from ...
3969         (canonicalize_loop_closed_ssa_form): ... this and amended to
3970         swap successor edges for loop exit blocks to make us use
3971         the RPO order we need for initial schedule generation.
3972
3973 2018-01-09  Joseph Myers  <joseph@codesourcery.com>
3974
3975         PR tree-optimization/64811
3976         * match.pd: When optimizing comparisons with Inf, avoid
3977         introducing or losing exceptions from comparisons with NaN.
3978
3979 2018-01-09  Martin Liska  <mliska@suse.cz>
3980
3981         PR sanitizer/82517
3982         * asan.c (shadow_mem_size): Add gcc_assert.
3983
3984 2018-01-09  Georg-Johann Lay  <avr@gjlay.de>
3985
3986         Don't save registers in main().
3987
3988         PR target/83738
3989         * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3990         * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3991         * config/avr/avr.c (avr_set_current_function): Don't error if
3992         naked, OS_task or OS_main are specified at the same time.
3993         (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3994         OS_main.
3995         (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3996         attribute.
3997         * common/config/avr/avr-common.c (avr_option_optimization_table):
3998         Switch on -mmain-is-OS_task for optimizing compilations.
3999
4000 2018-01-09  Richard Biener  <rguenther@suse.de>
4001
4002         PR tree-optimization/83572
4003         * graphite.c: Include cfganal.h.
4004         (graphite_transform_loops): Connect infinite loops to exit
4005         and remove fake edges at the end.
4006
4007 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
4008
4009         * ipa-inline.c (edge_badness): Revert accidental checkin.
4010
4011 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
4012
4013         PR ipa/80763
4014         * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4015         symbols; not inline clones.
4016
4017 2018-01-09  Jakub Jelinek  <jakub@redhat.com>
4018
4019         PR target/83507
4020         * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4021         hard registers.  Formatting fixes.
4022
4023         PR preprocessor/83722
4024         * gcc.c (try_generate_repro): Pass
4025         &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4026         &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4027         do_report_bug.
4028
4029 2018-01-08  Monk Chiang  <sh.chiang04@gmail.com>
4030             Kito Cheng  <kito.cheng@gmail.com>
4031
4032         * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4033         (riscv_leaf_function_p): Delete.
4034         (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4035
4036 2018-01-08  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
4037
4038         * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4039         function.
4040         (do_ifelse): New function.
4041         (do_isel): New function.
4042         (do_sub3): New function.
4043         (do_add3): New function.
4044         (do_load_mask_compare): New function.
4045         (do_overlap_load_compare): New function.
4046         (expand_compare_loop): New function.
4047         (expand_block_compare): Call expand_compare_loop() when appropriate.
4048         * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4049         option description.
4050         (-mblock-compare-inline-loop-limit): New option.
4051
4052 2018-01-08  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
4053
4054         PR target/83677
4055         * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4056         Reverse order of second and third operands in first alternative.
4057         * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4058         of first and second elements in UNSPEC_VPERMR vector.
4059         (altivec_expand_vec_perm_le): Likewise.
4060
4061 2018-01-08  Jeff Law  <law@redhat.com>
4062
4063         PR rtl-optimizatin/81308
4064         * tree-switch-conversion.c (cfg_altered): New file scoped static.
4065         (process_switch): If group_case_labels makes a change, then set
4066         cfg_altered.
4067         (pass_convert_switch::execute): If a switch is converted, then
4068         set cfg_altered.  Return TODO_cfg_cleanup if cfg_altered is true.
4069
4070         PR rtl-optimization/81308
4071         * recog.c (split_all_insns): Conditionally cleanup the CFG after
4072         splitting insns.
4073
4074 2018-01-08  Vidya Praveen  <vidyapraveen@arm.com>
4075
4076         PR target/83663 - Revert r255946
4077         * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4078         generation for cases where splatting a value is not useful.
4079         * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4080         across a vec_duplicate and a paradoxical subreg forming a vector
4081         mode to a vec_concat.
4082
4083 2018-01-08  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
4084
4085         * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4086         -march=armv8.3-a variants.
4087         * config/arm/t-multilib: Likewise.
4088         * config/arm/t-arm-elf: Likewise.  Handle dotprod extension.
4089
4090 2018-01-08  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
4091
4092         * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4093         to generate rtl.
4094         (cceq_ior_compare_complement): Give it a name so I can use it, and
4095         change boolean_or_operator predicate to boolean_operator so it can
4096         be used to generate a crand.
4097         (eqne): New code iterator.
4098         (bd/bd_neg): New code_attrs.
4099         (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4100         a single define_insn.
4101         (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4102         decrement (bdnzt/bdnzf/bdzt/bdzf).
4103         * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4104         with the new names of the branch decrement patterns, and added the
4105         names of the branch decrement conditional patterns.
4106
4107 2018-01-08  Richard Biener  <rguenther@suse.de>
4108
4109         PR tree-optimization/83563
4110         * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4111         cache.
4112
4113 2018-01-08  Richard Biener  <rguenther@suse.de>
4114
4115         PR middle-end/83713
4116         * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4117
4118 2018-01-08  Richard Biener  <rguenther@suse.de>
4119
4120         PR tree-optimization/83685
4121         * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4122         references to abnormals.
4123
4124 2018-01-08  Richard Biener  <rguenther@suse.de>
4125
4126         PR lto/83719
4127         * dwarf2out.c (output_indirect_strings): Handle empty
4128         skeleton_debug_str_hash.
4129         (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4130
4131 2018-01-08  Claudiu Zissulescu  <claziss@synopsys.com>
4132
4133         * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4134         (emit_store_direct): Likewise.
4135         (arc_trampoline_adjust_address): Likewise.
4136         (arc_asm_trampoline_template): New function.
4137         (arc_initialize_trampoline): Use asm_trampoline_template.
4138         (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4139         * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4140         * config/arc/arc.md (flush_icache): Delete pattern.
4141
4142 2018-01-08  Claudiu Zissulescu  <claziss@synopsys.com>
4143
4144         * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4145         * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4146         munaligned-access.
4147
4148 2018-01-08  Sebastian Huber  <sebastian.huber@embedded-brains.de>
4149
4150         PR target/83681
4151         * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4152         by not USED_FOR_TARGET.
4153         (make_pass_resolve_sw_modes): Likewise.
4154
4155 2018-01-08  Sebastian Huber  <sebastian.huber@embedded-brains.de>
4156
4157         * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4158         USED_FOR_TARGET.
4159
4160 2018-01-08  Richard Biener  <rguenther@suse.de>
4161
4162         PR middle-end/83580
4163         * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4164
4165 2018-01-08  Richard Biener  <rguenther@suse.de>
4166
4167         PR middle-end/83517
4168         * match.pd ((t * 2) / 2) -> t): Add missing :c.
4169
4170 2018-01-06  Aldy Hernandez  <aldyh@redhat.com>
4171
4172         PR middle-end/81897
4173         * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4174         basic blocks with a small number of successors.
4175         (convert_control_dep_chain_into_preds): Improve handling of
4176         forwarder blocks.
4177         (dump_predicates): Split apart into...
4178         (dump_pred_chain): ...here...
4179         (dump_pred_info): ...and here.
4180         (can_one_predicate_be_invalidated_p): Add debugging printfs.
4181         (can_chain_union_be_invalidated_p): Improve check for invalidation
4182         of paths.
4183         (uninit_uses_cannot_happen): Avoid unnecessary if
4184         convert_control_dep_chain_into_preds yielded nothing.
4185
4186 2018-01-06  Martin Sebor  <msebor@redhat.com>
4187
4188         PR tree-optimization/83640
4189         * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4190         subtracting negative offset from size.
4191         (builtin_access::overlap): Adjust offset bounds of the access to fall
4192         within the size of the object if possible.
4193
4194 2018-01-06  Richard Sandiford  <richard.sandiford@linaro.org>
4195
4196         PR rtl-optimization/83699
4197         * expmed.c (extract_bit_field_1): Restrict the vector usage of
4198         extract_bit_field_as_subreg to cases in which the extracted
4199         value is also a vector.
4200
4201         * lra-constraints.c (process_alt_operands): Test for the equivalence
4202         substitutions when detecting a possible reload cycle.
4203
4204 2018-01-06  Jakub Jelinek  <jakub@redhat.com>
4205
4206         PR debug/83480
4207         * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4208         by default if flag_selective_schedling{,2}.  Formatting fixes.
4209
4210         PR rtl-optimization/83682
4211         * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4212         if it has non-VECTOR_MODE element mode.
4213         (vec_duplicate_p): Likewise.
4214
4215         PR middle-end/83694
4216         * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4217         and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4218
4219 2018-01-05  Jakub Jelinek  <jakub@redhat.com>
4220
4221         PR target/83604
4222         * config/i386/i386-builtin.def
4223         (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4224         __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4225         Require also OPTION_MASK_ISA_AVX512F in addition to
4226         OPTION_MASK_ISA_GFNI.
4227         (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4228         __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4229         OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4230         to OPTION_MASK_ISA_GFNI.
4231         (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4232         OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4233         OPTION_MASK_ISA_AVX512BW.
4234         (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4235         OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4236         addition to OPTION_MASK_ISA_GFNI.
4237         (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4238         __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4239         Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4240         to OPTION_MASK_ISA_GFNI.
4241         * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4242         a requirement for all ISAs rather than any of them with a few
4243         exceptions.
4244         (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4245         processing.
4246         (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4247         bitmasks to be enabled with 3 exceptions, instead of requiring any
4248         enabled ISA with lots of exceptions.
4249         * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4250         vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4251         Change avx512bw in isa attribute to avx512f.
4252         * config/i386/sgxintrin.h: Add license boilerplate.
4253         * config/i386/vaesintrin.h: Likewise.  Fix macro spelling __AVX512F
4254         to __AVX512F__ and __AVX512VL to __AVX512VL__.
4255         (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4256         _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4257         defined.
4258         * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4259         _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4260         temporarily sse2 rather than sse if not enabled already.
4261
4262         PR target/83604
4263         * config/i386/sse.md (VI248_VLBW): Rename to ...
4264         (VI248_AVX512VL): ... this.  Don't guard V32HI with TARGET_AVX512BW.
4265         (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4266         vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4267         vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4268         vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4269         mode iterator instead of VI248_VLBW.
4270
4271 2018-01-05  Jan Hubicka  <hubicka@ucw.cz>
4272
4273         * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4274         (record_modified): Skip clobbers; add debug output.
4275         (param_change_prob): Use sreal frequencies.
4276
4277 2018-01-05  Richard Sandiford  <richard.sandiford@linaro.org>
4278
4279         * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4280         punt for user-aligned variables.
4281
4282 2018-01-05  Richard Sandiford  <richard.sandiford@linaro.org>
4283
4284         * tree-chrec.c (chrec_contains_symbols): Return true for
4285         POLY_INT_CST.
4286
4287 2018-01-05  Sudakshina Das  <sudi.das@arm.com>
4288
4289         PR target/82439
4290         * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4291         of (x|y) == x for BICS pattern.
4292
4293 2018-01-05  Jakub Jelinek  <jakub@redhat.com>
4294
4295         PR tree-optimization/83605
4296         * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4297         (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4298         can throw.
4299
4300 2018-01-05  Sebastian Huber  <sebastian.huber@embedded-brains.de>
4301
4302         * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4303         * config/epiphany/rtems.h: New file.
4304
4305 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
4306             Uros Bizjak  <ubizjak@gmail.com>
4307
4308         PR target/83554
4309         * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4310         QIreg_operand instead of register_operand predicate.
4311         * config/i386/i386.c (ix86_rop_should_change_byte_p,
4312         set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4313         comments instead of -fmitigate[-_]rop.
4314
4315 2018-01-04  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
4316
4317         PR bootstrap/81926
4318         * cgraphunit.c (symbol_table::compile): Switch to text_section
4319         before calling assembly_start debug hook.
4320         * run-rtl-passes.c (run_rtl_passes): Likewise.
4321         Include output.h.
4322
4323 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
4324
4325         * tree-vrp.c (extract_range_from_binary_expr_1): Check
4326         range_int_cst_p rather than !symbolic_range_p before calling
4327         extract_range_from_multiplicative_op_1.
4328
4329 2018-01-04  Jeff Law  <law@redhat.com>
4330
4331         * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4332         redundant test in assertion.
4333
4334 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
4335
4336         * doc/rtl.texi: Document machine_mode wrapper classes.
4337
4338 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
4339
4340         * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4341         using tree_to_uhwi.
4342
4343 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
4344
4345         * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4346         the VEC_PERM_EXPR fold to fail.
4347
4348 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
4349
4350         PR debug/83585
4351         * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4352         to switched_sections.
4353
4354 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
4355
4356         PR target/83680
4357         * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4358         test for d.testing.
4359
4360 2018-01-04  Peter Bergner  <bergner@vnet.ibm.com>
4361
4362         PR target/83387
4363         * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4364         allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4365
4366 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
4367
4368         PR debug/83666
4369         * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4370         is BLKmode and bitpos not zero or mode change is needed.
4371
4372 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
4373
4374         PR target/83675
4375         * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4376         TARGET_VIS2.
4377
4378 2018-01-04  Uros Bizjak  <ubizjak@gmail.com>
4379
4380         PR target/83628
4381         * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4382         instead of MULT rtx.  Update all corresponding splitters.
4383         (*saddl_se): Ditto.
4384         (*ssub<modesuffix>): Ditto.
4385         (*ssubl_se): Ditto.
4386         (*cmp_sadd_di): Update split patterns.
4387         (*cmp_sadd_si): Ditto.
4388         (*cmp_sadd_sidi): Ditto.
4389         (*cmp_ssub_di): Ditto.
4390         (*cmp_ssub_si): Ditto.
4391         (*cmp_ssub_sidi): Ditto.
4392         * config/alpha/predicates.md (const23_operand): New predicate.
4393         * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4394         Look for ASHIFT, not MULT inner operand.
4395         (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4396
4397 2018-01-04  Martin Liska  <mliska@suse.cz>
4398
4399         PR gcov-profile/83669
4400         * gcov.c (output_intermediate_file): Add version to intermediate
4401         gcov file.
4402         * doc/gcov.texi: Document new field 'version' in intermediate
4403         file format. Fix location of '-k' option of gcov command.
4404
4405 2018-01-04  Martin Liska  <mliska@suse.cz>
4406
4407         PR ipa/82352
4408         * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4409
4410 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
4411
4412         * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4413
4414 2018-01-03  Martin Sebor  <msebor@redhat.com>
4415
4416         PR tree-optimization/83655
4417         * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4418         checking calls with invalid arguments.
4419
4420 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4421
4422         * tree-vect-stmts.c (vect_get_store_rhs): New function.
4423         (vectorizable_mask_load_store): Delete.
4424         (vectorizable_call): Return false for masked loads and stores.
4425         (vectorizable_store): Handle IFN_MASK_STORE.  Use vect_get_store_rhs
4426         instead of gimple_assign_rhs1.
4427         (vectorizable_load): Handle IFN_MASK_LOAD.
4428         (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4429
4430 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4431
4432         * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4433         split out from..,
4434         (vectorizable_mask_load_store): ...here.
4435         (vectorizable_load): ...and here.
4436
4437 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4438
4439         * tree-vect-stmts.c (vect_build_all_ones_mask)
4440         (vect_build_zero_merge_argument): New functions, split out from...
4441         (vectorizable_load): ...here.
4442
4443 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4444
4445         * tree-vect-stmts.c (vect_check_store_rhs): New function,
4446         split out from...
4447         (vectorizable_mask_load_store): ...here.
4448         (vectorizable_store): ...and here.
4449
4450 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4451
4452         * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4453         split out from...
4454         (vectorizable_mask_load_store): ...here.
4455
4456 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4457
4458         * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4459         (vect_model_store_cost): Take a vec_load_store_type instead of a
4460         vect_def_type.
4461         * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4462         (vect_model_store_cost): Take a vec_load_store_type instead of a
4463         vect_def_type.
4464         (vectorizable_mask_load_store): Update accordingly.
4465         (vectorizable_store): Likewise.
4466         * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4467
4468 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4469
4470         * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4471         IFN_MASK_LOAD calls here rather than...
4472         * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4473
4474 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4475             Alan Hayward  <alan.hayward@arm.com>
4476             David Sherwood  <david.sherwood@arm.com>
4477
4478         * expmed.c (extract_bit_field_1): For vector extracts,
4479         fall back to extract_bit_field_as_subreg if vec_extract
4480         isn't available.
4481
4482 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4483             Alan Hayward  <alan.hayward@arm.com>
4484             David Sherwood  <david.sherwood@arm.com>
4485
4486         * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4487         they are variable or constant sized.
4488         (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4489         slots for constant-sized data.
4490
4491 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4492             Alan Hayward  <alan.hayward@arm.com>
4493             David Sherwood  <david.sherwood@arm.com>
4494
4495         * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4496         handling COND_EXPRs with boolean comparisons, try to find a better
4497         basis for the mask type than the boolean itself.
4498
4499 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4500
4501         * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4502         is calculated and how it can be overridden.
4503         * genmodes.c (max_bitsize_mode_any_mode): New variable.
4504         (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4505         if defined.
4506         (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4507         if nonzero.
4508
4509 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4510             Alan Hayward  <alan.hayward@arm.com>
4511             David Sherwood  <david.sherwood@arm.com>
4512
4513         * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4514         Remove the mode argument.
4515         (aarch64_simd_valid_immediate): Remove the mode and inverse
4516         arguments.
4517         * config/aarch64/iterators.md (bitsize): New iterator.
4518         * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4519         (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4520         * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4521         aarch64_simd_valid_immediate.
4522         * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4523         (aarch64_reg_or_bic_imm): Likewise.
4524         * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4525         with an insn_type enum and msl with a modifier_type enum.
4526         Replace element_width with a scalar_mode.  Change the shift
4527         to unsigned int.  Add constructors for scalar_float_mode and
4528         scalar_int_mode elements.
4529         (aarch64_vect_float_const_representable_p): Delete.
4530         (aarch64_can_const_movi_rtx_p)
4531         (aarch64_simd_scalar_immediate_valid_for_move)
4532         (aarch64_simd_make_constant): Update call to
4533         aarch64_simd_valid_immediate.
4534         (aarch64_advsimd_valid_immediate_hs): New function.
4535         (aarch64_advsimd_valid_immediate): Likewise.
4536         (aarch64_simd_valid_immediate): Remove mode and inverse
4537         arguments.  Rewrite to use the above.  Use const_vec_duplicate_p
4538         to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4539         and aarch64_float_const_representable_p on the result.
4540         (aarch64_output_simd_mov_immediate): Remove mode argument.
4541         Update call to aarch64_simd_valid_immediate and use of
4542         simd_immediate_info.
4543         (aarch64_output_scalar_simd_mov_immediate): Update call
4544         accordingly.
4545
4546 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4547             Alan Hayward  <alan.hayward@arm.com>
4548             David Sherwood  <david.sherwood@arm.com>
4549
4550         * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4551         (mode_nunits): Likewise CONST_MODE_NUNITS.
4552         * machmode.def (ADJUST_NUNITS): Document.
4553         * genmodes.c (mode_data::need_nunits_adj): New field.
4554         (blank_mode): Update accordingly.
4555         (adj_nunits): New variable.
4556         (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4557         parameter.
4558         (emit_mode_size_inline): Set need_bytesize_adj for all modes
4559         listed in adj_nunits.
4560         (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4561         listed in adj_nunits.  Don't emit case statements for such modes.
4562         (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4563         and CONST_MODE_PRECISION.  Make CONST_MODE_SIZE expand to
4564         nothing if adj_nunits is nonnull.
4565         (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4566         (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4567         (emit_mode_fbit): Update use of print_maybe_const_decl.
4568         (emit_move_size): Likewise.  Treat the array as non-const
4569         if adj_nunits.
4570         (emit_mode_adjustments): Handle adj_nunits.
4571
4572 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4573
4574         * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4575         * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4576         (VECTOR_MODES): Use it.
4577         (make_vector_modes): Take the prefix as an argument.
4578
4579 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4580             Alan Hayward  <alan.hayward@arm.com>
4581             David Sherwood  <david.sherwood@arm.com>
4582
4583         * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4584         * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4585         for MODE_VECTOR_BOOL.
4586         * machmode.def (VECTOR_BOOL_MODE): Document.
4587         * genmodes.c (VECTOR_BOOL_MODE): New macro.
4588         (make_vector_bool_mode): New function.
4589         (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4590         MODE_VECTOR_BOOL.
4591         * lto-streamer-in.c (lto_input_mode_table): Likewise.
4592         * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4593         Likewise.
4594         * stor-layout.c (int_mode_for_mode): Likewise.
4595         * tree.c (build_vector_type_for_mode): Likewise.
4596         * varasm.c (output_constant_pool_2): Likewise.
4597         * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4598         CONSTM1_RTX (BImode) are the same thing.  Initialize const_tiny_rtx
4599         for MODE_VECTOR_BOOL.
4600         * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4601         of mode class checks.
4602         * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4603         instead of a list of mode class checks.
4604         (expand_vector_scalar_condition): Likewise.
4605         (type_for_widest_vector_mode): Handle BImode as an inner mode.
4606
4607 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4608             Alan Hayward  <alan.hayward@arm.com>
4609             David Sherwood  <david.sherwood@arm.com>
4610
4611         * machmode.h (mode_size): Change from unsigned short to
4612         poly_uint16_pod.
4613         (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4614         (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4615         or if measurement_type is not polynomial.
4616         (fixed_size_mode::includes_p): Check for constant-sized modes.
4617         * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4618         return a poly_uint16 rather than an unsigned short.
4619         (emit_mode_size): Change the type of mode_size from unsigned short
4620         to poly_uint16_pod.  Use ZERO_COEFFS for the initializer.
4621         (emit_mode_adjustments): Cope with polynomial vector sizes.
4622         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4623         for GET_MODE_SIZE.
4624         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4625         for GET_MODE_SIZE.
4626         * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4627         * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4628         * caller-save.c (setup_save_areas): Likewise.
4629         (replace_reg_with_saved_mem): Likewise.
4630         * calls.c (emit_library_call_value_1): Likewise.
4631         * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4632         * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4633         (gen_lowpart_for_combine): Likewise.
4634         * convert.c (convert_to_integer_1): Likewise.
4635         * cse.c (equiv_constant, cse_insn): Likewise.
4636         * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4637         (cselib_subst_to_values): Likewise.
4638         * dce.c (word_dce_process_block): Likewise.
4639         * df-problems.c (df_word_lr_mark_ref): Likewise.
4640         * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4641         * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4642         (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4643         (rtl_for_decl_location): Likewise.
4644         * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4645         * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4646         * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4647         (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4648         (expand_expr_real_1): Likewise.
4649         * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4650         (pad_below): Likewise.
4651         * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4652         * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4653         * ira.c (get_subreg_tracking_sizes): Likewise.
4654         * ira-build.c (ira_create_allocno_objects): Likewise.
4655         * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4656         (ira_sort_regnos_for_alter_reg): Likewise.
4657         * ira-costs.c (record_operand_costs): Likewise.
4658         * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4659         (resolve_simple_move): Likewise.
4660         * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4661         (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4662         (lra_constraints): Likewise.
4663         (CONST_POOL_OK_P): Reject variable-sized modes.
4664         * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4665         (add_pseudo_to_slot, lra_spill): Likewise.
4666         * omp-low.c (omp_clause_aligned_alignment): Likewise.
4667         * optabs-query.c (get_best_extraction_insn): Likewise.
4668         * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4669         * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4670         (expand_mult_highpart, valid_multiword_target_p): Likewise.
4671         * recog.c (offsettable_address_addr_space_p): Likewise.
4672         * regcprop.c (maybe_mode_change): Likewise.
4673         * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4674         * regrename.c (build_def_use): Likewise.
4675         * regstat.c (dump_reg_info): Likewise.
4676         * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4677         (find_reloads, find_reloads_subreg_address): Likewise.
4678         * reload1.c (eliminate_regs_1): Likewise.
4679         * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4680         * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4681         (simplify_binary_operation_1, simplify_subreg): Likewise.
4682         * targhooks.c (default_function_arg_padding): Likewise.
4683         (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4684         * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4685         (verify_gimple_assign_ternary): Likewise.
4686         * tree-inline.c (estimate_move_cost): Likewise.
4687         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4688         * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4689         (get_address_cost_ainc): Likewise.
4690         * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4691         (vect_supportable_dr_alignment): Likewise.
4692         * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4693         (vectorizable_reduction): Likewise.
4694         * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4695         (vectorizable_operation, vectorizable_load): Likewise.
4696         * tree.c (build_same_sized_truth_vector_type): Likewise.
4697         * valtrack.c (cleanup_auto_inc_dec): Likewise.
4698         * var-tracking.c (emit_note_insn_var_location): Likewise.
4699         * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4700         (ADDR_VEC_ALIGN): Likewise.
4701
4702 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4703             Alan Hayward  <alan.hayward@arm.com>
4704             David Sherwood  <david.sherwood@arm.com>
4705
4706         * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4707         unsigned short.
4708         (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4709         or if measurement_type is polynomial.
4710         * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4711         * combine.c (make_extraction): Likewise.
4712         * dse.c (find_shift_sequence): Likewise.
4713         * dwarf2out.c (mem_loc_descriptor): Likewise.
4714         * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4715         (extract_bit_field, extract_low_bits): Likewise.
4716         * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4717         (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4718         (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4719         * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4720         * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4721         * reload.c (find_reloads): Likewise.
4722         * reload1.c (alter_reg): Likewise.
4723         * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4724         * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4725         * tree-if-conv.c (predicate_mem_writes): Likewise.
4726         * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4727         * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4728         * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4729         * valtrack.c (dead_debug_insert_temp): Likewise.
4730         * varasm.c (mergeable_constant_section): Likewise.
4731         * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4732
4733 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4734             Alan Hayward  <alan.hayward@arm.com>
4735             David Sherwood  <david.sherwood@arm.com>
4736
4737         * expr.c (expand_assignment): Cope with polynomial mode sizes
4738         when assigning to a CONCAT.
4739
4740 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4741             Alan Hayward  <alan.hayward@arm.com>
4742             David Sherwood  <david.sherwood@arm.com>
4743
4744         * machmode.h (mode_precision): Change from unsigned short to
4745         poly_uint16_pod.
4746         (mode_to_precision): Return a poly_uint16 rather than an unsigned
4747         short.
4748         (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4749         or if measurement_type is not polynomial.
4750         (HWI_COMPUTABLE_MODE_P): Turn into a function.  Optimize the case
4751         in which the mode is already known to be a scalar_int_mode.
4752         * genmodes.c (emit_mode_precision): Change the type of mode_precision
4753         from unsigned short to poly_uint16_pod.  Use ZERO_COEFFS for the
4754         initializer.
4755         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4756         for GET_MODE_PRECISION.
4757         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4758         for GET_MODE_PRECISION.
4759         * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4760         as polynomial.
4761         (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4762         (expand_field_assignment, make_extraction): Likewise.
4763         (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4764         (get_last_value): Likewise.
4765         * convert.c (convert_to_integer_1): Likewise.
4766         * cse.c (cse_insn): Likewise.
4767         * expr.c (expand_expr_real_1): Likewise.
4768         * lra-constraints.c (simplify_operand_subreg): Likewise.
4769         * optabs-query.c (can_atomic_load_p): Likewise.
4770         * optabs.c (expand_atomic_load): Likewise.
4771         (expand_atomic_store): Likewise.
4772         * ree.c (combine_reaching_defs): Likewise.
4773         * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4774         * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4775         * tree.h (type_has_mode_precision_p): Likewise.
4776         * ubsan.c (instrument_si_overflow): Likewise.
4777
4778 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4779             Alan Hayward  <alan.hayward@arm.com>
4780             David Sherwood  <david.sherwood@arm.com>
4781
4782         * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4783         polynomial numbers of units.
4784         (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4785         (valid_vector_subparts_p): New function.
4786         (build_vector_type): Remove temporary shim and take the number
4787         of units as a poly_uint64 rather than an int.
4788         (build_opaque_vector_type): Take the number of units as a
4789         poly_uint64 rather than an int.
4790         * tree.c (build_vector_from_ctor): Handle polynomial
4791         TYPE_VECTOR_SUBPARTS.
4792         (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4793         (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4794         (build_vector_from_val): If the number of units is variable,
4795         use build_vec_duplicate_cst for constant operands and
4796         VEC_DUPLICATE_EXPR otherwise.
4797         (make_vector_type): Remove temporary is_constant ().
4798         (build_vector_type, build_opaque_vector_type): Take the number of
4799         units as a poly_uint64 rather than an int.
4800         (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4801         VECTOR_CST_NELTS.
4802         * cfgexpand.c (expand_debug_expr): Likewise.
4803         * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4804         (store_constructor, expand_expr_real_1): Likewise.
4805         (const_scalar_mask_from_tree): Likewise.
4806         * fold-const-call.c (fold_const_reduction): Likewise.
4807         * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4808         (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4809         (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4810         (fold_relational_const): Likewise.
4811         (native_interpret_vector): Likewise.  Change the size from an
4812         int to an unsigned int.
4813         * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4814         TYPE_VECTOR_SUBPARTS.
4815         (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4816         (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4817         duplicating a non-constant operand into a variable-length vector.
4818         * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4819         TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4820         * ipa-icf.c (sem_variable::equals): Likewise.
4821         * match.pd: Likewise.
4822         * omp-simd-clone.c (simd_clone_subparts): Likewise.
4823         * print-tree.c (print_node): Likewise.
4824         * stor-layout.c (layout_type): Likewise.
4825         * targhooks.c (default_builtin_vectorization_cost): Likewise.
4826         * tree-cfg.c (verify_gimple_comparison): Likewise.
4827         (verify_gimple_assign_binary): Likewise.
4828         (verify_gimple_assign_ternary): Likewise.
4829         (verify_gimple_assign_single): Likewise.
4830         * tree-pretty-print.c (dump_generic_node): Likewise.
4831         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4832         (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4833         * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4834         (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4835         (vect_shift_permute_load_chain): Likewise.
4836         * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4837         (expand_vector_condition, optimize_vector_constructor): Likewise.
4838         (lower_vec_perm, get_compute_type): Likewise.
4839         * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4840         (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4841         * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4842         (vect_recog_mask_conversion_pattern): Likewise.
4843         * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4844         (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4845         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4846         (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4847         (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4848         (vectorizable_shift, vectorizable_operation, vectorizable_store)
4849         (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4850         (supportable_widening_operation): Likewise.
4851         (supportable_narrowing_operation): Likewise.
4852         * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4853         Likewise.
4854         * varasm.c (output_constant): Likewise.
4855
4856 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4857             Alan Hayward  <alan.hayward@arm.com>
4858             David Sherwood  <david.sherwood@arm.com>
4859
4860         * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4861         so that both the length == 3 and length != 3 cases set up their
4862         own permute vectors.  Add comments explaining why we know the
4863         number of elements is constant.
4864         (vect_permute_load_chain): Likewise.
4865
4866 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4867             Alan Hayward  <alan.hayward@arm.com>
4868             David Sherwood  <david.sherwood@arm.com>
4869
4870         * machmode.h (mode_nunits): Change from unsigned char to
4871         poly_uint16_pod.
4872         (ONLY_FIXED_SIZE_MODES): New macro.
4873         (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4874         (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4875         (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4876         New typedefs.
4877         (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4878         (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4879         or if measurement_type is not polynomial.
4880         * genmodes.c (ZERO_COEFFS): New macro.
4881         (emit_mode_nunits_inline): Make mode_nunits_inline return a
4882         poly_uint16.
4883         (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4884         Use ZERO_COEFFS when emitting initializers.
4885         * data-streamer.h (bp_pack_poly_value): New function.
4886         (bp_unpack_poly_value): Likewise.
4887         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4888         for GET_MODE_NUNITS.
4889         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4890         for GET_MODE_NUNITS.
4891         * tree.c (make_vector_type): Remove temporary shim and make
4892         the real function take the number of units as a poly_uint64
4893         rather than an int.
4894         (build_vector_type_for_mode): Handle polynomial nunits.
4895         * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4896         * emit-rtl.c (const_vec_series_p_1): Likewise.
4897         (gen_rtx_CONST_VECTOR): Likewise.
4898         * fold-const.c (test_vec_duplicate_folding): Likewise.
4899         * genrecog.c (validate_pattern): Likewise.
4900         * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4901         * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4902         * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4903         (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4904         (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4905         * rtlanal.c (subreg_get_info): Likewise.
4906         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4907         (vect_grouped_load_supported): Likewise.
4908         * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4909         * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4910         * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4911         (simplify_const_unary_operation, simplify_binary_operation_1)
4912         (simplify_const_binary_operation, simplify_ternary_operation)
4913         (test_vector_ops_duplicate, test_vector_ops): Likewise.
4914         (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4915         instead of CONST_VECTOR_NUNITS.
4916         * varasm.c (output_constant_pool_2): Likewise.
4917         * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4918         explicit-encoded elements in the XVEC for variable-length vectors.
4919
4920 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4921
4922         * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4923
4924 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4925             Alan Hayward  <alan.hayward@arm.com>
4926             David Sherwood  <david.sherwood@arm.com>
4927
4928         * coretypes.h (fixed_size_mode): Declare.
4929         (fixed_size_mode_pod): New typedef.
4930         * builtins.h (target_builtins::x_apply_args_mode)
4931         (target_builtins::x_apply_result_mode): Change type to
4932         fixed_size_mode_pod.
4933         * builtins.c (apply_args_size, apply_result_size, result_vector)
4934         (expand_builtin_apply_args_1, expand_builtin_apply)
4935         (expand_builtin_return): Update accordingly.
4936
4937 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
4938
4939         * cse.c (hash_rtx_cb): Hash only the encoded elements.
4940         * cselib.c (cselib_hash_rtx): Likewise.
4941         * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4942         CONST_VECTOR encoding.
4943
4944 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
4945             Jeff Law  <law@redhat.com>
4946
4947         PR target/83641
4948         * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4949         noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4950         only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4951         and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4952
4953         PR target/83641
4954         * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4955         explicitly probe *sp in a noreturn function if there were any callee
4956         register saves or frame pointer is needed.
4957
4958 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
4959
4960         PR debug/83621
4961         * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4962         BLKmode for ternary, binary or unary expressions.
4963
4964         PR debug/83645
4965         * var-tracking.c (delete_vta_debug_insn): New inline function.
4966         (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4967         insns from get_insns () to NULL instead of each bb separately.
4968         Use delete_vta_debug_insn.  No longer static.
4969         (vt_debug_insns_local, variable_tracking_main_1): Adjust
4970         delete_vta_debug_insns callers.
4971         * rtl.h (delete_vta_debug_insns): Declare.
4972         * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4973         instead of variable_tracking_main.
4974
4975 2018-01-03  Martin Sebor  <msebor@redhat.com>
4976
4977         PR tree-optimization/83603
4978         * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4979         arguments past the endof the argument list in functions declared
4980         without a prototype.
4981         * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4982         Avoid checking when arguments are null.
4983
4984 2018-01-03  Martin Sebor  <msebor@redhat.com>
4985
4986         PR c/83559
4987         * doc/extend.texi (attribute const): Fix a typo.
4988         * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4989         issuing -Wsuggest-attribute for void functions.
4990
4991 2018-01-03  Martin Sebor  <msebor@redhat.com>
4992
4993         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4994         offset_int::from instead of wide_int::to_shwi.
4995         (maybe_diag_overlap): Remove assertion.
4996         Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4997         * gimple-ssa-sprintf.c (format_directive): Same.
4998         (parse_directive): Same.
4999         (sprintf_dom_walker::compute_format_length): Same.
5000         (try_substitute_return_value): Same.
5001
5002 2018-01-03  Jeff Law  <law@redhat.com>
5003
5004         PR middle-end/83654
5005         * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5006         non-constant residual for zero at runtime and avoid probing in
5007         that case.  Reorganize code for trailing problem to mirror handling
5008         of the residual.
5009
5010 2018-01-03  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
5011
5012         PR tree-optimization/83501
5013         * tree-ssa-strlen.c (get_string_cst): New.
5014         (handle_char_store): Call get_string_cst.
5015
5016 2018-01-03  Martin Liska  <mliska@suse.cz>
5017
5018         PR tree-optimization/83593
5019         * tree-ssa-strlen.c: Include tree-cfg.h.
5020         (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5021         (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5022         (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5023         to false.
5024         (strlen_dom_walker::before_dom_children): Call
5025         gimple_purge_dead_eh_edges. Dump tranformation with details
5026         dump flags.
5027         (strlen_dom_walker::before_dom_children): Update call by adding
5028         new argument cleanup_eh.
5029         (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5030
5031 2018-01-03  Martin Liska  <mliska@suse.cz>
5032
5033         PR ipa/83549
5034         * cif-code.def (VARIADIC_THUNK): New enum value.
5035         * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5036         thunks.
5037
5038 2018-01-03  Jan Beulich  <jbeulich@suse.com>
5039
5040         * sse.md (mov<mode>_internal): Tighten condition for when to use
5041         vmovdqu<ssescalarsize> for TI and OI modes.
5042
5043 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
5044
5045         Update copyright years.
5046
5047 2018-01-03  Martin Liska  <mliska@suse.cz>
5048
5049         PR ipa/83594
5050         * ipa-visibility.c (function_and_variable_visibility): Skip
5051         functions with noipa attribure.
5052
5053 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
5054
5055         * gcc.c (process_command): Update copyright notice dates.
5056         * gcov-dump.c (print_version): Ditto.
5057         * gcov.c (print_version): Ditto.
5058         * gcov-tool.c (print_version): Ditto.
5059         * gengtype.c (create_file): Ditto.
5060         * doc/cpp.texi: Bump @copying's copyright year.
5061         * doc/cppinternals.texi: Ditto.
5062         * doc/gcc.texi: Ditto.
5063         * doc/gccint.texi: Ditto.
5064         * doc/gcov.texi: Ditto.
5065         * doc/install.texi: Ditto.
5066         * doc/invoke.texi: Ditto.
5067
5068 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5069
5070         * vector-builder.h (vector_builder::m_full_nelts): Change from
5071         unsigned int to poly_uint64.
5072         (vector_builder::full_nelts): Update prototype accordingly.
5073         (vector_builder::new_vector): Likewise.
5074         (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5075         (vector_builder::operator ==): Likewise.
5076         (vector_builder::finalize): Likewise.
5077         * int-vector-builder.h (int_vector_builder::int_vector_builder):
5078         Take the number of elements as a poly_uint64 rather than an
5079         unsigned int.
5080         * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5081         from unsigned int to poly_uint64.
5082         (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5083         (vec_perm_indices::new_vector): Likewise.
5084         (vec_perm_indices::length): Likewise.
5085         (vec_perm_indices::nelts_per_input): Likewise.
5086         (vec_perm_indices::input_nelts): Likewise.
5087         * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5088         number of elements per input as a poly_uint64 rather than an
5089         unsigned int.  Use the original encoding for variable-length
5090         vectors, rather than clamping each individual element.
5091         For the second and subsequent elements in each pattern,
5092         clamp the step and base before clamping their sum.
5093         (vec_perm_indices::series_p): Handle polynomial element counts.
5094         (vec_perm_indices::all_in_range_p): Likewise.
5095         (vec_perm_indices_to_tree): Likewise.
5096         (vec_perm_indices_to_rtx): Likewise.
5097         * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5098         * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5099         (tree_vector_builder::new_binary_operation): Handle polynomial
5100         element counts.  Return false if we need to know the number
5101         of elements at compile time.
5102         * fold-const.c (fold_vec_perm): Punt if the number of elements
5103         isn't known at compile time.
5104
5105 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5106
5107         * vec-perm-indices.h (vec_perm_builder): Change element type
5108         from HOST_WIDE_INT to poly_int64.
5109         (vec_perm_indices::element_type): Update accordingly.
5110         (vec_perm_indices::clamp): Handle polynomial element_types.
5111         * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5112         (vec_perm_indices::all_in_range_p): Likewise.
5113         (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5114         than shwi trees.
5115         * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5116         polynomial vec_perm_indices element types.
5117         * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5118         * fold-const.c (fold_vec_perm): Likewise.
5119         * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5120         * tree-vect-generic.c (lower_vec_perm): Likewise.
5121         * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5122         * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5123         element type to HOST_WIDE_INT.
5124
5125 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5126             Alan Hayward  <alan.hayward@arm.com>
5127             David Sherwood  <david.sherwood@arm.com>
5128
5129         * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5130         rather than an int.  Use plus_constant.
5131         (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5132         Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5133
5134 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5135             Alan Hayward  <alan.hayward@arm.com>
5136             David Sherwood  <david.sherwood@arm.com>
5137
5138         * calls.c (emit_call_1, expand_call): Change struct_value_size from
5139         a HOST_WIDE_INT to a poly_int64.
5140
5141 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5142             Alan Hayward  <alan.hayward@arm.com>
5143             David Sherwood  <david.sherwood@arm.com>
5144
5145         * calls.c (load_register_parameters): Cope with polynomial
5146         mode sizes.  Require a constant size for BLKmode parameters
5147         that aren't described by a PARALLEL.  If BLOCK_REG_PADDING
5148         forces a parameter to be padded at the lsb end in order to
5149         fill a complete number of words, require the parameter size
5150         to be ordered wrt UNITS_PER_WORD.
5151
5152 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5153             Alan Hayward  <alan.hayward@arm.com>
5154             David Sherwood  <david.sherwood@arm.com>
5155
5156         * reload1.c (spill_stack_slot_width): Change element type
5157         from unsigned int to poly_uint64_pod.
5158         (alter_reg): Treat mode sizes as polynomial.
5159
5160 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5161             Alan Hayward  <alan.hayward@arm.com>
5162             David Sherwood  <david.sherwood@arm.com>
5163
5164         * reload.c (complex_word_subreg_p): New function.
5165         (reload_inner_reg_of_subreg, push_reload): Use it.
5166
5167 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5168             Alan Hayward  <alan.hayward@arm.com>
5169             David Sherwood  <david.sherwood@arm.com>
5170
5171         * lra-constraints.c (process_alt_operands): Reject matched
5172         operands whose sizes aren't ordered.
5173         (match_reload): Refer to this check here.
5174
5175 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5176             Alan Hayward  <alan.hayward@arm.com>
5177             David Sherwood  <david.sherwood@arm.com>
5178
5179         * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5180         that the mode size is in the set {1, 2, 4, 8, 16}.
5181
5182 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5183             Alan Hayward  <alan.hayward@arm.com>
5184             David Sherwood  <david.sherwood@arm.com>
5185
5186         * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5187         Use plus_constant instead of gen_rtx_PLUS.
5188
5189 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5190             Alan Hayward  <alan.hayward@arm.com>
5191             David Sherwood  <david.sherwood@arm.com>
5192
5193         * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5194         * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5195         * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5196         * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5197         * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5198         * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5199         * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5200         * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5201         * config/i386/i386.c (ix86_push_rounding): ...this new function.
5202         * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5203         a poly_int64.
5204         * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5205         * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5206         * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5207         * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5208         * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5209         * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5210         * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5211         * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5212         * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5213         * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5214         function.
5215         * expr.c (emit_move_resolve_push): Treat the input and result
5216         of PUSH_ROUNDING as a poly_int64.
5217         (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5218         (emit_push_insn): Likewise.
5219         * lra-eliminations.c (mark_not_eliminable): Likewise.
5220         * recog.c (push_operand): Likewise.
5221         * reload1.c (elimination_effects): Likewise.
5222         * rtlanal.c (nonzero_bits1): Likewise.
5223         * calls.c (store_one_arg): Likewise.  Require the padding to be
5224         known at compile time.
5225
5226 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5227             Alan Hayward  <alan.hayward@arm.com>
5228             David Sherwood  <david.sherwood@arm.com>
5229
5230         * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5231         Use plus_constant instead of gen_rtx_PLUS.
5232
5233 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5234             Alan Hayward  <alan.hayward@arm.com>
5235             David Sherwood  <david.sherwood@arm.com>
5236
5237         * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5238         rather than an int.
5239
5240 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5241             Alan Hayward  <alan.hayward@arm.com>
5242             David Sherwood  <david.sherwood@arm.com>
5243
5244         * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5245         instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5246         via stack temporaries.  Treat the mode size as polynomial too.
5247
5248 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5249             Alan Hayward  <alan.hayward@arm.com>
5250             David Sherwood  <david.sherwood@arm.com>
5251
5252         * expr.c (expand_expr_real_2): When handling conversions involving
5253         unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5254         multiplying int_size_in_bytes by BITS_PER_UNIT.  Treat GET_MODE_BISIZE
5255         as a poly_uint64 too.
5256
5257 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5258             Alan Hayward  <alan.hayward@arm.com>
5259             David Sherwood  <david.sherwood@arm.com>
5260
5261         * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5262
5263 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5264             Alan Hayward  <alan.hayward@arm.com>
5265             David Sherwood  <david.sherwood@arm.com>
5266
5267         * combine.c (can_change_dest_mode): Handle polynomial
5268         REGMODE_NATURAL_SIZE.
5269         * expmed.c (store_bit_field_1): Likewise.
5270         * expr.c (store_constructor): Likewise.
5271         * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5272         and polynomial REGMODE_NATURAL_SIZE.
5273         (gen_lowpart_common): Likewise.
5274         * reginfo.c (record_subregs_of_mode): Likewise.
5275         * rtlanal.c (read_modify_subreg_p): Likewise.
5276
5277 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5278             Alan Hayward  <alan.hayward@arm.com>
5279             David Sherwood  <david.sherwood@arm.com>
5280
5281         * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5282         numbers of elements.
5283
5284 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5285             Alan Hayward  <alan.hayward@arm.com>
5286             David Sherwood  <david.sherwood@arm.com>
5287
5288         * match.pd: Cope with polynomial numbers of vector elements.
5289
5290 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5291             Alan Hayward  <alan.hayward@arm.com>
5292             David Sherwood  <david.sherwood@arm.com>
5293
5294         * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5295         in a POINTER_PLUS_EXPR.
5296
5297 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5298             Alan Hayward  <alan.hayward@arm.com>
5299             David Sherwood  <david.sherwood@arm.com>
5300
5301         * omp-simd-clone.c (simd_clone_subparts): New function.
5302         (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5303         (ipa_simd_modify_function_body): Likewise.
5304
5305 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5306             Alan Hayward  <alan.hayward@arm.com>
5307             David Sherwood  <david.sherwood@arm.com>
5308
5309         * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5310         (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5311         (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5312         (expand_vector_condition, vector_element): Likewise.
5313         (subparts_gt): New function.
5314         (get_compute_type): Use subparts_gt.
5315         (count_type_subparts): Delete.
5316         (expand_vector_operations_1): Use subparts_gt instead of
5317         count_type_subparts.
5318
5319 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5320             Alan Hayward  <alan.hayward@arm.com>
5321             David Sherwood  <david.sherwood@arm.com>
5322
5323         * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5324         (vect_compile_time_alias): ...this new function.  Do the calculation
5325         on poly_ints rather than trees.
5326         (vect_prune_runtime_alias_test_list): Update call accordingly.
5327
5328 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5329             Alan Hayward  <alan.hayward@arm.com>
5330             David Sherwood  <david.sherwood@arm.com>
5331
5332         * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5333         numbers of units.
5334         (vect_schedule_slp_instance): Likewise.
5335
5336 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5337             Alan Hayward  <alan.hayward@arm.com>
5338             David Sherwood  <david.sherwood@arm.com>
5339
5340         * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5341         constant and extern definitions for variable-length vectors.
5342         (vect_get_constant_vectors): Note that the number of units
5343         is known to be constant.
5344
5345 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5346             Alan Hayward  <alan.hayward@arm.com>
5347             David Sherwood  <david.sherwood@arm.com>
5348
5349         * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5350         of units as polynomial.  Choose between WIDE and NARROW based
5351         on multiple_p.
5352
5353 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5354             Alan Hayward  <alan.hayward@arm.com>
5355             David Sherwood  <david.sherwood@arm.com>
5356
5357         * tree-vect-stmts.c (simd_clone_subparts): New function.
5358         (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5359
5360 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5361             Alan Hayward  <alan.hayward@arm.com>
5362             David Sherwood  <david.sherwood@arm.com>
5363
5364         * tree-vect-stmts.c (vectorizable_call): Treat the number of
5365         vectors as polynomial.  Use build_index_vector for
5366         IFN_GOMP_SIMD_LANE.
5367
5368 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5369             Alan Hayward  <alan.hayward@arm.com>
5370             David Sherwood  <david.sherwood@arm.com>
5371
5372         * tree-vect-stmts.c (get_load_store_type): Treat the number of
5373         units as polynomial.  Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5374         for variable-length vectors.
5375         (vectorizable_mask_load_store): Treat the number of units as
5376         polynomial, asserting that it is constant if the condition has
5377         already been enforced.
5378         (vectorizable_store, vectorizable_load): Likewise.
5379
5380 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5381             Alan Hayward  <alan.hayward@arm.com>
5382             David Sherwood  <david.sherwood@arm.com>
5383
5384         * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5385         of units as polynomial.  Punt if we can't tell at compile time
5386         which vector contains the final result.
5387
5388 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5389             Alan Hayward  <alan.hayward@arm.com>
5390             David Sherwood  <david.sherwood@arm.com>
5391
5392         * tree-vect-loop.c (vectorizable_induction): Treat the number
5393         of units as polynomial.  Punt on SLP inductions.  Use an integer
5394         VEC_SERIES_EXPR for variable-length integer reductions.  Use a
5395         cast of such a series for variable-length floating-point
5396         reductions.
5397
5398 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5399             Alan Hayward  <alan.hayward@arm.com>
5400             David Sherwood  <david.sherwood@arm.com>
5401
5402         * tree.h (build_index_vector): Declare.
5403         * tree.c (build_index_vector): New function.
5404         * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5405         of units as polynomial, forcibly converting it to a constant if
5406         vectorizable_reduction has already enforced the condition.
5407         (vect_create_epilog_for_reduction): Likewise.  Use build_index_vector
5408         to create a {1,2,3,...} vector.
5409         (vectorizable_reduction): Treat the number of units as polynomial.
5410         Choose vectype_in based on the largest scalar element size rather
5411         than the smallest number of units.  Enforce the restrictions
5412         relied on above.
5413
5414 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5415             Alan Hayward  <alan.hayward@arm.com>
5416             David Sherwood  <david.sherwood@arm.com>
5417
5418         * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5419         number of units as polynomial.
5420
5421 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5422             Alan Hayward  <alan.hayward@arm.com>
5423             David Sherwood  <david.sherwood@arm.com>
5424
5425         * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5426         * target.def (autovectorize_vector_sizes): Return the vector sizes
5427         by pointer, using vector_sizes rather than a bitmask.
5428         * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5429         * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5430         * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5431         Likewise.
5432         * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5433         * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5434         * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5435         * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5436         * omp-general.c (omp_max_vf): Likewise.
5437         * omp-low.c (omp_clause_aligned_alignment): Likewise.
5438         * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5439         * tree-vect-loop.c (vect_analyze_loop): Likewise.
5440         * tree-vect-slp.c (vect_slp_bb): Likewise.
5441         * doc/tm.texi: Regenerate.
5442         * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5443         to a poly_uint64.
5444         * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5445         the vector size as a poly_uint64 rather than an unsigned int.
5446         (current_vector_size): Change from an unsigned int to a poly_uint64.
5447         (get_vectype_for_scalar_type): Update accordingly.
5448         * tree.h (build_truth_vector_type): Take the size and number of
5449         units as a poly_uint64 rather than an unsigned int.
5450         (build_vector_type): Add a temporary overload that takes
5451         the number of units as a poly_uint64 rather than an unsigned int.
5452         * tree.c (make_vector_type): Likewise.
5453         (build_truth_vector_type): Take the number of units as a poly_uint64
5454         rather than an unsigned int.
5455
5456 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5457             Alan Hayward  <alan.hayward@arm.com>
5458             David Sherwood  <david.sherwood@arm.com>
5459
5460         * target.def (get_mask_mode): Take the number of units and length
5461         as poly_uint64s rather than unsigned ints.
5462         * targhooks.h (default_get_mask_mode): Update accordingly.
5463         * targhooks.c (default_get_mask_mode): Likewise.
5464         * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5465         * doc/tm.texi: Regenerate.
5466
5467 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5468             Alan Hayward  <alan.hayward@arm.com>
5469             David Sherwood  <david.sherwood@arm.com>
5470
5471         * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5472         * omp-general.c (omp_max_vf): Likewise.
5473         * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5474         (expand_omp_simd): Handle polynomial safelen.
5475         * omp-low.c (omplow_simd_context): Add a default constructor.
5476         (omplow_simd_context::max_vf): Change from int to poly_uint64.
5477         (lower_rec_simd_input_clauses): Update accordingly.
5478         (lower_rec_input_clauses): Likewise.
5479
5480 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5481             Alan Hayward  <alan.hayward@arm.com>
5482             David Sherwood  <david.sherwood@arm.com>
5483
5484         * tree-vectorizer.h (vect_nunits_for_cost): New function.
5485         * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5486         * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5487         (vect_analyze_slp_cost): Likewise.
5488         * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5489         (vect_model_load_cost): Likewise.
5490
5491 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5492             Alan Hayward  <alan.hayward@arm.com>
5493             David Sherwood  <david.sherwood@arm.com>
5494
5495         * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5496         (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5497         from an unsigned int * to a poly_uint64_pod *.
5498         (calculate_unrolling_factor): New function.
5499         (vect_analyze_slp_instance): Use it.  Track polynomial max_nunits.
5500
5501 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5502             Alan Hayward  <alan.hayward@arm.com>
5503             David Sherwood  <david.sherwood@arm.com>
5504
5505         * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5506         from an unsigned int to a poly_uint64.
5507         (_loop_vec_info::slp_unrolling_factor): Likewise.
5508         (_loop_vec_info::vectorization_factor): Change from an int
5509         to a poly_uint64.
5510         (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5511         (vect_get_num_vectors): New function.
5512         (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5513         (vect_get_num_copies): Use vect_get_num_vectors.
5514         (vect_analyze_data_ref_dependences): Change max_vf from an int *
5515         to an unsigned int *.
5516         (vect_analyze_data_refs): Change min_vf from an int * to a
5517         poly_uint64 *.
5518         (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5519         than an unsigned HOST_WIDE_INT.
5520         * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5521         (vect_analyze_data_ref_dependence): Change max_vf from an int *
5522         to an unsigned int *.
5523         (vect_analyze_data_ref_dependences): Likewise.
5524         (vect_compute_data_ref_alignment): Handle polynomial vf.
5525         (vect_enhance_data_refs_alignment): Likewise.
5526         (vect_prune_runtime_alias_test_list): Likewise.
5527         (vect_shift_permute_load_chain): Likewise.
5528         (vect_supportable_dr_alignment): Likewise.
5529         (dependence_distance_ge_vf): Take the vectorization factor as a
5530         poly_uint64 rather than an unsigned HOST_WIDE_INT.
5531         (vect_analyze_data_refs): Change min_vf from an int * to a
5532         poly_uint64 *.
5533         * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5534         vfm1 as a poly_uint64 rather than an int.  Make the same change
5535         for the returned bound_scalar.
5536         (vect_gen_vector_loop_niters): Handle polynomial vf.
5537         (vect_do_peeling): Likewise.  Update call to
5538         vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5539         (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5540         be constant.
5541         * tree-vect-loop.c (vect_determine_vectorization_factor)
5542         (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5543         (vect_get_known_peeling_cost): Likewise.
5544         (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5545         (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5546         (vect_transform_loop): Likewise.  Use the lowest possible VF when
5547         updating the upper bounds of the loop.
5548         (vect_min_worthwhile_factor): Make static.  Return an unsigned int
5549         rather than an int.
5550         * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5551         polynomial unroll factors.
5552         (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5553         (vect_make_slp_decision): Likewise.
5554         (vect_supported_load_permutation_p): Likewise, and polynomial
5555         vf too.
5556         (vect_analyze_slp_cost): Handle polynomial vf.
5557         (vect_slp_analyze_node_operations): Likewise.
5558         (vect_slp_analyze_bb_1): Likewise.
5559         (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5560         than an unsigned HOST_WIDE_INT.
5561         * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5562         (vectorizable_load): Handle polynomial vf.
5563         * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5564         a poly_uint64.
5565         (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5566
5567 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5568             Alan Hayward  <alan.hayward@arm.com>
5569             David Sherwood  <david.sherwood@arm.com>
5570
5571         * match.pd: Handle bit operations involving three constants
5572         and try to fold one pair.
5573
5574 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
5575
5576         * tree-vect-loop-manip.c: Include gimple-fold.h.
5577         (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5578         niters_maybe_zero parameters.  Handle other cases besides a step of 1.
5579         (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5580         Add a path that uses a step of VF instead of 1, but disable it
5581         for now.
5582         (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5583         and niters_no_overflow parameters.  Update calls to
5584         slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5585         Create a new SSA name if the latter choses to use a ste other
5586         than zero, and return it via niters_vector_mult_vf_var.
5587         * tree-vect-loop.c (vect_transform_loop): Update calls to
5588         vect_do_peeling, vect_gen_vector_loop_niters and
5589         slpeel_make_loop_iterate_ntimes.
5590         * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5591         (vect_gen_vector_loop_niters): Update declarations after above changes.
5592
5593 2018-01-02  Michael Meissner  <meissner@linux.vnet.ibm.com>
5594
5595         * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5596         128-bit round to integer instructions.
5597         (ceil<mode>2): Likewise.
5598         (btrunc<mode>2): Likewise.
5599         (round<mode>2): Likewise.
5600
5601 2018-01-02  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
5602
5603         * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5604         unaligned VSX load/store on P8/P9.
5605         (expand_block_clear): Allow the use of unaligned VSX
5606         load/store on P8/P9.
5607
5608 2018-01-02  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
5609
5610         * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5611         New function.
5612         (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5613         swap associated with both a load and a store.
5614
5615 2018-01-02  Andrew Waterman  <andrew@sifive.com>
5616
5617         * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5618         * config/riscv/riscv.md (clear_cache): Use it.
5619
5620 2018-01-02  Artyom Skrobov  <tyomitch@gmail.com>
5621
5622         * web.c: Remove out-of-date comment.
5623
5624 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
5625
5626         * expr.c (fixup_args_size_notes): Check that any existing
5627         REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5628         (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5629         (emit_single_push_insn): ...here.
5630
5631 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
5632
5633         * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5634         (const_vector_encoded_nelts): New function.
5635         (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5636         (const_vector_int_elt, const_vector_elt): Declare.
5637         * emit-rtl.c (const_vector_int_elt_1): New function.
5638         (const_vector_elt): Likewise.
5639         * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5640         of CONST_VECTOR_ELT.
5641
5642 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
5643
5644         * expr.c: Include rtx-vector-builder.h.
5645         (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5646         directly on the tree encoding.
5647         (const_vector_from_tree): Likewise.
5648         * optabs.c: Include rtx-vector-builder.h.
5649         (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5650         sequence of "u" values.
5651         * vec-perm-indices.c: Include rtx-vector-builder.h.
5652         (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5653         directly on the vec_perm_indices encoding.
5654
5655 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
5656
5657         * doc/rtl.texi (const_vector): Describe new encoding scheme.
5658         * Makefile.in (OBJS): Add rtx-vector-builder.o.
5659         * rtx-vector-builder.h: New file.
5660         * rtx-vector-builder.c: Likewise.
5661         * rtl.h (rtx_def::u2): Add a const_vector field.
5662         (CONST_VECTOR_NPATTERNS): New macro.
5663         (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5664         (CONST_VECTOR_DUPLICATE_P): Likewise.
5665         (CONST_VECTOR_STEPPED_P): Likewise.
5666         (CONST_VECTOR_ENCODED_ELT): Likewise.
5667         (const_vec_duplicate_p): Check for a duplicated vector encoding.
5668         (unwrap_const_vec_duplicate): Likewise.
5669         (const_vec_series_p): Check for a non-duplicated vector encoding.
5670         Say that the function only returns true for integer vectors.
5671         * emit-rtl.c: Include rtx-vector-builder.h.
5672         (gen_const_vec_duplicate_1): Delete.
5673         (gen_const_vector): Call gen_const_vec_duplicate instead of
5674         gen_const_vec_duplicate_1.
5675         (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5676         (gen_const_vec_duplicate): Use rtx_vector_builder.
5677         (gen_const_vec_series): Likewise.
5678         (gen_rtx_CONST_VECTOR): Likewise.
5679         * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5680         (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5681         Build a new vector rather than modifying a CONST_VECTOR in-place.
5682         (handle_special_swappables): Update call accordingly.
5683         * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5684         (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5685         Build a new vector rather than modifying a CONST_VECTOR in-place.
5686         (handle_special_swappables): Update call accordingly.
5687
5688 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
5689
5690         * simplify-rtx.c (simplify_const_binary_operation): Use
5691         CONST_VECTOR_ELT instead of XVECEXP.
5692
5693 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
5694
5695         * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5696         the selector elements to be different from the data elements
5697         if the selector is a VECTOR_CST.
5698         * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5699         ssizetype for the selector.
5700
5701 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
5702
5703         * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5704         before testing each element individually.
5705         * tree-vect-generic.c (lower_vec_perm): Likewise.
5706
5707 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
5708
5709         * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5710         * selftest-run-tests.c (selftest::run_tests): Call it.
5711         * vector-builder.h (vector_builder::operator ==): New function.
5712         (vector_builder::operator !=): Likewise.
5713         * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5714         (vec_perm_indices::all_from_input_p): New function.
5715         * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5716         (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5717         * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5718         instead of reading the VECTOR_CST directly.  Detect whether both
5719         vector inputs are the same before constructing the vec_perm_indices,
5720         and update the number of inputs argument accordingly.  Use the
5721         utility functions added above.  Only construct sel2 if we need to.
5722
5723 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
5724
5725         * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5726         the broadcast of the low byte.
5727         (expand_mult_highpart): Use an explicit encoding for the permutes.
5728         * optabs-query.c (can_mult_highpart_p): Likewise.
5729         * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5730         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5731         (vectorizable_bswap): Likewise.
5732         * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5733         explicit encoding for the power-of-2 permutes.
5734         (vect_permute_store_chain): Likewise.
5735         (vect_grouped_load_supported): Likewise.
5736         (vect_permute_load_chain): Likewise.
5737
5738 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
5739
5740         * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5741         * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5742         * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5743         * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5744         * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5745         (vect_gen_perm_mask_any): Likewise.
5746
5747 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
5748
5749         * int-vector-builder.h: New file.
5750         * vec-perm-indices.h: Include int-vector-builder.h.
5751         (vec_perm_indices): Redefine as an int_vector_builder.
5752         (auto_vec_perm_indices): Delete.
5753         (vec_perm_builder): Redefine as a stand-alone class.
5754         (vec_perm_indices::vec_perm_indices): New function.
5755         (vec_perm_indices::clamp): Likewise.
5756         * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5757         (vec_perm_indices::new_vector): New function.
5758         (vec_perm_indices::new_expanded_vector): Update for new
5759         vec_perm_indices class.
5760         (vec_perm_indices::rotate_inputs): New function.
5761         (vec_perm_indices::all_in_range_p): Operate directly on the
5762         encoded form, without computing elided elements.
5763         (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5764         encoding.  Update for new vec_perm_indices class.
5765         * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5766         the given vec_perm_builder.
5767         (expand_vec_perm_var): Update vec_perm_builder constructor.
5768         (expand_mult_highpart): Use vec_perm_builder instead of
5769         auto_vec_perm_indices.
5770         * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5771         vec_perm_indices instead of auto_vec_perm_indices.  Use a single
5772         or double series encoding as appropriate.
5773         * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5774         vec_perm_indices instead of auto_vec_perm_indices.
5775         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5776         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5777         (vect_permute_store_chain): Likewise.
5778         (vect_grouped_load_supported): Likewise.
5779         (vect_permute_load_chain): Likewise.
5780         (vect_shift_permute_load_chain): Likewise.
5781         * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5782         (vect_transform_slp_perm_load): Likewise.
5783         (vect_schedule_slp_instance): Likewise.
5784         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5785         (vectorizable_mask_load_store): Likewise.
5786         (vectorizable_bswap): Likewise.
5787         (vectorizable_store): Likewise.
5788         (vectorizable_load): Likewise.
5789         * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5790         vec_perm_indices instead of auto_vec_perm_indices.  Use
5791         tree_to_vec_perm_builder to read the vector from a tree.
5792         * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5793         vec_perm_builder instead of a vec_perm_indices.
5794         (have_whole_vector_shift): Use vec_perm_builder and
5795         vec_perm_indices instead of auto_vec_perm_indices.  Leave the
5796         truncation to calc_vec_perm_mask_for_shift.
5797         (vect_create_epilog_for_reduction): Likewise.
5798         * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5799         from auto_vec_perm_indices to vec_perm_indices.
5800         (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5801         instead of changing individual elements.
5802         (aarch64_vectorize_vec_perm_const): Use new_vector to install
5803         the vector in d.perm.
5804         * config/arm/arm.c (expand_vec_perm_d::perm): Change
5805         from auto_vec_perm_indices to vec_perm_indices.
5806         (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5807         instead of changing individual elements.
5808         (arm_vectorize_vec_perm_const): Use new_vector to install
5809         the vector in d.perm.
5810         * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5811         Update vec_perm_builder constructor.
5812         (rs6000_expand_interleave): Likewise.
5813         * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5814         (rs6000_expand_interleave): Likewise.
5815
5816 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
5817
5818         * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5819         to qimode could truncate the indices.
5820         * optabs.c (expand_vec_perm_var): Likewise.
5821
5822 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
5823
5824         * Makefile.in (OBJS): Add vec-perm-indices.o.
5825         * vec-perm-indices.h: New file.
5826         * vec-perm-indices.c: Likewise.
5827         * target.h (vec_perm_indices): Replace with a forward class
5828         declaration.
5829         (auto_vec_perm_indices): Move to vec-perm-indices.h.
5830         * optabs.h: Include vec-perm-indices.h.
5831         (expand_vec_perm): Delete.
5832         (selector_fits_mode_p, expand_vec_perm_var): Declare.
5833         (expand_vec_perm_const): Declare.
5834         * target.def (vec_perm_const_ok): Replace with...
5835         (vec_perm_const): ...this new hook.
5836         * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5837         (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5838         * doc/tm.texi: Regenerate.
5839         * optabs.def (vec_perm_const): Delete.
5840         * doc/md.texi (vec_perm_const): Likewise.
5841         (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5842         * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5843         expand_vec_perm for constant permutation vectors.  Assert that
5844         the mode of variable permutation vectors is the integer equivalent
5845         of the mode that is being permuted.
5846         * optabs-query.h (selector_fits_mode_p): Declare.
5847         * optabs-query.c: Include vec-perm-indices.h.
5848         (selector_fits_mode_p): New function.
5849         (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5850         is defined, instead of checking whether the vec_perm_const_optab
5851         exists.  Use targetm.vectorize.vec_perm_const instead of
5852         targetm.vectorize.vec_perm_const_ok.  Check whether the indices
5853         fit in the vector mode before using a variable permute.
5854         * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5855         vec_perm_indices instead of an rtx.
5856         (expand_vec_perm): Replace with...
5857         (expand_vec_perm_const): ...this new function.  Take the selector
5858         as a vec_perm_indices rather than an rtx.  Also take the mode of
5859         the selector.  Update call to shift_amt_for_vec_perm_mask.
5860         Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5861         Use vec_perm_indices::new_expanded_vector to expand the original
5862         selector into bytes.  Check whether the indices fit in the vector
5863         mode before using a variable permute.
5864         (expand_vec_perm_var): Make global.
5865         (expand_mult_highpart): Use expand_vec_perm_const.
5866         * fold-const.c: Includes vec-perm-indices.h.
5867         * tree-ssa-forwprop.c: Likewise.
5868         * tree-vect-data-refs.c: Likewise.
5869         * tree-vect-generic.c: Likewise.
5870         * tree-vect-loop.c: Likewise.
5871         * tree-vect-slp.c: Likewise.
5872         * tree-vect-stmts.c: Likewise.
5873         * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5874         Delete.
5875         * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5876         * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5877         (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5878         (aarch64_vectorize_vec_perm_const): ...this new function.
5879         (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5880         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5881         * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5882         * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5883         * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5884         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5885         (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5886         into...
5887         (arm_vectorize_vec_perm_const): ...this new function.  Explicitly
5888         check for NEON modes.
5889         * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5890         * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5891         * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5892         (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5893         into...
5894         (ix86_vectorize_vec_perm_const): ...this new function.  Incorporate
5895         the old VEC_PERM_CONST conditions.
5896         * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5897         * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5898         * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5899         (ia64_vectorize_vec_perm_const_ok): Merge into...
5900         (ia64_vectorize_vec_perm_const): ...this new function.
5901         * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5902         * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5903         * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5904         * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5905         * config/mips/mips.c (mips_expand_vec_perm_const)
5906         (mips_vectorize_vec_perm_const_ok): Merge into...
5907         (mips_vectorize_vec_perm_const): ...this new function.
5908         * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5909         * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5910         * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5911         * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5912         * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5913         (rs6000_expand_vec_perm_const): Delete.
5914         * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5915         Delete.
5916         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5917         (altivec_expand_vec_perm_const_le): Take each operand individually.
5918         Operate on constant selectors rather than rtxes.
5919         (altivec_expand_vec_perm_const): Likewise.  Update call to
5920         altivec_expand_vec_perm_const_le.
5921         (rs6000_expand_vec_perm_const): Delete.
5922         (rs6000_vectorize_vec_perm_const_ok): Delete.
5923         (rs6000_vectorize_vec_perm_const): New function.
5924         (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5925         an element count and rtx array.
5926         (rs6000_expand_extract_even): Update call accordingly.
5927         (rs6000_expand_interleave): Likewise.
5928         * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5929         * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5930         * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5931         * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5932         (rs6000_expand_vec_perm_const): Delete.
5933         * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5934         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5935         (altivec_expand_vec_perm_const_le): Take each operand individually.
5936         Operate on constant selectors rather than rtxes.
5937         (altivec_expand_vec_perm_const): Likewise.  Update call to
5938         altivec_expand_vec_perm_const_le.
5939         (rs6000_expand_vec_perm_const): Delete.
5940         (rs6000_vectorize_vec_perm_const_ok): Delete.
5941         (rs6000_vectorize_vec_perm_const): New function.  Remove stray
5942         reference to the SPE evmerge intructions.
5943         (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5944         an element count and rtx array.
5945         (rs6000_expand_extract_even): Update call accordingly.
5946         (rs6000_expand_interleave): Likewise.
5947         * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5948         * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5949         new function.
5950         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5951
5952 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
5953
5954         * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5955         vector mode and that that mode matches the mode of the data
5956         being permuted.
5957         (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5958         out into expand_vec_perm_var.  Do all CONST_VECTOR handling here,
5959         directly using expand_vec_perm_1 when forcing selectors into
5960         registers.
5961         (expand_vec_perm_var): New function, split out from expand_vec_perm.
5962
5963 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
5964
5965         * optabs-query.h (can_vec_perm_p): Delete.
5966         (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5967         * optabs-query.c (can_vec_perm_p): Split into...
5968         (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5969         (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5970         particular selector is valid.
5971         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5972         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5973         (vect_grouped_load_supported): Likewise.
5974         (vect_shift_permute_load_chain): Likewise.
5975         * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5976         (vect_transform_slp_perm_load): Likewise.
5977         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5978         (vectorizable_bswap): Likewise.
5979         (vect_gen_perm_mask_checked): Likewise.
5980         * fold-const.c (fold_ternary_loc): Likewise.  Don't take
5981         implementations of variable permutation vectors into account
5982         when deciding which selector to use.
5983         * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5984         vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5985         with a false third argument.
5986         * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5987         to test whether the constant selector is valid and can_vec_perm_var_p
5988         to test whether a variable selector is valid.
5989
5990 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
5991
5992         * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5993         * optabs-query.c (can_vec_perm_p): Likewise.
5994         * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5995         instead of vec_perm_indices.
5996         * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5997         (vect_gen_perm_mask_checked): Likewise,
5998         * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5999         (vect_gen_perm_mask_checked): Likewise,
6000
6001 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
6002
6003         * optabs-query.h (qimode_for_vec_perm): Declare.
6004         * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6005         (qimode_for_vec_perm): ...this new function.
6006         * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6007
6008 2018-01-02  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
6009
6010         * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6011         does not have a conditional at the top.
6012
6013 2018-01-02  Richard Biener  <rguenther@suse.de>
6014
6015         * ipa-inline.c (big_speedup_p): Fix expression.
6016
6017 2018-01-02  Jan Hubicka  <hubicka@ucw.cz>
6018
6019         PR target/81616
6020         * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6021         for generic 4->6.
6022
6023 2018-01-02  Jan Hubicka  <hubicka@ucw.cz>
6024
6025         PR target/81616
6026         Generic tuning.
6027         * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6028         cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6029         and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6030         cond_taken_branch_cost 3->4.
6031
6032 2018-01-01  Jakub Jelinek  <jakub@redhat.com>
6033
6034         PR tree-optimization/83581
6035         * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6036         TODO_cleanup_cfg if any changes have been made.
6037
6038         PR middle-end/83608
6039         * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6040         convert_modes if target mode has the right side, but different mode
6041         class.
6042
6043         PR middle-end/83609
6044         * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6045         last argument when extracting from CONCAT.  If either from_real or
6046         from_imag is NULL, use expansion through memory.  If result is not
6047         a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6048         the parts directly to inner mode, if even that fails, use expansion
6049         through memory.
6050
6051         PR middle-end/83623
6052         * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6053         check for bswap in mode rather than HImode and use that in expand_unop
6054         too.
6055 \f
6056 Copyright (C) 2018 Free Software Foundation, Inc.
6057
6058 Copying and distribution of this file, with or without modification,
6059 are permitted in any medium without royalty provided the copyright
6060 notice and this notice are preserved.